xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision f935ed4b645a8760884500a0e936d50e3c0a35c7)
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 	rte_spinlock_lock(&thub->sl);
673 	LIST_FOREACH(tun, &thub->tunnels, chain) {
674 		if (&tun->item == pmd_items) {
675 			LIST_REMOVE(tun, chain);
676 			break;
677 		}
678 	}
679 	rte_spinlock_unlock(&thub->sl);
680 	if (!tun || num_items != 1)
681 		return rte_flow_error_set(err, EINVAL,
682 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
683 					  "invalid argument");
684 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
685 		mlx5_flow_tunnel_free(dev, tun);
686 	return 0;
687 }
688 
689 static int
690 mlx5_flow_action_release(struct rte_eth_dev *dev,
691 			 struct rte_flow_action *pmd_actions,
692 			 uint32_t num_actions, struct rte_flow_error *err)
693 {
694 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
695 	struct mlx5_flow_tunnel *tun;
696 
697 	rte_spinlock_lock(&thub->sl);
698 	LIST_FOREACH(tun, &thub->tunnels, chain) {
699 		if (&tun->action == pmd_actions) {
700 			LIST_REMOVE(tun, chain);
701 			break;
702 		}
703 	}
704 	rte_spinlock_unlock(&thub->sl);
705 	if (!tun || num_actions != 1)
706 		return rte_flow_error_set(err, EINVAL,
707 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
708 					  "invalid argument");
709 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
710 		mlx5_flow_tunnel_free(dev, tun);
711 
712 	return 0;
713 }
714 
715 static int
716 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
717 				  struct rte_mbuf *m,
718 				  struct rte_flow_restore_info *info,
719 				  struct rte_flow_error *err)
720 {
721 	uint64_t ol_flags = m->ol_flags;
722 	const struct mlx5_flow_tbl_data_entry *tble;
723 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
724 
725 	if ((ol_flags & mask) != mask)
726 		goto err;
727 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
728 	if (!tble) {
729 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
730 			dev->data->port_id, m->hash.fdir.hi);
731 		goto err;
732 	}
733 	MLX5_ASSERT(tble->tunnel);
734 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
735 	info->group_id = tble->group_id;
736 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
737 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
738 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
739 
740 	return 0;
741 
742 err:
743 	return rte_flow_error_set(err, EINVAL,
744 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
745 				  "failed to get restore info");
746 }
747 
748 static const struct rte_flow_ops mlx5_flow_ops = {
749 	.validate = mlx5_flow_validate,
750 	.create = mlx5_flow_create,
751 	.destroy = mlx5_flow_destroy,
752 	.flush = mlx5_flow_flush,
753 	.isolate = mlx5_flow_isolate,
754 	.query = mlx5_flow_query,
755 	.dev_dump = mlx5_flow_dev_dump,
756 	.get_aged_flows = mlx5_flow_get_aged_flows,
757 	.shared_action_create = mlx5_shared_action_create,
758 	.shared_action_destroy = mlx5_shared_action_destroy,
759 	.shared_action_update = mlx5_shared_action_update,
760 	.shared_action_query = mlx5_shared_action_query,
761 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
762 	.tunnel_match = mlx5_flow_tunnel_match,
763 	.tunnel_action_decap_release = mlx5_flow_action_release,
764 	.tunnel_item_release = mlx5_flow_item_release,
765 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
766 };
767 
768 /* Tunnel information. */
769 struct mlx5_flow_tunnel_info {
770 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
771 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
772 };
773 
774 static struct mlx5_flow_tunnel_info tunnels_info[] = {
775 	{
776 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
777 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
778 	},
779 	{
780 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
781 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
782 	},
783 	{
784 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
785 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
786 	},
787 	{
788 		.tunnel = MLX5_FLOW_LAYER_GRE,
789 		.ptype = RTE_PTYPE_TUNNEL_GRE,
790 	},
791 	{
792 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
793 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
794 	},
795 	{
796 		.tunnel = MLX5_FLOW_LAYER_MPLS,
797 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
798 	},
799 	{
800 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
801 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
802 	},
803 	{
804 		.tunnel = MLX5_FLOW_LAYER_IPIP,
805 		.ptype = RTE_PTYPE_TUNNEL_IP,
806 	},
807 	{
808 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
809 		.ptype = RTE_PTYPE_TUNNEL_IP,
810 	},
811 	{
812 		.tunnel = MLX5_FLOW_LAYER_GTP,
813 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
814 	},
815 };
816 
817 /* Key of thread specific flow workspace data. */
818 static pthread_key_t key_workspace;
819 
820 /* Thread specific flow workspace data once initialization data. */
821 static pthread_once_t key_workspace_init;
822 
823 
824 /**
825  * Translate tag ID to register.
826  *
827  * @param[in] dev
828  *   Pointer to the Ethernet device structure.
829  * @param[in] feature
830  *   The feature that request the register.
831  * @param[in] id
832  *   The request register ID.
833  * @param[out] error
834  *   Error description in case of any.
835  *
836  * @return
837  *   The request register on success, a negative errno
838  *   value otherwise and rte_errno is set.
839  */
840 int
841 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
842 		     enum mlx5_feature_name feature,
843 		     uint32_t id,
844 		     struct rte_flow_error *error)
845 {
846 	struct mlx5_priv *priv = dev->data->dev_private;
847 	struct mlx5_dev_config *config = &priv->config;
848 	enum modify_reg start_reg;
849 	bool skip_mtr_reg = false;
850 
851 	switch (feature) {
852 	case MLX5_HAIRPIN_RX:
853 		return REG_B;
854 	case MLX5_HAIRPIN_TX:
855 		return REG_A;
856 	case MLX5_METADATA_RX:
857 		switch (config->dv_xmeta_en) {
858 		case MLX5_XMETA_MODE_LEGACY:
859 			return REG_B;
860 		case MLX5_XMETA_MODE_META16:
861 			return REG_C_0;
862 		case MLX5_XMETA_MODE_META32:
863 			return REG_C_1;
864 		}
865 		break;
866 	case MLX5_METADATA_TX:
867 		return REG_A;
868 	case MLX5_METADATA_FDB:
869 		switch (config->dv_xmeta_en) {
870 		case MLX5_XMETA_MODE_LEGACY:
871 			return REG_NON;
872 		case MLX5_XMETA_MODE_META16:
873 			return REG_C_0;
874 		case MLX5_XMETA_MODE_META32:
875 			return REG_C_1;
876 		}
877 		break;
878 	case MLX5_FLOW_MARK:
879 		switch (config->dv_xmeta_en) {
880 		case MLX5_XMETA_MODE_LEGACY:
881 			return REG_NON;
882 		case MLX5_XMETA_MODE_META16:
883 			return REG_C_1;
884 		case MLX5_XMETA_MODE_META32:
885 			return REG_C_0;
886 		}
887 		break;
888 	case MLX5_MTR_SFX:
889 		/*
890 		 * If meter color and flow match share one register, flow match
891 		 * should use the meter color register for match.
892 		 */
893 		if (priv->mtr_reg_share)
894 			return priv->mtr_color_reg;
895 		else
896 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
897 			       REG_C_3;
898 	case MLX5_MTR_COLOR:
899 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
900 		return priv->mtr_color_reg;
901 	case MLX5_COPY_MARK:
902 		/*
903 		 * Metadata COPY_MARK register using is in meter suffix sub
904 		 * flow while with meter. It's safe to share the same register.
905 		 */
906 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
907 	case MLX5_APP_TAG:
908 		/*
909 		 * If meter is enable, it will engage the register for color
910 		 * match and flow match. If meter color match is not using the
911 		 * REG_C_2, need to skip the REG_C_x be used by meter color
912 		 * match.
913 		 * If meter is disable, free to use all available registers.
914 		 */
915 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
916 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
917 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
918 		if (id > (REG_C_7 - start_reg))
919 			return rte_flow_error_set(error, EINVAL,
920 						  RTE_FLOW_ERROR_TYPE_ITEM,
921 						  NULL, "invalid tag id");
922 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
923 			return rte_flow_error_set(error, ENOTSUP,
924 						  RTE_FLOW_ERROR_TYPE_ITEM,
925 						  NULL, "unsupported tag id");
926 		/*
927 		 * This case means meter is using the REG_C_x great than 2.
928 		 * Take care not to conflict with meter color REG_C_x.
929 		 * If the available index REG_C_y >= REG_C_x, skip the
930 		 * color register.
931 		 */
932 		if (skip_mtr_reg && config->flow_mreg_c
933 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
934 			if (id >= (REG_C_7 - start_reg))
935 				return rte_flow_error_set(error, EINVAL,
936 						       RTE_FLOW_ERROR_TYPE_ITEM,
937 							NULL, "invalid tag id");
938 			if (config->flow_mreg_c
939 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
940 				return config->flow_mreg_c
941 					       [id + 1 + start_reg - REG_C_0];
942 			return rte_flow_error_set(error, ENOTSUP,
943 						  RTE_FLOW_ERROR_TYPE_ITEM,
944 						  NULL, "unsupported tag id");
945 		}
946 		return config->flow_mreg_c[id + start_reg - REG_C_0];
947 	}
948 	MLX5_ASSERT(false);
949 	return rte_flow_error_set(error, EINVAL,
950 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
951 				  NULL, "invalid feature name");
952 }
953 
954 /**
955  * Check extensive flow metadata register support.
