xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b7fe612ac1de393f869c9818d5503633c8e96b36)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rx.h"
32 #include "mlx5_tx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 struct tunnel_default_miss_ctx {
37 	uint16_t *queue;
38 	__extension__
39 	union {
40 		struct rte_flow_action_rss action_rss;
41 		struct rte_flow_action_queue miss_queue;
42 		struct rte_flow_action_jump miss_jump;
43 		uint8_t raw[0];
44 	};
45 };
46 
47 static int
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 			     struct rte_flow *flow,
50 			     const struct rte_flow_attr *attr,
51 			     const struct rte_flow_action *app_actions,
52 			     uint32_t flow_idx,
53 			     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 	return lsize;
389 }
390 
391 enum mlx5_expansion {
392 	MLX5_EXPANSION_ROOT,
393 	MLX5_EXPANSION_ROOT_OUTER,
394 	MLX5_EXPANSION_ROOT_ETH_VLAN,
395 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
396 	MLX5_EXPANSION_OUTER_ETH,
397 	MLX5_EXPANSION_OUTER_ETH_VLAN,
398 	MLX5_EXPANSION_OUTER_VLAN,
399 	MLX5_EXPANSION_OUTER_IPV4,
400 	MLX5_EXPANSION_OUTER_IPV4_UDP,
401 	MLX5_EXPANSION_OUTER_IPV4_TCP,
402 	MLX5_EXPANSION_OUTER_IPV6,
403 	MLX5_EXPANSION_OUTER_IPV6_UDP,
404 	MLX5_EXPANSION_OUTER_IPV6_TCP,
405 	MLX5_EXPANSION_VXLAN,
406 	MLX5_EXPANSION_VXLAN_GPE,
407 	MLX5_EXPANSION_GRE,
408 	MLX5_EXPANSION_MPLS,
409 	MLX5_EXPANSION_ETH,
410 	MLX5_EXPANSION_ETH_VLAN,
411 	MLX5_EXPANSION_VLAN,
412 	MLX5_EXPANSION_IPV4,
413 	MLX5_EXPANSION_IPV4_UDP,
414 	MLX5_EXPANSION_IPV4_TCP,
415 	MLX5_EXPANSION_IPV6,
416 	MLX5_EXPANSION_IPV6_UDP,
417 	MLX5_EXPANSION_IPV6_TCP,
418 };
419 
420 /** Supported expansion of items. */
421 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
422 	[MLX5_EXPANSION_ROOT] = {
423 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
424 						  MLX5_EXPANSION_IPV4,
425 						  MLX5_EXPANSION_IPV6),
426 		.type = RTE_FLOW_ITEM_TYPE_END,
427 	},
428 	[MLX5_EXPANSION_ROOT_OUTER] = {
429 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
430 						  MLX5_EXPANSION_OUTER_IPV4,
431 						  MLX5_EXPANSION_OUTER_IPV6),
432 		.type = RTE_FLOW_ITEM_TYPE_END,
433 	},
434 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
435 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
436 		.type = RTE_FLOW_ITEM_TYPE_END,
437 	},
438 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
439 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
440 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
441 		.type = RTE_FLOW_ITEM_TYPE_END,
442 	},
443 	[MLX5_EXPANSION_OUTER_ETH] = {
444 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
445 						  MLX5_EXPANSION_OUTER_IPV6,
446 						  MLX5_EXPANSION_MPLS),
447 		.type = RTE_FLOW_ITEM_TYPE_ETH,
448 		.rss_types = 0,
449 	},
450 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
451 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
452 		.type = RTE_FLOW_ITEM_TYPE_ETH,
453 		.rss_types = 0,
454 	},
455 	[MLX5_EXPANSION_OUTER_VLAN] = {
456 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
457 						  MLX5_EXPANSION_OUTER_IPV6),
458 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
459 	},
460 	[MLX5_EXPANSION_OUTER_IPV4] = {
461 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
462 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
463 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
464 			 MLX5_EXPANSION_GRE,
465 			 MLX5_EXPANSION_IPV4,
466 			 MLX5_EXPANSION_IPV6),
467 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
468 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
469 			ETH_RSS_NONFRAG_IPV4_OTHER,
470 	},
471 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
472 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
473 						  MLX5_EXPANSION_VXLAN_GPE),
474 		.type = RTE_FLOW_ITEM_TYPE_UDP,
475 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
478 		.type = RTE_FLOW_ITEM_TYPE_TCP,
479 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
480 	},
481 	[MLX5_EXPANSION_OUTER_IPV6] = {
482 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
483 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
484 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
485 			 MLX5_EXPANSION_IPV4,
486 			 MLX5_EXPANSION_IPV6,
487 			 MLX5_EXPANSION_GRE),
488 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
489 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
490 			ETH_RSS_NONFRAG_IPV6_OTHER,
491 	},
492 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
493 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
494 						  MLX5_EXPANSION_VXLAN_GPE),
495 		.type = RTE_FLOW_ITEM_TYPE_UDP,
496 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
497 	},
498 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
499 		.type = RTE_FLOW_ITEM_TYPE_TCP,
500 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
501 	},
502 	[MLX5_EXPANSION_VXLAN] = {
503 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
504 						  MLX5_EXPANSION_IPV4,
505 						  MLX5_EXPANSION_IPV6),
506 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
507 	},
508 	[MLX5_EXPANSION_VXLAN_GPE] = {
509 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
510 						  MLX5_EXPANSION_IPV4,
511 						  MLX5_EXPANSION_IPV6),
512 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
513 	},
514 	[MLX5_EXPANSION_GRE] = {
515 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
516 						  MLX5_EXPANSION_IPV6),
517 		.type = RTE_FLOW_ITEM_TYPE_GRE,
518 	},
519 	[MLX5_EXPANSION_MPLS] = {
520 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
521 						  MLX5_EXPANSION_IPV6),
522 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
523 	},
524 	[MLX5_EXPANSION_ETH] = {
525 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
526 						  MLX5_EXPANSION_IPV6),
527 		.type = RTE_FLOW_ITEM_TYPE_ETH,
528 	},
529 	[MLX5_EXPANSION_ETH_VLAN] = {
530 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
531 		.type = RTE_FLOW_ITEM_TYPE_ETH,
532 	},
533 	[MLX5_EXPANSION_VLAN] = {
534 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 						  MLX5_EXPANSION_IPV6),
536 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
537 	},
538 	[MLX5_EXPANSION_IPV4] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
540 						  MLX5_EXPANSION_IPV4_TCP),
541 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
542 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
543 			ETH_RSS_NONFRAG_IPV4_OTHER,
544 	},
545 	[MLX5_EXPANSION_IPV4_UDP] = {
546 		.type = RTE_FLOW_ITEM_TYPE_UDP,
547 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
548 	},
549 	[MLX5_EXPANSION_IPV4_TCP] = {
550 		.type = RTE_FLOW_ITEM_TYPE_TCP,
551 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
552 	},
553 	[MLX5_EXPANSION_IPV6] = {
554 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
555 						  MLX5_EXPANSION_IPV6_TCP),
556 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
557 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
558 			ETH_RSS_NONFRAG_IPV6_OTHER,
559 	},
560 	[MLX5_EXPANSION_IPV6_UDP] = {
561 		.type = RTE_FLOW_ITEM_TYPE_UDP,
562 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
563 	},
564 	[MLX5_EXPANSION_IPV6_TCP] = {
565 		.type = RTE_FLOW_ITEM_TYPE_TCP,
566 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
567 	},
568 };
569 
570 static struct rte_flow_action_handle *
571 mlx5_action_handle_create(struct rte_eth_dev *dev,
572 			  const struct rte_flow_indir_action_conf *conf,
573 			  const struct rte_flow_action *action,
574 			  struct rte_flow_error *error);
575 static int mlx5_action_handle_destroy
576 				(struct rte_eth_dev *dev,
577 				 struct rte_flow_action_handle *handle,
578 				 struct rte_flow_error *error);
579 static int mlx5_action_handle_update
580 				(struct rte_eth_dev *dev,
581 				 struct rte_flow_action_handle *handle,
582 				 const void *update,
583 				 struct rte_flow_error *error);
584 static int mlx5_action_handle_query
585 				(struct rte_eth_dev *dev,
586 				 const struct rte_flow_action_handle *handle,
587 				 void *data,
588 				 struct rte_flow_error *error);
589 static int
590 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
591 		    struct rte_flow_tunnel *app_tunnel,
592 		    struct rte_flow_action **actions,
593 		    uint32_t *num_of_actions,
594 		    struct rte_flow_error *error);
595 static int
596 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
597 		       struct rte_flow_tunnel *app_tunnel,
598 		       struct rte_flow_item **items,
599 		       uint32_t *num_of_items,
600 		       struct rte_flow_error *error);
601 static int
602 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
603 			      struct rte_flow_item *pmd_items,
604 			      uint32_t num_items, struct rte_flow_error *err);
605 static int
606 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
607 				struct rte_flow_action *pmd_actions,
608 				uint32_t num_actions,
609 				struct rte_flow_error *err);
610 static int
611 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
612 				  struct rte_mbuf *m,
613 				  struct rte_flow_restore_info *info,
614 				  struct rte_flow_error *err);
615 
616 static const struct rte_flow_ops mlx5_flow_ops = {
617 	.validate = mlx5_flow_validate,
618 	.create = mlx5_flow_create,
619 	.destroy = mlx5_flow_destroy,
620 	.flush = mlx5_flow_flush,
621 	.isolate = mlx5_flow_isolate,
622 	.query = mlx5_flow_query,
623 	.dev_dump = mlx5_flow_dev_dump,
624 	.get_aged_flows = mlx5_flow_get_aged_flows,
625 	.action_handle_create = mlx5_action_handle_create,
626 	.action_handle_destroy = mlx5_action_handle_destroy,
627 	.action_handle_update = mlx5_action_handle_update,
628 	.action_handle_query = mlx5_action_handle_query,
629 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
630 	.tunnel_match = mlx5_flow_tunnel_match,
631 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
632 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
633 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
634 };
635 
636 /* Tunnel information. */
637 struct mlx5_flow_tunnel_info {
638 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
639 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
640 };
641 
642 static struct mlx5_flow_tunnel_info tunnels_info[] = {
643 	{
644 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
645 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
646 	},
647 	{
648 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
649 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
650 	},
651 	{
652 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
653 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
654 	},
655 	{
656 		.tunnel = MLX5_FLOW_LAYER_GRE,
657 		.ptype = RTE_PTYPE_TUNNEL_GRE,
658 	},
659 	{
660 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
661 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
662 	},
663 	{
664 		.tunnel = MLX5_FLOW_LAYER_MPLS,
665 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
666 	},
667 	{
668 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
669 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
670 	},
671 	{
672 		.tunnel = MLX5_FLOW_LAYER_IPIP,
673 		.ptype = RTE_PTYPE_TUNNEL_IP,
674 	},
675 	{
676 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
677 		.ptype = RTE_PTYPE_TUNNEL_IP,
678 	},
679 	{
680 		.tunnel = MLX5_FLOW_LAYER_GTP,
681 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
682 	},
683 };
684 
685 
686 
687 /**
688  * Translate tag ID to register.
689  *
690  * @param[in] dev
691  *   Pointer to the Ethernet device structure.
692  * @param[in] feature
693  *   The feature that request the register.
694  * @param[in] id
695  *   The request register ID.
696  * @param[out] error
697  *   Error description in case of any.
698  *
699  * @return
700  *   The request register on success, a negative errno
701  *   value otherwise and rte_errno is set.
702  */
703 int
704 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
705 		     enum mlx5_feature_name feature,
706 		     uint32_t id,
707 		     struct rte_flow_error *error)
708 {
709 	struct mlx5_priv *priv = dev->data->dev_private;
710 	struct mlx5_dev_config *config = &priv->config;
711 	enum modify_reg start_reg;
712 	bool skip_mtr_reg = false;
713 
714 	switch (feature) {
715 	case MLX5_HAIRPIN_RX:
716 		return REG_B;
717 	case MLX5_HAIRPIN_TX:
718 		return REG_A;
719 	case MLX5_METADATA_RX:
720 		switch (config->dv_xmeta_en) {
721 		case MLX5_XMETA_MODE_LEGACY:
722 			return REG_B;
723 		case MLX5_XMETA_MODE_META16:
724 			return REG_C_0;
725 		case MLX5_XMETA_MODE_META32:
726 			return REG_C_1;
727 		}
728 		break;
729 	case MLX5_METADATA_TX:
730 		return REG_A;
731 	case MLX5_METADATA_FDB:
732 		switch (config->dv_xmeta_en) {
733 		case MLX5_XMETA_MODE_LEGACY:
734 			return REG_NON;
735 		case MLX5_XMETA_MODE_META16:
736 			return REG_C_0;
737 		case MLX5_XMETA_MODE_META32:
738 			return REG_C_1;
739 		}
740 		break;
741 	case MLX5_FLOW_MARK:
742 		switch (config->dv_xmeta_en) {
743 		case MLX5_XMETA_MODE_LEGACY:
744 			return REG_NON;
745 		case MLX5_XMETA_MODE_META16:
746 			return REG_C_1;
747 		case MLX5_XMETA_MODE_META32:
748 			return REG_C_0;
749 		}
750 		break;
751 	case MLX5_MTR_ID:
752 		/*
753 		 * If meter color and meter id share one register, flow match
754 		 * should use the meter color register for match.
755 		 */
756 		if (priv->mtr_reg_share)
757 			return priv->mtr_color_reg;
758 		else
759 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
760 			       REG_C_3;
761 	case MLX5_MTR_COLOR:
762 	case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
763 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
764 		return priv->mtr_color_reg;
765 	case MLX5_COPY_MARK:
766 		/*
767 		 * Metadata COPY_MARK register using is in meter suffix sub
768 		 * flow while with meter. It's safe to share the same register.
769 		 */
770 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
771 	case MLX5_APP_TAG:
772 		/*
773 		 * If meter is enable, it will engage the register for color
774 		 * match and flow match. If meter color match is not using the
775 		 * REG_C_2, need to skip the REG_C_x be used by meter color
776 		 * match.
777 		 * If meter is disable, free to use all available registers.
778 		 */
779 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
780 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
781 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
782 		if (id > (uint32_t)(REG_C_7 - start_reg))
783 			return rte_flow_error_set(error, EINVAL,
784 						  RTE_FLOW_ERROR_TYPE_ITEM,
785 						  NULL, "invalid tag id");
786 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
787 			return rte_flow_error_set(error, ENOTSUP,
788 						  RTE_FLOW_ERROR_TYPE_ITEM,
789 						  NULL, "unsupported tag id");
790 		/*
791 		 * This case means meter is using the REG_C_x great than 2.
792 		 * Take care not to conflict with meter color REG_C_x.
793 		 * If the available index REG_C_y >= REG_C_x, skip the
794 		 * color register.
795 		 */
796 		if (skip_mtr_reg && config->flow_mreg_c
797 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
798 			if (id >= (uint32_t)(REG_C_7 - start_reg))
799 				return rte_flow_error_set(error, EINVAL,
800 						       RTE_FLOW_ERROR_TYPE_ITEM,
801 							NULL, "invalid tag id");
802 			if (config->flow_mreg_c
803 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
804 				return config->flow_mreg_c
805 					       [id + 1 + start_reg - REG_C_0];
806 			return rte_flow_error_set(error, ENOTSUP,
807 						  RTE_FLOW_ERROR_TYPE_ITEM,
808 						  NULL, "unsupported tag id");
809 		}
810 		return config->flow_mreg_c[id + start_reg - REG_C_0];
811 	}
812 	MLX5_ASSERT(false);
813 	return rte_flow_error_set(error, EINVAL,
814 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
815 				  NULL, "invalid feature name");
816 }
817 
818 /**
819  * Check extensive flow metadata register support.
820  *
821  * @param dev
822  *   Pointer to rte_eth_dev structure.
823  *
824  * @return
825  *   True if device supports extensive flow metadata register, otherwise false.
826  */
827 bool
828 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
829 {
830 	struct mlx5_priv *priv = dev->data->dev_private;
831 	struct mlx5_dev_config *config = &priv->config;
832 
833 	/*
834 	 * Having available reg_c can be regarded inclusively as supporting
835 	 * extensive flow metadata register, which could mean,
836 	 * - metadata register copy action by modify header.
837 	 * - 16 modify header actions is supported.
838 	 * - reg_c's are preserved across different domain (FDB and NIC) on
839 	 *   packet loopback by flow lookup miss.
840 	 */
841 	return config->flow_mreg_c[2] != REG_NON;
842 }
843 
844 /**
845  * Get the lowest priority.
846  *
847  * @param[in] dev
848  *   Pointer to the Ethernet device structure.
849  * @param[in] attributes
850  *   Pointer to device flow rule attributes.
851  *
852  * @return
853  *   The value of lowest priority of flow.
854  */
855 uint32_t
856 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
857 			  const struct rte_flow_attr *attr)
858 {
859 	struct mlx5_priv *priv = dev->data->dev_private;
860 
861 	if (!attr->group && !attr->transfer)
862 		return priv->config.flow_prio - 2;
863 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
864 }
865 
866 /**
867  * Calculate matcher priority of the flow.
868  *
869  * @param[in] dev
870  *   Pointer to the Ethernet device structure.
871  * @param[in] attr
872  *   Pointer to device flow rule attributes.
873  * @param[in] subpriority
874  *   The priority based on the items.
875  * @return
876  *   The matcher priority of the flow.
877  */
878 uint16_t
879 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
880 			  const struct rte_flow_attr *attr,
881 			  uint32_t subpriority)
882 {
883 	uint16_t priority = (uint16_t)attr->priority;
884 	struct mlx5_priv *priv = dev->data->dev_private;
885 
886 	if (!attr->group && !attr->transfer) {
887 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
888 			priority = priv->config.flow_prio - 1;
889 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
890 	}
891 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
892 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
893 	return priority * 3 + subpriority;
894 }
895 
896 /**
897  * Verify the @p item specifications (spec, last, mask) are compatible with the
898  * NIC capabilities.
899  *
900  * @param[in] item
901  *   Item specification.
902  * @param[in] mask
903  *   @p item->mask or flow default bit-masks.
904  * @param[in] nic_mask
905  *   Bit-masks covering supported fields by the NIC to compare with user mask.
906  * @param[in] size
907  *   Bit-masks size in bytes.
908  * @param[in] range_accepted
909  *   True if range of values is accepted for specific fields, false otherwise.
910  * @param[out] error
911  *   Pointer to error structure.
912  *
913  * @return
914  *   0 on success, a negative errno value otherwise and rte_errno is set.
915  */
916 int
917 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
918 			  const uint8_t *mask,
919 			  const uint8_t *nic_mask,
920 			  unsigned int size,
921 			  bool range_accepted,
922 			  struct rte_flow_error *error)
923 {
924 	unsigned int i;
925 
926 	MLX5_ASSERT(nic_mask);
927 	for (i = 0; i < size; ++i)
928 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
929 			return rte_flow_error_set(error, ENOTSUP,
930 						  RTE_FLOW_ERROR_TYPE_ITEM,
931 						  item,
932 						  "mask enables non supported"
933 						  " bits");
934 	if (!item->spec && (item->mask || item->last))
935 		return rte_flow_error_set(error, EINVAL,
936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
937 					  "mask/last without a spec is not"
938 					  " supported");
939 	if (item->spec && item->last && !range_accepted) {
940 		uint8_t spec[size];
941 		uint8_t last[size];
942 		unsigned int i;
943 		int ret;
944 
945 		for (i = 0; i < size; ++i) {
946 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
947 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
948 		}
949 		ret = memcmp(spec, last, size);
950 		if (ret != 0)
951 			return rte_flow_error_set(error, EINVAL,
952 						  RTE_FLOW_ERROR_TYPE_ITEM,
953 						  item,
954 						  "range is not valid");
955 	}
956 	return 0;
957 }
958 
959 /**
960  * Adjust the hash fields according to the @p flow information.
961  *
962  * @param[in] dev_flow.
963  *   Pointer to the mlx5_flow.
964  * @param[in] tunnel
965  *   1 when the hash field is for a tunnel item.
966  * @param[in] layer_types
967  *   ETH_RSS_* types.
968  * @param[in] hash_fields
969  *   Item hash fields.
970  *
971  * @return
972  *   The hash fields that should be used.
973  */
974 uint64_t
975 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
976 			    int tunnel __rte_unused, uint64_t layer_types,
977 			    uint64_t hash_fields)
978 {
979 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
980 	int rss_request_inner = rss_desc->level >= 2;
981 
982 	/* Check RSS hash level for tunnel. */
983 	if (tunnel && rss_request_inner)
984 		hash_fields |= IBV_RX_HASH_INNER;
985 	else if (tunnel || rss_request_inner)
986 		return 0;
987 #endif
988 	/* Check if requested layer matches RSS hash fields. */
989 	if (!(rss_desc->types & layer_types))
990 		return 0;
991 	return hash_fields;
992 }
993 
994 /**
995  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
996  * if several tunnel rules are used on this queue, the tunnel ptype will be
997  * cleared.
998  *
999  * @param rxq_ctrl
1000  *   Rx queue to update.
1001  */
1002 static void
1003 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1004 {
1005 	unsigned int i;
1006 	uint32_t tunnel_ptype = 0;
1007 
1008 	/* Look up for the ptype to use. */
1009 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1010 		if (!rxq_ctrl->flow_tunnels_n[i])
1011 			continue;
1012 		if (!tunnel_ptype) {
1013 			tunnel_ptype = tunnels_info[i].ptype;
1014 		} else {
1015 			tunnel_ptype = 0;
1016 			break;
1017 		}
1018 	}
1019 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1020 }
1021 
1022 /**
1023  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1024  * flow.
1025  *
1026  * @param[in] dev
1027  *   Pointer to the Ethernet device structure.
1028  * @param[in] dev_handle
1029  *   Pointer to device flow handle structure.
1030  */
1031 static void
1032 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1033 		       struct mlx5_flow_handle *dev_handle)
1034 {
1035 	struct mlx5_priv *priv = dev->data->dev_private;
1036 	const int mark = dev_handle->mark;
1037 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1038 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1039 	unsigned int i;
1040 
1041 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1042 		struct mlx5_hrxq *hrxq;
1043 
1044 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1045 			      dev_handle->rix_hrxq);
1046 		if (hrxq)
1047 			ind_tbl = hrxq->ind_table;
1048 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1049 		struct mlx5_shared_action_rss *shared_rss;
1050 
1051 		shared_rss = mlx5_ipool_get
1052 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1053 			 dev_handle->rix_srss);
1054 		if (shared_rss)
1055 			ind_tbl = shared_rss->ind_tbl;
1056 	}
1057 	if (!ind_tbl)
1058 		return;
1059 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1060 		int idx = ind_tbl->queues[i];
1061 		struct mlx5_rxq_ctrl *rxq_ctrl =
1062 			container_of((*priv->rxqs)[idx],
1063 				     struct mlx5_rxq_ctrl, rxq);
1064 
1065 		/*
1066 		 * To support metadata register copy on Tx loopback,
1067 		 * this must be always enabled (metadata may arive
1068 		 * from other port - not from local flows only.
1069 		 */
1070 		if (priv->config.dv_flow_en &&
1071 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1072 		    mlx5_flow_ext_mreg_supported(dev)) {
1073 			rxq_ctrl->rxq.mark = 1;
1074 			rxq_ctrl->flow_mark_n = 1;
1075 		} else if (mark) {
1076 			rxq_ctrl->rxq.mark = 1;
1077 			rxq_ctrl->flow_mark_n++;
1078 		}
1079 		if (tunnel) {
1080 			unsigned int j;
1081 
1082 			/* Increase the counter matching the flow. */
1083 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1084 				if ((tunnels_info[j].tunnel &
1085 				     dev_handle->layers) ==
1086 				    tunnels_info[j].tunnel) {
1087 					rxq_ctrl->flow_tunnels_n[j]++;
1088 					break;
1089 				}
1090 			}
1091 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1092 		}
1093 	}
1094 }
1095 
1096 /**
1097  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1098  *
1099  * @param[in] dev
1100  *   Pointer to the Ethernet device structure.
1101  * @param[in] flow
1102  *   Pointer to flow structure.
1103  */
1104 static void
1105 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1106 {
1107 	struct mlx5_priv *priv = dev->data->dev_private;
1108 	uint32_t handle_idx;
1109 	struct mlx5_flow_handle *dev_handle;
1110 
1111 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1112 		       handle_idx, dev_handle, next)
1113 		flow_drv_rxq_flags_set(dev, dev_handle);
1114 }
1115 
1116 /**
1117  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1118  * device flow if no other flow uses it with the same kind of request.
1119  *
1120  * @param dev
1121  *   Pointer to Ethernet device.
1122  * @param[in] dev_handle
1123  *   Pointer to the device flow handle structure.
1124  */
1125 static void
1126 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1127 			struct mlx5_flow_handle *dev_handle)
1128 {
1129 	struct mlx5_priv *priv = dev->data->dev_private;
1130 	const int mark = dev_handle->mark;
1131 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1132 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1133 	unsigned int i;
1134 
1135 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1136 		struct mlx5_hrxq *hrxq;
1137 
1138 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1139 			      dev_handle->rix_hrxq);
1140 		if (hrxq)
1141 			ind_tbl = hrxq->ind_table;
1142 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1143 		struct mlx5_shared_action_rss *shared_rss;
1144 
1145 		shared_rss = mlx5_ipool_get
1146 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1147 			 dev_handle->rix_srss);
1148 		if (shared_rss)
1149 			ind_tbl = shared_rss->ind_tbl;
1150 	}
1151 	if (!ind_tbl)
1152 		return;
1153 	MLX5_ASSERT(dev->data->dev_started);
1154 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1155 		int idx = ind_tbl->queues[i];
1156 		struct mlx5_rxq_ctrl *rxq_ctrl =
1157 			container_of((*priv->rxqs)[idx],
1158 				     struct mlx5_rxq_ctrl, rxq);
1159 
1160 		if (priv->config.dv_flow_en &&
1161 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1162 		    mlx5_flow_ext_mreg_supported(dev)) {
1163 			rxq_ctrl->rxq.mark = 1;
1164 			rxq_ctrl->flow_mark_n = 1;
1165 		} else if (mark) {
1166 			rxq_ctrl->flow_mark_n--;
1167 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1168 		}
1169 		if (tunnel) {
1170 			unsigned int j;
1171 
1172 			/* Decrease the counter matching the flow. */
1173 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1174 				if ((tunnels_info[j].tunnel &
1175 				     dev_handle->layers) ==
1176 				    tunnels_info[j].tunnel) {
1177 					rxq_ctrl->flow_tunnels_n[j]--;
1178 					break;
1179 				}
1180 			}
1181 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1182 		}
1183 	}
1184 }
1185 
1186 /**
1187  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1188  * @p flow if no other flow uses it with the same kind of request.
1189  *
1190  * @param dev
1191  *   Pointer to Ethernet device.
1192  * @param[in] flow
1193  *   Pointer to the flow.
1194  */
1195 static void
1196 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1197 {
1198 	struct mlx5_priv *priv = dev->data->dev_private;
1199 	uint32_t handle_idx;
1200 	struct mlx5_flow_handle *dev_handle;
1201 
1202 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1203 		       handle_idx, dev_handle, next)
1204 		flow_drv_rxq_flags_trim(dev, dev_handle);
1205 }
1206 
1207 /**
1208  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1209  *
1210  * @param dev
1211  *   Pointer to Ethernet device.
1212  */
1213 static void
1214 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1215 {
1216 	struct mlx5_priv *priv = dev->data->dev_private;
1217 	unsigned int i;
1218 
1219 	for (i = 0; i != priv->rxqs_n; ++i) {
1220 		struct mlx5_rxq_ctrl *rxq_ctrl;
1221 		unsigned int j;
1222 
1223 		if (!(*priv->rxqs)[i])
1224 			continue;
1225 		rxq_ctrl = container_of((*priv->rxqs)[i],
1226 					struct mlx5_rxq_ctrl, rxq);
1227 		rxq_ctrl->flow_mark_n = 0;
1228 		rxq_ctrl->rxq.mark = 0;
1229 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1230 			rxq_ctrl->flow_tunnels_n[j] = 0;
1231 		rxq_ctrl->rxq.tunnel = 0;
1232 	}
1233 }
1234 
1235 /**
1236  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1237  *
1238  * @param[in] dev
1239  *   Pointer to the Ethernet device structure.
1240  */
1241 void
1242 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1243 {
1244 	struct mlx5_priv *priv = dev->data->dev_private;
1245 	struct mlx5_rxq_data *data;
1246 	unsigned int i;
1247 
1248 	for (i = 0; i != priv->rxqs_n; ++i) {
1249 		if (!(*priv->rxqs)[i])
1250 			continue;
1251 		data = (*priv->rxqs)[i];
1252 		if (!rte_flow_dynf_metadata_avail()) {
1253 			data->dynf_meta = 0;
1254 			data->flow_meta_mask = 0;
1255 			data->flow_meta_offset = -1;
1256 			data->flow_meta_port_mask = 0;
1257 		} else {
1258 			data->dynf_meta = 1;
1259 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1260 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1261 			data->flow_meta_port_mask = (uint32_t)~0;
1262 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1263 				data->flow_meta_port_mask >>= 16;
1264 		}
1265 	}
1266 }
1267 
1268 /*
1269  * return a pointer to the desired action in the list of actions.
1270  *
1271  * @param[in] actions
1272  *   The list of actions to search the action in.
1273  * @param[in] action
1274  *   The action to find.
1275  *
1276  * @return
1277  *   Pointer to the action in the list, if found. NULL otherwise.
1278  */
1279 const struct rte_flow_action *
1280 mlx5_flow_find_action(const struct rte_flow_action *actions,
1281 		      enum rte_flow_action_type action)
1282 {
1283 	if (actions == NULL)
1284 		return NULL;
1285 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1286 		if (actions->type == action)
1287 			return actions;
1288 	return NULL;
1289 }
1290 
1291 /*
1292  * Validate the flag action.
1293  *
1294  * @param[in] action_flags
1295  *   Bit-fields that holds the actions detected until now.
1296  * @param[in] attr
1297  *   Attributes of flow that includes this action.
1298  * @param[out] error
1299  *   Pointer to error structure.
1300  *
1301  * @return
1302  *   0 on success, a negative errno value otherwise and rte_errno is set.
1303  */
1304 int
1305 mlx5_flow_validate_action_flag(uint64_t action_flags,
1306 			       const struct rte_flow_attr *attr,
1307 			       struct rte_flow_error *error)
1308 {
1309 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1310 		return rte_flow_error_set(error, EINVAL,
1311 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1312 					  "can't mark and flag in same flow");
1313 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1314 		return rte_flow_error_set(error, EINVAL,
1315 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1316 					  "can't have 2 flag"
1317 					  " actions in same flow");
1318 	if (attr->egress)
1319 		return rte_flow_error_set(error, ENOTSUP,
1320 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1321 					  "flag action not supported for "
1322 					  "egress");
1323 	return 0;
1324 }
1325 
1326 /*
1327  * Validate the mark action.
1328  *
1329  * @param[in] action
1330  *   Pointer to the queue action.
1331  * @param[in] action_flags
1332  *   Bit-fields that holds the actions detected until now.
1333  * @param[in] attr
1334  *   Attributes of flow that includes this action.
1335  * @param[out] error
1336  *   Pointer to error structure.
1337  *
1338  * @return
1339  *   0 on success, a negative errno value otherwise and rte_errno is set.
1340  */
1341 int
1342 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1343 			       uint64_t action_flags,
1344 			       const struct rte_flow_attr *attr,
1345 			       struct rte_flow_error *error)
1346 {
1347 	const struct rte_flow_action_mark *mark = action->conf;
1348 
1349 	if (!mark)
1350 		return rte_flow_error_set(error, EINVAL,
1351 					  RTE_FLOW_ERROR_TYPE_ACTION,
1352 					  action,
1353 					  "configuration cannot be null");
1354 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1357 					  &mark->id,
1358 					  "mark id must in 0 <= id < "
1359 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1360 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1361 		return rte_flow_error_set(error, EINVAL,
1362 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1363 					  "can't flag and mark in same flow");
1364 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1365 		return rte_flow_error_set(error, EINVAL,
1366 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1367 					  "can't have 2 mark actions in same"
1368 					  " flow");
1369 	if (attr->egress)
1370 		return rte_flow_error_set(error, ENOTSUP,
1371 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1372 					  "mark action not supported for "
1373 					  "egress");
1374 	return 0;
1375 }
1376 
1377 /*
1378  * Validate the drop action.
