xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 5a36ec735130f81a187ef65c4bc43c445a89568d)
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 enum rte_flow_item_type
118 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
119 {
120 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
121 	uint16_t ether_type = 0;
122 	uint16_t ether_type_m;
123 	uint8_t ip_next_proto = 0;
124 	uint8_t ip_next_proto_m;
125 
126 	if (item == NULL || item->spec == NULL)
127 		return ret;
128 	switch (item->type) {
129 	case RTE_FLOW_ITEM_TYPE_ETH:
130 		if (item->mask)
131 			ether_type_m = ((const struct rte_flow_item_eth *)
132 						(item->mask))->type;
133 		else
134 			ether_type_m = rte_flow_item_eth_mask.type;
135 		if (ether_type_m != RTE_BE16(0xFFFF))
136 			break;
137 		ether_type = ((const struct rte_flow_item_eth *)
138 				(item->spec))->type;
139 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
140 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
141 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
142 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
143 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
144 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
145 		else
146 			ret = RTE_FLOW_ITEM_TYPE_END;
147 		break;
148 	case RTE_FLOW_ITEM_TYPE_VLAN:
149 		if (item->mask)
150 			ether_type_m = ((const struct rte_flow_item_vlan *)
151 						(item->mask))->inner_type;
152 		else
153 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
154 		if (ether_type_m != RTE_BE16(0xFFFF))
155 			break;
156 		ether_type = ((const struct rte_flow_item_vlan *)
157 				(item->spec))->inner_type;
158 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
159 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
160 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
161 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
162 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
163 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
164 		else
165 			ret = RTE_FLOW_ITEM_TYPE_END;
166 		break;
167 	case RTE_FLOW_ITEM_TYPE_IPV4:
168 		if (item->mask)
169 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
170 					(item->mask))->hdr.next_proto_id;
171 		else
172 			ip_next_proto_m =
173 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
174 		if (ip_next_proto_m != 0xFF)
175 			break;
176 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
177 				(item->spec))->hdr.next_proto_id;
178 		if (ip_next_proto == IPPROTO_UDP)
179 			ret = RTE_FLOW_ITEM_TYPE_UDP;
180 		else if (ip_next_proto == IPPROTO_TCP)
181 			ret = RTE_FLOW_ITEM_TYPE_TCP;
182 		else if (ip_next_proto == IPPROTO_IP)
183 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
184 		else if (ip_next_proto == IPPROTO_IPV6)
185 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
186 		else
187 			ret = RTE_FLOW_ITEM_TYPE_END;
188 		break;
189 	case RTE_FLOW_ITEM_TYPE_IPV6:
190 		if (item->mask)
191 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
192 						(item->mask))->hdr.proto;
193 		else
194 			ip_next_proto_m =
195 				rte_flow_item_ipv6_mask.hdr.proto;
196 		if (ip_next_proto_m != 0xFF)
197 			break;
198 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
199 				(item->spec))->hdr.proto;
200 		if (ip_next_proto == IPPROTO_UDP)
201 			ret = RTE_FLOW_ITEM_TYPE_UDP;
202 		else if (ip_next_proto == IPPROTO_TCP)
203 			ret = RTE_FLOW_ITEM_TYPE_TCP;
204 		else if (ip_next_proto == IPPROTO_IP)
205 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
206 		else if (ip_next_proto == IPPROTO_IPV6)
207 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
208 		else
209 			ret = RTE_FLOW_ITEM_TYPE_END;
210 		break;
211 	default:
212 		ret = RTE_FLOW_ITEM_TYPE_VOID;
213 		break;
214 	}
215 	return ret;
216 }
217 
218 #define MLX5_RSS_EXP_ELT_N 16
219 
220 /**
221  * Expand RSS flows into several possible flows according to the RSS hash
222  * fields requested and the driver capabilities.
223  *
224  * @param[out] buf
225  *   Buffer to store the result expansion.
226  * @param[in] size
227  *   Buffer size in bytes. If 0, @p buf can be NULL.
228  * @param[in] pattern
229  *   User flow pattern.
230  * @param[in] types
231  *   RSS types to expand (see ETH_RSS_* definitions).
232  * @param[in] graph
233  *   Input graph to expand @p pattern according to @p types.
234  * @param[in] graph_root_index
235  *   Index of root node in @p graph, typically 0.
236  *
237  * @return
238  *   A positive value representing the size of @p buf in bytes regardless of
239  *   @p size on success, a negative errno value otherwise and rte_errno is
240  *   set, the following errors are defined:
241  *
242  *   -E2BIG: graph-depth @p graph is too deep.
243  */
244 static int
245 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
246 		     const struct rte_flow_item *pattern, uint64_t types,
247 		     const struct mlx5_flow_expand_node graph[],
248 		     int graph_root_index)
249 {
250 	const struct rte_flow_item *item;
251 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
252 	const int *next_node;
253 	const int *stack[MLX5_RSS_EXP_ELT_N];
254 	int stack_pos = 0;
255 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
256 	unsigned int i;
257 	size_t lsize;
258 	size_t user_pattern_size = 0;
259 	void *addr = NULL;
260 	const struct mlx5_flow_expand_node *next = NULL;
261 	struct rte_flow_item missed_item;
262 	int missed = 0;
263 	int elt = 0;
264 	const struct rte_flow_item *last_item = NULL;
265 
266 	memset(&missed_item, 0, sizeof(missed_item));
267 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
268 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
269 	if (lsize <= size) {
270 		buf->entry[0].priority = 0;
271 		buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
272 		buf->entries = 0;
273 		addr = buf->entry[0].pattern;
274 	}
275 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
276 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
277 			last_item = item;
278 		for (i = 0; node->next && node->next[i]; ++i) {
279 			next = &graph[node->next[i]];
280 			if (next->type == item->type)
281 				break;
282 		}
283 		if (next)
284 			node = next;
285 		user_pattern_size += sizeof(*item);
286 	}
287 	user_pattern_size += sizeof(*item); /* Handle END item. */
288 	lsize += user_pattern_size;
289 	/* Copy the user pattern in the first entry of the buffer. */
290 	if (lsize <= size) {
291 		rte_memcpy(addr, pattern, user_pattern_size);
292 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
293 		buf->entries = 1;
294 	}
295 	/* Start expanding. */
296 	memset(flow_items, 0, sizeof(flow_items));
297 	user_pattern_size -= sizeof(*item);
298 	/*
299 	 * Check if the last valid item has spec set, need complete pattern,
300 	 * and the pattern can be used for expansion.
301 	 */
302 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
303 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
304 		/* Item type END indicates expansion is not required. */
305 		return lsize;
306 	}
307 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
308 		next = NULL;
309 		missed = 1;
310 		for (i = 0; node->next && node->next[i]; ++i) {
311 			next = &graph[node->next[i]];
312 			if (next->type == missed_item.type) {
313 				flow_items[0].type = missed_item.type;
314 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
315 				break;
316 			}
317 			next = NULL;
318 		}
319 	}
320 	if (next && missed) {
321 		elt = 2; /* missed item + item end. */
322 		node = next;
323 		lsize += elt * sizeof(*item) + user_pattern_size;
324 		if ((node->rss_types & types) && lsize <= size) {
325 			buf->entry[buf->entries].priority = 1;
326 			buf->entry[buf->entries].pattern = addr;
327 			buf->entries++;
328 			rte_memcpy(addr, buf->entry[0].pattern,
329 				   user_pattern_size);
330 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
331 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
332 			addr = (void *)(((uintptr_t)addr) +
333 					elt * sizeof(*item));
334 		}
335 	}
336 	memset(flow_items, 0, sizeof(flow_items));
337 	next_node = node->next;
338 	stack[stack_pos] = next_node;
339 	node = next_node ? &graph[*next_node] : NULL;
340 	while (node) {
341 		flow_items[stack_pos].type = node->type;
342 		if (node->rss_types & types) {
343 			/*
344 			 * compute the number of items to copy from the
345 			 * expansion and copy it.
346 			 * When the stack_pos is 0, there are 1 element in it,
347 			 * plus the addition END item.
348 			 */
349 			elt = stack_pos + 2;
350 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
351 			lsize += elt * sizeof(*item) + user_pattern_size;
352 			if (lsize <= size) {
353 				size_t n = elt * sizeof(*item);
354 
355 				buf->entry[buf->entries].priority =
356 					stack_pos + 1 + missed;
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) +
362 						user_pattern_size);
363 				rte_memcpy(addr, &missed_item,
364 					   missed * sizeof(*item));
365 				addr = (void *)(((uintptr_t)addr) +
366 					missed * sizeof(*item));
367 				rte_memcpy(addr, flow_items, n);
368 				addr = (void *)(((uintptr_t)addr) + n);
369 			}
370 		}
371 		/* Go deeper. */
372 		if (!node->optional && node->next) {
373 			next_node = node->next;
374 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
375 				rte_errno = E2BIG;
376 				return -rte_errno;
377 			}
378 			stack[stack_pos] = next_node;
379 		} else if (*(next_node + 1)) {
380 			/* Follow up with the next possibility. */
381 			++next_node;
382 		} else {
383 			/* Move to the next path. */
384 			if (stack_pos)
385 				next_node = stack[--stack_pos];
386 			next_node++;
387 			stack[stack_pos] = next_node;
388 		}
389 		node = *next_node ? &graph[*next_node] : NULL;
390 	};
391 	return lsize;
392 }
393 
394 enum mlx5_expansion {
395 	MLX5_EXPANSION_ROOT,
396 	MLX5_EXPANSION_ROOT_OUTER,
397 	MLX5_EXPANSION_ROOT_ETH_VLAN,
398 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
399 	MLX5_EXPANSION_OUTER_ETH,
400 	MLX5_EXPANSION_OUTER_ETH_VLAN,
401 	MLX5_EXPANSION_OUTER_VLAN,
402 	MLX5_EXPANSION_OUTER_IPV4,
403 	MLX5_EXPANSION_OUTER_IPV4_UDP,
404 	MLX5_EXPANSION_OUTER_IPV4_TCP,
405 	MLX5_EXPANSION_OUTER_IPV6,
406 	MLX5_EXPANSION_OUTER_IPV6_UDP,
407 	MLX5_EXPANSION_OUTER_IPV6_TCP,
408 	MLX5_EXPANSION_VXLAN,
409 	MLX5_EXPANSION_VXLAN_GPE,
410 	MLX5_EXPANSION_GRE,
411 	MLX5_EXPANSION_NVGRE,
412 	MLX5_EXPANSION_GRE_KEY,
413 	MLX5_EXPANSION_MPLS,
414 	MLX5_EXPANSION_ETH,
415 	MLX5_EXPANSION_ETH_VLAN,
416 	MLX5_EXPANSION_VLAN,
417 	MLX5_EXPANSION_IPV4,
418 	MLX5_EXPANSION_IPV4_UDP,
419 	MLX5_EXPANSION_IPV4_TCP,
420 	MLX5_EXPANSION_IPV6,
421 	MLX5_EXPANSION_IPV6_UDP,
422 	MLX5_EXPANSION_IPV6_TCP,
423 };
424 
425 /** Supported expansion of items. */
426 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
427 	[MLX5_EXPANSION_ROOT] = {
428 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
429 						  MLX5_EXPANSION_IPV4,
430 						  MLX5_EXPANSION_IPV6),
431 		.type = RTE_FLOW_ITEM_TYPE_END,
432 	},
433 	[MLX5_EXPANSION_ROOT_OUTER] = {
434 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
435 						  MLX5_EXPANSION_OUTER_IPV4,
436 						  MLX5_EXPANSION_OUTER_IPV6),
437 		.type = RTE_FLOW_ITEM_TYPE_END,
438 	},
439 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
440 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
441 		.type = RTE_FLOW_ITEM_TYPE_END,
442 	},
443 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
444 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
445 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
446 		.type = RTE_FLOW_ITEM_TYPE_END,
447 	},
448 	[MLX5_EXPANSION_OUTER_ETH] = {
449 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
450 						  MLX5_EXPANSION_OUTER_IPV6,
451 						  MLX5_EXPANSION_MPLS),
452 		.type = RTE_FLOW_ITEM_TYPE_ETH,
453 		.rss_types = 0,
454 	},
455 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
456 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
457 		.type = RTE_FLOW_ITEM_TYPE_ETH,
458 		.rss_types = 0,
459 	},
460 	[MLX5_EXPANSION_OUTER_VLAN] = {
461 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
462 						  MLX5_EXPANSION_OUTER_IPV6),
463 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
464 	},
465 	[MLX5_EXPANSION_OUTER_IPV4] = {
466 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
467 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
468 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
469 			 MLX5_EXPANSION_GRE,
470 			 MLX5_EXPANSION_NVGRE,
471 			 MLX5_EXPANSION_IPV4,
472 			 MLX5_EXPANSION_IPV6),
473 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
474 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
475 			ETH_RSS_NONFRAG_IPV4_OTHER,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
478 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
479 						  MLX5_EXPANSION_VXLAN_GPE),
480 		.type = RTE_FLOW_ITEM_TYPE_UDP,
481 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
482 	},
483 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
484 		.type = RTE_FLOW_ITEM_TYPE_TCP,
485 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV6] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
489 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
490 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
491 			 MLX5_EXPANSION_IPV4,
492 			 MLX5_EXPANSION_IPV6,
493 			 MLX5_EXPANSION_GRE,
494 			 MLX5_EXPANSION_NVGRE),
495 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
496 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
497 			ETH_RSS_NONFRAG_IPV6_OTHER,
498 	},
499 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
500 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
501 						  MLX5_EXPANSION_VXLAN_GPE),
502 		.type = RTE_FLOW_ITEM_TYPE_UDP,
503 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
504 	},
505 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
506 		.type = RTE_FLOW_ITEM_TYPE_TCP,
507 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
508 	},
509 	[MLX5_EXPANSION_VXLAN] = {
510 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
511 						  MLX5_EXPANSION_IPV4,
512 						  MLX5_EXPANSION_IPV6),
513 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
514 	},
515 	[MLX5_EXPANSION_VXLAN_GPE] = {
516 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
517 						  MLX5_EXPANSION_IPV4,
518 						  MLX5_EXPANSION_IPV6),
519 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
520 	},
521 	[MLX5_EXPANSION_GRE] = {
522 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
523 						  MLX5_EXPANSION_IPV6,
524 						  MLX5_EXPANSION_GRE_KEY),
525 		.type = RTE_FLOW_ITEM_TYPE_GRE,
526 	},
527 	[MLX5_EXPANSION_GRE_KEY] = {
528 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 						  MLX5_EXPANSION_IPV6),
530 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
531 		.optional = 1,
532 	},
533 	[MLX5_EXPANSION_NVGRE] = {
534 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
535 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
536 	},
537 	[MLX5_EXPANSION_MPLS] = {
538 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
539 						  MLX5_EXPANSION_IPV6),
540 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
541 	},
542 	[MLX5_EXPANSION_ETH] = {
543 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
544 						  MLX5_EXPANSION_IPV6),
545 		.type = RTE_FLOW_ITEM_TYPE_ETH,
546 	},
547 	[MLX5_EXPANSION_ETH_VLAN] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
549 		.type = RTE_FLOW_ITEM_TYPE_ETH,
550 	},
551 	[MLX5_EXPANSION_VLAN] = {
552 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
553 						  MLX5_EXPANSION_IPV6),
554 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
555 	},
556 	[MLX5_EXPANSION_IPV4] = {
557 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
558 						  MLX5_EXPANSION_IPV4_TCP),
559 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
560 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
561 			ETH_RSS_NONFRAG_IPV4_OTHER,
562 	},
563 	[MLX5_EXPANSION_IPV4_UDP] = {
564 		.type = RTE_FLOW_ITEM_TYPE_UDP,
565 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
566 	},
567 	[MLX5_EXPANSION_IPV4_TCP] = {
568 		.type = RTE_FLOW_ITEM_TYPE_TCP,
569 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
570 	},
571 	[MLX5_EXPANSION_IPV6] = {
572 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
573 						  MLX5_EXPANSION_IPV6_TCP),
574 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
575 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
576 			ETH_RSS_NONFRAG_IPV6_OTHER,
577 	},
578 	[MLX5_EXPANSION_IPV6_UDP] = {
579 		.type = RTE_FLOW_ITEM_TYPE_UDP,
580 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
581 	},
582 	[MLX5_EXPANSION_IPV6_TCP] = {
583 		.type = RTE_FLOW_ITEM_TYPE_TCP,
584 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
585 	},
586 };
587 
588 static struct rte_flow_action_handle *
589 mlx5_action_handle_create(struct rte_eth_dev *dev,
590 			  const struct rte_flow_indir_action_conf *conf,
591 			  const struct rte_flow_action *action,
592 			  struct rte_flow_error *error);
593 static int mlx5_action_handle_destroy
594 				(struct rte_eth_dev *dev,
595 				 struct rte_flow_action_handle *handle,
596 				 struct rte_flow_error *error);
597 static int mlx5_action_handle_update
598 				(struct rte_eth_dev *dev,
599 				 struct rte_flow_action_handle *handle,
600 				 const void *update,
601 				 struct rte_flow_error *error);
602 static int mlx5_action_handle_query
603 				(struct rte_eth_dev *dev,
604 				 const struct rte_flow_action_handle *handle,
605 				 void *data,
606 				 struct rte_flow_error *error);
607 static int
608 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
609 		    struct rte_flow_tunnel *app_tunnel,
610 		    struct rte_flow_action **actions,
611 		    uint32_t *num_of_actions,
612 		    struct rte_flow_error *error);
613 static int
614 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
615 		       struct rte_flow_tunnel *app_tunnel,
616 		       struct rte_flow_item **items,
617 		       uint32_t *num_of_items,
618 		       struct rte_flow_error *error);
619 static int
620 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
621 			      struct rte_flow_item *pmd_items,
622 			      uint32_t num_items, struct rte_flow_error *err);
623 static int
624 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
625 				struct rte_flow_action *pmd_actions,
626 				uint32_t num_actions,
627 				struct rte_flow_error *err);
628 static int
629 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
630 				  struct rte_mbuf *m,
631 				  struct rte_flow_restore_info *info,
632 				  struct rte_flow_error *err);
633 
634 static const struct rte_flow_ops mlx5_flow_ops = {
635 	.validate = mlx5_flow_validate,
636 	.create = mlx5_flow_create,
637 	.destroy = mlx5_flow_destroy,
638 	.flush = mlx5_flow_flush,
639 	.isolate = mlx5_flow_isolate,
640 	.query = mlx5_flow_query,
641 	.dev_dump = mlx5_flow_dev_dump,
642 	.get_aged_flows = mlx5_flow_get_aged_flows,
643 	.action_handle_create = mlx5_action_handle_create,
644 	.action_handle_destroy = mlx5_action_handle_destroy,
645 	.action_handle_update = mlx5_action_handle_update,
646 	.action_handle_query = mlx5_action_handle_query,
647 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
648 	.tunnel_match = mlx5_flow_tunnel_match,
649 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
650 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
651 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
652 };
653 
654 /* Tunnel information. */
655 struct mlx5_flow_tunnel_info {
656 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
657 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
658 };
659 
660 static struct mlx5_flow_tunnel_info tunnels_info[] = {
661 	{
662 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
663 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
664 	},
665 	{
666 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
667 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
668 	},
669 	{
670 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
671 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
672 	},
673 	{
674 		.tunnel = MLX5_FLOW_LAYER_GRE,
675 		.ptype = RTE_PTYPE_TUNNEL_GRE,
676 	},
677 	{
678 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
679 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
680 	},
681 	{
682 		.tunnel = MLX5_FLOW_LAYER_MPLS,
683 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
684 	},
685 	{
686 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
687 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
688 	},
689 	{
690 		.tunnel = MLX5_FLOW_LAYER_IPIP,
691 		.ptype = RTE_PTYPE_TUNNEL_IP,
692 	},
693 	{
694 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
695 		.ptype = RTE_PTYPE_TUNNEL_IP,
696 	},
697 	{
698 		.tunnel = MLX5_FLOW_LAYER_GTP,
699 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
700 	},
701 };
702 
703 
704 
705 /**
706  * Translate tag ID to register.
707  *
708  * @param[in] dev
709  *   Pointer to the Ethernet device structure.
710  * @param[in] feature
711  *   The feature that request the register.
712  * @param[in] id
713  *   The request register ID.
714  * @param[out] error
715  *   Error description in case of any.
716  *
717  * @return
718  *   The request register on success, a negative errno
719  *   value otherwise and rte_errno is set.
720  */
721 int
722 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
723 		     enum mlx5_feature_name feature,
724 		     uint32_t id,
725 		     struct rte_flow_error *error)
726 {
727 	struct mlx5_priv *priv = dev->data->dev_private;
728 	struct mlx5_dev_config *config = &priv->config;
729 	enum modify_reg start_reg;
730 	bool skip_mtr_reg = false;
731 
732 	switch (feature) {
733 	case MLX5_HAIRPIN_RX:
734 		return REG_B;
735 	case MLX5_HAIRPIN_TX:
736 		return REG_A;
737 	case MLX5_METADATA_RX:
738 		switch (config->dv_xmeta_en) {
739 		case MLX5_XMETA_MODE_LEGACY:
740 			return REG_B;
741 		case MLX5_XMETA_MODE_META16:
742 			return REG_C_0;
743 		case MLX5_XMETA_MODE_META32:
744 			return REG_C_1;
745 		}
746 		break;
747 	case MLX5_METADATA_TX:
748 		return REG_A;
749 	case MLX5_METADATA_FDB:
750 		switch (config->dv_xmeta_en) {
751 		case MLX5_XMETA_MODE_LEGACY:
752 			return REG_NON;
753 		case MLX5_XMETA_MODE_META16:
754 			return REG_C_0;
755 		case MLX5_XMETA_MODE_META32:
756 			return REG_C_1;
757 		}
758 		break;
759 	case MLX5_FLOW_MARK:
760 		switch (config->dv_xmeta_en) {
761 		case MLX5_XMETA_MODE_LEGACY:
762 			return REG_NON;
763 		case MLX5_XMETA_MODE_META16:
764 			return REG_C_1;
765 		case MLX5_XMETA_MODE_META32:
766 			return REG_C_0;
767 		}
768 		break;
769 	case MLX5_MTR_ID:
770 		/*
771 		 * If meter color and meter id share one register, flow match
772 		 * should use the meter color register for match.
773 		 */
774 		if (priv->mtr_reg_share)
775 			return priv->mtr_color_reg;
776 		else
777 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
778 			       REG_C_3;
779 	case MLX5_MTR_COLOR:
780 	case MLX5_ASO_FLOW_HIT:
781 	case MLX5_ASO_CONNTRACK:
782 		/* All features use the same REG_C. */
783 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
784 		return priv->mtr_color_reg;
785 	case MLX5_COPY_MARK:
786 		/*
787 		 * Metadata COPY_MARK register using is in meter suffix sub
788 		 * flow while with meter. It's safe to share the same register.
789 		 */
790 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
791 	case MLX5_APP_TAG:
792 		/*
793 		 * If meter is enable, it will engage the register for color
794 		 * match and flow match. If meter color match is not using the
795 		 * REG_C_2, need to skip the REG_C_x be used by meter color
796 		 * match.
797 		 * If meter is disable, free to use all available registers.
798 		 */
799 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
800 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
801 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
802 		if (id > (uint32_t)(REG_C_7 - start_reg))
803 			return rte_flow_error_set(error, EINVAL,
804 						  RTE_FLOW_ERROR_TYPE_ITEM,
805 						  NULL, "invalid tag id");
806 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
807 			return rte_flow_error_set(error, ENOTSUP,
808 						  RTE_FLOW_ERROR_TYPE_ITEM,
809 						  NULL, "unsupported tag id");
810 		/*
811 		 * This case means meter is using the REG_C_x great than 2.
812 		 * Take care not to conflict with meter color REG_C_x.
813 		 * If the available index REG_C_y >= REG_C_x, skip the
814 		 * color register.
815 		 */
816 		if (skip_mtr_reg && config->flow_mreg_c
817 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
818 			if (id >= (uint32_t)(REG_C_7 - start_reg))
819 				return rte_flow_error_set(error, EINVAL,
820 						       RTE_FLOW_ERROR_TYPE_ITEM,
821 							NULL, "invalid tag id");
822 			if (config->flow_mreg_c
823 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
824 				return config->flow_mreg_c
825 					       [id + 1 + start_reg - REG_C_0];
826 			return rte_flow_error_set(error, ENOTSUP,
827 						  RTE_FLOW_ERROR_TYPE_ITEM,
828 						  NULL, "unsupported tag id");
829 		}
830 		return config->flow_mreg_c[id + start_reg - REG_C_0];
831 	}
832 	MLX5_ASSERT(false);
833 	return rte_flow_error_set(error, EINVAL,
834 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
835 				  NULL, "invalid feature name");
836 }
837 
838 /**
839  * Check extensive flow metadata register support.
840  *
841  * @param dev
842  *   Pointer to rte_eth_dev structure.
843  *
844  * @return
845  *   True if device supports extensive flow metadata register, otherwise false.
846  */
847 bool
848 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
849 {
850 	struct mlx5_priv *priv = dev->data->dev_private;
851 	struct mlx5_dev_config *config = &priv->config;
852 
853 	/*
854 	 * Having available reg_c can be regarded inclusively as supporting
855 	 * extensive flow metadata register, which could mean,
856 	 * - metadata register copy action by modify header.
857 	 * - 16 modify header actions is supported.
858 	 * - reg_c's are preserved across different domain (FDB and NIC) on
859 	 *   packet loopback by flow lookup miss.
860 	 */
861 	return config->flow_mreg_c[2] != REG_NON;
862 }
863 
864 /**
865  * Get the lowest priority.
866  *
867  * @param[in] dev
868  *   Pointer to the Ethernet device structure.
869  * @param[in] attributes
870  *   Pointer to device flow rule attributes.
871  *
872  * @return
873  *   The value of lowest priority of flow.
874  */
875 uint32_t
876 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
877 			  const struct rte_flow_attr *attr)
878 {
879 	struct mlx5_priv *priv = dev->data->dev_private;
880 
881 	if (!attr->group && !attr->transfer)
882 		return priv->config.flow_prio - 2;
883 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
884 }
885 
886 /**
887  * Calculate matcher priority of the flow.
888  *
889  * @param[in] dev
890  *   Pointer to the Ethernet device structure.
891  * @param[in] attr
892  *   Pointer to device flow rule attributes.
893  * @param[in] subpriority
894  *   The priority based on the items.
895  * @return
896  *   The matcher priority of the flow.
897  */
898 uint16_t
899 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
900 			  const struct rte_flow_attr *attr,
901 			  uint32_t subpriority)
902 {
903 	uint16_t priority = (uint16_t)attr->priority;
904 	struct mlx5_priv *priv = dev->data->dev_private;
905 
906 	if (!attr->group && !attr->transfer) {
907 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
908 			priority = priv->config.flow_prio - 1;
909 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
910 	}
911 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
912 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
913 	return priority * 3 + subpriority;
914 }
915 
916 /**
917  * Verify the @p item specifications (spec, last, mask) are compatible with the
918  * NIC capabilities.
919  *
920  * @param[in] item
921  *   Item specification.
922  * @param[in] mask
923  *   @p item->mask or flow default bit-masks.
924  * @param[in] nic_mask
925  *   Bit-masks covering supported fields by the NIC to compare with user mask.
926  * @param[in] size
927  *   Bit-masks size in bytes.
928  * @param[in] range_accepted
929  *   True if range of values is accepted for specific fields, false otherwise.
930  * @param[out] error
931  *   Pointer to error structure.
932  *
933  * @return
934  *   0 on success, a negative errno value otherwise and rte_errno is set.
935  */
936 int
937 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
938 			  const uint8_t *mask,
939 			  const uint8_t *nic_mask,
940 			  unsigned int size,
941 			  bool range_accepted,
942 			  struct rte_flow_error *error)
943 {
944 	unsigned int i;
945 
946 	MLX5_ASSERT(nic_mask);
947 	for (i = 0; i < size; ++i)
948 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
949 			return rte_flow_error_set(error, ENOTSUP,
950 						  RTE_FLOW_ERROR_TYPE_ITEM,
951 						  item,
952 						  "mask enables non supported"
953 						  " bits");
954 	if (!item->spec && (item->mask || item->last))
955 		return rte_flow_error_set(error, EINVAL,
956 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
957 					  "mask/last without a spec is not"
958 					  " supported");
959 	if (item->spec && item->last && !range_accepted) {
960 		uint8_t spec[size];
961 		uint8_t last[size];
962 		unsigned int i;
963 		int ret;
964 
965 		for (i = 0; i < size; ++i) {
966 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
967 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
968 		}
969 		ret = memcmp(spec, last, size);
970 		if (ret != 0)
971 			return rte_flow_error_set(error, EINVAL,
972 						  RTE_FLOW_ERROR_TYPE_ITEM,
973 						  item,
974 						  "range is not valid");
975 	}
976 	return 0;
977 }
978 
979 /**
980  * Adjust the hash fields according to the @p flow information.
981  *
982  * @param[in] dev_flow.
983  *   Pointer to the mlx5_flow.
984  * @param[in] tunnel
985  *   1 when the hash field is for a tunnel item.
986  * @param[in] layer_types
987  *   ETH_RSS_* types.
988  * @param[in] hash_fields
989  *   Item hash fields.
990  *
991  * @return
992  *   The hash fields that should be used.
993  */
994 uint64_t
995 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
996 			    int tunnel __rte_unused, uint64_t layer_types,
997 			    uint64_t hash_fields)
998 {
999 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1000 	int rss_request_inner = rss_desc->level >= 2;
1001 
1002 	/* Check RSS hash level for tunnel. */
1003 	if (tunnel && rss_request_inner)
1004 		hash_fields |= IBV_RX_HASH_INNER;
1005 	else if (tunnel || rss_request_inner)
1006 		return 0;
1007 #endif
1008 	/* Check if requested layer matches RSS hash fields. */
1009 	if (!(rss_desc->types & layer_types))
1010 		return 0;
1011 	return hash_fields;
1012 }
1013 
1014 /**
1015  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1016  * if several tunnel rules are used on this queue, the tunnel ptype will be
1017  * cleared.
1018  *
1019  * @param rxq_ctrl
1020  *   Rx queue to update.
1021  */
1022 static void
1023 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1024 {
1025 	unsigned int i;
1026 	uint32_t tunnel_ptype = 0;
1027 
1028 	/* Look up for the ptype to use. */
1029 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1030 		if (!rxq_ctrl->flow_tunnels_n[i])
1031 			continue;
1032 		if (!tunnel_ptype) {
1033 			tunnel_ptype = tunnels_info[i].ptype;
1034 		} else {
1035 			tunnel_ptype = 0;
1036 			break;
1037 		}
1038 	}
1039 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1040 }
1041 
1042 /**
1043  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1044  * flow.
1045  *
1046  * @param[in] dev
1047  *   Pointer to the Ethernet device structure.
1048  * @param[in] dev_handle
1049  *   Pointer to device flow handle structure.
1050  */
1051 void
1052 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1053 		       struct mlx5_flow_handle *dev_handle)
1054 {
1055 	struct mlx5_priv *priv = dev->data->dev_private;
1056 	const int mark = dev_handle->mark;
1057 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1058 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1059 	unsigned int i;
1060 
1061 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1062 		struct mlx5_hrxq *hrxq;
1063 
1064 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1065 			      dev_handle->rix_hrxq);
1066 		if (hrxq)
1067 			ind_tbl = hrxq->ind_table;
1068 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1069 		struct mlx5_shared_action_rss *shared_rss;
1070 
1071 		shared_rss = mlx5_ipool_get
1072 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1073 			 dev_handle->rix_srss);
1074 		if (shared_rss)
1075 			ind_tbl = shared_rss->ind_tbl;
1076 	}
1077 	if (!ind_tbl)
1078 		return;
1079 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1080 		int idx = ind_tbl->queues[i];
1081 		struct mlx5_rxq_ctrl *rxq_ctrl =
1082 			container_of((*priv->rxqs)[idx],
1083 				     struct mlx5_rxq_ctrl, rxq);
1084 
1085 		/*
1086 		 * To support metadata register copy on Tx loopback,
1087 		 * this must be always enabled (metadata may arive
1088 		 * from other port - not from local flows only.
1089 		 */
1090 		if (priv->config.dv_flow_en &&
1091 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1092 		    mlx5_flow_ext_mreg_supported(dev)) {
1093 			rxq_ctrl->rxq.mark = 1;
1094 			rxq_ctrl->flow_mark_n = 1;
1095 		} else if (mark) {
1096 			rxq_ctrl->rxq.mark = 1;
1097 			rxq_ctrl->flow_mark_n++;
1098 		}
1099 		if (tunnel) {
1100 			unsigned int j;
1101 
1102 			/* Increase the counter matching the flow. */
1103 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1104 				if ((tunnels_info[j].tunnel &
1105 				     dev_handle->layers) ==
1106 				    tunnels_info[j].tunnel) {
1107 					rxq_ctrl->flow_tunnels_n[j]++;
1108 					break;
1109 				}
1110 			}
1111 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1112 		}
1113 	}
1114 }
1115 
1116 /**
1117  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1118  *
1119  * @param[in] dev
1120  *   Pointer to the Ethernet device structure.
1121  * @param[in] flow
1122  *   Pointer to flow structure.
1123  */
1124 static void
1125 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1126 {
1127 	struct mlx5_priv *priv = dev->data->dev_private;
1128 	uint32_t handle_idx;
1129 	struct mlx5_flow_handle *dev_handle;
1130 
1131 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1132 		       handle_idx, dev_handle, next)
1133 		flow_drv_rxq_flags_set(dev, dev_handle);
1134 }
1135 
1136 /**
1137  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1138  * device flow if no other flow uses it with the same kind of request.
1139  *
1140  * @param dev
1141  *   Pointer to Ethernet device.
1142  * @param[in] dev_handle
1143  *   Pointer to the device flow handle structure.
1144  */
1145 static void
1146 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1147 			struct mlx5_flow_handle *dev_handle)
1148 {
1149 	struct mlx5_priv *priv = dev->data->dev_private;
1150 	const int mark = dev_handle->mark;
1151 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1152 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1153 	unsigned int i;
1154 
1155 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1156 		struct mlx5_hrxq *hrxq;
1157 
1158 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1159 			      dev_handle->rix_hrxq);
1160 		if (hrxq)
1161 			ind_tbl = hrxq->ind_table;
1162 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1163 		struct mlx5_shared_action_rss *shared_rss;
1164 
1165 		shared_rss = mlx5_ipool_get
1166 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1167 			 dev_handle->rix_srss);
1168 		if (shared_rss)
1169 			ind_tbl = shared_rss->ind_tbl;
1170 	}
1171 	if (!ind_tbl)
1172 		return;
1173 	MLX5_ASSERT(dev->data->dev_started);
1174 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1175 		int idx = ind_tbl->queues[i];
1176 		struct mlx5_rxq_ctrl *rxq_ctrl =
1177 			container_of((*priv->rxqs)[idx],
1178 				     struct mlx5_rxq_ctrl, rxq);
1179 
1180 		if (priv->config.dv_flow_en &&
1181 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1182 		    mlx5_flow_ext_mreg_supported(dev)) {
1183 			rxq_ctrl->rxq.mark = 1;
1184 			rxq_ctrl->flow_mark_n = 1;
1185 		} else if (mark) {
1186 			rxq_ctrl->flow_mark_n--;
1187 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1188 		}
1189 		if (tunnel) {
1190 			unsigned int j;
1191 
1192 			/* Decrease the counter matching the flow. */
1193 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1194 				if ((tunnels_info[j].tunnel &
1195 				     dev_handle->layers) ==
1196 				    tunnels_info[j].tunnel) {
1197 					rxq_ctrl->flow_tunnels_n[j]--;
1198 					break;
1199 				}
1200 			}
1201 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1202 		}
1203 	}
1204 }
1205 
1206 /**
1207  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1208  * @p flow if no other flow uses it with the same kind of request.