956  *
957  * @param dev
958  *   Pointer to rte_eth_dev structure.
959  *
960  * @return
961  *   True if device supports extensive flow metadata register, otherwise false.
962  */
963 bool
964 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
965 {
966 	struct mlx5_priv *priv = dev->data->dev_private;
967 	struct mlx5_dev_config *config = &priv->config;
968 
969 	/*
970 	 * Having available reg_c can be regarded inclusively as supporting
971 	 * extensive flow metadata register, which could mean,
972 	 * - metadata register copy action by modify header.
973 	 * - 16 modify header actions is supported.
974 	 * - reg_c's are preserved across different domain (FDB and NIC) on
975 	 *   packet loopback by flow lookup miss.
976 	 */
977 	return config->flow_mreg_c[2] != REG_NON;
978 }
979 
980 /**
981  * Verify the @p item specifications (spec, last, mask) are compatible with the
982  * NIC capabilities.
983  *
984  * @param[in] item
985  *   Item specification.
986  * @param[in] mask
987  *   @p item->mask or flow default bit-masks.
988  * @param[in] nic_mask
989  *   Bit-masks covering supported fields by the NIC to compare with user mask.
990  * @param[in] size
991  *   Bit-masks size in bytes.
992  * @param[in] range_accepted
993  *   True if range of values is accepted for specific fields, false otherwise.
994  * @param[out] error
995  *   Pointer to error structure.
996  *
997  * @return
998  *   0 on success, a negative errno value otherwise and rte_errno is set.
999  */
1000 int
1001 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1002 			  const uint8_t *mask,
1003 			  const uint8_t *nic_mask,
1004 			  unsigned int size,
1005 			  bool range_accepted,
1006 			  struct rte_flow_error *error)
1007 {
1008 	unsigned int i;
1009 
1010 	MLX5_ASSERT(nic_mask);
1011 	for (i = 0; i < size; ++i)
1012 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1013 			return rte_flow_error_set(error, ENOTSUP,
1014 						  RTE_FLOW_ERROR_TYPE_ITEM,
1015 						  item,
1016 						  "mask enables non supported"
1017 						  " bits");
1018 	if (!item->spec && (item->mask || item->last))
1019 		return rte_flow_error_set(error, EINVAL,
1020 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1021 					  "mask/last without a spec is not"
1022 					  " supported");
1023 	if (item->spec && item->last && !range_accepted) {
1024 		uint8_t spec[size];
1025 		uint8_t last[size];
1026 		unsigned int i;
1027 		int ret;
1028 
1029 		for (i = 0; i < size; ++i) {
1030 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1031 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1032 		}
1033 		ret = memcmp(spec, last, size);
1034 		if (ret != 0)
1035 			return rte_flow_error_set(error, EINVAL,
1036 						  RTE_FLOW_ERROR_TYPE_ITEM,
1037 						  item,
1038 						  "range is not valid");
1039 	}
1040 	return 0;
1041 }
1042 
1043 /**
1044  * Adjust the hash fields according to the @p flow information.
1045  *
1046  * @param[in] dev_flow.
1047  *   Pointer to the mlx5_flow.
1048  * @param[in] tunnel
1049  *   1 when the hash field is for a tunnel item.
1050  * @param[in] layer_types
1051  *   ETH_RSS_* types.
1052  * @param[in] hash_fields
1053  *   Item hash fields.
1054  *
1055  * @return
1056  *   The hash fields that should be used.
1057  */
1058 uint64_t
1059 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1060 			    int tunnel __rte_unused, uint64_t layer_types,
1061 			    uint64_t hash_fields)
1062 {
1063 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1064 	int rss_request_inner = rss_desc->level >= 2;
1065 
1066 	/* Check RSS hash level for tunnel. */
1067 	if (tunnel && rss_request_inner)
1068 		hash_fields |= IBV_RX_HASH_INNER;
1069 	else if (tunnel || rss_request_inner)
1070 		return 0;
1071 #endif
1072 	/* Check if requested layer matches RSS hash fields. */
1073 	if (!(rss_desc->types & layer_types))
1074 		return 0;
1075 	return hash_fields;
1076 }
1077 
1078 /**
1079  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1080  * if several tunnel rules are used on this queue, the tunnel ptype will be
1081  * cleared.
1082  *
1083  * @param rxq_ctrl
1084  *   Rx queue to update.
1085  */
1086 static void
1087 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1088 {
1089 	unsigned int i;
1090 	uint32_t tunnel_ptype = 0;
1091 
1092 	/* Look up for the ptype to use. */
1093 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1094 		if (!rxq_ctrl->flow_tunnels_n[i])
1095 			continue;
1096 		if (!tunnel_ptype) {
1097 			tunnel_ptype = tunnels_info[i].ptype;
1098 		} else {
1099 			tunnel_ptype = 0;
1100 			break;
1101 		}
1102 	}
1103 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1104 }
1105 
1106 /**
1107  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1108  * flow.
1109  *
1110  * @param[in] dev
1111  *   Pointer to the Ethernet device structure.
1112  * @param[in] dev_handle
1113  *   Pointer to device flow handle structure.
1114  */
1115 static void
1116 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1117 		       struct mlx5_flow_handle *dev_handle)
1118 {
1119 	struct mlx5_priv *priv = dev->data->dev_private;
1120 	const int mark = dev_handle->mark;
1121 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1122 	struct mlx5_hrxq *hrxq;
1123 	unsigned int i;
1124 
1125 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1126 		return;
1127 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1128 			      dev_handle->rix_hrxq);
1129 	if (!hrxq)
1130 		return;
1131 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1132 		int idx = hrxq->ind_table->queues[i];
1133 		struct mlx5_rxq_ctrl *rxq_ctrl =
1134 			container_of((*priv->rxqs)[idx],
1135 				     struct mlx5_rxq_ctrl, rxq);
1136 
1137 		/*
1138 		 * To support metadata register copy on Tx loopback,
1139 		 * this must be always enabled (metadata may arive
1140 		 * from other port - not from local flows only.
1141 		 */
1142 		if (priv->config.dv_flow_en &&
1143 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1144 		    mlx5_flow_ext_mreg_supported(dev)) {
1145 			rxq_ctrl->rxq.mark = 1;
1146 			rxq_ctrl->flow_mark_n = 1;
1147 		} else if (mark) {
1148 			rxq_ctrl->rxq.mark = 1;
1149 			rxq_ctrl->flow_mark_n++;
1150 		}
1151 		if (tunnel) {
1152 			unsigned int j;
1153 
1154 			/* Increase the counter matching the flow. */
1155 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1156 				if ((tunnels_info[j].tunnel &
1157 				     dev_handle->layers) ==
1158 				    tunnels_info[j].tunnel) {
1159 					rxq_ctrl->flow_tunnels_n[j]++;
1160 					break;
1161 				}
1162 			}
1163 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1164 		}
1165 	}
1166 }
1167 
1168 /**
1169  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1170  *
1171  * @param[in] dev
1172  *   Pointer to the Ethernet device structure.
1173  * @param[in] flow
1174  *   Pointer to flow structure.
1175  */
1176 static void
1177 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1178 {
1179 	struct mlx5_priv *priv = dev->data->dev_private;
1180 	uint32_t handle_idx;
1181 	struct mlx5_flow_handle *dev_handle;
1182 
1183 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1184 		       handle_idx, dev_handle, next)
1185 		flow_drv_rxq_flags_set(dev, dev_handle);
1186 }
1187 
1188 /**
1189  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1190  * device flow if no other flow uses it with the same kind of request.
1191  *
1192  * @param dev
1193  *   Pointer to Ethernet device.
1194  * @param[in] dev_handle
1195  *   Pointer to the device flow handle structure.
1196  */
1197 static void
1198 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1199 			struct mlx5_flow_handle *dev_handle)
1200 {
1201 	struct mlx5_priv *priv = dev->data->dev_private;
1202 	const int mark = dev_handle->mark;
1203 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1204 	struct mlx5_hrxq *hrxq;
1205 	unsigned int i;
1206 
1207 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1208 		return;
1209 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1210 			      dev_handle->rix_hrxq);
1211 	if (!hrxq)
1212 		return;
1213 	MLX5_ASSERT(dev->data->dev_started);
1214 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1215 		int idx = hrxq->ind_table->queues[i];
1216 		struct mlx5_rxq_ctrl *rxq_ctrl =
1217 			container_of((*priv->rxqs)[idx],
1218 				     struct mlx5_rxq_ctrl, rxq);
1219 
1220 		if (priv->config.dv_flow_en &&
1221 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1222 		    mlx5_flow_ext_mreg_supported(dev)) {
1223 			rxq_ctrl->rxq.mark = 1;
1224 			rxq_ctrl->flow_mark_n = 1;
1225 		} else if (mark) {
1226 			rxq_ctrl->flow_mark_n--;
1227 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1228 		}
1229 		if (tunnel) {
1230 			unsigned int j;
1231 
1232 			/* Decrease the counter matching the flow. */
1233 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1234 				if ((tunnels_info[j].tunnel &
1235 				     dev_handle->layers) ==
1236 				    tunnels_info[j].tunnel) {
1237 					rxq_ctrl->flow_tunnels_n[j]--;
1238 					break;
1239 				}
1240 			}
1241 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1242 		}
1243 	}
1244 }
1245 
1246 /**
1247  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1248  * @p flow if no other flow uses it with the same kind of request.