1379  *
1380  * @param[in] action_flags
1381  *   Bit-fields that holds the actions detected until now.
1382  * @param[in] attr
1383  *   Attributes of flow that includes this action.
1384  * @param[out] error
1385  *   Pointer to error structure.
1386  *
1387  * @return
1388  *   0 on success, a negative errno value otherwise and rte_errno is set.
1389  */
1390 int
1391 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1392 			       const struct rte_flow_attr *attr,
1393 			       struct rte_flow_error *error)
1394 {
1395 	if (attr->egress)
1396 		return rte_flow_error_set(error, ENOTSUP,
1397 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1398 					  "drop action not supported for "
1399 					  "egress");
1400 	return 0;
1401 }
1402 
1403 /*
1404  * Validate the queue action.
1405  *
1406  * @param[in] action
1407  *   Pointer to the queue action.
1408  * @param[in] action_flags
1409  *   Bit-fields that holds the actions detected until now.
1410  * @param[in] dev
1411  *   Pointer to the Ethernet device structure.
1412  * @param[in] attr
1413  *   Attributes of flow that includes this action.
1414  * @param[out] error
1415  *   Pointer to error structure.
1416  *
1417  * @return
1418  *   0 on success, a negative errno value otherwise and rte_errno is set.
1419  */
1420 int
1421 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1422 				uint64_t action_flags,
1423 				struct rte_eth_dev *dev,
1424 				const struct rte_flow_attr *attr,
1425 				struct rte_flow_error *error)
1426 {
1427 	struct mlx5_priv *priv = dev->data->dev_private;
1428 	const struct rte_flow_action_queue *queue = action->conf;
1429 
1430 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1431 		return rte_flow_error_set(error, EINVAL,
1432 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1433 					  "can't have 2 fate actions in"
1434 					  " same flow");
1435 	if (!priv->rxqs_n)
1436 		return rte_flow_error_set(error, EINVAL,
1437 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1438 					  NULL, "No Rx queues configured");
1439 	if (queue->index >= priv->rxqs_n)
1440 		return rte_flow_error_set(error, EINVAL,
1441 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1442 					  &queue->index,
1443 					  "queue index out of range");
1444 	if (!(*priv->rxqs)[queue->index])
1445 		return rte_flow_error_set(error, EINVAL,
1446 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1447 					  &queue->index,
1448 					  "queue is not configured");
1449 	if (attr->egress)
1450 		return rte_flow_error_set(error, ENOTSUP,
1451 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1452 					  "queue action not supported for "
1453 					  "egress");
1454 	return 0;
1455 }
1456 
1457 /*
1458  * Validate the rss action.
1459  *
1460  * @param[in] dev
1461  *   Pointer to the Ethernet device structure.
1462  * @param[in] action
1463  *   Pointer to the queue action.
1464  * @param[out] error
1465  *   Pointer to error structure.
1466  *
1467  * @return
1468  *   0 on success, a negative errno value otherwise and rte_errno is set.
1469  */
1470 int
1471 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1472 			 const struct rte_flow_action *action,
1473 			 struct rte_flow_error *error)
1474 {
1475 	struct mlx5_priv *priv = dev->data->dev_private;
1476 	const struct rte_flow_action_rss *rss = action->conf;
1477 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1478 	unsigned int i;
1479 
1480 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1481 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1482 		return rte_flow_error_set(error, ENOTSUP,
1483 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1484 					  &rss->func,
1485 					  "RSS hash function not supported");
1486 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1487 	if (rss->level > 2)
1488 #else
1489 	if (rss->level > 1)
1490 #endif
1491 		return rte_flow_error_set(error, ENOTSUP,
1492 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1493 					  &rss->level,
1494 					  "tunnel RSS is not supported");
1495 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1496 	if (rss->key_len == 0 && rss->key != NULL)
1497 		return rte_flow_error_set(error, ENOTSUP,
1498 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499 					  &rss->key_len,
1500 					  "RSS hash key length 0");
1501 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1502 		return rte_flow_error_set(error, ENOTSUP,
1503 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1504 					  &rss->key_len,
1505 					  "RSS hash key too small");
1506 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1507 		return rte_flow_error_set(error, ENOTSUP,
1508 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1509 					  &rss->key_len,
1510 					  "RSS hash key too large");
1511 	if (rss->queue_num > priv->config.ind_table_max_size)
1512 		return rte_flow_error_set(error, ENOTSUP,
1513 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1514 					  &rss->queue_num,
1515 					  "number of queues too large");
1516 	if (rss->types & MLX5_RSS_HF_MASK)
1517 		return rte_flow_error_set(error, ENOTSUP,
1518 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1519 					  &rss->types,
1520 					  "some RSS protocols are not"
1521 					  " supported");
1522 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1523 	    !(rss->types & ETH_RSS_IP))
1524 		return rte_flow_error_set(error, EINVAL,
1525 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1526 					  "L3 partial RSS requested but L3 RSS"
1527 					  " type not specified");
1528 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1529 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1530 		return rte_flow_error_set(error, EINVAL,
1531 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1532 					  "L4 partial RSS requested but L4 RSS"
1533 					  " type not specified");
1534 	if (!priv->rxqs_n)
1535 		return rte_flow_error_set(error, EINVAL,
1536 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1537 					  NULL, "No Rx queues configured");
1538 	if (!rss->queue_num)
1539 		return rte_flow_error_set(error, EINVAL,
1540 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1541 					  NULL, "No queues configured");
1542 	for (i = 0; i != rss->queue_num; ++i) {
1543 		struct mlx5_rxq_ctrl *rxq_ctrl;
1544 
1545 		if (rss->queue[i] >= priv->rxqs_n)
1546 			return rte_flow_error_set
1547 				(error, EINVAL,
1548 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1549 				 &rss->queue[i], "queue index out of range");
1550 		if (!(*priv->rxqs)[rss->queue[i]])
1551 			return rte_flow_error_set
1552 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1553 				 &rss->queue[i], "queue is not configured");
1554 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1555 					struct mlx5_rxq_ctrl, rxq);
1556 		if (i == 0)
1557 			rxq_type = rxq_ctrl->type;
1558 		if (rxq_type != rxq_ctrl->type)
1559 			return rte_flow_error_set
1560 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1561 				 &rss->queue[i],
1562 				 "combining hairpin and regular RSS queues is not supported");
1563 	}
1564 	return 0;
1565 }
1566 
1567 /*
1568  * Validate the rss action.
1569  *
1570  * @param[in] action
1571  *   Pointer to the queue action.
1572  * @param[in] action_flags
1573  *   Bit-fields that holds the actions detected until now.
1574  * @param[in] dev
1575  *   Pointer to the Ethernet device structure.
1576  * @param[in] attr
1577  *   Attributes of flow that includes this action.
1578  * @param[in] item_flags
1579  *   Items that were detected.
1580  * @param[out] error
1581  *   Pointer to error structure.
1582  *
1583  * @return
1584  *   0 on success, a negative errno value otherwise and rte_errno is set.
1585  */
1586 int
1587 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1588 			      uint64_t action_flags,
1589 			      struct rte_eth_dev *dev,
1590 			      const struct rte_flow_attr *attr,
1591 			      uint64_t item_flags,
1592 			      struct rte_flow_error *error)
1593 {
1594 	const struct rte_flow_action_rss *rss = action->conf;
1595 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1596 	int ret;
1597 
1598 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1599 		return rte_flow_error_set(error, EINVAL,
1600 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1601 					  "can't have 2 fate actions"
1602 					  " in same flow");
1603 	ret = mlx5_validate_action_rss(dev, action, error);
1604 	if (ret)
1605 		return ret;
1606 	if (attr->egress)
1607 		return rte_flow_error_set(error, ENOTSUP,
1608 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1609 					  "rss action not supported for "
1610 					  "egress");
1611 	if (rss->level > 1 && !tunnel)
1612 		return rte_flow_error_set(error, EINVAL,
1613 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1614 					  "inner RSS is not supported for "
1615 					  "non-tunnel flows");
1616 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1617 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1618 		return rte_flow_error_set(error, EINVAL,
1619 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1620 					  "RSS on eCPRI is not supported now");
1621 	}
1622 	return 0;
1623 }
1624 
1625 /*
1626  * Validate the default miss action.
1627  *
1628  * @param[in] action_flags
1629  *   Bit-fields that holds the actions detected until now.
1630  * @param[out] error
1631  *   Pointer to error structure.
1632  *
1633  * @return
1634  *   0 on success, a negative errno value otherwise and rte_errno is set.
1635  */
1636 int
1637 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1638 				const struct rte_flow_attr *attr,
1639 				struct rte_flow_error *error)
1640 {
1641 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1642 		return rte_flow_error_set(error, EINVAL,
1643 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1644 					  "can't have 2 fate actions in"
1645 					  " same flow");
1646 	if (attr->egress)
1647 		return rte_flow_error_set(error, ENOTSUP,
1648 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1649 					  "default miss action not supported "
1650 					  "for egress");
1651 	if (attr->group)
1652 		return rte_flow_error_set(error, ENOTSUP,
1653 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1654 					  "only group 0 is supported");
1655 	if (attr->transfer)
1656 		return rte_flow_error_set(error, ENOTSUP,
1657 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1658 					  NULL, "transfer is not supported");
1659 	return 0;
1660 }
1661 
1662 /*
1663  * Validate the count action.
1664  *
1665  * @param[in] dev
1666  *   Pointer to the Ethernet device structure.
1667  * @param[in] attr
1668  *   Attributes of flow that includes this action.
1669  * @param[out] error
1670  *   Pointer to error structure.
1671  *
1672  * @return
1673  *   0 on success, a negative errno value otherwise and rte_errno is set.
1674  */
1675 int
1676 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1677 				const struct rte_flow_attr *attr,
1678 				struct rte_flow_error *error)
1679 {
1680 	if (attr->egress)
1681 		return rte_flow_error_set(error, ENOTSUP,
1682 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1683 					  "count action not supported for "
1684 					  "egress");
1685 	return 0;
1686 }
1687 
1688 /**
1689  * Verify the @p attributes will be correctly understood by the NIC and store
1690  * them in the @p flow if everything is correct.
1691  *
1692  * @param[in] dev
1693  *   Pointer to the Ethernet device structure.
1694  * @param[in] attributes
1695  *   Pointer to flow attributes
1696  * @param[out] error
1697  *   Pointer to error structure.
1698  *
1699  * @return
1700  *   0 on success, a negative errno value otherwise and rte_errno is set.
1701  */
1702 int
1703 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1704 			      const struct rte_flow_attr *attributes,
1705 			      struct rte_flow_error *error)
1706 {
1707 	struct mlx5_priv *priv = dev->data->dev_private;
1708 	uint32_t priority_max = priv->config.flow_prio - 1;
1709 
1710 	if (attributes->group)
1711 		return rte_flow_error_set(error, ENOTSUP,
1712 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1713 					  NULL, "groups is not supported");
1714 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1715 	    attributes->priority >= priority_max)
1716 		return rte_flow_error_set(error, ENOTSUP,
1717 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1718 					  NULL, "priority out of range");
1719 	if (attributes->egress)
1720 		return rte_flow_error_set(error, ENOTSUP,
1721 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1722 					  "egress is not supported");
1723 	if (attributes->transfer && !priv->config.dv_esw_en)
1724 		return rte_flow_error_set(error, ENOTSUP,
1725 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1726 					  NULL, "transfer is not supported");
1727 	if (!attributes->ingress)
1728 		return rte_flow_error_set(error, EINVAL,
1729 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1730 					  NULL,
1731 					  "ingress attribute is mandatory");
1732 	return 0;
1733 }
1734 
1735 /**
1736  * Validate ICMP6 item.
1737  *
1738  * @param[in] item
1739  *   Item specification.
1740  * @param[in] item_flags
1741  *   Bit-fields that holds the items detected until now.
1742  * @param[in] ext_vlan_sup
1743  *   Whether extended VLAN features are supported or not.
1744  * @param[out] error
1745  *   Pointer to error structure.
1746  *
1747  * @return
1748  *   0 on success, a negative errno value otherwise and rte_errno is set.
1749  */
1750 int
1751 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1752 			       uint64_t item_flags,
1753 			       uint8_t target_protocol,
1754 			       struct rte_flow_error *error)
1755 {
1756 	const struct rte_flow_item_icmp6 *mask = item->mask;
1757 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1758 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1759 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1760 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1761 				      MLX5_FLOW_LAYER_OUTER_L4;
1762 	int ret;
1763 
1764 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1765 		return rte_flow_error_set(error, EINVAL,
1766 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1767 					  "protocol filtering not compatible"
1768 					  " with ICMP6 layer");
1769 	if (!(item_flags & l3m))
1770 		return rte_flow_error_set(error, EINVAL,
1771 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1772 					  "IPv6 is mandatory to filter on"
1773 					  " ICMP6");
1774 	if (item_flags & l4m)
1775 		return rte_flow_error_set(error, EINVAL,
1776 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1777 					  "multiple L4 layers not supported");
1778 	if (!mask)
1779 		mask = &rte_flow_item_icmp6_mask;
1780 	ret = mlx5_flow_item_acceptable
1781 		(item, (const uint8_t *)mask,
1782 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1783 		 sizeof(struct rte_flow_item_icmp6),
1784 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1785 	if (ret < 0)
1786 		return ret;
1787 	return 0;
1788 }
1789 
1790 /**
1791  * Validate ICMP item.
1792  *
1793  * @param[in] item
1794  *   Item specification.
1795  * @param[in] item_flags
1796  *   Bit-fields that holds the items detected until now.
1797  * @param[out] error
1798  *   Pointer to error structure.
1799  *
1800  * @return
1801  *   0 on success, a negative errno value otherwise and rte_errno is set.
1802  */
1803 int
1804 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1805 			     uint64_t item_flags,
1806 			     uint8_t target_protocol,
1807 			     struct rte_flow_error *error)
1808 {
1809 	const struct rte_flow_item_icmp *mask = item->mask;
1810 	const struct rte_flow_item_icmp nic_mask = {
1811 		.hdr.icmp_type = 0xff,
1812 		.hdr.icmp_code = 0xff,
1813 		.hdr.icmp_ident = RTE_BE16(0xffff),
1814 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1815 	};
1816 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1817 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1818 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1819 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1820 				      MLX5_FLOW_LAYER_OUTER_L4;
1821 	int ret;
1822 
1823 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1824 		return rte_flow_error_set(error, EINVAL,
1825 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1826 					  "protocol filtering not compatible"
1827 					  " with ICMP layer");
1828 	if (!(item_flags & l3m))
1829 		return rte_flow_error_set(error, EINVAL,
1830 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1831 					  "IPv4 is mandatory to filter"
1832 					  " on ICMP");
1833 	if (item_flags & l4m)
1834 		return rte_flow_error_set(error, EINVAL,
1835 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1836 					  "multiple L4 layers not supported");
1837 	if (!mask)
1838 		mask = &nic_mask;
1839 	ret = mlx5_flow_item_acceptable
1840 		(item, (const uint8_t *)mask,
1841 		 (const uint8_t *)&nic_mask,
1842 		 sizeof(struct rte_flow_item_icmp),
1843 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1844 	if (ret < 0)
1845 		return ret;
1846 	return 0;
1847 }
1848 
1849 /**
1850  * Validate Ethernet item.
1851  *
1852  * @param[in] item
1853  *   Item specification.
1854  * @param[in] item_flags
1855  *   Bit-fields that holds the items detected until now.
1856  * @param[out] error
1857  *   Pointer to error structure.
1858  *
1859  * @return
1860  *   0 on success, a negative errno value otherwise and rte_errno is set.
1861  */
1862 int
1863 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1864 			    uint64_t item_flags, bool ext_vlan_sup,
1865 			    struct rte_flow_error *error)
1866 {
1867 	const struct rte_flow_item_eth *mask = item->mask;
1868 	const struct rte_flow_item_eth nic_mask = {
1869 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1870 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1871 		.type = RTE_BE16(0xffff),
1872 		.has_vlan = ext_vlan_sup ? 1 : 0,
1873 	};
1874 	int ret;
1875 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1876 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1877 				       MLX5_FLOW_LAYER_OUTER_L2;
1878 
1879 	if (item_flags & ethm)
1880 		return rte_flow_error_set(error, ENOTSUP,
1881 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1882 					  "multiple L2 layers not supported");
1883 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1884 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1885 		return rte_flow_error_set(error, EINVAL,
1886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1887 					  "L2 layer should not follow "
1888 					  "L3 layers");
1889 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1890 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1891 		return rte_flow_error_set(error, EINVAL,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "L2 layer should not follow VLAN");
1894 	if (!mask)
1895 		mask = &rte_flow_item_eth_mask;
1896 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1897 					(const uint8_t *)&nic_mask,
1898 					sizeof(struct rte_flow_item_eth),
1899 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1900 	return ret;
1901 }
1902 
1903 /**
1904  * Validate VLAN item.
1905  *
1906  * @param[in] item
1907  *   Item specification.
1908  * @param[in] item_flags
1909  *   Bit-fields that holds the items detected until now.
1910  * @param[in] dev
1911  *   Ethernet device flow is being created on.
1912  * @param[out] error
1913  *   Pointer to error structure.
1914  *
1915  * @return
1916  *   0 on success, a negative errno value otherwise and rte_errno is set.
1917  */
1918 int
1919 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1920 			     uint64_t item_flags,
1921 			     struct rte_eth_dev *dev,
1922 			     struct rte_flow_error *error)
1923 {
1924 	const struct rte_flow_item_vlan *spec = item->spec;
1925 	const struct rte_flow_item_vlan *mask = item->mask;
1926 	const struct rte_flow_item_vlan nic_mask = {
1927 		.tci = RTE_BE16(UINT16_MAX),
1928 		.inner_type = RTE_BE16(UINT16_MAX),
1929 	};
1930 	uint16_t vlan_tag = 0;
1931 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1932 	int ret;
1933 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1934 					MLX5_FLOW_LAYER_INNER_L4) :
1935 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1936 					MLX5_FLOW_LAYER_OUTER_L4);
1937 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1938 					MLX5_FLOW_LAYER_OUTER_VLAN;
1939 
1940 	if (item_flags & vlanm)
1941 		return rte_flow_error_set(error, EINVAL,
1942 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1943 					  "multiple VLAN layers not supported");
1944 	else if ((item_flags & l34m) != 0)
1945 		return rte_flow_error_set(error, EINVAL,
1946 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1947 					  "VLAN cannot follow L3/L4 layer");
1948 	if (!mask)
1949 		mask = &rte_flow_item_vlan_mask;
1950 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1951 					(const uint8_t *)&nic_mask,
1952 					sizeof(struct rte_flow_item_vlan),
1953 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1954 	if (ret)
1955 		return ret;
1956 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1957 		struct mlx5_priv *priv = dev->data->dev_private;
1958 
1959 		if (priv->vmwa_context) {
1960 			/*
1961 			 * Non-NULL context means we have a virtual machine
1962 			 * and SR-IOV enabled, we have to create VLAN interface
1963 			 * to make hypervisor to setup E-Switch vport
1964 			 * context correctly. We avoid creating the multiple
1965 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1966 			 */
1967 			return rte_flow_error_set(error, EINVAL,
1968 						  RTE_FLOW_ERROR_TYPE_ITEM,
1969 						  item,
1970 						  "VLAN tag mask is not"
1971 						  " supported in virtual"
1972 						  " environment");
1973 		}
1974 	}
1975 	if (spec) {
1976 		vlan_tag = spec->tci;
1977 		vlan_tag &= mask->tci;
1978 	}
1979 	/*
1980 	 * From verbs perspective an empty VLAN is equivalent
1981 	 * to a packet without VLAN layer.
1982 	 */
1983 	if (!vlan_tag)
1984 		return rte_flow_error_set(error, EINVAL,
1985 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1986 					  item->spec,
1987 					  "VLAN cannot be empty");
1988 	return 0;
1989 }
1990 
1991 /**
1992  * Validate IPV4 item.
1993  *
1994  * @param[in] item
1995  *   Item specification.
1996  * @param[in] item_flags
1997  *   Bit-fields that holds the items detected until now.
1998  * @param[in] last_item
1999  *   Previous validated item in the pattern items.
2000  * @param[in] ether_type
2001  *   Type in the ethernet layer header (including dot1q).
2002  * @param[in] acc_mask
2003  *   Acceptable mask, if NULL default internal default mask
2004  *   will be used to check whether item fields are supported.
2005  * @param[in] range_accepted
2006  *   True if range of values is accepted for specific fields, false otherwise.
2007  * @param[out] error
2008  *   Pointer to error structure.
2009  *
2010  * @return
2011  *   0 on success, a negative errno value otherwise and rte_errno is set.
2012  */
2013 int
2014 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2015 			     uint64_t item_flags,
2016 			     uint64_t last_item,
2017 			     uint16_t ether_type,
2018 			     const struct rte_flow_item_ipv4 *acc_mask,
2019 			     bool range_accepted,
2020 			     struct rte_flow_error *error)
2021 {
2022 	const struct rte_flow_item_ipv4 *mask = item->mask;
2023 	const struct rte_flow_item_ipv4 *spec = item->spec;
2024 	const struct rte_flow_item_ipv4 nic_mask = {
2025 		.hdr = {
2026 			.src_addr = RTE_BE32(0xffffffff),
2027 			.dst_addr = RTE_BE32(0xffffffff),
2028 			.type_of_service = 0xff,
2029 			.next_proto_id = 0xff,
2030 		},
2031 	};
2032 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2034 				      MLX5_FLOW_LAYER_OUTER_L3;
2035 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2036 				      MLX5_FLOW_LAYER_OUTER_L4;
2037 	int ret;
2038 	uint8_t next_proto = 0xFF;
2039 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2040 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2041 				  MLX5_FLOW_LAYER_INNER_VLAN);
2042 
2043 	if ((last_item & l2_vlan) && ether_type &&
2044 	    ether_type != RTE_ETHER_TYPE_IPV4)
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "IPv4 cannot follow L2/VLAN layer "
2048 					  "which ether type is not IPv4");
2049 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2050 		if (mask && spec)
2051 			next_proto = mask->hdr.next_proto_id &
2052 				     spec->hdr.next_proto_id;
2053 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2054 			return rte_flow_error_set(error, EINVAL,
2055 						  RTE_FLOW_ERROR_TYPE_ITEM,
2056 						  item,
2057 						  "multiple tunnel "
2058 						  "not supported");
2059 	}
2060 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2061 		return rte_flow_error_set(error, EINVAL,
2062 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2063 					  "wrong tunnel type - IPv6 specified "
2064 					  "but IPv4 item provided");
2065 	if (item_flags & l3m)
2066 		return rte_flow_error_set(error, ENOTSUP,
2067 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2068 					  "multiple L3 layers not supported");
2069 	else if (item_flags & l4m)
2070 		return rte_flow_error_set(error, EINVAL,
2071 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2072 					  "L3 cannot follow an L4 layer.");
2073 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2074 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2075 		return rte_flow_error_set(error, EINVAL,
2076 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2077 					  "L3 cannot follow an NVGRE layer.");
2078 	if (!mask)
2079 		mask = &rte_flow_item_ipv4_mask;
2080 	else if (mask->hdr.next_proto_id != 0 &&
2081 		 mask->hdr.next_proto_id != 0xff)
2082 		return rte_flow_error_set(error, EINVAL,
2083 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2084 					  "partial mask is not supported"
2085 					  " for protocol");
2086 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2087 					acc_mask ? (const uint8_t *)acc_mask
2088 						 : (const uint8_t *)&nic_mask,
2089 					sizeof(struct rte_flow_item_ipv4),
2090 					range_accepted, error);
2091 	if (ret < 0)
2092 		return ret;
2093 	return 0;
2094 }
2095 
2096 /**
2097  * Validate IPV6 item.
2098  *
2099  * @param[in] item
2100  *   Item specification.
2101  * @param[in] item_flags
2102  *   Bit-fields that holds the items detected until now.
2103  * @param[in] last_item
2104  *   Previous validated item in the pattern items.
2105  * @param[in] ether_type
2106  *   Type in the ethernet layer header (including dot1q).
2107  * @param[in] acc_mask
2108  *   Acceptable mask, if NULL default internal default mask
2109  *   will be used to check whether item fields are supported.
2110  * @param[out] error
2111  *   Pointer to error structure.
2112  *
2113  * @return
2114  *   0 on success, a negative errno value otherwise and rte_errno is set.
2115  */
2116 int
2117 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2118 			     uint64_t item_flags,
2119 			     uint64_t last_item,
2120 			     uint16_t ether_type,
2121 			     const struct rte_flow_item_ipv6 *acc_mask,
2122 			     struct rte_flow_error *error)
2123 {
2124 	const struct rte_flow_item_ipv6 *mask = item->mask;
2125 	const struct rte_flow_item_ipv6 *spec = item->spec;
2126 	const struct rte_flow_item_ipv6 nic_mask = {
2127 		.hdr = {
2128 			.src_addr =
2129 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2130 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2131 			.dst_addr =
2132 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2133 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2134 			.vtc_flow = RTE_BE32(0xffffffff),
2135 			.proto = 0xff,
2136 		},
2137 	};
2138 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2139 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2140 				      MLX5_FLOW_LAYER_OUTER_L3;
2141 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2142 				      MLX5_FLOW_LAYER_OUTER_L4;
2143 	int ret;
2144 	uint8_t next_proto = 0xFF;
2145 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2146 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2147 				  MLX5_FLOW_LAYER_INNER_VLAN);
2148 
2149 	if ((last_item & l2_vlan) && ether_type &&
2150 	    ether_type != RTE_ETHER_TYPE_IPV6)
2151 		return rte_flow_error_set(error, EINVAL,
2152 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2153 					  "IPv6 cannot follow L2/VLAN layer "
2154 					  "which ether type is not IPv6");
2155 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2156 		next_proto = spec->hdr.proto;
2157 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2158 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2159 			return rte_flow_error_set(error, EINVAL,
2160 						  RTE_FLOW_ERROR_TYPE_ITEM,
2161 						  item,
2162 						  "multiple tunnel "
2163 						  "not supported");
2164 	}
2165 	if (next_proto == IPPROTO_HOPOPTS  ||
2166 	    next_proto == IPPROTO_ROUTING  ||
2167 	    next_proto == IPPROTO_FRAGMENT ||
2168 	    next_proto == IPPROTO_ESP	   ||
2169 	    next_proto == IPPROTO_AH	   ||
2170 	    next_proto == IPPROTO_DSTOPTS)
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "IPv6 proto (next header) should "
2174 					  "not be set as extension header");
2175 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2176 		return rte_flow_error_set(error, EINVAL,
2177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2178 					  "wrong tunnel type - IPv4 specified "
2179 					  "but IPv6 item provided");
2180 	if (item_flags & l3m)
2181 		return rte_flow_error_set(error, ENOTSUP,
2182 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2183 					  "multiple L3 layers not supported");
2184 	else if (item_flags & l4m)
2185 		return rte_flow_error_set(error, EINVAL,
2186 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2187 					  "L3 cannot follow an L4 layer.");
2188 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2189 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2190 		return rte_flow_error_set(error, EINVAL,
2191 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2192 					  "L3 cannot follow an NVGRE layer.");
2193 	if (!mask)
2194 		mask = &rte_flow_item_ipv6_mask;
2195 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2196 					acc_mask ? (const uint8_t *)acc_mask
2197 						 : (const uint8_t *)&nic_mask,
2198 					sizeof(struct rte_flow_item_ipv6),
2199 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2200 	if (ret < 0)
2201 		return ret;
2202 	return 0;
2203 }
2204 
2205 /**
2206  * Validate UDP item.
2207  *
2208  * @param[in] item
2209  *   Item specification.
2210  * @param[in] item_flags
2211  *   Bit-fields that holds the items detected until now.
2212  * @param[in] target_protocol
2213  *   The next protocol in the previous item.
2214  * @param[in] flow_mask
2215  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2216  * @param[out] error
2217  *   Pointer to error structure.
2218  *
2219  * @return
2220  *   0 on success, a negative errno value otherwise and rte_errno is set.
2221  */
2222 int
2223 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2224 			    uint64_t item_flags,
2225 			    uint8_t target_protocol,
2226 			    struct rte_flow_error *error)
2227 {
2228 	const struct rte_flow_item_udp *mask = item->mask;
2229 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2230 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2231 				      MLX5_FLOW_LAYER_OUTER_L3;
2232 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2233 				      MLX5_FLOW_LAYER_OUTER_L4;
2234 	int ret;
2235 
2236 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2237 		return rte_flow_error_set(error, EINVAL,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "protocol filtering not compatible"
2240 					  " with UDP layer");
2241 	if (!(item_flags & l3m))
2242 		return rte_flow_error_set(error, EINVAL,
2243 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2244 					  "L3 is mandatory to filter on L4");
2245 	if (item_flags & l4m)
2246 		return rte_flow_error_set(error, EINVAL,
2247 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2248 					  "multiple L4 layers not supported");
2249 	if (!mask)
2250 		mask = &rte_flow_item_udp_mask;
2251 	ret = mlx5_flow_item_acceptable
2252 		(item, (const uint8_t *)mask,
2253 		 (const uint8_t *)&rte_flow_item_udp_mask,
2254 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2255 		 error);
2256 	if (ret < 0)
2257 		return ret;
2258 	return 0;
2259 }
2260 
2261 /**
2262  * Validate TCP item.
2263  *
2264  * @param[in] item
2265  *   Item specification.
2266  * @param[in] item_flags
2267  *   Bit-fields that holds the items detected until now.
2268  * @param[in] target_protocol
2269  *   The next protocol in the previous item.
2270  * @param[out] error
2271  *   Pointer to error structure.
2272  *
2273  * @return
2274  *   0 on success, a negative errno value otherwise and rte_errno is set.
2275  */
2276 int
2277 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2278 			    uint64_t item_flags,
2279 			    uint8_t target_protocol,
2280 			    const struct rte_flow_item_tcp *flow_mask,
2281 			    struct rte_flow_error *error)
2282 {
2283 	const struct rte_flow_item_tcp *mask = item->mask;
2284 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2285 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2286 				      MLX5_FLOW_LAYER_OUTER_L3;
2287 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2288 				      MLX5_FLOW_LAYER_OUTER_L4;
2289 	int ret;
2290 
2291 	MLX5_ASSERT(flow_mask);
2292 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2293 		return rte_flow_error_set(error, EINVAL,
2294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 					  "protocol filtering not compatible"
2296 					  " with TCP layer");
2297 	if (!(item_flags & l3m))
2298 		return rte_flow_error_set(error, EINVAL,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "L3 is mandatory to filter on L4");
2301 	if (item_flags & l4m)
2302 		return rte_flow_error_set(error, EINVAL,
2303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2304 					  "multiple L4 layers not supported");
2305 	if (!mask)
2306 		mask = &rte_flow_item_tcp_mask;
2307 	ret = mlx5_flow_item_acceptable
2308 		(item, (const uint8_t *)mask,
2309 		 (const uint8_t *)flow_mask,
2310 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2311 		 error);
2312 	if (ret < 0)
2313 		return ret;
2314 	return 0;
2315 }
2316 
2317 /**
2318  * Validate VXLAN item.
2319  *
2320  * @param[in] item
2321  *   Item specification.
2322  * @param[in] item_flags
2323  *   Bit-fields that holds the items detected until now.
2324  * @param[in] target_protocol
2325  *   The next protocol in the previous item.
2326  * @param[out] error
2327  *   Pointer to error structure.
2328  *
2329  * @return
2330  *   0 on success, a negative errno value otherwise and rte_errno is set.
