xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision ff7ab341aff9891e40910aaa0c7a7d9c1c815475)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <rte_ethdev_driver.h>
16 #include <rte_eal_paging.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
21 #include <rte_ip.h>
22 
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
25 #include <mlx5_prm.h>
26 #include <mlx5_malloc.h>
27 
28 #include "mlx5_defs.h"
29 #include "mlx5.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 static struct mlx5_flow_tunnel *
37 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
38 static void
39 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
40 static const struct mlx5_flow_tbl_data_entry  *
41 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark);
42 static int
43 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
44 		     const struct rte_flow_tunnel *app_tunnel,
45 		     struct mlx5_flow_tunnel **tunnel);
46 
47 
48 /** Device flow drivers. */
49 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
50 
51 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
52 
53 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
54 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
55 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
56 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
57 #endif
58 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
59 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
60 };
61 
62 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
63 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
64 	(const int []){ \
65 		__VA_ARGS__, 0, \
66 	}
67 
68 /** Node object of input graph for mlx5_flow_expand_rss(). */
69 struct mlx5_flow_expand_node {
70 	const int *const next;
71 	/**<
72 	 * List of next node indexes. Index 0 is interpreted as a terminator.
73 	 */
74 	const enum rte_flow_item_type type;
75 	/**< Pattern item type of current node. */
76 	uint64_t rss_types;
77 	/**<
78 	 * RSS types bit-field associated with this node
79 	 * (see ETH_RSS_* definitions).
80 	 */
81 };
82 
83 /** Object returned by mlx5_flow_expand_rss(). */
84 struct mlx5_flow_expand_rss {
85 	uint32_t entries;
86 	/**< Number of entries @p patterns and @p priorities. */
87 	struct {
88 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
89 		uint32_t priority; /**< Priority offset for each expansion. */
90 	} entry[];
91 };
92 
93 static enum rte_flow_item_type
94 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
95 {
96 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
97 	uint16_t ether_type = 0;
98 	uint16_t ether_type_m;
99 	uint8_t ip_next_proto = 0;
100 	uint8_t ip_next_proto_m;
101 
102 	if (item == NULL || item->spec == NULL)
103 		return ret;
104 	switch (item->type) {
105 	case RTE_FLOW_ITEM_TYPE_ETH:
106 		if (item->mask)
107 			ether_type_m = ((const struct rte_flow_item_eth *)
108 						(item->mask))->type;
109 		else
110 			ether_type_m = rte_flow_item_eth_mask.type;
111 		if (ether_type_m != RTE_BE16(0xFFFF))
112 			break;
113 		ether_type = ((const struct rte_flow_item_eth *)
114 				(item->spec))->type;
115 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
116 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
117 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
118 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
119 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
120 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
121 		else
122 			ret = RTE_FLOW_ITEM_TYPE_END;
123 		break;
124 	case RTE_FLOW_ITEM_TYPE_VLAN:
125 		if (item->mask)
126 			ether_type_m = ((const struct rte_flow_item_vlan *)
127 						(item->mask))->inner_type;
128 		else
129 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
130 		if (ether_type_m != RTE_BE16(0xFFFF))
131 			break;
132 		ether_type = ((const struct rte_flow_item_vlan *)
133 				(item->spec))->inner_type;
134 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
135 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
136 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
139 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
140 		else
141 			ret = RTE_FLOW_ITEM_TYPE_END;
142 		break;
143 	case RTE_FLOW_ITEM_TYPE_IPV4:
144 		if (item->mask)
145 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
146 					(item->mask))->hdr.next_proto_id;
147 		else
148 			ip_next_proto_m =
149 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
150 		if (ip_next_proto_m != 0xFF)
151 			break;
152 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
153 				(item->spec))->hdr.next_proto_id;
154 		if (ip_next_proto == IPPROTO_UDP)
155 			ret = RTE_FLOW_ITEM_TYPE_UDP;
156 		else if (ip_next_proto == IPPROTO_TCP)
157 			ret = RTE_FLOW_ITEM_TYPE_TCP;
158 		else if (ip_next_proto == IPPROTO_IP)
159 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
160 		else if (ip_next_proto == IPPROTO_IPV6)
161 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
162 		else
163 			ret = RTE_FLOW_ITEM_TYPE_END;
164 		break;
165 	case RTE_FLOW_ITEM_TYPE_IPV6:
166 		if (item->mask)
167 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
168 						(item->mask))->hdr.proto;
169 		else
170 			ip_next_proto_m =
171 				rte_flow_item_ipv6_mask.hdr.proto;
172 		if (ip_next_proto_m != 0xFF)
173 			break;
174 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
175 				(item->spec))->hdr.proto;
176 		if (ip_next_proto == IPPROTO_UDP)
177 			ret = RTE_FLOW_ITEM_TYPE_UDP;
178 		else if (ip_next_proto == IPPROTO_TCP)
179 			ret = RTE_FLOW_ITEM_TYPE_TCP;
180 		else if (ip_next_proto == IPPROTO_IP)
181 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
182 		else if (ip_next_proto == IPPROTO_IPV6)
183 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
184 		else
185 			ret = RTE_FLOW_ITEM_TYPE_END;
186 		break;
187 	default:
188 		ret = RTE_FLOW_ITEM_TYPE_VOID;
189 		break;
190 	}
191 	return ret;
192 }
193 
194 /**
195  * Expand RSS flows into several possible flows according to the RSS hash
196  * fields requested and the driver capabilities.
197  *
198  * @param[out] buf
199  *   Buffer to store the result expansion.
200  * @param[in] size
201  *   Buffer size in bytes. If 0, @p buf can be NULL.
202  * @param[in] pattern
203  *   User flow pattern.
204  * @param[in] types
205  *   RSS types to expand (see ETH_RSS_* definitions).
206  * @param[in] graph
207  *   Input graph to expand @p pattern according to @p types.
208  * @param[in] graph_root_index
209  *   Index of root node in @p graph, typically 0.
210  *
211  * @return
212  *   A positive value representing the size of @p buf in bytes regardless of
213  *   @p size on success, a negative errno value otherwise and rte_errno is
214  *   set, the following errors are defined:
215  *
216  *   -E2BIG: graph-depth @p graph is too deep.
217  */
218 static int
219 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
220 		     const struct rte_flow_item *pattern, uint64_t types,
221 		     const struct mlx5_flow_expand_node graph[],
222 		     int graph_root_index)
223 {
224 	const int elt_n = 8;
225 	const struct rte_flow_item *item;
226 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
227 	const int *next_node;
228 	const int *stack[elt_n];
229 	int stack_pos = 0;
230 	struct rte_flow_item flow_items[elt_n];
231 	unsigned int i;
232 	size_t lsize;
233 	size_t user_pattern_size = 0;
234 	void *addr = NULL;
235 	const struct mlx5_flow_expand_node *next = NULL;
236 	struct rte_flow_item missed_item;
237 	int missed = 0;
238 	int elt = 0;
239 	const struct rte_flow_item *last_item = NULL;
240 
241 	memset(&missed_item, 0, sizeof(missed_item));
242 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
243 		elt_n * sizeof(buf->entry[0]);
244 	if (lsize <= size) {
245 		buf->entry[0].priority = 0;
246 		buf->entry[0].pattern = (void *)&buf->entry[elt_n];
247 		buf->entries = 0;
248 		addr = buf->entry[0].pattern;
249 	}
250 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
251 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
252 			last_item = item;
253 		for (i = 0; node->next && node->next[i]; ++i) {
254 			next = &graph[node->next[i]];
255 			if (next->type == item->type)
256 				break;
257 		}
258 		if (next)
259 			node = next;
260 		user_pattern_size += sizeof(*item);
261 	}
262 	user_pattern_size += sizeof(*item); /* Handle END item. */
263 	lsize += user_pattern_size;
264 	/* Copy the user pattern in the first entry of the buffer. */
265 	if (lsize <= size) {
266 		rte_memcpy(addr, pattern, user_pattern_size);
267 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
268 		buf->entries = 1;
269 	}
270 	/* Start expanding. */
271 	memset(flow_items, 0, sizeof(flow_items));
272 	user_pattern_size -= sizeof(*item);
273 	/*
274 	 * Check if the last valid item has spec set, need complete pattern,
275 	 * and the pattern can be used for expansion.
276 	 */
277 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
278 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
279 		/* Item type END indicates expansion is not required. */
280 		return lsize;
281 	}
282 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
283 		next = NULL;
284 		missed = 1;
285 		for (i = 0; node->next && node->next[i]; ++i) {
286 			next = &graph[node->next[i]];
287 			if (next->type == missed_item.type) {
288 				flow_items[0].type = missed_item.type;
289 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
290 				break;
291 			}
292 			next = NULL;
293 		}
294 	}
295 	if (next && missed) {
296 		elt = 2; /* missed item + item end. */
297 		node = next;
298 		lsize += elt * sizeof(*item) + user_pattern_size;
299 		if ((node->rss_types & types) && lsize <= size) {
300 			buf->entry[buf->entries].priority = 1;
301 			buf->entry[buf->entries].pattern = addr;
302 			buf->entries++;
303 			rte_memcpy(addr, buf->entry[0].pattern,
304 				   user_pattern_size);
305 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
306 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
307 			addr = (void *)(((uintptr_t)addr) +
308 					elt * sizeof(*item));
309 		}
310 	}
311 	memset(flow_items, 0, sizeof(flow_items));
312 	next_node = node->next;
313 	stack[stack_pos] = next_node;
314 	node = next_node ? &graph[*next_node] : NULL;
315 	while (node) {
316 		flow_items[stack_pos].type = node->type;
317 		if (node->rss_types & types) {
318 			/*
319 			 * compute the number of items to copy from the
320 			 * expansion and copy it.
321 			 * When the stack_pos is 0, there are 1 element in it,
322 			 * plus the addition END item.
323 			 */
324 			elt = stack_pos + 2;
325 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
326 			lsize += elt * sizeof(*item) + user_pattern_size;
327 			if (lsize <= size) {
328 				size_t n = elt * sizeof(*item);
329 
330 				buf->entry[buf->entries].priority =
331 					stack_pos + 1 + missed;
332 				buf->entry[buf->entries].pattern = addr;
333 				buf->entries++;
334 				rte_memcpy(addr, buf->entry[0].pattern,
335 					   user_pattern_size);
336 				addr = (void *)(((uintptr_t)addr) +
337 						user_pattern_size);
338 				rte_memcpy(addr, &missed_item,
339 					   missed * sizeof(*item));
340 				addr = (void *)(((uintptr_t)addr) +
341 					missed * sizeof(*item));
342 				rte_memcpy(addr, flow_items, n);
343 				addr = (void *)(((uintptr_t)addr) + n);
344 			}
345 		}
346 		/* Go deeper. */
347 		if (node->next) {
348 			next_node = node->next;
349 			if (stack_pos++ == elt_n) {
350 				rte_errno = E2BIG;
351 				return -rte_errno;
352 			}
353 			stack[stack_pos] = next_node;
354 		} else if (*(next_node + 1)) {
355 			/* Follow up with the next possibility. */
356 			++next_node;
357 		} else {
358 			/* Move to the next path. */
359 			if (stack_pos)
360 				next_node = stack[--stack_pos];
361 			next_node++;
362 			stack[stack_pos] = next_node;
363 		}
364 		node = *next_node ? &graph[*next_node] : NULL;
365 	};
366 	/* no expanded flows but we have missed item, create one rule for it */
367 	if (buf->entries == 1 && missed != 0) {
368 		elt = 2;
369 		lsize += elt * sizeof(*item) + user_pattern_size;
370 		if (lsize <= size) {
371 			buf->entry[buf->entries].priority = 1;
372 			buf->entry[buf->entries].pattern = addr;
373 			buf->entries++;
374 			flow_items[0].type = missed_item.type;
375 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
376 			rte_memcpy(addr, buf->entry[0].pattern,
377 				   user_pattern_size);
378 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
379 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
380 			addr = (void *)(((uintptr_t)addr) +
381 					elt * sizeof(*item));
382 		}
383 	}
384 	return lsize;
385 }
386 
387 enum mlx5_expansion {
388 	MLX5_EXPANSION_ROOT,
389 	MLX5_EXPANSION_ROOT_OUTER,
390 	MLX5_EXPANSION_ROOT_ETH_VLAN,
391 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
392 	MLX5_EXPANSION_OUTER_ETH,
393 	MLX5_EXPANSION_OUTER_ETH_VLAN,
394 	MLX5_EXPANSION_OUTER_VLAN,
395 	MLX5_EXPANSION_OUTER_IPV4,
396 	MLX5_EXPANSION_OUTER_IPV4_UDP,
397 	MLX5_EXPANSION_OUTER_IPV4_TCP,
398 	MLX5_EXPANSION_OUTER_IPV6,
399 	MLX5_EXPANSION_OUTER_IPV6_UDP,
400 	MLX5_EXPANSION_OUTER_IPV6_TCP,
401 	MLX5_EXPANSION_VXLAN,
402 	MLX5_EXPANSION_VXLAN_GPE,
403 	MLX5_EXPANSION_GRE,
404 	MLX5_EXPANSION_MPLS,
405 	MLX5_EXPANSION_ETH,
406 	MLX5_EXPANSION_ETH_VLAN,
407 	MLX5_EXPANSION_VLAN,
408 	MLX5_EXPANSION_IPV4,
409 	MLX5_EXPANSION_IPV4_UDP,
410 	MLX5_EXPANSION_IPV4_TCP,
411 	MLX5_EXPANSION_IPV6,
412 	MLX5_EXPANSION_IPV6_UDP,
413 	MLX5_EXPANSION_IPV6_TCP,
414 };
415 
416 /** Supported expansion of items. */
417 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
418 	[MLX5_EXPANSION_ROOT] = {
419 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
420 						  MLX5_EXPANSION_IPV4,
421 						  MLX5_EXPANSION_IPV6),
422 		.type = RTE_FLOW_ITEM_TYPE_END,
423 	},
424 	[MLX5_EXPANSION_ROOT_OUTER] = {
425 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
426 						  MLX5_EXPANSION_OUTER_IPV4,
427 						  MLX5_EXPANSION_OUTER_IPV6),
428 		.type = RTE_FLOW_ITEM_TYPE_END,
429 	},
430 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
431 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
432 		.type = RTE_FLOW_ITEM_TYPE_END,
433 	},
434 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
435 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
436 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
437 		.type = RTE_FLOW_ITEM_TYPE_END,
438 	},
439 	[MLX5_EXPANSION_OUTER_ETH] = {
440 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
441 						  MLX5_EXPANSION_OUTER_IPV6,
442 						  MLX5_EXPANSION_MPLS),
443 		.type = RTE_FLOW_ITEM_TYPE_ETH,
444 		.rss_types = 0,
445 	},
446 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
447 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
448 		.type = RTE_FLOW_ITEM_TYPE_ETH,
449 		.rss_types = 0,
450 	},
451 	[MLX5_EXPANSION_OUTER_VLAN] = {
452 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
453 						  MLX5_EXPANSION_OUTER_IPV6),
454 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
455 	},
456 	[MLX5_EXPANSION_OUTER_IPV4] = {
457 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
458 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
459 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
460 			 MLX5_EXPANSION_GRE,
461 			 MLX5_EXPANSION_IPV4,
462 			 MLX5_EXPANSION_IPV6),
463 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
464 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
465 			ETH_RSS_NONFRAG_IPV4_OTHER,
466 	},
467 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
468 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
469 						  MLX5_EXPANSION_VXLAN_GPE),
470 		.type = RTE_FLOW_ITEM_TYPE_UDP,
471 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
472 	},
473 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
474 		.type = RTE_FLOW_ITEM_TYPE_TCP,
475 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV6] = {
478 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
479 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
480 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
481 			 MLX5_EXPANSION_IPV4,
482 			 MLX5_EXPANSION_IPV6),
483 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
484 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
485 			ETH_RSS_NONFRAG_IPV6_OTHER,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 						  MLX5_EXPANSION_VXLAN_GPE),
490 		.type = RTE_FLOW_ITEM_TYPE_UDP,
491 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
492 	},
493 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
494 		.type = RTE_FLOW_ITEM_TYPE_TCP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
496 	},
497 	[MLX5_EXPANSION_VXLAN] = {
498 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
499 						  MLX5_EXPANSION_IPV4,
500 						  MLX5_EXPANSION_IPV6),
501 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
502 	},
503 	[MLX5_EXPANSION_VXLAN_GPE] = {
504 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
505 						  MLX5_EXPANSION_IPV4,
506 						  MLX5_EXPANSION_IPV6),
507 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
508 	},
509 	[MLX5_EXPANSION_GRE] = {
510 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
511 		.type = RTE_FLOW_ITEM_TYPE_GRE,
512 	},
513 	[MLX5_EXPANSION_MPLS] = {
514 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
515 						  MLX5_EXPANSION_IPV6),
516 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
517 	},
518 	[MLX5_EXPANSION_ETH] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_ETH,
522 	},
523 	[MLX5_EXPANSION_ETH_VLAN] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
525 		.type = RTE_FLOW_ITEM_TYPE_ETH,
526 	},
527 	[MLX5_EXPANSION_VLAN] = {
528 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 						  MLX5_EXPANSION_IPV6),
530 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
531 	},
532 	[MLX5_EXPANSION_IPV4] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
534 						  MLX5_EXPANSION_IPV4_TCP),
535 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
536 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
537 			ETH_RSS_NONFRAG_IPV4_OTHER,
538 	},
539 	[MLX5_EXPANSION_IPV4_UDP] = {
540 		.type = RTE_FLOW_ITEM_TYPE_UDP,
541 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
542 	},
543 	[MLX5_EXPANSION_IPV4_TCP] = {
544 		.type = RTE_FLOW_ITEM_TYPE_TCP,
545 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
546 	},
547 	[MLX5_EXPANSION_IPV6] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
549 						  MLX5_EXPANSION_IPV6_TCP),
550 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
551 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
552 			ETH_RSS_NONFRAG_IPV6_OTHER,
553 	},
554 	[MLX5_EXPANSION_IPV6_UDP] = {
555 		.type = RTE_FLOW_ITEM_TYPE_UDP,
556 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
557 	},
558 	[MLX5_EXPANSION_IPV6_TCP] = {
559 		.type = RTE_FLOW_ITEM_TYPE_TCP,
560 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
561 	},
562 };
563 
564 static struct rte_flow_shared_action *
565 mlx5_shared_action_create(struct rte_eth_dev *dev,
566 			  const struct rte_flow_shared_action_conf *conf,
567 			  const struct rte_flow_action *action,
568 			  struct rte_flow_error *error);
569 static int mlx5_shared_action_destroy
570 				(struct rte_eth_dev *dev,
571 				 struct rte_flow_shared_action *shared_action,
572 				 struct rte_flow_error *error);
573 static int mlx5_shared_action_update
574 				(struct rte_eth_dev *dev,
575 				 struct rte_flow_shared_action *shared_action,
576 				 const struct rte_flow_action *action,
577 				 struct rte_flow_error *error);
578 static int mlx5_shared_action_query
579 				(struct rte_eth_dev *dev,
580 				 const struct rte_flow_shared_action *action,
581 				 void *data,
582 				 struct rte_flow_error *error);
583 static inline bool
584 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
585 			  struct rte_flow_tunnel *tunnel,
586 			  const char *err_msg)
587 {
588 	err_msg = NULL;
589 	if (!is_tunnel_offload_active(dev)) {
590 		err_msg = "tunnel offload was not activated";
591 		goto out;
592 	} else if (!tunnel) {
593 		err_msg = "no application tunnel";
594 		goto out;
595 	}
596 
597 	switch (tunnel->type) {
598 	default:
599 		err_msg = "unsupported tunnel type";
600 		goto out;
601 	case RTE_FLOW_ITEM_TYPE_VXLAN:
602 		break;
603 	}
604 
605 out:
606 	return !err_msg;
607 }
608 
609 
610 static int
611 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
612 		    struct rte_flow_tunnel *app_tunnel,
613 		    struct rte_flow_action **actions,
614 		    uint32_t *num_of_actions,
615 		    struct rte_flow_error *error)
616 {
617 	int ret;
618 	struct mlx5_flow_tunnel *tunnel;
619 	const char *err_msg = NULL;
620 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
621 
622 	if (!verdict)
623 		return rte_flow_error_set(error, EINVAL,
624 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
625 					  err_msg);
626 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
627 	if (ret < 0) {
628 		return rte_flow_error_set(error, ret,
629 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
630 					  "failed to initialize pmd tunnel");
631 	}
632 	*actions = &tunnel->action;
633 	*num_of_actions = 1;
634 	return 0;
635 }
636 
637 static int
638 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
639 		       struct rte_flow_tunnel *app_tunnel,
640 		       struct rte_flow_item **items,
641 		       uint32_t *num_of_items,
642 		       struct rte_flow_error *error)
643 {
644 	int ret;
645 	struct mlx5_flow_tunnel *tunnel;
646 	const char *err_msg = NULL;
647 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
648 
649 	if (!verdict)
650 		return rte_flow_error_set(error, EINVAL,
651 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
652 					  err_msg);
653 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
654 	if (ret < 0) {
655 		return rte_flow_error_set(error, ret,
656 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
657 					  "failed to initialize pmd tunnel");
658 	}
659 	*items = &tunnel->item;
660 	*num_of_items = 1;
661 	return 0;
662 }
663 
664 static int
665 mlx5_flow_item_release(struct rte_eth_dev *dev,
666 		       struct rte_flow_item *pmd_items,
667 		       uint32_t num_items, struct rte_flow_error *err)
668 {
669 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
670 	struct mlx5_flow_tunnel *tun;
671 
672 	LIST_FOREACH(tun, &thub->tunnels, chain) {
673 		if (&tun->item == pmd_items)
674 			break;
675 	}
676 	if (!tun || num_items != 1)
677 		return rte_flow_error_set(err, EINVAL,
678 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
679 					  "invalid argument");
680 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
681 		mlx5_flow_tunnel_free(dev, tun);
682 	return 0;
683 }
684 
685 static int
686 mlx5_flow_action_release(struct rte_eth_dev *dev,
687 			 struct rte_flow_action *pmd_actions,
688 			 uint32_t num_actions, struct rte_flow_error *err)
689 {
690 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
691 	struct mlx5_flow_tunnel *tun;
692 
693 	LIST_FOREACH(tun, &thub->tunnels, chain) {
694 		if (&tun->action == pmd_actions)
695 			break;
696 	}
697 	if (!tun || num_actions != 1)
698 		return rte_flow_error_set(err, EINVAL,
699 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
700 					  "invalid argument");
701 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
702 		mlx5_flow_tunnel_free(dev, tun);
703 
704 	return 0;
705 }
706 
707 static int
708 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
709 				  struct rte_mbuf *m,
710 				  struct rte_flow_restore_info *info,
711 				  struct rte_flow_error *err)
712 {
713 	uint64_t ol_flags = m->ol_flags;
714 	const struct mlx5_flow_tbl_data_entry *tble;
715 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
716 
717 	if ((ol_flags & mask) != mask)
718 		goto err;
719 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
720 	if (!tble) {
721 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
722 			dev->data->port_id, m->hash.fdir.hi);
723 		goto err;
724 	}
725 	MLX5_ASSERT(tble->tunnel);
726 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
727 	info->group_id = tble->group_id;
728 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
729 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
730 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
731 
732 	return 0;
733 
734 err:
735 	return rte_flow_error_set(err, EINVAL,
736 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
737 				  "failed to get restore info");
738 }
739 
740 static const struct rte_flow_ops mlx5_flow_ops = {
741 	.validate = mlx5_flow_validate,
742 	.create = mlx5_flow_create,
743 	.destroy = mlx5_flow_destroy,
744 	.flush = mlx5_flow_flush,
745 	.isolate = mlx5_flow_isolate,
746 	.query = mlx5_flow_query,
747 	.dev_dump = mlx5_flow_dev_dump,
748 	.get_aged_flows = mlx5_flow_get_aged_flows,
749 	.shared_action_create = mlx5_shared_action_create,
750 	.shared_action_destroy = mlx5_shared_action_destroy,
751 	.shared_action_update = mlx5_shared_action_update,
752 	.shared_action_query = mlx5_shared_action_query,
753 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
754 	.tunnel_match = mlx5_flow_tunnel_match,
755 	.tunnel_action_decap_release = mlx5_flow_action_release,
756 	.tunnel_item_release = mlx5_flow_item_release,
757 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
758 };
759 
760 /* Convert FDIR request to Generic flow. */
761 struct mlx5_fdir {
762 	struct rte_flow_attr attr;
763 	struct rte_flow_item items[4];
764 	struct rte_flow_item_eth l2;
765 	struct rte_flow_item_eth l2_mask;
766 	union {
767 		struct rte_flow_item_ipv4 ipv4;
768 		struct rte_flow_item_ipv6 ipv6;
769 	} l3;
770 	union {
771 		struct rte_flow_item_ipv4 ipv4;
772 		struct rte_flow_item_ipv6 ipv6;
773 	} l3_mask;
774 	union {
775 		struct rte_flow_item_udp udp;
776 		struct rte_flow_item_tcp tcp;
777 	} l4;
778 	union {
779 		struct rte_flow_item_udp udp;
780 		struct rte_flow_item_tcp tcp;
781 	} l4_mask;
782 	struct rte_flow_action actions[2];
783 	struct rte_flow_action_queue queue;
784 };
785 
786 /* Tunnel information. */
787 struct mlx5_flow_tunnel_info {
788 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
789 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
790 };
791 
792 static struct mlx5_flow_tunnel_info tunnels_info[] = {
793 	{
794 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
795 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
796 	},
797 	{
798 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
799 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
800 	},
801 	{
802 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
803 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
804 	},
805 	{
806 		.tunnel = MLX5_FLOW_LAYER_GRE,
807 		.ptype = RTE_PTYPE_TUNNEL_GRE,
808 	},
809 	{
810 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
811 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
812 	},
813 	{
814 		.tunnel = MLX5_FLOW_LAYER_MPLS,
815 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
816 	},
817 	{
818 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
819 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
820 	},
821 	{
822 		.tunnel = MLX5_FLOW_LAYER_IPIP,
823 		.ptype = RTE_PTYPE_TUNNEL_IP,
824 	},
825 	{
826 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
827 		.ptype = RTE_PTYPE_TUNNEL_IP,
828 	},
829 	{
830 		.tunnel = MLX5_FLOW_LAYER_GTP,
831 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
832 	},
833 };
834 
835 /* Key of thread specific flow workspace data. */
836 static pthread_key_t key_workspace;
837 
838 /* Thread specific flow workspace data once initialization data. */
839 static pthread_once_t key_workspace_init;
840 
841 
842 /**
843  * Translate tag ID to register.
844  *
845  * @param[in] dev
846  *   Pointer to the Ethernet device structure.
847  * @param[in] feature
848  *   The feature that request the register.
849  * @param[in] id
850  *   The request register ID.
851  * @param[out] error
852  *   Error description in case of any.
853  *
854  * @return
855  *   The request register on success, a negative errno
856  *   value otherwise and rte_errno is set.
857  */
858 int
859 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
860 		     enum mlx5_feature_name feature,
861 		     uint32_t id,
862 		     struct rte_flow_error *error)
863 {
864 	struct mlx5_priv *priv = dev->data->dev_private;
865 	struct mlx5_dev_config *config = &priv->config;
866 	enum modify_reg start_reg;
867 	bool skip_mtr_reg = false;
868 
869 	switch (feature) {
870 	case MLX5_HAIRPIN_RX:
871 		return REG_B;
872 	case MLX5_HAIRPIN_TX:
873 		return REG_A;
874 	case MLX5_METADATA_RX:
875 		switch (config->dv_xmeta_en) {
876 		case MLX5_XMETA_MODE_LEGACY:
877 			return REG_B;
878 		case MLX5_XMETA_MODE_META16:
879 			return REG_C_0;
880 		case MLX5_XMETA_MODE_META32:
881 			return REG_C_1;
882 		}
883 		break;
884 	case MLX5_METADATA_TX:
885 		return REG_A;
886 	case MLX5_METADATA_FDB:
887 		switch (config->dv_xmeta_en) {
888 		case MLX5_XMETA_MODE_LEGACY:
889 			return REG_NON;
890 		case MLX5_XMETA_MODE_META16:
891 			return REG_C_0;
892 		case MLX5_XMETA_MODE_META32:
893 			return REG_C_1;
894 		}
895 		break;
896 	case MLX5_FLOW_MARK:
897 		switch (config->dv_xmeta_en) {
898 		case MLX5_XMETA_MODE_LEGACY:
899 			return REG_NON;
900 		case MLX5_XMETA_MODE_META16:
901 			return REG_C_1;
902 		case MLX5_XMETA_MODE_META32:
903 			return REG_C_0;
904 		}
905 		break;
906 	case MLX5_MTR_SFX:
907 		/*
908 		 * If meter color and flow match share one register, flow match
909 		 * should use the meter color register for match.
910 		 */
911 		if (priv->mtr_reg_share)
912 			return priv->mtr_color_reg;
913 		else
914 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
915 			       REG_C_3;
916 	case MLX5_MTR_COLOR:
917 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
918 		return priv->mtr_color_reg;
919 	case MLX5_COPY_MARK:
920 		/*
921 		 * Metadata COPY_MARK register using is in meter suffix sub
922 		 * flow while with meter. It's safe to share the same register.
923 		 */
924 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
925 	case MLX5_APP_TAG:
926 		/*
927 		 * If meter is enable, it will engage the register for color
928 		 * match and flow match. If meter color match is not using the
929 		 * REG_C_2, need to skip the REG_C_x be used by meter color
930 		 * match.
931 		 * If meter is disable, free to use all available registers.
932 		 */
933 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
934 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
935 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
936 		if (id > (REG_C_7 - start_reg))
937 			return rte_flow_error_set(error, EINVAL,
938 						  RTE_FLOW_ERROR_TYPE_ITEM,
939 						  NULL, "invalid tag id");
940 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
941 			return rte_flow_error_set(error, ENOTSUP,
942 						  RTE_FLOW_ERROR_TYPE_ITEM,
943 						  NULL, "unsupported tag id");
944 		/*
945 		 * This case means meter is using the REG_C_x great than 2.
946 		 * Take care not to conflict with meter color REG_C_x.
947 		 * If the available index REG_C_y >= REG_C_x, skip the
948 		 * color register.
949 		 */
950 		if (skip_mtr_reg && config->flow_mreg_c
951 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
952 			if (id >= (REG_C_7 - start_reg))
953 				return rte_flow_error_set(error, EINVAL,
954 						       RTE_FLOW_ERROR_TYPE_ITEM,
955 							NULL, "invalid tag id");
956 			if (config->flow_mreg_c
957 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
958 				return config->flow_mreg_c
959 					       [id + 1 + start_reg - REG_C_0];
960 			return rte_flow_error_set(error, ENOTSUP,
961 						  RTE_FLOW_ERROR_TYPE_ITEM,
962 						  NULL, "unsupported tag id");
963 		}
964 		return config->flow_mreg_c[id + start_reg - REG_C_0];
965 	}
966 	MLX5_ASSERT(false);
967 	return rte_flow_error_set(error, EINVAL,
968 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
969 				  NULL, "invalid feature name");
970 }
971 
972 /**
973  * Check extensive flow metadata register support.
974  *
975  * @param dev
976  *   Pointer to rte_eth_dev structure.
977  *
978  * @return
979  *   True if device supports extensive flow metadata register, otherwise false.
980  */
981 bool
982 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
983 {
984 	struct mlx5_priv *priv = dev->data->dev_private;
985 	struct mlx5_dev_config *config = &priv->config;
986 
987 	/*
988 	 * Having available reg_c can be regarded inclusively as supporting
989 	 * extensive flow metadata register, which could mean,
990 	 * - metadata register copy action by modify header.
991 	 * - 16 modify header actions is supported.
992 	 * - reg_c's are preserved across different domain (FDB and NIC) on
993 	 *   packet loopback by flow lookup miss.
994 	 */
995 	return config->flow_mreg_c[2] != REG_NON;
996 }
997 
998 /**
999  * Verify the @p item specifications (spec, last, mask) are compatible with the
1000  * NIC capabilities.
1001  *
1002  * @param[in] item
1003  *   Item specification.
1004  * @param[in] mask
1005  *   @p item->mask or flow default bit-masks.
1006  * @param[in] nic_mask
1007  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1008  * @param[in] size
1009  *   Bit-masks size in bytes.
1010  * @param[in] range_accepted
1011  *   True if range of values is accepted for specific fields, false otherwise.
1012  * @param[out] error
1013  *   Pointer to error structure.
1014  *
1015  * @return
1016  *   0 on success, a negative errno value otherwise and rte_errno is set.
