xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b752fb4d)
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 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  * Flow driver find RSS policy tbl API. This abstracts calling driver
3348  * specific functions. Parent flow (rte_flow) should have driver
3349  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3350  *
3351  * @param[in] dev
3352  *   Pointer to Ethernet device.
3353  * @param[in, out] flow
3354  *   Pointer to flow structure.
3355  * @param[in] policy
3356  *   Pointer to meter policy table.
3357  * @param[in] rss_desc
3358  *   Pointer to rss_desc
3359  */
3360 static struct mlx5_flow_meter_sub_policy *
3361 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3362 		struct rte_flow *flow,
3363 		struct mlx5_flow_meter_policy *policy,
3364 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3365 {
3366 	const struct mlx5_flow_driver_ops *fops;
3367 	enum mlx5_flow_drv_type type = flow->drv_type;
3368 
3369 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3370 	fops = flow_get_drv_ops(type);
3371 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3372 }
3373 
3374 /**
3375  * Get RSS action from the action list.
3376  *
3377  * @param[in] dev
3378  *   Pointer to Ethernet device.
3379  * @param[in] actions
3380  *   Pointer to the list of actions.
3381  * @param[in] flow
3382  *   Parent flow structure pointer.
3383  *
3384  * @return
3385  *   Pointer to the RSS action if exist, else return NULL.
3386  */
3387 static const struct rte_flow_action_rss*
3388 flow_get_rss_action(struct rte_eth_dev *dev,
3389 		    const struct rte_flow_action actions[])
3390 {
3391 	struct mlx5_priv *priv = dev->data->dev_private;
3392 	const struct rte_flow_action_rss *rss = NULL;
3393 
3394 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3395 		switch (actions->type) {
3396 		case RTE_FLOW_ACTION_TYPE_RSS:
3397 			rss = actions->conf;
3398 			break;
3399 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3400 		{
3401 			const struct rte_flow_action_sample *sample =
3402 								actions->conf;
3403 			const struct rte_flow_action *act = sample->actions;
3404 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3405 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3406 					rss = act->conf;
3407 			break;
3408 		}
3409 		case RTE_FLOW_ACTION_TYPE_METER:
3410 		{
3411 			uint32_t mtr_idx;
3412 			struct mlx5_flow_meter_info *fm;
3413 			struct mlx5_flow_meter_policy *policy;
3414 			const struct rte_flow_action_meter *mtr = actions->conf;
3415 
3416 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3417 			if (fm) {
3418 				policy = mlx5_flow_meter_policy_find(dev,
3419 						fm->policy_id, NULL);
3420 				if (policy && policy->is_rss)
3421 					rss =
3422 				policy->act_cnt[RTE_COLOR_GREEN].rss->conf;
3423 			}
3424 			break;
3425 		}
3426 		default:
3427 			break;
3428 		}
3429 	}
3430 	return rss;
3431 }
3432 
3433 /**
3434  * Get ASO age action by index.
3435  *
3436  * @param[in] dev
3437  *   Pointer to the Ethernet device structure.
3438  * @param[in] age_idx
3439  *   Index to the ASO age action.
3440  *
3441  * @return
3442  *   The specified ASO age action.
3443  */
3444 struct mlx5_aso_age_action*
3445 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3446 {
3447 	uint16_t pool_idx = age_idx & UINT16_MAX;
3448 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3449 	struct mlx5_priv *priv = dev->data->dev_private;
3450 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3451 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3452 
3453 	return &pool->actions[offset - 1];
3454 }
3455 
3456 /* maps indirect action to translated direct in some actions array */
3457 struct mlx5_translated_action_handle {
3458 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3459 	int index; /**< Index in related array of rte_flow_action. */
3460 };
3461 
3462 /**
3463  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3464  * direct action if translation possible.
3465  * This functionality used to run same execution path for both direct and
3466  * indirect actions on flow create. All necessary preparations for indirect
3467  * action handling should be performed on *handle* actions list returned
3468  * from this call.
3469  *
3470  * @param[in] dev
3471  *   Pointer to Ethernet device.
3472  * @param[in] actions
3473  *   List of actions to translate.
3474  * @param[out] handle
3475  *   List to store translated indirect action object handles.
3476  * @param[in, out] indir_n
3477  *   Size of *handle* array. On return should be updated with number of
3478  *   indirect actions retrieved from the *actions* list.
3479  * @param[out] translated_actions
3480  *   List of actions where all indirect actions were translated to direct
3481  *   if possible. NULL if no translation took place.
3482  * @param[out] error
3483  *   Pointer to the error structure.
3484  *
3485  * @return
3486  *   0 on success, a negative errno value otherwise and rte_errno is set.
3487  */
3488 static int
3489 flow_action_handles_translate(struct rte_eth_dev *dev,
3490 			      const struct rte_flow_action actions[],
3491 			      struct mlx5_translated_action_handle *handle,
3492 			      int *indir_n,
3493 			      struct rte_flow_action **translated_actions,
3494 			      struct rte_flow_error *error)
3495 {
3496 	struct mlx5_priv *priv = dev->data->dev_private;
3497 	struct rte_flow_action *translated = NULL;
3498 	size_t actions_size;
3499 	int n;
3500 	int copied_n = 0;
3501 	struct mlx5_translated_action_handle *handle_end = NULL;
3502 
3503 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3504 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3505 			continue;
3506 		if (copied_n == *indir_n) {
3507 			return rte_flow_error_set
3508 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3509 				 NULL, "too many shared actions");
3510 		}
3511 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3512 			   sizeof(actions[n].conf));
3513 		handle[copied_n].index = n;
3514 		copied_n++;
3515 	}
3516 	n++;
3517 	*indir_n = copied_n;
3518 	if (!copied_n)
3519 		return 0;
3520 	actions_size = sizeof(struct rte_flow_action) * n;
3521 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3522 	if (!translated) {
3523 		rte_errno = ENOMEM;
3524 		return -ENOMEM;
3525 	}
3526 	memcpy(translated, actions, actions_size);
3527 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3528 		struct mlx5_shared_action_rss *shared_rss;
3529 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3530 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3531 		uint32_t idx = act_idx &
3532 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3533 
3534 		switch (type) {
3535 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3536 			shared_rss = mlx5_ipool_get
3537 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3538 			translated[handle->index].type =
3539 				RTE_FLOW_ACTION_TYPE_RSS;
3540 			translated[handle->index].conf =
3541 				&shared_rss->origin;
3542 			break;
3543 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3544 			if (priv->sh->flow_hit_aso_en) {
3545 				translated[handle->index].type =
3546 					(enum rte_flow_action_type)
3547 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3548 				translated[handle->index].conf =
3549 							 (void *)(uintptr_t)idx;
3550 				break;
3551 			}
3552 			/* Fall-through */
3553 		default:
3554 			mlx5_free(translated);
3555 			return rte_flow_error_set
3556 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3557 				 NULL, "invalid indirect action type");
3558 		}
3559 	}
3560 	*translated_actions = translated;
3561 	return 0;
3562 }
3563 
3564 /**
3565  * Get Shared RSS action from the action list.
3566  *
3567  * @param[in] dev
3568  *   Pointer to Ethernet device.
3569  * @param[in] shared
3570  *   Pointer to the list of actions.
3571  * @param[in] shared_n
3572  *   Actions list length.
3573  *
3574  * @return
3575  *   The MLX5 RSS action ID if exists, otherwise return 0.
3576  */
3577 static uint32_t
3578 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3579 			   struct mlx5_translated_action_handle *handle,
3580 			   int shared_n)
3581 {
3582 	struct mlx5_translated_action_handle *handle_end;
3583 	struct mlx5_priv *priv = dev->data->dev_private;
3584 	struct mlx5_shared_action_rss *shared_rss;
3585 
3586 
3587 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3588 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3589 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3590 		uint32_t idx = act_idx &
3591 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3592 		switch (type) {
3593 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3594 			shared_rss = mlx5_ipool_get
3595 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3596 									   idx);
3597 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3598 					   __ATOMIC_RELAXED);
3599 			return idx;
3600 		default:
3601 			break;
3602 		}
3603 	}
3604 	return 0;
3605 }
3606 
3607 static unsigned int
3608 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3609 {
3610 	const struct rte_flow_item *item;
3611 	unsigned int has_vlan = 0;
3612 
3613 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3614 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3615 			has_vlan = 1;
3616 			break;
3617 		}
3618 	}
3619 	if (has_vlan)
3620 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3621 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3622 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3623 			       MLX5_EXPANSION_ROOT_OUTER;
3624 }
3625 
3626 /**
3627  *  Get layer flags from the prefix flow.
3628  *
3629  *  Some flows may be split to several subflows, the prefix subflow gets the
3630  *  match items and the suffix sub flow gets the actions.
3631  *  Some actions need the user defined match item flags to get the detail for
3632  *  the action.
3633  *  This function helps the suffix flow to get the item layer flags from prefix
3634  *  subflow.
3635  *
3636  * @param[in] dev_flow
3637  *   Pointer the created preifx subflow.
3638  *
3639  * @return
3640  *   The layers get from prefix subflow.
3641  */
3642 static inline uint64_t
3643 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3644 {
3645 	uint64_t layers = 0;
3646 
3647 	/*
3648 	 * Layers bits could be localization, but usually the compiler will
3649 	 * help to do the optimization work for source code.
3650 	 * If no decap actions, use the layers directly.
3651 	 */
3652 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3653 		return dev_flow->handle->layers;
3654 	/* Convert L3 layers with decap action. */
3655 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3656 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3657 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3658 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3659 	/* Convert L4 layers with decap action.  */
3660 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3661 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3662 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3663 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3664 	return layers;
3665 }
3666 
3667 /**
3668  * Get metadata split action information.
3669  *
3670  * @param[in] actions
3671  *   Pointer to the list of actions.
3672  * @param[out] qrss
3673  *   Pointer to the return pointer.
3674  * @param[out] qrss_type
3675  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3676  *   if no QUEUE/RSS is found.
3677  * @param[out] encap_idx
3678  *   Pointer to the index of the encap action if exists, otherwise the last
3679  *   action index.
3680  *
3681  * @return
3682  *   Total number of actions.
3683  */
3684 static int
3685 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3686 				       const struct rte_flow_action **qrss,
3687 				       int *encap_idx)
3688 {
3689 	const struct rte_flow_action_raw_encap *raw_encap;
3690 	int actions_n = 0;
3691 	int raw_decap_idx = -1;
3692 
3693 	*encap_idx = -1;
3694 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3695 		switch (actions->type) {
3696 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3697 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3698 			*encap_idx = actions_n;
3699 			break;
3700 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3701 			raw_decap_idx = actions_n;
3702 			break;
3703 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3704 			raw_encap = actions->conf;
3705 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3706 				*encap_idx = raw_decap_idx != -1 ?
3707 						      raw_decap_idx : actions_n;
3708 			break;
3709 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3710 		case RTE_FLOW_ACTION_TYPE_RSS:
3711 			*qrss = actions;
3712 			break;
3713 		default:
3714 			break;
3715 		}
3716 		actions_n++;
3717 	}
3718 	if (*encap_idx == -1)
3719 		*encap_idx = actions_n;
3720 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3721 	return actions_n + 1;
3722 }
3723 
3724 /**
3725  * Check if the action will change packet.
3726  *
3727  * @param dev
3728  *   Pointer to Ethernet device.
3729  * @param[in] type
3730  *   action type.
3731  *
3732  * @return
3733  *   true if action will change packet, false otherwise.
3734  */
3735 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
3736 					  enum rte_flow_action_type type)
3737 {
3738 	struct mlx5_priv *priv = dev->data->dev_private;
3739 
3740 	switch (type) {
3741 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3742 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3743 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3744 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3745 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3746 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3747 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3748 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3749 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3750 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
3751 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
3752 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
3753 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
3754 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
3755 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
3756 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
3757 	case RTE_FLOW_ACTION_TYPE_SET_META:
3758 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
3759 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3760 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3761 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3762 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3763 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3764 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3765 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3766 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3767 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3768 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3769 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
3770 		return true;
3771 	case RTE_FLOW_ACTION_TYPE_FLAG:
3772 	case RTE_FLOW_ACTION_TYPE_MARK:
3773 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
3774 			return true;
3775 		else
3776 			return false;
3777 	default:
3778 		return false;
3779 	}
3780 }
3781 
3782 /**
3783  * Check meter action from the action list.
3784  *
3785  * @param dev
3786  *   Pointer to Ethernet device.
3787  * @param[in] actions
3788  *   Pointer to the list of actions.
3789  * @param[out] has_mtr
3790  *   Pointer to the meter exist flag.
3791  * @param[out] has_modify
3792  *   Pointer to the flag showing there's packet change action.
3793  * @param[out] meter_id
3794  *   Pointer to the meter id.
3795  *
3796  * @return
3797  *   Total number of actions.
3798  */
3799 static int
3800 flow_check_meter_action(struct rte_eth_dev *dev,
3801 			const struct rte_flow_action actions[],
3802 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
3803 {
3804 	const struct rte_flow_action_meter *mtr = NULL;
3805 	int actions_n = 0;
3806 
3807 	MLX5_ASSERT(has_mtr);
3808 	*has_mtr = false;
3809 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3810 		switch (actions->type) {
3811 		case RTE_FLOW_ACTION_TYPE_METER:
3812 			mtr = actions->conf;
3813 			*meter_id = mtr->mtr_id;
3814 			*has_mtr = true;
3815 			break;
3816 		default:
3817 			break;
3818 		}
3819 		if (!*has_mtr)
3820 			*has_modify |= flow_check_modify_action_type(dev,
3821 								actions->type);
3822 		actions_n++;
3823 	}
3824 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3825 	return actions_n + 1;
3826 }
3827 
3828 /**
3829  * Check if the flow should be split due to hairpin.
3830  * The reason for the split is that in current HW we can't
3831  * support encap and push-vlan on Rx, so if a flow contains
3832  * these actions we move it to Tx.
3833  *
3834  * @param dev
3835  *   Pointer to Ethernet device.
3836  * @param[in] attr
3837  *   Flow rule attributes.
3838  * @param[in] actions
3839  *   Associated actions (list terminated by the END action).
3840  *
3841  * @return
3842  *   > 0 the number of actions and the flow should be split,
3843  *   0 when no split required.
3844  */
3845 static int
3846 flow_check_hairpin_split(struct rte_eth_dev *dev,
3847 			 const struct rte_flow_attr *attr,
3848 			 const struct rte_flow_action actions[])
3849 {
3850 	int queue_action = 0;
3851 	int action_n = 0;
3852 	int split = 0;
3853 	const struct rte_flow_action_queue *queue;
3854 	const struct rte_flow_action_rss *rss;
3855 	const struct rte_flow_action_raw_encap *raw_encap;
3856 	const struct rte_eth_hairpin_conf *conf;
3857 
3858 	if (!attr->ingress)
3859 		return 0;
3860 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3861 		switch (actions->type) {
3862 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3863 			queue = actions->conf;
3864 			if (queue == NULL)
3865 				return 0;
3866 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3867 			if (conf == NULL || conf->tx_explicit != 0)
3868 				return 0;
3869 			queue_action = 1;
3870 			action_n++;
3871 			break;
3872 		case RTE_FLOW_ACTION_TYPE_RSS:
3873 			rss = actions->conf;
3874 			if (rss == NULL || rss->queue_num == 0)
3875 				return 0;
3876 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3877 			if (conf == NULL || conf->tx_explicit != 0)
3878 				return 0;
3879 			queue_action = 1;
3880 			action_n++;
3881 			break;
3882 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3883 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3884 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3885 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3886 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3887 			split++;
3888 			action_n++;
3889 			break;
3890 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3891 			raw_encap = actions->conf;
3892 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3893 				split++;
3894 			action_n++;
3895 			break;
3896 		default:
3897 			action_n++;
3898 			break;
3899 		}
3900 	}
3901 	if (split && queue_action)
3902 		return action_n;
3903 	return 0;
3904 }
3905 
3906 /* Declare flow create/destroy prototype in advance. */
3907 static uint32_t
3908 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3909 		 const struct rte_flow_attr *attr,
3910 		 const struct rte_flow_item items[],
3911 		 const struct rte_flow_action actions[],
3912 		 bool external, struct rte_flow_error *error);
3913 
3914 static void
3915 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3916 		  uint32_t flow_idx);
3917 
3918 int
3919 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3920 		      struct mlx5_hlist_entry *entry,
3921 		      uint64_t key, void *cb_ctx __rte_unused)
3922 {
3923 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3924 		container_of(entry, typeof(*mcp_res), hlist_ent);
3925 
3926 	return mcp_res->mark_id != key;
3927 }
3928 
3929 struct mlx5_hlist_entry *
3930 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3931 		       void *cb_ctx)
3932 {
3933 	struct rte_eth_dev *dev = list->ctx;
3934 	struct mlx5_priv *priv = dev->data->dev_private;
3935 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3936 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3937 	struct rte_flow_error *error = ctx->error;
3938 	uint32_t idx = 0;
3939 	int ret;
3940 	uint32_t mark_id = key;
3941 	struct rte_flow_attr attr = {
3942 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3943 		.ingress = 1,
3944 	};
3945 	struct mlx5_rte_flow_item_tag tag_spec = {
3946 		.data = mark_id,
3947 	};
3948 	struct rte_flow_item items[] = {
3949 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3950 	};
3951 	struct rte_flow_action_mark ftag = {
3952 		.id = mark_id,
3953 	};
3954 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3955 		.dst = REG_B,
3956 		.src = REG_NON,
3957 	};
3958 	struct rte_flow_action_jump jump = {
3959 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3960 	};
3961 	struct rte_flow_action actions[] = {
3962 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3963 	};
3964 
3965 	/* Fill the register fileds in the flow. */
3966 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3967 	if (ret < 0)
3968 		return NULL;
3969 	tag_spec.id = ret;
3970 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3971 	if (ret < 0)
3972 		return NULL;
3973 	cp_mreg.src = ret;
3974 	/* Provide the full width of FLAG specific value. */
3975 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3976 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3977 	/* Build a new flow. */
3978 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3979 		items[0] = (struct rte_flow_item){
3980 			.type = (enum rte_flow_item_type)
3981 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3982 			.spec = &tag_spec,
3983 		};
3984 		items[1] = (struct rte_flow_item){
3985 			.type = RTE_FLOW_ITEM_TYPE_END,
3986 		};
3987 		actions[0] = (struct rte_flow_action){
3988 			.type = (enum rte_flow_action_type)
3989 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3990 			.conf = &ftag,
3991 		};
3992 		actions[1] = (struct rte_flow_action){
3993 			.type = (enum rte_flow_action_type)
3994 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3995 			.conf = &cp_mreg,
3996 		};
3997 		actions[2] = (struct rte_flow_action){
3998 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3999 			.conf = &jump,
4000 		};
4001 		actions[3] = (struct rte_flow_action){
4002 			.type = RTE_FLOW_ACTION_TYPE_END,
4003 		};
4004 	} else {
4005 		/* Default rule, wildcard match. */
4006 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4007 		items[0] = (struct rte_flow_item){
4008 			.type = RTE_FLOW_ITEM_TYPE_END,
4009 		};
4010 		actions[0] = (struct rte_flow_action){
4011 			.type = (enum rte_flow_action_type)
4012 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4013 			.conf = &cp_mreg,
4014 		};
4015 		actions[1] = (struct rte_flow_action){
4016 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4017 			.conf = &jump,
4018 		};
4019 		actions[2] = (struct rte_flow_action){
4020 			.type = RTE_FLOW_ACTION_TYPE_END,
4021 		};
4022 	}
4023 	/* Build a new entry. */
4024 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4025 	if (!mcp_res) {
4026 		rte_errno = ENOMEM;
4027 		return NULL;
4028 	}
4029 	mcp_res->idx = idx;
4030 	mcp_res->mark_id = mark_id;
4031 	/*
4032 	 * The copy Flows are not included in any list. There
4033 	 * ones are referenced from other Flows and can not
4034 	 * be applied, removed, deleted in ardbitrary order
4035 	 * by list traversing.
4036 	 */
4037 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
4038 					 actions, false, error);
4039 	if (!mcp_res->rix_flow) {
4040 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4041 		return NULL;
4042 	}
4043 	return &mcp_res->hlist_ent;
4044 }
4045 
4046 /**
4047  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4048  *
4049  * As mark_id is unique, if there's already a registered flow for the mark_id,
4050  * return by increasing the reference counter of the resource. Otherwise, create
4051  * the resource (mcp_res) and flow.
4052  *
4053  * Flow looks like,
4054  *   - If ingress port is ANY and reg_c[1] is mark_id,
4055  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4056  *
4057  * For default flow (zero mark_id), flow is like,
4058  *   - If ingress port is ANY,
4059  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4060  *
4061  * @param dev
4062  *   Pointer to Ethernet device.
4063  * @param mark_id
4064  *   ID of MARK action, zero means default flow for META.
4065  * @param[out] error
4066  *   Perform verbose error reporting if not NULL.
4067  *
4068  * @return
4069  *   Associated resource on success, NULL otherwise and rte_errno is set.
4070  */
4071 static struct mlx5_flow_mreg_copy_resource *
4072 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4073 			  struct rte_flow_error *error)
4074 {
4075 	struct mlx5_priv *priv = dev->data->dev_private;
4076 	struct mlx5_hlist_entry *entry;
4077 	struct mlx5_flow_cb_ctx ctx = {
4078 		.dev = dev,
4079 		.error = error,
4080 	};
4081 
4082 	/* Check if already registered. */
4083 	MLX5_ASSERT(priv->mreg_cp_tbl);
4084 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4085 	if (!entry)
4086 		return NULL;
4087 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4088 			    hlist_ent);
4089 }
4090 
4091 void
4092 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
4093 {
4094 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4095 		container_of(entry, typeof(*mcp_res), hlist_ent);
4096 	struct rte_eth_dev *dev = list->ctx;
4097 	struct mlx5_priv *priv = dev->data->dev_private;
4098 
4099 	MLX5_ASSERT(mcp_res->rix_flow);
4100 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
4101 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4102 }
4103 
4104 /**
4105  * Release flow in RX_CP_TBL.