1249  *
1250  * @param dev
1251  *   Pointer to Ethernet device.
1252  * @param[in] flow
1253  *   Pointer to the flow.
1254  */
1255 static void
1256 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1257 {
1258 	struct mlx5_priv *priv = dev->data->dev_private;
1259 	uint32_t handle_idx;
1260 	struct mlx5_flow_handle *dev_handle;
1261 
1262 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1263 		       handle_idx, dev_handle, next)
1264 		flow_drv_rxq_flags_trim(dev, dev_handle);
1265 }
1266 
1267 /**
1268  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1269  *
1270  * @param dev
1271  *   Pointer to Ethernet device.
1272  */
1273 static void
1274 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1275 {
1276 	struct mlx5_priv *priv = dev->data->dev_private;
1277 	unsigned int i;
1278 
1279 	for (i = 0; i != priv->rxqs_n; ++i) {
1280 		struct mlx5_rxq_ctrl *rxq_ctrl;
1281 		unsigned int j;
1282 
1283 		if (!(*priv->rxqs)[i])
1284 			continue;
1285 		rxq_ctrl = container_of((*priv->rxqs)[i],
1286 					struct mlx5_rxq_ctrl, rxq);
1287 		rxq_ctrl->flow_mark_n = 0;
1288 		rxq_ctrl->rxq.mark = 0;
1289 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1290 			rxq_ctrl->flow_tunnels_n[j] = 0;
1291 		rxq_ctrl->rxq.tunnel = 0;
1292 	}
1293 }
1294 
1295 /**
1296  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1297  *
1298  * @param[in] dev
1299  *   Pointer to the Ethernet device structure.
1300  */
1301 void
1302 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1303 {
1304 	struct mlx5_priv *priv = dev->data->dev_private;
1305 	struct mlx5_rxq_data *data;
1306 	unsigned int i;
1307 
1308 	for (i = 0; i != priv->rxqs_n; ++i) {
1309 		if (!(*priv->rxqs)[i])
1310 			continue;
1311 		data = (*priv->rxqs)[i];
1312 		if (!rte_flow_dynf_metadata_avail()) {
1313 			data->dynf_meta = 0;
1314 			data->flow_meta_mask = 0;
1315 			data->flow_meta_offset = -1;
1316 		} else {
1317 			data->dynf_meta = 1;
1318 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1319 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1320 		}
1321 	}
1322 }
1323 
1324 /*
1325  * return a pointer to the desired action in the list of actions.
1326  *
1327  * @param[in] actions
1328  *   The list of actions to search the action in.
1329  * @param[in] action
1330  *   The action to find.
1331  *
1332  * @return
1333  *   Pointer to the action in the list, if found. NULL otherwise.
1334  */
1335 const struct rte_flow_action *
1336 mlx5_flow_find_action(const struct rte_flow_action *actions,
1337 		      enum rte_flow_action_type action)
1338 {
1339 	if (actions == NULL)
1340 		return NULL;
1341 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1342 		if (actions->type == action)
1343 			return actions;
1344 	return NULL;
1345 }
1346 
1347 /*
1348  * Validate the flag action.
1349  *
1350  * @param[in] action_flags
1351  *   Bit-fields that holds the actions detected until now.
1352  * @param[in] attr
1353  *   Attributes of flow that includes this action.
1354  * @param[out] error
1355  *   Pointer to error structure.
1356  *
1357  * @return
1358  *   0 on success, a negative errno value otherwise and rte_errno is set.
1359  */
1360 int
1361 mlx5_flow_validate_action_flag(uint64_t action_flags,
1362 			       const struct rte_flow_attr *attr,
1363 			       struct rte_flow_error *error)
1364 {
1365 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1366 		return rte_flow_error_set(error, EINVAL,
1367 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1368 					  "can't mark and flag in same flow");
1369 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1370 		return rte_flow_error_set(error, EINVAL,
1371 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1372 					  "can't have 2 flag"
1373 					  " actions in same flow");
1374 	if (attr->egress)
1375 		return rte_flow_error_set(error, ENOTSUP,
1376 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1377 					  "flag action not supported for "
1378 					  "egress");
1379 	return 0;
1380 }
1381 
1382 /*
1383  * Validate the mark action.
1384  *
1385  * @param[in] action
1386  *   Pointer to the queue action.
1387  * @param[in] action_flags
1388  *   Bit-fields that holds the actions detected until now.
1389  * @param[in] attr
1390  *   Attributes of flow that includes this action.
1391  * @param[out] error
1392  *   Pointer to error structure.
1393  *
1394  * @return
1395  *   0 on success, a negative errno value otherwise and rte_errno is set.
1396  */
1397 int
1398 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1399 			       uint64_t action_flags,
1400 			       const struct rte_flow_attr *attr,
1401 			       struct rte_flow_error *error)
1402 {
1403 	const struct rte_flow_action_mark *mark = action->conf;
1404 
1405 	if (!mark)
1406 		return rte_flow_error_set(error, EINVAL,
1407 					  RTE_FLOW_ERROR_TYPE_ACTION,
1408 					  action,
1409 					  "configuration cannot be null");
1410 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1411 		return rte_flow_error_set(error, EINVAL,
1412 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1413 					  &mark->id,
1414 					  "mark id must in 0 <= id < "
1415 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1416 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1417 		return rte_flow_error_set(error, EINVAL,
1418 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1419 					  "can't flag and mark in same flow");
1420 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1421 		return rte_flow_error_set(error, EINVAL,
1422 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1423 					  "can't have 2 mark actions in same"
1424 					  " flow");
1425 	if (attr->egress)
1426 		return rte_flow_error_set(error, ENOTSUP,
1427 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1428 					  "mark action not supported for "
1429 					  "egress");
1430 	return 0;
1431 }
1432 
1433 /*
1434  * Validate the drop action.
1435  *
1436  * @param[in] action_flags
1437  *   Bit-fields that holds the actions detected until now.
1438  * @param[in] attr
1439  *   Attributes of flow that includes this action.
1440  * @param[out] error
1441  *   Pointer to error structure.
1442  *
1443  * @return
1444  *   0 on success, a negative errno value otherwise and rte_errno is set.
1445  */
1446 int
1447 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1448 			       const struct rte_flow_attr *attr,
1449 			       struct rte_flow_error *error)
1450 {
1451 	if (attr->egress)
1452 		return rte_flow_error_set(error, ENOTSUP,
1453 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1454 					  "drop action not supported for "
1455 					  "egress");
1456 	return 0;
1457 }
1458 
1459 /*
1460  * Validate the queue action.
1461  *
1462  * @param[in] action
1463  *   Pointer to the queue action.
1464  * @param[in] action_flags
1465  *   Bit-fields that holds the actions detected until now.
1466  * @param[in] dev
1467  *   Pointer to the Ethernet device structure.
1468  * @param[in] attr
1469  *   Attributes of flow that includes this action.
1470  * @param[out] error
1471  *   Pointer to error structure.
1472  *
1473  * @return
1474  *   0 on success, a negative errno value otherwise and rte_errno is set.
1475  */
1476 int
1477 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1478 				uint64_t action_flags,
1479 				struct rte_eth_dev *dev,
1480 				const struct rte_flow_attr *attr,
1481 				struct rte_flow_error *error)
1482 {
1483 	struct mlx5_priv *priv = dev->data->dev_private;
1484 	const struct rte_flow_action_queue *queue = action->conf;
1485 
1486 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1487 		return rte_flow_error_set(error, EINVAL,
1488 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1489 					  "can't have 2 fate actions in"
1490 					  " same flow");
1491 	if (!priv->rxqs_n)
1492 		return rte_flow_error_set(error, EINVAL,
1493 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1494 					  NULL, "No Rx queues configured");
1495 	if (queue->index >= priv->rxqs_n)
1496 		return rte_flow_error_set(error, EINVAL,
1497 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 					  &queue->index,
1499 					  "queue index out of range");
1500 	if (!(*priv->rxqs)[queue->index])
1501 		return rte_flow_error_set(error, EINVAL,
1502 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1503 					  &queue->index,
1504 					  "queue is not configured");
1505 	if (attr->egress)
1506 		return rte_flow_error_set(error, ENOTSUP,
1507 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1508 					  "queue action not supported for "
1509 					  "egress");
1510 	return 0;
1511 }
1512 
1513 /*
1514  * Validate the rss action.
