xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision ae2927cd26e93077bc827f95b3e436e1aa63b1ba)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <ethdev_driver.h>
16 #include <rte_eal_paging.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
21 #include <rte_ip.h>
22 
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
25 #include <mlx5_prm.h>
26 #include <mlx5_malloc.h>
27 
28 #include "mlx5_defs.h"
29 #include "mlx5.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.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 			     struct tunnel_default_miss_ctx *ctx,
54 			     struct rte_flow_error *error);
55 static struct mlx5_flow_tunnel *
56 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
57 static void
58 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
59 static uint32_t
60 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
61 				const struct mlx5_flow_tunnel *tunnel,
62 				uint32_t group, uint32_t *table,
63 				struct rte_flow_error *error);
64 
65 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
66 static void mlx5_flow_pop_thread_workspace(void);
67 
68 
69 /** Device flow drivers. */
70 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
71 
72 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
73 
74 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
75 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
76 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
77 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
78 #endif
79 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
80 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
81 };
82 
83 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
84 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
85 	(const int []){ \
86 		__VA_ARGS__, 0, \
87 	}
88 
89 /** Node object of input graph for mlx5_flow_expand_rss(). */
90 struct mlx5_flow_expand_node {
91 	const int *const next;
92 	/**<
93 	 * List of next node indexes. Index 0 is interpreted as a terminator.
94 	 */
95 	const enum rte_flow_item_type type;
96 	/**< Pattern item type of current node. */
97 	uint64_t rss_types;
98 	/**<
99 	 * RSS types bit-field associated with this node
100 	 * (see ETH_RSS_* definitions).
101 	 */
102 };
103 
104 /** Object returned by mlx5_flow_expand_rss(). */
105 struct mlx5_flow_expand_rss {
106 	uint32_t entries;
107 	/**< Number of entries @p patterns and @p priorities. */
108 	struct {
109 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
110 		uint32_t priority; /**< Priority offset for each expansion. */
111 	} entry[];
112 };
113 
114 static enum rte_flow_item_type
115 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
116 {
117 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
118 	uint16_t ether_type = 0;
119 	uint16_t ether_type_m;
120 	uint8_t ip_next_proto = 0;
121 	uint8_t ip_next_proto_m;
122 
123 	if (item == NULL || item->spec == NULL)
124 		return ret;
125 	switch (item->type) {
126 	case RTE_FLOW_ITEM_TYPE_ETH:
127 		if (item->mask)
128 			ether_type_m = ((const struct rte_flow_item_eth *)
129 						(item->mask))->type;
130 		else
131 			ether_type_m = rte_flow_item_eth_mask.type;
132 		if (ether_type_m != RTE_BE16(0xFFFF))
133 			break;
134 		ether_type = ((const struct rte_flow_item_eth *)
135 				(item->spec))->type;
136 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
139 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
140 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
141 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
142 		else
143 			ret = RTE_FLOW_ITEM_TYPE_END;
144 		break;
145 	case RTE_FLOW_ITEM_TYPE_VLAN:
146 		if (item->mask)
147 			ether_type_m = ((const struct rte_flow_item_vlan *)
148 						(item->mask))->inner_type;
149 		else
150 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
151 		if (ether_type_m != RTE_BE16(0xFFFF))
152 			break;
153 		ether_type = ((const struct rte_flow_item_vlan *)
154 				(item->spec))->inner_type;
155 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
156 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
157 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
158 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
159 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
160 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
161 		else
162 			ret = RTE_FLOW_ITEM_TYPE_END;
163 		break;
164 	case RTE_FLOW_ITEM_TYPE_IPV4:
165 		if (item->mask)
166 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
167 					(item->mask))->hdr.next_proto_id;
168 		else
169 			ip_next_proto_m =
170 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
171 		if (ip_next_proto_m != 0xFF)
172 			break;
173 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
174 				(item->spec))->hdr.next_proto_id;
175 		if (ip_next_proto == IPPROTO_UDP)
176 			ret = RTE_FLOW_ITEM_TYPE_UDP;
177 		else if (ip_next_proto == IPPROTO_TCP)
178 			ret = RTE_FLOW_ITEM_TYPE_TCP;
179 		else if (ip_next_proto == IPPROTO_IP)
180 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
181 		else if (ip_next_proto == IPPROTO_IPV6)
182 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
183 		else
184 			ret = RTE_FLOW_ITEM_TYPE_END;
185 		break;
186 	case RTE_FLOW_ITEM_TYPE_IPV6:
187 		if (item->mask)
188 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
189 						(item->mask))->hdr.proto;
190 		else
191 			ip_next_proto_m =
192 				rte_flow_item_ipv6_mask.hdr.proto;
193 		if (ip_next_proto_m != 0xFF)
194 			break;
195 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
196 				(item->spec))->hdr.proto;
197 		if (ip_next_proto == IPPROTO_UDP)
198 			ret = RTE_FLOW_ITEM_TYPE_UDP;
199 		else if (ip_next_proto == IPPROTO_TCP)
200 			ret = RTE_FLOW_ITEM_TYPE_TCP;
201 		else if (ip_next_proto == IPPROTO_IP)
202 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
203 		else if (ip_next_proto == IPPROTO_IPV6)
204 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
205 		else
206 			ret = RTE_FLOW_ITEM_TYPE_END;
207 		break;
208 	default:
209 		ret = RTE_FLOW_ITEM_TYPE_VOID;
210 		break;
211 	}
212 	return ret;
213 }
214 
215 #define MLX5_RSS_EXP_ELT_N 8
216 
217 /**
218  * Expand RSS flows into several possible flows according to the RSS hash
219  * fields requested and the driver capabilities.
220  *
221  * @param[out] buf
222  *   Buffer to store the result expansion.
223  * @param[in] size
224  *   Buffer size in bytes. If 0, @p buf can be NULL.
225  * @param[in] pattern
226  *   User flow pattern.
227  * @param[in] types
228  *   RSS types to expand (see ETH_RSS_* definitions).
229  * @param[in] graph
230  *   Input graph to expand @p pattern according to @p types.
231  * @param[in] graph_root_index
232  *   Index of root node in @p graph, typically 0.
233  *
234  * @return
235  *   A positive value representing the size of @p buf in bytes regardless of
236  *   @p size on success, a negative errno value otherwise and rte_errno is
237  *   set, the following errors are defined:
238  *
239  *   -E2BIG: graph-depth @p graph is too deep.
240  */
241 static int
242 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
243 		     const struct rte_flow_item *pattern, uint64_t types,
244 		     const struct mlx5_flow_expand_node graph[],
245 		     int graph_root_index)
246 {
247 	const struct rte_flow_item *item;
248 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
249 	const int *next_node;
250 	const int *stack[MLX5_RSS_EXP_ELT_N];
251 	int stack_pos = 0;
252 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
253 	unsigned int i;
254 	size_t lsize;
255 	size_t user_pattern_size = 0;
256 	void *addr = NULL;
257 	const struct mlx5_flow_expand_node *next = NULL;
258 	struct rte_flow_item missed_item;
259 	int missed = 0;
260 	int elt = 0;
261 	const struct rte_flow_item *last_item = NULL;
262 
263 	memset(&missed_item, 0, sizeof(missed_item));
264 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
265 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
266 	if (lsize <= size) {
267 		buf->entry[0].priority = 0;
268 		buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
269 		buf->entries = 0;
270 		addr = buf->entry[0].pattern;
271 	}
272 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
273 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
274 			last_item = item;
275 		for (i = 0; node->next && node->next[i]; ++i) {
276 			next = &graph[node->next[i]];
277 			if (next->type == item->type)
278 				break;
279 		}
280 		if (next)
281 			node = next;
282 		user_pattern_size += sizeof(*item);
283 	}
284 	user_pattern_size += sizeof(*item); /* Handle END item. */
285 	lsize += user_pattern_size;
286 	/* Copy the user pattern in the first entry of the buffer. */
287 	if (lsize <= size) {
288 		rte_memcpy(addr, pattern, user_pattern_size);
289 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
290 		buf->entries = 1;
291 	}
292 	/* Start expanding. */
293 	memset(flow_items, 0, sizeof(flow_items));
294 	user_pattern_size -= sizeof(*item);
295 	/*
296 	 * Check if the last valid item has spec set, need complete pattern,
297 	 * and the pattern can be used for expansion.
298 	 */
299 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
300 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
301 		/* Item type END indicates expansion is not required. */
302 		return lsize;
303 	}
304 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
305 		next = NULL;
306 		missed = 1;
307 		for (i = 0; node->next && node->next[i]; ++i) {
308 			next = &graph[node->next[i]];
309 			if (next->type == missed_item.type) {
310 				flow_items[0].type = missed_item.type;
311 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
312 				break;
313 			}
314 			next = NULL;
315 		}
316 	}
317 	if (next && missed) {
318 		elt = 2; /* missed item + item end. */
319 		node = next;
320 		lsize += elt * sizeof(*item) + user_pattern_size;
321 		if ((node->rss_types & types) && lsize <= size) {
322 			buf->entry[buf->entries].priority = 1;
323 			buf->entry[buf->entries].pattern = addr;
324 			buf->entries++;
325 			rte_memcpy(addr, buf->entry[0].pattern,
326 				   user_pattern_size);
327 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
328 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
329 			addr = (void *)(((uintptr_t)addr) +
330 					elt * sizeof(*item));
331 		}
332 	}
333 	memset(flow_items, 0, sizeof(flow_items));
334 	next_node = node->next;
335 	stack[stack_pos] = next_node;
336 	node = next_node ? &graph[*next_node] : NULL;
337 	while (node) {
338 		flow_items[stack_pos].type = node->type;
339 		if (node->rss_types & types) {
340 			/*
341 			 * compute the number of items to copy from the
342 			 * expansion and copy it.
343 			 * When the stack_pos is 0, there are 1 element in it,
344 			 * plus the addition END item.
345 			 */
346 			elt = stack_pos + 2;
347 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
348 			lsize += elt * sizeof(*item) + user_pattern_size;
349 			if (lsize <= size) {
350 				size_t n = elt * sizeof(*item);
351 
352 				buf->entry[buf->entries].priority =
353 					stack_pos + 1 + missed;
354 				buf->entry[buf->entries].pattern = addr;
355 				buf->entries++;
356 				rte_memcpy(addr, buf->entry[0].pattern,
357 					   user_pattern_size);
358 				addr = (void *)(((uintptr_t)addr) +
359 						user_pattern_size);
360 				rte_memcpy(addr, &missed_item,
361 					   missed * sizeof(*item));
362 				addr = (void *)(((uintptr_t)addr) +
363 					missed * sizeof(*item));
364 				rte_memcpy(addr, flow_items, n);
365 				addr = (void *)(((uintptr_t)addr) + n);
366 			}
367 		}
368 		/* Go deeper. */
369 		if (node->next) {
370 			next_node = node->next;
371 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
372 				rte_errno = E2BIG;
373 				return -rte_errno;
374 			}
375 			stack[stack_pos] = next_node;
376 		} else if (*(next_node + 1)) {
377 			/* Follow up with the next possibility. */
378 			++next_node;
379 		} else {
380 			/* Move to the next path. */
381 			if (stack_pos)
382 				next_node = stack[--stack_pos];
383 			next_node++;
384 			stack[stack_pos] = next_node;
385 		}
386 		node = *next_node ? &graph[*next_node] : NULL;
387 	};
388 	/* no expanded flows but we have missed item, create one rule for it */
389 	if (buf->entries == 1 && missed != 0) {
390 		elt = 2;
391 		lsize += elt * sizeof(*item) + user_pattern_size;
392 		if (lsize <= size) {
393 			buf->entry[buf->entries].priority = 1;
394 			buf->entry[buf->entries].pattern = addr;
395 			buf->entries++;
396 			flow_items[0].type = missed_item.type;
397 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
398 			rte_memcpy(addr, buf->entry[0].pattern,
399 				   user_pattern_size);
400 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
401 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
402 		}
403 	}
404 	return lsize;
405 }
406 
407 enum mlx5_expansion {
408 	MLX5_EXPANSION_ROOT,
409 	MLX5_EXPANSION_ROOT_OUTER,
410 	MLX5_EXPANSION_ROOT_ETH_VLAN,
411 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
412 	MLX5_EXPANSION_OUTER_ETH,
413 	MLX5_EXPANSION_OUTER_ETH_VLAN,
414 	MLX5_EXPANSION_OUTER_VLAN,
415 	MLX5_EXPANSION_OUTER_IPV4,
416 	MLX5_EXPANSION_OUTER_IPV4_UDP,
417 	MLX5_EXPANSION_OUTER_IPV4_TCP,
418 	MLX5_EXPANSION_OUTER_IPV6,
419 	MLX5_EXPANSION_OUTER_IPV6_UDP,
420 	MLX5_EXPANSION_OUTER_IPV6_TCP,
421 	MLX5_EXPANSION_VXLAN,
422 	MLX5_EXPANSION_VXLAN_GPE,
423 	MLX5_EXPANSION_GRE,
424 	MLX5_EXPANSION_MPLS,
425 	MLX5_EXPANSION_ETH,
426 	MLX5_EXPANSION_ETH_VLAN,
427 	MLX5_EXPANSION_VLAN,
428 	MLX5_EXPANSION_IPV4,
429 	MLX5_EXPANSION_IPV4_UDP,
430 	MLX5_EXPANSION_IPV4_TCP,
431 	MLX5_EXPANSION_IPV6,
432 	MLX5_EXPANSION_IPV6_UDP,
433 	MLX5_EXPANSION_IPV6_TCP,
434 };
435 
436 /** Supported expansion of items. */
437 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
438 	[MLX5_EXPANSION_ROOT] = {
439 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
440 						  MLX5_EXPANSION_IPV4,
441 						  MLX5_EXPANSION_IPV6),
442 		.type = RTE_FLOW_ITEM_TYPE_END,
443 	},
444 	[MLX5_EXPANSION_ROOT_OUTER] = {
445 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
446 						  MLX5_EXPANSION_OUTER_IPV4,
447 						  MLX5_EXPANSION_OUTER_IPV6),
448 		.type = RTE_FLOW_ITEM_TYPE_END,
449 	},
450 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
451 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
452 		.type = RTE_FLOW_ITEM_TYPE_END,
453 	},
454 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
455 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
456 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
457 		.type = RTE_FLOW_ITEM_TYPE_END,
458 	},
459 	[MLX5_EXPANSION_OUTER_ETH] = {
460 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
461 						  MLX5_EXPANSION_OUTER_IPV6,
462 						  MLX5_EXPANSION_MPLS),
463 		.type = RTE_FLOW_ITEM_TYPE_ETH,
464 		.rss_types = 0,
465 	},
466 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
467 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
468 		.type = RTE_FLOW_ITEM_TYPE_ETH,
469 		.rss_types = 0,
470 	},
471 	[MLX5_EXPANSION_OUTER_VLAN] = {
472 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
473 						  MLX5_EXPANSION_OUTER_IPV6),
474 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
475 	},
476 	[MLX5_EXPANSION_OUTER_IPV4] = {
477 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
478 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
479 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
480 			 MLX5_EXPANSION_GRE,
481 			 MLX5_EXPANSION_IPV4,
482 			 MLX5_EXPANSION_IPV6),
483 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
484 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
485 			ETH_RSS_NONFRAG_IPV4_OTHER,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 						  MLX5_EXPANSION_VXLAN_GPE),
490 		.type = RTE_FLOW_ITEM_TYPE_UDP,
491 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
492 	},
493 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
494 		.type = RTE_FLOW_ITEM_TYPE_TCP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
496 	},
497 	[MLX5_EXPANSION_OUTER_IPV6] = {
498 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
499 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
500 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
501 			 MLX5_EXPANSION_IPV4,
502 			 MLX5_EXPANSION_IPV6),
503 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
504 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
505 			ETH_RSS_NONFRAG_IPV6_OTHER,
506 	},
507 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
508 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 						  MLX5_EXPANSION_VXLAN_GPE),
510 		.type = RTE_FLOW_ITEM_TYPE_UDP,
511 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
512 	},
513 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
514 		.type = RTE_FLOW_ITEM_TYPE_TCP,
515 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
516 	},
517 	[MLX5_EXPANSION_VXLAN] = {
518 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
519 						  MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
522 	},
523 	[MLX5_EXPANSION_VXLAN_GPE] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
525 						  MLX5_EXPANSION_IPV4,
526 						  MLX5_EXPANSION_IPV6),
527 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
528 	},
529 	[MLX5_EXPANSION_GRE] = {
530 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
531 		.type = RTE_FLOW_ITEM_TYPE_GRE,
532 	},
533 	[MLX5_EXPANSION_MPLS] = {
534 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 						  MLX5_EXPANSION_IPV6),
536 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
537 	},
538 	[MLX5_EXPANSION_ETH] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
540 						  MLX5_EXPANSION_IPV6),
541 		.type = RTE_FLOW_ITEM_TYPE_ETH,
542 	},
543 	[MLX5_EXPANSION_ETH_VLAN] = {
544 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
545 		.type = RTE_FLOW_ITEM_TYPE_ETH,
546 	},
547 	[MLX5_EXPANSION_VLAN] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
549 						  MLX5_EXPANSION_IPV6),
550 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
551 	},
552 	[MLX5_EXPANSION_IPV4] = {
553 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
554 						  MLX5_EXPANSION_IPV4_TCP),
555 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
556 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
557 			ETH_RSS_NONFRAG_IPV4_OTHER,
558 	},
559 	[MLX5_EXPANSION_IPV4_UDP] = {
560 		.type = RTE_FLOW_ITEM_TYPE_UDP,
561 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
562 	},
563 	[MLX5_EXPANSION_IPV4_TCP] = {
564 		.type = RTE_FLOW_ITEM_TYPE_TCP,
565 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
566 	},
567 	[MLX5_EXPANSION_IPV6] = {
568 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
569 						  MLX5_EXPANSION_IPV6_TCP),
570 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
571 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
572 			ETH_RSS_NONFRAG_IPV6_OTHER,
573 	},
574 	[MLX5_EXPANSION_IPV6_UDP] = {
575 		.type = RTE_FLOW_ITEM_TYPE_UDP,
576 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
577 	},
578 	[MLX5_EXPANSION_IPV6_TCP] = {
579 		.type = RTE_FLOW_ITEM_TYPE_TCP,
580 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
581 	},
582 };
583 
584 static struct rte_flow_shared_action *
585 mlx5_shared_action_create(struct rte_eth_dev *dev,
586 			  const struct rte_flow_shared_action_conf *conf,
587 			  const struct rte_flow_action *action,
588 			  struct rte_flow_error *error);
589 static int mlx5_shared_action_destroy
590 				(struct rte_eth_dev *dev,
591 				 struct rte_flow_shared_action *shared_action,
592 				 struct rte_flow_error *error);
593 static int mlx5_shared_action_update
594 				(struct rte_eth_dev *dev,
595 				 struct rte_flow_shared_action *shared_action,
596 				 const struct rte_flow_action *action,
597 				 struct rte_flow_error *error);
598 static int mlx5_shared_action_query
599 				(struct rte_eth_dev *dev,
600 				 const struct rte_flow_shared_action *action,
601 				 void *data,
602 				 struct rte_flow_error *error);
603 static int
604 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
605 		    struct rte_flow_tunnel *app_tunnel,
606 		    struct rte_flow_action **actions,
607 		    uint32_t *num_of_actions,
608 		    struct rte_flow_error *error);
609 static int
610 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
611 		       struct rte_flow_tunnel *app_tunnel,
612 		       struct rte_flow_item **items,
613 		       uint32_t *num_of_items,
614 		       struct rte_flow_error *error);
615 static int
616 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
617 			      struct rte_flow_item *pmd_items,
618 			      uint32_t num_items, struct rte_flow_error *err);
619 static int
620 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
621 				struct rte_flow_action *pmd_actions,
622 				uint32_t num_actions,
623 				struct rte_flow_error *err);
624 static int
625 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
626 				  struct rte_mbuf *m,
627 				  struct rte_flow_restore_info *info,
628 				  struct rte_flow_error *err);
629 
630 static const struct rte_flow_ops mlx5_flow_ops = {
631 	.validate = mlx5_flow_validate,
632 	.create = mlx5_flow_create,
633 	.destroy = mlx5_flow_destroy,
634 	.flush = mlx5_flow_flush,
635 	.isolate = mlx5_flow_isolate,
636 	.query = mlx5_flow_query,
637 	.dev_dump = mlx5_flow_dev_dump,
638 	.get_aged_flows = mlx5_flow_get_aged_flows,
639 	.shared_action_create = mlx5_shared_action_create,
640 	.shared_action_destroy = mlx5_shared_action_destroy,
641 	.shared_action_update = mlx5_shared_action_update,
642 	.shared_action_query = mlx5_shared_action_query,
643 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
644 	.tunnel_match = mlx5_flow_tunnel_match,
645 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
646 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
647 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
648 };
649 
650 /* Tunnel information. */
651 struct mlx5_flow_tunnel_info {
652 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
653 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
654 };
655 
656 static struct mlx5_flow_tunnel_info tunnels_info[] = {
657 	{
658 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
659 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
660 	},
661 	{
662 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
663 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
664 	},
665 	{
666 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
667 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
668 	},
669 	{
670 		.tunnel = MLX5_FLOW_LAYER_GRE,
671 		.ptype = RTE_PTYPE_TUNNEL_GRE,
672 	},
673 	{
674 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
675 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
676 	},
677 	{
678 		.tunnel = MLX5_FLOW_LAYER_MPLS,
679 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
680 	},
681 	{
682 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
683 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
684 	},
685 	{
686 		.tunnel = MLX5_FLOW_LAYER_IPIP,
687 		.ptype = RTE_PTYPE_TUNNEL_IP,
688 	},
689 	{
690 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
691 		.ptype = RTE_PTYPE_TUNNEL_IP,
692 	},
693 	{
694 		.tunnel = MLX5_FLOW_LAYER_GTP,
695 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
696 	},
697 };
698 
699 
700 
701 /**
702  * Translate tag ID to register.
703  *
704  * @param[in] dev
705  *   Pointer to the Ethernet device structure.
706  * @param[in] feature
707  *   The feature that request the register.
708  * @param[in] id
709  *   The request register ID.
710  * @param[out] error
711  *   Error description in case of any.
712  *
713  * @return
714  *   The request register on success, a negative errno
715  *   value otherwise and rte_errno is set.
716  */
717 int
718 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
719 		     enum mlx5_feature_name feature,
720 		     uint32_t id,
721 		     struct rte_flow_error *error)
722 {
723 	struct mlx5_priv *priv = dev->data->dev_private;
724 	struct mlx5_dev_config *config = &priv->config;
725 	enum modify_reg start_reg;
726 	bool skip_mtr_reg = false;
727 
728 	switch (feature) {
729 	case MLX5_HAIRPIN_RX:
730 		return REG_B;
731 	case MLX5_HAIRPIN_TX:
732 		return REG_A;
733 	case MLX5_METADATA_RX:
734 		switch (config->dv_xmeta_en) {
735 		case MLX5_XMETA_MODE_LEGACY:
736 			return REG_B;
737 		case MLX5_XMETA_MODE_META16:
738 			return REG_C_0;
739 		case MLX5_XMETA_MODE_META32:
740 			return REG_C_1;
741 		}
742 		break;
743 	case MLX5_METADATA_TX:
744 		return REG_A;
745 	case MLX5_METADATA_FDB:
746 		switch (config->dv_xmeta_en) {
747 		case MLX5_XMETA_MODE_LEGACY:
748 			return REG_NON;
749 		case MLX5_XMETA_MODE_META16:
750 			return REG_C_0;
751 		case MLX5_XMETA_MODE_META32:
752 			return REG_C_1;
753 		}
754 		break;
755 	case MLX5_FLOW_MARK:
756 		switch (config->dv_xmeta_en) {
757 		case MLX5_XMETA_MODE_LEGACY:
758 			return REG_NON;
759 		case MLX5_XMETA_MODE_META16:
760 			return REG_C_1;
761 		case MLX5_XMETA_MODE_META32:
762 			return REG_C_0;
763 		}
764 		break;
765 	case MLX5_MTR_SFX:
766 		/*
767 		 * If meter color and flow match share one register, flow match
768 		 * should use the meter color register for match.
769 		 */
770 		if (priv->mtr_reg_share)
771 			return priv->mtr_color_reg;
772 		else
773 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
774 			       REG_C_3;
775 	case MLX5_MTR_COLOR:
776 	case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
777 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
778 		return priv->mtr_color_reg;
779 	case MLX5_COPY_MARK:
780 		/*
781 		 * Metadata COPY_MARK register using is in meter suffix sub
782 		 * flow while with meter. It's safe to share the same register.
783 		 */
784 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
785 	case MLX5_APP_TAG:
786 		/*
787 		 * If meter is enable, it will engage the register for color
788 		 * match and flow match. If meter color match is not using the
789 		 * REG_C_2, need to skip the REG_C_x be used by meter color
790 		 * match.
791 		 * If meter is disable, free to use all available registers.
792 		 */
793 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
794 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
795 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
796 		if (id > (uint32_t)(REG_C_7 - start_reg))
797 			return rte_flow_error_set(error, EINVAL,
798 						  RTE_FLOW_ERROR_TYPE_ITEM,
799 						  NULL, "invalid tag id");
800 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
801 			return rte_flow_error_set(error, ENOTSUP,
802 						  RTE_FLOW_ERROR_TYPE_ITEM,
803 						  NULL, "unsupported tag id");
804 		/*
805 		 * This case means meter is using the REG_C_x great than 2.
806 		 * Take care not to conflict with meter color REG_C_x.
807 		 * If the available index REG_C_y >= REG_C_x, skip the
808 		 * color register.
809 		 */
810 		if (skip_mtr_reg && config->flow_mreg_c
811 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
812 			if (id >= (uint32_t)(REG_C_7 - start_reg))
813 				return rte_flow_error_set(error, EINVAL,
814 						       RTE_FLOW_ERROR_TYPE_ITEM,
815 							NULL, "invalid tag id");
816 			if (config->flow_mreg_c
817 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
818 				return config->flow_mreg_c
819 					       [id + 1 + start_reg - REG_C_0];
820 			return rte_flow_error_set(error, ENOTSUP,
821 						  RTE_FLOW_ERROR_TYPE_ITEM,
822 						  NULL, "unsupported tag id");
823 		}
824 		return config->flow_mreg_c[id + start_reg - REG_C_0];
825 	}
826 	MLX5_ASSERT(false);
827 	return rte_flow_error_set(error, EINVAL,
828 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
829 				  NULL, "invalid feature name");
830 }
831 
832 /**
833  * Check extensive flow metadata register support.
834  *
835  * @param dev
836  *   Pointer to rte_eth_dev structure.
837  *
838  * @return
839  *   True if device supports extensive flow metadata register, otherwise false.
840  */
841 bool
842 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
843 {
844 	struct mlx5_priv *priv = dev->data->dev_private;
845 	struct mlx5_dev_config *config = &priv->config;
846 
847 	/*
848 	 * Having available reg_c can be regarded inclusively as supporting
849 	 * extensive flow metadata register, which could mean,
850 	 * - metadata register copy action by modify header.
851 	 * - 16 modify header actions is supported.
852 	 * - reg_c's are preserved across different domain (FDB and NIC) on
853 	 *   packet loopback by flow lookup miss.
854 	 */
855 	return config->flow_mreg_c[2] != REG_NON;
856 }
857 
858 /**
859  * Verify the @p item specifications (spec, last, mask) are compatible with the
860  * NIC capabilities.
861  *
862  * @param[in] item
863  *   Item specification.
864  * @param[in] mask
865  *   @p item->mask or flow default bit-masks.
866  * @param[in] nic_mask
867  *   Bit-masks covering supported fields by the NIC to compare with user mask.
868  * @param[in] size
869  *   Bit-masks size in bytes.
870  * @param[in] range_accepted
871  *   True if range of values is accepted for specific fields, false otherwise.
872  * @param[out] error
873  *   Pointer to error structure.
874  *
875  * @return
876  *   0 on success, a negative errno value otherwise and rte_errno is set.
877  */
878 int
879 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
880 			  const uint8_t *mask,
881 			  const uint8_t *nic_mask,
882 			  unsigned int size,
883 			  bool range_accepted,
884 			  struct rte_flow_error *error)
885 {
886 	unsigned int i;
887 
888 	MLX5_ASSERT(nic_mask);
889 	for (i = 0; i < size; ++i)
890 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
891 			return rte_flow_error_set(error, ENOTSUP,
892 						  RTE_FLOW_ERROR_TYPE_ITEM,
893 						  item,
894 						  "mask enables non supported"
895 						  " bits");
896 	if (!item->spec && (item->mask || item->last))
897 		return rte_flow_error_set(error, EINVAL,
898 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
899 					  "mask/last without a spec is not"
900 					  " supported");
901 	if (item->spec && item->last && !range_accepted) {
902 		uint8_t spec[size];
903 		uint8_t last[size];
904 		unsigned int i;
905 		int ret;
906 
907 		for (i = 0; i < size; ++i) {
908 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
909 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
910 		}
911 		ret = memcmp(spec, last, size);
912 		if (ret != 0)
913 			return rte_flow_error_set(error, EINVAL,
914 						  RTE_FLOW_ERROR_TYPE_ITEM,
915 						  item,
916 						  "range is not valid");
917 	}
918 	return 0;
919 }
920 
921 /**
922  * Adjust the hash fields according to the @p flow information.
923  *
924  * @param[in] dev_flow.
925  *   Pointer to the mlx5_flow.
926  * @param[in] tunnel
927  *   1 when the hash field is for a tunnel item.
928  * @param[in] layer_types
929  *   ETH_RSS_* types.
930  * @param[in] hash_fields
931  *   Item hash fields.
932  *
933  * @return
934  *   The hash fields that should be used.
935  */
936 uint64_t
937 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
938 			    int tunnel __rte_unused, uint64_t layer_types,
939 			    uint64_t hash_fields)
940 {
941 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
942 	int rss_request_inner = rss_desc->level >= 2;
943 
944 	/* Check RSS hash level for tunnel. */
945 	if (tunnel && rss_request_inner)
946 		hash_fields |= IBV_RX_HASH_INNER;
947 	else if (tunnel || rss_request_inner)
948 		return 0;
949 #endif
950 	/* Check if requested layer matches RSS hash fields. */
951 	if (!(rss_desc->types & layer_types))
952 		return 0;
953 	return hash_fields;
954 }
955 
956 /**
957  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
958  * if several tunnel rules are used on this queue, the tunnel ptype will be
959  * cleared.
960  *
961  * @param rxq_ctrl
962  *   Rx queue to update.
963  */
964 static void
965 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
966 {
967 	unsigned int i;
968 	uint32_t tunnel_ptype = 0;
969 
970 	/* Look up for the ptype to use. */
971 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
972 		if (!rxq_ctrl->flow_tunnels_n[i])
973 			continue;
974 		if (!tunnel_ptype) {
975 			tunnel_ptype = tunnels_info[i].ptype;
976 		} else {
977 			tunnel_ptype = 0;
978 			break;
979 		}
980 	}
981 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
982 }
983 
984 /**
985  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
986  * flow.
987  *
988  * @param[in] dev
989  *   Pointer to the Ethernet device structure.
990  * @param[in] dev_handle
991  *   Pointer to device flow handle structure.
992  */
993 static void
994 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
995 		       struct mlx5_flow_handle *dev_handle)
996 {
997 	struct mlx5_priv *priv = dev->data->dev_private;
998 	const int mark = dev_handle->mark;
999 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1000 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1001 	unsigned int i;
1002 
1003 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1004 		struct mlx5_hrxq *hrxq;
1005 
1006 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1007 			      dev_handle->rix_hrxq);
1008 		if (hrxq)
1009 			ind_tbl = hrxq->ind_table;
1010 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1011 		struct mlx5_shared_action_rss *shared_rss;
1012 
1013 		shared_rss = mlx5_ipool_get
1014 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1015 			 dev_handle->rix_srss);
1016 		if (shared_rss)
1017 			ind_tbl = shared_rss->ind_tbl;
1018 	}
1019 	if (!ind_tbl)
1020 		return;
1021 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1022 		int idx = ind_tbl->queues[i];
1023 		struct mlx5_rxq_ctrl *rxq_ctrl =
1024 			container_of((*priv->rxqs)[idx],
1025 				     struct mlx5_rxq_ctrl, rxq);
1026 
1027 		/*
1028 		 * To support metadata register copy on Tx loopback,
1029 		 * this must be always enabled (metadata may arive
1030 		 * from other port - not from local flows only.