2331  */
2332 int
2333 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2334 			      uint64_t item_flags,
2335 			      struct rte_flow_error *error)
2336 {
2337 	const struct rte_flow_item_vxlan *spec = item->spec;
2338 	const struct rte_flow_item_vxlan *mask = item->mask;
2339 	int ret;
2340 	union vni {
2341 		uint32_t vlan_id;
2342 		uint8_t vni[4];
2343 	} id = { .vlan_id = 0, };
2344 
2345 
2346 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2347 		return rte_flow_error_set(error, ENOTSUP,
2348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2349 					  "multiple tunnel layers not"
2350 					  " supported");
2351 	/*
2352 	 * Verify only UDPv4 is present as defined in
2353 	 * https://tools.ietf.org/html/rfc7348
2354 	 */
2355 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2356 		return rte_flow_error_set(error, EINVAL,
2357 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2358 					  "no outer UDP layer found");
2359 	if (!mask)
2360 		mask = &rte_flow_item_vxlan_mask;
2361 	ret = mlx5_flow_item_acceptable
2362 		(item, (const uint8_t *)mask,
2363 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2364 		 sizeof(struct rte_flow_item_vxlan),
2365 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2366 	if (ret < 0)
2367 		return ret;
2368 	if (spec) {
2369 		memcpy(&id.vni[1], spec->vni, 3);
2370 		memcpy(&id.vni[1], mask->vni, 3);
2371 	}
2372 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2373 		return rte_flow_error_set(error, ENOTSUP,
2374 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2375 					  "VXLAN tunnel must be fully defined");
2376 	return 0;
2377 }
2378 
2379 /**
2380  * Validate VXLAN_GPE item.
2381  *
2382  * @param[in] item
2383  *   Item specification.
2384  * @param[in] item_flags
2385  *   Bit-fields that holds the items detected until now.
2386  * @param[in] priv
2387  *   Pointer to the private data structure.
2388  * @param[in] target_protocol
2389  *   The next protocol in the previous item.
2390  * @param[out] error
2391  *   Pointer to error structure.
2392  *
2393  * @return
2394  *   0 on success, a negative errno value otherwise and rte_errno is set.
2395  */
2396 int
2397 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2398 				  uint64_t item_flags,
2399 				  struct rte_eth_dev *dev,
2400 				  struct rte_flow_error *error)
2401 {
2402 	struct mlx5_priv *priv = dev->data->dev_private;
2403 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2404 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2405 	int ret;
2406 	union vni {
2407 		uint32_t vlan_id;
2408 		uint8_t vni[4];
2409 	} id = { .vlan_id = 0, };
2410 
2411 	if (!priv->config.l3_vxlan_en)
2412 		return rte_flow_error_set(error, ENOTSUP,
2413 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2414 					  "L3 VXLAN is not enabled by device"
2415 					  " parameter and/or not configured in"
2416 					  " firmware");
2417 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2418 		return rte_flow_error_set(error, ENOTSUP,
2419 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2420 					  "multiple tunnel layers not"
2421 					  " supported");
2422 	/*
2423 	 * Verify only UDPv4 is present as defined in
2424 	 * https://tools.ietf.org/html/rfc7348
2425 	 */
2426 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2427 		return rte_flow_error_set(error, EINVAL,
2428 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2429 					  "no outer UDP layer found");
2430 	if (!mask)
2431 		mask = &rte_flow_item_vxlan_gpe_mask;
2432 	ret = mlx5_flow_item_acceptable
2433 		(item, (const uint8_t *)mask,
2434 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2435 		 sizeof(struct rte_flow_item_vxlan_gpe),
2436 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2437 	if (ret < 0)
2438 		return ret;
2439 	if (spec) {
2440 		if (spec->protocol)
2441 			return rte_flow_error_set(error, ENOTSUP,
2442 						  RTE_FLOW_ERROR_TYPE_ITEM,
2443 						  item,
2444 						  "VxLAN-GPE protocol"
2445 						  " not supported");
2446 		memcpy(&id.vni[1], spec->vni, 3);
2447 		memcpy(&id.vni[1], mask->vni, 3);
2448 	}
2449 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2450 		return rte_flow_error_set(error, ENOTSUP,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 					  "VXLAN-GPE tunnel must be fully"
2453 					  " defined");
2454 	return 0;
2455 }
2456 /**
2457  * Validate GRE Key item.
2458  *
2459  * @param[in] item
2460  *   Item specification.
2461  * @param[in] item_flags
2462  *   Bit flags to mark detected items.
2463  * @param[in] gre_item
2464  *   Pointer to gre_item
2465  * @param[out] error
2466  *   Pointer to error structure.
2467  *
2468  * @return
2469  *   0 on success, a negative errno value otherwise and rte_errno is set.
2470  */
2471 int
2472 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2473 				uint64_t item_flags,
2474 				const struct rte_flow_item *gre_item,
2475 				struct rte_flow_error *error)
2476 {
2477 	const rte_be32_t *mask = item->mask;
2478 	int ret = 0;
2479 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2480 	const struct rte_flow_item_gre *gre_spec;
2481 	const struct rte_flow_item_gre *gre_mask;
2482 
2483 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2484 		return rte_flow_error_set(error, ENOTSUP,
2485 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2486 					  "Multiple GRE key not support");
2487 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2488 		return rte_flow_error_set(error, ENOTSUP,
2489 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2490 					  "No preceding GRE header");
2491 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2492 		return rte_flow_error_set(error, ENOTSUP,
2493 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2494 					  "GRE key following a wrong item");
2495 	gre_mask = gre_item->mask;
2496 	if (!gre_mask)
2497 		gre_mask = &rte_flow_item_gre_mask;
2498 	gre_spec = gre_item->spec;
2499 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2500 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2501 		return rte_flow_error_set(error, EINVAL,
2502 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2503 					  "Key bit must be on");
2504 
2505 	if (!mask)
2506 		mask = &gre_key_default_mask;
2507 	ret = mlx5_flow_item_acceptable
2508 		(item, (const uint8_t *)mask,
2509 		 (const uint8_t *)&gre_key_default_mask,
2510 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2511 	return ret;
2512 }
2513 
2514 /**
2515  * Validate GRE item.
2516  *
2517  * @param[in] item
2518  *   Item specification.
2519  * @param[in] item_flags
2520  *   Bit flags to mark detected items.
2521  * @param[in] target_protocol
2522  *   The next protocol in the previous item.
2523  * @param[out] error
2524  *   Pointer to error structure.
2525  *
2526  * @return
2527  *   0 on success, a negative errno value otherwise and rte_errno is set.
2528  */
2529 int
2530 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2531 			    uint64_t item_flags,
2532 			    uint8_t target_protocol,
2533 			    struct rte_flow_error *error)
2534 {
2535 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2536 	const struct rte_flow_item_gre *mask = item->mask;
2537 	int ret;
2538 	const struct rte_flow_item_gre nic_mask = {
2539 		.c_rsvd0_ver = RTE_BE16(0xB000),
2540 		.protocol = RTE_BE16(UINT16_MAX),
2541 	};
2542 
2543 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2544 		return rte_flow_error_set(error, EINVAL,
2545 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2546 					  "protocol filtering not compatible"
2547 					  " with this GRE layer");
2548 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2549 		return rte_flow_error_set(error, ENOTSUP,
2550 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2551 					  "multiple tunnel layers not"
2552 					  " supported");
2553 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2554 		return rte_flow_error_set(error, ENOTSUP,
2555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2556 					  "L3 Layer is missing");
2557 	if (!mask)
2558 		mask = &rte_flow_item_gre_mask;
2559 	ret = mlx5_flow_item_acceptable
2560 		(item, (const uint8_t *)mask,
2561 		 (const uint8_t *)&nic_mask,
2562 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2563 		 error);
2564 	if (ret < 0)
2565 		return ret;
2566 #ifndef HAVE_MLX5DV_DR
2567 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2568 	if (spec && (spec->protocol & mask->protocol))
2569 		return rte_flow_error_set(error, ENOTSUP,
2570 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2571 					  "without MPLS support the"
2572 					  " specification cannot be used for"
2573 					  " filtering");
2574 #endif
2575 #endif
2576 	return 0;
2577 }
2578 
2579 /**
2580  * Validate Geneve item.
2581  *
2582  * @param[in] item
2583  *   Item specification.
2584  * @param[in] itemFlags
2585  *   Bit-fields that holds the items detected until now.
2586  * @param[in] enPriv
2587  *   Pointer to the private data structure.
2588  * @param[out] error
2589  *   Pointer to error structure.
2590  *
2591  * @return
2592  *   0 on success, a negative errno value otherwise and rte_errno is set.
2593  */
2594 
2595 int
2596 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2597 			       uint64_t item_flags,
2598 			       struct rte_eth_dev *dev,
2599 			       struct rte_flow_error *error)
2600 {
2601 	struct mlx5_priv *priv = dev->data->dev_private;
2602 	const struct rte_flow_item_geneve *spec = item->spec;
2603 	const struct rte_flow_item_geneve *mask = item->mask;
2604 	int ret;
2605 	uint16_t gbhdr;
2606 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2607 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2608 	const struct rte_flow_item_geneve nic_mask = {
2609 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2610 		.vni = "\xff\xff\xff",
2611 		.protocol = RTE_BE16(UINT16_MAX),
2612 	};
2613 
2614 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2615 		return rte_flow_error_set(error, ENOTSUP,
2616 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2617 					  "L3 Geneve is not enabled by device"
2618 					  " parameter and/or not configured in"
2619 					  " firmware");
2620 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2621 		return rte_flow_error_set(error, ENOTSUP,
2622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2623 					  "multiple tunnel layers not"
2624 					  " supported");
2625 	/*
2626 	 * Verify only UDPv4 is present as defined in
2627 	 * https://tools.ietf.org/html/rfc7348
2628 	 */
2629 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2630 		return rte_flow_error_set(error, EINVAL,
2631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2632 					  "no outer UDP layer found");
2633 	if (!mask)
2634 		mask = &rte_flow_item_geneve_mask;
2635 	ret = mlx5_flow_item_acceptable
2636 				  (item, (const uint8_t *)mask,
2637 				   (const uint8_t *)&nic_mask,
2638 				   sizeof(struct rte_flow_item_geneve),
2639 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2640 	if (ret)
2641 		return ret;
2642 	if (spec) {
2643 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2644 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2645 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2646 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2647 			return rte_flow_error_set(error, ENOTSUP,
2648 						  RTE_FLOW_ERROR_TYPE_ITEM,
2649 						  item,
2650 						  "Geneve protocol unsupported"
2651 						  " fields are being used");
2652 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2653 			return rte_flow_error_set
2654 					(error, ENOTSUP,
2655 					 RTE_FLOW_ERROR_TYPE_ITEM,
2656 					 item,
2657 					 "Unsupported Geneve options length");
2658 	}
2659 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2660 		return rte_flow_error_set
2661 				    (error, ENOTSUP,
2662 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2663 				     "Geneve tunnel must be fully defined");
2664 	return 0;
2665 }
2666 
2667 /**
2668  * Validate Geneve TLV option item.
2669  *
2670  * @param[in] item
2671  *   Item specification.
2672  * @param[in] last_item
2673  *   Previous validated item in the pattern items.
2674  * @param[in] geneve_item
2675  *   Previous GENEVE item specification.
2676  * @param[in] dev
2677  *   Pointer to the rte_eth_dev structure.
2678  * @param[out] error
2679  *   Pointer to error structure.
2680  *
2681  * @return
2682  *   0 on success, a negative errno value otherwise and rte_errno is set.
2683  */
2684 int
2685 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2686 				   uint64_t last_item,
2687 				   const struct rte_flow_item *geneve_item,
2688 				   struct rte_eth_dev *dev,
2689 				   struct rte_flow_error *error)
2690 {
2691 	struct mlx5_priv *priv = dev->data->dev_private;
2692 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2693 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2694 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2695 	uint8_t data_max_supported =
2696 			hca_attr->max_geneve_tlv_option_data_len * 4;
2697 	struct mlx5_dev_config *config = &priv->config;
2698 	const struct rte_flow_item_geneve *geneve_spec;
2699 	const struct rte_flow_item_geneve *geneve_mask;
2700 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2701 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2702 	unsigned int i;
2703 	unsigned int data_len;
2704 	uint8_t tlv_option_len;
2705 	uint16_t optlen_m, optlen_v;
2706 	const struct rte_flow_item_geneve_opt full_mask = {
2707 		.option_class = RTE_BE16(0xffff),
2708 		.option_type = 0xff,
2709 		.option_len = 0x1f,
2710 	};
2711 
2712 	if (!mask)
2713 		mask = &rte_flow_item_geneve_opt_mask;
2714 	if (!spec)
2715 		return rte_flow_error_set
2716 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2717 			"Geneve TLV opt class/type/length must be specified");
2718 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2719 		return rte_flow_error_set
2720 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2721 			"Geneve TLV opt length exceeeds the limit (31)");
2722 	/* Check if class type and length masks are full. */
2723 	if (full_mask.option_class != mask->option_class ||
2724 	    full_mask.option_type != mask->option_type ||
2725 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2726 		return rte_flow_error_set
2727 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2728 			"Geneve TLV opt class/type/length masks must be full");
2729 	/* Check if length is supported */
2730 	if ((uint32_t)spec->option_len >
2731 			config->hca_attr.max_geneve_tlv_option_data_len)
2732 		return rte_flow_error_set
2733 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2734 			"Geneve TLV opt length not supported");
2735 	if (config->hca_attr.max_geneve_tlv_options > 1)
2736 		DRV_LOG(DEBUG,
2737 			"max_geneve_tlv_options supports more than 1 option");
2738 	/* Check GENEVE item preceding. */
2739 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2740 		return rte_flow_error_set
2741 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 			"Geneve opt item must be preceded with Geneve item");
2743 	geneve_spec = geneve_item->spec;
2744 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2745 					  &rte_flow_item_geneve_mask;
2746 	/* Check if GENEVE TLV option size doesn't exceed option length */
2747 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2748 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2749 		tlv_option_len = spec->option_len & mask->option_len;
2750 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2751 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2752 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2753 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2754 		if ((optlen_v & optlen_m) <= tlv_option_len)
2755 			return rte_flow_error_set
2756 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2757 				 "GENEVE TLV option length exceeds optlen");
2758 	}
2759 	/* Check if length is 0 or data is 0. */
2760 	if (spec->data == NULL || spec->option_len == 0)
2761 		return rte_flow_error_set
2762 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2763 			"Geneve TLV opt with zero data/length not supported");
2764 	/* Check not all data & mask are 0. */
2765 	data_len = spec->option_len * 4;
2766 	if (mask->data == NULL) {
2767 		for (i = 0; i < data_len; i++)
2768 			if (spec->data[i])
2769 				break;
2770 		if (i == data_len)
2771 			return rte_flow_error_set(error, ENOTSUP,
2772 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2773 				"Can't match on Geneve option data 0");
2774 	} else {
2775 		for (i = 0; i < data_len; i++)
2776 			if (spec->data[i] & mask->data[i])
2777 				break;
2778 		if (i == data_len)
2779 			return rte_flow_error_set(error, ENOTSUP,
2780 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2781 				"Can't match on Geneve option data and mask 0");
2782 		/* Check data mask supported. */
2783 		for (i = data_max_supported; i < data_len ; i++)
2784 			if (mask->data[i])
2785 				return rte_flow_error_set(error, ENOTSUP,
2786 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2787 					"Data mask is of unsupported size");
2788 	}
2789 	/* Check GENEVE option is supported in NIC. */
2790 	if (!config->hca_attr.geneve_tlv_opt)
2791 		return rte_flow_error_set
2792 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 			"Geneve TLV opt not supported");
2794 	/* Check if we already have geneve option with different type/class. */
2795 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2796 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2797 	if (geneve_opt_resource != NULL)
2798 		if (geneve_opt_resource->option_class != spec->option_class ||
2799 		    geneve_opt_resource->option_type != spec->option_type ||
2800 		    geneve_opt_resource->length != spec->option_len) {
2801 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2802 			return rte_flow_error_set(error, ENOTSUP,
2803 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2804 				"Only one Geneve TLV option supported");
2805 		}
2806 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2807 	return 0;
2808 }
2809 
2810 /**
2811  * Validate MPLS item.
2812  *
2813  * @param[in] dev
2814  *   Pointer to the rte_eth_dev structure.
2815  * @param[in] item
2816  *   Item specification.
2817  * @param[in] item_flags
2818  *   Bit-fields that holds the items detected until now.
2819  * @param[in] prev_layer
2820  *   The protocol layer indicated in previous item.
2821  * @param[out] error
2822  *   Pointer to error structure.
2823  *
2824  * @return
2825  *   0 on success, a negative errno value otherwise and rte_errno is set.
2826  */
2827 int
2828 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2829 			     const struct rte_flow_item *item __rte_unused,
2830 			     uint64_t item_flags __rte_unused,
2831 			     uint64_t prev_layer __rte_unused,
2832 			     struct rte_flow_error *error)
2833 {
2834 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2835 	const struct rte_flow_item_mpls *mask = item->mask;
2836 	struct mlx5_priv *priv = dev->data->dev_private;
2837 	int ret;
2838 
2839 	if (!priv->config.mpls_en)
2840 		return rte_flow_error_set(error, ENOTSUP,
2841 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2842 					  "MPLS not supported or"
2843 					  " disabled in firmware"
2844 					  " configuration.");
2845 	/* MPLS over IP, UDP, GRE is allowed */
2846 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2847 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2848 			    MLX5_FLOW_LAYER_GRE |
2849 			    MLX5_FLOW_LAYER_GRE_KEY)))
2850 		return rte_flow_error_set(error, EINVAL,
2851 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2852 					  "protocol filtering not compatible"
2853 					  " with MPLS layer");
2854 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2855 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2856 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2857 		return rte_flow_error_set(error, ENOTSUP,
2858 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 					  "multiple tunnel layers not"
2860 					  " supported");
2861 	if (!mask)
2862 		mask = &rte_flow_item_mpls_mask;
2863 	ret = mlx5_flow_item_acceptable
2864 		(item, (const uint8_t *)mask,
2865 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2866 		 sizeof(struct rte_flow_item_mpls),
2867 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2868 	if (ret < 0)
2869 		return ret;
2870 	return 0;
2871 #else
2872 	return rte_flow_error_set(error, ENOTSUP,
2873 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 				  "MPLS is not supported by Verbs, please"
2875 				  " update.");
2876 #endif
2877 }
2878 
2879 /**
2880  * Validate NVGRE item.
2881  *
2882  * @param[in] item
2883  *   Item specification.
2884  * @param[in] item_flags
2885  *   Bit flags to mark detected items.
2886  * @param[in] target_protocol
2887  *   The next protocol in the previous item.
2888  * @param[out] error
2889  *   Pointer to error structure.
2890  *
2891  * @return
2892  *   0 on success, a negative errno value otherwise and rte_errno is set.
2893  */
2894 int
2895 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2896 			      uint64_t item_flags,
2897 			      uint8_t target_protocol,
2898 			      struct rte_flow_error *error)
2899 {
2900 	const struct rte_flow_item_nvgre *mask = item->mask;
2901 	int ret;
2902 
2903 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2904 		return rte_flow_error_set(error, EINVAL,
2905 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2906 					  "protocol filtering not compatible"
2907 					  " with this GRE layer");
2908 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2909 		return rte_flow_error_set(error, ENOTSUP,
2910 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2911 					  "multiple tunnel layers not"
2912 					  " supported");
2913 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2914 		return rte_flow_error_set(error, ENOTSUP,
2915 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2916 					  "L3 Layer is missing");
2917 	if (!mask)
2918 		mask = &rte_flow_item_nvgre_mask;
2919 	ret = mlx5_flow_item_acceptable
2920 		(item, (const uint8_t *)mask,
2921 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2922 		 sizeof(struct rte_flow_item_nvgre),
2923 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2924 	if (ret < 0)
2925 		return ret;
2926 	return 0;
2927 }
2928 
2929 /**
2930  * Validate eCPRI item.
2931  *
2932  * @param[in] item
2933  *   Item specification.
2934  * @param[in] item_flags
2935  *   Bit-fields that holds the items detected until now.
2936  * @param[in] last_item
2937  *   Previous validated item in the pattern items.
2938  * @param[in] ether_type
2939  *   Type in the ethernet layer header (including dot1q).
2940  * @param[in] acc_mask
2941  *   Acceptable mask, if NULL default internal default mask
2942  *   will be used to check whether item fields are supported.
2943  * @param[out] error
2944  *   Pointer to error structure.
2945  *
2946  * @return
2947  *   0 on success, a negative errno value otherwise and rte_errno is set.
2948  */
2949 int
2950 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2951 			      uint64_t item_flags,
2952 			      uint64_t last_item,
2953 			      uint16_t ether_type,
2954 			      const struct rte_flow_item_ecpri *acc_mask,
2955 			      struct rte_flow_error *error)
2956 {
2957 	const struct rte_flow_item_ecpri *mask = item->mask;
2958 	const struct rte_flow_item_ecpri nic_mask = {
2959 		.hdr = {
2960 			.common = {
2961 				.u32 =
2962 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2963 					.type = 0xFF,
2964 					}).u32),
2965 			},
2966 			.dummy[0] = 0xFFFFFFFF,
2967 		},
2968 	};
2969 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2970 					MLX5_FLOW_LAYER_OUTER_VLAN);
2971 	struct rte_flow_item_ecpri mask_lo;
2972 
2973 	if (!(last_item & outer_l2_vlan) &&
2974 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2975 		return rte_flow_error_set(error, EINVAL,
2976 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2977 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2978 	if ((last_item & outer_l2_vlan) && ether_type &&
2979 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2980 		return rte_flow_error_set(error, EINVAL,
2981 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2982 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2983 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2984 		return rte_flow_error_set(error, EINVAL,
2985 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2986 					  "eCPRI with tunnel is not supported right now");
2987 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2988 		return rte_flow_error_set(error, ENOTSUP,
2989 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2990 					  "multiple L3 layers not supported");
2991 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2992 		return rte_flow_error_set(error, EINVAL,
2993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2994 					  "eCPRI cannot coexist with a TCP layer");
2995 	/* In specification, eCPRI could be over UDP layer. */
2996 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2997 		return rte_flow_error_set(error, EINVAL,
2998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2999 					  "eCPRI over UDP layer is not yet supported right now");
3000 	/* Mask for type field in common header could be zero. */
3001 	if (!mask)
3002 		mask = &rte_flow_item_ecpri_mask;
3003 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3004 	/* Input mask is in big-endian format. */
3005 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3006 		return rte_flow_error_set(error, EINVAL,
3007 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3008 					  "partial mask is not supported for protocol");
3009 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3010 		return rte_flow_error_set(error, EINVAL,
3011 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3012 					  "message header mask must be after a type mask");
3013 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3014 					 acc_mask ? (const uint8_t *)acc_mask
3015 						  : (const uint8_t *)&nic_mask,
3016 					 sizeof(struct rte_flow_item_ecpri),
3017 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3018 }
3019 
3020 /**
3021  * Release resource related QUEUE/RSS action split.
3022  *
3023  * @param dev
3024  *   Pointer to Ethernet device.
3025  * @param flow
3026  *   Flow to release id's from.
3027  */
3028 static void
3029 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3030 			     struct rte_flow *flow)
3031 {
3032 	struct mlx5_priv *priv = dev->data->dev_private;
3033 	uint32_t handle_idx;
3034 	struct mlx5_flow_handle *dev_handle;
3035 
3036 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3037 		       handle_idx, dev_handle, next)
3038 		if (dev_handle->split_flow_id &&
3039 		    !dev_handle->is_meter_flow_id)
3040 			mlx5_ipool_free(priv->sh->ipool
3041 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3042 					dev_handle->split_flow_id);
3043 }
3044 
3045 static int
3046 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3047 		   const struct rte_flow_attr *attr __rte_unused,
3048 		   const struct rte_flow_item items[] __rte_unused,
3049 		   const struct rte_flow_action actions[] __rte_unused,
3050 		   bool external __rte_unused,
3051 		   int hairpin __rte_unused,
3052 		   struct rte_flow_error *error)
3053 {
3054 	return rte_flow_error_set(error, ENOTSUP,
3055 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3056 }
3057 
3058 static struct mlx5_flow *
3059 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3060 		  const struct rte_flow_attr *attr __rte_unused,
3061 		  const struct rte_flow_item items[] __rte_unused,
3062 		  const struct rte_flow_action actions[] __rte_unused,
3063 		  struct rte_flow_error *error)
3064 {
3065 	rte_flow_error_set(error, ENOTSUP,
3066 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3067 	return NULL;
3068 }
3069 
3070 static int
3071 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3072 		    struct mlx5_flow *dev_flow __rte_unused,
3073 		    const struct rte_flow_attr *attr __rte_unused,
3074 		    const struct rte_flow_item items[] __rte_unused,
3075 		    const struct rte_flow_action actions[] __rte_unused,
3076 		    struct rte_flow_error *error)
3077 {
3078 	return rte_flow_error_set(error, ENOTSUP,
3079 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3080 }
3081 
3082 static int
3083 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3084 		struct rte_flow *flow __rte_unused,
3085 		struct rte_flow_error *error)
3086 {
3087 	return rte_flow_error_set(error, ENOTSUP,
3088 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3089 }
3090 
3091 static void
3092 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3093 		 struct rte_flow *flow __rte_unused)
3094 {
3095 }
3096 
3097 static void
3098 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3099 		  struct rte_flow *flow __rte_unused)
3100 {
3101 }
3102 
3103 static int
3104 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3105 		struct rte_flow *flow __rte_unused,
3106 		const struct rte_flow_action *actions __rte_unused,
3107 		void *data __rte_unused,
3108 		struct rte_flow_error *error)
3109 {
3110 	return rte_flow_error_set(error, ENOTSUP,
3111 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3112 }
3113 
3114 static int
3115 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3116 		      uint32_t domains __rte_unused,
3117 		      uint32_t flags __rte_unused)
3118 {
3119 	return 0;
3120 }
3121 
3122 /* Void driver to protect from null pointer reference. */
3123 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3124 	.validate = flow_null_validate,
3125 	.prepare = flow_null_prepare,
3126 	.translate = flow_null_translate,
3127 	.apply = flow_null_apply,
3128 	.remove = flow_null_remove,
3129 	.destroy = flow_null_destroy,
3130 	.query = flow_null_query,
3131 	.sync_domain = flow_null_sync_domain,
3132 };
3133 
3134 /**
3135  * Select flow driver type according to flow attributes and device
3136  * configuration.
3137  *
3138  * @param[in] dev
3139  *   Pointer to the dev structure.
3140  * @param[in] attr
3141  *   Pointer to the flow attributes.
3142  *
3143  * @return
3144  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3145  */
3146 static enum mlx5_flow_drv_type
3147 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3148 {
3149 	struct mlx5_priv *priv = dev->data->dev_private;
3150 	/* The OS can determine first a specific flow type (DV, VERBS) */
3151 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3152 
3153 	if (type != MLX5_FLOW_TYPE_MAX)
3154 		return type;
3155 	/* If no OS specific type - continue with DV/VERBS selection */
3156 	if (attr->transfer && priv->config.dv_esw_en)
3157 		type = MLX5_FLOW_TYPE_DV;
3158 	if (!attr->transfer)
3159 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3160 						 MLX5_FLOW_TYPE_VERBS;
3161 	return type;
3162 }
3163 
3164 #define flow_get_drv_ops(type) flow_drv_ops[type]
3165 
3166 /**
3167  * Flow driver validation API. This abstracts calling driver specific functions.
3168  * The type of flow driver is determined according to flow attributes.
3169  *
3170  * @param[in] dev
3171  *   Pointer to the dev structure.
3172  * @param[in] attr
3173  *   Pointer to the flow attributes.
3174  * @param[in] items
3175  *   Pointer to the list of items.
3176  * @param[in] actions
3177  *   Pointer to the list of actions.
3178  * @param[in] external
3179  *   This flow rule is created by request external to PMD.
3180  * @param[in] hairpin
3181  *   Number of hairpin TX actions, 0 means classic flow.
3182  * @param[out] error
3183  *   Pointer to the error structure.
3184  *
3185  * @return
3186  *   0 on success, a negative errno value otherwise and rte_errno is set.
3187  */
3188 static inline int
3189 flow_drv_validate(struct rte_eth_dev *dev,
3190 		  const struct rte_flow_attr *attr,
3191 		  const struct rte_flow_item items[],
3192 		  const struct rte_flow_action actions[],
3193 		  bool external, int hairpin, struct rte_flow_error *error)
3194 {
3195 	const struct mlx5_flow_driver_ops *fops;
3196 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3197 
3198 	fops = flow_get_drv_ops(type);
3199 	return fops->validate(dev, attr, items, actions, external,
3200 			      hairpin, error);
3201 }
3202 
3203 /**
3204  * Flow driver preparation API. This abstracts calling driver specific
3205  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3206  * calculates the size of memory required for device flow, allocates the memory,
3207  * initializes the device flow and returns the pointer.
3208  *
3209  * @note
3210  *   This function initializes device flow structure such as dv or verbs in
3211  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3212  *   rest. For example, adding returning device flow to flow->dev_flow list and
3213  *   setting backward reference to the flow should be done out of this function.
3214  *   layers field is not filled either.
3215  *
3216  * @param[in] dev
3217  *   Pointer to the dev structure.
3218  * @param[in] attr
3219  *   Pointer to the flow attributes.
3220  * @param[in] items
3221  *   Pointer to the list of items.
3222  * @param[in] actions
3223  *   Pointer to the list of actions.
3224  * @param[in] flow_idx
3225  *   This memory pool index to the flow.
3226  * @param[out] error
3227  *   Pointer to the error structure.
3228  *
3229  * @return
3230  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3231  */
3232 static inline struct mlx5_flow *
3233 flow_drv_prepare(struct rte_eth_dev *dev,
3234 		 const struct rte_flow *flow,
3235 		 const struct rte_flow_attr *attr,
3236 		 const struct rte_flow_item items[],
3237 		 const struct rte_flow_action actions[],
3238 		 uint32_t flow_idx,
3239 		 struct rte_flow_error *error)
3240 {
3241 	const struct mlx5_flow_driver_ops *fops;
3242 	enum mlx5_flow_drv_type type = flow->drv_type;
3243 	struct mlx5_flow *mlx5_flow = NULL;
3244 
3245 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3246 	fops = flow_get_drv_ops(type);
3247 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3248 	if (mlx5_flow)
3249 		mlx5_flow->flow_idx = flow_idx;
3250 	return mlx5_flow;
3251 }
3252 
3253 /**
3254  * Flow driver translation API. This abstracts calling driver specific
3255  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3256  * translates a generic flow into a driver flow. flow_drv_prepare() must
3257  * precede.
3258  *
3259  * @note
3260  *   dev_flow->layers could be filled as a result of parsing during translation
3261  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3262  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3263  *   flow->actions could be overwritten even though all the expanded dev_flows
3264  *   have the same actions.
3265  *
3266  * @param[in] dev
3267  *   Pointer to the rte dev structure.
3268  * @param[in, out] dev_flow
3269  *   Pointer to the mlx5 flow.
3270  * @param[in] attr
3271  *   Pointer to the flow attributes.
3272  * @param[in] items
3273  *   Pointer to the list of items.
3274  * @param[in] actions
3275  *   Pointer to the list of actions.
3276  * @param[out] error
3277  *   Pointer to the error structure.
3278  *
3279  * @return
3280  *   0 on success, a negative errno value otherwise and rte_errno is set.
3281  */
3282 static inline int
3283 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3284 		   const struct rte_flow_attr *attr,
3285 		   const struct rte_flow_item items[],
3286 		   const struct rte_flow_action actions[],
3287 		   struct rte_flow_error *error)
3288 {
3289 	const struct mlx5_flow_driver_ops *fops;
3290 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3291 
3292 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3293 	fops = flow_get_drv_ops(type);
3294 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3295 }
3296 
3297 /**
3298  * Flow driver apply API. This abstracts calling driver specific functions.
3299  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3300  * translated driver flows on to device. flow_drv_translate() must precede.
3301  *
3302  * @param[in] dev
3303  *   Pointer to Ethernet device structure.
3304  * @param[in, out] flow
3305  *   Pointer to flow structure.
3306  * @param[out] error
3307  *   Pointer to error structure.
3308  *
3309  * @return
3310  *   0 on success, a negative errno value otherwise and rte_errno is set.
3311  */
3312 static inline int
3313 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3314 	       struct rte_flow_error *error)
3315 {
3316 	const struct mlx5_flow_driver_ops *fops;
3317 	enum mlx5_flow_drv_type type = flow->drv_type;
3318 
3319 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3320 	fops = flow_get_drv_ops(type);
3321 	return fops->apply(dev, flow, error);
3322 }
3323 
3324 /**
3325  * Flow driver destroy API. This abstracts calling driver specific functions.