4106  *
4107  * @param dev
4108  *   Pointer to Ethernet device.
4109  * @flow
4110  *   Parent flow for wich copying is provided.
4111  */
4112 static void
4113 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4114 			  struct rte_flow *flow)
4115 {
4116 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4117 	struct mlx5_priv *priv = dev->data->dev_private;
4118 
4119 	if (!flow->rix_mreg_copy)
4120 		return;
4121 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4122 				 flow->rix_mreg_copy);
4123 	if (!mcp_res || !priv->mreg_cp_tbl)
4124 		return;
4125 	MLX5_ASSERT(mcp_res->rix_flow);
4126 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4127 	flow->rix_mreg_copy = 0;
4128 }
4129 
4130 /**
4131  * Remove the default copy action from RX_CP_TBL.
4132  *
4133  * This functions is called in the mlx5_dev_start(). No thread safe
4134  * is guaranteed.
4135  *
4136  * @param dev
4137  *   Pointer to Ethernet device.
4138  */
4139 static void
4140 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4141 {
4142 	struct mlx5_hlist_entry *entry;
4143 	struct mlx5_priv *priv = dev->data->dev_private;
4144 
4145 	/* Check if default flow is registered. */
4146 	if (!priv->mreg_cp_tbl)
4147 		return;
4148 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4149 				  MLX5_DEFAULT_COPY_ID, NULL);
4150 	if (!entry)
4151 		return;
4152 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4153 }
4154 
4155 /**
4156  * Add the default copy action in in RX_CP_TBL.
4157  *
4158  * This functions is called in the mlx5_dev_start(). No thread safe
4159  * is guaranteed.
4160  *
4161  * @param dev
4162  *   Pointer to Ethernet device.
4163  * @param[out] error
4164  *   Perform verbose error reporting if not NULL.
4165  *
4166  * @return
4167  *   0 for success, negative value otherwise and rte_errno is set.
4168  */
4169 static int
4170 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4171 				  struct rte_flow_error *error)
4172 {
4173 	struct mlx5_priv *priv = dev->data->dev_private;
4174 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4175 
4176 	/* Check whether extensive metadata feature is engaged. */
4177 	if (!priv->config.dv_flow_en ||
4178 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4179 	    !mlx5_flow_ext_mreg_supported(dev) ||
4180 	    !priv->sh->dv_regc0_mask)
4181 		return 0;
4182 	/*
4183 	 * Add default mreg copy flow may be called multiple time, but
4184 	 * only be called once in stop. Avoid register it twice.
4185 	 */
4186 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4187 		return 0;
4188 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4189 	if (!mcp_res)
4190 		return -rte_errno;
4191 	return 0;
4192 }
4193 
4194 /**
4195  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4196  *
4197  * All the flow having Q/RSS action should be split by
4198  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4199  * performs the following,
4200  *   - CQE->flow_tag := reg_c[1] (MARK)
4201  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4202  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4203  * but there should be a flow per each MARK ID set by MARK action.
4204  *
4205  * For the aforementioned reason, if there's a MARK action in flow's action
4206  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4207  * the MARK ID to CQE's flow_tag like,
4208  *   - If reg_c[1] is mark_id,
4209  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4210  *
4211  * For SET_META action which stores value in reg_c[0], as the destination is
4212  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4213  * MARK ID means the default flow. The default flow looks like,
4214  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4215  *
4216  * @param dev
4217  *   Pointer to Ethernet device.
4218  * @param flow
4219  *   Pointer to flow structure.
4220  * @param[in] actions
4221  *   Pointer to the list of actions.
4222  * @param[out] error
4223  *   Perform verbose error reporting if not NULL.
4224  *
4225  * @return
4226  *   0 on success, negative value otherwise and rte_errno is set.
4227  */
4228 static int
4229 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4230 			    struct rte_flow *flow,
4231 			    const struct rte_flow_action *actions,
4232 			    struct rte_flow_error *error)
4233 {
4234 	struct mlx5_priv *priv = dev->data->dev_private;
4235 	struct mlx5_dev_config *config = &priv->config;
4236 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4237 	const struct rte_flow_action_mark *mark;
4238 
4239 	/* Check whether extensive metadata feature is engaged. */
4240 	if (!config->dv_flow_en ||
4241 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4242 	    !mlx5_flow_ext_mreg_supported(dev) ||
4243 	    !priv->sh->dv_regc0_mask)
4244 		return 0;
4245 	/* Find MARK action. */
4246 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4247 		switch (actions->type) {
4248 		case RTE_FLOW_ACTION_TYPE_FLAG:
4249 			mcp_res = flow_mreg_add_copy_action
4250 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4251 			if (!mcp_res)
4252 				return -rte_errno;
4253 			flow->rix_mreg_copy = mcp_res->idx;
4254 			return 0;
4255 		case RTE_FLOW_ACTION_TYPE_MARK:
4256 			mark = (const struct rte_flow_action_mark *)
4257 				actions->conf;
4258 			mcp_res =
4259 				flow_mreg_add_copy_action(dev, mark->id, error);
4260 			if (!mcp_res)
4261 				return -rte_errno;
4262 			flow->rix_mreg_copy = mcp_res->idx;
4263 			return 0;
4264 		default:
4265 			break;
4266 		}
4267 	}
4268 	return 0;
4269 }
4270 
4271 #define MLX5_MAX_SPLIT_ACTIONS 24
4272 #define MLX5_MAX_SPLIT_ITEMS 24
4273 
4274 /**
4275  * Split the hairpin flow.
4276  * Since HW can't support encap and push-vlan on Rx, we move these
4277  * actions to Tx.
4278  * If the count action is after the encap then we also
4279  * move the count action. in this case the count will also measure
4280  * the outer bytes.
4281  *
4282  * @param dev
4283  *   Pointer to Ethernet device.
4284  * @param[in] actions
4285  *   Associated actions (list terminated by the END action).
4286  * @param[out] actions_rx
4287  *   Rx flow actions.
4288  * @param[out] actions_tx
4289  *   Tx flow actions..
4290  * @param[out] pattern_tx
4291  *   The pattern items for the Tx flow.
4292  * @param[out] flow_id
4293  *   The flow ID connected to this flow.
4294  *
4295  * @return
4296  *   0 on success.
4297  */
4298 static int
4299 flow_hairpin_split(struct rte_eth_dev *dev,
4300 		   const struct rte_flow_action actions[],
4301 		   struct rte_flow_action actions_rx[],
4302 		   struct rte_flow_action actions_tx[],
4303 		   struct rte_flow_item pattern_tx[],
4304 		   uint32_t flow_id)
4305 {
4306 	const struct rte_flow_action_raw_encap *raw_encap;
4307 	const struct rte_flow_action_raw_decap *raw_decap;
4308 	struct mlx5_rte_flow_action_set_tag *set_tag;
4309 	struct rte_flow_action *tag_action;
4310 	struct mlx5_rte_flow_item_tag *tag_item;
4311 	struct rte_flow_item *item;
4312 	char *addr;
4313 	int encap = 0;
4314 
4315 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4316 		switch (actions->type) {
4317 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4318 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4319 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4320 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4321 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4322 			rte_memcpy(actions_tx, actions,
4323 			       sizeof(struct rte_flow_action));
4324 			actions_tx++;
4325 			break;
4326 		case RTE_FLOW_ACTION_TYPE_COUNT:
4327 			if (encap) {
4328 				rte_memcpy(actions_tx, actions,
4329 					   sizeof(struct rte_flow_action));
4330 				actions_tx++;
4331 			} else {
4332 				rte_memcpy(actions_rx, actions,
4333 					   sizeof(struct rte_flow_action));
4334 				actions_rx++;
4335 			}
4336 			break;
4337 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4338 			raw_encap = actions->conf;
4339 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4340 				memcpy(actions_tx, actions,
4341 				       sizeof(struct rte_flow_action));
4342 				actions_tx++;
4343 				encap = 1;
4344 			} else {
4345 				rte_memcpy(actions_rx, actions,
4346 					   sizeof(struct rte_flow_action));
4347 				actions_rx++;
4348 			}
4349 			break;
4350 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4351 			raw_decap = actions->conf;
4352 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4353 				memcpy(actions_tx, actions,
4354 				       sizeof(struct rte_flow_action));
4355 				actions_tx++;
4356 			} else {
4357 				rte_memcpy(actions_rx, actions,
4358 					   sizeof(struct rte_flow_action));
4359 				actions_rx++;
4360 			}
4361 			break;
4362 		default:
4363 			rte_memcpy(actions_rx, actions,
4364 				   sizeof(struct rte_flow_action));
4365 			actions_rx++;
4366 			break;
4367 		}
4368 	}
4369 	/* Add set meta action and end action for the Rx flow. */
4370 	tag_action = actions_rx;
4371 	tag_action->type = (enum rte_flow_action_type)
4372 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4373 	actions_rx++;
4374 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4375 	actions_rx++;
4376 	set_tag = (void *)actions_rx;
4377 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4378 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4379 		.data = flow_id,
4380 	};
4381 	MLX5_ASSERT(set_tag->id > REG_NON);
4382 	tag_action->conf = set_tag;
4383 	/* Create Tx item list. */
4384 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4385 	addr = (void *)&pattern_tx[2];
4386 	item = pattern_tx;
4387 	item->type = (enum rte_flow_item_type)
4388 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4389 	tag_item = (void *)addr;
4390 	tag_item->data = flow_id;
4391 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4392 	MLX5_ASSERT(set_tag->id > REG_NON);
4393 	item->spec = tag_item;
4394 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4395 	tag_item = (void *)addr;
4396 	tag_item->data = UINT32_MAX;
4397 	tag_item->id = UINT16_MAX;
4398 	item->mask = tag_item;
4399 	item->last = NULL;
4400 	item++;
4401 	item->type = RTE_FLOW_ITEM_TYPE_END;
4402 	return 0;
4403 }
4404 
4405 /**
4406  * The last stage of splitting chain, just creates the subflow
4407  * without any modification.
4408  *
4409  * @param[in] dev
4410  *   Pointer to Ethernet device.
4411  * @param[in] flow
4412  *   Parent flow structure pointer.
4413  * @param[in, out] sub_flow
4414  *   Pointer to return the created subflow, may be NULL.
4415  * @param[in] attr
4416  *   Flow rule attributes.
4417  * @param[in] items
4418  *   Pattern specification (list terminated by the END pattern item).
4419  * @param[in] actions
4420  *   Associated actions (list terminated by the END action).
4421  * @param[in] flow_split_info
4422  *   Pointer to flow split info structure.
4423  * @param[out] error
4424  *   Perform verbose error reporting if not NULL.
4425  * @return
4426  *   0 on success, negative value otherwise
4427  */
4428 static int
4429 flow_create_split_inner(struct rte_eth_dev *dev,
4430 			struct rte_flow *flow,
4431 			struct mlx5_flow **sub_flow,
4432 			const struct rte_flow_attr *attr,
4433 			const struct rte_flow_item items[],
4434 			const struct rte_flow_action actions[],
4435 			struct mlx5_flow_split_info *flow_split_info,
4436 			struct rte_flow_error *error)
4437 {
4438 	struct mlx5_flow *dev_flow;
4439 
4440 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4441 				    flow_split_info->flow_idx, error);
4442 	if (!dev_flow)
4443 		return -rte_errno;
4444 	dev_flow->flow = flow;
4445 	dev_flow->external = flow_split_info->external;
4446 	dev_flow->skip_scale = flow_split_info->skip_scale;
4447 	/* Subflow object was created, we must include one in the list. */
4448 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4449 		      dev_flow->handle, next);
4450 	/*
4451 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4452 	 * flow may need some user defined item layer flags, and pass the
4453 	 * Metadate rxq mark flag to suffix flow as well.
4454 	 */
4455 	if (flow_split_info->prefix_layers)
4456 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4457 	if (flow_split_info->prefix_mark)
4458 		dev_flow->handle->mark = 1;
4459 	if (sub_flow)
4460 		*sub_flow = dev_flow;
4461 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4462 	dev_flow->dv.table_id = flow_split_info->table_id;
4463 #endif
4464 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4465 }
4466 
4467 /**
4468  * Get the sub policy of a meter.
4469  *
4470  * @param[in] dev
4471  *   Pointer to Ethernet device.
4472  * @param[in] flow
4473  *   Parent flow structure pointer.
4474  * @param[in] policy_id;
4475  *   Meter Policy id.
4476  * @param[in] attr
4477  *   Flow rule attributes.
4478  * @param[in] items
4479  *   Pattern specification (list terminated by the END pattern item).
4480  * @param[out] error
4481  *   Perform verbose error reporting if not NULL.
4482  *
4483  * @return
4484  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4485  */
4486 static struct mlx5_flow_meter_sub_policy *
4487 get_meter_sub_policy(struct rte_eth_dev *dev,
4488 		     struct rte_flow *flow,
4489 		     uint32_t policy_id,
4490 		     const struct rte_flow_attr *attr,
4491 		     const struct rte_flow_item items[],
4492 		     struct rte_flow_error *error)
4493 {
4494 	struct mlx5_flow_meter_policy *policy;
4495 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4496 
4497 	policy = mlx5_flow_meter_policy_find(dev, policy_id, NULL);
4498 	if (!policy) {
4499 		rte_flow_error_set(error, EINVAL,
4500 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4501 				   "Failed to find Meter Policy.");
4502 		goto exit;
4503 	}
4504 	if (policy->is_rss) {
4505 		struct mlx5_flow_workspace *wks =
4506 				mlx5_flow_get_thread_workspace();
4507 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4508 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4509 		uint32_t i;
4510 
4511 		MLX5_ASSERT(wks);
4512 		/**
4513 		 * This is a tmp dev_flow,
4514 		 * no need to register any matcher for it in translate.
4515 		 */
4516 		wks->skip_matcher_reg = 1;
4517 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4518 			struct mlx5_flow dev_flow = {0};
4519 			struct mlx5_flow_handle dev_handle = { {0} };
4520 			const void *rss_act = policy->act_cnt[i].rss->conf;
4521 			struct rte_flow_action rss_actions[2] = {
4522 				[0] = {
4523 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4524 					.conf = rss_act
4525 				},
4526 				[1] = {
4527 					.type = RTE_FLOW_ACTION_TYPE_END,
4528 					.conf = NULL
4529 				}
4530 			};
4531 
4532 			dev_flow.handle = &dev_handle;
4533 			dev_flow.ingress = attr->ingress;
4534 			dev_flow.flow = flow;
4535 			dev_flow.external = 0;
4536 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4537 			dev_flow.dv.transfer = attr->transfer;
4538 #endif
4539 			/* Translate RSS action to get rss hash fields. */
4540 			if (flow_drv_translate(dev, &dev_flow, attr,
4541 						items, rss_actions, error))
4542 				goto exit;
4543 			rss_desc_v[i] = wks->rss_desc;
4544 			rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4545 			rss_desc_v[i].hash_fields = dev_flow.hash_fields;
4546 			rss_desc_v[i].queue_num = rss_desc_v[i].hash_fields ?
4547 						  rss_desc_v[i].queue_num : 1;
4548 			rss_desc[i] = &rss_desc_v[i];
4549 		}
4550 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4551 						flow, policy, rss_desc);
4552 	} else {
4553 		enum mlx5_meter_domain mtr_domain =
4554 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4555 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4556 					MLX5_MTR_DOMAIN_INGRESS;
4557 		sub_policy = policy->sub_policys[mtr_domain][0];
4558 	}
4559 	if (!sub_policy) {
4560 		rte_flow_error_set(error, EINVAL,
4561 			RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4562 			"Failed to get meter sub-policy.");
4563 		goto exit;
4564 	}
4565 exit:
4566 	return sub_policy;
4567 }
4568 
4569 /**
4570  * Split the meter flow.
4571  *
4572  * As meter flow will split to three sub flow, other than meter
4573  * action, the other actions make sense to only meter accepts
4574  * the packet. If it need to be dropped, no other additional
4575  * actions should be take.
4576  *
4577  * One kind of special action which decapsulates the L3 tunnel
4578  * header will be in the prefix sub flow, as not to take the
4579  * L3 tunnel header into account.
4580  *
4581  * @param[in] dev
4582  *   Pointer to Ethernet device.
4583  * @param[in] flow
4584  *   Parent flow structure pointer.
4585  * @param[in] fm
4586  *   Pointer to flow meter structure.
4587  * @param[in] attr
4588  *   Flow rule attributes.
4589  * @param[in] items
4590  *   Pattern specification (list terminated by the END pattern item).
4591  * @param[out] sfx_items
4592  *   Suffix flow match items (list terminated by the END pattern item).
4593  * @param[in] actions
4594  *   Associated actions (list terminated by the END action).
4595  * @param[out] actions_sfx
4596  *   Suffix flow actions.
4597  * @param[out] actions_pre
4598  *   Prefix flow actions.
4599  * @param[out] mtr_flow_id
4600  *   Pointer to meter flow id.
4601  * @param[out] error
4602  *   Perform verbose error reporting if not NULL.
4603  *
4604  * @return
4605  *   0 on success, a negative errno value otherwise and rte_errno is set.
4606  */
4607 static int
4608 flow_meter_split_prep(struct rte_eth_dev *dev,
4609 		      struct rte_flow *flow,
4610 		      struct mlx5_flow_meter_info *fm,
4611 		      const struct rte_flow_attr *attr,
4612 		      const struct rte_flow_item items[],
4613 		      struct rte_flow_item sfx_items[],
4614 		      const struct rte_flow_action actions[],
4615 		      struct rte_flow_action actions_sfx[],
4616 		      struct rte_flow_action actions_pre[],
4617 		      uint32_t *mtr_flow_id,
4618 		      struct rte_flow_error *error)
4619 {
4620 	struct mlx5_priv *priv = dev->data->dev_private;
4621 	struct rte_flow_action *tag_action = NULL;
4622 	struct rte_flow_item *tag_item;
4623 	struct mlx5_rte_flow_action_set_tag *set_tag;
4624 	const struct rte_flow_action_raw_encap *raw_encap;
4625 	const struct rte_flow_action_raw_decap *raw_decap;
4626 	struct mlx5_rte_flow_item_tag *tag_item_spec;
4627 	struct mlx5_rte_flow_item_tag *tag_item_mask;
4628 	uint32_t tag_id = 0;
4629 	bool copy_vlan = false;
4630 	struct rte_flow_action *hw_mtr_action;
4631 	struct rte_flow_action *action_pre_head = NULL;
4632 	bool mtr_first = priv->sh->meter_aso_en &&
4633 			(attr->egress ||
4634 			(attr->transfer && priv->representor_id != UINT16_MAX));
4635 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
4636 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
4637 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
4638 	uint32_t flow_id = 0;
4639 	uint32_t flow_id_reversed = 0;
4640 	uint8_t flow_id_bits = 0;
4641 	int shift;
4642 
4643 	/* For ASO meter, meter must be before tag in TX direction. */
4644 	if (mtr_first) {
4645 		action_pre_head = actions_pre++;
4646 		/* Leave space for tag action. */
4647 		tag_action = actions_pre++;
4648 	}
4649 	/* Prepare the actions for prefix and suffix flow. */
4650 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4651 		struct rte_flow_action *action_cur = NULL;
4652 
4653 		switch (actions->type) {
4654 		case RTE_FLOW_ACTION_TYPE_METER:
4655 			if (mtr_first) {
4656 				action_cur = action_pre_head;
4657 			} else {
4658 				/* Leave space for tag action. */
4659 				tag_action = actions_pre++;
4660 				action_cur = actions_pre++;
4661 			}
4662 			break;
4663 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4664 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4665 			action_cur = actions_pre++;
4666 			break;
4667 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4668 			raw_encap = actions->conf;
4669 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4670 				action_cur = actions_pre++;
4671 			break;
4672 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4673 			raw_decap = actions->conf;
4674 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4675 				action_cur = actions_pre++;
4676 			break;
4677 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4678 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4679 			copy_vlan = true;
4680 			break;
4681 		default:
4682 			break;
4683 		}
4684 		if (!action_cur)
4685 			action_cur = (fm->def_policy) ?
4686 					actions_sfx++ : actions_pre++;
4687 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
4688 	}
4689 	/* Add end action to the actions. */
4690 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4691 	if (priv->sh->meter_aso_en) {
4692 		/**
4693 		 * For ASO meter, need to add an extra jump action explicitly,
4694 		 * to jump from meter to policer table.
4695 		 */
4696 		struct mlx5_flow_meter_sub_policy *sub_policy;
4697 		struct mlx5_flow_tbl_data_entry *tbl_data;
4698 
4699 		if (!fm->def_policy) {
4700 			sub_policy = get_meter_sub_policy(dev, flow,
4701 							  fm->policy_id, attr,
4702 							  items, error);
4703 			if (!sub_policy)
4704 				return -rte_errno;
4705 		} else {
4706 			enum mlx5_meter_domain mtr_domain =
4707 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4708 				attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4709 					MLX5_MTR_DOMAIN_INGRESS;
4710 
4711 			sub_policy =
4712 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
4713 		}
4714 		tbl_data = container_of(sub_policy->tbl_rsc,
4715 					struct mlx5_flow_tbl_data_entry, tbl);
4716 		hw_mtr_action = actions_pre++;
4717 		hw_mtr_action->type = (enum rte_flow_action_type)
4718 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
4719 		hw_mtr_action->conf = tbl_data->jump.action;
4720 	}
4721 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4722 	actions_pre++;
4723 	if (!tag_action)
4724 		return rte_flow_error_set(error, ENOMEM,
4725 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4726 					"No tag action space.");
4727 	if (!mtr_flow_id) {
4728 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
4729 		goto exit;
4730 	}
4731 	/* Only default-policy Meter creates mtr flow id. */
4732 	if (fm->def_policy) {
4733 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
4734 		if (!tag_id)
4735 			return rte_flow_error_set(error, ENOMEM,
4736 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4737 					"Failed to allocate meter flow id.");
4738 		flow_id = tag_id - 1;
4739 		flow_id_bits = MLX5_REG_BITS - __builtin_clz(flow_id);
4740 		flow_id_bits = flow_id_bits ? flow_id_bits : 1;
4741 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
4742 		    mtr_reg_bits) {
4743 			mlx5_ipool_free(fm->flow_ipool, tag_id);
4744 			return rte_flow_error_set(error, EINVAL,
4745 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4746 					"Meter flow id exceeds max limit.");
4747 		}
4748 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
4749 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
4750 	}
4751 	/* Prepare the suffix subflow items. */
4752 	tag_item = sfx_items++;
4753 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4754 		int item_type = items->type;
4755 
4756 		switch (item_type) {
4757 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4758 			memcpy(sfx_items, items, sizeof(*sfx_items));
4759 			sfx_items++;
4760 			break;
4761 		case RTE_FLOW_ITEM_TYPE_VLAN:
4762 			if (copy_vlan) {
4763 				memcpy(sfx_items, items, sizeof(*sfx_items));
4764 				/*
4765 				 * Convert to internal match item, it is used
4766 				 * for vlan push and set vid.