1515  *
1516  * @param[in] dev
1517  *   Pointer to the Ethernet device structure.
1518  * @param[in] action
1519  *   Pointer to the queue action.
1520  * @param[out] error
1521  *   Pointer to error structure.
1522  *
1523  * @return
1524  *   0 on success, a negative errno value otherwise and rte_errno is set.
1525  */
1526 int
1527 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1528 			 const struct rte_flow_action *action,
1529 			 struct rte_flow_error *error)
1530 {
1531 	struct mlx5_priv *priv = dev->data->dev_private;
1532 	const struct rte_flow_action_rss *rss = action->conf;
1533 	unsigned int i;
1534 
1535 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1536 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1537 		return rte_flow_error_set(error, ENOTSUP,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 					  &rss->func,
1540 					  "RSS hash function not supported");
1541 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1542 	if (rss->level > 2)
1543 #else
1544 	if (rss->level > 1)
1545 #endif
1546 		return rte_flow_error_set(error, ENOTSUP,
1547 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1548 					  &rss->level,
1549 					  "tunnel RSS is not supported");
1550 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1551 	if (rss->key_len == 0 && rss->key != NULL)
1552 		return rte_flow_error_set(error, ENOTSUP,
1553 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1554 					  &rss->key_len,
1555 					  "RSS hash key length 0");
1556 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1559 					  &rss->key_len,
1560 					  "RSS hash key too small");
1561 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1562 		return rte_flow_error_set(error, ENOTSUP,
1563 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1564 					  &rss->key_len,
1565 					  "RSS hash key too large");
1566 	if (rss->queue_num > priv->config.ind_table_max_size)
1567 		return rte_flow_error_set(error, ENOTSUP,
1568 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 					  &rss->queue_num,
1570 					  "number of queues too large");
1571 	if (rss->types & MLX5_RSS_HF_MASK)
1572 		return rte_flow_error_set(error, ENOTSUP,
1573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1574 					  &rss->types,
1575 					  "some RSS protocols are not"
1576 					  " supported");
1577 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1578 	    !(rss->types & ETH_RSS_IP))
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1581 					  "L3 partial RSS requested but L3 RSS"
1582 					  " type not specified");
1583 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1584 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1585 		return rte_flow_error_set(error, EINVAL,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1587 					  "L4 partial RSS requested but L4 RSS"
1588 					  " type not specified");
1589 	if (!priv->rxqs_n)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 					  NULL, "No Rx queues configured");
1593 	if (!rss->queue_num)
1594 		return rte_flow_error_set(error, EINVAL,
1595 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1596 					  NULL, "No queues configured");
1597 	for (i = 0; i != rss->queue_num; ++i) {
1598 		if (rss->queue[i] >= priv->rxqs_n)
1599 			return rte_flow_error_set
1600 				(error, EINVAL,
1601 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1602 				 &rss->queue[i], "queue index out of range");
1603 		if (!(*priv->rxqs)[rss->queue[i]])
1604 			return rte_flow_error_set
1605 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1606 				 &rss->queue[i], "queue is not configured");
1607 	}
1608 	return 0;
1609 }
1610 
1611 /*
1612  * Validate the rss action.
1613  *
1614  * @param[in] action
1615  *   Pointer to the queue action.
1616  * @param[in] action_flags
1617  *   Bit-fields that holds the actions detected until now.
1618  * @param[in] dev
1619  *   Pointer to the Ethernet device structure.
1620  * @param[in] attr
1621  *   Attributes of flow that includes this action.
1622  * @param[in] item_flags
1623  *   Items that were detected.
1624  * @param[out] error
1625  *   Pointer to error structure.
1626  *
1627  * @return
1628  *   0 on success, a negative errno value otherwise and rte_errno is set.
1629  */
1630 int
1631 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1632 			      uint64_t action_flags,
1633 			      struct rte_eth_dev *dev,
1634 			      const struct rte_flow_attr *attr,
1635 			      uint64_t item_flags,
1636 			      struct rte_flow_error *error)
1637 {
1638 	const struct rte_flow_action_rss *rss = action->conf;
1639 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1640 	int ret;
1641 
1642 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1643 		return rte_flow_error_set(error, EINVAL,
1644 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1645 					  "can't have 2 fate actions"
1646 					  " in same flow");
1647 	ret = mlx5_validate_action_rss(dev, action, error);
1648 	if (ret)
1649 		return ret;
1650 	if (attr->egress)
1651 		return rte_flow_error_set(error, ENOTSUP,
1652 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1653 					  "rss action not supported for "
1654 					  "egress");
1655 	if (rss->level > 1 && !tunnel)
1656 		return rte_flow_error_set(error, EINVAL,
1657 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1658 					  "inner RSS is not supported for "
1659 					  "non-tunnel flows");
1660 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1661 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1662 		return rte_flow_error_set(error, EINVAL,
1663 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1664 					  "RSS on eCPRI is not supported now");
1665 	}
1666 	return 0;
1667 }
1668 
1669 /*
1670  * Validate the default miss action.
1671  *
1672  * @param[in] action_flags
1673  *   Bit-fields that holds the actions detected until now.
1674  * @param[out] error
1675  *   Pointer to error structure.
1676  *
1677  * @return
1678  *   0 on success, a negative errno value otherwise and rte_errno is set.
1679  */
1680 int
1681 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1682 				const struct rte_flow_attr *attr,
1683 				struct rte_flow_error *error)
1684 {
1685 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1688 					  "can't have 2 fate actions in"
1689 					  " same flow");
1690 	if (attr->egress)
1691 		return rte_flow_error_set(error, ENOTSUP,
1692 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1693 					  "default miss action not supported "
1694 					  "for egress");
1695 	if (attr->group)
1696 		return rte_flow_error_set(error, ENOTSUP,
1697 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1698 					  "only group 0 is supported");
1699 	if (attr->transfer)
1700 		return rte_flow_error_set(error, ENOTSUP,
1701 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1702 					  NULL, "transfer is not supported");
1703 	return 0;
1704 }
1705 
1706 /*
1707  * Validate the count action.
1708  *
1709  * @param[in] dev
1710  *   Pointer to the Ethernet device structure.
1711  * @param[in] attr
1712  *   Attributes of flow that includes this action.
1713  * @param[out] error
1714  *   Pointer to error structure.
1715  *
1716  * @return
1717  *   0 on success, a negative errno value otherwise and rte_errno is set.
1718  */
1719 int
1720 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1721 				const struct rte_flow_attr *attr,
1722 				struct rte_flow_error *error)
1723 {
1724 	if (attr->egress)
1725 		return rte_flow_error_set(error, ENOTSUP,
1726 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1727 					  "count action not supported for "
1728 					  "egress");
1729 	return 0;
1730 }
1731 
1732 /**
1733  * Verify the @p attributes will be correctly understood by the NIC and store
1734  * them in the @p flow if everything is correct.
1735  *
1736  * @param[in] dev
1737  *   Pointer to the Ethernet device structure.
1738  * @param[in] attributes
1739  *   Pointer to flow attributes
1740  * @param[out] error
1741  *   Pointer to error structure.
1742  *
1743  * @return
1744  *   0 on success, a negative errno value otherwise and rte_errno is set.
1745  */
1746 int
1747 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1748 			      const struct rte_flow_attr *attributes,
1749 			      struct rte_flow_error *error)
1750 {
1751 	struct mlx5_priv *priv = dev->data->dev_private;
1752 	uint32_t priority_max = priv->config.flow_prio - 1;
1753 
1754 	if (attributes->group)
1755 		return rte_flow_error_set(error, ENOTSUP,
1756 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1757 					  NULL, "groups is not supported");
1758 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1759 	    attributes->priority >= priority_max)
1760 		return rte_flow_error_set(error, ENOTSUP,
1761 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1762 					  NULL, "priority out of range");
1763 	if (attributes->egress)
1764 		return rte_flow_error_set(error, ENOTSUP,
1765 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1766 					  "egress is not supported");
1767 	if (attributes->transfer && !priv->config.dv_esw_en)
1768 		return rte_flow_error_set(error, ENOTSUP,
1769 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1770 					  NULL, "transfer is not supported");
1771 	if (!attributes->ingress)
1772 		return rte_flow_error_set(error, EINVAL,
1773 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1774 					  NULL,
1775 					  "ingress attribute is mandatory");
1776 	return 0;
1777 }
1778 
1779 /**
1780  * Validate ICMP6 item.
1781  *
1782  * @param[in] item
1783  *   Item specification.
1784  * @param[in] item_flags
1785  *   Bit-fields that holds the items detected until now.
1786  * @param[in] ext_vlan_sup
1787  *   Whether extended VLAN features are supported or not.
1788  * @param[out] error
1789  *   Pointer to error structure.
1790  *
1791  * @return
1792  *   0 on success, a negative errno value otherwise and rte_errno is set.