3326  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3327  * on device and releases resources of the flow.
3328  *
3329  * @param[in] dev
3330  *   Pointer to Ethernet device.
3331  * @param[in, out] flow
3332  *   Pointer to flow structure.
3333  */
3334 static inline void
3335 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3336 {
3337 	const struct mlx5_flow_driver_ops *fops;
3338 	enum mlx5_flow_drv_type type = flow->drv_type;
3339 
3340 	flow_mreg_split_qrss_release(dev, flow);
3341 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3342 	fops = flow_get_drv_ops(type);
3343 	fops->destroy(dev, flow);
3344 }
3345 
3346 /**
3347  * Get RSS action from the action list.
3348  *
3349  * @param[in] actions
3350  *   Pointer to the list of actions.
3351  *
3352  * @return
3353  *   Pointer to the RSS action if exist, else return NULL.
3354  */
3355 static const struct rte_flow_action_rss*
3356 flow_get_rss_action(const struct rte_flow_action actions[])
3357 {
3358 	const struct rte_flow_action_rss *rss = NULL;
3359 
3360 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3361 		switch (actions->type) {
3362 		case RTE_FLOW_ACTION_TYPE_RSS:
3363 			rss = actions->conf;
3364 			break;
3365 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3366 		{
3367 			const struct rte_flow_action_sample *sample =
3368 								actions->conf;
3369 			const struct rte_flow_action *act = sample->actions;
3370 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3371 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3372 					rss = act->conf;
3373 			break;
3374 		}
3375 		default:
3376 			break;
3377 		}
3378 	}
3379 	return rss;
3380 }
3381 
3382 /**
3383  * Get ASO age action by index.
3384  *
3385  * @param[in] dev
3386  *   Pointer to the Ethernet device structure.
3387  * @param[in] age_idx
3388  *   Index to the ASO age action.
3389  *
3390  * @return
3391  *   The specified ASO age action.
3392  */
3393 struct mlx5_aso_age_action*
3394 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3395 {
3396 	uint16_t pool_idx = age_idx & UINT16_MAX;
3397 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3398 	struct mlx5_priv *priv = dev->data->dev_private;
3399 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3400 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3401 
3402 	return &pool->actions[offset - 1];
3403 }
3404 
3405 /* maps indirect action to translated direct in some actions array */
3406 struct mlx5_translated_action_handle {
3407 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3408 	int index; /**< Index in related array of rte_flow_action. */
3409 };
3410 
3411 /**
3412  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3413  * direct action if translation possible.
3414  * This functionality used to run same execution path for both direct and
3415  * indirect actions on flow create. All necessary preparations for indirect
3416  * action handling should be performed on *handle* actions list returned
3417  * from this call.
3418  *
3419  * @param[in] dev
3420  *   Pointer to Ethernet device.
3421  * @param[in] actions
3422  *   List of actions to translate.
3423  * @param[out] handle
3424  *   List to store translated indirect action object handles.
3425  * @param[in, out] indir_n
3426  *   Size of *handle* array. On return should be updated with number of
3427  *   indirect actions retrieved from the *actions* list.
3428  * @param[out] translated_actions
3429  *   List of actions where all indirect actions were translated to direct
3430  *   if possible. NULL if no translation took place.
3431  * @param[out] error
3432  *   Pointer to the error structure.
3433  *
3434  * @return
3435  *   0 on success, a negative errno value otherwise and rte_errno is set.
3436  */
3437 static int
3438 flow_action_handles_translate(struct rte_eth_dev *dev,
3439 			      const struct rte_flow_action actions[],
3440 			      struct mlx5_translated_action_handle *handle,
3441 			      int *indir_n,
3442 			      struct rte_flow_action **translated_actions,
3443 			      struct rte_flow_error *error)
3444 {
3445 	struct mlx5_priv *priv = dev->data->dev_private;
3446 	struct rte_flow_action *translated = NULL;
3447 	size_t actions_size;
3448 	int n;
3449 	int copied_n = 0;
3450 	struct mlx5_translated_action_handle *handle_end = NULL;
3451 
3452 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3453 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3454 			continue;
3455 		if (copied_n == *indir_n) {
3456 			return rte_flow_error_set
3457 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3458 				 NULL, "too many shared actions");
3459 		}
3460 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3461 			   sizeof(actions[n].conf));
3462 		handle[copied_n].index = n;
3463 		copied_n++;
3464 	}
3465 	n++;
3466 	*indir_n = copied_n;
3467 	if (!copied_n)
3468 		return 0;
3469 	actions_size = sizeof(struct rte_flow_action) * n;
3470 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3471 	if (!translated) {
3472 		rte_errno = ENOMEM;
3473 		return -ENOMEM;
3474 	}
3475 	memcpy(translated, actions, actions_size);
3476 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3477 		struct mlx5_shared_action_rss *shared_rss;
3478 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3479 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3480 		uint32_t idx = act_idx &
3481 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3482 
3483 		switch (type) {
3484 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3485 			shared_rss = mlx5_ipool_get
3486 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3487 			translated[handle->index].type =
3488 				RTE_FLOW_ACTION_TYPE_RSS;
3489 			translated[handle->index].conf =
3490 				&shared_rss->origin;
3491 			break;
3492 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3493 			if (priv->sh->flow_hit_aso_en) {
3494 				translated[handle->index].type =
3495 					(enum rte_flow_action_type)
3496 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3497 				translated[handle->index].conf =
3498 							 (void *)(uintptr_t)idx;
3499 				break;
3500 			}
3501 			/* Fall-through */
3502 		default:
3503 			mlx5_free(translated);
3504 			return rte_flow_error_set
3505 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3506 				 NULL, "invalid indirect action type");
3507 		}
3508 	}
3509 	*translated_actions = translated;
3510 	return 0;
3511 }
3512 
3513 /**
3514  * Get Shared RSS action from the action list.
3515  *
3516  * @param[in] dev
3517  *   Pointer to Ethernet device.
3518  * @param[in] shared
3519  *   Pointer to the list of actions.
3520  * @param[in] shared_n
3521  *   Actions list length.
3522  *
3523  * @return
3524  *   The MLX5 RSS action ID if exists, otherwise return 0.
3525  */
3526 static uint32_t
3527 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3528 			   struct mlx5_translated_action_handle *handle,
3529 			   int shared_n)
3530 {
3531 	struct mlx5_translated_action_handle *handle_end;
3532 	struct mlx5_priv *priv = dev->data->dev_private;
3533 	struct mlx5_shared_action_rss *shared_rss;
3534 
3535 
3536 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3537 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3538 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3539 		uint32_t idx = act_idx &
3540 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3541 		switch (type) {
3542 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3543 			shared_rss = mlx5_ipool_get
3544 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3545 									   idx);
3546 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3547 					   __ATOMIC_RELAXED);
3548 			return idx;
3549 		default:
3550 			break;
3551 		}
3552 	}
3553 	return 0;
3554 }
3555 
3556 static unsigned int
3557 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3558 {
3559 	const struct rte_flow_item *item;
3560 	unsigned int has_vlan = 0;
3561 
3562 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3563 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3564 			has_vlan = 1;
3565 			break;
3566 		}
3567 	}
3568 	if (has_vlan)
3569 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3570 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3571 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3572 			       MLX5_EXPANSION_ROOT_OUTER;
3573 }
3574 
3575 /**
3576  *  Get layer flags from the prefix flow.
3577  *
3578  *  Some flows may be split to several subflows, the prefix subflow gets the
3579  *  match items and the suffix sub flow gets the actions.
3580  *  Some actions need the user defined match item flags to get the detail for
3581  *  the action.
3582  *  This function helps the suffix flow to get the item layer flags from prefix
3583  *  subflow.
3584  *
3585  * @param[in] dev_flow
3586  *   Pointer the created preifx subflow.
3587  *
3588  * @return
3589  *   The layers get from prefix subflow.
3590  */
3591 static inline uint64_t
3592 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3593 {
3594 	uint64_t layers = 0;
3595 
3596 	/*
3597 	 * Layers bits could be localization, but usually the compiler will
3598 	 * help to do the optimization work for source code.
3599 	 * If no decap actions, use the layers directly.
3600 	 */
3601 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3602 		return dev_flow->handle->layers;
3603 	/* Convert L3 layers with decap action. */
3604 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3605 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3606 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3607 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3608 	/* Convert L4 layers with decap action.  */
3609 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3610 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3611 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3612 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3613 	return layers;
3614 }
3615 
3616 /**
3617  * Get metadata split action information.
3618  *
3619  * @param[in] actions
3620  *   Pointer to the list of actions.
3621  * @param[out] qrss
3622  *   Pointer to the return pointer.
3623  * @param[out] qrss_type
3624  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3625  *   if no QUEUE/RSS is found.
3626  * @param[out] encap_idx
3627  *   Pointer to the index of the encap action if exists, otherwise the last
3628  *   action index.
3629  *
3630  * @return
3631  *   Total number of actions.
3632  */
3633 static int
3634 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3635 				       const struct rte_flow_action **qrss,
3636 				       int *encap_idx)
3637 {
3638 	const struct rte_flow_action_raw_encap *raw_encap;
3639 	int actions_n = 0;
3640 	int raw_decap_idx = -1;
3641 
3642 	*encap_idx = -1;
3643 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3644 		switch (actions->type) {
3645 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3646 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3647 			*encap_idx = actions_n;
3648 			break;
3649 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3650 			raw_decap_idx = actions_n;
3651 			break;
3652 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3653 			raw_encap = actions->conf;
3654 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3655 				*encap_idx = raw_decap_idx != -1 ?
3656 						      raw_decap_idx : actions_n;
3657 			break;
3658 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3659 		case RTE_FLOW_ACTION_TYPE_RSS:
3660 			*qrss = actions;
3661 			break;
3662 		default:
3663 			break;
3664 		}
3665 		actions_n++;
3666 	}
3667 	if (*encap_idx == -1)
3668 		*encap_idx = actions_n;
3669 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3670 	return actions_n + 1;
3671 }
3672 
3673 /**
3674  * Check meter action from the action list.
3675  *
3676  * @param[in] actions
3677  *   Pointer to the list of actions.
3678  * @param[out] has_mtr
3679  *   Pointer to the meter exist flag.
3680  * @param[out] meter_id
3681  *   Pointer to the meter id.
3682  *
3683  * @return
3684  *   Total number of actions.
3685  */
3686 static int
3687 flow_check_meter_action(const struct rte_flow_action actions[],
3688 			bool *has_mtr,
3689 			uint32_t *meter_id)
3690 {
3691 	const struct rte_flow_action_meter *mtr = NULL;
3692 	int actions_n = 0;
3693 
3694 	MLX5_ASSERT(has_mtr);
3695 	*has_mtr = false;
3696 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3697 		switch (actions->type) {
3698 		case RTE_FLOW_ACTION_TYPE_METER:
3699 			mtr = actions->conf;
3700 			*meter_id = mtr->mtr_id;
3701 			*has_mtr = true;
3702 			break;
3703 		default:
3704 			break;
3705 		}
3706 		actions_n++;
3707 	}
3708 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3709 	return actions_n + 1;
3710 }
3711 
3712 /**
3713  * Check if the flow should be split due to hairpin.
3714  * The reason for the split is that in current HW we can't
3715  * support encap and push-vlan on Rx, so if a flow contains
3716  * these actions we move it to Tx.
3717  *
3718  * @param dev
3719  *   Pointer to Ethernet device.
3720  * @param[in] attr
3721  *   Flow rule attributes.
3722  * @param[in] actions
3723  *   Associated actions (list terminated by the END action).
3724  *
3725  * @return
3726  *   > 0 the number of actions and the flow should be split,
3727  *   0 when no split required.
3728  */
3729 static int
3730 flow_check_hairpin_split(struct rte_eth_dev *dev,
3731 			 const struct rte_flow_attr *attr,
3732 			 const struct rte_flow_action actions[])
3733 {
3734 	int queue_action = 0;
3735 	int action_n = 0;
3736 	int split = 0;
3737 	const struct rte_flow_action_queue *queue;
3738 	const struct rte_flow_action_rss *rss;
3739 	const struct rte_flow_action_raw_encap *raw_encap;
3740 	const struct rte_eth_hairpin_conf *conf;
3741 
3742 	if (!attr->ingress)
3743 		return 0;
3744 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3745 		switch (actions->type) {
3746 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3747 			queue = actions->conf;
3748 			if (queue == NULL)
3749 				return 0;
3750 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3751 			if (conf == NULL || conf->tx_explicit != 0)
3752 				return 0;
3753 			queue_action = 1;
3754 			action_n++;
3755 			break;
3756 		case RTE_FLOW_ACTION_TYPE_RSS:
3757 			rss = actions->conf;
3758 			if (rss == NULL || rss->queue_num == 0)
3759 				return 0;
3760 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3761 			if (conf == NULL || conf->tx_explicit != 0)
3762 				return 0;
3763 			queue_action = 1;
3764 			action_n++;
3765 			break;
3766 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3767 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3768 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3769 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3770 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3771 			split++;
3772 			action_n++;
3773 			break;
3774 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3775 			raw_encap = actions->conf;
3776 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3777 				split++;
3778 			action_n++;
3779 			break;
3780 		default:
3781 			action_n++;
3782 			break;
3783 		}
3784 	}
3785 	if (split && queue_action)
3786 		return action_n;
3787 	return 0;
3788 }
3789 
3790 /* Declare flow create/destroy prototype in advance. */
3791 static uint32_t
3792 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3793 		 const struct rte_flow_attr *attr,
3794 		 const struct rte_flow_item items[],
3795 		 const struct rte_flow_action actions[],
3796 		 bool external, struct rte_flow_error *error);
3797 
3798 static void
3799 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3800 		  uint32_t flow_idx);
3801 
3802 int
3803 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3804 		      struct mlx5_hlist_entry *entry,
3805 		      uint64_t key, void *cb_ctx __rte_unused)
3806 {
3807 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3808 		container_of(entry, typeof(*mcp_res), hlist_ent);
3809 
3810 	return mcp_res->mark_id != key;
3811 }
3812 
3813 struct mlx5_hlist_entry *
3814 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3815 		       void *cb_ctx)
3816 {
3817 	struct rte_eth_dev *dev = list->ctx;
3818 	struct mlx5_priv *priv = dev->data->dev_private;
3819 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3820 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3821 	struct rte_flow_error *error = ctx->error;
3822 	uint32_t idx = 0;
3823 	int ret;
3824 	uint32_t mark_id = key;
3825 	struct rte_flow_attr attr = {
3826 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3827 		.ingress = 1,
3828 	};
3829 	struct mlx5_rte_flow_item_tag tag_spec = {
3830 		.data = mark_id,
3831 	};
3832 	struct rte_flow_item items[] = {
3833 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3834 	};
3835 	struct rte_flow_action_mark ftag = {
3836 		.id = mark_id,
3837 	};
3838 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3839 		.dst = REG_B,
3840 		.src = REG_NON,
3841 	};
3842 	struct rte_flow_action_jump jump = {
3843 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3844 	};
3845 	struct rte_flow_action actions[] = {
3846 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3847 	};
3848 
3849 	/* Fill the register fileds in the flow. */
3850 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3851 	if (ret < 0)
3852 		return NULL;
3853 	tag_spec.id = ret;
3854 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3855 	if (ret < 0)
3856 		return NULL;
3857 	cp_mreg.src = ret;
3858 	/* Provide the full width of FLAG specific value. */
3859 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3860 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3861 	/* Build a new flow. */
3862 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3863 		items[0] = (struct rte_flow_item){
3864 			.type = (enum rte_flow_item_type)
3865 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3866 			.spec = &tag_spec,
3867 		};
3868 		items[1] = (struct rte_flow_item){
3869 			.type = RTE_FLOW_ITEM_TYPE_END,
3870 		};
3871 		actions[0] = (struct rte_flow_action){
3872 			.type = (enum rte_flow_action_type)
3873 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3874 			.conf = &ftag,
3875 		};
3876 		actions[1] = (struct rte_flow_action){
3877 			.type = (enum rte_flow_action_type)
3878 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3879 			.conf = &cp_mreg,
3880 		};
3881 		actions[2] = (struct rte_flow_action){
3882 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3883 			.conf = &jump,
3884 		};
3885 		actions[3] = (struct rte_flow_action){
3886 			.type = RTE_FLOW_ACTION_TYPE_END,
3887 		};
3888 	} else {
3889 		/* Default rule, wildcard match. */
3890 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
3891 		items[0] = (struct rte_flow_item){
3892 			.type = RTE_FLOW_ITEM_TYPE_END,
3893 		};
3894 		actions[0] = (struct rte_flow_action){
3895 			.type = (enum rte_flow_action_type)
3896 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3897 			.conf = &cp_mreg,
3898 		};
3899 		actions[1] = (struct rte_flow_action){
3900 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3901 			.conf = &jump,
3902 		};
3903 		actions[2] = (struct rte_flow_action){
3904 			.type = RTE_FLOW_ACTION_TYPE_END,
3905 		};
3906 	}
3907 	/* Build a new entry. */
3908 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3909 	if (!mcp_res) {
3910 		rte_errno = ENOMEM;
3911 		return NULL;
3912 	}
3913 	mcp_res->idx = idx;
3914 	mcp_res->mark_id = mark_id;
3915 	/*
3916 	 * The copy Flows are not included in any list. There
3917 	 * ones are referenced from other Flows and can not
3918 	 * be applied, removed, deleted in ardbitrary order
3919 	 * by list traversing.
3920 	 */
3921 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3922 					 actions, false, error);
3923 	if (!mcp_res->rix_flow) {
3924 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3925 		return NULL;
3926 	}
3927 	return &mcp_res->hlist_ent;
3928 }
3929 
3930 /**
3931  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3932  *
3933  * As mark_id is unique, if there's already a registered flow for the mark_id,
3934  * return by increasing the reference counter of the resource. Otherwise, create
3935  * the resource (mcp_res) and flow.
3936  *
3937  * Flow looks like,
3938  *   - If ingress port is ANY and reg_c[1] is mark_id,
3939  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3940  *
3941  * For default flow (zero mark_id), flow is like,
3942  *   - If ingress port is ANY,
3943  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3944  *
3945  * @param dev
3946  *   Pointer to Ethernet device.
3947  * @param mark_id
3948  *   ID of MARK action, zero means default flow for META.
3949  * @param[out] error
3950  *   Perform verbose error reporting if not NULL.
3951  *
3952  * @return
3953  *   Associated resource on success, NULL otherwise and rte_errno is set.
3954  */
3955 static struct mlx5_flow_mreg_copy_resource *
3956 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3957 			  struct rte_flow_error *error)
3958 {
3959 	struct mlx5_priv *priv = dev->data->dev_private;
3960 	struct mlx5_hlist_entry *entry;
3961 	struct mlx5_flow_cb_ctx ctx = {
3962 		.dev = dev,
3963 		.error = error,
3964 	};
3965 
3966 	/* Check if already registered. */
3967 	MLX5_ASSERT(priv->mreg_cp_tbl);
3968 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3969 	if (!entry)
3970 		return NULL;
3971 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3972 			    hlist_ent);
3973 }
3974 
3975 void
3976 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3977 {
3978 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3979 		container_of(entry, typeof(*mcp_res), hlist_ent);
3980 	struct rte_eth_dev *dev = list->ctx;
3981 	struct mlx5_priv *priv = dev->data->dev_private;
3982 
3983 	MLX5_ASSERT(mcp_res->rix_flow);
3984 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3985 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3986 }
3987 
3988 /**
3989  * Release flow in RX_CP_TBL.
3990  *
3991  * @param dev
3992  *   Pointer to Ethernet device.
3993  * @flow
3994  *   Parent flow for wich copying is provided.
3995  */
3996 static void
3997 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3998 			  struct rte_flow *flow)
3999 {
4000 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4001 	struct mlx5_priv *priv = dev->data->dev_private;
4002 
4003 	if (!flow->rix_mreg_copy)
4004 		return;
4005 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4006 				 flow->rix_mreg_copy);
4007 	if (!mcp_res || !priv->mreg_cp_tbl)
4008 		return;
4009 	MLX5_ASSERT(mcp_res->rix_flow);
4010 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4011 	flow->rix_mreg_copy = 0;
4012 }
4013 
4014 /**
4015  * Remove the default copy action from RX_CP_TBL.
4016  *
4017  * This functions is called in the mlx5_dev_start(). No thread safe
4018  * is guaranteed.
4019  *
4020  * @param dev
4021  *   Pointer to Ethernet device.
4022  */
4023 static void
4024 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4025 {
4026 	struct mlx5_hlist_entry *entry;
4027 	struct mlx5_priv *priv = dev->data->dev_private;
4028 
4029 	/* Check if default flow is registered. */
4030 	if (!priv->mreg_cp_tbl)
4031 		return;
4032 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4033 				  MLX5_DEFAULT_COPY_ID, NULL);
4034 	if (!entry)
4035 		return;
4036 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4037 }
4038 
4039 /**
4040  * Add the default copy action in in RX_CP_TBL.
4041  *
4042  * This functions is called in the mlx5_dev_start(). No thread safe
4043  * is guaranteed.
4044  *
4045  * @param dev
4046  *   Pointer to Ethernet device.
4047  * @param[out] error
4048  *   Perform verbose error reporting if not NULL.
4049  *
4050  * @return
4051  *   0 for success, negative value otherwise and rte_errno is set.
4052  */
4053 static int
4054 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4055 				  struct rte_flow_error *error)
4056 {
4057 	struct mlx5_priv *priv = dev->data->dev_private;
4058 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4059 
4060 	/* Check whether extensive metadata feature is engaged. */
4061 	if (!priv->config.dv_flow_en ||
4062 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4063 	    !mlx5_flow_ext_mreg_supported(dev) ||
4064 	    !priv->sh->dv_regc0_mask)
4065 		return 0;
4066 	/*
4067 	 * Add default mreg copy flow may be called multiple time, but
4068 	 * only be called once in stop. Avoid register it twice.
4069 	 */
4070 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4071 		return 0;
4072 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4073 	if (!mcp_res)
4074 		return -rte_errno;
4075 	return 0;
4076 }
4077 
4078 /**
4079  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4080  *
4081  * All the flow having Q/RSS action should be split by
4082  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4083  * performs the following,
4084  *   - CQE->flow_tag := reg_c[1] (MARK)
4085  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4086  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4087  * but there should be a flow per each MARK ID set by MARK action.
4088  *
4089  * For the aforementioned reason, if there's a MARK action in flow's action
4090  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4091  * the MARK ID to CQE's flow_tag like,
4092  *   - If reg_c[1] is mark_id,
4093  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4094  *
4095  * For SET_META action which stores value in reg_c[0], as the destination is
4096  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4097  * MARK ID means the default flow. The default flow looks like,
4098  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4099  *
4100  * @param dev
4101  *   Pointer to Ethernet device.
4102  * @param flow
4103  *   Pointer to flow structure.
4104  * @param[in] actions
4105  *   Pointer to the list of actions.
4106  * @param[out] error
4107  *   Perform verbose error reporting if not NULL.
4108  *
4109  * @return
4110  *   0 on success, negative value otherwise and rte_errno is set.
4111  */
4112 static int
4113 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4114 			    struct rte_flow *flow,
4115 			    const struct rte_flow_action *actions,
4116 			    struct rte_flow_error *error)
4117 {
4118 	struct mlx5_priv *priv = dev->data->dev_private;
4119 	struct mlx5_dev_config *config = &priv->config;
4120 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4121 	const struct rte_flow_action_mark *mark;
4122 
4123 	/* Check whether extensive metadata feature is engaged. */
4124 	if (!config->dv_flow_en ||
4125 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4126 	    !mlx5_flow_ext_mreg_supported(dev) ||
4127 	    !priv->sh->dv_regc0_mask)
4128 		return 0;
4129 	/* Find MARK action. */
4130 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4131 		switch (actions->type) {
4132 		case RTE_FLOW_ACTION_TYPE_FLAG:
4133 			mcp_res = flow_mreg_add_copy_action
4134 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4135 			if (!mcp_res)
4136 				return -rte_errno;
4137 			flow->rix_mreg_copy = mcp_res->idx;
4138 			return 0;
4139 		case RTE_FLOW_ACTION_TYPE_MARK:
4140 			mark = (const struct rte_flow_action_mark *)
4141 				actions->conf;
4142 			mcp_res =
4143 				flow_mreg_add_copy_action(dev, mark->id, error);
4144 			if (!mcp_res)
4145 				return -rte_errno;
4146 			flow->rix_mreg_copy = mcp_res->idx;
4147 			return 0;
4148 		default:
4149 			break;
4150 		}
4151 	}
4152 	return 0;
4153 }
4154 
4155 #define MLX5_MAX_SPLIT_ACTIONS 24
4156 #define MLX5_MAX_SPLIT_ITEMS 24
4157 
4158 /**
4159  * Split the hairpin flow.
4160  * Since HW can't support encap and push-vlan on Rx, we move these
4161  * actions to Tx.
4162  * If the count action is after the encap then we also
4163  * move the count action. in this case the count will also measure
4164  * the outer bytes.
4165  *
4166  * @param dev
4167  *   Pointer to Ethernet device.
4168  * @param[in] actions
4169  *   Associated actions (list terminated by the END action).
4170  * @param[out] actions_rx
4171  *   Rx flow actions.
4172  * @param[out] actions_tx
4173  *   Tx flow actions..
4174  * @param[out] pattern_tx
4175  *   The pattern items for the Tx flow.
4176  * @param[out] flow_id
4177  *   The flow ID connected to this flow.
4178  *
4179  * @return
4180  *   0 on success.
4181  */
4182 static int
4183 flow_hairpin_split(struct rte_eth_dev *dev,
4184 		   const struct rte_flow_action actions[],
4185 		   struct rte_flow_action actions_rx[],
4186 		   struct rte_flow_action actions_tx[],
4187 		   struct rte_flow_item pattern_tx[],
4188 		   uint32_t flow_id)
4189 {
4190 	const struct rte_flow_action_raw_encap *raw_encap;
4191 	const struct rte_flow_action_raw_decap *raw_decap;
4192 	struct mlx5_rte_flow_action_set_tag *set_tag;
4193 	struct rte_flow_action *tag_action;
4194 	struct mlx5_rte_flow_item_tag *tag_item;
4195 	struct rte_flow_item *item;
4196 	char *addr;
4197 	int encap = 0;
4198 
4199 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4200 		switch (actions->type) {
4201 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4202 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4203 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4204 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4205 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4206 			rte_memcpy(actions_tx, actions,
4207 			       sizeof(struct rte_flow_action));
4208 			actions_tx++;
4209 			break;
4210 		case RTE_FLOW_ACTION_TYPE_COUNT:
4211 			if (encap) {
4212 				rte_memcpy(actions_tx, actions,
4213 					   sizeof(struct rte_flow_action));
4214 				actions_tx++;
4215 			} else {
4216 				rte_memcpy(actions_rx, actions,
4217 					   sizeof(struct rte_flow_action));
4218 				actions_rx++;
4219 			}
4220 			break;
4221 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4222 			raw_encap = actions->conf;
4223 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4224 				memcpy(actions_tx, actions,
4225 				       sizeof(struct rte_flow_action));
4226 				actions_tx++;
4227 				encap = 1;
4228 			} else {
4229 				rte_memcpy(actions_rx, actions,
4230 					   sizeof(struct rte_flow_action));
4231 				actions_rx++;
4232 			}
4233 			break;
4234 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4235 			raw_decap = actions->conf;
4236 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4237 				memcpy(actions_tx, actions,
4238 				       sizeof(struct rte_flow_action));
4239 				actions_tx++;
4240 			} else {
4241 				rte_memcpy(actions_rx, actions,
4242 					   sizeof(struct rte_flow_action));
4243 				actions_rx++;
4244 			}
4245 			break;
4246 		default:
4247 			rte_memcpy(actions_rx, actions,
4248 				   sizeof(struct rte_flow_action));
4249 			actions_rx++;
4250 			break;
4251 		}
4252 	}
4253 	/* Add set meta action and end action for the Rx flow. */
4254 	tag_action = actions_rx;
4255 	tag_action->type = (enum rte_flow_action_type)
4256 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4257 	actions_rx++;
4258 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4259 	actions_rx++;
4260 	set_tag = (void *)actions_rx;
4261 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4262 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4263 		.data = flow_id,
4264 	};
4265 	MLX5_ASSERT(set_tag->id > REG_NON);
4266 	tag_action->conf = set_tag;
4267 	/* Create Tx item list. */
4268 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4269 	addr = (void *)&pattern_tx[2];
4270 	item = pattern_tx;
4271 	item->type = (enum rte_flow_item_type)
4272 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4273 	tag_item = (void *)addr;
4274 	tag_item->data = flow_id;
4275 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4276 	MLX5_ASSERT(set_tag->id > REG_NON);
4277 	item->spec = tag_item;
4278 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4279 	tag_item = (void *)addr;
4280 	tag_item->data = UINT32_MAX;
4281 	tag_item->id = UINT16_MAX;
4282 	item->mask = tag_item;
4283 	item->last = NULL;
4284 	item++;
4285 	item->type = RTE_FLOW_ITEM_TYPE_END;
4286 	return 0;
4287 }
4288 
4289 /**
4290  * The last stage of splitting chain, just creates the subflow
4291  * without any modification.
4292  *
4293  * @param[in] dev
4294  *   Pointer to Ethernet device.
4295  * @param[in] flow
4296  *   Parent flow structure pointer.
4297  * @param[in, out] sub_flow
4298  *   Pointer to return the created subflow, may be NULL.
4299  * @param[in] attr
4300  *   Flow rule attributes.
4301  * @param[in] items
4302  *   Pattern specification (list terminated by the END pattern item).
4303  * @param[in] actions
4304  *   Associated actions (list terminated by the END action).
4305  * @param[in] flow_split_info
4306  *   Pointer to flow split info structure.
4307  * @param[out] error
4308  *   Perform verbose error reporting if not NULL.
4309  * @return
4310  *   0 on success, negative value otherwise
4311  */
4312 static int
4313 flow_create_split_inner(struct rte_eth_dev *dev,
4314 			struct rte_flow *flow,
4315 			struct mlx5_flow **sub_flow,
4316 			const struct rte_flow_attr *attr,
4317 			const struct rte_flow_item items[],
4318 			const struct rte_flow_action actions[],
4319 			struct mlx5_flow_split_info *flow_split_info,
4320 			struct rte_flow_error *error)
4321 {
4322 	struct mlx5_flow *dev_flow;
4323 
4324 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4325 				    flow_split_info->flow_idx, error);
4326 	if (!dev_flow)
4327 		return -rte_errno;
4328 	dev_flow->flow = flow;
4329 	dev_flow->external = flow_split_info->external;
4330 	dev_flow->skip_scale = flow_split_info->skip_scale;
4331 	/* Subflow object was created, we must include one in the list. */
4332 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4333 		      dev_flow->handle, next);
4334 	/*
4335 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4336 	 * flow may need some user defined item layer flags, and pass the
4337 	 * Metadate rxq mark flag to suffix flow as well.
4338 	 */
4339 	if (flow_split_info->prefix_layers)
4340 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4341 	if (flow_split_info->prefix_mark)
4342 		dev_flow->handle->mark = 1;
4343 	if (sub_flow)
4344 		*sub_flow = dev_flow;
4345 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4346 	dev_flow->dv.table_id = flow_split_info->table_id;
4347 #endif
4348 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4349 }
4350 
4351 /**
4352  * Split the meter flow.
4353  *
4354  * As meter flow will split to three sub flow, other than meter
4355  * action, the other actions make sense to only meter accepts
4356  * the packet. If it need to be dropped, no other additional
4357  * actions should be take.
4358  *
4359  * One kind of special action which decapsulates the L3 tunnel
4360  * header will be in the prefix sub flow, as not to take the
4361  * L3 tunnel header into account.
4362  *
4363  * @param[in] dev
4364  *   Pointer to Ethernet device.
4365  * @param[in] flow
4366  *   Parent flow structure pointer.
4367  * @param[in] fm
4368  *   Pointer to flow meter structure.
4369  * @param[in] attr
4370  *   Flow rule attributes.
4371  * @param[in] items
4372  *   Pattern specification (list terminated by the END pattern item).