4767 				 */
4768 				sfx_items->type = (enum rte_flow_item_type)
4769 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4770 				sfx_items++;
4771 			}
4772 			break;
4773 		default:
4774 			break;
4775 		}
4776 	}
4777 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4778 	sfx_items++;
4779 	/* Build tag actions and items for meter_id/meter flow_id. */
4780 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
4781 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4782 	tag_item_mask = tag_item_spec + 1;
4783 	/* Both flow_id and meter_id share the same register. */
4784 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4785 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
4786 							    0, error),
4787 		.offset = mtr_id_offset,
4788 		.length = mtr_reg_bits,
4789 		.data = flow->meter,
4790 	};
4791 	/*
4792 	 * The color Reg bits used by flow_id are growing from
4793 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
4794 	 */
4795 	for (shift = 0; shift < flow_id_bits; shift++)
4796 		flow_id_reversed = (flow_id_reversed << 1) |
4797 				((flow_id >> shift) & 0x1);
4798 	set_tag->data |=
4799 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
4800 	tag_item_spec->id = set_tag->id;
4801 	tag_item_spec->data = set_tag->data << mtr_id_offset;
4802 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
4803 	tag_action->type = (enum rte_flow_action_type)
4804 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4805 	tag_action->conf = set_tag;
4806 	tag_item->type = (enum rte_flow_item_type)
4807 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4808 	tag_item->spec = tag_item_spec;
4809 	tag_item->last = NULL;
4810 	tag_item->mask = tag_item_mask;
4811 exit:
4812 	if (mtr_flow_id)
4813 		*mtr_flow_id = tag_id;
4814 	return 0;
4815 }
4816 
4817 /**
4818  * Split action list having QUEUE/RSS for metadata register copy.
4819  *
4820  * Once Q/RSS action is detected in user's action list, the flow action
4821  * should be split in order to copy metadata registers, which will happen in
4822  * RX_CP_TBL like,
4823  *   - CQE->flow_tag := reg_c[1] (MARK)
4824  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4825  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4826  * This is because the last action of each flow must be a terminal action
4827  * (QUEUE, RSS or DROP).
4828  *
4829  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4830  * stored and kept in the mlx5_flow structure per each sub_flow.
4831  *
4832  * The Q/RSS action is replaced with,
4833  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4834  * And the following JUMP action is added at the end,
4835  *   - JUMP, to RX_CP_TBL.
4836  *
4837  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4838  * flow_create_split_metadata() routine. The flow will look like,
4839  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4840  *
4841  * @param dev
4842  *   Pointer to Ethernet device.
4843  * @param[out] split_actions
4844  *   Pointer to store split actions to jump to CP_TBL.
4845  * @param[in] actions
4846  *   Pointer to the list of original flow actions.
4847  * @param[in] qrss
4848  *   Pointer to the Q/RSS action.
4849  * @param[in] actions_n
4850  *   Number of original actions.
4851  * @param[out] error
4852  *   Perform verbose error reporting if not NULL.
4853  *
4854  * @return
4855  *   non-zero unique flow_id on success, otherwise 0 and
4856  *   error/rte_error are set.
4857  */
4858 static uint32_t
4859 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4860 			  struct rte_flow_action *split_actions,
4861 			  const struct rte_flow_action *actions,
4862 			  const struct rte_flow_action *qrss,
4863 			  int actions_n, struct rte_flow_error *error)
4864 {
4865 	struct mlx5_priv *priv = dev->data->dev_private;
4866 	struct mlx5_rte_flow_action_set_tag *set_tag;
4867 	struct rte_flow_action_jump *jump;
4868 	const int qrss_idx = qrss - actions;
4869 	uint32_t flow_id = 0;
4870 	int ret = 0;
4871 
4872 	/*
4873 	 * Given actions will be split
4874 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4875 	 * - Add jump to mreg CP_TBL.
4876 	 * As a result, there will be one more action.
4877 	 */
4878 	++actions_n;
4879 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4880 	set_tag = (void *)(split_actions + actions_n);
4881 	/*
4882 	 * If tag action is not set to void(it means we are not the meter
4883 	 * suffix flow), add the tag action. Since meter suffix flow already
4884 	 * has the tag added.
4885 	 */
4886 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4887 		/*
4888 		 * Allocate the new subflow ID. This one is unique within
4889 		 * device and not shared with representors. Otherwise,
4890 		 * we would have to resolve multi-thread access synch
4891 		 * issue. Each flow on the shared device is appended
4892 		 * with source vport identifier, so the resulting
4893 		 * flows will be unique in the shared (by master and
4894 		 * representors) domain even if they have coinciding
4895 		 * IDs.
4896 		 */
4897 		mlx5_ipool_malloc(priv->sh->ipool
4898 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4899 		if (!flow_id)
4900 			return rte_flow_error_set(error, ENOMEM,
4901 						  RTE_FLOW_ERROR_TYPE_ACTION,
4902 						  NULL, "can't allocate id "
4903 						  "for split Q/RSS subflow");
4904 		/* Internal SET_TAG action to set flow ID. */
4905 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4906 			.data = flow_id,
4907 		};
4908 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4909 		if (ret < 0)
4910 			return ret;
4911 		set_tag->id = ret;
4912 		/* Construct new actions array. */
4913 		/* Replace QUEUE/RSS action. */
4914 		split_actions[qrss_idx] = (struct rte_flow_action){
4915 			.type = (enum rte_flow_action_type)
4916 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4917 			.conf = set_tag,
4918 		};
4919 	}
4920 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4921 	jump = (void *)(set_tag + 1);
4922 	*jump = (struct rte_flow_action_jump){
4923 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4924 	};
4925 	split_actions[actions_n - 2] = (struct rte_flow_action){
4926 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4927 		.conf = jump,
4928 	};
4929 	split_actions[actions_n - 1] = (struct rte_flow_action){
4930 		.type = RTE_FLOW_ACTION_TYPE_END,
4931 	};
4932 	return flow_id;
4933 }
4934 
4935 /**
4936  * Extend the given action list for Tx metadata copy.
4937  *
4938  * Copy the given action list to the ext_actions and add flow metadata register
4939  * copy action in order to copy reg_a set by WQE to reg_c[0].
4940  *
4941  * @param[out] ext_actions
4942  *   Pointer to the extended action list.
4943  * @param[in] actions
4944  *   Pointer to the list of actions.
4945  * @param[in] actions_n
4946  *   Number of actions in the list.
4947  * @param[out] error
4948  *   Perform verbose error reporting if not NULL.
4949  * @param[in] encap_idx
4950  *   The encap action inndex.
4951  *
4952  * @return
4953  *   0 on success, negative value otherwise
4954  */
4955 static int
4956 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4957 		       struct rte_flow_action *ext_actions,
4958 		       const struct rte_flow_action *actions,
4959 		       int actions_n, struct rte_flow_error *error,
4960 		       int encap_idx)
4961 {
4962 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4963 		(struct mlx5_flow_action_copy_mreg *)
4964 			(ext_actions + actions_n + 1);
4965 	int ret;
4966 
4967 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4968 	if (ret < 0)
4969 		return ret;
4970 	cp_mreg->dst = ret;
4971 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4972 	if (ret < 0)
4973 		return ret;
4974 	cp_mreg->src = ret;
4975 	if (encap_idx != 0)
4976 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4977 	if (encap_idx == actions_n - 1) {
4978 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4979 			.type = (enum rte_flow_action_type)
4980 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4981 			.conf = cp_mreg,
4982 		};
4983 		ext_actions[actions_n] = (struct rte_flow_action){
4984 			.type = RTE_FLOW_ACTION_TYPE_END,
4985 		};
4986 	} else {
4987 		ext_actions[encap_idx] = (struct rte_flow_action){
4988 			.type = (enum rte_flow_action_type)
4989 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4990 			.conf = cp_mreg,
4991 		};
4992 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4993 				sizeof(*ext_actions) * (actions_n - encap_idx));
4994 	}
4995 	return 0;
4996 }
4997 
4998 /**
4999  * Check the match action from the action list.
5000  *
5001  * @param[in] actions
5002  *   Pointer to the list of actions.
5003  * @param[in] attr
5004  *   Flow rule attributes.
5005  * @param[in] action
5006  *   The action to be check if exist.
5007  * @param[out] match_action_pos
5008  *   Pointer to the position of the matched action if exists, otherwise is -1.
5009  * @param[out] qrss_action_pos
5010  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5011  * @param[out] modify_after_mirror
5012  *   Pointer to the flag of modify action after FDB mirroring.
5013  *
5014  * @return
5015  *   > 0 the total number of actions.
5016  *   0 if not found match action in action list.
5017  */
5018 static int
5019 flow_check_match_action(const struct rte_flow_action actions[],
5020 			const struct rte_flow_attr *attr,
5021 			enum rte_flow_action_type action,
5022 			int *match_action_pos, int *qrss_action_pos,
5023 			int *modify_after_mirror)
5024 {
5025 	const struct rte_flow_action_sample *sample;
5026 	int actions_n = 0;
5027 	uint32_t ratio = 0;
5028 	int sub_type = 0;
5029 	int flag = 0;
5030 	int fdb_mirror = 0;
5031 
5032 	*match_action_pos = -1;
5033 	*qrss_action_pos = -1;
5034 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5035 		if (actions->type == action) {
5036 			flag = 1;
5037 			*match_action_pos = actions_n;
5038 		}
5039 		switch (actions->type) {
5040 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5041 		case RTE_FLOW_ACTION_TYPE_RSS:
5042 			*qrss_action_pos = actions_n;
5043 			break;
5044 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5045 			sample = actions->conf;
5046 			ratio = sample->ratio;
5047 			sub_type = ((const struct rte_flow_action *)
5048 					(sample->actions))->type;
5049 			if (ratio == 1 && attr->transfer)
5050 				fdb_mirror = 1;
5051 			break;
5052 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5053 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5054 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5055 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5056 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5057 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5058 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5059 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5060 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5061 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5062 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5063 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5064 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5065 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5066 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5067 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5068 		case RTE_FLOW_ACTION_TYPE_FLAG:
5069 		case RTE_FLOW_ACTION_TYPE_MARK:
5070 		case RTE_FLOW_ACTION_TYPE_SET_META:
5071 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5072 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5073 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5074 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5075 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5076 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5077 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5078 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5079 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5080 			if (fdb_mirror)
5081 				*modify_after_mirror = 1;
5082 			break;
5083 		default:
5084 			break;
5085 		}
5086 		actions_n++;
5087 	}
5088 	if (flag && fdb_mirror && !*modify_after_mirror) {
5089 		/* FDB mirroring uses the destination array to implement
5090 		 * instead of FLOW_SAMPLER object.
5091 		 */
5092 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5093 			flag = 0;
5094 	}
5095 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5096 	return flag ? actions_n + 1 : 0;
5097 }
5098 
5099 #define SAMPLE_SUFFIX_ITEM 2
5100 
5101 /**
5102  * Split the sample flow.
5103  *
5104  * As sample flow will split to two sub flow, sample flow with
5105  * sample action, the other actions will move to new suffix flow.
5106  *
5107  * Also add unique tag id with tag action in the sample flow,
5108  * the same tag id will be as match in the suffix flow.
5109  *
5110  * @param dev
5111  *   Pointer to Ethernet device.
5112  * @param[in] add_tag
5113  *   Add extra tag action flag.
5114  * @param[out] sfx_items
5115  *   Suffix flow match items (list terminated by the END pattern item).
5116  * @param[in] actions
5117  *   Associated actions (list terminated by the END action).
5118  * @param[out] actions_sfx
5119  *   Suffix flow actions.
5120  * @param[out] actions_pre
5121  *   Prefix flow actions.
5122  * @param[in] actions_n
5123  *  The total number of actions.
5124  * @param[in] sample_action_pos
5125  *   The sample action position.
5126  * @param[in] qrss_action_pos
5127  *   The Queue/RSS action position.
5128  * @param[in] jump_table
5129  *   Add extra jump action flag.
5130  * @param[out] error
5131  *   Perform verbose error reporting if not NULL.
5132  *
5133  * @return
5134  *   0 on success, or unique flow_id, a negative errno value
5135  *   otherwise and rte_errno is set.
5136  */
5137 static int
5138 flow_sample_split_prep(struct rte_eth_dev *dev,
5139 		       int add_tag,
5140 		       struct rte_flow_item sfx_items[],
5141 		       const struct rte_flow_action actions[],
5142 		       struct rte_flow_action actions_sfx[],
5143 		       struct rte_flow_action actions_pre[],
5144 		       int actions_n,
5145 		       int sample_action_pos,
5146 		       int qrss_action_pos,
5147 		       int jump_table,
5148 		       struct rte_flow_error *error)
5149 {
5150 	struct mlx5_priv *priv = dev->data->dev_private;
5151 	struct mlx5_rte_flow_action_set_tag *set_tag;
5152 	struct mlx5_rte_flow_item_tag *tag_spec;
5153 	struct mlx5_rte_flow_item_tag *tag_mask;
5154 	struct rte_flow_action_jump *jump_action;
5155 	uint32_t tag_id = 0;
5156 	int index;
5157 	int append_index = 0;
5158 	int ret;
5159 
5160 	if (sample_action_pos < 0)
5161 		return rte_flow_error_set(error, EINVAL,
5162 					  RTE_FLOW_ERROR_TYPE_ACTION,
5163 					  NULL, "invalid position of sample "
5164 					  "action in list");
5165 	/* Prepare the actions for prefix and suffix flow. */
5166 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5167 		index = qrss_action_pos;
5168 		/* Put the preceding the Queue/RSS action into prefix flow. */
5169 		if (index != 0)
5170 			memcpy(actions_pre, actions,
5171 			       sizeof(struct rte_flow_action) * index);
5172 		/* Put others preceding the sample action into prefix flow. */
5173 		if (sample_action_pos > index + 1)
5174 			memcpy(actions_pre + index, actions + index + 1,
5175 			       sizeof(struct rte_flow_action) *
5176 			       (sample_action_pos - index - 1));
5177 		index = sample_action_pos - 1;
5178 		/* Put Queue/RSS action into Suffix flow. */
5179 		memcpy(actions_sfx, actions + qrss_action_pos,
5180 		       sizeof(struct rte_flow_action));
5181 		actions_sfx++;
5182 	} else {
5183 		index = sample_action_pos;
5184 		if (index != 0)
5185 			memcpy(actions_pre, actions,
5186 			       sizeof(struct rte_flow_action) * index);
5187 	}
5188 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5189 	 * For CX6DX and above, metadata registers Cx preserve their value,
5190 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5191 	 */
5192 	if (add_tag) {
5193 		/* Prepare the prefix tag action. */
5194 		append_index++;
5195 		set_tag = (void *)(actions_pre + actions_n + append_index);
5196 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5197 		if (ret < 0)
5198 			return ret;
5199 		mlx5_ipool_malloc(priv->sh->ipool
5200 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5201 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5202 			.id = ret,
5203 			.data = tag_id,
5204 		};
5205 		/* Prepare the suffix subflow items. */
5206 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5207 		tag_spec->data = tag_id;
5208 		tag_spec->id = set_tag->id;
5209 		tag_mask = tag_spec + 1;
5210 		tag_mask->data = UINT32_MAX;
5211 		sfx_items[0] = (struct rte_flow_item){
5212 			.type = (enum rte_flow_item_type)
5213 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5214 			.spec = tag_spec,
5215 			.last = NULL,
5216 			.mask = tag_mask,
5217 		};
5218 		sfx_items[1] = (struct rte_flow_item){
5219 			.type = (enum rte_flow_item_type)
5220 				RTE_FLOW_ITEM_TYPE_END,
5221 		};
5222 		/* Prepare the tag action in prefix subflow. */
5223 		actions_pre[index++] =
5224 			(struct rte_flow_action){
5225 			.type = (enum rte_flow_action_type)
5226 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5227 			.conf = set_tag,
5228 		};
5229 	}
5230 	memcpy(actions_pre + index, actions + sample_action_pos,
5231 	       sizeof(struct rte_flow_action));
5232 	index += 1;
5233 	/* For the modify action after the sample action in E-Switch mirroring,
5234 	 * Add the extra jump action in prefix subflow and jump into the next
5235 	 * table, then do the modify action in the new table.
5236 	 */
5237 	if (jump_table) {
5238 		/* Prepare the prefix jump action. */
5239 		append_index++;
5240 		jump_action = (void *)(actions_pre + actions_n + append_index);
5241 		jump_action->group = jump_table;
5242 		actions_pre[index++] =
5243 			(struct rte_flow_action){
5244 			.type = (enum rte_flow_action_type)
5245 				RTE_FLOW_ACTION_TYPE_JUMP,
5246 			.conf = jump_action,
5247 		};
5248 	}
5249 	actions_pre[index] = (struct rte_flow_action){
5250 		.type = (enum rte_flow_action_type)
5251 			RTE_FLOW_ACTION_TYPE_END,
5252 	};
5253 	/* Put the actions after sample into Suffix flow. */
5254 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5255 	       sizeof(struct rte_flow_action) *
5256 	       (actions_n - sample_action_pos - 1));
5257 	return tag_id;
5258 }
5259 
5260 /**
5261  * The splitting for metadata feature.
5262  *
5263  * - Q/RSS action on NIC Rx should be split in order to pass by
5264  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5265  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5266  *
5267  * - All the actions on NIC Tx should have a mreg copy action to
5268  *   copy reg_a from WQE to reg_c[0].
5269  *
5270  * @param dev
5271  *   Pointer to Ethernet device.
5272  * @param[in] flow
5273  *   Parent flow structure pointer.
5274  * @param[in] attr
5275  *   Flow rule attributes.
5276  * @param[in] items
5277  *   Pattern specification (list terminated by the END pattern item).
5278  * @param[in] actions
5279  *   Associated actions (list terminated by the END action).
5280  * @param[in] flow_split_info
5281  *   Pointer to flow split info structure.
5282  * @param[out] error
5283  *   Perform verbose error reporting if not NULL.
5284  * @return
5285  *   0 on success, negative value otherwise
5286  */
5287 static int
5288 flow_create_split_metadata(struct rte_eth_dev *dev,
5289 			   struct rte_flow *flow,
5290 			   const struct rte_flow_attr *attr,
5291 			   const struct rte_flow_item items[],
5292 			   const struct rte_flow_action actions[],
5293 			   struct mlx5_flow_split_info *flow_split_info,
5294 			   struct rte_flow_error *error)
5295 {
5296 	struct mlx5_priv *priv = dev->data->dev_private;
5297 	struct mlx5_dev_config *config = &priv->config;
5298 	const struct rte_flow_action *qrss = NULL;
5299 	struct rte_flow_action *ext_actions = NULL;
5300 	struct mlx5_flow *dev_flow = NULL;
5301 	uint32_t qrss_id = 0;
5302 	int mtr_sfx = 0;
5303 	size_t act_size;
5304 	int actions_n;
5305 	int encap_idx;
5306 	int ret;
5307 
5308 	/* Check whether extensive metadata feature is engaged. */
5309 	if (!config->dv_flow_en ||
5310 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5311 	    !mlx5_flow_ext_mreg_supported(dev))
5312 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5313 					       actions, flow_split_info, error);
5314 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5315 							   &encap_idx);
5316 	if (qrss) {
5317 		/* Exclude hairpin flows from splitting. */
5318 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5319 			const struct rte_flow_action_queue *queue;
5320 
5321 			queue = qrss->conf;
5322 			if (mlx5_rxq_get_type(dev, queue->index) ==
5323 			    MLX5_RXQ_TYPE_HAIRPIN)
5324 				qrss = NULL;
5325 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5326 			const struct rte_flow_action_rss *rss;
5327 
5328 			rss = qrss->conf;
5329 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5330 			    MLX5_RXQ_TYPE_HAIRPIN)
5331 				qrss = NULL;
5332 		}
5333 	}
5334 	if (qrss) {
5335 		/* Check if it is in meter suffix table. */
5336 		mtr_sfx = attr->group == (attr->transfer ?
5337 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5338 			  MLX5_FLOW_TABLE_LEVEL_METER);
5339 		/*
5340 		 * Q/RSS action on NIC Rx should be split in order to pass by
5341 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5342 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5343 		 */
5344 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5345 			   sizeof(struct rte_flow_action_set_tag) +
5346 			   sizeof(struct rte_flow_action_jump);
5347 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5348 					  SOCKET_ID_ANY);
5349 		if (!ext_actions)
5350 			return rte_flow_error_set(error, ENOMEM,
5351 						  RTE_FLOW_ERROR_TYPE_ACTION,
5352 						  NULL, "no memory to split "
5353 						  "metadata flow");
5354 		/*
5355 		 * If we are the suffix flow of meter, tag already exist.
5356 		 * Set the tag action to void.
5357 		 */
5358 		if (mtr_sfx)
5359 			ext_actions[qrss - actions].type =
5360 						RTE_FLOW_ACTION_TYPE_VOID;
5361 		else
5362 			ext_actions[qrss - actions].type =
5363 						(enum rte_flow_action_type)
5364 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5365 		/*
5366 		 * Create the new actions list with removed Q/RSS action
5367 		 * and appended set tag and jump to register copy table
5368 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5369 		 * in advance, because it is needed for set tag action.