1031 		 */
1032 		if (priv->config.dv_flow_en &&
1033 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1034 		    mlx5_flow_ext_mreg_supported(dev)) {
1035 			rxq_ctrl->rxq.mark = 1;
1036 			rxq_ctrl->flow_mark_n = 1;
1037 		} else if (mark) {
1038 			rxq_ctrl->rxq.mark = 1;
1039 			rxq_ctrl->flow_mark_n++;
1040 		}
1041 		if (tunnel) {
1042 			unsigned int j;
1043 
1044 			/* Increase the counter matching the flow. */
1045 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1046 				if ((tunnels_info[j].tunnel &
1047 				     dev_handle->layers) ==
1048 				    tunnels_info[j].tunnel) {
1049 					rxq_ctrl->flow_tunnels_n[j]++;
1050 					break;
1051 				}
1052 			}
1053 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1054 		}
1055 	}
1056 }
1057 
1058 /**
1059  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1060  *
1061  * @param[in] dev
1062  *   Pointer to the Ethernet device structure.
1063  * @param[in] flow
1064  *   Pointer to flow structure.
1065  */
1066 static void
1067 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1068 {
1069 	struct mlx5_priv *priv = dev->data->dev_private;
1070 	uint32_t handle_idx;
1071 	struct mlx5_flow_handle *dev_handle;
1072 
1073 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1074 		       handle_idx, dev_handle, next)
1075 		flow_drv_rxq_flags_set(dev, dev_handle);
1076 }
1077 
1078 /**
1079  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1080  * device flow if no other flow uses it with the same kind of request.
1081  *
1082  * @param dev
1083  *   Pointer to Ethernet device.
1084  * @param[in] dev_handle
1085  *   Pointer to the device flow handle structure.
1086  */
1087 static void
1088 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1089 			struct mlx5_flow_handle *dev_handle)
1090 {
1091 	struct mlx5_priv *priv = dev->data->dev_private;
1092 	const int mark = dev_handle->mark;
1093 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1094 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1095 	unsigned int i;
1096 
1097 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1098 		struct mlx5_hrxq *hrxq;
1099 
1100 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1101 			      dev_handle->rix_hrxq);
1102 		if (hrxq)
1103 			ind_tbl = hrxq->ind_table;
1104 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1105 		struct mlx5_shared_action_rss *shared_rss;
1106 
1107 		shared_rss = mlx5_ipool_get
1108 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1109 			 dev_handle->rix_srss);
1110 		if (shared_rss)
1111 			ind_tbl = shared_rss->ind_tbl;
1112 	}
1113 	if (!ind_tbl)
1114 		return;
1115 	MLX5_ASSERT(dev->data->dev_started);
1116 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1117 		int idx = ind_tbl->queues[i];
1118 		struct mlx5_rxq_ctrl *rxq_ctrl =
1119 			container_of((*priv->rxqs)[idx],
1120 				     struct mlx5_rxq_ctrl, rxq);
1121 
1122 		if (priv->config.dv_flow_en &&
1123 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1124 		    mlx5_flow_ext_mreg_supported(dev)) {
1125 			rxq_ctrl->rxq.mark = 1;
1126 			rxq_ctrl->flow_mark_n = 1;
1127 		} else if (mark) {
1128 			rxq_ctrl->flow_mark_n--;
1129 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1130 		}
1131 		if (tunnel) {
1132 			unsigned int j;
1133 
1134 			/* Decrease the counter matching the flow. */
1135 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1136 				if ((tunnels_info[j].tunnel &
1137 				     dev_handle->layers) ==
1138 				    tunnels_info[j].tunnel) {
1139 					rxq_ctrl->flow_tunnels_n[j]--;
1140 					break;
1141 				}
1142 			}
1143 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1144 		}
1145 	}
1146 }
1147 
1148 /**
1149  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1150  * @p flow if no other flow uses it with the same kind of request.
1151  *
1152  * @param dev
1153  *   Pointer to Ethernet device.
1154  * @param[in] flow
1155  *   Pointer to the flow.
1156  */
1157 static void
1158 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1159 {
1160 	struct mlx5_priv *priv = dev->data->dev_private;
1161 	uint32_t handle_idx;
1162 	struct mlx5_flow_handle *dev_handle;
1163 
1164 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1165 		       handle_idx, dev_handle, next)
1166 		flow_drv_rxq_flags_trim(dev, dev_handle);
1167 }
1168 
1169 /**
1170  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1171  *
1172  * @param dev
1173  *   Pointer to Ethernet device.
1174  */
1175 static void
1176 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1177 {
1178 	struct mlx5_priv *priv = dev->data->dev_private;
1179 	unsigned int i;
1180 
1181 	for (i = 0; i != priv->rxqs_n; ++i) {
1182 		struct mlx5_rxq_ctrl *rxq_ctrl;
1183 		unsigned int j;
1184 
1185 		if (!(*priv->rxqs)[i])
1186 			continue;
1187 		rxq_ctrl = container_of((*priv->rxqs)[i],
1188 					struct mlx5_rxq_ctrl, rxq);
1189 		rxq_ctrl->flow_mark_n = 0;
1190 		rxq_ctrl->rxq.mark = 0;
1191 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1192 			rxq_ctrl->flow_tunnels_n[j] = 0;
1193 		rxq_ctrl->rxq.tunnel = 0;
1194 	}
1195 }
1196 
1197 /**
1198  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1199  *
1200  * @param[in] dev
1201  *   Pointer to the Ethernet device structure.
1202  */
1203 void
1204 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1205 {
1206 	struct mlx5_priv *priv = dev->data->dev_private;
1207 	struct mlx5_rxq_data *data;
1208 	unsigned int i;
1209 
1210 	for (i = 0; i != priv->rxqs_n; ++i) {
1211 		if (!(*priv->rxqs)[i])
1212 			continue;
1213 		data = (*priv->rxqs)[i];
1214 		if (!rte_flow_dynf_metadata_avail()) {
1215 			data->dynf_meta = 0;
1216 			data->flow_meta_mask = 0;
1217 			data->flow_meta_offset = -1;
1218 		} else {
1219 			data->dynf_meta = 1;
1220 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1221 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1222 		}
1223 	}
1224 }
1225 
1226 /*
1227  * return a pointer to the desired action in the list of actions.
1228  *
1229  * @param[in] actions
1230  *   The list of actions to search the action in.
1231  * @param[in] action
1232  *   The action to find.
1233  *
1234  * @return
1235  *   Pointer to the action in the list, if found. NULL otherwise.
1236  */
1237 const struct rte_flow_action *
1238 mlx5_flow_find_action(const struct rte_flow_action *actions,
1239 		      enum rte_flow_action_type action)
1240 {
1241 	if (actions == NULL)
1242 		return NULL;
1243 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1244 		if (actions->type == action)
1245 			return actions;
1246 	return NULL;
1247 }
1248 
1249 /*
1250  * Validate the flag action.
1251  *
1252  * @param[in] action_flags
1253  *   Bit-fields that holds the actions detected until now.
1254  * @param[in] attr
1255  *   Attributes of flow that includes this action.
1256  * @param[out] error
1257  *   Pointer to error structure.
1258  *
1259  * @return
1260  *   0 on success, a negative errno value otherwise and rte_errno is set.
1261  */
1262 int
1263 mlx5_flow_validate_action_flag(uint64_t action_flags,
1264 			       const struct rte_flow_attr *attr,
1265 			       struct rte_flow_error *error)
1266 {
1267 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1268 		return rte_flow_error_set(error, EINVAL,
1269 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1270 					  "can't mark and flag in same flow");
1271 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1272 		return rte_flow_error_set(error, EINVAL,
1273 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1274 					  "can't have 2 flag"
1275 					  " actions in same flow");
1276 	if (attr->egress)
1277 		return rte_flow_error_set(error, ENOTSUP,
1278 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1279 					  "flag action not supported for "
1280 					  "egress");
1281 	return 0;
1282 }
1283 
1284 /*
1285  * Validate the mark action.
1286  *
1287  * @param[in] action
1288  *   Pointer to the queue action.
1289  * @param[in] action_flags
1290  *   Bit-fields that holds the actions detected until now.
1291  * @param[in] attr
1292  *   Attributes of flow that includes this action.
1293  * @param[out] error
1294  *   Pointer to error structure.
1295  *
1296  * @return
1297  *   0 on success, a negative errno value otherwise and rte_errno is set.
1298  */
1299 int
1300 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1301 			       uint64_t action_flags,
1302 			       const struct rte_flow_attr *attr,
1303 			       struct rte_flow_error *error)
1304 {
1305 	const struct rte_flow_action_mark *mark = action->conf;
1306 
1307 	if (!mark)
1308 		return rte_flow_error_set(error, EINVAL,
1309 					  RTE_FLOW_ERROR_TYPE_ACTION,
1310 					  action,
1311 					  "configuration cannot be null");
1312 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1313 		return rte_flow_error_set(error, EINVAL,
1314 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1315 					  &mark->id,
1316 					  "mark id must in 0 <= id < "
1317 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1318 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1319 		return rte_flow_error_set(error, EINVAL,
1320 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1321 					  "can't flag and mark in same flow");
1322 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1323 		return rte_flow_error_set(error, EINVAL,
1324 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1325 					  "can't have 2 mark actions in same"
1326 					  " flow");
1327 	if (attr->egress)
1328 		return rte_flow_error_set(error, ENOTSUP,
1329 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1330 					  "mark action not supported for "
1331 					  "egress");
1332 	return 0;
1333 }
1334 
1335 /*
1336  * Validate the drop action.
1337  *
1338  * @param[in] action_flags
1339  *   Bit-fields that holds the actions detected until now.
1340  * @param[in] attr
1341  *   Attributes of flow that includes this action.
1342  * @param[out] error
1343  *   Pointer to error structure.
1344  *
1345  * @return
1346  *   0 on success, a negative errno value otherwise and rte_errno is set.
1347  */
1348 int
1349 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1350 			       const struct rte_flow_attr *attr,
1351 			       struct rte_flow_error *error)
1352 {
1353 	if (attr->egress)
1354 		return rte_flow_error_set(error, ENOTSUP,
1355 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1356 					  "drop action not supported for "
1357 					  "egress");
1358 	return 0;
1359 }
1360 
1361 /*
1362  * Validate the queue action.
1363  *
1364  * @param[in] action
1365  *   Pointer to the queue action.
1366  * @param[in] action_flags
1367  *   Bit-fields that holds the actions detected until now.
1368  * @param[in] dev
1369  *   Pointer to the Ethernet device structure.
1370  * @param[in] attr
1371  *   Attributes of flow that includes this action.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1380 				uint64_t action_flags,
1381 				struct rte_eth_dev *dev,
1382 				const struct rte_flow_attr *attr,
1383 				struct rte_flow_error *error)
1384 {
1385 	struct mlx5_priv *priv = dev->data->dev_private;
1386 	const struct rte_flow_action_queue *queue = action->conf;
1387 
1388 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1389 		return rte_flow_error_set(error, EINVAL,
1390 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1391 					  "can't have 2 fate actions in"
1392 					  " same flow");
1393 	if (!priv->rxqs_n)
1394 		return rte_flow_error_set(error, EINVAL,
1395 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1396 					  NULL, "No Rx queues configured");
1397 	if (queue->index >= priv->rxqs_n)
1398 		return rte_flow_error_set(error, EINVAL,
1399 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1400 					  &queue->index,
1401 					  "queue index out of range");
1402 	if (!(*priv->rxqs)[queue->index])
1403 		return rte_flow_error_set(error, EINVAL,
1404 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1405 					  &queue->index,
1406 					  "queue is not configured");
1407 	if (attr->egress)
1408 		return rte_flow_error_set(error, ENOTSUP,
1409 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1410 					  "queue action not supported for "
1411 					  "egress");
1412 	return 0;
1413 }
1414 
1415 /*
1416  * Validate the rss action.
1417  *
1418  * @param[in] dev
1419  *   Pointer to the Ethernet device structure.
1420  * @param[in] action
1421  *   Pointer to the queue action.
1422  * @param[out] error
1423  *   Pointer to error structure.
1424  *
1425  * @return
1426  *   0 on success, a negative errno value otherwise and rte_errno is set.
1427  */
1428 int
1429 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1430 			 const struct rte_flow_action *action,
1431 			 struct rte_flow_error *error)
1432 {
1433 	struct mlx5_priv *priv = dev->data->dev_private;
1434 	const struct rte_flow_action_rss *rss = action->conf;
1435 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1436 	unsigned int i;
1437 
1438 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1439 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1440 		return rte_flow_error_set(error, ENOTSUP,
1441 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1442 					  &rss->func,
1443 					  "RSS hash function not supported");
1444 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1445 	if (rss->level > 2)
1446 #else
1447 	if (rss->level > 1)
1448 #endif
1449 		return rte_flow_error_set(error, ENOTSUP,
1450 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1451 					  &rss->level,
1452 					  "tunnel RSS is not supported");
1453 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1454 	if (rss->key_len == 0 && rss->key != NULL)
1455 		return rte_flow_error_set(error, ENOTSUP,
1456 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1457 					  &rss->key_len,
1458 					  "RSS hash key length 0");
1459 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1460 		return rte_flow_error_set(error, ENOTSUP,
1461 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1462 					  &rss->key_len,
1463 					  "RSS hash key too small");
1464 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1465 		return rte_flow_error_set(error, ENOTSUP,
1466 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1467 					  &rss->key_len,
1468 					  "RSS hash key too large");
1469 	if (rss->queue_num > priv->config.ind_table_max_size)
1470 		return rte_flow_error_set(error, ENOTSUP,
1471 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1472 					  &rss->queue_num,
1473 					  "number of queues too large");
1474 	if (rss->types & MLX5_RSS_HF_MASK)
1475 		return rte_flow_error_set(error, ENOTSUP,
1476 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1477 					  &rss->types,
1478 					  "some RSS protocols are not"
1479 					  " supported");
1480 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1481 	    !(rss->types & ETH_RSS_IP))
1482 		return rte_flow_error_set(error, EINVAL,
1483 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1484 					  "L3 partial RSS requested but L3 RSS"
1485 					  " type not specified");
1486 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1487 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1488 		return rte_flow_error_set(error, EINVAL,
1489 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1490 					  "L4 partial RSS requested but L4 RSS"
1491 					  " type not specified");
1492 	if (!priv->rxqs_n)
1493 		return rte_flow_error_set(error, EINVAL,
1494 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1495 					  NULL, "No Rx queues configured");
1496 	if (!rss->queue_num)
1497 		return rte_flow_error_set(error, EINVAL,
1498 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499 					  NULL, "No queues configured");
1500 	for (i = 0; i != rss->queue_num; ++i) {
1501 		struct mlx5_rxq_ctrl *rxq_ctrl;
1502 
1503 		if (rss->queue[i] >= priv->rxqs_n)
1504 			return rte_flow_error_set
1505 				(error, EINVAL,
1506 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1507 				 &rss->queue[i], "queue index out of range");
1508 		if (!(*priv->rxqs)[rss->queue[i]])
1509 			return rte_flow_error_set
1510 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1511 				 &rss->queue[i], "queue is not configured");
1512 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1513 					struct mlx5_rxq_ctrl, rxq);
1514 		if (i == 0)
1515 			rxq_type = rxq_ctrl->type;
1516 		if (rxq_type != rxq_ctrl->type)
1517 			return rte_flow_error_set
1518 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1519 				 &rss->queue[i],
1520 				 "combining hairpin and regular RSS queues is not supported");
1521 	}
1522 	return 0;
1523 }
1524 
1525 /*
1526  * Validate the rss action.
1527  *
1528  * @param[in] action
1529  *   Pointer to the queue action.
1530  * @param[in] action_flags
1531  *   Bit-fields that holds the actions detected until now.
1532  * @param[in] dev
1533  *   Pointer to the Ethernet device structure.
1534  * @param[in] attr
1535  *   Attributes of flow that includes this action.
1536  * @param[in] item_flags
1537  *   Items that were detected.
1538  * @param[out] error
1539  *   Pointer to error structure.
1540  *
1541  * @return
1542  *   0 on success, a negative errno value otherwise and rte_errno is set.
1543  */
1544 int
1545 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1546 			      uint64_t action_flags,
1547 			      struct rte_eth_dev *dev,
1548 			      const struct rte_flow_attr *attr,
1549 			      uint64_t item_flags,
1550 			      struct rte_flow_error *error)
1551 {
1552 	const struct rte_flow_action_rss *rss = action->conf;
1553 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1554 	int ret;
1555 
1556 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1557 		return rte_flow_error_set(error, EINVAL,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1559 					  "can't have 2 fate actions"
1560 					  " in same flow");
1561 	ret = mlx5_validate_action_rss(dev, action, error);
1562 	if (ret)
1563 		return ret;
1564 	if (attr->egress)
1565 		return rte_flow_error_set(error, ENOTSUP,
1566 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1567 					  "rss action not supported for "
1568 					  "egress");
1569 	if (rss->level > 1 && !tunnel)
1570 		return rte_flow_error_set(error, EINVAL,
1571 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1572 					  "inner RSS is not supported for "
1573 					  "non-tunnel flows");
1574 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1575 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1576 		return rte_flow_error_set(error, EINVAL,
1577 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1578 					  "RSS on eCPRI is not supported now");
1579 	}
1580 	return 0;
1581 }
1582 
1583 /*
1584  * Validate the default miss action.
1585  *
1586  * @param[in] action_flags
1587  *   Bit-fields that holds the actions detected until now.
1588  * @param[out] error
1589  *   Pointer to error structure.
1590  *
1591  * @return
1592  *   0 on success, a negative errno value otherwise and rte_errno is set.
1593  */
1594 int
1595 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1596 				const struct rte_flow_attr *attr,
1597 				struct rte_flow_error *error)
1598 {
1599 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1600 		return rte_flow_error_set(error, EINVAL,
1601 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1602 					  "can't have 2 fate actions in"
1603 					  " same flow");
1604 	if (attr->egress)
1605 		return rte_flow_error_set(error, ENOTSUP,
1606 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1607 					  "default miss action not supported "
1608 					  "for egress");
1609 	if (attr->group)
1610 		return rte_flow_error_set(error, ENOTSUP,
1611 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1612 					  "only group 0 is supported");
1613 	if (attr->transfer)
1614 		return rte_flow_error_set(error, ENOTSUP,
1615 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1616 					  NULL, "transfer is not supported");
1617 	return 0;
1618 }
1619 
1620 /*
1621  * Validate the count action.
1622  *
1623  * @param[in] dev
1624  *   Pointer to the Ethernet device structure.
1625  * @param[in] attr
1626  *   Attributes of flow that includes this action.
1627  * @param[out] error
1628  *   Pointer to error structure.
1629  *
1630  * @return
1631  *   0 on success, a negative errno value otherwise and rte_errno is set.
1632  */
1633 int
1634 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1635 				const struct rte_flow_attr *attr,
1636 				struct rte_flow_error *error)
1637 {
1638 	if (attr->egress)
1639 		return rte_flow_error_set(error, ENOTSUP,
1640 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1641 					  "count action not supported for "
1642 					  "egress");
1643 	return 0;
1644 }
1645 
1646 /**
1647  * Verify the @p attributes will be correctly understood by the NIC and store
1648  * them in the @p flow if everything is correct.
1649  *
1650  * @param[in] dev
1651  *   Pointer to the Ethernet device structure.
1652  * @param[in] attributes
1653  *   Pointer to flow attributes
1654  * @param[out] error
1655  *   Pointer to error structure.
1656  *
1657  * @return
1658  *   0 on success, a negative errno value otherwise and rte_errno is set.
1659  */
1660 int
1661 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1662 			      const struct rte_flow_attr *attributes,
1663 			      struct rte_flow_error *error)
1664 {
1665 	struct mlx5_priv *priv = dev->data->dev_private;
1666 	uint32_t priority_max = priv->config.flow_prio - 1;
1667 
1668 	if (attributes->group)
1669 		return rte_flow_error_set(error, ENOTSUP,
1670 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1671 					  NULL, "groups is not supported");
1672 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1673 	    attributes->priority >= priority_max)
1674 		return rte_flow_error_set(error, ENOTSUP,
1675 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1676 					  NULL, "priority out of range");
1677 	if (attributes->egress)
1678 		return rte_flow_error_set(error, ENOTSUP,
1679 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1680 					  "egress is not supported");
1681 	if (attributes->transfer && !priv->config.dv_esw_en)
1682 		return rte_flow_error_set(error, ENOTSUP,
1683 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1684 					  NULL, "transfer is not supported");
1685 	if (!attributes->ingress)
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1688 					  NULL,
1689 					  "ingress attribute is mandatory");
1690 	return 0;
1691 }
1692 
1693 /**
1694  * Validate ICMP6 item.
1695  *
1696  * @param[in] item
1697  *   Item specification.
1698  * @param[in] item_flags
1699  *   Bit-fields that holds the items detected until now.
1700  * @param[in] ext_vlan_sup
1701  *   Whether extended VLAN features are supported or not.
1702  * @param[out] error
1703  *   Pointer to error structure.
1704  *
1705  * @return
1706  *   0 on success, a negative errno value otherwise and rte_errno is set.
1707  */
1708 int
1709 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1710 			       uint64_t item_flags,
1711 			       uint8_t target_protocol,
1712 			       struct rte_flow_error *error)
1713 {
1714 	const struct rte_flow_item_icmp6 *mask = item->mask;
1715 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1716 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1717 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1718 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1719 				      MLX5_FLOW_LAYER_OUTER_L4;
1720 	int ret;
1721 
1722 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1723 		return rte_flow_error_set(error, EINVAL,
1724 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1725 					  "protocol filtering not compatible"
1726 					  " with ICMP6 layer");
1727 	if (!(item_flags & l3m))
1728 		return rte_flow_error_set(error, EINVAL,
1729 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1730 					  "IPv6 is mandatory to filter on"
1731 					  " ICMP6");
1732 	if (item_flags & l4m)
1733 		return rte_flow_error_set(error, EINVAL,
1734 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1735 					  "multiple L4 layers not supported");
1736 	if (!mask)
1737 		mask = &rte_flow_item_icmp6_mask;
1738 	ret = mlx5_flow_item_acceptable
1739 		(item, (const uint8_t *)mask,
1740 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1741 		 sizeof(struct rte_flow_item_icmp6),
1742 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1743 	if (ret < 0)
1744 		return ret;
1745 	return 0;
1746 }
1747 
1748 /**
1749  * Validate ICMP item.
1750  *
1751  * @param[in] item
1752  *   Item specification.
1753  * @param[in] item_flags
1754  *   Bit-fields that holds the items detected until now.
1755  * @param[out] error
1756  *   Pointer to error structure.
1757  *
1758  * @return
1759  *   0 on success, a negative errno value otherwise and rte_errno is set.
1760  */
1761 int
1762 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1763 			     uint64_t item_flags,
1764 			     uint8_t target_protocol,
1765 			     struct rte_flow_error *error)
1766 {
1767 	const struct rte_flow_item_icmp *mask = item->mask;
1768 	const struct rte_flow_item_icmp nic_mask = {
1769 		.hdr.icmp_type = 0xff,
1770 		.hdr.icmp_code = 0xff,
1771 		.hdr.icmp_ident = RTE_BE16(0xffff),
1772 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1773 	};
1774 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1775 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1776 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1777 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1778 				      MLX5_FLOW_LAYER_OUTER_L4;
1779 	int ret;
1780 
1781 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1782 		return rte_flow_error_set(error, EINVAL,
1783 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1784 					  "protocol filtering not compatible"
1785 					  " with ICMP layer");
1786 	if (!(item_flags & l3m))
1787 		return rte_flow_error_set(error, EINVAL,
1788 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1789 					  "IPv4 is mandatory to filter"
1790 					  " on ICMP");
1791 	if (item_flags & l4m)
1792 		return rte_flow_error_set(error, EINVAL,
1793 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1794 					  "multiple L4 layers not supported");
1795 	if (!mask)
1796 		mask = &nic_mask;
1797 	ret = mlx5_flow_item_acceptable
1798 		(item, (const uint8_t *)mask,
1799 		 (const uint8_t *)&nic_mask,
1800 		 sizeof(struct rte_flow_item_icmp),
1801 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1802 	if (ret < 0)
1803 		return ret;
1804 	return 0;
1805 }
1806 
1807 /**
1808  * Validate Ethernet item.
1809  *
1810  * @param[in] item
1811  *   Item specification.
1812  * @param[in] item_flags
1813  *   Bit-fields that holds the items detected until now.
1814  * @param[out] error
1815  *   Pointer to error structure.
1816  *
1817  * @return
1818  *   0 on success, a negative errno value otherwise and rte_errno is set.
1819  */
1820 int
1821 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1822 			    uint64_t item_flags, bool ext_vlan_sup,
1823 			    struct rte_flow_error *error)
1824 {
1825 	const struct rte_flow_item_eth *mask = item->mask;
1826 	const struct rte_flow_item_eth nic_mask = {
1827 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1828 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1829 		.type = RTE_BE16(0xffff),
1830 		.has_vlan = ext_vlan_sup ? 1 : 0,
1831 	};
1832 	int ret;
1833 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1834 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1835 				       MLX5_FLOW_LAYER_OUTER_L2;
1836 
1837 	if (item_flags & ethm)
1838 		return rte_flow_error_set(error, ENOTSUP,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "multiple L2 layers not supported");
1841 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1842 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1843 		return rte_flow_error_set(error, EINVAL,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "L2 layer should not follow "
1846 					  "L3 layers");
1847 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1848 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1849 		return rte_flow_error_set(error, EINVAL,
1850 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1851 					  "L2 layer should not follow VLAN");
1852 	if (!mask)
1853 		mask = &rte_flow_item_eth_mask;
1854 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1855 					(const uint8_t *)&nic_mask,
1856 					sizeof(struct rte_flow_item_eth),
1857 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1858 	return ret;
1859 }
1860 
1861 /**
1862  * Validate VLAN item.
1863  *
1864  * @param[in] item
1865  *   Item specification.
1866  * @param[in] item_flags
1867  *   Bit-fields that holds the items detected until now.
1868  * @param[in] dev
1869  *   Ethernet device flow is being created on.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 int
1877 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1878 			     uint64_t item_flags,
1879 			     struct rte_eth_dev *dev,
1880 			     struct rte_flow_error *error)
1881 {
1882 	const struct rte_flow_item_vlan *spec = item->spec;
1883 	const struct rte_flow_item_vlan *mask = item->mask;
1884 	const struct rte_flow_item_vlan nic_mask = {
1885 		.tci = RTE_BE16(UINT16_MAX),
1886 		.inner_type = RTE_BE16(UINT16_MAX),
1887 	};
1888 	uint16_t vlan_tag = 0;
1889 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1890 	int ret;
1891 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1892 					MLX5_FLOW_LAYER_INNER_L4) :
1893 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1894 					MLX5_FLOW_LAYER_OUTER_L4);
1895 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1896 					MLX5_FLOW_LAYER_OUTER_VLAN;
1897 
1898 	if (item_flags & vlanm)
1899 		return rte_flow_error_set(error, EINVAL,
1900 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1901 					  "multiple VLAN layers not supported");
1902 	else if ((item_flags & l34m) != 0)
1903 		return rte_flow_error_set(error, EINVAL,
1904 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1905 					  "VLAN cannot follow L3/L4 layer");
1906 	if (!mask)
1907 		mask = &rte_flow_item_vlan_mask;
1908 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1909 					(const uint8_t *)&nic_mask,
1910 					sizeof(struct rte_flow_item_vlan),
1911 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1912 	if (ret)
1913 		return ret;
1914 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1915 		struct mlx5_priv *priv = dev->data->dev_private;
1916 
1917 		if (priv->vmwa_context) {
1918 			/*
1919 			 * Non-NULL context means we have a virtual machine
1920 			 * and SR-IOV enabled, we have to create VLAN interface
1921 			 * to make hypervisor to setup E-Switch vport
1922 			 * context correctly. We avoid creating the multiple
1923 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1924 			 */
1925 			return rte_flow_error_set(error, EINVAL,
1926 						  RTE_FLOW_ERROR_TYPE_ITEM,
1927 						  item,
1928 						  "VLAN tag mask is not"
1929 						  " supported in virtual"
1930 						  " environment");
1931 		}
1932 	}
1933 	if (spec) {
1934 		vlan_tag = spec->tci;
1935 		vlan_tag &= mask->tci;
1936 	}
1937 	/*
1938 	 * From verbs perspective an empty VLAN is equivalent
1939 	 * to a packet without VLAN layer.
1940 	 */
1941 	if (!vlan_tag)
1942 		return rte_flow_error_set(error, EINVAL,
1943 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1944 					  item->spec,
1945 					  "VLAN cannot be empty");
1946 	return 0;
1947 }
1948 
1949 /**
1950  * Validate IPV4 item.
1951  *
1952  * @param[in] item
1953  *   Item specification.
1954  * @param[in] item_flags
1955  *   Bit-fields that holds the items detected until now.
1956  * @param[in] last_item
1957  *   Previous validated item in the pattern items.
1958  * @param[in] ether_type
1959  *   Type in the ethernet layer header (including dot1q).
1960  * @param[in] acc_mask
1961  *   Acceptable mask, if NULL default internal default mask
1962  *   will be used to check whether item fields are supported.
1963  * @param[in] range_accepted
1964  *   True if range of values is accepted for specific fields, false otherwise.
1965  * @param[out] error
1966  *   Pointer to error structure.
1967  *
1968  * @return
1969  *   0 on success, a negative errno value otherwise and rte_errno is set.
1970  */
1971 int
1972 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1973 			     uint64_t item_flags,
1974 			     uint64_t last_item,
1975 			     uint16_t ether_type,
1976 			     const struct rte_flow_item_ipv4 *acc_mask,
1977 			     bool range_accepted,
1978 			     struct rte_flow_error *error)
1979 {
1980 	const struct rte_flow_item_ipv4 *mask = item->mask;
1981 	const struct rte_flow_item_ipv4 *spec = item->spec;
1982 	const struct rte_flow_item_ipv4 nic_mask = {
1983 		.hdr = {
1984 			.src_addr = RTE_BE32(0xffffffff),
1985 			.dst_addr = RTE_BE32(0xffffffff),
1986 			.type_of_service = 0xff,
1987 			.next_proto_id = 0xff,
1988 		},
1989 	};
1990 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1991 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1992 				      MLX5_FLOW_LAYER_OUTER_L3;
1993 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1994 				      MLX5_FLOW_LAYER_OUTER_L4;
1995 	int ret;
1996 	uint8_t next_proto = 0xFF;
1997 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1998 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1999 				  MLX5_FLOW_LAYER_INNER_VLAN);
2000 
2001 	if ((last_item & l2_vlan) && ether_type &&
2002 	    ether_type != RTE_ETHER_TYPE_IPV4)
2003 		return rte_flow_error_set(error, EINVAL,
2004 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2005 					  "IPv4 cannot follow L2/VLAN layer "
2006 					  "which ether type is not IPv4");
2007 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2008 		if (mask && spec)
2009 			next_proto = mask->hdr.next_proto_id &
2010 				     spec->hdr.next_proto_id;
2011 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2012 			return rte_flow_error_set(error, EINVAL,
2013 						  RTE_FLOW_ERROR_TYPE_ITEM,
2014 						  item,
2015 						  "multiple tunnel "
2016 						  "not supported");
2017 	}
2018 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2019 		return rte_flow_error_set(error, EINVAL,
2020 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2021 					  "wrong tunnel type - IPv6 specified "
2022 					  "but IPv4 item provided");
2023 	if (item_flags & l3m)
2024 		return rte_flow_error_set(error, ENOTSUP,
2025 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2026 					  "multiple L3 layers not supported");
2027 	else if (item_flags & l4m)
2028 		return rte_flow_error_set(error, EINVAL,
2029 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2030 					  "L3 cannot follow an L4 layer.");
2031 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2032 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2033 		return rte_flow_error_set(error, EINVAL,
2034 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2035 					  "L3 cannot follow an NVGRE layer.");
2036 	if (!mask)
2037 		mask = &rte_flow_item_ipv4_mask;
2038 	else if (mask->hdr.next_proto_id != 0 &&
2039 		 mask->hdr.next_proto_id != 0xff)
2040 		return rte_flow_error_set(error, EINVAL,
2041 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2042 					  "partial mask is not supported"
2043 					  " for protocol");
2044 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2045 					acc_mask ? (const uint8_t *)acc_mask
2046 						 : (const uint8_t *)&nic_mask,
2047 					sizeof(struct rte_flow_item_ipv4),
2048 					range_accepted, error);
2049 	if (ret < 0)
2050 		return ret;
2051 	return 0;
2052 }
2053 
2054 /**
2055  * Validate IPV6 item.