1793  */
1794 int
1795 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1796 			       uint64_t item_flags,
1797 			       uint8_t target_protocol,
1798 			       struct rte_flow_error *error)
1799 {
1800 	const struct rte_flow_item_icmp6 *mask = item->mask;
1801 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1802 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1803 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1804 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1805 				      MLX5_FLOW_LAYER_OUTER_L4;
1806 	int ret;
1807 
1808 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1809 		return rte_flow_error_set(error, EINVAL,
1810 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1811 					  "protocol filtering not compatible"
1812 					  " with ICMP6 layer");
1813 	if (!(item_flags & l3m))
1814 		return rte_flow_error_set(error, EINVAL,
1815 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1816 					  "IPv6 is mandatory to filter on"
1817 					  " ICMP6");
1818 	if (item_flags & l4m)
1819 		return rte_flow_error_set(error, EINVAL,
1820 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1821 					  "multiple L4 layers not supported");
1822 	if (!mask)
1823 		mask = &rte_flow_item_icmp6_mask;
1824 	ret = mlx5_flow_item_acceptable
1825 		(item, (const uint8_t *)mask,
1826 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1827 		 sizeof(struct rte_flow_item_icmp6),
1828 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1829 	if (ret < 0)
1830 		return ret;
1831 	return 0;
1832 }
1833 
1834 /**
1835  * Validate ICMP item.
1836  *
1837  * @param[in] item
1838  *   Item specification.
1839  * @param[in] item_flags
1840  *   Bit-fields that holds the items detected until now.
1841  * @param[out] error
1842  *   Pointer to error structure.
1843  *
1844  * @return
1845  *   0 on success, a negative errno value otherwise and rte_errno is set.
1846  */
1847 int
1848 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1849 			     uint64_t item_flags,
1850 			     uint8_t target_protocol,
1851 			     struct rte_flow_error *error)
1852 {
1853 	const struct rte_flow_item_icmp *mask = item->mask;
1854 	const struct rte_flow_item_icmp nic_mask = {
1855 		.hdr.icmp_type = 0xff,
1856 		.hdr.icmp_code = 0xff,
1857 		.hdr.icmp_ident = RTE_BE16(0xffff),
1858 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1859 	};
1860 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1861 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1862 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1863 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1864 				      MLX5_FLOW_LAYER_OUTER_L4;
1865 	int ret;
1866 
1867 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1868 		return rte_flow_error_set(error, EINVAL,
1869 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1870 					  "protocol filtering not compatible"
1871 					  " with ICMP layer");
1872 	if (!(item_flags & l3m))
1873 		return rte_flow_error_set(error, EINVAL,
1874 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1875 					  "IPv4 is mandatory to filter"
1876 					  " on ICMP");
1877 	if (item_flags & l4m)
1878 		return rte_flow_error_set(error, EINVAL,
1879 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1880 					  "multiple L4 layers not supported");
1881 	if (!mask)
1882 		mask = &nic_mask;
1883 	ret = mlx5_flow_item_acceptable
1884 		(item, (const uint8_t *)mask,
1885 		 (const uint8_t *)&nic_mask,
1886 		 sizeof(struct rte_flow_item_icmp),
1887 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1888 	if (ret < 0)
1889 		return ret;
1890 	return 0;
1891 }
1892 
1893 /**
1894  * Validate Ethernet item.
1895  *
1896  * @param[in] item
1897  *   Item specification.
1898  * @param[in] item_flags
1899  *   Bit-fields that holds the items detected until now.
1900  * @param[out] error
1901  *   Pointer to error structure.
1902  *
1903  * @return
1904  *   0 on success, a negative errno value otherwise and rte_errno is set.
1905  */
1906 int
1907 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1908 			    uint64_t item_flags, bool ext_vlan_sup,
1909 			    struct rte_flow_error *error)
1910 {
1911 	const struct rte_flow_item_eth *mask = item->mask;
1912 	const struct rte_flow_item_eth nic_mask = {
1913 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1914 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1915 		.type = RTE_BE16(0xffff),
1916 		.has_vlan = ext_vlan_sup ? 1 : 0,
1917 	};
1918 	int ret;
1919 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1920 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1921 				       MLX5_FLOW_LAYER_OUTER_L2;
1922 
1923 	if (item_flags & ethm)
1924 		return rte_flow_error_set(error, ENOTSUP,
1925 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1926 					  "multiple L2 layers not supported");
1927 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1928 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1929 		return rte_flow_error_set(error, EINVAL,
1930 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1931 					  "L2 layer should not follow "
1932 					  "L3 layers");
1933 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1934 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1935 		return rte_flow_error_set(error, EINVAL,
1936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1937 					  "L2 layer should not follow VLAN");
1938 	if (!mask)
1939 		mask = &rte_flow_item_eth_mask;
1940 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1941 					(const uint8_t *)&nic_mask,
1942 					sizeof(struct rte_flow_item_eth),
1943 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1944 	return ret;
1945 }
1946 
1947 /**
1948  * Validate VLAN item.
1949  *
1950  * @param[in] item
1951  *   Item specification.
1952  * @param[in] item_flags
1953  *   Bit-fields that holds the items detected until now.
1954  * @param[in] dev
1955  *   Ethernet device flow is being created on.
1956  * @param[out] error
1957  *   Pointer to error structure.
1958  *
1959  * @return
1960  *   0 on success, a negative errno value otherwise and rte_errno is set.
1961  */
1962 int
1963 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1964 			     uint64_t item_flags,
1965 			     struct rte_eth_dev *dev,
1966 			     struct rte_flow_error *error)
1967 {
1968 	const struct rte_flow_item_vlan *spec = item->spec;
1969 	const struct rte_flow_item_vlan *mask = item->mask;
1970 	const struct rte_flow_item_vlan nic_mask = {
1971 		.tci = RTE_BE16(UINT16_MAX),
1972 		.inner_type = RTE_BE16(UINT16_MAX),
1973 	};
1974 	uint16_t vlan_tag = 0;
1975 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1976 	int ret;
1977 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1978 					MLX5_FLOW_LAYER_INNER_L4) :
1979 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1980 					MLX5_FLOW_LAYER_OUTER_L4);
1981 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1982 					MLX5_FLOW_LAYER_OUTER_VLAN;
1983 
1984 	if (item_flags & vlanm)
1985 		return rte_flow_error_set(error, EINVAL,
1986 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1987 					  "multiple VLAN layers not supported");
1988 	else if ((item_flags & l34m) != 0)
1989 		return rte_flow_error_set(error, EINVAL,
1990 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1991 					  "VLAN cannot follow L3/L4 layer");
1992 	if (!mask)
1993 		mask = &rte_flow_item_vlan_mask;
1994 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1995 					(const uint8_t *)&nic_mask,
1996 					sizeof(struct rte_flow_item_vlan),
1997 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1998 	if (ret)
1999 		return ret;
2000 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2001 		struct mlx5_priv *priv = dev->data->dev_private;
2002 
2003 		if (priv->vmwa_context) {
2004 			/*
2005 			 * Non-NULL context means we have a virtual machine
2006 			 * and SR-IOV enabled, we have to create VLAN interface
2007 			 * to make hypervisor to setup E-Switch vport
2008 			 * context correctly. We avoid creating the multiple
2009 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2010 			 */
2011 			return rte_flow_error_set(error, EINVAL,
2012 						  RTE_FLOW_ERROR_TYPE_ITEM,
2013 						  item,
2014 						  "VLAN tag mask is not"
2015 						  " supported in virtual"
2016 						  " environment");
2017 		}
2018 	}
2019 	if (spec) {
2020 		vlan_tag = spec->tci;
2021 		vlan_tag &= mask->tci;
2022 	}
2023 	/*
2024 	 * From verbs perspective an empty VLAN is equivalent
2025 	 * to a packet without VLAN layer.
2026 	 */
2027 	if (!vlan_tag)
2028 		return rte_flow_error_set(error, EINVAL,
2029 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2030 					  item->spec,
2031 					  "VLAN cannot be empty");
2032 	return 0;
2033 }
2034 
2035 /**
2036  * Validate IPV4 item.
2037  *
2038  * @param[in] item
2039  *   Item specification.
2040  * @param[in] item_flags
2041  *   Bit-fields that holds the items detected until now.
2042  * @param[in] last_item
2043  *   Previous validated item in the pattern items.
2044  * @param[in] ether_type
2045  *   Type in the ethernet layer header (including dot1q).
2046  * @param[in] acc_mask
2047  *   Acceptable mask, if NULL default internal default mask
2048  *   will be used to check whether item fields are supported.
2049  * @param[in] range_accepted
2050  *   True if range of values is accepted for specific fields, false otherwise.
2051  * @param[out] error
2052  *   Pointer to error structure.
2053  *
2054  * @return
2055  *   0 on success, a negative errno value otherwise and rte_errno is set.