5370 		 */
5371 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5372 						    qrss, actions_n, error);
5373 		if (!mtr_sfx && !qrss_id) {
5374 			ret = -rte_errno;
5375 			goto exit;
5376 		}
5377 	} else if (attr->egress && !attr->transfer) {
5378 		/*
5379 		 * All the actions on NIC Tx should have a metadata register
5380 		 * copy action to copy reg_a from WQE to reg_c[meta]
5381 		 */
5382 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5383 			   sizeof(struct mlx5_flow_action_copy_mreg);
5384 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5385 					  SOCKET_ID_ANY);
5386 		if (!ext_actions)
5387 			return rte_flow_error_set(error, ENOMEM,
5388 						  RTE_FLOW_ERROR_TYPE_ACTION,
5389 						  NULL, "no memory to split "
5390 						  "metadata flow");
5391 		/* Create the action list appended with copy register. */
5392 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5393 					     actions_n, error, encap_idx);
5394 		if (ret < 0)
5395 			goto exit;
5396 	}
5397 	/* Add the unmodified original or prefix subflow. */
5398 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5399 				      items, ext_actions ? ext_actions :
5400 				      actions, flow_split_info, error);
5401 	if (ret < 0)
5402 		goto exit;
5403 	MLX5_ASSERT(dev_flow);
5404 	if (qrss) {
5405 		const struct rte_flow_attr q_attr = {
5406 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5407 			.ingress = 1,
5408 		};
5409 		/* Internal PMD action to set register. */
5410 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5411 			.data = qrss_id,
5412 			.id = REG_NON,
5413 		};
5414 		struct rte_flow_item q_items[] = {
5415 			{
5416 				.type = (enum rte_flow_item_type)
5417 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5418 				.spec = &q_tag_spec,
5419 				.last = NULL,
5420 				.mask = NULL,
5421 			},
5422 			{
5423 				.type = RTE_FLOW_ITEM_TYPE_END,
5424 			},
5425 		};
5426 		struct rte_flow_action q_actions[] = {
5427 			{
5428 				.type = qrss->type,
5429 				.conf = qrss->conf,
5430 			},
5431 			{
5432 				.type = RTE_FLOW_ACTION_TYPE_END,
5433 			},
5434 		};
5435 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5436 
5437 		/*
5438 		 * Configure the tag item only if there is no meter subflow.
5439 		 * Since tag is already marked in the meter suffix subflow
5440 		 * we can just use the meter suffix items as is.
5441 		 */
5442 		if (qrss_id) {
5443 			/* Not meter subflow. */
5444 			MLX5_ASSERT(!mtr_sfx);
5445 			/*
5446 			 * Put unique id in prefix flow due to it is destroyed
5447 			 * after suffix flow and id will be freed after there
5448 			 * is no actual flows with this id and identifier
5449 			 * reallocation becomes possible (for example, for
5450 			 * other flows in other threads).
5451 			 */
5452 			dev_flow->handle->split_flow_id = qrss_id;
5453 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5454 						   error);
5455 			if (ret < 0)
5456 				goto exit;
5457 			q_tag_spec.id = ret;
5458 		}
5459 		dev_flow = NULL;
5460 		/* Add suffix subflow to execute Q/RSS. */
5461 		flow_split_info->prefix_layers = layers;
5462 		flow_split_info->prefix_mark = 0;
5463 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5464 					      &q_attr, mtr_sfx ? items :
5465 					      q_items, q_actions,
5466 					      flow_split_info, error);
5467 		if (ret < 0)
5468 			goto exit;
5469 		/* qrss ID should be freed if failed. */
5470 		qrss_id = 0;
5471 		MLX5_ASSERT(dev_flow);
5472 	}
5473 
5474 exit:
5475 	/*
5476 	 * We do not destroy the partially created sub_flows in case of error.
5477 	 * These ones are included into parent flow list and will be destroyed
5478 	 * by flow_drv_destroy.
5479 	 */
5480 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5481 			qrss_id);
5482 	mlx5_free(ext_actions);
5483 	return ret;
5484 }
5485 
5486 /**
5487  * Create meter internal drop flow with the original pattern.
5488  *
5489  * @param dev
5490  *   Pointer to Ethernet device.
5491  * @param[in] flow
5492  *   Parent flow structure pointer.
5493  * @param[in] attr
5494  *   Flow rule attributes.
5495  * @param[in] items
5496  *   Pattern specification (list terminated by the END pattern item).
5497  * @param[in] flow_split_info
5498  *   Pointer to flow split info structure.
5499  * @param[in] fm
5500  *   Pointer to flow meter structure.
5501  * @param[out] error
5502  *   Perform verbose error reporting if not NULL.
5503  * @return
5504  *   0 on success, negative value otherwise
5505  */
5506 static uint32_t
5507 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5508 			struct rte_flow *flow,
5509 			const struct rte_flow_attr *attr,
5510 			const struct rte_flow_item items[],
5511 			struct mlx5_flow_split_info *flow_split_info,
5512 			struct mlx5_flow_meter_info *fm,
5513 			struct rte_flow_error *error)
5514 {
5515 	struct mlx5_flow *dev_flow = NULL;
5516 	struct rte_flow_attr drop_attr = *attr;
5517 	struct rte_flow_action drop_actions[3];
5518 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5519 
5520 	MLX5_ASSERT(fm->drop_cnt);
5521 	drop_actions[0].type =
5522 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5523 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5524 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5525 	drop_actions[1].conf = NULL;
5526 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5527 	drop_actions[2].conf = NULL;
5528 	drop_split_info.external = false;
5529 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5530 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5531 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5532 	return flow_create_split_inner(dev, flow, &dev_flow,
5533 				&drop_attr, items, drop_actions,
5534 				&drop_split_info, error);
5535 }
5536 
5537 /**
5538  * The splitting for meter feature.
5539  *
5540  * - The meter flow will be split to two flows as prefix and
5541  *   suffix flow. The packets make sense only it pass the prefix
5542  *   meter action.
5543  *
5544  * - Reg_C_5 is used for the packet to match betweend prefix and
5545  *   suffix flow.
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_meter(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 	struct mlx5_priv *priv = dev->data->dev_private;
5574 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5575 	struct rte_flow_action *sfx_actions = NULL;
5576 	struct rte_flow_action *pre_actions = NULL;
5577 	struct rte_flow_item *sfx_items = NULL;
5578 	struct mlx5_flow *dev_flow = NULL;
5579 	struct rte_flow_attr sfx_attr = *attr;
5580 	struct mlx5_flow_meter_info *fm = NULL;
5581 	uint8_t skip_scale_restore;
5582 	bool has_mtr = false;
5583 	bool has_modify = false;
5584 	bool set_mtr_reg = true;
5585 	uint32_t meter_id = 0;
5586 	uint32_t mtr_idx = 0;
5587 	uint32_t mtr_flow_id = 0;
5588 	size_t act_size;
5589 	size_t item_size;
5590 	int actions_n = 0;
5591 	int ret = 0;
5592 
5593 	if (priv->mtr_en)
5594 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
5595 						    &has_modify, &meter_id);
5596 	if (has_mtr) {
5597 		if (flow->meter) {
5598 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
5599 			if (!fm)
5600 				return rte_flow_error_set(error, EINVAL,
5601 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5602 						NULL, "Meter not found.");
5603 		} else {
5604 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
5605 			if (!fm)
5606 				return rte_flow_error_set(error, EINVAL,
5607 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5608 						NULL, "Meter not found.");
5609 			ret = mlx5_flow_meter_attach(priv, fm,
5610 						     &sfx_attr, error);
5611 			if (ret)
5612 				return -rte_errno;
5613 			flow->meter = mtr_idx;
5614 		}
5615 		MLX5_ASSERT(wks);
5616 		wks->fm = fm;
5617 		/*
5618 		 * If it isn't default-policy Meter, and
5619 		 * 1. There's no action in flow to change
5620 		 *    packet (modify/encap/decap etc.), OR
5621 		 * 2. No drop count needed for this meter.
5622 		 * no need to use regC to save meter id anymore.
5623 		 */
5624 		if (!fm->def_policy && (!has_modify || !fm->drop_cnt))
5625 			set_mtr_reg = false;
5626 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
5627 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
5628 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5629 		/* Suffix items: tag, vlan, port id, end. */
5630 #define METER_SUFFIX_ITEM 4
5631 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5632 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5633 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5634 					  0, SOCKET_ID_ANY);
5635 		if (!sfx_actions)
5636 			return rte_flow_error_set(error, ENOMEM,
5637 						  RTE_FLOW_ERROR_TYPE_ACTION,
5638 						  NULL, "no memory to split "
5639 						  "meter flow");
5640 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5641 			     act_size);
5642 		/* There's no suffix flow for meter of non-default policy. */
5643 		if (!fm->def_policy)
5644 			pre_actions = sfx_actions + 1;
5645 		else
5646 			pre_actions = sfx_actions + actions_n;
5647 		ret = flow_meter_split_prep(dev, flow, fm, &sfx_attr,
5648 					    items, sfx_items, actions,
5649 					    sfx_actions, pre_actions,
5650 					    (set_mtr_reg ? &mtr_flow_id : NULL),
5651 					    error);
5652 		if (ret) {
5653 			ret = -rte_errno;
5654 			goto exit;
5655 		}
5656 		/* Add the prefix subflow. */
5657 		flow_split_info->prefix_mark = 0;
5658 		skip_scale_restore = flow_split_info->skip_scale;
5659 		flow_split_info->skip_scale |=
5660 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5661 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5662 					      attr, items, pre_actions,
5663 					      flow_split_info, error);
5664 		flow_split_info->skip_scale = skip_scale_restore;
5665 		if (ret) {
5666 			if (mtr_flow_id)
5667 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
5668 			ret = -rte_errno;
5669 			goto exit;
5670 		}
5671 		if (mtr_flow_id) {
5672 			dev_flow->handle->split_flow_id = mtr_flow_id;
5673 			dev_flow->handle->is_meter_flow_id = 1;
5674 		}
5675 		if (!fm->def_policy) {
5676 			if (!set_mtr_reg && fm->drop_cnt)
5677 				ret =
5678 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
5679 							&sfx_attr, items,
5680 							flow_split_info,
5681 							fm, error);
5682 			goto exit;
5683 		}
5684 		/* Setting the sfx group atrr. */
5685 		sfx_attr.group = sfx_attr.transfer ?
5686 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5687 				 MLX5_FLOW_TABLE_LEVEL_METER;
5688 		flow_split_info->prefix_layers =
5689 				flow_get_prefix_layer_flags(dev_flow);
5690 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5691 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
5692 	}
5693 	/* Add the prefix subflow. */
5694 	ret = flow_create_split_metadata(dev, flow,
5695 					 &sfx_attr, sfx_items ?
5696 					 sfx_items : items,
5697 					 sfx_actions ? sfx_actions : actions,
5698 					 flow_split_info, error);
5699 exit:
5700 	if (sfx_actions)
5701 		mlx5_free(sfx_actions);
5702 	return ret;
5703 }
5704 
5705 /**
5706  * The splitting for sample feature.
5707  *
5708  * Once Sample action is detected in the action list, the flow actions should
5709  * be split into prefix sub flow and suffix sub flow.
5710  *
5711  * The original items remain in the prefix sub flow, all actions preceding the
5712  * sample action and the sample action itself will be copied to the prefix
5713  * sub flow, the actions following the sample action will be copied to the
5714  * suffix sub flow, Queue action always be located in the suffix sub flow.
5715  *
5716  * In order to make the packet from prefix sub flow matches with suffix sub
5717  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5718  * flow uses tag item with the unique flow id.
5719  *
5720  * @param dev
5721  *   Pointer to Ethernet device.
5722  * @param[in] flow
5723  *   Parent flow structure pointer.
5724  * @param[in] attr
5725  *   Flow rule attributes.
5726  * @param[in] items
5727  *   Pattern specification (list terminated by the END pattern item).
5728  * @param[in] actions
5729  *   Associated actions (list terminated by the END action).
5730  * @param[in] flow_split_info
5731  *   Pointer to flow split info structure.
5732  * @param[out] error
5733  *   Perform verbose error reporting if not NULL.
5734  * @return
5735  *   0 on success, negative value otherwise
5736  */
5737 static int
5738 flow_create_split_sample(struct rte_eth_dev *dev,
5739 			 struct rte_flow *flow,
5740 			 const struct rte_flow_attr *attr,
5741 			 const struct rte_flow_item items[],
5742 			 const struct rte_flow_action actions[],
5743 			 struct mlx5_flow_split_info *flow_split_info,
5744 			 struct rte_flow_error *error)
5745 {
5746 	struct mlx5_priv *priv = dev->data->dev_private;
5747 	struct rte_flow_action *sfx_actions = NULL;
5748 	struct rte_flow_action *pre_actions = NULL;
5749 	struct rte_flow_item *sfx_items = NULL;
5750 	struct mlx5_flow *dev_flow = NULL;
5751 	struct rte_flow_attr sfx_attr = *attr;
5752 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5753 	struct mlx5_flow_dv_sample_resource *sample_res;
5754 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5755 	struct mlx5_flow_tbl_resource *sfx_tbl;
5756 #endif
5757 	size_t act_size;
5758 	size_t item_size;
5759 	uint32_t fdb_tx = 0;
5760 	int32_t tag_id = 0;
5761 	int actions_n = 0;
5762 	int sample_action_pos;
5763 	int qrss_action_pos;
5764 	int add_tag = 0;
5765 	int modify_after_mirror = 0;
5766 	uint16_t jump_table = 0;
5767 	const uint32_t next_ft_step = 1;
5768 	int ret = 0;
5769 
5770 	if (priv->sampler_en)
5771 		actions_n = flow_check_match_action(actions, attr,
5772 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5773 					&sample_action_pos, &qrss_action_pos,
5774 					&modify_after_mirror);
5775 	if (actions_n) {
5776 		/* The prefix actions must includes sample, tag, end. */
5777 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5778 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5779 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5780 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5781 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5782 					  item_size), 0, SOCKET_ID_ANY);
5783 		if (!sfx_actions)
5784 			return rte_flow_error_set(error, ENOMEM,
5785 						  RTE_FLOW_ERROR_TYPE_ACTION,
5786 						  NULL, "no memory to split "
5787 						  "sample flow");
5788 		/* The representor_id is -1 for uplink. */
5789 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5790 		/*
5791 		 * When reg_c_preserve is set, metadata registers Cx preserve
5792 		 * their value even through packet duplication.
5793 		 */
5794 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5795 		if (add_tag)
5796 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5797 					+ act_size);
5798 		if (modify_after_mirror)
5799 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5800 				     next_ft_step;
5801 		pre_actions = sfx_actions + actions_n;
5802 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5803 						actions, sfx_actions,
5804 						pre_actions, actions_n,
5805 						sample_action_pos,
5806 						qrss_action_pos, jump_table,
5807 						error);
5808 		if (tag_id < 0 || (add_tag && !tag_id)) {
5809 			ret = -rte_errno;
5810 			goto exit;
5811 		}
5812 		if (modify_after_mirror)
5813 			flow_split_info->skip_scale =
5814 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5815 		/* Add the prefix subflow. */
5816 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5817 					      items, pre_actions,
5818 					      flow_split_info, error);
5819 		if (ret) {
5820 			ret = -rte_errno;
5821 			goto exit;
5822 		}
5823 		dev_flow->handle->split_flow_id = tag_id;
5824 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5825 		if (!modify_after_mirror) {
5826 			/* Set the sfx group attr. */
5827 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5828 						dev_flow->dv.sample_res;
5829 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5830 						sample_res->normal_path_tbl;
5831 			sfx_tbl_data = container_of(sfx_tbl,
5832 						struct mlx5_flow_tbl_data_entry,
5833 						tbl);
5834 			sfx_attr.group = sfx_attr.transfer ?
5835 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
5836 		} else {
5837 			MLX5_ASSERT(attr->transfer);
5838 			sfx_attr.group = jump_table;
5839 		}
5840 		flow_split_info->prefix_layers =
5841 				flow_get_prefix_layer_flags(dev_flow);
5842 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5843 		/* Suffix group level already be scaled with factor, set
5844 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5845 		 * again in translation.
5846 		 */
5847 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5848 #endif
5849 	}
5850 	/* Add the suffix subflow. */
5851 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5852 				      sfx_items ? sfx_items : items,
5853 				      sfx_actions ? sfx_actions : actions,
5854 				      flow_split_info, error);
5855 exit:
5856 	if (sfx_actions)
5857 		mlx5_free(sfx_actions);
5858 	return ret;
5859 }
5860 
5861 /**
5862  * Split the flow to subflow set. The splitters might be linked
5863  * in the chain, like this:
5864  * flow_create_split_outer() calls:
5865  *   flow_create_split_meter() calls:
5866  *     flow_create_split_metadata(meter_subflow_0) calls:
5867  *       flow_create_split_inner(metadata_subflow_0)
5868  *       flow_create_split_inner(metadata_subflow_1)
5869  *       flow_create_split_inner(metadata_subflow_2)
5870  *     flow_create_split_metadata(meter_subflow_1) calls:
5871  *       flow_create_split_inner(metadata_subflow_0)
5872  *       flow_create_split_inner(metadata_subflow_1)
5873  *       flow_create_split_inner(metadata_subflow_2)
5874  *
5875  * This provide flexible way to add new levels of flow splitting.
5876  * The all of successfully created subflows are included to the
5877  * parent flow dev_flow list.
5878  *
5879  * @param dev
5880  *   Pointer to Ethernet device.
5881  * @param[in] flow
5882  *   Parent flow structure pointer.
5883  * @param[in] attr
5884  *   Flow rule attributes.
5885  * @param[in] items
5886  *   Pattern specification (list terminated by the END pattern item).
5887  * @param[in] actions
5888  *   Associated actions (list terminated by the END action).
5889  * @param[in] flow_split_info
5890  *   Pointer to flow split info structure.
5891  * @param[out] error
5892  *   Perform verbose error reporting if not NULL.
5893  * @return
5894  *   0 on success, negative value otherwise
5895  */
5896 static int
5897 flow_create_split_outer(struct rte_eth_dev *dev,
5898 			struct rte_flow *flow,
5899 			const struct rte_flow_attr *attr,
5900 			const struct rte_flow_item items[],
5901 			const struct rte_flow_action actions[],
5902 			struct mlx5_flow_split_info *flow_split_info,
5903 			struct rte_flow_error *error)
5904 {
5905 	int ret;
5906 
5907 	ret = flow_create_split_sample(dev, flow, attr, items,
5908 				       actions, flow_split_info, error);
5909 	MLX5_ASSERT(ret <= 0);
5910 	return ret;
5911 }
5912 
5913 static struct mlx5_flow_tunnel *
5914 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5915 		      const struct rte_flow_attr *attr,
5916 		      const struct rte_flow_item items[],
5917 		      const struct rte_flow_action actions[])
5918 {
5919 	struct mlx5_flow_tunnel *tunnel;
5920 
5921 #pragma GCC diagnostic push
5922 #pragma GCC diagnostic ignored "-Wcast-qual"
5923 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5924 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5925 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5926 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5927 	else
5928 		tunnel = NULL;
5929 #pragma GCC diagnostic pop
5930 
5931 	return tunnel;
5932 }
5933 
5934 /**
5935  * Adjust flow RSS workspace if needed.
5936  *
5937  * @param wks
5938  *   Pointer to thread flow work space.
5939  * @param rss_desc
5940  *   Pointer to RSS descriptor.
5941  * @param[in] nrssq_num
5942  *   New RSS queue number.
5943  *
5944  * @return
5945  *   0 on success, -1 otherwise and rte_errno is set.
5946  */
5947 static int
5948 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5949 			  struct mlx5_flow_rss_desc *rss_desc,
5950 			  uint32_t nrssq_num)
5951 {
5952 	if (likely(nrssq_num <= wks->rssq_num))
5953 		return 0;
5954 	rss_desc->queue = realloc(rss_desc->queue,
5955 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5956 	if (!rss_desc->queue) {
5957 		rte_errno = ENOMEM;
5958 		return -1;
5959 	}
5960 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5961 	return 0;
5962 }
5963 
5964 /**
5965  * Create a flow and add it to @p list.
5966  *
5967  * @param dev
5968  *   Pointer to Ethernet device.
5969  * @param list
5970  *   Pointer to a TAILQ flow list. If this parameter NULL,
5971  *   no list insertion occurred, flow is just created,
5972  *   this is caller's responsibility to track the
5973  *   created flow.
5974  * @param[in] attr
5975  *   Flow rule attributes.
5976  * @param[in] items
5977  *   Pattern specification (list terminated by the END pattern item).
5978  * @param[in] actions
5979  *   Associated actions (list terminated by the END action).
5980  * @param[in] external
5981  *   This flow rule is created by request external to PMD.
5982  * @param[out] error
5983  *   Perform verbose error reporting if not NULL.
5984  *
5985  * @return
5986  *   A flow index on success, 0 otherwise and rte_errno is set.