2056  *
2057  * @param[in] item
2058  *   Item specification.
2059  * @param[in] item_flags
2060  *   Bit-fields that holds the items detected until now.
2061  * @param[in] last_item
2062  *   Previous validated item in the pattern items.
2063  * @param[in] ether_type
2064  *   Type in the ethernet layer header (including dot1q).
2065  * @param[in] acc_mask
2066  *   Acceptable mask, if NULL default internal default mask
2067  *   will be used to check whether item fields are supported.
2068  * @param[out] error
2069  *   Pointer to error structure.
2070  *
2071  * @return
2072  *   0 on success, a negative errno value otherwise and rte_errno is set.
2073  */
2074 int
2075 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2076 			     uint64_t item_flags,
2077 			     uint64_t last_item,
2078 			     uint16_t ether_type,
2079 			     const struct rte_flow_item_ipv6 *acc_mask,
2080 			     struct rte_flow_error *error)
2081 {
2082 	const struct rte_flow_item_ipv6 *mask = item->mask;
2083 	const struct rte_flow_item_ipv6 *spec = item->spec;
2084 	const struct rte_flow_item_ipv6 nic_mask = {
2085 		.hdr = {
2086 			.src_addr =
2087 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2088 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2089 			.dst_addr =
2090 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2091 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2092 			.vtc_flow = RTE_BE32(0xffffffff),
2093 			.proto = 0xff,
2094 		},
2095 	};
2096 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2097 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2098 				      MLX5_FLOW_LAYER_OUTER_L3;
2099 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2100 				      MLX5_FLOW_LAYER_OUTER_L4;
2101 	int ret;
2102 	uint8_t next_proto = 0xFF;
2103 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2104 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2105 				  MLX5_FLOW_LAYER_INNER_VLAN);
2106 
2107 	if ((last_item & l2_vlan) && ether_type &&
2108 	    ether_type != RTE_ETHER_TYPE_IPV6)
2109 		return rte_flow_error_set(error, EINVAL,
2110 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2111 					  "IPv6 cannot follow L2/VLAN layer "
2112 					  "which ether type is not IPv6");
2113 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2114 		next_proto = spec->hdr.proto;
2115 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2116 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2117 			return rte_flow_error_set(error, EINVAL,
2118 						  RTE_FLOW_ERROR_TYPE_ITEM,
2119 						  item,
2120 						  "multiple tunnel "
2121 						  "not supported");
2122 	}
2123 	if (next_proto == IPPROTO_HOPOPTS  ||
2124 	    next_proto == IPPROTO_ROUTING  ||
2125 	    next_proto == IPPROTO_FRAGMENT ||
2126 	    next_proto == IPPROTO_ESP	   ||
2127 	    next_proto == IPPROTO_AH	   ||
2128 	    next_proto == IPPROTO_DSTOPTS)
2129 		return rte_flow_error_set(error, EINVAL,
2130 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2131 					  "IPv6 proto (next header) should "
2132 					  "not be set as extension header");
2133 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2134 		return rte_flow_error_set(error, EINVAL,
2135 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2136 					  "wrong tunnel type - IPv4 specified "
2137 					  "but IPv6 item provided");
2138 	if (item_flags & l3m)
2139 		return rte_flow_error_set(error, ENOTSUP,
2140 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2141 					  "multiple L3 layers not supported");
2142 	else if (item_flags & l4m)
2143 		return rte_flow_error_set(error, EINVAL,
2144 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2145 					  "L3 cannot follow an L4 layer.");
2146 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2147 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2148 		return rte_flow_error_set(error, EINVAL,
2149 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2150 					  "L3 cannot follow an NVGRE layer.");
2151 	if (!mask)
2152 		mask = &rte_flow_item_ipv6_mask;
2153 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2154 					acc_mask ? (const uint8_t *)acc_mask
2155 						 : (const uint8_t *)&nic_mask,
2156 					sizeof(struct rte_flow_item_ipv6),
2157 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2158 	if (ret < 0)
2159 		return ret;
2160 	return 0;
2161 }
2162 
2163 /**
2164  * Validate UDP item.
2165  *
2166  * @param[in] item
2167  *   Item specification.
2168  * @param[in] item_flags
2169  *   Bit-fields that holds the items detected until now.
2170  * @param[in] target_protocol
2171  *   The next protocol in the previous item.
2172  * @param[in] flow_mask
2173  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2174  * @param[out] error
2175  *   Pointer to error structure.
2176  *
2177  * @return
2178  *   0 on success, a negative errno value otherwise and rte_errno is set.
2179  */
2180 int
2181 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2182 			    uint64_t item_flags,
2183 			    uint8_t target_protocol,
2184 			    struct rte_flow_error *error)
2185 {
2186 	const struct rte_flow_item_udp *mask = item->mask;
2187 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2188 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2189 				      MLX5_FLOW_LAYER_OUTER_L3;
2190 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2191 				      MLX5_FLOW_LAYER_OUTER_L4;
2192 	int ret;
2193 
2194 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2195 		return rte_flow_error_set(error, EINVAL,
2196 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 					  "protocol filtering not compatible"
2198 					  " with UDP layer");
2199 	if (!(item_flags & l3m))
2200 		return rte_flow_error_set(error, EINVAL,
2201 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2202 					  "L3 is mandatory to filter on L4");
2203 	if (item_flags & l4m)
2204 		return rte_flow_error_set(error, EINVAL,
2205 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2206 					  "multiple L4 layers not supported");
2207 	if (!mask)
2208 		mask = &rte_flow_item_udp_mask;
2209 	ret = mlx5_flow_item_acceptable
2210 		(item, (const uint8_t *)mask,
2211 		 (const uint8_t *)&rte_flow_item_udp_mask,
2212 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2213 		 error);
2214 	if (ret < 0)
2215 		return ret;
2216 	return 0;
2217 }
2218 
2219 /**
2220  * Validate TCP item.
2221  *
2222  * @param[in] item
2223  *   Item specification.
2224  * @param[in] item_flags
2225  *   Bit-fields that holds the items detected until now.
2226  * @param[in] target_protocol
2227  *   The next protocol in the previous item.
2228  * @param[out] error
2229  *   Pointer to error structure.
2230  *
2231  * @return
2232  *   0 on success, a negative errno value otherwise and rte_errno is set.
2233  */
2234 int
2235 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2236 			    uint64_t item_flags,
2237 			    uint8_t target_protocol,
2238 			    const struct rte_flow_item_tcp *flow_mask,
2239 			    struct rte_flow_error *error)
2240 {
2241 	const struct rte_flow_item_tcp *mask = item->mask;
2242 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2243 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2244 				      MLX5_FLOW_LAYER_OUTER_L3;
2245 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2246 				      MLX5_FLOW_LAYER_OUTER_L4;
2247 	int ret;
2248 
2249 	MLX5_ASSERT(flow_mask);
2250 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2251 		return rte_flow_error_set(error, EINVAL,
2252 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2253 					  "protocol filtering not compatible"
2254 					  " with TCP layer");
2255 	if (!(item_flags & l3m))
2256 		return rte_flow_error_set(error, EINVAL,
2257 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2258 					  "L3 is mandatory to filter on L4");
2259 	if (item_flags & l4m)
2260 		return rte_flow_error_set(error, EINVAL,
2261 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2262 					  "multiple L4 layers not supported");
2263 	if (!mask)
2264 		mask = &rte_flow_item_tcp_mask;
2265 	ret = mlx5_flow_item_acceptable
2266 		(item, (const uint8_t *)mask,
2267 		 (const uint8_t *)flow_mask,
2268 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2269 		 error);
2270 	if (ret < 0)
2271 		return ret;
2272 	return 0;
2273 }
2274 
2275 /**
2276  * Validate VXLAN item.
2277  *
2278  * @param[in] item
2279  *   Item specification.
2280  * @param[in] item_flags
2281  *   Bit-fields that holds the items detected until now.
2282  * @param[in] target_protocol
2283  *   The next protocol in the previous item.
2284  * @param[out] error
2285  *   Pointer to error structure.
2286  *
2287  * @return
2288  *   0 on success, a negative errno value otherwise and rte_errno is set.
2289  */
2290 int
2291 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2292 			      uint64_t item_flags,
2293 			      struct rte_flow_error *error)
2294 {
2295 	const struct rte_flow_item_vxlan *spec = item->spec;
2296 	const struct rte_flow_item_vxlan *mask = item->mask;
2297 	int ret;
2298 	union vni {
2299 		uint32_t vlan_id;
2300 		uint8_t vni[4];
2301 	} id = { .vlan_id = 0, };
2302 
2303 
2304 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2305 		return rte_flow_error_set(error, ENOTSUP,
2306 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2307 					  "multiple tunnel layers not"
2308 					  " supported");
2309 	/*
2310 	 * Verify only UDPv4 is present as defined in
2311 	 * https://tools.ietf.org/html/rfc7348
2312 	 */
2313 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2314 		return rte_flow_error_set(error, EINVAL,
2315 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2316 					  "no outer UDP layer found");
2317 	if (!mask)
2318 		mask = &rte_flow_item_vxlan_mask;
2319 	ret = mlx5_flow_item_acceptable
2320 		(item, (const uint8_t *)mask,
2321 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2322 		 sizeof(struct rte_flow_item_vxlan),
2323 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2324 	if (ret < 0)
2325 		return ret;
2326 	if (spec) {
2327 		memcpy(&id.vni[1], spec->vni, 3);
2328 		memcpy(&id.vni[1], mask->vni, 3);
2329 	}
2330 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2331 		return rte_flow_error_set(error, ENOTSUP,
2332 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2333 					  "VXLAN tunnel must be fully defined");
2334 	return 0;
2335 }
2336 
2337 /**
2338  * Validate VXLAN_GPE item.
2339  *
2340  * @param[in] item
2341  *   Item specification.
2342  * @param[in] item_flags
2343  *   Bit-fields that holds the items detected until now.
2344  * @param[in] priv
2345  *   Pointer to the private data structure.
2346  * @param[in] target_protocol
2347  *   The next protocol in the previous item.
2348  * @param[out] error
2349  *   Pointer to error structure.
2350  *
2351  * @return
2352  *   0 on success, a negative errno value otherwise and rte_errno is set.
2353  */
2354 int
2355 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2356 				  uint64_t item_flags,
2357 				  struct rte_eth_dev *dev,
2358 				  struct rte_flow_error *error)
2359 {
2360 	struct mlx5_priv *priv = dev->data->dev_private;
2361 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2362 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2363 	int ret;
2364 	union vni {
2365 		uint32_t vlan_id;
2366 		uint8_t vni[4];
2367 	} id = { .vlan_id = 0, };
2368 
2369 	if (!priv->config.l3_vxlan_en)
2370 		return rte_flow_error_set(error, ENOTSUP,
2371 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2372 					  "L3 VXLAN is not enabled by device"
2373 					  " parameter and/or not configured in"
2374 					  " firmware");
2375 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2376 		return rte_flow_error_set(error, ENOTSUP,
2377 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2378 					  "multiple tunnel layers not"
2379 					  " supported");
2380 	/*
2381 	 * Verify only UDPv4 is present as defined in
2382 	 * https://tools.ietf.org/html/rfc7348
2383 	 */
2384 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2385 		return rte_flow_error_set(error, EINVAL,
2386 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2387 					  "no outer UDP layer found");
2388 	if (!mask)
2389 		mask = &rte_flow_item_vxlan_gpe_mask;
2390 	ret = mlx5_flow_item_acceptable
2391 		(item, (const uint8_t *)mask,
2392 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2393 		 sizeof(struct rte_flow_item_vxlan_gpe),
2394 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2395 	if (ret < 0)
2396 		return ret;
2397 	if (spec) {
2398 		if (spec->protocol)
2399 			return rte_flow_error_set(error, ENOTSUP,
2400 						  RTE_FLOW_ERROR_TYPE_ITEM,
2401 						  item,
2402 						  "VxLAN-GPE protocol"
2403 						  " not supported");
2404 		memcpy(&id.vni[1], spec->vni, 3);
2405 		memcpy(&id.vni[1], mask->vni, 3);
2406 	}
2407 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2408 		return rte_flow_error_set(error, ENOTSUP,
2409 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2410 					  "VXLAN-GPE tunnel must be fully"
2411 					  " defined");
2412 	return 0;
2413 }
2414 /**
2415  * Validate GRE Key item.
2416  *
2417  * @param[in] item
2418  *   Item specification.
2419  * @param[in] item_flags
2420  *   Bit flags to mark detected items.
2421  * @param[in] gre_item
2422  *   Pointer to gre_item
2423  * @param[out] error
2424  *   Pointer to error structure.
2425  *
2426  * @return
2427  *   0 on success, a negative errno value otherwise and rte_errno is set.
2428  */
2429 int
2430 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2431 				uint64_t item_flags,
2432 				const struct rte_flow_item *gre_item,
2433 				struct rte_flow_error *error)
2434 {
2435 	const rte_be32_t *mask = item->mask;
2436 	int ret = 0;
2437 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2438 	const struct rte_flow_item_gre *gre_spec;
2439 	const struct rte_flow_item_gre *gre_mask;
2440 
2441 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2442 		return rte_flow_error_set(error, ENOTSUP,
2443 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2444 					  "Multiple GRE key not support");
2445 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2446 		return rte_flow_error_set(error, ENOTSUP,
2447 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2448 					  "No preceding GRE header");
2449 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2450 		return rte_flow_error_set(error, ENOTSUP,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 					  "GRE key following a wrong item");
2453 	gre_mask = gre_item->mask;
2454 	if (!gre_mask)
2455 		gre_mask = &rte_flow_item_gre_mask;
2456 	gre_spec = gre_item->spec;
2457 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2458 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2459 		return rte_flow_error_set(error, EINVAL,
2460 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2461 					  "Key bit must be on");
2462 
2463 	if (!mask)
2464 		mask = &gre_key_default_mask;
2465 	ret = mlx5_flow_item_acceptable
2466 		(item, (const uint8_t *)mask,
2467 		 (const uint8_t *)&gre_key_default_mask,
2468 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2469 	return ret;
2470 }
2471 
2472 /**
2473  * Validate GRE item.
2474  *
2475  * @param[in] item
2476  *   Item specification.
2477  * @param[in] item_flags
2478  *   Bit flags to mark detected items.
2479  * @param[in] target_protocol
2480  *   The next protocol in the previous item.
2481  * @param[out] error
2482  *   Pointer to error structure.
2483  *
2484  * @return
2485  *   0 on success, a negative errno value otherwise and rte_errno is set.
2486  */
2487 int
2488 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2489 			    uint64_t item_flags,
2490 			    uint8_t target_protocol,
2491 			    struct rte_flow_error *error)
2492 {
2493 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2494 	const struct rte_flow_item_gre *mask = item->mask;
2495 	int ret;
2496 	const struct rte_flow_item_gre nic_mask = {
2497 		.c_rsvd0_ver = RTE_BE16(0xB000),
2498 		.protocol = RTE_BE16(UINT16_MAX),
2499 	};
2500 
2501 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2502 		return rte_flow_error_set(error, EINVAL,
2503 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2504 					  "protocol filtering not compatible"
2505 					  " with this GRE layer");
2506 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2507 		return rte_flow_error_set(error, ENOTSUP,
2508 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2509 					  "multiple tunnel layers not"
2510 					  " supported");
2511 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2512 		return rte_flow_error_set(error, ENOTSUP,
2513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2514 					  "L3 Layer is missing");
2515 	if (!mask)
2516 		mask = &rte_flow_item_gre_mask;
2517 	ret = mlx5_flow_item_acceptable
2518 		(item, (const uint8_t *)mask,
2519 		 (const uint8_t *)&nic_mask,
2520 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2521 		 error);
2522 	if (ret < 0)
2523 		return ret;
2524 #ifndef HAVE_MLX5DV_DR
2525 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2526 	if (spec && (spec->protocol & mask->protocol))
2527 		return rte_flow_error_set(error, ENOTSUP,
2528 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2529 					  "without MPLS support the"
2530 					  " specification cannot be used for"
2531 					  " filtering");
2532 #endif
2533 #endif
2534 	return 0;
2535 }
2536 
2537 /**
2538  * Validate Geneve item.
2539  *
2540  * @param[in] item
2541  *   Item specification.
2542  * @param[in] itemFlags
2543  *   Bit-fields that holds the items detected until now.
2544  * @param[in] enPriv
2545  *   Pointer to the private data structure.
2546  * @param[out] error
2547  *   Pointer to error structure.
2548  *
2549  * @return
2550  *   0 on success, a negative errno value otherwise and rte_errno is set.
2551  */
2552 
2553 int
2554 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2555 			       uint64_t item_flags,
2556 			       struct rte_eth_dev *dev,
2557 			       struct rte_flow_error *error)
2558 {
2559 	struct mlx5_priv *priv = dev->data->dev_private;
2560 	const struct rte_flow_item_geneve *spec = item->spec;
2561 	const struct rte_flow_item_geneve *mask = item->mask;
2562 	int ret;
2563 	uint16_t gbhdr;
2564 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2565 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2566 	const struct rte_flow_item_geneve nic_mask = {
2567 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2568 		.vni = "\xff\xff\xff",
2569 		.protocol = RTE_BE16(UINT16_MAX),
2570 	};
2571 
2572 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2573 		return rte_flow_error_set(error, ENOTSUP,
2574 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2575 					  "L3 Geneve is not enabled by device"
2576 					  " parameter and/or not configured in"
2577 					  " firmware");
2578 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2579 		return rte_flow_error_set(error, ENOTSUP,
2580 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2581 					  "multiple tunnel layers not"
2582 					  " supported");
2583 	/*
2584 	 * Verify only UDPv4 is present as defined in
2585 	 * https://tools.ietf.org/html/rfc7348
2586 	 */
2587 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2588 		return rte_flow_error_set(error, EINVAL,
2589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2590 					  "no outer UDP layer found");
2591 	if (!mask)
2592 		mask = &rte_flow_item_geneve_mask;
2593 	ret = mlx5_flow_item_acceptable
2594 				  (item, (const uint8_t *)mask,
2595 				   (const uint8_t *)&nic_mask,
2596 				   sizeof(struct rte_flow_item_geneve),
2597 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2598 	if (ret)
2599 		return ret;
2600 	if (spec) {
2601 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2602 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2603 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2604 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2605 			return rte_flow_error_set(error, ENOTSUP,
2606 						  RTE_FLOW_ERROR_TYPE_ITEM,
2607 						  item,
2608 						  "Geneve protocol unsupported"
2609 						  " fields are being used");
2610 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2611 			return rte_flow_error_set
2612 					(error, ENOTSUP,
2613 					 RTE_FLOW_ERROR_TYPE_ITEM,
2614 					 item,
2615 					 "Unsupported Geneve options length");
2616 	}
2617 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2618 		return rte_flow_error_set
2619 				    (error, ENOTSUP,
2620 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2621 				     "Geneve tunnel must be fully defined");
2622 	return 0;
2623 }
2624 
2625 /**
2626  * Validate Geneve TLV option item.
2627  *
2628  * @param[in] item
2629  *   Item specification.
2630  * @param[in] last_item
2631  *   Previous validated item in the pattern items.
2632  * @param[in] geneve_item
2633  *   Previous GENEVE item specification.
2634  * @param[in] dev
2635  *   Pointer to the rte_eth_dev structure.
2636  * @param[out] error
2637  *   Pointer to error structure.
2638  *
2639  * @return
2640  *   0 on success, a negative errno value otherwise and rte_errno is set.
2641  */
2642 int
2643 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2644 				   uint64_t last_item,
2645 				   const struct rte_flow_item *geneve_item,
2646 				   struct rte_eth_dev *dev,
2647 				   struct rte_flow_error *error)
2648 {
2649 	struct mlx5_priv *priv = dev->data->dev_private;
2650 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2651 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2652 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2653 	uint8_t data_max_supported =
2654 			hca_attr->max_geneve_tlv_option_data_len * 4;
2655 	struct mlx5_dev_config *config = &priv->config;
2656 	const struct rte_flow_item_geneve *geneve_spec;
2657 	const struct rte_flow_item_geneve *geneve_mask;
2658 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2659 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2660 	unsigned int i;
2661 	unsigned int data_len;
2662 	uint8_t tlv_option_len;
2663 	uint16_t optlen_m, optlen_v;
2664 	const struct rte_flow_item_geneve_opt full_mask = {
2665 		.option_class = RTE_BE16(0xffff),
2666 		.option_type = 0xff,
2667 		.option_len = 0x1f,
2668 	};
2669 
2670 	if (!mask)
2671 		mask = &rte_flow_item_geneve_opt_mask;
2672 	if (!spec)
2673 		return rte_flow_error_set
2674 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2675 			"Geneve TLV opt class/type/length must be specified");
2676 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2677 		return rte_flow_error_set
2678 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 			"Geneve TLV opt length exceeeds the limit (31)");
2680 	/* Check if class type and length masks are full. */
2681 	if (full_mask.option_class != mask->option_class ||
2682 	    full_mask.option_type != mask->option_type ||
2683 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2684 		return rte_flow_error_set
2685 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2686 			"Geneve TLV opt class/type/length masks must be full");
2687 	/* Check if length is supported */
2688 	if ((uint32_t)spec->option_len >
2689 			config->hca_attr.max_geneve_tlv_option_data_len)
2690 		return rte_flow_error_set
2691 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2692 			"Geneve TLV opt length not supported");
2693 	if (config->hca_attr.max_geneve_tlv_options > 1)
2694 		DRV_LOG(DEBUG,
2695 			"max_geneve_tlv_options supports more than 1 option");
2696 	/* Check GENEVE item preceding. */
2697 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2698 		return rte_flow_error_set
2699 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2700 			"Geneve opt item must be preceded with Geneve item");
2701 	geneve_spec = geneve_item->spec;
2702 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2703 					  &rte_flow_item_geneve_mask;
2704 	/* Check if GENEVE TLV option size doesn't exceed option length */
2705 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2706 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2707 		tlv_option_len = spec->option_len & mask->option_len;
2708 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2709 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2710 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2711 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2712 		if ((optlen_v & optlen_m) <= tlv_option_len)
2713 			return rte_flow_error_set
2714 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2715 				 "GENEVE TLV option length exceeds optlen");
2716 	}
2717 	/* Check if length is 0 or data is 0. */
2718 	if (spec->data == NULL || spec->option_len == 0)
2719 		return rte_flow_error_set
2720 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2721 			"Geneve TLV opt with zero data/length not supported");
2722 	/* Check not all data & mask are 0. */
2723 	data_len = spec->option_len * 4;
2724 	if (mask->data == NULL) {
2725 		for (i = 0; i < data_len; i++)
2726 			if (spec->data[i])
2727 				break;
2728 		if (i == data_len)
2729 			return rte_flow_error_set(error, ENOTSUP,
2730 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2731 				"Can't match on Geneve option data 0");
2732 	} else {
2733 		for (i = 0; i < data_len; i++)
2734 			if (spec->data[i] & mask->data[i])
2735 				break;
2736 		if (i == data_len)
2737 			return rte_flow_error_set(error, ENOTSUP,
2738 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2739 				"Can't match on Geneve option data and mask 0");
2740 		/* Check data mask supported. */
2741 		for (i = data_max_supported; i < data_len ; i++)
2742 			if (mask->data[i])
2743 				return rte_flow_error_set(error, ENOTSUP,
2744 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2745 					"Data mask is of unsupported size");
2746 	}
2747 	/* Check GENEVE option is supported in NIC. */
2748 	if (!config->hca_attr.geneve_tlv_opt)
2749 		return rte_flow_error_set
2750 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2751 			"Geneve TLV opt not supported");
2752 	/* Check if we already have geneve option with different type/class. */
2753 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2754 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2755 	if (geneve_opt_resource != NULL)
2756 		if (geneve_opt_resource->option_class != spec->option_class ||
2757 		    geneve_opt_resource->option_type != spec->option_type ||
2758 		    geneve_opt_resource->length != spec->option_len) {
2759 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2760 			return rte_flow_error_set(error, ENOTSUP,
2761 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2762 				"Only one Geneve TLV option supported");
2763 		}
2764 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2765 	return 0;
2766 }
2767 
2768 /**
2769  * Validate MPLS item.
2770  *
2771  * @param[in] dev
2772  *   Pointer to the rte_eth_dev structure.
2773  * @param[in] item
2774  *   Item specification.
2775  * @param[in] item_flags
2776  *   Bit-fields that holds the items detected until now.
2777  * @param[in] prev_layer
2778  *   The protocol layer indicated in previous item.
2779  * @param[out] error
2780  *   Pointer to error structure.
2781  *
2782  * @return
2783  *   0 on success, a negative errno value otherwise and rte_errno is set.
2784  */
2785 int
2786 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2787 			     const struct rte_flow_item *item __rte_unused,
2788 			     uint64_t item_flags __rte_unused,
2789 			     uint64_t prev_layer __rte_unused,
2790 			     struct rte_flow_error *error)
2791 {
2792 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2793 	const struct rte_flow_item_mpls *mask = item->mask;
2794 	struct mlx5_priv *priv = dev->data->dev_private;
2795 	int ret;
2796 
2797 	if (!priv->config.mpls_en)
2798 		return rte_flow_error_set(error, ENOTSUP,
2799 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2800 					  "MPLS not supported or"
2801 					  " disabled in firmware"
2802 					  " configuration.");
2803 	/* MPLS over IP, UDP, GRE is allowed */
2804 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2805 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2806 			    MLX5_FLOW_LAYER_GRE |
2807 			    MLX5_FLOW_LAYER_GRE_KEY)))
2808 		return rte_flow_error_set(error, EINVAL,
2809 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2810 					  "protocol filtering not compatible"
2811 					  " with MPLS layer");
2812 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2813 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2814 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2815 		return rte_flow_error_set(error, ENOTSUP,
2816 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2817 					  "multiple tunnel layers not"
2818 					  " supported");
2819 	if (!mask)
2820 		mask = &rte_flow_item_mpls_mask;
2821 	ret = mlx5_flow_item_acceptable
2822 		(item, (const uint8_t *)mask,
2823 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2824 		 sizeof(struct rte_flow_item_mpls),
2825 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2826 	if (ret < 0)
2827 		return ret;
2828 	return 0;
2829 #else
2830 	return rte_flow_error_set(error, ENOTSUP,
2831 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2832 				  "MPLS is not supported by Verbs, please"
2833 				  " update.");
2834 #endif
2835 }
2836 
2837 /**
2838  * Validate NVGRE item.
2839  *
2840  * @param[in] item
2841  *   Item specification.
2842  * @param[in] item_flags
2843  *   Bit flags to mark detected items.
2844  * @param[in] target_protocol
2845  *   The next protocol in the previous item.
2846  * @param[out] error
2847  *   Pointer to error structure.
2848  *
2849  * @return
2850  *   0 on success, a negative errno value otherwise and rte_errno is set.
2851  */
2852 int
2853 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2854 			      uint64_t item_flags,
2855 			      uint8_t target_protocol,
2856 			      struct rte_flow_error *error)
2857 {
2858 	const struct rte_flow_item_nvgre *mask = item->mask;
2859 	int ret;
2860 
2861 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2862 		return rte_flow_error_set(error, EINVAL,
2863 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2864 					  "protocol filtering not compatible"
2865 					  " with this GRE layer");
2866 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2867 		return rte_flow_error_set(error, ENOTSUP,
2868 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2869 					  "multiple tunnel layers not"
2870 					  " supported");
2871 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2872 		return rte_flow_error_set(error, ENOTSUP,
2873 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 					  "L3 Layer is missing");
2875 	if (!mask)
2876 		mask = &rte_flow_item_nvgre_mask;
2877 	ret = mlx5_flow_item_acceptable
2878 		(item, (const uint8_t *)mask,
2879 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2880 		 sizeof(struct rte_flow_item_nvgre),
2881 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2882 	if (ret < 0)
2883 		return ret;
2884 	return 0;
2885 }
2886 
2887 /**
2888  * Validate eCPRI item.
2889  *
2890  * @param[in] item
2891  *   Item specification.
2892  * @param[in] item_flags
2893  *   Bit-fields that holds the items detected until now.
2894  * @param[in] last_item
2895  *   Previous validated item in the pattern items.
2896  * @param[in] ether_type
2897  *   Type in the ethernet layer header (including dot1q).
2898  * @param[in] acc_mask
2899  *   Acceptable mask, if NULL default internal default mask
2900  *   will be used to check whether item fields are supported.
2901  * @param[out] error
2902  *   Pointer to error structure.
2903  *
2904  * @return
2905  *   0 on success, a negative errno value otherwise and rte_errno is set.
2906  */
2907 int
2908 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2909 			      uint64_t item_flags,
2910 			      uint64_t last_item,
2911 			      uint16_t ether_type,
2912 			      const struct rte_flow_item_ecpri *acc_mask,
2913 			      struct rte_flow_error *error)
2914 {
2915 	const struct rte_flow_item_ecpri *mask = item->mask;
2916 	const struct rte_flow_item_ecpri nic_mask = {
2917 		.hdr = {
2918 			.common = {
2919 				.u32 =
2920 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2921 					.type = 0xFF,
2922 					}).u32),
2923 			},
2924 			.dummy[0] = 0xFFFFFFFF,
2925 		},
2926 	};
2927 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2928 					MLX5_FLOW_LAYER_OUTER_VLAN);
2929 	struct rte_flow_item_ecpri mask_lo;
2930 
2931 	if (!(last_item & outer_l2_vlan) &&
2932 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2933 		return rte_flow_error_set(error, EINVAL,
2934 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2935 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2936 	if ((last_item & outer_l2_vlan) && ether_type &&
2937 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2938 		return rte_flow_error_set(error, EINVAL,
2939 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2940 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2941 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2942 		return rte_flow_error_set(error, EINVAL,
2943 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2944 					  "eCPRI with tunnel is not supported right now");
2945 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2946 		return rte_flow_error_set(error, ENOTSUP,
2947 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2948 					  "multiple L3 layers not supported");
2949 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2950 		return rte_flow_error_set(error, EINVAL,
2951 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2952 					  "eCPRI cannot coexist with a TCP layer");
2953 	/* In specification, eCPRI could be over UDP layer. */
2954 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2955 		return rte_flow_error_set(error, EINVAL,
2956 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2957 					  "eCPRI over UDP layer is not yet supported right now");
2958 	/* Mask for type field in common header could be zero. */
2959 	if (!mask)
2960 		mask = &rte_flow_item_ecpri_mask;
2961 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2962 	/* Input mask is in big-endian format. */
2963 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2964 		return rte_flow_error_set(error, EINVAL,
2965 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2966 					  "partial mask is not supported for protocol");
2967 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2968 		return rte_flow_error_set(error, EINVAL,
2969 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2970 					  "message header mask must be after a type mask");
2971 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2972 					 acc_mask ? (const uint8_t *)acc_mask
2973 						  : (const uint8_t *)&nic_mask,
2974 					 sizeof(struct rte_flow_item_ecpri),
2975 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2976 }
2977 
2978 /**
2979  * Release resource related QUEUE/RSS action split.
2980  *
2981  * @param dev
2982  *   Pointer to Ethernet device.
2983  * @param flow
2984  *   Flow to release id's from.