2056  */
2057 int
2058 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2059 			     uint64_t item_flags,
2060 			     uint64_t last_item,
2061 			     uint16_t ether_type,
2062 			     const struct rte_flow_item_ipv4 *acc_mask,
2063 			     bool range_accepted,
2064 			     struct rte_flow_error *error)
2065 {
2066 	const struct rte_flow_item_ipv4 *mask = item->mask;
2067 	const struct rte_flow_item_ipv4 *spec = item->spec;
2068 	const struct rte_flow_item_ipv4 nic_mask = {
2069 		.hdr = {
2070 			.src_addr = RTE_BE32(0xffffffff),
2071 			.dst_addr = RTE_BE32(0xffffffff),
2072 			.type_of_service = 0xff,
2073 			.next_proto_id = 0xff,
2074 		},
2075 	};
2076 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2077 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2078 				      MLX5_FLOW_LAYER_OUTER_L3;
2079 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2080 				      MLX5_FLOW_LAYER_OUTER_L4;
2081 	int ret;
2082 	uint8_t next_proto = 0xFF;
2083 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2084 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2085 				  MLX5_FLOW_LAYER_INNER_VLAN);
2086 
2087 	if ((last_item & l2_vlan) && ether_type &&
2088 	    ether_type != RTE_ETHER_TYPE_IPV4)
2089 		return rte_flow_error_set(error, EINVAL,
2090 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2091 					  "IPv4 cannot follow L2/VLAN layer "
2092 					  "which ether type is not IPv4");
2093 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2094 		if (mask && spec)
2095 			next_proto = mask->hdr.next_proto_id &
2096 				     spec->hdr.next_proto_id;
2097 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2098 			return rte_flow_error_set(error, EINVAL,
2099 						  RTE_FLOW_ERROR_TYPE_ITEM,
2100 						  item,
2101 						  "multiple tunnel "
2102 						  "not supported");
2103 	}
2104 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2105 		return rte_flow_error_set(error, EINVAL,
2106 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2107 					  "wrong tunnel type - IPv6 specified "
2108 					  "but IPv4 item provided");
2109 	if (item_flags & l3m)
2110 		return rte_flow_error_set(error, ENOTSUP,
2111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2112 					  "multiple L3 layers not supported");
2113 	else if (item_flags & l4m)
2114 		return rte_flow_error_set(error, EINVAL,
2115 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2116 					  "L3 cannot follow an L4 layer.");
2117 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2118 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2119 		return rte_flow_error_set(error, EINVAL,
2120 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2121 					  "L3 cannot follow an NVGRE layer.");
2122 	if (!mask)
2123 		mask = &rte_flow_item_ipv4_mask;
2124 	else if (mask->hdr.next_proto_id != 0 &&
2125 		 mask->hdr.next_proto_id != 0xff)
2126 		return rte_flow_error_set(error, EINVAL,
2127 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2128 					  "partial mask is not supported"
2129 					  " for protocol");
2130 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2131 					acc_mask ? (const uint8_t *)acc_mask
2132 						 : (const uint8_t *)&nic_mask,
2133 					sizeof(struct rte_flow_item_ipv4),
2134 					range_accepted, error);
2135 	if (ret < 0)
2136 		return ret;
2137 	return 0;
2138 }
2139 
2140 /**
2141  * Validate IPV6 item.
2142  *
2143  * @param[in] item
2144  *   Item specification.
2145  * @param[in] item_flags
2146  *   Bit-fields that holds the items detected until now.
2147  * @param[in] last_item
2148  *   Previous validated item in the pattern items.
2149  * @param[in] ether_type
2150  *   Type in the ethernet layer header (including dot1q).
2151  * @param[in] acc_mask
2152  *   Acceptable mask, if NULL default internal default mask
2153  *   will be used to check whether item fields are supported.
2154  * @param[out] error
2155  *   Pointer to error structure.
2156  *
2157  * @return
2158  *   0 on success, a negative errno value otherwise and rte_errno is set.
2159  */
2160 int
2161 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2162 			     uint64_t item_flags,
2163 			     uint64_t last_item,
2164 			     uint16_t ether_type,
2165 			     const struct rte_flow_item_ipv6 *acc_mask,
2166 			     struct rte_flow_error *error)
2167 {
2168 	const struct rte_flow_item_ipv6 *mask = item->mask;
2169 	const struct rte_flow_item_ipv6 *spec = item->spec;
2170 	const struct rte_flow_item_ipv6 nic_mask = {
2171 		.hdr = {
2172 			.src_addr =
2173 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2174 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2175 			.dst_addr =
2176 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2177 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2178 			.vtc_flow = RTE_BE32(0xffffffff),
2179 			.proto = 0xff,
2180 		},
2181 	};
2182 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2183 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2184 				      MLX5_FLOW_LAYER_OUTER_L3;
2185 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2186 				      MLX5_FLOW_LAYER_OUTER_L4;
2187 	int ret;
2188 	uint8_t next_proto = 0xFF;
2189 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2190 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2191 				  MLX5_FLOW_LAYER_INNER_VLAN);
2192 
2193 	if ((last_item & l2_vlan) && ether_type &&
2194 	    ether_type != RTE_ETHER_TYPE_IPV6)
2195 		return rte_flow_error_set(error, EINVAL,
2196 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 					  "IPv6 cannot follow L2/VLAN layer "
2198 					  "which ether type is not IPv6");
2199 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2200 		next_proto = spec->hdr.proto;
2201 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2202 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2203 			return rte_flow_error_set(error, EINVAL,
2204 						  RTE_FLOW_ERROR_TYPE_ITEM,
2205 						  item,
2206 						  "multiple tunnel "
2207 						  "not supported");
2208 	}
2209 	if (next_proto == IPPROTO_HOPOPTS  ||
2210 	    next_proto == IPPROTO_ROUTING  ||
2211 	    next_proto == IPPROTO_FRAGMENT ||
2212 	    next_proto == IPPROTO_ESP	   ||
2213 	    next_proto == IPPROTO_AH	   ||
2214 	    next_proto == IPPROTO_DSTOPTS)
2215 		return rte_flow_error_set(error, EINVAL,
2216 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2217 					  "IPv6 proto (next header) should "
2218 					  "not be set as extension header");
2219 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2220 		return rte_flow_error_set(error, EINVAL,
2221 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 					  "wrong tunnel type - IPv4 specified "
2223 					  "but IPv6 item provided");
2224 	if (item_flags & l3m)
2225 		return rte_flow_error_set(error, ENOTSUP,
2226 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2227 					  "multiple L3 layers not supported");
2228 	else if (item_flags & l4m)
2229 		return rte_flow_error_set(error, EINVAL,
2230 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2231 					  "L3 cannot follow an L4 layer.");
2232 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2233 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2234 		return rte_flow_error_set(error, EINVAL,
2235 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2236 					  "L3 cannot follow an NVGRE layer.");
2237 	if (!mask)
2238 		mask = &rte_flow_item_ipv6_mask;
2239 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2240 					acc_mask ? (const uint8_t *)acc_mask
2241 						 : (const uint8_t *)&nic_mask,
2242 					sizeof(struct rte_flow_item_ipv6),
2243 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2244 	if (ret < 0)
2245 		return ret;
2246 	return 0;
2247 }
2248 
2249 /**
2250  * Validate UDP item.
2251  *
2252  * @param[in] item
2253  *   Item specification.
2254  * @param[in] item_flags
2255  *   Bit-fields that holds the items detected until now.
2256  * @param[in] target_protocol
2257  *   The next protocol in the previous item.
2258  * @param[in] flow_mask
2259  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2260  * @param[out] error
2261  *   Pointer to error structure.
2262  *
2263  * @return
2264  *   0 on success, a negative errno value otherwise and rte_errno is set.
2265  */
2266 int
2267 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2268 			    uint64_t item_flags,
2269 			    uint8_t target_protocol,
2270 			    struct rte_flow_error *error)
2271 {
2272 	const struct rte_flow_item_udp *mask = item->mask;
2273 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2274 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2275 				      MLX5_FLOW_LAYER_OUTER_L3;
2276 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2277 				      MLX5_FLOW_LAYER_OUTER_L4;
2278 	int ret;
2279 
2280 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2281 		return rte_flow_error_set(error, EINVAL,
2282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2283 					  "protocol filtering not compatible"
2284 					  " with UDP layer");
2285 	if (!(item_flags & l3m))
2286 		return rte_flow_error_set(error, EINVAL,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "L3 is mandatory to filter on L4");
2289 	if (item_flags & l4m)
2290 		return rte_flow_error_set(error, EINVAL,
2291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2292 					  "multiple L4 layers not supported");
2293 	if (!mask)
2294 		mask = &rte_flow_item_udp_mask;
2295 	ret = mlx5_flow_item_acceptable
2296 		(item, (const uint8_t *)mask,
2297 		 (const uint8_t *)&rte_flow_item_udp_mask,
2298 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2299 		 error);
2300 	if (ret < 0)
2301 		return ret;
2302 	return 0;
2303 }
2304 
2305 /**
2306  * Validate TCP item.
2307  *
2308  * @param[in] item
2309  *   Item specification.
2310  * @param[in] item_flags
2311  *   Bit-fields that holds the items detected until now.