5987  */
5988 static uint32_t
5989 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5990 		 const struct rte_flow_attr *attr,
5991 		 const struct rte_flow_item items[],
5992 		 const struct rte_flow_action original_actions[],
5993 		 bool external, struct rte_flow_error *error)
5994 {
5995 	struct mlx5_priv *priv = dev->data->dev_private;
5996 	struct rte_flow *flow = NULL;
5997 	struct mlx5_flow *dev_flow;
5998 	const struct rte_flow_action_rss *rss = NULL;
5999 	struct mlx5_translated_action_handle
6000 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6001 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6002 	union {
6003 		struct mlx5_flow_expand_rss buf;
6004 		uint8_t buffer[2048];
6005 	} expand_buffer;
6006 	union {
6007 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6008 		uint8_t buffer[2048];
6009 	} actions_rx;
6010 	union {
6011 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6012 		uint8_t buffer[2048];
6013 	} actions_hairpin_tx;
6014 	union {
6015 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6016 		uint8_t buffer[2048];
6017 	} items_tx;
6018 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6019 	struct mlx5_flow_rss_desc *rss_desc;
6020 	const struct rte_flow_action *p_actions_rx;
6021 	uint32_t i;
6022 	uint32_t idx = 0;
6023 	int hairpin_flow;
6024 	struct rte_flow_attr attr_tx = { .priority = 0 };
6025 	const struct rte_flow_action *actions;
6026 	struct rte_flow_action *translated_actions = NULL;
6027 	struct mlx5_flow_tunnel *tunnel;
6028 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6029 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6030 	struct mlx5_flow_split_info flow_split_info = {
6031 		.external = !!external,
6032 		.skip_scale = 0,
6033 		.flow_idx = 0,
6034 		.prefix_mark = 0,
6035 		.prefix_layers = 0,
6036 		.table_id = 0
6037 	};
6038 	int ret;
6039 
6040 	MLX5_ASSERT(wks);
6041 	rss_desc = &wks->rss_desc;
6042 	ret = flow_action_handles_translate(dev, original_actions,
6043 					    indir_actions,
6044 					    &indir_actions_n,
6045 					    &translated_actions, error);
6046 	if (ret < 0) {
6047 		MLX5_ASSERT(translated_actions == NULL);
6048 		return 0;
6049 	}
6050 	actions = translated_actions ? translated_actions : original_actions;
6051 	p_actions_rx = actions;
6052 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6053 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6054 				external, hairpin_flow, error);
6055 	if (ret < 0)
6056 		goto error_before_hairpin_split;
6057 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
6058 	if (!flow) {
6059 		rte_errno = ENOMEM;
6060 		goto error_before_hairpin_split;
6061 	}
6062 	if (hairpin_flow > 0) {
6063 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6064 			rte_errno = EINVAL;
6065 			goto error_before_hairpin_split;
6066 		}
6067 		flow_hairpin_split(dev, actions, actions_rx.actions,
6068 				   actions_hairpin_tx.actions, items_tx.items,
6069 				   idx);
6070 		p_actions_rx = actions_rx.actions;
6071 	}
6072 	flow_split_info.flow_idx = idx;
6073 	flow->drv_type = flow_get_drv_type(dev, attr);
6074 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6075 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6076 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6077 	/* RSS Action only works on NIC RX domain */
6078 	if (attr->ingress && !attr->transfer)
6079 		rss = flow_get_rss_action(dev, p_actions_rx);
6080 	if (rss) {
6081 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6082 			return 0;
6083 		/*
6084 		 * The following information is required by
6085 		 * mlx5_flow_hashfields_adjust() in advance.
6086 		 */
6087 		rss_desc->level = rss->level;
6088 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
6089 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
6090 	}
6091 	flow->dev_handles = 0;
6092 	if (rss && rss->types) {
6093 		unsigned int graph_root;
6094 
6095 		graph_root = find_graph_root(items, rss->level);
6096 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6097 					   items, rss->types,
6098 					   mlx5_support_expansion, graph_root);
6099 		MLX5_ASSERT(ret > 0 &&
6100 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6101 	} else {
6102 		buf->entries = 1;
6103 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6104 	}
6105 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6106 						      indir_actions_n);
6107 	for (i = 0; i < buf->entries; ++i) {
6108 		/* Initialize flow split data. */
6109 		flow_split_info.prefix_layers = 0;
6110 		flow_split_info.prefix_mark = 0;
6111 		flow_split_info.skip_scale = 0;
6112 		/*
6113 		 * The splitter may create multiple dev_flows,
6114 		 * depending on configuration. In the simplest
6115 		 * case it just creates unmodified original flow.
6116 		 */
6117 		ret = flow_create_split_outer(dev, flow, attr,
6118 					      buf->entry[i].pattern,
6119 					      p_actions_rx, &flow_split_info,
6120 					      error);
6121 		if (ret < 0)
6122 			goto error;
6123 		if (is_flow_tunnel_steer_rule(dev, attr,
6124 					      buf->entry[i].pattern,
6125 					      p_actions_rx)) {
6126 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6127 							   p_actions_rx,
6128 							   idx,
6129 							   &default_miss_ctx,
6130 							   error);
6131 			if (ret < 0) {
6132 				mlx5_free(default_miss_ctx.queue);
6133 				goto error;
6134 			}
6135 		}
6136 	}
6137 	/* Create the tx flow. */
6138 	if (hairpin_flow) {
6139 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6140 		attr_tx.ingress = 0;
6141 		attr_tx.egress = 1;
6142 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6143 					 actions_hairpin_tx.actions,
6144 					 idx, error);
6145 		if (!dev_flow)
6146 			goto error;
6147 		dev_flow->flow = flow;
6148 		dev_flow->external = 0;
6149 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6150 			      dev_flow->handle, next);
6151 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6152 					 items_tx.items,
6153 					 actions_hairpin_tx.actions, error);
6154 		if (ret < 0)
6155 			goto error;
6156 	}
6157 	/*
6158 	 * Update the metadata register copy table. If extensive
6159 	 * metadata feature is enabled and registers are supported
6160 	 * we might create the extra rte_flow for each unique
6161 	 * MARK/FLAG action ID.
6162 	 *
6163 	 * The table is updated for ingress Flows only, because
6164 	 * the egress Flows belong to the different device and
6165 	 * copy table should be updated in peer NIC Rx domain.
6166 	 */
6167 	if (attr->ingress &&
6168 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6169 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6170 		if (ret)
6171 			goto error;
6172 	}
6173 	/*
6174 	 * If the flow is external (from application) OR device is started,
6175 	 * OR mreg discover, then apply immediately.
6176 	 */
6177 	if (external || dev->data->dev_started ||
6178 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6179 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6180 		ret = flow_drv_apply(dev, flow, error);
6181 		if (ret < 0)
6182 			goto error;
6183 	}
6184 	if (list) {
6185 		rte_spinlock_lock(&priv->flow_list_lock);
6186 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
6187 			     flow, next);
6188 		rte_spinlock_unlock(&priv->flow_list_lock);
6189 	}
6190 	flow_rxq_flags_set(dev, flow);
6191 	rte_free(translated_actions);
6192 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
6193 	if (tunnel) {
6194 		flow->tunnel = 1;
6195 		flow->tunnel_id = tunnel->tunnel_id;
6196 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6197 		mlx5_free(default_miss_ctx.queue);
6198 	}
6199 	mlx5_flow_pop_thread_workspace();
6200 	return idx;
6201 error:
6202 	MLX5_ASSERT(flow);
6203 	ret = rte_errno; /* Save rte_errno before cleanup. */
6204 	flow_mreg_del_copy_action(dev, flow);
6205 	flow_drv_destroy(dev, flow);
6206 	if (rss_desc->shared_rss)
6207 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6208 			mlx5_ipool_get
6209 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6210 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6211 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
6212 	rte_errno = ret; /* Restore rte_errno. */
6213 	ret = rte_errno;
6214 	rte_errno = ret;
6215 	mlx5_flow_pop_thread_workspace();
6216 error_before_hairpin_split:
6217 	rte_free(translated_actions);
6218 	return 0;
6219 }
6220 
6221 /**
6222  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6223  * incoming packets to table 1.
6224  *
6225  * Other flow rules, requested for group n, will be created in
6226  * e-switch table n+1.
6227  * Jump action to e-switch group n will be created to group n+1.
6228  *
6229  * Used when working in switchdev mode, to utilise advantages of table 1
6230  * and above.
6231  *
6232  * @param dev
6233  *   Pointer to Ethernet device.
6234  *
6235  * @return
6236  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6237  */
6238 struct rte_flow *
6239 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6240 {
6241 	const struct rte_flow_attr attr = {
6242 		.group = 0,
6243 		.priority = 0,
6244 		.ingress = 1,
6245 		.egress = 0,
6246 		.transfer = 1,
6247 	};
6248 	const struct rte_flow_item pattern = {
6249 		.type = RTE_FLOW_ITEM_TYPE_END,
6250 	};
6251 	struct rte_flow_action_jump jump = {
6252 		.group = 1,
6253 	};
6254 	const struct rte_flow_action actions[] = {
6255 		{
6256 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6257 			.conf = &jump,
6258 		},
6259 		{
6260 			.type = RTE_FLOW_ACTION_TYPE_END,
6261 		},
6262 	};
6263 	struct mlx5_priv *priv = dev->data->dev_private;
6264 	struct rte_flow_error error;
6265 
6266 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
6267 						   &attr, &pattern,
6268 						   actions, false, &error);
6269 }
6270 
6271 /**
6272  * Validate a flow supported by the NIC.
6273  *
6274  * @see rte_flow_validate()
6275  * @see rte_flow_ops
6276  */
6277 int
6278 mlx5_flow_validate(struct rte_eth_dev *dev,
6279 		   const struct rte_flow_attr *attr,
6280 		   const struct rte_flow_item items[],
6281 		   const struct rte_flow_action original_actions[],
6282 		   struct rte_flow_error *error)
6283 {
6284 	int hairpin_flow;
6285 	struct mlx5_translated_action_handle
6286 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6287 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6288 	const struct rte_flow_action *actions;
6289 	struct rte_flow_action *translated_actions = NULL;
6290 	int ret = flow_action_handles_translate(dev, original_actions,
6291 						indir_actions,
6292 						&indir_actions_n,
6293 						&translated_actions, error);
6294 
6295 	if (ret)
6296 		return ret;
6297 	actions = translated_actions ? translated_actions : original_actions;
6298 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6299 	ret = flow_drv_validate(dev, attr, items, actions,
6300 				true, hairpin_flow, error);
6301 	rte_free(translated_actions);
6302 	return ret;
6303 }
6304 
6305 /**
6306  * Create a flow.
6307  *
6308  * @see rte_flow_create()
6309  * @see rte_flow_ops
6310  */
6311 struct rte_flow *
6312 mlx5_flow_create(struct rte_eth_dev *dev,
6313 		 const struct rte_flow_attr *attr,
6314 		 const struct rte_flow_item items[],
6315 		 const struct rte_flow_action actions[],
6316 		 struct rte_flow_error *error)
6317 {
6318 	struct mlx5_priv *priv = dev->data->dev_private;
6319 
6320 	/*
6321 	 * If the device is not started yet, it is not allowed to created a
6322 	 * flow from application. PMD default flows and traffic control flows
6323 	 * are not affected.
6324 	 */
6325 	if (unlikely(!dev->data->dev_started)) {
6326 		DRV_LOG(DEBUG, "port %u is not started when "
6327 			"inserting a flow", dev->data->port_id);
6328 		rte_flow_error_set(error, ENODEV,
6329 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6330 				   NULL,
6331 				   "port not started");
6332 		return NULL;
6333 	}
6334 
6335 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
6336 				  attr, items, actions, true, error);
6337 }
6338 
6339 /**
6340  * Destroy a flow in a list.
6341  *
6342  * @param dev
6343  *   Pointer to Ethernet device.
6344  * @param list
6345  *   Pointer to the Indexed flow list. If this parameter NULL,
6346  *   there is no flow removal from the list. Be noted that as
6347  *   flow is add to the indexed list, memory of the indexed
6348  *   list points to maybe changed as flow destroyed.
6349  * @param[in] flow_idx
6350  *   Index of flow to destroy.
6351  */
6352 static void
6353 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
6354 		  uint32_t flow_idx)
6355 {
6356 	struct mlx5_priv *priv = dev->data->dev_private;
6357 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6358 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
6359 
6360 	if (!flow)
6361 		return;
6362 	/*
6363 	 * Update RX queue flags only if port is started, otherwise it is
6364 	 * already clean.
6365 	 */
6366 	if (dev->data->dev_started)
6367 		flow_rxq_flags_trim(dev, flow);
6368 	flow_drv_destroy(dev, flow);
6369 	if (list) {
6370 		rte_spinlock_lock(&priv->flow_list_lock);
6371 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
6372 			     flow_idx, flow, next);
6373 		rte_spinlock_unlock(&priv->flow_list_lock);
6374 	}
6375 	if (flow->tunnel) {
6376 		struct mlx5_flow_tunnel *tunnel;
6377 
6378 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6379 		RTE_VERIFY(tunnel);
6380 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6381 			mlx5_flow_tunnel_free(dev, tunnel);
6382 	}
6383 	flow_mreg_del_copy_action(dev, flow);
6384 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6385 }
6386 
6387 /**
6388  * Destroy all flows.
6389  *
6390  * @param dev
6391  *   Pointer to Ethernet device.
6392  * @param list
6393  *   Pointer to the Indexed flow list.
6394  * @param active
6395  *   If flushing is called avtively.
6396  */
6397 void
6398 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
6399 {
6400 	uint32_t num_flushed = 0;
6401 
6402 	while (*list) {
6403 		flow_list_destroy(dev, list, *list);
6404 		num_flushed++;
6405 	}
6406 	if (active) {
6407 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6408 			dev->data->port_id, num_flushed);
6409 	}
6410 }
6411 
6412 /**
6413  * Stop all default actions for flows.
6414  *
6415  * @param dev
6416  *   Pointer to Ethernet device.
6417  */
6418 void
6419 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6420 {
6421 	flow_mreg_del_default_copy_action(dev);
6422 	flow_rxq_flags_clear(dev);
6423 }
6424 
6425 /**
6426  * Start all default actions for flows.
6427  *
6428  * @param dev
6429  *   Pointer to Ethernet device.
6430  * @return
6431  *   0 on success, a negative errno value otherwise and rte_errno is set.
6432  */
6433 int
6434 mlx5_flow_start_default(struct rte_eth_dev *dev)
6435 {
6436 	struct rte_flow_error error;
6437 
6438 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6439 	return flow_mreg_add_default_copy_action(dev, &error);
6440 }
6441 
6442 /**
6443  * Release key of thread specific flow workspace data.
6444  */
6445 void
6446 flow_release_workspace(void *data)
6447 {
6448 	struct mlx5_flow_workspace *wks = data;
6449 	struct mlx5_flow_workspace *next;
6450 
6451 	while (wks) {
6452 		next = wks->next;
6453 		free(wks->rss_desc.queue);
6454 		free(wks);
6455 		wks = next;
6456 	}
6457 }
6458 
6459 /**
6460  * Get thread specific current flow workspace.
6461  *
6462  * @return pointer to thread specific flow workspace data, NULL on error.
6463  */
6464 struct mlx5_flow_workspace*
6465 mlx5_flow_get_thread_workspace(void)
6466 {
6467 	struct mlx5_flow_workspace *data;
6468 
6469 	data = mlx5_flow_os_get_specific_workspace();
6470 	MLX5_ASSERT(data && data->inuse);
6471 	if (!data || !data->inuse)
6472 		DRV_LOG(ERR, "flow workspace not initialized.");
6473 	return data;
6474 }
6475 
6476 /**
6477  * Allocate and init new flow workspace.
6478  *
6479  * @return pointer to flow workspace data, NULL on error.
6480  */
6481 static struct mlx5_flow_workspace*
6482 flow_alloc_thread_workspace(void)
6483 {
6484 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6485 
6486 	if (!data) {
6487 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6488 			"memory.");
6489 		return NULL;
6490 	}
6491 	data->rss_desc.queue = calloc(1,
6492 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6493 	if (!data->rss_desc.queue)
6494 		goto err;
6495 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6496 	return data;
6497 err:
6498 	if (data->rss_desc.queue)
6499 		free(data->rss_desc.queue);
6500 	free(data);
6501 	return NULL;
6502 }
6503 
6504 /**
6505  * Get new thread specific flow workspace.
6506  *
6507  * If current workspace inuse, create new one and set as current.
6508  *
6509  * @return pointer to thread specific flow workspace data, NULL on error.
6510  */
6511 static struct mlx5_flow_workspace*
6512 mlx5_flow_push_thread_workspace(void)
6513 {
6514 	struct mlx5_flow_workspace *curr;
6515 	struct mlx5_flow_workspace *data;
6516 
6517 	curr = mlx5_flow_os_get_specific_workspace();
6518 	if (!curr) {
6519 		data = flow_alloc_thread_workspace();
6520 		if (!data)
6521 			return NULL;
6522 	} else if (!curr->inuse) {
6523 		data = curr;
6524 	} else if (curr->next) {
6525 		data = curr->next;
6526 	} else {
6527 		data = flow_alloc_thread_workspace();
6528 		if (!data)
6529 			return NULL;
6530 		curr->next = data;
6531 		data->prev = curr;
6532 	}
6533 	data->inuse = 1;
6534 	data->flow_idx = 0;
6535 	/* Set as current workspace */
6536 	if (mlx5_flow_os_set_specific_workspace(data))
6537 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6538 	return data;
6539 }
6540 
6541 /**
6542  * Close current thread specific flow workspace.
6543  *
6544  * If previous workspace available, set it as current.
6545  *
6546  * @return pointer to thread specific flow workspace data, NULL on error.
6547  */
6548 static void
6549 mlx5_flow_pop_thread_workspace(void)
6550 {
6551 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6552 
6553 	if (!data)
6554 		return;
6555 	if (!data->inuse) {
6556 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6557 		return;
6558 	}
6559 	data->inuse = 0;
6560 	if (!data->prev)
6561 		return;
6562 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6563 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6564 }
6565 
6566 /**
6567  * Verify the flow list is empty
6568  *
6569  * @param dev
6570  *  Pointer to Ethernet device.
6571  *
6572  * @return the number of flows not released.
6573  */
6574 int
6575 mlx5_flow_verify(struct rte_eth_dev *dev)
6576 {
6577 	struct mlx5_priv *priv = dev->data->dev_private;
6578 	struct rte_flow *flow;
6579 	uint32_t idx;
6580 	int ret = 0;
6581 
6582 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6583 		      flow, next) {
6584 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6585 			dev->data->port_id, (void *)flow);
6586 		++ret;
6587 	}
6588 	return ret;
6589 }
6590 
6591 /**
6592  * Enable default hairpin egress flow.
6593  *
6594  * @param dev
6595  *   Pointer to Ethernet device.
6596  * @param queue
6597  *   The queue index.
6598  *
6599  * @return
6600  *   0 on success, a negative errno value otherwise and rte_errno is set.
6601  */
6602 int
6603 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6604 			    uint32_t queue)
6605 {
6606 	struct mlx5_priv *priv = dev->data->dev_private;
6607 	const struct rte_flow_attr attr = {
6608 		.egress = 1,
6609 		.priority = 0,
6610 	};
6611 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6612 		.queue = queue,
6613 	};
6614 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6615 		.queue = UINT32_MAX,
6616 	};
6617 	struct rte_flow_item items[] = {
6618 		{
6619 			.type = (enum rte_flow_item_type)
6620 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6621 			.spec = &queue_spec,
6622 			.last = NULL,
6623 			.mask = &queue_mask,
6624 		},
6625 		{
6626 			.type = RTE_FLOW_ITEM_TYPE_END,
6627 		},
6628 	};
6629 	struct rte_flow_action_jump jump = {
6630 		.group = MLX5_HAIRPIN_TX_TABLE,
6631 	};
6632 	struct rte_flow_action actions[2];
6633 	uint32_t flow_idx;
6634 	struct rte_flow_error error;
6635 
6636 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6637 	actions[0].conf = &jump;
6638 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6639 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6640 				&attr, items, actions, false, &error);
6641 	if (!flow_idx) {
6642 		DRV_LOG(DEBUG,
6643 			"Failed to create ctrl flow: rte_errno(%d),"
6644 			" type(%d), message(%s)",
6645 			rte_errno, error.type,
6646 			error.message ? error.message : " (no stated reason)");
6647 		return -rte_errno;
6648 	}
6649 	return 0;
6650 }
6651 
6652 /**
6653  * Enable a control flow configured from the control plane.
6654  *
6655  * @param dev
6656  *   Pointer to Ethernet device.
6657  * @param eth_spec
6658  *   An Ethernet flow spec to apply.
6659  * @param eth_mask
6660  *   An Ethernet flow mask to apply.
6661  * @param vlan_spec
6662  *   A VLAN flow spec to apply.
6663  * @param vlan_mask
6664  *   A VLAN flow mask to apply.
6665  *
6666  * @return
6667  *   0 on success, a negative errno value otherwise and rte_errno is set.
6668  */
6669 int
6670 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6671 		    struct rte_flow_item_eth *eth_spec,
6672 		    struct rte_flow_item_eth *eth_mask,
6673 		    struct rte_flow_item_vlan *vlan_spec,
6674 		    struct rte_flow_item_vlan *vlan_mask)
6675 {
6676 	struct mlx5_priv *priv = dev->data->dev_private;
6677 	const struct rte_flow_attr attr = {
6678 		.ingress = 1,
6679 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6680 	};
6681 	struct rte_flow_item items[] = {
6682 		{
6683 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6684 			.spec = eth_spec,
6685 			.last = NULL,
6686 			.mask = eth_mask,
6687 		},
6688 		{
6689 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6690 					      RTE_FLOW_ITEM_TYPE_END,
6691 			.spec = vlan_spec,
6692 			.last = NULL,
6693 			.mask = vlan_mask,
6694 		},
6695 		{
6696 			.type = RTE_FLOW_ITEM_TYPE_END,
6697 		},
6698 	};
6699 	uint16_t queue[priv->reta_idx_n];
6700 	struct rte_flow_action_rss action_rss = {
6701 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6702 		.level = 0,
6703 		.types = priv->rss_conf.rss_hf,
6704 		.key_len = priv->rss_conf.rss_key_len,
6705 		.queue_num = priv->reta_idx_n,
6706 		.key = priv->rss_conf.rss_key,
6707 		.queue = queue,
6708 	};
6709 	struct rte_flow_action actions[] = {
6710 		{
6711 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6712 			.conf = &action_rss,
6713 		},
6714 		{
6715 			.type = RTE_FLOW_ACTION_TYPE_END,
6716 		},
6717 	};
6718 	uint32_t flow_idx;
6719 	struct rte_flow_error error;
6720 	unsigned int i;
6721 
6722 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6723 		return 0;
6724 	}
6725 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6726 		action_rss.types = 0;
6727 	for (i = 0; i != priv->reta_idx_n; ++i)
6728 		queue[i] = (*priv->reta_idx)[i];
6729 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6730 				&attr, items, actions, false, &error);
6731 	if (!flow_idx)
6732 		return -rte_errno;
6733 	return 0;
6734 }
6735 
6736 /**
6737  * Enable a flow control configured from the control plane.