2985  */
2986 static void
2987 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2988 			     struct rte_flow *flow)
2989 {
2990 	struct mlx5_priv *priv = dev->data->dev_private;
2991 	uint32_t handle_idx;
2992 	struct mlx5_flow_handle *dev_handle;
2993 
2994 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2995 		       handle_idx, dev_handle, next)
2996 		if (dev_handle->split_flow_id)
2997 			mlx5_ipool_free(priv->sh->ipool
2998 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2999 					dev_handle->split_flow_id);
3000 }
3001 
3002 static int
3003 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3004 		   const struct rte_flow_attr *attr __rte_unused,
3005 		   const struct rte_flow_item items[] __rte_unused,
3006 		   const struct rte_flow_action actions[] __rte_unused,
3007 		   bool external __rte_unused,
3008 		   int hairpin __rte_unused,
3009 		   struct rte_flow_error *error)
3010 {
3011 	return rte_flow_error_set(error, ENOTSUP,
3012 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3013 }
3014 
3015 static struct mlx5_flow *
3016 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3017 		  const struct rte_flow_attr *attr __rte_unused,
3018 		  const struct rte_flow_item items[] __rte_unused,
3019 		  const struct rte_flow_action actions[] __rte_unused,
3020 		  struct rte_flow_error *error)
3021 {
3022 	rte_flow_error_set(error, ENOTSUP,
3023 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3024 	return NULL;
3025 }
3026 
3027 static int
3028 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3029 		    struct mlx5_flow *dev_flow __rte_unused,
3030 		    const struct rte_flow_attr *attr __rte_unused,
3031 		    const struct rte_flow_item items[] __rte_unused,
3032 		    const struct rte_flow_action actions[] __rte_unused,
3033 		    struct rte_flow_error *error)
3034 {
3035 	return rte_flow_error_set(error, ENOTSUP,
3036 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3037 }
3038 
3039 static int
3040 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3041 		struct rte_flow *flow __rte_unused,
3042 		struct rte_flow_error *error)
3043 {
3044 	return rte_flow_error_set(error, ENOTSUP,
3045 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3046 }
3047 
3048 static void
3049 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3050 		 struct rte_flow *flow __rte_unused)
3051 {
3052 }
3053 
3054 static void
3055 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3056 		  struct rte_flow *flow __rte_unused)
3057 {
3058 }
3059 
3060 static int
3061 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3062 		struct rte_flow *flow __rte_unused,
3063 		const struct rte_flow_action *actions __rte_unused,
3064 		void *data __rte_unused,
3065 		struct rte_flow_error *error)
3066 {
3067 	return rte_flow_error_set(error, ENOTSUP,
3068 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3069 }
3070 
3071 static int
3072 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3073 		      uint32_t domains __rte_unused,
3074 		      uint32_t flags __rte_unused)
3075 {
3076 	return 0;
3077 }
3078 
3079 /* Void driver to protect from null pointer reference. */
3080 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3081 	.validate = flow_null_validate,
3082 	.prepare = flow_null_prepare,
3083 	.translate = flow_null_translate,
3084 	.apply = flow_null_apply,
3085 	.remove = flow_null_remove,
3086 	.destroy = flow_null_destroy,
3087 	.query = flow_null_query,
3088 	.sync_domain = flow_null_sync_domain,
3089 };
3090 
3091 /**
3092  * Select flow driver type according to flow attributes and device
3093  * configuration.
3094  *
3095  * @param[in] dev
3096  *   Pointer to the dev structure.
3097  * @param[in] attr
3098  *   Pointer to the flow attributes.
3099  *
3100  * @return
3101  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3102  */
3103 static enum mlx5_flow_drv_type
3104 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3105 {
3106 	struct mlx5_priv *priv = dev->data->dev_private;
3107 	/* The OS can determine first a specific flow type (DV, VERBS) */
3108 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3109 
3110 	if (type != MLX5_FLOW_TYPE_MAX)
3111 		return type;
3112 	/* If no OS specific type - continue with DV/VERBS selection */
3113 	if (attr->transfer && priv->config.dv_esw_en)
3114 		type = MLX5_FLOW_TYPE_DV;
3115 	if (!attr->transfer)
3116 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3117 						 MLX5_FLOW_TYPE_VERBS;
3118 	return type;
3119 }
3120 
3121 #define flow_get_drv_ops(type) flow_drv_ops[type]
3122 
3123 /**
3124  * Flow driver validation API. This abstracts calling driver specific functions.
3125  * The type of flow driver is determined according to flow attributes.
3126  *
3127  * @param[in] dev
3128  *   Pointer to the dev structure.
3129  * @param[in] attr
3130  *   Pointer to the flow attributes.
3131  * @param[in] items
3132  *   Pointer to the list of items.
3133  * @param[in] actions
3134  *   Pointer to the list of actions.
3135  * @param[in] external
3136  *   This flow rule is created by request external to PMD.
3137  * @param[in] hairpin
3138  *   Number of hairpin TX actions, 0 means classic flow.
3139  * @param[out] error
3140  *   Pointer to the error structure.
3141  *
3142  * @return
3143  *   0 on success, a negative errno value otherwise and rte_errno is set.
3144  */
3145 static inline int
3146 flow_drv_validate(struct rte_eth_dev *dev,
3147 		  const struct rte_flow_attr *attr,
3148 		  const struct rte_flow_item items[],
3149 		  const struct rte_flow_action actions[],
3150 		  bool external, int hairpin, struct rte_flow_error *error)
3151 {
3152 	const struct mlx5_flow_driver_ops *fops;
3153 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3154 
3155 	fops = flow_get_drv_ops(type);
3156 	return fops->validate(dev, attr, items, actions, external,
3157 			      hairpin, error);
3158 }
3159 
3160 /**
3161  * Flow driver preparation API. This abstracts calling driver specific
3162  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3163  * calculates the size of memory required for device flow, allocates the memory,
3164  * initializes the device flow and returns the pointer.
3165  *
3166  * @note
3167  *   This function initializes device flow structure such as dv or verbs in
3168  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3169  *   rest. For example, adding returning device flow to flow->dev_flow list and
3170  *   setting backward reference to the flow should be done out of this function.
3171  *   layers field is not filled either.
3172  *
3173  * @param[in] dev
3174  *   Pointer to the dev structure.
3175  * @param[in] attr
3176  *   Pointer to the flow attributes.
3177  * @param[in] items
3178  *   Pointer to the list of items.
3179  * @param[in] actions
3180  *   Pointer to the list of actions.
3181  * @param[in] flow_idx
3182  *   This memory pool index to the flow.
3183  * @param[out] error
3184  *   Pointer to the error structure.
3185  *
3186  * @return
3187  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3188  */
3189 static inline struct mlx5_flow *
3190 flow_drv_prepare(struct rte_eth_dev *dev,
3191 		 const struct rte_flow *flow,
3192 		 const struct rte_flow_attr *attr,
3193 		 const struct rte_flow_item items[],
3194 		 const struct rte_flow_action actions[],
3195 		 uint32_t flow_idx,
3196 		 struct rte_flow_error *error)
3197 {
3198 	const struct mlx5_flow_driver_ops *fops;
3199 	enum mlx5_flow_drv_type type = flow->drv_type;
3200 	struct mlx5_flow *mlx5_flow = NULL;
3201 
3202 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3203 	fops = flow_get_drv_ops(type);
3204 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3205 	if (mlx5_flow)
3206 		mlx5_flow->flow_idx = flow_idx;
3207 	return mlx5_flow;
3208 }
3209 
3210 /**
3211  * Flow driver translation API. This abstracts calling driver specific
3212  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3213  * translates a generic flow into a driver flow. flow_drv_prepare() must
3214  * precede.
3215  *
3216  * @note
3217  *   dev_flow->layers could be filled as a result of parsing during translation
3218  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3219  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3220  *   flow->actions could be overwritten even though all the expanded dev_flows
3221  *   have the same actions.
3222  *
3223  * @param[in] dev
3224  *   Pointer to the rte dev structure.
3225  * @param[in, out] dev_flow
3226  *   Pointer to the mlx5 flow.
3227  * @param[in] attr
3228  *   Pointer to the flow attributes.
3229  * @param[in] items
3230  *   Pointer to the list of items.
3231  * @param[in] actions
3232  *   Pointer to the list of actions.
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_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3241 		   const struct rte_flow_attr *attr,
3242 		   const struct rte_flow_item items[],
3243 		   const struct rte_flow_action actions[],
3244 		   struct rte_flow_error *error)
3245 {
3246 	const struct mlx5_flow_driver_ops *fops;
3247 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3248 
3249 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3250 	fops = flow_get_drv_ops(type);
3251 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3252 }
3253 
3254 /**
3255  * Flow driver apply API. This abstracts calling driver specific functions.
3256  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3257  * translated driver flows on to device. flow_drv_translate() must precede.
3258  *
3259  * @param[in] dev
3260  *   Pointer to Ethernet device structure.
3261  * @param[in, out] flow
3262  *   Pointer to flow structure.
3263  * @param[out] error
3264  *   Pointer to error structure.
3265  *
3266  * @return
3267  *   0 on success, a negative errno value otherwise and rte_errno is set.
3268  */
3269 static inline int
3270 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3271 	       struct rte_flow_error *error)
3272 {
3273 	const struct mlx5_flow_driver_ops *fops;
3274 	enum mlx5_flow_drv_type type = flow->drv_type;
3275 
3276 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3277 	fops = flow_get_drv_ops(type);
3278 	return fops->apply(dev, flow, error);
3279 }
3280 
3281 /**
3282  * Flow driver destroy API. This abstracts calling driver specific functions.
3283  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3284  * on device and releases resources of the flow.
3285  *
3286  * @param[in] dev
3287  *   Pointer to Ethernet device.
3288  * @param[in, out] flow
3289  *   Pointer to flow structure.
3290  */
3291 static inline void
3292 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3293 {
3294 	const struct mlx5_flow_driver_ops *fops;
3295 	enum mlx5_flow_drv_type type = flow->drv_type;
3296 
3297 	flow_mreg_split_qrss_release(dev, flow);
3298 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3299 	fops = flow_get_drv_ops(type);
3300 	fops->destroy(dev, flow);
3301 }
3302 
3303 /**
3304  * Get RSS action from the action list.
3305  *
3306  * @param[in] actions
3307  *   Pointer to the list of actions.
3308  *
3309  * @return
3310  *   Pointer to the RSS action if exist, else return NULL.
3311  */
3312 static const struct rte_flow_action_rss*
3313 flow_get_rss_action(const struct rte_flow_action actions[])
3314 {
3315 	const struct rte_flow_action_rss *rss = NULL;
3316 
3317 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3318 		switch (actions->type) {
3319 		case RTE_FLOW_ACTION_TYPE_RSS:
3320 			rss = actions->conf;
3321 			break;
3322 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3323 		{
3324 			const struct rte_flow_action_sample *sample =
3325 								actions->conf;
3326 			const struct rte_flow_action *act = sample->actions;
3327 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3328 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3329 					rss = act->conf;
3330 			break;
3331 		}
3332 		default:
3333 			break;
3334 		}
3335 	}
3336 	return rss;
3337 }
3338 
3339 /**
3340  * Get ASO age action by index.
3341  *
3342  * @param[in] dev
3343  *   Pointer to the Ethernet device structure.
3344  * @param[in] age_idx
3345  *   Index to the ASO age action.
3346  *
3347  * @return
3348  *   The specified ASO age action.
3349  */
3350 struct mlx5_aso_age_action*
3351 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3352 {
3353 	uint16_t pool_idx = age_idx & UINT16_MAX;
3354 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3355 	struct mlx5_priv *priv = dev->data->dev_private;
3356 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3357 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3358 
3359 	return &pool->actions[offset - 1];
3360 }
3361 
3362 /* maps shared action to translated non shared in some actions array */
3363 struct mlx5_translated_shared_action {
3364 	struct rte_flow_shared_action *action; /**< Shared action */
3365 	int index; /**< Index in related array of rte_flow_action */
3366 };
3367 
3368 /**
3369  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3370  * non shared action if translation possible.
3371  * This functionality used to run same execution path for both shared & non
3372  * shared actions on flow create. All necessary preparations for shared
3373  * action handling should be preformed on *shared* actions list returned
3374  * from this call.
3375  *
3376  * @param[in] dev
3377  *   Pointer to Ethernet device.
3378  * @param[in] actions
3379  *   List of actions to translate.
3380  * @param[out] shared
3381  *   List to store translated shared actions.
3382  * @param[in, out] shared_n
3383  *   Size of *shared* array. On return should be updated with number of shared
3384  *   actions retrieved from the *actions* list.
3385  * @param[out] translated_actions
3386  *   List of actions where all shared actions were translated to non shared
3387  *   if possible. NULL if no translation took place.
3388  * @param[out] error
3389  *   Pointer to the error structure.
3390  *
3391  * @return
3392  *   0 on success, a negative errno value otherwise and rte_errno is set.
3393  */
3394 static int
3395 flow_shared_actions_translate(struct rte_eth_dev *dev,
3396 			      const struct rte_flow_action actions[],
3397 			      struct mlx5_translated_shared_action *shared,
3398 			      int *shared_n,
3399 			      struct rte_flow_action **translated_actions,
3400 			      struct rte_flow_error *error)
3401 {
3402 	struct mlx5_priv *priv = dev->data->dev_private;
3403 	struct rte_flow_action *translated = NULL;
3404 	size_t actions_size;
3405 	int n;
3406 	int copied_n = 0;
3407 	struct mlx5_translated_shared_action *shared_end = NULL;
3408 
3409 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3410 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3411 			continue;
3412 		if (copied_n == *shared_n) {
3413 			return rte_flow_error_set
3414 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3415 				 NULL, "too many shared actions");
3416 		}
3417 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3418 			   sizeof(actions[n].conf));
3419 		shared[copied_n].index = n;
3420 		copied_n++;
3421 	}
3422 	n++;
3423 	*shared_n = copied_n;
3424 	if (!copied_n)
3425 		return 0;
3426 	actions_size = sizeof(struct rte_flow_action) * n;
3427 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3428 	if (!translated) {
3429 		rte_errno = ENOMEM;
3430 		return -ENOMEM;
3431 	}
3432 	memcpy(translated, actions, actions_size);
3433 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3434 		struct mlx5_shared_action_rss *shared_rss;
3435 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3436 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3437 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3438 									   - 1);
3439 
3440 		switch (type) {
3441 		case MLX5_SHARED_ACTION_TYPE_RSS:
3442 			shared_rss = mlx5_ipool_get
3443 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3444 			translated[shared->index].type =
3445 				RTE_FLOW_ACTION_TYPE_RSS;
3446 			translated[shared->index].conf =
3447 				&shared_rss->origin;
3448 			break;
3449 		case MLX5_SHARED_ACTION_TYPE_AGE:
3450 			if (priv->sh->flow_hit_aso_en) {
3451 				translated[shared->index].type =
3452 					(enum rte_flow_action_type)
3453 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3454 				translated[shared->index].conf =
3455 							 (void *)(uintptr_t)idx;
3456 				break;
3457 			}
3458 			/* Fall-through */
3459 		default:
3460 			mlx5_free(translated);
3461 			return rte_flow_error_set
3462 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3463 				 NULL, "invalid shared action type");
3464 		}
3465 	}
3466 	*translated_actions = translated;
3467 	return 0;
3468 }
3469 
3470 /**
3471  * Get Shared RSS action from the action list.
3472  *
3473  * @param[in] dev
3474  *   Pointer to Ethernet device.
3475  * @param[in] shared
3476  *   Pointer to the list of actions.
3477  * @param[in] shared_n
3478  *   Actions list length.
3479  *
3480  * @return
3481  *   The MLX5 RSS action ID if exists, otherwise return 0.
3482  */
3483 static uint32_t
3484 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3485 			   struct mlx5_translated_shared_action *shared,
3486 			   int shared_n)
3487 {
3488 	struct mlx5_translated_shared_action *shared_end;
3489 	struct mlx5_priv *priv = dev->data->dev_private;
3490 	struct mlx5_shared_action_rss *shared_rss;
3491 
3492 
3493 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3494 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3495 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3496 		uint32_t idx = act_idx &
3497 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3498 		switch (type) {
3499 		case MLX5_SHARED_ACTION_TYPE_RSS:
3500 			shared_rss = mlx5_ipool_get
3501 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3502 									   idx);
3503 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3504 					   __ATOMIC_RELAXED);
3505 			return idx;
3506 		default:
3507 			break;
3508 		}
3509 	}
3510 	return 0;
3511 }
3512 
3513 static unsigned int
3514 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3515 {
3516 	const struct rte_flow_item *item;
3517 	unsigned int has_vlan = 0;
3518 
3519 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3520 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3521 			has_vlan = 1;
3522 			break;
3523 		}
3524 	}
3525 	if (has_vlan)
3526 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3527 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3528 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3529 			       MLX5_EXPANSION_ROOT_OUTER;
3530 }
3531 
3532 /**
3533  *  Get layer flags from the prefix flow.
3534  *
3535  *  Some flows may be split to several subflows, the prefix subflow gets the
3536  *  match items and the suffix sub flow gets the actions.
3537  *  Some actions need the user defined match item flags to get the detail for
3538  *  the action.
3539  *  This function helps the suffix flow to get the item layer flags from prefix
3540  *  subflow.
3541  *
3542  * @param[in] dev_flow
3543  *   Pointer the created preifx subflow.
3544  *
3545  * @return
3546  *   The layers get from prefix subflow.
3547  */
3548 static inline uint64_t
3549 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3550 {
3551 	uint64_t layers = 0;
3552 
3553 	/*
3554 	 * Layers bits could be localization, but usually the compiler will
3555 	 * help to do the optimization work for source code.
3556 	 * If no decap actions, use the layers directly.
3557 	 */
3558 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3559 		return dev_flow->handle->layers;
3560 	/* Convert L3 layers with decap action. */
3561 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3562 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3563 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3564 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3565 	/* Convert L4 layers with decap action.  */
3566 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3567 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3568 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3569 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3570 	return layers;
3571 }
3572 
3573 /**
3574  * Get metadata split action information.
3575  *
3576  * @param[in] actions
3577  *   Pointer to the list of actions.
3578  * @param[out] qrss
3579  *   Pointer to the return pointer.
3580  * @param[out] qrss_type
3581  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3582  *   if no QUEUE/RSS is found.
3583  * @param[out] encap_idx
3584  *   Pointer to the index of the encap action if exists, otherwise the last
3585  *   action index.
3586  *
3587  * @return
3588  *   Total number of actions.
3589  */
3590 static int
3591 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3592 				       const struct rte_flow_action **qrss,
3593 				       int *encap_idx)
3594 {
3595 	const struct rte_flow_action_raw_encap *raw_encap;
3596 	int actions_n = 0;
3597 	int raw_decap_idx = -1;
3598 
3599 	*encap_idx = -1;
3600 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3601 		switch (actions->type) {
3602 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3603 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3604 			*encap_idx = actions_n;
3605 			break;
3606 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3607 			raw_decap_idx = actions_n;
3608 			break;
3609 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3610 			raw_encap = actions->conf;
3611 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3612 				*encap_idx = raw_decap_idx != -1 ?
3613 						      raw_decap_idx : actions_n;
3614 			break;
3615 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3616 		case RTE_FLOW_ACTION_TYPE_RSS:
3617 			*qrss = actions;
3618 			break;
3619 		default:
3620 			break;
3621 		}
3622 		actions_n++;
3623 	}
3624 	if (*encap_idx == -1)
3625 		*encap_idx = actions_n;
3626 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3627 	return actions_n + 1;
3628 }
3629 
3630 /**
3631  * Check meter action from the action list.
3632  *
3633  * @param[in] actions
3634  *   Pointer to the list of actions.
3635  * @param[out] mtr
3636  *   Pointer to the meter exist flag.
3637  *
3638  * @return
3639  *   Total number of actions.
3640  */
3641 static int
3642 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3643 {
3644 	int actions_n = 0;
3645 
3646 	MLX5_ASSERT(mtr);
3647 	*mtr = 0;
3648 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3649 		switch (actions->type) {
3650 		case RTE_FLOW_ACTION_TYPE_METER:
3651 			*mtr = 1;
3652 			break;
3653 		default:
3654 			break;
3655 		}
3656 		actions_n++;
3657 	}
3658 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3659 	return actions_n + 1;
3660 }
3661 
3662 /**
3663  * Check if the flow should be split due to hairpin.
3664  * The reason for the split is that in current HW we can't
3665  * support encap and push-vlan on Rx, so if a flow contains
3666  * these actions we move it to Tx.
3667  *
3668  * @param dev
3669  *   Pointer to Ethernet device.
3670  * @param[in] attr
3671  *   Flow rule attributes.
3672  * @param[in] actions
3673  *   Associated actions (list terminated by the END action).
3674  *
3675  * @return
3676  *   > 0 the number of actions and the flow should be split,
3677  *   0 when no split required.
3678  */
3679 static int
3680 flow_check_hairpin_split(struct rte_eth_dev *dev,
3681 			 const struct rte_flow_attr *attr,
3682 			 const struct rte_flow_action actions[])
3683 {
3684 	int queue_action = 0;
3685 	int action_n = 0;
3686 	int split = 0;
3687 	const struct rte_flow_action_queue *queue;
3688 	const struct rte_flow_action_rss *rss;
3689 	const struct rte_flow_action_raw_encap *raw_encap;
3690 	const struct rte_eth_hairpin_conf *conf;
3691 
3692 	if (!attr->ingress)
3693 		return 0;
3694 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3695 		switch (actions->type) {
3696 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3697 			queue = actions->conf;
3698 			if (queue == NULL)
3699 				return 0;
3700 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3701 			if (conf == NULL || conf->tx_explicit != 0)
3702 				return 0;
3703 			queue_action = 1;
3704 			action_n++;
3705 			break;
3706 		case RTE_FLOW_ACTION_TYPE_RSS:
3707 			rss = actions->conf;
3708 			if (rss == NULL || rss->queue_num == 0)
3709 				return 0;
3710 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3711 			if (conf == NULL || conf->tx_explicit != 0)
3712 				return 0;
3713 			queue_action = 1;
3714 			action_n++;
3715 			break;
3716 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3717 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3718 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3719 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3720 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3721 			split++;
3722 			action_n++;
3723 			break;
3724 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3725 			raw_encap = actions->conf;
3726 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3727 				split++;
3728 			action_n++;
3729 			break;
3730 		default:
3731 			action_n++;
3732 			break;
3733 		}
3734 	}
3735 	if (split && queue_action)
3736 		return action_n;
3737 	return 0;
3738 }
3739 
3740 /* Declare flow create/destroy prototype in advance. */
3741 static uint32_t
3742 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3743 		 const struct rte_flow_attr *attr,
3744 		 const struct rte_flow_item items[],
3745 		 const struct rte_flow_action actions[],
3746 		 bool external, struct rte_flow_error *error);
3747 
3748 static void
3749 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3750 		  uint32_t flow_idx);
3751 
3752 int
3753 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3754 		      struct mlx5_hlist_entry *entry,
3755 		      uint64_t key, void *cb_ctx __rte_unused)
3756 {
3757 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3758 		container_of(entry, typeof(*mcp_res), hlist_ent);
3759 
3760 	return mcp_res->mark_id != key;
3761 }
3762 
3763 struct mlx5_hlist_entry *
3764 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3765 		       void *cb_ctx)
3766 {
3767 	struct rte_eth_dev *dev = list->ctx;
3768 	struct mlx5_priv *priv = dev->data->dev_private;
3769 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3770 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3771 	struct rte_flow_error *error = ctx->error;
3772 	uint32_t idx = 0;
3773 	int ret;
3774 	uint32_t mark_id = key;
3775 	struct rte_flow_attr attr = {
3776 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3777 		.ingress = 1,
3778 	};
3779 	struct mlx5_rte_flow_item_tag tag_spec = {
3780 		.data = mark_id,
3781 	};
3782 	struct rte_flow_item items[] = {
3783 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3784 	};
3785 	struct rte_flow_action_mark ftag = {
3786 		.id = mark_id,
3787 	};
3788 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3789 		.dst = REG_B,
3790 		.src = REG_NON,
3791 	};
3792 	struct rte_flow_action_jump jump = {
3793 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3794 	};
3795 	struct rte_flow_action actions[] = {
3796 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3797 	};
3798 
3799 	/* Fill the register fileds in the flow. */
3800 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3801 	if (ret < 0)
3802 		return NULL;
3803 	tag_spec.id = ret;
3804 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3805 	if (ret < 0)
3806 		return NULL;
3807 	cp_mreg.src = ret;
3808 	/* Provide the full width of FLAG specific value. */
3809 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3810 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3811 	/* Build a new flow. */
3812 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3813 		items[0] = (struct rte_flow_item){
3814 			.type = (enum rte_flow_item_type)
3815 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3816 			.spec = &tag_spec,
3817 		};
3818 		items[1] = (struct rte_flow_item){
3819 			.type = RTE_FLOW_ITEM_TYPE_END,
3820 		};
3821 		actions[0] = (struct rte_flow_action){
3822 			.type = (enum rte_flow_action_type)
3823 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3824 			.conf = &ftag,
3825 		};
3826 		actions[1] = (struct rte_flow_action){
3827 			.type = (enum rte_flow_action_type)
3828 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3829 			.conf = &cp_mreg,
3830 		};
3831 		actions[2] = (struct rte_flow_action){
3832 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3833 			.conf = &jump,
3834 		};
3835 		actions[3] = (struct rte_flow_action){
3836 			.type = RTE_FLOW_ACTION_TYPE_END,
3837 		};
3838 	} else {
3839 		/* Default rule, wildcard match. */
3840 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3841 		items[0] = (struct rte_flow_item){
3842 			.type = RTE_FLOW_ITEM_TYPE_END,
3843 		};
3844 		actions[0] = (struct rte_flow_action){
3845 			.type = (enum rte_flow_action_type)
3846 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3847 			.conf = &cp_mreg,
3848 		};
3849 		actions[1] = (struct rte_flow_action){
3850 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3851 			.conf = &jump,
3852 		};
3853 		actions[2] = (struct rte_flow_action){
3854 			.type = RTE_FLOW_ACTION_TYPE_END,
3855 		};
3856 	}
3857 	/* Build a new entry. */
3858 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3859 	if (!mcp_res) {
3860 		rte_errno = ENOMEM;
3861 		return NULL;
3862 	}
3863 	mcp_res->idx = idx;
3864 	mcp_res->mark_id = mark_id;
3865 	/*
3866 	 * The copy Flows are not included in any list. There
3867 	 * ones are referenced from other Flows and can not
3868 	 * be applied, removed, deleted in ardbitrary order
3869 	 * by list traversing.
3870 	 */
3871 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3872 					 actions, false, error);
3873 	if (!mcp_res->rix_flow) {
3874 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3875 		return NULL;
3876 	}
3877 	return &mcp_res->hlist_ent;
3878 }
3879 
3880 /**
3881  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3882  *
3883  * As mark_id is unique, if there's already a registered flow for the mark_id,
3884  * return by increasing the reference counter of the resource. Otherwise, create
3885  * the resource (mcp_res) and flow.
3886  *
3887  * Flow looks like,
3888  *   - If ingress port is ANY and reg_c[1] is mark_id,
3889  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3890  *
3891  * For default flow (zero mark_id), flow is like,
3892  *   - If ingress port is ANY,
3893  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3894  *
3895  * @param dev
3896  *   Pointer to Ethernet device.
3897  * @param mark_id
3898  *   ID of MARK action, zero means default flow for META.
3899  * @param[out] error
3900  *   Perform verbose error reporting if not NULL.
3901  *
3902  * @return
3903  *   Associated resource on success, NULL otherwise and rte_errno is set.
3904  */
3905 static struct mlx5_flow_mreg_copy_resource *
3906 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3907 			  struct rte_flow_error *error)
3908 {
3909 	struct mlx5_priv *priv = dev->data->dev_private;
3910 	struct mlx5_hlist_entry *entry;
3911 	struct mlx5_flow_cb_ctx ctx = {
3912 		.dev = dev,
3913 		.error = error,
3914 	};
3915 
3916 	/* Check if already registered. */
3917 	MLX5_ASSERT(priv->mreg_cp_tbl);
3918 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3919 	if (!entry)
3920 		return NULL;
3921 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3922 			    hlist_ent);
3923 }
3924 
3925 void
3926 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3927 {
3928 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3929 		container_of(entry, typeof(*mcp_res), hlist_ent);
3930 	struct rte_eth_dev *dev = list->ctx;
3931 	struct mlx5_priv *priv = dev->data->dev_private;
3932 
3933 	MLX5_ASSERT(mcp_res->rix_flow);
3934 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3935 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3936 }
3937 
3938 /**
3939  * Release flow in RX_CP_TBL.
3940  *
3941  * @param dev
3942  *   Pointer to Ethernet device.
3943  * @flow
3944  *   Parent flow for wich copying is provided.
3945  */
3946 static void
3947 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3948 			  struct rte_flow *flow)
3949 {
3950 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3951 	struct mlx5_priv *priv = dev->data->dev_private;
3952 
3953 	if (!flow->rix_mreg_copy)
3954 		return;
3955 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3956 				 flow->rix_mreg_copy);
3957 	if (!mcp_res || !priv->mreg_cp_tbl)
3958 		return;
3959 	MLX5_ASSERT(mcp_res->rix_flow);
3960 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3961 	flow->rix_mreg_copy = 0;
3962 }
3963 
3964 /**
3965  * Remove the default copy action from RX_CP_TBL.
3966  *
3967  * This functions is called in the mlx5_dev_start(). No thread safe
3968  * is guaranteed.
3969  *
3970  * @param dev
3971  *   Pointer to Ethernet device.
3972  */
3973 static void
3974 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3975 {
3976 	struct mlx5_hlist_entry *entry;
3977 	struct mlx5_priv *priv = dev->data->dev_private;
3978 
3979 	/* Check if default flow is registered. */
3980 	if (!priv->mreg_cp_tbl)
3981 		return;
3982 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3983 				  MLX5_DEFAULT_COPY_ID, NULL);
3984 	if (!entry)
3985 		return;
3986 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3987 }
3988 
3989 /**
3990  * Add the default copy action in in RX_CP_TBL.
3991  *
3992  * This functions is called in the mlx5_dev_start(). No thread safe
3993  * is guaranteed.
3994  *
3995  * @param dev
3996  *   Pointer to Ethernet device.
3997  * @param[out] error
3998  *   Perform verbose error reporting if not NULL.
3999  *
4000  * @return
4001  *   0 for success, negative value otherwise and rte_errno is set.
4002  */
4003 static int
4004 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4005 				  struct rte_flow_error *error)
4006 {
4007 	struct mlx5_priv *priv = dev->data->dev_private;
4008 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4009 
4010 	/* Check whether extensive metadata feature is engaged. */
4011 	if (!priv->config.dv_flow_en ||
4012 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4013 	    !mlx5_flow_ext_mreg_supported(dev) ||
4014 	    !priv->sh->dv_regc0_mask)
4015 		return 0;
4016 	/*
4017 	 * Add default mreg copy flow may be called multiple time, but
4018 	 * only be called once in stop. Avoid register it twice.
4019 	 */
4020 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4021 		return 0;
4022 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4023 	if (!mcp_res)
4024 		return -rte_errno;
4025 	return 0;
4026 }
4027 
4028 /**
4029  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4030  *
4031  * All the flow having Q/RSS action should be split by
4032  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4033  * performs the following,
4034  *   - CQE->flow_tag := reg_c[1] (MARK)
4035  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4036  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4037  * but there should be a flow per each MARK ID set by MARK action.
4038  *
4039  * For the aforementioned reason, if there's a MARK action in flow's action
4040  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4041  * the MARK ID to CQE's flow_tag like,
4042  *   - If reg_c[1] is mark_id,
4043  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4044  *
4045  * For SET_META action which stores value in reg_c[0], as the destination is
4046  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4047  * MARK ID means the default flow. The default flow looks like,
4048  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4049  *
4050  * @param dev
4051  *   Pointer to Ethernet device.
4052  * @param flow
4053  *   Pointer to flow structure.
4054  * @param[in] actions
4055  *   Pointer to the list of actions.
4056  * @param[out] error
4057  *   Perform verbose error reporting if not NULL.
4058  *
4059  * @return
4060  *   0 on success, negative value otherwise and rte_errno is set.
4061  */
4062 static int
4063 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4064 			    struct rte_flow *flow,
4065 			    const struct rte_flow_action *actions,
4066 			    struct rte_flow_error *error)
4067 {
4068 	struct mlx5_priv *priv = dev->data->dev_private;
4069 	struct mlx5_dev_config *config = &priv->config;
4070 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4071 	const struct rte_flow_action_mark *mark;
4072 
4073 	/* Check whether extensive metadata feature is engaged. */
4074 	if (!config->dv_flow_en ||
4075 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4076 	    !mlx5_flow_ext_mreg_supported(dev) ||
4077 	    !priv->sh->dv_regc0_mask)
4078 		return 0;
4079 	/* Find MARK action. */
4080 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4081 		switch (actions->type) {
4082 		case RTE_FLOW_ACTION_TYPE_FLAG:
4083 			mcp_res = flow_mreg_add_copy_action
4084 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4085 			if (!mcp_res)
4086 				return -rte_errno;
4087 			flow->rix_mreg_copy = mcp_res->idx;
4088 			return 0;
4089 		case RTE_FLOW_ACTION_TYPE_MARK:
4090 			mark = (const struct rte_flow_action_mark *)
4091 				actions->conf;
4092 			mcp_res =
4093 				flow_mreg_add_copy_action(dev, mark->id, error);
4094 			if (!mcp_res)
4095 				return -rte_errno;
4096 			flow->rix_mreg_copy = mcp_res->idx;
4097 			return 0;
4098 		default:
4099 			break;
4100 		}
4101 	}
4102 	return 0;
4103 }
4104 
4105 #define MLX5_MAX_SPLIT_ACTIONS 24
4106 #define MLX5_MAX_SPLIT_ITEMS 24
4107 
4108 /**
4109  * Split the hairpin flow.