1209  *
1210  * @param dev
1211  *   Pointer to Ethernet device.
1212  * @param[in] flow
1213  *   Pointer to the flow.
1214  */
1215 static void
1216 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1217 {
1218 	struct mlx5_priv *priv = dev->data->dev_private;
1219 	uint32_t handle_idx;
1220 	struct mlx5_flow_handle *dev_handle;
1221 
1222 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1223 		       handle_idx, dev_handle, next)
1224 		flow_drv_rxq_flags_trim(dev, dev_handle);
1225 }
1226 
1227 /**
1228  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1229  *
1230  * @param dev
1231  *   Pointer to Ethernet device.
1232  */
1233 static void
1234 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1235 {
1236 	struct mlx5_priv *priv = dev->data->dev_private;
1237 	unsigned int i;
1238 
1239 	for (i = 0; i != priv->rxqs_n; ++i) {
1240 		struct mlx5_rxq_ctrl *rxq_ctrl;
1241 		unsigned int j;
1242 
1243 		if (!(*priv->rxqs)[i])
1244 			continue;
1245 		rxq_ctrl = container_of((*priv->rxqs)[i],
1246 					struct mlx5_rxq_ctrl, rxq);
1247 		rxq_ctrl->flow_mark_n = 0;
1248 		rxq_ctrl->rxq.mark = 0;
1249 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1250 			rxq_ctrl->flow_tunnels_n[j] = 0;
1251 		rxq_ctrl->rxq.tunnel = 0;
1252 	}
1253 }
1254 
1255 /**
1256  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1257  *
1258  * @param[in] dev
1259  *   Pointer to the Ethernet device structure.
1260  */
1261 void
1262 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1263 {
1264 	struct mlx5_priv *priv = dev->data->dev_private;
1265 	struct mlx5_rxq_data *data;
1266 	unsigned int i;
1267 
1268 	for (i = 0; i != priv->rxqs_n; ++i) {
1269 		if (!(*priv->rxqs)[i])
1270 			continue;
1271 		data = (*priv->rxqs)[i];
1272 		if (!rte_flow_dynf_metadata_avail()) {
1273 			data->dynf_meta = 0;
1274 			data->flow_meta_mask = 0;
1275 			data->flow_meta_offset = -1;
1276 			data->flow_meta_port_mask = 0;
1277 		} else {
1278 			data->dynf_meta = 1;
1279 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1280 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1281 			data->flow_meta_port_mask = (uint32_t)~0;
1282 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1283 				data->flow_meta_port_mask >>= 16;
1284 		}
1285 	}
1286 }
1287 
1288 /*
1289  * return a pointer to the desired action in the list of actions.
1290  *
1291  * @param[in] actions
1292  *   The list of actions to search the action in.
1293  * @param[in] action
1294  *   The action to find.
1295  *
1296  * @return
1297  *   Pointer to the action in the list, if found. NULL otherwise.
1298  */
1299 const struct rte_flow_action *
1300 mlx5_flow_find_action(const struct rte_flow_action *actions,
1301 		      enum rte_flow_action_type action)
1302 {
1303 	if (actions == NULL)
1304 		return NULL;
1305 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1306 		if (actions->type == action)
1307 			return actions;
1308 	return NULL;
1309 }
1310 
1311 /*
1312  * Validate the flag action.
1313  *
1314  * @param[in] action_flags
1315  *   Bit-fields that holds the actions detected until now.
1316  * @param[in] attr
1317  *   Attributes of flow that includes this action.
1318  * @param[out] error
1319  *   Pointer to error structure.
1320  *
1321  * @return
1322  *   0 on success, a negative errno value otherwise and rte_errno is set.
1323  */
1324 int
1325 mlx5_flow_validate_action_flag(uint64_t action_flags,
1326 			       const struct rte_flow_attr *attr,
1327 			       struct rte_flow_error *error)
1328 {
1329 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1330 		return rte_flow_error_set(error, EINVAL,
1331 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1332 					  "can't mark and flag in same flow");
1333 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1334 		return rte_flow_error_set(error, EINVAL,
1335 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1336 					  "can't have 2 flag"
1337 					  " actions in same flow");
1338 	if (attr->egress)
1339 		return rte_flow_error_set(error, ENOTSUP,
1340 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1341 					  "flag action not supported for "
1342 					  "egress");
1343 	return 0;
1344 }
1345 
1346 /*
1347  * Validate the mark action.
1348  *
1349  * @param[in] action
1350  *   Pointer to the queue action.
1351  * @param[in] action_flags
1352  *   Bit-fields that holds the actions detected until now.
1353  * @param[in] attr
1354  *   Attributes of flow that includes this action.
1355  * @param[out] error
1356  *   Pointer to error structure.
1357  *
1358  * @return
1359  *   0 on success, a negative errno value otherwise and rte_errno is set.
1360  */
1361 int
1362 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1363 			       uint64_t action_flags,
1364 			       const struct rte_flow_attr *attr,
1365 			       struct rte_flow_error *error)
1366 {
1367 	const struct rte_flow_action_mark *mark = action->conf;
1368 
1369 	if (!mark)
1370 		return rte_flow_error_set(error, EINVAL,
1371 					  RTE_FLOW_ERROR_TYPE_ACTION,
1372 					  action,
1373 					  "configuration cannot be null");
1374 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1375 		return rte_flow_error_set(error, EINVAL,
1376 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1377 					  &mark->id,
1378 					  "mark id must in 0 <= id < "
1379 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1380 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1381 		return rte_flow_error_set(error, EINVAL,
1382 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1383 					  "can't flag and mark in same flow");
1384 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1385 		return rte_flow_error_set(error, EINVAL,
1386 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1387 					  "can't have 2 mark actions in same"
1388 					  " flow");
1389 	if (attr->egress)
1390 		return rte_flow_error_set(error, ENOTSUP,
1391 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1392 					  "mark action not supported for "
1393 					  "egress");
1394 	return 0;
1395 }
1396 
1397 /*
1398  * Validate the drop action.
1399  *
1400  * @param[in] action_flags
1401  *   Bit-fields that holds the actions detected until now.
1402  * @param[in] attr
1403  *   Attributes of flow that includes this action.
1404  * @param[out] error
1405  *   Pointer to error structure.
1406  *
1407  * @return
1408  *   0 on success, a negative errno value otherwise and rte_errno is set.
1409  */
1410 int
1411 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1412 			       const struct rte_flow_attr *attr,
1413 			       struct rte_flow_error *error)
1414 {
1415 	if (attr->egress)
1416 		return rte_flow_error_set(error, ENOTSUP,
1417 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1418 					  "drop action not supported for "
1419 					  "egress");
1420 	return 0;
1421 }
1422 
1423 /*
1424  * Validate the queue action.
1425  *
1426  * @param[in] action
1427  *   Pointer to the queue action.
1428  * @param[in] action_flags
1429  *   Bit-fields that holds the actions detected until now.
1430  * @param[in] dev
1431  *   Pointer to the Ethernet device structure.
1432  * @param[in] attr
1433  *   Attributes of flow that includes this action.
1434  * @param[out] error
1435  *   Pointer to error structure.
1436  *
1437  * @return
1438  *   0 on success, a negative errno value otherwise and rte_errno is set.
1439  */
1440 int
1441 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1442 				uint64_t action_flags,
1443 				struct rte_eth_dev *dev,
1444 				const struct rte_flow_attr *attr,
1445 				struct rte_flow_error *error)
1446 {
1447 	struct mlx5_priv *priv = dev->data->dev_private;
1448 	const struct rte_flow_action_queue *queue = action->conf;
1449 
1450 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1451 		return rte_flow_error_set(error, EINVAL,
1452 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1453 					  "can't have 2 fate actions in"
1454 					  " same flow");
1455 	if (!priv->rxqs_n)
1456 		return rte_flow_error_set(error, EINVAL,
1457 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1458 					  NULL, "No Rx queues configured");
1459 	if (queue->index >= priv->rxqs_n)
1460 		return rte_flow_error_set(error, EINVAL,
1461 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1462 					  &queue->index,
1463 					  "queue index out of range");
1464 	if (!(*priv->rxqs)[queue->index])
1465 		return rte_flow_error_set(error, EINVAL,
1466 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1467 					  &queue->index,
1468 					  "queue is not configured");
1469 	if (attr->egress)
1470 		return rte_flow_error_set(error, ENOTSUP,
1471 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1472 					  "queue action not supported for "
1473 					  "egress");
1474 	return 0;
1475 }
1476 
1477 /*
1478  * Validate the rss action.
1479  *
1480  * @param[in] dev
1481  *   Pointer to the Ethernet device structure.
1482  * @param[in] action
1483  *   Pointer to the queue action.
1484  * @param[out] error
1485  *   Pointer to error structure.
1486  *
1487  * @return
1488  *   0 on success, a negative errno value otherwise and rte_errno is set.
1489  */
1490 int
1491 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1492 			 const struct rte_flow_action *action,
1493 			 struct rte_flow_error *error)
1494 {
1495 	struct mlx5_priv *priv = dev->data->dev_private;
1496 	const struct rte_flow_action_rss *rss = action->conf;
1497 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1498 	unsigned int i;
1499 
1500 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1501 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1502 		return rte_flow_error_set(error, ENOTSUP,
1503 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1504 					  &rss->func,
1505 					  "RSS hash function not supported");
1506 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1507 	if (rss->level > 2)
1508 #else
1509 	if (rss->level > 1)
1510 #endif
1511 		return rte_flow_error_set(error, ENOTSUP,
1512 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1513 					  &rss->level,
1514 					  "tunnel RSS is not supported");
1515 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1516 	if (rss->key_len == 0 && rss->key != NULL)
1517 		return rte_flow_error_set(error, ENOTSUP,
1518 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1519 					  &rss->key_len,
1520 					  "RSS hash key length 0");
1521 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1522 		return rte_flow_error_set(error, ENOTSUP,
1523 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1524 					  &rss->key_len,
1525 					  "RSS hash key too small");
1526 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1527 		return rte_flow_error_set(error, ENOTSUP,
1528 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1529 					  &rss->key_len,
1530 					  "RSS hash key too large");
1531 	if (rss->queue_num > priv->config.ind_table_max_size)
1532 		return rte_flow_error_set(error, ENOTSUP,
1533 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1534 					  &rss->queue_num,
1535 					  "number of queues too large");
1536 	if (rss->types & MLX5_RSS_HF_MASK)
1537 		return rte_flow_error_set(error, ENOTSUP,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 					  &rss->types,
1540 					  "some RSS protocols are not"
1541 					  " supported");
1542 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1543 	    !(rss->types & ETH_RSS_IP))
1544 		return rte_flow_error_set(error, EINVAL,
1545 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1546 					  "L3 partial RSS requested but L3 RSS"
1547 					  " type not specified");
1548 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1549 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1550 		return rte_flow_error_set(error, EINVAL,
1551 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1552 					  "L4 partial RSS requested but L4 RSS"
1553 					  " type not specified");
1554 	if (!priv->rxqs_n)
1555 		return rte_flow_error_set(error, EINVAL,
1556 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1557 					  NULL, "No Rx queues configured");
1558 	if (!rss->queue_num)
1559 		return rte_flow_error_set(error, EINVAL,
1560 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1561 					  NULL, "No queues configured");
1562 	for (i = 0; i != rss->queue_num; ++i) {
1563 		struct mlx5_rxq_ctrl *rxq_ctrl;
1564 
1565 		if (rss->queue[i] >= priv->rxqs_n)
1566 			return rte_flow_error_set
1567 				(error, EINVAL,
1568 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 				 &rss->queue[i], "queue index out of range");
1570 		if (!(*priv->rxqs)[rss->queue[i]])
1571 			return rte_flow_error_set
1572 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1573 				 &rss->queue[i], "queue is not configured");
1574 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1575 					struct mlx5_rxq_ctrl, rxq);
1576 		if (i == 0)
1577 			rxq_type = rxq_ctrl->type;
1578 		if (rxq_type != rxq_ctrl->type)
1579 			return rte_flow_error_set
1580 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1581 				 &rss->queue[i],
1582 				 "combining hairpin and regular RSS queues is not supported");
1583 	}
1584 	return 0;
1585 }
1586 
1587 /*
1588  * Validate the rss action.
1589  *
1590  * @param[in] action
1591  *   Pointer to the queue action.
1592  * @param[in] action_flags
1593  *   Bit-fields that holds the actions detected until now.
1594  * @param[in] dev
1595  *   Pointer to the Ethernet device structure.
1596  * @param[in] attr
1597  *   Attributes of flow that includes this action.
1598  * @param[in] item_flags
1599  *   Items that were detected.
1600  * @param[out] error
1601  *   Pointer to error structure.
1602  *
1603  * @return
1604  *   0 on success, a negative errno value otherwise and rte_errno is set.
1605  */
1606 int
1607 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1608 			      uint64_t action_flags,
1609 			      struct rte_eth_dev *dev,
1610 			      const struct rte_flow_attr *attr,
1611 			      uint64_t item_flags,
1612 			      struct rte_flow_error *error)
1613 {
1614 	const struct rte_flow_action_rss *rss = action->conf;
1615 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1616 	int ret;
1617 
1618 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1619 		return rte_flow_error_set(error, EINVAL,
1620 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1621 					  "can't have 2 fate actions"
1622 					  " in same flow");
1623 	ret = mlx5_validate_action_rss(dev, action, error);
1624 	if (ret)
1625 		return ret;
1626 	if (attr->egress)
1627 		return rte_flow_error_set(error, ENOTSUP,
1628 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1629 					  "rss action not supported for "
1630 					  "egress");
1631 	if (rss->level > 1 && !tunnel)
1632 		return rte_flow_error_set(error, EINVAL,
1633 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1634 					  "inner RSS is not supported for "
1635 					  "non-tunnel flows");
1636 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1637 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1638 		return rte_flow_error_set(error, EINVAL,
1639 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1640 					  "RSS on eCPRI is not supported now");
1641 	}
1642 	return 0;
1643 }
1644 
1645 /*
1646  * Validate the default miss action.
1647  *
1648  * @param[in] action_flags
1649  *   Bit-fields that holds the actions detected until now.
1650  * @param[out] error
1651  *   Pointer to error structure.
1652  *
1653  * @return
1654  *   0 on success, a negative errno value otherwise and rte_errno is set.
1655  */
1656 int
1657 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1658 				const struct rte_flow_attr *attr,
1659 				struct rte_flow_error *error)
1660 {
1661 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1662 		return rte_flow_error_set(error, EINVAL,
1663 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1664 					  "can't have 2 fate actions in"
1665 					  " same flow");
1666 	if (attr->egress)
1667 		return rte_flow_error_set(error, ENOTSUP,
1668 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1669 					  "default miss action not supported "
1670 					  "for egress");
1671 	if (attr->group)
1672 		return rte_flow_error_set(error, ENOTSUP,
1673 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1674 					  "only group 0 is supported");
1675 	if (attr->transfer)
1676 		return rte_flow_error_set(error, ENOTSUP,
1677 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1678 					  NULL, "transfer is not supported");
1679 	return 0;
1680 }
1681 
1682 /*
1683  * Validate the count action.
1684  *
1685  * @param[in] dev
1686  *   Pointer to the Ethernet device structure.
1687  * @param[in] attr
1688  *   Attributes of flow that includes this action.
1689  * @param[out] error
1690  *   Pointer to error structure.
1691  *
1692  * @return
1693  *   0 on success, a negative errno value otherwise and rte_errno is set.
1694  */
1695 int
1696 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1697 				const struct rte_flow_attr *attr,
1698 				struct rte_flow_error *error)
1699 {
1700 	if (attr->egress)
1701 		return rte_flow_error_set(error, ENOTSUP,
1702 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1703 					  "count action not supported for "
1704 					  "egress");
1705 	return 0;
1706 }
1707 
1708 /*
1709  * Validate the ASO CT action.
1710  *
1711  * @param[in] dev
1712  *   Pointer to the Ethernet device structure.
1713  * @param[in] conntrack
1714  *   Pointer to the CT action profile.
1715  * @param[out] error
1716  *   Pointer to error structure.
1717  *
1718  * @return
1719  *   0 on success, a negative errno value otherwise and rte_errno is set.
1720  */
1721 int
1722 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1723 			const struct rte_flow_action_conntrack *conntrack,
1724 			struct rte_flow_error *error)
1725 {
1726 	RTE_SET_USED(dev);
1727 
1728 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1729 		return rte_flow_error_set(error, EINVAL,
1730 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1731 					  "Invalid CT state");
1732 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1733 		return rte_flow_error_set(error, EINVAL,
1734 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1735 					  "Invalid last TCP packet flag");
1736 	return 0;
1737 }
1738 
1739 /**
1740  * Verify the @p attributes will be correctly understood by the NIC and store
1741  * them in the @p flow if everything is correct.
1742  *
1743  * @param[in] dev
1744  *   Pointer to the Ethernet device structure.
1745  * @param[in] attributes
1746  *   Pointer to flow attributes
1747  * @param[out] error
1748  *   Pointer to error structure.
1749  *
1750  * @return
1751  *   0 on success, a negative errno value otherwise and rte_errno is set.
1752  */
1753 int
1754 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1755 			      const struct rte_flow_attr *attributes,
1756 			      struct rte_flow_error *error)
1757 {
1758 	struct mlx5_priv *priv = dev->data->dev_private;
1759 	uint32_t priority_max = priv->config.flow_prio - 1;
1760 
1761 	if (attributes->group)
1762 		return rte_flow_error_set(error, ENOTSUP,
1763 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1764 					  NULL, "groups is not supported");
1765 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1766 	    attributes->priority >= priority_max)
1767 		return rte_flow_error_set(error, ENOTSUP,
1768 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1769 					  NULL, "priority out of range");
1770 	if (attributes->egress)
1771 		return rte_flow_error_set(error, ENOTSUP,
1772 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1773 					  "egress is not supported");
1774 	if (attributes->transfer && !priv->config.dv_esw_en)
1775 		return rte_flow_error_set(error, ENOTSUP,
1776 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1777 					  NULL, "transfer is not supported");
1778 	if (!attributes->ingress)
1779 		return rte_flow_error_set(error, EINVAL,
1780 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1781 					  NULL,
1782 					  "ingress attribute is mandatory");
1783 	return 0;
1784 }
1785 
1786 /**
1787  * Validate ICMP6 item.
1788  *
1789  * @param[in] item
1790  *   Item specification.
1791  * @param[in] item_flags
1792  *   Bit-fields that holds the items detected until now.
1793  * @param[in] ext_vlan_sup
1794  *   Whether extended VLAN features are supported or not.
1795  * @param[out] error
1796  *   Pointer to error structure.
1797  *
1798  * @return
1799  *   0 on success, a negative errno value otherwise and rte_errno is set.
1800  */
1801 int
1802 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1803 			       uint64_t item_flags,
1804 			       uint8_t target_protocol,
1805 			       struct rte_flow_error *error)
1806 {
1807 	const struct rte_flow_item_icmp6 *mask = item->mask;
1808 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1809 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1810 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1811 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1812 				      MLX5_FLOW_LAYER_OUTER_L4;
1813 	int ret;
1814 
1815 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1816 		return rte_flow_error_set(error, EINVAL,
1817 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1818 					  "protocol filtering not compatible"
1819 					  " with ICMP6 layer");
1820 	if (!(item_flags & l3m))
1821 		return rte_flow_error_set(error, EINVAL,
1822 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1823 					  "IPv6 is mandatory to filter on"
1824 					  " ICMP6");
1825 	if (item_flags & l4m)
1826 		return rte_flow_error_set(error, EINVAL,
1827 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1828 					  "multiple L4 layers not supported");
1829 	if (!mask)
1830 		mask = &rte_flow_item_icmp6_mask;
1831 	ret = mlx5_flow_item_acceptable
1832 		(item, (const uint8_t *)mask,
1833 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1834 		 sizeof(struct rte_flow_item_icmp6),
1835 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1836 	if (ret < 0)
1837 		return ret;
1838 	return 0;
1839 }
1840 
1841 /**
1842  * Validate ICMP item.
1843  *
1844  * @param[in] item
1845  *   Item specification.
1846  * @param[in] item_flags
1847  *   Bit-fields that holds the items detected until now.
1848  * @param[out] error
1849  *   Pointer to error structure.
1850  *
1851  * @return
1852  *   0 on success, a negative errno value otherwise and rte_errno is set.
1853  */
1854 int
1855 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1856 			     uint64_t item_flags,
1857 			     uint8_t target_protocol,
1858 			     struct rte_flow_error *error)
1859 {
1860 	const struct rte_flow_item_icmp *mask = item->mask;
1861 	const struct rte_flow_item_icmp nic_mask = {
1862 		.hdr.icmp_type = 0xff,
1863 		.hdr.icmp_code = 0xff,
1864 		.hdr.icmp_ident = RTE_BE16(0xffff),
1865 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1866 	};
1867 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1868 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1869 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1870 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1871 				      MLX5_FLOW_LAYER_OUTER_L4;
1872 	int ret;
1873 
1874 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1875 		return rte_flow_error_set(error, EINVAL,
1876 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1877 					  "protocol filtering not compatible"
1878 					  " with ICMP layer");
1879 	if (!(item_flags & l3m))
1880 		return rte_flow_error_set(error, EINVAL,
1881 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1882 					  "IPv4 is mandatory to filter"
1883 					  " on ICMP");
1884 	if (item_flags & l4m)
1885 		return rte_flow_error_set(error, EINVAL,
1886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1887 					  "multiple L4 layers not supported");
1888 	if (!mask)
1889 		mask = &nic_mask;
1890 	ret = mlx5_flow_item_acceptable
1891 		(item, (const uint8_t *)mask,
1892 		 (const uint8_t *)&nic_mask,
1893 		 sizeof(struct rte_flow_item_icmp),
1894 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1895 	if (ret < 0)
1896 		return ret;
1897 	return 0;
1898 }
1899 
1900 /**
1901  * Validate Ethernet item.
1902  *
1903  * @param[in] item
1904  *   Item specification.
1905  * @param[in] item_flags
1906  *   Bit-fields that holds the items detected until now.
1907  * @param[out] error
1908  *   Pointer to error structure.
1909  *
1910  * @return
1911  *   0 on success, a negative errno value otherwise and rte_errno is set.
1912  */
1913 int
1914 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1915 			    uint64_t item_flags, bool ext_vlan_sup,
1916 			    struct rte_flow_error *error)
1917 {
1918 	const struct rte_flow_item_eth *mask = item->mask;
1919 	const struct rte_flow_item_eth nic_mask = {
1920 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1921 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1922 		.type = RTE_BE16(0xffff),
1923 		.has_vlan = ext_vlan_sup ? 1 : 0,
1924 	};
1925 	int ret;
1926 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1927 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1928 				       MLX5_FLOW_LAYER_OUTER_L2;
1929 
1930 	if (item_flags & ethm)
1931 		return rte_flow_error_set(error, ENOTSUP,
1932 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 					  "multiple L2 layers not supported");
1934 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1935 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1936 		return rte_flow_error_set(error, EINVAL,
1937 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1938 					  "L2 layer should not follow "
1939 					  "L3 layers");
1940 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1941 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1942 		return rte_flow_error_set(error, EINVAL,
1943 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 					  "L2 layer should not follow VLAN");
1945 	if (!mask)
1946 		mask = &rte_flow_item_eth_mask;
1947 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1948 					(const uint8_t *)&nic_mask,
1949 					sizeof(struct rte_flow_item_eth),
1950 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1951 	return ret;
1952 }
1953 
1954 /**
1955  * Validate VLAN item.
1956  *
1957  * @param[in] item
1958  *   Item specification.
1959  * @param[in] item_flags
1960  *   Bit-fields that holds the items detected until now.
1961  * @param[in] dev
1962  *   Ethernet device flow is being created on.
1963  * @param[out] error
1964  *   Pointer to error structure.
1965  *
1966  * @return
1967  *   0 on success, a negative errno value otherwise and rte_errno is set.
1968  */
1969 int
1970 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1971 			     uint64_t item_flags,
1972 			     struct rte_eth_dev *dev,
1973 			     struct rte_flow_error *error)
1974 {
1975 	const struct rte_flow_item_vlan *spec = item->spec;
1976 	const struct rte_flow_item_vlan *mask = item->mask;
1977 	const struct rte_flow_item_vlan nic_mask = {
1978 		.tci = RTE_BE16(UINT16_MAX),
1979 		.inner_type = RTE_BE16(UINT16_MAX),
1980 	};
1981 	uint16_t vlan_tag = 0;
1982 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1983 	int ret;
1984 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1985 					MLX5_FLOW_LAYER_INNER_L4) :
1986 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1987 					MLX5_FLOW_LAYER_OUTER_L4);
1988 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1989 					MLX5_FLOW_LAYER_OUTER_VLAN;
1990 
1991 	if (item_flags & vlanm)
1992 		return rte_flow_error_set(error, EINVAL,
1993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1994 					  "multiple VLAN layers not supported");
1995 	else if ((item_flags & l34m) != 0)
1996 		return rte_flow_error_set(error, EINVAL,
1997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 					  "VLAN cannot follow L3/L4 layer");
1999 	if (!mask)
2000 		mask = &rte_flow_item_vlan_mask;
2001 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2002 					(const uint8_t *)&nic_mask,
2003 					sizeof(struct rte_flow_item_vlan),
2004 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2005 	if (ret)
2006 		return ret;
2007 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2008 		struct mlx5_priv *priv = dev->data->dev_private;
2009 
2010 		if (priv->vmwa_context) {
2011 			/*
2012 			 * Non-NULL context means we have a virtual machine
2013 			 * and SR-IOV enabled, we have to create VLAN interface
2014 			 * to make hypervisor to setup E-Switch vport
2015 			 * context correctly. We avoid creating the multiple
2016 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2017 			 */
2018 			return rte_flow_error_set(error, EINVAL,
2019 						  RTE_FLOW_ERROR_TYPE_ITEM,
2020 						  item,
2021 						  "VLAN tag mask is not"
2022 						  " supported in virtual"
2023 						  " environment");
2024 		}
2025 	}
2026 	if (spec) {
2027 		vlan_tag = spec->tci;
2028 		vlan_tag &= mask->tci;
2029 	}
2030 	/*
2031 	 * From verbs perspective an empty VLAN is equivalent
2032 	 * to a packet without VLAN layer.
2033 	 */
2034 	if (!vlan_tag)
2035 		return rte_flow_error_set(error, EINVAL,
2036 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2037 					  item->spec,
2038 					  "VLAN cannot be empty");
2039 	return 0;
2040 }
2041 
2042 /**
2043  * Validate IPV4 item.
2044  *
2045  * @param[in] item
2046  *   Item specification.
2047  * @param[in] item_flags
2048  *   Bit-fields that holds the items detected until now.
2049  * @param[in] last_item
2050  *   Previous validated item in the pattern items.
2051  * @param[in] ether_type
2052  *   Type in the ethernet layer header (including dot1q).
2053  * @param[in] acc_mask
2054  *   Acceptable mask, if NULL default internal default mask
2055  *   will be used to check whether item fields are supported.
2056  * @param[in] range_accepted
2057  *   True if range of values is accepted for specific fields, false otherwise.
2058  * @param[out] error
2059  *   Pointer to error structure.
2060  *
2061  * @return
2062  *   0 on success, a negative errno value otherwise and rte_errno is set.
2063  */
2064 int
2065 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2066 			     uint64_t item_flags,
2067 			     uint64_t last_item,
2068 			     uint16_t ether_type,
2069 			     const struct rte_flow_item_ipv4 *acc_mask,
2070 			     bool range_accepted,
2071 			     struct rte_flow_error *error)
2072 {
2073 	const struct rte_flow_item_ipv4 *mask = item->mask;
2074 	const struct rte_flow_item_ipv4 *spec = item->spec;
2075 	const struct rte_flow_item_ipv4 nic_mask = {
2076 		.hdr = {
2077 			.src_addr = RTE_BE32(0xffffffff),
2078 			.dst_addr = RTE_BE32(0xffffffff),
2079 			.type_of_service = 0xff,
2080 			.next_proto_id = 0xff,
2081 		},
2082 	};
2083 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2084 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2085 				      MLX5_FLOW_LAYER_OUTER_L3;
2086 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2087 				      MLX5_FLOW_LAYER_OUTER_L4;
2088 	int ret;
2089 	uint8_t next_proto = 0xFF;
2090 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2091 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2092 				  MLX5_FLOW_LAYER_INNER_VLAN);
2093 
2094 	if ((last_item & l2_vlan) && ether_type &&
2095 	    ether_type != RTE_ETHER_TYPE_IPV4)
2096 		return rte_flow_error_set(error, EINVAL,
2097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 					  "IPv4 cannot follow L2/VLAN layer "
2099 					  "which ether type is not IPv4");
2100 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2101 		if (mask && spec)
2102 			next_proto = mask->hdr.next_proto_id &
2103 				     spec->hdr.next_proto_id;
2104 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2105 			return rte_flow_error_set(error, EINVAL,
2106 						  RTE_FLOW_ERROR_TYPE_ITEM,
2107 						  item,
2108 						  "multiple tunnel "
2109 						  "not supported");
2110 	}
2111 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2112 		return rte_flow_error_set(error, EINVAL,
2113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 					  "wrong tunnel type - IPv6 specified "
2115 					  "but IPv4 item provided");
2116 	if (item_flags & l3m)
2117 		return rte_flow_error_set(error, ENOTSUP,
2118 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2119 					  "multiple L3 layers not supported");
2120 	else if (item_flags & l4m)
2121 		return rte_flow_error_set(error, EINVAL,
2122 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2123 					  "L3 cannot follow an L4 layer.");
2124 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2125 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2126 		return rte_flow_error_set(error, EINVAL,
2127 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2128 					  "L3 cannot follow an NVGRE layer.");
2129 	if (!mask)
2130 		mask = &rte_flow_item_ipv4_mask;
2131 	else if (mask->hdr.next_proto_id != 0 &&
2132 		 mask->hdr.next_proto_id != 0xff)
2133 		return rte_flow_error_set(error, EINVAL,
2134 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2135 					  "partial mask is not supported"
2136 					  " for protocol");
2137 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2138 					acc_mask ? (const uint8_t *)acc_mask
2139 						 : (const uint8_t *)&nic_mask,
2140 					sizeof(struct rte_flow_item_ipv4),
2141 					range_accepted, error);
2142 	if (ret < 0)
2143 		return ret;
2144 	return 0;
2145 }
2146 
2147 /**
2148  * Validate IPV6 item.
2149  *
2150  * @param[in] item
2151  *   Item specification.
2152  * @param[in] item_flags
2153  *   Bit-fields that holds the items detected until now.
2154  * @param[in] last_item
2155  *   Previous validated item in the pattern items.
2156  * @param[in] ether_type
2157  *   Type in the ethernet layer header (including dot1q).
2158  * @param[in] acc_mask
2159  *   Acceptable mask, if NULL default internal default mask
2160  *   will be used to check whether item fields are supported.
2161  * @param[out] error
2162  *   Pointer to error structure.
2163  *
2164  * @return
2165  *   0 on success, a negative errno value otherwise and rte_errno is set.
2166  */
2167 int
2168 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2169 			     uint64_t item_flags,
2170 			     uint64_t last_item,
2171 			     uint16_t ether_type,
2172 			     const struct rte_flow_item_ipv6 *acc_mask,
2173 			     struct rte_flow_error *error)
2174 {
2175 	const struct rte_flow_item_ipv6 *mask = item->mask;
2176 	const struct rte_flow_item_ipv6 *spec = item->spec;
2177 	const struct rte_flow_item_ipv6 nic_mask = {
2178 		.hdr = {
2179 			.src_addr =
2180 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2181 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2182 			.dst_addr =
2183 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2184 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2185 			.vtc_flow = RTE_BE32(0xffffffff),
2186 			.proto = 0xff,
2187 		},
2188 	};
2189 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2190 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2191 				      MLX5_FLOW_LAYER_OUTER_L3;
2192 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2193 				      MLX5_FLOW_LAYER_OUTER_L4;
2194 	int ret;
2195 	uint8_t next_proto = 0xFF;
2196 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2197 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2198 				  MLX5_FLOW_LAYER_INNER_VLAN);
2199 
2200 	if ((last_item & l2_vlan) && ether_type &&
2201 	    ether_type != RTE_ETHER_TYPE_IPV6)
2202 		return rte_flow_error_set(error, EINVAL,
2203 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2204 					  "IPv6 cannot follow L2/VLAN layer "
2205 					  "which ether type is not IPv6");
2206 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2207 		next_proto = spec->hdr.proto;
2208 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2209 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2210 			return rte_flow_error_set(error, EINVAL,
2211 						  RTE_FLOW_ERROR_TYPE_ITEM,
2212 						  item,
2213 						  "multiple tunnel "
2214 						  "not supported");
2215 	}
2216 	if (next_proto == IPPROTO_HOPOPTS  ||
2217 	    next_proto == IPPROTO_ROUTING  ||
2218 	    next_proto == IPPROTO_FRAGMENT ||
2219 	    next_proto == IPPROTO_ESP	   ||
2220 	    next_proto == IPPROTO_AH	   ||
2221 	    next_proto == IPPROTO_DSTOPTS)
2222 		return rte_flow_error_set(error, EINVAL,
2223 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2224 					  "IPv6 proto (next header) should "
2225 					  "not be set as extension header");
2226 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2227 		return rte_flow_error_set(error, EINVAL,
2228 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 					  "wrong tunnel type - IPv4 specified "
2230 					  "but IPv6 item provided");
2231 	if (item_flags & l3m)
2232 		return rte_flow_error_set(error, ENOTSUP,
2233 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2234 					  "multiple L3 layers not supported");
2235 	else if (item_flags & l4m)
2236 		return rte_flow_error_set(error, EINVAL,
2237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 					  "L3 cannot follow an L4 layer.");
2239 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2240 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2241 		return rte_flow_error_set(error, EINVAL,
2242 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2243 					  "L3 cannot follow an NVGRE layer.");
2244 	if (!mask)
2245 		mask = &rte_flow_item_ipv6_mask;
2246 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2247 					acc_mask ? (const uint8_t *)acc_mask
2248 						 : (const uint8_t *)&nic_mask,
2249 					sizeof(struct rte_flow_item_ipv6),
2250 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2251 	if (ret < 0)
2252 		return ret;
2253 	return 0;
2254 }
2255 
2256 /**
2257  * Validate UDP item.
2258  *
2259  * @param[in] item
2260  *   Item specification.
2261  * @param[in] item_flags
2262  *   Bit-fields that holds the items detected until now.
2263  * @param[in] target_protocol
2264  *   The next protocol in the previous item.
2265  * @param[in] flow_mask
2266  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2267  * @param[out] error
2268  *   Pointer to error structure.
2269  *
2270  * @return
2271  *   0 on success, a negative errno value otherwise and rte_errno is set.
2272  */
2273 int
2274 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2275 			    uint64_t item_flags,
2276 			    uint8_t target_protocol,
2277 			    struct rte_flow_error *error)
2278 {
2279 	const struct rte_flow_item_udp *mask = item->mask;
2280 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2281 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2282 				      MLX5_FLOW_LAYER_OUTER_L3;
2283 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2284 				      MLX5_FLOW_LAYER_OUTER_L4;
2285 	int ret;
2286 
2287 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2288 		return rte_flow_error_set(error, EINVAL,
2289 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2290 					  "protocol filtering not compatible"
2291 					  " with UDP layer");
2292 	if (!(item_flags & l3m))
2293 		return rte_flow_error_set(error, EINVAL,
2294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 					  "L3 is mandatory to filter on L4");
2296 	if (item_flags & l4m)
2297 		return rte_flow_error_set(error, EINVAL,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "multiple L4 layers not supported");
2300 	if (!mask)
2301 		mask = &rte_flow_item_udp_mask;
2302 	ret = mlx5_flow_item_acceptable
2303 		(item, (const uint8_t *)mask,
2304 		 (const uint8_t *)&rte_flow_item_udp_mask,
2305 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2306 		 error);
2307 	if (ret < 0)
2308 		return ret;
2309 	return 0;
2310 }
2311 
2312 /**
2313  * Validate TCP item.