1017  */
1018 int
1019 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1020 			  const uint8_t *mask,
1021 			  const uint8_t *nic_mask,
1022 			  unsigned int size,
1023 			  bool range_accepted,
1024 			  struct rte_flow_error *error)
1025 {
1026 	unsigned int i;
1027 
1028 	MLX5_ASSERT(nic_mask);
1029 	for (i = 0; i < size; ++i)
1030 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1031 			return rte_flow_error_set(error, ENOTSUP,
1032 						  RTE_FLOW_ERROR_TYPE_ITEM,
1033 						  item,
1034 						  "mask enables non supported"
1035 						  " bits");
1036 	if (!item->spec && (item->mask || item->last))
1037 		return rte_flow_error_set(error, EINVAL,
1038 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1039 					  "mask/last without a spec is not"
1040 					  " supported");
1041 	if (item->spec && item->last && !range_accepted) {
1042 		uint8_t spec[size];
1043 		uint8_t last[size];
1044 		unsigned int i;
1045 		int ret;
1046 
1047 		for (i = 0; i < size; ++i) {
1048 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1049 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1050 		}
1051 		ret = memcmp(spec, last, size);
1052 		if (ret != 0)
1053 			return rte_flow_error_set(error, EINVAL,
1054 						  RTE_FLOW_ERROR_TYPE_ITEM,
1055 						  item,
1056 						  "range is not valid");
1057 	}
1058 	return 0;
1059 }
1060 
1061 /**
1062  * Adjust the hash fields according to the @p flow information.
1063  *
1064  * @param[in] dev_flow.
1065  *   Pointer to the mlx5_flow.
1066  * @param[in] tunnel
1067  *   1 when the hash field is for a tunnel item.
1068  * @param[in] layer_types
1069  *   ETH_RSS_* types.
1070  * @param[in] hash_fields
1071  *   Item hash fields.
1072  *
1073  * @return
1074  *   The hash fields that should be used.
1075  */
1076 uint64_t
1077 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1078 			    int tunnel __rte_unused, uint64_t layer_types,
1079 			    uint64_t hash_fields)
1080 {
1081 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1082 	int rss_request_inner = rss_desc->level >= 2;
1083 
1084 	/* Check RSS hash level for tunnel. */
1085 	if (tunnel && rss_request_inner)
1086 		hash_fields |= IBV_RX_HASH_INNER;
1087 	else if (tunnel || rss_request_inner)
1088 		return 0;
1089 #endif
1090 	/* Check if requested layer matches RSS hash fields. */
1091 	if (!(rss_desc->types & layer_types))
1092 		return 0;
1093 	return hash_fields;
1094 }
1095 
1096 /**
1097  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1098  * if several tunnel rules are used on this queue, the tunnel ptype will be
1099  * cleared.
1100  *
1101  * @param rxq_ctrl
1102  *   Rx queue to update.
1103  */
1104 static void
1105 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1106 {
1107 	unsigned int i;
1108 	uint32_t tunnel_ptype = 0;
1109 
1110 	/* Look up for the ptype to use. */
1111 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1112 		if (!rxq_ctrl->flow_tunnels_n[i])
1113 			continue;
1114 		if (!tunnel_ptype) {
1115 			tunnel_ptype = tunnels_info[i].ptype;
1116 		} else {
1117 			tunnel_ptype = 0;
1118 			break;
1119 		}
1120 	}
1121 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1122 }
1123 
1124 /**
1125  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1126  * flow.
1127  *
1128  * @param[in] dev
1129  *   Pointer to the Ethernet device structure.
1130  * @param[in] dev_handle
1131  *   Pointer to device flow handle structure.
1132  */
1133 static void
1134 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1135 		       struct mlx5_flow_handle *dev_handle)
1136 {
1137 	struct mlx5_priv *priv = dev->data->dev_private;
1138 	const int mark = dev_handle->mark;
1139 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1140 	struct mlx5_hrxq *hrxq;
1141 	unsigned int i;
1142 
1143 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1144 		return;
1145 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1146 			      dev_handle->rix_hrxq);
1147 	if (!hrxq)
1148 		return;
1149 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1150 		int idx = hrxq->ind_table->queues[i];
1151 		struct mlx5_rxq_ctrl *rxq_ctrl =
1152 			container_of((*priv->rxqs)[idx],
1153 				     struct mlx5_rxq_ctrl, rxq);
1154 
1155 		/*
1156 		 * To support metadata register copy on Tx loopback,
1157 		 * this must be always enabled (metadata may arive
1158 		 * from other port - not from local flows only.
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->rxq.mark = 1;
1167 			rxq_ctrl->flow_mark_n++;
1168 		}
1169 		if (tunnel) {
1170 			unsigned int j;
1171 
1172 			/* Increase 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  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1188  *
1189  * @param[in] dev
1190  *   Pointer to the Ethernet device structure.
1191  * @param[in] flow
1192  *   Pointer to flow structure.
1193  */
1194 static void
1195 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1196 {
1197 	struct mlx5_priv *priv = dev->data->dev_private;
1198 	uint32_t handle_idx;
1199 	struct mlx5_flow_handle *dev_handle;
1200 
1201 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1202 		       handle_idx, dev_handle, next)
1203 		flow_drv_rxq_flags_set(dev, dev_handle);
1204 }
1205 
1206 /**
1207  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1208  * device flow if no other flow uses it with the same kind of request.
1209  *
1210  * @param dev
1211  *   Pointer to Ethernet device.
1212  * @param[in] dev_handle
1213  *   Pointer to the device flow handle structure.
1214  */
1215 static void
1216 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1217 			struct mlx5_flow_handle *dev_handle)
1218 {
1219 	struct mlx5_priv *priv = dev->data->dev_private;
1220 	const int mark = dev_handle->mark;
1221 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1222 	struct mlx5_hrxq *hrxq;
1223 	unsigned int i;
1224 
1225 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1226 		return;
1227 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1228 			      dev_handle->rix_hrxq);
1229 	if (!hrxq)
1230 		return;
1231 	MLX5_ASSERT(dev->data->dev_started);
1232 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1233 		int idx = hrxq->ind_table->queues[i];
1234 		struct mlx5_rxq_ctrl *rxq_ctrl =
1235 			container_of((*priv->rxqs)[idx],
1236 				     struct mlx5_rxq_ctrl, rxq);
1237 
1238 		if (priv->config.dv_flow_en &&
1239 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1240 		    mlx5_flow_ext_mreg_supported(dev)) {
1241 			rxq_ctrl->rxq.mark = 1;
1242 			rxq_ctrl->flow_mark_n = 1;
1243 		} else if (mark) {
1244 			rxq_ctrl->flow_mark_n--;
1245 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1246 		}
1247 		if (tunnel) {
1248 			unsigned int j;
1249 
1250 			/* Decrease the counter matching the flow. */
1251 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1252 				if ((tunnels_info[j].tunnel &
1253 				     dev_handle->layers) ==
1254 				    tunnels_info[j].tunnel) {
1255 					rxq_ctrl->flow_tunnels_n[j]--;
1256 					break;
1257 				}
1258 			}
1259 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1260 		}
1261 	}
1262 }
1263 
1264 /**
1265  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1266  * @p flow if no other flow uses it with the same kind of request.
1267  *
1268  * @param dev
1269  *   Pointer to Ethernet device.
1270  * @param[in] flow
1271  *   Pointer to the flow.
1272  */
1273 static void
1274 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1275 {
1276 	struct mlx5_priv *priv = dev->data->dev_private;
1277 	uint32_t handle_idx;
1278 	struct mlx5_flow_handle *dev_handle;
1279 
1280 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1281 		       handle_idx, dev_handle, next)
1282 		flow_drv_rxq_flags_trim(dev, dev_handle);
1283 }
1284 
1285 /**
1286  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1287  *
1288  * @param dev
1289  *   Pointer to Ethernet device.
1290  */
1291 static void
1292 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1293 {
1294 	struct mlx5_priv *priv = dev->data->dev_private;
1295 	unsigned int i;
1296 
1297 	for (i = 0; i != priv->rxqs_n; ++i) {
1298 		struct mlx5_rxq_ctrl *rxq_ctrl;
1299 		unsigned int j;
1300 
1301 		if (!(*priv->rxqs)[i])
1302 			continue;
1303 		rxq_ctrl = container_of((*priv->rxqs)[i],
1304 					struct mlx5_rxq_ctrl, rxq);
1305 		rxq_ctrl->flow_mark_n = 0;
1306 		rxq_ctrl->rxq.mark = 0;
1307 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1308 			rxq_ctrl->flow_tunnels_n[j] = 0;
1309 		rxq_ctrl->rxq.tunnel = 0;
1310 	}
1311 }
1312 
1313 /**
1314  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1315  *
1316  * @param[in] dev
1317  *   Pointer to the Ethernet device structure.
1318  */
1319 void
1320 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1321 {
1322 	struct mlx5_priv *priv = dev->data->dev_private;
1323 	struct mlx5_rxq_data *data;
1324 	unsigned int i;
1325 
1326 	for (i = 0; i != priv->rxqs_n; ++i) {
1327 		if (!(*priv->rxqs)[i])
1328 			continue;
1329 		data = (*priv->rxqs)[i];
1330 		if (!rte_flow_dynf_metadata_avail()) {
1331 			data->dynf_meta = 0;
1332 			data->flow_meta_mask = 0;
1333 			data->flow_meta_offset = -1;
1334 		} else {
1335 			data->dynf_meta = 1;
1336 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1337 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1338 		}
1339 	}
1340 }
1341 
1342 /*
1343  * return a pointer to the desired action in the list of actions.
1344  *
1345  * @param[in] actions
1346  *   The list of actions to search the action in.
1347  * @param[in] action
1348  *   The action to find.
1349  *
1350  * @return
1351  *   Pointer to the action in the list, if found. NULL otherwise.
1352  */
1353 const struct rte_flow_action *
1354 mlx5_flow_find_action(const struct rte_flow_action *actions,
1355 		      enum rte_flow_action_type action)
1356 {
1357 	if (actions == NULL)
1358 		return NULL;
1359 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1360 		if (actions->type == action)
1361 			return actions;
1362 	return NULL;
1363 }
1364 
1365 /*
1366  * Validate the flag action.
1367  *
1368  * @param[in] action_flags
1369  *   Bit-fields that holds the actions detected until now.
1370  * @param[in] attr
1371  *   Attributes of flow that includes this action.
1372  * @param[out] error
1373  *   Pointer to error structure.
1374  *
1375  * @return
1376  *   0 on success, a negative errno value otherwise and rte_errno is set.
1377  */
1378 int
1379 mlx5_flow_validate_action_flag(uint64_t action_flags,
1380 			       const struct rte_flow_attr *attr,
1381 			       struct rte_flow_error *error)
1382 {
1383 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1384 		return rte_flow_error_set(error, EINVAL,
1385 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1386 					  "can't mark and flag in same flow");
1387 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1388 		return rte_flow_error_set(error, EINVAL,
1389 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1390 					  "can't have 2 flag"
1391 					  " actions in same flow");
1392 	if (attr->egress)
1393 		return rte_flow_error_set(error, ENOTSUP,
1394 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1395 					  "flag action not supported for "
1396 					  "egress");
1397 	return 0;
1398 }
1399 
1400 /*
1401  * Validate the mark action.
1402  *
1403  * @param[in] action
1404  *   Pointer to the queue action.
1405  * @param[in] action_flags
1406  *   Bit-fields that holds the actions detected until now.
1407  * @param[in] attr
1408  *   Attributes of flow that includes this action.
1409  * @param[out] error
1410  *   Pointer to error structure.
1411  *
1412  * @return
1413  *   0 on success, a negative errno value otherwise and rte_errno is set.
1414  */
1415 int
1416 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1417 			       uint64_t action_flags,
1418 			       const struct rte_flow_attr *attr,
1419 			       struct rte_flow_error *error)
1420 {
1421 	const struct rte_flow_action_mark *mark = action->conf;
1422 
1423 	if (!mark)
1424 		return rte_flow_error_set(error, EINVAL,
1425 					  RTE_FLOW_ERROR_TYPE_ACTION,
1426 					  action,
1427 					  "configuration cannot be null");
1428 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1429 		return rte_flow_error_set(error, EINVAL,
1430 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1431 					  &mark->id,
1432 					  "mark id must in 0 <= id < "
1433 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1434 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1435 		return rte_flow_error_set(error, EINVAL,
1436 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1437 					  "can't flag and mark in same flow");
1438 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1439 		return rte_flow_error_set(error, EINVAL,
1440 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1441 					  "can't have 2 mark actions in same"
1442 					  " flow");
1443 	if (attr->egress)
1444 		return rte_flow_error_set(error, ENOTSUP,
1445 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1446 					  "mark action not supported for "
1447 					  "egress");
1448 	return 0;
1449 }
1450 
1451 /*
1452  * Validate the drop action.
1453  *
1454  * @param[in] action_flags
1455  *   Bit-fields that holds the actions detected until now.
1456  * @param[in] attr
1457  *   Attributes of flow that includes this action.
1458  * @param[out] error
1459  *   Pointer to error structure.
1460  *
1461  * @return
1462  *   0 on success, a negative errno value otherwise and rte_errno is set.
1463  */
1464 int
1465 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1466 			       const struct rte_flow_attr *attr,
1467 			       struct rte_flow_error *error)
1468 {
1469 	if (attr->egress)
1470 		return rte_flow_error_set(error, ENOTSUP,
1471 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1472 					  "drop action not supported for "
1473 					  "egress");
1474 	return 0;
1475 }
1476 
1477 /*
1478  * Validate the queue action.
1479  *
1480  * @param[in] action
1481  *   Pointer to the queue action.
1482  * @param[in] action_flags
1483  *   Bit-fields that holds the actions detected until now.
1484  * @param[in] dev
1485  *   Pointer to the Ethernet device structure.
1486  * @param[in] attr
1487  *   Attributes of flow that includes this action.
1488  * @param[out] error
1489  *   Pointer to error structure.
1490  *
1491  * @return
1492  *   0 on success, a negative errno value otherwise and rte_errno is set.
1493  */
1494 int
1495 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1496 				uint64_t action_flags,
1497 				struct rte_eth_dev *dev,
1498 				const struct rte_flow_attr *attr,
1499 				struct rte_flow_error *error)
1500 {
1501 	struct mlx5_priv *priv = dev->data->dev_private;
1502 	const struct rte_flow_action_queue *queue = action->conf;
1503 
1504 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1505 		return rte_flow_error_set(error, EINVAL,
1506 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1507 					  "can't have 2 fate actions in"
1508 					  " same flow");
1509 	if (!priv->rxqs_n)
1510 		return rte_flow_error_set(error, EINVAL,
1511 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1512 					  NULL, "No Rx queues configured");
1513 	if (queue->index >= priv->rxqs_n)
1514 		return rte_flow_error_set(error, EINVAL,
1515 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1516 					  &queue->index,
1517 					  "queue index out of range");
1518 	if (!(*priv->rxqs)[queue->index])
1519 		return rte_flow_error_set(error, EINVAL,
1520 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1521 					  &queue->index,
1522 					  "queue is not configured");
1523 	if (attr->egress)
1524 		return rte_flow_error_set(error, ENOTSUP,
1525 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1526 					  "queue action not supported for "
1527 					  "egress");
1528 	return 0;
1529 }
1530 
1531 /*
1532  * Validate the rss action.
1533  *
1534  * @param[in] dev
1535  *   Pointer to the Ethernet device structure.
1536  * @param[in] action
1537  *   Pointer to the queue action.
1538  * @param[out] error
1539  *   Pointer to error structure.
1540  *
1541  * @return
1542  *   0 on success, a negative errno value otherwise and rte_errno is set.
1543  */
1544 int
1545 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1546 			 const struct rte_flow_action *action,
1547 			 struct rte_flow_error *error)
1548 {
1549 	struct mlx5_priv *priv = dev->data->dev_private;
1550 	const struct rte_flow_action_rss *rss = action->conf;
1551 	unsigned int i;
1552 
1553 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1554 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1555 		return rte_flow_error_set(error, ENOTSUP,
1556 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1557 					  &rss->func,
1558 					  "RSS hash function not supported");
1559 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1560 	if (rss->level > 2)
1561 #else
1562 	if (rss->level > 1)
1563 #endif
1564 		return rte_flow_error_set(error, ENOTSUP,
1565 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1566 					  &rss->level,
1567 					  "tunnel RSS is not supported");
1568 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1569 	if (rss->key_len == 0 && rss->key != NULL)
1570 		return rte_flow_error_set(error, ENOTSUP,
1571 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1572 					  &rss->key_len,
1573 					  "RSS hash key length 0");
1574 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1575 		return rte_flow_error_set(error, ENOTSUP,
1576 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1577 					  &rss->key_len,
1578 					  "RSS hash key too small");
1579 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1580 		return rte_flow_error_set(error, ENOTSUP,
1581 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1582 					  &rss->key_len,
1583 					  "RSS hash key too large");
1584 	if (rss->queue_num > priv->config.ind_table_max_size)
1585 		return rte_flow_error_set(error, ENOTSUP,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1587 					  &rss->queue_num,
1588 					  "number of queues too large");
1589 	if (rss->types & MLX5_RSS_HF_MASK)
1590 		return rte_flow_error_set(error, ENOTSUP,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 					  &rss->types,
1593 					  "some RSS protocols are not"
1594 					  " supported");
1595 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1596 	    !(rss->types & ETH_RSS_IP))
1597 		return rte_flow_error_set(error, EINVAL,
1598 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1599 					  "L3 partial RSS requested but L3 RSS"
1600 					  " type not specified");
1601 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1602 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1603 		return rte_flow_error_set(error, EINVAL,
1604 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1605 					  "L4 partial RSS requested but L4 RSS"
1606 					  " type not specified");
1607 	if (!priv->rxqs_n)
1608 		return rte_flow_error_set(error, EINVAL,
1609 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1610 					  NULL, "No Rx queues configured");
1611 	if (!rss->queue_num)
1612 		return rte_flow_error_set(error, EINVAL,
1613 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1614 					  NULL, "No queues configured");
1615 	for (i = 0; i != rss->queue_num; ++i) {
1616 		if (rss->queue[i] >= priv->rxqs_n)
1617 			return rte_flow_error_set
1618 				(error, EINVAL,
1619 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1620 				 &rss->queue[i], "queue index out of range");
1621 		if (!(*priv->rxqs)[rss->queue[i]])
1622 			return rte_flow_error_set
1623 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1624 				 &rss->queue[i], "queue is not configured");
1625 	}
1626 	return 0;
1627 }
1628 
1629 /*
1630  * Validate the rss action.
1631  *
1632  * @param[in] action
1633  *   Pointer to the queue action.
1634  * @param[in] action_flags
1635  *   Bit-fields that holds the actions detected until now.
1636  * @param[in] dev
1637  *   Pointer to the Ethernet device structure.
1638  * @param[in] attr
1639  *   Attributes of flow that includes this action.
1640  * @param[in] item_flags
1641  *   Items that were detected.
1642  * @param[out] error
1643  *   Pointer to error structure.
1644  *
1645  * @return
1646  *   0 on success, a negative errno value otherwise and rte_errno is set.
1647  */
1648 int
1649 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1650 			      uint64_t action_flags,
1651 			      struct rte_eth_dev *dev,
1652 			      const struct rte_flow_attr *attr,
1653 			      uint64_t item_flags,
1654 			      struct rte_flow_error *error)
1655 {
1656 	const struct rte_flow_action_rss *rss = action->conf;
1657 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1658 	int ret;
1659 
1660 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1661 		return rte_flow_error_set(error, EINVAL,
1662 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1663 					  "can't have 2 fate actions"
1664 					  " in same flow");
1665 	ret = mlx5_validate_action_rss(dev, action, error);
1666 	if (ret)
1667 		return ret;
1668 	if (attr->egress)
1669 		return rte_flow_error_set(error, ENOTSUP,
1670 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1671 					  "rss action not supported for "
1672 					  "egress");
1673 	if (rss->level > 1 && !tunnel)
1674 		return rte_flow_error_set(error, EINVAL,
1675 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1676 					  "inner RSS is not supported for "
1677 					  "non-tunnel flows");
1678 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1679 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1680 		return rte_flow_error_set(error, EINVAL,
1681 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1682 					  "RSS on eCPRI is not supported now");
1683 	}
1684 	return 0;
1685 }
1686 
1687 /*
1688  * Validate the default miss action.
1689  *
1690  * @param[in] action_flags
1691  *   Bit-fields that holds the actions detected until now.
1692  * @param[out] error
1693  *   Pointer to error structure.
1694  *
1695  * @return
1696  *   0 on success, a negative errno value otherwise and rte_errno is set.
1697  */
1698 int
1699 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1700 				const struct rte_flow_attr *attr,
1701 				struct rte_flow_error *error)
1702 {
1703 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1704 		return rte_flow_error_set(error, EINVAL,
1705 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1706 					  "can't have 2 fate actions in"
1707 					  " same flow");
1708 	if (attr->egress)
1709 		return rte_flow_error_set(error, ENOTSUP,
1710 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1711 					  "default miss action not supported "
1712 					  "for egress");
1713 	if (attr->group)
1714 		return rte_flow_error_set(error, ENOTSUP,
1715 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1716 					  "only group 0 is supported");
1717 	if (attr->transfer)
1718 		return rte_flow_error_set(error, ENOTSUP,
1719 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1720 					  NULL, "transfer is not supported");
1721 	return 0;
1722 }
1723 
1724 /*
1725  * Validate the count action.
1726  *
1727  * @param[in] dev
1728  *   Pointer to the Ethernet device structure.
1729  * @param[in] attr
1730  *   Attributes of flow that includes this action.
1731  * @param[out] error
1732  *   Pointer to error structure.
1733  *
1734  * @return
1735  *   0 on success, a negative errno value otherwise and rte_errno is set.
1736  */
1737 int
1738 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1739 				const struct rte_flow_attr *attr,
1740 				struct rte_flow_error *error)
1741 {
1742 	if (attr->egress)
1743 		return rte_flow_error_set(error, ENOTSUP,
1744 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1745 					  "count action not supported for "
1746 					  "egress");
1747 	return 0;
1748 }
1749 
1750 /**
1751  * Verify the @p attributes will be correctly understood by the NIC and store
1752  * them in the @p flow if everything is correct.
1753  *
1754  * @param[in] dev
1755  *   Pointer to the Ethernet device structure.
1756  * @param[in] attributes
1757  *   Pointer to flow attributes
1758  * @param[out] error
1759  *   Pointer to error structure.
1760  *
1761  * @return
1762  *   0 on success, a negative errno value otherwise and rte_errno is set.
1763  */
1764 int
1765 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1766 			      const struct rte_flow_attr *attributes,
1767 			      struct rte_flow_error *error)
1768 {
1769 	struct mlx5_priv *priv = dev->data->dev_private;
1770 	uint32_t priority_max = priv->config.flow_prio - 1;
1771 
1772 	if (attributes->group)
1773 		return rte_flow_error_set(error, ENOTSUP,
1774 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1775 					  NULL, "groups is not supported");
1776 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1777 	    attributes->priority >= priority_max)
1778 		return rte_flow_error_set(error, ENOTSUP,
1779 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1780 					  NULL, "priority out of range");
1781 	if (attributes->egress)
1782 		return rte_flow_error_set(error, ENOTSUP,
1783 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1784 					  "egress is not supported");
1785 	if (attributes->transfer && !priv->config.dv_esw_en)
1786 		return rte_flow_error_set(error, ENOTSUP,
1787 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1788 					  NULL, "transfer is not supported");
1789 	if (!attributes->ingress)
1790 		return rte_flow_error_set(error, EINVAL,
1791 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1792 					  NULL,
1793 					  "ingress attribute is mandatory");
1794 	return 0;
1795 }
1796 
1797 /**
1798  * Validate ICMP6 item.
1799  *
1800  * @param[in] item
1801  *   Item specification.
1802  * @param[in] item_flags
1803  *   Bit-fields that holds the items detected until now.
1804  * @param[in] ext_vlan_sup
1805  *   Whether extended VLAN features are supported or not.
1806  * @param[out] error
1807  *   Pointer to error structure.
1808  *
1809  * @return
1810  *   0 on success, a negative errno value otherwise and rte_errno is set.
1811  */
1812 int
1813 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1814 			       uint64_t item_flags,
1815 			       uint8_t target_protocol,
1816 			       struct rte_flow_error *error)
1817 {
1818 	const struct rte_flow_item_icmp6 *mask = item->mask;
1819 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1820 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1821 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1822 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1823 				      MLX5_FLOW_LAYER_OUTER_L4;
1824 	int ret;
1825 
1826 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1827 		return rte_flow_error_set(error, EINVAL,
1828 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1829 					  "protocol filtering not compatible"
1830 					  " with ICMP6 layer");
1831 	if (!(item_flags & l3m))
1832 		return rte_flow_error_set(error, EINVAL,
1833 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1834 					  "IPv6 is mandatory to filter on"
1835 					  " ICMP6");
1836 	if (item_flags & l4m)
1837 		return rte_flow_error_set(error, EINVAL,
1838 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1839 					  "multiple L4 layers not supported");
1840 	if (!mask)
1841 		mask = &rte_flow_item_icmp6_mask;
1842 	ret = mlx5_flow_item_acceptable
1843 		(item, (const uint8_t *)mask,
1844 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1845 		 sizeof(struct rte_flow_item_icmp6),
1846 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1847 	if (ret < 0)
1848 		return ret;
1849 	return 0;
1850 }
1851 
1852 /**
1853  * Validate ICMP item.
1854  *
1855  * @param[in] item
1856  *   Item specification.
1857  * @param[in] item_flags
1858  *   Bit-fields that holds the items detected until now.
1859  * @param[out] error
1860  *   Pointer to error structure.
1861  *
1862  * @return
1863  *   0 on success, a negative errno value otherwise and rte_errno is set.
1864  */
1865 int
1866 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1867 			     uint64_t item_flags,
1868 			     uint8_t target_protocol,
1869 			     struct rte_flow_error *error)
1870 {
1871 	const struct rte_flow_item_icmp *mask = item->mask;
1872 	const struct rte_flow_item_icmp nic_mask = {
1873 		.hdr.icmp_type = 0xff,
1874 		.hdr.icmp_code = 0xff,
1875 		.hdr.icmp_ident = RTE_BE16(0xffff),
1876 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1877 	};
1878 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1879 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1880 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1881 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1882 				      MLX5_FLOW_LAYER_OUTER_L4;
1883 	int ret;
1884 
1885 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1886 		return rte_flow_error_set(error, EINVAL,
1887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 					  "protocol filtering not compatible"
1889 					  " with ICMP layer");
1890 	if (!(item_flags & l3m))
1891 		return rte_flow_error_set(error, EINVAL,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "IPv4 is mandatory to filter"
1894 					  " on ICMP");
1895 	if (item_flags & l4m)
1896 		return rte_flow_error_set(error, EINVAL,
1897 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1898 					  "multiple L4 layers not supported");
1899 	if (!mask)
1900 		mask = &nic_mask;
1901 	ret = mlx5_flow_item_acceptable
1902 		(item, (const uint8_t *)mask,
1903 		 (const uint8_t *)&nic_mask,
1904 		 sizeof(struct rte_flow_item_icmp),
1905 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1906 	if (ret < 0)
1907 		return ret;
1908 	return 0;
1909 }
1910 
1911 /**
1912  * Validate Ethernet item.
1913  *
1914  * @param[in] item
1915  *   Item specification.
1916  * @param[in] item_flags
1917  *   Bit-fields that holds the items detected until now.
1918  * @param[out] error
1919  *   Pointer to error structure.
1920  *
1921  * @return
1922  *   0 on success, a negative errno value otherwise and rte_errno is set.
1923  */
1924 int
1925 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1926 			    uint64_t item_flags, bool ext_vlan_sup,
1927 			    struct rte_flow_error *error)
1928 {
1929 	const struct rte_flow_item_eth *mask = item->mask;
1930 	const struct rte_flow_item_eth nic_mask = {
1931 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1932 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1933 		.type = RTE_BE16(0xffff),
1934 		.has_vlan = ext_vlan_sup ? 1 : 0,
1935 	};
1936 	int ret;
1937 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1938 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1939 				       MLX5_FLOW_LAYER_OUTER_L2;
1940 
1941 	if (item_flags & ethm)
1942 		return rte_flow_error_set(error, ENOTSUP,
1943 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1944 					  "multiple L2 layers not supported");
1945 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1946 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1947 		return rte_flow_error_set(error, EINVAL,
1948 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1949 					  "L2 layer should not follow "
1950 					  "L3 layers");
1951 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1952 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1953 		return rte_flow_error_set(error, EINVAL,
1954 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1955 					  "L2 layer should not follow VLAN");
1956 	if (!mask)
1957 		mask = &rte_flow_item_eth_mask;
1958 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1959 					(const uint8_t *)&nic_mask,
1960 					sizeof(struct rte_flow_item_eth),
1961 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1962 	return ret;
1963 }
1964 
1965 /**
1966  * Validate VLAN item.
1967  *
1968  * @param[in] item
1969  *   Item specification.
1970  * @param[in] item_flags
1971  *   Bit-fields that holds the items detected until now.
1972  * @param[in] dev
1973  *   Ethernet device flow is being created on.
1974  * @param[out] error
1975  *   Pointer to error structure.
1976  *
1977  * @return
1978  *   0 on success, a negative errno value otherwise and rte_errno is set.
1979  */
1980 int
1981 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1982 			     uint64_t item_flags,
1983 			     struct rte_eth_dev *dev,
1984 			     struct rte_flow_error *error)
1985 {
1986 	const struct rte_flow_item_vlan *spec = item->spec;
1987 	const struct rte_flow_item_vlan *mask = item->mask;
1988 	const struct rte_flow_item_vlan nic_mask = {
1989 		.tci = RTE_BE16(UINT16_MAX),
1990 		.inner_type = RTE_BE16(UINT16_MAX),
1991 	};
1992 	uint16_t vlan_tag = 0;
1993 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1994 	int ret;
1995 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1996 					MLX5_FLOW_LAYER_INNER_L4) :
1997 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1998 					MLX5_FLOW_LAYER_OUTER_L4);
1999 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2000 					MLX5_FLOW_LAYER_OUTER_VLAN;
2001 
2002 	if (item_flags & vlanm)
2003 		return rte_flow_error_set(error, EINVAL,
2004 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2005 					  "multiple VLAN layers not supported");
2006 	else if ((item_flags & l34m) != 0)
2007 		return rte_flow_error_set(error, EINVAL,
2008 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 					  "VLAN cannot follow L3/L4 layer");
2010 	if (!mask)
2011 		mask = &rte_flow_item_vlan_mask;
2012 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2013 					(const uint8_t *)&nic_mask,
2014 					sizeof(struct rte_flow_item_vlan),
2015 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2016 	if (ret)
2017 		return ret;
2018 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2019 		struct mlx5_priv *priv = dev->data->dev_private;
2020 
2021 		if (priv->vmwa_context) {
2022 			/*
2023 			 * Non-NULL context means we have a virtual machine
2024 			 * and SR-IOV enabled, we have to create VLAN interface
2025 			 * to make hypervisor to setup E-Switch vport
2026 			 * context correctly. We avoid creating the multiple
2027 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2028 			 */
2029 			return rte_flow_error_set(error, EINVAL,
2030 						  RTE_FLOW_ERROR_TYPE_ITEM,
2031 						  item,
2032 						  "VLAN tag mask is not"
2033 						  " supported in virtual"
2034 						  " environment");
2035 		}
2036 	}
2037 	if (spec) {
2038 		vlan_tag = spec->tci;
2039 		vlan_tag &= mask->tci;
2040 	}
2041 	/*
2042 	 * From verbs perspective an empty VLAN is equivalent
2043 	 * to a packet without VLAN layer.
2044 	 */
2045 	if (!vlan_tag)
2046 		return rte_flow_error_set(error, EINVAL,
2047 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2048 					  item->spec,
2049 					  "VLAN cannot be empty");
2050 	return 0;
2051 }
2052 
2053 /**
2054  * Validate IPV4 item.
2055  *
2056  * @param[in] item
2057  *   Item specification.
2058  * @param[in] item_flags
2059  *   Bit-fields that holds the items detected until now.
2060  * @param[in] last_item
2061  *   Previous validated item in the pattern items.
2062  * @param[in] ether_type
2063  *   Type in the ethernet layer header (including dot1q).
2064  * @param[in] acc_mask
2065  *   Acceptable mask, if NULL default internal default mask
2066  *   will be used to check whether item fields are supported.
2067  * @param[in] range_accepted
2068  *   True if range of values is accepted for specific fields, false otherwise.