4110  * Since HW can't support encap and push-vlan on Rx, we move these
4111  * actions to Tx.
4112  * If the count action is after the encap then we also
4113  * move the count action. in this case the count will also measure
4114  * the outer bytes.
4115  *
4116  * @param dev
4117  *   Pointer to Ethernet device.
4118  * @param[in] actions
4119  *   Associated actions (list terminated by the END action).
4120  * @param[out] actions_rx
4121  *   Rx flow actions.
4122  * @param[out] actions_tx
4123  *   Tx flow actions..
4124  * @param[out] pattern_tx
4125  *   The pattern items for the Tx flow.
4126  * @param[out] flow_id
4127  *   The flow ID connected to this flow.
4128  *
4129  * @return
4130  *   0 on success.
4131  */
4132 static int
4133 flow_hairpin_split(struct rte_eth_dev *dev,
4134 		   const struct rte_flow_action actions[],
4135 		   struct rte_flow_action actions_rx[],
4136 		   struct rte_flow_action actions_tx[],
4137 		   struct rte_flow_item pattern_tx[],
4138 		   uint32_t flow_id)
4139 {
4140 	const struct rte_flow_action_raw_encap *raw_encap;
4141 	const struct rte_flow_action_raw_decap *raw_decap;
4142 	struct mlx5_rte_flow_action_set_tag *set_tag;
4143 	struct rte_flow_action *tag_action;
4144 	struct mlx5_rte_flow_item_tag *tag_item;
4145 	struct rte_flow_item *item;
4146 	char *addr;
4147 	int encap = 0;
4148 
4149 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4150 		switch (actions->type) {
4151 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4152 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4153 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4154 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4155 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4156 			rte_memcpy(actions_tx, actions,
4157 			       sizeof(struct rte_flow_action));
4158 			actions_tx++;
4159 			break;
4160 		case RTE_FLOW_ACTION_TYPE_COUNT:
4161 			if (encap) {
4162 				rte_memcpy(actions_tx, actions,
4163 					   sizeof(struct rte_flow_action));
4164 				actions_tx++;
4165 			} else {
4166 				rte_memcpy(actions_rx, actions,
4167 					   sizeof(struct rte_flow_action));
4168 				actions_rx++;
4169 			}
4170 			break;
4171 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4172 			raw_encap = actions->conf;
4173 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4174 				memcpy(actions_tx, actions,
4175 				       sizeof(struct rte_flow_action));
4176 				actions_tx++;
4177 				encap = 1;
4178 			} else {
4179 				rte_memcpy(actions_rx, actions,
4180 					   sizeof(struct rte_flow_action));
4181 				actions_rx++;
4182 			}
4183 			break;
4184 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4185 			raw_decap = actions->conf;
4186 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4187 				memcpy(actions_tx, actions,
4188 				       sizeof(struct rte_flow_action));
4189 				actions_tx++;
4190 			} else {
4191 				rte_memcpy(actions_rx, actions,
4192 					   sizeof(struct rte_flow_action));
4193 				actions_rx++;
4194 			}
4195 			break;
4196 		default:
4197 			rte_memcpy(actions_rx, actions,
4198 				   sizeof(struct rte_flow_action));
4199 			actions_rx++;
4200 			break;
4201 		}
4202 	}
4203 	/* Add set meta action and end action for the Rx flow. */
4204 	tag_action = actions_rx;
4205 	tag_action->type = (enum rte_flow_action_type)
4206 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4207 	actions_rx++;
4208 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4209 	actions_rx++;
4210 	set_tag = (void *)actions_rx;
4211 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4212 	MLX5_ASSERT(set_tag->id > REG_NON);
4213 	set_tag->data = flow_id;
4214 	tag_action->conf = set_tag;
4215 	/* Create Tx item list. */
4216 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4217 	addr = (void *)&pattern_tx[2];
4218 	item = pattern_tx;
4219 	item->type = (enum rte_flow_item_type)
4220 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4221 	tag_item = (void *)addr;
4222 	tag_item->data = flow_id;
4223 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4224 	MLX5_ASSERT(set_tag->id > REG_NON);
4225 	item->spec = tag_item;
4226 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4227 	tag_item = (void *)addr;
4228 	tag_item->data = UINT32_MAX;
4229 	tag_item->id = UINT16_MAX;
4230 	item->mask = tag_item;
4231 	item->last = NULL;
4232 	item++;
4233 	item->type = RTE_FLOW_ITEM_TYPE_END;
4234 	return 0;
4235 }
4236 
4237 /**
4238  * The last stage of splitting chain, just creates the subflow
4239  * without any modification.
4240  *
4241  * @param[in] dev
4242  *   Pointer to Ethernet device.
4243  * @param[in] flow
4244  *   Parent flow structure pointer.
4245  * @param[in, out] sub_flow
4246  *   Pointer to return the created subflow, may be NULL.
4247  * @param[in] attr
4248  *   Flow rule attributes.
4249  * @param[in] items
4250  *   Pattern specification (list terminated by the END pattern item).
4251  * @param[in] actions
4252  *   Associated actions (list terminated by the END action).
4253  * @param[in] flow_split_info
4254  *   Pointer to flow split info structure.
4255  * @param[out] error
4256  *   Perform verbose error reporting if not NULL.
4257  * @return
4258  *   0 on success, negative value otherwise
4259  */
4260 static int
4261 flow_create_split_inner(struct rte_eth_dev *dev,
4262 			struct rte_flow *flow,
4263 			struct mlx5_flow **sub_flow,
4264 			const struct rte_flow_attr *attr,
4265 			const struct rte_flow_item items[],
4266 			const struct rte_flow_action actions[],
4267 			struct mlx5_flow_split_info *flow_split_info,
4268 			struct rte_flow_error *error)
4269 {
4270 	struct mlx5_flow *dev_flow;
4271 
4272 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4273 				    flow_split_info->flow_idx, error);
4274 	if (!dev_flow)
4275 		return -rte_errno;
4276 	dev_flow->flow = flow;
4277 	dev_flow->external = flow_split_info->external;
4278 	dev_flow->skip_scale = flow_split_info->skip_scale;
4279 	/* Subflow object was created, we must include one in the list. */
4280 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4281 		      dev_flow->handle, next);
4282 	/*
4283 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4284 	 * flow may need some user defined item layer flags, and pass the
4285 	 * Metadate rxq mark flag to suffix flow as well.
4286 	 */
4287 	if (flow_split_info->prefix_layers)
4288 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4289 	if (flow_split_info->prefix_mark)
4290 		dev_flow->handle->mark = 1;
4291 	if (sub_flow)
4292 		*sub_flow = dev_flow;
4293 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4294 }
4295 
4296 /**
4297  * Split the meter flow.
4298  *
4299  * As meter flow will split to three sub flow, other than meter
4300  * action, the other actions make sense to only meter accepts
4301  * the packet. If it need to be dropped, no other additional
4302  * actions should be take.
4303  *
4304  * One kind of special action which decapsulates the L3 tunnel
4305  * header will be in the prefix sub flow, as not to take the
4306  * L3 tunnel header into account.
4307  *
4308  * @param dev
4309  *   Pointer to Ethernet device.
4310  * @param[in] items
4311  *   Pattern specification (list terminated by the END pattern item).
4312  * @param[out] sfx_items
4313  *   Suffix flow match items (list terminated by the END pattern item).
4314  * @param[in] actions
4315  *   Associated actions (list terminated by the END action).
4316  * @param[out] actions_sfx
4317  *   Suffix flow actions.
4318  * @param[out] actions_pre
4319  *   Prefix flow actions.
4320  * @param[out] pattern_sfx
4321  *   The pattern items for the suffix flow.
4322  * @param[out] tag_sfx
4323  *   Pointer to suffix flow tag.
4324  *
4325  * @return
4326  *   0 on success.
4327  */
4328 static int
4329 flow_meter_split_prep(struct rte_eth_dev *dev,
4330 		 const struct rte_flow_item items[],
4331 		 struct rte_flow_item sfx_items[],
4332 		 const struct rte_flow_action actions[],
4333 		 struct rte_flow_action actions_sfx[],
4334 		 struct rte_flow_action actions_pre[])
4335 {
4336 	struct mlx5_priv *priv = dev->data->dev_private;
4337 	struct rte_flow_action *tag_action = NULL;
4338 	struct rte_flow_item *tag_item;
4339 	struct mlx5_rte_flow_action_set_tag *set_tag;
4340 	struct rte_flow_error error;
4341 	const struct rte_flow_action_raw_encap *raw_encap;
4342 	const struct rte_flow_action_raw_decap *raw_decap;
4343 	struct mlx5_rte_flow_item_tag *tag_spec;
4344 	struct mlx5_rte_flow_item_tag *tag_mask;
4345 	uint32_t tag_id = 0;
4346 	bool copy_vlan = false;
4347 
4348 	/* Prepare the actions for prefix and suffix flow. */
4349 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4350 		struct rte_flow_action **action_cur = NULL;
4351 
4352 		switch (actions->type) {
4353 		case RTE_FLOW_ACTION_TYPE_METER:
4354 			/* Add the extra tag action first. */
4355 			tag_action = actions_pre;
4356 			tag_action->type = (enum rte_flow_action_type)
4357 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4358 			actions_pre++;
4359 			action_cur = &actions_pre;
4360 			break;
4361 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4362 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4363 			action_cur = &actions_pre;
4364 			break;
4365 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4366 			raw_encap = actions->conf;
4367 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4368 				action_cur = &actions_pre;
4369 			break;
4370 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4371 			raw_decap = actions->conf;
4372 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4373 				action_cur = &actions_pre;
4374 			break;
4375 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4376 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4377 			copy_vlan = true;
4378 			break;
4379 		default:
4380 			break;
4381 		}
4382 		if (!action_cur)
4383 			action_cur = &actions_sfx;
4384 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4385 		(*action_cur)++;
4386 	}
4387 	/* Add end action to the actions. */
4388 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4389 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4390 	actions_pre++;
4391 	/* Set the tag. */
4392 	set_tag = (void *)actions_pre;
4393 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4394 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4395 			  &tag_id);
4396 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4397 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4398 			dev->data->port_id);
4399 		mlx5_ipool_free(priv->sh->ipool
4400 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4401 		return 0;
4402 	} else if (!tag_id) {
4403 		return 0;
4404 	}
4405 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4406 	assert(tag_action);
4407 	tag_action->conf = set_tag;
4408 	/* Prepare the suffix subflow items. */
4409 	tag_item = sfx_items++;
4410 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4411 		int item_type = items->type;
4412 
4413 		switch (item_type) {
4414 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4415 			memcpy(sfx_items, items, sizeof(*sfx_items));
4416 			sfx_items++;
4417 			break;
4418 		case RTE_FLOW_ITEM_TYPE_VLAN:
4419 			if (copy_vlan) {
4420 				memcpy(sfx_items, items, sizeof(*sfx_items));
4421 				/*
4422 				 * Convert to internal match item, it is used
4423 				 * for vlan push and set vid.
4424 				 */
4425 				sfx_items->type = (enum rte_flow_item_type)
4426 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4427 				sfx_items++;
4428 			}
4429 			break;
4430 		default:
4431 			break;
4432 		}
4433 	}
4434 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4435 	sfx_items++;
4436 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4437 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4438 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4439 	tag_mask = tag_spec + 1;
4440 	tag_mask->data = 0xffffff00;
4441 	tag_item->type = (enum rte_flow_item_type)
4442 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4443 	tag_item->spec = tag_spec;
4444 	tag_item->last = NULL;
4445 	tag_item->mask = tag_mask;
4446 	return tag_id;
4447 }
4448 
4449 /**
4450  * Split action list having QUEUE/RSS for metadata register copy.
4451  *
4452  * Once Q/RSS action is detected in user's action list, the flow action
4453  * should be split in order to copy metadata registers, which will happen in
4454  * RX_CP_TBL like,
4455  *   - CQE->flow_tag := reg_c[1] (MARK)
4456  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4457  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4458  * This is because the last action of each flow must be a terminal action
4459  * (QUEUE, RSS or DROP).
4460  *
4461  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4462  * stored and kept in the mlx5_flow structure per each sub_flow.
4463  *
4464  * The Q/RSS action is replaced with,
4465  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4466  * And the following JUMP action is added at the end,
4467  *   - JUMP, to RX_CP_TBL.
4468  *
4469  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4470  * flow_create_split_metadata() routine. The flow will look like,
4471  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4472  *
4473  * @param dev
4474  *   Pointer to Ethernet device.
4475  * @param[out] split_actions
4476  *   Pointer to store split actions to jump to CP_TBL.
4477  * @param[in] actions
4478  *   Pointer to the list of original flow actions.
4479  * @param[in] qrss
4480  *   Pointer to the Q/RSS action.
4481  * @param[in] actions_n
4482  *   Number of original actions.
4483  * @param[out] error
4484  *   Perform verbose error reporting if not NULL.
4485  *
4486  * @return
4487  *   non-zero unique flow_id on success, otherwise 0 and
4488  *   error/rte_error are set.
4489  */
4490 static uint32_t
4491 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4492 			  struct rte_flow_action *split_actions,
4493 			  const struct rte_flow_action *actions,
4494 			  const struct rte_flow_action *qrss,
4495 			  int actions_n, struct rte_flow_error *error)
4496 {
4497 	struct mlx5_priv *priv = dev->data->dev_private;
4498 	struct mlx5_rte_flow_action_set_tag *set_tag;
4499 	struct rte_flow_action_jump *jump;
4500 	const int qrss_idx = qrss - actions;
4501 	uint32_t flow_id = 0;
4502 	int ret = 0;
4503 
4504 	/*
4505 	 * Given actions will be split
4506 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4507 	 * - Add jump to mreg CP_TBL.
4508 	 * As a result, there will be one more action.
4509 	 */
4510 	++actions_n;
4511 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4512 	set_tag = (void *)(split_actions + actions_n);
4513 	/*
4514 	 * If tag action is not set to void(it means we are not the meter
4515 	 * suffix flow), add the tag action. Since meter suffix flow already
4516 	 * has the tag added.
4517 	 */
4518 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4519 		/*
4520 		 * Allocate the new subflow ID. This one is unique within
4521 		 * device and not shared with representors. Otherwise,
4522 		 * we would have to resolve multi-thread access synch
4523 		 * issue. Each flow on the shared device is appended
4524 		 * with source vport identifier, so the resulting
4525 		 * flows will be unique in the shared (by master and
4526 		 * representors) domain even if they have coinciding
4527 		 * IDs.
4528 		 */
4529 		mlx5_ipool_malloc(priv->sh->ipool
4530 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4531 		if (!flow_id)
4532 			return rte_flow_error_set(error, ENOMEM,
4533 						  RTE_FLOW_ERROR_TYPE_ACTION,
4534 						  NULL, "can't allocate id "
4535 						  "for split Q/RSS subflow");
4536 		/* Internal SET_TAG action to set flow ID. */
4537 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4538 			.data = flow_id,
4539 		};
4540 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4541 		if (ret < 0)
4542 			return ret;
4543 		set_tag->id = ret;
4544 		/* Construct new actions array. */
4545 		/* Replace QUEUE/RSS action. */
4546 		split_actions[qrss_idx] = (struct rte_flow_action){
4547 			.type = (enum rte_flow_action_type)
4548 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4549 			.conf = set_tag,
4550 		};
4551 	}
4552 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4553 	jump = (void *)(set_tag + 1);
4554 	*jump = (struct rte_flow_action_jump){
4555 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4556 	};
4557 	split_actions[actions_n - 2] = (struct rte_flow_action){
4558 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4559 		.conf = jump,
4560 	};
4561 	split_actions[actions_n - 1] = (struct rte_flow_action){
4562 		.type = RTE_FLOW_ACTION_TYPE_END,
4563 	};
4564 	return flow_id;
4565 }
4566 
4567 /**
4568  * Extend the given action list for Tx metadata copy.
4569  *
4570  * Copy the given action list to the ext_actions and add flow metadata register
4571  * copy action in order to copy reg_a set by WQE to reg_c[0].
4572  *
4573  * @param[out] ext_actions
4574  *   Pointer to the extended action list.
4575  * @param[in] actions
4576  *   Pointer to the list of actions.
4577  * @param[in] actions_n
4578  *   Number of actions in the list.
4579  * @param[out] error
4580  *   Perform verbose error reporting if not NULL.
4581  * @param[in] encap_idx
4582  *   The encap action inndex.
4583  *
4584  * @return
4585  *   0 on success, negative value otherwise
4586  */
4587 static int
4588 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4589 		       struct rte_flow_action *ext_actions,
4590 		       const struct rte_flow_action *actions,
4591 		       int actions_n, struct rte_flow_error *error,
4592 		       int encap_idx)
4593 {
4594 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4595 		(struct mlx5_flow_action_copy_mreg *)
4596 			(ext_actions + actions_n + 1);
4597 	int ret;
4598 
4599 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4600 	if (ret < 0)
4601 		return ret;
4602 	cp_mreg->dst = ret;
4603 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4604 	if (ret < 0)
4605 		return ret;
4606 	cp_mreg->src = ret;
4607 	if (encap_idx != 0)
4608 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4609 	if (encap_idx == actions_n - 1) {
4610 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4611 			.type = (enum rte_flow_action_type)
4612 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4613 			.conf = cp_mreg,
4614 		};
4615 		ext_actions[actions_n] = (struct rte_flow_action){
4616 			.type = RTE_FLOW_ACTION_TYPE_END,
4617 		};
4618 	} else {
4619 		ext_actions[encap_idx] = (struct rte_flow_action){
4620 			.type = (enum rte_flow_action_type)
4621 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4622 			.conf = cp_mreg,
4623 		};
4624 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4625 				sizeof(*ext_actions) * (actions_n - encap_idx));
4626 	}
4627 	return 0;
4628 }
4629 
4630 /**
4631  * Check the match action from the action list.
4632  *
4633  * @param[in] actions
4634  *   Pointer to the list of actions.
4635  * @param[in] attr
4636  *   Flow rule attributes.
4637  * @param[in] action
4638  *   The action to be check if exist.
4639  * @param[out] match_action_pos
4640  *   Pointer to the position of the matched action if exists, otherwise is -1.
4641  * @param[out] qrss_action_pos
4642  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4643  *
4644  * @return
4645  *   > 0 the total number of actions.
4646  *   0 if not found match action in action list.
4647  */
4648 static int
4649 flow_check_match_action(const struct rte_flow_action actions[],
4650 			const struct rte_flow_attr *attr,
4651 			enum rte_flow_action_type action,
4652 			int *match_action_pos, int *qrss_action_pos)
4653 {
4654 	const struct rte_flow_action_sample *sample;
4655 	int actions_n = 0;
4656 	uint32_t ratio = 0;
4657 	int sub_type = 0;
4658 	int flag = 0;
4659 
4660 	*match_action_pos = -1;
4661 	*qrss_action_pos = -1;
4662 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4663 		if (actions->type == action) {
4664 			flag = 1;
4665 			*match_action_pos = actions_n;
4666 		}
4667 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4668 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4669 			*qrss_action_pos = actions_n;
4670 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4671 			sample = actions->conf;
4672 			ratio = sample->ratio;
4673 			sub_type = ((const struct rte_flow_action *)
4674 					(sample->actions))->type;
4675 		}
4676 		actions_n++;
4677 	}
4678 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4679 		if (ratio == 1) {
4680 			/* FDB mirroring uses the destination array to implement
4681 			 * instead of FLOW_SAMPLER object.
4682 			 */
4683 			if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4684 				flag = 0;
4685 		}
4686 	}
4687 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4688 	return flag ? actions_n + 1 : 0;
4689 }
4690 
4691 #define SAMPLE_SUFFIX_ITEM 2
4692 
4693 /**
4694  * Split the sample flow.
4695  *
4696  * As sample flow will split to two sub flow, sample flow with
4697  * sample action, the other actions will move to new suffix flow.
4698  *
4699  * Also add unique tag id with tag action in the sample flow,
4700  * the same tag id will be as match in the suffix flow.
4701  *
4702  * @param dev
4703  *   Pointer to Ethernet device.
4704  * @param[in] add_tag
4705  *   Add extra tag action flag.
4706  * @param[out] sfx_items
4707  *   Suffix flow match items (list terminated by the END pattern item).
4708  * @param[in] actions
4709  *   Associated actions (list terminated by the END action).
4710  * @param[out] actions_sfx
4711  *   Suffix flow actions.
4712  * @param[out] actions_pre
4713  *   Prefix flow actions.
4714  * @param[in] actions_n
4715  *  The total number of actions.
4716  * @param[in] sample_action_pos
4717  *   The sample action position.
4718  * @param[in] qrss_action_pos
4719  *   The Queue/RSS action position.
4720  * @param[out] error
4721  *   Perform verbose error reporting if not NULL.
4722  *
4723  * @return
4724  *   0 on success, or unique flow_id, a negative errno value
4725  *   otherwise and rte_errno is set.
4726  */
4727 static int
4728 flow_sample_split_prep(struct rte_eth_dev *dev,
4729 		       int add_tag,
4730 		       struct rte_flow_item sfx_items[],
4731 		       const struct rte_flow_action actions[],
4732 		       struct rte_flow_action actions_sfx[],
4733 		       struct rte_flow_action actions_pre[],
4734 		       int actions_n,
4735 		       int sample_action_pos,
4736 		       int qrss_action_pos,
4737 		       struct rte_flow_error *error)
4738 {
4739 	struct mlx5_priv *priv = dev->data->dev_private;
4740 	struct mlx5_rte_flow_action_set_tag *set_tag;
4741 	struct mlx5_rte_flow_item_tag *tag_spec;
4742 	struct mlx5_rte_flow_item_tag *tag_mask;
4743 	uint32_t tag_id = 0;
4744 	int index;
4745 	int ret;
4746 
4747 	if (sample_action_pos < 0)
4748 		return rte_flow_error_set(error, EINVAL,
4749 					  RTE_FLOW_ERROR_TYPE_ACTION,
4750 					  NULL, "invalid position of sample "
4751 					  "action in list");
4752 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4753 	 * For CX6DX and above, metadata registers Cx preserve their value,
4754 	 * add an extra tag action for NIC-RX and E-Switch ingress and egress.
4755 	 */
4756 	if (add_tag) {
4757 		/* Prepare the prefix tag action. */
4758 		set_tag = (void *)(actions_pre + actions_n + 1);
4759 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4760 		if (ret < 0)
4761 			return ret;
4762 		set_tag->id = ret;
4763 		mlx5_ipool_malloc(priv->sh->ipool
4764 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4765 		set_tag->data = tag_id;
4766 		/* Prepare the suffix subflow items. */
4767 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4768 		tag_spec->data = tag_id;
4769 		tag_spec->id = set_tag->id;
4770 		tag_mask = tag_spec + 1;
4771 		tag_mask->data = UINT32_MAX;
4772 		sfx_items[0] = (struct rte_flow_item){
4773 			.type = (enum rte_flow_item_type)
4774 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4775 			.spec = tag_spec,
4776 			.last = NULL,
4777 			.mask = tag_mask,
4778 		};
4779 		sfx_items[1] = (struct rte_flow_item){
4780 			.type = (enum rte_flow_item_type)
4781 				RTE_FLOW_ITEM_TYPE_END,
4782 		};
4783 	}
4784 	/* Prepare the actions for prefix and suffix flow. */
4785 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4786 		index = qrss_action_pos;
4787 		/* Put the preceding the Queue/RSS action into prefix flow. */
4788 		if (index != 0)
4789 			memcpy(actions_pre, actions,
4790 			       sizeof(struct rte_flow_action) * index);
4791 		/* Put others preceding the sample action into prefix flow. */
4792 		if (sample_action_pos > index + 1)
4793 			memcpy(actions_pre + index, actions + index + 1,
4794 			       sizeof(struct rte_flow_action) *
4795 			       (sample_action_pos - index - 1));
4796 		index = sample_action_pos - 1;
4797 		/* Put Queue/RSS action into Suffix flow. */
4798 		memcpy(actions_sfx, actions + qrss_action_pos,
4799 		       sizeof(struct rte_flow_action));
4800 		actions_sfx++;
4801 	} else {
4802 		index = sample_action_pos;
4803 		if (index != 0)
4804 			memcpy(actions_pre, actions,
4805 			       sizeof(struct rte_flow_action) * index);
4806 	}
4807 	if (add_tag) {
4808 		actions_pre[index++] =
4809 			(struct rte_flow_action){
4810 			.type = (enum rte_flow_action_type)
4811 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4812 			.conf = set_tag,
4813 		};
4814 	}
4815 	memcpy(actions_pre + index, actions + sample_action_pos,
4816 	       sizeof(struct rte_flow_action));
4817 	index += 1;
4818 	actions_pre[index] = (struct rte_flow_action){
4819 		.type = (enum rte_flow_action_type)
4820 			RTE_FLOW_ACTION_TYPE_END,
4821 	};
4822 	/* Put the actions after sample into Suffix flow. */
4823 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4824 	       sizeof(struct rte_flow_action) *
4825 	       (actions_n - sample_action_pos - 1));
4826 	return tag_id;
4827 }
4828 
4829 /**
4830  * The splitting for metadata feature.
4831  *
4832  * - Q/RSS action on NIC Rx should be split in order to pass by
4833  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4834  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4835  *
4836  * - All the actions on NIC Tx should have a mreg copy action to
4837  *   copy reg_a from WQE to reg_c[0].
4838  *
4839  * @param dev
4840  *   Pointer to Ethernet device.
4841  * @param[in] flow
4842  *   Parent flow structure pointer.
4843  * @param[in] attr
4844  *   Flow rule attributes.
4845  * @param[in] items
4846  *   Pattern specification (list terminated by the END pattern item).
4847  * @param[in] actions
4848  *   Associated actions (list terminated by the END action).
4849  * @param[in] flow_split_info
4850  *   Pointer to flow split info structure.
4851  * @param[out] error
4852  *   Perform verbose error reporting if not NULL.
4853  * @return
4854  *   0 on success, negative value otherwise
4855  */
4856 static int
4857 flow_create_split_metadata(struct rte_eth_dev *dev,
4858 			   struct rte_flow *flow,
4859 			   const struct rte_flow_attr *attr,
4860 			   const struct rte_flow_item items[],
4861 			   const struct rte_flow_action actions[],
4862 			   struct mlx5_flow_split_info *flow_split_info,
4863 			   struct rte_flow_error *error)
4864 {
4865 	struct mlx5_priv *priv = dev->data->dev_private;
4866 	struct mlx5_dev_config *config = &priv->config;
4867 	const struct rte_flow_action *qrss = NULL;
4868 	struct rte_flow_action *ext_actions = NULL;
4869 	struct mlx5_flow *dev_flow = NULL;
4870 	uint32_t qrss_id = 0;
4871 	int mtr_sfx = 0;
4872 	size_t act_size;
4873 	int actions_n;
4874 	int encap_idx;
4875 	int ret;
4876 
4877 	/* Check whether extensive metadata feature is engaged. */
4878 	if (!config->dv_flow_en ||
4879 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4880 	    !mlx5_flow_ext_mreg_supported(dev))
4881 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4882 					       actions, flow_split_info, error);
4883 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4884 							   &encap_idx);
4885 	if (qrss) {
4886 		/* Exclude hairpin flows from splitting. */
4887 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4888 			const struct rte_flow_action_queue *queue;
4889 
4890 			queue = qrss->conf;
4891 			if (mlx5_rxq_get_type(dev, queue->index) ==
4892 			    MLX5_RXQ_TYPE_HAIRPIN)
4893 				qrss = NULL;
4894 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4895 			const struct rte_flow_action_rss *rss;
4896 
4897 			rss = qrss->conf;
4898 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4899 			    MLX5_RXQ_TYPE_HAIRPIN)
4900 				qrss = NULL;
4901 		}
4902 	}
4903 	if (qrss) {
4904 		/* Check if it is in meter suffix table. */
4905 		mtr_sfx = attr->group == (attr->transfer ?
4906 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4907 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4908 		/*
4909 		 * Q/RSS action on NIC Rx should be split in order to pass by
4910 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4911 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4912 		 */
4913 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4914 			   sizeof(struct rte_flow_action_set_tag) +
4915 			   sizeof(struct rte_flow_action_jump);
4916 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4917 					  SOCKET_ID_ANY);
4918 		if (!ext_actions)
4919 			return rte_flow_error_set(error, ENOMEM,
4920 						  RTE_FLOW_ERROR_TYPE_ACTION,
4921 						  NULL, "no memory to split "
4922 						  "metadata flow");
4923 		/*
4924 		 * If we are the suffix flow of meter, tag already exist.
4925 		 * Set the tag action to void.
4926 		 */
4927 		if (mtr_sfx)
4928 			ext_actions[qrss - actions].type =
4929 						RTE_FLOW_ACTION_TYPE_VOID;
4930 		else
4931 			ext_actions[qrss - actions].type =
4932 						(enum rte_flow_action_type)
4933 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4934 		/*
4935 		 * Create the new actions list with removed Q/RSS action
4936 		 * and appended set tag and jump to register copy table
4937 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4938 		 * in advance, because it is needed for set tag action.
4939 		 */
4940 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4941 						    qrss, actions_n, error);
4942 		if (!mtr_sfx && !qrss_id) {
4943 			ret = -rte_errno;
4944 			goto exit;
4945 		}
4946 	} else if (attr->egress && !attr->transfer) {
4947 		/*
4948 		 * All the actions on NIC Tx should have a metadata register
4949 		 * copy action to copy reg_a from WQE to reg_c[meta]
4950 		 */
4951 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4952 			   sizeof(struct mlx5_flow_action_copy_mreg);
4953 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4954 					  SOCKET_ID_ANY);
4955 		if (!ext_actions)
4956 			return rte_flow_error_set(error, ENOMEM,
4957 						  RTE_FLOW_ERROR_TYPE_ACTION,
4958 						  NULL, "no memory to split "
4959 						  "metadata flow");
4960 		/* Create the action list appended with copy register. */
4961 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4962 					     actions_n, error, encap_idx);
4963 		if (ret < 0)
4964 			goto exit;
4965 	}
4966 	/* Add the unmodified original or prefix subflow. */
4967 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
4968 				      items, ext_actions ? ext_actions :
4969 				      actions, flow_split_info, error);
4970 	if (ret < 0)
4971 		goto exit;
4972 	MLX5_ASSERT(dev_flow);
4973 	if (qrss) {
4974 		const struct rte_flow_attr q_attr = {
4975 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4976 			.ingress = 1,
4977 		};
4978 		/* Internal PMD action to set register. */
4979 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4980 			.data = qrss_id,
4981 			.id = REG_NON,
4982 		};
4983 		struct rte_flow_item q_items[] = {
4984 			{
4985 				.type = (enum rte_flow_item_type)
4986 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4987 				.spec = &q_tag_spec,
4988 				.last = NULL,
4989 				.mask = NULL,
4990 			},
4991 			{
4992 				.type = RTE_FLOW_ITEM_TYPE_END,
4993 			},
4994 		};
4995 		struct rte_flow_action q_actions[] = {
4996 			{
4997 				.type = qrss->type,
4998 				.conf = qrss->conf,
4999 			},
5000 			{
5001 				.type = RTE_FLOW_ACTION_TYPE_END,
5002 			},
5003 		};
5004 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5005 
5006 		/*
5007 		 * Configure the tag item only if there is no meter subflow.
5008 		 * Since tag is already marked in the meter suffix subflow
5009 		 * we can just use the meter suffix items as is.
5010 		 */
5011 		if (qrss_id) {
5012 			/* Not meter subflow. */
5013 			MLX5_ASSERT(!mtr_sfx);
5014 			/*
5015 			 * Put unique id in prefix flow due to it is destroyed
5016 			 * after suffix flow and id will be freed after there
5017 			 * is no actual flows with this id and identifier
5018 			 * reallocation becomes possible (for example, for
5019 			 * other flows in other threads).