2314  *
2315  * @param[in] item
2316  *   Item specification.
2317  * @param[in] item_flags
2318  *   Bit-fields that holds the items detected until now.
2319  * @param[in] target_protocol
2320  *   The next protocol in the previous item.
2321  * @param[out] error
2322  *   Pointer to error structure.
2323  *
2324  * @return
2325  *   0 on success, a negative errno value otherwise and rte_errno is set.
2326  */
2327 int
2328 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2329 			    uint64_t item_flags,
2330 			    uint8_t target_protocol,
2331 			    const struct rte_flow_item_tcp *flow_mask,
2332 			    struct rte_flow_error *error)
2333 {
2334 	const struct rte_flow_item_tcp *mask = item->mask;
2335 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2336 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2337 				      MLX5_FLOW_LAYER_OUTER_L3;
2338 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2339 				      MLX5_FLOW_LAYER_OUTER_L4;
2340 	int ret;
2341 
2342 	MLX5_ASSERT(flow_mask);
2343 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2344 		return rte_flow_error_set(error, EINVAL,
2345 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2346 					  "protocol filtering not compatible"
2347 					  " with TCP layer");
2348 	if (!(item_flags & l3m))
2349 		return rte_flow_error_set(error, EINVAL,
2350 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2351 					  "L3 is mandatory to filter on L4");
2352 	if (item_flags & l4m)
2353 		return rte_flow_error_set(error, EINVAL,
2354 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 					  "multiple L4 layers not supported");
2356 	if (!mask)
2357 		mask = &rte_flow_item_tcp_mask;
2358 	ret = mlx5_flow_item_acceptable
2359 		(item, (const uint8_t *)mask,
2360 		 (const uint8_t *)flow_mask,
2361 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2362 		 error);
2363 	if (ret < 0)
2364 		return ret;
2365 	return 0;
2366 }
2367 
2368 /**
2369  * Validate VXLAN item.
2370  *
2371  * @param[in] item
2372  *   Item specification.
2373  * @param[in] item_flags
2374  *   Bit-fields that holds the items detected until now.
2375  * @param[in] target_protocol
2376  *   The next protocol in the previous item.
2377  * @param[out] error
2378  *   Pointer to error structure.
2379  *
2380  * @return
2381  *   0 on success, a negative errno value otherwise and rte_errno is set.
2382  */
2383 int
2384 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2385 			      uint64_t item_flags,
2386 			      struct rte_flow_error *error)
2387 {
2388 	const struct rte_flow_item_vxlan *spec = item->spec;
2389 	const struct rte_flow_item_vxlan *mask = item->mask;
2390 	int ret;
2391 	union vni {
2392 		uint32_t vlan_id;
2393 		uint8_t vni[4];
2394 	} id = { .vlan_id = 0, };
2395 
2396 
2397 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2398 		return rte_flow_error_set(error, ENOTSUP,
2399 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2400 					  "multiple tunnel layers not"
2401 					  " supported");
2402 	/*
2403 	 * Verify only UDPv4 is present as defined in
2404 	 * https://tools.ietf.org/html/rfc7348
2405 	 */
2406 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2407 		return rte_flow_error_set(error, EINVAL,
2408 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2409 					  "no outer UDP layer found");
2410 	if (!mask)
2411 		mask = &rte_flow_item_vxlan_mask;
2412 	ret = mlx5_flow_item_acceptable
2413 		(item, (const uint8_t *)mask,
2414 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2415 		 sizeof(struct rte_flow_item_vxlan),
2416 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2417 	if (ret < 0)
2418 		return ret;
2419 	if (spec) {
2420 		memcpy(&id.vni[1], spec->vni, 3);
2421 		memcpy(&id.vni[1], mask->vni, 3);
2422 	}
2423 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2424 		return rte_flow_error_set(error, ENOTSUP,
2425 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2426 					  "VXLAN tunnel must be fully defined");
2427 	return 0;
2428 }
2429 
2430 /**
2431  * Validate VXLAN_GPE item.
2432  *
2433  * @param[in] item
2434  *   Item specification.
2435  * @param[in] item_flags
2436  *   Bit-fields that holds the items detected until now.
2437  * @param[in] priv
2438  *   Pointer to the private data structure.
2439  * @param[in] target_protocol
2440  *   The next protocol in the previous item.
2441  * @param[out] error
2442  *   Pointer to error structure.
2443  *
2444  * @return
2445  *   0 on success, a negative errno value otherwise and rte_errno is set.
2446  */
2447 int
2448 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2449 				  uint64_t item_flags,
2450 				  struct rte_eth_dev *dev,
2451 				  struct rte_flow_error *error)
2452 {
2453 	struct mlx5_priv *priv = dev->data->dev_private;
2454 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2455 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2456 	int ret;
2457 	union vni {
2458 		uint32_t vlan_id;
2459 		uint8_t vni[4];
2460 	} id = { .vlan_id = 0, };
2461 
2462 	if (!priv->config.l3_vxlan_en)
2463 		return rte_flow_error_set(error, ENOTSUP,
2464 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2465 					  "L3 VXLAN is not enabled by device"
2466 					  " parameter and/or not configured in"
2467 					  " firmware");
2468 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2469 		return rte_flow_error_set(error, ENOTSUP,
2470 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 					  "multiple tunnel layers not"
2472 					  " supported");
2473 	/*
2474 	 * Verify only UDPv4 is present as defined in
2475 	 * https://tools.ietf.org/html/rfc7348
2476 	 */
2477 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2478 		return rte_flow_error_set(error, EINVAL,
2479 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2480 					  "no outer UDP layer found");
2481 	if (!mask)
2482 		mask = &rte_flow_item_vxlan_gpe_mask;
2483 	ret = mlx5_flow_item_acceptable
2484 		(item, (const uint8_t *)mask,
2485 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2486 		 sizeof(struct rte_flow_item_vxlan_gpe),
2487 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2488 	if (ret < 0)
2489 		return ret;
2490 	if (spec) {
2491 		if (spec->protocol)
2492 			return rte_flow_error_set(error, ENOTSUP,
2493 						  RTE_FLOW_ERROR_TYPE_ITEM,
2494 						  item,
2495 						  "VxLAN-GPE protocol"
2496 						  " not supported");
2497 		memcpy(&id.vni[1], spec->vni, 3);
2498 		memcpy(&id.vni[1], mask->vni, 3);
2499 	}
2500 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2501 		return rte_flow_error_set(error, ENOTSUP,
2502 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2503 					  "VXLAN-GPE tunnel must be fully"
2504 					  " defined");
2505 	return 0;
2506 }
2507 /**
2508  * Validate GRE Key item.
2509  *
2510  * @param[in] item
2511  *   Item specification.
2512  * @param[in] item_flags
2513  *   Bit flags to mark detected items.
2514  * @param[in] gre_item
2515  *   Pointer to gre_item
2516  * @param[out] error
2517  *   Pointer to error structure.
2518  *
2519  * @return
2520  *   0 on success, a negative errno value otherwise and rte_errno is set.
2521  */
2522 int
2523 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2524 				uint64_t item_flags,
2525 				const struct rte_flow_item *gre_item,
2526 				struct rte_flow_error *error)
2527 {
2528 	const rte_be32_t *mask = item->mask;
2529 	int ret = 0;
2530 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2531 	const struct rte_flow_item_gre *gre_spec;
2532 	const struct rte_flow_item_gre *gre_mask;
2533 
2534 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2535 		return rte_flow_error_set(error, ENOTSUP,
2536 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2537 					  "Multiple GRE key not support");
2538 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2539 		return rte_flow_error_set(error, ENOTSUP,
2540 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2541 					  "No preceding GRE header");
2542 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2543 		return rte_flow_error_set(error, ENOTSUP,
2544 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2545 					  "GRE key following a wrong item");
2546 	gre_mask = gre_item->mask;
2547 	if (!gre_mask)
2548 		gre_mask = &rte_flow_item_gre_mask;
2549 	gre_spec = gre_item->spec;
2550 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2551 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2552 		return rte_flow_error_set(error, EINVAL,
2553 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2554 					  "Key bit must be on");
2555 
2556 	if (!mask)
2557 		mask = &gre_key_default_mask;
2558 	ret = mlx5_flow_item_acceptable
2559 		(item, (const uint8_t *)mask,
2560 		 (const uint8_t *)&gre_key_default_mask,
2561 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2562 	return ret;
2563 }
2564 
2565 /**
2566  * Validate GRE item.
2567  *
2568  * @param[in] item
2569  *   Item specification.
2570  * @param[in] item_flags
2571  *   Bit flags to mark detected items.
2572  * @param[in] target_protocol
2573  *   The next protocol in the previous item.
2574  * @param[out] error
2575  *   Pointer to error structure.
2576  *
2577  * @return
2578  *   0 on success, a negative errno value otherwise and rte_errno is set.
2579  */
2580 int
2581 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2582 			    uint64_t item_flags,
2583 			    uint8_t target_protocol,
2584 			    struct rte_flow_error *error)
2585 {
2586 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2587 	const struct rte_flow_item_gre *mask = item->mask;
2588 	int ret;
2589 	const struct rte_flow_item_gre nic_mask = {
2590 		.c_rsvd0_ver = RTE_BE16(0xB000),
2591 		.protocol = RTE_BE16(UINT16_MAX),
2592 	};
2593 
2594 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2595 		return rte_flow_error_set(error, EINVAL,
2596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2597 					  "protocol filtering not compatible"
2598 					  " with this GRE layer");
2599 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2600 		return rte_flow_error_set(error, ENOTSUP,
2601 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2602 					  "multiple tunnel layers not"
2603 					  " supported");
2604 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2605 		return rte_flow_error_set(error, ENOTSUP,
2606 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2607 					  "L3 Layer is missing");
2608 	if (!mask)
2609 		mask = &rte_flow_item_gre_mask;
2610 	ret = mlx5_flow_item_acceptable
2611 		(item, (const uint8_t *)mask,
2612 		 (const uint8_t *)&nic_mask,
2613 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2614 		 error);
2615 	if (ret < 0)
2616 		return ret;
2617 #ifndef HAVE_MLX5DV_DR
2618 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2619 	if (spec && (spec->protocol & mask->protocol))
2620 		return rte_flow_error_set(error, ENOTSUP,
2621 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2622 					  "without MPLS support the"
2623 					  " specification cannot be used for"
2624 					  " filtering");
2625 #endif
2626 #endif
2627 	return 0;
2628 }
2629 
2630 /**
2631  * Validate Geneve item.
2632  *
2633  * @param[in] item
2634  *   Item specification.
2635  * @param[in] itemFlags
2636  *   Bit-fields that holds the items detected until now.
2637  * @param[in] enPriv
2638  *   Pointer to the private data structure.
2639  * @param[out] error
2640  *   Pointer to error structure.
2641  *
2642  * @return
2643  *   0 on success, a negative errno value otherwise and rte_errno is set.
2644  */
2645 
2646 int
2647 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2648 			       uint64_t item_flags,
2649 			       struct rte_eth_dev *dev,
2650 			       struct rte_flow_error *error)
2651 {
2652 	struct mlx5_priv *priv = dev->data->dev_private;
2653 	const struct rte_flow_item_geneve *spec = item->spec;
2654 	const struct rte_flow_item_geneve *mask = item->mask;
2655 	int ret;
2656 	uint16_t gbhdr;
2657 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2658 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2659 	const struct rte_flow_item_geneve nic_mask = {
2660 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2661 		.vni = "\xff\xff\xff",
2662 		.protocol = RTE_BE16(UINT16_MAX),
2663 	};
2664 
2665 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2666 		return rte_flow_error_set(error, ENOTSUP,
2667 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2668 					  "L3 Geneve is not enabled by device"
2669 					  " parameter and/or not configured in"
2670 					  " firmware");
2671 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2672 		return rte_flow_error_set(error, ENOTSUP,
2673 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2674 					  "multiple tunnel layers not"
2675 					  " supported");
2676 	/*
2677 	 * Verify only UDPv4 is present as defined in
2678 	 * https://tools.ietf.org/html/rfc7348
2679 	 */
2680 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2681 		return rte_flow_error_set(error, EINVAL,
2682 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2683 					  "no outer UDP layer found");
2684 	if (!mask)
2685 		mask = &rte_flow_item_geneve_mask;
2686 	ret = mlx5_flow_item_acceptable
2687 				  (item, (const uint8_t *)mask,
2688 				   (const uint8_t *)&nic_mask,
2689 				   sizeof(struct rte_flow_item_geneve),
2690 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2691 	if (ret)
2692 		return ret;
2693 	if (spec) {
2694 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2695 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2696 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2697 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2698 			return rte_flow_error_set(error, ENOTSUP,
2699 						  RTE_FLOW_ERROR_TYPE_ITEM,
2700 						  item,
2701 						  "Geneve protocol unsupported"
2702 						  " fields are being used");
2703 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2704 			return rte_flow_error_set
2705 					(error, ENOTSUP,
2706 					 RTE_FLOW_ERROR_TYPE_ITEM,
2707 					 item,
2708 					 "Unsupported Geneve options length");
2709 	}
2710 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2711 		return rte_flow_error_set
2712 				    (error, ENOTSUP,
2713 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2714 				     "Geneve tunnel must be fully defined");
2715 	return 0;
2716 }
2717 
2718 /**
2719  * Validate Geneve TLV option item.
2720  *
2721  * @param[in] item
2722  *   Item specification.
2723  * @param[in] last_item
2724  *   Previous validated item in the pattern items.
2725  * @param[in] geneve_item
2726  *   Previous GENEVE item specification.
2727  * @param[in] dev
2728  *   Pointer to the rte_eth_dev structure.
2729  * @param[out] error
2730  *   Pointer to error structure.
2731  *
2732  * @return
2733  *   0 on success, a negative errno value otherwise and rte_errno is set.
2734  */
2735 int
2736 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2737 				   uint64_t last_item,
2738 				   const struct rte_flow_item *geneve_item,
2739 				   struct rte_eth_dev *dev,
2740 				   struct rte_flow_error *error)
2741 {
2742 	struct mlx5_priv *priv = dev->data->dev_private;
2743 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2744 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2745 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2746 	uint8_t data_max_supported =
2747 			hca_attr->max_geneve_tlv_option_data_len * 4;
2748 	struct mlx5_dev_config *config = &priv->config;
2749 	const struct rte_flow_item_geneve *geneve_spec;
2750 	const struct rte_flow_item_geneve *geneve_mask;
2751 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2752 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2753 	unsigned int i;
2754 	unsigned int data_len;
2755 	uint8_t tlv_option_len;
2756 	uint16_t optlen_m, optlen_v;
2757 	const struct rte_flow_item_geneve_opt full_mask = {
2758 		.option_class = RTE_BE16(0xffff),
2759 		.option_type = 0xff,
2760 		.option_len = 0x1f,
2761 	};
2762 
2763 	if (!mask)
2764 		mask = &rte_flow_item_geneve_opt_mask;
2765 	if (!spec)
2766 		return rte_flow_error_set
2767 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2768 			"Geneve TLV opt class/type/length must be specified");
2769 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2770 		return rte_flow_error_set
2771 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 			"Geneve TLV opt length exceeeds the limit (31)");
2773 	/* Check if class type and length masks are full. */
2774 	if (full_mask.option_class != mask->option_class ||
2775 	    full_mask.option_type != mask->option_type ||
2776 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2777 		return rte_flow_error_set
2778 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2779 			"Geneve TLV opt class/type/length masks must be full");
2780 	/* Check if length is supported */
2781 	if ((uint32_t)spec->option_len >
2782 			config->hca_attr.max_geneve_tlv_option_data_len)
2783 		return rte_flow_error_set
2784 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2785 			"Geneve TLV opt length not supported");
2786 	if (config->hca_attr.max_geneve_tlv_options > 1)
2787 		DRV_LOG(DEBUG,
2788 			"max_geneve_tlv_options supports more than 1 option");
2789 	/* Check GENEVE item preceding. */
2790 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2791 		return rte_flow_error_set
2792 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 			"Geneve opt item must be preceded with Geneve item");
2794 	geneve_spec = geneve_item->spec;
2795 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2796 					  &rte_flow_item_geneve_mask;
2797 	/* Check if GENEVE TLV option size doesn't exceed option length */
2798 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2799 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2800 		tlv_option_len = spec->option_len & mask->option_len;
2801 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2802 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2803 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2804 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2805 		if ((optlen_v & optlen_m) <= tlv_option_len)
2806 			return rte_flow_error_set
2807 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2808 				 "GENEVE TLV option length exceeds optlen");
2809 	}
2810 	/* Check if length is 0 or data is 0. */
2811 	if (spec->data == NULL || spec->option_len == 0)
2812 		return rte_flow_error_set
2813 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2814 			"Geneve TLV opt with zero data/length not supported");
2815 	/* Check not all data & mask are 0. */
2816 	data_len = spec->option_len * 4;
2817 	if (mask->data == NULL) {
2818 		for (i = 0; i < data_len; i++)
2819 			if (spec->data[i])
2820 				break;
2821 		if (i == data_len)
2822 			return rte_flow_error_set(error, ENOTSUP,
2823 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2824 				"Can't match on Geneve option data 0");
2825 	} else {
2826 		for (i = 0; i < data_len; i++)
2827 			if (spec->data[i] & mask->data[i])
2828 				break;
2829 		if (i == data_len)
2830 			return rte_flow_error_set(error, ENOTSUP,
2831 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2832 				"Can't match on Geneve option data and mask 0");
2833 		/* Check data mask supported. */
2834 		for (i = data_max_supported; i < data_len ; i++)
2835 			if (mask->data[i])
2836 				return rte_flow_error_set(error, ENOTSUP,
2837 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2838 					"Data mask is of unsupported size");
2839 	}
2840 	/* Check GENEVE option is supported in NIC. */
2841 	if (!config->hca_attr.geneve_tlv_opt)
2842 		return rte_flow_error_set
2843 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2844 			"Geneve TLV opt not supported");
2845 	/* Check if we already have geneve option with different type/class. */
2846 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2847 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2848 	if (geneve_opt_resource != NULL)
2849 		if (geneve_opt_resource->option_class != spec->option_class ||
2850 		    geneve_opt_resource->option_type != spec->option_type ||
2851 		    geneve_opt_resource->length != spec->option_len) {
2852 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2853 			return rte_flow_error_set(error, ENOTSUP,
2854 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2855 				"Only one Geneve TLV option supported");
2856 		}
2857 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2858 	return 0;
2859 }
2860 
2861 /**
2862  * Validate MPLS item.
2863  *
2864  * @param[in] dev
2865  *   Pointer to the rte_eth_dev structure.
2866  * @param[in] item
2867  *   Item specification.
2868  * @param[in] item_flags
2869  *   Bit-fields that holds the items detected until now.
2870  * @param[in] prev_layer
2871  *   The protocol layer indicated in previous item.
2872  * @param[out] error
2873  *   Pointer to error structure.
2874  *
2875  * @return
2876  *   0 on success, a negative errno value otherwise and rte_errno is set.
2877  */
2878 int
2879 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2880 			     const struct rte_flow_item *item __rte_unused,
2881 			     uint64_t item_flags __rte_unused,
2882 			     uint64_t prev_layer __rte_unused,
2883 			     struct rte_flow_error *error)
2884 {
2885 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2886 	const struct rte_flow_item_mpls *mask = item->mask;
2887 	struct mlx5_priv *priv = dev->data->dev_private;
2888 	int ret;
2889 
2890 	if (!priv->config.mpls_en)
2891 		return rte_flow_error_set(error, ENOTSUP,
2892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2893 					  "MPLS not supported or"
2894 					  " disabled in firmware"
2895 					  " configuration.");
2896 	/* MPLS over IP, UDP, GRE is allowed */
2897 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2898 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2899 			    MLX5_FLOW_LAYER_GRE |
2900 			    MLX5_FLOW_LAYER_GRE_KEY)))
2901 		return rte_flow_error_set(error, EINVAL,
2902 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2903 					  "protocol filtering not compatible"
2904 					  " with MPLS layer");
2905 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2906 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2907 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2908 		return rte_flow_error_set(error, ENOTSUP,
2909 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2910 					  "multiple tunnel layers not"
2911 					  " supported");
2912 	if (!mask)
2913 		mask = &rte_flow_item_mpls_mask;
2914 	ret = mlx5_flow_item_acceptable
2915 		(item, (const uint8_t *)mask,
2916 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2917 		 sizeof(struct rte_flow_item_mpls),
2918 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2919 	if (ret < 0)
2920 		return ret;
2921 	return 0;
2922 #else
2923 	return rte_flow_error_set(error, ENOTSUP,
2924 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2925 				  "MPLS is not supported by Verbs, please"
2926 				  " update.");
2927 #endif
2928 }
2929 
2930 /**
2931  * Validate NVGRE item.
2932  *
2933  * @param[in] item
2934  *   Item specification.
2935  * @param[in] item_flags
2936  *   Bit flags to mark detected items.
2937  * @param[in] target_protocol
2938  *   The next protocol in the previous item.
2939  * @param[out] error
2940  *   Pointer to error structure.
2941  *
2942  * @return
2943  *   0 on success, a negative errno value otherwise and rte_errno is set.
2944  */
2945 int
2946 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2947 			      uint64_t item_flags,
2948 			      uint8_t target_protocol,
2949 			      struct rte_flow_error *error)
2950 {
2951 	const struct rte_flow_item_nvgre *mask = item->mask;
2952 	int ret;
2953 
2954 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2955 		return rte_flow_error_set(error, EINVAL,
2956 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2957 					  "protocol filtering not compatible"
2958 					  " with this GRE layer");
2959 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2960 		return rte_flow_error_set(error, ENOTSUP,
2961 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2962 					  "multiple tunnel layers not"
2963 					  " supported");
2964 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2965 		return rte_flow_error_set(error, ENOTSUP,
2966 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 					  "L3 Layer is missing");
2968 	if (!mask)
2969 		mask = &rte_flow_item_nvgre_mask;
2970 	ret = mlx5_flow_item_acceptable
2971 		(item, (const uint8_t *)mask,
2972 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2973 		 sizeof(struct rte_flow_item_nvgre),
2974 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2975 	if (ret < 0)
2976 		return ret;
2977 	return 0;
2978 }
2979 
2980 /**
2981  * Validate eCPRI item.
2982  *
2983  * @param[in] item
2984  *   Item specification.
2985  * @param[in] item_flags
2986  *   Bit-fields that holds the items detected until now.
2987  * @param[in] last_item
2988  *   Previous validated item in the pattern items.
2989  * @param[in] ether_type
2990  *   Type in the ethernet layer header (including dot1q).
2991  * @param[in] acc_mask
2992  *   Acceptable mask, if NULL default internal default mask
2993  *   will be used to check whether item fields are supported.
2994  * @param[out] error
2995  *   Pointer to error structure.
2996  *
2997  * @return
2998  *   0 on success, a negative errno value otherwise and rte_errno is set.
2999  */
3000 int
3001 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3002 			      uint64_t item_flags,
3003 			      uint64_t last_item,
3004 			      uint16_t ether_type,
3005 			      const struct rte_flow_item_ecpri *acc_mask,
3006 			      struct rte_flow_error *error)
3007 {
3008 	const struct rte_flow_item_ecpri *mask = item->mask;
3009 	const struct rte_flow_item_ecpri nic_mask = {
3010 		.hdr = {
3011 			.common = {
3012 				.u32 =
3013 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3014 					.type = 0xFF,
3015 					}).u32),
3016 			},
3017 			.dummy[0] = 0xFFFFFFFF,
3018 		},
3019 	};
3020 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3021 					MLX5_FLOW_LAYER_OUTER_VLAN);
3022 	struct rte_flow_item_ecpri mask_lo;
3023 
3024 	if (!(last_item & outer_l2_vlan) &&
3025 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3026 		return rte_flow_error_set(error, EINVAL,
3027 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3028 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3029 	if ((last_item & outer_l2_vlan) && ether_type &&
3030 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3031 		return rte_flow_error_set(error, EINVAL,
3032 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3033 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3034 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3035 		return rte_flow_error_set(error, EINVAL,
3036 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 					  "eCPRI with tunnel is not supported right now");
3038 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3039 		return rte_flow_error_set(error, ENOTSUP,
3040 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3041 					  "multiple L3 layers not supported");
3042 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3043 		return rte_flow_error_set(error, EINVAL,
3044 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3045 					  "eCPRI cannot coexist with a TCP layer");
3046 	/* In specification, eCPRI could be over UDP layer. */
3047 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3048 		return rte_flow_error_set(error, EINVAL,
3049 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3050 					  "eCPRI over UDP layer is not yet supported right now");
3051 	/* Mask for type field in common header could be zero. */
3052 	if (!mask)
3053 		mask = &rte_flow_item_ecpri_mask;
3054 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3055 	/* Input mask is in big-endian format. */
3056 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3057 		return rte_flow_error_set(error, EINVAL,
3058 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3059 					  "partial mask is not supported for protocol");
3060 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3061 		return rte_flow_error_set(error, EINVAL,
3062 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3063 					  "message header mask must be after a type mask");
3064 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3065 					 acc_mask ? (const uint8_t *)acc_mask
3066 						  : (const uint8_t *)&nic_mask,
3067 					 sizeof(struct rte_flow_item_ecpri),
3068 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3069 }
3070 
3071 /**
3072  * Release resource related QUEUE/RSS action split.
3073  *
3074  * @param dev
3075  *   Pointer to Ethernet device.
3076  * @param flow
3077  *   Flow to release id's from.
3078  */
3079 static void
3080 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3081 			     struct rte_flow *flow)
3082 {
3083 	struct mlx5_priv *priv = dev->data->dev_private;
3084 	uint32_t handle_idx;
3085 	struct mlx5_flow_handle *dev_handle;
3086 
3087 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3088 		       handle_idx, dev_handle, next)
3089 		if (dev_handle->split_flow_id &&
3090 		    !dev_handle->is_meter_flow_id)
3091 			mlx5_ipool_free(priv->sh->ipool
3092 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3093 					dev_handle->split_flow_id);
3094 }
3095 
3096 static int
3097 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3098 		   const struct rte_flow_attr *attr __rte_unused,
3099 		   const struct rte_flow_item items[] __rte_unused,
3100 		   const struct rte_flow_action actions[] __rte_unused,
3101 		   bool external __rte_unused,
3102 		   int hairpin __rte_unused,
3103 		   struct rte_flow_error *error)
3104 {
3105 	return rte_flow_error_set(error, ENOTSUP,
3106 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3107 }
3108 
3109 static struct mlx5_flow *
3110 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3111 		  const struct rte_flow_attr *attr __rte_unused,
3112 		  const struct rte_flow_item items[] __rte_unused,
3113 		  const struct rte_flow_action actions[] __rte_unused,
3114 		  struct rte_flow_error *error)
3115 {
3116 	rte_flow_error_set(error, ENOTSUP,
3117 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3118 	return NULL;
3119 }
3120 
3121 static int
3122 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3123 		    struct mlx5_flow *dev_flow __rte_unused,
3124 		    const struct rte_flow_attr *attr __rte_unused,
3125 		    const struct rte_flow_item items[] __rte_unused,
3126 		    const struct rte_flow_action actions[] __rte_unused,
3127 		    struct rte_flow_error *error)
3128 {
3129 	return rte_flow_error_set(error, ENOTSUP,
3130 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3131 }
3132 
3133 static int
3134 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3135 		struct rte_flow *flow __rte_unused,
3136 		struct rte_flow_error *error)
3137 {
3138 	return rte_flow_error_set(error, ENOTSUP,
3139 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3140 }
3141 
3142 static void
3143 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3144 		 struct rte_flow *flow __rte_unused)
3145 {
3146 }
3147 
3148 static void
3149 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3150 		  struct rte_flow *flow __rte_unused)
3151 {
3152 }
3153 
3154 static int
3155 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3156 		struct rte_flow *flow __rte_unused,
3157 		const struct rte_flow_action *actions __rte_unused,
3158 		void *data __rte_unused,
3159 		struct rte_flow_error *error)
3160 {
3161 	return rte_flow_error_set(error, ENOTSUP,
3162 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3163 }
3164 
3165 static int
3166 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3167 		      uint32_t domains __rte_unused,
3168 		      uint32_t flags __rte_unused)
3169 {
3170 	return 0;
3171 }
3172 
3173 /* Void driver to protect from null pointer reference. */
3174 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3175 	.validate = flow_null_validate,
3176 	.prepare = flow_null_prepare,
3177 	.translate = flow_null_translate,
3178 	.apply = flow_null_apply,
3179 	.remove = flow_null_remove,
3180 	.destroy = flow_null_destroy,
3181 	.query = flow_null_query,
3182 	.sync_domain = flow_null_sync_domain,
3183 };
3184 
3185 /**
3186  * Select flow driver type according to flow attributes and device
3187  * configuration.
3188  *
3189  * @param[in] dev
3190  *   Pointer to the dev structure.
3191  * @param[in] attr
3192  *   Pointer to the flow attributes.
3193  *
3194  * @return
3195  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3196  */
3197 static enum mlx5_flow_drv_type
3198 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3199 {
3200 	struct mlx5_priv *priv = dev->data->dev_private;
3201 	/* The OS can determine first a specific flow type (DV, VERBS) */
3202 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3203 
3204 	if (type != MLX5_FLOW_TYPE_MAX)
3205 		return type;
3206 	/* If no OS specific type - continue with DV/VERBS selection */
3207 	if (attr->transfer && priv->config.dv_esw_en)
3208 		type = MLX5_FLOW_TYPE_DV;
3209 	if (!attr->transfer)
3210 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3211 						 MLX5_FLOW_TYPE_VERBS;
3212 	return type;
3213 }
3214 
3215 #define flow_get_drv_ops(type) flow_drv_ops[type]
3216 
3217 /**
3218  * Flow driver validation API. This abstracts calling driver specific functions.
3219  * The type of flow driver is determined according to flow attributes.
3220  *
3221  * @param[in] dev
3222  *   Pointer to the dev structure.
3223  * @param[in] attr
3224  *   Pointer to the flow attributes.
3225  * @param[in] items
3226  *   Pointer to the list of items.
3227  * @param[in] actions
3228  *   Pointer to the list of actions.
3229  * @param[in] external
3230  *   This flow rule is created by request external to PMD.
3231  * @param[in] hairpin
3232  *   Number of hairpin TX actions, 0 means classic flow.
3233  * @param[out] error
3234  *   Pointer to the error structure.
3235  *
3236  * @return
3237  *   0 on success, a negative errno value otherwise and rte_errno is set.
3238  */
3239 static inline int
3240 flow_drv_validate(struct rte_eth_dev *dev,
3241 		  const struct rte_flow_attr *attr,
3242 		  const struct rte_flow_item items[],
3243 		  const struct rte_flow_action actions[],
3244 		  bool external, int hairpin, struct rte_flow_error *error)
3245 {
3246 	const struct mlx5_flow_driver_ops *fops;
3247 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3248 
3249 	fops = flow_get_drv_ops(type);
3250 	return fops->validate(dev, attr, items, actions, external,
3251 			      hairpin, error);
3252 }
3253 
3254 /**
3255  * Flow driver preparation API. This abstracts calling driver specific
3256  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3257  * calculates the size of memory required for device flow, allocates the memory,
3258  * initializes the device flow and returns the pointer.
3259  *
3260  * @note
3261  *   This function initializes device flow structure such as dv or verbs in
3262  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3263  *   rest. For example, adding returning device flow to flow->dev_flow list and
3264  *   setting backward reference to the flow should be done out of this function.
3265  *   layers field is not filled either.
3266  *
3267  * @param[in] dev
3268  *   Pointer to the dev structure.
3269  * @param[in] attr
3270  *   Pointer to the flow attributes.
3271  * @param[in] items
3272  *   Pointer to the list of items.
3273  * @param[in] actions
3274  *   Pointer to the list of actions.
3275  * @param[in] flow_idx
3276  *   This memory pool index to the flow.
3277  * @param[out] error
3278  *   Pointer to the error structure.
3279  *
3280  * @return
3281  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3282  */
3283 static inline struct mlx5_flow *
3284 flow_drv_prepare(struct rte_eth_dev *dev,
3285 		 const struct rte_flow *flow,
3286 		 const struct rte_flow_attr *attr,
3287 		 const struct rte_flow_item items[],
3288 		 const struct rte_flow_action actions[],
3289 		 uint32_t flow_idx,
3290 		 struct rte_flow_error *error)
3291 {
3292 	const struct mlx5_flow_driver_ops *fops;
3293 	enum mlx5_flow_drv_type type = flow->drv_type;
3294 	struct mlx5_flow *mlx5_flow = NULL;
3295 
3296 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3297 	fops = flow_get_drv_ops(type);
3298 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3299 	if (mlx5_flow)
3300 		mlx5_flow->flow_idx = flow_idx;
3301 	return mlx5_flow;
3302 }
3303 
3304 /**
3305  * Flow driver translation API. This abstracts calling driver specific
3306  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3307  * translates a generic flow into a driver flow. flow_drv_prepare() must
3308  * precede.
3309  *
3310  * @note
3311  *   dev_flow->layers could be filled as a result of parsing during translation
3312  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3313  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3314  *   flow->actions could be overwritten even though all the expanded dev_flows
3315  *   have the same actions.
3316  *
3317  * @param[in] dev
3318  *   Pointer to the rte dev structure.
3319  * @param[in, out] dev_flow
3320  *   Pointer to the mlx5 flow.
3321  * @param[in] attr
3322  *   Pointer to the flow attributes.
3323  * @param[in] items
3324  *   Pointer to the list of items.
3325  * @param[in] actions
3326  *   Pointer to the list of actions.
3327  * @param[out] error
3328  *   Pointer to the error structure.
3329  *
3330  * @return
3331  *   0 on success, a negative errno value otherwise and rte_errno is set.
3332  */
3333 static inline int
3334 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3335 		   const struct rte_flow_attr *attr,
3336 		   const struct rte_flow_item items[],
3337 		   const struct rte_flow_action actions[],
3338 		   struct rte_flow_error *error)
3339 {
3340 	const struct mlx5_flow_driver_ops *fops;
3341 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3342 
3343 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3344 	fops = flow_get_drv_ops(type);
3345 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3346 }
3347 
3348 /**
3349  * Flow driver apply API. This abstracts calling driver specific functions.
3350  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3351  * translated driver flows on to device. flow_drv_translate() must precede.
3352  *
3353  * @param[in] dev
3354  *   Pointer to Ethernet device structure.
3355  * @param[in, out] flow
3356  *   Pointer to flow structure.
3357  * @param[out] error
3358  *   Pointer to error structure.
3359  *
3360  * @return
3361  *   0 on success, a negative errno value otherwise and rte_errno is set.
3362  */
3363 static inline int
3364 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3365 	       struct rte_flow_error *error)
3366 {
3367 	const struct mlx5_flow_driver_ops *fops;
3368 	enum mlx5_flow_drv_type type = flow->drv_type;
3369 
3370 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3371 	fops = flow_get_drv_ops(type);
3372 	return fops->apply(dev, flow, error);
3373 }
3374 
3375 /**
3376  * Flow driver destroy API. This abstracts calling driver specific functions.