2069  * @param[out] error
2070  *   Pointer to error structure.
2071  *
2072  * @return
2073  *   0 on success, a negative errno value otherwise and rte_errno is set.
2074  */
2075 int
2076 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2077 			     uint64_t item_flags,
2078 			     uint64_t last_item,
2079 			     uint16_t ether_type,
2080 			     const struct rte_flow_item_ipv4 *acc_mask,
2081 			     bool range_accepted,
2082 			     struct rte_flow_error *error)
2083 {
2084 	const struct rte_flow_item_ipv4 *mask = item->mask;
2085 	const struct rte_flow_item_ipv4 *spec = item->spec;
2086 	const struct rte_flow_item_ipv4 nic_mask = {
2087 		.hdr = {
2088 			.src_addr = RTE_BE32(0xffffffff),
2089 			.dst_addr = RTE_BE32(0xffffffff),
2090 			.type_of_service = 0xff,
2091 			.next_proto_id = 0xff,
2092 		},
2093 	};
2094 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2095 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2096 				      MLX5_FLOW_LAYER_OUTER_L3;
2097 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2098 				      MLX5_FLOW_LAYER_OUTER_L4;
2099 	int ret;
2100 	uint8_t next_proto = 0xFF;
2101 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2102 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2103 				  MLX5_FLOW_LAYER_INNER_VLAN);
2104 
2105 	if ((last_item & l2_vlan) && ether_type &&
2106 	    ether_type != RTE_ETHER_TYPE_IPV4)
2107 		return rte_flow_error_set(error, EINVAL,
2108 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2109 					  "IPv4 cannot follow L2/VLAN layer "
2110 					  "which ether type is not IPv4");
2111 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2112 		if (mask && spec)
2113 			next_proto = mask->hdr.next_proto_id &
2114 				     spec->hdr.next_proto_id;
2115 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2116 			return rte_flow_error_set(error, EINVAL,
2117 						  RTE_FLOW_ERROR_TYPE_ITEM,
2118 						  item,
2119 						  "multiple tunnel "
2120 						  "not supported");
2121 	}
2122 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2123 		return rte_flow_error_set(error, EINVAL,
2124 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2125 					  "wrong tunnel type - IPv6 specified "
2126 					  "but IPv4 item provided");
2127 	if (item_flags & l3m)
2128 		return rte_flow_error_set(error, ENOTSUP,
2129 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2130 					  "multiple L3 layers not supported");
2131 	else if (item_flags & l4m)
2132 		return rte_flow_error_set(error, EINVAL,
2133 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2134 					  "L3 cannot follow an L4 layer.");
2135 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2136 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2137 		return rte_flow_error_set(error, EINVAL,
2138 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2139 					  "L3 cannot follow an NVGRE layer.");
2140 	if (!mask)
2141 		mask = &rte_flow_item_ipv4_mask;
2142 	else if (mask->hdr.next_proto_id != 0 &&
2143 		 mask->hdr.next_proto_id != 0xff)
2144 		return rte_flow_error_set(error, EINVAL,
2145 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2146 					  "partial mask is not supported"
2147 					  " for protocol");
2148 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2149 					acc_mask ? (const uint8_t *)acc_mask
2150 						 : (const uint8_t *)&nic_mask,
2151 					sizeof(struct rte_flow_item_ipv4),
2152 					range_accepted, error);
2153 	if (ret < 0)
2154 		return ret;
2155 	return 0;
2156 }
2157 
2158 /**
2159  * Validate IPV6 item.
2160  *
2161  * @param[in] item
2162  *   Item specification.
2163  * @param[in] item_flags
2164  *   Bit-fields that holds the items detected until now.
2165  * @param[in] last_item
2166  *   Previous validated item in the pattern items.
2167  * @param[in] ether_type
2168  *   Type in the ethernet layer header (including dot1q).
2169  * @param[in] acc_mask
2170  *   Acceptable mask, if NULL default internal default mask
2171  *   will be used to check whether item fields are supported.
2172  * @param[out] error
2173  *   Pointer to error structure.
2174  *
2175  * @return
2176  *   0 on success, a negative errno value otherwise and rte_errno is set.
2177  */
2178 int
2179 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2180 			     uint64_t item_flags,
2181 			     uint64_t last_item,
2182 			     uint16_t ether_type,
2183 			     const struct rte_flow_item_ipv6 *acc_mask,
2184 			     struct rte_flow_error *error)
2185 {
2186 	const struct rte_flow_item_ipv6 *mask = item->mask;
2187 	const struct rte_flow_item_ipv6 *spec = item->spec;
2188 	const struct rte_flow_item_ipv6 nic_mask = {
2189 		.hdr = {
2190 			.src_addr =
2191 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2192 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2193 			.dst_addr =
2194 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2195 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2196 			.vtc_flow = RTE_BE32(0xffffffff),
2197 			.proto = 0xff,
2198 		},
2199 	};
2200 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2201 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2202 				      MLX5_FLOW_LAYER_OUTER_L3;
2203 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2204 				      MLX5_FLOW_LAYER_OUTER_L4;
2205 	int ret;
2206 	uint8_t next_proto = 0xFF;
2207 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2208 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2209 				  MLX5_FLOW_LAYER_INNER_VLAN);
2210 
2211 	if ((last_item & l2_vlan) && ether_type &&
2212 	    ether_type != RTE_ETHER_TYPE_IPV6)
2213 		return rte_flow_error_set(error, EINVAL,
2214 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2215 					  "IPv6 cannot follow L2/VLAN layer "
2216 					  "which ether type is not IPv6");
2217 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2218 		next_proto = spec->hdr.proto;
2219 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2220 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2221 			return rte_flow_error_set(error, EINVAL,
2222 						  RTE_FLOW_ERROR_TYPE_ITEM,
2223 						  item,
2224 						  "multiple tunnel "
2225 						  "not supported");
2226 	}
2227 	if (next_proto == IPPROTO_HOPOPTS  ||
2228 	    next_proto == IPPROTO_ROUTING  ||
2229 	    next_proto == IPPROTO_FRAGMENT ||
2230 	    next_proto == IPPROTO_ESP	   ||
2231 	    next_proto == IPPROTO_AH	   ||
2232 	    next_proto == IPPROTO_DSTOPTS)
2233 		return rte_flow_error_set(error, EINVAL,
2234 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2235 					  "IPv6 proto (next header) should "
2236 					  "not be set as extension header");
2237 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2238 		return rte_flow_error_set(error, EINVAL,
2239 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 					  "wrong tunnel type - IPv4 specified "
2241 					  "but IPv6 item provided");
2242 	if (item_flags & l3m)
2243 		return rte_flow_error_set(error, ENOTSUP,
2244 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2245 					  "multiple L3 layers not supported");
2246 	else if (item_flags & l4m)
2247 		return rte_flow_error_set(error, EINVAL,
2248 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2249 					  "L3 cannot follow an L4 layer.");
2250 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2251 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2252 		return rte_flow_error_set(error, EINVAL,
2253 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 					  "L3 cannot follow an NVGRE layer.");
2255 	if (!mask)
2256 		mask = &rte_flow_item_ipv6_mask;
2257 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2258 					acc_mask ? (const uint8_t *)acc_mask
2259 						 : (const uint8_t *)&nic_mask,
2260 					sizeof(struct rte_flow_item_ipv6),
2261 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2262 	if (ret < 0)
2263 		return ret;
2264 	return 0;
2265 }
2266 
2267 /**
2268  * Validate UDP item.
2269  *
2270  * @param[in] item
2271  *   Item specification.
2272  * @param[in] item_flags
2273  *   Bit-fields that holds the items detected until now.
2274  * @param[in] target_protocol
2275  *   The next protocol in the previous item.
2276  * @param[in] flow_mask
2277  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2278  * @param[out] error
2279  *   Pointer to error structure.
2280  *
2281  * @return
2282  *   0 on success, a negative errno value otherwise and rte_errno is set.
2283  */
2284 int
2285 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2286 			    uint64_t item_flags,
2287 			    uint8_t target_protocol,
2288 			    struct rte_flow_error *error)
2289 {
2290 	const struct rte_flow_item_udp *mask = item->mask;
2291 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2292 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2293 				      MLX5_FLOW_LAYER_OUTER_L3;
2294 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2295 				      MLX5_FLOW_LAYER_OUTER_L4;
2296 	int ret;
2297 
2298 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2299 		return rte_flow_error_set(error, EINVAL,
2300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2301 					  "protocol filtering not compatible"
2302 					  " with UDP layer");
2303 	if (!(item_flags & l3m))
2304 		return rte_flow_error_set(error, EINVAL,
2305 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2306 					  "L3 is mandatory to filter on L4");
2307 	if (item_flags & l4m)
2308 		return rte_flow_error_set(error, EINVAL,
2309 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2310 					  "multiple L4 layers not supported");
2311 	if (!mask)
2312 		mask = &rte_flow_item_udp_mask;
2313 	ret = mlx5_flow_item_acceptable
2314 		(item, (const uint8_t *)mask,
2315 		 (const uint8_t *)&rte_flow_item_udp_mask,
2316 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2317 		 error);
2318 	if (ret < 0)
2319 		return ret;
2320 	return 0;
2321 }
2322 
2323 /**
2324  * Validate TCP item.
2325  *
2326  * @param[in] item
2327  *   Item specification.
2328  * @param[in] item_flags
2329  *   Bit-fields that holds the items detected until now.
2330  * @param[in] target_protocol
2331  *   The next protocol in the previous item.
2332  * @param[out] error
2333  *   Pointer to error structure.
2334  *
2335  * @return
2336  *   0 on success, a negative errno value otherwise and rte_errno is set.
2337  */
2338 int
2339 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2340 			    uint64_t item_flags,
2341 			    uint8_t target_protocol,
2342 			    const struct rte_flow_item_tcp *flow_mask,
2343 			    struct rte_flow_error *error)
2344 {
2345 	const struct rte_flow_item_tcp *mask = item->mask;
2346 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2347 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2348 				      MLX5_FLOW_LAYER_OUTER_L3;
2349 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2350 				      MLX5_FLOW_LAYER_OUTER_L4;
2351 	int ret;
2352 
2353 	MLX5_ASSERT(flow_mask);
2354 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2355 		return rte_flow_error_set(error, EINVAL,
2356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2357 					  "protocol filtering not compatible"
2358 					  " with TCP layer");
2359 	if (!(item_flags & l3m))
2360 		return rte_flow_error_set(error, EINVAL,
2361 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2362 					  "L3 is mandatory to filter on L4");
2363 	if (item_flags & l4m)
2364 		return rte_flow_error_set(error, EINVAL,
2365 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2366 					  "multiple L4 layers not supported");
2367 	if (!mask)
2368 		mask = &rte_flow_item_tcp_mask;
2369 	ret = mlx5_flow_item_acceptable
2370 		(item, (const uint8_t *)mask,
2371 		 (const uint8_t *)flow_mask,
2372 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2373 		 error);
2374 	if (ret < 0)
2375 		return ret;
2376 	return 0;
2377 }
2378 
2379 /**
2380  * Validate VXLAN 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] target_protocol
2387  *   The next protocol in the previous item.
2388  * @param[out] error
2389  *   Pointer to error structure.
2390  *
2391  * @return
2392  *   0 on success, a negative errno value otherwise and rte_errno is set.
2393  */
2394 int
2395 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2396 			      uint64_t item_flags,
2397 			      struct rte_flow_error *error)
2398 {
2399 	const struct rte_flow_item_vxlan *spec = item->spec;
2400 	const struct rte_flow_item_vxlan *mask = item->mask;
2401 	int ret;
2402 	union vni {
2403 		uint32_t vlan_id;
2404 		uint8_t vni[4];
2405 	} id = { .vlan_id = 0, };
2406 
2407 
2408 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2409 		return rte_flow_error_set(error, ENOTSUP,
2410 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2411 					  "multiple tunnel layers not"
2412 					  " supported");
2413 	/*
2414 	 * Verify only UDPv4 is present as defined in
2415 	 * https://tools.ietf.org/html/rfc7348
2416 	 */
2417 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2418 		return rte_flow_error_set(error, EINVAL,
2419 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2420 					  "no outer UDP layer found");
2421 	if (!mask)
2422 		mask = &rte_flow_item_vxlan_mask;
2423 	ret = mlx5_flow_item_acceptable
2424 		(item, (const uint8_t *)mask,
2425 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2426 		 sizeof(struct rte_flow_item_vxlan),
2427 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2428 	if (ret < 0)
2429 		return ret;
2430 	if (spec) {
2431 		memcpy(&id.vni[1], spec->vni, 3);
2432 		memcpy(&id.vni[1], mask->vni, 3);
2433 	}
2434 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2435 		return rte_flow_error_set(error, ENOTSUP,
2436 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2437 					  "VXLAN tunnel must be fully defined");
2438 	return 0;
2439 }
2440 
2441 /**
2442  * Validate VXLAN_GPE item.
2443  *
2444  * @param[in] item
2445  *   Item specification.
2446  * @param[in] item_flags
2447  *   Bit-fields that holds the items detected until now.
2448  * @param[in] priv
2449  *   Pointer to the private data structure.
2450  * @param[in] target_protocol
2451  *   The next protocol in the previous item.
2452  * @param[out] error
2453  *   Pointer to error structure.
2454  *
2455  * @return
2456  *   0 on success, a negative errno value otherwise and rte_errno is set.
2457  */
2458 int
2459 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2460 				  uint64_t item_flags,
2461 				  struct rte_eth_dev *dev,
2462 				  struct rte_flow_error *error)
2463 {
2464 	struct mlx5_priv *priv = dev->data->dev_private;
2465 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2466 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2467 	int ret;
2468 	union vni {
2469 		uint32_t vlan_id;
2470 		uint8_t vni[4];
2471 	} id = { .vlan_id = 0, };
2472 
2473 	if (!priv->config.l3_vxlan_en)
2474 		return rte_flow_error_set(error, ENOTSUP,
2475 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2476 					  "L3 VXLAN is not enabled by device"
2477 					  " parameter and/or not configured in"
2478 					  " firmware");
2479 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2480 		return rte_flow_error_set(error, ENOTSUP,
2481 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2482 					  "multiple tunnel layers not"
2483 					  " supported");
2484 	/*
2485 	 * Verify only UDPv4 is present as defined in
2486 	 * https://tools.ietf.org/html/rfc7348
2487 	 */
2488 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2489 		return rte_flow_error_set(error, EINVAL,
2490 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2491 					  "no outer UDP layer found");
2492 	if (!mask)
2493 		mask = &rte_flow_item_vxlan_gpe_mask;
2494 	ret = mlx5_flow_item_acceptable
2495 		(item, (const uint8_t *)mask,
2496 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2497 		 sizeof(struct rte_flow_item_vxlan_gpe),
2498 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2499 	if (ret < 0)
2500 		return ret;
2501 	if (spec) {
2502 		if (spec->protocol)
2503 			return rte_flow_error_set(error, ENOTSUP,
2504 						  RTE_FLOW_ERROR_TYPE_ITEM,
2505 						  item,
2506 						  "VxLAN-GPE protocol"
2507 						  " not supported");
2508 		memcpy(&id.vni[1], spec->vni, 3);
2509 		memcpy(&id.vni[1], mask->vni, 3);
2510 	}
2511 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2512 		return rte_flow_error_set(error, ENOTSUP,
2513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2514 					  "VXLAN-GPE tunnel must be fully"
2515 					  " defined");
2516 	return 0;
2517 }
2518 /**
2519  * Validate GRE Key item.
2520  *
2521  * @param[in] item
2522  *   Item specification.
2523  * @param[in] item_flags
2524  *   Bit flags to mark detected items.
2525  * @param[in] gre_item
2526  *   Pointer to gre_item
2527  * @param[out] error
2528  *   Pointer to error structure.
2529  *
2530  * @return
2531  *   0 on success, a negative errno value otherwise and rte_errno is set.
2532  */
2533 int
2534 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2535 				uint64_t item_flags,
2536 				const struct rte_flow_item *gre_item,
2537 				struct rte_flow_error *error)
2538 {
2539 	const rte_be32_t *mask = item->mask;
2540 	int ret = 0;
2541 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2542 	const struct rte_flow_item_gre *gre_spec;
2543 	const struct rte_flow_item_gre *gre_mask;
2544 
2545 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2546 		return rte_flow_error_set(error, ENOTSUP,
2547 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2548 					  "Multiple GRE key not support");
2549 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2550 		return rte_flow_error_set(error, ENOTSUP,
2551 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2552 					  "No preceding GRE header");
2553 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2554 		return rte_flow_error_set(error, ENOTSUP,
2555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2556 					  "GRE key following a wrong item");
2557 	gre_mask = gre_item->mask;
2558 	if (!gre_mask)
2559 		gre_mask = &rte_flow_item_gre_mask;
2560 	gre_spec = gre_item->spec;
2561 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2562 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2563 		return rte_flow_error_set(error, EINVAL,
2564 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2565 					  "Key bit must be on");
2566 
2567 	if (!mask)
2568 		mask = &gre_key_default_mask;
2569 	ret = mlx5_flow_item_acceptable
2570 		(item, (const uint8_t *)mask,
2571 		 (const uint8_t *)&gre_key_default_mask,
2572 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2573 	return ret;
2574 }
2575 
2576 /**
2577  * Validate GRE item.
2578  *
2579  * @param[in] item
2580  *   Item specification.
2581  * @param[in] item_flags
2582  *   Bit flags to mark detected items.
2583  * @param[in] target_protocol
2584  *   The next protocol in the previous item.
2585  * @param[out] error
2586  *   Pointer to error structure.
2587  *
2588  * @return
2589  *   0 on success, a negative errno value otherwise and rte_errno is set.
2590  */
2591 int
2592 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2593 			    uint64_t item_flags,
2594 			    uint8_t target_protocol,
2595 			    struct rte_flow_error *error)
2596 {
2597 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2598 	const struct rte_flow_item_gre *mask = item->mask;
2599 	int ret;
2600 	const struct rte_flow_item_gre nic_mask = {
2601 		.c_rsvd0_ver = RTE_BE16(0xB000),
2602 		.protocol = RTE_BE16(UINT16_MAX),
2603 	};
2604 
2605 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2606 		return rte_flow_error_set(error, EINVAL,
2607 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2608 					  "protocol filtering not compatible"
2609 					  " with this GRE layer");
2610 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2611 		return rte_flow_error_set(error, ENOTSUP,
2612 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2613 					  "multiple tunnel layers not"
2614 					  " supported");
2615 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2616 		return rte_flow_error_set(error, ENOTSUP,
2617 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2618 					  "L3 Layer is missing");
2619 	if (!mask)
2620 		mask = &rte_flow_item_gre_mask;
2621 	ret = mlx5_flow_item_acceptable
2622 		(item, (const uint8_t *)mask,
2623 		 (const uint8_t *)&nic_mask,
2624 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2625 		 error);
2626 	if (ret < 0)
2627 		return ret;
2628 #ifndef HAVE_MLX5DV_DR
2629 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2630 	if (spec && (spec->protocol & mask->protocol))
2631 		return rte_flow_error_set(error, ENOTSUP,
2632 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2633 					  "without MPLS support the"
2634 					  " specification cannot be used for"
2635 					  " filtering");
2636 #endif
2637 #endif
2638 	return 0;
2639 }
2640 
2641 /**
2642  * Validate Geneve item.
2643  *
2644  * @param[in] item
2645  *   Item specification.
2646  * @param[in] itemFlags
2647  *   Bit-fields that holds the items detected until now.
2648  * @param[in] enPriv
2649  *   Pointer to the private data structure.
2650  * @param[out] error
2651  *   Pointer to error structure.
2652  *
2653  * @return
2654  *   0 on success, a negative errno value otherwise and rte_errno is set.
2655  */
2656 
2657 int
2658 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2659 			       uint64_t item_flags,
2660 			       struct rte_eth_dev *dev,
2661 			       struct rte_flow_error *error)
2662 {
2663 	struct mlx5_priv *priv = dev->data->dev_private;
2664 	const struct rte_flow_item_geneve *spec = item->spec;
2665 	const struct rte_flow_item_geneve *mask = item->mask;
2666 	int ret;
2667 	uint16_t gbhdr;
2668 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2669 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2670 	const struct rte_flow_item_geneve nic_mask = {
2671 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2672 		.vni = "\xff\xff\xff",
2673 		.protocol = RTE_BE16(UINT16_MAX),
2674 	};
2675 
2676 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2677 		return rte_flow_error_set(error, ENOTSUP,
2678 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 					  "L3 Geneve is not enabled by device"
2680 					  " parameter and/or not configured in"
2681 					  " firmware");
2682 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2683 		return rte_flow_error_set(error, ENOTSUP,
2684 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2685 					  "multiple tunnel layers not"
2686 					  " supported");
2687 	/*
2688 	 * Verify only UDPv4 is present as defined in
2689 	 * https://tools.ietf.org/html/rfc7348
2690 	 */
2691 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2692 		return rte_flow_error_set(error, EINVAL,
2693 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2694 					  "no outer UDP layer found");
2695 	if (!mask)
2696 		mask = &rte_flow_item_geneve_mask;
2697 	ret = mlx5_flow_item_acceptable
2698 				  (item, (const uint8_t *)mask,
2699 				   (const uint8_t *)&nic_mask,
2700 				   sizeof(struct rte_flow_item_geneve),
2701 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2702 	if (ret)
2703 		return ret;
2704 	if (spec) {
2705 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2706 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2707 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2708 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2709 			return rte_flow_error_set(error, ENOTSUP,
2710 						  RTE_FLOW_ERROR_TYPE_ITEM,
2711 						  item,
2712 						  "Geneve protocol unsupported"
2713 						  " fields are being used");
2714 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2715 			return rte_flow_error_set
2716 					(error, ENOTSUP,
2717 					 RTE_FLOW_ERROR_TYPE_ITEM,
2718 					 item,
2719 					 "Unsupported Geneve options length");
2720 	}
2721 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2722 		return rte_flow_error_set
2723 				    (error, ENOTSUP,
2724 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2725 				     "Geneve tunnel must be fully defined");
2726 	return 0;
2727 }
2728 
2729 /**
2730  * Validate MPLS item.
2731  *
2732  * @param[in] dev
2733  *   Pointer to the rte_eth_dev structure.
2734  * @param[in] item
2735  *   Item specification.
2736  * @param[in] item_flags
2737  *   Bit-fields that holds the items detected until now.
2738  * @param[in] prev_layer
2739  *   The protocol layer indicated in previous item.
2740  * @param[out] error
2741  *   Pointer to error structure.
2742  *
2743  * @return
2744  *   0 on success, a negative errno value otherwise and rte_errno is set.
2745  */
2746 int
2747 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2748 			     const struct rte_flow_item *item __rte_unused,
2749 			     uint64_t item_flags __rte_unused,
2750 			     uint64_t prev_layer __rte_unused,
2751 			     struct rte_flow_error *error)
2752 {
2753 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2754 	const struct rte_flow_item_mpls *mask = item->mask;
2755 	struct mlx5_priv *priv = dev->data->dev_private;
2756 	int ret;
2757 
2758 	if (!priv->config.mpls_en)
2759 		return rte_flow_error_set(error, ENOTSUP,
2760 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2761 					  "MPLS not supported or"
2762 					  " disabled in firmware"
2763 					  " configuration.");
2764 	/* MPLS over IP, UDP, GRE is allowed */
2765 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2766 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2767 			    MLX5_FLOW_LAYER_GRE)))
2768 		return rte_flow_error_set(error, EINVAL,
2769 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2770 					  "protocol filtering not compatible"
2771 					  " with MPLS layer");
2772 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2773 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2774 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2775 		return rte_flow_error_set(error, ENOTSUP,
2776 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2777 					  "multiple tunnel layers not"
2778 					  " supported");
2779 	if (!mask)
2780 		mask = &rte_flow_item_mpls_mask;
2781 	ret = mlx5_flow_item_acceptable
2782 		(item, (const uint8_t *)mask,
2783 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2784 		 sizeof(struct rte_flow_item_mpls),
2785 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2786 	if (ret < 0)
2787 		return ret;
2788 	return 0;
2789 #else
2790 	return rte_flow_error_set(error, ENOTSUP,
2791 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2792 				  "MPLS is not supported by Verbs, please"
2793 				  " update.");
2794 #endif
2795 }
2796 
2797 /**
2798  * Validate NVGRE item.
2799  *
2800  * @param[in] item
2801  *   Item specification.
2802  * @param[in] item_flags
2803  *   Bit flags to mark detected items.
2804  * @param[in] target_protocol
2805  *   The next protocol in the previous item.
2806  * @param[out] error
2807  *   Pointer to error structure.
2808  *
2809  * @return
2810  *   0 on success, a negative errno value otherwise and rte_errno is set.
2811  */
2812 int
2813 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2814 			      uint64_t item_flags,
2815 			      uint8_t target_protocol,
2816 			      struct rte_flow_error *error)
2817 {
2818 	const struct rte_flow_item_nvgre *mask = item->mask;
2819 	int ret;
2820 
2821 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2822 		return rte_flow_error_set(error, EINVAL,
2823 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2824 					  "protocol filtering not compatible"
2825 					  " with this GRE layer");
2826 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2827 		return rte_flow_error_set(error, ENOTSUP,
2828 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2829 					  "multiple tunnel layers not"
2830 					  " supported");
2831 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2832 		return rte_flow_error_set(error, ENOTSUP,
2833 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2834 					  "L3 Layer is missing");
2835 	if (!mask)
2836 		mask = &rte_flow_item_nvgre_mask;
2837 	ret = mlx5_flow_item_acceptable
2838 		(item, (const uint8_t *)mask,
2839 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2840 		 sizeof(struct rte_flow_item_nvgre),
2841 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2842 	if (ret < 0)
2843 		return ret;
2844 	return 0;
2845 }
2846 
2847 /**
2848  * Validate eCPRI item.
2849  *
2850  * @param[in] item
2851  *   Item specification.
2852  * @param[in] item_flags
2853  *   Bit-fields that holds the items detected until now.
2854  * @param[in] last_item
2855  *   Previous validated item in the pattern items.
2856  * @param[in] ether_type
2857  *   Type in the ethernet layer header (including dot1q).
2858  * @param[in] acc_mask
2859  *   Acceptable mask, if NULL default internal default mask
2860  *   will be used to check whether item fields are supported.
2861  * @param[out] error
2862  *   Pointer to error structure.
2863  *
2864  * @return
2865  *   0 on success, a negative errno value otherwise and rte_errno is set.
2866  */
2867 int
2868 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2869 			      uint64_t item_flags,
2870 			      uint64_t last_item,
2871 			      uint16_t ether_type,
2872 			      const struct rte_flow_item_ecpri *acc_mask,
2873 			      struct rte_flow_error *error)
2874 {
2875 	const struct rte_flow_item_ecpri *mask = item->mask;
2876 	const struct rte_flow_item_ecpri nic_mask = {
2877 		.hdr = {
2878 			.common = {
2879 				.u32 =
2880 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2881 					.type = 0xFF,
2882 					}).u32),
2883 			},
2884 			.dummy[0] = 0xFFFFFFFF,
2885 		},
2886 	};
2887 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2888 					MLX5_FLOW_LAYER_OUTER_VLAN);
2889 	struct rte_flow_item_ecpri mask_lo;
2890 
2891 	if ((last_item & outer_l2_vlan) && ether_type &&
2892 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2893 		return rte_flow_error_set(error, EINVAL,
2894 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2895 					  "eCPRI cannot follow L2/VLAN layer "
2896 					  "which ether type is not 0xAEFE.");
2897 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2898 		return rte_flow_error_set(error, EINVAL,
2899 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2900 					  "eCPRI with tunnel is not supported "
2901 					  "right now.");
2902 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2903 		return rte_flow_error_set(error, ENOTSUP,
2904 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2905 					  "multiple L3 layers not supported");
2906 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2907 		return rte_flow_error_set(error, EINVAL,
2908 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2909 					  "eCPRI cannot follow a TCP layer.");
2910 	/* In specification, eCPRI could be over UDP layer. */
2911 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2912 		return rte_flow_error_set(error, EINVAL,
2913 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2914 					  "eCPRI over UDP layer is not yet "
2915 					  "supported right now.");
2916 	/* Mask for type field in common header could be zero. */
2917 	if (!mask)
2918 		mask = &rte_flow_item_ecpri_mask;
2919 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2920 	/* Input mask is in big-endian format. */
2921 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2922 		return rte_flow_error_set(error, EINVAL,
2923 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2924 					  "partial mask is not supported "
2925 					  "for protocol");
2926 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2927 		return rte_flow_error_set(error, EINVAL,
2928 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2929 					  "message header mask must be after "
2930 					  "a type mask");
2931 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2932 					 acc_mask ? (const uint8_t *)acc_mask
2933 						  : (const uint8_t *)&nic_mask,
2934 					 sizeof(struct rte_flow_item_ecpri),
2935 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2936 }
2937 
2938 /**
2939  * Release resource related QUEUE/RSS action split.
2940  *
2941  * @param dev
2942  *   Pointer to Ethernet device.
2943  * @param flow
2944  *   Flow to release id's from.
2945  */
2946 static void
2947 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2948 			     struct rte_flow *flow)
2949 {
2950 	struct mlx5_priv *priv = dev->data->dev_private;
2951 	uint32_t handle_idx;
2952 	struct mlx5_flow_handle *dev_handle;
2953 
2954 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2955 		       handle_idx, dev_handle, next)
2956 		if (dev_handle->split_flow_id)
2957 			mlx5_ipool_free(priv->sh->ipool
2958 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2959 					dev_handle->split_flow_id);
2960 }
2961 
2962 static int
2963 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2964 		   const struct rte_flow_attr *attr __rte_unused,
2965 		   const struct rte_flow_item items[] __rte_unused,
2966 		   const struct rte_flow_action actions[] __rte_unused,
2967 		   bool external __rte_unused,
2968 		   int hairpin __rte_unused,
2969 		   struct rte_flow_error *error)
2970 {
2971 	return rte_flow_error_set(error, ENOTSUP,
2972 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2973 }
2974 
2975 static struct mlx5_flow *
2976 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2977 		  const struct rte_flow_attr *attr __rte_unused,
2978 		  const struct rte_flow_item items[] __rte_unused,
2979 		  const struct rte_flow_action actions[] __rte_unused,
2980 		  struct rte_flow_error *error)
2981 {
2982 	rte_flow_error_set(error, ENOTSUP,
2983 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2984 	return NULL;
2985 }
2986 
2987 static int
2988 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2989 		    struct mlx5_flow *dev_flow __rte_unused,
2990 		    const struct rte_flow_attr *attr __rte_unused,
2991 		    const struct rte_flow_item items[] __rte_unused,
2992 		    const struct rte_flow_action actions[] __rte_unused,
2993 		    struct rte_flow_error *error)
2994 {
2995 	return rte_flow_error_set(error, ENOTSUP,
2996 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2997 }
2998 
2999 static int
3000 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3001 		struct rte_flow *flow __rte_unused,
3002 		struct rte_flow_error *error)
3003 {
3004 	return rte_flow_error_set(error, ENOTSUP,
3005 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3006 }
3007 
3008 static void
3009 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3010 		 struct rte_flow *flow __rte_unused)
3011 {
3012 }
3013 
3014 static void
3015 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3016 		  struct rte_flow *flow __rte_unused)
3017 {
3018 }
3019 
3020 static int
3021 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3022 		struct rte_flow *flow __rte_unused,
3023 		const struct rte_flow_action *actions __rte_unused,
3024 		void *data __rte_unused,
3025 		struct rte_flow_error *error)
3026 {
3027 	return rte_flow_error_set(error, ENOTSUP,
3028 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3029 }
3030 
3031 static int
3032 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3033 		      uint32_t domains __rte_unused,
3034 		      uint32_t flags __rte_unused)
3035 {
3036 	return 0;
3037 }
3038 
3039 /* Void driver to protect from null pointer reference. */
3040 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3041 	.validate = flow_null_validate,
3042 	.prepare = flow_null_prepare,
3043 	.translate = flow_null_translate,
3044 	.apply = flow_null_apply,
3045 	.remove = flow_null_remove,
3046 	.destroy = flow_null_destroy,
3047 	.query = flow_null_query,
3048 	.sync_domain = flow_null_sync_domain,
3049 };
3050 
3051 /**
3052  * Select flow driver type according to flow attributes and device
3053  * configuration.
3054  *
3055  * @param[in] dev
3056  *   Pointer to the dev structure.
3057  * @param[in] attr
3058  *   Pointer to the flow attributes.