4373  * @param[out] sfx_items
4374  *   Suffix flow match items (list terminated by the END pattern item).
4375  * @param[in] actions
4376  *   Associated actions (list terminated by the END action).
4377  * @param[out] actions_sfx
4378  *   Suffix flow actions.
4379  * @param[out] actions_pre
4380  *   Prefix flow actions.
4381  * @param[out] error
4382  *   Perform verbose error reporting if not NULL.
4383  *
4384  * @return
4385  *   The flow id, 0 otherwise and rte_errno is set.
4386  */
4387 static uint32_t
4388 flow_meter_split_prep(struct rte_eth_dev *dev,
4389 		      struct rte_flow *flow,
4390 		      struct mlx5_flow_meter_info *fm,
4391 		      const struct rte_flow_attr *attr,
4392 		      const struct rte_flow_item items[],
4393 		      struct rte_flow_item sfx_items[],
4394 		      const struct rte_flow_action actions[],
4395 		      struct rte_flow_action actions_sfx[],
4396 		      struct rte_flow_action actions_pre[],
4397 		      struct rte_flow_error *error)
4398 {
4399 	struct mlx5_priv *priv = dev->data->dev_private;
4400 	struct rte_flow_action *tag_action = NULL;
4401 	struct rte_flow_item *tag_item;
4402 	struct mlx5_rte_flow_action_set_tag *set_tag;
4403 	const struct rte_flow_action_raw_encap *raw_encap;
4404 	const struct rte_flow_action_raw_decap *raw_decap;
4405 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4406 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4407 	uint32_t tag_id = 0;
4408 	bool copy_vlan = false;
4409 	struct rte_flow_action *hw_mtr_action;
4410 	struct rte_flow_action_jump *jump_data;
4411 	struct rte_flow_action *action_pre_head = NULL;
4412 	bool mtr_first = priv->sh->meter_aso_en &&
4413 			(attr->egress ||
4414 			(attr->transfer && priv->representor_id != UINT16_MAX));
4415 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4416 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4417 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4418 	uint32_t flow_id = 0;
4419 	uint32_t flow_id_reversed = 0;
4420 	uint8_t flow_id_bits = 0;
4421 	int shift;
4422 
4423 	/* For ASO meter, meter must be before tag in TX direction. */
4424 	if (mtr_first) {
4425 		action_pre_head = actions_pre++;
4426 		/* Leave space for tag action. */
4427 		tag_action = actions_pre++;
4428 	}
4429 	/* Prepare the actions for prefix and suffix flow. */
4430 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4431 		struct rte_flow_action *action_cur = NULL;
4432 
4433 		switch (actions->type) {
4434 		case RTE_FLOW_ACTION_TYPE_METER:
4435 			if (mtr_first) {
4436 				action_cur = action_pre_head;
4437 			} else {
4438 				/* Leave space for tag action. */
4439 				tag_action = actions_pre++;
4440 				action_cur = actions_pre++;
4441 			}
4442 			break;
4443 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4444 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4445 			action_cur = actions_pre++;
4446 			break;
4447 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4448 			raw_encap = actions->conf;
4449 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4450 				action_cur = actions_pre++;
4451 			break;
4452 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4453 			raw_decap = actions->conf;
4454 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4455 				action_cur = actions_pre++;
4456 			break;
4457 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4458 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4459 			copy_vlan = true;
4460 			break;
4461 		default:
4462 			break;
4463 		}
4464 		if (!action_cur)
4465 			action_cur = actions_sfx++;
4466 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4467 	}
4468 	/* Add end action to the actions. */
4469 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4470 	if (priv->sh->meter_aso_en) {
4471 		/**
4472 		 * For ASO meter, need to add an extra jump action explicitly,
4473 		 * to jump from meter to policer table.
4474 		 */
4475 		hw_mtr_action = actions_pre;
4476 		hw_mtr_action->type = RTE_FLOW_ACTION_TYPE_JUMP;
4477 		actions_pre++;
4478 		actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4479 		actions_pre++;
4480 		jump_data = (struct rte_flow_action_jump *)actions_pre;
4481 		jump_data->group = attr->transfer ?
4482 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
4483 				 MLX5_FLOW_TABLE_LEVEL_METER;
4484 		hw_mtr_action->conf = jump_data;
4485 		actions_pre = (struct rte_flow_action *)(jump_data + 1);
4486 	} else {
4487 		actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4488 		actions_pre++;
4489 	}
4490 	/* Generate meter flow_id only if support multiple flows per meter. */
4491 	mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4492 	if (!tag_id)
4493 		return rte_flow_error_set(error, ENOMEM,
4494 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4495 				"Failed to allocate meter flow id.");
4496 	flow_id = tag_id - 1;
4497 	flow_id_bits = MLX5_REG_BITS - __builtin_clz(flow_id);
4498 	flow_id_bits = flow_id_bits ? flow_id_bits : 1;
4499 	if ((flow_id_bits + priv->max_mtr_bits) > mtr_reg_bits) {
4500 		mlx5_ipool_free(fm->flow_ipool, tag_id);
4501 		return rte_flow_error_set(error, EINVAL,
4502 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4503 				"Meter flow id exceeds max limit.");
4504 	}
4505 	if (flow_id_bits > priv->max_mtr_flow_bits)
4506 		priv->max_mtr_flow_bits = flow_id_bits;
4507 	/* Prepare the suffix subflow items. */
4508 	tag_item = sfx_items++;
4509 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4510 		int item_type = items->type;
4511 
4512 		switch (item_type) {
4513 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4514 			memcpy(sfx_items, items, sizeof(*sfx_items));
4515 			sfx_items++;
4516 			break;
4517 		case RTE_FLOW_ITEM_TYPE_VLAN:
4518 			if (copy_vlan) {
4519 				memcpy(sfx_items, items, sizeof(*sfx_items));
4520 				/*
4521 				 * Convert to internal match item, it is used
4522 				 * for vlan push and set vid.
4523 				 */
4524 				sfx_items->type = (enum rte_flow_item_type)
4525 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4526 				sfx_items++;
4527 			}
4528 			break;
4529 		default:
4530 			break;
4531 		}
4532 	}
4533 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4534 	sfx_items++;
4535 	/* Build tag actions and items for meter_id/meter flow_id. */
4536 	assert(tag_action);
4537 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4538 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4539 	tag_item_mask = tag_item_spec + 1;
4540 	/* Both flow_id and meter_id share the same register. */
4541 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4542 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4543 							    0, error),
4544 		.offset = mtr_id_offset,
4545 		.length = mtr_reg_bits,
4546 		.data = flow->meter,
4547 	};
4548 	/*
4549 	 * The color Reg bits used by flow_id are growing from
4550 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4551 	 */
4552 	for (shift = 0; shift < flow_id_bits; shift++)
4553 		flow_id_reversed = (flow_id_reversed << 1) |
4554 			      ((flow_id >> shift) & 0x1);
4555 	set_tag->data |= flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4556 	tag_item_spec->id = set_tag->id;
4557 	tag_item_spec->data = set_tag->data << mtr_id_offset;
4558 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4559 	tag_action->type = (enum rte_flow_action_type)
4560 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4561 	tag_action->conf = set_tag;
4562 	tag_item->type = (enum rte_flow_item_type)
4563 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4564 	tag_item->spec = tag_item_spec;
4565 	tag_item->last = NULL;
4566 	tag_item->mask = tag_item_mask;
4567 	return tag_id;
4568 }
4569 
4570 /**
4571  * Split action list having QUEUE/RSS for metadata register copy.
4572  *
4573  * Once Q/RSS action is detected in user's action list, the flow action
4574  * should be split in order to copy metadata registers, which will happen in
4575  * RX_CP_TBL like,
4576  *   - CQE->flow_tag := reg_c[1] (MARK)
4577  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4578  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4579  * This is because the last action of each flow must be a terminal action
4580  * (QUEUE, RSS or DROP).
4581  *
4582  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4583  * stored and kept in the mlx5_flow structure per each sub_flow.
4584  *
4585  * The Q/RSS action is replaced with,
4586  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4587  * And the following JUMP action is added at the end,
4588  *   - JUMP, to RX_CP_TBL.
4589  *
4590  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4591  * flow_create_split_metadata() routine. The flow will look like,
4592  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4593  *
4594  * @param dev
4595  *   Pointer to Ethernet device.
4596  * @param[out] split_actions
4597  *   Pointer to store split actions to jump to CP_TBL.
4598  * @param[in] actions
4599  *   Pointer to the list of original flow actions.
4600  * @param[in] qrss
4601  *   Pointer to the Q/RSS action.
4602  * @param[in] actions_n
4603  *   Number of original actions.
4604  * @param[out] error
4605  *   Perform verbose error reporting if not NULL.
4606  *
4607  * @return
4608  *   non-zero unique flow_id on success, otherwise 0 and
4609  *   error/rte_error are set.
4610  */
4611 static uint32_t
4612 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4613 			  struct rte_flow_action *split_actions,
4614 			  const struct rte_flow_action *actions,
4615 			  const struct rte_flow_action *qrss,
4616 			  int actions_n, struct rte_flow_error *error)
4617 {
4618 	struct mlx5_priv *priv = dev->data->dev_private;
4619 	struct mlx5_rte_flow_action_set_tag *set_tag;
4620 	struct rte_flow_action_jump *jump;
4621 	const int qrss_idx = qrss - actions;
4622 	uint32_t flow_id = 0;
4623 	int ret = 0;
4624 
4625 	/*
4626 	 * Given actions will be split
4627 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4628 	 * - Add jump to mreg CP_TBL.
4629 	 * As a result, there will be one more action.
4630 	 */
4631 	++actions_n;
4632 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4633 	set_tag = (void *)(split_actions + actions_n);
4634 	/*
4635 	 * If tag action is not set to void(it means we are not the meter
4636 	 * suffix flow), add the tag action. Since meter suffix flow already
4637 	 * has the tag added.
4638 	 */
4639 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4640 		/*
4641 		 * Allocate the new subflow ID. This one is unique within
4642 		 * device and not shared with representors. Otherwise,
4643 		 * we would have to resolve multi-thread access synch
4644 		 * issue. Each flow on the shared device is appended
4645 		 * with source vport identifier, so the resulting
4646 		 * flows will be unique in the shared (by master and
4647 		 * representors) domain even if they have coinciding
4648 		 * IDs.
4649 		 */
4650 		mlx5_ipool_malloc(priv->sh->ipool
4651 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4652 		if (!flow_id)
4653 			return rte_flow_error_set(error, ENOMEM,
4654 						  RTE_FLOW_ERROR_TYPE_ACTION,
4655 						  NULL, "can't allocate id "
4656 						  "for split Q/RSS subflow");
4657 		/* Internal SET_TAG action to set flow ID. */
4658 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4659 			.data = flow_id,
4660 		};
4661 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4662 		if (ret < 0)
4663 			return ret;
4664 		set_tag->id = ret;
4665 		/* Construct new actions array. */
4666 		/* Replace QUEUE/RSS action. */
4667 		split_actions[qrss_idx] = (struct rte_flow_action){
4668 			.type = (enum rte_flow_action_type)
4669 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4670 			.conf = set_tag,
4671 		};
4672 	}
4673 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4674 	jump = (void *)(set_tag + 1);
4675 	*jump = (struct rte_flow_action_jump){
4676 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4677 	};
4678 	split_actions[actions_n - 2] = (struct rte_flow_action){
4679 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4680 		.conf = jump,
4681 	};
4682 	split_actions[actions_n - 1] = (struct rte_flow_action){
4683 		.type = RTE_FLOW_ACTION_TYPE_END,
4684 	};
4685 	return flow_id;
4686 }
4687 
4688 /**
4689  * Extend the given action list for Tx metadata copy.
4690  *
4691  * Copy the given action list to the ext_actions and add flow metadata register
4692  * copy action in order to copy reg_a set by WQE to reg_c[0].
4693  *
4694  * @param[out] ext_actions
4695  *   Pointer to the extended action list.
4696  * @param[in] actions
4697  *   Pointer to the list of actions.
4698  * @param[in] actions_n
4699  *   Number of actions in the list.
4700  * @param[out] error
4701  *   Perform verbose error reporting if not NULL.
4702  * @param[in] encap_idx
4703  *   The encap action inndex.
4704  *
4705  * @return
4706  *   0 on success, negative value otherwise
4707  */
4708 static int
4709 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4710 		       struct rte_flow_action *ext_actions,
4711 		       const struct rte_flow_action *actions,
4712 		       int actions_n, struct rte_flow_error *error,
4713 		       int encap_idx)
4714 {
4715 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4716 		(struct mlx5_flow_action_copy_mreg *)
4717 			(ext_actions + actions_n + 1);
4718 	int ret;
4719 
4720 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4721 	if (ret < 0)
4722 		return ret;
4723 	cp_mreg->dst = ret;
4724 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4725 	if (ret < 0)
4726 		return ret;
4727 	cp_mreg->src = ret;
4728 	if (encap_idx != 0)
4729 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4730 	if (encap_idx == actions_n - 1) {
4731 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4732 			.type = (enum rte_flow_action_type)
4733 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4734 			.conf = cp_mreg,
4735 		};
4736 		ext_actions[actions_n] = (struct rte_flow_action){
4737 			.type = RTE_FLOW_ACTION_TYPE_END,
4738 		};
4739 	} else {
4740 		ext_actions[encap_idx] = (struct rte_flow_action){
4741 			.type = (enum rte_flow_action_type)
4742 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4743 			.conf = cp_mreg,
4744 		};
4745 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4746 				sizeof(*ext_actions) * (actions_n - encap_idx));
4747 	}
4748 	return 0;
4749 }
4750 
4751 /**
4752  * Check the match action from the action list.
4753  *
4754  * @param[in] actions
4755  *   Pointer to the list of actions.
4756  * @param[in] attr
4757  *   Flow rule attributes.
4758  * @param[in] action
4759  *   The action to be check if exist.
4760  * @param[out] match_action_pos
4761  *   Pointer to the position of the matched action if exists, otherwise is -1.
4762  * @param[out] qrss_action_pos
4763  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4764  * @param[out] modify_after_mirror
4765  *   Pointer to the flag of modify action after FDB mirroring.
4766  *
4767  * @return
4768  *   > 0 the total number of actions.
4769  *   0 if not found match action in action list.
4770  */
4771 static int
4772 flow_check_match_action(const struct rte_flow_action actions[],
4773 			const struct rte_flow_attr *attr,
4774 			enum rte_flow_action_type action,
4775 			int *match_action_pos, int *qrss_action_pos,
4776 			int *modify_after_mirror)
4777 {
4778 	const struct rte_flow_action_sample *sample;
4779 	int actions_n = 0;
4780 	uint32_t ratio = 0;
4781 	int sub_type = 0;
4782 	int flag = 0;
4783 	int fdb_mirror = 0;
4784 
4785 	*match_action_pos = -1;
4786 	*qrss_action_pos = -1;
4787 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4788 		if (actions->type == action) {
4789 			flag = 1;
4790 			*match_action_pos = actions_n;
4791 		}
4792 		switch (actions->type) {
4793 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4794 		case RTE_FLOW_ACTION_TYPE_RSS:
4795 			*qrss_action_pos = actions_n;
4796 			break;
4797 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
4798 			sample = actions->conf;
4799 			ratio = sample->ratio;
4800 			sub_type = ((const struct rte_flow_action *)
4801 					(sample->actions))->type;
4802 			if (ratio == 1 && attr->transfer)
4803 				fdb_mirror = 1;
4804 			break;
4805 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4806 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4807 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4808 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4809 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4810 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4811 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4812 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4813 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4814 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
4815 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4816 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4817 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4818 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4819 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4820 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4821 		case RTE_FLOW_ACTION_TYPE_FLAG:
4822 		case RTE_FLOW_ACTION_TYPE_MARK:
4823 		case RTE_FLOW_ACTION_TYPE_SET_META:
4824 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
4825 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4826 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4827 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4828 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4829 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4830 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4831 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4832 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4833 			if (fdb_mirror)
4834 				*modify_after_mirror = 1;
4835 			break;
4836 		default:
4837 			break;
4838 		}
4839 		actions_n++;
4840 	}
4841 	if (flag && fdb_mirror && !*modify_after_mirror) {
4842 		/* FDB mirroring uses the destination array to implement
4843 		 * instead of FLOW_SAMPLER object.
4844 		 */
4845 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4846 			flag = 0;
4847 	}
4848 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4849 	return flag ? actions_n + 1 : 0;
4850 }
4851 
4852 #define SAMPLE_SUFFIX_ITEM 2
4853 
4854 /**
4855  * Split the sample flow.
4856  *
4857  * As sample flow will split to two sub flow, sample flow with
4858  * sample action, the other actions will move to new suffix flow.
4859  *
4860  * Also add unique tag id with tag action in the sample flow,
4861  * the same tag id will be as match in the suffix flow.
4862  *
4863  * @param dev
4864  *   Pointer to Ethernet device.
4865  * @param[in] add_tag
4866  *   Add extra tag action flag.
4867  * @param[out] sfx_items
4868  *   Suffix flow match items (list terminated by the END pattern item).
4869  * @param[in] actions
4870  *   Associated actions (list terminated by the END action).
4871  * @param[out] actions_sfx
4872  *   Suffix flow actions.
4873  * @param[out] actions_pre
4874  *   Prefix flow actions.
4875  * @param[in] actions_n
4876  *  The total number of actions.
4877  * @param[in] sample_action_pos
4878  *   The sample action position.
4879  * @param[in] qrss_action_pos
4880  *   The Queue/RSS action position.
4881  * @param[in] jump_table
4882  *   Add extra jump action flag.
4883  * @param[out] error
4884  *   Perform verbose error reporting if not NULL.
4885  *
4886  * @return
4887  *   0 on success, or unique flow_id, a negative errno value
4888  *   otherwise and rte_errno is set.
4889  */
4890 static int
4891 flow_sample_split_prep(struct rte_eth_dev *dev,
4892 		       int add_tag,
4893 		       struct rte_flow_item sfx_items[],
4894 		       const struct rte_flow_action actions[],
4895 		       struct rte_flow_action actions_sfx[],
4896 		       struct rte_flow_action actions_pre[],
4897 		       int actions_n,
4898 		       int sample_action_pos,
4899 		       int qrss_action_pos,
4900 		       int jump_table,
4901 		       struct rte_flow_error *error)
4902 {
4903 	struct mlx5_priv *priv = dev->data->dev_private;
4904 	struct mlx5_rte_flow_action_set_tag *set_tag;
4905 	struct mlx5_rte_flow_item_tag *tag_spec;
4906 	struct mlx5_rte_flow_item_tag *tag_mask;
4907 	struct rte_flow_action_jump *jump_action;
4908 	uint32_t tag_id = 0;
4909 	int index;
4910 	int append_index = 0;
4911 	int ret;
4912 
4913 	if (sample_action_pos < 0)
4914 		return rte_flow_error_set(error, EINVAL,
4915 					  RTE_FLOW_ERROR_TYPE_ACTION,
4916 					  NULL, "invalid position of sample "
4917 					  "action in list");
4918 	/* Prepare the actions for prefix and suffix flow. */
4919 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4920 		index = qrss_action_pos;
4921 		/* Put the preceding the Queue/RSS action into prefix flow. */
4922 		if (index != 0)
4923 			memcpy(actions_pre, actions,
4924 			       sizeof(struct rte_flow_action) * index);
4925 		/* Put others preceding the sample action into prefix flow. */
4926 		if (sample_action_pos > index + 1)
4927 			memcpy(actions_pre + index, actions + index + 1,
4928 			       sizeof(struct rte_flow_action) *
4929 			       (sample_action_pos - index - 1));
4930 		index = sample_action_pos - 1;
4931 		/* Put Queue/RSS action into Suffix flow. */
4932 		memcpy(actions_sfx, actions + qrss_action_pos,
4933 		       sizeof(struct rte_flow_action));
4934 		actions_sfx++;
4935 	} else {
4936 		index = sample_action_pos;
4937 		if (index != 0)
4938 			memcpy(actions_pre, actions,
4939 			       sizeof(struct rte_flow_action) * index);
4940 	}
4941 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4942 	 * For CX6DX and above, metadata registers Cx preserve their value,
4943 	 * add an extra tag action for NIC-RX and E-Switch Domain.
4944 	 */
4945 	if (add_tag) {
4946 		/* Prepare the prefix tag action. */
4947 		append_index++;
4948 		set_tag = (void *)(actions_pre + actions_n + append_index);
4949 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4950 		if (ret < 0)
4951 			return ret;
4952 		mlx5_ipool_malloc(priv->sh->ipool
4953 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4954 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4955 			.id = ret,
4956 			.data = tag_id,
4957 		};
4958 		/* Prepare the suffix subflow items. */
4959 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4960 		tag_spec->data = tag_id;
4961 		tag_spec->id = set_tag->id;
4962 		tag_mask = tag_spec + 1;
4963 		tag_mask->data = UINT32_MAX;
4964 		sfx_items[0] = (struct rte_flow_item){
4965 			.type = (enum rte_flow_item_type)
4966 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4967 			.spec = tag_spec,
4968 			.last = NULL,
4969 			.mask = tag_mask,
4970 		};
4971 		sfx_items[1] = (struct rte_flow_item){
4972 			.type = (enum rte_flow_item_type)
4973 				RTE_FLOW_ITEM_TYPE_END,
4974 		};
4975 		/* Prepare the tag action in prefix subflow. */
4976 		actions_pre[index++] =
4977 			(struct rte_flow_action){
4978 			.type = (enum rte_flow_action_type)
4979 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4980 			.conf = set_tag,
4981 		};
4982 	}
4983 	memcpy(actions_pre + index, actions + sample_action_pos,
4984 	       sizeof(struct rte_flow_action));
4985 	index += 1;
4986 	/* For the modify action after the sample action in E-Switch mirroring,
4987 	 * Add the extra jump action in prefix subflow and jump into the next
4988 	 * table, then do the modify action in the new table.
4989 	 */
4990 	if (jump_table) {
4991 		/* Prepare the prefix jump action. */
4992 		append_index++;
4993 		jump_action = (void *)(actions_pre + actions_n + append_index);
4994 		jump_action->group = jump_table;
4995 		actions_pre[index++] =
4996 			(struct rte_flow_action){
4997 			.type = (enum rte_flow_action_type)
4998 				RTE_FLOW_ACTION_TYPE_JUMP,
4999 			.conf = jump_action,
5000 		};
5001 	}
5002 	actions_pre[index] = (struct rte_flow_action){
5003 		.type = (enum rte_flow_action_type)
5004 			RTE_FLOW_ACTION_TYPE_END,
5005 	};
5006 	/* Put the actions after sample into Suffix flow. */
5007 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5008 	       sizeof(struct rte_flow_action) *
5009 	       (actions_n - sample_action_pos - 1));
5010 	return tag_id;
5011 }
5012 
5013 /**
5014  * The splitting for metadata feature.
5015  *
5016  * - Q/RSS action on NIC Rx should be split in order to pass by
5017  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5018  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5019  *
5020  * - All the actions on NIC Tx should have a mreg copy action to
5021  *   copy reg_a from WQE to reg_c[0].
5022  *
5023  * @param dev
5024  *   Pointer to Ethernet device.
5025  * @param[in] flow
5026  *   Parent flow structure pointer.
5027  * @param[in] attr
5028  *   Flow rule attributes.
5029  * @param[in] items
5030  *   Pattern specification (list terminated by the END pattern item).
5031  * @param[in] actions
5032  *   Associated actions (list terminated by the END action).
5033  * @param[in] flow_split_info
5034  *   Pointer to flow split info structure.
5035  * @param[out] error
5036  *   Perform verbose error reporting if not NULL.
5037  * @return
5038  *   0 on success, negative value otherwise
5039  */
5040 static int
5041 flow_create_split_metadata(struct rte_eth_dev *dev,
5042 			   struct rte_flow *flow,
5043 			   const struct rte_flow_attr *attr,
5044 			   const struct rte_flow_item items[],
5045 			   const struct rte_flow_action actions[],
5046 			   struct mlx5_flow_split_info *flow_split_info,
5047 			   struct rte_flow_error *error)
5048 {
5049 	struct mlx5_priv *priv = dev->data->dev_private;
5050 	struct mlx5_dev_config *config = &priv->config;
5051 	const struct rte_flow_action *qrss = NULL;
5052 	struct rte_flow_action *ext_actions = NULL;
5053 	struct mlx5_flow *dev_flow = NULL;
5054 	uint32_t qrss_id = 0;
5055 	int mtr_sfx = 0;
5056 	size_t act_size;
5057 	int actions_n;
5058 	int encap_idx;
5059 	int ret;
5060 
5061 	/* Check whether extensive metadata feature is engaged. */
5062 	if (!config->dv_flow_en ||
5063 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5064 	    !mlx5_flow_ext_mreg_supported(dev))
5065 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5066 					       actions, flow_split_info, error);
5067 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5068 							   &encap_idx);
5069 	if (qrss) {
5070 		/* Exclude hairpin flows from splitting. */
5071 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5072 			const struct rte_flow_action_queue *queue;
5073 
5074 			queue = qrss->conf;
5075 			if (mlx5_rxq_get_type(dev, queue->index) ==
5076 			    MLX5_RXQ_TYPE_HAIRPIN)
5077 				qrss = NULL;
5078 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5079 			const struct rte_flow_action_rss *rss;
5080 
5081 			rss = qrss->conf;
5082 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5083 			    MLX5_RXQ_TYPE_HAIRPIN)
5084 				qrss = NULL;
5085 		}
5086 	}
5087 	if (qrss) {
5088 		/* Check if it is in meter suffix table. */
5089 		mtr_sfx = attr->group == (attr->transfer ?
5090 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5091 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5092 		/*
5093 		 * Q/RSS action on NIC Rx should be split in order to pass by
5094 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5095 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5096 		 */
5097 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5098 			   sizeof(struct rte_flow_action_set_tag) +
5099 			   sizeof(struct rte_flow_action_jump);
5100 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5101 					  SOCKET_ID_ANY);
5102 		if (!ext_actions)
5103 			return rte_flow_error_set(error, ENOMEM,
5104 						  RTE_FLOW_ERROR_TYPE_ACTION,
5105 						  NULL, "no memory to split "
5106 						  "metadata flow");
5107 		/*
5108 		 * If we are the suffix flow of meter, tag already exist.
5109 		 * Set the tag action to void.
5110 		 */
5111 		if (mtr_sfx)
5112 			ext_actions[qrss - actions].type =
5113 						RTE_FLOW_ACTION_TYPE_VOID;
5114 		else
5115 			ext_actions[qrss - actions].type =
5116 						(enum rte_flow_action_type)
5117 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5118 		/*
5119 		 * Create the new actions list with removed Q/RSS action
5120 		 * and appended set tag and jump to register copy table
5121 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5122 		 * in advance, because it is needed for set tag action.
5123 		 */
5124 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5125 						    qrss, actions_n, error);
5126 		if (!mtr_sfx && !qrss_id) {
5127 			ret = -rte_errno;
5128 			goto exit;
5129 		}
5130 	} else if (attr->egress && !attr->transfer) {
5131 		/*
5132 		 * All the actions on NIC Tx should have a metadata register
5133 		 * copy action to copy reg_a from WQE to reg_c[meta]
5134 		 */
5135 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5136 			   sizeof(struct mlx5_flow_action_copy_mreg);
5137 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5138 					  SOCKET_ID_ANY);
5139 		if (!ext_actions)
5140 			return rte_flow_error_set(error, ENOMEM,
5141 						  RTE_FLOW_ERROR_TYPE_ACTION,
5142 						  NULL, "no memory to split "
5143 						  "metadata flow");
5144 		/* Create the action list appended with copy register. */
5145 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5146 					     actions_n, error, encap_idx);
5147 		if (ret < 0)
5148 			goto exit;
5149 	}
5150 	/* Add the unmodified original or prefix subflow. */
5151 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5152 				      items, ext_actions ? ext_actions :
5153 				      actions, flow_split_info, error);
5154 	if (ret < 0)
5155 		goto exit;
5156 	MLX5_ASSERT(dev_flow);
5157 	if (qrss) {
5158 		const struct rte_flow_attr q_attr = {
5159 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5160 			.ingress = 1,
5161 		};
5162 		/* Internal PMD action to set register. */
5163 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5164 			.data = qrss_id,
5165 			.id = REG_NON,
5166 		};
5167 		struct rte_flow_item q_items[] = {
5168 			{
5169 				.type = (enum rte_flow_item_type)
5170 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5171 				.spec = &q_tag_spec,
5172 				.last = NULL,
5173 				.mask = NULL,
5174 			},
5175 			{
5176 				.type = RTE_FLOW_ITEM_TYPE_END,
5177 			},
5178 		};
5179 		struct rte_flow_action q_actions[] = {
5180 			{
5181 				.type = qrss->type,
5182 				.conf = qrss->conf,
5183 			},
5184 			{
5185 				.type = RTE_FLOW_ACTION_TYPE_END,
5186 			},
5187 		};
5188 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5189 
5190 		/*
5191 		 * Configure the tag item only if there is no meter subflow.
5192 		 * Since tag is already marked in the meter suffix subflow
5193 		 * we can just use the meter suffix items as is.
5194 		 */
5195 		if (qrss_id) {
5196 			/* Not meter subflow. */
5197 			MLX5_ASSERT(!mtr_sfx);
5198 			/*
5199 			 * Put unique id in prefix flow due to it is destroyed
5200 			 * after suffix flow and id will be freed after there
5201 			 * is no actual flows with this id and identifier
5202 			 * reallocation becomes possible (for example, for
5203 			 * other flows in other threads).
5204 			 */
5205 			dev_flow->handle->split_flow_id = qrss_id;
5206 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5207 						   error);
5208 			if (ret < 0)
5209 				goto exit;
5210 			q_tag_spec.id = ret;
5211 		}
5212 		dev_flow = NULL;
5213 		/* Add suffix subflow to execute Q/RSS. */
5214 		flow_split_info->prefix_layers = layers;
5215 		flow_split_info->prefix_mark = 0;
5216 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5217 					      &q_attr, mtr_sfx ? items :
5218 					      q_items, q_actions,
5219 					      flow_split_info, error);
5220 		if (ret < 0)
5221 			goto exit;
5222 		/* qrss ID should be freed if failed. */
5223 		qrss_id = 0;
5224 		MLX5_ASSERT(dev_flow);
5225 	}
5226 
5227 exit:
5228 	/*
5229 	 * We do not destroy the partially created sub_flows in case of error.
5230 	 * These ones are included into parent flow list and will be destroyed
5231 	 * by flow_drv_destroy.
5232 	 */
5233 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5234 			qrss_id);
5235 	mlx5_free(ext_actions);
5236 	return ret;
5237 }
5238 
5239 /**
5240  * The splitting for meter feature.
5241  *
5242  * - The meter flow will be split to two flows as prefix and
5243  *   suffix flow. The packets make sense only it pass the prefix
5244  *   meter action.
5245  *
5246  * - Reg_C_5 is used for the packet to match betweend prefix and
5247  *   suffix flow.
5248  *
5249  * @param dev
5250  *   Pointer to Ethernet device.
5251  * @param[in] flow
5252  *   Parent flow structure pointer.
5253  * @param[in] attr
5254  *   Flow rule attributes.
5255  * @param[in] items
5256  *   Pattern specification (list terminated by the END pattern item).
5257  * @param[in] actions
5258  *   Associated actions (list terminated by the END action).
5259  * @param[in] flow_split_info
5260  *   Pointer to flow split info structure.
5261  * @param[out] error
5262  *   Perform verbose error reporting if not NULL.