3377  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3378  * on device and releases resources of the flow.
3379  *
3380  * @param[in] dev
3381  *   Pointer to Ethernet device.
3382  * @param[in, out] flow
3383  *   Pointer to flow structure.
3384  */
3385 static inline void
3386 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3387 {
3388 	const struct mlx5_flow_driver_ops *fops;
3389 	enum mlx5_flow_drv_type type = flow->drv_type;
3390 
3391 	flow_mreg_split_qrss_release(dev, flow);
3392 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3393 	fops = flow_get_drv_ops(type);
3394 	fops->destroy(dev, flow);
3395 }
3396 
3397 /**
3398  * Flow driver find RSS policy tbl API. This abstracts calling driver
3399  * specific functions. Parent flow (rte_flow) should have driver
3400  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3401  *
3402  * @param[in] dev
3403  *   Pointer to Ethernet device.
3404  * @param[in, out] flow
3405  *   Pointer to flow structure.
3406  * @param[in] policy
3407  *   Pointer to meter policy table.
3408  * @param[in] rss_desc
3409  *   Pointer to rss_desc
3410  */
3411 static struct mlx5_flow_meter_sub_policy *
3412 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3413 		struct rte_flow *flow,
3414 		struct mlx5_flow_meter_policy *policy,
3415 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3416 {
3417 	const struct mlx5_flow_driver_ops *fops;
3418 	enum mlx5_flow_drv_type type = flow->drv_type;
3419 
3420 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3421 	fops = flow_get_drv_ops(type);
3422 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3423 }
3424 
3425 /**
3426  * Get RSS action from the action list.
3427  *
3428  * @param[in] dev
3429  *   Pointer to Ethernet device.
3430  * @param[in] actions
3431  *   Pointer to the list of actions.
3432  * @param[in] flow
3433  *   Parent flow structure pointer.
3434  *
3435  * @return
3436  *   Pointer to the RSS action if exist, else return NULL.
3437  */
3438 static const struct rte_flow_action_rss*
3439 flow_get_rss_action(struct rte_eth_dev *dev,
3440 		    const struct rte_flow_action actions[])
3441 {
3442 	struct mlx5_priv *priv = dev->data->dev_private;
3443 	const struct rte_flow_action_rss *rss = NULL;
3444 
3445 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3446 		switch (actions->type) {
3447 		case RTE_FLOW_ACTION_TYPE_RSS:
3448 			rss = actions->conf;
3449 			break;
3450 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3451 		{
3452 			const struct rte_flow_action_sample *sample =
3453 								actions->conf;
3454 			const struct rte_flow_action *act = sample->actions;
3455 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3456 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3457 					rss = act->conf;
3458 			break;
3459 		}
3460 		case RTE_FLOW_ACTION_TYPE_METER:
3461 		{
3462 			uint32_t mtr_idx;
3463 			struct mlx5_flow_meter_info *fm;
3464 			struct mlx5_flow_meter_policy *policy;
3465 			const struct rte_flow_action_meter *mtr = actions->conf;
3466 
3467 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3468 			if (fm) {
3469 				policy = mlx5_flow_meter_policy_find(dev,
3470 						fm->policy_id, NULL);
3471 				if (policy && policy->is_rss)
3472 					rss =
3473 				policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3474 			}
3475 			break;
3476 		}
3477 		default:
3478 			break;
3479 		}
3480 	}
3481 	return rss;
3482 }
3483 
3484 /**
3485  * Get ASO age action by index.
3486  *
3487  * @param[in] dev
3488  *   Pointer to the Ethernet device structure.
3489  * @param[in] age_idx
3490  *   Index to the ASO age action.
3491  *
3492  * @return
3493  *   The specified ASO age action.
3494  */
3495 struct mlx5_aso_age_action*
3496 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3497 {
3498 	uint16_t pool_idx = age_idx & UINT16_MAX;
3499 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3500 	struct mlx5_priv *priv = dev->data->dev_private;
3501 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3502 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3503 
3504 	return &pool->actions[offset - 1];
3505 }
3506 
3507 /* maps indirect action to translated direct in some actions array */
3508 struct mlx5_translated_action_handle {
3509 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3510 	int index; /**< Index in related array of rte_flow_action. */
3511 };
3512 
3513 /**
3514  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3515  * direct action if translation possible.
3516  * This functionality used to run same execution path for both direct and
3517  * indirect actions on flow create. All necessary preparations for indirect
3518  * action handling should be performed on *handle* actions list returned
3519  * from this call.
3520  *
3521  * @param[in] dev
3522  *   Pointer to Ethernet device.
3523  * @param[in] actions
3524  *   List of actions to translate.
3525  * @param[out] handle
3526  *   List to store translated indirect action object handles.
3527  * @param[in, out] indir_n
3528  *   Size of *handle* array. On return should be updated with number of
3529  *   indirect actions retrieved from the *actions* list.
3530  * @param[out] translated_actions
3531  *   List of actions where all indirect actions were translated to direct
3532  *   if possible. NULL if no translation took place.
3533  * @param[out] error
3534  *   Pointer to the error structure.
3535  *
3536  * @return
3537  *   0 on success, a negative errno value otherwise and rte_errno is set.
3538  */
3539 static int
3540 flow_action_handles_translate(struct rte_eth_dev *dev,
3541 			      const struct rte_flow_action actions[],
3542 			      struct mlx5_translated_action_handle *handle,
3543 			      int *indir_n,
3544 			      struct rte_flow_action **translated_actions,
3545 			      struct rte_flow_error *error)
3546 {
3547 	struct mlx5_priv *priv = dev->data->dev_private;
3548 	struct rte_flow_action *translated = NULL;
3549 	size_t actions_size;
3550 	int n;
3551 	int copied_n = 0;
3552 	struct mlx5_translated_action_handle *handle_end = NULL;
3553 
3554 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3555 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3556 			continue;
3557 		if (copied_n == *indir_n) {
3558 			return rte_flow_error_set
3559 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3560 				 NULL, "too many shared actions");
3561 		}
3562 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3563 			   sizeof(actions[n].conf));
3564 		handle[copied_n].index = n;
3565 		copied_n++;
3566 	}
3567 	n++;
3568 	*indir_n = copied_n;
3569 	if (!copied_n)
3570 		return 0;
3571 	actions_size = sizeof(struct rte_flow_action) * n;
3572 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3573 	if (!translated) {
3574 		rte_errno = ENOMEM;
3575 		return -ENOMEM;
3576 	}
3577 	memcpy(translated, actions, actions_size);
3578 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3579 		struct mlx5_shared_action_rss *shared_rss;
3580 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3581 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3582 		uint32_t idx = act_idx &
3583 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3584 
3585 		switch (type) {
3586 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3587 			shared_rss = mlx5_ipool_get
3588 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3589 			translated[handle->index].type =
3590 				RTE_FLOW_ACTION_TYPE_RSS;
3591 			translated[handle->index].conf =
3592 				&shared_rss->origin;
3593 			break;
3594 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3595 			translated[handle->index].type =
3596 						(enum rte_flow_action_type)
3597 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3598 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3599 			break;
3600 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3601 			if (priv->sh->flow_hit_aso_en) {
3602 				translated[handle->index].type =
3603 					(enum rte_flow_action_type)
3604 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3605 				translated[handle->index].conf =
3606 							 (void *)(uintptr_t)idx;
3607 				break;
3608 			}
3609 			/* Fall-through */
3610 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3611 			if (priv->sh->ct_aso_en) {
3612 				translated[handle->index].type =
3613 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3614 				translated[handle->index].conf =
3615 							 (void *)(uintptr_t)idx;
3616 				break;
3617 			}
3618 			/* Fall-through */
3619 		default:
3620 			mlx5_free(translated);
3621 			return rte_flow_error_set
3622 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3623 				 NULL, "invalid indirect action type");
3624 		}
3625 	}
3626 	*translated_actions = translated;
3627 	return 0;
3628 }
3629 
3630 /**
3631  * Get Shared RSS action from the action list.
3632  *
3633  * @param[in] dev
3634  *   Pointer to Ethernet device.
3635  * @param[in] shared
3636  *   Pointer to the list of actions.
3637  * @param[in] shared_n
3638  *   Actions list length.
3639  *
3640  * @return
3641  *   The MLX5 RSS action ID if exists, otherwise return 0.
3642  */
3643 static uint32_t
3644 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3645 			   struct mlx5_translated_action_handle *handle,
3646 			   int shared_n)
3647 {
3648 	struct mlx5_translated_action_handle *handle_end;
3649 	struct mlx5_priv *priv = dev->data->dev_private;
3650 	struct mlx5_shared_action_rss *shared_rss;
3651 
3652 
3653 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3654 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3655 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3656 		uint32_t idx = act_idx &
3657 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3658 		switch (type) {
3659 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3660 			shared_rss = mlx5_ipool_get
3661 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3662 									   idx);
3663 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3664 					   __ATOMIC_RELAXED);
3665 			return idx;
3666 		default:
3667 			break;
3668 		}
3669 	}
3670 	return 0;
3671 }
3672 
3673 static unsigned int
3674 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3675 {
3676 	const struct rte_flow_item *item;
3677 	unsigned int has_vlan = 0;
3678 
3679 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3680 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3681 			has_vlan = 1;
3682 			break;
3683 		}
3684 	}
3685 	if (has_vlan)
3686 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3687 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3688 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3689 			       MLX5_EXPANSION_ROOT_OUTER;
3690 }
3691 
3692 /**
3693  *  Get layer flags from the prefix flow.
3694  *
3695  *  Some flows may be split to several subflows, the prefix subflow gets the
3696  *  match items and the suffix sub flow gets the actions.
3697  *  Some actions need the user defined match item flags to get the detail for
3698  *  the action.
3699  *  This function helps the suffix flow to get the item layer flags from prefix
3700  *  subflow.
3701  *
3702  * @param[in] dev_flow
3703  *   Pointer the created preifx subflow.
3704  *
3705  * @return
3706  *   The layers get from prefix subflow.
3707  */
3708 static inline uint64_t
3709 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3710 {
3711 	uint64_t layers = 0;
3712 
3713 	/*
3714 	 * Layers bits could be localization, but usually the compiler will
3715 	 * help to do the optimization work for source code.
3716 	 * If no decap actions, use the layers directly.
3717 	 */
3718 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3719 		return dev_flow->handle->layers;
3720 	/* Convert L3 layers with decap action. */
3721 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3722 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3723 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3724 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3725 	/* Convert L4 layers with decap action.  */
3726 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3727 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3728 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3729 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3730 	return layers;
3731 }
3732 
3733 /**
3734  * Get metadata split action information.
3735  *
3736  * @param[in] actions
3737  *   Pointer to the list of actions.
3738  * @param[out] qrss
3739  *   Pointer to the return pointer.
3740  * @param[out] qrss_type
3741  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3742  *   if no QUEUE/RSS is found.
3743  * @param[out] encap_idx
3744  *   Pointer to the index of the encap action if exists, otherwise the last
3745  *   action index.
3746  *
3747  * @return
3748  *   Total number of actions.
3749  */
3750 static int
3751 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3752 				       const struct rte_flow_action **qrss,
3753 				       int *encap_idx)
3754 {
3755 	const struct rte_flow_action_raw_encap *raw_encap;
3756 	int actions_n = 0;
3757 	int raw_decap_idx = -1;
3758 
3759 	*encap_idx = -1;
3760 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3761 		switch (actions->type) {
3762 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3763 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3764 			*encap_idx = actions_n;
3765 			break;
3766 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3767 			raw_decap_idx = actions_n;
3768 			break;
3769 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3770 			raw_encap = actions->conf;
3771 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3772 				*encap_idx = raw_decap_idx != -1 ?
3773 						      raw_decap_idx : actions_n;
3774 			break;
3775 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3776 		case RTE_FLOW_ACTION_TYPE_RSS:
3777 			*qrss = actions;
3778 			break;
3779 		default:
3780 			break;
3781 		}
3782 		actions_n++;
3783 	}
3784 	if (*encap_idx == -1)
3785 		*encap_idx = actions_n;
3786 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3787 	return actions_n + 1;
3788 }
3789 
3790 /**
3791  * Check if the action will change packet.
3792  *
3793  * @param dev
3794  *   Pointer to Ethernet device.
3795  * @param[in] type
3796  *   action type.
3797  *
3798  * @return
3799  *   true if action will change packet, false otherwise.
3800  */
3801 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3802 					  enum rte_flow_action_type type)
3803 {
3804 	struct mlx5_priv *priv = dev->data->dev_private;
3805 
3806 	switch (type) {
3807 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3808 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3809 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3810 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3811 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3812 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3813 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3814 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3815 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3816 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
3817 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3818 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3819 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3820 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3821 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3822 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3823 	case RTE_FLOW_ACTION_TYPE_SET_META:
3824 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
3825 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3826 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3827 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3828 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3829 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3830 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3831 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3832 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3833 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3834 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3835 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3836 		return true;
3837 	case RTE_FLOW_ACTION_TYPE_FLAG:
3838 	case RTE_FLOW_ACTION_TYPE_MARK:
3839 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3840 			return true;
3841 		else
3842 			return false;
3843 	default:
3844 		return false;
3845 	}
3846 }
3847 
3848 /**
3849  * Check meter action from the action list.
3850  *
3851  * @param dev
3852  *   Pointer to Ethernet device.
3853  * @param[in] actions
3854  *   Pointer to the list of actions.
3855  * @param[out] has_mtr
3856  *   Pointer to the meter exist flag.
3857  * @param[out] has_modify
3858  *   Pointer to the flag showing there's packet change action.
3859  * @param[out] meter_id
3860  *   Pointer to the meter id.
3861  *
3862  * @return
3863  *   Total number of actions.
3864  */
3865 static int
3866 flow_check_meter_action(struct rte_eth_dev *dev,
3867 			const struct rte_flow_action actions[],
3868 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3869 {
3870 	const struct rte_flow_action_meter *mtr = NULL;
3871 	int actions_n = 0;
3872 
3873 	MLX5_ASSERT(has_mtr);
3874 	*has_mtr = false;
3875 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3876 		switch (actions->type) {
3877 		case RTE_FLOW_ACTION_TYPE_METER:
3878 			mtr = actions->conf;
3879 			*meter_id = mtr->mtr_id;
3880 			*has_mtr = true;
3881 			break;
3882 		default:
3883 			break;
3884 		}
3885 		if (!*has_mtr)
3886 			*has_modify |= flow_check_modify_action_type(dev,
3887 								actions->type);
3888 		actions_n++;
3889 	}
3890 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3891 	return actions_n + 1;
3892 }
3893 
3894 /**
3895  * Check if the flow should be split due to hairpin.
3896  * The reason for the split is that in current HW we can't
3897  * support encap and push-vlan on Rx, so if a flow contains
3898  * these actions we move it to Tx.
3899  *
3900  * @param dev
3901  *   Pointer to Ethernet device.
3902  * @param[in] attr
3903  *   Flow rule attributes.
3904  * @param[in] actions
3905  *   Associated actions (list terminated by the END action).
3906  *
3907  * @return
3908  *   > 0 the number of actions and the flow should be split,
3909  *   0 when no split required.
3910  */
3911 static int
3912 flow_check_hairpin_split(struct rte_eth_dev *dev,
3913 			 const struct rte_flow_attr *attr,
3914 			 const struct rte_flow_action actions[])
3915 {
3916 	int queue_action = 0;
3917 	int action_n = 0;
3918 	int split = 0;
3919 	const struct rte_flow_action_queue *queue;
3920 	const struct rte_flow_action_rss *rss;
3921 	const struct rte_flow_action_raw_encap *raw_encap;
3922 	const struct rte_eth_hairpin_conf *conf;
3923 
3924 	if (!attr->ingress)
3925 		return 0;
3926 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3927 		switch (actions->type) {
3928 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3929 			queue = actions->conf;
3930 			if (queue == NULL)
3931 				return 0;
3932 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3933 			if (conf == NULL || conf->tx_explicit != 0)
3934 				return 0;
3935 			queue_action = 1;
3936 			action_n++;
3937 			break;
3938 		case RTE_FLOW_ACTION_TYPE_RSS:
3939 			rss = actions->conf;
3940 			if (rss == NULL || rss->queue_num == 0)
3941 				return 0;
3942 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3943 			if (conf == NULL || conf->tx_explicit != 0)
3944 				return 0;
3945 			queue_action = 1;
3946 			action_n++;
3947 			break;
3948 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3949 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3950 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3951 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3952 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3953 			split++;
3954 			action_n++;
3955 			break;
3956 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3957 			raw_encap = actions->conf;
3958 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3959 				split++;
3960 			action_n++;
3961 			break;
3962 		default:
3963 			action_n++;
3964 			break;
3965 		}
3966 	}
3967 	if (split && queue_action)
3968 		return action_n;
3969 	return 0;
3970 }
3971 
3972 /* Declare flow create/destroy prototype in advance. */
3973 static uint32_t
3974 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3975 		 const struct rte_flow_attr *attr,
3976 		 const struct rte_flow_item items[],
3977 		 const struct rte_flow_action actions[],
3978 		 bool external, struct rte_flow_error *error);
3979 
3980 static void
3981 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3982 		  uint32_t flow_idx);
3983 
3984 int
3985 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3986 		      struct mlx5_hlist_entry *entry,
3987 		      uint64_t key, void *cb_ctx __rte_unused)
3988 {
3989 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3990 		container_of(entry, typeof(*mcp_res), hlist_ent);
3991 
3992 	return mcp_res->mark_id != key;
3993 }
3994 
3995 struct mlx5_hlist_entry *
3996 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3997 		       void *cb_ctx)
3998 {
3999 	struct rte_eth_dev *dev = list->ctx;
4000 	struct mlx5_priv *priv = dev->data->dev_private;
4001 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4002 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4003 	struct rte_flow_error *error = ctx->error;
4004 	uint32_t idx = 0;
4005 	int ret;
4006 	uint32_t mark_id = key;
4007 	struct rte_flow_attr attr = {
4008 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4009 		.ingress = 1,
4010 	};
4011 	struct mlx5_rte_flow_item_tag tag_spec = {
4012 		.data = mark_id,
4013 	};
4014 	struct rte_flow_item items[] = {
4015 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4016 	};
4017 	struct rte_flow_action_mark ftag = {
4018 		.id = mark_id,
4019 	};
4020 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4021 		.dst = REG_B,
4022 		.src = REG_NON,
4023 	};
4024 	struct rte_flow_action_jump jump = {
4025 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4026 	};
4027 	struct rte_flow_action actions[] = {
4028 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4029 	};
4030 
4031 	/* Fill the register fileds in the flow. */
4032 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4033 	if (ret < 0)
4034 		return NULL;
4035 	tag_spec.id = ret;
4036 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4037 	if (ret < 0)
4038 		return NULL;
4039 	cp_mreg.src = ret;
4040 	/* Provide the full width of FLAG specific value. */
4041 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4042 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4043 	/* Build a new flow. */
4044 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4045 		items[0] = (struct rte_flow_item){
4046 			.type = (enum rte_flow_item_type)
4047 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4048 			.spec = &tag_spec,
4049 		};
4050 		items[1] = (struct rte_flow_item){
4051 			.type = RTE_FLOW_ITEM_TYPE_END,
4052 		};
4053 		actions[0] = (struct rte_flow_action){
4054 			.type = (enum rte_flow_action_type)
4055 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4056 			.conf = &ftag,
4057 		};
4058 		actions[1] = (struct rte_flow_action){
4059 			.type = (enum rte_flow_action_type)
4060 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4061 			.conf = &cp_mreg,
4062 		};
4063 		actions[2] = (struct rte_flow_action){
4064 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4065 			.conf = &jump,
4066 		};
4067 		actions[3] = (struct rte_flow_action){
4068 			.type = RTE_FLOW_ACTION_TYPE_END,
4069 		};
4070 	} else {
4071 		/* Default rule, wildcard match. */
4072 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4073 		items[0] = (struct rte_flow_item){
4074 			.type = RTE_FLOW_ITEM_TYPE_END,
4075 		};
4076 		actions[0] = (struct rte_flow_action){
4077 			.type = (enum rte_flow_action_type)
4078 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4079 			.conf = &cp_mreg,
4080 		};
4081 		actions[1] = (struct rte_flow_action){
4082 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4083 			.conf = &jump,
4084 		};
4085 		actions[2] = (struct rte_flow_action){
4086 			.type = RTE_FLOW_ACTION_TYPE_END,
4087 		};
4088 	}
4089 	/* Build a new entry. */
4090 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4091 	if (!mcp_res) {
4092 		rte_errno = ENOMEM;
4093 		return NULL;
4094 	}
4095 	mcp_res->idx = idx;
4096 	mcp_res->mark_id = mark_id;
4097 	/*
4098 	 * The copy Flows are not included in any list. There
4099 	 * ones are referenced from other Flows and can not
4100 	 * be applied, removed, deleted in ardbitrary order
4101 	 * by list traversing.
4102 	 */
4103 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4104 					 actions, false, error);
4105 	if (!mcp_res->rix_flow) {
4106 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4107 		return NULL;
4108 	}
4109 	return &mcp_res->hlist_ent;
4110 }
4111 
4112 /**
4113  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4114  *
4115  * As mark_id is unique, if there's already a registered flow for the mark_id,
4116  * return by increasing the reference counter of the resource. Otherwise, create
4117  * the resource (mcp_res) and flow.
4118  *
4119  * Flow looks like,
4120  *   - If ingress port is ANY and reg_c[1] is mark_id,
4121  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4122  *
4123  * For default flow (zero mark_id), flow is like,
4124  *   - If ingress port is ANY,
4125  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4126  *
4127  * @param dev
4128  *   Pointer to Ethernet device.
4129  * @param mark_id
4130  *   ID of MARK action, zero means default flow for META.
4131  * @param[out] error
4132  *   Perform verbose error reporting if not NULL.
4133  *
4134  * @return
4135  *   Associated resource on success, NULL otherwise and rte_errno is set.
4136  */
4137 static struct mlx5_flow_mreg_copy_resource *
4138 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4139 			  struct rte_flow_error *error)
4140 {
4141 	struct mlx5_priv *priv = dev->data->dev_private;
4142 	struct mlx5_hlist_entry *entry;
4143 	struct mlx5_flow_cb_ctx ctx = {
4144 		.dev = dev,
4145 		.error = error,
4146 	};
4147 
4148 	/* Check if already registered. */
4149 	MLX5_ASSERT(priv->mreg_cp_tbl);
4150 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4151 	if (!entry)
4152 		return NULL;
4153 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4154 			    hlist_ent);
4155 }
4156 
4157 void
4158 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4159 {
4160 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4161 		container_of(entry, typeof(*mcp_res), hlist_ent);
4162 	struct rte_eth_dev *dev = list->ctx;
4163 	struct mlx5_priv *priv = dev->data->dev_private;
4164 
4165 	MLX5_ASSERT(mcp_res->rix_flow);
4166 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4167 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4168 }
4169 
4170 /**
4171  * Release flow in RX_CP_TBL.
4172  *
4173  * @param dev
4174  *   Pointer to Ethernet device.
4175  * @flow
4176  *   Parent flow for wich copying is provided.
4177  */
4178 static void
4179 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4180 			  struct rte_flow *flow)
4181 {
4182 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4183 	struct mlx5_priv *priv = dev->data->dev_private;
4184 
4185 	if (!flow->rix_mreg_copy)
4186 		return;
4187 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4188 				 flow->rix_mreg_copy);
4189 	if (!mcp_res || !priv->mreg_cp_tbl)
4190 		return;
4191 	MLX5_ASSERT(mcp_res->rix_flow);
4192 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4193 	flow->rix_mreg_copy = 0;
4194 }
4195 
4196 /**
4197  * Remove the default copy action from RX_CP_TBL.
4198  *
4199  * This functions is called in the mlx5_dev_start(). No thread safe
4200  * is guaranteed.
4201  *
4202  * @param dev
4203  *   Pointer to Ethernet device.
4204  */
4205 static void
4206 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4207 {
4208 	struct mlx5_hlist_entry *entry;
4209 	struct mlx5_priv *priv = dev->data->dev_private;
4210 
4211 	/* Check if default flow is registered. */
4212 	if (!priv->mreg_cp_tbl)
4213 		return;
4214 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4215 				  MLX5_DEFAULT_COPY_ID, NULL);
4216 	if (!entry)
4217 		return;
4218 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4219 }
4220 
4221 /**
4222  * Add the default copy action in in RX_CP_TBL.
4223  *
4224  * This functions is called in the mlx5_dev_start(). No thread safe
4225  * is guaranteed.
4226  *
4227  * @param dev
4228  *   Pointer to Ethernet device.
4229  * @param[out] error
4230  *   Perform verbose error reporting if not NULL.
4231  *
4232  * @return
4233  *   0 for success, negative value otherwise and rte_errno is set.
4234  */
4235 static int
4236 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4237 				  struct rte_flow_error *error)
4238 {
4239 	struct mlx5_priv *priv = dev->data->dev_private;
4240 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4241 
4242 	/* Check whether extensive metadata feature is engaged. */
4243 	if (!priv->config.dv_flow_en ||
4244 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4245 	    !mlx5_flow_ext_mreg_supported(dev) ||
4246 	    !priv->sh->dv_regc0_mask)
4247 		return 0;
4248 	/*
4249 	 * Add default mreg copy flow may be called multiple time, but
4250 	 * only be called once in stop. Avoid register it twice.
4251 	 */
4252 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4253 		return 0;
4254 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4255 	if (!mcp_res)
4256 		return -rte_errno;
4257 	return 0;
4258 }
4259 
4260 /**
4261  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4262  *
4263  * All the flow having Q/RSS action should be split by
4264  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4265  * performs the following,
4266  *   - CQE->flow_tag := reg_c[1] (MARK)
4267  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4268  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4269  * but there should be a flow per each MARK ID set by MARK action.
4270  *
4271  * For the aforementioned reason, if there's a MARK action in flow's action
4272  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4273  * the MARK ID to CQE's flow_tag like,
4274  *   - If reg_c[1] is mark_id,
4275  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4276  *
4277  * For SET_META action which stores value in reg_c[0], as the destination is
4278  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4279  * MARK ID means the default flow. The default flow looks like,
4280  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4281  *
4282  * @param dev
4283  *   Pointer to Ethernet device.
4284  * @param flow
4285  *   Pointer to flow structure.
4286  * @param[in] actions
4287  *   Pointer to the list of actions.
4288  * @param[out] error
4289  *   Perform verbose error reporting if not NULL.
4290  *
4291  * @return
4292  *   0 on success, negative value otherwise and rte_errno is set.
4293  */
4294 static int
4295 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4296 			    struct rte_flow *flow,
4297 			    const struct rte_flow_action *actions,
4298 			    struct rte_flow_error *error)
4299 {
4300 	struct mlx5_priv *priv = dev->data->dev_private;
4301 	struct mlx5_dev_config *config = &priv->config;
4302 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4303 	const struct rte_flow_action_mark *mark;
4304 
4305 	/* Check whether extensive metadata feature is engaged. */
4306 	if (!config->dv_flow_en ||
4307 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4308 	    !mlx5_flow_ext_mreg_supported(dev) ||
4309 	    !priv->sh->dv_regc0_mask)
4310 		return 0;
4311 	/* Find MARK action. */
4312 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4313 		switch (actions->type) {
4314 		case RTE_FLOW_ACTION_TYPE_FLAG:
4315 			mcp_res = flow_mreg_add_copy_action
4316 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4317 			if (!mcp_res)
4318 				return -rte_errno;
4319 			flow->rix_mreg_copy = mcp_res->idx;
4320 			return 0;
4321 		case RTE_FLOW_ACTION_TYPE_MARK:
4322 			mark = (const struct rte_flow_action_mark *)
4323 				actions->conf;
4324 			mcp_res =
4325 				flow_mreg_add_copy_action(dev, mark->id, error);
4326 			if (!mcp_res)
4327 				return -rte_errno;
4328 			flow->rix_mreg_copy = mcp_res->idx;
4329 			return 0;
4330 		default:
4331 			break;
4332 		}
4333 	}
4334 	return 0;
4335 }
4336 
4337 #define MLX5_MAX_SPLIT_ACTIONS 24
4338 #define MLX5_MAX_SPLIT_ITEMS 24
4339 
4340 /**
4341  * Split the hairpin flow.
4342  * Since HW can't support encap and push-vlan on Rx, we move these
4343  * actions to Tx.
4344  * If the count action is after the encap then we also
4345  * move the count action. in this case the count will also measure
4346  * the outer bytes.
4347  *
4348  * @param dev
4349  *   Pointer to Ethernet device.
4350  * @param[in] actions
4351  *   Associated actions (list terminated by the END action).
4352  * @param[out] actions_rx
4353  *   Rx flow actions.
4354  * @param[out] actions_tx
4355  *   Tx flow actions..
4356  * @param[out] pattern_tx
4357  *   The pattern items for the Tx flow.
4358  * @param[out] flow_id
4359  *   The flow ID connected to this flow.
4360  *
4361  * @return
4362  *   0 on success.
4363  */
4364 static int
4365 flow_hairpin_split(struct rte_eth_dev *dev,
4366 		   const struct rte_flow_action actions[],
4367 		   struct rte_flow_action actions_rx[],
4368 		   struct rte_flow_action actions_tx[],
4369 		   struct rte_flow_item pattern_tx[],
4370 		   uint32_t flow_id)
4371 {
4372 	const struct rte_flow_action_raw_encap *raw_encap;
4373 	const struct rte_flow_action_raw_decap *raw_decap;
4374 	struct mlx5_rte_flow_action_set_tag *set_tag;
4375 	struct rte_flow_action *tag_action;
4376 	struct mlx5_rte_flow_item_tag *tag_item;
4377 	struct rte_flow_item *item;
4378 	char *addr;
4379 	int encap = 0;
4380 
4381 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4382 		switch (actions->type) {
4383 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4384 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4385 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4386 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4387 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4388 			rte_memcpy(actions_tx, actions,
4389 			       sizeof(struct rte_flow_action));
4390 			actions_tx++;
4391 			break;
4392 		case RTE_FLOW_ACTION_TYPE_COUNT:
4393 			if (encap) {
4394 				rte_memcpy(actions_tx, actions,
4395 					   sizeof(struct rte_flow_action));
4396 				actions_tx++;
4397 			} else {
4398 				rte_memcpy(actions_rx, actions,
4399 					   sizeof(struct rte_flow_action));
4400 				actions_rx++;
4401 			}
4402 			break;
4403 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4404 			raw_encap = actions->conf;
4405 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4406 				memcpy(actions_tx, actions,
4407 				       sizeof(struct rte_flow_action));
4408 				actions_tx++;
4409 				encap = 1;
4410 			} else {
4411 				rte_memcpy(actions_rx, actions,
4412 					   sizeof(struct rte_flow_action));
4413 				actions_rx++;
4414 			}
4415 			break;
4416 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4417 			raw_decap = actions->conf;
4418 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4419 				memcpy(actions_tx, actions,
4420 				       sizeof(struct rte_flow_action));
4421 				actions_tx++;
4422 			} else {
4423 				rte_memcpy(actions_rx, actions,
4424 					   sizeof(struct rte_flow_action));
4425 				actions_rx++;
4426 			}
4427 			break;
4428 		default:
4429 			rte_memcpy(actions_rx, actions,
4430 				   sizeof(struct rte_flow_action));
4431 			actions_rx++;
4432 			break;
4433 		}
4434 	}
4435 	/* Add set meta action and end action for the Rx flow. */
4436 	tag_action = actions_rx;
4437 	tag_action->type = (enum rte_flow_action_type)
4438 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4439 	actions_rx++;
4440 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4441 	actions_rx++;
4442 	set_tag = (void *)actions_rx;
4443 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4444 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4445 		.data = flow_id,
4446 	};
4447 	MLX5_ASSERT(set_tag->id > REG_NON);
4448 	tag_action->conf = set_tag;
4449 	/* Create Tx item list. */
4450 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4451 	addr = (void *)&pattern_tx[2];
4452 	item = pattern_tx;
4453 	item->type = (enum rte_flow_item_type)
4454 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4455 	tag_item = (void *)addr;
4456 	tag_item->data = flow_id;
4457 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4458 	MLX5_ASSERT(set_tag->id > REG_NON);
4459 	item->spec = tag_item;
4460 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4461 	tag_item = (void *)addr;
4462 	tag_item->data = UINT32_MAX;
4463 	tag_item->id = UINT16_MAX;
4464 	item->mask = tag_item;
4465 	item->last = NULL;
4466 	item++;
4467 	item->type = RTE_FLOW_ITEM_TYPE_END;
4468 	return 0;
4469 }
4470 
4471 /**
4472  * The last stage of splitting chain, just creates the subflow
4473  * without any modification.
4474  *
4475  * @param[in] dev
4476  *   Pointer to Ethernet device.
4477  * @param[in] flow
4478  *   Parent flow structure pointer.
4479  * @param[in, out] sub_flow
4480  *   Pointer to return the created subflow, may be NULL.
4481  * @param[in] attr
4482  *   Flow rule attributes.
4483  * @param[in] items
4484  *   Pattern specification (list terminated by the END pattern item).
4485  * @param[in] actions
4486  *   Associated actions (list terminated by the END action).
4487  * @param[in] flow_split_info
4488  *   Pointer to flow split info structure.
4489  * @param[out] error
4490  *   Perform verbose error reporting if not NULL.
4491  * @return
4492  *   0 on success, negative value otherwise
4493  */
4494 static int
4495 flow_create_split_inner(struct rte_eth_dev *dev,
4496 			struct rte_flow *flow,
4497 			struct mlx5_flow **sub_flow,
4498 			const struct rte_flow_attr *attr,
4499 			const struct rte_flow_item items[],
4500 			const struct rte_flow_action actions[],
4501 			struct mlx5_flow_split_info *flow_split_info,
4502 			struct rte_flow_error *error)
4503 {
4504 	struct mlx5_flow *dev_flow;
4505 
4506 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4507 				    flow_split_info->flow_idx, error);
4508 	if (!dev_flow)
4509 		return -rte_errno;
4510 	dev_flow->flow = flow;
4511 	dev_flow->external = flow_split_info->external;
4512 	dev_flow->skip_scale = flow_split_info->skip_scale;
4513 	/* Subflow object was created, we must include one in the list. */
4514 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4515 		      dev_flow->handle, next);
4516 	/*
4517 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4518 	 * flow may need some user defined item layer flags, and pass the
4519 	 * Metadate rxq mark flag to suffix flow as well.
4520 	 */
4521 	if (flow_split_info->prefix_layers)
4522 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4523 	if (flow_split_info->prefix_mark)
4524 		dev_flow->handle->mark = 1;
4525 	if (sub_flow)
4526 		*sub_flow = dev_flow;
4527 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4528 	dev_flow->dv.table_id = flow_split_info->table_id;
4529 #endif
4530 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4531 }
4532 
4533 /**
4534  * Get the sub policy of a meter.
4535  *
4536  * @param[in] dev
4537  *   Pointer to Ethernet device.
4538  * @param[in] flow
4539  *   Parent flow structure pointer.
4540  * @param[in] policy_id;
4541  *   Meter Policy id.
4542  * @param[in] attr
4543  *   Flow rule attributes.
4544  * @param[in] items
4545  *   Pattern specification (list terminated by the END pattern item).
4546  * @param[out] error
4547  *   Perform verbose error reporting if not NULL.