6738  *
6739  * @param dev
6740  *   Pointer to Ethernet device.
6741  * @param eth_spec
6742  *   An Ethernet flow spec to apply.
6743  * @param eth_mask
6744  *   An Ethernet flow mask to apply.
6745  *
6746  * @return
6747  *   0 on success, a negative errno value otherwise and rte_errno is set.
6748  */
6749 int
6750 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6751 	       struct rte_flow_item_eth *eth_spec,
6752 	       struct rte_flow_item_eth *eth_mask)
6753 {
6754 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6755 }
6756 
6757 /**
6758  * Create default miss flow rule matching lacp traffic
6759  *
6760  * @param dev
6761  *   Pointer to Ethernet device.
6762  * @param eth_spec
6763  *   An Ethernet flow spec to apply.
6764  *
6765  * @return
6766  *   0 on success, a negative errno value otherwise and rte_errno is set.
6767  */
6768 int
6769 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6770 {
6771 	struct mlx5_priv *priv = dev->data->dev_private;
6772 	/*
6773 	 * The LACP matching is done by only using ether type since using
6774 	 * a multicast dst mac causes kernel to give low priority to this flow.
6775 	 */
6776 	static const struct rte_flow_item_eth lacp_spec = {
6777 		.type = RTE_BE16(0x8809),
6778 	};
6779 	static const struct rte_flow_item_eth lacp_mask = {
6780 		.type = 0xffff,
6781 	};
6782 	const struct rte_flow_attr attr = {
6783 		.ingress = 1,
6784 	};
6785 	struct rte_flow_item items[] = {
6786 		{
6787 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6788 			.spec = &lacp_spec,
6789 			.mask = &lacp_mask,
6790 		},
6791 		{
6792 			.type = RTE_FLOW_ITEM_TYPE_END,
6793 		},
6794 	};
6795 	struct rte_flow_action actions[] = {
6796 		{
6797 			.type = (enum rte_flow_action_type)
6798 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6799 		},
6800 		{
6801 			.type = RTE_FLOW_ACTION_TYPE_END,
6802 		},
6803 	};
6804 	struct rte_flow_error error;
6805 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6806 				&attr, items, actions, false, &error);
6807 
6808 	if (!flow_idx)
6809 		return -rte_errno;
6810 	return 0;
6811 }
6812 
6813 /**
6814  * Destroy a flow.
6815  *
6816  * @see rte_flow_destroy()
6817  * @see rte_flow_ops
6818  */
6819 int
6820 mlx5_flow_destroy(struct rte_eth_dev *dev,
6821 		  struct rte_flow *flow,
6822 		  struct rte_flow_error *error __rte_unused)
6823 {
6824 	struct mlx5_priv *priv = dev->data->dev_private;
6825 
6826 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6827 	return 0;
6828 }
6829 
6830 /**
6831  * Destroy all flows.
6832  *
6833  * @see rte_flow_flush()
6834  * @see rte_flow_ops
6835  */
6836 int
6837 mlx5_flow_flush(struct rte_eth_dev *dev,
6838 		struct rte_flow_error *error __rte_unused)
6839 {
6840 	struct mlx5_priv *priv = dev->data->dev_private;
6841 
6842 	mlx5_flow_list_flush(dev, &priv->flows, false);
6843 	return 0;
6844 }
6845 
6846 /**
6847  * Isolated mode.
6848  *
6849  * @see rte_flow_isolate()
6850  * @see rte_flow_ops
6851  */
6852 int
6853 mlx5_flow_isolate(struct rte_eth_dev *dev,
6854 		  int enable,
6855 		  struct rte_flow_error *error)
6856 {
6857 	struct mlx5_priv *priv = dev->data->dev_private;
6858 
6859 	if (dev->data->dev_started) {
6860 		rte_flow_error_set(error, EBUSY,
6861 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6862 				   NULL,
6863 				   "port must be stopped first");
6864 		return -rte_errno;
6865 	}
6866 	priv->isolated = !!enable;
6867 	if (enable)
6868 		dev->dev_ops = &mlx5_dev_ops_isolate;
6869 	else
6870 		dev->dev_ops = &mlx5_dev_ops;
6871 
6872 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6873 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6874 
6875 	return 0;
6876 }
6877 
6878 /**
6879  * Query a flow.
6880  *
6881  * @see rte_flow_query()
6882  * @see rte_flow_ops
6883  */
6884 static int
6885 flow_drv_query(struct rte_eth_dev *dev,
6886 	       uint32_t flow_idx,
6887 	       const struct rte_flow_action *actions,
6888 	       void *data,
6889 	       struct rte_flow_error *error)
6890 {
6891 	struct mlx5_priv *priv = dev->data->dev_private;
6892 	const struct mlx5_flow_driver_ops *fops;
6893 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6894 					       [MLX5_IPOOL_RTE_FLOW],
6895 					       flow_idx);
6896 	enum mlx5_flow_drv_type ftype;
6897 
6898 	if (!flow) {
6899 		return rte_flow_error_set(error, ENOENT,
6900 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6901 			  NULL,
6902 			  "invalid flow handle");
6903 	}
6904 	ftype = flow->drv_type;
6905 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6906 	fops = flow_get_drv_ops(ftype);
6907 
6908 	return fops->query(dev, flow, actions, data, error);
6909 }
6910 
6911 /**
6912  * Query a flow.
6913  *
6914  * @see rte_flow_query()
6915  * @see rte_flow_ops
6916  */
6917 int
6918 mlx5_flow_query(struct rte_eth_dev *dev,
6919 		struct rte_flow *flow,
6920 		const struct rte_flow_action *actions,
6921 		void *data,
6922 		struct rte_flow_error *error)
6923 {
6924 	int ret;
6925 
6926 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6927 			     error);
6928 	if (ret < 0)
6929 		return ret;
6930 	return 0;
6931 }
6932 
6933 /**
6934  * Get rte_flow callbacks.
6935  *
6936  * @param dev
6937  *   Pointer to Ethernet device structure.
6938  * @param ops
6939  *   Pointer to operation-specific structure.
6940  *
6941  * @return 0
6942  */
6943 int
6944 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
6945 		  const struct rte_flow_ops **ops)
6946 {
6947 	*ops = &mlx5_flow_ops;
6948 	return 0;
6949 }
6950 
6951 /**
6952  * Validate meter policy actions.
6953  * Dispatcher for action type specific validation.
6954  *
6955  * @param[in] dev
6956  *   Pointer to the Ethernet device structure.
6957  * @param[in] action
6958  *   The meter policy action object to validate.
6959  * @param[in] attr
6960  *   Attributes of flow to determine steering domain.
6961  * @param[out] is_rss
6962  *   Is RSS or not.
6963  * @param[out] domain_bitmap
6964  *   Domain bitmap.
6965  * @param[out] is_def_policy
6966  *   Is default policy or not.
6967  * @param[out] error
6968  *   Perform verbose error reporting if not NULL. Initialized in case of
6969  *   error only.
6970  *
6971  * @return
6972  *   0 on success, otherwise negative errno value.
6973  */
6974 int
6975 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
6976 			const struct rte_flow_action *actions[RTE_COLORS],
6977 			struct rte_flow_attr *attr,
6978 			bool *is_rss,
6979 			uint8_t *domain_bitmap,
6980 			bool *is_def_policy,
6981 			struct rte_mtr_error *error)
6982 {
6983 	const struct mlx5_flow_driver_ops *fops;
6984 
6985 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6986 	return fops->validate_mtr_acts(dev, actions, attr,
6987 			is_rss, domain_bitmap, is_def_policy, error);
6988 }
6989 
6990 /**
6991  * Destroy the meter table set.
6992  *
6993  * @param[in] dev
6994  *   Pointer to Ethernet device.
6995  * @param[in] mtr_policy
6996  *   Meter policy struct.
6997  */
6998 void
6999 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7000 		      struct mlx5_flow_meter_policy *mtr_policy)
7001 {
7002 	const struct mlx5_flow_driver_ops *fops;
7003 
7004 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7005 	fops->destroy_mtr_acts(dev, mtr_policy);
7006 }
7007 
7008 /**
7009  * Create policy action, lock free,
7010  * (mutex should be acquired by caller).
7011  * Dispatcher for action type specific call.
7012  *
7013  * @param[in] dev
7014  *   Pointer to the Ethernet device structure.
7015  * @param[in] mtr_policy
7016  *   Meter policy struct.
7017  * @param[in] action
7018  *   Action specification used to create meter actions.
7019  * @param[out] error
7020  *   Perform verbose error reporting if not NULL. Initialized in case of
7021  *   error only.
7022  *
7023  * @return
7024  *   0 on success, otherwise negative errno value.
7025  */
7026 int
7027 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7028 		      struct mlx5_flow_meter_policy *mtr_policy,
7029 		      const struct rte_flow_action *actions[RTE_COLORS],
7030 		      struct rte_mtr_error *error)
7031 {
7032 	const struct mlx5_flow_driver_ops *fops;
7033 
7034 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7035 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7036 }
7037 
7038 /**
7039  * Create policy rules, lock free,
7040  * (mutex should be acquired by caller).
7041  * Dispatcher for action type specific call.
7042  *
7043  * @param[in] dev
7044  *   Pointer to the Ethernet device structure.
7045  * @param[in] mtr_policy
7046  *   Meter policy struct.
7047  *
7048  * @return
7049  *   0 on success, -1 otherwise.
7050  */
7051 int
7052 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7053 			     struct mlx5_flow_meter_policy *mtr_policy)
7054 {
7055 	const struct mlx5_flow_driver_ops *fops;
7056 
7057 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7058 	return fops->create_policy_rules(dev, mtr_policy);
7059 }
7060 
7061 /**
7062  * Destroy policy rules, lock free,
7063  * (mutex should be acquired by caller).
7064  * Dispatcher for action type specific call.
7065  *
7066  * @param[in] dev
7067  *   Pointer to the Ethernet device structure.
7068  * @param[in] mtr_policy
7069  *   Meter policy struct.
7070  */
7071 void
7072 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7073 			     struct mlx5_flow_meter_policy *mtr_policy)
7074 {
7075 	const struct mlx5_flow_driver_ops *fops;
7076 
7077 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7078 	fops->destroy_policy_rules(dev, mtr_policy);
7079 }
7080 
7081 /**
7082  * Destroy the default policy table set.
7083  *
7084  * @param[in] dev
7085  *   Pointer to Ethernet device.
7086  */
7087 void
7088 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7089 {
7090 	const struct mlx5_flow_driver_ops *fops;
7091 
7092 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7093 	fops->destroy_def_policy(dev);
7094 }
7095 
7096 /**
7097  * Destroy the default policy table set.
7098  *
7099  * @param[in] dev
7100  *   Pointer to Ethernet device.
7101  *
7102  * @return
7103  *   0 on success, -1 otherwise.
7104  */
7105 int
7106 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7107 {
7108 	const struct mlx5_flow_driver_ops *fops;
7109 
7110 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7111 	return fops->create_def_policy(dev);
7112 }
7113 
7114 /**
7115  * Create the needed meter and suffix tables.
7116  *
7117  * @param[in] dev
7118  *   Pointer to Ethernet device.
7119  *
7120  * @return
7121  *   0 on success, -1 otherwise.
7122  */
7123 int
7124 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7125 			struct mlx5_flow_meter_info *fm,
7126 			uint32_t mtr_idx,
7127 			uint8_t domain_bitmap)
7128 {
7129 	const struct mlx5_flow_driver_ops *fops;
7130 
7131 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7132 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7133 }
7134 
7135 /**
7136  * Destroy the meter table set.
7137  *
7138  * @param[in] dev
7139  *   Pointer to Ethernet device.
7140  * @param[in] tbl
7141  *   Pointer to the meter table set.
7142  */
7143 void
7144 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7145 			   struct mlx5_flow_meter_info *fm)
7146 {
7147 	const struct mlx5_flow_driver_ops *fops;
7148 
7149 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7150 	fops->destroy_mtr_tbls(dev, fm);
7151 }
7152 
7153 /**
7154  * Destroy the global meter drop table.
7155  *
7156  * @param[in] dev
7157  *   Pointer to Ethernet device.
7158  */
7159 void
7160 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7161 {
7162 	const struct mlx5_flow_driver_ops *fops;
7163 
7164 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7165 	fops->destroy_mtr_drop_tbls(dev);
7166 }
7167 
7168 /**
7169  * Allocate the needed aso flow meter id.
7170  *
7171  * @param[in] dev
7172  *   Pointer to Ethernet device.
7173  *
7174  * @return
7175  *   Index to aso flow meter on success, NULL otherwise.
7176  */
7177 uint32_t
7178 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7179 {
7180 	const struct mlx5_flow_driver_ops *fops;
7181 
7182 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7183 	return fops->create_meter(dev);
7184 }
7185 
7186 /**
7187  * Free the aso flow meter id.
7188  *
7189  * @param[in] dev
7190  *   Pointer to Ethernet device.
7191  * @param[in] mtr_idx
7192  *  Index to aso flow meter to be free.
7193  *
7194  * @return
7195  *   0 on success.
7196  */
7197 void
7198 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7199 {
7200 	const struct mlx5_flow_driver_ops *fops;
7201 
7202 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7203 	fops->free_meter(dev, mtr_idx);
7204 }
7205 
7206 /**
7207  * Allocate a counter.
7208  *
7209  * @param[in] dev
7210  *   Pointer to Ethernet device structure.
7211  *
7212  * @return
7213  *   Index to allocated counter  on success, 0 otherwise.
7214  */
7215 uint32_t
7216 mlx5_counter_alloc(struct rte_eth_dev *dev)
7217 {
7218 	const struct mlx5_flow_driver_ops *fops;
7219 	struct rte_flow_attr attr = { .transfer = 0 };
7220 
7221 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7222 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7223 		return fops->counter_alloc(dev);
7224 	}
7225 	DRV_LOG(ERR,
7226 		"port %u counter allocate is not supported.",
7227 		 dev->data->port_id);
7228 	return 0;
7229 }
7230 
7231 /**
7232  * Free a counter.
7233  *
7234  * @param[in] dev
7235  *   Pointer to Ethernet device structure.
7236  * @param[in] cnt
7237  *   Index to counter to be free.
7238  */
7239 void
7240 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7241 {
7242 	const struct mlx5_flow_driver_ops *fops;
7243 	struct rte_flow_attr attr = { .transfer = 0 };
7244 
7245 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7246 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7247 		fops->counter_free(dev, cnt);
7248 		return;
7249 	}
7250 	DRV_LOG(ERR,
7251 		"port %u counter free is not supported.",
7252 		 dev->data->port_id);
7253 }
7254 
7255 /**
7256  * Query counter statistics.
7257  *
7258  * @param[in] dev
7259  *   Pointer to Ethernet device structure.
7260  * @param[in] cnt
7261  *   Index to counter to query.
7262  * @param[in] clear
7263  *   Set to clear counter statistics.
7264  * @param[out] pkts
7265  *   The counter hits packets number to save.
7266  * @param[out] bytes
7267  *   The counter hits bytes number to save.
7268  *
7269  * @return
7270  *   0 on success, a negative errno value otherwise.
7271  */
7272 int
7273 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7274 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7275 {
7276 	const struct mlx5_flow_driver_ops *fops;
7277 	struct rte_flow_attr attr = { .transfer = 0 };
7278 
7279 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7280 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7281 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7282 	}
7283 	DRV_LOG(ERR,
7284 		"port %u counter query is not supported.",
7285 		 dev->data->port_id);
7286 	return -ENOTSUP;
7287 }
7288 
7289 /**
7290  * Allocate a new memory for the counter values wrapped by all the needed
7291  * management.
7292  *
7293  * @param[in] sh
7294  *   Pointer to mlx5_dev_ctx_shared object.
7295  *
7296  * @return
7297  *   0 on success, a negative errno value otherwise.
7298  */
7299 static int
7300 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7301 {
7302 	struct mlx5_devx_mkey_attr mkey_attr;
7303 	struct mlx5_counter_stats_mem_mng *mem_mng;
7304 	volatile struct flow_counter_stats *raw_data;
7305 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7306 	int size = (sizeof(struct flow_counter_stats) *
7307 			MLX5_COUNTERS_PER_POOL +
7308 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7309 			sizeof(struct mlx5_counter_stats_mem_mng);
7310 	size_t pgsize = rte_mem_page_size();
7311 	uint8_t *mem;
7312 	int i;
7313 
7314 	if (pgsize == (size_t)-1) {
7315 		DRV_LOG(ERR, "Failed to get mem page size");
7316 		rte_errno = ENOMEM;
7317 		return -ENOMEM;
7318 	}
7319 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7320 	if (!mem) {
7321 		rte_errno = ENOMEM;
7322 		return -ENOMEM;
7323 	}
7324 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7325 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7326 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
7327 						 IBV_ACCESS_LOCAL_WRITE);
7328 	if (!mem_mng->umem) {
7329 		rte_errno = errno;
7330 		mlx5_free(mem);
7331 		return -rte_errno;
7332 	}
7333 	mkey_attr.addr = (uintptr_t)mem;
7334 	mkey_attr.size = size;
7335 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7336 	mkey_attr.pd = sh->pdn;
7337 	mkey_attr.log_entity_size = 0;
7338 	mkey_attr.pg_access = 0;
7339 	mkey_attr.klm_array = NULL;
7340 	mkey_attr.klm_num = 0;
7341 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
7342 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
7343 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7344 	if (!mem_mng->dm) {
7345 		mlx5_os_umem_dereg(mem_mng->umem);
7346 		rte_errno = errno;
7347 		mlx5_free(mem);
7348 		return -rte_errno;
7349 	}
7350 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7351 	raw_data = (volatile struct flow_counter_stats *)mem;
7352 	for (i = 0; i < raws_n; ++i) {
7353 		mem_mng->raws[i].mem_mng = mem_mng;
7354 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7355 	}
7356 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7357 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7358 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7359 				 next);
7360 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7361 	sh->cmng.mem_mng = mem_mng;
7362 	return 0;
7363 }
7364 
7365 /**
7366  * Set the statistic memory to the new counter pool.
7367  *
7368  * @param[in] sh
7369  *   Pointer to mlx5_dev_ctx_shared object.
7370  * @param[in] pool
7371  *   Pointer to the pool to set the statistic memory.
7372  *
7373  * @return
7374  *   0 on success, a negative errno value otherwise.
7375  */
7376 static int
7377 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7378 			       struct mlx5_flow_counter_pool *pool)
7379 {
7380 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7381 	/* Resize statistic memory once used out. */
7382 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7383 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7384 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7385 		return -1;
7386 	}
7387 	rte_spinlock_lock(&pool->sl);
7388 	pool->raw = cmng->mem_mng->raws + pool->index %
7389 		    MLX5_CNT_CONTAINER_RESIZE;
7390 	rte_spinlock_unlock(&pool->sl);
7391 	pool->raw_hw = NULL;
7392 	return 0;
7393 }
7394 
7395 #define MLX5_POOL_QUERY_FREQ_US 1000000
7396 
7397 /**
7398  * Set the periodic procedure for triggering asynchronous batch queries for all
7399  * the counter pools.
7400  *
7401  * @param[in] sh
7402  *   Pointer to mlx5_dev_ctx_shared object.
7403  */
7404 void
7405 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7406 {
7407 	uint32_t pools_n, us;
7408 
7409 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7410 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7411 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7412 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7413 		sh->cmng.query_thread_on = 0;
7414 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7415 	} else {
7416 		sh->cmng.query_thread_on = 1;
7417 	}
7418 }
7419 
7420 /**
7421  * The periodic procedure for triggering asynchronous batch queries for all the
7422  * counter pools. This function is probably called by the host thread.
7423  *
7424  * @param[in] arg
7425  *   The parameter for the alarm process.
7426  */
7427 void
7428 mlx5_flow_query_alarm(void *arg)
7429 {
7430 	struct mlx5_dev_ctx_shared *sh = arg;
7431 	int ret;
7432 	uint16_t pool_index = sh->cmng.pool_index;
7433 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7434 	struct mlx5_flow_counter_pool *pool;
7435 	uint16_t n_valid;
7436 
7437 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7438 		goto set_alarm;
7439 	rte_spinlock_lock(&cmng->pool_update_sl);
7440 	pool = cmng->pools[pool_index];
7441 	n_valid = cmng->n_valid;
7442 	rte_spinlock_unlock(&cmng->pool_update_sl);
7443 	/* Set the statistic memory to the new created pool. */
7444 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7445 		goto set_alarm;
7446 	if (pool->raw_hw)
7447 		/* There is a pool query in progress. */
7448 		goto set_alarm;
7449 	pool->raw_hw =
7450 		LIST_FIRST(&sh->cmng.free_stat_raws);
7451 	if (!pool->raw_hw)
7452 		/* No free counter statistics raw memory. */
7453 		goto set_alarm;
7454 	/*
7455 	 * Identify the counters released between query trigger and query
7456 	 * handle more efficiently. The counter released in this gap period
7457 	 * should wait for a new round of query as the new arrived packets
7458 	 * will not be taken into account.
7459 	 */
7460 	pool->query_gen++;
7461 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7462 					       MLX5_COUNTERS_PER_POOL,
7463 					       NULL, NULL,
7464 					       pool->raw_hw->mem_mng->dm->id,
7465 					       (void *)(uintptr_t)
7466 					       pool->raw_hw->data,
7467 					       sh->devx_comp,
7468 					       (uint64_t)(uintptr_t)pool);
7469 	if (ret) {
7470 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7471 			" %d", pool->min_dcs->id);
7472 		pool->raw_hw = NULL;
7473 		goto set_alarm;
7474 	}
7475 	LIST_REMOVE(pool->raw_hw, next);
7476 	sh->cmng.pending_queries++;
7477 	pool_index++;
7478 	if (pool_index >= n_valid)
7479 		pool_index = 0;
7480 set_alarm:
7481 	sh->cmng.pool_index = pool_index;
7482 	mlx5_set_query_alarm(sh);
7483 }
7484 
7485 /**
7486  * Check and callback event for new aged flow in the counter pool
7487  *
7488  * @param[in] sh
7489  *   Pointer to mlx5_dev_ctx_shared object.