3059  *
3060  * @return
3061  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3062  */
3063 static enum mlx5_flow_drv_type
3064 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3065 {
3066 	struct mlx5_priv *priv = dev->data->dev_private;
3067 	/* The OS can determine first a specific flow type (DV, VERBS) */
3068 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3069 
3070 	if (type != MLX5_FLOW_TYPE_MAX)
3071 		return type;
3072 	/* If no OS specific type - continue with DV/VERBS selection */
3073 	if (attr->transfer && priv->config.dv_esw_en)
3074 		type = MLX5_FLOW_TYPE_DV;
3075 	if (!attr->transfer)
3076 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3077 						 MLX5_FLOW_TYPE_VERBS;
3078 	return type;
3079 }
3080 
3081 #define flow_get_drv_ops(type) flow_drv_ops[type]
3082 
3083 /**
3084  * Flow driver validation API. This abstracts calling driver specific functions.
3085  * The type of flow driver is determined according to flow attributes.
3086  *
3087  * @param[in] dev
3088  *   Pointer to the dev structure.
3089  * @param[in] attr
3090  *   Pointer to the flow attributes.
3091  * @param[in] items
3092  *   Pointer to the list of items.
3093  * @param[in] actions
3094  *   Pointer to the list of actions.
3095  * @param[in] external
3096  *   This flow rule is created by request external to PMD.
3097  * @param[in] hairpin
3098  *   Number of hairpin TX actions, 0 means classic flow.
3099  * @param[out] error
3100  *   Pointer to the error structure.
3101  *
3102  * @return
3103  *   0 on success, a negative errno value otherwise and rte_errno is set.
3104  */
3105 static inline int
3106 flow_drv_validate(struct rte_eth_dev *dev,
3107 		  const struct rte_flow_attr *attr,
3108 		  const struct rte_flow_item items[],
3109 		  const struct rte_flow_action actions[],
3110 		  bool external, int hairpin, struct rte_flow_error *error)
3111 {
3112 	const struct mlx5_flow_driver_ops *fops;
3113 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3114 
3115 	fops = flow_get_drv_ops(type);
3116 	return fops->validate(dev, attr, items, actions, external,
3117 			      hairpin, error);
3118 }
3119 
3120 /**
3121  * Flow driver preparation API. This abstracts calling driver specific
3122  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3123  * calculates the size of memory required for device flow, allocates the memory,
3124  * initializes the device flow and returns the pointer.
3125  *
3126  * @note
3127  *   This function initializes device flow structure such as dv or verbs in
3128  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3129  *   rest. For example, adding returning device flow to flow->dev_flow list and
3130  *   setting backward reference to the flow should be done out of this function.
3131  *   layers field is not filled either.
3132  *
3133  * @param[in] dev
3134  *   Pointer to the dev structure.
3135  * @param[in] attr
3136  *   Pointer to the flow attributes.
3137  * @param[in] items
3138  *   Pointer to the list of items.
3139  * @param[in] actions
3140  *   Pointer to the list of actions.
3141  * @param[in] flow_idx
3142  *   This memory pool index to the flow.
3143  * @param[out] error
3144  *   Pointer to the error structure.
3145  *
3146  * @return
3147  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3148  */
3149 static inline struct mlx5_flow *
3150 flow_drv_prepare(struct rte_eth_dev *dev,
3151 		 const struct rte_flow *flow,
3152 		 const struct rte_flow_attr *attr,
3153 		 const struct rte_flow_item items[],
3154 		 const struct rte_flow_action actions[],
3155 		 uint32_t flow_idx,
3156 		 struct rte_flow_error *error)
3157 {
3158 	const struct mlx5_flow_driver_ops *fops;
3159 	enum mlx5_flow_drv_type type = flow->drv_type;
3160 	struct mlx5_flow *mlx5_flow = NULL;
3161 
3162 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3163 	fops = flow_get_drv_ops(type);
3164 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3165 	if (mlx5_flow)
3166 		mlx5_flow->flow_idx = flow_idx;
3167 	return mlx5_flow;
3168 }
3169 
3170 /**
3171  * Flow driver translation API. This abstracts calling driver specific
3172  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3173  * translates a generic flow into a driver flow. flow_drv_prepare() must
3174  * precede.
3175  *
3176  * @note
3177  *   dev_flow->layers could be filled as a result of parsing during translation
3178  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3179  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3180  *   flow->actions could be overwritten even though all the expanded dev_flows
3181  *   have the same actions.
3182  *
3183  * @param[in] dev
3184  *   Pointer to the rte dev structure.
3185  * @param[in, out] dev_flow
3186  *   Pointer to the mlx5 flow.
3187  * @param[in] attr
3188  *   Pointer to the flow attributes.
3189  * @param[in] items
3190  *   Pointer to the list of items.
3191  * @param[in] actions
3192  *   Pointer to the list of actions.
3193  * @param[out] error
3194  *   Pointer to the error structure.
3195  *
3196  * @return
3197  *   0 on success, a negative errno value otherwise and rte_errno is set.
3198  */
3199 static inline int
3200 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3201 		   const struct rte_flow_attr *attr,
3202 		   const struct rte_flow_item items[],
3203 		   const struct rte_flow_action actions[],
3204 		   struct rte_flow_error *error)
3205 {
3206 	const struct mlx5_flow_driver_ops *fops;
3207 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3208 
3209 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3210 	fops = flow_get_drv_ops(type);
3211 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3212 }
3213 
3214 /**
3215  * Flow driver apply API. This abstracts calling driver specific functions.
3216  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3217  * translated driver flows on to device. flow_drv_translate() must precede.
3218  *
3219  * @param[in] dev
3220  *   Pointer to Ethernet device structure.
3221  * @param[in, out] flow
3222  *   Pointer to flow structure.
3223  * @param[out] error
3224  *   Pointer to error structure.
3225  *
3226  * @return
3227  *   0 on success, a negative errno value otherwise and rte_errno is set.
3228  */
3229 static inline int
3230 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3231 	       struct rte_flow_error *error)
3232 {
3233 	const struct mlx5_flow_driver_ops *fops;
3234 	enum mlx5_flow_drv_type type = flow->drv_type;
3235 
3236 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3237 	fops = flow_get_drv_ops(type);
3238 	return fops->apply(dev, flow, error);
3239 }
3240 
3241 /**
3242  * Flow driver destroy API. This abstracts calling driver specific functions.
3243  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3244  * on device and releases resources of the flow.
3245  *
3246  * @param[in] dev
3247  *   Pointer to Ethernet device.
3248  * @param[in, out] flow
3249  *   Pointer to flow structure.
3250  */
3251 static inline void
3252 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3253 {
3254 	const struct mlx5_flow_driver_ops *fops;
3255 	enum mlx5_flow_drv_type type = flow->drv_type;
3256 
3257 	flow_mreg_split_qrss_release(dev, flow);
3258 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3259 	fops = flow_get_drv_ops(type);
3260 	fops->destroy(dev, flow);
3261 }
3262 
3263 /**
3264  * Get RSS action from the action list.
3265  *
3266  * @param[in] actions
3267  *   Pointer to the list of actions.
3268  *
3269  * @return
3270  *   Pointer to the RSS action if exist, else return NULL.
3271  */
3272 static const struct rte_flow_action_rss*
3273 flow_get_rss_action(const struct rte_flow_action actions[])
3274 {
3275 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3276 		switch (actions->type) {
3277 		case RTE_FLOW_ACTION_TYPE_RSS:
3278 			return (const struct rte_flow_action_rss *)
3279 			       actions->conf;
3280 		default:
3281 			break;
3282 		}
3283 	}
3284 	return NULL;
3285 }
3286 
3287 /* maps shared action to translated non shared in some actions array */
3288 struct mlx5_translated_shared_action {
3289 	struct rte_flow_shared_action *action; /**< Shared action */
3290 	int index; /**< Index in related array of rte_flow_action */
3291 };
3292 
3293 /**
3294  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3295  * non shared action if translation possible.
3296  * This functionality used to run same execution path for both shared & non
3297  * shared actions on flow create. All necessary preparations for shared
3298  * action handling should be preformed on *shared* actions list returned
3299  * from this call.
3300  *
3301  * @param[in] actions
3302  *   List of actions to translate.
3303  * @param[out] shared
3304  *   List to store translated shared actions.
3305  * @param[in, out] shared_n
3306  *   Size of *shared* array. On return should be updated with number of shared
3307  *   actions retrieved from the *actions* list.
3308  * @param[out] translated_actions
3309  *   List of actions where all shared actions were translated to non shared
3310  *   if possible. NULL if no translation took place.
3311  * @param[out] error
3312  *   Pointer to the error structure.
3313  *
3314  * @return
3315  *   0 on success, a negative errno value otherwise and rte_errno is set.
3316  */
3317 static int
3318 flow_shared_actions_translate(const struct rte_flow_action actions[],
3319 	struct mlx5_translated_shared_action *shared,
3320 	int *shared_n,
3321 	struct rte_flow_action **translated_actions,
3322 	struct rte_flow_error *error)
3323 {
3324 	struct rte_flow_action *translated = NULL;
3325 	size_t actions_size;
3326 	int n;
3327 	int copied_n = 0;
3328 	struct mlx5_translated_shared_action *shared_end = NULL;
3329 
3330 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3331 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3332 			continue;
3333 		if (copied_n == *shared_n) {
3334 			return rte_flow_error_set
3335 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3336 				 NULL, "too many shared actions");
3337 		}
3338 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3339 			   sizeof(actions[n].conf));
3340 		shared[copied_n].index = n;
3341 		copied_n++;
3342 	}
3343 	n++;
3344 	*shared_n = copied_n;
3345 	if (!copied_n)
3346 		return 0;
3347 	actions_size = sizeof(struct rte_flow_action) * n;
3348 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3349 	if (!translated) {
3350 		rte_errno = ENOMEM;
3351 		return -ENOMEM;
3352 	}
3353 	memcpy(translated, actions, actions_size);
3354 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3355 		const struct rte_flow_shared_action *shared_action;
3356 
3357 		shared_action = shared->action;
3358 		switch (shared_action->type) {
3359 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3360 			translated[shared->index].type =
3361 				RTE_FLOW_ACTION_TYPE_RSS;
3362 			translated[shared->index].conf =
3363 				&shared_action->rss.origin;
3364 			break;
3365 		default:
3366 			mlx5_free(translated);
3367 			return rte_flow_error_set
3368 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3369 				 NULL, "invalid shared action type");
3370 		}
3371 	}
3372 	*translated_actions = translated;
3373 	return 0;
3374 }
3375 
3376 /**
3377  * Get Shared RSS action from the action list.
3378  *
3379  * @param[in] shared
3380  *   Pointer to the list of actions.
3381  * @param[in] shared_n
3382  *   Actions list length.
3383  *
3384  * @return
3385  *   Pointer to the MLX5 RSS action if exists, otherwise return NULL.
3386  */
3387 static struct mlx5_shared_action_rss *
3388 flow_get_shared_rss_action(struct mlx5_translated_shared_action *shared,
3389 			   int shared_n)
3390 {
3391 	struct mlx5_translated_shared_action *shared_end;
3392 
3393 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3394 		struct rte_flow_shared_action *shared_action;
3395 
3396 		shared_action = shared->action;
3397 		switch (shared_action->type) {
3398 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3399 			__atomic_add_fetch(&shared_action->refcnt, 1,
3400 					   __ATOMIC_RELAXED);
3401 			return &shared_action->rss;
3402 		default:
3403 			break;
3404 		}
3405 	}
3406 	return NULL;
3407 }
3408 
3409 struct rte_flow_shared_action *
3410 mlx5_flow_get_shared_rss(struct rte_flow *flow)
3411 {
3412 	if (flow->shared_rss)
3413 		return container_of(flow->shared_rss,
3414 				    struct rte_flow_shared_action, rss);
3415 	else
3416 		return NULL;
3417 }
3418 
3419 static unsigned int
3420 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3421 {
3422 	const struct rte_flow_item *item;
3423 	unsigned int has_vlan = 0;
3424 
3425 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3426 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3427 			has_vlan = 1;
3428 			break;
3429 		}
3430 	}
3431 	if (has_vlan)
3432 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3433 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3434 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3435 			       MLX5_EXPANSION_ROOT_OUTER;
3436 }
3437 
3438 /**
3439  *  Get layer flags from the prefix flow.
3440  *
3441  *  Some flows may be split to several subflows, the prefix subflow gets the
3442  *  match items and the suffix sub flow gets the actions.
3443  *  Some actions need the user defined match item flags to get the detail for
3444  *  the action.
3445  *  This function helps the suffix flow to get the item layer flags from prefix
3446  *  subflow.
3447  *
3448  * @param[in] dev_flow
3449  *   Pointer the created preifx subflow.
3450  *
3451  * @return
3452  *   The layers get from prefix subflow.
3453  */
3454 static inline uint64_t
3455 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3456 {
3457 	uint64_t layers = 0;
3458 
3459 	/*
3460 	 * Layers bits could be localization, but usually the compiler will
3461 	 * help to do the optimization work for source code.
3462 	 * If no decap actions, use the layers directly.
3463 	 */
3464 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3465 		return dev_flow->handle->layers;
3466 	/* Convert L3 layers with decap action. */
3467 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3468 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3469 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3470 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3471 	/* Convert L4 layers with decap action.  */
3472 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3473 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3474 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3475 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3476 	return layers;
3477 }
3478 
3479 /**
3480  * Get metadata split action information.
3481  *
3482  * @param[in] actions
3483  *   Pointer to the list of actions.
3484  * @param[out] qrss
3485  *   Pointer to the return pointer.
3486  * @param[out] qrss_type
3487  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3488  *   if no QUEUE/RSS is found.
3489  * @param[out] encap_idx
3490  *   Pointer to the index of the encap action if exists, otherwise the last
3491  *   action index.
3492  *
3493  * @return
3494  *   Total number of actions.
3495  */
3496 static int
3497 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3498 				       const struct rte_flow_action **qrss,
3499 				       int *encap_idx)
3500 {
3501 	const struct rte_flow_action_raw_encap *raw_encap;
3502 	int actions_n = 0;
3503 	int raw_decap_idx = -1;
3504 
3505 	*encap_idx = -1;
3506 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3507 		switch (actions->type) {
3508 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3509 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3510 			*encap_idx = actions_n;
3511 			break;
3512 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3513 			raw_decap_idx = actions_n;
3514 			break;
3515 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3516 			raw_encap = actions->conf;
3517 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3518 				*encap_idx = raw_decap_idx != -1 ?
3519 						      raw_decap_idx : actions_n;
3520 			break;
3521 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3522 		case RTE_FLOW_ACTION_TYPE_RSS:
3523 			*qrss = actions;
3524 			break;
3525 		default:
3526 			break;
3527 		}
3528 		actions_n++;
3529 	}
3530 	if (*encap_idx == -1)
3531 		*encap_idx = actions_n;
3532 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3533 	return actions_n + 1;
3534 }
3535 
3536 /**
3537  * Check meter action from the action list.
3538  *
3539  * @param[in] actions
3540  *   Pointer to the list of actions.
3541  * @param[out] mtr
3542  *   Pointer to the meter exist flag.
3543  *
3544  * @return
3545  *   Total number of actions.
3546  */
3547 static int
3548 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3549 {
3550 	int actions_n = 0;
3551 
3552 	MLX5_ASSERT(mtr);
3553 	*mtr = 0;
3554 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3555 		switch (actions->type) {
3556 		case RTE_FLOW_ACTION_TYPE_METER:
3557 			*mtr = 1;
3558 			break;
3559 		default:
3560 			break;
3561 		}
3562 		actions_n++;
3563 	}
3564 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3565 	return actions_n + 1;
3566 }
3567 
3568 /**
3569  * Check if the flow should be split due to hairpin.
3570  * The reason for the split is that in current HW we can't
3571  * support encap and push-vlan on Rx, so if a flow contains
3572  * these actions we move it to Tx.
3573  *
3574  * @param dev
3575  *   Pointer to Ethernet device.
3576  * @param[in] attr
3577  *   Flow rule attributes.
3578  * @param[in] actions
3579  *   Associated actions (list terminated by the END action).
3580  *
3581  * @return
3582  *   > 0 the number of actions and the flow should be split,
3583  *   0 when no split required.
3584  */
3585 static int
3586 flow_check_hairpin_split(struct rte_eth_dev *dev,
3587 			 const struct rte_flow_attr *attr,
3588 			 const struct rte_flow_action actions[])
3589 {
3590 	int queue_action = 0;
3591 	int action_n = 0;
3592 	int split = 0;
3593 	const struct rte_flow_action_queue *queue;
3594 	const struct rte_flow_action_rss *rss;
3595 	const struct rte_flow_action_raw_encap *raw_encap;
3596 	const struct rte_eth_hairpin_conf *conf;
3597 
3598 	if (!attr->ingress)
3599 		return 0;
3600 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3601 		switch (actions->type) {
3602 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3603 			queue = actions->conf;
3604 			if (queue == NULL)
3605 				return 0;
3606 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3607 			if (conf != NULL && !!conf->tx_explicit)
3608 				return 0;
3609 			queue_action = 1;
3610 			action_n++;
3611 			break;
3612 		case RTE_FLOW_ACTION_TYPE_RSS:
3613 			rss = actions->conf;
3614 			if (rss == NULL || rss->queue_num == 0)
3615 				return 0;
3616 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3617 			if (conf != NULL && !!conf->tx_explicit)
3618 				return 0;
3619 			queue_action = 1;
3620 			action_n++;
3621 			break;
3622 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3623 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3624 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3625 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3626 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3627 			split++;
3628 			action_n++;
3629 			break;
3630 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3631 			raw_encap = actions->conf;
3632 			if (raw_encap->size >
3633 			    (sizeof(struct rte_flow_item_eth) +
3634 			     sizeof(struct rte_flow_item_ipv4)))
3635 				split++;
3636 			action_n++;
3637 			break;
3638 		default:
3639 			action_n++;
3640 			break;
3641 		}
3642 	}
3643 	if (split && queue_action)
3644 		return action_n;
3645 	return 0;
3646 }
3647 
3648 /* Declare flow create/destroy prototype in advance. */
3649 static uint32_t
3650 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3651 		 const struct rte_flow_attr *attr,
3652 		 const struct rte_flow_item items[],
3653 		 const struct rte_flow_action actions[],
3654 		 bool external, struct rte_flow_error *error);
3655 
3656 static void
3657 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3658 		  uint32_t flow_idx);
3659 
3660 /**
3661  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3662  *
3663  * As mark_id is unique, if there's already a registered flow for the mark_id,
3664  * return by increasing the reference counter of the resource. Otherwise, create
3665  * the resource (mcp_res) and flow.
3666  *
3667  * Flow looks like,
3668  *   - If ingress port is ANY and reg_c[1] is mark_id,
3669  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3670  *
3671  * For default flow (zero mark_id), flow is like,
3672  *   - If ingress port is ANY,
3673  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3674  *
3675  * @param dev
3676  *   Pointer to Ethernet device.
3677  * @param mark_id
3678  *   ID of MARK action, zero means default flow for META.
3679  * @param[out] error
3680  *   Perform verbose error reporting if not NULL.
3681  *
3682  * @return
3683  *   Associated resource on success, NULL otherwise and rte_errno is set.
3684  */
3685 static struct mlx5_flow_mreg_copy_resource *
3686 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3687 			  struct rte_flow_error *error)
3688 {
3689 	struct mlx5_priv *priv = dev->data->dev_private;
3690 	struct rte_flow_attr attr = {
3691 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3692 		.ingress = 1,
3693 	};
3694 	struct mlx5_rte_flow_item_tag tag_spec = {
3695 		.data = mark_id,
3696 	};
3697 	struct rte_flow_item items[] = {
3698 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3699 	};
3700 	struct rte_flow_action_mark ftag = {
3701 		.id = mark_id,
3702 	};
3703 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3704 		.dst = REG_B,
3705 		.src = REG_NON,
3706 	};
3707 	struct rte_flow_action_jump jump = {
3708 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3709 	};
3710 	struct rte_flow_action actions[] = {
3711 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3712 	};
3713 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3714 	uint32_t idx = 0;
3715 	int ret;
3716 
3717 	/* Fill the register fileds in the flow. */
3718 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3719 	if (ret < 0)
3720 		return NULL;
3721 	tag_spec.id = ret;
3722 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3723 	if (ret < 0)
3724 		return NULL;
3725 	cp_mreg.src = ret;
3726 	/* Check if already registered. */
3727 	MLX5_ASSERT(priv->mreg_cp_tbl);
3728 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, NULL);
3729 	if (mcp_res) {
3730 		/* For non-default rule. */
3731 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3732 			mcp_res->refcnt++;
3733 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3734 			    mcp_res->refcnt == 1);
3735 		return mcp_res;
3736 	}
3737 	/* Provide the full width of FLAG specific value. */
3738 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3739 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3740 	/* Build a new flow. */
3741 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3742 		items[0] = (struct rte_flow_item){
3743 			.type = (enum rte_flow_item_type)
3744 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3745 			.spec = &tag_spec,
3746 		};
3747 		items[1] = (struct rte_flow_item){
3748 			.type = RTE_FLOW_ITEM_TYPE_END,
3749 		};
3750 		actions[0] = (struct rte_flow_action){
3751 			.type = (enum rte_flow_action_type)
3752 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3753 			.conf = &ftag,
3754 		};
3755 		actions[1] = (struct rte_flow_action){
3756 			.type = (enum rte_flow_action_type)
3757 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3758 			.conf = &cp_mreg,
3759 		};
3760 		actions[2] = (struct rte_flow_action){
3761 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3762 			.conf = &jump,
3763 		};
3764 		actions[3] = (struct rte_flow_action){
3765 			.type = RTE_FLOW_ACTION_TYPE_END,
3766 		};
3767 	} else {
3768 		/* Default rule, wildcard match. */
3769 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3770 		items[0] = (struct rte_flow_item){
3771 			.type = RTE_FLOW_ITEM_TYPE_END,
3772 		};
3773 		actions[0] = (struct rte_flow_action){
3774 			.type = (enum rte_flow_action_type)
3775 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3776 			.conf = &cp_mreg,
3777 		};
3778 		actions[1] = (struct rte_flow_action){
3779 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3780 			.conf = &jump,
3781 		};
3782 		actions[2] = (struct rte_flow_action){
3783 			.type = RTE_FLOW_ACTION_TYPE_END,
3784 		};
3785 	}
3786 	/* Build a new entry. */
3787 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3788 	if (!mcp_res) {
3789 		rte_errno = ENOMEM;
3790 		return NULL;
3791 	}
3792 	mcp_res->idx = idx;
3793 	/*
3794 	 * The copy Flows are not included in any list. There
3795 	 * ones are referenced from other Flows and can not
3796 	 * be applied, removed, deleted in ardbitrary order
3797 	 * by list traversing.
3798 	 */
3799 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3800 					 actions, false, error);
3801 	if (!mcp_res->rix_flow)
3802 		goto error;
3803 	mcp_res->refcnt++;
3804 	mcp_res->hlist_ent.key = mark_id;
3805 	ret = !mlx5_hlist_insert(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3806 	MLX5_ASSERT(!ret);
3807 	if (ret)
3808 		goto error;
3809 	return mcp_res;
3810 error:
3811 	if (mcp_res->rix_flow)
3812 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3813 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3814 	return NULL;
3815 }
3816 
3817 /**
3818  * Release flow in RX_CP_TBL.
3819  *
3820  * @param dev
3821  *   Pointer to Ethernet device.
3822  * @flow
3823  *   Parent flow for wich copying is provided.
3824  */
3825 static void
3826 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3827 			  struct rte_flow *flow)
3828 {
3829 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3830 	struct mlx5_priv *priv = dev->data->dev_private;
3831 
3832 	if (!flow->rix_mreg_copy)
3833 		return;
3834 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3835 				 flow->rix_mreg_copy);
3836 	if (!mcp_res || !priv->mreg_cp_tbl)
3837 		return;
3838 	/*
3839 	 * We do not check availability of metadata registers here,
3840 	 * because copy resources are not allocated in this case.
3841 	 */
3842 	if (--mcp_res->refcnt)
3843 		return;
3844 	MLX5_ASSERT(mcp_res->rix_flow);
3845 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3846 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3847 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3848 	flow->rix_mreg_copy = 0;
3849 }
3850 
3851 /**
3852  * Remove the default copy action from RX_CP_TBL.
3853  *
3854  * @param dev
3855  *   Pointer to Ethernet device.
3856  */
3857 static void
3858 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3859 {
3860 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3861 	struct mlx5_priv *priv = dev->data->dev_private;
3862 
3863 	/* Check if default flow is registered. */
3864 	if (!priv->mreg_cp_tbl)
3865 		return;
3866 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3867 					    MLX5_DEFAULT_COPY_ID, NULL);
3868 	if (!mcp_res)
3869 		return;
3870 	MLX5_ASSERT(mcp_res->rix_flow);
3871 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3872 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3873 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3874 }
3875 
3876 /**
3877  * Add the default copy action in in RX_CP_TBL.
3878  *
3879  * @param dev
3880  *   Pointer to Ethernet device.
3881  * @param[out] error
3882  *   Perform verbose error reporting if not NULL.
3883  *
3884  * @return
3885  *   0 for success, negative value otherwise and rte_errno is set.
3886  */
3887 static int
3888 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3889 				  struct rte_flow_error *error)
3890 {
3891 	struct mlx5_priv *priv = dev->data->dev_private;
3892 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3893 
3894 	/* Check whether extensive metadata feature is engaged. */
3895 	if (!priv->config.dv_flow_en ||
3896 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3897 	    !mlx5_flow_ext_mreg_supported(dev) ||
3898 	    !priv->sh->dv_regc0_mask)
3899 		return 0;
3900 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3901 	if (!mcp_res)
3902 		return -rte_errno;
3903 	return 0;
3904 }
3905 
3906 /**
3907  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3908  *
3909  * All the flow having Q/RSS action should be split by
3910  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3911  * performs the following,
3912  *   - CQE->flow_tag := reg_c[1] (MARK)
3913  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3914  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3915  * but there should be a flow per each MARK ID set by MARK action.
3916  *
3917  * For the aforementioned reason, if there's a MARK action in flow's action
3918  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3919  * the MARK ID to CQE's flow_tag like,
3920  *   - If reg_c[1] is mark_id,
3921  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3922  *
3923  * For SET_META action which stores value in reg_c[0], as the destination is
3924  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3925  * MARK ID means the default flow. The default flow looks like,
3926  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3927  *
3928  * @param dev
3929  *   Pointer to Ethernet device.
3930  * @param flow
3931  *   Pointer to flow structure.
3932  * @param[in] actions
3933  *   Pointer to the list of actions.
3934  * @param[out] error
3935  *   Perform verbose error reporting if not NULL.
3936  *
3937  * @return
3938  *   0 on success, negative value otherwise and rte_errno is set.
3939  */
3940 static int
3941 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3942 			    struct rte_flow *flow,
3943 			    const struct rte_flow_action *actions,
3944 			    struct rte_flow_error *error)
3945 {
3946 	struct mlx5_priv *priv = dev->data->dev_private;
3947 	struct mlx5_dev_config *config = &priv->config;
3948 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3949 	const struct rte_flow_action_mark *mark;
3950 
3951 	/* Check whether extensive metadata feature is engaged. */
3952 	if (!config->dv_flow_en ||
3953 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3954 	    !mlx5_flow_ext_mreg_supported(dev) ||
3955 	    !priv->sh->dv_regc0_mask)
3956 		return 0;
3957 	/* Find MARK action. */
3958 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3959 		switch (actions->type) {
3960 		case RTE_FLOW_ACTION_TYPE_FLAG:
3961 			mcp_res = flow_mreg_add_copy_action
3962 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3963 			if (!mcp_res)
3964 				return -rte_errno;
3965 			flow->rix_mreg_copy = mcp_res->idx;
3966 			return 0;
3967 		case RTE_FLOW_ACTION_TYPE_MARK:
3968 			mark = (const struct rte_flow_action_mark *)
3969 				actions->conf;
3970 			mcp_res =
3971 				flow_mreg_add_copy_action(dev, mark->id, error);
3972 			if (!mcp_res)
3973 				return -rte_errno;
3974 			flow->rix_mreg_copy = mcp_res->idx;
3975 			return 0;
3976 		default:
3977 			break;
3978 		}
3979 	}
3980 	return 0;
3981 }
3982 
3983 #define MLX5_MAX_SPLIT_ACTIONS 24
3984 #define MLX5_MAX_SPLIT_ITEMS 24
3985 
3986 /**
3987  * Split the hairpin flow.
3988  * Since HW can't support encap and push-vlan on Rx, we move these
3989  * actions to Tx.
3990  * If the count action is after the encap then we also
3991  * move the count action. in this case the count will also measure
3992  * the outer bytes.
3993  *
3994  * @param dev
3995  *   Pointer to Ethernet device.
3996  * @param[in] actions
3997  *   Associated actions (list terminated by the END action).
3998  * @param[out] actions_rx
3999  *   Rx flow actions.
4000  * @param[out] actions_tx
4001  *   Tx flow actions..
4002  * @param[out] pattern_tx
4003  *   The pattern items for the Tx flow.
4004  * @param[out] flow_id
4005  *   The flow ID connected to this flow.
4006  *
4007  * @return
4008  *   0 on success.