4548  *
4549  * @return
4550  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4551  */
4552 static struct mlx5_flow_meter_sub_policy *
4553 get_meter_sub_policy(struct rte_eth_dev *dev,
4554 		     struct rte_flow *flow,
4555 		     uint32_t policy_id,
4556 		     const struct rte_flow_attr *attr,
4557 		     const struct rte_flow_item items[],
4558 		     struct rte_flow_error *error)
4559 {
4560 	struct mlx5_flow_meter_policy *policy;
4561 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4562 
4563 	policy = mlx5_flow_meter_policy_find(dev, policy_id, NULL);
4564 	if (!policy) {
4565 		rte_flow_error_set(error, EINVAL,
4566 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4567 				   "Failed to find Meter Policy.");
4568 		goto exit;
4569 	}
4570 	if (policy->is_rss ||
4571 		(policy->is_queue &&
4572 	!policy->sub_policys[MLX5_MTR_DOMAIN_INGRESS][0]->rix_hrxq[0])) {
4573 		struct mlx5_flow_workspace *wks =
4574 				mlx5_flow_get_thread_workspace();
4575 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4576 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4577 		uint32_t i;
4578 
4579 		MLX5_ASSERT(wks);
4580 		/**
4581 		 * This is a tmp dev_flow,
4582 		 * no need to register any matcher for it in translate.
4583 		 */
4584 		wks->skip_matcher_reg = 1;
4585 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4586 			struct mlx5_flow dev_flow = {0};
4587 			struct mlx5_flow_handle dev_handle = { {0} };
4588 
4589 			rss_desc_v[i] = wks->rss_desc;
4590 			if (policy->is_rss) {
4591 				const void *rss_act =
4592 					policy->act_cnt[i].rss->conf;
4593 				struct rte_flow_action rss_actions[2] = {
4594 					[0] = {
4595 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4596 					.conf = rss_act
4597 					},
4598 					[1] = {
4599 					.type = RTE_FLOW_ACTION_TYPE_END,
4600 					.conf = NULL
4601 					}
4602 				};
4603 
4604 				dev_flow.handle = &dev_handle;
4605 				dev_flow.ingress = attr->ingress;
4606 				dev_flow.flow = flow;
4607 				dev_flow.external = 0;
4608 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4609 				dev_flow.dv.transfer = attr->transfer;
4610 #endif
4611 				/**
4612 				 * Translate RSS action to get rss hash fields.
4613 				 */
4614 				if (flow_drv_translate(dev, &dev_flow, attr,
4615 						items, rss_actions, error))
4616 					goto exit;
4617 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4618 				rss_desc_v[i].hash_fields =
4619 						dev_flow.hash_fields;
4620 				rss_desc_v[i].queue_num =
4621 						rss_desc_v[i].hash_fields ?
4622 						rss_desc_v[i].queue_num : 1;
4623 			} else {
4624 				/* This is queue action. */
4625 				rss_desc_v[i].key_len = 0;
4626 				rss_desc_v[i].hash_fields = 0;
4627 				rss_desc_v[i].queue =
4628 					&policy->act_cnt[i].queue;
4629 				rss_desc_v[i].queue_num = 1;
4630 			}
4631 			rss_desc[i] = &rss_desc_v[i];
4632 		}
4633 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4634 						flow, policy, rss_desc);
4635 	} else {
4636 		enum mlx5_meter_domain mtr_domain =
4637 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4638 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4639 					MLX5_MTR_DOMAIN_INGRESS;
4640 		sub_policy = policy->sub_policys[mtr_domain][0];
4641 	}
4642 	if (!sub_policy) {
4643 		rte_flow_error_set(error, EINVAL,
4644 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4645 			"Failed to get meter sub-policy.");
4646 		goto exit;
4647 	}
4648 exit:
4649 	return sub_policy;
4650 }
4651 
4652 /**
4653  * Split the meter flow.
4654  *
4655  * As meter flow will split to three sub flow, other than meter
4656  * action, the other actions make sense to only meter accepts
4657  * the packet. If it need to be dropped, no other additional
4658  * actions should be take.
4659  *
4660  * One kind of special action which decapsulates the L3 tunnel
4661  * header will be in the prefix sub flow, as not to take the
4662  * L3 tunnel header into account.
4663  *
4664  * @param[in] dev
4665  *   Pointer to Ethernet device.
4666  * @param[in] flow
4667  *   Parent flow structure pointer.
4668  * @param[in] fm
4669  *   Pointer to flow meter structure.
4670  * @param[in] attr
4671  *   Flow rule attributes.
4672  * @param[in] items
4673  *   Pattern specification (list terminated by the END pattern item).
4674  * @param[out] sfx_items
4675  *   Suffix flow match items (list terminated by the END pattern item).
4676  * @param[in] actions
4677  *   Associated actions (list terminated by the END action).
4678  * @param[out] actions_sfx
4679  *   Suffix flow actions.
4680  * @param[out] actions_pre
4681  *   Prefix flow actions.
4682  * @param[out] mtr_flow_id
4683  *   Pointer to meter flow id.
4684  * @param[out] error
4685  *   Perform verbose error reporting if not NULL.
4686  *
4687  * @return
4688  *   0 on success, a negative errno value otherwise and rte_errno is set.
4689  */
4690 static int
4691 flow_meter_split_prep(struct rte_eth_dev *dev,
4692 		      struct rte_flow *flow,
4693 		      struct mlx5_flow_meter_info *fm,
4694 		      const struct rte_flow_attr *attr,
4695 		      const struct rte_flow_item items[],
4696 		      struct rte_flow_item sfx_items[],
4697 		      const struct rte_flow_action actions[],
4698 		      struct rte_flow_action actions_sfx[],
4699 		      struct rte_flow_action actions_pre[],
4700 		      uint32_t *mtr_flow_id,
4701 		      struct rte_flow_error *error)
4702 {
4703 	struct mlx5_priv *priv = dev->data->dev_private;
4704 	struct rte_flow_action *tag_action = NULL;
4705 	struct rte_flow_item *tag_item;
4706 	struct mlx5_rte_flow_action_set_tag *set_tag;
4707 	const struct rte_flow_action_raw_encap *raw_encap;
4708 	const struct rte_flow_action_raw_decap *raw_decap;
4709 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4710 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4711 	uint32_t tag_id = 0;
4712 	bool copy_vlan = false;
4713 	struct rte_flow_action *hw_mtr_action;
4714 	struct rte_flow_action *action_pre_head = NULL;
4715 	bool mtr_first = priv->sh->meter_aso_en &&
4716 			(attr->egress ||
4717 			(attr->transfer && priv->representor_id != UINT16_MAX));
4718 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4719 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4720 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4721 	uint32_t flow_id = 0;
4722 	uint32_t flow_id_reversed = 0;
4723 	uint8_t flow_id_bits = 0;
4724 	int shift;
4725 
4726 	/* For ASO meter, meter must be before tag in TX direction. */
4727 	if (mtr_first) {
4728 		action_pre_head = actions_pre++;
4729 		/* Leave space for tag action. */
4730 		tag_action = actions_pre++;
4731 	}
4732 	/* Prepare the actions for prefix and suffix flow. */
4733 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4734 		struct rte_flow_action *action_cur = NULL;
4735 
4736 		switch (actions->type) {
4737 		case RTE_FLOW_ACTION_TYPE_METER:
4738 			if (mtr_first) {
4739 				action_cur = action_pre_head;
4740 			} else {
4741 				/* Leave space for tag action. */
4742 				tag_action = actions_pre++;
4743 				action_cur = actions_pre++;
4744 			}
4745 			break;
4746 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4747 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4748 			action_cur = actions_pre++;
4749 			break;
4750 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4751 			raw_encap = actions->conf;
4752 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4753 				action_cur = actions_pre++;
4754 			break;
4755 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4756 			raw_decap = actions->conf;
4757 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4758 				action_cur = actions_pre++;
4759 			break;
4760 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4761 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4762 			copy_vlan = true;
4763 			break;
4764 		default:
4765 			break;
4766 		}
4767 		if (!action_cur)
4768 			action_cur = (fm->def_policy) ?
4769 					actions_sfx++ : actions_pre++;
4770 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4771 	}
4772 	/* Add end action to the actions. */
4773 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4774 	if (priv->sh->meter_aso_en) {
4775 		/**
4776 		 * For ASO meter, need to add an extra jump action explicitly,
4777 		 * to jump from meter to policer table.
4778 		 */
4779 		struct mlx5_flow_meter_sub_policy *sub_policy;
4780 		struct mlx5_flow_tbl_data_entry *tbl_data;
4781 
4782 		if (!fm->def_policy) {
4783 			sub_policy = get_meter_sub_policy(dev, flow,
4784 							  fm->policy_id, attr,
4785 							  items, error);
4786 			if (!sub_policy)
4787 				return -rte_errno;
4788 		} else {
4789 			enum mlx5_meter_domain mtr_domain =
4790 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4791 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4792 					MLX5_MTR_DOMAIN_INGRESS;
4793 
4794 			sub_policy =
4795 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4796 		}
4797 		tbl_data = container_of(sub_policy->tbl_rsc,
4798 					struct mlx5_flow_tbl_data_entry, tbl);
4799 		hw_mtr_action = actions_pre++;
4800 		hw_mtr_action->type = (enum rte_flow_action_type)
4801 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4802 		hw_mtr_action->conf = tbl_data->jump.action;
4803 	}
4804 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4805 	actions_pre++;
4806 	if (!tag_action)
4807 		return rte_flow_error_set(error, ENOMEM,
4808 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4809 					"No tag action space.");
4810 	if (!mtr_flow_id) {
4811 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4812 		goto exit;
4813 	}
4814 	/* Only default-policy Meter creates mtr flow id. */
4815 	if (fm->def_policy) {
4816 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4817 		if (!tag_id)
4818 			return rte_flow_error_set(error, ENOMEM,
4819 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4820 					"Failed to allocate meter flow id.");
4821 		flow_id = tag_id - 1;
4822 		flow_id_bits = (!flow_id) ? 1 :
4823 				(MLX5_REG_BITS - __builtin_clz(flow_id));
4824 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4825 		    mtr_reg_bits) {
4826 			mlx5_ipool_free(fm->flow_ipool, tag_id);
4827 			return rte_flow_error_set(error, EINVAL,
4828 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4829 					"Meter flow id exceeds max limit.");
4830 		}
4831 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4832 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4833 	}
4834 	/* Prepare the suffix subflow items. */
4835 	tag_item = sfx_items++;
4836 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4837 		int item_type = items->type;
4838 
4839 		switch (item_type) {
4840 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4841 			memcpy(sfx_items, items, sizeof(*sfx_items));
4842 			sfx_items++;
4843 			break;
4844 		case RTE_FLOW_ITEM_TYPE_VLAN:
4845 			if (copy_vlan) {
4846 				memcpy(sfx_items, items, sizeof(*sfx_items));
4847 				/*
4848 				 * Convert to internal match item, it is used
4849 				 * for vlan push and set vid.
4850 				 */
4851 				sfx_items->type = (enum rte_flow_item_type)
4852 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4853 				sfx_items++;
4854 			}
4855 			break;
4856 		default:
4857 			break;
4858 		}
4859 	}
4860 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4861 	sfx_items++;
4862 	/* Build tag actions and items for meter_id/meter flow_id. */
4863 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4864 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4865 	tag_item_mask = tag_item_spec + 1;
4866 	/* Both flow_id and meter_id share the same register. */
4867 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4868 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4869 							    0, error),
4870 		.offset = mtr_id_offset,
4871 		.length = mtr_reg_bits,
4872 		.data = flow->meter,
4873 	};
4874 	/*
4875 	 * The color Reg bits used by flow_id are growing from
4876 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4877 	 */
4878 	for (shift = 0; shift < flow_id_bits; shift++)
4879 		flow_id_reversed = (flow_id_reversed << 1) |
4880 				((flow_id >> shift) & 0x1);
4881 	set_tag->data |=
4882 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4883 	tag_item_spec->id = set_tag->id;
4884 	tag_item_spec->data = set_tag->data << mtr_id_offset;
4885 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4886 	tag_action->type = (enum rte_flow_action_type)
4887 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4888 	tag_action->conf = set_tag;
4889 	tag_item->type = (enum rte_flow_item_type)
4890 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4891 	tag_item->spec = tag_item_spec;
4892 	tag_item->last = NULL;
4893 	tag_item->mask = tag_item_mask;
4894 exit:
4895 	if (mtr_flow_id)
4896 		*mtr_flow_id = tag_id;
4897 	return 0;
4898 }
4899 
4900 /**
4901  * Split action list having QUEUE/RSS for metadata register copy.
4902  *
4903  * Once Q/RSS action is detected in user's action list, the flow action
4904  * should be split in order to copy metadata registers, which will happen in
4905  * RX_CP_TBL like,
4906  *   - CQE->flow_tag := reg_c[1] (MARK)
4907  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4908  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4909  * This is because the last action of each flow must be a terminal action
4910  * (QUEUE, RSS or DROP).
4911  *
4912  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4913  * stored and kept in the mlx5_flow structure per each sub_flow.
4914  *
4915  * The Q/RSS action is replaced with,
4916  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4917  * And the following JUMP action is added at the end,
4918  *   - JUMP, to RX_CP_TBL.
4919  *
4920  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4921  * flow_create_split_metadata() routine. The flow will look like,
4922  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4923  *
4924  * @param dev
4925  *   Pointer to Ethernet device.
4926  * @param[out] split_actions
4927  *   Pointer to store split actions to jump to CP_TBL.
4928  * @param[in] actions
4929  *   Pointer to the list of original flow actions.
4930  * @param[in] qrss
4931  *   Pointer to the Q/RSS action.
4932  * @param[in] actions_n
4933  *   Number of original actions.
4934  * @param[out] error
4935  *   Perform verbose error reporting if not NULL.
4936  *
4937  * @return
4938  *   non-zero unique flow_id on success, otherwise 0 and
4939  *   error/rte_error are set.
4940  */
4941 static uint32_t
4942 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4943 			  struct rte_flow_action *split_actions,
4944 			  const struct rte_flow_action *actions,
4945 			  const struct rte_flow_action *qrss,
4946 			  int actions_n, struct rte_flow_error *error)
4947 {
4948 	struct mlx5_priv *priv = dev->data->dev_private;
4949 	struct mlx5_rte_flow_action_set_tag *set_tag;
4950 	struct rte_flow_action_jump *jump;
4951 	const int qrss_idx = qrss - actions;
4952 	uint32_t flow_id = 0;
4953 	int ret = 0;
4954 
4955 	/*
4956 	 * Given actions will be split
4957 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4958 	 * - Add jump to mreg CP_TBL.
4959 	 * As a result, there will be one more action.
4960 	 */
4961 	++actions_n;
4962 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4963 	set_tag = (void *)(split_actions + actions_n);
4964 	/*
4965 	 * If tag action is not set to void(it means we are not the meter
4966 	 * suffix flow), add the tag action. Since meter suffix flow already
4967 	 * has the tag added.
4968 	 */
4969 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4970 		/*
4971 		 * Allocate the new subflow ID. This one is unique within
4972 		 * device and not shared with representors. Otherwise,
4973 		 * we would have to resolve multi-thread access synch
4974 		 * issue. Each flow on the shared device is appended
4975 		 * with source vport identifier, so the resulting
4976 		 * flows will be unique in the shared (by master and
4977 		 * representors) domain even if they have coinciding
4978 		 * IDs.
4979 		 */
4980 		mlx5_ipool_malloc(priv->sh->ipool
4981 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4982 		if (!flow_id)
4983 			return rte_flow_error_set(error, ENOMEM,
4984 						  RTE_FLOW_ERROR_TYPE_ACTION,
4985 						  NULL, "can't allocate id "
4986 						  "for split Q/RSS subflow");
4987 		/* Internal SET_TAG action to set flow ID. */
4988 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4989 			.data = flow_id,
4990 		};
4991 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4992 		if (ret < 0)
4993 			return ret;
4994 		set_tag->id = ret;
4995 		/* Construct new actions array. */
4996 		/* Replace QUEUE/RSS action. */
4997 		split_actions[qrss_idx] = (struct rte_flow_action){
4998 			.type = (enum rte_flow_action_type)
4999 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5000 			.conf = set_tag,
5001 		};
5002 	}
5003 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5004 	jump = (void *)(set_tag + 1);
5005 	*jump = (struct rte_flow_action_jump){
5006 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5007 	};
5008 	split_actions[actions_n - 2] = (struct rte_flow_action){
5009 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5010 		.conf = jump,
5011 	};
5012 	split_actions[actions_n - 1] = (struct rte_flow_action){
5013 		.type = RTE_FLOW_ACTION_TYPE_END,
5014 	};
5015 	return flow_id;
5016 }
5017 
5018 /**
5019  * Extend the given action list for Tx metadata copy.
5020  *
5021  * Copy the given action list to the ext_actions and add flow metadata register
5022  * copy action in order to copy reg_a set by WQE to reg_c[0].
5023  *
5024  * @param[out] ext_actions
5025  *   Pointer to the extended action list.
5026  * @param[in] actions
5027  *   Pointer to the list of actions.
5028  * @param[in] actions_n
5029  *   Number of actions in the list.
5030  * @param[out] error
5031  *   Perform verbose error reporting if not NULL.
5032  * @param[in] encap_idx
5033  *   The encap action inndex.
5034  *
5035  * @return
5036  *   0 on success, negative value otherwise
5037  */
5038 static int
5039 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5040 		       struct rte_flow_action *ext_actions,
5041 		       const struct rte_flow_action *actions,
5042 		       int actions_n, struct rte_flow_error *error,
5043 		       int encap_idx)
5044 {
5045 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5046 		(struct mlx5_flow_action_copy_mreg *)
5047 			(ext_actions + actions_n + 1);
5048 	int ret;
5049 
5050 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5051 	if (ret < 0)
5052 		return ret;
5053 	cp_mreg->dst = ret;
5054 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5055 	if (ret < 0)
5056 		return ret;
5057 	cp_mreg->src = ret;
5058 	if (encap_idx != 0)
5059 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5060 	if (encap_idx == actions_n - 1) {
5061 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5062 			.type = (enum rte_flow_action_type)
5063 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5064 			.conf = cp_mreg,
5065 		};
5066 		ext_actions[actions_n] = (struct rte_flow_action){
5067 			.type = RTE_FLOW_ACTION_TYPE_END,
5068 		};
5069 	} else {
5070 		ext_actions[encap_idx] = (struct rte_flow_action){
5071 			.type = (enum rte_flow_action_type)
5072 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5073 			.conf = cp_mreg,
5074 		};
5075 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5076 				sizeof(*ext_actions) * (actions_n - encap_idx));
5077 	}
5078 	return 0;
5079 }
5080 
5081 /**
5082  * Check the match action from the action list.
5083  *
5084  * @param[in] actions
5085  *   Pointer to the list of actions.
5086  * @param[in] attr
5087  *   Flow rule attributes.
5088  * @param[in] action
5089  *   The action to be check if exist.
5090  * @param[out] match_action_pos
5091  *   Pointer to the position of the matched action if exists, otherwise is -1.
5092  * @param[out] qrss_action_pos
5093  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5094  * @param[out] modify_after_mirror
5095  *   Pointer to the flag of modify action after FDB mirroring.
5096  *
5097  * @return
5098  *   > 0 the total number of actions.
5099  *   0 if not found match action in action list.
5100  */
5101 static int
5102 flow_check_match_action(const struct rte_flow_action actions[],
5103 			const struct rte_flow_attr *attr,
5104 			enum rte_flow_action_type action,
5105 			int *match_action_pos, int *qrss_action_pos,
5106 			int *modify_after_mirror)
5107 {
5108 	const struct rte_flow_action_sample *sample;
5109 	int actions_n = 0;
5110 	uint32_t ratio = 0;
5111 	int sub_type = 0;
5112 	int flag = 0;
5113 	int fdb_mirror = 0;
5114 
5115 	*match_action_pos = -1;
5116 	*qrss_action_pos = -1;
5117 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5118 		if (actions->type == action) {
5119 			flag = 1;
5120 			*match_action_pos = actions_n;
5121 		}
5122 		switch (actions->type) {
5123 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5124 		case RTE_FLOW_ACTION_TYPE_RSS:
5125 			*qrss_action_pos = actions_n;
5126 			break;
5127 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5128 			sample = actions->conf;
5129 			ratio = sample->ratio;
5130 			sub_type = ((const struct rte_flow_action *)
5131 					(sample->actions))->type;
5132 			if (ratio == 1 && attr->transfer)
5133 				fdb_mirror = 1;
5134 			break;
5135 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5136 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5137 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5138 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5139 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5140 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5141 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5142 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5143 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5144 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5145 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5146 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5147 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5148 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5149 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5150 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5151 		case RTE_FLOW_ACTION_TYPE_FLAG:
5152 		case RTE_FLOW_ACTION_TYPE_MARK:
5153 		case RTE_FLOW_ACTION_TYPE_SET_META:
5154 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5155 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5156 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5157 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5158 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5159 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5160 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5161 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5162 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5163 		case RTE_FLOW_ACTION_TYPE_METER:
5164 			if (fdb_mirror)
5165 				*modify_after_mirror = 1;
5166 			break;
5167 		default:
5168 			break;
5169 		}
5170 		actions_n++;
5171 	}
5172 	if (flag && fdb_mirror && !*modify_after_mirror) {
5173 		/* FDB mirroring uses the destination array to implement
5174 		 * instead of FLOW_SAMPLER object.
5175 		 */
5176 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5177 			flag = 0;
5178 	}
5179 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5180 	return flag ? actions_n + 1 : 0;
5181 }
5182 
5183 #define SAMPLE_SUFFIX_ITEM 2
5184 
5185 /**
5186  * Split the sample flow.
5187  *
5188  * As sample flow will split to two sub flow, sample flow with
5189  * sample action, the other actions will move to new suffix flow.
5190  *
5191  * Also add unique tag id with tag action in the sample flow,
5192  * the same tag id will be as match in the suffix flow.
5193  *
5194  * @param dev
5195  *   Pointer to Ethernet device.
5196  * @param[in] add_tag
5197  *   Add extra tag action flag.
5198  * @param[out] sfx_items
5199  *   Suffix flow match items (list terminated by the END pattern item).
5200  * @param[in] actions
5201  *   Associated actions (list terminated by the END action).
5202  * @param[out] actions_sfx
5203  *   Suffix flow actions.
5204  * @param[out] actions_pre
5205  *   Prefix flow actions.
5206  * @param[in] actions_n
5207  *  The total number of actions.
5208  * @param[in] sample_action_pos
5209  *   The sample action position.
5210  * @param[in] qrss_action_pos
5211  *   The Queue/RSS action position.
5212  * @param[in] jump_table
5213  *   Add extra jump action flag.
5214  * @param[out] error
5215  *   Perform verbose error reporting if not NULL.
5216  *
5217  * @return
5218  *   0 on success, or unique flow_id, a negative errno value
5219  *   otherwise and rte_errno is set.
5220  */
5221 static int
5222 flow_sample_split_prep(struct rte_eth_dev *dev,
5223 		       int add_tag,
5224 		       struct rte_flow_item sfx_items[],
5225 		       const struct rte_flow_action actions[],
5226 		       struct rte_flow_action actions_sfx[],
5227 		       struct rte_flow_action actions_pre[],
5228 		       int actions_n,
5229 		       int sample_action_pos,
5230 		       int qrss_action_pos,
5231 		       int jump_table,
5232 		       struct rte_flow_error *error)
5233 {
5234 	struct mlx5_priv *priv = dev->data->dev_private;
5235 	struct mlx5_rte_flow_action_set_tag *set_tag;
5236 	struct mlx5_rte_flow_item_tag *tag_spec;
5237 	struct mlx5_rte_flow_item_tag *tag_mask;
5238 	struct rte_flow_action_jump *jump_action;
5239 	uint32_t tag_id = 0;
5240 	int index;
5241 	int append_index = 0;
5242 	int ret;
5243 
5244 	if (sample_action_pos < 0)
5245 		return rte_flow_error_set(error, EINVAL,
5246 					  RTE_FLOW_ERROR_TYPE_ACTION,
5247 					  NULL, "invalid position of sample "
5248 					  "action in list");
5249 	/* Prepare the actions for prefix and suffix flow. */
5250 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5251 		index = qrss_action_pos;
5252 		/* Put the preceding the Queue/RSS action into prefix flow. */
5253 		if (index != 0)
5254 			memcpy(actions_pre, actions,
5255 			       sizeof(struct rte_flow_action) * index);
5256 		/* Put others preceding the sample action into prefix flow. */
5257 		if (sample_action_pos > index + 1)
5258 			memcpy(actions_pre + index, actions + index + 1,
5259 			       sizeof(struct rte_flow_action) *
5260 			       (sample_action_pos - index - 1));
5261 		index = sample_action_pos - 1;
5262 		/* Put Queue/RSS action into Suffix flow. */
5263 		memcpy(actions_sfx, actions + qrss_action_pos,
5264 		       sizeof(struct rte_flow_action));
5265 		actions_sfx++;
5266 	} else {
5267 		index = sample_action_pos;
5268 		if (index != 0)
5269 			memcpy(actions_pre, actions,
5270 			       sizeof(struct rte_flow_action) * index);
5271 	}
5272 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5273 	 * For CX6DX and above, metadata registers Cx preserve their value,
5274 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5275 	 */
5276 	if (add_tag) {
5277 		/* Prepare the prefix tag action. */
5278 		append_index++;
5279 		set_tag = (void *)(actions_pre + actions_n + append_index);
5280 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5281 		if (ret < 0)
5282 			return ret;
5283 		mlx5_ipool_malloc(priv->sh->ipool
5284 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5285 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5286 			.id = ret,
5287 			.data = tag_id,
5288 		};
5289 		/* Prepare the suffix subflow items. */
5290 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5291 		tag_spec->data = tag_id;
5292 		tag_spec->id = set_tag->id;
5293 		tag_mask = tag_spec + 1;
5294 		tag_mask->data = UINT32_MAX;
5295 		sfx_items[0] = (struct rte_flow_item){
5296 			.type = (enum rte_flow_item_type)
5297 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5298 			.spec = tag_spec,
5299 			.last = NULL,
5300 			.mask = tag_mask,
5301 		};
5302 		sfx_items[1] = (struct rte_flow_item){
5303 			.type = (enum rte_flow_item_type)
5304 				RTE_FLOW_ITEM_TYPE_END,
5305 		};
5306 		/* Prepare the tag action in prefix subflow. */
5307 		actions_pre[index++] =
5308 			(struct rte_flow_action){
5309 			.type = (enum rte_flow_action_type)
5310 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5311 			.conf = set_tag,
5312 		};
5313 	}
5314 	memcpy(actions_pre + index, actions + sample_action_pos,
5315 	       sizeof(struct rte_flow_action));
5316 	index += 1;
5317 	/* For the modify action after the sample action in E-Switch mirroring,
5318 	 * Add the extra jump action in prefix subflow and jump into the next
5319 	 * table, then do the modify action in the new table.
5320 	 */
5321 	if (jump_table) {
5322 		/* Prepare the prefix jump action. */
5323 		append_index++;
5324 		jump_action = (void *)(actions_pre + actions_n + append_index);
5325 		jump_action->group = jump_table;
5326 		actions_pre[index++] =
5327 			(struct rte_flow_action){
5328 			.type = (enum rte_flow_action_type)
5329 				RTE_FLOW_ACTION_TYPE_JUMP,
5330 			.conf = jump_action,
5331 		};
5332 	}
5333 	actions_pre[index] = (struct rte_flow_action){
5334 		.type = (enum rte_flow_action_type)
5335 			RTE_FLOW_ACTION_TYPE_END,
5336 	};
5337 	/* Put the actions after sample into Suffix flow. */
5338 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5339 	       sizeof(struct rte_flow_action) *
5340 	       (actions_n - sample_action_pos - 1));
5341 	return tag_id;
5342 }
5343 
5344 /**
5345  * The splitting for metadata feature.
5346  *
5347  * - Q/RSS action on NIC Rx should be split in order to pass by
5348  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5349  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5350  *
5351  * - All the actions on NIC Tx should have a mreg copy action to
5352  *   copy reg_a from WQE to reg_c[0].
5353  *
5354  * @param dev
5355  *   Pointer to Ethernet device.
5356  * @param[in] flow
5357  *   Parent flow structure pointer.
5358  * @param[in] attr
5359  *   Flow rule attributes.
5360  * @param[in] items
5361  *   Pattern specification (list terminated by the END pattern item).
5362  * @param[in] actions
5363  *   Associated actions (list terminated by the END action).
5364  * @param[in] flow_split_info
5365  *   Pointer to flow split info structure.
5366  * @param[out] error
5367  *   Perform verbose error reporting if not NULL.
5368  * @return
5369  *   0 on success, negative value otherwise
5370  */
5371 static int
5372 flow_create_split_metadata(struct rte_eth_dev *dev,
5373 			   struct rte_flow *flow,
5374 			   const struct rte_flow_attr *attr,
5375 			   const struct rte_flow_item items[],
5376 			   const struct rte_flow_action actions[],
5377 			   struct mlx5_flow_split_info *flow_split_info,
5378 			   struct rte_flow_error *error)
5379 {
5380 	struct mlx5_priv *priv = dev->data->dev_private;
5381 	struct mlx5_dev_config *config = &priv->config;
5382 	const struct rte_flow_action *qrss = NULL;
5383 	struct rte_flow_action *ext_actions = NULL;
5384 	struct mlx5_flow *dev_flow = NULL;
5385 	uint32_t qrss_id = 0;
5386 	int mtr_sfx = 0;
5387 	size_t act_size;
5388 	int actions_n;
5389 	int encap_idx;
5390 	int ret;
5391 
5392 	/* Check whether extensive metadata feature is engaged. */
5393 	if (!config->dv_flow_en ||
5394 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5395 	    !mlx5_flow_ext_mreg_supported(dev))
5396 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5397 					       actions, flow_split_info, error);
5398 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5399 							   &encap_idx);
5400 	if (qrss) {
5401 		/* Exclude hairpin flows from splitting. */
5402 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5403 			const struct rte_flow_action_queue *queue;
5404 
5405 			queue = qrss->conf;
5406 			if (mlx5_rxq_get_type(dev, queue->index) ==
5407 			    MLX5_RXQ_TYPE_HAIRPIN)
5408 				qrss = NULL;
5409 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5410 			const struct rte_flow_action_rss *rss;
5411 
5412 			rss = qrss->conf;
5413 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5414 			    MLX5_RXQ_TYPE_HAIRPIN)
5415 				qrss = NULL;
5416 		}
5417 	}
5418 	if (qrss) {
5419 		/* Check if it is in meter suffix table. */
5420 		mtr_sfx = attr->group == (attr->transfer ?
5421 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5422 			  MLX5_FLOW_TABLE_LEVEL_METER);
5423 		/*
5424 		 * Q/RSS action on NIC Rx should be split in order to pass by
5425 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5426 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5427 		 */
5428 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5429 			   sizeof(struct rte_flow_action_set_tag) +
5430 			   sizeof(struct rte_flow_action_jump);
5431 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5432 					  SOCKET_ID_ANY);
5433 		if (!ext_actions)
5434 			return rte_flow_error_set(error, ENOMEM,
5435 						  RTE_FLOW_ERROR_TYPE_ACTION,
5436 						  NULL, "no memory to split "
5437 						  "metadata flow");
5438 		/*
5439 		 * If we are the suffix flow of meter, tag already exist.
5440 		 * Set the tag action to void.
5441 		 */
5442 		if (mtr_sfx)
5443 			ext_actions[qrss - actions].type =
5444 						RTE_FLOW_ACTION_TYPE_VOID;
5445 		else
5446 			ext_actions[qrss - actions].type =
5447 						(enum rte_flow_action_type)
5448 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5449 		/*
5450 		 * Create the new actions list with removed Q/RSS action
5451 		 * and appended set tag and jump to register copy table
5452 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5453 		 * in advance, because it is needed for set tag action.
5454 		 */
5455 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5456 						    qrss, actions_n, error);
5457 		if (!mtr_sfx && !qrss_id) {
5458 			ret = -rte_errno;
5459 			goto exit;
5460 		}
5461 	} else if (attr->egress && !attr->transfer) {
5462 		/*
5463 		 * All the actions on NIC Tx should have a metadata register
5464 		 * copy action to copy reg_a from WQE to reg_c[meta]
5465 		 */
5466 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5467 			   sizeof(struct mlx5_flow_action_copy_mreg);
5468 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5469 					  SOCKET_ID_ANY);
5470 		if (!ext_actions)
5471 			return rte_flow_error_set(error, ENOMEM,
5472 						  RTE_FLOW_ERROR_TYPE_ACTION,
5473 						  NULL, "no memory to split "
5474 						  "metadata flow");
5475 		/* Create the action list appended with copy register. */
5476 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5477 					     actions_n, error, encap_idx);
5478 		if (ret < 0)
5479 			goto exit;
5480 	}
5481 	/* Add the unmodified original or prefix subflow. */
5482 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5483 				      items, ext_actions ? ext_actions :
5484 				      actions, flow_split_info, error);
5485 	if (ret < 0)
5486 		goto exit;
5487 	MLX5_ASSERT(dev_flow);
5488 	if (qrss) {
5489 		const struct rte_flow_attr q_attr = {
5490 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5491 			.ingress = 1,
5492 		};
5493 		/* Internal PMD action to set register. */
5494 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5495 			.data = qrss_id,
5496 			.id = REG_NON,
5497 		};
5498 		struct rte_flow_item q_items[] = {
5499 			{
5500 				.type = (enum rte_flow_item_type)
5501 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5502 				.spec = &q_tag_spec,
5503 				.last = NULL,
5504 				.mask = NULL,
5505 			},
5506 			{
5507 				.type = RTE_FLOW_ITEM_TYPE_END,
5508 			},
5509 		};
5510 		struct rte_flow_action q_actions[] = {
5511 			{
5512 				.type = qrss->type,
5513 				.conf = qrss->conf,
5514 			},
5515 			{
5516 				.type = RTE_FLOW_ACTION_TYPE_END,
5517 			},
5518 		};
5519 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5520 
5521 		/*
5522 		 * Configure the tag item only if there is no meter subflow.
5523 		 * Since tag is already marked in the meter suffix subflow
5524 		 * we can just use the meter suffix items as is.
5525 		 */
5526 		if (qrss_id) {
5527 			/* Not meter subflow. */
5528 			MLX5_ASSERT(!mtr_sfx);
5529 			/*
5530 			 * Put unique id in prefix flow due to it is destroyed
5531 			 * after suffix flow and id will be freed after there
5532 			 * is no actual flows with this id and identifier
5533 			 * reallocation becomes possible (for example, for
5534 			 * other flows in other threads).
5535 			 */
5536 			dev_flow->handle->split_flow_id = qrss_id;
5537 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5538 						   error);
5539 			if (ret < 0)
5540 				goto exit;
5541 			q_tag_spec.id = ret;
5542 		}
5543 		dev_flow = NULL;
5544 		/* Add suffix subflow to execute Q/RSS. */
5545 		flow_split_info->prefix_layers = layers;
5546 		flow_split_info->prefix_mark = 0;
5547 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5548 					      &q_attr, mtr_sfx ? items :
5549 					      q_items, q_actions,
5550 					      flow_split_info, error);
5551 		if (ret < 0)
5552 			goto exit;
5553 		/* qrss ID should be freed if failed. */
5554 		qrss_id = 0;
5555 		MLX5_ASSERT(dev_flow);
5556 	}
5557 
5558 exit:
5559 	/*
5560 	 * We do not destroy the partially created sub_flows in case of error.
5561 	 * These ones are included into parent flow list and will be destroyed
5562 	 * by flow_drv_destroy.
5563 	 */
5564 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5565 			qrss_id);
5566 	mlx5_free(ext_actions);
5567 	return ret;
5568 }
5569 
5570 /**
5571  * Create meter internal drop flow with the original pattern.
5572  *
5573  * @param dev
5574  *   Pointer to Ethernet device.
5575  * @param[in] flow
5576  *   Parent flow structure pointer.
5577  * @param[in] attr
5578  *   Flow rule attributes.