7490  * @param[in] pool
7491  *   Pointer to Current counter pool.
7492  */
7493 static void
7494 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7495 		   struct mlx5_flow_counter_pool *pool)
7496 {
7497 	struct mlx5_priv *priv;
7498 	struct mlx5_flow_counter *cnt;
7499 	struct mlx5_age_info *age_info;
7500 	struct mlx5_age_param *age_param;
7501 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7502 	struct mlx5_counter_stats_raw *prev = pool->raw;
7503 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7504 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7505 	uint16_t expected = AGE_CANDIDATE;
7506 	uint32_t i;
7507 
7508 	pool->time_of_last_age_check = curr_time;
7509 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7510 		cnt = MLX5_POOL_GET_CNT(pool, i);
7511 		age_param = MLX5_CNT_TO_AGE(cnt);
7512 		if (__atomic_load_n(&age_param->state,
7513 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7514 			continue;
7515 		if (cur->data[i].hits != prev->data[i].hits) {
7516 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7517 					 __ATOMIC_RELAXED);
7518 			continue;
7519 		}
7520 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7521 				       time_delta,
7522 				       __ATOMIC_RELAXED) <= age_param->timeout)
7523 			continue;
7524 		/**
7525 		 * Hold the lock first, or if between the
7526 		 * state AGE_TMOUT and tailq operation the
7527 		 * release happened, the release procedure
7528 		 * may delete a non-existent tailq node.
7529 		 */
7530 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7531 		age_info = GET_PORT_AGE_INFO(priv);
7532 		rte_spinlock_lock(&age_info->aged_sl);
7533 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7534 						AGE_TMOUT, false,
7535 						__ATOMIC_RELAXED,
7536 						__ATOMIC_RELAXED)) {
7537 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7538 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7539 		}
7540 		rte_spinlock_unlock(&age_info->aged_sl);
7541 	}
7542 	mlx5_age_event_prepare(sh);
7543 }
7544 
7545 /**
7546  * Handler for the HW respond about ready values from an asynchronous batch
7547  * query. This function is probably called by the host thread.
7548  *
7549  * @param[in] sh
7550  *   The pointer to the shared device context.
7551  * @param[in] async_id
7552  *   The Devx async ID.
7553  * @param[in] status
7554  *   The status of the completion.
7555  */
7556 void
7557 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7558 				  uint64_t async_id, int status)
7559 {
7560 	struct mlx5_flow_counter_pool *pool =
7561 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7562 	struct mlx5_counter_stats_raw *raw_to_free;
7563 	uint8_t query_gen = pool->query_gen ^ 1;
7564 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7565 	enum mlx5_counter_type cnt_type =
7566 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7567 				MLX5_COUNTER_TYPE_ORIGIN;
7568 
7569 	if (unlikely(status)) {
7570 		raw_to_free = pool->raw_hw;
7571 	} else {
7572 		raw_to_free = pool->raw;
7573 		if (pool->is_aged)
7574 			mlx5_flow_aging_check(sh, pool);
7575 		rte_spinlock_lock(&pool->sl);
7576 		pool->raw = pool->raw_hw;
7577 		rte_spinlock_unlock(&pool->sl);
7578 		/* Be sure the new raw counters data is updated in memory. */
7579 		rte_io_wmb();
7580 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7581 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7582 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7583 				     &pool->counters[query_gen], next);
7584 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7585 		}
7586 	}
7587 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7588 	pool->raw_hw = NULL;
7589 	sh->cmng.pending_queries--;
7590 }
7591 
7592 static int
7593 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7594 		    const struct flow_grp_info *grp_info,
7595 		    struct rte_flow_error *error)
7596 {
7597 	if (grp_info->transfer && grp_info->external &&
7598 	    grp_info->fdb_def_rule) {
7599 		if (group == UINT32_MAX)
7600 			return rte_flow_error_set
7601 						(error, EINVAL,
7602 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7603 						 NULL,
7604 						 "group index not supported");
7605 		*table = group + 1;
7606 	} else {
7607 		*table = group;
7608 	}
7609 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7610 	return 0;
7611 }
7612 
7613 /**
7614  * Translate the rte_flow group index to HW table value.
7615  *
7616  * If tunnel offload is disabled, all group ids converted to flow table
7617  * id using the standard method.
7618  * If tunnel offload is enabled, group id can be converted using the
7619  * standard or tunnel conversion method. Group conversion method
7620  * selection depends on flags in `grp_info` parameter:
7621  * - Internal (grp_info.external == 0) groups conversion uses the
7622  *   standard method.
7623  * - Group ids in JUMP action converted with the tunnel conversion.
7624  * - Group id in rule attribute conversion depends on a rule type and
7625  *   group id value:
7626  *   ** non zero group attributes converted with the tunnel method
7627  *   ** zero group attribute in non-tunnel rule is converted using the
7628  *      standard method - there's only one root table
7629  *   ** zero group attribute in steer tunnel rule is converted with the
7630  *      standard method - single root table
7631  *   ** zero group attribute in match tunnel rule is a special OvS
7632  *      case: that value is used for portability reasons. That group
7633  *      id is converted with the tunnel conversion method.
7634  *
7635  * @param[in] dev
7636  *   Port device
7637  * @param[in] tunnel
7638  *   PMD tunnel offload object
7639  * @param[in] group
7640  *   rte_flow group index value.
7641  * @param[out] table
7642  *   HW table value.
7643  * @param[in] grp_info
7644  *   flags used for conversion
7645  * @param[out] error
7646  *   Pointer to error structure.
7647  *
7648  * @return
7649  *   0 on success, a negative errno value otherwise and rte_errno is set.
7650  */
7651 int
7652 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7653 			 const struct mlx5_flow_tunnel *tunnel,
7654 			 uint32_t group, uint32_t *table,
7655 			 const struct flow_grp_info *grp_info,
7656 			 struct rte_flow_error *error)
7657 {
7658 	int ret;
7659 	bool standard_translation;
7660 
7661 	if (!grp_info->skip_scale && grp_info->external &&
7662 	    group < MLX5_MAX_TABLES_EXTERNAL)
7663 		group *= MLX5_FLOW_TABLE_FACTOR;
7664 	if (is_tunnel_offload_active(dev)) {
7665 		standard_translation = !grp_info->external ||
7666 					grp_info->std_tbl_fix;
7667 	} else {
7668 		standard_translation = true;
7669 	}
7670 	DRV_LOG(DEBUG,
7671 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7672 		dev->data->port_id, group, grp_info->transfer,
7673 		grp_info->external, grp_info->fdb_def_rule,
7674 		standard_translation ? "STANDARD" : "TUNNEL");
7675 	if (standard_translation)
7676 		ret = flow_group_to_table(dev->data->port_id, group, table,
7677 					  grp_info, error);
7678 	else
7679 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7680 						      table, error);
7681 
7682 	return ret;
7683 }
7684 
7685 /**
7686  * Discover availability of metadata reg_c's.
7687  *
7688  * Iteratively use test flows to check availability.
7689  *
7690  * @param[in] dev
7691  *   Pointer to the Ethernet device structure.
7692  *
7693  * @return
7694  *   0 on success, a negative errno value otherwise and rte_errno is set.
7695  */
7696 int
7697 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7698 {
7699 	struct mlx5_priv *priv = dev->data->dev_private;
7700 	struct mlx5_dev_config *config = &priv->config;
7701 	enum modify_reg idx;
7702 	int n = 0;
7703 
7704 	/* reg_c[0] and reg_c[1] are reserved. */
7705 	config->flow_mreg_c[n++] = REG_C_0;
7706 	config->flow_mreg_c[n++] = REG_C_1;
7707 	/* Discover availability of other reg_c's. */
7708 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7709 		struct rte_flow_attr attr = {
7710 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7711 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7712 			.ingress = 1,
7713 		};
7714 		struct rte_flow_item items[] = {
7715 			[0] = {
7716 				.type = RTE_FLOW_ITEM_TYPE_END,
7717 			},
7718 		};
7719 		struct rte_flow_action actions[] = {
7720 			[0] = {
7721 				.type = (enum rte_flow_action_type)
7722 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7723 				.conf = &(struct mlx5_flow_action_copy_mreg){
7724 					.src = REG_C_1,
7725 					.dst = idx,
7726 				},
7727 			},
7728 			[1] = {
7729 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7730 				.conf = &(struct rte_flow_action_jump){
7731 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7732 				},
7733 			},
7734 			[2] = {
7735 				.type = RTE_FLOW_ACTION_TYPE_END,
7736 			},
7737 		};
7738 		uint32_t flow_idx;
7739 		struct rte_flow *flow;
7740 		struct rte_flow_error error;
7741 
7742 		if (!config->dv_flow_en)
7743 			break;
7744 		/* Create internal flow, validation skips copy action. */
7745 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7746 					    actions, false, &error);
7747 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7748 				      flow_idx);
7749 		if (!flow)
7750 			continue;
7751 		config->flow_mreg_c[n++] = idx;
7752 		flow_list_destroy(dev, NULL, flow_idx);
7753 	}
7754 	for (; n < MLX5_MREG_C_NUM; ++n)
7755 		config->flow_mreg_c[n] = REG_NON;
7756 	return 0;
7757 }
7758 
7759 /**
7760  * Dump flow raw hw data to file
7761  *
7762  * @param[in] dev
7763  *    The pointer to Ethernet device.
7764  * @param[in] file
7765  *   A pointer to a file for output.
7766  * @param[out] error
7767  *   Perform verbose error reporting if not NULL. PMDs initialize this
7768  *   structure in case of error only.
7769  * @return
7770  *   0 on success, a nagative value otherwise.
7771  */
7772 int
7773 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7774 		   FILE *file,
7775 		   struct rte_flow_error *error __rte_unused)
7776 {
7777 	struct mlx5_priv *priv = dev->data->dev_private;
7778 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7779 	uint32_t handle_idx;
7780 	int ret;
7781 	struct mlx5_flow_handle *dh;
7782 	struct rte_flow *flow;
7783 
7784 	if (!priv->config.dv_flow_en) {
7785 		if (fputs("device dv flow disabled\n", file) <= 0)
7786 			return -errno;
7787 		return -ENOTSUP;
7788 	}
7789 
7790 	/* dump all */
7791 	if (!flow_idx)
7792 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7793 					sh->rx_domain,
7794 					sh->tx_domain, file);
7795 	/* dump one */
7796 	flow = mlx5_ipool_get(priv->sh->ipool
7797 			[MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
7798 	if (!flow)
7799 		return -ENOENT;
7800 
7801 	handle_idx = flow->dev_handles;
7802 	while (handle_idx) {
7803 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
7804 				handle_idx);
7805 		if (!dh)
7806 			return -ENOENT;
7807 		if (dh->drv_flow) {
7808 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
7809 					file);
7810 			if (ret)
7811 				return -ENOENT;
7812 		}
7813 		handle_idx = dh->next.next;
7814 	}
7815 	return 0;
7816 }
7817 
7818 /**
7819  * Get aged-out flows.
7820  *
7821  * @param[in] dev
7822  *   Pointer to the Ethernet device structure.
7823  * @param[in] context
7824  *   The address of an array of pointers to the aged-out flows contexts.
7825  * @param[in] nb_countexts
7826  *   The length of context array pointers.
7827  * @param[out] error
7828  *   Perform verbose error reporting if not NULL. Initialized in case of
7829  *   error only.
7830  *
7831  * @return
7832  *   how many contexts get in success, otherwise negative errno value.
7833  *   if nb_contexts is 0, return the amount of all aged contexts.
7834  *   if nb_contexts is not 0 , return the amount of aged flows reported
7835  *   in the context array.
7836  */
7837 int
7838 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7839 			uint32_t nb_contexts, struct rte_flow_error *error)
7840 {
7841 	const struct mlx5_flow_driver_ops *fops;
7842 	struct rte_flow_attr attr = { .transfer = 0 };
7843 
7844 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7845 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7846 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7847 						    error);
7848 	}
7849 	DRV_LOG(ERR,
7850 		"port %u get aged flows is not supported.",
7851 		 dev->data->port_id);
7852 	return -ENOTSUP;
7853 }
7854 
7855 /* Wrapper for driver action_validate op callback */
7856 static int
7857 flow_drv_action_validate(struct rte_eth_dev *dev,
7858 			 const struct rte_flow_indir_action_conf *conf,
7859 			 const struct rte_flow_action *action,
7860 			 const struct mlx5_flow_driver_ops *fops,
7861 			 struct rte_flow_error *error)
7862 {
7863 	static const char err_msg[] = "indirect action validation unsupported";
7864 
7865 	if (!fops->action_validate) {
7866 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7867 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7868 				   NULL, err_msg);
7869 		return -rte_errno;
7870 	}
7871 	return fops->action_validate(dev, conf, action, error);
7872 }
7873 
7874 /**
7875  * Destroys the shared action by handle.
7876  *
7877  * @param dev
7878  *   Pointer to Ethernet device structure.
7879  * @param[in] handle
7880  *   Handle for the indirect action object to be destroyed.
7881  * @param[out] error
7882  *   Perform verbose error reporting if not NULL. PMDs initialize this
7883  *   structure in case of error only.
7884  *
7885  * @return
7886  *   0 on success, a negative errno value otherwise and rte_errno is set.
7887  *
7888  * @note: wrapper for driver action_create op callback.
7889  */
7890 static int
7891 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7892 			   struct rte_flow_action_handle *handle,
7893 			   struct rte_flow_error *error)
7894 {
7895 	static const char err_msg[] = "indirect action destruction unsupported";
7896 	struct rte_flow_attr attr = { .transfer = 0 };
7897 	const struct mlx5_flow_driver_ops *fops =
7898 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7899 
7900 	if (!fops->action_destroy) {
7901 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7902 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7903 				   NULL, err_msg);
7904 		return -rte_errno;
7905 	}
7906 	return fops->action_destroy(dev, handle, error);
7907 }
7908 
7909 /* Wrapper for driver action_destroy op callback */
7910 static int
7911 flow_drv_action_update(struct rte_eth_dev *dev,
7912 		       struct rte_flow_action_handle *handle,
7913 		       const void *update,
7914 		       const struct mlx5_flow_driver_ops *fops,
7915 		       struct rte_flow_error *error)
7916 {
7917 	static const char err_msg[] = "indirect action update unsupported";
7918 
7919 	if (!fops->action_update) {
7920 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7921 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7922 				   NULL, err_msg);
7923 		return -rte_errno;
7924 	}
7925 	return fops->action_update(dev, handle, update, error);
7926 }
7927 
7928 /* Wrapper for driver action_destroy op callback */
7929 static int
7930 flow_drv_action_query(struct rte_eth_dev *dev,
7931 		      const struct rte_flow_action_handle *handle,
7932 		      void *data,
7933 		      const struct mlx5_flow_driver_ops *fops,
7934 		      struct rte_flow_error *error)
7935 {
7936 	static const char err_msg[] = "indirect action query unsupported";
7937 
7938 	if (!fops->action_query) {
7939 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7940 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7941 				   NULL, err_msg);
7942 		return -rte_errno;
7943 	}
7944 	return fops->action_query(dev, handle, data, error);
7945 }
7946 
7947 /**
7948  * Create indirect action for reuse in multiple flow rules.
7949  *
7950  * @param dev
7951  *   Pointer to Ethernet device structure.
7952  * @param conf
7953  *   Pointer to indirect action object configuration.
7954  * @param[in] action
7955  *   Action configuration for indirect action object creation.
7956  * @param[out] error
7957  *   Perform verbose error reporting if not NULL. PMDs initialize this
7958  *   structure in case of error only.
7959  * @return
7960  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7961  */
7962 static struct rte_flow_action_handle *
7963 mlx5_action_handle_create(struct rte_eth_dev *dev,
7964 			  const struct rte_flow_indir_action_conf *conf,
7965 			  const struct rte_flow_action *action,
7966 			  struct rte_flow_error *error)
7967 {
7968 	static const char err_msg[] = "indirect action creation unsupported";
7969 	struct rte_flow_attr attr = { .transfer = 0 };
7970 	const struct mlx5_flow_driver_ops *fops =
7971 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7972 
7973 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7974 		return NULL;
7975 	if (!fops->action_create) {
7976 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7977 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7978 				   NULL, err_msg);
7979 		return NULL;
7980 	}
7981 	return fops->action_create(dev, conf, action, error);
7982 }
7983 
7984 /**
7985  * Updates inplace the indirect action configuration pointed by *handle*
7986  * with the configuration provided as *update* argument.
7987  * The update of the indirect action configuration effects all flow rules
7988  * reusing the action via handle.
7989  *
7990  * @param dev
7991  *   Pointer to Ethernet device structure.
7992  * @param[in] handle
7993  *   Handle for the indirect action to be updated.
7994  * @param[in] update
7995  *   Action specification used to modify the action pointed by handle.
7996  *   *update* could be of same type with the action pointed by the *handle*
7997  *   handle argument, or some other structures like a wrapper, depending on
7998  *   the indirect action type.
7999  * @param[out] error
8000  *   Perform verbose error reporting if not NULL. PMDs initialize this
8001  *   structure in case of error only.
8002  *
8003  * @return
8004  *   0 on success, a negative errno value otherwise and rte_errno is set.
8005  */
8006 static int
8007 mlx5_action_handle_update(struct rte_eth_dev *dev,
8008 		struct rte_flow_action_handle *handle,
8009 		const void *update,
8010 		struct rte_flow_error *error)
8011 {
8012 	struct rte_flow_attr attr = { .transfer = 0 };
8013 	const struct mlx5_flow_driver_ops *fops =
8014 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8015 	int ret;
8016 
8017 	ret = flow_drv_action_validate(dev, NULL,
8018 			(const struct rte_flow_action *)update, fops, error);
8019 	if (ret)
8020 		return ret;
8021 	return flow_drv_action_update(dev, handle, update, fops,
8022 				      error);
8023 }
8024 
8025 /**
8026  * Query the indirect action by handle.
8027  *
8028  * This function allows retrieving action-specific data such as counters.
8029  * Data is gathered by special action which may be present/referenced in
8030  * more than one flow rule definition.
8031  *
8032  * see @RTE_FLOW_ACTION_TYPE_COUNT
8033  *
8034  * @param dev
8035  *   Pointer to Ethernet device structure.
8036  * @param[in] handle
8037  *   Handle for the indirect action to query.
8038  * @param[in, out] data
8039  *   Pointer to storage for the associated query data type.
8040  * @param[out] error
8041  *   Perform verbose error reporting if not NULL. PMDs initialize this
8042  *   structure in case of error only.
8043  *
8044  * @return
8045  *   0 on success, a negative errno value otherwise and rte_errno is set.
8046  */
8047 static int
8048 mlx5_action_handle_query(struct rte_eth_dev *dev,
8049 			 const struct rte_flow_action_handle *handle,
8050 			 void *data,
8051 			 struct rte_flow_error *error)
8052 {
8053 	struct rte_flow_attr attr = { .transfer = 0 };
8054 	const struct mlx5_flow_driver_ops *fops =
8055 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8056 
8057 	return flow_drv_action_query(dev, handle, data, fops, error);
8058 }
8059 
8060 /**
8061  * Destroy all indirect actions (shared RSS).
8062  *
8063  * @param dev
8064  *   Pointer to Ethernet device.
8065  *
8066  * @return
8067  *   0 on success, a negative errno value otherwise and rte_errno is set.