5263  * @return
5264  *   0 on success, negative value otherwise
5265  */
5266 static int
5267 flow_create_split_meter(struct rte_eth_dev *dev,
5268 			struct rte_flow *flow,
5269 			const struct rte_flow_attr *attr,
5270 			const struct rte_flow_item items[],
5271 			const struct rte_flow_action actions[],
5272 			struct mlx5_flow_split_info *flow_split_info,
5273 			struct rte_flow_error *error)
5274 {
5275 	struct mlx5_priv *priv = dev->data->dev_private;
5276 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5277 	struct rte_flow_action *sfx_actions = NULL;
5278 	struct rte_flow_action *pre_actions = NULL;
5279 	struct rte_flow_item *sfx_items = NULL;
5280 	struct mlx5_flow *dev_flow = NULL;
5281 	struct rte_flow_attr sfx_attr = *attr;
5282 	struct mlx5_flow_meter_info *fm = NULL;
5283 	bool has_mtr = false;
5284 	uint32_t meter_id;
5285 	uint32_t mtr_idx = 0;
5286 	uint32_t mtr_tag_id = 0;
5287 	size_t act_size;
5288 	size_t item_size;
5289 	int actions_n = 0;
5290 	int ret = 0;
5291 
5292 	if (priv->mtr_en)
5293 		actions_n = flow_check_meter_action(actions, &has_mtr,
5294 						    &meter_id);
5295 	if (has_mtr) {
5296 		if (flow->meter) {
5297 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5298 			if (!fm)
5299 				return rte_flow_error_set(error, EINVAL,
5300 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5301 						NULL, "Meter not found.");
5302 		} else {
5303 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5304 			if (!fm)
5305 				return rte_flow_error_set(error, EINVAL,
5306 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5307 						NULL, "Meter not found.");
5308 			ret = mlx5_flow_meter_attach(priv, fm,
5309 						     &sfx_attr, error);
5310 			if (ret)
5311 				return -rte_errno;
5312 			flow->meter = mtr_idx;
5313 		}
5314 		wks->fm = fm;
5315 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5316 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5317 			   sizeof(struct mlx5_rte_flow_action_set_tag) +
5318 			   sizeof(struct rte_flow_action_jump);
5319 		/* Suffix items: tag, vlan, port id, end. */
5320 #define METER_SUFFIX_ITEM 4
5321 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5322 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5323 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5324 					  0, SOCKET_ID_ANY);
5325 		if (!sfx_actions)
5326 			return rte_flow_error_set(error, ENOMEM,
5327 						  RTE_FLOW_ERROR_TYPE_ACTION,
5328 						  NULL, "no memory to split "
5329 						  "meter flow");
5330 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5331 			     act_size);
5332 		pre_actions = sfx_actions + actions_n;
5333 		mtr_tag_id = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5334 						   items, sfx_items, actions,
5335 						   sfx_actions, pre_actions,
5336 						   error);
5337 		if (!mtr_tag_id) {
5338 			ret = -rte_errno;
5339 			goto exit;
5340 		}
5341 		/* Add the prefix subflow. */
5342 		flow_split_info->prefix_mark = 0;
5343 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5344 					      attr, items, pre_actions,
5345 					      flow_split_info, error);
5346 		if (ret) {
5347 			mlx5_ipool_free(fm->flow_ipool, mtr_tag_id);
5348 			ret = -rte_errno;
5349 			goto exit;
5350 		}
5351 		dev_flow->handle->split_flow_id = mtr_tag_id;
5352 		dev_flow->handle->is_meter_flow_id = 1;
5353 		/* Setting the sfx group atrr. */
5354 		sfx_attr.group = sfx_attr.transfer ?
5355 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5356 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5357 		flow_split_info->prefix_layers =
5358 				flow_get_prefix_layer_flags(dev_flow);
5359 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5360 	}
5361 	/* Add the prefix subflow. */
5362 	ret = flow_create_split_metadata(dev, flow,
5363 					 &sfx_attr, sfx_items ?
5364 					 sfx_items : items,
5365 					 sfx_actions ? sfx_actions : actions,
5366 					 flow_split_info, error);
5367 exit:
5368 	if (sfx_actions)
5369 		mlx5_free(sfx_actions);
5370 	return ret;
5371 }
5372 
5373 /**
5374  * The splitting for sample feature.
5375  *
5376  * Once Sample action is detected in the action list, the flow actions should
5377  * be split into prefix sub flow and suffix sub flow.
5378  *
5379  * The original items remain in the prefix sub flow, all actions preceding the
5380  * sample action and the sample action itself will be copied to the prefix
5381  * sub flow, the actions following the sample action will be copied to the
5382  * suffix sub flow, Queue action always be located in the suffix sub flow.
5383  *
5384  * In order to make the packet from prefix sub flow matches with suffix sub
5385  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5386  * flow uses tag item with the unique flow id.
5387  *
5388  * @param dev
5389  *   Pointer to Ethernet device.
5390  * @param[in] flow
5391  *   Parent flow structure pointer.
5392  * @param[in] attr
5393  *   Flow rule attributes.
5394  * @param[in] items
5395  *   Pattern specification (list terminated by the END pattern item).
5396  * @param[in] actions
5397  *   Associated actions (list terminated by the END action).
5398  * @param[in] flow_split_info
5399  *   Pointer to flow split info structure.
5400  * @param[out] error
5401  *   Perform verbose error reporting if not NULL.
5402  * @return
5403  *   0 on success, negative value otherwise
5404  */
5405 static int
5406 flow_create_split_sample(struct rte_eth_dev *dev,
5407 			 struct rte_flow *flow,
5408 			 const struct rte_flow_attr *attr,
5409 			 const struct rte_flow_item items[],
5410 			 const struct rte_flow_action actions[],
5411 			 struct mlx5_flow_split_info *flow_split_info,
5412 			 struct rte_flow_error *error)
5413 {
5414 	struct mlx5_priv *priv = dev->data->dev_private;
5415 	struct rte_flow_action *sfx_actions = NULL;
5416 	struct rte_flow_action *pre_actions = NULL;
5417 	struct rte_flow_item *sfx_items = NULL;
5418 	struct mlx5_flow *dev_flow = NULL;
5419 	struct rte_flow_attr sfx_attr = *attr;
5420 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5421 	struct mlx5_flow_dv_sample_resource *sample_res;
5422 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5423 	struct mlx5_flow_tbl_resource *sfx_tbl;
5424 #endif
5425 	size_t act_size;
5426 	size_t item_size;
5427 	uint32_t fdb_tx = 0;
5428 	int32_t tag_id = 0;
5429 	int actions_n = 0;
5430 	int sample_action_pos;
5431 	int qrss_action_pos;
5432 	int add_tag = 0;
5433 	int modify_after_mirror = 0;
5434 	uint16_t jump_table = 0;
5435 	const uint32_t next_ft_step = 1;
5436 	int ret = 0;
5437 
5438 	if (priv->sampler_en)
5439 		actions_n = flow_check_match_action(actions, attr,
5440 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5441 					&sample_action_pos, &qrss_action_pos,
5442 					&modify_after_mirror);
5443 	if (actions_n) {
5444 		/* The prefix actions must includes sample, tag, end. */
5445 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5446 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5447 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5448 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5449 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5450 					  item_size), 0, SOCKET_ID_ANY);
5451 		if (!sfx_actions)
5452 			return rte_flow_error_set(error, ENOMEM,
5453 						  RTE_FLOW_ERROR_TYPE_ACTION,
5454 						  NULL, "no memory to split "
5455 						  "sample flow");
5456 		/* The representor_id is -1 for uplink. */
5457 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5458 		/*
5459 		 * When reg_c_preserve is set, metadata registers Cx preserve
5460 		 * their value even through packet duplication.
5461 		 */
5462 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5463 		if (add_tag)
5464 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5465 					+ act_size);
5466 		if (modify_after_mirror)
5467 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5468 				     next_ft_step;
5469 		pre_actions = sfx_actions + actions_n;
5470 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5471 						actions, sfx_actions,
5472 						pre_actions, actions_n,
5473 						sample_action_pos,
5474 						qrss_action_pos, jump_table,
5475 						error);
5476 		if (tag_id < 0 || (add_tag && !tag_id)) {
5477 			ret = -rte_errno;
5478 			goto exit;
5479 		}
5480 		if (modify_after_mirror)
5481 			flow_split_info->skip_scale =
5482 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5483 		/* Add the prefix subflow. */
5484 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5485 					      items, pre_actions,
5486 					      flow_split_info, error);
5487 		if (ret) {
5488 			ret = -rte_errno;
5489 			goto exit;
5490 		}
5491 		dev_flow->handle->split_flow_id = tag_id;
5492 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5493 		if (!modify_after_mirror) {
5494 			/* Set the sfx group attr. */
5495 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5496 						dev_flow->dv.sample_res;
5497 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5498 						sample_res->normal_path_tbl;
5499 			sfx_tbl_data = container_of(sfx_tbl,
5500 						struct mlx5_flow_tbl_data_entry,
5501 						tbl);
5502 			sfx_attr.group = sfx_attr.transfer ?
5503 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5504 		} else {
5505 			MLX5_ASSERT(attr->transfer);
5506 			sfx_attr.group = jump_table;
5507 		}
5508 		flow_split_info->prefix_layers =
5509 				flow_get_prefix_layer_flags(dev_flow);
5510 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5511 		/* Suffix group level already be scaled with factor, set
5512 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5513 		 * again in translation.
5514 		 */
5515 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5516 #endif
5517 	}
5518 	/* Add the suffix subflow. */
5519 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5520 				      sfx_items ? sfx_items : items,
5521 				      sfx_actions ? sfx_actions : actions,
5522 				      flow_split_info, error);
5523 exit:
5524 	if (sfx_actions)
5525 		mlx5_free(sfx_actions);
5526 	return ret;
5527 }
5528 
5529 /**
5530  * Split the flow to subflow set. The splitters might be linked
5531  * in the chain, like this:
5532  * flow_create_split_outer() calls:
5533  *   flow_create_split_meter() calls:
5534  *     flow_create_split_metadata(meter_subflow_0) calls:
5535  *       flow_create_split_inner(metadata_subflow_0)
5536  *       flow_create_split_inner(metadata_subflow_1)
5537  *       flow_create_split_inner(metadata_subflow_2)
5538  *     flow_create_split_metadata(meter_subflow_1) calls:
5539  *       flow_create_split_inner(metadata_subflow_0)
5540  *       flow_create_split_inner(metadata_subflow_1)
5541  *       flow_create_split_inner(metadata_subflow_2)
5542  *
5543  * This provide flexible way to add new levels of flow splitting.
5544  * The all of successfully created subflows are included to the
5545  * parent flow dev_flow list.
5546  *
5547  * @param dev
5548  *   Pointer to Ethernet device.
5549  * @param[in] flow
5550  *   Parent flow structure pointer.
5551  * @param[in] attr
5552  *   Flow rule attributes.
5553  * @param[in] items
5554  *   Pattern specification (list terminated by the END pattern item).
5555  * @param[in] actions
5556  *   Associated actions (list terminated by the END action).
5557  * @param[in] flow_split_info
5558  *   Pointer to flow split info structure.
5559  * @param[out] error
5560  *   Perform verbose error reporting if not NULL.
5561  * @return
5562  *   0 on success, negative value otherwise
5563  */
5564 static int
5565 flow_create_split_outer(struct rte_eth_dev *dev,
5566 			struct rte_flow *flow,
5567 			const struct rte_flow_attr *attr,
5568 			const struct rte_flow_item items[],
5569 			const struct rte_flow_action actions[],
5570 			struct mlx5_flow_split_info *flow_split_info,
5571 			struct rte_flow_error *error)
5572 {
5573 	int ret;
5574 
5575 	ret = flow_create_split_sample(dev, flow, attr, items,
5576 				       actions, flow_split_info, error);
5577 	MLX5_ASSERT(ret <= 0);
5578 	return ret;
5579 }
5580 
5581 static struct mlx5_flow_tunnel *
5582 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5583 		      const struct rte_flow_attr *attr,
5584 		      const struct rte_flow_item items[],
5585 		      const struct rte_flow_action actions[])
5586 {
5587 	struct mlx5_flow_tunnel *tunnel;
5588 
5589 #pragma GCC diagnostic push
5590 #pragma GCC diagnostic ignored "-Wcast-qual"
5591 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5592 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5593 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5594 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5595 	else
5596 		tunnel = NULL;
5597 #pragma GCC diagnostic pop
5598 
5599 	return tunnel;
5600 }
5601 
5602 /**
5603  * Adjust flow RSS workspace if needed.
5604  *
5605  * @param wks
5606  *   Pointer to thread flow work space.
5607  * @param rss_desc
5608  *   Pointer to RSS descriptor.
5609  * @param[in] nrssq_num
5610  *   New RSS queue number.
5611  *
5612  * @return
5613  *   0 on success, -1 otherwise and rte_errno is set.
5614  */
5615 static int
5616 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5617 			  struct mlx5_flow_rss_desc *rss_desc,
5618 			  uint32_t nrssq_num)
5619 {
5620 	if (likely(nrssq_num <= wks->rssq_num))
5621 		return 0;
5622 	rss_desc->queue = realloc(rss_desc->queue,
5623 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5624 	if (!rss_desc->queue) {
5625 		rte_errno = ENOMEM;
5626 		return -1;
5627 	}
5628 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5629 	return 0;
5630 }
5631 
5632 /**
5633  * Create a flow and add it to @p list.
5634  *
5635  * @param dev
5636  *   Pointer to Ethernet device.
5637  * @param list
5638  *   Pointer to a TAILQ flow list. If this parameter NULL,
5639  *   no list insertion occurred, flow is just created,
5640  *   this is caller's responsibility to track the
5641  *   created flow.
5642  * @param[in] attr
5643  *   Flow rule attributes.
5644  * @param[in] items
5645  *   Pattern specification (list terminated by the END pattern item).
5646  * @param[in] actions
5647  *   Associated actions (list terminated by the END action).
5648  * @param[in] external
5649  *   This flow rule is created by request external to PMD.
5650  * @param[out] error
5651  *   Perform verbose error reporting if not NULL.
5652  *
5653  * @return
5654  *   A flow index on success, 0 otherwise and rte_errno is set.
5655  */
5656 static uint32_t
5657 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5658 		 const struct rte_flow_attr *attr,
5659 		 const struct rte_flow_item items[],
5660 		 const struct rte_flow_action original_actions[],
5661 		 bool external, struct rte_flow_error *error)
5662 {
5663 	struct mlx5_priv *priv = dev->data->dev_private;
5664 	struct rte_flow *flow = NULL;
5665 	struct mlx5_flow *dev_flow;
5666 	const struct rte_flow_action_rss *rss = NULL;
5667 	struct mlx5_translated_action_handle
5668 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
5669 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
5670 	union {
5671 		struct mlx5_flow_expand_rss buf;
5672 		uint8_t buffer[2048];
5673 	} expand_buffer;
5674 	union {
5675 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5676 		uint8_t buffer[2048];
5677 	} actions_rx;
5678 	union {
5679 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5680 		uint8_t buffer[2048];
5681 	} actions_hairpin_tx;
5682 	union {
5683 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5684 		uint8_t buffer[2048];
5685 	} items_tx;
5686 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5687 	struct mlx5_flow_rss_desc *rss_desc;
5688 	const struct rte_flow_action *p_actions_rx;
5689 	uint32_t i;
5690 	uint32_t idx = 0;
5691 	int hairpin_flow;
5692 	struct rte_flow_attr attr_tx = { .priority = 0 };
5693 	const struct rte_flow_action *actions;
5694 	struct rte_flow_action *translated_actions = NULL;
5695 	struct mlx5_flow_tunnel *tunnel;
5696 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5697 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5698 	struct mlx5_flow_split_info flow_split_info = {
5699 		.external = !!external,
5700 		.skip_scale = 0,
5701 		.flow_idx = 0,
5702 		.prefix_mark = 0,
5703 		.prefix_layers = 0,
5704 		.table_id = 0
5705 	};
5706 	int ret;
5707 
5708 	MLX5_ASSERT(wks);
5709 	rss_desc = &wks->rss_desc;
5710 	ret = flow_action_handles_translate(dev, original_actions,
5711 					    indir_actions,
5712 					    &indir_actions_n,
5713 					    &translated_actions, error);
5714 	if (ret < 0) {
5715 		MLX5_ASSERT(translated_actions == NULL);
5716 		return 0;
5717 	}
5718 	actions = translated_actions ? translated_actions : original_actions;
5719 	p_actions_rx = actions;
5720 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5721 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5722 				external, hairpin_flow, error);
5723 	if (ret < 0)
5724 		goto error_before_hairpin_split;
5725 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5726 	if (!flow) {
5727 		rte_errno = ENOMEM;
5728 		goto error_before_hairpin_split;
5729 	}
5730 	if (hairpin_flow > 0) {
5731 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5732 			rte_errno = EINVAL;
5733 			goto error_before_hairpin_split;
5734 		}
5735 		flow_hairpin_split(dev, actions, actions_rx.actions,
5736 				   actions_hairpin_tx.actions, items_tx.items,
5737 				   idx);
5738 		p_actions_rx = actions_rx.actions;
5739 	}
5740 	flow_split_info.flow_idx = idx;
5741 	flow->drv_type = flow_get_drv_type(dev, attr);
5742 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5743 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5744 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5745 	/* RSS Action only works on NIC RX domain */
5746 	if (attr->ingress && !attr->transfer)
5747 		rss = flow_get_rss_action(p_actions_rx);
5748 	if (rss) {
5749 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5750 			return 0;
5751 		/*
5752 		 * The following information is required by
5753 		 * mlx5_flow_hashfields_adjust() in advance.
5754 		 */
5755 		rss_desc->level = rss->level;
5756 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5757 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5758 	}
5759 	flow->dev_handles = 0;
5760 	if (rss && rss->types) {
5761 		unsigned int graph_root;
5762 
5763 		graph_root = find_graph_root(items, rss->level);
5764 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5765 					   items, rss->types,
5766 					   mlx5_support_expansion, graph_root);
5767 		MLX5_ASSERT(ret > 0 &&
5768 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5769 	} else {
5770 		buf->entries = 1;
5771 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5772 	}
5773 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
5774 						      indir_actions_n);
5775 	for (i = 0; i < buf->entries; ++i) {
5776 		/* Initialize flow split data. */
5777 		flow_split_info.prefix_layers = 0;
5778 		flow_split_info.prefix_mark = 0;
5779 		flow_split_info.skip_scale = 0;
5780 		/*
5781 		 * The splitter may create multiple dev_flows,
5782 		 * depending on configuration. In the simplest
5783 		 * case it just creates unmodified original flow.
5784 		 */
5785 		ret = flow_create_split_outer(dev, flow, attr,
5786 					      buf->entry[i].pattern,
5787 					      p_actions_rx, &flow_split_info,
5788 					      error);
5789 		if (ret < 0)
5790 			goto error;
5791 		if (is_flow_tunnel_steer_rule(dev, attr,
5792 					      buf->entry[i].pattern,
5793 					      p_actions_rx)) {
5794 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5795 							   p_actions_rx,
5796 							   idx,
5797 							   &default_miss_ctx,
5798 							   error);
5799 			if (ret < 0) {
5800 				mlx5_free(default_miss_ctx.queue);
5801 				goto error;
5802 			}
5803 		}
5804 	}
5805 	/* Create the tx flow. */
5806 	if (hairpin_flow) {
5807 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5808 		attr_tx.ingress = 0;
5809 		attr_tx.egress = 1;
5810 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5811 					 actions_hairpin_tx.actions,
5812 					 idx, error);
5813 		if (!dev_flow)
5814 			goto error;
5815 		dev_flow->flow = flow;
5816 		dev_flow->external = 0;
5817 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5818 			      dev_flow->handle, next);
5819 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5820 					 items_tx.items,
5821 					 actions_hairpin_tx.actions, error);
5822 		if (ret < 0)
5823 			goto error;
5824 	}
5825 	/*
5826 	 * Update the metadata register copy table. If extensive
5827 	 * metadata feature is enabled and registers are supported
5828 	 * we might create the extra rte_flow for each unique
5829 	 * MARK/FLAG action ID.
5830 	 *
5831 	 * The table is updated for ingress Flows only, because
5832 	 * the egress Flows belong to the different device and
5833 	 * copy table should be updated in peer NIC Rx domain.
5834 	 */
5835 	if (attr->ingress &&
5836 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5837 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5838 		if (ret)
5839 			goto error;
5840 	}
5841 	/*
5842 	 * If the flow is external (from application) OR device is started,
5843 	 * OR mreg discover, then apply immediately.
5844 	 */
5845 	if (external || dev->data->dev_started ||
5846 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5847 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
5848 		ret = flow_drv_apply(dev, flow, error);
5849 		if (ret < 0)
5850 			goto error;
5851 	}
5852 	if (list) {
5853 		rte_spinlock_lock(&priv->flow_list_lock);
5854 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5855 			     flow, next);
5856 		rte_spinlock_unlock(&priv->flow_list_lock);
5857 	}
5858 	flow_rxq_flags_set(dev, flow);
5859 	rte_free(translated_actions);
5860 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5861 	if (tunnel) {
5862 		flow->tunnel = 1;
5863 		flow->tunnel_id = tunnel->tunnel_id;
5864 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5865 		mlx5_free(default_miss_ctx.queue);
5866 	}
5867 	mlx5_flow_pop_thread_workspace();
5868 	return idx;
5869 error:
5870 	MLX5_ASSERT(flow);
5871 	ret = rte_errno; /* Save rte_errno before cleanup. */
5872 	flow_mreg_del_copy_action(dev, flow);
5873 	flow_drv_destroy(dev, flow);
5874 	if (rss_desc->shared_rss)
5875 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5876 			mlx5_ipool_get
5877 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5878 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5879 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5880 	rte_errno = ret; /* Restore rte_errno. */
5881 	ret = rte_errno;
5882 	rte_errno = ret;
5883 	mlx5_flow_pop_thread_workspace();
5884 error_before_hairpin_split:
5885 	rte_free(translated_actions);
5886 	return 0;
5887 }
5888 
5889 /**
5890  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5891  * incoming packets to table 1.
5892  *
5893  * Other flow rules, requested for group n, will be created in
5894  * e-switch table n+1.
5895  * Jump action to e-switch group n will be created to group n+1.
5896  *
5897  * Used when working in switchdev mode, to utilise advantages of table 1
5898  * and above.
5899  *
5900  * @param dev
5901  *   Pointer to Ethernet device.
5902  *
5903  * @return
5904  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5905  */
5906 struct rte_flow *
5907 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5908 {
5909 	const struct rte_flow_attr attr = {
5910 		.group = 0,
5911 		.priority = 0,
5912 		.ingress = 1,
5913 		.egress = 0,
5914 		.transfer = 1,
5915 	};
5916 	const struct rte_flow_item pattern = {
5917 		.type = RTE_FLOW_ITEM_TYPE_END,
5918 	};
5919 	struct rte_flow_action_jump jump = {
5920 		.group = 1,
5921 	};
5922 	const struct rte_flow_action actions[] = {
5923 		{
5924 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5925 			.conf = &jump,
5926 		},
5927 		{
5928 			.type = RTE_FLOW_ACTION_TYPE_END,
5929 		},
5930 	};
5931 	struct mlx5_priv *priv = dev->data->dev_private;
5932 	struct rte_flow_error error;
5933 
5934 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5935 						   &attr, &pattern,
5936 						   actions, false, &error);
5937 }
5938 
5939 /**
5940  * Validate a flow supported by the NIC.
5941  *
5942  * @see rte_flow_validate()
5943  * @see rte_flow_ops
5944  */
5945 int
5946 mlx5_flow_validate(struct rte_eth_dev *dev,
5947 		   const struct rte_flow_attr *attr,
5948 		   const struct rte_flow_item items[],
5949 		   const struct rte_flow_action original_actions[],
5950 		   struct rte_flow_error *error)
5951 {
5952 	int hairpin_flow;
5953 	struct mlx5_translated_action_handle
5954 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
5955 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
5956 	const struct rte_flow_action *actions;
5957 	struct rte_flow_action *translated_actions = NULL;
5958 	int ret = flow_action_handles_translate(dev, original_actions,
5959 						indir_actions,
5960 						&indir_actions_n,
5961 						&translated_actions, error);
5962 
5963 	if (ret)
5964 		return ret;
5965 	actions = translated_actions ? translated_actions : original_actions;
5966 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5967 	ret = flow_drv_validate(dev, attr, items, actions,
5968 				true, hairpin_flow, error);
5969 	rte_free(translated_actions);
5970 	return ret;
5971 }
5972 
5973 /**
5974  * Create a flow.
5975  *
5976  * @see rte_flow_create()
5977  * @see rte_flow_ops
5978  */
5979 struct rte_flow *
5980 mlx5_flow_create(struct rte_eth_dev *dev,
5981 		 const struct rte_flow_attr *attr,
5982 		 const struct rte_flow_item items[],
5983 		 const struct rte_flow_action actions[],
5984 		 struct rte_flow_error *error)
5985 {
5986 	struct mlx5_priv *priv = dev->data->dev_private;
5987 
5988 	/*
5989 	 * If the device is not started yet, it is not allowed to created a
5990 	 * flow from application. PMD default flows and traffic control flows
5991 	 * are not affected.
5992 	 */
5993 	if (unlikely(!dev->data->dev_started)) {
5994 		DRV_LOG(DEBUG, "port %u is not started when "
5995 			"inserting a flow", dev->data->port_id);
5996 		rte_flow_error_set(error, ENODEV,
5997 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5998 				   NULL,
5999 				   "port not started");
6000 		return NULL;
6001 	}
6002 
6003 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6004 				  attr, items, actions, true, error);
6005 }
6006 
6007 /**
6008  * Destroy a flow in a list.
6009  *
6010  * @param dev
6011  *   Pointer to Ethernet device.
6012  * @param list
6013  *   Pointer to the Indexed flow list. If this parameter NULL,
6014  *   there is no flow removal from the list. Be noted that as
6015  *   flow is add to the indexed list, memory of the indexed
6016  *   list points to maybe changed as flow destroyed.
6017  * @param[in] flow_idx
6018  *   Index of flow to destroy.
6019  */
6020 static void
6021 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6022 		  uint32_t flow_idx)
6023 {
6024 	struct mlx5_priv *priv = dev->data->dev_private;
6025 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6026 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
6027 
6028 	if (!flow)
6029 		return;
6030 	/*
6031 	 * Update RX queue flags only if port is started, otherwise it is
6032 	 * already clean.
6033 	 */
6034 	if (dev->data->dev_started)
6035 		flow_rxq_flags_trim(dev, flow);
6036 	flow_drv_destroy(dev, flow);
6037 	if (list) {
6038 		rte_spinlock_lock(&priv->flow_list_lock);
6039 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6040 			     flow_idx, flow, next);
6041 		rte_spinlock_unlock(&priv->flow_list_lock);
6042 	}
6043 	if (flow->tunnel) {
6044 		struct mlx5_flow_tunnel *tunnel;
6045 
6046 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6047 		RTE_VERIFY(tunnel);
6048 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6049 			mlx5_flow_tunnel_free(dev, tunnel);
6050 	}
6051 	flow_mreg_del_copy_action(dev, flow);
6052 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6053 }
6054 
6055 /**
6056  * Destroy all flows.
6057  *
6058  * @param dev
6059  *   Pointer to Ethernet device.
6060  * @param list
6061  *   Pointer to the Indexed flow list.
6062  * @param active
6063  *   If flushing is called avtively.
6064  */
6065 void
6066 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6067 {
6068 	uint32_t num_flushed = 0;
6069 
6070 	while (*list) {
6071 		flow_list_destroy(dev, list, *list);
6072 		num_flushed++;
6073 	}
6074 	if (active) {
6075 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6076 			dev->data->port_id, num_flushed);
6077 	}
6078 }
6079 
6080 /**
6081  * Stop all default actions for flows.
6082  *
6083  * @param dev
6084  *   Pointer to Ethernet device.
6085  */
6086 void
6087 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6088 {
6089 	flow_mreg_del_default_copy_action(dev);
6090 	flow_rxq_flags_clear(dev);
6091 }
6092 
6093 /**
6094  * Start all default actions for flows.
6095  *
6096  * @param dev
6097  *   Pointer to Ethernet device.
6098  * @return
6099  *   0 on success, a negative errno value otherwise and rte_errno is set.
6100  */
6101 int
6102 mlx5_flow_start_default(struct rte_eth_dev *dev)
6103 {
6104 	struct rte_flow_error error;
6105 
6106 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6107 	return flow_mreg_add_default_copy_action(dev, &error);
6108 }
6109 
6110 /**
6111  * Release key of thread specific flow workspace data.
6112  */
6113 void
6114 flow_release_workspace(void *data)
6115 {
6116 	struct mlx5_flow_workspace *wks = data;
6117 	struct mlx5_flow_workspace *next;
6118 
6119 	while (wks) {
6120 		next = wks->next;
6121 		free(wks->rss_desc.queue);
6122 		free(wks);
6123 		wks = next;
6124 	}
6125 }
6126 
6127 /**
6128  * Get thread specific current flow workspace.
6129  *
6130  * @return pointer to thread specific flow workspace data, NULL on error.
6131  */
6132 struct mlx5_flow_workspace*
6133 mlx5_flow_get_thread_workspace(void)
6134 {
6135 	struct mlx5_flow_workspace *data;
6136 
6137 	data = mlx5_flow_os_get_specific_workspace();
6138 	MLX5_ASSERT(data && data->inuse);
6139 	if (!data || !data->inuse)
6140 		DRV_LOG(ERR, "flow workspace not initialized.");
6141 	return data;
6142 }
6143 
6144 /**
6145  * Allocate and init new flow workspace.
6146  *
6147  * @return pointer to flow workspace data, NULL on error.
6148  */
6149 static struct mlx5_flow_workspace*
6150 flow_alloc_thread_workspace(void)
6151 {
6152 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6153 
6154 	if (!data) {
6155 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6156 			"memory.");
6157 		return NULL;
6158 	}
6159 	data->rss_desc.queue = calloc(1,
6160 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6161 	if (!data->rss_desc.queue)
6162 		goto err;
6163 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6164 	return data;
6165 err:
6166 	if (data->rss_desc.queue)
6167 		free(data->rss_desc.queue);
6168 	free(data);
6169 	return NULL;
6170 }
6171 
6172 /**
6173  * Get new thread specific flow workspace.
6174  *
6175  * If current workspace inuse, create new one and set as current.
6176  *
6177  * @return pointer to thread specific flow workspace data, NULL on error.
6178  */
6179 static struct mlx5_flow_workspace*
6180 mlx5_flow_push_thread_workspace(void)
6181 {
6182 	struct mlx5_flow_workspace *curr;
6183 	struct mlx5_flow_workspace *data;
6184 
6185 	curr = mlx5_flow_os_get_specific_workspace();
6186 	if (!curr) {
6187 		data = flow_alloc_thread_workspace();
6188 		if (!data)
6189 			return NULL;
6190 	} else if (!curr->inuse) {
6191 		data = curr;
6192 	} else if (curr->next) {
6193 		data = curr->next;
6194 	} else {
6195 		data = flow_alloc_thread_workspace();
6196 		if (!data)
6197 			return NULL;
6198 		curr->next = data;
6199 		data->prev = curr;
6200 	}
6201 	data->inuse = 1;
6202 	data->flow_idx = 0;
6203 	/* Set as current workspace */
6204 	if (mlx5_flow_os_set_specific_workspace(data))
6205 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6206 	return data;
6207 }
6208 
6209 /**
6210  * Close current thread specific flow workspace.
6211  *
6212  * If previous workspace available, set it as current.
6213  *
6214  * @return pointer to thread specific flow workspace data, NULL on error.
6215  */
6216 static void
6217 mlx5_flow_pop_thread_workspace(void)
6218 {
6219 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6220 
6221 	if (!data)
6222 		return;
6223 	if (!data->inuse) {
6224 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6225 		return;
6226 	}
6227 	data->inuse = 0;
6228 	if (!data->prev)
6229 		return;
6230 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6231 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6232 }
6233 
6234 /**
6235  * Verify the flow list is empty
6236  *
6237  * @param dev
6238  *  Pointer to Ethernet device.
6239  *
6240  * @return the number of flows not released.
6241  */
6242 int
6243 mlx5_flow_verify(struct rte_eth_dev *dev)
6244 {
6245 	struct mlx5_priv *priv = dev->data->dev_private;
6246 	struct rte_flow *flow;
6247 	uint32_t idx;
6248 	int ret = 0;
6249 
6250 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6251 		      flow, next) {
6252 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6253 			dev->data->port_id, (void *)flow);
6254 		++ret;
6255 	}
6256 	return ret;
6257 }
6258 
6259 /**
6260  * Enable default hairpin egress flow.
6261  *
6262  * @param dev
6263  *   Pointer to Ethernet device.
6264  * @param queue
6265  *   The queue index.
6266  *
6267  * @return
6268  *   0 on success, a negative errno value otherwise and rte_errno is set.