5579  * @param[in] items
5580  *   Pattern specification (list terminated by the END pattern item).
5581  * @param[in] flow_split_info
5582  *   Pointer to flow split info structure.
5583  * @param[in] fm
5584  *   Pointer to flow meter structure.
5585  * @param[out] error
5586  *   Perform verbose error reporting if not NULL.
5587  * @return
5588  *   0 on success, negative value otherwise
5589  */
5590 static uint32_t
5591 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5592 			struct rte_flow *flow,
5593 			const struct rte_flow_attr *attr,
5594 			const struct rte_flow_item items[],
5595 			struct mlx5_flow_split_info *flow_split_info,
5596 			struct mlx5_flow_meter_info *fm,
5597 			struct rte_flow_error *error)
5598 {
5599 	struct mlx5_flow *dev_flow = NULL;
5600 	struct rte_flow_attr drop_attr = *attr;
5601 	struct rte_flow_action drop_actions[3];
5602 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5603 
5604 	MLX5_ASSERT(fm->drop_cnt);
5605 	drop_actions[0].type =
5606 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5607 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5608 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5609 	drop_actions[1].conf = NULL;
5610 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5611 	drop_actions[2].conf = NULL;
5612 	drop_split_info.external = false;
5613 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5614 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5615 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5616 	return flow_create_split_inner(dev, flow, &dev_flow,
5617 				&drop_attr, items, drop_actions,
5618 				&drop_split_info, error);
5619 }
5620 
5621 /**
5622  * The splitting for meter feature.
5623  *
5624  * - The meter flow will be split to two flows as prefix and
5625  *   suffix flow. The packets make sense only it pass the prefix
5626  *   meter action.
5627  *
5628  * - Reg_C_5 is used for the packet to match betweend prefix and
5629  *   suffix flow.
5630  *
5631  * @param dev
5632  *   Pointer to Ethernet device.
5633  * @param[in] flow
5634  *   Parent flow structure pointer.
5635  * @param[in] attr
5636  *   Flow rule attributes.
5637  * @param[in] items
5638  *   Pattern specification (list terminated by the END pattern item).
5639  * @param[in] actions
5640  *   Associated actions (list terminated by the END action).
5641  * @param[in] flow_split_info
5642  *   Pointer to flow split info structure.
5643  * @param[out] error
5644  *   Perform verbose error reporting if not NULL.
5645  * @return
5646  *   0 on success, negative value otherwise
5647  */
5648 static int
5649 flow_create_split_meter(struct rte_eth_dev *dev,
5650 			struct rte_flow *flow,
5651 			const struct rte_flow_attr *attr,
5652 			const struct rte_flow_item items[],
5653 			const struct rte_flow_action actions[],
5654 			struct mlx5_flow_split_info *flow_split_info,
5655 			struct rte_flow_error *error)
5656 {
5657 	struct mlx5_priv *priv = dev->data->dev_private;
5658 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5659 	struct rte_flow_action *sfx_actions = NULL;
5660 	struct rte_flow_action *pre_actions = NULL;
5661 	struct rte_flow_item *sfx_items = NULL;
5662 	struct mlx5_flow *dev_flow = NULL;
5663 	struct rte_flow_attr sfx_attr = *attr;
5664 	struct mlx5_flow_meter_info *fm = NULL;
5665 	uint8_t skip_scale_restore;
5666 	bool has_mtr = false;
5667 	bool has_modify = false;
5668 	bool set_mtr_reg = true;
5669 	uint32_t meter_id = 0;
5670 	uint32_t mtr_idx = 0;
5671 	uint32_t mtr_flow_id = 0;
5672 	size_t act_size;
5673 	size_t item_size;
5674 	int actions_n = 0;
5675 	int ret = 0;
5676 
5677 	if (priv->mtr_en)
5678 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5679 						    &has_modify, &meter_id);
5680 	if (has_mtr) {
5681 		if (flow->meter) {
5682 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5683 			if (!fm)
5684 				return rte_flow_error_set(error, EINVAL,
5685 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5686 						NULL, "Meter not found.");
5687 		} else {
5688 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5689 			if (!fm)
5690 				return rte_flow_error_set(error, EINVAL,
5691 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5692 						NULL, "Meter not found.");
5693 			ret = mlx5_flow_meter_attach(priv, fm,
5694 						     &sfx_attr, error);
5695 			if (ret)
5696 				return -rte_errno;
5697 			flow->meter = mtr_idx;
5698 		}
5699 		MLX5_ASSERT(wks);
5700 		wks->fm = fm;
5701 		/*
5702 		 * If it isn't default-policy Meter, and
5703 		 * 1. There's no action in flow to change
5704 		 *    packet (modify/encap/decap etc.), OR
5705 		 * 2. No drop count needed for this meter.
5706 		 * no need to use regC to save meter id anymore.
5707 		 */
5708 		if (!fm->def_policy && (!has_modify || !fm->drop_cnt))
5709 			set_mtr_reg = false;
5710 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5711 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5712 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5713 		/* Suffix items: tag, vlan, port id, end. */
5714 #define METER_SUFFIX_ITEM 4
5715 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5716 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5717 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5718 					  0, SOCKET_ID_ANY);
5719 		if (!sfx_actions)
5720 			return rte_flow_error_set(error, ENOMEM,
5721 						  RTE_FLOW_ERROR_TYPE_ACTION,
5722 						  NULL, "no memory to split "
5723 						  "meter flow");
5724 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5725 			     act_size);
5726 		/* There's no suffix flow for meter of non-default policy. */
5727 		if (!fm->def_policy)
5728 			pre_actions = sfx_actions + 1;
5729 		else
5730 			pre_actions = sfx_actions + actions_n;
5731 		ret = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5732 					    items, sfx_items, actions,
5733 					    sfx_actions, pre_actions,
5734 					    (set_mtr_reg ? &mtr_flow_id : NULL),
5735 					    error);
5736 		if (ret) {
5737 			ret = -rte_errno;
5738 			goto exit;
5739 		}
5740 		/* Add the prefix subflow. */
5741 		flow_split_info->prefix_mark = 0;
5742 		skip_scale_restore = flow_split_info->skip_scale;
5743 		flow_split_info->skip_scale |=
5744 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5745 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5746 					      attr, items, pre_actions,
5747 					      flow_split_info, error);
5748 		flow_split_info->skip_scale = skip_scale_restore;
5749 		if (ret) {
5750 			if (mtr_flow_id)
5751 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5752 			ret = -rte_errno;
5753 			goto exit;
5754 		}
5755 		if (mtr_flow_id) {
5756 			dev_flow->handle->split_flow_id = mtr_flow_id;
5757 			dev_flow->handle->is_meter_flow_id = 1;
5758 		}
5759 		if (!fm->def_policy) {
5760 			if (!set_mtr_reg && fm->drop_cnt)
5761 				ret =
5762 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5763 							&sfx_attr, items,
5764 							flow_split_info,
5765 							fm, error);
5766 			goto exit;
5767 		}
5768 		/* Setting the sfx group atrr. */
5769 		sfx_attr.group = sfx_attr.transfer ?
5770 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5771 				 MLX5_FLOW_TABLE_LEVEL_METER;
5772 		flow_split_info->prefix_layers =
5773 				flow_get_prefix_layer_flags(dev_flow);
5774 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5775 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5776 	}
5777 	/* Add the prefix subflow. */
5778 	ret = flow_create_split_metadata(dev, flow,
5779 					 &sfx_attr, sfx_items ?
5780 					 sfx_items : items,
5781 					 sfx_actions ? sfx_actions : actions,
5782 					 flow_split_info, error);
5783 exit:
5784 	if (sfx_actions)
5785 		mlx5_free(sfx_actions);
5786 	return ret;
5787 }
5788 
5789 /**
5790  * The splitting for sample feature.
5791  *
5792  * Once Sample action is detected in the action list, the flow actions should
5793  * be split into prefix sub flow and suffix sub flow.
5794  *
5795  * The original items remain in the prefix sub flow, all actions preceding the
5796  * sample action and the sample action itself will be copied to the prefix
5797  * sub flow, the actions following the sample action will be copied to the
5798  * suffix sub flow, Queue action always be located in the suffix sub flow.
5799  *
5800  * In order to make the packet from prefix sub flow matches with suffix sub
5801  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5802  * flow uses tag item with the unique flow id.
5803  *
5804  * @param dev
5805  *   Pointer to Ethernet device.
5806  * @param[in] flow
5807  *   Parent flow structure pointer.
5808  * @param[in] attr
5809  *   Flow rule attributes.
5810  * @param[in] items
5811  *   Pattern specification (list terminated by the END pattern item).
5812  * @param[in] actions
5813  *   Associated actions (list terminated by the END action).
5814  * @param[in] flow_split_info
5815  *   Pointer to flow split info structure.
5816  * @param[out] error
5817  *   Perform verbose error reporting if not NULL.
5818  * @return
5819  *   0 on success, negative value otherwise
5820  */
5821 static int
5822 flow_create_split_sample(struct rte_eth_dev *dev,
5823 			 struct rte_flow *flow,
5824 			 const struct rte_flow_attr *attr,
5825 			 const struct rte_flow_item items[],
5826 			 const struct rte_flow_action actions[],
5827 			 struct mlx5_flow_split_info *flow_split_info,
5828 			 struct rte_flow_error *error)
5829 {
5830 	struct mlx5_priv *priv = dev->data->dev_private;
5831 	struct rte_flow_action *sfx_actions = NULL;
5832 	struct rte_flow_action *pre_actions = NULL;
5833 	struct rte_flow_item *sfx_items = NULL;
5834 	struct mlx5_flow *dev_flow = NULL;
5835 	struct rte_flow_attr sfx_attr = *attr;
5836 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5837 	struct mlx5_flow_dv_sample_resource *sample_res;
5838 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5839 	struct mlx5_flow_tbl_resource *sfx_tbl;
5840 #endif
5841 	size_t act_size;
5842 	size_t item_size;
5843 	uint32_t fdb_tx = 0;
5844 	int32_t tag_id = 0;
5845 	int actions_n = 0;
5846 	int sample_action_pos;
5847 	int qrss_action_pos;
5848 	int add_tag = 0;
5849 	int modify_after_mirror = 0;
5850 	uint16_t jump_table = 0;
5851 	const uint32_t next_ft_step = 1;
5852 	int ret = 0;
5853 
5854 	if (priv->sampler_en)
5855 		actions_n = flow_check_match_action(actions, attr,
5856 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5857 					&sample_action_pos, &qrss_action_pos,
5858 					&modify_after_mirror);
5859 	if (actions_n) {
5860 		/* The prefix actions must includes sample, tag, end. */
5861 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5862 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5863 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5864 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5865 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5866 					  item_size), 0, SOCKET_ID_ANY);
5867 		if (!sfx_actions)
5868 			return rte_flow_error_set(error, ENOMEM,
5869 						  RTE_FLOW_ERROR_TYPE_ACTION,
5870 						  NULL, "no memory to split "
5871 						  "sample flow");
5872 		/* The representor_id is -1 for uplink. */
5873 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5874 		/*
5875 		 * When reg_c_preserve is set, metadata registers Cx preserve
5876 		 * their value even through packet duplication.
5877 		 */
5878 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5879 		if (add_tag)
5880 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5881 					+ act_size);
5882 		if (modify_after_mirror)
5883 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5884 				     next_ft_step;
5885 		pre_actions = sfx_actions + actions_n;
5886 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5887 						actions, sfx_actions,
5888 						pre_actions, actions_n,
5889 						sample_action_pos,
5890 						qrss_action_pos, jump_table,
5891 						error);
5892 		if (tag_id < 0 || (add_tag && !tag_id)) {
5893 			ret = -rte_errno;
5894 			goto exit;
5895 		}
5896 		if (modify_after_mirror)
5897 			flow_split_info->skip_scale =
5898 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5899 		/* Add the prefix subflow. */
5900 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5901 					      items, pre_actions,
5902 					      flow_split_info, error);
5903 		if (ret) {
5904 			ret = -rte_errno;
5905 			goto exit;
5906 		}
5907 		dev_flow->handle->split_flow_id = tag_id;
5908 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5909 		if (!modify_after_mirror) {
5910 			/* Set the sfx group attr. */
5911 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5912 						dev_flow->dv.sample_res;
5913 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5914 						sample_res->normal_path_tbl;
5915 			sfx_tbl_data = container_of(sfx_tbl,
5916 						struct mlx5_flow_tbl_data_entry,
5917 						tbl);
5918 			sfx_attr.group = sfx_attr.transfer ?
5919 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5920 		} else {
5921 			MLX5_ASSERT(attr->transfer);
5922 			sfx_attr.group = jump_table;
5923 		}
5924 		flow_split_info->prefix_layers =
5925 				flow_get_prefix_layer_flags(dev_flow);
5926 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5927 		/* Suffix group level already be scaled with factor, set
5928 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5929 		 * again in translation.
5930 		 */
5931 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5932 #endif
5933 	}
5934 	/* Add the suffix subflow. */
5935 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5936 				      sfx_items ? sfx_items : items,
5937 				      sfx_actions ? sfx_actions : actions,
5938 				      flow_split_info, error);
5939 exit:
5940 	if (sfx_actions)
5941 		mlx5_free(sfx_actions);
5942 	return ret;
5943 }
5944 
5945 /**
5946  * Split the flow to subflow set. The splitters might be linked
5947  * in the chain, like this:
5948  * flow_create_split_outer() calls:
5949  *   flow_create_split_meter() calls:
5950  *     flow_create_split_metadata(meter_subflow_0) calls:
5951  *       flow_create_split_inner(metadata_subflow_0)
5952  *       flow_create_split_inner(metadata_subflow_1)
5953  *       flow_create_split_inner(metadata_subflow_2)
5954  *     flow_create_split_metadata(meter_subflow_1) calls:
5955  *       flow_create_split_inner(metadata_subflow_0)
5956  *       flow_create_split_inner(metadata_subflow_1)
5957  *       flow_create_split_inner(metadata_subflow_2)
5958  *
5959  * This provide flexible way to add new levels of flow splitting.
5960  * The all of successfully created subflows are included to the
5961  * parent flow dev_flow list.
5962  *
5963  * @param dev
5964  *   Pointer to Ethernet device.
5965  * @param[in] flow
5966  *   Parent flow structure pointer.
5967  * @param[in] attr
5968  *   Flow rule attributes.
5969  * @param[in] items
5970  *   Pattern specification (list terminated by the END pattern item).
5971  * @param[in] actions
5972  *   Associated actions (list terminated by the END action).
5973  * @param[in] flow_split_info
5974  *   Pointer to flow split info structure.
5975  * @param[out] error
5976  *   Perform verbose error reporting if not NULL.
5977  * @return
5978  *   0 on success, negative value otherwise
5979  */
5980 static int
5981 flow_create_split_outer(struct rte_eth_dev *dev,
5982 			struct rte_flow *flow,
5983 			const struct rte_flow_attr *attr,
5984 			const struct rte_flow_item items[],
5985 			const struct rte_flow_action actions[],
5986 			struct mlx5_flow_split_info *flow_split_info,
5987 			struct rte_flow_error *error)
5988 {
5989 	int ret;
5990 
5991 	ret = flow_create_split_sample(dev, flow, attr, items,
5992 				       actions, flow_split_info, error);
5993 	MLX5_ASSERT(ret <= 0);
5994 	return ret;
5995 }
5996 
5997 static inline struct mlx5_flow_tunnel *
5998 flow_tunnel_from_rule(const struct mlx5_flow *flow)
5999 {
6000 	struct mlx5_flow_tunnel *tunnel;
6001 
6002 #pragma GCC diagnostic push
6003 #pragma GCC diagnostic ignored "-Wcast-qual"
6004 	tunnel = (typeof(tunnel))flow->tunnel;
6005 #pragma GCC diagnostic pop
6006 
6007 	return tunnel;
6008 }
6009 
6010 /**
6011  * Adjust flow RSS workspace if needed.
6012  *
6013  * @param wks
6014  *   Pointer to thread flow work space.
6015  * @param rss_desc
6016  *   Pointer to RSS descriptor.
6017  * @param[in] nrssq_num
6018  *   New RSS queue number.
6019  *
6020  * @return
6021  *   0 on success, -1 otherwise and rte_errno is set.
6022  */
6023 static int
6024 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6025 			  struct mlx5_flow_rss_desc *rss_desc,
6026 			  uint32_t nrssq_num)
6027 {
6028 	if (likely(nrssq_num <= wks->rssq_num))
6029 		return 0;
6030 	rss_desc->queue = realloc(rss_desc->queue,
6031 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6032 	if (!rss_desc->queue) {
6033 		rte_errno = ENOMEM;
6034 		return -1;
6035 	}
6036 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6037 	return 0;
6038 }
6039 
6040 /**
6041  * Create a flow and add it to @p list.
6042  *
6043  * @param dev
6044  *   Pointer to Ethernet device.
6045  * @param list
6046  *   Pointer to a TAILQ flow list. If this parameter NULL,
6047  *   no list insertion occurred, flow is just created,
6048  *   this is caller's responsibility to track the
6049  *   created flow.
6050  * @param[in] attr
6051  *   Flow rule attributes.
6052  * @param[in] items
6053  *   Pattern specification (list terminated by the END pattern item).
6054  * @param[in] actions
6055  *   Associated actions (list terminated by the END action).
6056  * @param[in] external
6057  *   This flow rule is created by request external to PMD.
6058  * @param[out] error
6059  *   Perform verbose error reporting if not NULL.
6060  *
6061  * @return
6062  *   A flow index on success, 0 otherwise and rte_errno is set.
6063  */
6064 static uint32_t
6065 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
6066 		 const struct rte_flow_attr *attr,
6067 		 const struct rte_flow_item items[],
6068 		 const struct rte_flow_action original_actions[],
6069 		 bool external, struct rte_flow_error *error)
6070 {
6071 	struct mlx5_priv *priv = dev->data->dev_private;
6072 	struct rte_flow *flow = NULL;
6073 	struct mlx5_flow *dev_flow;
6074 	const struct rte_flow_action_rss *rss = NULL;
6075 	struct mlx5_translated_action_handle
6076 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6077 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6078 	union {
6079 		struct mlx5_flow_expand_rss buf;
6080 		uint8_t buffer[2048];
6081 	} expand_buffer;
6082 	union {
6083 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6084 		uint8_t buffer[2048];
6085 	} actions_rx;
6086 	union {
6087 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6088 		uint8_t buffer[2048];
6089 	} actions_hairpin_tx;
6090 	union {
6091 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6092 		uint8_t buffer[2048];
6093 	} items_tx;
6094 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6095 	struct mlx5_flow_rss_desc *rss_desc;
6096 	const struct rte_flow_action *p_actions_rx;
6097 	uint32_t i;
6098 	uint32_t idx = 0;
6099 	int hairpin_flow;
6100 	struct rte_flow_attr attr_tx = { .priority = 0 };
6101 	const struct rte_flow_action *actions;
6102 	struct rte_flow_action *translated_actions = NULL;
6103 	struct mlx5_flow_tunnel *tunnel;
6104 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6105 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6106 	struct mlx5_flow_split_info flow_split_info = {
6107 		.external = !!external,
6108 		.skip_scale = 0,
6109 		.flow_idx = 0,
6110 		.prefix_mark = 0,
6111 		.prefix_layers = 0,
6112 		.table_id = 0
6113 	};
6114 	int ret;
6115 
6116 	MLX5_ASSERT(wks);
6117 	rss_desc = &wks->rss_desc;
6118 	ret = flow_action_handles_translate(dev, original_actions,
6119 					    indir_actions,
6120 					    &indir_actions_n,
6121 					    &translated_actions, error);
6122 	if (ret < 0) {
6123 		MLX5_ASSERT(translated_actions == NULL);
6124 		return 0;
6125 	}
6126 	actions = translated_actions ? translated_actions : original_actions;
6127 	p_actions_rx = actions;
6128 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6129 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6130 				external, hairpin_flow, error);
6131 	if (ret < 0)
6132 		goto error_before_hairpin_split;
6133 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6134 	if (!flow) {
6135 		rte_errno = ENOMEM;
6136 		goto error_before_hairpin_split;
6137 	}
6138 	if (hairpin_flow > 0) {
6139 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6140 			rte_errno = EINVAL;
6141 			goto error_before_hairpin_split;
6142 		}
6143 		flow_hairpin_split(dev, actions, actions_rx.actions,
6144 				   actions_hairpin_tx.actions, items_tx.items,
6145 				   idx);
6146 		p_actions_rx = actions_rx.actions;
6147 	}
6148 	flow_split_info.flow_idx = idx;
6149 	flow->drv_type = flow_get_drv_type(dev, attr);
6150 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6151 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6152 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6153 	/* RSS Action only works on NIC RX domain */
6154 	if (attr->ingress && !attr->transfer)
6155 		rss = flow_get_rss_action(dev, p_actions_rx);
6156 	if (rss) {
6157 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6158 			return 0;
6159 		/*
6160 		 * The following information is required by
6161 		 * mlx5_flow_hashfields_adjust() in advance.
6162 		 */
6163 		rss_desc->level = rss->level;
6164 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6165 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6166 	}
6167 	flow->dev_handles = 0;
6168 	if (rss && rss->types) {
6169 		unsigned int graph_root;
6170 
6171 		graph_root = find_graph_root(items, rss->level);
6172 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6173 					   items, rss->types,
6174 					   mlx5_support_expansion, graph_root);
6175 		MLX5_ASSERT(ret > 0 &&
6176 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6177 	} else {
6178 		buf->entries = 1;
6179 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6180 	}
6181 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6182 						      indir_actions_n);
6183 	for (i = 0; i < buf->entries; ++i) {
6184 		/* Initialize flow split data. */
6185 		flow_split_info.prefix_layers = 0;
6186 		flow_split_info.prefix_mark = 0;
6187 		flow_split_info.skip_scale = 0;
6188 		/*
6189 		 * The splitter may create multiple dev_flows,
6190 		 * depending on configuration. In the simplest
6191 		 * case it just creates unmodified original flow.
6192 		 */
6193 		ret = flow_create_split_outer(dev, flow, attr,
6194 					      buf->entry[i].pattern,
6195 					      p_actions_rx, &flow_split_info,
6196 					      error);
6197 		if (ret < 0)
6198 			goto error;
6199 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6200 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6201 							   p_actions_rx,
6202 							   idx,
6203 							   wks->flows[0].tunnel,
6204 							   &default_miss_ctx,
6205 							   error);
6206 			if (ret < 0) {
6207 				mlx5_free(default_miss_ctx.queue);
6208 				goto error;
6209 			}
6210 		}
6211 	}
6212 	/* Create the tx flow. */
6213 	if (hairpin_flow) {
6214 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6215 		attr_tx.ingress = 0;
6216 		attr_tx.egress = 1;
6217 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6218 					 actions_hairpin_tx.actions,
6219 					 idx, error);
6220 		if (!dev_flow)
6221 			goto error;
6222 		dev_flow->flow = flow;
6223 		dev_flow->external = 0;
6224 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6225 			      dev_flow->handle, next);
6226 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6227 					 items_tx.items,
6228 					 actions_hairpin_tx.actions, error);
6229 		if (ret < 0)
6230 			goto error;
6231 	}
6232 	/*
6233 	 * Update the metadata register copy table. If extensive
6234 	 * metadata feature is enabled and registers are supported
6235 	 * we might create the extra rte_flow for each unique
6236 	 * MARK/FLAG action ID.
6237 	 *
6238 	 * The table is updated for ingress Flows only, because
6239 	 * the egress Flows belong to the different device and
6240 	 * copy table should be updated in peer NIC Rx domain.
6241 	 */
6242 	if (attr->ingress &&
6243 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6244 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6245 		if (ret)
6246 			goto error;
6247 	}
6248 	/*
6249 	 * If the flow is external (from application) OR device is started,
6250 	 * OR mreg discover, then apply immediately.
6251 	 */
6252 	if (external || dev->data->dev_started ||
6253 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6254 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6255 		ret = flow_drv_apply(dev, flow, error);
6256 		if (ret < 0)
6257 			goto error;
6258 	}
6259 	if (list) {
6260 		rte_spinlock_lock(&priv->flow_list_lock);
6261 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6262 			     flow, next);
6263 		rte_spinlock_unlock(&priv->flow_list_lock);
6264 	}
6265 	flow_rxq_flags_set(dev, flow);
6266 	rte_free(translated_actions);
6267 	tunnel = flow_tunnel_from_rule(wks->flows);
6268 	if (tunnel) {
6269 		flow->tunnel = 1;
6270 		flow->tunnel_id = tunnel->tunnel_id;
6271 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6272 		mlx5_free(default_miss_ctx.queue);
6273 	}
6274 	mlx5_flow_pop_thread_workspace();
6275 	return idx;
6276 error:
6277 	MLX5_ASSERT(flow);
6278 	ret = rte_errno; /* Save rte_errno before cleanup. */
6279 	flow_mreg_del_copy_action(dev, flow);
6280 	flow_drv_destroy(dev, flow);
6281 	if (rss_desc->shared_rss)
6282 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6283 			mlx5_ipool_get
6284 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6285 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6286 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6287 	rte_errno = ret; /* Restore rte_errno. */
6288 	ret = rte_errno;
6289 	rte_errno = ret;
6290 	mlx5_flow_pop_thread_workspace();
6291 error_before_hairpin_split:
6292 	rte_free(translated_actions);
6293 	return 0;
6294 }
6295 
6296 /**
6297  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6298  * incoming packets to table 1.
6299  *
6300  * Other flow rules, requested for group n, will be created in
6301  * e-switch table n+1.
6302  * Jump action to e-switch group n will be created to group n+1.
6303  *
6304  * Used when working in switchdev mode, to utilise advantages of table 1
6305  * and above.
6306  *
6307  * @param dev
6308  *   Pointer to Ethernet device.
6309  *
6310  * @return
6311  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6312  */
6313 struct rte_flow *
6314 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6315 {
6316 	const struct rte_flow_attr attr = {
6317 		.group = 0,
6318 		.priority = 0,
6319 		.ingress = 1,
6320 		.egress = 0,
6321 		.transfer = 1,
6322 	};
6323 	const struct rte_flow_item pattern = {
6324 		.type = RTE_FLOW_ITEM_TYPE_END,
6325 	};
6326 	struct rte_flow_action_jump jump = {
6327 		.group = 1,
6328 	};
6329 	const struct rte_flow_action actions[] = {
6330 		{
6331 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6332 			.conf = &jump,
6333 		},
6334 		{
6335 			.type = RTE_FLOW_ACTION_TYPE_END,
6336 		},
6337 	};
6338 	struct mlx5_priv *priv = dev->data->dev_private;
6339 	struct rte_flow_error error;
6340 
6341 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6342 						   &attr, &pattern,
6343 						   actions, false, &error);
6344 }
6345 
6346 /**
6347  * Validate a flow supported by the NIC.
6348  *
6349  * @see rte_flow_validate()
6350  * @see rte_flow_ops
6351  */
6352 int
6353 mlx5_flow_validate(struct rte_eth_dev *dev,
6354 		   const struct rte_flow_attr *attr,
6355 		   const struct rte_flow_item items[],
6356 		   const struct rte_flow_action original_actions[],
6357 		   struct rte_flow_error *error)
6358 {
6359 	int hairpin_flow;
6360 	struct mlx5_translated_action_handle
6361 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6362 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6363 	const struct rte_flow_action *actions;
6364 	struct rte_flow_action *translated_actions = NULL;
6365 	int ret = flow_action_handles_translate(dev, original_actions,
6366 						indir_actions,
6367 						&indir_actions_n,
6368 						&translated_actions, error);
6369 
6370 	if (ret)
6371 		return ret;
6372 	actions = translated_actions ? translated_actions : original_actions;
6373 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6374 	ret = flow_drv_validate(dev, attr, items, actions,
6375 				true, hairpin_flow, error);
6376 	rte_free(translated_actions);
6377 	return ret;
6378 }
6379 
6380 /**
6381  * Create a flow.
6382  *
6383  * @see rte_flow_create()
6384  * @see rte_flow_ops
6385  */
6386 struct rte_flow *
6387 mlx5_flow_create(struct rte_eth_dev *dev,
6388 		 const struct rte_flow_attr *attr,
6389 		 const struct rte_flow_item items[],
6390 		 const struct rte_flow_action actions[],
6391 		 struct rte_flow_error *error)
6392 {
6393 	struct mlx5_priv *priv = dev->data->dev_private;
6394 
6395 	/*
6396 	 * If the device is not started yet, it is not allowed to created a
6397 	 * flow from application. PMD default flows and traffic control flows
6398 	 * are not affected.
6399 	 */
6400 	if (unlikely(!dev->data->dev_started)) {
6401 		DRV_LOG(DEBUG, "port %u is not started when "
6402 			"inserting a flow", dev->data->port_id);
6403 		rte_flow_error_set(error, ENODEV,
6404 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6405 				   NULL,
6406 				   "port not started");
6407 		return NULL;
6408 	}
6409 
6410 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6411 				  attr, items, actions, true, error);
6412 }
6413 
6414 /**
6415  * Destroy a flow in a list.
6416  *
6417  * @param dev
6418  *   Pointer to Ethernet device.
6419  * @param list
6420  *   Pointer to the Indexed flow list. If this parameter NULL,
6421  *   there is no flow removal from the list. Be noted that as
6422  *   flow is add to the indexed list, memory of the indexed
6423  *   list points to maybe changed as flow destroyed.
6424  * @param[in] flow_idx
6425  *   Index of flow to destroy.
6426  */
6427 static void
6428 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6429 		  uint32_t flow_idx)
6430 {
6431 	struct mlx5_priv *priv = dev->data->dev_private;
6432 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6433 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
6434 
6435 	if (!flow)
6436 		return;
6437 	/*
6438 	 * Update RX queue flags only if port is started, otherwise it is
6439 	 * already clean.
6440 	 */
6441 	if (dev->data->dev_started)
6442 		flow_rxq_flags_trim(dev, flow);
6443 	flow_drv_destroy(dev, flow);
6444 	if (list) {
6445 		rte_spinlock_lock(&priv->flow_list_lock);
6446 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6447 			     flow_idx, flow, next);
6448 		rte_spinlock_unlock(&priv->flow_list_lock);
6449 	}
6450 	if (flow->tunnel) {
6451 		struct mlx5_flow_tunnel *tunnel;
6452 
6453 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6454 		RTE_VERIFY(tunnel);
6455 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6456 			mlx5_flow_tunnel_free(dev, tunnel);
6457 	}
6458 	flow_mreg_del_copy_action(dev, flow);
6459 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6460 }
6461 
6462 /**
6463  * Destroy all flows.
6464  *
6465  * @param dev
6466  *   Pointer to Ethernet device.
6467  * @param list
6468  *   Pointer to the Indexed flow list.
6469  * @param active
6470  *   If flushing is called avtively.
6471  */
6472 void
6473 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6474 {
6475 	uint32_t num_flushed = 0;
6476 
6477 	while (*list) {
6478 		flow_list_destroy(dev, list, *list);
6479 		num_flushed++;
6480 	}
6481 	if (active) {
6482 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6483 			dev->data->port_id, num_flushed);
6484 	}
6485 }
6486 
6487 /**
6488  * Stop all default actions for flows.
6489  *
6490  * @param dev
6491  *   Pointer to Ethernet device.
6492  */
6493 void
6494 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6495 {
6496 	flow_mreg_del_default_copy_action(dev);
6497 	flow_rxq_flags_clear(dev);
6498 }
6499 
6500 /**
6501  * Start all default actions for flows.
6502  *
6503  * @param dev
6504  *   Pointer to Ethernet device.
6505  * @return
6506  *   0 on success, a negative errno value otherwise and rte_errno is set.
6507  */
6508 int
6509 mlx5_flow_start_default(struct rte_eth_dev *dev)
6510 {
6511 	struct rte_flow_error error;
6512 
6513 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6514 	return flow_mreg_add_default_copy_action(dev, &error);
6515 }
6516 
6517 /**
6518  * Release key of thread specific flow workspace data.
6519  */
6520 void
6521 flow_release_workspace(void *data)
6522 {
6523 	struct mlx5_flow_workspace *wks = data;
6524 	struct mlx5_flow_workspace *next;
6525 
6526 	while (wks) {
6527 		next = wks->next;
6528 		free(wks->rss_desc.queue);
6529 		free(wks);
6530 		wks = next;
6531 	}
6532 }
6533 
6534 /**
6535  * Get thread specific current flow workspace.
6536  *
6537  * @return pointer to thread specific flow workspace data, NULL on error.
6538  */
6539 struct mlx5_flow_workspace*
6540 mlx5_flow_get_thread_workspace(void)
6541 {
6542 	struct mlx5_flow_workspace *data;
6543 
6544 	data = mlx5_flow_os_get_specific_workspace();
6545 	MLX5_ASSERT(data && data->inuse);
6546 	if (!data || !data->inuse)
6547 		DRV_LOG(ERR, "flow workspace not initialized.");
6548 	return data;
6549 }
6550 
6551 /**
6552  * Allocate and init new flow workspace.
6553  *
6554  * @return pointer to flow workspace data, NULL on error.
6555  */
6556 static struct mlx5_flow_workspace*
6557 flow_alloc_thread_workspace(void)
6558 {
6559 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6560 
6561 	if (!data) {
6562 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6563 			"memory.");
6564 		return NULL;
6565 	}
6566 	data->rss_desc.queue = calloc(1,
6567 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6568 	if (!data->rss_desc.queue)
6569 		goto err;
6570 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6571 	return data;
6572 err:
6573 	if (data->rss_desc.queue)
6574 		free(data->rss_desc.queue);
6575 	free(data);
6576 	return NULL;
6577 }
6578 
6579 /**
6580  * Get new thread specific flow workspace.
6581  *
6582  * If current workspace inuse, create new one and set as current.
6583  *
6584  * @return pointer to thread specific flow workspace data, NULL on error.
6585  */
6586 static struct mlx5_flow_workspace*
6587 mlx5_flow_push_thread_workspace(void)
6588 {
6589 	struct mlx5_flow_workspace *curr;
6590 	struct mlx5_flow_workspace *data;
6591 
6592 	curr = mlx5_flow_os_get_specific_workspace();
6593 	if (!curr) {
6594 		data = flow_alloc_thread_workspace();
6595 		if (!data)
6596 			return NULL;
6597 	} else if (!curr->inuse) {
6598 		data = curr;
6599 	} else if (curr->next) {
6600 		data = curr->next;
6601 	} else {
6602 		data = flow_alloc_thread_workspace();
6603 		if (!data)
6604 			return NULL;
6605 		curr->next = data;
6606 		data->prev = curr;
6607 	}
6608 	data->inuse = 1;
6609 	data->flow_idx = 0;
6610 	/* Set as current workspace */
6611 	if (mlx5_flow_os_set_specific_workspace(data))
6612 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6613 	return data;
6614 }
6615 
6616 /**
6617  * Close current thread specific flow workspace.
6618  *
6619  * If previous workspace available, set it as current.
6620  *
6621  * @return pointer to thread specific flow workspace data, NULL on error.
6622  */
6623 static void
6624 mlx5_flow_pop_thread_workspace(void)
6625 {
6626 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6627 
6628 	if (!data)
6629 		return;
6630 	if (!data->inuse) {
6631 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6632 		return;
6633 	}
6634 	data->inuse = 0;
6635 	if (!data->prev)
6636 		return;
6637 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6638 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6639 }
6640 
6641 /**
6642  * Verify the flow list is empty
6643  *
6644  * @param dev
6645  *  Pointer to Ethernet device.
6646  *
6647  * @return the number of flows not released.