8068  */
8069 int
8070 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8071 {
8072 	struct rte_flow_error error;
8073 	struct mlx5_priv *priv = dev->data->dev_private;
8074 	struct mlx5_shared_action_rss *shared_rss;
8075 	int ret = 0;
8076 	uint32_t idx;
8077 
8078 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8079 		      priv->rss_shared_actions, idx, shared_rss, next) {
8080 		ret |= mlx5_action_handle_destroy(dev,
8081 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8082 	}
8083 	return ret;
8084 }
8085 
8086 #ifndef HAVE_MLX5DV_DR
8087 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8088 #else
8089 #define MLX5_DOMAIN_SYNC_FLOW \
8090 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8091 #endif
8092 
8093 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8094 {
8095 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8096 	const struct mlx5_flow_driver_ops *fops;
8097 	int ret;
8098 	struct rte_flow_attr attr = { .transfer = 0 };
8099 
8100 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8101 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8102 	if (ret > 0)
8103 		ret = -ret;
8104 	return ret;
8105 }
8106 
8107 /**
8108  * tunnel offload functionalilty is defined for DV environment only
8109  */
8110 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8111 __extension__
8112 union tunnel_offload_mark {
8113 	uint32_t val;
8114 	struct {
8115 		uint32_t app_reserve:8;
8116 		uint32_t table_id:15;
8117 		uint32_t transfer:1;
8118 		uint32_t _unused_:8;
8119 	};
8120 };
8121 
8122 static bool
8123 mlx5_access_tunnel_offload_db
8124 	(struct rte_eth_dev *dev,
8125 	 bool (*match)(struct rte_eth_dev *,
8126 		       struct mlx5_flow_tunnel *, const void *),
8127 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8128 	 void (*miss)(struct rte_eth_dev *, void *),
8129 	 void *ctx, bool lock_op);
8130 
8131 static int
8132 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
8133 			     struct rte_flow *flow,
8134 			     const struct rte_flow_attr *attr,
8135 			     const struct rte_flow_action *app_actions,
8136 			     uint32_t flow_idx,
8137 			     struct tunnel_default_miss_ctx *ctx,
8138 			     struct rte_flow_error *error)
8139 {
8140 	struct mlx5_priv *priv = dev->data->dev_private;
8141 	struct mlx5_flow *dev_flow;
8142 	struct rte_flow_attr miss_attr = *attr;
8143 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
8144 	const struct rte_flow_item miss_items[2] = {
8145 		{
8146 			.type = RTE_FLOW_ITEM_TYPE_ETH,
8147 			.spec = NULL,
8148 			.last = NULL,
8149 			.mask = NULL
8150 		},
8151 		{
8152 			.type = RTE_FLOW_ITEM_TYPE_END,
8153 			.spec = NULL,
8154 			.last = NULL,
8155 			.mask = NULL
8156 		}
8157 	};
8158 	union tunnel_offload_mark mark_id;
8159 	struct rte_flow_action_mark miss_mark;
8160 	struct rte_flow_action miss_actions[3] = {
8161 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
8162 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
8163 	};
8164 	const struct rte_flow_action_jump *jump_data;
8165 	uint32_t i, flow_table = 0; /* prevent compilation warning */
8166 	struct flow_grp_info grp_info = {
8167 		.external = 1,
8168 		.transfer = attr->transfer,
8169 		.fdb_def_rule = !!priv->fdb_def_rule,
8170 		.std_tbl_fix = 0,
8171 	};
8172 	int ret;
8173 
8174 	if (!attr->transfer) {
8175 		uint32_t q_size;
8176 
8177 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
8178 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
8179 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
8180 					 0, SOCKET_ID_ANY);
8181 		if (!ctx->queue)
8182 			return rte_flow_error_set
8183 				(error, ENOMEM,
8184 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8185 				NULL, "invalid default miss RSS");
8186 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
8187 		ctx->action_rss.level = 0,
8188 		ctx->action_rss.types = priv->rss_conf.rss_hf,
8189 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
8190 		ctx->action_rss.queue_num = priv->reta_idx_n,
8191 		ctx->action_rss.key = priv->rss_conf.rss_key,
8192 		ctx->action_rss.queue = ctx->queue;
8193 		if (!priv->reta_idx_n || !priv->rxqs_n)
8194 			return rte_flow_error_set
8195 				(error, EINVAL,
8196 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8197 				NULL, "invalid port configuration");
8198 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
8199 			ctx->action_rss.types = 0;
8200 		for (i = 0; i != priv->reta_idx_n; ++i)
8201 			ctx->queue[i] = (*priv->reta_idx)[i];
8202 	} else {
8203 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
8204 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
8205 	}
8206 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
8207 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
8208 	jump_data = app_actions->conf;
8209 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
8210 	miss_attr.group = jump_data->group;
8211 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
8212 				       &flow_table, &grp_info, error);
8213 	if (ret)
8214 		return rte_flow_error_set(error, EINVAL,
8215 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
8216 					  NULL, "invalid tunnel id");
8217 	mark_id.app_reserve = 0;
8218 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
8219 	mark_id.transfer = !!attr->transfer;
8220 	mark_id._unused_ = 0;
8221 	miss_mark.id = mark_id.val;
8222 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
8223 				    miss_items, miss_actions, flow_idx, error);
8224 	if (!dev_flow)
8225 		return -rte_errno;
8226 	dev_flow->flow = flow;
8227 	dev_flow->external = true;
8228 	dev_flow->tunnel = tunnel;
8229 	/* Subflow object was created, we must include one in the list. */
8230 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
8231 		      dev_flow->handle, next);
8232 	DRV_LOG(DEBUG,
8233 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
8234 		dev->data->port_id, tunnel->app_tunnel.type,
8235 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
8236 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
8237 				  miss_actions, error);
8238 	if (!ret)
8239 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
8240 						  error);
8241 
8242 	return ret;
8243 }
8244 
8245 static const struct mlx5_flow_tbl_data_entry  *
8246 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
8247 {
8248 	struct mlx5_priv *priv = dev->data->dev_private;
8249 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8250 	struct mlx5_hlist_entry *he;
8251 	union tunnel_offload_mark mbits = { .val = mark };
8252 	union mlx5_flow_tbl_key table_key = {
8253 		{
8254 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
8255 			.id = 0,
8256 			.reserved = 0,
8257 			.dummy = 0,
8258 			.is_fdb = !!mbits.transfer,
8259 			.is_egress = 0,
8260 		}
8261 	};
8262 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
8263 	return he ?
8264 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
8265 }
8266 
8267 static void
8268 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
8269 				   struct mlx5_hlist_entry *entry)
8270 {
8271 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8272 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8273 
8274 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8275 			tunnel_flow_tbl_to_id(tte->flow_table));
8276 	mlx5_free(tte);
8277 }
8278 
8279 static int
8280 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
8281 				  struct mlx5_hlist_entry *entry,
8282 				  uint64_t key, void *cb_ctx __rte_unused)
8283 {
8284 	union tunnel_tbl_key tbl = {
8285 		.val = key,
8286 	};
8287 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
8288 
8289 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
8290 }
8291 
8292 static struct mlx5_hlist_entry *
8293 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
8294 				   void *ctx __rte_unused)
8295 {
8296 	struct mlx5_dev_ctx_shared *sh = list->ctx;
8297 	struct tunnel_tbl_entry *tte;
8298 	union tunnel_tbl_key tbl = {
8299 		.val = key,
8300 	};
8301 
8302 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
8303 			  sizeof(*tte), 0,
8304 			  SOCKET_ID_ANY);
8305 	if (!tte)
8306 		goto err;
8307 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8308 			  &tte->flow_table);
8309 	if (tte->flow_table >= MLX5_MAX_TABLES) {
8310 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
8311 			tte->flow_table);
8312 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
8313 				tte->flow_table);
8314 		goto err;
8315 	} else if (!tte->flow_table) {
8316 		goto err;
8317 	}
8318 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
8319 	tte->tunnel_id = tbl.tunnel_id;
8320 	tte->group = tbl.group;
8321 	return &tte->hash;
8322 err:
8323 	if (tte)
8324 		mlx5_free(tte);
8325 	return NULL;
8326 }
8327 
8328 static uint32_t
8329 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
8330 				const struct mlx5_flow_tunnel *tunnel,
8331 				uint32_t group, uint32_t *table,
8332 				struct rte_flow_error *error)
8333 {
8334 	struct mlx5_hlist_entry *he;
8335 	struct tunnel_tbl_entry *tte;
8336 	union tunnel_tbl_key key = {
8337 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
8338 		.group = group
8339 	};
8340 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8341 	struct mlx5_hlist *group_hash;
8342 
8343 	group_hash = tunnel ? tunnel->groups : thub->groups;
8344 	he = mlx5_hlist_register(group_hash, key.val, NULL);
8345 	if (!he)
8346 		return rte_flow_error_set(error, EINVAL,
8347 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8348 					  NULL,
8349 					  "tunnel group index not supported");
8350 	tte = container_of(he, typeof(*tte), hash);
8351 	*table = tte->flow_table;
8352 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
8353 		dev->data->port_id, key.tunnel_id, group, *table);
8354 	return 0;
8355 }
8356 
8357 static void
8358 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8359 		      struct mlx5_flow_tunnel *tunnel)
8360 {
8361 	struct mlx5_priv *priv = dev->data->dev_private;
8362 	struct mlx5_indexed_pool *ipool;
8363 
8364 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8365 		dev->data->port_id, tunnel->tunnel_id);
8366 	LIST_REMOVE(tunnel, chain);
8367 	mlx5_hlist_destroy(tunnel->groups);
8368 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8369 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
8370 }
8371 
8372 static bool
8373 mlx5_access_tunnel_offload_db
8374 	(struct rte_eth_dev *dev,
8375 	 bool (*match)(struct rte_eth_dev *,
8376 		       struct mlx5_flow_tunnel *, const void *),
8377 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
8378 	 void (*miss)(struct rte_eth_dev *, void *),
8379 	 void *ctx, bool lock_op)
8380 {
8381 	bool verdict = false;
8382 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8383 	struct mlx5_flow_tunnel *tunnel;
8384 
8385 	rte_spinlock_lock(&thub->sl);
8386 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
8387 		verdict = match(dev, tunnel, (const void *)ctx);
8388 		if (verdict)
8389 			break;
8390 	}
8391 	if (!lock_op)
8392 		rte_spinlock_unlock(&thub->sl);
8393 	if (verdict && hit)
8394 		hit(dev, tunnel, ctx);
8395 	if (!verdict && miss)
8396 		miss(dev, ctx);
8397 	if (lock_op)
8398 		rte_spinlock_unlock(&thub->sl);
8399 
8400 	return verdict;
8401 }
8402 
8403 struct tunnel_db_find_tunnel_id_ctx {
8404 	uint32_t tunnel_id;
8405 	struct mlx5_flow_tunnel *tunnel;
8406 };
8407 
8408 static bool
8409 find_tunnel_id_match(struct rte_eth_dev *dev,
8410 		     struct mlx5_flow_tunnel *tunnel, const void *x)
8411 {
8412 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8413 
8414 	RTE_SET_USED(dev);
8415 	return tunnel->tunnel_id == ctx->tunnel_id;
8416 }
8417 
8418 static void
8419 find_tunnel_id_hit(struct rte_eth_dev *dev,
8420 		   struct mlx5_flow_tunnel *tunnel, void *x)
8421 {
8422 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
8423 	RTE_SET_USED(dev);
8424 	ctx->tunnel = tunnel;
8425 }
8426 
8427 static struct mlx5_flow_tunnel *
8428 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8429 {
8430 	struct tunnel_db_find_tunnel_id_ctx ctx = {
8431 		.tunnel_id = id,
8432 	};
8433 
8434 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
8435 				      find_tunnel_id_hit, NULL, &ctx, true);
8436 
8437 	return ctx.tunnel;
8438 }
8439 
8440 static struct mlx5_flow_tunnel *
8441 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8442 			  const struct rte_flow_tunnel *app_tunnel)
8443 {
8444 	struct mlx5_priv *priv = dev->data->dev_private;
8445 	struct mlx5_indexed_pool *ipool;
8446 	struct mlx5_flow_tunnel *tunnel;
8447 	uint32_t id;
8448 
8449 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
8450 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
8451 	if (!tunnel)
8452 		return NULL;
8453 	if (id >= MLX5_MAX_TUNNELS) {
8454 		mlx5_ipool_free(ipool, id);
8455 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
8456 		return NULL;
8457 	}
8458 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
8459 					   mlx5_flow_tunnel_grp2tbl_create_cb,
8460 					   mlx5_flow_tunnel_grp2tbl_match_cb,
8461 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
8462 	if (!tunnel->groups) {
8463 		mlx5_ipool_free(ipool, id);
8464 		return NULL;
8465 	}
8466 	tunnel->groups->ctx = priv->sh;
8467 	/* initiate new PMD tunnel */
8468 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8469 	tunnel->tunnel_id = id;
8470 	tunnel->action.type = (typeof(tunnel->action.type))
8471 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8472 	tunnel->action.conf = tunnel;
8473 	tunnel->item.type = (typeof(tunnel->item.type))
8474 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8475 	tunnel->item.spec = tunnel;
8476 	tunnel->item.last = NULL;
8477 	tunnel->item.mask = NULL;
8478 
8479 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8480 		dev->data->port_id, tunnel->tunnel_id);
8481 
8482 	return tunnel;
8483 }
8484 
8485 struct tunnel_db_get_tunnel_ctx {
8486 	const struct rte_flow_tunnel *app_tunnel;
8487 	struct mlx5_flow_tunnel *tunnel;
8488 };
8489 
8490 static bool get_tunnel_match(struct rte_eth_dev *dev,
8491 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8492 {
8493 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
8494 
8495 	RTE_SET_USED(dev);
8496 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
8497 		       sizeof(*ctx->app_tunnel));
8498 }
8499 
8500 static void get_tunnel_hit(struct rte_eth_dev *dev,
8501 			   struct mlx5_flow_tunnel *tunnel, void *x)
8502 {
8503 	/* called under tunnel spinlock protection */
8504 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8505 
8506 	RTE_SET_USED(dev);
8507 	tunnel->refctn++;
8508 	ctx->tunnel = tunnel;
8509 }
8510 
8511 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
8512 {
8513 	/* called under tunnel spinlock protection */
8514 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8515 	struct tunnel_db_get_tunnel_ctx *ctx = x;
8516 
8517 	rte_spinlock_unlock(&thub->sl);
8518 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
8519 	rte_spinlock_lock(&thub->sl);
8520 	if (ctx->tunnel) {
8521 		ctx->tunnel->refctn = 1;
8522 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
8523 	}
8524 }
8525 
8526 
8527 static int
8528 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8529 		     const struct rte_flow_tunnel *app_tunnel,
8530 		     struct mlx5_flow_tunnel **tunnel)
8531 {
8532 	struct tunnel_db_get_tunnel_ctx ctx = {
8533 		.app_tunnel = app_tunnel,
8534 	};
8535 
8536 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
8537 				      get_tunnel_miss, &ctx, true);
8538 	*tunnel = ctx.tunnel;
8539 	return ctx.tunnel ? 0 : -ENOMEM;
8540 }
8541 
8542 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8543 {
8544 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8545 
8546 	if (!thub)
8547 		return;
8548 	if (!LIST_EMPTY(&thub->tunnels))
8549 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
8550 	mlx5_hlist_destroy(thub->groups);
8551 	mlx5_free(thub);
8552 }
8553 
8554 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8555 {
8556 	int err;
8557 	struct mlx5_flow_tunnel_hub *thub;
8558 
8559 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8560 			   0, SOCKET_ID_ANY);
8561 	if (!thub)
8562 		return -ENOMEM;
8563 	LIST_INIT(&thub->tunnels);
8564 	rte_spinlock_init(&thub->sl);
8565 	thub->groups = mlx5_hlist_create("flow groups",
8566 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
8567 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
8568 					 mlx5_flow_tunnel_grp2tbl_match_cb,
8569 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
8570 	if (!thub->groups) {
8571 		err = -rte_errno;
8572 		goto err;
8573 	}
8574 	thub->groups->ctx = sh;
8575 	sh->tunnel_hub = thub;
8576 
8577 	return 0;
8578 
8579 err:
8580 	if (thub->groups)
8581 		mlx5_hlist_destroy(thub->groups);
8582 	if (thub)
8583 		mlx5_free(thub);
8584 	return err;
8585 }
8586 
8587 static inline bool
8588 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
8589 			  struct rte_flow_tunnel *tunnel,
8590 			  const char *err_msg)
8591 {
8592 	err_msg = NULL;
8593 	if (!is_tunnel_offload_active(dev)) {
8594 		err_msg = "tunnel offload was not activated";
8595 		goto out;
8596 	} else if (!tunnel) {
8597 		err_msg = "no application tunnel";
8598 		goto out;
8599 	}
8600 
8601 	switch (tunnel->type) {
8602 	default:
8603 		err_msg = "unsupported tunnel type";
8604 		goto out;
8605 	case RTE_FLOW_ITEM_TYPE_VXLAN:
8606 		break;
8607 	}
8608 
8609 out:
8610 	return !err_msg;
8611 }
8612 
8613 static int
8614 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
8615 		    struct rte_flow_tunnel *app_tunnel,
8616 		    struct rte_flow_action **actions,
8617 		    uint32_t *num_of_actions,
8618 		    struct rte_flow_error *error)
8619 {
8620 	int ret;
8621 	struct mlx5_flow_tunnel *tunnel;
8622 	const char *err_msg = NULL;
8623 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8624 
8625 	if (!verdict)
8626 		return rte_flow_error_set(error, EINVAL,
8627 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8628 					  err_msg);
8629 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8630 	if (ret < 0) {
8631 		return rte_flow_error_set(error, ret,
8632 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8633 					  "failed to initialize pmd tunnel");
8634 	}
8635 	*actions = &tunnel->action;
8636 	*num_of_actions = 1;
8637 	return 0;
8638 }
8639 
8640 static int
8641 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8642 		       struct rte_flow_tunnel *app_tunnel,
8643 		       struct rte_flow_item **items,
8644 		       uint32_t *num_of_items,
8645 		       struct rte_flow_error *error)
8646 {
8647 	int ret;
8648 	struct mlx5_flow_tunnel *tunnel;
8649 	const char *err_msg = NULL;
8650 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8651 
8652 	if (!verdict)
8653 		return rte_flow_error_set(error, EINVAL,
8654 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8655 					  err_msg);
8656 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8657 	if (ret < 0) {
8658 		return rte_flow_error_set(error, ret,
8659 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8660 					  "failed to initialize pmd tunnel");
8661 	}
8662 	*items = &tunnel->item;
8663 	*num_of_items = 1;
8664 	return 0;
8665 }
8666 
8667 struct tunnel_db_element_release_ctx {
8668 	struct rte_flow_item *items;
8669 	struct rte_flow_action *actions;
8670 	uint32_t num_elements;
8671 	struct rte_flow_error *error;
8672 	int ret;
8673 };
8674 
8675 static bool
8676 tunnel_element_release_match(struct rte_eth_dev *dev,
8677 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8678 {
8679 	const struct tunnel_db_element_release_ctx *ctx = x;
8680 
8681 	RTE_SET_USED(dev);
8682 	if (ctx->num_elements != 1)
8683 		return false;
8684 	else if (ctx->items)
8685 		return ctx->items == &tunnel->item;
8686 	else if (ctx->actions)
8687 		return ctx->actions == &tunnel->action;
8688 
8689 	return false;
8690 }
8691 
8692 static void
8693 tunnel_element_release_hit(struct rte_eth_dev *dev,
8694 			   struct mlx5_flow_tunnel *tunnel, void *x)
8695 {
8696 	struct tunnel_db_element_release_ctx *ctx = x;
8697 	ctx->ret = 0;
8698 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8699 		mlx5_flow_tunnel_free(dev, tunnel);
8700 }
8701 
8702 static void
8703 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8704 {
8705 	struct tunnel_db_element_release_ctx *ctx = x;
8706 	RTE_SET_USED(dev);
8707 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8708 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8709 				      "invalid argument");
8710 }
8711 
8712 static int
8713 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8714 		       struct rte_flow_item *pmd_items,
8715 		       uint32_t num_items, struct rte_flow_error *err)
8716 {
8717 	struct tunnel_db_element_release_ctx ctx = {
8718 		.items = pmd_items,
8719 		.actions = NULL,
8720 		.num_elements = num_items,
8721 		.error = err,
8722 	};
8723 
8724 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8725 				      tunnel_element_release_hit,
8726 				      tunnel_element_release_miss, &ctx, false);
8727 
8728 	return ctx.ret;
8729 }
8730 
8731 static int
8732 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8733 			 struct rte_flow_action *pmd_actions,
8734 			 uint32_t num_actions, struct rte_flow_error *err)
8735 {
8736 	struct tunnel_db_element_release_ctx ctx = {
8737 		.items = NULL,
8738 		.actions = pmd_actions,
8739 		.num_elements = num_actions,
8740 		.error = err,
8741 	};
8742 
8743 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8744 				      tunnel_element_release_hit,
8745 				      tunnel_element_release_miss, &ctx, false);
8746 
8747 	return ctx.ret;
8748 }
8749 
8750 static int
8751 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8752 				  struct rte_mbuf *m,
8753 				  struct rte_flow_restore_info *info,
8754 				  struct rte_flow_error *err)
8755 {
8756 	uint64_t ol_flags = m->ol_flags;
8757 	const struct mlx5_flow_tbl_data_entry *tble;
8758 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8759 
8760 	if (!is_tunnel_offload_active(dev)) {
8761 		info->flags = 0;
8762 		return 0;
8763 	}
8764 
8765 	if ((ol_flags & mask) != mask)
8766 		goto err;
8767 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8768 	if (!tble) {
8769 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8770 			dev->data->port_id, m->hash.fdir.hi);
8771 		goto err;
8772 	}
8773 	MLX5_ASSERT(tble->tunnel);
8774 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8775 	info->group_id = tble->group_id;
8776 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8777 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8778 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8779 
8780 	return 0;
8781 
8782 err:
8783 	return rte_flow_error_set(err, EINVAL,
8784 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8785 				  "failed to get restore info");
8786 }
8787 
8788 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8789 static int
8790 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8791 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8792 			   __rte_unused struct rte_flow_action **actions,
8793 			   __rte_unused uint32_t *num_of_actions,
8794 			   __rte_unused struct rte_flow_error *error)
8795 {
8796 	return -ENOTSUP;
8797 }
8798 
8799 static int
8800 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8801 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8802 		       __rte_unused struct rte_flow_item **items,
8803 		       __rte_unused uint32_t *num_of_items,
8804 		       __rte_unused struct rte_flow_error *error)
8805 {
8806 	return -ENOTSUP;
8807 }
8808 
8809 static int
8810 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8811 			      __rte_unused struct rte_flow_item *pmd_items,
8812 			      __rte_unused uint32_t num_items,
8813 			      __rte_unused struct rte_flow_error *err)
8814 {
8815 	return -ENOTSUP;
8816 }
8817 
8818 static int
8819 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8820 				__rte_unused struct rte_flow_action *pmd_action,
8821 				__rte_unused uint32_t num_actions,
8822 				__rte_unused struct rte_flow_error *err)
8823 {
8824 	return -ENOTSUP;
8825 }
8826 
8827 static int
8828 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8829 				  __rte_unused struct rte_mbuf *m,
8830 				  __rte_unused struct rte_flow_restore_info *i,
8831 				  __rte_unused struct rte_flow_error *err)
8832 {
8833 	return -ENOTSUP;
8834 }
8835 
8836 static int
8837 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8838 			     __rte_unused struct rte_flow *flow,
8839 			     __rte_unused const struct rte_flow_attr *attr,
8840 			     __rte_unused const struct rte_flow_action *actions,
8841 			     __rte_unused uint32_t flow_idx,
8842 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8843 			     __rte_unused struct rte_flow_error *error)
8844 {
8845 	return -ENOTSUP;
8846 }
8847 
8848 static struct mlx5_flow_tunnel *
8849 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8850 		    __rte_unused uint32_t id)
8851 {
8852 	return NULL;
8853 }
8854 
8855 static void
8856 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8857 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8858 {
8859 }
8860 
8861 static uint32_t
8862 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8863 				__rte_unused const struct mlx5_flow_tunnel *t,
8864 				__rte_unused uint32_t group,
8865 				__rte_unused uint32_t *table,
8866 				struct rte_flow_error *error)
8867 {
8868 	return rte_flow_error_set(error, ENOTSUP,
8869 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8870 				  "tunnel offload requires DV support");
8871 }
8872 
8873 void
8874 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8875 			__rte_unused  uint16_t port_id)
8876 {
8877 }
8878 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8879