6269  */
6270 int
6271 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6272 			    uint32_t queue)
6273 {
6274 	struct mlx5_priv *priv = dev->data->dev_private;
6275 	const struct rte_flow_attr attr = {
6276 		.egress = 1,
6277 		.priority = 0,
6278 	};
6279 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6280 		.queue = queue,
6281 	};
6282 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6283 		.queue = UINT32_MAX,
6284 	};
6285 	struct rte_flow_item items[] = {
6286 		{
6287 			.type = (enum rte_flow_item_type)
6288 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6289 			.spec = &queue_spec,
6290 			.last = NULL,
6291 			.mask = &queue_mask,
6292 		},
6293 		{
6294 			.type = RTE_FLOW_ITEM_TYPE_END,
6295 		},
6296 	};
6297 	struct rte_flow_action_jump jump = {
6298 		.group = MLX5_HAIRPIN_TX_TABLE,
6299 	};
6300 	struct rte_flow_action actions[2];
6301 	uint32_t flow_idx;
6302 	struct rte_flow_error error;
6303 
6304 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6305 	actions[0].conf = &jump;
6306 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6307 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6308 				&attr, items, actions, false, &error);
6309 	if (!flow_idx) {
6310 		DRV_LOG(DEBUG,
6311 			"Failed to create ctrl flow: rte_errno(%d),"
6312 			" type(%d), message(%s)",
6313 			rte_errno, error.type,
6314 			error.message ? error.message : " (no stated reason)");
6315 		return -rte_errno;
6316 	}
6317 	return 0;
6318 }
6319 
6320 /**
6321  * Enable a control flow configured from the control plane.
6322  *
6323  * @param dev
6324  *   Pointer to Ethernet device.
6325  * @param eth_spec
6326  *   An Ethernet flow spec to apply.
6327  * @param eth_mask
6328  *   An Ethernet flow mask to apply.
6329  * @param vlan_spec
6330  *   A VLAN flow spec to apply.
6331  * @param vlan_mask
6332  *   A VLAN flow mask to apply.
6333  *
6334  * @return
6335  *   0 on success, a negative errno value otherwise and rte_errno is set.
6336  */
6337 int
6338 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6339 		    struct rte_flow_item_eth *eth_spec,
6340 		    struct rte_flow_item_eth *eth_mask,
6341 		    struct rte_flow_item_vlan *vlan_spec,
6342 		    struct rte_flow_item_vlan *vlan_mask)
6343 {
6344 	struct mlx5_priv *priv = dev->data->dev_private;
6345 	const struct rte_flow_attr attr = {
6346 		.ingress = 1,
6347 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6348 	};
6349 	struct rte_flow_item items[] = {
6350 		{
6351 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6352 			.spec = eth_spec,
6353 			.last = NULL,
6354 			.mask = eth_mask,
6355 		},
6356 		{
6357 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6358 					      RTE_FLOW_ITEM_TYPE_END,
6359 			.spec = vlan_spec,
6360 			.last = NULL,
6361 			.mask = vlan_mask,
6362 		},
6363 		{
6364 			.type = RTE_FLOW_ITEM_TYPE_END,
6365 		},
6366 	};
6367 	uint16_t queue[priv->reta_idx_n];
6368 	struct rte_flow_action_rss action_rss = {
6369 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6370 		.level = 0,
6371 		.types = priv->rss_conf.rss_hf,
6372 		.key_len = priv->rss_conf.rss_key_len,
6373 		.queue_num = priv->reta_idx_n,
6374 		.key = priv->rss_conf.rss_key,
6375 		.queue = queue,
6376 	};
6377 	struct rte_flow_action actions[] = {
6378 		{
6379 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6380 			.conf = &action_rss,
6381 		},
6382 		{
6383 			.type = RTE_FLOW_ACTION_TYPE_END,
6384 		},
6385 	};
6386 	uint32_t flow_idx;
6387 	struct rte_flow_error error;
6388 	unsigned int i;
6389 
6390 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6391 		return 0;
6392 	}
6393 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6394 		action_rss.types = 0;
6395 	for (i = 0; i != priv->reta_idx_n; ++i)
6396 		queue[i] = (*priv->reta_idx)[i];
6397 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6398 				&attr, items, actions, false, &error);
6399 	if (!flow_idx)
6400 		return -rte_errno;
6401 	return 0;
6402 }
6403 
6404 /**
6405  * Enable a flow control configured from the control plane.
6406  *
6407  * @param dev
6408  *   Pointer to Ethernet device.
6409  * @param eth_spec
6410  *   An Ethernet flow spec to apply.
6411  * @param eth_mask
6412  *   An Ethernet flow mask to apply.
6413  *
6414  * @return
6415  *   0 on success, a negative errno value otherwise and rte_errno is set.
6416  */
6417 int
6418 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6419 	       struct rte_flow_item_eth *eth_spec,
6420 	       struct rte_flow_item_eth *eth_mask)
6421 {
6422 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6423 }
6424 
6425 /**
6426  * Create default miss flow rule matching lacp traffic
6427  *
6428  * @param dev
6429  *   Pointer to Ethernet device.
6430  * @param eth_spec
6431  *   An Ethernet flow spec to apply.
6432  *
6433  * @return
6434  *   0 on success, a negative errno value otherwise and rte_errno is set.
6435  */
6436 int
6437 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6438 {
6439 	struct mlx5_priv *priv = dev->data->dev_private;
6440 	/*
6441 	 * The LACP matching is done by only using ether type since using
6442 	 * a multicast dst mac causes kernel to give low priority to this flow.
6443 	 */
6444 	static const struct rte_flow_item_eth lacp_spec = {
6445 		.type = RTE_BE16(0x8809),
6446 	};
6447 	static const struct rte_flow_item_eth lacp_mask = {
6448 		.type = 0xffff,
6449 	};
6450 	const struct rte_flow_attr attr = {
6451 		.ingress = 1,
6452 	};
6453 	struct rte_flow_item items[] = {
6454 		{
6455 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6456 			.spec = &lacp_spec,
6457 			.mask = &lacp_mask,
6458 		},
6459 		{
6460 			.type = RTE_FLOW_ITEM_TYPE_END,
6461 		},
6462 	};
6463 	struct rte_flow_action actions[] = {
6464 		{
6465 			.type = (enum rte_flow_action_type)
6466 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6467 		},
6468 		{
6469 			.type = RTE_FLOW_ACTION_TYPE_END,
6470 		},
6471 	};
6472 	struct rte_flow_error error;
6473 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6474 				&attr, items, actions, false, &error);
6475 
6476 	if (!flow_idx)
6477 		return -rte_errno;
6478 	return 0;
6479 }
6480 
6481 /**
6482  * Destroy a flow.
6483  *
6484  * @see rte_flow_destroy()
6485  * @see rte_flow_ops
6486  */
6487 int
6488 mlx5_flow_destroy(struct rte_eth_dev *dev,
6489 		  struct rte_flow *flow,
6490 		  struct rte_flow_error *error __rte_unused)
6491 {
6492 	struct mlx5_priv *priv = dev->data->dev_private;
6493 
6494 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6495 	return 0;
6496 }
6497 
6498 /**
6499  * Destroy all flows.
6500  *
6501  * @see rte_flow_flush()
6502  * @see rte_flow_ops
6503  */
6504 int
6505 mlx5_flow_flush(struct rte_eth_dev *dev,
6506 		struct rte_flow_error *error __rte_unused)
6507 {
6508 	struct mlx5_priv *priv = dev->data->dev_private;
6509 
6510 	mlx5_flow_list_flush(dev, &priv->flows, false);
6511 	return 0;
6512 }
6513 
6514 /**
6515  * Isolated mode.
6516  *
6517  * @see rte_flow_isolate()
6518  * @see rte_flow_ops
6519  */
6520 int
6521 mlx5_flow_isolate(struct rte_eth_dev *dev,
6522 		  int enable,
6523 		  struct rte_flow_error *error)
6524 {
6525 	struct mlx5_priv *priv = dev->data->dev_private;
6526 
6527 	if (dev->data->dev_started) {
6528 		rte_flow_error_set(error, EBUSY,
6529 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6530 				   NULL,
6531 				   "port must be stopped first");
6532 		return -rte_errno;
6533 	}
6534 	priv->isolated = !!enable;
6535 	if (enable)
6536 		dev->dev_ops = &mlx5_dev_ops_isolate;
6537 	else
6538 		dev->dev_ops = &mlx5_dev_ops;
6539 
6540 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6541 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6542 
6543 	return 0;
6544 }
6545 
6546 /**
6547  * Query a flow.
6548  *
6549  * @see rte_flow_query()
6550  * @see rte_flow_ops
6551  */
6552 static int
6553 flow_drv_query(struct rte_eth_dev *dev,
6554 	       uint32_t flow_idx,
6555 	       const struct rte_flow_action *actions,
6556 	       void *data,
6557 	       struct rte_flow_error *error)
6558 {
6559 	struct mlx5_priv *priv = dev->data->dev_private;
6560 	const struct mlx5_flow_driver_ops *fops;
6561 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6562 					       [MLX5_IPOOL_RTE_FLOW],
6563 					       flow_idx);
6564 	enum mlx5_flow_drv_type ftype;
6565 
6566 	if (!flow) {
6567 		return rte_flow_error_set(error, ENOENT,
6568 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6569 			  NULL,
6570 			  "invalid flow handle");
6571 	}
6572 	ftype = flow->drv_type;
6573 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6574 	fops = flow_get_drv_ops(ftype);
6575 
6576 	return fops->query(dev, flow, actions, data, error);
6577 }
6578 
6579 /**
6580  * Query a flow.
6581  *
6582  * @see rte_flow_query()
6583  * @see rte_flow_ops
6584  */
6585 int
6586 mlx5_flow_query(struct rte_eth_dev *dev,
6587 		struct rte_flow *flow,
6588 		const struct rte_flow_action *actions,
6589 		void *data,
6590 		struct rte_flow_error *error)
6591 {
6592 	int ret;
6593 
6594 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6595 			     error);
6596 	if (ret < 0)
6597 		return ret;
6598 	return 0;
6599 }
6600 
6601 /**
6602  * Get rte_flow callbacks.
6603  *
6604  * @param dev
6605  *   Pointer to Ethernet device structure.
6606  * @param ops
6607  *   Pointer to operation-specific structure.
6608  *
6609  * @return 0
6610  */
6611 int
6612 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
6613 		  const struct rte_flow_ops **ops)
6614 {
6615 	*ops = &mlx5_flow_ops;
6616 	return 0;
6617 }
6618 
6619 /**
6620  * Create the needed meter and suffix tables.
6621  *
6622  * @param[in] dev
6623  *   Pointer to Ethernet device.
6624  *
6625  * @return
6626  *   Pointer to table set on success, NULL otherwise.
6627  */
6628 struct mlx5_meter_domains_infos *
6629 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev)
6630 {
6631 	const struct mlx5_flow_driver_ops *fops;
6632 
6633 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6634 	return fops->create_mtr_tbls(dev);
6635 }
6636 
6637 /**
6638  * Destroy the meter table set.
6639  *
6640  * @param[in] dev
6641  *   Pointer to Ethernet device.
6642  * @param[in] tbl
6643  *   Pointer to the meter table set.
6644  *
6645  * @return
6646  *   0 on success.
6647  */
6648 int
6649 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6650 			   struct mlx5_meter_domains_infos *tbls)
6651 {
6652 	const struct mlx5_flow_driver_ops *fops;
6653 
6654 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6655 	return fops->destroy_mtr_tbls(dev, tbls);
6656 }
6657 
6658 /**
6659  * Allocate the needed aso flow meter id.
6660  *
6661  * @param[in] dev
6662  *   Pointer to Ethernet device.
6663  *
6664  * @return
6665  *   Index to aso flow meter on success, NULL otherwise.
6666  */
6667 uint32_t
6668 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
6669 {
6670 	const struct mlx5_flow_driver_ops *fops;
6671 
6672 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6673 	return fops->create_meter(dev);
6674 }
6675 
6676 /**
6677  * Free the aso flow meter id.
6678  *
6679  * @param[in] dev
6680  *   Pointer to Ethernet device.
6681  * @param[in] mtr_idx
6682  *  Index to aso flow meter to be free.
6683  *
6684  * @return
6685  *   0 on success.
6686  */
6687 void
6688 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
6689 {
6690 	const struct mlx5_flow_driver_ops *fops;
6691 
6692 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6693 	fops->free_meter(dev, mtr_idx);
6694 }
6695 
6696 /**
6697  * Allocate a counter.
6698  *
6699  * @param[in] dev
6700  *   Pointer to Ethernet device structure.
6701  *
6702  * @return
6703  *   Index to allocated counter  on success, 0 otherwise.
6704  */
6705 uint32_t
6706 mlx5_counter_alloc(struct rte_eth_dev *dev)
6707 {
6708 	const struct mlx5_flow_driver_ops *fops;
6709 	struct rte_flow_attr attr = { .transfer = 0 };
6710 
6711 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6712 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6713 		return fops->counter_alloc(dev);
6714 	}
6715 	DRV_LOG(ERR,
6716 		"port %u counter allocate is not supported.",
6717 		 dev->data->port_id);
6718 	return 0;
6719 }
6720 
6721 /**
6722  * Free a counter.
6723  *
6724  * @param[in] dev
6725  *   Pointer to Ethernet device structure.
6726  * @param[in] cnt
6727  *   Index to counter to be free.
6728  */
6729 void
6730 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6731 {
6732 	const struct mlx5_flow_driver_ops *fops;
6733 	struct rte_flow_attr attr = { .transfer = 0 };
6734 
6735 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6736 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6737 		fops->counter_free(dev, cnt);
6738 		return;
6739 	}
6740 	DRV_LOG(ERR,
6741 		"port %u counter free is not supported.",
6742 		 dev->data->port_id);
6743 }
6744 
6745 /**
6746  * Query counter statistics.
6747  *
6748  * @param[in] dev
6749  *   Pointer to Ethernet device structure.
6750  * @param[in] cnt
6751  *   Index to counter to query.
6752  * @param[in] clear
6753  *   Set to clear counter statistics.
6754  * @param[out] pkts
6755  *   The counter hits packets number to save.
6756  * @param[out] bytes
6757  *   The counter hits bytes number to save.
6758  *
6759  * @return
6760  *   0 on success, a negative errno value otherwise.
6761  */
6762 int
6763 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6764 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6765 {
6766 	const struct mlx5_flow_driver_ops *fops;
6767 	struct rte_flow_attr attr = { .transfer = 0 };
6768 
6769 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6770 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6771 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6772 	}
6773 	DRV_LOG(ERR,
6774 		"port %u counter query is not supported.",
6775 		 dev->data->port_id);
6776 	return -ENOTSUP;
6777 }
6778 
6779 /**
6780  * Allocate a new memory for the counter values wrapped by all the needed
6781  * management.
6782  *
6783  * @param[in] sh
6784  *   Pointer to mlx5_dev_ctx_shared object.
6785  *
6786  * @return
6787  *   0 on success, a negative errno value otherwise.
6788  */
6789 static int
6790 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6791 {
6792 	struct mlx5_devx_mkey_attr mkey_attr;
6793 	struct mlx5_counter_stats_mem_mng *mem_mng;
6794 	volatile struct flow_counter_stats *raw_data;
6795 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6796 	int size = (sizeof(struct flow_counter_stats) *
6797 			MLX5_COUNTERS_PER_POOL +
6798 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6799 			sizeof(struct mlx5_counter_stats_mem_mng);
6800 	size_t pgsize = rte_mem_page_size();
6801 	uint8_t *mem;
6802 	int i;
6803 
6804 	if (pgsize == (size_t)-1) {
6805 		DRV_LOG(ERR, "Failed to get mem page size");
6806 		rte_errno = ENOMEM;
6807 		return -ENOMEM;
6808 	}
6809 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6810 	if (!mem) {
6811 		rte_errno = ENOMEM;
6812 		return -ENOMEM;
6813 	}
6814 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6815 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6816 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6817 						 IBV_ACCESS_LOCAL_WRITE);
6818 	if (!mem_mng->umem) {
6819 		rte_errno = errno;
6820 		mlx5_free(mem);
6821 		return -rte_errno;
6822 	}
6823 	mkey_attr.addr = (uintptr_t)mem;
6824 	mkey_attr.size = size;
6825 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6826 	mkey_attr.pd = sh->pdn;
6827 	mkey_attr.log_entity_size = 0;
6828 	mkey_attr.pg_access = 0;
6829 	mkey_attr.klm_array = NULL;
6830 	mkey_attr.klm_num = 0;
6831 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6832 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6833 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6834 	if (!mem_mng->dm) {
6835 		mlx5_os_umem_dereg(mem_mng->umem);
6836 		rte_errno = errno;
6837 		mlx5_free(mem);
6838 		return -rte_errno;
6839 	}
6840 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6841 	raw_data = (volatile struct flow_counter_stats *)mem;
6842 	for (i = 0; i < raws_n; ++i) {
6843 		mem_mng->raws[i].mem_mng = mem_mng;
6844 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6845 	}
6846 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6847 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6848 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6849 				 next);
6850 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6851 	sh->cmng.mem_mng = mem_mng;
6852 	return 0;
6853 }
6854 
6855 /**
6856  * Set the statistic memory to the new counter pool.
6857  *
6858  * @param[in] sh
6859  *   Pointer to mlx5_dev_ctx_shared object.
6860  * @param[in] pool
6861  *   Pointer to the pool to set the statistic memory.
6862  *
6863  * @return
6864  *   0 on success, a negative errno value otherwise.
6865  */
6866 static int
6867 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6868 			       struct mlx5_flow_counter_pool *pool)
6869 {
6870 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6871 	/* Resize statistic memory once used out. */
6872 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6873 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6874 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6875 		return -1;
6876 	}
6877 	rte_spinlock_lock(&pool->sl);
6878 	pool->raw = cmng->mem_mng->raws + pool->index %
6879 		    MLX5_CNT_CONTAINER_RESIZE;
6880 	rte_spinlock_unlock(&pool->sl);
6881 	pool->raw_hw = NULL;
6882 	return 0;
6883 }
6884 
6885 #define MLX5_POOL_QUERY_FREQ_US 1000000
6886 
6887 /**
6888  * Set the periodic procedure for triggering asynchronous batch queries for all
6889  * the counter pools.
6890  *
6891  * @param[in] sh
6892  *   Pointer to mlx5_dev_ctx_shared object.
6893  */
6894 void
6895 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6896 {
6897 	uint32_t pools_n, us;
6898 
6899 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6900 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6901 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6902 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6903 		sh->cmng.query_thread_on = 0;
6904 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6905 	} else {
6906 		sh->cmng.query_thread_on = 1;
6907 	}
6908 }
6909 
6910 /**
6911  * The periodic procedure for triggering asynchronous batch queries for all the
6912  * counter pools. This function is probably called by the host thread.
6913  *
6914  * @param[in] arg
6915  *   The parameter for the alarm process.
6916  */
6917 void
6918 mlx5_flow_query_alarm(void *arg)
6919 {
6920 	struct mlx5_dev_ctx_shared *sh = arg;
6921 	int ret;
6922 	uint16_t pool_index = sh->cmng.pool_index;
6923 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6924 	struct mlx5_flow_counter_pool *pool;
6925 	uint16_t n_valid;
6926 
6927 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6928 		goto set_alarm;
6929 	rte_spinlock_lock(&cmng->pool_update_sl);
6930 	pool = cmng->pools[pool_index];
6931 	n_valid = cmng->n_valid;
6932 	rte_spinlock_unlock(&cmng->pool_update_sl);
6933 	/* Set the statistic memory to the new created pool. */
6934 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6935 		goto set_alarm;
6936 	if (pool->raw_hw)
6937 		/* There is a pool query in progress. */
6938 		goto set_alarm;
6939 	pool->raw_hw =
6940 		LIST_FIRST(&sh->cmng.free_stat_raws);
6941 	if (!pool->raw_hw)
6942 		/* No free counter statistics raw memory. */
6943 		goto set_alarm;
6944 	/*
6945 	 * Identify the counters released between query trigger and query
6946 	 * handle more efficiently. The counter released in this gap period
6947 	 * should wait for a new round of query as the new arrived packets
6948 	 * will not be taken into account.
6949 	 */
6950 	pool->query_gen++;
6951 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6952 					       MLX5_COUNTERS_PER_POOL,
6953 					       NULL, NULL,
6954 					       pool->raw_hw->mem_mng->dm->id,
6955 					       (void *)(uintptr_t)
6956 					       pool->raw_hw->data,
6957 					       sh->devx_comp,
6958 					       (uint64_t)(uintptr_t)pool);
6959 	if (ret) {
6960 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6961 			" %d", pool->min_dcs->id);
6962 		pool->raw_hw = NULL;
6963 		goto set_alarm;
6964 	}
6965 	LIST_REMOVE(pool->raw_hw, next);
6966 	sh->cmng.pending_queries++;
6967 	pool_index++;
6968 	if (pool_index >= n_valid)
6969 		pool_index = 0;
6970 set_alarm:
6971 	sh->cmng.pool_index = pool_index;
6972 	mlx5_set_query_alarm(sh);
6973 }
6974 
6975 /**
6976  * Check and callback event for new aged flow in the counter pool
6977  *
6978  * @param[in] sh
6979  *   Pointer to mlx5_dev_ctx_shared object.
6980  * @param[in] pool
6981  *   Pointer to Current counter pool.
6982  */
6983 static void
6984 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6985 		   struct mlx5_flow_counter_pool *pool)
6986 {
6987 	struct mlx5_priv *priv;
6988 	struct mlx5_flow_counter *cnt;
6989 	struct mlx5_age_info *age_info;
6990 	struct mlx5_age_param *age_param;
6991 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6992 	struct mlx5_counter_stats_raw *prev = pool->raw;
6993 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6994 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6995 	uint16_t expected = AGE_CANDIDATE;
6996 	uint32_t i;
6997 
6998 	pool->time_of_last_age_check = curr_time;
6999 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7000 		cnt = MLX5_POOL_GET_CNT(pool, i);
7001 		age_param = MLX5_CNT_TO_AGE(cnt);
7002 		if (__atomic_load_n(&age_param->state,
7003 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7004 			continue;
7005 		if (cur->data[i].hits != prev->data[i].hits) {
7006 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7007 					 __ATOMIC_RELAXED);
7008 			continue;
7009 		}
7010 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7011 				       time_delta,
7012 				       __ATOMIC_RELAXED) <= age_param->timeout)
7013 			continue;
7014 		/**
7015 		 * Hold the lock first, or if between the
7016 		 * state AGE_TMOUT and tailq operation the
7017 		 * release happened, the release procedure
7018 		 * may delete a non-existent tailq node.
7019 		 */
7020 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7021 		age_info = GET_PORT_AGE_INFO(priv);
7022 		rte_spinlock_lock(&age_info->aged_sl);
7023 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7024 						AGE_TMOUT, false,
7025 						__ATOMIC_RELAXED,
7026 						__ATOMIC_RELAXED)) {
7027 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7028 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7029 		}
7030 		rte_spinlock_unlock(&age_info->aged_sl);
7031 	}
7032 	mlx5_age_event_prepare(sh);
7033 }
7034 
7035 /**
7036  * Handler for the HW respond about ready values from an asynchronous batch
7037  * query. This function is probably called by the host thread.
7038  *
7039  * @param[in] sh
7040  *   The pointer to the shared device context.
7041  * @param[in] async_id
7042  *   The Devx async ID.
7043  * @param[in] status
7044  *   The status of the completion.
7045  */
7046 void
7047 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7048 				  uint64_t async_id, int status)
7049 {
7050 	struct mlx5_flow_counter_pool *pool =
7051 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7052 	struct mlx5_counter_stats_raw *raw_to_free;
7053 	uint8_t query_gen = pool->query_gen ^ 1;
7054 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7055 	enum mlx5_counter_type cnt_type =
7056 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7057 				MLX5_COUNTER_TYPE_ORIGIN;
7058 
7059 	if (unlikely(status)) {
7060 		raw_to_free = pool->raw_hw;
7061 	} else {
7062 		raw_to_free = pool->raw;
7063 		if (pool->is_aged)
7064 			mlx5_flow_aging_check(sh, pool);
7065 		rte_spinlock_lock(&pool->sl);
7066 		pool->raw = pool->raw_hw;
7067 		rte_spinlock_unlock(&pool->sl);
7068 		/* Be sure the new raw counters data is updated in memory. */
7069 		rte_io_wmb();
7070 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7071 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7072 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7073 				     &pool->counters[query_gen], next);
7074 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7075 		}
7076 	}
7077 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7078 	pool->raw_hw = NULL;
7079 	sh->cmng.pending_queries--;
7080 }
7081 
7082 static int
7083 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7084 		    const struct flow_grp_info *grp_info,
7085 		    struct rte_flow_error *error)
7086 {
7087 	if (grp_info->transfer && grp_info->external &&
7088 	    grp_info->fdb_def_rule) {
7089 		if (group == UINT32_MAX)
7090 			return rte_flow_error_set
7091 						(error, EINVAL,
7092 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7093 						 NULL,
7094 						 "group index not supported");
7095 		*table = group + 1;
7096 	} else {
7097 		*table = group;
7098 	}
7099 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7100 	return 0;
7101 }
7102 
7103 /**
7104  * Translate the rte_flow group index to HW table value.
7105  *
7106  * If tunnel offload is disabled, all group ids converted to flow table
7107  * id using the standard method.
7108  * If tunnel offload is enabled, group id can be converted using the
7109  * standard or tunnel conversion method. Group conversion method
7110  * selection depends on flags in `grp_info` parameter:
7111  * - Internal (grp_info.external == 0) groups conversion uses the
7112  *   standard method.
7113  * - Group ids in JUMP action converted with the tunnel conversion.
7114  * - Group id in rule attribute conversion depends on a rule type and
7115  *   group id value:
7116  *   ** non zero group attributes converted with the tunnel method
7117  *   ** zero group attribute in non-tunnel rule is converted using the
7118  *      standard method - there's only one root table
7119  *   ** zero group attribute in steer tunnel rule is converted with the
7120  *      standard method - single root table
7121  *   ** zero group attribute in match tunnel rule is a special OvS
7122  *      case: that value is used for portability reasons. That group
7123  *      id is converted with the tunnel conversion method.
7124  *
7125  * @param[in] dev
7126  *   Port device
7127  * @param[in] tunnel
7128  *   PMD tunnel offload object
7129  * @param[in] group
7130  *   rte_flow group index value.
7131  * @param[out] table
7132  *   HW table value.
7133  * @param[in] grp_info
7134  *   flags used for conversion
7135  * @param[out] error
7136  *   Pointer to error structure.
7137  *
7138  * @return
7139  *   0 on success, a negative errno value otherwise and rte_errno is set.
7140  */
7141 int
7142 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7143 			 const struct mlx5_flow_tunnel *tunnel,
7144 			 uint32_t group, uint32_t *table,
7145 			 const struct flow_grp_info *grp_info,
7146 			 struct rte_flow_error *error)
7147 {
7148 	int ret;
7149 	bool standard_translation;
7150 
7151 	if (!grp_info->skip_scale && grp_info->external &&
7152 	    group < MLX5_MAX_TABLES_EXTERNAL)
7153 		group *= MLX5_FLOW_TABLE_FACTOR;
7154 	if (is_tunnel_offload_active(dev)) {
7155 		standard_translation = !grp_info->external ||
7156 					grp_info->std_tbl_fix;
7157 	} else {
7158 		standard_translation = true;
7159 	}
7160 	DRV_LOG(DEBUG,
7161 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7162 		dev->data->port_id, group, grp_info->transfer,
7163 		grp_info->external, grp_info->fdb_def_rule,
7164 		standard_translation ? "STANDARD" : "TUNNEL");
7165 	if (standard_translation)
7166 		ret = flow_group_to_table(dev->data->port_id, group, table,
7167 					  grp_info, error);
7168 	else
7169 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7170 						      table, error);
7171 
7172 	return ret;
7173 }
7174 
7175 /**
7176  * Discover availability of metadata reg_c's.
7177  *
7178  * Iteratively use test flows to check availability.
7179  *
7180  * @param[in] dev
7181  *   Pointer to the Ethernet device structure.
7182  *
7183  * @return
7184  *   0 on success, a negative errno value otherwise and rte_errno is set.
7185  */
7186 int
7187 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7188 {
7189 	struct mlx5_priv *priv = dev->data->dev_private;
7190 	struct mlx5_dev_config *config = &priv->config;
7191 	enum modify_reg idx;
7192 	int n = 0;
7193 
7194 	/* reg_c[0] and reg_c[1] are reserved. */
7195 	config->flow_mreg_c[n++] = REG_C_0;
7196 	config->flow_mreg_c[n++] = REG_C_1;
7197 	/* Discover availability of other reg_c's. */
7198 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7199 		struct rte_flow_attr attr = {
7200 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7201 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7202 			.ingress = 1,
7203 		};
7204 		struct rte_flow_item items[] = {
7205 			[0] = {
7206 				.type = RTE_FLOW_ITEM_TYPE_END,
7207 			},
7208 		};
7209 		struct rte_flow_action actions[] = {
7210 			[0] = {
7211 				.type = (enum rte_flow_action_type)
7212 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7213 				.conf = &(struct mlx5_flow_action_copy_mreg){
7214 					.src = REG_C_1,
7215 					.dst = idx,
7216 				},
7217 			},
7218 			[1] = {
7219 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7220 				.conf = &(struct rte_flow_action_jump){
7221 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7222 				},
7223 			},
7224 			[2] = {
7225 				.type = RTE_FLOW_ACTION_TYPE_END,
7226 			},
7227 		};
7228 		uint32_t flow_idx;
7229 		struct rte_flow *flow;
7230 		struct rte_flow_error error;
7231 
7232 		if (!config->dv_flow_en)
7233 			break;
7234 		/* Create internal flow, validation skips copy action. */
7235 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7236 					    actions, false, &error);
7237 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7238 				      flow_idx);
7239 		if (!flow)
7240 			continue;
7241 		config->flow_mreg_c[n++] = idx;
7242 		flow_list_destroy(dev, NULL, flow_idx);
7243 	}
7244 	for (; n < MLX5_MREG_C_NUM; ++n)
7245 		config->flow_mreg_c[n] = REG_NON;
7246 	return 0;
7247 }
7248 
7249 /**
7250  * Dump flow raw hw data to file
7251  *
7252  * @param[in] dev
7253  *    The pointer to Ethernet device.
7254  * @param[in] file
7255  *   A pointer to a file for output.
7256  * @param[out] error
7257  *   Perform verbose error reporting if not NULL. PMDs initialize this
7258  *   structure in case of error only.
7259  * @return
7260  *   0 on success, a nagative value otherwise.
7261  */
7262 int
7263 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7264 		   FILE *file,
7265 		   struct rte_flow_error *error __rte_unused)
7266 {
7267 	struct mlx5_priv *priv = dev->data->dev_private;
7268 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7269 	uint32_t handle_idx;
7270 	int ret;
7271 	struct mlx5_flow_handle *dh;
7272 	struct rte_flow *flow;
7273 
7274 	if (!priv->config.dv_flow_en) {
7275 		if (fputs("device dv flow disabled\n", file) <= 0)
7276 			return -errno;
7277 		return -ENOTSUP;
7278 	}
7279 
7280 	/* dump all */
7281 	if (!flow_idx)
7282 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7283 					sh->rx_domain,
7284 					sh->tx_domain, file);
7285 	/* dump one */
7286 	flow = mlx5_ipool_get(priv->sh->ipool
7287 			[MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
7288 	if (!flow)
7289 		return -ENOENT;
7290 
7291 	handle_idx = flow->dev_handles;
7292 	while (handle_idx) {
7293 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
7294 				handle_idx);
7295 		if (!dh)
7296 			return -ENOENT;
7297 		if (dh->drv_flow) {
7298 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
7299 					file);
7300 			if (ret)
7301 				return -ENOENT;
7302 		}
7303 		handle_idx = dh->next.next;
7304 	}
7305 	return 0;
7306 }
7307 
7308 /**
7309  * Get aged-out flows.
7310  *
7311  * @param[in] dev
7312  *   Pointer to the Ethernet device structure.
7313  * @param[in] context
7314  *   The address of an array of pointers to the aged-out flows contexts.