6648  */
6649 int
6650 mlx5_flow_verify(struct rte_eth_dev *dev)
6651 {
6652 	struct mlx5_priv *priv = dev->data->dev_private;
6653 	struct rte_flow *flow;
6654 	uint32_t idx;
6655 	int ret = 0;
6656 
6657 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6658 		      flow, next) {
6659 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6660 			dev->data->port_id, (void *)flow);
6661 		++ret;
6662 	}
6663 	return ret;
6664 }
6665 
6666 /**
6667  * Enable default hairpin egress flow.
6668  *
6669  * @param dev
6670  *   Pointer to Ethernet device.
6671  * @param queue
6672  *   The queue index.
6673  *
6674  * @return
6675  *   0 on success, a negative errno value otherwise and rte_errno is set.
6676  */
6677 int
6678 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6679 			    uint32_t queue)
6680 {
6681 	struct mlx5_priv *priv = dev->data->dev_private;
6682 	const struct rte_flow_attr attr = {
6683 		.egress = 1,
6684 		.priority = 0,
6685 	};
6686 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6687 		.queue = queue,
6688 	};
6689 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6690 		.queue = UINT32_MAX,
6691 	};
6692 	struct rte_flow_item items[] = {
6693 		{
6694 			.type = (enum rte_flow_item_type)
6695 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6696 			.spec = &queue_spec,
6697 			.last = NULL,
6698 			.mask = &queue_mask,
6699 		},
6700 		{
6701 			.type = RTE_FLOW_ITEM_TYPE_END,
6702 		},
6703 	};
6704 	struct rte_flow_action_jump jump = {
6705 		.group = MLX5_HAIRPIN_TX_TABLE,
6706 	};
6707 	struct rte_flow_action actions[2];
6708 	uint32_t flow_idx;
6709 	struct rte_flow_error error;
6710 
6711 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6712 	actions[0].conf = &jump;
6713 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6714 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6715 				&attr, items, actions, false, &error);
6716 	if (!flow_idx) {
6717 		DRV_LOG(DEBUG,
6718 			"Failed to create ctrl flow: rte_errno(%d),"
6719 			" type(%d), message(%s)",
6720 			rte_errno, error.type,
6721 			error.message ? error.message : " (no stated reason)");
6722 		return -rte_errno;
6723 	}
6724 	return 0;
6725 }
6726 
6727 /**
6728  * Enable a control flow configured from the control plane.
6729  *
6730  * @param dev
6731  *   Pointer to Ethernet device.
6732  * @param eth_spec
6733  *   An Ethernet flow spec to apply.
6734  * @param eth_mask
6735  *   An Ethernet flow mask to apply.
6736  * @param vlan_spec
6737  *   A VLAN flow spec to apply.
6738  * @param vlan_mask
6739  *   A VLAN flow mask to apply.
6740  *
6741  * @return
6742  *   0 on success, a negative errno value otherwise and rte_errno is set.
6743  */
6744 int
6745 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6746 		    struct rte_flow_item_eth *eth_spec,
6747 		    struct rte_flow_item_eth *eth_mask,
6748 		    struct rte_flow_item_vlan *vlan_spec,
6749 		    struct rte_flow_item_vlan *vlan_mask)
6750 {
6751 	struct mlx5_priv *priv = dev->data->dev_private;
6752 	const struct rte_flow_attr attr = {
6753 		.ingress = 1,
6754 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6755 	};
6756 	struct rte_flow_item items[] = {
6757 		{
6758 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6759 			.spec = eth_spec,
6760 			.last = NULL,
6761 			.mask = eth_mask,
6762 		},
6763 		{
6764 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6765 					      RTE_FLOW_ITEM_TYPE_END,
6766 			.spec = vlan_spec,
6767 			.last = NULL,
6768 			.mask = vlan_mask,
6769 		},
6770 		{
6771 			.type = RTE_FLOW_ITEM_TYPE_END,
6772 		},
6773 	};
6774 	uint16_t queue[priv->reta_idx_n];
6775 	struct rte_flow_action_rss action_rss = {
6776 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6777 		.level = 0,
6778 		.types = priv->rss_conf.rss_hf,
6779 		.key_len = priv->rss_conf.rss_key_len,
6780 		.queue_num = priv->reta_idx_n,
6781 		.key = priv->rss_conf.rss_key,
6782 		.queue = queue,
6783 	};
6784 	struct rte_flow_action actions[] = {
6785 		{
6786 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6787 			.conf = &action_rss,
6788 		},
6789 		{
6790 			.type = RTE_FLOW_ACTION_TYPE_END,
6791 		},
6792 	};
6793 	uint32_t flow_idx;
6794 	struct rte_flow_error error;
6795 	unsigned int i;
6796 
6797 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6798 		return 0;
6799 	}
6800 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6801 		action_rss.types = 0;
6802 	for (i = 0; i != priv->reta_idx_n; ++i)
6803 		queue[i] = (*priv->reta_idx)[i];
6804 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6805 				&attr, items, actions, false, &error);
6806 	if (!flow_idx)
6807 		return -rte_errno;
6808 	return 0;
6809 }
6810 
6811 /**
6812  * Enable a flow control configured from the control plane.
6813  *
6814  * @param dev
6815  *   Pointer to Ethernet device.
6816  * @param eth_spec
6817  *   An Ethernet flow spec to apply.
6818  * @param eth_mask
6819  *   An Ethernet flow mask to apply.
6820  *
6821  * @return
6822  *   0 on success, a negative errno value otherwise and rte_errno is set.
6823  */
6824 int
6825 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6826 	       struct rte_flow_item_eth *eth_spec,
6827 	       struct rte_flow_item_eth *eth_mask)
6828 {
6829 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6830 }
6831 
6832 /**
6833  * Create default miss flow rule matching lacp traffic
6834  *
6835  * @param dev
6836  *   Pointer to Ethernet device.
6837  * @param eth_spec
6838  *   An Ethernet flow spec to apply.
6839  *
6840  * @return
6841  *   0 on success, a negative errno value otherwise and rte_errno is set.
6842  */
6843 int
6844 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6845 {
6846 	struct mlx5_priv *priv = dev->data->dev_private;
6847 	/*
6848 	 * The LACP matching is done by only using ether type since using
6849 	 * a multicast dst mac causes kernel to give low priority to this flow.
6850 	 */
6851 	static const struct rte_flow_item_eth lacp_spec = {
6852 		.type = RTE_BE16(0x8809),
6853 	};
6854 	static const struct rte_flow_item_eth lacp_mask = {
6855 		.type = 0xffff,
6856 	};
6857 	const struct rte_flow_attr attr = {
6858 		.ingress = 1,
6859 	};
6860 	struct rte_flow_item items[] = {
6861 		{
6862 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6863 			.spec = &lacp_spec,
6864 			.mask = &lacp_mask,
6865 		},
6866 		{
6867 			.type = RTE_FLOW_ITEM_TYPE_END,
6868 		},
6869 	};
6870 	struct rte_flow_action actions[] = {
6871 		{
6872 			.type = (enum rte_flow_action_type)
6873 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6874 		},
6875 		{
6876 			.type = RTE_FLOW_ACTION_TYPE_END,
6877 		},
6878 	};
6879 	struct rte_flow_error error;
6880 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6881 				&attr, items, actions, false, &error);
6882 
6883 	if (!flow_idx)
6884 		return -rte_errno;
6885 	return 0;
6886 }
6887 
6888 /**
6889  * Destroy a flow.
6890  *
6891  * @see rte_flow_destroy()
6892  * @see rte_flow_ops
6893  */
6894 int
6895 mlx5_flow_destroy(struct rte_eth_dev *dev,
6896 		  struct rte_flow *flow,
6897 		  struct rte_flow_error *error __rte_unused)
6898 {
6899 	struct mlx5_priv *priv = dev->data->dev_private;
6900 
6901 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6902 	return 0;
6903 }
6904 
6905 /**
6906  * Destroy all flows.
6907  *
6908  * @see rte_flow_flush()
6909  * @see rte_flow_ops
6910  */
6911 int
6912 mlx5_flow_flush(struct rte_eth_dev *dev,
6913 		struct rte_flow_error *error __rte_unused)
6914 {
6915 	struct mlx5_priv *priv = dev->data->dev_private;
6916 
6917 	mlx5_flow_list_flush(dev, &priv->flows, false);
6918 	return 0;
6919 }
6920 
6921 /**
6922  * Isolated mode.
6923  *
6924  * @see rte_flow_isolate()
6925  * @see rte_flow_ops
6926  */
6927 int
6928 mlx5_flow_isolate(struct rte_eth_dev *dev,
6929 		  int enable,
6930 		  struct rte_flow_error *error)
6931 {
6932 	struct mlx5_priv *priv = dev->data->dev_private;
6933 
6934 	if (dev->data->dev_started) {
6935 		rte_flow_error_set(error, EBUSY,
6936 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6937 				   NULL,
6938 				   "port must be stopped first");
6939 		return -rte_errno;
6940 	}
6941 	priv->isolated = !!enable;
6942 	if (enable)
6943 		dev->dev_ops = &mlx5_dev_ops_isolate;
6944 	else
6945 		dev->dev_ops = &mlx5_dev_ops;
6946 
6947 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6948 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6949 
6950 	return 0;
6951 }
6952 
6953 /**
6954  * Query a flow.
6955  *
6956  * @see rte_flow_query()
6957  * @see rte_flow_ops
6958  */
6959 static int
6960 flow_drv_query(struct rte_eth_dev *dev,
6961 	       uint32_t flow_idx,
6962 	       const struct rte_flow_action *actions,
6963 	       void *data,
6964 	       struct rte_flow_error *error)
6965 {
6966 	struct mlx5_priv *priv = dev->data->dev_private;
6967 	const struct mlx5_flow_driver_ops *fops;
6968 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6969 					       [MLX5_IPOOL_RTE_FLOW],
6970 					       flow_idx);
6971 	enum mlx5_flow_drv_type ftype;
6972 
6973 	if (!flow) {
6974 		return rte_flow_error_set(error, ENOENT,
6975 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6976 			  NULL,
6977 			  "invalid flow handle");
6978 	}
6979 	ftype = flow->drv_type;
6980 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6981 	fops = flow_get_drv_ops(ftype);
6982 
6983 	return fops->query(dev, flow, actions, data, error);
6984 }
6985 
6986 /**
6987  * Query a flow.
6988  *
6989  * @see rte_flow_query()
6990  * @see rte_flow_ops
6991  */
6992 int
6993 mlx5_flow_query(struct rte_eth_dev *dev,
6994 		struct rte_flow *flow,
6995 		const struct rte_flow_action *actions,
6996 		void *data,
6997 		struct rte_flow_error *error)
6998 {
6999 	int ret;
7000 
7001 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7002 			     error);
7003 	if (ret < 0)
7004 		return ret;
7005 	return 0;
7006 }
7007 
7008 /**
7009  * Get rte_flow callbacks.
7010  *
7011  * @param dev
7012  *   Pointer to Ethernet device structure.
7013  * @param ops
7014  *   Pointer to operation-specific structure.
7015  *
7016  * @return 0
7017  */
7018 int
7019 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7020 		  const struct rte_flow_ops **ops)
7021 {
7022 	*ops = &mlx5_flow_ops;
7023 	return 0;
7024 }
7025 
7026 /**
7027  * Validate meter policy actions.
7028  * Dispatcher for action type specific validation.
7029  *
7030  * @param[in] dev
7031  *   Pointer to the Ethernet device structure.
7032  * @param[in] action
7033  *   The meter policy action object to validate.
7034  * @param[in] attr
7035  *   Attributes of flow to determine steering domain.
7036  * @param[out] is_rss
7037  *   Is RSS or not.
7038  * @param[out] domain_bitmap
7039  *   Domain bitmap.
7040  * @param[out] is_def_policy
7041  *   Is default policy or not.
7042  * @param[out] error
7043  *   Perform verbose error reporting if not NULL. Initialized in case of
7044  *   error only.
7045  *
7046  * @return
7047  *   0 on success, otherwise negative errno value.
7048  */
7049 int
7050 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7051 			const struct rte_flow_action *actions[RTE_COLORS],
7052 			struct rte_flow_attr *attr,
7053 			bool *is_rss,
7054 			uint8_t *domain_bitmap,
7055 			bool *is_def_policy,
7056 			struct rte_mtr_error *error)
7057 {
7058 	const struct mlx5_flow_driver_ops *fops;
7059 
7060 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7061 	return fops->validate_mtr_acts(dev, actions, attr,
7062 			is_rss, domain_bitmap, is_def_policy, error);
7063 }
7064 
7065 /**
7066  * Destroy the meter table set.
7067  *
7068  * @param[in] dev
7069  *   Pointer to Ethernet device.
7070  * @param[in] mtr_policy
7071  *   Meter policy struct.
7072  */
7073 void
7074 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7075 		      struct mlx5_flow_meter_policy *mtr_policy)
7076 {
7077 	const struct mlx5_flow_driver_ops *fops;
7078 
7079 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7080 	fops->destroy_mtr_acts(dev, mtr_policy);
7081 }
7082 
7083 /**
7084  * Create policy action, lock free,
7085  * (mutex should be acquired by caller).
7086  * Dispatcher for action type specific call.
7087  *
7088  * @param[in] dev
7089  *   Pointer to the Ethernet device structure.
7090  * @param[in] mtr_policy
7091  *   Meter policy struct.
7092  * @param[in] action
7093  *   Action specification used to create meter actions.
7094  * @param[out] error
7095  *   Perform verbose error reporting if not NULL. Initialized in case of
7096  *   error only.
7097  *
7098  * @return
7099  *   0 on success, otherwise negative errno value.
7100  */
7101 int
7102 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7103 		      struct mlx5_flow_meter_policy *mtr_policy,
7104 		      const struct rte_flow_action *actions[RTE_COLORS],
7105 		      struct rte_mtr_error *error)
7106 {
7107 	const struct mlx5_flow_driver_ops *fops;
7108 
7109 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7110 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7111 }
7112 
7113 /**
7114  * Create policy rules, lock free,
7115  * (mutex should be acquired by caller).
7116  * Dispatcher for action type specific call.
7117  *
7118  * @param[in] dev
7119  *   Pointer to the Ethernet device structure.
7120  * @param[in] mtr_policy
7121  *   Meter policy struct.
7122  *
7123  * @return
7124  *   0 on success, -1 otherwise.
7125  */
7126 int
7127 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7128 			     struct mlx5_flow_meter_policy *mtr_policy)
7129 {
7130 	const struct mlx5_flow_driver_ops *fops;
7131 
7132 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7133 	return fops->create_policy_rules(dev, mtr_policy);
7134 }
7135 
7136 /**
7137  * Destroy policy rules, lock free,
7138  * (mutex should be acquired by caller).
7139  * Dispatcher for action type specific call.
7140  *
7141  * @param[in] dev
7142  *   Pointer to the Ethernet device structure.
7143  * @param[in] mtr_policy
7144  *   Meter policy struct.
7145  */
7146 void
7147 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7148 			     struct mlx5_flow_meter_policy *mtr_policy)
7149 {
7150 	const struct mlx5_flow_driver_ops *fops;
7151 
7152 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7153 	fops->destroy_policy_rules(dev, mtr_policy);
7154 }
7155 
7156 /**
7157  * Destroy the default policy table set.
7158  *
7159  * @param[in] dev
7160  *   Pointer to Ethernet device.
7161  */
7162 void
7163 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7164 {
7165 	const struct mlx5_flow_driver_ops *fops;
7166 
7167 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7168 	fops->destroy_def_policy(dev);
7169 }
7170 
7171 /**
7172  * Destroy the default policy table set.
7173  *
7174  * @param[in] dev
7175  *   Pointer to Ethernet device.
7176  *
7177  * @return
7178  *   0 on success, -1 otherwise.
7179  */
7180 int
7181 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7182 {
7183 	const struct mlx5_flow_driver_ops *fops;
7184 
7185 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7186 	return fops->create_def_policy(dev);
7187 }
7188 
7189 /**
7190  * Create the needed meter and suffix tables.
7191  *
7192  * @param[in] dev
7193  *   Pointer to Ethernet device.
7194  *
7195  * @return
7196  *   0 on success, -1 otherwise.
7197  */
7198 int
7199 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7200 			struct mlx5_flow_meter_info *fm,
7201 			uint32_t mtr_idx,
7202 			uint8_t domain_bitmap)
7203 {
7204 	const struct mlx5_flow_driver_ops *fops;
7205 
7206 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7207 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7208 }
7209 
7210 /**
7211  * Destroy the meter table set.
7212  *
7213  * @param[in] dev
7214  *   Pointer to Ethernet device.
7215  * @param[in] tbl
7216  *   Pointer to the meter table set.
7217  */
7218 void
7219 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7220 			   struct mlx5_flow_meter_info *fm)
7221 {
7222 	const struct mlx5_flow_driver_ops *fops;
7223 
7224 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7225 	fops->destroy_mtr_tbls(dev, fm);
7226 }
7227 
7228 /**
7229  * Destroy the global meter drop table.
7230  *
7231  * @param[in] dev
7232  *   Pointer to Ethernet device.
7233  */
7234 void
7235 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7236 {
7237 	const struct mlx5_flow_driver_ops *fops;
7238 
7239 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7240 	fops->destroy_mtr_drop_tbls(dev);
7241 }
7242 
7243 /**
7244  * Destroy the sub policy table with RX queue.
7245  *
7246  * @param[in] dev
7247  *   Pointer to Ethernet device.
7248  * @param[in] mtr_policy
7249  *   Pointer to meter policy table.
7250  */
7251 void
7252 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7253 		struct mlx5_flow_meter_policy *mtr_policy)
7254 {
7255 	const struct mlx5_flow_driver_ops *fops;
7256 
7257 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7258 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7259 }
7260 
7261 /**
7262  * Allocate the needed aso flow meter id.
7263  *
7264  * @param[in] dev
7265  *   Pointer to Ethernet device.
7266  *
7267  * @return
7268  *   Index to aso flow meter on success, NULL otherwise.
7269  */
7270 uint32_t
7271 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7272 {
7273 	const struct mlx5_flow_driver_ops *fops;
7274 
7275 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7276 	return fops->create_meter(dev);
7277 }
7278 
7279 /**
7280  * Free the aso flow meter id.
7281  *
7282  * @param[in] dev
7283  *   Pointer to Ethernet device.
7284  * @param[in] mtr_idx
7285  *  Index to aso flow meter to be free.
7286  *
7287  * @return
7288  *   0 on success.
7289  */
7290 void
7291 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7292 {
7293 	const struct mlx5_flow_driver_ops *fops;
7294 
7295 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7296 	fops->free_meter(dev, mtr_idx);
7297 }
7298 
7299 /**
7300  * Allocate a counter.
7301  *
7302  * @param[in] dev
7303  *   Pointer to Ethernet device structure.
7304  *
7305  * @return
7306  *   Index to allocated counter  on success, 0 otherwise.
7307  */
7308 uint32_t
7309 mlx5_counter_alloc(struct rte_eth_dev *dev)
7310 {
7311 	const struct mlx5_flow_driver_ops *fops;
7312 	struct rte_flow_attr attr = { .transfer = 0 };
7313 
7314 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7315 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7316 		return fops->counter_alloc(dev);
7317 	}
7318 	DRV_LOG(ERR,
7319 		"port %u counter allocate is not supported.",
7320 		 dev->data->port_id);
7321 	return 0;
7322 }
7323 
7324 /**
7325  * Free a counter.
7326  *
7327  * @param[in] dev
7328  *   Pointer to Ethernet device structure.
7329  * @param[in] cnt
7330  *   Index to counter to be free.
7331  */
7332 void
7333 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7334 {
7335 	const struct mlx5_flow_driver_ops *fops;
7336 	struct rte_flow_attr attr = { .transfer = 0 };
7337 
7338 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7339 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7340 		fops->counter_free(dev, cnt);
7341 		return;
7342 	}
7343 	DRV_LOG(ERR,
7344 		"port %u counter free is not supported.",
7345 		 dev->data->port_id);
7346 }
7347 
7348 /**
7349  * Query counter statistics.
7350  *
7351  * @param[in] dev
7352  *   Pointer to Ethernet device structure.
7353  * @param[in] cnt
7354  *   Index to counter to query.
7355  * @param[in] clear
7356  *   Set to clear counter statistics.
7357  * @param[out] pkts
7358  *   The counter hits packets number to save.
7359  * @param[out] bytes
7360  *   The counter hits bytes number to save.
7361  *
7362  * @return
7363  *   0 on success, a negative errno value otherwise.
7364  */
7365 int
7366 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7367 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7368 {
7369 	const struct mlx5_flow_driver_ops *fops;
7370 	struct rte_flow_attr attr = { .transfer = 0 };
7371 
7372 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7373 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7374 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7375 	}
7376 	DRV_LOG(ERR,
7377 		"port %u counter query is not supported.",
7378 		 dev->data->port_id);
7379 	return -ENOTSUP;
7380 }
7381 
7382 /**
7383  * Allocate a new memory for the counter values wrapped by all the needed
7384  * management.
7385  *
7386  * @param[in] sh
7387  *   Pointer to mlx5_dev_ctx_shared object.
7388  *
7389  * @return
7390  *   0 on success, a negative errno value otherwise.
7391  */
7392 static int
7393 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7394 {
7395 	struct mlx5_devx_mkey_attr mkey_attr;
7396 	struct mlx5_counter_stats_mem_mng *mem_mng;
7397 	volatile struct flow_counter_stats *raw_data;
7398 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7399 	int size = (sizeof(struct flow_counter_stats) *
7400 			MLX5_COUNTERS_PER_POOL +
7401 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7402 			sizeof(struct mlx5_counter_stats_mem_mng);
7403 	size_t pgsize = rte_mem_page_size();
7404 	uint8_t *mem;
7405 	int i;
7406 
7407 	if (pgsize == (size_t)-1) {
7408 		DRV_LOG(ERR, "Failed to get mem page size");
7409 		rte_errno = ENOMEM;
7410 		return -ENOMEM;
7411 	}
7412 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7413 	if (!mem) {
7414 		rte_errno = ENOMEM;
7415 		return -ENOMEM;
7416 	}
7417 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7418 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7419 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7420 						 IBV_ACCESS_LOCAL_WRITE);
7421 	if (!mem_mng->umem) {
7422 		rte_errno = errno;
7423 		mlx5_free(mem);
7424 		return -rte_errno;
7425 	}
7426 	memset(&mkey_attr, 0, sizeof(mkey_attr));
7427 	mkey_attr.addr = (uintptr_t)mem;
7428 	mkey_attr.size = size;
7429 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7430 	mkey_attr.pd = sh->pdn;
7431 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7432 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7433 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7434 	if (!mem_mng->dm) {
7435 		mlx5_os_umem_dereg(mem_mng->umem);
7436 		rte_errno = errno;
7437 		mlx5_free(mem);
7438 		return -rte_errno;
7439 	}
7440 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7441 	raw_data = (volatile struct flow_counter_stats *)mem;
7442 	for (i = 0; i < raws_n; ++i) {
7443 		mem_mng->raws[i].mem_mng = mem_mng;
7444 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7445 	}
7446 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7447 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7448 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7449 				 next);
7450 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7451 	sh->cmng.mem_mng = mem_mng;
7452 	return 0;
7453 }
7454 
7455 /**
7456  * Set the statistic memory to the new counter pool.
7457  *
7458  * @param[in] sh
7459  *   Pointer to mlx5_dev_ctx_shared object.
7460  * @param[in] pool
7461  *   Pointer to the pool to set the statistic memory.
7462  *
7463  * @return
7464  *   0 on success, a negative errno value otherwise.
7465  */
7466 static int
7467 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7468 			       struct mlx5_flow_counter_pool *pool)
7469 {
7470 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7471 	/* Resize statistic memory once used out. */
7472 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7473 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7474 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7475 		return -1;
7476 	}
7477 	rte_spinlock_lock(&pool->sl);
7478 	pool->raw = cmng->mem_mng->raws + pool->index %
7479 		    MLX5_CNT_CONTAINER_RESIZE;
7480 	rte_spinlock_unlock(&pool->sl);
7481 	pool->raw_hw = NULL;
7482 	return 0;
7483 }
7484 
7485 #define MLX5_POOL_QUERY_FREQ_US 1000000
7486 
7487 /**
7488  * Set the periodic procedure for triggering asynchronous batch queries for all
7489  * the counter pools.
7490  *
7491  * @param[in] sh
7492  *   Pointer to mlx5_dev_ctx_shared object.
7493  */
7494 void
7495 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7496 {
7497 	uint32_t pools_n, us;
7498 
7499 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7500 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7501 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7502 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7503 		sh->cmng.query_thread_on = 0;
7504 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7505 	} else {
7506 		sh->cmng.query_thread_on = 1;
7507 	}
7508 }
7509 
7510 /**
7511  * The periodic procedure for triggering asynchronous batch queries for all the
7512  * counter pools. This function is probably called by the host thread.
7513  *
7514  * @param[in] arg
7515  *   The parameter for the alarm process.
7516  */
7517 void
7518 mlx5_flow_query_alarm(void *arg)
7519 {
7520 	struct mlx5_dev_ctx_shared *sh = arg;
7521 	int ret;
7522 	uint16_t pool_index = sh->cmng.pool_index;
7523 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7524 	struct mlx5_flow_counter_pool *pool;
7525 	uint16_t n_valid;
7526 
7527 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7528 		goto set_alarm;
7529 	rte_spinlock_lock(&cmng->pool_update_sl);
7530 	pool = cmng->pools[pool_index];
7531 	n_valid = cmng->n_valid;
7532 	rte_spinlock_unlock(&cmng->pool_update_sl);
7533 	/* Set the statistic memory to the new created pool. */
7534 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7535 		goto set_alarm;
7536 	if (pool->raw_hw)
7537 		/* There is a pool query in progress. */
7538 		goto set_alarm;
7539 	pool->raw_hw =
7540 		LIST_FIRST(&sh->cmng.free_stat_raws);
7541 	if (!pool->raw_hw)
7542 		/* No free counter statistics raw memory. */
7543 		goto set_alarm;
7544 	/*
7545 	 * Identify the counters released between query trigger and query
7546 	 * handle more efficiently. The counter released in this gap period
7547 	 * should wait for a new round of query as the new arrived packets
7548 	 * will not be taken into account.
7549 	 */
7550 	pool->query_gen++;
7551 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7552 					       MLX5_COUNTERS_PER_POOL,
7553 					       NULL, NULL,
7554 					       pool->raw_hw->mem_mng->dm->id,
7555 					       (void *)(uintptr_t)
7556 					       pool->raw_hw->data,
7557 					       sh->devx_comp,
7558 					       (uint64_t)(uintptr_t)pool);
7559 	if (ret) {
7560 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7561 			" %d", pool->min_dcs->id);
7562 		pool->raw_hw = NULL;
7563 		goto set_alarm;
7564 	}
7565 	LIST_REMOVE(pool->raw_hw, next);
7566 	sh->cmng.pending_queries++;
7567 	pool_index++;
7568 	if (pool_index >= n_valid)
7569 		pool_index = 0;
7570 set_alarm:
7571 	sh->cmng.pool_index = pool_index;
7572 	mlx5_set_query_alarm(sh);
7573 }
7574 
7575 /**
7576  * Check and callback event for new aged flow in the counter pool
7577  *
7578  * @param[in] sh
7579  *   Pointer to mlx5_dev_ctx_shared object.
7580  * @param[in] pool
7581  *   Pointer to Current counter pool.
7582  */
7583 static void
7584 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7585 		   struct mlx5_flow_counter_pool *pool)
7586 {
7587 	struct mlx5_priv *priv;
7588 	struct mlx5_flow_counter *cnt;
7589 	struct mlx5_age_info *age_info;
7590 	struct mlx5_age_param *age_param;
7591 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7592 	struct mlx5_counter_stats_raw *prev = pool->raw;
7593 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7594 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7595 	uint16_t expected = AGE_CANDIDATE;
7596 	uint32_t i;
7597 
7598 	pool->time_of_last_age_check = curr_time;
7599 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7600 		cnt = MLX5_POOL_GET_CNT(pool, i);
7601 		age_param = MLX5_CNT_TO_AGE(cnt);
7602 		if (__atomic_load_n(&age_param->state,
7603 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7604 			continue;
7605 		if (cur->data[i].hits != prev->data[i].hits) {
7606 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7607 					 __ATOMIC_RELAXED);
7608 			continue;
7609 		}
7610 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7611 				       time_delta,
7612 				       __ATOMIC_RELAXED) <= age_param->timeout)
7613 			continue;
7614 		/**
7615 		 * Hold the lock first, or if between the
7616 		 * state AGE_TMOUT and tailq operation the
7617 		 * release happened, the release procedure
7618 		 * may delete a non-existent tailq node.
7619 		 */
7620 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7621 		age_info = GET_PORT_AGE_INFO(priv);
7622 		rte_spinlock_lock(&age_info->aged_sl);
7623 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7624 						AGE_TMOUT, false,
7625 						__ATOMIC_RELAXED,
7626 						__ATOMIC_RELAXED)) {
7627 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7628 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7629 		}
7630 		rte_spinlock_unlock(&age_info->aged_sl);
7631 	}
7632 	mlx5_age_event_prepare(sh);
7633 }
7634 
7635 /**
7636  * Handler for the HW respond about ready values from an asynchronous batch
7637  * query. This function is probably called by the host thread.
7638  *
7639  * @param[in] sh
7640  *   The pointer to the shared device context.
7641  * @param[in] async_id
7642  *   The Devx async ID.
7643  * @param[in] status
7644  *   The status of the completion.
7645  */
7646 void
7647 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7648 				  uint64_t async_id, int status)
7649 {
7650 	struct mlx5_flow_counter_pool *pool =
7651 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7652 	struct mlx5_counter_stats_raw *raw_to_free;
7653 	uint8_t query_gen = pool->query_gen ^ 1;
7654 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7655 	enum mlx5_counter_type cnt_type =
7656 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7657 				MLX5_COUNTER_TYPE_ORIGIN;
7658 
7659 	if (unlikely(status)) {
7660 		raw_to_free = pool->raw_hw;
7661 	} else {
7662 		raw_to_free = pool->raw;
7663 		if (pool->is_aged)
7664 			mlx5_flow_aging_check(sh, pool);
7665 		rte_spinlock_lock(&pool->sl);
7666 		pool->raw = pool->raw_hw;
7667 		rte_spinlock_unlock(&pool->sl);
7668 		/* Be sure the new raw counters data is updated in memory. */
7669 		rte_io_wmb();
7670 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7671 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7672 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7673 				     &pool->counters[query_gen], next);
7674 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7675 		}
7676 	}
7677 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7678 	pool->raw_hw = NULL;
7679 	sh->cmng.pending_queries--;
7680 }
7681 
7682 static int
7683 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7684 		    const struct flow_grp_info *grp_info,
7685 		    struct rte_flow_error *error)
7686 {
7687 	if (grp_info->transfer && grp_info->external &&
7688 	    grp_info->fdb_def_rule) {
7689 		if (group == UINT32_MAX)
7690 			return rte_flow_error_set
7691 						(error, EINVAL,
7692 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7693 						 NULL,
7694 						 "group index not supported");
7695 		*table = group + 1;
7696 	} else {
7697 		*table = group;
7698 	}
7699 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7700 	return 0;
7701 }
7702 
7703 /**
7704  * Translate the rte_flow group index to HW table value.
7705  *
7706  * If tunnel offload is disabled, all group ids converted to flow table
7707  * id using the standard method.
7708  * If tunnel offload is enabled, group id can be converted using the
7709  * standard or tunnel conversion method. Group conversion method
7710  * selection depends on flags in `grp_info` parameter:
7711  * - Internal (grp_info.external == 0) groups conversion uses the
7712  *   standard method.
7713  * - Group ids in JUMP action converted with the tunnel conversion.
7714  * - Group id in rule attribute conversion depends on a rule type and
7715  *   group id value:
7716  *   ** non zero group attributes converted with the tunnel method
7717  *   ** zero group attribute in non-tunnel rule is converted using the
7718  *      standard method - there's only one root table
7719  *   ** zero group attribute in steer tunnel rule is converted with the
7720  *      standard method - single root table
7721  *   ** zero group attribute in match tunnel rule is a special OvS
7722  *      case: that value is used for portability reasons. That group
7723  *      id is converted with the tunnel conversion method.
7724  *
7725  * @param[in] dev
7726  *   Port device
7727  * @param[in] tunnel
7728  *   PMD tunnel offload object
7729  * @param[in] group
7730  *   rte_flow group index value.
7731  * @param[out] table
7732  *   HW table value.
7733  * @param[in] grp_info
7734  *   flags used for conversion
7735  * @param[out] error
7736  *   Pointer to error structure.
7737  *
7738  * @return
7739  *   0 on success, a negative errno value otherwise and rte_errno is set.
7740  */
7741 int
7742 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7743 			 const struct mlx5_flow_tunnel *tunnel,
7744 			 uint32_t group, uint32_t *table,
7745 			 const struct flow_grp_info *grp_info,
7746 			 struct rte_flow_error *error)
7747 {
7748 	int ret;
7749 	bool standard_translation;
7750 
7751 	if (!grp_info->skip_scale && grp_info->external &&
7752 	    group < MLX5_MAX_TABLES_EXTERNAL)
7753 		group *= MLX5_FLOW_TABLE_FACTOR;
7754 	if (is_tunnel_offload_active(dev)) {
7755 		standard_translation = !grp_info->external ||
7756 					grp_info->std_tbl_fix;
7757 	} else {
7758 		standard_translation = true;
7759 	}
7760 	DRV_LOG(DEBUG,
7761 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7762 		dev->data->port_id, group, grp_info->transfer,
7763 		grp_info->external, grp_info->fdb_def_rule,
7764 		standard_translation ? "STANDARD" : "TUNNEL");
7765 	if (standard_translation)
7766 		ret = flow_group_to_table(dev->data->port_id, group, table,
7767 					  grp_info, error);
7768 	else
7769 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7770 						      table, error);
7771 
7772 	return ret;
7773 }
7774 
7775 /**
7776  * Discover availability of metadata reg_c's.
7777  *
7778  * Iteratively use test flows to check availability.
7779  *
7780  * @param[in] dev
7781  *   Pointer to the Ethernet device structure.
7782  *
7783  * @return
7784  *   0 on success, a negative errno value otherwise and rte_errno is set.
7785  */
7786 int
7787 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7788 {
7789 	struct mlx5_priv *priv = dev->data->dev_private;
7790 	struct mlx5_dev_config *config = &priv->config;
7791 	enum modify_reg idx;
7792 	int n = 0;
7793 
7794 	/* reg_c[0] and reg_c[1] are reserved. */
7795 	config->flow_mreg_c[n++] = REG_C_0;
7796 	config->flow_mreg_c[n++] = REG_C_1;
7797 	/* Discover availability of other reg_c's. */
7798 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7799 		struct rte_flow_attr attr = {
7800 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7801 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7802 			.ingress = 1,
7803 		};
7804 		struct rte_flow_item items[] = {
7805 			[0] = {
7806 				.type = RTE_FLOW_ITEM_TYPE_END,
7807 			},
7808 		};
7809 		struct rte_flow_action actions[] = {
7810 			[0] = {
7811 				.type = (enum rte_flow_action_type)
7812 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7813 				.conf = &(struct mlx5_flow_action_copy_mreg){
7814 					.src = REG_C_1,
7815 					.dst = idx,
7816 				},
7817 			},
7818 			[1] = {
7819 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7820 				.conf = &(struct rte_flow_action_jump){
7821 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7822 				},
7823 			},
7824 			[2] = {
7825 				.type = RTE_FLOW_ACTION_TYPE_END,
7826 			},
7827 		};
7828 		uint32_t flow_idx;
7829 		struct rte_flow *flow;
7830 		struct rte_flow_error error;
7831 
7832 		if (!config->dv_flow_en)
7833 			break;
7834 		/* Create internal flow, validation skips copy action. */
7835 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7836 					    actions, false, &error);
7837 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7838 				      flow_idx);
7839 		if (!flow)
7840 			continue;
7841 		config->flow_mreg_c[n++] = idx;
7842 		flow_list_destroy(dev, NULL, flow_idx);
7843 	}
7844 	for (; n < MLX5_MREG_C_NUM; ++n)
7845 		config->flow_mreg_c[n] = REG_NON;
7846 	return 0;
7847 }
7848 
7849 /**
7850  * Dump flow raw hw data to file
7851  *
7852  * @param[in] dev
7853  *    The pointer to Ethernet device.