4009  */
4010 static int
4011 flow_hairpin_split(struct rte_eth_dev *dev,
4012 		   const struct rte_flow_action actions[],
4013 		   struct rte_flow_action actions_rx[],
4014 		   struct rte_flow_action actions_tx[],
4015 		   struct rte_flow_item pattern_tx[],
4016 		   uint32_t flow_id)
4017 {
4018 	const struct rte_flow_action_raw_encap *raw_encap;
4019 	const struct rte_flow_action_raw_decap *raw_decap;
4020 	struct mlx5_rte_flow_action_set_tag *set_tag;
4021 	struct rte_flow_action *tag_action;
4022 	struct mlx5_rte_flow_item_tag *tag_item;
4023 	struct rte_flow_item *item;
4024 	char *addr;
4025 	int encap = 0;
4026 
4027 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4028 		switch (actions->type) {
4029 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4030 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4031 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4032 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4033 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4034 			rte_memcpy(actions_tx, actions,
4035 			       sizeof(struct rte_flow_action));
4036 			actions_tx++;
4037 			break;
4038 		case RTE_FLOW_ACTION_TYPE_COUNT:
4039 			if (encap) {
4040 				rte_memcpy(actions_tx, actions,
4041 					   sizeof(struct rte_flow_action));
4042 				actions_tx++;
4043 			} else {
4044 				rte_memcpy(actions_rx, actions,
4045 					   sizeof(struct rte_flow_action));
4046 				actions_rx++;
4047 			}
4048 			break;
4049 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4050 			raw_encap = actions->conf;
4051 			if (raw_encap->size >
4052 			    (sizeof(struct rte_flow_item_eth) +
4053 			     sizeof(struct rte_flow_item_ipv4))) {
4054 				memcpy(actions_tx, actions,
4055 				       sizeof(struct rte_flow_action));
4056 				actions_tx++;
4057 				encap = 1;
4058 			} else {
4059 				rte_memcpy(actions_rx, actions,
4060 					   sizeof(struct rte_flow_action));
4061 				actions_rx++;
4062 			}
4063 			break;
4064 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4065 			raw_decap = actions->conf;
4066 			if (raw_decap->size <
4067 			    (sizeof(struct rte_flow_item_eth) +
4068 			     sizeof(struct rte_flow_item_ipv4))) {
4069 				memcpy(actions_tx, actions,
4070 				       sizeof(struct rte_flow_action));
4071 				actions_tx++;
4072 			} else {
4073 				rte_memcpy(actions_rx, actions,
4074 					   sizeof(struct rte_flow_action));
4075 				actions_rx++;
4076 			}
4077 			break;
4078 		default:
4079 			rte_memcpy(actions_rx, actions,
4080 				   sizeof(struct rte_flow_action));
4081 			actions_rx++;
4082 			break;
4083 		}
4084 	}
4085 	/* Add set meta action and end action for the Rx flow. */
4086 	tag_action = actions_rx;
4087 	tag_action->type = (enum rte_flow_action_type)
4088 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4089 	actions_rx++;
4090 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4091 	actions_rx++;
4092 	set_tag = (void *)actions_rx;
4093 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4094 	MLX5_ASSERT(set_tag->id > REG_NON);
4095 	set_tag->data = flow_id;
4096 	tag_action->conf = set_tag;
4097 	/* Create Tx item list. */
4098 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4099 	addr = (void *)&pattern_tx[2];
4100 	item = pattern_tx;
4101 	item->type = (enum rte_flow_item_type)
4102 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4103 	tag_item = (void *)addr;
4104 	tag_item->data = flow_id;
4105 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4106 	MLX5_ASSERT(set_tag->id > REG_NON);
4107 	item->spec = tag_item;
4108 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4109 	tag_item = (void *)addr;
4110 	tag_item->data = UINT32_MAX;
4111 	tag_item->id = UINT16_MAX;
4112 	item->mask = tag_item;
4113 	item->last = NULL;
4114 	item++;
4115 	item->type = RTE_FLOW_ITEM_TYPE_END;
4116 	return 0;
4117 }
4118 
4119 __extension__
4120 union tunnel_offload_mark {
4121 	uint32_t val;
4122 	struct {
4123 		uint32_t app_reserve:8;
4124 		uint32_t table_id:15;
4125 		uint32_t transfer:1;
4126 		uint32_t _unused_:8;
4127 	};
4128 };
4129 
4130 struct tunnel_default_miss_ctx {
4131 	uint16_t *queue;
4132 	__extension__
4133 	union {
4134 		struct rte_flow_action_rss action_rss;
4135 		struct rte_flow_action_queue miss_queue;
4136 		struct rte_flow_action_jump miss_jump;
4137 		uint8_t raw[0];
4138 	};
4139 };
4140 
4141 static int
4142 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4143 			     struct rte_flow *flow,
4144 			     const struct rte_flow_attr *attr,
4145 			     const struct rte_flow_action *app_actions,
4146 			     uint32_t flow_idx,
4147 			     struct tunnel_default_miss_ctx *ctx,
4148 			     struct rte_flow_error *error)
4149 {
4150 	struct mlx5_priv *priv = dev->data->dev_private;
4151 	struct mlx5_flow *dev_flow;
4152 	struct rte_flow_attr miss_attr = *attr;
4153 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4154 	const struct rte_flow_item miss_items[2] = {
4155 		{
4156 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4157 			.spec = NULL,
4158 			.last = NULL,
4159 			.mask = NULL
4160 		},
4161 		{
4162 			.type = RTE_FLOW_ITEM_TYPE_END,
4163 			.spec = NULL,
4164 			.last = NULL,
4165 			.mask = NULL
4166 		}
4167 	};
4168 	union tunnel_offload_mark mark_id;
4169 	struct rte_flow_action_mark miss_mark;
4170 	struct rte_flow_action miss_actions[3] = {
4171 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4172 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
4173 	};
4174 	const struct rte_flow_action_jump *jump_data;
4175 	uint32_t i, flow_table = 0; /* prevent compilation warning */
4176 	struct flow_grp_info grp_info = {
4177 		.external = 1,
4178 		.transfer = attr->transfer,
4179 		.fdb_def_rule = !!priv->fdb_def_rule,
4180 		.std_tbl_fix = 0,
4181 	};
4182 	int ret;
4183 
4184 	if (!attr->transfer) {
4185 		uint32_t q_size;
4186 
4187 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4188 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4189 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4190 					 0, SOCKET_ID_ANY);
4191 		if (!ctx->queue)
4192 			return rte_flow_error_set
4193 				(error, ENOMEM,
4194 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4195 				NULL, "invalid default miss RSS");
4196 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4197 		ctx->action_rss.level = 0,
4198 		ctx->action_rss.types = priv->rss_conf.rss_hf,
4199 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4200 		ctx->action_rss.queue_num = priv->reta_idx_n,
4201 		ctx->action_rss.key = priv->rss_conf.rss_key,
4202 		ctx->action_rss.queue = ctx->queue;
4203 		if (!priv->reta_idx_n || !priv->rxqs_n)
4204 			return rte_flow_error_set
4205 				(error, EINVAL,
4206 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4207 				NULL, "invalid port configuration");
4208 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4209 			ctx->action_rss.types = 0;
4210 		for (i = 0; i != priv->reta_idx_n; ++i)
4211 			ctx->queue[i] = (*priv->reta_idx)[i];
4212 	} else {
4213 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4214 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4215 	}
4216 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4217 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4218 	jump_data = app_actions->conf;
4219 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4220 	miss_attr.group = jump_data->group;
4221 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4222 				       &flow_table, grp_info, error);
4223 	if (ret)
4224 		return rte_flow_error_set(error, EINVAL,
4225 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4226 					  NULL, "invalid tunnel id");
4227 	mark_id.app_reserve = 0;
4228 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4229 	mark_id.transfer = !!attr->transfer;
4230 	mark_id._unused_ = 0;
4231 	miss_mark.id = mark_id.val;
4232 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4233 				    miss_items, miss_actions, flow_idx, error);
4234 	if (!dev_flow)
4235 		return -rte_errno;
4236 	dev_flow->flow = flow;
4237 	dev_flow->external = true;
4238 	dev_flow->tunnel = tunnel;
4239 	/* Subflow object was created, we must include one in the list. */
4240 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4241 		      dev_flow->handle, next);
4242 	DRV_LOG(DEBUG,
4243 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4244 		dev->data->port_id, tunnel->app_tunnel.type,
4245 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4246 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4247 				  miss_actions, error);
4248 	if (!ret)
4249 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4250 						  error);
4251 
4252 	return ret;
4253 }
4254 
4255 /**
4256  * The last stage of splitting chain, just creates the subflow
4257  * without any modification.
4258  *
4259  * @param[in] dev
4260  *   Pointer to Ethernet device.
4261  * @param[in] flow
4262  *   Parent flow structure pointer.
4263  * @param[in, out] sub_flow
4264  *   Pointer to return the created subflow, may be NULL.
4265  * @param[in] prefix_layers
4266  *   Prefix subflow layers, may be 0.
4267  * @param[in] prefix_mark
4268  *   Prefix subflow mark flag, may be 0.
4269  * @param[in] attr
4270  *   Flow rule attributes.
4271  * @param[in] items
4272  *   Pattern specification (list terminated by the END pattern item).
4273  * @param[in] actions
4274  *   Associated actions (list terminated by the END action).
4275  * @param[in] external
4276  *   This flow rule is created by request external to PMD.
4277  * @param[in] flow_idx
4278  *   This memory pool index to the flow.
4279  * @param[out] error
4280  *   Perform verbose error reporting if not NULL.
4281  * @return
4282  *   0 on success, negative value otherwise
4283  */
4284 static int
4285 flow_create_split_inner(struct rte_eth_dev *dev,
4286 			struct rte_flow *flow,
4287 			struct mlx5_flow **sub_flow,
4288 			uint64_t prefix_layers,
4289 			uint32_t prefix_mark,
4290 			const struct rte_flow_attr *attr,
4291 			const struct rte_flow_item items[],
4292 			const struct rte_flow_action actions[],
4293 			bool external, uint32_t flow_idx,
4294 			struct rte_flow_error *error)
4295 {
4296 	struct mlx5_flow *dev_flow;
4297 
4298 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4299 		flow_idx, error);
4300 	if (!dev_flow)
4301 		return -rte_errno;
4302 	dev_flow->flow = flow;
4303 	dev_flow->external = external;
4304 	/* Subflow object was created, we must include one in the list. */
4305 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4306 		      dev_flow->handle, next);
4307 	/*
4308 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4309 	 * flow may need some user defined item layer flags, and pass the
4310 	 * Metadate rxq mark flag to suffix flow as well.
4311 	 */
4312 	if (prefix_layers)
4313 		dev_flow->handle->layers = prefix_layers;
4314 	if (prefix_mark)
4315 		dev_flow->handle->mark = 1;
4316 	if (sub_flow)
4317 		*sub_flow = dev_flow;
4318 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4319 }
4320 
4321 /**
4322  * Split the meter flow.
4323  *
4324  * As meter flow will split to three sub flow, other than meter
4325  * action, the other actions make sense to only meter accepts
4326  * the packet. If it need to be dropped, no other additional
4327  * actions should be take.
4328  *
4329  * One kind of special action which decapsulates the L3 tunnel
4330  * header will be in the prefix sub flow, as not to take the
4331  * L3 tunnel header into account.
4332  *
4333  * @param dev
4334  *   Pointer to Ethernet device.
4335  * @param[in] items
4336  *   Pattern specification (list terminated by the END pattern item).
4337  * @param[out] sfx_items
4338  *   Suffix flow match items (list terminated by the END pattern item).
4339  * @param[in] actions
4340  *   Associated actions (list terminated by the END action).
4341  * @param[out] actions_sfx
4342  *   Suffix flow actions.
4343  * @param[out] actions_pre
4344  *   Prefix flow actions.
4345  * @param[out] pattern_sfx
4346  *   The pattern items for the suffix flow.
4347  * @param[out] tag_sfx
4348  *   Pointer to suffix flow tag.
4349  *
4350  * @return
4351  *   0 on success.
4352  */
4353 static int
4354 flow_meter_split_prep(struct rte_eth_dev *dev,
4355 		 const struct rte_flow_item items[],
4356 		 struct rte_flow_item sfx_items[],
4357 		 const struct rte_flow_action actions[],
4358 		 struct rte_flow_action actions_sfx[],
4359 		 struct rte_flow_action actions_pre[])
4360 {
4361 	struct mlx5_priv *priv = dev->data->dev_private;
4362 	struct rte_flow_action *tag_action = NULL;
4363 	struct rte_flow_item *tag_item;
4364 	struct mlx5_rte_flow_action_set_tag *set_tag;
4365 	struct rte_flow_error error;
4366 	const struct rte_flow_action_raw_encap *raw_encap;
4367 	const struct rte_flow_action_raw_decap *raw_decap;
4368 	struct mlx5_rte_flow_item_tag *tag_spec;
4369 	struct mlx5_rte_flow_item_tag *tag_mask;
4370 	uint32_t tag_id = 0;
4371 	bool copy_vlan = false;
4372 
4373 	/* Prepare the actions for prefix and suffix flow. */
4374 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4375 		struct rte_flow_action **action_cur = NULL;
4376 
4377 		switch (actions->type) {
4378 		case RTE_FLOW_ACTION_TYPE_METER:
4379 			/* Add the extra tag action first. */
4380 			tag_action = actions_pre;
4381 			tag_action->type = (enum rte_flow_action_type)
4382 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4383 			actions_pre++;
4384 			action_cur = &actions_pre;
4385 			break;
4386 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4387 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4388 			action_cur = &actions_pre;
4389 			break;
4390 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4391 			raw_encap = actions->conf;
4392 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4393 				action_cur = &actions_pre;
4394 			break;
4395 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4396 			raw_decap = actions->conf;
4397 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4398 				action_cur = &actions_pre;
4399 			break;
4400 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4401 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4402 			copy_vlan = true;
4403 			break;
4404 		default:
4405 			break;
4406 		}
4407 		if (!action_cur)
4408 			action_cur = &actions_sfx;
4409 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4410 		(*action_cur)++;
4411 	}
4412 	/* Add end action to the actions. */
4413 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4414 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4415 	actions_pre++;
4416 	/* Set the tag. */
4417 	set_tag = (void *)actions_pre;
4418 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4419 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4420 			  &tag_id);
4421 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4422 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4423 			dev->data->port_id);
4424 		mlx5_ipool_free(priv->sh->ipool
4425 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4426 		return 0;
4427 	} else if (!tag_id) {
4428 		return 0;
4429 	}
4430 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4431 	assert(tag_action);
4432 	tag_action->conf = set_tag;
4433 	/* Prepare the suffix subflow items. */
4434 	tag_item = sfx_items++;
4435 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4436 		int item_type = items->type;
4437 
4438 		switch (item_type) {
4439 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4440 			memcpy(sfx_items, items, sizeof(*sfx_items));
4441 			sfx_items++;
4442 			break;
4443 		case RTE_FLOW_ITEM_TYPE_VLAN:
4444 			if (copy_vlan) {
4445 				memcpy(sfx_items, items, sizeof(*sfx_items));
4446 				/*
4447 				 * Convert to internal match item, it is used
4448 				 * for vlan push and set vid.
4449 				 */
4450 				sfx_items->type = (enum rte_flow_item_type)
4451 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4452 				sfx_items++;
4453 			}
4454 			break;
4455 		default:
4456 			break;
4457 		}
4458 	}
4459 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4460 	sfx_items++;
4461 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4462 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4463 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4464 	tag_mask = tag_spec + 1;
4465 	tag_mask->data = 0xffffff00;
4466 	tag_item->type = (enum rte_flow_item_type)
4467 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4468 	tag_item->spec = tag_spec;
4469 	tag_item->last = NULL;
4470 	tag_item->mask = tag_mask;
4471 	return tag_id;
4472 }
4473 
4474 /**
4475  * Split action list having QUEUE/RSS for metadata register copy.
4476  *
4477  * Once Q/RSS action is detected in user's action list, the flow action
4478  * should be split in order to copy metadata registers, which will happen in
4479  * RX_CP_TBL like,
4480  *   - CQE->flow_tag := reg_c[1] (MARK)
4481  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4482  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4483  * This is because the last action of each flow must be a terminal action
4484  * (QUEUE, RSS or DROP).
4485  *
4486  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4487  * stored and kept in the mlx5_flow structure per each sub_flow.
4488  *
4489  * The Q/RSS action is replaced with,
4490  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4491  * And the following JUMP action is added at the end,
4492  *   - JUMP, to RX_CP_TBL.
4493  *
4494  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4495  * flow_create_split_metadata() routine. The flow will look like,
4496  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4497  *
4498  * @param dev
4499  *   Pointer to Ethernet device.
4500  * @param[out] split_actions
4501  *   Pointer to store split actions to jump to CP_TBL.
4502  * @param[in] actions
4503  *   Pointer to the list of original flow actions.
4504  * @param[in] qrss
4505  *   Pointer to the Q/RSS action.
4506  * @param[in] actions_n
4507  *   Number of original actions.
4508  * @param[out] error
4509  *   Perform verbose error reporting if not NULL.
4510  *
4511  * @return
4512  *   non-zero unique flow_id on success, otherwise 0 and
4513  *   error/rte_error are set.
4514  */
4515 static uint32_t
4516 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4517 			  struct rte_flow_action *split_actions,
4518 			  const struct rte_flow_action *actions,
4519 			  const struct rte_flow_action *qrss,
4520 			  int actions_n, struct rte_flow_error *error)
4521 {
4522 	struct mlx5_priv *priv = dev->data->dev_private;
4523 	struct mlx5_rte_flow_action_set_tag *set_tag;
4524 	struct rte_flow_action_jump *jump;
4525 	const int qrss_idx = qrss - actions;
4526 	uint32_t flow_id = 0;
4527 	int ret = 0;
4528 
4529 	/*
4530 	 * Given actions will be split
4531 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4532 	 * - Add jump to mreg CP_TBL.
4533 	 * As a result, there will be one more action.
4534 	 */
4535 	++actions_n;
4536 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4537 	set_tag = (void *)(split_actions + actions_n);
4538 	/*
4539 	 * If tag action is not set to void(it means we are not the meter
4540 	 * suffix flow), add the tag action. Since meter suffix flow already
4541 	 * has the tag added.
4542 	 */
4543 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4544 		/*
4545 		 * Allocate the new subflow ID. This one is unique within
4546 		 * device and not shared with representors. Otherwise,
4547 		 * we would have to resolve multi-thread access synch
4548 		 * issue. Each flow on the shared device is appended
4549 		 * with source vport identifier, so the resulting
4550 		 * flows will be unique in the shared (by master and
4551 		 * representors) domain even if they have coinciding
4552 		 * IDs.
4553 		 */
4554 		mlx5_ipool_malloc(priv->sh->ipool
4555 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4556 		if (!flow_id)
4557 			return rte_flow_error_set(error, ENOMEM,
4558 						  RTE_FLOW_ERROR_TYPE_ACTION,
4559 						  NULL, "can't allocate id "
4560 						  "for split Q/RSS subflow");
4561 		/* Internal SET_TAG action to set flow ID. */
4562 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4563 			.data = flow_id,
4564 		};
4565 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4566 		if (ret < 0)
4567 			return ret;
4568 		set_tag->id = ret;
4569 		/* Construct new actions array. */
4570 		/* Replace QUEUE/RSS action. */
4571 		split_actions[qrss_idx] = (struct rte_flow_action){
4572 			.type = (enum rte_flow_action_type)
4573 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4574 			.conf = set_tag,
4575 		};
4576 	}
4577 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4578 	jump = (void *)(set_tag + 1);
4579 	*jump = (struct rte_flow_action_jump){
4580 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4581 	};
4582 	split_actions[actions_n - 2] = (struct rte_flow_action){
4583 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4584 		.conf = jump,
4585 	};
4586 	split_actions[actions_n - 1] = (struct rte_flow_action){
4587 		.type = RTE_FLOW_ACTION_TYPE_END,
4588 	};
4589 	return flow_id;
4590 }
4591 
4592 /**
4593  * Extend the given action list for Tx metadata copy.
4594  *
4595  * Copy the given action list to the ext_actions and add flow metadata register
4596  * copy action in order to copy reg_a set by WQE to reg_c[0].
4597  *
4598  * @param[out] ext_actions
4599  *   Pointer to the extended action list.
4600  * @param[in] actions
4601  *   Pointer to the list of actions.
4602  * @param[in] actions_n
4603  *   Number of actions in the list.
4604  * @param[out] error
4605  *   Perform verbose error reporting if not NULL.
4606  * @param[in] encap_idx
4607  *   The encap action inndex.
4608  *
4609  * @return
4610  *   0 on success, negative value otherwise
4611  */
4612 static int
4613 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4614 		       struct rte_flow_action *ext_actions,
4615 		       const struct rte_flow_action *actions,
4616 		       int actions_n, struct rte_flow_error *error,
4617 		       int encap_idx)
4618 {
4619 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4620 		(struct mlx5_flow_action_copy_mreg *)
4621 			(ext_actions + actions_n + 1);
4622 	int ret;
4623 
4624 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4625 	if (ret < 0)
4626 		return ret;
4627 	cp_mreg->dst = ret;
4628 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4629 	if (ret < 0)
4630 		return ret;
4631 	cp_mreg->src = ret;
4632 	if (encap_idx != 0)
4633 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4634 	if (encap_idx == actions_n - 1) {
4635 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4636 			.type = (enum rte_flow_action_type)
4637 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4638 			.conf = cp_mreg,
4639 		};
4640 		ext_actions[actions_n] = (struct rte_flow_action){
4641 			.type = RTE_FLOW_ACTION_TYPE_END,
4642 		};
4643 	} else {
4644 		ext_actions[encap_idx] = (struct rte_flow_action){
4645 			.type = (enum rte_flow_action_type)
4646 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4647 			.conf = cp_mreg,
4648 		};
4649 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4650 				sizeof(*ext_actions) * (actions_n - encap_idx));
4651 	}
4652 	return 0;
4653 }
4654 
4655 /**
4656  * Check the match action from the action list.
4657  *
4658  * @param[in] actions
4659  *   Pointer to the list of actions.
4660  * @param[in] attr
4661  *   Flow rule attributes.
4662  * @param[in] action
4663  *   The action to be check if exist.
4664  * @param[out] match_action_pos
4665  *   Pointer to the position of the matched action if exists, otherwise is -1.
4666  * @param[out] qrss_action_pos
4667  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4668  *
4669  * @return
4670  *   > 0 the total number of actions.
4671  *   0 if not found match action in action list.
4672  */
4673 static int
4674 flow_check_match_action(const struct rte_flow_action actions[],
4675 			const struct rte_flow_attr *attr,
4676 			enum rte_flow_action_type action,
4677 			int *match_action_pos, int *qrss_action_pos)
4678 {
4679 	const struct rte_flow_action_sample *sample;
4680 	int actions_n = 0;
4681 	int jump_flag = 0;
4682 	uint32_t ratio = 0;
4683 	int sub_type = 0;
4684 	int flag = 0;
4685 
4686 	*match_action_pos = -1;
4687 	*qrss_action_pos = -1;
4688 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4689 		if (actions->type == action) {
4690 			flag = 1;
4691 			*match_action_pos = actions_n;
4692 		}
4693 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4694 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4695 			*qrss_action_pos = actions_n;
4696 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4697 			jump_flag = 1;
4698 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4699 			sample = actions->conf;
4700 			ratio = sample->ratio;
4701 			sub_type = ((const struct rte_flow_action *)
4702 					(sample->actions))->type;
4703 		}
4704 		actions_n++;
4705 	}
4706 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4707 		if (ratio == 1) {
4708 			/* JUMP Action not support for Mirroring;
4709 			 * Mirroring support multi-destination;
4710 			 */
4711 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4712 				flag = 0;
4713 		}
4714 	}
4715 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4716 	return flag ? actions_n + 1 : 0;
4717 }
4718 
4719 #define SAMPLE_SUFFIX_ITEM 2
4720 
4721 /**
4722  * Split the sample flow.
4723  *
4724  * As sample flow will split to two sub flow, sample flow with
4725  * sample action, the other actions will move to new suffix flow.
4726  *
4727  * Also add unique tag id with tag action in the sample flow,
4728  * the same tag id will be as match in the suffix flow.
4729  *
4730  * @param dev
4731  *   Pointer to Ethernet device.
4732  * @param[in] fdb_tx
4733  *   FDB egress flow flag.
4734  * @param[out] sfx_items
4735  *   Suffix flow match items (list terminated by the END pattern item).
4736  * @param[in] actions
4737  *   Associated actions (list terminated by the END action).
4738  * @param[out] actions_sfx
4739  *   Suffix flow actions.
4740  * @param[out] actions_pre
4741  *   Prefix flow actions.
4742  * @param[in] actions_n
4743  *  The total number of actions.
4744  * @param[in] sample_action_pos
4745  *   The sample action position.
4746  * @param[in] qrss_action_pos
4747  *   The Queue/RSS action position.
4748  * @param[out] error
4749  *   Perform verbose error reporting if not NULL.
4750  *
4751  * @return
4752  *   0 on success, or unique flow_id, a negative errno value
4753  *   otherwise and rte_errno is set.
4754  */
4755 static int
4756 flow_sample_split_prep(struct rte_eth_dev *dev,
4757 		       uint32_t fdb_tx,
4758 		       struct rte_flow_item sfx_items[],
4759 		       const struct rte_flow_action actions[],
4760 		       struct rte_flow_action actions_sfx[],
4761 		       struct rte_flow_action actions_pre[],
4762 		       int actions_n,
4763 		       int sample_action_pos,
4764 		       int qrss_action_pos,
4765 		       struct rte_flow_error *error)
4766 {
4767 	struct mlx5_priv *priv = dev->data->dev_private;
4768 	struct mlx5_rte_flow_action_set_tag *set_tag;
4769 	struct mlx5_rte_flow_item_tag *tag_spec;
4770 	struct mlx5_rte_flow_item_tag *tag_mask;
4771 	uint32_t tag_id = 0;
4772 	int index;
4773 	int ret;
4774 
4775 	if (sample_action_pos < 0)
4776 		return rte_flow_error_set(error, EINVAL,
4777 					  RTE_FLOW_ERROR_TYPE_ACTION,
4778 					  NULL, "invalid position of sample "
4779 					  "action in list");
4780 	if (!fdb_tx) {
4781 		/* Prepare the prefix tag action. */
4782 		set_tag = (void *)(actions_pre + actions_n + 1);
4783 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4784 		if (ret < 0)
4785 			return ret;
4786 		set_tag->id = ret;
4787 		mlx5_ipool_malloc(priv->sh->ipool
4788 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4789 		set_tag->data = tag_id;
4790 		/* Prepare the suffix subflow items. */
4791 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4792 		tag_spec->data = tag_id;
4793 		tag_spec->id = set_tag->id;
4794 		tag_mask = tag_spec + 1;
4795 		tag_mask->data = UINT32_MAX;
4796 		sfx_items[0] = (struct rte_flow_item){
4797 			.type = (enum rte_flow_item_type)
4798 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4799 			.spec = tag_spec,
4800 			.last = NULL,
4801 			.mask = tag_mask,
4802 		};
4803 		sfx_items[1] = (struct rte_flow_item){
4804 			.type = (enum rte_flow_item_type)
4805 				RTE_FLOW_ITEM_TYPE_END,
4806 		};
4807 	}
4808 	/* Prepare the actions for prefix and suffix flow. */
4809 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4810 		index = qrss_action_pos;
4811 		/* Put the preceding the Queue/RSS action into prefix flow. */
4812 		if (index != 0)
4813 			memcpy(actions_pre, actions,
4814 			       sizeof(struct rte_flow_action) * index);
4815 		/* Put others preceding the sample action into prefix flow. */
4816 		if (sample_action_pos > index + 1)
4817 			memcpy(actions_pre + index, actions + index + 1,
4818 			       sizeof(struct rte_flow_action) *
4819 			       (sample_action_pos - index - 1));
4820 		index = sample_action_pos - 1;
4821 		/* Put Queue/RSS action into Suffix flow. */
4822 		memcpy(actions_sfx, actions + qrss_action_pos,
4823 		       sizeof(struct rte_flow_action));
4824 		actions_sfx++;
4825 	} else {
4826 		index = sample_action_pos;
4827 		if (index != 0)
4828 			memcpy(actions_pre, actions,
4829 			       sizeof(struct rte_flow_action) * index);
4830 	}
4831 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4832 	if (!fdb_tx) {
4833 		actions_pre[index++] =
4834 			(struct rte_flow_action){
4835 			.type = (enum rte_flow_action_type)
4836 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4837 			.conf = set_tag,
4838 		};
4839 	}
4840 	memcpy(actions_pre + index, actions + sample_action_pos,
4841 	       sizeof(struct rte_flow_action));
4842 	index += 1;
4843 	actions_pre[index] = (struct rte_flow_action){
4844 		.type = (enum rte_flow_action_type)
4845 			RTE_FLOW_ACTION_TYPE_END,
4846 	};
4847 	/* Put the actions after sample into Suffix flow. */
4848 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4849 	       sizeof(struct rte_flow_action) *
4850 	       (actions_n - sample_action_pos - 1));
4851 	return tag_id;
4852 }
4853 
4854 /**
4855  * The splitting for metadata feature.
4856  *
4857  * - Q/RSS action on NIC Rx should be split in order to pass by
4858  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4859  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4860  *
4861  * - All the actions on NIC Tx should have a mreg copy action to
4862  *   copy reg_a from WQE to reg_c[0].
4863  *
4864  * @param dev
4865  *   Pointer to Ethernet device.
4866  * @param[in] flow
4867  *   Parent flow structure pointer.
4868  * @param[in] prefix_layers
4869  *   Prefix flow layer flags.
4870  * @param[in] prefix_mark
4871  *   Prefix subflow mark flag, may be 0.
4872  * @param[in] attr
4873  *   Flow rule attributes.
4874  * @param[in] items
4875  *   Pattern specification (list terminated by the END pattern item).
4876  * @param[in] actions
4877  *   Associated actions (list terminated by the END action).
4878  * @param[in] external
4879  *   This flow rule is created by request external to PMD.
4880  * @param[in] flow_idx
4881  *   This memory pool index to the flow.
4882  * @param[out] error
4883  *   Perform verbose error reporting if not NULL.
4884  * @return
4885  *   0 on success, negative value otherwise
4886  */
4887 static int
4888 flow_create_split_metadata(struct rte_eth_dev *dev,
4889 			   struct rte_flow *flow,
4890 			   uint64_t prefix_layers,
4891 			   uint32_t prefix_mark,
4892 			   const struct rte_flow_attr *attr,
4893 			   const struct rte_flow_item items[],
4894 			   const struct rte_flow_action actions[],
4895 			   bool external, uint32_t flow_idx,
4896 			   struct rte_flow_error *error)
4897 {
4898 	struct mlx5_priv *priv = dev->data->dev_private;
4899 	struct mlx5_dev_config *config = &priv->config;
4900 	const struct rte_flow_action *qrss = NULL;
4901 	struct rte_flow_action *ext_actions = NULL;
4902 	struct mlx5_flow *dev_flow = NULL;
4903 	uint32_t qrss_id = 0;
4904 	int mtr_sfx = 0;
4905 	size_t act_size;
4906 	int actions_n;
4907 	int encap_idx;
4908 	int ret;
4909 
4910 	/* Check whether extensive metadata feature is engaged. */
4911 	if (!config->dv_flow_en ||
4912 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4913 	    !mlx5_flow_ext_mreg_supported(dev))
4914 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
4915 					       prefix_mark, attr, items,
4916 					       actions, external, flow_idx,
4917 					       error);
4918 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4919 							   &encap_idx);
4920 	if (qrss) {
4921 		/* Exclude hairpin flows from splitting. */
4922 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4923 			const struct rte_flow_action_queue *queue;
4924 
4925 			queue = qrss->conf;
4926 			if (mlx5_rxq_get_type(dev, queue->index) ==
4927 			    MLX5_RXQ_TYPE_HAIRPIN)
4928 				qrss = NULL;
4929 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4930 			const struct rte_flow_action_rss *rss;
4931 
4932 			rss = qrss->conf;
4933 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4934 			    MLX5_RXQ_TYPE_HAIRPIN)
4935 				qrss = NULL;
4936 		}
4937 	}
4938 	if (qrss) {
4939 		/* Check if it is in meter suffix table. */
4940 		mtr_sfx = attr->group == (attr->transfer ?
4941 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4942 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4943 		/*
4944 		 * Q/RSS action on NIC Rx should be split in order to pass by
4945 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4946 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4947 		 */
4948 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4949 			   sizeof(struct rte_flow_action_set_tag) +
4950 			   sizeof(struct rte_flow_action_jump);
4951 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4952 					  SOCKET_ID_ANY);
4953 		if (!ext_actions)
4954 			return rte_flow_error_set(error, ENOMEM,
4955 						  RTE_FLOW_ERROR_TYPE_ACTION,
4956 						  NULL, "no memory to split "
4957 						  "metadata flow");
4958 		/*
4959 		 * If we are the suffix flow of meter, tag already exist.
4960 		 * Set the tag action to void.
4961 		 */
4962 		if (mtr_sfx)
4963 			ext_actions[qrss - actions].type =
4964 						RTE_FLOW_ACTION_TYPE_VOID;
4965 		else
4966 			ext_actions[qrss - actions].type =
4967 						(enum rte_flow_action_type)
4968 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4969 		/*
4970 		 * Create the new actions list with removed Q/RSS action
4971 		 * and appended set tag and jump to register copy table
4972 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4973 		 * in advance, because it is needed for set tag action.
4974 		 */
4975 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4976 						    qrss, actions_n, error);
4977 		if (!mtr_sfx && !qrss_id) {
4978 			ret = -rte_errno;
4979 			goto exit;
4980 		}
4981 	} else if (attr->egress && !attr->transfer) {
4982 		/*
4983 		 * All the actions on NIC Tx should have a metadata register
4984 		 * copy action to copy reg_a from WQE to reg_c[meta]
4985 		 */
4986 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4987 			   sizeof(struct mlx5_flow_action_copy_mreg);
4988 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4989 					  SOCKET_ID_ANY);
4990 		if (!ext_actions)
4991 			return rte_flow_error_set(error, ENOMEM,
4992 						  RTE_FLOW_ERROR_TYPE_ACTION,
4993 						  NULL, "no memory to split "
4994 						  "metadata flow");
4995 		/* Create the action list appended with copy register. */
4996 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4997 					     actions_n, error, encap_idx);
4998 		if (ret < 0)
4999 			goto exit;
5000 	}
5001 	/* Add the unmodified original or prefix subflow. */
5002 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers,
5003 				      prefix_mark, attr,
5004 				      items, ext_actions ? ext_actions :
5005 				      actions, external, flow_idx, error);
5006 	if (ret < 0)
5007 		goto exit;
5008 	MLX5_ASSERT(dev_flow);
5009 	if (qrss) {
5010 		const struct rte_flow_attr q_attr = {
5011 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5012 			.ingress = 1,
5013 		};
5014 		/* Internal PMD action to set register. */
5015 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5016 			.data = qrss_id,
5017 			.id = REG_NON,
5018 		};
5019 		struct rte_flow_item q_items[] = {
5020 			{
5021 				.type = (enum rte_flow_item_type)
5022 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5023 				.spec = &q_tag_spec,
5024 				.last = NULL,
5025 				.mask = NULL,
5026 			},
5027 			{
5028 				.type = RTE_FLOW_ITEM_TYPE_END,
5029 			},
5030 		};
5031 		struct rte_flow_action q_actions[] = {
5032 			{
5033 				.type = qrss->type,
5034 				.conf = qrss->conf,
5035 			},
5036 			{
5037 				.type = RTE_FLOW_ACTION_TYPE_END,
5038 			},
5039 		};
5040 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5041 
5042 		/*
5043 		 * Configure the tag item only if there is no meter subflow.
5044 		 * Since tag is already marked in the meter suffix subflow
5045 		 * we can just use the meter suffix items as is.