2312  * @param[in] target_protocol
2313  *   The next protocol in the previous item.
2314  * @param[out] error
2315  *   Pointer to error structure.
2316  *
2317  * @return
2318  *   0 on success, a negative errno value otherwise and rte_errno is set.
2319  */
2320 int
2321 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2322 			    uint64_t item_flags,
2323 			    uint8_t target_protocol,
2324 			    const struct rte_flow_item_tcp *flow_mask,
2325 			    struct rte_flow_error *error)
2326 {
2327 	const struct rte_flow_item_tcp *mask = item->mask;
2328 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2329 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2330 				      MLX5_FLOW_LAYER_OUTER_L3;
2331 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2332 				      MLX5_FLOW_LAYER_OUTER_L4;
2333 	int ret;
2334 
2335 	MLX5_ASSERT(flow_mask);
2336 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2337 		return rte_flow_error_set(error, EINVAL,
2338 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2339 					  "protocol filtering not compatible"
2340 					  " with TCP layer");
2341 	if (!(item_flags & l3m))
2342 		return rte_flow_error_set(error, EINVAL,
2343 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2344 					  "L3 is mandatory to filter on L4");
2345 	if (item_flags & l4m)
2346 		return rte_flow_error_set(error, EINVAL,
2347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2348 					  "multiple L4 layers not supported");
2349 	if (!mask)
2350 		mask = &rte_flow_item_tcp_mask;
2351 	ret = mlx5_flow_item_acceptable
2352 		(item, (const uint8_t *)mask,
2353 		 (const uint8_t *)flow_mask,
2354 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2355 		 error);
2356 	if (ret < 0)
2357 		return ret;
2358 	return 0;
2359 }
2360 
2361 /**
2362  * Validate VXLAN item.
2363  *
2364  * @param[in] item
2365  *   Item specification.
2366  * @param[in] item_flags
2367  *   Bit-fields that holds the items detected until now.
2368  * @param[in] target_protocol
2369  *   The next protocol in the previous item.
2370  * @param[out] error
2371  *   Pointer to error structure.
2372  *
2373  * @return
2374  *   0 on success, a negative errno value otherwise and rte_errno is set.
2375  */
2376 int
2377 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2378 			      uint64_t item_flags,
2379 			      struct rte_flow_error *error)
2380 {
2381 	const struct rte_flow_item_vxlan *spec = item->spec;
2382 	const struct rte_flow_item_vxlan *mask = item->mask;
2383 	int ret;
2384 	union vni {
2385 		uint32_t vlan_id;
2386 		uint8_t vni[4];
2387 	} id = { .vlan_id = 0, };
2388 
2389 
2390 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2391 		return rte_flow_error_set(error, ENOTSUP,
2392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2393 					  "multiple tunnel layers not"
2394 					  " supported");
2395 	/*
2396 	 * Verify only UDPv4 is present as defined in
2397 	 * https://tools.ietf.org/html/rfc7348
2398 	 */
2399 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2400 		return rte_flow_error_set(error, EINVAL,
2401 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2402 					  "no outer UDP layer found");
2403 	if (!mask)
2404 		mask = &rte_flow_item_vxlan_mask;
2405 	ret = mlx5_flow_item_acceptable
2406 		(item, (const uint8_t *)mask,
2407 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2408 		 sizeof(struct rte_flow_item_vxlan),
2409 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2410 	if (ret < 0)
2411 		return ret;
2412 	if (spec) {
2413 		memcpy(&id.vni[1], spec->vni, 3);
2414 		memcpy(&id.vni[1], mask->vni, 3);
2415 	}
2416 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2417 		return rte_flow_error_set(error, ENOTSUP,
2418 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 					  "VXLAN tunnel must be fully defined");
2420 	return 0;
2421 }
2422 
2423 /**
2424  * Validate VXLAN_GPE item.
2425  *
2426  * @param[in] item
2427  *   Item specification.
2428  * @param[in] item_flags
2429  *   Bit-fields that holds the items detected until now.
2430  * @param[in] priv
2431  *   Pointer to the private data structure.
2432  * @param[in] target_protocol
2433  *   The next protocol in the previous item.
2434  * @param[out] error
2435  *   Pointer to error structure.
2436  *
2437  * @return
2438  *   0 on success, a negative errno value otherwise and rte_errno is set.
2439  */
2440 int
2441 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2442 				  uint64_t item_flags,
2443 				  struct rte_eth_dev *dev,
2444 				  struct rte_flow_error *error)
2445 {
2446 	struct mlx5_priv *priv = dev->data->dev_private;
2447 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2448 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2449 	int ret;
2450 	union vni {
2451 		uint32_t vlan_id;
2452 		uint8_t vni[4];
2453 	} id = { .vlan_id = 0, };
2454 
2455 	if (!priv->config.l3_vxlan_en)
2456 		return rte_flow_error_set(error, ENOTSUP,
2457 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2458 					  "L3 VXLAN is not enabled by device"
2459 					  " parameter and/or not configured in"
2460 					  " firmware");
2461 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2462 		return rte_flow_error_set(error, ENOTSUP,
2463 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2464 					  "multiple tunnel layers not"
2465 					  " supported");
2466 	/*
2467 	 * Verify only UDPv4 is present as defined in
2468 	 * https://tools.ietf.org/html/rfc7348
2469 	 */
2470 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2471 		return rte_flow_error_set(error, EINVAL,
2472 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2473 					  "no outer UDP layer found");
2474 	if (!mask)
2475 		mask = &rte_flow_item_vxlan_gpe_mask;
2476 	ret = mlx5_flow_item_acceptable
2477 		(item, (const uint8_t *)mask,
2478 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2479 		 sizeof(struct rte_flow_item_vxlan_gpe),
2480 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2481 	if (ret < 0)
2482 		return ret;
2483 	if (spec) {
2484 		if (spec->protocol)
2485 			return rte_flow_error_set(error, ENOTSUP,
2486 						  RTE_FLOW_ERROR_TYPE_ITEM,
2487 						  item,
2488 						  "VxLAN-GPE protocol"
2489 						  " not supported");
2490 		memcpy(&id.vni[1], spec->vni, 3);
2491 		memcpy(&id.vni[1], mask->vni, 3);
2492 	}
2493 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2494 		return rte_flow_error_set(error, ENOTSUP,
2495 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2496 					  "VXLAN-GPE tunnel must be fully"
2497 					  " defined");
2498 	return 0;
2499 }
2500 /**
2501  * Validate GRE Key item.
2502  *
2503  * @param[in] item
2504  *   Item specification.
2505  * @param[in] item_flags
2506  *   Bit flags to mark detected items.
2507  * @param[in] gre_item
2508  *   Pointer to gre_item
2509  * @param[out] error
2510  *   Pointer to error structure.
2511  *
2512  * @return
2513  *   0 on success, a negative errno value otherwise and rte_errno is set.
2514  */
2515 int
2516 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2517 				uint64_t item_flags,
2518 				const struct rte_flow_item *gre_item,
2519 				struct rte_flow_error *error)
2520 {
2521 	const rte_be32_t *mask = item->mask;
2522 	int ret = 0;
2523 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2524 	const struct rte_flow_item_gre *gre_spec;
2525 	const struct rte_flow_item_gre *gre_mask;
2526 
2527 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2528 		return rte_flow_error_set(error, ENOTSUP,
2529 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2530 					  "Multiple GRE key not support");
2531 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2532 		return rte_flow_error_set(error, ENOTSUP,
2533 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2534 					  "No preceding GRE header");
2535 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2536 		return rte_flow_error_set(error, ENOTSUP,
2537 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2538 					  "GRE key following a wrong item");
2539 	gre_mask = gre_item->mask;
2540 	if (!gre_mask)
2541 		gre_mask = &rte_flow_item_gre_mask;
2542 	gre_spec = gre_item->spec;
2543 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2544 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2545 		return rte_flow_error_set(error, EINVAL,
2546 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2547 					  "Key bit must be on");
2548 
2549 	if (!mask)
2550 		mask = &gre_key_default_mask;
2551 	ret = mlx5_flow_item_acceptable
2552 		(item, (const uint8_t *)mask,
2553 		 (const uint8_t *)&gre_key_default_mask,
2554 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2555 	return ret;
2556 }
2557 
2558 /**
2559  * Validate GRE item.
2560  *
2561  * @param[in] item
2562  *   Item specification.
2563  * @param[in] item_flags
2564  *   Bit flags to mark detected items.
2565  * @param[in] target_protocol
2566  *   The next protocol in the previous item.
2567  * @param[out] error
2568  *   Pointer to error structure.
2569  *
2570  * @return
2571  *   0 on success, a negative errno value otherwise and rte_errno is set.