7854  * @param[in] file
7855  *   A pointer to a file for output.
7856  * @param[out] error
7857  *   Perform verbose error reporting if not NULL. PMDs initialize this
7858  *   structure in case of error only.
7859  * @return
7860  *   0 on success, a nagative value otherwise.
7861  */
7862 int
7863 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7864 		   FILE *file,
7865 		   struct rte_flow_error *error __rte_unused)
7866 {
7867 	struct mlx5_priv *priv = dev->data->dev_private;
7868 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7869 	uint32_t handle_idx;
7870 	int ret;
7871 	struct mlx5_flow_handle *dh;
7872 	struct rte_flow *flow;
7873 
7874 	if (!priv->config.dv_flow_en) {
7875 		if (fputs("device dv flow disabled\n", file) <= 0)
7876 			return -errno;
7877 		return -ENOTSUP;
7878 	}
7879 
7880 	/* dump all */
7881 	if (!flow_idx)
7882 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7883 					sh->rx_domain,
7884 					sh->tx_domain, file);
7885 	/* dump one */
7886 	flow = mlx5_ipool_get(priv->sh->ipool
7887 			[MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
7888 	if (!flow)
7889 		return -ENOENT;
7890 
7891 	handle_idx = flow->dev_handles;
7892 	while (handle_idx) {
7893 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
7894 				handle_idx);
7895 		if (!dh)
7896 			return -ENOENT;
7897 		if (dh->drv_flow) {
7898 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
7899 					file);
7900 			if (ret)
7901 				return -ENOENT;
7902 		}
7903 		handle_idx = dh->next.next;
7904 	}
7905 	return 0;
7906 }
7907 
7908 /**
7909  * Get aged-out flows.
7910  *
7911  * @param[in] dev
7912  *   Pointer to the Ethernet device structure.
7913  * @param[in] context
7914  *   The address of an array of pointers to the aged-out flows contexts.
7915  * @param[in] nb_countexts
7916  *   The length of context array pointers.
7917  * @param[out] error
7918  *   Perform verbose error reporting if not NULL. Initialized in case of
7919  *   error only.
7920  *
7921  * @return
7922  *   how many contexts get in success, otherwise negative errno value.
7923  *   if nb_contexts is 0, return the amount of all aged contexts.
7924  *   if nb_contexts is not 0 , return the amount of aged flows reported
7925  *   in the context array.
7926  */
7927 int
7928 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7929 			uint32_t nb_contexts, struct rte_flow_error *error)
7930 {
7931 	const struct mlx5_flow_driver_ops *fops;
7932 	struct rte_flow_attr attr = { .transfer = 0 };
7933 
7934 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7935 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7936 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7937 						    error);
7938 	}
7939 	DRV_LOG(ERR,
7940 		"port %u get aged flows is not supported.",
7941 		 dev->data->port_id);
7942 	return -ENOTSUP;
7943 }
7944 
7945 /* Wrapper for driver action_validate op callback */
7946 static int
7947 flow_drv_action_validate(struct rte_eth_dev *dev,
7948 			 const struct rte_flow_indir_action_conf *conf,
7949 			 const struct rte_flow_action *action,
7950 			 const struct mlx5_flow_driver_ops *fops,
7951 			 struct rte_flow_error *error)
7952 {
7953 	static const char err_msg[] = "indirect action validation unsupported";
7954 
7955 	if (!fops->action_validate) {
7956 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7957 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7958 				   NULL, err_msg);
7959 		return -rte_errno;
7960 	}
7961 	return fops->action_validate(dev, conf, action, error);
7962 }
7963 
7964 /**
7965  * Destroys the shared action by handle.
7966  *
7967  * @param dev
7968  *   Pointer to Ethernet device structure.
7969  * @param[in] handle
7970  *   Handle for the indirect action object to be destroyed.
7971  * @param[out] error
7972  *   Perform verbose error reporting if not NULL. PMDs initialize this
7973  *   structure in case of error only.
7974  *
7975  * @return
7976  *   0 on success, a negative errno value otherwise and rte_errno is set.
7977  *
7978  * @note: wrapper for driver action_create op callback.
7979  */
7980 static int
7981 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7982 			   struct rte_flow_action_handle *handle,
7983 			   struct rte_flow_error *error)
7984 {
7985 	static const char err_msg[] = "indirect action destruction unsupported";
7986 	struct rte_flow_attr attr = { .transfer = 0 };
7987 	const struct mlx5_flow_driver_ops *fops =
7988 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7989 
7990 	if (!fops->action_destroy) {
7991 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7992 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7993 				   NULL, err_msg);
7994 		return -rte_errno;
7995 	}
7996 	return fops->action_destroy(dev, handle, error);
7997 }
7998 
7999 /* Wrapper for driver action_destroy op callback */
8000 static int
8001 flow_drv_action_update(struct rte_eth_dev *dev,
8002 		       struct rte_flow_action_handle *handle,
8003 		       const void *update,
8004 		       const struct mlx5_flow_driver_ops *fops,
8005 		       struct rte_flow_error *error)
8006 {
8007 	static const char err_msg[] = "indirect action update unsupported";
8008 
8009 	if (!fops->action_update) {
8010 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8011 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8012 				   NULL, err_msg);
8013 		return -rte_errno;
8014 	}
8015 	return fops->action_update(dev, handle, update, error);
8016 }
8017 
8018 /* Wrapper for driver action_destroy op callback */
8019 static int
8020 flow_drv_action_query(struct rte_eth_dev *dev,
8021 		      const struct rte_flow_action_handle *handle,
8022 		      void *data,
8023 		      const struct mlx5_flow_driver_ops *fops,
8024 		      struct rte_flow_error *error)
8025 {
8026 	static const char err_msg[] = "indirect action query unsupported";
8027 
8028 	if (!fops->action_query) {
8029 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8030 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8031 				   NULL, err_msg);
8032 		return -rte_errno;
8033 	}
8034 	return fops->action_query(dev, handle, data, error);
8035 }
8036 
8037 /**
8038  * Create indirect action for reuse in multiple flow rules.
8039  *
8040  * @param dev
8041  *   Pointer to Ethernet device structure.
8042  * @param conf
8043  *   Pointer to indirect action object configuration.
8044  * @param[in] action
8045  *   Action configuration for indirect action object creation.
8046  * @param[out] error
8047  *   Perform verbose error reporting if not NULL. PMDs initialize this
8048  *   structure in case of error only.
8049  * @return
8050  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8051  */
8052 static struct rte_flow_action_handle *
8053 mlx5_action_handle_create(struct rte_eth_dev *dev,
8054 			  const struct rte_flow_indir_action_conf *conf,
8055 			  const struct rte_flow_action *action,
8056 			  struct rte_flow_error *error)
8057 {
8058 	static const char err_msg[] = "indirect action creation unsupported";
8059 	struct rte_flow_attr attr = { .transfer = 0 };
8060 	const struct mlx5_flow_driver_ops *fops =
8061 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8062 
8063 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8064 		return NULL;
8065 	if (!fops->action_create) {
8066 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8067 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8068 				   NULL, err_msg);
8069 		return NULL;
8070 	}
8071 	return fops->action_create(dev, conf, action, error);
8072 }
8073 
8074 /**
8075  * Updates inplace the indirect action configuration pointed by *handle*
8076  * with the configuration provided as *update* argument.
8077  * The update of the indirect action configuration effects all flow rules
8078  * reusing the action via handle.
8079  *
8080  * @param dev
8081  *   Pointer to Ethernet device structure.
8082  * @param[in] handle
8083  *   Handle for the indirect action to be updated.
8084  * @param[in] update
8085  *   Action specification used to modify the action pointed by handle.
8086  *   *update* could be of same type with the action pointed by the *handle*
8087  *   handle argument, or some other structures like a wrapper, depending on
8088  *   the indirect action type.
8089  * @param[out] error
8090  *   Perform verbose error reporting if not NULL. PMDs initialize this
8091  *   structure in case of error only.
8092  *
8093  * @return
8094  *   0 on success, a negative errno value otherwise and rte_errno is set.
8095  */
8096 static int
8097 mlx5_action_handle_update(struct rte_eth_dev *dev,
8098 		struct rte_flow_action_handle *handle,
8099 		const void *update,
8100 		struct rte_flow_error *error)
8101 {
8102 	struct rte_flow_attr attr = { .transfer = 0 };
8103 	const struct mlx5_flow_driver_ops *fops =
8104 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8105 	int ret;
8106 
8107 	ret = flow_drv_action_validate(dev, NULL,
8108 			(const struct rte_flow_action *)update, fops, error);
8109 	if (ret)
8110 		return ret;
8111 	return flow_drv_action_update(dev, handle, update, fops,
8112 				      error);
8113 }
8114 
8115 /**
8116  * Query the indirect action by handle.
8117  *
8118  * This function allows retrieving action-specific data such as counters.
8119  * Data is gathered by special action which may be present/referenced in
8120  * more than one flow rule definition.
8121  *
8122  * see @RTE_FLOW_ACTION_TYPE_COUNT
8123  *
8124  * @param dev
8125  *   Pointer to Ethernet device structure.
8126  * @param[in] handle
8127  *   Handle for the indirect action to query.
8128  * @param[in, out] data
8129  *   Pointer to storage for the associated query data type.
8130  * @param[out] error
8131  *   Perform verbose error reporting if not NULL. PMDs initialize this
8132  *   structure in case of error only.
8133  *
8134  * @return
8135  *   0 on success, a negative errno value otherwise and rte_errno is set.
8136  */
8137 static int
8138 mlx5_action_handle_query(struct rte_eth_dev *dev,
8139 			 const struct rte_flow_action_handle *handle,
8140 			 void *data,
8141 			 struct rte_flow_error *error)
8142 {
8143 	struct rte_flow_attr attr = { .transfer = 0 };
8144 	const struct mlx5_flow_driver_ops *fops =
8145 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8146 
8147 	return flow_drv_action_query(dev, handle, data, fops, error);
8148 }
8149 
8150 /**
8151  * Destroy all indirect actions (shared RSS).
8152  *
8153  * @param dev
8154  *   Pointer to Ethernet device.
8155  *
8156  * @return
8157  *   0 on success, a negative errno value otherwise and rte_errno is set.
8158  */
8159 int
8160 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8161 {
8162 	struct rte_flow_error error;
8163 	struct mlx5_priv *priv = dev->data->dev_private;
8164 	struct mlx5_shared_action_rss *shared_rss;
8165 	int ret = 0;
8166 	uint32_t idx;
8167 
8168 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8169 		      priv->rss_shared_actions, idx, shared_rss, next) {
8170 		ret |= mlx5_action_handle_destroy(dev,
8171 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8172 	}
8173 	return ret;
8174 }
8175 
8176 #ifndef HAVE_MLX5DV_DR
8177 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8178 #else
8179 #define MLX5_DOMAIN_SYNC_FLOW \
8180 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8181 #endif
8182 
8183 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8184 {
8185 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8186 	const struct mlx5_flow_driver_ops *fops;
8187 	int ret;
8188 	struct rte_flow_attr attr = { .transfer = 0 };
8189 
8190 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8191 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8192 	if (ret > 0)
8193 		ret = -ret;
8194 	return ret;
8195 }
8196 
8197 const struct mlx5_flow_tunnel *
8198 mlx5_get_tof(const struct rte_flow_item *item,
8199 	     const struct rte_flow_action *action,
8200 	     enum mlx5_tof_rule_type *rule_type)
8201 {
8202 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
8203 		if (item->type == (typeof(item->type))
8204 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
8205 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
8206 			return flow_items_to_tunnel(item);
8207 		}
8208 	}
8209 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
8210 		if (action->type == (typeof(action->type))
8211 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
8212 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
8213 			return flow_actions_to_tunnel(action);
8214 		}
8215 	}
8216 	return NULL;
8217 }
8218 
8219 /**
8220  * tunnel offload functionalilty is defined for DV environment only
8221  */
8222 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8223 __extension__
8224 union tunnel_offload_mark {
8225 	uint32_t val;
8226 	struct {
8227 		uint32_t app_reserve:8;
8228 		uint32_t table_id:15;
8229 		uint32_t transfer:1;
8230 		uint32_t _unused_:8;
8231 	};
8232 };
8233 
8234 static bool
8235 mlx5_access_tunnel_offload_db
8236 	(struct rte_eth_dev *dev,
8237 	 bool (*match)(struct rte_eth_dev *,
8238 		       struct mlx5_flow_tunnel *, const void *),
8239 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8240 	 void (*miss)(struct rte_eth_dev *, void *),
8241 	 void *ctx, bool lock_op);
8242 
8243 static int
8244 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8245 			     struct rte_flow *flow,
8246 			     const struct rte_flow_attr *attr,
8247 			     const struct rte_flow_action *app_actions,
8248 			     uint32_t flow_idx,
8249 			     const struct mlx5_flow_tunnel *tunnel,
8250 			     struct tunnel_default_miss_ctx *ctx,
8251 			     struct rte_flow_error *error)
8252 {
8253 	struct mlx5_priv *priv = dev->data->dev_private;
8254 	struct mlx5_flow *dev_flow;
8255 	struct rte_flow_attr miss_attr = *attr;
8256 	const struct rte_flow_item miss_items[2] = {
8257 		{
8258 			.type = RTE_FLOW_ITEM_TYPE_ETH,
8259 			.spec = NULL,
8260 			.last = NULL,
8261 			.mask = NULL
8262 		},
8263 		{
8264 			.type = RTE_FLOW_ITEM_TYPE_END,
8265 			.spec = NULL,
8266 			.last = NULL,
8267 			.mask = NULL
8268 		}
8269 	};
8270 	union tunnel_offload_mark mark_id;
8271 	struct rte_flow_action_mark miss_mark;
8272 	struct rte_flow_action miss_actions[3] = {
8273 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8274 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
8275 	};
8276 	const struct rte_flow_action_jump *jump_data;
8277 	uint32_t i, flow_table = 0; /* prevent compilation warning */
8278 	struct flow_grp_info grp_info = {
8279 		.external = 1,
8280 		.transfer = attr->transfer,
8281 		.fdb_def_rule = !!priv->fdb_def_rule,
8282 		.std_tbl_fix = 0,
8283 	};
8284 	int ret;
8285 
8286 	if (!attr->transfer) {
8287 		uint32_t q_size;
8288 
8289 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8290 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8291 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8292 					 0, SOCKET_ID_ANY);
8293 		if (!ctx->queue)
8294 			return rte_flow_error_set
8295 				(error, ENOMEM,
8296 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8297 				NULL, "invalid default miss RSS");
8298 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8299 		ctx->action_rss.level = 0,
8300 		ctx->action_rss.types = priv->rss_conf.rss_hf,
8301 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8302 		ctx->action_rss.queue_num = priv->reta_idx_n,
8303 		ctx->action_rss.key = priv->rss_conf.rss_key,
8304 		ctx->action_rss.queue = ctx->queue;
8305 		if (!priv->reta_idx_n || !priv->rxqs_n)
8306 			return rte_flow_error_set
8307 				(error, EINVAL,
8308 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8309 				NULL, "invalid port configuration");
8310 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8311 			ctx->action_rss.types = 0;
8312 		for (i = 0; i != priv->reta_idx_n; ++i)
8313 			ctx->queue[i] = (*priv->reta_idx)[i];
8314 	} else {
8315 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8316 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8317 	}
8318 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8319 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8320 	jump_data = app_actions->conf;
8321 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8322 	miss_attr.group = jump_data->group;
8323 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8324 				       &flow_table, &grp_info, error);
8325 	if (ret)
8326 		return rte_flow_error_set(error, EINVAL,
8327 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8328 					  NULL, "invalid tunnel id");
8329 	mark_id.app_reserve = 0;
8330 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8331 	mark_id.transfer = !!attr->transfer;
8332 	mark_id._unused_ = 0;
8333 	miss_mark.id = mark_id.val;
8334 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8335 				    miss_items, miss_actions, flow_idx, error);
8336 	if (!dev_flow)
8337 		return -rte_errno;
8338 	dev_flow->flow = flow;
8339 	dev_flow->external = true;
8340 	dev_flow->tunnel = tunnel;
8341 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
8342 	/* Subflow object was created, we must include one in the list. */
8343 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8344 		      dev_flow->handle, next);
8345 	DRV_LOG(DEBUG,
8346 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8347 		dev->data->port_id, tunnel->app_tunnel.type,
8348 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8349 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8350 				  miss_actions, error);
8351 	if (!ret)
8352 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8353 						  error);
8354 
8355 	return ret;
8356 }
8357 
8358 static const struct mlx5_flow_tbl_data_entry  *
8359 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8360 {
8361 	struct mlx5_priv *priv = dev->data->dev_private;
8362 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8363 	struct mlx5_hlist_entry *he;
8364 	union tunnel_offload_mark mbits = { .val = mark };
8365 	union mlx5_flow_tbl_key table_key = {
8366 		{
8367 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
8368 			.id = 0,
8369 			.reserved = 0,
8370 			.dummy = 0,
8371 			.is_fdb = !!mbits.transfer,
8372 			.is_egress = 0,
8373 		}
8374 	};
8375 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8376 	return he ?
8377 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8378 }
8379 
8380 static void
8381 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8382 				   struct mlx5_hlist_entry *entry)
8383 {
8384 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8385 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8386 
8387 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8388 			tunnel_flow_tbl_to_id(tte->flow_table));
8389 	mlx5_free(tte);
8390 }
8391 
8392 static int
8393 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8394 				  struct mlx5_hlist_entry *entry,
8395 				  uint64_t key, void *cb_ctx __rte_unused)
8396 {
8397 	union tunnel_tbl_key tbl = {
8398 		.val = key,
8399 	};
8400 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8401 
8402 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8403 }
8404 
8405 static struct mlx5_hlist_entry *
8406 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8407 				   void *ctx __rte_unused)
8408 {
8409 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8410 	struct tunnel_tbl_entry *tte;
8411 	union tunnel_tbl_key tbl = {
8412 		.val = key,
8413 	};
8414 
8415 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8416 			  sizeof(*tte), 0,
8417 			  SOCKET_ID_ANY);
8418 	if (!tte)
8419 		goto err;
8420 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8421 			  &tte->flow_table);
8422 	if (tte->flow_table >= MLX5_MAX_TABLES) {
8423 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8424 			tte->flow_table);
8425 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8426 				tte->flow_table);
8427 		goto err;
8428 	} else if (!tte->flow_table) {
8429 		goto err;
8430 	}
8431 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8432 	tte->tunnel_id = tbl.tunnel_id;
8433 	tte->group = tbl.group;
8434 	return &tte->hash;
8435 err:
8436 	if (tte)
8437 		mlx5_free(tte);
8438 	return NULL;
8439 }
8440 
8441 static uint32_t
8442 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8443 				const struct mlx5_flow_tunnel *tunnel,
8444 				uint32_t group, uint32_t *table,
8445 				struct rte_flow_error *error)
8446 {
8447 	struct mlx5_hlist_entry *he;
8448 	struct tunnel_tbl_entry *tte;
8449 	union tunnel_tbl_key key = {
8450 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8451 		.group = group
8452 	};
8453 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8454 	struct mlx5_hlist *group_hash;
8455 
8456 	group_hash = tunnel ? tunnel->groups : thub->groups;
8457 	he = mlx5_hlist_register(group_hash, key.val, NULL);
8458 	if (!he)
8459 		return rte_flow_error_set(error, EINVAL,
8460 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8461 					  NULL,
8462 					  "tunnel group index not supported");
8463 	tte = container_of(he, typeof(*tte), hash);
8464 	*table = tte->flow_table;
8465 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8466 		dev->data->port_id, key.tunnel_id, group, *table);
8467 	return 0;
8468 }
8469 
8470 static void
8471 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8472 		      struct mlx5_flow_tunnel *tunnel)
8473 {
8474 	struct mlx5_priv *priv = dev->data->dev_private;
8475 	struct mlx5_indexed_pool *ipool;
8476 
8477 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8478 		dev->data->port_id, tunnel->tunnel_id);
8479 	LIST_REMOVE(tunnel, chain);
8480 	mlx5_hlist_destroy(tunnel->groups);
8481 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8482 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
8483 }
8484 
8485 static bool
8486 mlx5_access_tunnel_offload_db
8487 	(struct rte_eth_dev *dev,
8488 	 bool (*match)(struct rte_eth_dev *,
8489 		       struct mlx5_flow_tunnel *, const void *),
8490 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8491 	 void (*miss)(struct rte_eth_dev *, void *),
8492 	 void *ctx, bool lock_op)
8493 {
8494 	bool verdict = false;
8495 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8496 	struct mlx5_flow_tunnel *tunnel;
8497 
8498 	rte_spinlock_lock(&thub->sl);
8499 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8500 		verdict = match(dev, tunnel, (const void *)ctx);
8501 		if (verdict)
8502 			break;
8503 	}
8504 	if (!lock_op)
8505 		rte_spinlock_unlock(&thub->sl);
8506 	if (verdict && hit)
8507 		hit(dev, tunnel, ctx);
8508 	if (!verdict && miss)
8509 		miss(dev, ctx);
8510 	if (lock_op)
8511 		rte_spinlock_unlock(&thub->sl);
8512 
8513 	return verdict;
8514 }
8515 
8516 struct tunnel_db_find_tunnel_id_ctx {
8517 	uint32_t tunnel_id;
8518 	struct mlx5_flow_tunnel *tunnel;
8519 };
8520 
8521 static bool
8522 find_tunnel_id_match(struct rte_eth_dev *dev,
8523 		     struct mlx5_flow_tunnel *tunnel, const void *x)
8524 {
8525 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8526 
8527 	RTE_SET_USED(dev);
8528 	return tunnel->tunnel_id == ctx->tunnel_id;
8529 }
8530 
8531 static void
8532 find_tunnel_id_hit(struct rte_eth_dev *dev,
8533 		   struct mlx5_flow_tunnel *tunnel, void *x)
8534 {
8535 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8536 	RTE_SET_USED(dev);
8537 	ctx->tunnel = tunnel;
8538 }
8539 
8540 static struct mlx5_flow_tunnel *
8541 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8542 {
8543 	struct tunnel_db_find_tunnel_id_ctx ctx = {
8544 		.tunnel_id = id,
8545 	};
8546 
8547 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8548 				      find_tunnel_id_hit, NULL, &ctx, true);
8549 
8550 	return ctx.tunnel;
8551 }
8552 
8553 static struct mlx5_flow_tunnel *
8554 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8555 			  const struct rte_flow_tunnel *app_tunnel)
8556 {
8557 	struct mlx5_priv *priv = dev->data->dev_private;
8558 	struct mlx5_indexed_pool *ipool;
8559 	struct mlx5_flow_tunnel *tunnel;
8560 	uint32_t id;
8561 
8562 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8563 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
8564 	if (!tunnel)
8565 		return NULL;
8566 	if (id >= MLX5_MAX_TUNNELS) {
8567 		mlx5_ipool_free(ipool, id);
8568 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8569 		return NULL;
8570 	}
8571 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8572 					   mlx5_flow_tunnel_grp2tbl_create_cb,
8573 					   mlx5_flow_tunnel_grp2tbl_match_cb,
8574 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
8575 	if (!tunnel->groups) {
8576 		mlx5_ipool_free(ipool, id);
8577 		return NULL;
8578 	}
8579 	tunnel->groups->ctx = priv->sh;
8580 	/* initiate new PMD tunnel */
8581 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8582 	tunnel->tunnel_id = id;
8583 	tunnel->action.type = (typeof(tunnel->action.type))
8584 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8585 	tunnel->action.conf = tunnel;
8586 	tunnel->item.type = (typeof(tunnel->item.type))
8587 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8588 	tunnel->item.spec = tunnel;
8589 	tunnel->item.last = NULL;
8590 	tunnel->item.mask = NULL;
8591 
8592 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8593 		dev->data->port_id, tunnel->tunnel_id);
8594 
8595 	return tunnel;
8596 }
8597 
8598 struct tunnel_db_get_tunnel_ctx {
8599 	const struct rte_flow_tunnel *app_tunnel;
8600 	struct mlx5_flow_tunnel *tunnel;
8601 };
8602 
8603 static bool get_tunnel_match(struct rte_eth_dev *dev,
8604 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8605 {
8606 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
8607 
8608 	RTE_SET_USED(dev);
8609 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8610 		       sizeof(*ctx->app_tunnel));
8611 }
8612 
8613 static void get_tunnel_hit(struct rte_eth_dev *dev,
8614 			   struct mlx5_flow_tunnel *tunnel, void *x)
8615 {
8616 	/* called under tunnel spinlock protection */
8617 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8618 
8619 	RTE_SET_USED(dev);
8620 	tunnel->refctn++;
8621 	ctx->tunnel = tunnel;
8622 }
8623 
8624 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8625 {
8626 	/* called under tunnel spinlock protection */
8627 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8628 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8629 
8630 	rte_spinlock_unlock(&thub->sl);
8631 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8632 	rte_spinlock_lock(&thub->sl);
8633 	if (ctx->tunnel) {
8634 		ctx->tunnel->refctn = 1;
8635 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8636 	}
8637 }
8638 
8639 
8640 static int
8641 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8642 		     const struct rte_flow_tunnel *app_tunnel,
8643 		     struct mlx5_flow_tunnel **tunnel)
8644 {
8645 	struct tunnel_db_get_tunnel_ctx ctx = {
8646 		.app_tunnel = app_tunnel,
8647 	};
8648 
8649 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8650 				      get_tunnel_miss, &ctx, true);
8651 	*tunnel = ctx.tunnel;
8652 	return ctx.tunnel ? 0 : -ENOMEM;
8653 }
8654 
8655 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8656 {
8657 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8658 
8659 	if (!thub)
8660 		return;
8661 	if (!LIST_EMPTY(&thub->tunnels))
8662 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
8663 	mlx5_hlist_destroy(thub->groups);
8664 	mlx5_free(thub);
8665 }
8666 
8667 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8668 {
8669 	int err;
8670 	struct mlx5_flow_tunnel_hub *thub;
8671 
8672 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8673 			   0, SOCKET_ID_ANY);
8674 	if (!thub)
8675 		return -ENOMEM;
8676 	LIST_INIT(&thub->tunnels);
8677 	rte_spinlock_init(&thub->sl);
8678 	thub->groups = mlx5_hlist_create("flow groups",
8679 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
8680 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8681 					 mlx5_flow_tunnel_grp2tbl_match_cb,
8682 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
8683 	if (!thub->groups) {
8684 		err = -rte_errno;
8685 		goto err;
8686 	}
8687 	thub->groups->ctx = sh;
8688 	sh->tunnel_hub = thub;
8689 
8690 	return 0;
8691 
8692 err:
8693 	if (thub->groups)
8694 		mlx5_hlist_destroy(thub->groups);
8695 	if (thub)
8696 		mlx5_free(thub);
8697 	return err;
8698 }
8699 
8700 static inline bool
8701 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8702 			  struct rte_flow_tunnel *tunnel,
8703 			  const char *err_msg)
8704 {
8705 	err_msg = NULL;
8706 	if (!is_tunnel_offload_active(dev)) {
8707 		err_msg = "tunnel offload was not activated";
8708 		goto out;
8709 	} else if (!tunnel) {
8710 		err_msg = "no application tunnel";
8711 		goto out;
8712 	}
8713 
8714 	switch (tunnel->type) {
8715 	default:
8716 		err_msg = "unsupported tunnel type";
8717 		goto out;
8718 	case RTE_FLOW_ITEM_TYPE_VXLAN:
8719 		break;
8720 	}
8721 
8722 out:
8723 	return !err_msg;
8724 }
8725 
8726 static int
8727 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8728 		    struct rte_flow_tunnel *app_tunnel,
8729 		    struct rte_flow_action **actions,
8730 		    uint32_t *num_of_actions,
8731 		    struct rte_flow_error *error)
8732 {
8733 	int ret;
8734 	struct mlx5_flow_tunnel *tunnel;
8735 	const char *err_msg = NULL;
8736 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8737 
8738 	if (!verdict)
8739 		return rte_flow_error_set(error, EINVAL,
8740 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8741 					  err_msg);
8742 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8743 	if (ret < 0) {
8744 		return rte_flow_error_set(error, ret,
8745 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8746 					  "failed to initialize pmd tunnel");
8747 	}
8748 	*actions = &tunnel->action;
8749 	*num_of_actions = 1;
8750 	return 0;
8751 }
8752 
8753 static int
8754 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8755 		       struct rte_flow_tunnel *app_tunnel,
8756 		       struct rte_flow_item **items,
8757 		       uint32_t *num_of_items,
8758 		       struct rte_flow_error *error)
8759 {
8760 	int ret;
8761 	struct mlx5_flow_tunnel *tunnel;
8762 	const char *err_msg = NULL;
8763 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8764 
8765 	if (!verdict)
8766 		return rte_flow_error_set(error, EINVAL,
8767 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8768 					  err_msg);
8769 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8770 	if (ret < 0) {
8771 		return rte_flow_error_set(error, ret,
8772 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8773 					  "failed to initialize pmd tunnel");
8774 	}
8775 	*items = &tunnel->item;
8776 	*num_of_items = 1;
8777 	return 0;
8778 }
8779 
8780 struct tunnel_db_element_release_ctx {
8781 	struct rte_flow_item *items;
8782 	struct rte_flow_action *actions;
8783 	uint32_t num_elements;
8784 	struct rte_flow_error *error;
8785 	int ret;
8786 };
8787 
8788 static bool
8789 tunnel_element_release_match(struct rte_eth_dev *dev,
8790 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8791 {
8792 	const struct tunnel_db_element_release_ctx *ctx = x;
8793 
8794 	RTE_SET_USED(dev);
8795 	if (ctx->num_elements != 1)
8796 		return false;
8797 	else if (ctx->items)
8798 		return ctx->items == &tunnel->item;
8799 	else if (ctx->actions)
8800 		return ctx->actions == &tunnel->action;
8801 
8802 	return false;
8803 }
8804 
8805 static void
8806 tunnel_element_release_hit(struct rte_eth_dev *dev,
8807 			   struct mlx5_flow_tunnel *tunnel, void *x)
8808 {
8809 	struct tunnel_db_element_release_ctx *ctx = x;
8810 	ctx->ret = 0;
8811 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8812 		mlx5_flow_tunnel_free(dev, tunnel);
8813 }
8814 
8815 static void
8816 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8817 {
8818 	struct tunnel_db_element_release_ctx *ctx = x;
8819 	RTE_SET_USED(dev);
8820 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8821 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8822 				      "invalid argument");
8823 }
8824 
8825 static int
8826 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8827 		       struct rte_flow_item *pmd_items,
8828 		       uint32_t num_items, struct rte_flow_error *err)
8829 {
8830 	struct tunnel_db_element_release_ctx ctx = {
8831 		.items = pmd_items,
8832 		.actions = NULL,
8833 		.num_elements = num_items,
8834 		.error = err,
8835 	};
8836 
8837 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8838 				      tunnel_element_release_hit,
8839 				      tunnel_element_release_miss, &ctx, false);
8840 
8841 	return ctx.ret;
8842 }
8843 
8844 static int
8845 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8846 			 struct rte_flow_action *pmd_actions,
8847 			 uint32_t num_actions, struct rte_flow_error *err)
8848 {
8849 	struct tunnel_db_element_release_ctx ctx = {
8850 		.items = NULL,
8851 		.actions = pmd_actions,
8852 		.num_elements = num_actions,
8853 		.error = err,
8854 	};
8855 
8856 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8857 				      tunnel_element_release_hit,
8858 				      tunnel_element_release_miss, &ctx, false);
8859 
8860 	return ctx.ret;
8861 }
8862 
8863 static int
8864 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8865 				  struct rte_mbuf *m,
8866 				  struct rte_flow_restore_info *info,
8867 				  struct rte_flow_error *err)
8868 {
8869 	uint64_t ol_flags = m->ol_flags;
8870 	const struct mlx5_flow_tbl_data_entry *tble;
8871 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8872 
8873 	if (!is_tunnel_offload_active(dev)) {
8874 		info->flags = 0;
8875 		return 0;
8876 	}
8877 
8878 	if ((ol_flags & mask) != mask)
8879 		goto err;
8880 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8881 	if (!tble) {
8882 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8883 			dev->data->port_id, m->hash.fdir.hi);
8884 		goto err;
8885 	}
8886 	MLX5_ASSERT(tble->tunnel);
8887 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8888 	info->group_id = tble->group_id;
8889 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8890 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8891 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8892 
8893 	return 0;
8894 
8895 err:
8896 	return rte_flow_error_set(err, EINVAL,
8897 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8898 				  "failed to get restore info");
8899 }
8900 
8901 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8902 static int
8903 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8904 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8905 			   __rte_unused struct rte_flow_action **actions,
8906 			   __rte_unused uint32_t *num_of_actions,
8907 			   __rte_unused struct rte_flow_error *error)
8908 {
8909 	return -ENOTSUP;
8910 }
8911 
8912 static int
8913 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8914 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8915 		       __rte_unused struct rte_flow_item **items,
8916 		       __rte_unused uint32_t *num_of_items,
8917 		       __rte_unused struct rte_flow_error *error)
8918 {
8919 	return -ENOTSUP;
8920 }
8921 
8922 static int
8923 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8924 			      __rte_unused struct rte_flow_item *pmd_items,
8925 			      __rte_unused uint32_t num_items,
8926 			      __rte_unused struct rte_flow_error *err)
8927 {
8928 	return -ENOTSUP;
8929 }
8930 
8931 static int
8932 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8933 				__rte_unused struct rte_flow_action *pmd_action,
8934 				__rte_unused uint32_t num_actions,
8935 				__rte_unused struct rte_flow_error *err)
8936 {
8937 	return -ENOTSUP;
8938 }
8939 
8940 static int
8941 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8942 				  __rte_unused struct rte_mbuf *m,
8943 				  __rte_unused struct rte_flow_restore_info *i,
8944 				  __rte_unused struct rte_flow_error *err)
8945 {
8946 	return -ENOTSUP;
8947 }
8948 
8949 static int
8950 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8951 			     __rte_unused struct rte_flow *flow,
8952 			     __rte_unused const struct rte_flow_attr *attr,
8953 			     __rte_unused const struct rte_flow_action *actions,
8954 			     __rte_unused uint32_t flow_idx,
8955 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
8956 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8957 			     __rte_unused struct rte_flow_error *error)
8958 {
8959 	return -ENOTSUP;
8960 }
8961 
8962 static struct mlx5_flow_tunnel *
8963 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8964 		    __rte_unused uint32_t id)
8965 {
8966 	return NULL;
8967 }
8968 
8969 static void
8970 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8971 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8972 {
8973 }
8974 
8975 static uint32_t
8976 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8977 				__rte_unused const struct mlx5_flow_tunnel *t,
8978 				__rte_unused uint32_t group,
8979 				__rte_unused uint32_t *table,
8980 				struct rte_flow_error *error)
8981 {
8982 	return rte_flow_error_set(error, ENOTSUP,
8983 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8984 				  "tunnel offload requires DV support");
8985 }
8986 
8987 void
8988 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8989 			__rte_unused  uint16_t port_id)
8990 {
8991 }
8992 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8993