5020 			 */
5021 			dev_flow->handle->split_flow_id = qrss_id;
5022 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5023 						   error);
5024 			if (ret < 0)
5025 				goto exit;
5026 			q_tag_spec.id = ret;
5027 		}
5028 		dev_flow = NULL;
5029 		/* Add suffix subflow to execute Q/RSS. */
5030 		flow_split_info->prefix_layers = layers;
5031 		flow_split_info->prefix_mark = 0;
5032 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5033 					      &q_attr, mtr_sfx ? items :
5034 					      q_items, q_actions,
5035 					      flow_split_info, error);
5036 		if (ret < 0)
5037 			goto exit;
5038 		/* qrss ID should be freed if failed. */
5039 		qrss_id = 0;
5040 		MLX5_ASSERT(dev_flow);
5041 	}
5042 
5043 exit:
5044 	/*
5045 	 * We do not destroy the partially created sub_flows in case of error.
5046 	 * These ones are included into parent flow list and will be destroyed
5047 	 * by flow_drv_destroy.
5048 	 */
5049 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5050 			qrss_id);
5051 	mlx5_free(ext_actions);
5052 	return ret;
5053 }
5054 
5055 /**
5056  * The splitting for meter feature.
5057  *
5058  * - The meter flow will be split to two flows as prefix and
5059  *   suffix flow. The packets make sense only it pass the prefix
5060  *   meter action.
5061  *
5062  * - Reg_C_5 is used for the packet to match betweend prefix and
5063  *   suffix flow.
5064  *
5065  * @param dev
5066  *   Pointer to Ethernet device.
5067  * @param[in] flow
5068  *   Parent flow structure pointer.
5069  * @param[in] attr
5070  *   Flow rule attributes.
5071  * @param[in] items
5072  *   Pattern specification (list terminated by the END pattern item).
5073  * @param[in] actions
5074  *   Associated actions (list terminated by the END action).
5075  * @param[in] flow_split_info
5076  *   Pointer to flow split info structure.
5077  * @param[out] error
5078  *   Perform verbose error reporting if not NULL.
5079  * @return
5080  *   0 on success, negative value otherwise
5081  */
5082 static int
5083 flow_create_split_meter(struct rte_eth_dev *dev,
5084 			struct rte_flow *flow,
5085 			const struct rte_flow_attr *attr,
5086 			const struct rte_flow_item items[],
5087 			const struct rte_flow_action actions[],
5088 			struct mlx5_flow_split_info *flow_split_info,
5089 			struct rte_flow_error *error)
5090 {
5091 	struct mlx5_priv *priv = dev->data->dev_private;
5092 	struct rte_flow_action *sfx_actions = NULL;
5093 	struct rte_flow_action *pre_actions = NULL;
5094 	struct rte_flow_item *sfx_items = NULL;
5095 	struct mlx5_flow *dev_flow = NULL;
5096 	struct rte_flow_attr sfx_attr = *attr;
5097 	uint32_t mtr = 0;
5098 	uint32_t mtr_tag_id = 0;
5099 	size_t act_size;
5100 	size_t item_size;
5101 	int actions_n = 0;
5102 	int ret;
5103 
5104 	if (priv->mtr_en)
5105 		actions_n = flow_check_meter_action(actions, &mtr);
5106 	if (mtr) {
5107 		/* The five prefix actions: meter, decap, encap, tag, end. */
5108 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5109 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5110 		/* tag, vlan, port id, end. */
5111 #define METER_SUFFIX_ITEM 4
5112 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5113 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5114 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5115 					  0, SOCKET_ID_ANY);
5116 		if (!sfx_actions)
5117 			return rte_flow_error_set(error, ENOMEM,
5118 						  RTE_FLOW_ERROR_TYPE_ACTION,
5119 						  NULL, "no memory to split "
5120 						  "meter flow");
5121 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5122 			     act_size);
5123 		pre_actions = sfx_actions + actions_n;
5124 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5125 						   actions, sfx_actions,
5126 						   pre_actions);
5127 		if (!mtr_tag_id) {
5128 			ret = -rte_errno;
5129 			goto exit;
5130 		}
5131 		/* Add the prefix subflow. */
5132 		flow_split_info->prefix_mark = 0;
5133 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5134 					      attr, items, pre_actions,
5135 					      flow_split_info, error);
5136 		if (ret) {
5137 			ret = -rte_errno;
5138 			goto exit;
5139 		}
5140 		dev_flow->handle->split_flow_id = mtr_tag_id;
5141 		/* Setting the sfx group atrr. */
5142 		sfx_attr.group = sfx_attr.transfer ?
5143 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5144 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5145 		flow_split_info->prefix_layers =
5146 				flow_get_prefix_layer_flags(dev_flow);
5147 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5148 	}
5149 	/* Add the prefix subflow. */
5150 	ret = flow_create_split_metadata(dev, flow,
5151 					 &sfx_attr, sfx_items ?
5152 					 sfx_items : items,
5153 					 sfx_actions ? sfx_actions : actions,
5154 					 flow_split_info, error);
5155 exit:
5156 	if (sfx_actions)
5157 		mlx5_free(sfx_actions);
5158 	return ret;
5159 }
5160 
5161 /**
5162  * The splitting for sample feature.
5163  *
5164  * Once Sample action is detected in the action list, the flow actions should
5165  * be split into prefix sub flow and suffix sub flow.
5166  *
5167  * The original items remain in the prefix sub flow, all actions preceding the
5168  * sample action and the sample action itself will be copied to the prefix
5169  * sub flow, the actions following the sample action will be copied to the
5170  * suffix sub flow, Queue action always be located in the suffix sub flow.
5171  *
5172  * In order to make the packet from prefix sub flow matches with suffix sub
5173  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5174  * flow uses tag item with the unique flow id.
5175  *
5176  * @param dev
5177  *   Pointer to Ethernet device.
5178  * @param[in] flow
5179  *   Parent flow structure pointer.
5180  * @param[in] attr
5181  *   Flow rule attributes.
5182  * @param[in] items
5183  *   Pattern specification (list terminated by the END pattern item).
5184  * @param[in] actions
5185  *   Associated actions (list terminated by the END action).
5186  * @param[in] flow_split_info
5187  *   Pointer to flow split info structure.
5188  * @param[out] error
5189  *   Perform verbose error reporting if not NULL.
5190  * @return
5191  *   0 on success, negative value otherwise
5192  */
5193 static int
5194 flow_create_split_sample(struct rte_eth_dev *dev,
5195 			 struct rte_flow *flow,
5196 			 const struct rte_flow_attr *attr,
5197 			 const struct rte_flow_item items[],
5198 			 const struct rte_flow_action actions[],
5199 			 struct mlx5_flow_split_info *flow_split_info,
5200 			 struct rte_flow_error *error)
5201 {
5202 	struct mlx5_priv *priv = dev->data->dev_private;
5203 	struct rte_flow_action *sfx_actions = NULL;
5204 	struct rte_flow_action *pre_actions = NULL;
5205 	struct rte_flow_item *sfx_items = NULL;
5206 	struct mlx5_flow *dev_flow = NULL;
5207 	struct rte_flow_attr sfx_attr = *attr;
5208 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5209 	struct mlx5_flow_dv_sample_resource *sample_res;
5210 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5211 	struct mlx5_flow_tbl_resource *sfx_tbl;
5212 #endif
5213 	size_t act_size;
5214 	size_t item_size;
5215 	uint32_t fdb_tx = 0;
5216 	int32_t tag_id = 0;
5217 	int actions_n = 0;
5218 	int sample_action_pos;
5219 	int qrss_action_pos;
5220 	int add_tag = 0;
5221 	int ret = 0;
5222 
5223 	if (priv->sampler_en)
5224 		actions_n = flow_check_match_action(actions, attr,
5225 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5226 					&sample_action_pos, &qrss_action_pos);
5227 	if (actions_n) {
5228 		/* The prefix actions must includes sample, tag, end. */
5229 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5230 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5231 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5232 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5233 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5234 					  item_size), 0, SOCKET_ID_ANY);
5235 		if (!sfx_actions)
5236 			return rte_flow_error_set(error, ENOMEM,
5237 						  RTE_FLOW_ERROR_TYPE_ACTION,
5238 						  NULL, "no memory to split "
5239 						  "sample flow");
5240 		/* The representor_id is -1 for uplink. */
5241 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5242 		/*
5243 		 * When reg_c_preserve is set, metadata registers Cx preserve
5244 		 * their value even through packet duplication.
5245 		 */
5246 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5247 		if (add_tag)
5248 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5249 					+ act_size);
5250 		pre_actions = sfx_actions + actions_n;
5251 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5252 						actions, sfx_actions,
5253 						pre_actions, actions_n,
5254 						sample_action_pos,
5255 						qrss_action_pos, error);
5256 		if (tag_id < 0 || (add_tag && !tag_id)) {
5257 			ret = -rte_errno;
5258 			goto exit;
5259 		}
5260 		/* Add the prefix subflow. */
5261 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5262 					      items, pre_actions,
5263 					      flow_split_info, error);
5264 		if (ret) {
5265 			ret = -rte_errno;
5266 			goto exit;
5267 		}
5268 		dev_flow->handle->split_flow_id = tag_id;
5269 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5270 		/* Set the sfx group attr. */
5271 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5272 					dev_flow->dv.sample_res;
5273 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5274 					sample_res->normal_path_tbl;
5275 		sfx_tbl_data = container_of(sfx_tbl,
5276 					struct mlx5_flow_tbl_data_entry, tbl);
5277 		sfx_attr.group = sfx_attr.transfer ?
5278 					(sfx_tbl_data->table_id - 1) :
5279 					 sfx_tbl_data->table_id;
5280 		flow_split_info->prefix_layers =
5281 				flow_get_prefix_layer_flags(dev_flow);
5282 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5283 		/* Suffix group level already be scaled with factor, set
5284 		 * skip_scale to 1 to avoid scale again in translation.
5285 		 */
5286 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5287 #endif
5288 	}
5289 	/* Add the suffix subflow. */
5290 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5291 				      sfx_items ? sfx_items : items,
5292 				      sfx_actions ? sfx_actions : actions,
5293 				      flow_split_info, error);
5294 exit:
5295 	if (sfx_actions)
5296 		mlx5_free(sfx_actions);
5297 	return ret;
5298 }
5299 
5300 /**
5301  * Split the flow to subflow set. The splitters might be linked
5302  * in the chain, like this:
5303  * flow_create_split_outer() calls:
5304  *   flow_create_split_meter() calls:
5305  *     flow_create_split_metadata(meter_subflow_0) calls:
5306  *       flow_create_split_inner(metadata_subflow_0)
5307  *       flow_create_split_inner(metadata_subflow_1)
5308  *       flow_create_split_inner(metadata_subflow_2)
5309  *     flow_create_split_metadata(meter_subflow_1) calls:
5310  *       flow_create_split_inner(metadata_subflow_0)
5311  *       flow_create_split_inner(metadata_subflow_1)
5312  *       flow_create_split_inner(metadata_subflow_2)
5313  *
5314  * This provide flexible way to add new levels of flow splitting.
5315  * The all of successfully created subflows are included to the
5316  * parent flow dev_flow list.
5317  *
5318  * @param dev
5319  *   Pointer to Ethernet device.
5320  * @param[in] flow
5321  *   Parent flow structure pointer.
5322  * @param[in] attr
5323  *   Flow rule attributes.
5324  * @param[in] items
5325  *   Pattern specification (list terminated by the END pattern item).
5326  * @param[in] actions
5327  *   Associated actions (list terminated by the END action).
5328  * @param[in] flow_split_info
5329  *   Pointer to flow split info structure.
5330  * @param[out] error
5331  *   Perform verbose error reporting if not NULL.
5332  * @return
5333  *   0 on success, negative value otherwise
5334  */
5335 static int
5336 flow_create_split_outer(struct rte_eth_dev *dev,
5337 			struct rte_flow *flow,
5338 			const struct rte_flow_attr *attr,
5339 			const struct rte_flow_item items[],
5340 			const struct rte_flow_action actions[],
5341 			struct mlx5_flow_split_info *flow_split_info,
5342 			struct rte_flow_error *error)
5343 {
5344 	int ret;
5345 
5346 	ret = flow_create_split_sample(dev, flow, attr, items,
5347 				       actions, flow_split_info, error);
5348 	MLX5_ASSERT(ret <= 0);
5349 	return ret;
5350 }
5351 
5352 static struct mlx5_flow_tunnel *
5353 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5354 		      const struct rte_flow_attr *attr,
5355 		      const struct rte_flow_item items[],
5356 		      const struct rte_flow_action actions[])
5357 {
5358 	struct mlx5_flow_tunnel *tunnel;
5359 
5360 #pragma GCC diagnostic push
5361 #pragma GCC diagnostic ignored "-Wcast-qual"
5362 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5363 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5364 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5365 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5366 	else
5367 		tunnel = NULL;
5368 #pragma GCC diagnostic pop
5369 
5370 	return tunnel;
5371 }
5372 
5373 /**
5374  * Adjust flow RSS workspace if needed.
5375  *
5376  * @param wks
5377  *   Pointer to thread flow work space.
5378  * @param rss_desc
5379  *   Pointer to RSS descriptor.
5380  * @param[in] nrssq_num
5381  *   New RSS queue number.
5382  *
5383  * @return
5384  *   0 on success, -1 otherwise and rte_errno is set.
5385  */
5386 static int
5387 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5388 			  struct mlx5_flow_rss_desc *rss_desc,
5389 			  uint32_t nrssq_num)
5390 {
5391 	if (likely(nrssq_num <= wks->rssq_num))
5392 		return 0;
5393 	rss_desc->queue = realloc(rss_desc->queue,
5394 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5395 	if (!rss_desc->queue) {
5396 		rte_errno = ENOMEM;
5397 		return -1;
5398 	}
5399 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5400 	return 0;
5401 }
5402 
5403 /**
5404  * Create a flow and add it to @p list.
5405  *
5406  * @param dev
5407  *   Pointer to Ethernet device.
5408  * @param list
5409  *   Pointer to a TAILQ flow list. If this parameter NULL,
5410  *   no list insertion occurred, flow is just created,
5411  *   this is caller's responsibility to track the
5412  *   created flow.
5413  * @param[in] attr
5414  *   Flow rule attributes.
5415  * @param[in] items
5416  *   Pattern specification (list terminated by the END pattern item).
5417  * @param[in] actions
5418  *   Associated actions (list terminated by the END action).
5419  * @param[in] external
5420  *   This flow rule is created by request external to PMD.
5421  * @param[out] error
5422  *   Perform verbose error reporting if not NULL.
5423  *
5424  * @return
5425  *   A flow index on success, 0 otherwise and rte_errno is set.
5426  */
5427 static uint32_t
5428 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5429 		 const struct rte_flow_attr *attr,
5430 		 const struct rte_flow_item items[],
5431 		 const struct rte_flow_action original_actions[],
5432 		 bool external, struct rte_flow_error *error)
5433 {
5434 	struct mlx5_priv *priv = dev->data->dev_private;
5435 	struct rte_flow *flow = NULL;
5436 	struct mlx5_flow *dev_flow;
5437 	const struct rte_flow_action_rss *rss = NULL;
5438 	struct mlx5_translated_shared_action
5439 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5440 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5441 	union {
5442 		struct mlx5_flow_expand_rss buf;
5443 		uint8_t buffer[2048];
5444 	} expand_buffer;
5445 	union {
5446 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5447 		uint8_t buffer[2048];
5448 	} actions_rx;
5449 	union {
5450 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5451 		uint8_t buffer[2048];
5452 	} actions_hairpin_tx;
5453 	union {
5454 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5455 		uint8_t buffer[2048];
5456 	} items_tx;
5457 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5458 	struct mlx5_flow_rss_desc *rss_desc;
5459 	const struct rte_flow_action *p_actions_rx;
5460 	uint32_t i;
5461 	uint32_t idx = 0;
5462 	int hairpin_flow;
5463 	struct rte_flow_attr attr_tx = { .priority = 0 };
5464 	const struct rte_flow_action *actions;
5465 	struct rte_flow_action *translated_actions = NULL;
5466 	struct mlx5_flow_tunnel *tunnel;
5467 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5468 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5469 	struct mlx5_flow_split_info flow_split_info = {
5470 		.external = !!external,
5471 		.skip_scale = 0,
5472 		.flow_idx = 0,
5473 		.prefix_mark = 0,
5474 		.prefix_layers = 0
5475 	};
5476 	int ret;
5477 
5478 	MLX5_ASSERT(wks);
5479 	rss_desc = &wks->rss_desc;
5480 	ret = flow_shared_actions_translate(dev, original_actions,
5481 					    shared_actions,
5482 					    &shared_actions_n,
5483 					    &translated_actions, error);
5484 	if (ret < 0) {
5485 		MLX5_ASSERT(translated_actions == NULL);
5486 		return 0;
5487 	}
5488 	actions = translated_actions ? translated_actions : original_actions;
5489 	p_actions_rx = actions;
5490 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5491 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5492 				external, hairpin_flow, error);
5493 	if (ret < 0)
5494 		goto error_before_hairpin_split;
5495 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5496 	if (!flow) {
5497 		rte_errno = ENOMEM;
5498 		goto error_before_hairpin_split;
5499 	}
5500 	if (hairpin_flow > 0) {
5501 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5502 			rte_errno = EINVAL;
5503 			goto error_before_hairpin_split;
5504 		}
5505 		flow_hairpin_split(dev, actions, actions_rx.actions,
5506 				   actions_hairpin_tx.actions, items_tx.items,
5507 				   idx);
5508 		p_actions_rx = actions_rx.actions;
5509 	}
5510 	flow_split_info.flow_idx = idx;
5511 	flow->drv_type = flow_get_drv_type(dev, attr);
5512 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5513 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5514 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5515 	/* RSS Action only works on NIC RX domain */
5516 	if (attr->ingress && !attr->transfer)
5517 		rss = flow_get_rss_action(p_actions_rx);
5518 	if (rss) {
5519 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5520 			return 0;
5521 		/*
5522 		 * The following information is required by
5523 		 * mlx5_flow_hashfields_adjust() in advance.
5524 		 */
5525 		rss_desc->level = rss->level;
5526 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5527 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5528 	}
5529 	flow->dev_handles = 0;
5530 	if (rss && rss->types) {
5531 		unsigned int graph_root;
5532 
5533 		graph_root = find_graph_root(items, rss->level);
5534 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5535 					   items, rss->types,
5536 					   mlx5_support_expansion, graph_root);
5537 		MLX5_ASSERT(ret > 0 &&
5538 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5539 	} else {
5540 		buf->entries = 1;
5541 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5542 	}
5543 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5544 						      shared_actions_n);
5545 	for (i = 0; i < buf->entries; ++i) {
5546 		/* Initialize flow split data. */
5547 		flow_split_info.prefix_layers = 0;
5548 		flow_split_info.prefix_mark = 0;
5549 		flow_split_info.skip_scale = 0;
5550 		/*
5551 		 * The splitter may create multiple dev_flows,
5552 		 * depending on configuration. In the simplest
5553 		 * case it just creates unmodified original flow.
5554 		 */
5555 		ret = flow_create_split_outer(dev, flow, attr,
5556 					      buf->entry[i].pattern,
5557 					      p_actions_rx, &flow_split_info,
5558 					      error);
5559 		if (ret < 0)
5560 			goto error;
5561 		if (is_flow_tunnel_steer_rule(dev, attr,
5562 					      buf->entry[i].pattern,
5563 					      p_actions_rx)) {
5564 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5565 							   p_actions_rx,
5566 							   idx,
5567 							   &default_miss_ctx,
5568 							   error);
5569 			if (ret < 0) {
5570 				mlx5_free(default_miss_ctx.queue);
5571 				goto error;
5572 			}
5573 		}
5574 	}
5575 	/* Create the tx flow. */
5576 	if (hairpin_flow) {
5577 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5578 		attr_tx.ingress = 0;
5579 		attr_tx.egress = 1;
5580 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5581 					 actions_hairpin_tx.actions,
5582 					 idx, error);
5583 		if (!dev_flow)
5584 			goto error;
5585 		dev_flow->flow = flow;
5586 		dev_flow->external = 0;
5587 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5588 			      dev_flow->handle, next);
5589 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5590 					 items_tx.items,
5591 					 actions_hairpin_tx.actions, error);
5592 		if (ret < 0)
5593 			goto error;
5594 	}
5595 	/*
5596 	 * Update the metadata register copy table. If extensive
5597 	 * metadata feature is enabled and registers are supported
5598 	 * we might create the extra rte_flow for each unique
5599 	 * MARK/FLAG action ID.
5600 	 *
5601 	 * The table is updated for ingress Flows only, because
5602 	 * the egress Flows belong to the different device and
5603 	 * copy table should be updated in peer NIC Rx domain.
5604 	 */
5605 	if (attr->ingress &&
5606 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5607 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5608 		if (ret)
5609 			goto error;
5610 	}
5611 	/*
5612 	 * If the flow is external (from application) OR device is started,
5613 	 * OR mreg discover, then apply immediately.
5614 	 */
5615 	if (external || dev->data->dev_started ||
5616 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5617 	     attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5618 		ret = flow_drv_apply(dev, flow, error);
5619 		if (ret < 0)
5620 			goto error;
5621 	}
5622 	if (list) {
5623 		rte_spinlock_lock(&priv->flow_list_lock);
5624 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5625 			     flow, next);
5626 		rte_spinlock_unlock(&priv->flow_list_lock);
5627 	}
5628 	flow_rxq_flags_set(dev, flow);
5629 	rte_free(translated_actions);
5630 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5631 	if (tunnel) {
5632 		flow->tunnel = 1;
5633 		flow->tunnel_id = tunnel->tunnel_id;
5634 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5635 		mlx5_free(default_miss_ctx.queue);
5636 	}
5637 	mlx5_flow_pop_thread_workspace();
5638 	return idx;
5639 error:
5640 	MLX5_ASSERT(flow);
5641 	ret = rte_errno; /* Save rte_errno before cleanup. */
5642 	flow_mreg_del_copy_action(dev, flow);
5643 	flow_drv_destroy(dev, flow);
5644 	if (rss_desc->shared_rss)
5645 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5646 			mlx5_ipool_get
5647 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5648 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5649 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5650 	rte_errno = ret; /* Restore rte_errno. */
5651 	ret = rte_errno;
5652 	rte_errno = ret;
5653 	mlx5_flow_pop_thread_workspace();
5654 error_before_hairpin_split:
5655 	rte_free(translated_actions);
5656 	return 0;
5657 }
5658 
5659 /**
5660  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5661  * incoming packets to table 1.
5662  *
5663  * Other flow rules, requested for group n, will be created in
5664  * e-switch table n+1.
5665  * Jump action to e-switch group n will be created to group n+1.
5666  *
5667  * Used when working in switchdev mode, to utilise advantages of table 1
5668  * and above.
5669  *
5670  * @param dev
5671  *   Pointer to Ethernet device.
5672  *
5673  * @return
5674  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5675  */
5676 struct rte_flow *
5677 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5678 {
5679 	const struct rte_flow_attr attr = {
5680 		.group = 0,
5681 		.priority = 0,
5682 		.ingress = 1,
5683 		.egress = 0,
5684 		.transfer = 1,
5685 	};
5686 	const struct rte_flow_item pattern = {
5687 		.type = RTE_FLOW_ITEM_TYPE_END,
5688 	};
5689 	struct rte_flow_action_jump jump = {
5690 		.group = 1,
5691 	};
5692 	const struct rte_flow_action actions[] = {
5693 		{
5694 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5695 			.conf = &jump,
5696 		},
5697 		{
5698 			.type = RTE_FLOW_ACTION_TYPE_END,
5699 		},
5700 	};
5701 	struct mlx5_priv *priv = dev->data->dev_private;
5702 	struct rte_flow_error error;
5703 
5704 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5705 						   &attr, &pattern,
5706 						   actions, false, &error);
5707 }
5708 
5709 /**
5710  * Validate a flow supported by the NIC.
5711  *
5712  * @see rte_flow_validate()
5713  * @see rte_flow_ops
5714  */
5715 int
5716 mlx5_flow_validate(struct rte_eth_dev *dev,
5717 		   const struct rte_flow_attr *attr,
5718 		   const struct rte_flow_item items[],
5719 		   const struct rte_flow_action original_actions[],
5720 		   struct rte_flow_error *error)
5721 {
5722 	int hairpin_flow;
5723 	struct mlx5_translated_shared_action
5724 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5725 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5726 	const struct rte_flow_action *actions;
5727 	struct rte_flow_action *translated_actions = NULL;
5728 	int ret = flow_shared_actions_translate(dev, original_actions,
5729 						shared_actions,
5730 						&shared_actions_n,
5731 						&translated_actions, error);
5732 
5733 	if (ret)
5734 		return ret;
5735 	actions = translated_actions ? translated_actions : original_actions;
5736 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5737 	ret = flow_drv_validate(dev, attr, items, actions,
5738 				true, hairpin_flow, error);
5739 	rte_free(translated_actions);
5740 	return ret;
5741 }
5742 
5743 /**
5744  * Create a flow.
5745  *
5746  * @see rte_flow_create()
5747  * @see rte_flow_ops
5748  */
5749 struct rte_flow *
5750 mlx5_flow_create(struct rte_eth_dev *dev,
5751 		 const struct rte_flow_attr *attr,
5752 		 const struct rte_flow_item items[],
5753 		 const struct rte_flow_action actions[],
5754 		 struct rte_flow_error *error)
5755 {
5756 	struct mlx5_priv *priv = dev->data->dev_private;
5757 
5758 	/*
5759 	 * If the device is not started yet, it is not allowed to created a
5760 	 * flow from application. PMD default flows and traffic control flows
5761 	 * are not affected.
5762 	 */
5763 	if (unlikely(!dev->data->dev_started)) {
5764 		DRV_LOG(DEBUG, "port %u is not started when "
5765 			"inserting a flow", dev->data->port_id);
5766 		rte_flow_error_set(error, ENODEV,
5767 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5768 				   NULL,
5769 				   "port not started");
5770 		return NULL;
5771 	}
5772 
5773 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5774 				  attr, items, actions, true, error);
5775 }
5776 
5777 /**
5778  * Destroy a flow in a list.
5779  *
5780  * @param dev
5781  *   Pointer to Ethernet device.
5782  * @param list
5783  *   Pointer to the Indexed flow list. If this parameter NULL,
5784  *   there is no flow removal from the list. Be noted that as
5785  *   flow is add to the indexed list, memory of the indexed
5786  *   list points to maybe changed as flow destroyed.
5787  * @param[in] flow_idx
5788  *   Index of flow to destroy.
5789  */
5790 static void
5791 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5792 		  uint32_t flow_idx)
5793 {
5794 	struct mlx5_priv *priv = dev->data->dev_private;
5795 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5796 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5797 
5798 	if (!flow)
5799 		return;
5800 	/*
5801 	 * Update RX queue flags only if port is started, otherwise it is
5802 	 * already clean.
5803 	 */
5804 	if (dev->data->dev_started)
5805 		flow_rxq_flags_trim(dev, flow);
5806 	flow_drv_destroy(dev, flow);
5807 	if (list) {
5808 		rte_spinlock_lock(&priv->flow_list_lock);
5809 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5810 			     flow_idx, flow, next);
5811 		rte_spinlock_unlock(&priv->flow_list_lock);
5812 	}
5813 	if (flow->tunnel) {
5814 		struct mlx5_flow_tunnel *tunnel;
5815 
5816 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5817 		RTE_VERIFY(tunnel);
5818 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5819 			mlx5_flow_tunnel_free(dev, tunnel);
5820 	}
5821 	flow_mreg_del_copy_action(dev, flow);
5822 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5823 }
5824 
5825 /**
5826  * Destroy all flows.
5827  *
5828  * @param dev
5829  *   Pointer to Ethernet device.
5830  * @param list
5831  *   Pointer to the Indexed flow list.
5832  * @param active
5833  *   If flushing is called avtively.
5834  */
5835 void
5836 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5837 {
5838 	uint32_t num_flushed = 0;
5839 
5840 	while (*list) {
5841 		flow_list_destroy(dev, list, *list);
5842 		num_flushed++;
5843 	}
5844 	if (active) {
5845 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5846 			dev->data->port_id, num_flushed);
5847 	}
5848 }
5849 
5850 /**
5851  * Stop all default actions for flows.
5852  *
5853  * @param dev
5854  *   Pointer to Ethernet device.
5855  */
5856 void
5857 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5858 {
5859 	flow_mreg_del_default_copy_action(dev);
5860 	flow_rxq_flags_clear(dev);
5861 }
5862 
5863 /**
5864  * Start all default actions for flows.
5865  *
5866  * @param dev
5867  *   Pointer to Ethernet device.
5868  * @return
5869  *   0 on success, a negative errno value otherwise and rte_errno is set.
5870  */
5871 int
5872 mlx5_flow_start_default(struct rte_eth_dev *dev)
5873 {
5874 	struct rte_flow_error error;
5875 
5876 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5877 	return flow_mreg_add_default_copy_action(dev, &error);
5878 }
5879 
5880 /**
5881  * Release key of thread specific flow workspace data.
5882  */
5883 void
5884 flow_release_workspace(void *data)
5885 {
5886 	struct mlx5_flow_workspace *wks = data;
5887 	struct mlx5_flow_workspace *next;
5888 
5889 	while (wks) {
5890 		next = wks->next;
5891 		free(wks->rss_desc.queue);
5892 		free(wks);
5893 		wks = next;
5894 	}
5895 }
5896 
5897 /**
5898  * Get thread specific current flow workspace.
5899  *
5900  * @return pointer to thread specific flow workspace data, NULL on error.
5901  */
5902 struct mlx5_flow_workspace*
5903 mlx5_flow_get_thread_workspace(void)
5904 {
5905 	struct mlx5_flow_workspace *data;
5906 
5907 	data = mlx5_flow_os_get_specific_workspace();
5908 	MLX5_ASSERT(data && data->inuse);
5909 	if (!data || !data->inuse)
5910 		DRV_LOG(ERR, "flow workspace not initialized.");
5911 	return data;
5912 }
5913 
5914 /**
5915  * Allocate and init new flow workspace.
5916  *
5917  * @return pointer to flow workspace data, NULL on error.
5918  */
5919 static struct mlx5_flow_workspace*
5920 flow_alloc_thread_workspace(void)
5921 {
5922 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5923 
5924 	if (!data) {
5925 		DRV_LOG(ERR, "Failed to allocate flow workspace "
5926 			"memory.");
5927 		return NULL;
5928 	}
5929 	data->rss_desc.queue = calloc(1,
5930 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5931 	if (!data->rss_desc.queue)
5932 		goto err;
5933 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
5934 	return data;
5935 err:
5936 	if (data->rss_desc.queue)
5937 		free(data->rss_desc.queue);
5938 	free(data);
5939 	return NULL;
5940 }
5941 
5942 /**
5943  * Get new thread specific flow workspace.
5944  *
5945  * If current workspace inuse, create new one and set as current.
5946  *
5947  * @return pointer to thread specific flow workspace data, NULL on error.
5948  */
5949 static struct mlx5_flow_workspace*
5950 mlx5_flow_push_thread_workspace(void)
5951 {
5952 	struct mlx5_flow_workspace *curr;
5953 	struct mlx5_flow_workspace *data;
5954 
5955 	curr = mlx5_flow_os_get_specific_workspace();
5956 	if (!curr) {
5957 		data = flow_alloc_thread_workspace();
5958 		if (!data)
5959 			return NULL;
5960 	} else if (!curr->inuse) {
5961 		data = curr;
5962 	} else if (curr->next) {
5963 		data = curr->next;
5964 	} else {
5965 		data = flow_alloc_thread_workspace();
5966 		if (!data)
5967 			return NULL;
5968 		curr->next = data;
5969 		data->prev = curr;
5970 	}
5971 	data->inuse = 1;
5972 	data->flow_idx = 0;
5973 	/* Set as current workspace */
5974 	if (mlx5_flow_os_set_specific_workspace(data))
5975 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5976 	return data;
5977 }
5978 
5979 /**
5980  * Close current thread specific flow workspace.
5981  *
5982  * If previous workspace available, set it as current.
5983  *
5984  * @return pointer to thread specific flow workspace data, NULL on error.
5985  */
5986 static void
5987 mlx5_flow_pop_thread_workspace(void)
5988 {
5989 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
5990 
5991 	if (!data)
5992 		return;
5993 	if (!data->inuse) {
5994 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
5995 		return;
5996 	}
5997 	data->inuse = 0;
5998 	if (!data->prev)
5999 		return;
6000 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6001 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6002 }
6003 
6004 /**
6005  * Verify the flow list is empty
6006  *
6007  * @param dev
6008  *  Pointer to Ethernet device.
6009  *
6010  * @return the number of flows not released.