5046 		 */
5047 		if (qrss_id) {
5048 			/* Not meter subflow. */
5049 			MLX5_ASSERT(!mtr_sfx);
5050 			/*
5051 			 * Put unique id in prefix flow due to it is destroyed
5052 			 * after suffix flow and id will be freed after there
5053 			 * is no actual flows with this id and identifier
5054 			 * reallocation becomes possible (for example, for
5055 			 * other flows in other threads).
5056 			 */
5057 			dev_flow->handle->split_flow_id = qrss_id;
5058 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5059 						   error);
5060 			if (ret < 0)
5061 				goto exit;
5062 			q_tag_spec.id = ret;
5063 		}
5064 		dev_flow = NULL;
5065 		/* Add suffix subflow to execute Q/RSS. */
5066 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers, 0,
5067 					      &q_attr, mtr_sfx ? items :
5068 					      q_items, q_actions,
5069 					      external, flow_idx, error);
5070 		if (ret < 0)
5071 			goto exit;
5072 		/* qrss ID should be freed if failed. */
5073 		qrss_id = 0;
5074 		MLX5_ASSERT(dev_flow);
5075 	}
5076 
5077 exit:
5078 	/*
5079 	 * We do not destroy the partially created sub_flows in case of error.
5080 	 * These ones are included into parent flow list and will be destroyed
5081 	 * by flow_drv_destroy.
5082 	 */
5083 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5084 			qrss_id);
5085 	mlx5_free(ext_actions);
5086 	return ret;
5087 }
5088 
5089 /**
5090  * The splitting for meter feature.
5091  *
5092  * - The meter flow will be split to two flows as prefix and
5093  *   suffix flow. The packets make sense only it pass the prefix
5094  *   meter action.
5095  *
5096  * - Reg_C_5 is used for the packet to match betweend prefix and
5097  *   suffix flow.
5098  *
5099  * @param dev
5100  *   Pointer to Ethernet device.
5101  * @param[in] flow
5102  *   Parent flow structure pointer.
5103  * @param[in] prefix_layers
5104  *   Prefix subflow layers, may be 0.
5105  * @param[in] prefix_mark
5106  *   Prefix subflow mark flag, may be 0.
5107  * @param[in] attr
5108  *   Flow rule attributes.
5109  * @param[in] items
5110  *   Pattern specification (list terminated by the END pattern item).
5111  * @param[in] actions
5112  *   Associated actions (list terminated by the END action).
5113  * @param[in] external
5114  *   This flow rule is created by request external to PMD.
5115  * @param[in] flow_idx
5116  *   This memory pool index to the flow.
5117  * @param[out] error
5118  *   Perform verbose error reporting if not NULL.
5119  * @return
5120  *   0 on success, negative value otherwise
5121  */
5122 static int
5123 flow_create_split_meter(struct rte_eth_dev *dev,
5124 			struct rte_flow *flow,
5125 			uint64_t prefix_layers,
5126 			uint32_t prefix_mark,
5127 			const struct rte_flow_attr *attr,
5128 			const struct rte_flow_item items[],
5129 			const struct rte_flow_action actions[],
5130 			bool external, uint32_t flow_idx,
5131 			struct rte_flow_error *error)
5132 {
5133 	struct mlx5_priv *priv = dev->data->dev_private;
5134 	struct rte_flow_action *sfx_actions = NULL;
5135 	struct rte_flow_action *pre_actions = NULL;
5136 	struct rte_flow_item *sfx_items = NULL;
5137 	struct mlx5_flow *dev_flow = NULL;
5138 	struct rte_flow_attr sfx_attr = *attr;
5139 	uint32_t mtr = 0;
5140 	uint32_t mtr_tag_id = 0;
5141 	size_t act_size;
5142 	size_t item_size;
5143 	int actions_n = 0;
5144 	int ret;
5145 
5146 	if (priv->mtr_en)
5147 		actions_n = flow_check_meter_action(actions, &mtr);
5148 	if (mtr) {
5149 		/* The five prefix actions: meter, decap, encap, tag, end. */
5150 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5151 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5152 		/* tag, vlan, port id, end. */
5153 #define METER_SUFFIX_ITEM 4
5154 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5155 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5156 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5157 					  0, SOCKET_ID_ANY);
5158 		if (!sfx_actions)
5159 			return rte_flow_error_set(error, ENOMEM,
5160 						  RTE_FLOW_ERROR_TYPE_ACTION,
5161 						  NULL, "no memory to split "
5162 						  "meter flow");
5163 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5164 			     act_size);
5165 		pre_actions = sfx_actions + actions_n;
5166 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5167 						   actions, sfx_actions,
5168 						   pre_actions);
5169 		if (!mtr_tag_id) {
5170 			ret = -rte_errno;
5171 			goto exit;
5172 		}
5173 		/* Add the prefix subflow. */
5174 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5175 					      prefix_layers, 0,
5176 					      attr, items,
5177 					      pre_actions, external,
5178 					      flow_idx, error);
5179 		if (ret) {
5180 			ret = -rte_errno;
5181 			goto exit;
5182 		}
5183 		dev_flow->handle->split_flow_id = mtr_tag_id;
5184 		/* Setting the sfx group atrr. */
5185 		sfx_attr.group = sfx_attr.transfer ?
5186 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5187 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5188 	}
5189 	/* Add the prefix subflow. */
5190 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
5191 					 flow_get_prefix_layer_flags(dev_flow) :
5192 					 prefix_layers, dev_flow ?
5193 					 dev_flow->handle->mark : prefix_mark,
5194 					 &sfx_attr, sfx_items ?
5195 					 sfx_items : items,
5196 					 sfx_actions ? sfx_actions : actions,
5197 					 external, flow_idx, error);
5198 exit:
5199 	if (sfx_actions)
5200 		mlx5_free(sfx_actions);
5201 	return ret;
5202 }
5203 
5204 /**
5205  * The splitting for sample feature.
5206  *
5207  * Once Sample action is detected in the action list, the flow actions should
5208  * be split into prefix sub flow and suffix sub flow.
5209  *
5210  * The original items remain in the prefix sub flow, all actions preceding the
5211  * sample action and the sample action itself will be copied to the prefix
5212  * sub flow, the actions following the sample action will be copied to the
5213  * suffix sub flow, Queue action always be located in the suffix sub flow.
5214  *
5215  * In order to make the packet from prefix sub flow matches with suffix sub
5216  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5217  * flow uses tag item with the unique flow id.
5218  *
5219  * @param dev
5220  *   Pointer to Ethernet device.
5221  * @param[in] flow
5222  *   Parent flow structure pointer.
5223  * @param[in] attr
5224  *   Flow rule attributes.
5225  * @param[in] items
5226  *   Pattern specification (list terminated by the END pattern item).
5227  * @param[in] actions
5228  *   Associated actions (list terminated by the END action).
5229  * @param[in] external
5230  *   This flow rule is created by request external to PMD.
5231  * @param[in] flow_idx
5232  *   This memory pool index to the flow.
5233  * @param[out] error
5234  *   Perform verbose error reporting if not NULL.
5235  * @return
5236  *   0 on success, negative value otherwise
5237  */
5238 static int
5239 flow_create_split_sample(struct rte_eth_dev *dev,
5240 			 struct rte_flow *flow,
5241 			 const struct rte_flow_attr *attr,
5242 			 const struct rte_flow_item items[],
5243 			 const struct rte_flow_action actions[],
5244 			 bool external, uint32_t flow_idx,
5245 			 struct rte_flow_error *error)
5246 {
5247 	struct mlx5_priv *priv = dev->data->dev_private;
5248 	struct rte_flow_action *sfx_actions = NULL;
5249 	struct rte_flow_action *pre_actions = NULL;
5250 	struct rte_flow_item *sfx_items = NULL;
5251 	struct mlx5_flow *dev_flow = NULL;
5252 	struct rte_flow_attr sfx_attr = *attr;
5253 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5254 	struct mlx5_flow_dv_sample_resource *sample_res;
5255 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5256 	struct mlx5_flow_tbl_resource *sfx_tbl;
5257 	union mlx5_flow_tbl_key sfx_table_key;
5258 #endif
5259 	size_t act_size;
5260 	size_t item_size;
5261 	uint32_t fdb_tx = 0;
5262 	int32_t tag_id = 0;
5263 	int actions_n = 0;
5264 	int sample_action_pos;
5265 	int qrss_action_pos;
5266 	int ret = 0;
5267 
5268 	if (priv->sampler_en)
5269 		actions_n = flow_check_match_action(actions, attr,
5270 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5271 					&sample_action_pos, &qrss_action_pos);
5272 	if (actions_n) {
5273 		/* The prefix actions must includes sample, tag, end. */
5274 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5275 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5276 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5277 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5278 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5279 					  item_size), 0, SOCKET_ID_ANY);
5280 		if (!sfx_actions)
5281 			return rte_flow_error_set(error, ENOMEM,
5282 						  RTE_FLOW_ERROR_TYPE_ACTION,
5283 						  NULL, "no memory to split "
5284 						  "sample flow");
5285 		/* The representor_id is -1 for uplink. */
5286 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5287 		if (!fdb_tx)
5288 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5289 					+ act_size);
5290 		pre_actions = sfx_actions + actions_n;
5291 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5292 						actions, sfx_actions,
5293 						pre_actions, actions_n,
5294 						sample_action_pos,
5295 						qrss_action_pos, error);
5296 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5297 			ret = -rte_errno;
5298 			goto exit;
5299 		}
5300 		/* Add the prefix subflow. */
5301 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, 0, attr,
5302 					      items, pre_actions, external,
5303 					      flow_idx, error);
5304 		if (ret) {
5305 			ret = -rte_errno;
5306 			goto exit;
5307 		}
5308 		dev_flow->handle->split_flow_id = tag_id;
5309 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5310 		/* Set the sfx group attr. */
5311 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5312 					dev_flow->dv.sample_res;
5313 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5314 					sample_res->normal_path_tbl;
5315 		sfx_tbl_data = container_of(sfx_tbl,
5316 					struct mlx5_flow_tbl_data_entry, tbl);
5317 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5318 		sfx_attr.group = sfx_attr.transfer ?
5319 					(sfx_table_key.table_id - 1) :
5320 					 sfx_table_key.table_id;
5321 #endif
5322 	}
5323 	/* Add the suffix subflow. */
5324 	ret = flow_create_split_meter(dev, flow, dev_flow ?
5325 				 flow_get_prefix_layer_flags(dev_flow) : 0,
5326 				 dev_flow ? dev_flow->handle->mark : 0,
5327 				 &sfx_attr, sfx_items ? sfx_items : items,
5328 				 sfx_actions ? sfx_actions : actions,
5329 				 external, flow_idx, error);
5330 exit:
5331 	if (sfx_actions)
5332 		mlx5_free(sfx_actions);
5333 	return ret;
5334 }
5335 
5336 /**
5337  * Split the flow to subflow set. The splitters might be linked
5338  * in the chain, like this:
5339  * flow_create_split_outer() calls:
5340  *   flow_create_split_meter() calls:
5341  *     flow_create_split_metadata(meter_subflow_0) calls:
5342  *       flow_create_split_inner(metadata_subflow_0)
5343  *       flow_create_split_inner(metadata_subflow_1)
5344  *       flow_create_split_inner(metadata_subflow_2)
5345  *     flow_create_split_metadata(meter_subflow_1) calls:
5346  *       flow_create_split_inner(metadata_subflow_0)
5347  *       flow_create_split_inner(metadata_subflow_1)
5348  *       flow_create_split_inner(metadata_subflow_2)
5349  *
5350  * This provide flexible way to add new levels of flow splitting.
5351  * The all of successfully created subflows are included to the
5352  * parent flow dev_flow list.
5353  *
5354  * @param dev
5355  *   Pointer to Ethernet device.
5356  * @param[in] flow
5357  *   Parent flow structure pointer.
5358  * @param[in] attr
5359  *   Flow rule attributes.
5360  * @param[in] items
5361  *   Pattern specification (list terminated by the END pattern item).
5362  * @param[in] actions
5363  *   Associated actions (list terminated by the END action).
5364  * @param[in] external
5365  *   This flow rule is created by request external to PMD.
5366  * @param[in] flow_idx
5367  *   This memory pool index to the flow.
5368  * @param[out] error
5369  *   Perform verbose error reporting if not NULL.
5370  * @return
5371  *   0 on success, negative value otherwise
5372  */
5373 static int
5374 flow_create_split_outer(struct rte_eth_dev *dev,
5375 			struct rte_flow *flow,
5376 			const struct rte_flow_attr *attr,
5377 			const struct rte_flow_item items[],
5378 			const struct rte_flow_action actions[],
5379 			bool external, uint32_t flow_idx,
5380 			struct rte_flow_error *error)
5381 {
5382 	int ret;
5383 
5384 	ret = flow_create_split_sample(dev, flow, attr, items,
5385 				       actions, external, flow_idx, error);
5386 	MLX5_ASSERT(ret <= 0);
5387 	return ret;
5388 }
5389 
5390 static struct mlx5_flow_tunnel *
5391 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5392 		      const struct rte_flow_attr *attr,
5393 		      const struct rte_flow_item items[],
5394 		      const struct rte_flow_action actions[])
5395 {
5396 	struct mlx5_flow_tunnel *tunnel;
5397 
5398 #pragma GCC diagnostic push
5399 #pragma GCC diagnostic ignored "-Wcast-qual"
5400 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5401 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5402 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5403 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5404 	else
5405 		tunnel = NULL;
5406 #pragma GCC diagnostic pop
5407 
5408 	return tunnel;
5409 }
5410 
5411 /**
5412  * Adjust flow RSS workspace if needed.
5413  *
5414  * @param wks
5415  *   Pointer to thread flow work space.
5416  * @param rss_desc
5417  *   Pointer to RSS descriptor.
5418  * @param[in] nrssq_num
5419  *   New RSS queue number.
5420  *
5421  * @return
5422  *   0 on success, -1 otherwise and rte_errno is set.
5423  */
5424 static int
5425 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5426 			  struct mlx5_flow_rss_desc *rss_desc,
5427 			  uint32_t nrssq_num)
5428 {
5429 	bool fidx = !!wks->flow_idx;
5430 
5431 	if (likely(nrssq_num <= wks->rssq_num[fidx]))
5432 		return 0;
5433 	rss_desc->queue = realloc(rss_desc->queue,
5434 			  sizeof(rss_desc->queue[0]) * RTE_ALIGN(nrssq_num, 2));
5435 	if (!rss_desc->queue) {
5436 		rte_errno = ENOMEM;
5437 		return -1;
5438 	}
5439 	wks->rssq_num[fidx] = RTE_ALIGN(nrssq_num, 2);
5440 	return 0;
5441 }
5442 
5443 /**
5444  * Create a flow and add it to @p list.
5445  *
5446  * @param dev
5447  *   Pointer to Ethernet device.
5448  * @param list
5449  *   Pointer to a TAILQ flow list. If this parameter NULL,
5450  *   no list insertion occurred, flow is just created,
5451  *   this is caller's responsibility to track the
5452  *   created flow.
5453  * @param[in] attr
5454  *   Flow rule attributes.
5455  * @param[in] items
5456  *   Pattern specification (list terminated by the END pattern item).
5457  * @param[in] actions
5458  *   Associated actions (list terminated by the END action).
5459  * @param[in] external
5460  *   This flow rule is created by request external to PMD.
5461  * @param[out] error
5462  *   Perform verbose error reporting if not NULL.
5463  *
5464  * @return
5465  *   A flow index on success, 0 otherwise and rte_errno is set.
5466  */
5467 static uint32_t
5468 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5469 		 const struct rte_flow_attr *attr,
5470 		 const struct rte_flow_item items[],
5471 		 const struct rte_flow_action original_actions[],
5472 		 bool external, struct rte_flow_error *error)
5473 {
5474 	struct mlx5_priv *priv = dev->data->dev_private;
5475 	struct rte_flow *flow = NULL;
5476 	struct mlx5_flow *dev_flow;
5477 	const struct rte_flow_action_rss *rss;
5478 	struct mlx5_translated_shared_action
5479 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5480 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5481 	union {
5482 		struct mlx5_flow_expand_rss buf;
5483 		uint8_t buffer[2048];
5484 	} expand_buffer;
5485 	union {
5486 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5487 		uint8_t buffer[2048];
5488 	} actions_rx;
5489 	union {
5490 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5491 		uint8_t buffer[2048];
5492 	} actions_hairpin_tx;
5493 	union {
5494 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5495 		uint8_t buffer[2048];
5496 	} items_tx;
5497 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5498 	struct mlx5_flow_rss_desc *rss_desc;
5499 	const struct rte_flow_action *p_actions_rx;
5500 	uint32_t i;
5501 	uint32_t idx = 0;
5502 	int hairpin_flow;
5503 	struct rte_flow_attr attr_tx = { .priority = 0 };
5504 	struct rte_flow_attr attr_factor = {0};
5505 	const struct rte_flow_action *actions;
5506 	struct rte_flow_action *translated_actions = NULL;
5507 	struct mlx5_flow_tunnel *tunnel;
5508 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5509 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5510 	bool fidx = !!wks->flow_idx;
5511 	int ret;
5512 
5513 	MLX5_ASSERT(wks);
5514 	rss_desc = &wks->rss_desc[fidx];
5515 	ret = flow_shared_actions_translate(original_actions,
5516 					    shared_actions,
5517 					    &shared_actions_n,
5518 					    &translated_actions, error);
5519 	if (ret < 0) {
5520 		MLX5_ASSERT(translated_actions == NULL);
5521 		return 0;
5522 	}
5523 	actions = translated_actions ? translated_actions : original_actions;
5524 	memcpy((void *)&attr_factor, (const void *)attr, sizeof(*attr));
5525 	p_actions_rx = actions;
5526 	hairpin_flow = flow_check_hairpin_split(dev, &attr_factor, actions);
5527 	ret = flow_drv_validate(dev, &attr_factor, items, p_actions_rx,
5528 				external, hairpin_flow, error);
5529 	if (ret < 0)
5530 		goto error_before_hairpin_split;
5531 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5532 	if (!flow) {
5533 		rte_errno = ENOMEM;
5534 		goto error_before_hairpin_split;
5535 	}
5536 	if (hairpin_flow > 0) {
5537 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5538 			rte_errno = EINVAL;
5539 			goto error_before_hairpin_split;
5540 		}
5541 		flow_hairpin_split(dev, actions, actions_rx.actions,
5542 				   actions_hairpin_tx.actions, items_tx.items,
5543 				   idx);
5544 		p_actions_rx = actions_rx.actions;
5545 	}
5546 	flow->drv_type = flow_get_drv_type(dev, &attr_factor);
5547 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5548 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5549 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5550 	rss = flow_get_rss_action(p_actions_rx);
5551 	if (rss) {
5552 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5553 			return 0;
5554 		/*
5555 		 * The following information is required by
5556 		 * mlx5_flow_hashfields_adjust() in advance.
5557 		 */
5558 		rss_desc->level = rss->level;
5559 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5560 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5561 	}
5562 	flow->dev_handles = 0;
5563 	if (rss && rss->types) {
5564 		unsigned int graph_root;
5565 
5566 		graph_root = find_graph_root(items, rss->level);
5567 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5568 					   items, rss->types,
5569 					   mlx5_support_expansion, graph_root);
5570 		MLX5_ASSERT(ret > 0 &&
5571 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5572 	} else {
5573 		buf->entries = 1;
5574 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5575 	}
5576 	flow->shared_rss = flow_get_shared_rss_action(shared_actions,
5577 						      shared_actions_n);
5578 	/*
5579 	 * Record the start index when there is a nested call. All sub-flows
5580 	 * need to be translated before another calling.
5581 	 * No need to use ping-pong buffer to save memory here.
5582 	 */
5583 	if (fidx) {
5584 		MLX5_ASSERT(!wks->flow_nested_idx);
5585 		wks->flow_nested_idx = fidx;
5586 	}
5587 	for (i = 0; i < buf->entries; ++i) {
5588 		/*
5589 		 * The splitter may create multiple dev_flows,
5590 		 * depending on configuration. In the simplest
5591 		 * case it just creates unmodified original flow.
5592 		 */
5593 		ret = flow_create_split_outer(dev, flow, &attr_factor,
5594 					      buf->entry[i].pattern,
5595 					      p_actions_rx, external, idx,
5596 					      error);
5597 		if (ret < 0)
5598 			goto error;
5599 		if (is_flow_tunnel_steer_rule(dev, attr,
5600 					      buf->entry[i].pattern,
5601 					      p_actions_rx)) {
5602 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5603 							   p_actions_rx,
5604 							   idx,
5605 							   &default_miss_ctx,
5606 							   error);
5607 			if (ret < 0) {
5608 				mlx5_free(default_miss_ctx.queue);
5609 				goto error;
5610 			}
5611 		}
5612 	}
5613 	/* Create the tx flow. */
5614 	if (hairpin_flow) {
5615 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5616 		attr_tx.ingress = 0;
5617 		attr_tx.egress = 1;
5618 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5619 					 actions_hairpin_tx.actions,
5620 					 idx, error);
5621 		if (!dev_flow)
5622 			goto error;
5623 		dev_flow->flow = flow;
5624 		dev_flow->external = 0;
5625 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5626 			      dev_flow->handle, next);
5627 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5628 					 items_tx.items,
5629 					 actions_hairpin_tx.actions, error);
5630 		if (ret < 0)
5631 			goto error;
5632 	}
5633 	/*
5634 	 * Update the metadata register copy table. If extensive
5635 	 * metadata feature is enabled and registers are supported
5636 	 * we might create the extra rte_flow for each unique
5637 	 * MARK/FLAG action ID.
5638 	 *
5639 	 * The table is updated for ingress Flows only, because
5640 	 * the egress Flows belong to the different device and
5641 	 * copy table should be updated in peer NIC Rx domain.
5642 	 */
5643 	if (attr_factor.ingress &&
5644 	    (external || attr_factor.group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5645 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5646 		if (ret)
5647 			goto error;
5648 	}
5649 	/*
5650 	 * If the flow is external (from application) OR device is started, then
5651 	 * the flow will be applied immediately.
5652 	 */
5653 	if (external || dev->data->dev_started) {
5654 		ret = flow_drv_apply(dev, flow, error);
5655 		if (ret < 0)
5656 			goto error;
5657 	}
5658 	if (list) {
5659 		rte_spinlock_lock(&priv->flow_list_lock);
5660 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5661 			     flow, next);
5662 		rte_spinlock_unlock(&priv->flow_list_lock);
5663 	}
5664 	flow_rxq_flags_set(dev, flow);
5665 	rte_free(translated_actions);
5666 	/* Nested flow creation index recovery. */
5667 	wks->flow_idx = wks->flow_nested_idx;
5668 	if (wks->flow_nested_idx)
5669 		wks->flow_nested_idx = 0;
5670 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5671 	if (tunnel) {
5672 		flow->tunnel = 1;
5673 		flow->tunnel_id = tunnel->tunnel_id;
5674 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5675 		mlx5_free(default_miss_ctx.queue);
5676 	}
5677 	return idx;
5678 error:
5679 	MLX5_ASSERT(flow);
5680 	ret = rte_errno; /* Save rte_errno before cleanup. */
5681 	flow_mreg_del_copy_action(dev, flow);
5682 	flow_drv_destroy(dev, flow);
5683 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5684 	rte_errno = ret; /* Restore rte_errno. */
5685 	ret = rte_errno;
5686 	rte_errno = ret;
5687 	wks->flow_idx = wks->flow_nested_idx;
5688 	if (wks->flow_nested_idx)
5689 		wks->flow_nested_idx = 0;
5690 error_before_hairpin_split:
5691 	rte_free(translated_actions);
5692 	return 0;
5693 }
5694 
5695 /**
5696  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5697  * incoming packets to table 1.
5698  *
5699  * Other flow rules, requested for group n, will be created in
5700  * e-switch table n+1.
5701  * Jump action to e-switch group n will be created to group n+1.
5702  *
5703  * Used when working in switchdev mode, to utilise advantages of table 1
5704  * and above.
5705  *
5706  * @param dev
5707  *   Pointer to Ethernet device.
5708  *
5709  * @return
5710  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5711  */
5712 struct rte_flow *
5713 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5714 {
5715 	const struct rte_flow_attr attr = {
5716 		.group = 0,
5717 		.priority = 0,
5718 		.ingress = 1,
5719 		.egress = 0,
5720 		.transfer = 1,
5721 	};
5722 	const struct rte_flow_item pattern = {
5723 		.type = RTE_FLOW_ITEM_TYPE_END,
5724 	};
5725 	struct rte_flow_action_jump jump = {
5726 		.group = 1,
5727 	};
5728 	const struct rte_flow_action actions[] = {
5729 		{
5730 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5731 			.conf = &jump,
5732 		},
5733 		{
5734 			.type = RTE_FLOW_ACTION_TYPE_END,
5735 		},
5736 	};
5737 	struct mlx5_priv *priv = dev->data->dev_private;
5738 	struct rte_flow_error error;
5739 
5740 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5741 						   &attr, &pattern,
5742 						   actions, false, &error);
5743 }
5744 
5745 /**
5746  * Validate a flow supported by the NIC.
5747  *
5748  * @see rte_flow_validate()
5749  * @see rte_flow_ops
5750  */
5751 int
5752 mlx5_flow_validate(struct rte_eth_dev *dev,
5753 		   const struct rte_flow_attr *attr,
5754 		   const struct rte_flow_item items[],
5755 		   const struct rte_flow_action original_actions[],
5756 		   struct rte_flow_error *error)
5757 {
5758 	int hairpin_flow;
5759 	struct mlx5_translated_shared_action
5760 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5761 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5762 	const struct rte_flow_action *actions;
5763 	struct rte_flow_action *translated_actions = NULL;
5764 	int ret = flow_shared_actions_translate(original_actions,
5765 						shared_actions,
5766 						&shared_actions_n,
5767 						&translated_actions, error);
5768 
5769 	if (ret)
5770 		return ret;
5771 	actions = translated_actions ? translated_actions : original_actions;
5772 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5773 	ret = flow_drv_validate(dev, attr, items, actions,
5774 				true, hairpin_flow, error);
5775 	rte_free(translated_actions);
5776 	return ret;
5777 }
5778 
5779 /**
5780  * Create a flow.
5781  *
5782  * @see rte_flow_create()
5783  * @see rte_flow_ops
5784  */
5785 struct rte_flow *
5786 mlx5_flow_create(struct rte_eth_dev *dev,
5787 		 const struct rte_flow_attr *attr,
5788 		 const struct rte_flow_item items[],
5789 		 const struct rte_flow_action actions[],
5790 		 struct rte_flow_error *error)
5791 {
5792 	struct mlx5_priv *priv = dev->data->dev_private;
5793 
5794 	/*
5795 	 * If the device is not started yet, it is not allowed to created a
5796 	 * flow from application. PMD default flows and traffic control flows
5797 	 * are not affected.
5798 	 */
5799 	if (unlikely(!dev->data->dev_started)) {
5800 		DRV_LOG(DEBUG, "port %u is not started when "
5801 			"inserting a flow", dev->data->port_id);
5802 		rte_flow_error_set(error, ENODEV,
5803 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5804 				   NULL,
5805 				   "port not started");
5806 		return NULL;
5807 	}
5808 
5809 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5810 				  attr, items, actions, true, error);
5811 }
5812 
5813 /**
5814  * Destroy a flow in a list.
5815  *
5816  * @param dev
5817  *   Pointer to Ethernet device.
5818  * @param list
5819  *   Pointer to the Indexed flow list. If this parameter NULL,
5820  *   there is no flow removal from the list. Be noted that as
5821  *   flow is add to the indexed list, memory of the indexed
5822  *   list points to maybe changed as flow destroyed.
5823  * @param[in] flow_idx
5824  *   Index of flow to destroy.
5825  */
5826 static void
5827 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5828 		  uint32_t flow_idx)
5829 {
5830 	struct mlx5_priv *priv = dev->data->dev_private;
5831 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5832 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5833 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5834 
5835 	if (!flow)
5836 		return;
5837 	/*
5838 	 * Update RX queue flags only if port is started, otherwise it is
5839 	 * already clean.
5840 	 */
5841 	if (dev->data->dev_started)
5842 		flow_rxq_flags_trim(dev, flow);
5843 	flow_drv_destroy(dev, flow);
5844 	if (list) {
5845 		rte_spinlock_lock(&priv->flow_list_lock);
5846 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5847 			     flow_idx, flow, next);
5848 		rte_spinlock_unlock(&priv->flow_list_lock);
5849 	}
5850 	flow_mreg_del_copy_action(dev, flow);
5851 	if (flow->fdir) {
5852 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5853 			if (priv_fdir_flow->rix_flow == flow_idx)
5854 				break;
5855 		}
5856 		if (priv_fdir_flow) {
5857 			LIST_REMOVE(priv_fdir_flow, next);
5858 			mlx5_free(priv_fdir_flow->fdir);
5859 			mlx5_free(priv_fdir_flow);
5860 		}
5861 	}
5862 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5863 	if (flow->tunnel) {
5864 		struct mlx5_flow_tunnel *tunnel;
5865 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5866 		RTE_VERIFY(tunnel);
5867 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5868 			mlx5_flow_tunnel_free(dev, tunnel);
5869 	}
5870 }
5871 
5872 /**
5873  * Destroy all flows.
5874  *
5875  * @param dev
5876  *   Pointer to Ethernet device.
5877  * @param list
5878  *   Pointer to the Indexed flow list.
5879  * @param active
5880  *   If flushing is called avtively.
5881  */
5882 void
5883 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5884 {
5885 	uint32_t num_flushed = 0;
5886 
5887 	while (*list) {
5888 		flow_list_destroy(dev, list, *list);
5889 		num_flushed++;
5890 	}
5891 	if (active) {
5892 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5893 			dev->data->port_id, num_flushed);
5894 	}
5895 }
5896 
5897 /**
5898  * Stop all default actions for flows.
5899  *
5900  * @param dev
5901  *   Pointer to Ethernet device.
5902  */
5903 void
5904 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5905 {
5906 	flow_mreg_del_default_copy_action(dev);
5907 	flow_rxq_flags_clear(dev);
5908 }
5909 
5910 /**
5911  * Start all default actions for flows.
5912  *
5913  * @param dev
5914  *   Pointer to Ethernet device.
5915  * @return
5916  *   0 on success, a negative errno value otherwise and rte_errno is set.
5917  */
5918 int
5919 mlx5_flow_start_default(struct rte_eth_dev *dev)
5920 {
5921 	struct rte_flow_error error;
5922 
5923 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5924 	return flow_mreg_add_default_copy_action(dev, &error);
5925 }
5926 
5927 /**
5928  * Release key of thread specific flow workspace data.
5929  */
5930 static void
5931 flow_release_workspace(void *data)
5932 {
5933 	struct mlx5_flow_workspace *wks = data;
5934 
5935 	if (!wks)
5936 		return;
5937 	free(wks->rss_desc[0].queue);
5938 	free(wks->rss_desc[1].queue);
5939 	free(wks);
5940 }
5941 
5942 /**
5943  * Initialize key of thread specific flow workspace data.
5944  */
5945 static void
5946 flow_alloc_workspace(void)
5947 {
5948 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5949 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5950 }
5951 
5952 /**
5953  * Get thread specific flow workspace.
5954  *
5955  * @return pointer to thread specific flowworkspace data, NULL on error.
5956  */
5957 struct mlx5_flow_workspace*
5958 mlx5_flow_get_thread_workspace(void)
5959 {
5960 	struct mlx5_flow_workspace *data;
5961 
5962 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
5963 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
5964 		return NULL;
5965 	}
5966 	data = pthread_getspecific(key_workspace);
5967 	if (!data) {
5968 		data = calloc(1, sizeof(*data));
5969 		if (!data) {
5970 			DRV_LOG(ERR, "Failed to allocate flow workspace "
5971 				"memory.");
5972 			return NULL;
5973 		}
5974 		data->rss_desc[0].queue = calloc(1,
5975 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5976 		if (!data->rss_desc[0].queue)
5977 			goto err;
5978 		data->rss_desc[1].queue = calloc(1,
5979 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5980 		if (!data->rss_desc[1].queue)
5981 			goto err;
5982 		data->rssq_num[0] = MLX5_RSSQ_DEFAULT_NUM;
5983 		data->rssq_num[1] = MLX5_RSSQ_DEFAULT_NUM;
5984 		if (pthread_setspecific(key_workspace, data)) {
5985 			DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5986 			goto err;
5987 		}
5988 	}
5989 	return data;
5990 err:
5991 	if (data->rss_desc[0].queue)
5992 		free(data->rss_desc[0].queue);
5993 	if (data->rss_desc[1].queue)
5994 		free(data->rss_desc[1].queue);
5995 	free(data);
5996 	return NULL;
5997 }
5998 
5999 /**
6000  * Verify the flow list is empty
6001  *
6002  * @param dev
6003  *  Pointer to Ethernet device.