7315  * @param[in] nb_countexts
7316  *   The length of context array pointers.
7317  * @param[out] error
7318  *   Perform verbose error reporting if not NULL. Initialized in case of
7319  *   error only.
7320  *
7321  * @return
7322  *   how many contexts get in success, otherwise negative errno value.
7323  *   if nb_contexts is 0, return the amount of all aged contexts.
7324  *   if nb_contexts is not 0 , return the amount of aged flows reported
7325  *   in the context array.
7326  */
7327 int
7328 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7329 			uint32_t nb_contexts, struct rte_flow_error *error)
7330 {
7331 	const struct mlx5_flow_driver_ops *fops;
7332 	struct rte_flow_attr attr = { .transfer = 0 };
7333 
7334 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7335 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7336 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7337 						    error);
7338 	}
7339 	DRV_LOG(ERR,
7340 		"port %u get aged flows is not supported.",
7341 		 dev->data->port_id);
7342 	return -ENOTSUP;
7343 }
7344 
7345 /* Wrapper for driver action_validate op callback */
7346 static int
7347 flow_drv_action_validate(struct rte_eth_dev *dev,
7348 			 const struct rte_flow_indir_action_conf *conf,
7349 			 const struct rte_flow_action *action,
7350 			 const struct mlx5_flow_driver_ops *fops,
7351 			 struct rte_flow_error *error)
7352 {
7353 	static const char err_msg[] = "indirect action validation unsupported";
7354 
7355 	if (!fops->action_validate) {
7356 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7357 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7358 				   NULL, err_msg);
7359 		return -rte_errno;
7360 	}
7361 	return fops->action_validate(dev, conf, action, error);
7362 }
7363 
7364 /**
7365  * Destroys the shared action by handle.
7366  *
7367  * @param dev
7368  *   Pointer to Ethernet device structure.
7369  * @param[in] handle
7370  *   Handle for the indirect action object to be destroyed.
7371  * @param[out] error
7372  *   Perform verbose error reporting if not NULL. PMDs initialize this
7373  *   structure in case of error only.
7374  *
7375  * @return
7376  *   0 on success, a negative errno value otherwise and rte_errno is set.
7377  *
7378  * @note: wrapper for driver action_create op callback.
7379  */
7380 static int
7381 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7382 			   struct rte_flow_action_handle *handle,
7383 			   struct rte_flow_error *error)
7384 {
7385 	static const char err_msg[] = "indirect action destruction unsupported";
7386 	struct rte_flow_attr attr = { .transfer = 0 };
7387 	const struct mlx5_flow_driver_ops *fops =
7388 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7389 
7390 	if (!fops->action_destroy) {
7391 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7392 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7393 				   NULL, err_msg);
7394 		return -rte_errno;
7395 	}
7396 	return fops->action_destroy(dev, handle, error);
7397 }
7398 
7399 /* Wrapper for driver action_destroy op callback */
7400 static int
7401 flow_drv_action_update(struct rte_eth_dev *dev,
7402 		       struct rte_flow_action_handle *handle,
7403 		       const void *update,
7404 		       const struct mlx5_flow_driver_ops *fops,
7405 		       struct rte_flow_error *error)
7406 {
7407 	static const char err_msg[] = "indirect action update unsupported";
7408 
7409 	if (!fops->action_update) {
7410 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7411 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7412 				   NULL, err_msg);
7413 		return -rte_errno;
7414 	}
7415 	return fops->action_update(dev, handle, update, error);
7416 }
7417 
7418 /* Wrapper for driver action_destroy op callback */
7419 static int
7420 flow_drv_action_query(struct rte_eth_dev *dev,
7421 		      const struct rte_flow_action_handle *handle,
7422 		      void *data,
7423 		      const struct mlx5_flow_driver_ops *fops,
7424 		      struct rte_flow_error *error)
7425 {
7426 	static const char err_msg[] = "indirect action query unsupported";
7427 
7428 	if (!fops->action_query) {
7429 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7430 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7431 				   NULL, err_msg);
7432 		return -rte_errno;
7433 	}
7434 	return fops->action_query(dev, handle, data, error);
7435 }
7436 
7437 /**
7438  * Create indirect action for reuse in multiple flow rules.
7439  *
7440  * @param dev
7441  *   Pointer to Ethernet device structure.
7442  * @param conf
7443  *   Pointer to indirect action object configuration.
7444  * @param[in] action
7445  *   Action configuration for indirect action object creation.
7446  * @param[out] error
7447  *   Perform verbose error reporting if not NULL. PMDs initialize this
7448  *   structure in case of error only.
7449  * @return
7450  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7451  */
7452 static struct rte_flow_action_handle *
7453 mlx5_action_handle_create(struct rte_eth_dev *dev,
7454 			  const struct rte_flow_indir_action_conf *conf,
7455 			  const struct rte_flow_action *action,
7456 			  struct rte_flow_error *error)
7457 {
7458 	static const char err_msg[] = "indirect action creation unsupported";
7459 	struct rte_flow_attr attr = { .transfer = 0 };
7460 	const struct mlx5_flow_driver_ops *fops =
7461 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7462 
7463 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7464 		return NULL;
7465 	if (!fops->action_create) {
7466 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7467 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7468 				   NULL, err_msg);
7469 		return NULL;
7470 	}
7471 	return fops->action_create(dev, conf, action, error);
7472 }
7473 
7474 /**
7475  * Updates inplace the indirect action configuration pointed by *handle*
7476  * with the configuration provided as *update* argument.
7477  * The update of the indirect action configuration effects all flow rules
7478  * reusing the action via handle.
7479  *
7480  * @param dev
7481  *   Pointer to Ethernet device structure.
7482  * @param[in] handle
7483  *   Handle for the indirect action to be updated.
7484  * @param[in] update
7485  *   Action specification used to modify the action pointed by handle.
7486  *   *update* could be of same type with the action pointed by the *handle*
7487  *   handle argument, or some other structures like a wrapper, depending on
7488  *   the indirect action type.
7489  * @param[out] error
7490  *   Perform verbose error reporting if not NULL. PMDs initialize this
7491  *   structure in case of error only.
7492  *
7493  * @return
7494  *   0 on success, a negative errno value otherwise and rte_errno is set.
7495  */
7496 static int
7497 mlx5_action_handle_update(struct rte_eth_dev *dev,
7498 		struct rte_flow_action_handle *handle,
7499 		const void *update,
7500 		struct rte_flow_error *error)
7501 {
7502 	struct rte_flow_attr attr = { .transfer = 0 };
7503 	const struct mlx5_flow_driver_ops *fops =
7504 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7505 	int ret;
7506 
7507 	ret = flow_drv_action_validate(dev, NULL,
7508 			(const struct rte_flow_action *)update, fops, error);
7509 	if (ret)
7510 		return ret;
7511 	return flow_drv_action_update(dev, handle, update, fops,
7512 				      error);
7513 }
7514 
7515 /**
7516  * Query the indirect action by handle.
7517  *
7518  * This function allows retrieving action-specific data such as counters.
7519  * Data is gathered by special action which may be present/referenced in
7520  * more than one flow rule definition.
7521  *
7522  * see @RTE_FLOW_ACTION_TYPE_COUNT
7523  *
7524  * @param dev
7525  *   Pointer to Ethernet device structure.
7526  * @param[in] handle
7527  *   Handle for the indirect action to query.
7528  * @param[in, out] data
7529  *   Pointer to storage for the associated query data type.
7530  * @param[out] error
7531  *   Perform verbose error reporting if not NULL. PMDs initialize this
7532  *   structure in case of error only.
7533  *
7534  * @return
7535  *   0 on success, a negative errno value otherwise and rte_errno is set.
7536  */
7537 static int
7538 mlx5_action_handle_query(struct rte_eth_dev *dev,
7539 			 const struct rte_flow_action_handle *handle,
7540 			 void *data,
7541 			 struct rte_flow_error *error)
7542 {
7543 	struct rte_flow_attr attr = { .transfer = 0 };
7544 	const struct mlx5_flow_driver_ops *fops =
7545 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7546 
7547 	return flow_drv_action_query(dev, handle, data, fops, error);
7548 }
7549 
7550 /**
7551  * Destroy all indirect actions (shared RSS).
7552  *
7553  * @param dev
7554  *   Pointer to Ethernet device.
7555  *
7556  * @return
7557  *   0 on success, a negative errno value otherwise and rte_errno is set.
7558  */
7559 int
7560 mlx5_action_handle_flush(struct rte_eth_dev *dev)
7561 {
7562 	struct rte_flow_error error;
7563 	struct mlx5_priv *priv = dev->data->dev_private;
7564 	struct mlx5_shared_action_rss *shared_rss;
7565 	int ret = 0;
7566 	uint32_t idx;
7567 
7568 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7569 		      priv->rss_shared_actions, idx, shared_rss, next) {
7570 		ret |= mlx5_action_handle_destroy(dev,
7571 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
7572 	}
7573 	return ret;
7574 }
7575 
7576 #ifndef HAVE_MLX5DV_DR
7577 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7578 #else
7579 #define MLX5_DOMAIN_SYNC_FLOW \
7580 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7581 #endif
7582 
7583 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7584 {
7585 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7586 	const struct mlx5_flow_driver_ops *fops;
7587 	int ret;
7588 	struct rte_flow_attr attr = { .transfer = 0 };
7589 
7590 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7591 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7592 	if (ret > 0)
7593 		ret = -ret;
7594 	return ret;
7595 }
7596 
7597 /**
7598  * tunnel offload functionalilty is defined for DV environment only
7599  */
7600 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7601 __extension__
7602 union tunnel_offload_mark {
7603 	uint32_t val;
7604 	struct {
7605 		uint32_t app_reserve:8;
7606 		uint32_t table_id:15;
7607 		uint32_t transfer:1;
7608 		uint32_t _unused_:8;
7609 	};
7610 };
7611 
7612 static bool
7613 mlx5_access_tunnel_offload_db
7614 	(struct rte_eth_dev *dev,
7615 	 bool (*match)(struct rte_eth_dev *,
7616 		       struct mlx5_flow_tunnel *, const void *),
7617 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7618 	 void (*miss)(struct rte_eth_dev *, void *),
7619 	 void *ctx, bool lock_op);
7620 
7621 static int
7622 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7623 			     struct rte_flow *flow,
7624 			     const struct rte_flow_attr *attr,
7625 			     const struct rte_flow_action *app_actions,
7626 			     uint32_t flow_idx,
7627 			     struct tunnel_default_miss_ctx *ctx,
7628 			     struct rte_flow_error *error)
7629 {
7630 	struct mlx5_priv *priv = dev->data->dev_private;
7631 	struct mlx5_flow *dev_flow;
7632 	struct rte_flow_attr miss_attr = *attr;
7633 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7634 	const struct rte_flow_item miss_items[2] = {
7635 		{
7636 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7637 			.spec = NULL,
7638 			.last = NULL,
7639 			.mask = NULL
7640 		},
7641 		{
7642 			.type = RTE_FLOW_ITEM_TYPE_END,
7643 			.spec = NULL,
7644 			.last = NULL,
7645 			.mask = NULL
7646 		}
7647 	};
7648 	union tunnel_offload_mark mark_id;
7649 	struct rte_flow_action_mark miss_mark;
7650 	struct rte_flow_action miss_actions[3] = {
7651 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7652 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7653 	};
7654 	const struct rte_flow_action_jump *jump_data;
7655 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7656 	struct flow_grp_info grp_info = {
7657 		.external = 1,
7658 		.transfer = attr->transfer,
7659 		.fdb_def_rule = !!priv->fdb_def_rule,
7660 		.std_tbl_fix = 0,
7661 	};
7662 	int ret;
7663 
7664 	if (!attr->transfer) {
7665 		uint32_t q_size;
7666 
7667 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7668 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7669 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7670 					 0, SOCKET_ID_ANY);
7671 		if (!ctx->queue)
7672 			return rte_flow_error_set
7673 				(error, ENOMEM,
7674 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7675 				NULL, "invalid default miss RSS");
7676 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7677 		ctx->action_rss.level = 0,
7678 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7679 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7680 		ctx->action_rss.queue_num = priv->reta_idx_n,
7681 		ctx->action_rss.key = priv->rss_conf.rss_key,
7682 		ctx->action_rss.queue = ctx->queue;
7683 		if (!priv->reta_idx_n || !priv->rxqs_n)
7684 			return rte_flow_error_set
7685 				(error, EINVAL,
7686 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7687 				NULL, "invalid port configuration");
7688 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7689 			ctx->action_rss.types = 0;
7690 		for (i = 0; i != priv->reta_idx_n; ++i)
7691 			ctx->queue[i] = (*priv->reta_idx)[i];
7692 	} else {
7693 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7694 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7695 	}
7696 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7697 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7698 	jump_data = app_actions->conf;
7699 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7700 	miss_attr.group = jump_data->group;
7701 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7702 				       &flow_table, &grp_info, error);
7703 	if (ret)
7704 		return rte_flow_error_set(error, EINVAL,
7705 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7706 					  NULL, "invalid tunnel id");
7707 	mark_id.app_reserve = 0;
7708 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7709 	mark_id.transfer = !!attr->transfer;
7710 	mark_id._unused_ = 0;
7711 	miss_mark.id = mark_id.val;
7712 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7713 				    miss_items, miss_actions, flow_idx, error);
7714 	if (!dev_flow)
7715 		return -rte_errno;
7716 	dev_flow->flow = flow;
7717 	dev_flow->external = true;
7718 	dev_flow->tunnel = tunnel;
7719 	/* Subflow object was created, we must include one in the list. */
7720 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7721 		      dev_flow->handle, next);
7722 	DRV_LOG(DEBUG,
7723 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7724 		dev->data->port_id, tunnel->app_tunnel.type,
7725 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7726 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7727 				  miss_actions, error);
7728 	if (!ret)
7729 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7730 						  error);
7731 
7732 	return ret;
7733 }
7734 
7735 static const struct mlx5_flow_tbl_data_entry  *
7736 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7737 {
7738 	struct mlx5_priv *priv = dev->data->dev_private;
7739 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7740 	struct mlx5_hlist_entry *he;
7741 	union tunnel_offload_mark mbits = { .val = mark };
7742 	union mlx5_flow_tbl_key table_key = {
7743 		{
7744 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
7745 			.id = 0,
7746 			.reserved = 0,
7747 			.dummy = 0,
7748 			.is_fdb = !!mbits.transfer,
7749 			.is_egress = 0,
7750 		}
7751 	};
7752 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7753 	return he ?
7754 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7755 }
7756 
7757 static void
7758 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7759 				   struct mlx5_hlist_entry *entry)
7760 {
7761 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7762 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7763 
7764 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7765 			tunnel_flow_tbl_to_id(tte->flow_table));
7766 	mlx5_free(tte);
7767 }
7768 
7769 static int
7770 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7771 				  struct mlx5_hlist_entry *entry,
7772 				  uint64_t key, void *cb_ctx __rte_unused)
7773 {
7774 	union tunnel_tbl_key tbl = {
7775 		.val = key,
7776 	};
7777 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7778 
7779 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7780 }
7781 
7782 static struct mlx5_hlist_entry *
7783 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7784 				   void *ctx __rte_unused)
7785 {
7786 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7787 	struct tunnel_tbl_entry *tte;
7788 	union tunnel_tbl_key tbl = {
7789 		.val = key,
7790 	};
7791 
7792 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7793 			  sizeof(*tte), 0,
7794 			  SOCKET_ID_ANY);
7795 	if (!tte)
7796 		goto err;
7797 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7798 			  &tte->flow_table);
7799 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7800 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7801 			tte->flow_table);
7802 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7803 				tte->flow_table);
7804 		goto err;
7805 	} else if (!tte->flow_table) {
7806 		goto err;
7807 	}
7808 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7809 	tte->tunnel_id = tbl.tunnel_id;
7810 	tte->group = tbl.group;
7811 	return &tte->hash;
7812 err:
7813 	if (tte)
7814 		mlx5_free(tte);
7815 	return NULL;
7816 }
7817 
7818 static uint32_t
7819 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7820 				const struct mlx5_flow_tunnel *tunnel,
7821 				uint32_t group, uint32_t *table,
7822 				struct rte_flow_error *error)
7823 {
7824 	struct mlx5_hlist_entry *he;
7825 	struct tunnel_tbl_entry *tte;
7826 	union tunnel_tbl_key key = {
7827 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7828 		.group = group
7829 	};
7830 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7831 	struct mlx5_hlist *group_hash;
7832 
7833 	group_hash = tunnel ? tunnel->groups : thub->groups;
7834 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7835 	if (!he)
7836 		return rte_flow_error_set(error, EINVAL,
7837 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7838 					  NULL,
7839 					  "tunnel group index not supported");
7840 	tte = container_of(he, typeof(*tte), hash);
7841 	*table = tte->flow_table;
7842 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7843 		dev->data->port_id, key.tunnel_id, group, *table);
7844 	return 0;
7845 }
7846 
7847 static void
7848 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7849 		      struct mlx5_flow_tunnel *tunnel)
7850 {
7851 	struct mlx5_priv *priv = dev->data->dev_private;
7852 	struct mlx5_indexed_pool *ipool;
7853 
7854 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7855 		dev->data->port_id, tunnel->tunnel_id);
7856 	LIST_REMOVE(tunnel, chain);
7857 	mlx5_hlist_destroy(tunnel->groups);
7858 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7859 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7860 }
7861 
7862 static bool
7863 mlx5_access_tunnel_offload_db
7864 	(struct rte_eth_dev *dev,
7865 	 bool (*match)(struct rte_eth_dev *,
7866 		       struct mlx5_flow_tunnel *, const void *),
7867 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7868 	 void (*miss)(struct rte_eth_dev *, void *),
7869 	 void *ctx, bool lock_op)
7870 {
7871 	bool verdict = false;
7872 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7873 	struct mlx5_flow_tunnel *tunnel;
7874 
7875 	rte_spinlock_lock(&thub->sl);
7876 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7877 		verdict = match(dev, tunnel, (const void *)ctx);
7878 		if (verdict)
7879 			break;
7880 	}
7881 	if (!lock_op)
7882 		rte_spinlock_unlock(&thub->sl);
7883 	if (verdict && hit)
7884 		hit(dev, tunnel, ctx);
7885 	if (!verdict && miss)
7886 		miss(dev, ctx);
7887 	if (lock_op)
7888 		rte_spinlock_unlock(&thub->sl);
7889 
7890 	return verdict;
7891 }
7892 
7893 struct tunnel_db_find_tunnel_id_ctx {
7894 	uint32_t tunnel_id;
7895 	struct mlx5_flow_tunnel *tunnel;
7896 };
7897 
7898 static bool
7899 find_tunnel_id_match(struct rte_eth_dev *dev,
7900 		     struct mlx5_flow_tunnel *tunnel, const void *x)
7901 {
7902 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7903 
7904 	RTE_SET_USED(dev);
7905 	return tunnel->tunnel_id == ctx->tunnel_id;
7906 }
7907 
7908 static void
7909 find_tunnel_id_hit(struct rte_eth_dev *dev,
7910 		   struct mlx5_flow_tunnel *tunnel, void *x)
7911 {
7912 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7913 	RTE_SET_USED(dev);
7914 	ctx->tunnel = tunnel;
7915 }
7916 
7917 static struct mlx5_flow_tunnel *
7918 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7919 {
7920 	struct tunnel_db_find_tunnel_id_ctx ctx = {
7921 		.tunnel_id = id,
7922 	};
7923 
7924 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7925 				      find_tunnel_id_hit, NULL, &ctx, true);
7926 
7927 	return ctx.tunnel;
7928 }
7929 
7930 static struct mlx5_flow_tunnel *
7931 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7932 			  const struct rte_flow_tunnel *app_tunnel)
7933 {
7934 	struct mlx5_priv *priv = dev->data->dev_private;
7935 	struct mlx5_indexed_pool *ipool;
7936 	struct mlx5_flow_tunnel *tunnel;
7937 	uint32_t id;
7938 
7939 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7940 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7941 	if (!tunnel)
7942 		return NULL;
7943 	if (id >= MLX5_MAX_TUNNELS) {
7944 		mlx5_ipool_free(ipool, id);
7945 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7946 		return NULL;
7947 	}
7948 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7949 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7950 					   mlx5_flow_tunnel_grp2tbl_match_cb,
7951 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7952 	if (!tunnel->groups) {
7953 		mlx5_ipool_free(ipool, id);
7954 		return NULL;
7955 	}
7956 	tunnel->groups->ctx = priv->sh;
7957 	/* initiate new PMD tunnel */
7958 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7959 	tunnel->tunnel_id = id;
7960 	tunnel->action.type = (typeof(tunnel->action.type))
7961 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7962 	tunnel->action.conf = tunnel;
7963 	tunnel->item.type = (typeof(tunnel->item.type))
7964 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7965 	tunnel->item.spec = tunnel;
7966 	tunnel->item.last = NULL;
7967 	tunnel->item.mask = NULL;
7968 
7969 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7970 		dev->data->port_id, tunnel->tunnel_id);
7971 
7972 	return tunnel;
7973 }
7974 
7975 struct tunnel_db_get_tunnel_ctx {
7976 	const struct rte_flow_tunnel *app_tunnel;
7977 	struct mlx5_flow_tunnel *tunnel;
7978 };
7979 
7980 static bool get_tunnel_match(struct rte_eth_dev *dev,
7981 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7982 {
7983 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
7984 
7985 	RTE_SET_USED(dev);
7986 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7987 		       sizeof(*ctx->app_tunnel));
7988 }
7989 
7990 static void get_tunnel_hit(struct rte_eth_dev *dev,
7991 			   struct mlx5_flow_tunnel *tunnel, void *x)
7992 {
7993 	/* called under tunnel spinlock protection */
7994 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7995 
7996 	RTE_SET_USED(dev);
7997 	tunnel->refctn++;
7998 	ctx->tunnel = tunnel;
7999 }
8000 
8001 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8002 {
8003 	/* called under tunnel spinlock protection */
8004 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8005 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8006 
8007 	rte_spinlock_unlock(&thub->sl);
8008 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8009 	rte_spinlock_lock(&thub->sl);
8010 	if (ctx->tunnel) {
8011 		ctx->tunnel->refctn = 1;
8012 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8013 	}
8014 }
8015 
8016 
8017 static int
8018 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8019 		     const struct rte_flow_tunnel *app_tunnel,
8020 		     struct mlx5_flow_tunnel **tunnel)
8021 {
8022 	struct tunnel_db_get_tunnel_ctx ctx = {
8023 		.app_tunnel = app_tunnel,
8024 	};
8025 
8026 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8027 				      get_tunnel_miss, &ctx, true);
8028 	*tunnel = ctx.tunnel;
8029 	return ctx.tunnel ? 0 : -ENOMEM;
8030 }
8031 
8032 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8033 {
8034 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8035 
8036 	if (!thub)
8037 		return;
8038 	if (!LIST_EMPTY(&thub->tunnels))
8039 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
8040 	mlx5_hlist_destroy(thub->groups);
8041 	mlx5_free(thub);
8042 }
8043 
8044 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8045 {
8046 	int err;
8047 	struct mlx5_flow_tunnel_hub *thub;
8048 
8049 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8050 			   0, SOCKET_ID_ANY);
8051 	if (!thub)
8052 		return -ENOMEM;
8053 	LIST_INIT(&thub->tunnels);
8054 	rte_spinlock_init(&thub->sl);
8055 	thub->groups = mlx5_hlist_create("flow groups",
8056 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
8057 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8058 					 mlx5_flow_tunnel_grp2tbl_match_cb,
8059 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
8060 	if (!thub->groups) {
8061 		err = -rte_errno;
8062 		goto err;
8063 	}
8064 	thub->groups->ctx = sh;
8065 	sh->tunnel_hub = thub;
8066 
8067 	return 0;
8068 
8069 err:
8070 	if (thub->groups)
8071 		mlx5_hlist_destroy(thub->groups);
8072 	if (thub)
8073 		mlx5_free(thub);
8074 	return err;
8075 }
8076 
8077 static inline bool
8078 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8079 			  struct rte_flow_tunnel *tunnel,
8080 			  const char *err_msg)
8081 {
8082 	err_msg = NULL;
8083 	if (!is_tunnel_offload_active(dev)) {
8084 		err_msg = "tunnel offload was not activated";
8085 		goto out;
8086 	} else if (!tunnel) {
8087 		err_msg = "no application tunnel";
8088 		goto out;
8089 	}
8090 
8091 	switch (tunnel->type) {
8092 	default:
8093 		err_msg = "unsupported tunnel type";
8094 		goto out;
8095 	case RTE_FLOW_ITEM_TYPE_VXLAN:
8096 		break;
8097 	}
8098 
8099 out:
8100 	return !err_msg;
8101 }
8102 
8103 static int
8104 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8105 		    struct rte_flow_tunnel *app_tunnel,
8106 		    struct rte_flow_action **actions,
8107 		    uint32_t *num_of_actions,
8108 		    struct rte_flow_error *error)
8109 {
8110 	int ret;
8111 	struct mlx5_flow_tunnel *tunnel;
8112 	const char *err_msg = NULL;
8113 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8114 
8115 	if (!verdict)
8116 		return rte_flow_error_set(error, EINVAL,
8117 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8118 					  err_msg);
8119 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8120 	if (ret < 0) {
8121 		return rte_flow_error_set(error, ret,
8122 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8123 					  "failed to initialize pmd tunnel");
8124 	}
8125 	*actions = &tunnel->action;
8126 	*num_of_actions = 1;
8127 	return 0;
8128 }
8129 
8130 static int
8131 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8132 		       struct rte_flow_tunnel *app_tunnel,
8133 		       struct rte_flow_item **items,
8134 		       uint32_t *num_of_items,
8135 		       struct rte_flow_error *error)
8136 {
8137 	int ret;
8138 	struct mlx5_flow_tunnel *tunnel;
8139 	const char *err_msg = NULL;
8140 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8141 
8142 	if (!verdict)
8143 		return rte_flow_error_set(error, EINVAL,
8144 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8145 					  err_msg);
8146 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8147 	if (ret < 0) {
8148 		return rte_flow_error_set(error, ret,
8149 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8150 					  "failed to initialize pmd tunnel");
8151 	}
8152 	*items = &tunnel->item;
8153 	*num_of_items = 1;
8154 	return 0;
8155 }
8156 
8157 struct tunnel_db_element_release_ctx {
8158 	struct rte_flow_item *items;
8159 	struct rte_flow_action *actions;
8160 	uint32_t num_elements;
8161 	struct rte_flow_error *error;
8162 	int ret;
8163 };
8164 
8165 static bool
8166 tunnel_element_release_match(struct rte_eth_dev *dev,
8167 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8168 {
8169 	const struct tunnel_db_element_release_ctx *ctx = x;
8170 
8171 	RTE_SET_USED(dev);
8172 	if (ctx->num_elements != 1)
8173 		return false;
8174 	else if (ctx->items)
8175 		return ctx->items == &tunnel->item;
8176 	else if (ctx->actions)
8177 		return ctx->actions == &tunnel->action;
8178 
8179 	return false;
8180 }
8181 
8182 static void
8183 tunnel_element_release_hit(struct rte_eth_dev *dev,
8184 			   struct mlx5_flow_tunnel *tunnel, void *x)
8185 {
8186 	struct tunnel_db_element_release_ctx *ctx = x;
8187 	ctx->ret = 0;
8188 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8189 		mlx5_flow_tunnel_free(dev, tunnel);
8190 }
8191 
8192 static void
8193 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8194 {
8195 	struct tunnel_db_element_release_ctx *ctx = x;
8196 	RTE_SET_USED(dev);
8197 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8198 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8199 				      "invalid argument");
8200 }
8201 
8202 static int
8203 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8204 		       struct rte_flow_item *pmd_items,
8205 		       uint32_t num_items, struct rte_flow_error *err)
8206 {
8207 	struct tunnel_db_element_release_ctx ctx = {
8208 		.items = pmd_items,
8209 		.actions = NULL,
8210 		.num_elements = num_items,
8211 		.error = err,
8212 	};
8213 
8214 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8215 				      tunnel_element_release_hit,
8216 				      tunnel_element_release_miss, &ctx, false);
8217 
8218 	return ctx.ret;
8219 }
8220 
8221 static int
8222 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8223 			 struct rte_flow_action *pmd_actions,
8224 			 uint32_t num_actions, struct rte_flow_error *err)
8225 {
8226 	struct tunnel_db_element_release_ctx ctx = {
8227 		.items = NULL,
8228 		.actions = pmd_actions,
8229 		.num_elements = num_actions,
8230 		.error = err,
8231 	};
8232 
8233 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8234 				      tunnel_element_release_hit,
8235 				      tunnel_element_release_miss, &ctx, false);
8236 
8237 	return ctx.ret;
8238 }
8239 
8240 static int
8241 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8242 				  struct rte_mbuf *m,
8243 				  struct rte_flow_restore_info *info,
8244 				  struct rte_flow_error *err)
8245 {
8246 	uint64_t ol_flags = m->ol_flags;
8247 	const struct mlx5_flow_tbl_data_entry *tble;
8248 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8249 
8250 	if (!is_tunnel_offload_active(dev)) {
8251 		info->flags = 0;
8252 		return 0;
8253 	}
8254 
8255 	if ((ol_flags & mask) != mask)
8256 		goto err;
8257 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8258 	if (!tble) {
8259 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8260 			dev->data->port_id, m->hash.fdir.hi);
8261 		goto err;
8262 	}
8263 	MLX5_ASSERT(tble->tunnel);
8264 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8265 	info->group_id = tble->group_id;
8266 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8267 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8268 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8269 
8270 	return 0;
8271 
8272 err:
8273 	return rte_flow_error_set(err, EINVAL,
8274 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8275 				  "failed to get restore info");
8276 }
8277 
8278 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8279 static int
8280 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8281 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8282 			   __rte_unused struct rte_flow_action **actions,
8283 			   __rte_unused uint32_t *num_of_actions,
8284 			   __rte_unused struct rte_flow_error *error)
8285 {
8286 	return -ENOTSUP;
8287 }
8288 
8289 static int
8290 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8291 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8292 		       __rte_unused struct rte_flow_item **items,
8293 		       __rte_unused uint32_t *num_of_items,
8294 		       __rte_unused struct rte_flow_error *error)
8295 {
8296 	return -ENOTSUP;
8297 }
8298 
8299 static int
8300 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8301 			      __rte_unused struct rte_flow_item *pmd_items,
8302 			      __rte_unused uint32_t num_items,
8303 			      __rte_unused struct rte_flow_error *err)
8304 {
8305 	return -ENOTSUP;
8306 }
8307 
8308 static int
8309 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8310 				__rte_unused struct rte_flow_action *pmd_action,
8311 				__rte_unused uint32_t num_actions,
8312 				__rte_unused struct rte_flow_error *err)
8313 {
8314 	return -ENOTSUP;
8315 }
8316 
8317 static int
8318 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8319 				  __rte_unused struct rte_mbuf *m,
8320 				  __rte_unused struct rte_flow_restore_info *i,
8321 				  __rte_unused struct rte_flow_error *err)
8322 {
8323 	return -ENOTSUP;
8324 }
8325 
8326 static int
8327 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8328 			     __rte_unused struct rte_flow *flow,
8329 			     __rte_unused const struct rte_flow_attr *attr,
8330 			     __rte_unused const struct rte_flow_action *actions,
8331 			     __rte_unused uint32_t flow_idx,
8332 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8333 			     __rte_unused struct rte_flow_error *error)
8334 {
8335 	return -ENOTSUP;
8336 }
8337 
8338 static struct mlx5_flow_tunnel *
8339 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8340 		    __rte_unused uint32_t id)
8341 {
8342 	return NULL;
8343 }
8344 
8345 static void
8346 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8347 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8348 {
8349 }
8350 
8351 static uint32_t
8352 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8353 				__rte_unused const struct mlx5_flow_tunnel *t,
8354 				__rte_unused uint32_t group,
8355 				__rte_unused uint32_t *table,
8356 				struct rte_flow_error *error)
8357 {
8358 	return rte_flow_error_set(error, ENOTSUP,
8359 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8360 				  "tunnel offload requires DV support");
8361 }
8362 
8363 void
8364 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8365 			__rte_unused  uint16_t port_id)
8366 {
8367 }
8368 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8369