6011  */
6012 int
6013 mlx5_flow_verify(struct rte_eth_dev *dev)
6014 {
6015 	struct mlx5_priv *priv = dev->data->dev_private;
6016 	struct rte_flow *flow;
6017 	uint32_t idx;
6018 	int ret = 0;
6019 
6020 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6021 		      flow, next) {
6022 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6023 			dev->data->port_id, (void *)flow);
6024 		++ret;
6025 	}
6026 	return ret;
6027 }
6028 
6029 /**
6030  * Enable default hairpin egress flow.
6031  *
6032  * @param dev
6033  *   Pointer to Ethernet device.
6034  * @param queue
6035  *   The queue index.
6036  *
6037  * @return
6038  *   0 on success, a negative errno value otherwise and rte_errno is set.
6039  */
6040 int
6041 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6042 			    uint32_t queue)
6043 {
6044 	struct mlx5_priv *priv = dev->data->dev_private;
6045 	const struct rte_flow_attr attr = {
6046 		.egress = 1,
6047 		.priority = 0,
6048 	};
6049 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6050 		.queue = queue,
6051 	};
6052 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6053 		.queue = UINT32_MAX,
6054 	};
6055 	struct rte_flow_item items[] = {
6056 		{
6057 			.type = (enum rte_flow_item_type)
6058 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6059 			.spec = &queue_spec,
6060 			.last = NULL,
6061 			.mask = &queue_mask,
6062 		},
6063 		{
6064 			.type = RTE_FLOW_ITEM_TYPE_END,
6065 		},
6066 	};
6067 	struct rte_flow_action_jump jump = {
6068 		.group = MLX5_HAIRPIN_TX_TABLE,
6069 	};
6070 	struct rte_flow_action actions[2];
6071 	uint32_t flow_idx;
6072 	struct rte_flow_error error;
6073 
6074 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6075 	actions[0].conf = &jump;
6076 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6077 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6078 				&attr, items, actions, false, &error);
6079 	if (!flow_idx) {
6080 		DRV_LOG(DEBUG,
6081 			"Failed to create ctrl flow: rte_errno(%d),"
6082 			" type(%d), message(%s)",
6083 			rte_errno, error.type,
6084 			error.message ? error.message : " (no stated reason)");
6085 		return -rte_errno;
6086 	}
6087 	return 0;
6088 }
6089 
6090 /**
6091  * Enable a control flow configured from the control plane.
6092  *
6093  * @param dev
6094  *   Pointer to Ethernet device.
6095  * @param eth_spec
6096  *   An Ethernet flow spec to apply.
6097  * @param eth_mask
6098  *   An Ethernet flow mask to apply.
6099  * @param vlan_spec
6100  *   A VLAN flow spec to apply.
6101  * @param vlan_mask
6102  *   A VLAN flow mask to apply.
6103  *
6104  * @return
6105  *   0 on success, a negative errno value otherwise and rte_errno is set.
6106  */
6107 int
6108 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6109 		    struct rte_flow_item_eth *eth_spec,
6110 		    struct rte_flow_item_eth *eth_mask,
6111 		    struct rte_flow_item_vlan *vlan_spec,
6112 		    struct rte_flow_item_vlan *vlan_mask)
6113 {
6114 	struct mlx5_priv *priv = dev->data->dev_private;
6115 	const struct rte_flow_attr attr = {
6116 		.ingress = 1,
6117 		.priority = MLX5_FLOW_PRIO_RSVD,
6118 	};
6119 	struct rte_flow_item items[] = {
6120 		{
6121 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6122 			.spec = eth_spec,
6123 			.last = NULL,
6124 			.mask = eth_mask,
6125 		},
6126 		{
6127 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6128 					      RTE_FLOW_ITEM_TYPE_END,
6129 			.spec = vlan_spec,
6130 			.last = NULL,
6131 			.mask = vlan_mask,
6132 		},
6133 		{
6134 			.type = RTE_FLOW_ITEM_TYPE_END,
6135 		},
6136 	};
6137 	uint16_t queue[priv->reta_idx_n];
6138 	struct rte_flow_action_rss action_rss = {
6139 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6140 		.level = 0,
6141 		.types = priv->rss_conf.rss_hf,
6142 		.key_len = priv->rss_conf.rss_key_len,
6143 		.queue_num = priv->reta_idx_n,
6144 		.key = priv->rss_conf.rss_key,
6145 		.queue = queue,
6146 	};
6147 	struct rte_flow_action actions[] = {
6148 		{
6149 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6150 			.conf = &action_rss,
6151 		},
6152 		{
6153 			.type = RTE_FLOW_ACTION_TYPE_END,
6154 		},
6155 	};
6156 	uint32_t flow_idx;
6157 	struct rte_flow_error error;
6158 	unsigned int i;
6159 
6160 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6161 		return 0;
6162 	}
6163 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6164 		action_rss.types = 0;
6165 	for (i = 0; i != priv->reta_idx_n; ++i)
6166 		queue[i] = (*priv->reta_idx)[i];
6167 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6168 				&attr, items, actions, false, &error);
6169 	if (!flow_idx)
6170 		return -rte_errno;
6171 	return 0;
6172 }
6173 
6174 /**
6175  * Enable a flow control configured from the control plane.
6176  *
6177  * @param dev
6178  *   Pointer to Ethernet device.
6179  * @param eth_spec
6180  *   An Ethernet flow spec to apply.
6181  * @param eth_mask
6182  *   An Ethernet flow mask to apply.
6183  *
6184  * @return
6185  *   0 on success, a negative errno value otherwise and rte_errno is set.
6186  */
6187 int
6188 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6189 	       struct rte_flow_item_eth *eth_spec,
6190 	       struct rte_flow_item_eth *eth_mask)
6191 {
6192 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6193 }
6194 
6195 /**
6196  * Create default miss flow rule matching lacp traffic
6197  *
6198  * @param dev
6199  *   Pointer to Ethernet device.
6200  * @param eth_spec
6201  *   An Ethernet flow spec to apply.
6202  *
6203  * @return
6204  *   0 on success, a negative errno value otherwise and rte_errno is set.
6205  */
6206 int
6207 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6208 {
6209 	struct mlx5_priv *priv = dev->data->dev_private;
6210 	/*
6211 	 * The LACP matching is done by only using ether type since using
6212 	 * a multicast dst mac causes kernel to give low priority to this flow.
6213 	 */
6214 	static const struct rte_flow_item_eth lacp_spec = {
6215 		.type = RTE_BE16(0x8809),
6216 	};
6217 	static const struct rte_flow_item_eth lacp_mask = {
6218 		.type = 0xffff,
6219 	};
6220 	const struct rte_flow_attr attr = {
6221 		.ingress = 1,
6222 	};
6223 	struct rte_flow_item items[] = {
6224 		{
6225 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6226 			.spec = &lacp_spec,
6227 			.mask = &lacp_mask,
6228 		},
6229 		{
6230 			.type = RTE_FLOW_ITEM_TYPE_END,
6231 		},
6232 	};
6233 	struct rte_flow_action actions[] = {
6234 		{
6235 			.type = (enum rte_flow_action_type)
6236 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6237 		},
6238 		{
6239 			.type = RTE_FLOW_ACTION_TYPE_END,
6240 		},
6241 	};
6242 	struct rte_flow_error error;
6243 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6244 				&attr, items, actions, false, &error);
6245 
6246 	if (!flow_idx)
6247 		return -rte_errno;
6248 	return 0;
6249 }
6250 
6251 /**
6252  * Destroy a flow.
6253  *
6254  * @see rte_flow_destroy()
6255  * @see rte_flow_ops
6256  */
6257 int
6258 mlx5_flow_destroy(struct rte_eth_dev *dev,
6259 		  struct rte_flow *flow,
6260 		  struct rte_flow_error *error __rte_unused)
6261 {
6262 	struct mlx5_priv *priv = dev->data->dev_private;
6263 
6264 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6265 	return 0;
6266 }
6267 
6268 /**
6269  * Destroy all flows.
6270  *
6271  * @see rte_flow_flush()
6272  * @see rte_flow_ops
6273  */
6274 int
6275 mlx5_flow_flush(struct rte_eth_dev *dev,
6276 		struct rte_flow_error *error __rte_unused)
6277 {
6278 	struct mlx5_priv *priv = dev->data->dev_private;
6279 
6280 	mlx5_flow_list_flush(dev, &priv->flows, false);
6281 	return 0;
6282 }
6283 
6284 /**
6285  * Isolated mode.
6286  *
6287  * @see rte_flow_isolate()
6288  * @see rte_flow_ops
6289  */
6290 int
6291 mlx5_flow_isolate(struct rte_eth_dev *dev,
6292 		  int enable,
6293 		  struct rte_flow_error *error)
6294 {
6295 	struct mlx5_priv *priv = dev->data->dev_private;
6296 
6297 	if (dev->data->dev_started) {
6298 		rte_flow_error_set(error, EBUSY,
6299 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6300 				   NULL,
6301 				   "port must be stopped first");
6302 		return -rte_errno;
6303 	}
6304 	priv->isolated = !!enable;
6305 	if (enable)
6306 		dev->dev_ops = &mlx5_dev_ops_isolate;
6307 	else
6308 		dev->dev_ops = &mlx5_dev_ops;
6309 
6310 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6311 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6312 
6313 	return 0;
6314 }
6315 
6316 /**
6317  * Query a flow.
6318  *
6319  * @see rte_flow_query()
6320  * @see rte_flow_ops
6321  */
6322 static int
6323 flow_drv_query(struct rte_eth_dev *dev,
6324 	       uint32_t flow_idx,
6325 	       const struct rte_flow_action *actions,
6326 	       void *data,
6327 	       struct rte_flow_error *error)
6328 {
6329 	struct mlx5_priv *priv = dev->data->dev_private;
6330 	const struct mlx5_flow_driver_ops *fops;
6331 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6332 					       [MLX5_IPOOL_RTE_FLOW],
6333 					       flow_idx);
6334 	enum mlx5_flow_drv_type ftype;
6335 
6336 	if (!flow) {
6337 		return rte_flow_error_set(error, ENOENT,
6338 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6339 			  NULL,
6340 			  "invalid flow handle");
6341 	}
6342 	ftype = flow->drv_type;
6343 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6344 	fops = flow_get_drv_ops(ftype);
6345 
6346 	return fops->query(dev, flow, actions, data, error);
6347 }
6348 
6349 /**
6350  * Query a flow.
6351  *
6352  * @see rte_flow_query()
6353  * @see rte_flow_ops
6354  */
6355 int
6356 mlx5_flow_query(struct rte_eth_dev *dev,
6357 		struct rte_flow *flow,
6358 		const struct rte_flow_action *actions,
6359 		void *data,
6360 		struct rte_flow_error *error)
6361 {
6362 	int ret;
6363 
6364 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6365 			     error);
6366 	if (ret < 0)
6367 		return ret;
6368 	return 0;
6369 }
6370 
6371 /**
6372  * Manage filter operations.
6373  *
6374  * @param dev
6375  *   Pointer to Ethernet device structure.
6376  * @param filter_type
6377  *   Filter type.
6378  * @param filter_op
6379  *   Operation to perform.
6380  * @param arg
6381  *   Pointer to operation-specific structure.
6382  *
6383  * @return
6384  *   0 on success, a negative errno value otherwise and rte_errno is set.
6385  */
6386 int
6387 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6388 		     enum rte_filter_type filter_type,
6389 		     enum rte_filter_op filter_op,
6390 		     void *arg)
6391 {
6392 	switch (filter_type) {
6393 	case RTE_ETH_FILTER_GENERIC:
6394 		if (filter_op != RTE_ETH_FILTER_GET) {
6395 			rte_errno = EINVAL;
6396 			return -rte_errno;
6397 		}
6398 		*(const void **)arg = &mlx5_flow_ops;
6399 		return 0;
6400 	default:
6401 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6402 			dev->data->port_id, filter_type);
6403 		rte_errno = ENOTSUP;
6404 		return -rte_errno;
6405 	}
6406 	return 0;
6407 }
6408 
6409 /**
6410  * Create the needed meter and suffix tables.
6411  *
6412  * @param[in] dev
6413  *   Pointer to Ethernet device.
6414  * @param[in] fm
6415  *   Pointer to the flow meter.
6416  *
6417  * @return
6418  *   Pointer to table set on success, NULL otherwise.
6419  */
6420 struct mlx5_meter_domains_infos *
6421 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6422 			  const struct mlx5_flow_meter *fm)
6423 {
6424 	const struct mlx5_flow_driver_ops *fops;
6425 
6426 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6427 	return fops->create_mtr_tbls(dev, fm);
6428 }
6429 
6430 /**
6431  * Destroy the meter table set.
6432  *
6433  * @param[in] dev
6434  *   Pointer to Ethernet device.
6435  * @param[in] tbl
6436  *   Pointer to the meter table set.
6437  *
6438  * @return
6439  *   0 on success.
6440  */
6441 int
6442 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6443 			   struct mlx5_meter_domains_infos *tbls)
6444 {
6445 	const struct mlx5_flow_driver_ops *fops;
6446 
6447 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6448 	return fops->destroy_mtr_tbls(dev, tbls);
6449 }
6450 
6451 /**
6452  * Create policer rules.
6453  *
6454  * @param[in] dev
6455  *   Pointer to Ethernet device.
6456  * @param[in] fm
6457  *   Pointer to flow meter structure.
6458  * @param[in] attr
6459  *   Pointer to flow attributes.
6460  *
6461  * @return
6462  *   0 on success, -1 otherwise.
6463  */
6464 int
6465 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6466 			       struct mlx5_flow_meter *fm,
6467 			       const struct rte_flow_attr *attr)
6468 {
6469 	const struct mlx5_flow_driver_ops *fops;
6470 
6471 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6472 	return fops->create_policer_rules(dev, fm, attr);
6473 }
6474 
6475 /**
6476  * Destroy policer rules.
6477  *
6478  * @param[in] fm
6479  *   Pointer to flow meter structure.
6480  * @param[in] attr
6481  *   Pointer to flow attributes.
6482  *
6483  * @return
6484  *   0 on success, -1 otherwise.
6485  */
6486 int
6487 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6488 				struct mlx5_flow_meter *fm,
6489 				const struct rte_flow_attr *attr)
6490 {
6491 	const struct mlx5_flow_driver_ops *fops;
6492 
6493 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6494 	return fops->destroy_policer_rules(dev, fm, attr);
6495 }
6496 
6497 /**
6498  * Allocate a counter.
6499  *
6500  * @param[in] dev
6501  *   Pointer to Ethernet device structure.
6502  *
6503  * @return
6504  *   Index to allocated counter  on success, 0 otherwise.
6505  */
6506 uint32_t
6507 mlx5_counter_alloc(struct rte_eth_dev *dev)
6508 {
6509 	const struct mlx5_flow_driver_ops *fops;
6510 	struct rte_flow_attr attr = { .transfer = 0 };
6511 
6512 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6513 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6514 		return fops->counter_alloc(dev);
6515 	}
6516 	DRV_LOG(ERR,
6517 		"port %u counter allocate is not supported.",
6518 		 dev->data->port_id);
6519 	return 0;
6520 }
6521 
6522 /**
6523  * Free a counter.
6524  *
6525  * @param[in] dev
6526  *   Pointer to Ethernet device structure.
6527  * @param[in] cnt
6528  *   Index to counter to be free.
6529  */
6530 void
6531 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6532 {
6533 	const struct mlx5_flow_driver_ops *fops;
6534 	struct rte_flow_attr attr = { .transfer = 0 };
6535 
6536 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6537 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6538 		fops->counter_free(dev, cnt);
6539 		return;
6540 	}
6541 	DRV_LOG(ERR,
6542 		"port %u counter free is not supported.",
6543 		 dev->data->port_id);
6544 }
6545 
6546 /**
6547  * Query counter statistics.
6548  *
6549  * @param[in] dev
6550  *   Pointer to Ethernet device structure.
6551  * @param[in] cnt
6552  *   Index to counter to query.
6553  * @param[in] clear
6554  *   Set to clear counter statistics.
6555  * @param[out] pkts
6556  *   The counter hits packets number to save.
6557  * @param[out] bytes
6558  *   The counter hits bytes number to save.
6559  *
6560  * @return
6561  *   0 on success, a negative errno value otherwise.
6562  */
6563 int
6564 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6565 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6566 {
6567 	const struct mlx5_flow_driver_ops *fops;
6568 	struct rte_flow_attr attr = { .transfer = 0 };
6569 
6570 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6571 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6572 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6573 	}
6574 	DRV_LOG(ERR,
6575 		"port %u counter query is not supported.",
6576 		 dev->data->port_id);
6577 	return -ENOTSUP;
6578 }
6579 
6580 /**
6581  * Allocate a new memory for the counter values wrapped by all the needed
6582  * management.
6583  *
6584  * @param[in] sh
6585  *   Pointer to mlx5_dev_ctx_shared object.
6586  *
6587  * @return
6588  *   0 on success, a negative errno value otherwise.
6589  */
6590 static int
6591 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6592 {
6593 	struct mlx5_devx_mkey_attr mkey_attr;
6594 	struct mlx5_counter_stats_mem_mng *mem_mng;
6595 	volatile struct flow_counter_stats *raw_data;
6596 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6597 	int size = (sizeof(struct flow_counter_stats) *
6598 			MLX5_COUNTERS_PER_POOL +
6599 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6600 			sizeof(struct mlx5_counter_stats_mem_mng);
6601 	size_t pgsize = rte_mem_page_size();
6602 	uint8_t *mem;
6603 	int i;
6604 
6605 	if (pgsize == (size_t)-1) {
6606 		DRV_LOG(ERR, "Failed to get mem page size");
6607 		rte_errno = ENOMEM;
6608 		return -ENOMEM;
6609 	}
6610 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6611 	if (!mem) {
6612 		rte_errno = ENOMEM;
6613 		return -ENOMEM;
6614 	}
6615 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6616 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6617 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6618 						 IBV_ACCESS_LOCAL_WRITE);
6619 	if (!mem_mng->umem) {
6620 		rte_errno = errno;
6621 		mlx5_free(mem);
6622 		return -rte_errno;
6623 	}
6624 	mkey_attr.addr = (uintptr_t)mem;
6625 	mkey_attr.size = size;
6626 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6627 	mkey_attr.pd = sh->pdn;
6628 	mkey_attr.log_entity_size = 0;
6629 	mkey_attr.pg_access = 0;
6630 	mkey_attr.klm_array = NULL;
6631 	mkey_attr.klm_num = 0;
6632 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6633 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6634 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6635 	if (!mem_mng->dm) {
6636 		mlx5_os_umem_dereg(mem_mng->umem);
6637 		rte_errno = errno;
6638 		mlx5_free(mem);
6639 		return -rte_errno;
6640 	}
6641 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6642 	raw_data = (volatile struct flow_counter_stats *)mem;
6643 	for (i = 0; i < raws_n; ++i) {
6644 		mem_mng->raws[i].mem_mng = mem_mng;
6645 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6646 	}
6647 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6648 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6649 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6650 				 next);
6651 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6652 	sh->cmng.mem_mng = mem_mng;
6653 	return 0;
6654 }
6655 
6656 /**
6657  * Set the statistic memory to the new counter pool.
6658  *
6659  * @param[in] sh
6660  *   Pointer to mlx5_dev_ctx_shared object.
6661  * @param[in] pool
6662  *   Pointer to the pool to set the statistic memory.
6663  *
6664  * @return
6665  *   0 on success, a negative errno value otherwise.
6666  */
6667 static int
6668 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6669 			       struct mlx5_flow_counter_pool *pool)
6670 {
6671 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6672 	/* Resize statistic memory once used out. */
6673 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6674 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6675 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6676 		return -1;
6677 	}
6678 	rte_spinlock_lock(&pool->sl);
6679 	pool->raw = cmng->mem_mng->raws + pool->index %
6680 		    MLX5_CNT_CONTAINER_RESIZE;
6681 	rte_spinlock_unlock(&pool->sl);
6682 	pool->raw_hw = NULL;
6683 	return 0;
6684 }
6685 
6686 #define MLX5_POOL_QUERY_FREQ_US 1000000
6687 
6688 /**
6689  * Set the periodic procedure for triggering asynchronous batch queries for all
6690  * the counter pools.
6691  *
6692  * @param[in] sh
6693  *   Pointer to mlx5_dev_ctx_shared object.
6694  */
6695 void
6696 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6697 {
6698 	uint32_t pools_n, us;
6699 
6700 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6701 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6702 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6703 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6704 		sh->cmng.query_thread_on = 0;
6705 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6706 	} else {
6707 		sh->cmng.query_thread_on = 1;
6708 	}
6709 }
6710 
6711 /**
6712  * The periodic procedure for triggering asynchronous batch queries for all the
6713  * counter pools. This function is probably called by the host thread.
6714  *
6715  * @param[in] arg
6716  *   The parameter for the alarm process.
6717  */
6718 void
6719 mlx5_flow_query_alarm(void *arg)
6720 {
6721 	struct mlx5_dev_ctx_shared *sh = arg;
6722 	int ret;
6723 	uint16_t pool_index = sh->cmng.pool_index;
6724 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6725 	struct mlx5_flow_counter_pool *pool;
6726 	uint16_t n_valid;
6727 
6728 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6729 		goto set_alarm;
6730 	rte_spinlock_lock(&cmng->pool_update_sl);
6731 	pool = cmng->pools[pool_index];
6732 	n_valid = cmng->n_valid;
6733 	rte_spinlock_unlock(&cmng->pool_update_sl);
6734 	/* Set the statistic memory to the new created pool. */
6735 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6736 		goto set_alarm;
6737 	if (pool->raw_hw)
6738 		/* There is a pool query in progress. */
6739 		goto set_alarm;
6740 	pool->raw_hw =
6741 		LIST_FIRST(&sh->cmng.free_stat_raws);
6742 	if (!pool->raw_hw)
6743 		/* No free counter statistics raw memory. */
6744 		goto set_alarm;
6745 	/*
6746 	 * Identify the counters released between query trigger and query
6747 	 * handle more efficiently. The counter released in this gap period
6748 	 * should wait for a new round of query as the new arrived packets
6749 	 * will not be taken into account.
6750 	 */
6751 	pool->query_gen++;
6752 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6753 					       MLX5_COUNTERS_PER_POOL,
6754 					       NULL, NULL,
6755 					       pool->raw_hw->mem_mng->dm->id,
6756 					       (void *)(uintptr_t)
6757 					       pool->raw_hw->data,
6758 					       sh->devx_comp,
6759 					       (uint64_t)(uintptr_t)pool);
6760 	if (ret) {
6761 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6762 			" %d", pool->min_dcs->id);
6763 		pool->raw_hw = NULL;
6764 		goto set_alarm;
6765 	}
6766 	LIST_REMOVE(pool->raw_hw, next);
6767 	sh->cmng.pending_queries++;
6768 	pool_index++;
6769 	if (pool_index >= n_valid)
6770 		pool_index = 0;
6771 set_alarm:
6772 	sh->cmng.pool_index = pool_index;
6773 	mlx5_set_query_alarm(sh);
6774 }
6775 
6776 /**
6777  * Check and callback event for new aged flow in the counter pool
6778  *
6779  * @param[in] sh
6780  *   Pointer to mlx5_dev_ctx_shared object.
6781  * @param[in] pool
6782  *   Pointer to Current counter pool.
6783  */
6784 static void
6785 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6786 		   struct mlx5_flow_counter_pool *pool)
6787 {
6788 	struct mlx5_priv *priv;
6789 	struct mlx5_flow_counter *cnt;
6790 	struct mlx5_age_info *age_info;
6791 	struct mlx5_age_param *age_param;
6792 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6793 	struct mlx5_counter_stats_raw *prev = pool->raw;
6794 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6795 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6796 	uint16_t expected = AGE_CANDIDATE;
6797 	uint32_t i;
6798 
6799 	pool->time_of_last_age_check = curr_time;
6800 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6801 		cnt = MLX5_POOL_GET_CNT(pool, i);
6802 		age_param = MLX5_CNT_TO_AGE(cnt);
6803 		if (__atomic_load_n(&age_param->state,
6804 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6805 			continue;
6806 		if (cur->data[i].hits != prev->data[i].hits) {
6807 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6808 					 __ATOMIC_RELAXED);
6809 			continue;
6810 		}
6811 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6812 				       time_delta,
6813 				       __ATOMIC_RELAXED) <= age_param->timeout)
6814 			continue;
6815 		/**
6816 		 * Hold the lock first, or if between the
6817 		 * state AGE_TMOUT and tailq operation the
6818 		 * release happened, the release procedure
6819 		 * may delete a non-existent tailq node.
6820 		 */
6821 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6822 		age_info = GET_PORT_AGE_INFO(priv);
6823 		rte_spinlock_lock(&age_info->aged_sl);
6824 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6825 						AGE_TMOUT, false,
6826 						__ATOMIC_RELAXED,
6827 						__ATOMIC_RELAXED)) {
6828 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6829 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6830 		}
6831 		rte_spinlock_unlock(&age_info->aged_sl);
6832 	}
6833 	mlx5_age_event_prepare(sh);
6834 }
6835 
6836 /**
6837  * Handler for the HW respond about ready values from an asynchronous batch
6838  * query. This function is probably called by the host thread.
6839  *
6840  * @param[in] sh
6841  *   The pointer to the shared device context.
6842  * @param[in] async_id
6843  *   The Devx async ID.
6844  * @param[in] status
6845  *   The status of the completion.
6846  */
6847 void
6848 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6849 				  uint64_t async_id, int status)
6850 {
6851 	struct mlx5_flow_counter_pool *pool =
6852 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6853 	struct mlx5_counter_stats_raw *raw_to_free;
6854 	uint8_t query_gen = pool->query_gen ^ 1;
6855 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6856 	enum mlx5_counter_type cnt_type =
6857 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6858 				MLX5_COUNTER_TYPE_ORIGIN;
6859 
6860 	if (unlikely(status)) {
6861 		raw_to_free = pool->raw_hw;
6862 	} else {
6863 		raw_to_free = pool->raw;
6864 		if (pool->is_aged)
6865 			mlx5_flow_aging_check(sh, pool);
6866 		rte_spinlock_lock(&pool->sl);
6867 		pool->raw = pool->raw_hw;
6868 		rte_spinlock_unlock(&pool->sl);
6869 		/* Be sure the new raw counters data is updated in memory. */
6870 		rte_io_wmb();
6871 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6872 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6873 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6874 				     &pool->counters[query_gen], next);
6875 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6876 		}
6877 	}
6878 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6879 	pool->raw_hw = NULL;
6880 	sh->cmng.pending_queries--;
6881 }
6882 
6883 static int
6884 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6885 		    const struct flow_grp_info *grp_info,
6886 		    struct rte_flow_error *error)
6887 {
6888 	if (grp_info->transfer && grp_info->external &&
6889 	    grp_info->fdb_def_rule) {
6890 		if (group == UINT32_MAX)
6891 			return rte_flow_error_set
6892 						(error, EINVAL,
6893 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6894 						 NULL,
6895 						 "group index not supported");
6896 		*table = group + 1;
6897 	} else {
6898 		*table = group;
6899 	}
6900 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6901 	return 0;
6902 }
6903 
6904 /**
6905  * Translate the rte_flow group index to HW table value.
6906  *
6907  * If tunnel offload is disabled, all group ids converted to flow table
6908  * id using the standard method.
6909  * If tunnel offload is enabled, group id can be converted using the
6910  * standard or tunnel conversion method. Group conversion method
6911  * selection depends on flags in `grp_info` parameter:
6912  * - Internal (grp_info.external == 0) groups conversion uses the
6913  *   standard method.
6914  * - Group ids in JUMP action converted with the tunnel conversion.
6915  * - Group id in rule attribute conversion depends on a rule type and
6916  *   group id value:
6917  *   ** non zero group attributes converted with the tunnel method
6918  *   ** zero group attribute in non-tunnel rule is converted using the
6919  *      standard method - there's only one root table
6920  *   ** zero group attribute in steer tunnel rule is converted with the
6921  *      standard method - single root table
6922  *   ** zero group attribute in match tunnel rule is a special OvS
6923  *      case: that value is used for portability reasons. That group
6924  *      id is converted with the tunnel conversion method.
6925  *
6926  * @param[in] dev
6927  *   Port device
6928  * @param[in] tunnel
6929  *   PMD tunnel offload object
6930  * @param[in] group
6931  *   rte_flow group index value.
6932  * @param[out] table
6933  *   HW table value.
6934  * @param[in] grp_info
6935  *   flags used for conversion
6936  * @param[out] error
6937  *   Pointer to error structure.
6938  *
6939  * @return
6940  *   0 on success, a negative errno value otherwise and rte_errno is set.
6941  */
6942 int
6943 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
6944 			 const struct mlx5_flow_tunnel *tunnel,
6945 			 uint32_t group, uint32_t *table,
6946 			 const struct flow_grp_info *grp_info,
6947 			 struct rte_flow_error *error)
6948 {
6949 	int ret;
6950 	bool standard_translation;
6951 
6952 	if (!grp_info->skip_scale && grp_info->external &&
6953 	    group < MLX5_MAX_TABLES_EXTERNAL)
6954 		group *= MLX5_FLOW_TABLE_FACTOR;
6955 	if (is_tunnel_offload_active(dev)) {
6956 		standard_translation = !grp_info->external ||
6957 					grp_info->std_tbl_fix;
6958 	} else {
6959 		standard_translation = true;
6960 	}
6961 	DRV_LOG(DEBUG,
6962 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
6963 		dev->data->port_id, group, grp_info->transfer,
6964 		grp_info->external, grp_info->fdb_def_rule,
6965 		standard_translation ? "STANDARD" : "TUNNEL");
6966 	if (standard_translation)
6967 		ret = flow_group_to_table(dev->data->port_id, group, table,
6968 					  grp_info, error);
6969 	else
6970 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
6971 						      table, error);
6972 
6973 	return ret;
6974 }
6975 
6976 /**
6977  * Discover availability of metadata reg_c's.
6978  *
6979  * Iteratively use test flows to check availability.
6980  *
6981  * @param[in] dev
6982  *   Pointer to the Ethernet device structure.
6983  *
6984  * @return
6985  *   0 on success, a negative errno value otherwise and rte_errno is set.
6986  */
6987 int
6988 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6989 {
6990 	struct mlx5_priv *priv = dev->data->dev_private;
6991 	struct mlx5_dev_config *config = &priv->config;
6992 	enum modify_reg idx;
6993 	int n = 0;
6994 
6995 	/* reg_c[0] and reg_c[1] are reserved. */
6996 	config->flow_mreg_c[n++] = REG_C_0;
6997 	config->flow_mreg_c[n++] = REG_C_1;
6998 	/* Discover availability of other reg_c's. */
6999 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7000 		struct rte_flow_attr attr = {
7001 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7002 			.priority = MLX5_FLOW_PRIO_RSVD,
7003 			.ingress = 1,
7004 		};
7005 		struct rte_flow_item items[] = {
7006 			[0] = {
7007 				.type = RTE_FLOW_ITEM_TYPE_END,
7008 			},
7009 		};
7010 		struct rte_flow_action actions[] = {
7011 			[0] = {
7012 				.type = (enum rte_flow_action_type)
7013 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7014 				.conf = &(struct mlx5_flow_action_copy_mreg){
7015 					.src = REG_C_1,
7016 					.dst = idx,
7017 				},
7018 			},
7019 			[1] = {
7020 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7021 				.conf = &(struct rte_flow_action_jump){
7022 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7023 				},
7024 			},
7025 			[2] = {
7026 				.type = RTE_FLOW_ACTION_TYPE_END,
7027 			},
7028 		};
7029 		uint32_t flow_idx;
7030 		struct rte_flow *flow;
7031 		struct rte_flow_error error;
7032 
7033 		if (!config->dv_flow_en)
7034 			break;
7035 		/* Create internal flow, validation skips copy action. */
7036 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7037 					    actions, false, &error);
7038 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7039 				      flow_idx);
7040 		if (!flow)
7041 			continue;
7042 		config->flow_mreg_c[n++] = idx;
7043 		flow_list_destroy(dev, NULL, flow_idx);
7044 	}
7045 	for (; n < MLX5_MREG_C_NUM; ++n)
7046 		config->flow_mreg_c[n] = REG_NON;
7047 	return 0;
7048 }
7049 
7050 /**
7051  * Dump flow raw hw data to file
7052  *
7053  * @param[in] dev
7054  *    The pointer to Ethernet device.
7055  * @param[in] file
7056  *   A pointer to a file for output.
7057  * @param[out] error
7058  *   Perform verbose error reporting if not NULL. PMDs initialize this
7059  *   structure in case of error only.