6004  *
6005  * @return the number of flows not released.
6006  */
6007 int
6008 mlx5_flow_verify(struct rte_eth_dev *dev)
6009 {
6010 	struct mlx5_priv *priv = dev->data->dev_private;
6011 	struct rte_flow *flow;
6012 	uint32_t idx;
6013 	int ret = 0;
6014 
6015 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6016 		      flow, next) {
6017 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6018 			dev->data->port_id, (void *)flow);
6019 		++ret;
6020 	}
6021 	return ret;
6022 }
6023 
6024 /**
6025  * Enable default hairpin egress flow.
6026  *
6027  * @param dev
6028  *   Pointer to Ethernet device.
6029  * @param queue
6030  *   The queue index.
6031  *
6032  * @return
6033  *   0 on success, a negative errno value otherwise and rte_errno is set.
6034  */
6035 int
6036 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6037 			    uint32_t queue)
6038 {
6039 	struct mlx5_priv *priv = dev->data->dev_private;
6040 	const struct rte_flow_attr attr = {
6041 		.egress = 1,
6042 		.priority = 0,
6043 	};
6044 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6045 		.queue = queue,
6046 	};
6047 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6048 		.queue = UINT32_MAX,
6049 	};
6050 	struct rte_flow_item items[] = {
6051 		{
6052 			.type = (enum rte_flow_item_type)
6053 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6054 			.spec = &queue_spec,
6055 			.last = NULL,
6056 			.mask = &queue_mask,
6057 		},
6058 		{
6059 			.type = RTE_FLOW_ITEM_TYPE_END,
6060 		},
6061 	};
6062 	struct rte_flow_action_jump jump = {
6063 		.group = MLX5_HAIRPIN_TX_TABLE,
6064 	};
6065 	struct rte_flow_action actions[2];
6066 	uint32_t flow_idx;
6067 	struct rte_flow_error error;
6068 
6069 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6070 	actions[0].conf = &jump;
6071 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6072 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6073 				&attr, items, actions, false, &error);
6074 	if (!flow_idx) {
6075 		DRV_LOG(DEBUG,
6076 			"Failed to create ctrl flow: rte_errno(%d),"
6077 			" type(%d), message(%s)",
6078 			rte_errno, error.type,
6079 			error.message ? error.message : " (no stated reason)");
6080 		return -rte_errno;
6081 	}
6082 	return 0;
6083 }
6084 
6085 /**
6086  * Enable a control flow configured from the control plane.
6087  *
6088  * @param dev
6089  *   Pointer to Ethernet device.
6090  * @param eth_spec
6091  *   An Ethernet flow spec to apply.
6092  * @param eth_mask
6093  *   An Ethernet flow mask to apply.
6094  * @param vlan_spec
6095  *   A VLAN flow spec to apply.
6096  * @param vlan_mask
6097  *   A VLAN flow mask to apply.
6098  *
6099  * @return
6100  *   0 on success, a negative errno value otherwise and rte_errno is set.
6101  */
6102 int
6103 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6104 		    struct rte_flow_item_eth *eth_spec,
6105 		    struct rte_flow_item_eth *eth_mask,
6106 		    struct rte_flow_item_vlan *vlan_spec,
6107 		    struct rte_flow_item_vlan *vlan_mask)
6108 {
6109 	struct mlx5_priv *priv = dev->data->dev_private;
6110 	const struct rte_flow_attr attr = {
6111 		.ingress = 1,
6112 		.priority = MLX5_FLOW_PRIO_RSVD,
6113 	};
6114 	struct rte_flow_item items[] = {
6115 		{
6116 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6117 			.spec = eth_spec,
6118 			.last = NULL,
6119 			.mask = eth_mask,
6120 		},
6121 		{
6122 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6123 					      RTE_FLOW_ITEM_TYPE_END,
6124 			.spec = vlan_spec,
6125 			.last = NULL,
6126 			.mask = vlan_mask,
6127 		},
6128 		{
6129 			.type = RTE_FLOW_ITEM_TYPE_END,
6130 		},
6131 	};
6132 	uint16_t queue[priv->reta_idx_n];
6133 	struct rte_flow_action_rss action_rss = {
6134 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6135 		.level = 0,
6136 		.types = priv->rss_conf.rss_hf,
6137 		.key_len = priv->rss_conf.rss_key_len,
6138 		.queue_num = priv->reta_idx_n,
6139 		.key = priv->rss_conf.rss_key,
6140 		.queue = queue,
6141 	};
6142 	struct rte_flow_action actions[] = {
6143 		{
6144 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6145 			.conf = &action_rss,
6146 		},
6147 		{
6148 			.type = RTE_FLOW_ACTION_TYPE_END,
6149 		},
6150 	};
6151 	uint32_t flow_idx;
6152 	struct rte_flow_error error;
6153 	unsigned int i;
6154 
6155 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6156 		return 0;
6157 	}
6158 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6159 		action_rss.types = 0;
6160 	for (i = 0; i != priv->reta_idx_n; ++i)
6161 		queue[i] = (*priv->reta_idx)[i];
6162 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6163 				&attr, items, actions, false, &error);
6164 	if (!flow_idx)
6165 		return -rte_errno;
6166 	return 0;
6167 }
6168 
6169 /**
6170  * Enable a flow control configured from the control plane.
6171  *
6172  * @param dev
6173  *   Pointer to Ethernet device.
6174  * @param eth_spec
6175  *   An Ethernet flow spec to apply.
6176  * @param eth_mask
6177  *   An Ethernet flow mask to apply.
6178  *
6179  * @return
6180  *   0 on success, a negative errno value otherwise and rte_errno is set.
6181  */
6182 int
6183 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6184 	       struct rte_flow_item_eth *eth_spec,
6185 	       struct rte_flow_item_eth *eth_mask)
6186 {
6187 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6188 }
6189 
6190 /**
6191  * Create default miss flow rule matching lacp traffic
6192  *
6193  * @param dev
6194  *   Pointer to Ethernet device.
6195  * @param eth_spec
6196  *   An Ethernet flow spec to apply.
6197  *
6198  * @return
6199  *   0 on success, a negative errno value otherwise and rte_errno is set.
6200  */
6201 int
6202 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6203 {
6204 	struct mlx5_priv *priv = dev->data->dev_private;
6205 	/*
6206 	 * The LACP matching is done by only using ether type since using
6207 	 * a multicast dst mac causes kernel to give low priority to this flow.
6208 	 */
6209 	static const struct rte_flow_item_eth lacp_spec = {
6210 		.type = RTE_BE16(0x8809),
6211 	};
6212 	static const struct rte_flow_item_eth lacp_mask = {
6213 		.type = 0xffff,
6214 	};
6215 	const struct rte_flow_attr attr = {
6216 		.ingress = 1,
6217 	};
6218 	struct rte_flow_item items[] = {
6219 		{
6220 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6221 			.spec = &lacp_spec,
6222 			.mask = &lacp_mask,
6223 		},
6224 		{
6225 			.type = RTE_FLOW_ITEM_TYPE_END,
6226 		},
6227 	};
6228 	struct rte_flow_action actions[] = {
6229 		{
6230 			.type = (enum rte_flow_action_type)
6231 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6232 		},
6233 		{
6234 			.type = RTE_FLOW_ACTION_TYPE_END,
6235 		},
6236 	};
6237 	struct rte_flow_error error;
6238 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6239 				&attr, items, actions, false, &error);
6240 
6241 	if (!flow_idx)
6242 		return -rte_errno;
6243 	return 0;
6244 }
6245 
6246 /**
6247  * Destroy a flow.
6248  *
6249  * @see rte_flow_destroy()
6250  * @see rte_flow_ops
6251  */
6252 int
6253 mlx5_flow_destroy(struct rte_eth_dev *dev,
6254 		  struct rte_flow *flow,
6255 		  struct rte_flow_error *error __rte_unused)
6256 {
6257 	struct mlx5_priv *priv = dev->data->dev_private;
6258 
6259 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6260 	return 0;
6261 }
6262 
6263 /**
6264  * Destroy all flows.
6265  *
6266  * @see rte_flow_flush()
6267  * @see rte_flow_ops
6268  */
6269 int
6270 mlx5_flow_flush(struct rte_eth_dev *dev,
6271 		struct rte_flow_error *error __rte_unused)
6272 {
6273 	struct mlx5_priv *priv = dev->data->dev_private;
6274 
6275 	mlx5_flow_list_flush(dev, &priv->flows, false);
6276 	return 0;
6277 }
6278 
6279 /**
6280  * Isolated mode.
6281  *
6282  * @see rte_flow_isolate()
6283  * @see rte_flow_ops
6284  */
6285 int
6286 mlx5_flow_isolate(struct rte_eth_dev *dev,
6287 		  int enable,
6288 		  struct rte_flow_error *error)
6289 {
6290 	struct mlx5_priv *priv = dev->data->dev_private;
6291 
6292 	if (dev->data->dev_started) {
6293 		rte_flow_error_set(error, EBUSY,
6294 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6295 				   NULL,
6296 				   "port must be stopped first");
6297 		return -rte_errno;
6298 	}
6299 	priv->isolated = !!enable;
6300 	if (enable)
6301 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6302 	else
6303 		dev->dev_ops = &mlx5_os_dev_ops;
6304 
6305 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6306 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6307 
6308 	return 0;
6309 }
6310 
6311 /**
6312  * Query a flow.
6313  *
6314  * @see rte_flow_query()
6315  * @see rte_flow_ops
6316  */
6317 static int
6318 flow_drv_query(struct rte_eth_dev *dev,
6319 	       uint32_t flow_idx,
6320 	       const struct rte_flow_action *actions,
6321 	       void *data,
6322 	       struct rte_flow_error *error)
6323 {
6324 	struct mlx5_priv *priv = dev->data->dev_private;
6325 	const struct mlx5_flow_driver_ops *fops;
6326 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6327 					       [MLX5_IPOOL_RTE_FLOW],
6328 					       flow_idx);
6329 	enum mlx5_flow_drv_type ftype;
6330 
6331 	if (!flow) {
6332 		return rte_flow_error_set(error, ENOENT,
6333 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6334 			  NULL,
6335 			  "invalid flow handle");
6336 	}
6337 	ftype = flow->drv_type;
6338 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6339 	fops = flow_get_drv_ops(ftype);
6340 
6341 	return fops->query(dev, flow, actions, data, error);
6342 }
6343 
6344 /**
6345  * Query a flow.
6346  *
6347  * @see rte_flow_query()
6348  * @see rte_flow_ops
6349  */
6350 int
6351 mlx5_flow_query(struct rte_eth_dev *dev,
6352 		struct rte_flow *flow,
6353 		const struct rte_flow_action *actions,
6354 		void *data,
6355 		struct rte_flow_error *error)
6356 {
6357 	int ret;
6358 
6359 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6360 			     error);
6361 	if (ret < 0)
6362 		return ret;
6363 	return 0;
6364 }
6365 
6366 /**
6367  * Convert a flow director filter to a generic flow.
6368  *
6369  * @param dev
6370  *   Pointer to Ethernet device.
6371  * @param fdir_filter
6372  *   Flow director filter to add.
6373  * @param attributes
6374  *   Generic flow parameters structure.
6375  *
6376  * @return
6377  *   0 on success, a negative errno value otherwise and rte_errno is set.
6378  */
6379 static int
6380 flow_fdir_filter_convert(struct rte_eth_dev *dev,
6381 			 const struct rte_eth_fdir_filter *fdir_filter,
6382 			 struct mlx5_fdir *attributes)
6383 {
6384 	struct mlx5_priv *priv = dev->data->dev_private;
6385 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
6386 	const struct rte_eth_fdir_masks *mask =
6387 		&dev->data->dev_conf.fdir_conf.mask;
6388 
6389 	/* Validate queue number. */
6390 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
6391 		DRV_LOG(ERR, "port %u invalid queue number %d",
6392 			dev->data->port_id, fdir_filter->action.rx_queue);
6393 		rte_errno = EINVAL;
6394 		return -rte_errno;
6395 	}
6396 	attributes->attr.ingress = 1;
6397 	attributes->items[0] = (struct rte_flow_item) {
6398 		.type = RTE_FLOW_ITEM_TYPE_ETH,
6399 		.spec = &attributes->l2,
6400 		.mask = &attributes->l2_mask,
6401 	};
6402 	switch (fdir_filter->action.behavior) {
6403 	case RTE_ETH_FDIR_ACCEPT:
6404 		attributes->actions[0] = (struct rte_flow_action){
6405 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
6406 			.conf = &attributes->queue,
6407 		};
6408 		break;
6409 	case RTE_ETH_FDIR_REJECT:
6410 		attributes->actions[0] = (struct rte_flow_action){
6411 			.type = RTE_FLOW_ACTION_TYPE_DROP,
6412 		};
6413 		break;
6414 	default:
6415 		DRV_LOG(ERR, "port %u invalid behavior %d",
6416 			dev->data->port_id,
6417 			fdir_filter->action.behavior);
6418 		rte_errno = ENOTSUP;
6419 		return -rte_errno;
6420 	}
6421 	attributes->queue.index = fdir_filter->action.rx_queue;
6422 	/* Handle L3. */
6423 	switch (fdir_filter->input.flow_type) {
6424 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6425 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6426 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6427 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
6428 			.src_addr = input->flow.ip4_flow.src_ip,
6429 			.dst_addr = input->flow.ip4_flow.dst_ip,
6430 			.time_to_live = input->flow.ip4_flow.ttl,
6431 			.type_of_service = input->flow.ip4_flow.tos,
6432 		};
6433 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
6434 			.src_addr = mask->ipv4_mask.src_ip,
6435 			.dst_addr = mask->ipv4_mask.dst_ip,
6436 			.time_to_live = mask->ipv4_mask.ttl,
6437 			.type_of_service = mask->ipv4_mask.tos,
6438 			.next_proto_id = mask->ipv4_mask.proto,
6439 		};
6440 		attributes->items[1] = (struct rte_flow_item){
6441 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
6442 			.spec = &attributes->l3,
6443 			.mask = &attributes->l3_mask,
6444 		};
6445 		break;
6446 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6447 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6448 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6449 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
6450 			.hop_limits = input->flow.ipv6_flow.hop_limits,
6451 			.proto = input->flow.ipv6_flow.proto,
6452 		};
6453 
6454 		memcpy(attributes->l3.ipv6.hdr.src_addr,
6455 		       input->flow.ipv6_flow.src_ip,
6456 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6457 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
6458 		       input->flow.ipv6_flow.dst_ip,
6459 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6460 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
6461 		       mask->ipv6_mask.src_ip,
6462 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6463 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
6464 		       mask->ipv6_mask.dst_ip,
6465 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6466 		attributes->items[1] = (struct rte_flow_item){
6467 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
6468 			.spec = &attributes->l3,
6469 			.mask = &attributes->l3_mask,
6470 		};
6471 		break;
6472 	default:
6473 		DRV_LOG(ERR, "port %u invalid flow type%d",
6474 			dev->data->port_id, fdir_filter->input.flow_type);
6475 		rte_errno = ENOTSUP;
6476 		return -rte_errno;
6477 	}
6478 	/* Handle L4. */
6479 	switch (fdir_filter->input.flow_type) {
6480 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6481 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6482 			.src_port = input->flow.udp4_flow.src_port,
6483 			.dst_port = input->flow.udp4_flow.dst_port,
6484 		};
6485 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6486 			.src_port = mask->src_port_mask,
6487 			.dst_port = mask->dst_port_mask,
6488 		};
6489 		attributes->items[2] = (struct rte_flow_item){
6490 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6491 			.spec = &attributes->l4,
6492 			.mask = &attributes->l4_mask,
6493 		};
6494 		break;
6495 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6496 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6497 			.src_port = input->flow.tcp4_flow.src_port,
6498 			.dst_port = input->flow.tcp4_flow.dst_port,
6499 		};
6500 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6501 			.src_port = mask->src_port_mask,
6502 			.dst_port = mask->dst_port_mask,
6503 		};
6504 		attributes->items[2] = (struct rte_flow_item){
6505 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6506 			.spec = &attributes->l4,
6507 			.mask = &attributes->l4_mask,
6508 		};
6509 		break;
6510 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6511 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6512 			.src_port = input->flow.udp6_flow.src_port,
6513 			.dst_port = input->flow.udp6_flow.dst_port,
6514 		};
6515 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6516 			.src_port = mask->src_port_mask,
6517 			.dst_port = mask->dst_port_mask,
6518 		};
6519 		attributes->items[2] = (struct rte_flow_item){
6520 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6521 			.spec = &attributes->l4,
6522 			.mask = &attributes->l4_mask,
6523 		};
6524 		break;
6525 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6526 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6527 			.src_port = input->flow.tcp6_flow.src_port,
6528 			.dst_port = input->flow.tcp6_flow.dst_port,
6529 		};
6530 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6531 			.src_port = mask->src_port_mask,
6532 			.dst_port = mask->dst_port_mask,
6533 		};
6534 		attributes->items[2] = (struct rte_flow_item){
6535 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6536 			.spec = &attributes->l4,
6537 			.mask = &attributes->l4_mask,
6538 		};
6539 		break;
6540 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6541 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6542 		break;
6543 	default:
6544 		DRV_LOG(ERR, "port %u invalid flow type%d",
6545 			dev->data->port_id, fdir_filter->input.flow_type);
6546 		rte_errno = ENOTSUP;
6547 		return -rte_errno;
6548 	}
6549 	return 0;
6550 }
6551 
6552 #define FLOW_FDIR_CMP(f1, f2, fld) \
6553 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
6554 
6555 /**
6556  * Compare two FDIR flows. If items and actions are identical, the two flows are
6557  * regarded as same.
6558  *
6559  * @param dev
6560  *   Pointer to Ethernet device.
6561  * @param f1
6562  *   FDIR flow to compare.
6563  * @param f2
6564  *   FDIR flow to compare.
6565  *
6566  * @return
6567  *   Zero on match, 1 otherwise.
6568  */
6569 static int
6570 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
6571 {
6572 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
6573 	    FLOW_FDIR_CMP(f1, f2, l2) ||
6574 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
6575 	    FLOW_FDIR_CMP(f1, f2, l3) ||
6576 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
6577 	    FLOW_FDIR_CMP(f1, f2, l4) ||
6578 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
6579 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
6580 		return 1;
6581 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
6582 	    FLOW_FDIR_CMP(f1, f2, queue))
6583 		return 1;
6584 	return 0;
6585 }
6586 
6587 /**
6588  * Search device flow list to find out a matched FDIR flow.
6589  *
6590  * @param dev
6591  *   Pointer to Ethernet device.
6592  * @param fdir_flow
6593  *   FDIR flow to lookup.
6594  *
6595  * @return
6596  *   Index of flow if found, 0 otherwise.
6597  */
6598 static uint32_t
6599 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
6600 {
6601 	struct mlx5_priv *priv = dev->data->dev_private;
6602 	uint32_t flow_idx = 0;
6603 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6604 
6605 	MLX5_ASSERT(fdir_flow);
6606 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6607 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
6608 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
6609 				dev->data->port_id, flow_idx);
6610 			flow_idx = priv_fdir_flow->rix_flow;
6611 			break;
6612 		}
6613 	}
6614 	return flow_idx;
6615 }
6616 
6617 /**
6618  * Add new flow director filter and store it in list.
6619  *
6620  * @param dev
6621  *   Pointer to Ethernet device.
6622  * @param fdir_filter
6623  *   Flow director filter to add.
6624  *
6625  * @return
6626  *   0 on success, a negative errno value otherwise and rte_errno is set.
6627  */
6628 static int
6629 flow_fdir_filter_add(struct rte_eth_dev *dev,
6630 		     const struct rte_eth_fdir_filter *fdir_filter)
6631 {
6632 	struct mlx5_priv *priv = dev->data->dev_private;
6633 	struct mlx5_fdir *fdir_flow;
6634 	struct rte_flow *flow;
6635 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6636 	uint32_t flow_idx;
6637 	int ret;
6638 
6639 	fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
6640 				SOCKET_ID_ANY);
6641 	if (!fdir_flow) {
6642 		rte_errno = ENOMEM;
6643 		return -rte_errno;
6644 	}
6645 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
6646 	if (ret)
6647 		goto error;
6648 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
6649 	if (flow_idx) {
6650 		rte_errno = EEXIST;
6651 		goto error;
6652 	}
6653 	priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
6654 				     sizeof(struct mlx5_fdir_flow),
6655 				     0, SOCKET_ID_ANY);
6656 	if (!priv_fdir_flow) {
6657 		rte_errno = ENOMEM;
6658 		goto error;
6659 	}
6660 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
6661 				    fdir_flow->items, fdir_flow->actions, true,
6662 				    NULL);
6663 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6664 	if (!flow)
6665 		goto error;
6666 	flow->fdir = 1;
6667 	priv_fdir_flow->fdir = fdir_flow;
6668 	priv_fdir_flow->rix_flow = flow_idx;
6669 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
6670 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
6671 		dev->data->port_id, (void *)flow);
6672 	return 0;
6673 error:
6674 	mlx5_free(priv_fdir_flow);
6675 	mlx5_free(fdir_flow);
6676 	return -rte_errno;
6677 }
6678 
6679 /**
6680  * Delete specific filter.
6681  *
6682  * @param dev
6683  *   Pointer to Ethernet device.
6684  * @param fdir_filter
6685  *   Filter to be deleted.
6686  *
6687  * @return
6688  *   0 on success, a negative errno value otherwise and rte_errno is set.
6689  */
6690 static int
6691 flow_fdir_filter_delete(struct rte_eth_dev *dev,
6692 			const struct rte_eth_fdir_filter *fdir_filter)
6693 {
6694 	struct mlx5_priv *priv = dev->data->dev_private;
6695 	uint32_t flow_idx;
6696 	struct mlx5_fdir fdir_flow = {
6697 		.attr.group = 0,
6698 	};
6699 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6700 	int ret;
6701 
6702 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
6703 	if (ret)
6704 		return -rte_errno;
6705 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6706 		/* Find the fdir in priv list */
6707 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
6708 			break;
6709 	}
6710 	if (!priv_fdir_flow)
6711 		return 0;
6712 	LIST_REMOVE(priv_fdir_flow, next);
6713 	flow_idx = priv_fdir_flow->rix_flow;
6714 	flow_list_destroy(dev, &priv->flows, flow_idx);
6715 	mlx5_free(priv_fdir_flow->fdir);
6716 	mlx5_free(priv_fdir_flow);
6717 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
6718 		dev->data->port_id, flow_idx);
6719 	return 0;
6720 }
6721 
6722 /**
6723  * Update queue for specific filter.
6724  *
6725  * @param dev
6726  *   Pointer to Ethernet device.
6727  * @param fdir_filter
6728  *   Filter to be updated.
6729  *
6730  * @return
6731  *   0 on success, a negative errno value otherwise and rte_errno is set.
6732  */
6733 static int
6734 flow_fdir_filter_update(struct rte_eth_dev *dev,
6735 			const struct rte_eth_fdir_filter *fdir_filter)
6736 {
6737 	int ret;
6738 
6739 	ret = flow_fdir_filter_delete(dev, fdir_filter);
6740 	if (ret)
6741 		return ret;
6742 	return flow_fdir_filter_add(dev, fdir_filter);
6743 }
6744 
6745 /**
6746  * Flush all filters.
6747  *
6748  * @param dev
6749  *   Pointer to Ethernet device.
6750  */
6751 static void
6752 flow_fdir_filter_flush(struct rte_eth_dev *dev)
6753 {
6754 	struct mlx5_priv *priv = dev->data->dev_private;
6755 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6756 
6757 	while (!LIST_EMPTY(&priv->fdir_flows)) {
6758 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
6759 		LIST_REMOVE(priv_fdir_flow, next);
6760 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
6761 		mlx5_free(priv_fdir_flow->fdir);
6762 		mlx5_free(priv_fdir_flow);
6763 	}
6764 }
6765 
6766 /**
6767  * Get flow director information.
6768  *
6769  * @param dev
6770  *   Pointer to Ethernet device.
6771  * @param[out] fdir_info
6772  *   Resulting flow director information.
6773  */
6774 static void
6775 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
6776 {
6777 	struct rte_eth_fdir_masks *mask =
6778 		&dev->data->dev_conf.fdir_conf.mask;
6779 
6780 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
6781 	fdir_info->guarant_spc = 0;
6782 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
6783 	fdir_info->max_flexpayload = 0;
6784 	fdir_info->flow_types_mask[0] = 0;
6785 	fdir_info->flex_payload_unit = 0;
6786 	fdir_info->max_flex_payload_segment_num = 0;
6787 	fdir_info->flex_payload_limit = 0;
6788 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
6789 }
6790 
6791 /**
6792  * Deal with flow director operations.
6793  *
6794  * @param dev
6795  *   Pointer to Ethernet device.
6796  * @param filter_op
6797  *   Operation to perform.
6798  * @param arg
6799  *   Pointer to operation-specific structure.
6800  *
6801  * @return
6802  *   0 on success, a negative errno value otherwise and rte_errno is set.
6803  */
6804 static int
6805 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
6806 		    void *arg)
6807 {
6808 	enum rte_fdir_mode fdir_mode =
6809 		dev->data->dev_conf.fdir_conf.mode;
6810 
6811 	if (filter_op == RTE_ETH_FILTER_NOP)
6812 		return 0;
6813 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
6814 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
6815 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
6816 			dev->data->port_id, fdir_mode);
6817 		rte_errno = EINVAL;
6818 		return -rte_errno;
6819 	}
6820 	switch (filter_op) {
6821 	case RTE_ETH_FILTER_ADD:
6822 		return flow_fdir_filter_add(dev, arg);
6823 	case RTE_ETH_FILTER_UPDATE:
6824 		return flow_fdir_filter_update(dev, arg);
6825 	case RTE_ETH_FILTER_DELETE:
6826 		return flow_fdir_filter_delete(dev, arg);
6827 	case RTE_ETH_FILTER_FLUSH:
6828 		flow_fdir_filter_flush(dev);
6829 		break;
6830 	case RTE_ETH_FILTER_INFO:
6831 		flow_fdir_info_get(dev, arg);
6832 		break;
6833 	default:
6834 		DRV_LOG(DEBUG, "port %u unknown operation %u",
6835 			dev->data->port_id, filter_op);
6836 		rte_errno = EINVAL;
6837 		return -rte_errno;
6838 	}
6839 	return 0;
6840 }
6841 
6842 /**
6843  * Manage filter operations.
6844  *
6845  * @param dev
6846  *   Pointer to Ethernet device structure.
6847  * @param filter_type
6848  *   Filter type.
6849  * @param filter_op
6850  *   Operation to perform.
6851  * @param arg
6852  *   Pointer to operation-specific structure.
6853  *
6854  * @return
6855  *   0 on success, a negative errno value otherwise and rte_errno is set.
6856  */
6857 int
6858 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6859 		     enum rte_filter_type filter_type,
6860 		     enum rte_filter_op filter_op,
6861 		     void *arg)
6862 {
6863 	switch (filter_type) {
6864 	case RTE_ETH_FILTER_GENERIC:
6865 		if (filter_op != RTE_ETH_FILTER_GET) {
6866 			rte_errno = EINVAL;
6867 			return -rte_errno;
6868 		}
6869 		*(const void **)arg = &mlx5_flow_ops;
6870 		return 0;
6871 	case RTE_ETH_FILTER_FDIR:
6872 		return flow_fdir_ctrl_func(dev, filter_op, arg);
6873 	default:
6874 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6875 			dev->data->port_id, filter_type);
6876 		rte_errno = ENOTSUP;
6877 		return -rte_errno;
6878 	}
6879 	return 0;
6880 }
6881 
6882 /**
6883  * Create the needed meter and suffix tables.
6884  *
6885  * @param[in] dev
6886  *   Pointer to Ethernet device.
6887  * @param[in] fm
6888  *   Pointer to the flow meter.
6889  *
6890  * @return
6891  *   Pointer to table set on success, NULL otherwise.
6892  */
6893 struct mlx5_meter_domains_infos *
6894 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6895 			  const struct mlx5_flow_meter *fm)
6896 {
6897 	const struct mlx5_flow_driver_ops *fops;
6898 
6899 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6900 	return fops->create_mtr_tbls(dev, fm);
6901 }
6902 
6903 /**
6904  * Destroy the meter table set.
6905  *
6906  * @param[in] dev
6907  *   Pointer to Ethernet device.
6908  * @param[in] tbl
6909  *   Pointer to the meter table set.
6910  *
6911  * @return
6912  *   0 on success.
6913  */
6914 int
6915 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6916 			   struct mlx5_meter_domains_infos *tbls)
6917 {
6918 	const struct mlx5_flow_driver_ops *fops;
6919 
6920 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6921 	return fops->destroy_mtr_tbls(dev, tbls);
6922 }
6923 
6924 /**
6925  * Create policer rules.
6926  *
6927  * @param[in] dev
6928  *   Pointer to Ethernet device.
6929  * @param[in] fm
6930  *   Pointer to flow meter structure.
6931  * @param[in] attr
6932  *   Pointer to flow attributes.
6933  *
6934  * @return
6935  *   0 on success, -1 otherwise.
6936  */
6937 int
6938 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6939 			       struct mlx5_flow_meter *fm,
6940 			       const struct rte_flow_attr *attr)
6941 {
6942 	const struct mlx5_flow_driver_ops *fops;
6943 
6944 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6945 	return fops->create_policer_rules(dev, fm, attr);
6946 }
6947 
6948 /**
6949  * Destroy policer rules.
6950  *
6951  * @param[in] fm
6952  *   Pointer to flow meter structure.
6953  * @param[in] attr
6954  *   Pointer to flow attributes.
6955  *
6956  * @return
6957  *   0 on success, -1 otherwise.
6958  */
6959 int
6960 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6961 				struct mlx5_flow_meter *fm,
6962 				const struct rte_flow_attr *attr)
6963 {
6964 	const struct mlx5_flow_driver_ops *fops;
6965 
6966 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6967 	return fops->destroy_policer_rules(dev, fm, attr);
6968 }
6969 
6970 /**
6971  * Allocate a counter.
6972  *
6973  * @param[in] dev
6974  *   Pointer to Ethernet device structure.
6975  *
6976  * @return
6977  *   Index to allocated counter  on success, 0 otherwise.
6978  */
6979 uint32_t
6980 mlx5_counter_alloc(struct rte_eth_dev *dev)
6981 {
6982 	const struct mlx5_flow_driver_ops *fops;
6983 	struct rte_flow_attr attr = { .transfer = 0 };
6984 
6985 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6986 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6987 		return fops->counter_alloc(dev);
6988 	}
6989 	DRV_LOG(ERR,
6990 		"port %u counter allocate is not supported.",
6991 		 dev->data->port_id);
6992 	return 0;
6993 }
6994 
6995 /**
6996  * Free a counter.
6997  *
6998  * @param[in] dev
6999  *   Pointer to Ethernet device structure.
7000  * @param[in] cnt
7001  *   Index to counter to be free.
7002  */
7003 void
7004 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7005 {
7006 	const struct mlx5_flow_driver_ops *fops;
7007 	struct rte_flow_attr attr = { .transfer = 0 };
7008 
7009 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7010 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7011 		fops->counter_free(dev, cnt);
7012 		return;
7013 	}
7014 	DRV_LOG(ERR,
7015 		"port %u counter free is not supported.",
7016 		 dev->data->port_id);
7017 }
7018 
7019 /**
7020  * Query counter statistics.
7021  *
7022  * @param[in] dev
7023  *   Pointer to Ethernet device structure.
7024  * @param[in] cnt
7025  *   Index to counter to query.
7026  * @param[in] clear
7027  *   Set to clear counter statistics.
7028  * @param[out] pkts
7029  *   The counter hits packets number to save.
7030  * @param[out] bytes
7031  *   The counter hits bytes number to save.
7032  *
7033  * @return
7034  *   0 on success, a negative errno value otherwise.
7035  */
7036 int
7037 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7038 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7039 {
7040 	const struct mlx5_flow_driver_ops *fops;
7041 	struct rte_flow_attr attr = { .transfer = 0 };
7042 
7043 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7044 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7045 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7046 	}
7047 	DRV_LOG(ERR,
7048 		"port %u counter query is not supported.",
7049 		 dev->data->port_id);
7050 	return -ENOTSUP;
7051 }
7052 
7053 /**
7054  * Allocate a new memory for the counter values wrapped by all the needed
7055  * management.