7060  * @return
7061  *   0 on success, a nagative value otherwise.
7062  */
7063 int
7064 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7065 		   FILE *file,
7066 		   struct rte_flow_error *error __rte_unused)
7067 {
7068 	struct mlx5_priv *priv = dev->data->dev_private;
7069 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7070 
7071 	if (!priv->config.dv_flow_en) {
7072 		if (fputs("device dv flow disabled\n", file) <= 0)
7073 			return -errno;
7074 		return -ENOTSUP;
7075 	}
7076 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7077 				       sh->tx_domain, file);
7078 }
7079 
7080 /**
7081  * Get aged-out flows.
7082  *
7083  * @param[in] dev
7084  *   Pointer to the Ethernet device structure.
7085  * @param[in] context
7086  *   The address of an array of pointers to the aged-out flows contexts.
7087  * @param[in] nb_countexts
7088  *   The length of context array pointers.
7089  * @param[out] error
7090  *   Perform verbose error reporting if not NULL. Initialized in case of
7091  *   error only.
7092  *
7093  * @return
7094  *   how many contexts get in success, otherwise negative errno value.
7095  *   if nb_contexts is 0, return the amount of all aged contexts.
7096  *   if nb_contexts is not 0 , return the amount of aged flows reported
7097  *   in the context array.
7098  */
7099 int
7100 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7101 			uint32_t nb_contexts, struct rte_flow_error *error)
7102 {
7103 	const struct mlx5_flow_driver_ops *fops;
7104 	struct rte_flow_attr attr = { .transfer = 0 };
7105 
7106 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7107 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7108 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7109 						    error);
7110 	}
7111 	DRV_LOG(ERR,
7112 		"port %u get aged flows is not supported.",
7113 		 dev->data->port_id);
7114 	return -ENOTSUP;
7115 }
7116 
7117 /* Wrapper for driver action_validate op callback */
7118 static int
7119 flow_drv_action_validate(struct rte_eth_dev *dev,
7120 			 const struct rte_flow_shared_action_conf *conf,
7121 			 const struct rte_flow_action *action,
7122 			 const struct mlx5_flow_driver_ops *fops,
7123 			 struct rte_flow_error *error)
7124 {
7125 	static const char err_msg[] = "shared action validation unsupported";
7126 
7127 	if (!fops->action_validate) {
7128 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7129 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7130 				   NULL, err_msg);
7131 		return -rte_errno;
7132 	}
7133 	return fops->action_validate(dev, conf, action, error);
7134 }
7135 
7136 /**
7137  * Destroys the shared action by handle.
7138  *
7139  * @param dev
7140  *   Pointer to Ethernet device structure.
7141  * @param[in] action
7142  *   Handle for the shared action to be destroyed.
7143  * @param[out] error
7144  *   Perform verbose error reporting if not NULL. PMDs initialize this
7145  *   structure in case of error only.
7146  *
7147  * @return
7148  *   0 on success, a negative errno value otherwise and rte_errno is set.
7149  *
7150  * @note: wrapper for driver action_create op callback.
7151  */
7152 static int
7153 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7154 			   struct rte_flow_shared_action *action,
7155 			   struct rte_flow_error *error)
7156 {
7157 	static const char err_msg[] = "shared action destruction unsupported";
7158 	struct rte_flow_attr attr = { .transfer = 0 };
7159 	const struct mlx5_flow_driver_ops *fops =
7160 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7161 
7162 	if (!fops->action_destroy) {
7163 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7164 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7165 				   NULL, err_msg);
7166 		return -rte_errno;
7167 	}
7168 	return fops->action_destroy(dev, action, error);
7169 }
7170 
7171 /* Wrapper for driver action_destroy op callback */
7172 static int
7173 flow_drv_action_update(struct rte_eth_dev *dev,
7174 		       struct rte_flow_shared_action *action,
7175 		       const void *action_conf,
7176 		       const struct mlx5_flow_driver_ops *fops,
7177 		       struct rte_flow_error *error)
7178 {
7179 	static const char err_msg[] = "shared action update unsupported";
7180 
7181 	if (!fops->action_update) {
7182 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7183 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7184 				   NULL, err_msg);
7185 		return -rte_errno;
7186 	}
7187 	return fops->action_update(dev, action, action_conf, error);
7188 }
7189 
7190 /* Wrapper for driver action_destroy op callback */
7191 static int
7192 flow_drv_action_query(struct rte_eth_dev *dev,
7193 		      const struct rte_flow_shared_action *action,
7194 		      void *data,
7195 		      const struct mlx5_flow_driver_ops *fops,
7196 		      struct rte_flow_error *error)
7197 {
7198 	static const char err_msg[] = "shared action query unsupported";
7199 
7200 	if (!fops->action_query) {
7201 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7202 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7203 				   NULL, err_msg);
7204 		return -rte_errno;
7205 	}
7206 	return fops->action_query(dev, action, data, error);
7207 }
7208 
7209 /**
7210  * Create shared action for reuse in multiple flow rules.
7211  *
7212  * @param dev
7213  *   Pointer to Ethernet device structure.
7214  * @param[in] action
7215  *   Action configuration for shared action creation.
7216  * @param[out] error
7217  *   Perform verbose error reporting if not NULL. PMDs initialize this
7218  *   structure in case of error only.
7219  * @return
7220  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7221  */
7222 static struct rte_flow_shared_action *
7223 mlx5_shared_action_create(struct rte_eth_dev *dev,
7224 			  const struct rte_flow_shared_action_conf *conf,
7225 			  const struct rte_flow_action *action,
7226 			  struct rte_flow_error *error)
7227 {
7228 	static const char err_msg[] = "shared action creation unsupported";
7229 	struct rte_flow_attr attr = { .transfer = 0 };
7230 	const struct mlx5_flow_driver_ops *fops =
7231 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7232 
7233 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7234 		return NULL;
7235 	if (!fops->action_create) {
7236 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7237 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7238 				   NULL, err_msg);
7239 		return NULL;
7240 	}
7241 	return fops->action_create(dev, conf, action, error);
7242 }
7243 
7244 /**
7245  * Updates inplace the shared action configuration pointed by *action* handle
7246  * with the configuration provided as *action* argument.
7247  * The update of the shared action configuration effects all flow rules reusing
7248  * the action via handle.
7249  *
7250  * @param dev
7251  *   Pointer to Ethernet device structure.
7252  * @param[in] shared_action
7253  *   Handle for the shared action to be updated.
7254  * @param[in] action
7255  *   Action specification used to modify the action pointed by handle.
7256  *   *action* should be of same type with the action pointed by the *action*
7257  *   handle argument, otherwise considered as invalid.
7258  * @param[out] error
7259  *   Perform verbose error reporting if not NULL. PMDs initialize this
7260  *   structure in case of error only.
7261  *
7262  * @return
7263  *   0 on success, a negative errno value otherwise and rte_errno is set.
7264  */
7265 static int
7266 mlx5_shared_action_update(struct rte_eth_dev *dev,
7267 		struct rte_flow_shared_action *shared_action,
7268 		const struct rte_flow_action *action,
7269 		struct rte_flow_error *error)
7270 {
7271 	struct rte_flow_attr attr = { .transfer = 0 };
7272 	const struct mlx5_flow_driver_ops *fops =
7273 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7274 	int ret;
7275 
7276 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7277 	if (ret)
7278 		return ret;
7279 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7280 				      error);
7281 }
7282 
7283 /**
7284  * Query the shared action by handle.
7285  *
7286  * This function allows retrieving action-specific data such as counters.
7287  * Data is gathered by special action which may be present/referenced in
7288  * more than one flow rule definition.
7289  *
7290  * \see RTE_FLOW_ACTION_TYPE_COUNT
7291  *
7292  * @param dev
7293  *   Pointer to Ethernet device structure.
7294  * @param[in] action
7295  *   Handle for the shared action to query.
7296  * @param[in, out] data
7297  *   Pointer to storage for the associated query data type.
7298  * @param[out] error
7299  *   Perform verbose error reporting if not NULL. PMDs initialize this
7300  *   structure in case of error only.
7301  *
7302  * @return
7303  *   0 on success, a negative errno value otherwise and rte_errno is set.
7304  */
7305 static int
7306 mlx5_shared_action_query(struct rte_eth_dev *dev,
7307 			 const struct rte_flow_shared_action *action,
7308 			 void *data,
7309 			 struct rte_flow_error *error)
7310 {
7311 	struct rte_flow_attr attr = { .transfer = 0 };
7312 	const struct mlx5_flow_driver_ops *fops =
7313 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7314 
7315 	return flow_drv_action_query(dev, action, data, fops, error);
7316 }
7317 
7318 /**
7319  * Destroy all shared actions.
7320  *
7321  * @param dev
7322  *   Pointer to Ethernet device.
7323  *
7324  * @return
7325  *   0 on success, a negative errno value otherwise and rte_errno is set.
7326  */
7327 int
7328 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7329 {
7330 	struct rte_flow_error error;
7331 	struct mlx5_priv *priv = dev->data->dev_private;
7332 	struct mlx5_shared_action_rss *action;
7333 	int ret = 0;
7334 	uint32_t idx;
7335 
7336 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7337 		      priv->rss_shared_actions, idx, action, next) {
7338 		ret |= mlx5_shared_action_destroy(dev,
7339 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7340 	}
7341 	return ret;
7342 }
7343 
7344 #ifndef HAVE_MLX5DV_DR
7345 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7346 #else
7347 #define MLX5_DOMAIN_SYNC_FLOW \
7348 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7349 #endif
7350 
7351 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7352 {
7353 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7354 	const struct mlx5_flow_driver_ops *fops;
7355 	int ret;
7356 	struct rte_flow_attr attr = { .transfer = 0 };
7357 
7358 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7359 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7360 	if (ret > 0)
7361 		ret = -ret;
7362 	return ret;
7363 }
7364 
7365 /**
7366  * tunnel offload functionalilty is defined for DV environment only
7367  */
7368 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7369 __extension__
7370 union tunnel_offload_mark {
7371 	uint32_t val;
7372 	struct {
7373 		uint32_t app_reserve:8;
7374 		uint32_t table_id:15;
7375 		uint32_t transfer:1;
7376 		uint32_t _unused_:8;
7377 	};
7378 };
7379 
7380 static bool
7381 mlx5_access_tunnel_offload_db
7382 	(struct rte_eth_dev *dev,
7383 	 bool (*match)(struct rte_eth_dev *,
7384 		       struct mlx5_flow_tunnel *, const void *),
7385 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7386 	 void (*miss)(struct rte_eth_dev *, void *),
7387 	 void *ctx, bool lock_op);
7388 
7389 static int
7390 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7391 			     struct rte_flow *flow,
7392 			     const struct rte_flow_attr *attr,
7393 			     const struct rte_flow_action *app_actions,
7394 			     uint32_t flow_idx,
7395 			     struct tunnel_default_miss_ctx *ctx,
7396 			     struct rte_flow_error *error)
7397 {
7398 	struct mlx5_priv *priv = dev->data->dev_private;
7399 	struct mlx5_flow *dev_flow;
7400 	struct rte_flow_attr miss_attr = *attr;
7401 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7402 	const struct rte_flow_item miss_items[2] = {
7403 		{
7404 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7405 			.spec = NULL,
7406 			.last = NULL,
7407 			.mask = NULL
7408 		},
7409 		{
7410 			.type = RTE_FLOW_ITEM_TYPE_END,
7411 			.spec = NULL,
7412 			.last = NULL,
7413 			.mask = NULL
7414 		}
7415 	};
7416 	union tunnel_offload_mark mark_id;
7417 	struct rte_flow_action_mark miss_mark;
7418 	struct rte_flow_action miss_actions[3] = {
7419 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7420 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7421 	};
7422 	const struct rte_flow_action_jump *jump_data;
7423 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7424 	struct flow_grp_info grp_info = {
7425 		.external = 1,
7426 		.transfer = attr->transfer,
7427 		.fdb_def_rule = !!priv->fdb_def_rule,
7428 		.std_tbl_fix = 0,
7429 	};
7430 	int ret;
7431 
7432 	if (!attr->transfer) {
7433 		uint32_t q_size;
7434 
7435 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7436 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7437 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7438 					 0, SOCKET_ID_ANY);
7439 		if (!ctx->queue)
7440 			return rte_flow_error_set
7441 				(error, ENOMEM,
7442 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7443 				NULL, "invalid default miss RSS");
7444 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7445 		ctx->action_rss.level = 0,
7446 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7447 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7448 		ctx->action_rss.queue_num = priv->reta_idx_n,
7449 		ctx->action_rss.key = priv->rss_conf.rss_key,
7450 		ctx->action_rss.queue = ctx->queue;
7451 		if (!priv->reta_idx_n || !priv->rxqs_n)
7452 			return rte_flow_error_set
7453 				(error, EINVAL,
7454 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7455 				NULL, "invalid port configuration");
7456 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7457 			ctx->action_rss.types = 0;
7458 		for (i = 0; i != priv->reta_idx_n; ++i)
7459 			ctx->queue[i] = (*priv->reta_idx)[i];
7460 	} else {
7461 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7462 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7463 	}
7464 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7465 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7466 	jump_data = app_actions->conf;
7467 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7468 	miss_attr.group = jump_data->group;
7469 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7470 				       &flow_table, &grp_info, error);
7471 	if (ret)
7472 		return rte_flow_error_set(error, EINVAL,
7473 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7474 					  NULL, "invalid tunnel id");
7475 	mark_id.app_reserve = 0;
7476 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7477 	mark_id.transfer = !!attr->transfer;
7478 	mark_id._unused_ = 0;
7479 	miss_mark.id = mark_id.val;
7480 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7481 				    miss_items, miss_actions, flow_idx, error);
7482 	if (!dev_flow)
7483 		return -rte_errno;
7484 	dev_flow->flow = flow;
7485 	dev_flow->external = true;
7486 	dev_flow->tunnel = tunnel;
7487 	/* Subflow object was created, we must include one in the list. */
7488 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7489 		      dev_flow->handle, next);
7490 	DRV_LOG(DEBUG,
7491 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7492 		dev->data->port_id, tunnel->app_tunnel.type,
7493 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7494 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7495 				  miss_actions, error);
7496 	if (!ret)
7497 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7498 						  error);
7499 
7500 	return ret;
7501 }
7502 
7503 static const struct mlx5_flow_tbl_data_entry  *
7504 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7505 {
7506 	struct mlx5_priv *priv = dev->data->dev_private;
7507 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7508 	struct mlx5_hlist_entry *he;
7509 	union tunnel_offload_mark mbits = { .val = mark };
7510 	union mlx5_flow_tbl_key table_key = {
7511 		{
7512 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7513 			.dummy = 0,
7514 			.domain = !!mbits.transfer,
7515 			.direction = 0,
7516 		}
7517 	};
7518 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7519 	return he ?
7520 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7521 }
7522 
7523 static void
7524 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7525 				   struct mlx5_hlist_entry *entry)
7526 {
7527 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7528 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7529 
7530 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7531 			tunnel_flow_tbl_to_id(tte->flow_table));
7532 	mlx5_free(tte);
7533 }
7534 
7535 static int
7536 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7537 				  struct mlx5_hlist_entry *entry,
7538 				  uint64_t key, void *cb_ctx __rte_unused)
7539 {
7540 	union tunnel_tbl_key tbl = {
7541 		.val = key,
7542 	};
7543 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7544 
7545 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7546 }
7547 
7548 static struct mlx5_hlist_entry *
7549 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7550 				   void *ctx __rte_unused)
7551 {
7552 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7553 	struct tunnel_tbl_entry *tte;
7554 	union tunnel_tbl_key tbl = {
7555 		.val = key,
7556 	};
7557 
7558 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7559 			  sizeof(*tte), 0,
7560 			  SOCKET_ID_ANY);
7561 	if (!tte)
7562 		goto err;
7563 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7564 			  &tte->flow_table);
7565 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7566 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7567 			tte->flow_table);
7568 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7569 				tte->flow_table);
7570 		goto err;
7571 	} else if (!tte->flow_table) {
7572 		goto err;
7573 	}
7574 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7575 	tte->tunnel_id = tbl.tunnel_id;
7576 	tte->group = tbl.group;
7577 	return &tte->hash;
7578 err:
7579 	if (tte)
7580 		mlx5_free(tte);
7581 	return NULL;
7582 }
7583 
7584 static uint32_t
7585 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7586 				const struct mlx5_flow_tunnel *tunnel,
7587 				uint32_t group, uint32_t *table,
7588 				struct rte_flow_error *error)
7589 {
7590 	struct mlx5_hlist_entry *he;
7591 	struct tunnel_tbl_entry *tte;
7592 	union tunnel_tbl_key key = {
7593 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7594 		.group = group
7595 	};
7596 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7597 	struct mlx5_hlist *group_hash;
7598 
7599 	group_hash = tunnel ? tunnel->groups : thub->groups;
7600 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7601 	if (!he)
7602 		return rte_flow_error_set(error, EINVAL,
7603 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7604 					  NULL,
7605 					  "tunnel group index not supported");
7606 	tte = container_of(he, typeof(*tte), hash);
7607 	*table = tte->flow_table;
7608 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7609 		dev->data->port_id, key.tunnel_id, group, *table);
7610 	return 0;
7611 }
7612 
7613 static void
7614 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7615 		      struct mlx5_flow_tunnel *tunnel)
7616 {
7617 	struct mlx5_priv *priv = dev->data->dev_private;
7618 	struct mlx5_indexed_pool *ipool;
7619 
7620 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7621 		dev->data->port_id, tunnel->tunnel_id);
7622 	LIST_REMOVE(tunnel, chain);
7623 	mlx5_hlist_destroy(tunnel->groups);
7624 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7625 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7626 }
7627 
7628 static bool
7629 mlx5_access_tunnel_offload_db
7630 	(struct rte_eth_dev *dev,
7631 	 bool (*match)(struct rte_eth_dev *,
7632 		       struct mlx5_flow_tunnel *, const void *),
7633 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7634 	 void (*miss)(struct rte_eth_dev *, void *),
7635 	 void *ctx, bool lock_op)
7636 {
7637 	bool verdict = false;
7638 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7639 	struct mlx5_flow_tunnel *tunnel;
7640 
7641 	rte_spinlock_lock(&thub->sl);
7642 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7643 		verdict = match(dev, tunnel, (const void *)ctx);
7644 		if (verdict)
7645 			break;
7646 	}
7647 	if (!lock_op)
7648 		rte_spinlock_unlock(&thub->sl);
7649 	if (verdict && hit)
7650 		hit(dev, tunnel, ctx);
7651 	if (!verdict && miss)
7652 		miss(dev, ctx);
7653 	if (lock_op)
7654 		rte_spinlock_unlock(&thub->sl);
7655 
7656 	return verdict;
7657 }
7658 
7659 struct tunnel_db_find_tunnel_id_ctx {
7660 	uint32_t tunnel_id;
7661 	struct mlx5_flow_tunnel *tunnel;
7662 };
7663 
7664 static bool
7665 find_tunnel_id_match(struct rte_eth_dev *dev,
7666 		     struct mlx5_flow_tunnel *tunnel, const void *x)
7667 {
7668 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7669 
7670 	RTE_SET_USED(dev);
7671 	return tunnel->tunnel_id == ctx->tunnel_id;
7672 }
7673 
7674 static void
7675 find_tunnel_id_hit(struct rte_eth_dev *dev,
7676 		   struct mlx5_flow_tunnel *tunnel, void *x)
7677 {
7678 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7679 	RTE_SET_USED(dev);
7680 	ctx->tunnel = tunnel;
7681 }
7682 
7683 static struct mlx5_flow_tunnel *
7684 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7685 {
7686 	struct tunnel_db_find_tunnel_id_ctx ctx = {
7687 		.tunnel_id = id,
7688 	};
7689 
7690 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7691 				      find_tunnel_id_hit, NULL, &ctx, true);
7692 
7693 	return ctx.tunnel;
7694 }
7695 
7696 static struct mlx5_flow_tunnel *
7697 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7698 			  const struct rte_flow_tunnel *app_tunnel)
7699 {
7700 	struct mlx5_priv *priv = dev->data->dev_private;
7701 	struct mlx5_indexed_pool *ipool;
7702 	struct mlx5_flow_tunnel *tunnel;
7703 	uint32_t id;
7704 
7705 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7706 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7707 	if (!tunnel)
7708 		return NULL;
7709 	if (id >= MLX5_MAX_TUNNELS) {
7710 		mlx5_ipool_free(ipool, id);
7711 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7712 		return NULL;
7713 	}
7714 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7715 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7716 					   mlx5_flow_tunnel_grp2tbl_match_cb,
7717 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7718 	if (!tunnel->groups) {
7719 		mlx5_ipool_free(ipool, id);
7720 		return NULL;
7721 	}
7722 	tunnel->groups->ctx = priv->sh;
7723 	/* initiate new PMD tunnel */
7724 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7725 	tunnel->tunnel_id = id;
7726 	tunnel->action.type = (typeof(tunnel->action.type))
7727 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7728 	tunnel->action.conf = tunnel;
7729 	tunnel->item.type = (typeof(tunnel->item.type))
7730 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7731 	tunnel->item.spec = tunnel;
7732 	tunnel->item.last = NULL;
7733 	tunnel->item.mask = NULL;
7734 
7735 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7736 		dev->data->port_id, tunnel->tunnel_id);
7737 
7738 	return tunnel;
7739 }
7740 
7741 struct tunnel_db_get_tunnel_ctx {
7742 	const struct rte_flow_tunnel *app_tunnel;
7743 	struct mlx5_flow_tunnel *tunnel;
7744 };
7745 
7746 static bool get_tunnel_match(struct rte_eth_dev *dev,
7747 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7748 {
7749 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
7750 
7751 	RTE_SET_USED(dev);
7752 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7753 		       sizeof(*ctx->app_tunnel));
7754 }
7755 
7756 static void get_tunnel_hit(struct rte_eth_dev *dev,
7757 			   struct mlx5_flow_tunnel *tunnel, void *x)
7758 {
7759 	/* called under tunnel spinlock protection */
7760 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7761 
7762 	RTE_SET_USED(dev);
7763 	tunnel->refctn++;
7764 	ctx->tunnel = tunnel;
7765 }
7766 
7767 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7768 {
7769 	/* called under tunnel spinlock protection */
7770 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7771 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7772 
7773 	rte_spinlock_unlock(&thub->sl);
7774 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7775 	ctx->tunnel->refctn = 1;
7776 	rte_spinlock_lock(&thub->sl);
7777 	if (ctx->tunnel)
7778 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7779 }
7780 
7781 
7782 static int
7783 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7784 		     const struct rte_flow_tunnel *app_tunnel,
7785 		     struct mlx5_flow_tunnel **tunnel)
7786 {
7787 	struct tunnel_db_get_tunnel_ctx ctx = {
7788 		.app_tunnel = app_tunnel,
7789 	};
7790 
7791 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7792 				      get_tunnel_miss, &ctx, true);
7793 	*tunnel = ctx.tunnel;
7794 	return ctx.tunnel ? 0 : -ENOMEM;
7795 }
7796 
7797 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7798 {
7799 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7800 
7801 	if (!thub)
7802 		return;
7803 	if (!LIST_EMPTY(&thub->tunnels))
7804 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7805 	mlx5_hlist_destroy(thub->groups);
7806 	mlx5_free(thub);
7807 }
7808 
7809 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7810 {
7811 	int err;
7812 	struct mlx5_flow_tunnel_hub *thub;
7813 
7814 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7815 			   0, SOCKET_ID_ANY);
7816 	if (!thub)
7817 		return -ENOMEM;
7818 	LIST_INIT(&thub->tunnels);
7819 	rte_spinlock_init(&thub->sl);
7820 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7821 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7822 					 mlx5_flow_tunnel_grp2tbl_match_cb,
7823 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7824 	if (!thub->groups) {
7825 		err = -rte_errno;
7826 		goto err;
7827 	}
7828 	thub->groups->ctx = sh;
7829 	sh->tunnel_hub = thub;
7830 
7831 	return 0;
7832 
7833 err:
7834 	if (thub->groups)
7835 		mlx5_hlist_destroy(thub->groups);
7836 	if (thub)
7837 		mlx5_free(thub);
7838 	return err;
7839 }
7840 
7841 static inline bool
7842 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7843 			  struct rte_flow_tunnel *tunnel,
7844 			  const char *err_msg)
7845 {
7846 	err_msg = NULL;
7847 	if (!is_tunnel_offload_active(dev)) {
7848 		err_msg = "tunnel offload was not activated";
7849 		goto out;
7850 	} else if (!tunnel) {
7851 		err_msg = "no application tunnel";
7852 		goto out;
7853 	}
7854 
7855 	switch (tunnel->type) {
7856 	default:
7857 		err_msg = "unsupported tunnel type";
7858 		goto out;
7859 	case RTE_FLOW_ITEM_TYPE_VXLAN:
7860 		break;
7861 	}
7862 
7863 out:
7864 	return !err_msg;
7865 }
7866 
7867 static int
7868 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7869 		    struct rte_flow_tunnel *app_tunnel,
7870 		    struct rte_flow_action **actions,
7871 		    uint32_t *num_of_actions,
7872 		    struct rte_flow_error *error)
7873 {
7874 	int ret;
7875 	struct mlx5_flow_tunnel *tunnel;
7876 	const char *err_msg = NULL;
7877 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7878 
7879 	if (!verdict)
7880 		return rte_flow_error_set(error, EINVAL,
7881 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7882 					  err_msg);
7883 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7884 	if (ret < 0) {
7885 		return rte_flow_error_set(error, ret,
7886 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7887 					  "failed to initialize pmd tunnel");
7888 	}
7889 	*actions = &tunnel->action;
7890 	*num_of_actions = 1;
7891 	return 0;
7892 }
7893 
7894 static int
7895 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7896 		       struct rte_flow_tunnel *app_tunnel,
7897 		       struct rte_flow_item **items,
7898 		       uint32_t *num_of_items,
7899 		       struct rte_flow_error *error)
7900 {
7901 	int ret;
7902 	struct mlx5_flow_tunnel *tunnel;
7903 	const char *err_msg = NULL;
7904 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7905 
7906 	if (!verdict)
7907 		return rte_flow_error_set(error, EINVAL,
7908 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7909 					  err_msg);
7910 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7911 	if (ret < 0) {
7912 		return rte_flow_error_set(error, ret,
7913 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7914 					  "failed to initialize pmd tunnel");
7915 	}
7916 	*items = &tunnel->item;
7917 	*num_of_items = 1;
7918 	return 0;
7919 }
7920 
7921 struct tunnel_db_element_release_ctx {
7922 	struct rte_flow_item *items;
7923 	struct rte_flow_action *actions;
7924 	uint32_t num_elements;
7925 	struct rte_flow_error *error;
7926 	int ret;
7927 };
7928 
7929 static bool
7930 tunnel_element_release_match(struct rte_eth_dev *dev,
7931 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7932 {
7933 	const struct tunnel_db_element_release_ctx *ctx = x;
7934 
7935 	RTE_SET_USED(dev);
7936 	if (ctx->num_elements != 1)
7937 		return false;
7938 	else if (ctx->items)
7939 		return ctx->items == &tunnel->item;
7940 	else if (ctx->actions)
7941 		return ctx->actions == &tunnel->action;
7942 
7943 	return false;
7944 }
7945 
7946 static void
7947 tunnel_element_release_hit(struct rte_eth_dev *dev,
7948 			   struct mlx5_flow_tunnel *tunnel, void *x)
7949 {
7950 	struct tunnel_db_element_release_ctx *ctx = x;
7951 	ctx->ret = 0;
7952 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
7953 		mlx5_flow_tunnel_free(dev, tunnel);
7954 }
7955 
7956 static void
7957 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
7958 {
7959 	struct tunnel_db_element_release_ctx *ctx = x;
7960 	RTE_SET_USED(dev);
7961 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
7962 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7963 				      "invalid argument");
7964 }
7965 
7966 static int
7967 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
7968 		       struct rte_flow_item *pmd_items,
7969 		       uint32_t num_items, struct rte_flow_error *err)
7970 {
7971 	struct tunnel_db_element_release_ctx ctx = {
7972 		.items = pmd_items,
7973 		.actions = NULL,
7974 		.num_elements = num_items,
7975 		.error = err,
7976 	};
7977 
7978 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
7979 				      tunnel_element_release_hit,
7980 				      tunnel_element_release_miss, &ctx, false);
7981 
7982 	return ctx.ret;
7983 }
7984 
7985 static int
7986 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
7987 			 struct rte_flow_action *pmd_actions,
7988 			 uint32_t num_actions, struct rte_flow_error *err)
7989 {
7990 	struct tunnel_db_element_release_ctx ctx = {
7991 		.items = NULL,
7992 		.actions = pmd_actions,
7993 		.num_elements = num_actions,
7994 		.error = err,
7995 	};
7996 
7997 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
7998 				      tunnel_element_release_hit,
7999 				      tunnel_element_release_miss, &ctx, false);
8000 
8001 	return ctx.ret;
8002 }
8003 
8004 static int
8005 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8006 				  struct rte_mbuf *m,
8007 				  struct rte_flow_restore_info *info,
8008 				  struct rte_flow_error *err)
8009 {
8010 	uint64_t ol_flags = m->ol_flags;
8011 	const struct mlx5_flow_tbl_data_entry *tble;
8012 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8013 
8014 	if (!is_tunnel_offload_active(dev)) {
8015 		info->flags = 0;
8016 		return 0;
8017 	}
8018 
8019 	if ((ol_flags & mask) != mask)
8020 		goto err;
8021 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8022 	if (!tble) {
8023 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8024 			dev->data->port_id, m->hash.fdir.hi);
8025 		goto err;
8026 	}
8027 	MLX5_ASSERT(tble->tunnel);
8028 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8029 	info->group_id = tble->group_id;
8030 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8031 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8032 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8033 
8034 	return 0;
8035 
8036 err:
8037 	return rte_flow_error_set(err, EINVAL,
8038 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8039 				  "failed to get restore info");
8040 }
8041 
8042 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8043 static int
8044 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8045 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8046 			   __rte_unused struct rte_flow_action **actions,
8047 			   __rte_unused uint32_t *num_of_actions,
8048 			   __rte_unused struct rte_flow_error *error)
8049 {
8050 	return -ENOTSUP;
8051 }
8052 
8053 static int
8054 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8055 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8056 		       __rte_unused struct rte_flow_item **items,
8057 		       __rte_unused uint32_t *num_of_items,
8058 		       __rte_unused struct rte_flow_error *error)
8059 {
8060 	return -ENOTSUP;
8061 }
8062 
8063 static int
8064 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8065 			      __rte_unused struct rte_flow_item *pmd_items,
8066 			      __rte_unused uint32_t num_items,
8067 			      __rte_unused struct rte_flow_error *err)
8068 {
8069 	return -ENOTSUP;
8070 }
8071 
8072 static int
8073 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8074 				__rte_unused struct rte_flow_action *pmd_action,
8075 				__rte_unused uint32_t num_actions,
8076 				__rte_unused struct rte_flow_error *err)
8077 {
8078 	return -ENOTSUP;
8079 }
8080 
8081 static int
8082 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8083 				  __rte_unused struct rte_mbuf *m,
8084 				  __rte_unused struct rte_flow_restore_info *i,
8085 				  __rte_unused struct rte_flow_error *err)
8086 {
8087 	return -ENOTSUP;
8088 }
8089 
8090 static int
8091 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8092 			     __rte_unused struct rte_flow *flow,
8093 			     __rte_unused const struct rte_flow_attr *attr,
8094 			     __rte_unused const struct rte_flow_action *actions,
8095 			     __rte_unused uint32_t flow_idx,
8096 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8097 			     __rte_unused struct rte_flow_error *error)
8098 {
8099 	return -ENOTSUP;
8100 }
8101 
8102 static struct mlx5_flow_tunnel *
8103 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8104 		    __rte_unused uint32_t id)
8105 {
8106 	return NULL;
8107 }
8108 
8109 static void
8110 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8111 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8112 {
8113 }
8114 
8115 static uint32_t
8116 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8117 				__rte_unused const struct mlx5_flow_tunnel *t,
8118 				__rte_unused uint32_t group,
8119 				__rte_unused uint32_t *table,
8120 				struct rte_flow_error *error)
8121 {
8122 	return rte_flow_error_set(error, ENOTSUP,
8123 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8124 				  "tunnel offload requires DV support");
8125 }
8126 
8127 void
8128 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8129 			__rte_unused  uint16_t port_id)
8130 {
8131 }
8132 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8133 
8134