7056  *
7057  * @param[in] sh
7058  *   Pointer to mlx5_dev_ctx_shared object.
7059  *
7060  * @return
7061  *   0 on success, a negative errno value otherwise.
7062  */
7063 static int
7064 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7065 {
7066 	struct mlx5_devx_mkey_attr mkey_attr;
7067 	struct mlx5_counter_stats_mem_mng *mem_mng;
7068 	volatile struct flow_counter_stats *raw_data;
7069 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7070 	int size = (sizeof(struct flow_counter_stats) *
7071 			MLX5_COUNTERS_PER_POOL +
7072 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7073 			sizeof(struct mlx5_counter_stats_mem_mng);
7074 	size_t pgsize = rte_mem_page_size();
7075 	uint8_t *mem;
7076 	int i;
7077 
7078 	if (pgsize == (size_t)-1) {
7079 		DRV_LOG(ERR, "Failed to get mem page size");
7080 		rte_errno = ENOMEM;
7081 		return -ENOMEM;
7082 	}
7083 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7084 	if (!mem) {
7085 		rte_errno = ENOMEM;
7086 		return -ENOMEM;
7087 	}
7088 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7089 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7090 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
7091 						 IBV_ACCESS_LOCAL_WRITE);
7092 	if (!mem_mng->umem) {
7093 		rte_errno = errno;
7094 		mlx5_free(mem);
7095 		return -rte_errno;
7096 	}
7097 	mkey_attr.addr = (uintptr_t)mem;
7098 	mkey_attr.size = size;
7099 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7100 	mkey_attr.pd = sh->pdn;
7101 	mkey_attr.log_entity_size = 0;
7102 	mkey_attr.pg_access = 0;
7103 	mkey_attr.klm_array = NULL;
7104 	mkey_attr.klm_num = 0;
7105 	mkey_attr.relaxed_ordering = sh->cmng.relaxed_ordering;
7106 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7107 	if (!mem_mng->dm) {
7108 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
7109 		rte_errno = errno;
7110 		mlx5_free(mem);
7111 		return -rte_errno;
7112 	}
7113 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7114 	raw_data = (volatile struct flow_counter_stats *)mem;
7115 	for (i = 0; i < raws_n; ++i) {
7116 		mem_mng->raws[i].mem_mng = mem_mng;
7117 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7118 	}
7119 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7120 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7121 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7122 				 next);
7123 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7124 	sh->cmng.mem_mng = mem_mng;
7125 	return 0;
7126 }
7127 
7128 /**
7129  * Set the statistic memory to the new counter pool.
7130  *
7131  * @param[in] sh
7132  *   Pointer to mlx5_dev_ctx_shared object.
7133  * @param[in] pool
7134  *   Pointer to the pool to set the statistic memory.
7135  *
7136  * @return
7137  *   0 on success, a negative errno value otherwise.
7138  */
7139 static int
7140 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7141 			       struct mlx5_flow_counter_pool *pool)
7142 {
7143 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7144 	/* Resize statistic memory once used out. */
7145 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7146 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7147 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7148 		return -1;
7149 	}
7150 	rte_spinlock_lock(&pool->sl);
7151 	pool->raw = cmng->mem_mng->raws + pool->index %
7152 		    MLX5_CNT_CONTAINER_RESIZE;
7153 	rte_spinlock_unlock(&pool->sl);
7154 	pool->raw_hw = NULL;
7155 	return 0;
7156 }
7157 
7158 #define MLX5_POOL_QUERY_FREQ_US 1000000
7159 
7160 /**
7161  * Set the periodic procedure for triggering asynchronous batch queries for all
7162  * the counter pools.
7163  *
7164  * @param[in] sh
7165  *   Pointer to mlx5_dev_ctx_shared object.
7166  */
7167 void
7168 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7169 {
7170 	uint32_t pools_n, us;
7171 
7172 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7173 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7174 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7175 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7176 		sh->cmng.query_thread_on = 0;
7177 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7178 	} else {
7179 		sh->cmng.query_thread_on = 1;
7180 	}
7181 }
7182 
7183 /**
7184  * The periodic procedure for triggering asynchronous batch queries for all the
7185  * counter pools. This function is probably called by the host thread.
7186  *
7187  * @param[in] arg
7188  *   The parameter for the alarm process.
7189  */
7190 void
7191 mlx5_flow_query_alarm(void *arg)
7192 {
7193 	struct mlx5_dev_ctx_shared *sh = arg;
7194 	int ret;
7195 	uint16_t pool_index = sh->cmng.pool_index;
7196 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7197 	struct mlx5_flow_counter_pool *pool;
7198 	uint16_t n_valid;
7199 
7200 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7201 		goto set_alarm;
7202 	rte_spinlock_lock(&cmng->pool_update_sl);
7203 	pool = cmng->pools[pool_index];
7204 	n_valid = cmng->n_valid;
7205 	rte_spinlock_unlock(&cmng->pool_update_sl);
7206 	/* Set the statistic memory to the new created pool. */
7207 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7208 		goto set_alarm;
7209 	if (pool->raw_hw)
7210 		/* There is a pool query in progress. */
7211 		goto set_alarm;
7212 	pool->raw_hw =
7213 		LIST_FIRST(&sh->cmng.free_stat_raws);
7214 	if (!pool->raw_hw)
7215 		/* No free counter statistics raw memory. */
7216 		goto set_alarm;
7217 	/*
7218 	 * Identify the counters released between query trigger and query
7219 	 * handle more efficiently. The counter released in this gap period
7220 	 * should wait for a new round of query as the new arrived packets
7221 	 * will not be taken into account.
7222 	 */
7223 	pool->query_gen++;
7224 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7225 					       MLX5_COUNTERS_PER_POOL,
7226 					       NULL, NULL,
7227 					       pool->raw_hw->mem_mng->dm->id,
7228 					       (void *)(uintptr_t)
7229 					       pool->raw_hw->data,
7230 					       sh->devx_comp,
7231 					       (uint64_t)(uintptr_t)pool);
7232 	if (ret) {
7233 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7234 			" %d", pool->min_dcs->id);
7235 		pool->raw_hw = NULL;
7236 		goto set_alarm;
7237 	}
7238 	LIST_REMOVE(pool->raw_hw, next);
7239 	sh->cmng.pending_queries++;
7240 	pool_index++;
7241 	if (pool_index >= n_valid)
7242 		pool_index = 0;
7243 set_alarm:
7244 	sh->cmng.pool_index = pool_index;
7245 	mlx5_set_query_alarm(sh);
7246 }
7247 
7248 /**
7249  * Check and callback event for new aged flow in the counter pool
7250  *
7251  * @param[in] sh
7252  *   Pointer to mlx5_dev_ctx_shared object.
7253  * @param[in] pool
7254  *   Pointer to Current counter pool.
7255  */
7256 static void
7257 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7258 		   struct mlx5_flow_counter_pool *pool)
7259 {
7260 	struct mlx5_priv *priv;
7261 	struct mlx5_flow_counter *cnt;
7262 	struct mlx5_age_info *age_info;
7263 	struct mlx5_age_param *age_param;
7264 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7265 	struct mlx5_counter_stats_raw *prev = pool->raw;
7266 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7267 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7268 	uint16_t expected = AGE_CANDIDATE;
7269 	uint32_t i;
7270 
7271 	pool->time_of_last_age_check = curr_time;
7272 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7273 		cnt = MLX5_POOL_GET_CNT(pool, i);
7274 		age_param = MLX5_CNT_TO_AGE(cnt);
7275 		if (__atomic_load_n(&age_param->state,
7276 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7277 			continue;
7278 		if (cur->data[i].hits != prev->data[i].hits) {
7279 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7280 					 __ATOMIC_RELAXED);
7281 			continue;
7282 		}
7283 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7284 				       time_delta,
7285 				       __ATOMIC_RELAXED) <= age_param->timeout)
7286 			continue;
7287 		/**
7288 		 * Hold the lock first, or if between the
7289 		 * state AGE_TMOUT and tailq operation the
7290 		 * release happened, the release procedure
7291 		 * may delete a non-existent tailq node.
7292 		 */
7293 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7294 		age_info = GET_PORT_AGE_INFO(priv);
7295 		rte_spinlock_lock(&age_info->aged_sl);
7296 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7297 						AGE_TMOUT, false,
7298 						__ATOMIC_RELAXED,
7299 						__ATOMIC_RELAXED)) {
7300 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7301 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7302 		}
7303 		rte_spinlock_unlock(&age_info->aged_sl);
7304 	}
7305 	for (i = 0; i < sh->max_port; i++) {
7306 		age_info = &sh->port[i].age_info;
7307 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
7308 			continue;
7309 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
7310 			rte_eth_dev_callback_process
7311 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
7312 				RTE_ETH_EVENT_FLOW_AGED, NULL);
7313 		age_info->flags = 0;
7314 	}
7315 }
7316 
7317 /**
7318  * Handler for the HW respond about ready values from an asynchronous batch
7319  * query. This function is probably called by the host thread.
7320  *
7321  * @param[in] sh
7322  *   The pointer to the shared device context.
7323  * @param[in] async_id
7324  *   The Devx async ID.
7325  * @param[in] status
7326  *   The status of the completion.
7327  */
7328 void
7329 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7330 				  uint64_t async_id, int status)
7331 {
7332 	struct mlx5_flow_counter_pool *pool =
7333 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7334 	struct mlx5_counter_stats_raw *raw_to_free;
7335 	uint8_t query_gen = pool->query_gen ^ 1;
7336 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7337 	enum mlx5_counter_type cnt_type =
7338 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7339 				MLX5_COUNTER_TYPE_ORIGIN;
7340 
7341 	if (unlikely(status)) {
7342 		raw_to_free = pool->raw_hw;
7343 	} else {
7344 		raw_to_free = pool->raw;
7345 		if (pool->is_aged)
7346 			mlx5_flow_aging_check(sh, pool);
7347 		rte_spinlock_lock(&pool->sl);
7348 		pool->raw = pool->raw_hw;
7349 		rte_spinlock_unlock(&pool->sl);
7350 		/* Be sure the new raw counters data is updated in memory. */
7351 		rte_io_wmb();
7352 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7353 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7354 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7355 				     &pool->counters[query_gen], next);
7356 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7357 		}
7358 	}
7359 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7360 	pool->raw_hw = NULL;
7361 	sh->cmng.pending_queries--;
7362 }
7363 
7364 static const struct mlx5_flow_tbl_data_entry  *
7365 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7366 {
7367 	struct mlx5_priv *priv = dev->data->dev_private;
7368 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7369 	struct mlx5_hlist_entry *he;
7370 	union tunnel_offload_mark mbits = { .val = mark };
7371 	union mlx5_flow_tbl_key table_key = {
7372 		{
7373 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7374 			.dummy = 0,
7375 			.domain = !!mbits.transfer,
7376 			.direction = 0,
7377 		}
7378 	};
7379 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7380 	return he ?
7381 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7382 }
7383 
7384 static uint32_t
7385 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7386 				const struct mlx5_flow_tunnel *tunnel,
7387 				uint32_t group, uint32_t *table,
7388 				struct rte_flow_error *error)
7389 {
7390 	struct mlx5_priv *priv = dev->data->dev_private;
7391 	struct mlx5_hlist_entry *he;
7392 	struct tunnel_tbl_entry *tte;
7393 	union tunnel_tbl_key key = {
7394 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7395 		.group = group
7396 	};
7397 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7398 	struct mlx5_hlist *group_hash;
7399 
7400 	group_hash = tunnel ? tunnel->groups : thub->groups;
7401 	he = mlx5_hlist_lookup(group_hash, key.val, NULL);
7402 	if (!he) {
7403 		tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7404 				  sizeof(*tte), 0,
7405 				  SOCKET_ID_ANY);
7406 		if (!tte)
7407 			goto err;
7408 		tte->hash.key = key.val;
7409 		mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7410 				  &tte->flow_table);
7411 		if (tte->flow_table >= MLX5_MAX_TABLES) {
7412 			DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7413 				tte->flow_table);
7414 			mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7415 					tte->flow_table);
7416 			goto err;
7417 		} else if (!tte->flow_table) {
7418 			goto err;
7419 		}
7420 		tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7421 		mlx5_hlist_insert(group_hash, &tte->hash);
7422 	} else {
7423 		tte = container_of(he, typeof(*tte), hash);
7424 	}
7425 	*table = tte->flow_table;
7426 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7427 		dev->data->port_id, key.tunnel_id, group, *table);
7428 	return 0;
7429 
7430 err:
7431 	if (tte)
7432 		mlx5_free(tte);
7433 	return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7434 				  NULL, "tunnel group index not supported");
7435 }
7436 
7437 static int
7438 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7439 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
7440 {
7441 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7442 		if (group == UINT32_MAX)
7443 			return rte_flow_error_set
7444 						(error, EINVAL,
7445 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7446 						 NULL,
7447 						 "group index not supported");
7448 		*table = group + 1;
7449 	} else {
7450 		*table = group;
7451 	}
7452 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7453 	return 0;
7454 }
7455 
7456 /**
7457  * Translate the rte_flow group index to HW table value.
7458  *
7459  * If tunnel offload is disabled, all group ids converted to flow table
7460  * id using the standard method.
7461  * If tunnel offload is enabled, group id can be converted using the
7462  * standard or tunnel conversion method. Group conversion method
7463  * selection depends on flags in `grp_info` parameter:
7464  * - Internal (grp_info.external == 0) groups conversion uses the
7465  *   standard method.
7466  * - Group ids in JUMP action converted with the tunnel conversion.
7467  * - Group id in rule attribute conversion depends on a rule type and
7468  *   group id value:
7469  *   ** non zero group attributes converted with the tunnel method
7470  *   ** zero group attribute in non-tunnel rule is converted using the
7471  *      standard method - there's only one root table
7472  *   ** zero group attribute in steer tunnel rule is converted with the
7473  *      standard method - single root table
7474  *   ** zero group attribute in match tunnel rule is a special OvS
7475  *      case: that value is used for portability reasons. That group
7476  *      id is converted with the tunnel conversion method.
7477  *
7478  * @param[in] dev
7479  *   Port device
7480  * @param[in] tunnel
7481  *   PMD tunnel offload object
7482  * @param[in] group
7483  *   rte_flow group index value.
7484  * @param[out] table
7485  *   HW table value.
7486  * @param[in] grp_info
7487  *   flags used for conversion
7488  * @param[out] error
7489  *   Pointer to error structure.
7490  *
7491  * @return
7492  *   0 on success, a negative errno value otherwise and rte_errno is set.
7493  */
7494 int
7495 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7496 			 const struct mlx5_flow_tunnel *tunnel,
7497 			 uint32_t group, uint32_t *table,
7498 			 struct flow_grp_info grp_info,
7499 			 struct rte_flow_error *error)
7500 {
7501 	int ret;
7502 	bool standard_translation;
7503 
7504 	if (grp_info.external && group < MLX5_MAX_TABLES_EXTERNAL)
7505 		group *= MLX5_FLOW_TABLE_FACTOR;
7506 	if (is_tunnel_offload_active(dev)) {
7507 		standard_translation = !grp_info.external ||
7508 					grp_info.std_tbl_fix;
7509 	} else {
7510 		standard_translation = true;
7511 	}
7512 	DRV_LOG(DEBUG,
7513 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7514 		dev->data->port_id, group, grp_info.transfer,
7515 		grp_info.external, grp_info.fdb_def_rule,
7516 		standard_translation ? "STANDARD" : "TUNNEL");
7517 	if (standard_translation)
7518 		ret = flow_group_to_table(dev->data->port_id, group, table,
7519 					  grp_info, error);
7520 	else
7521 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7522 						      table, error);
7523 
7524 	return ret;
7525 }
7526 
7527 /**
7528  * Discover availability of metadata reg_c's.
7529  *
7530  * Iteratively use test flows to check availability.
7531  *
7532  * @param[in] dev
7533  *   Pointer to the Ethernet device structure.
7534  *
7535  * @return
7536  *   0 on success, a negative errno value otherwise and rte_errno is set.
7537  */
7538 int
7539 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7540 {
7541 	struct mlx5_priv *priv = dev->data->dev_private;
7542 	struct mlx5_dev_config *config = &priv->config;
7543 	enum modify_reg idx;
7544 	int n = 0;
7545 
7546 	/* reg_c[0] and reg_c[1] are reserved. */
7547 	config->flow_mreg_c[n++] = REG_C_0;
7548 	config->flow_mreg_c[n++] = REG_C_1;
7549 	/* Discover availability of other reg_c's. */
7550 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7551 		struct rte_flow_attr attr = {
7552 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7553 			.priority = MLX5_FLOW_PRIO_RSVD,
7554 			.ingress = 1,
7555 		};
7556 		struct rte_flow_item items[] = {
7557 			[0] = {
7558 				.type = RTE_FLOW_ITEM_TYPE_END,
7559 			},
7560 		};
7561 		struct rte_flow_action actions[] = {
7562 			[0] = {
7563 				.type = (enum rte_flow_action_type)
7564 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7565 				.conf = &(struct mlx5_flow_action_copy_mreg){
7566 					.src = REG_C_1,
7567 					.dst = idx,
7568 				},
7569 			},
7570 			[1] = {
7571 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7572 				.conf = &(struct rte_flow_action_jump){
7573 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7574 				},
7575 			},
7576 			[2] = {
7577 				.type = RTE_FLOW_ACTION_TYPE_END,
7578 			},
7579 		};
7580 		uint32_t flow_idx;
7581 		struct rte_flow *flow;
7582 		struct rte_flow_error error;
7583 
7584 		if (!config->dv_flow_en)
7585 			break;
7586 		/* Create internal flow, validation skips copy action. */
7587 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7588 					    actions, false, &error);
7589 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7590 				      flow_idx);
7591 		if (!flow)
7592 			continue;
7593 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
7594 			config->flow_mreg_c[n++] = idx;
7595 		flow_list_destroy(dev, NULL, flow_idx);
7596 	}
7597 	for (; n < MLX5_MREG_C_NUM; ++n)
7598 		config->flow_mreg_c[n] = REG_NON;
7599 	return 0;
7600 }
7601 
7602 /**
7603  * Dump flow raw hw data to file
7604  *
7605  * @param[in] dev
7606  *    The pointer to Ethernet device.
7607  * @param[in] file
7608  *   A pointer to a file for output.
7609  * @param[out] error
7610  *   Perform verbose error reporting if not NULL. PMDs initialize this
7611  *   structure in case of error only.
7612  * @return
7613  *   0 on success, a nagative value otherwise.
7614  */
7615 int
7616 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7617 		   FILE *file,
7618 		   struct rte_flow_error *error __rte_unused)
7619 {
7620 	struct mlx5_priv *priv = dev->data->dev_private;
7621 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7622 
7623 	if (!priv->config.dv_flow_en) {
7624 		if (fputs("device dv flow disabled\n", file) <= 0)
7625 			return -errno;
7626 		return -ENOTSUP;
7627 	}
7628 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7629 				       sh->tx_domain, file);
7630 }
7631 
7632 /**
7633  * Get aged-out flows.
7634  *
7635  * @param[in] dev
7636  *   Pointer to the Ethernet device structure.
7637  * @param[in] context
7638  *   The address of an array of pointers to the aged-out flows contexts.
7639  * @param[in] nb_countexts
7640  *   The length of context array pointers.
7641  * @param[out] error
7642  *   Perform verbose error reporting if not NULL. Initialized in case of
7643  *   error only.
7644  *
7645  * @return
7646  *   how many contexts get in success, otherwise negative errno value.
7647  *   if nb_contexts is 0, return the amount of all aged contexts.
7648  *   if nb_contexts is not 0 , return the amount of aged flows reported
7649  *   in the context array.
7650  */
7651 int
7652 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7653 			uint32_t nb_contexts, struct rte_flow_error *error)
7654 {
7655 	const struct mlx5_flow_driver_ops *fops;
7656 	struct rte_flow_attr attr = { .transfer = 0 };
7657 
7658 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7659 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7660 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7661 						    error);
7662 	}
7663 	DRV_LOG(ERR,
7664 		"port %u get aged flows is not supported.",
7665 		 dev->data->port_id);
7666 	return -ENOTSUP;
7667 }
7668 
7669 /* Wrapper for driver action_validate op callback */
7670 static int
7671 flow_drv_action_validate(struct rte_eth_dev *dev,
7672 			 const struct rte_flow_shared_action_conf *conf,
7673 			 const struct rte_flow_action *action,
7674 			 const struct mlx5_flow_driver_ops *fops,
7675 			 struct rte_flow_error *error)
7676 {
7677 	static const char err_msg[] = "shared action validation unsupported";
7678 
7679 	if (!fops->action_validate) {
7680 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7681 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7682 				   NULL, err_msg);
7683 		return -rte_errno;
7684 	}
7685 	return fops->action_validate(dev, conf, action, error);
7686 }
7687 
7688 /**
7689  * Destroys the shared action by handle.
7690  *
7691  * @param dev
7692  *   Pointer to Ethernet device structure.
7693  * @param[in] action
7694  *   Handle for the shared action to be destroyed.
7695  * @param[out] error
7696  *   Perform verbose error reporting if not NULL. PMDs initialize this
7697  *   structure in case of error only.
7698  *
7699  * @return
7700  *   0 on success, a negative errno value otherwise and rte_errno is set.
7701  *
7702  * @note: wrapper for driver action_create op callback.
7703  */
7704 static int
7705 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7706 			   struct rte_flow_shared_action *action,
7707 			   struct rte_flow_error *error)
7708 {
7709 	static const char err_msg[] = "shared action destruction unsupported";
7710 	struct rte_flow_attr attr = { .transfer = 0 };
7711 	const struct mlx5_flow_driver_ops *fops =
7712 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7713 
7714 	if (!fops->action_destroy) {
7715 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7716 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7717 				   NULL, err_msg);
7718 		return -rte_errno;
7719 	}
7720 	return fops->action_destroy(dev, action, error);
7721 }
7722 
7723 /* Wrapper for driver action_destroy op callback */
7724 static int
7725 flow_drv_action_update(struct rte_eth_dev *dev,
7726 		       struct rte_flow_shared_action *action,
7727 		       const void *action_conf,
7728 		       const struct mlx5_flow_driver_ops *fops,
7729 		       struct rte_flow_error *error)
7730 {
7731 	static const char err_msg[] = "shared action update unsupported";
7732 
7733 	if (!fops->action_update) {
7734 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7735 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7736 				   NULL, err_msg);
7737 		return -rte_errno;
7738 	}
7739 	return fops->action_update(dev, action, action_conf, error);
7740 }
7741 
7742 /**
7743  * Create shared action for reuse in multiple flow rules.
7744  *
7745  * @param dev
7746  *   Pointer to Ethernet device structure.
7747  * @param[in] action
7748  *   Action configuration for shared action creation.
7749  * @param[out] error
7750  *   Perform verbose error reporting if not NULL. PMDs initialize this
7751  *   structure in case of error only.
7752  * @return
7753  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7754  */
7755 static struct rte_flow_shared_action *
7756 mlx5_shared_action_create(struct rte_eth_dev *dev,
7757 			  const struct rte_flow_shared_action_conf *conf,
7758 			  const struct rte_flow_action *action,
7759 			  struct rte_flow_error *error)
7760 {
7761 	static const char err_msg[] = "shared action creation unsupported";
7762 	struct rte_flow_attr attr = { .transfer = 0 };
7763 	const struct mlx5_flow_driver_ops *fops =
7764 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7765 
7766 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7767 		return NULL;
7768 	if (!fops->action_create) {
7769 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7770 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7771 				   NULL, err_msg);
7772 		return NULL;
7773 	}
7774 	return fops->action_create(dev, conf, action, error);
7775 }
7776 
7777 /**
7778  * Updates inplace the shared action configuration pointed by *action* handle
7779  * with the configuration provided as *action* argument.
7780  * The update of the shared action configuration effects all flow rules reusing
7781  * the action via handle.
7782  *
7783  * @param dev
7784  *   Pointer to Ethernet device structure.
7785  * @param[in] shared_action
7786  *   Handle for the shared action to be updated.
7787  * @param[in] action
7788  *   Action specification used to modify the action pointed by handle.
7789  *   *action* should be of same type with the action pointed by the *action*
7790  *   handle argument, otherwise considered as invalid.
7791  * @param[out] error
7792  *   Perform verbose error reporting if not NULL. PMDs initialize this
7793  *   structure in case of error only.
7794  *
7795  * @return
7796  *   0 on success, a negative errno value otherwise and rte_errno is set.
7797  */
7798 static int
7799 mlx5_shared_action_update(struct rte_eth_dev *dev,
7800 		struct rte_flow_shared_action *shared_action,
7801 		const struct rte_flow_action *action,
7802 		struct rte_flow_error *error)
7803 {
7804 	struct rte_flow_attr attr = { .transfer = 0 };
7805 	const struct mlx5_flow_driver_ops *fops =
7806 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7807 	int ret;
7808 
7809 	switch (shared_action->type) {
7810 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7811 		if (action->type != RTE_FLOW_ACTION_TYPE_RSS) {
7812 			return rte_flow_error_set(error, EINVAL,
7813 						  RTE_FLOW_ERROR_TYPE_ACTION,
7814 						  NULL,
7815 						  "update action type invalid");
7816 		}
7817 		ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7818 		if (ret)
7819 			return ret;
7820 		return flow_drv_action_update(dev, shared_action, action->conf,
7821 					      fops, error);
7822 	default:
7823 		return rte_flow_error_set(error, ENOTSUP,
7824 					  RTE_FLOW_ERROR_TYPE_ACTION,
7825 					  NULL,
7826 					  "action type not supported");
7827 	}
7828 }
7829 
7830 /**
7831  * Query the shared action by handle.
7832  *
7833  * This function allows retrieving action-specific data such as counters.
7834  * Data is gathered by special action which may be present/referenced in
7835  * more than one flow rule definition.
7836  *
7837  * \see RTE_FLOW_ACTION_TYPE_COUNT
7838  *
7839  * @param dev
7840  *   Pointer to Ethernet device structure.
7841  * @param[in] action
7842  *   Handle for the shared action to query.
7843  * @param[in, out] data
7844  *   Pointer to storage for the associated query data type.
7845  * @param[out] error
7846  *   Perform verbose error reporting if not NULL. PMDs initialize this
7847  *   structure in case of error only.
7848  *
7849  * @return
7850  *   0 on success, a negative errno value otherwise and rte_errno is set.
7851  */
7852 static int
7853 mlx5_shared_action_query(struct rte_eth_dev *dev,
7854 			 const struct rte_flow_shared_action *action,
7855 			 void *data,
7856 			 struct rte_flow_error *error)
7857 {
7858 	(void)dev;
7859 	switch (action->type) {
7860 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7861 		__atomic_load(&action->refcnt, (uint32_t *)data,
7862 			      __ATOMIC_RELAXED);
7863 		return 0;
7864 	default:
7865 		return rte_flow_error_set(error, ENOTSUP,
7866 					  RTE_FLOW_ERROR_TYPE_ACTION,
7867 					  NULL,
7868 					  "action type not supported");
7869 	}
7870 }
7871 
7872 /**
7873  * Destroy all shared actions.
7874  *
7875  * @param dev
7876  *   Pointer to Ethernet device.
7877  *
7878  * @return
7879  *   0 on success, a negative errno value otherwise and rte_errno is set.
7880  */
7881 int
7882 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7883 {
7884 	struct rte_flow_error error;
7885 	struct mlx5_priv *priv = dev->data->dev_private;
7886 	struct rte_flow_shared_action *action;
7887 	int ret = 0;
7888 
7889 	while (!LIST_EMPTY(&priv->shared_actions)) {
7890 		action = LIST_FIRST(&priv->shared_actions);
7891 		ret = mlx5_shared_action_destroy(dev, action, &error);
7892 	}
7893 	return ret;
7894 }
7895 
7896 static void
7897 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7898 		      struct mlx5_flow_tunnel *tunnel)
7899 {
7900 	struct mlx5_priv *priv = dev->data->dev_private;
7901 
7902 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7903 		dev->data->port_id, tunnel->tunnel_id);
7904 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
7905 	LIST_REMOVE(tunnel, chain);
7906 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
7907 			tunnel->tunnel_id);
7908 	mlx5_hlist_destroy(tunnel->groups);
7909 	mlx5_free(tunnel);
7910 }
7911 
7912 static struct mlx5_flow_tunnel *
7913 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7914 {
7915 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7916 	struct mlx5_flow_tunnel *tun;
7917 
7918 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7919 		if (tun->tunnel_id == id)
7920 			break;
7921 	}
7922 
7923 	return tun;
7924 }
7925 
7926 static struct mlx5_flow_tunnel *
7927 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7928 			  const struct rte_flow_tunnel *app_tunnel)
7929 {
7930 	struct mlx5_priv *priv = dev->data->dev_private;
7931 	struct mlx5_flow_tunnel *tunnel;
7932 	uint32_t id;
7933 
7934 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
7935 			  &id);
7936 	if (id >= MLX5_MAX_TUNNELS) {
7937 		mlx5_ipool_free(priv->sh->ipool
7938 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7939 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7940 		return NULL;
7941 	} else if (!id) {
7942 		return NULL;
7943 	}
7944 	/**
7945 	 * mlx5 flow tunnel is an auxlilary data structure
7946 	 * It's not part of IO. No need to allocate it from
7947 	 * huge pages pools dedicated for IO
7948 	 */
7949 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
7950 			     0, SOCKET_ID_ANY);
7951 	if (!tunnel) {
7952 		mlx5_ipool_free(priv->sh->ipool
7953 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7954 		return NULL;
7955 	}
7956 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7957 					   NULL, NULL, NULL);
7958 	if (!tunnel->groups) {
7959 		mlx5_ipool_free(priv->sh->ipool
7960 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7961 		mlx5_free(tunnel);
7962 		return NULL;
7963 	}
7964 	/* initiate new PMD tunnel */
7965 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7966 	tunnel->tunnel_id = id;
7967 	tunnel->action.type = (typeof(tunnel->action.type))
7968 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7969 	tunnel->action.conf = tunnel;
7970 	tunnel->item.type = (typeof(tunnel->item.type))
7971 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7972 	tunnel->item.spec = tunnel;
7973 	tunnel->item.last = NULL;
7974 	tunnel->item.mask = NULL;
7975 
7976 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7977 		dev->data->port_id, tunnel->tunnel_id);
7978 
7979 	return tunnel;
7980 }
7981 
7982 static int
7983 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7984 		     const struct rte_flow_tunnel *app_tunnel,
7985 		     struct mlx5_flow_tunnel **tunnel)
7986 {
7987 	int ret;
7988 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7989 	struct mlx5_flow_tunnel *tun;
7990 
7991 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7992 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7993 			    sizeof(*app_tunnel))) {
7994 			*tunnel = tun;
7995 			ret = 0;
7996 			break;
7997 		}
7998 	}
7999 	if (!tun) {
8000 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
8001 		if (tun) {
8002 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
8003 			*tunnel = tun;
8004 		} else {
8005 			ret = -ENOMEM;
8006 		}
8007 	}
8008 	if (tun)
8009 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
8010 
8011 	return ret;
8012 }
8013 
8014 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8015 {
8016 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8017 
8018 	if (!thub)
8019 		return;
8020 	if (!LIST_EMPTY(&thub->tunnels))
8021 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
8022 	mlx5_hlist_destroy(thub->groups);
8023 	mlx5_free(thub);
8024 }
8025 
8026 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8027 {
8028 	int err;
8029 	struct mlx5_flow_tunnel_hub *thub;
8030 
8031 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8032 			   0, SOCKET_ID_ANY);
8033 	if (!thub)
8034 		return -ENOMEM;
8035 	LIST_INIT(&thub->tunnels);
8036 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
8037 					 0, NULL, NULL, NULL);
8038 	if (!thub->groups) {
8039 		err = -rte_errno;
8040 		goto err;
8041 	}
8042 	sh->tunnel_hub = thub;
8043 
8044 	return 0;
8045 
8046 err:
8047 	if (thub->groups)
8048 		mlx5_hlist_destroy(thub->groups);
8049 	if (thub)
8050 		mlx5_free(thub);
8051 	return err;
8052 }
8053 
8054 #ifndef HAVE_MLX5DV_DR
8055 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8056 #else
8057 #define MLX5_DOMAIN_SYNC_FLOW \
8058 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8059 #endif
8060 
8061 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8062 {
8063 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8064 	const struct mlx5_flow_driver_ops *fops;
8065 	int ret;
8066 	struct rte_flow_attr attr = { .transfer = 0 };
8067 
8068 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8069 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8070 	if (ret > 0)
8071 		ret = -ret;
8072 	return ret;
8073 }
8074