2572  */
2573 int
2574 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2575 			    uint64_t item_flags,
2576 			    uint8_t target_protocol,
2577 			    struct rte_flow_error *error)
2578 {
2579 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2580 	const struct rte_flow_item_gre *mask = item->mask;
2581 	int ret;
2582 	const struct rte_flow_item_gre nic_mask = {
2583 		.c_rsvd0_ver = RTE_BE16(0xB000),
2584 		.protocol = RTE_BE16(UINT16_MAX),
2585 	};
2586 
2587 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2588 		return rte_flow_error_set(error, EINVAL,
2589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2590 					  "protocol filtering not compatible"
2591 					  " with this GRE layer");
2592 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2593 		return rte_flow_error_set(error, ENOTSUP,
2594 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2595 					  "multiple tunnel layers not"
2596 					  " supported");
2597 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2598 		return rte_flow_error_set(error, ENOTSUP,
2599 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2600 					  "L3 Layer is missing");
2601 	if (!mask)
2602 		mask = &rte_flow_item_gre_mask;
2603 	ret = mlx5_flow_item_acceptable
2604 		(item, (const uint8_t *)mask,
2605 		 (const uint8_t *)&nic_mask,
2606 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2607 		 error);
2608 	if (ret < 0)
2609 		return ret;
2610 #ifndef HAVE_MLX5DV_DR
2611 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2612 	if (spec && (spec->protocol & mask->protocol))
2613 		return rte_flow_error_set(error, ENOTSUP,
2614 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2615 					  "without MPLS support the"
2616 					  " specification cannot be used for"
2617 					  " filtering");
2618 #endif
2619 #endif
2620 	return 0;
2621 }
2622 
2623 /**
2624  * Validate Geneve item.
2625  *
2626  * @param[in] item
2627  *   Item specification.
2628  * @param[in] itemFlags
2629  *   Bit-fields that holds the items detected until now.
2630  * @param[in] enPriv
2631  *   Pointer to the private data structure.
2632  * @param[out] error
2633  *   Pointer to error structure.
2634  *
2635  * @return
2636  *   0 on success, a negative errno value otherwise and rte_errno is set.
2637  */
2638 
2639 int
2640 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2641 			       uint64_t item_flags,
2642 			       struct rte_eth_dev *dev,
2643 			       struct rte_flow_error *error)
2644 {
2645 	struct mlx5_priv *priv = dev->data->dev_private;
2646 	const struct rte_flow_item_geneve *spec = item->spec;
2647 	const struct rte_flow_item_geneve *mask = item->mask;
2648 	int ret;
2649 	uint16_t gbhdr;
2650 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2651 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2652 	const struct rte_flow_item_geneve nic_mask = {
2653 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2654 		.vni = "\xff\xff\xff",
2655 		.protocol = RTE_BE16(UINT16_MAX),
2656 	};
2657 
2658 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2659 		return rte_flow_error_set(error, ENOTSUP,
2660 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2661 					  "L3 Geneve is not enabled by device"
2662 					  " parameter and/or not configured in"
2663 					  " firmware");
2664 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2665 		return rte_flow_error_set(error, ENOTSUP,
2666 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2667 					  "multiple tunnel layers not"
2668 					  " supported");
2669 	/*
2670 	 * Verify only UDPv4 is present as defined in
2671 	 * https://tools.ietf.org/html/rfc7348
2672 	 */
2673 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2674 		return rte_flow_error_set(error, EINVAL,
2675 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2676 					  "no outer UDP layer found");
2677 	if (!mask)
2678 		mask = &rte_flow_item_geneve_mask;
2679 	ret = mlx5_flow_item_acceptable
2680 				  (item, (const uint8_t *)mask,
2681 				   (const uint8_t *)&nic_mask,
2682 				   sizeof(struct rte_flow_item_geneve),
2683 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2684 	if (ret)
2685 		return ret;
2686 	if (spec) {
2687 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2688 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2689 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2690 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2691 			return rte_flow_error_set(error, ENOTSUP,
2692 						  RTE_FLOW_ERROR_TYPE_ITEM,
2693 						  item,
2694 						  "Geneve protocol unsupported"
2695 						  " fields are being used");
2696 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2697 			return rte_flow_error_set
2698 					(error, ENOTSUP,
2699 					 RTE_FLOW_ERROR_TYPE_ITEM,
2700 					 item,
2701 					 "Unsupported Geneve options length");
2702 	}
2703 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2704 		return rte_flow_error_set
2705 				    (error, ENOTSUP,
2706 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2707 				     "Geneve tunnel must be fully defined");
2708 	return 0;
2709 }
2710 
2711 /**
2712  * Validate MPLS item.
2713  *
2714  * @param[in] dev
2715  *   Pointer to the rte_eth_dev structure.
2716  * @param[in] item
2717  *   Item specification.
2718  * @param[in] item_flags
2719  *   Bit-fields that holds the items detected until now.
2720  * @param[in] prev_layer
2721  *   The protocol layer indicated in previous item.
2722  * @param[out] error
2723  *   Pointer to error structure.
2724  *
2725  * @return
2726  *   0 on success, a negative errno value otherwise and rte_errno is set.
2727  */
2728 int
2729 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2730 			     const struct rte_flow_item *item __rte_unused,
2731 			     uint64_t item_flags __rte_unused,
2732 			     uint64_t prev_layer __rte_unused,
2733 			     struct rte_flow_error *error)
2734 {
2735 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2736 	const struct rte_flow_item_mpls *mask = item->mask;
2737 	struct mlx5_priv *priv = dev->data->dev_private;
2738 	int ret;
2739 
2740 	if (!priv->config.mpls_en)
2741 		return rte_flow_error_set(error, ENOTSUP,
2742 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2743 					  "MPLS not supported or"
2744 					  " disabled in firmware"
2745 					  " configuration.");
2746 	/* MPLS over IP, UDP, GRE is allowed */
2747 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2748 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2749 			    MLX5_FLOW_LAYER_GRE)))
2750 		return rte_flow_error_set(error, EINVAL,
2751 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2752 					  "protocol filtering not compatible"
2753 					  " with MPLS layer");
2754 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2755 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2756 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2757 		return rte_flow_error_set(error, ENOTSUP,
2758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2759 					  "multiple tunnel layers not"
2760 					  " supported");
2761 	if (!mask)
2762 		mask = &rte_flow_item_mpls_mask;
2763 	ret = mlx5_flow_item_acceptable
2764 		(item, (const uint8_t *)mask,
2765 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2766 		 sizeof(struct rte_flow_item_mpls),
2767 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2768 	if (ret < 0)
2769 		return ret;
2770 	return 0;
2771 #else
2772 	return rte_flow_error_set(error, ENOTSUP,
2773 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2774 				  "MPLS is not supported by Verbs, please"
2775 				  " update.");
2776 #endif
2777 }
2778 
2779 /**
2780  * Validate NVGRE item.
2781  *
2782  * @param[in] item
2783  *   Item specification.
2784  * @param[in] item_flags
2785  *   Bit flags to mark detected items.
2786  * @param[in] target_protocol
2787  *   The next protocol in the previous item.
2788  * @param[out] error
2789  *   Pointer to error structure.
2790  *
2791  * @return
2792  *   0 on success, a negative errno value otherwise and rte_errno is set.
2793  */
2794 int
2795 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2796 			      uint64_t item_flags,
2797 			      uint8_t target_protocol,
2798 			      struct rte_flow_error *error)
2799 {
2800 	const struct rte_flow_item_nvgre *mask = item->mask;
2801 	int ret;
2802 
2803 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2804 		return rte_flow_error_set(error, EINVAL,
2805 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2806 					  "protocol filtering not compatible"
2807 					  " with this GRE layer");
2808 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2809 		return rte_flow_error_set(error, ENOTSUP,
2810 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2811 					  "multiple tunnel layers not"
2812 					  " supported");
2813 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2814 		return rte_flow_error_set(error, ENOTSUP,
2815 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2816 					  "L3 Layer is missing");
2817 	if (!mask)
2818 		mask = &rte_flow_item_nvgre_mask;
2819 	ret = mlx5_flow_item_acceptable
2820 		(item, (const uint8_t *)mask,
2821 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2822 		 sizeof(struct rte_flow_item_nvgre),
2823 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2824 	if (ret < 0)
2825 		return ret;
2826 	return 0;
2827 }
2828 
2829 /**
2830  * Validate eCPRI item.
2831  *
2832  * @param[in] item
2833  *   Item specification.
2834  * @param[in] item_flags
2835  *   Bit-fields that holds the items detected until now.
2836  * @param[in] last_item
2837  *   Previous validated item in the pattern items.
2838  * @param[in] ether_type
2839  *   Type in the ethernet layer header (including dot1q).
2840  * @param[in] acc_mask
2841  *   Acceptable mask, if NULL default internal default mask
2842  *   will be used to check whether item fields are supported.
2843  * @param[out] error
2844  *   Pointer to error structure.
2845  *
2846  * @return
2847  *   0 on success, a negative errno value otherwise and rte_errno is set.
2848  */
2849 int
2850 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2851 			      uint64_t item_flags,
2852 			      uint64_t last_item,
2853 			      uint16_t ether_type,
2854 			      const struct rte_flow_item_ecpri *acc_mask,
2855 			      struct rte_flow_error *error)
2856 {
2857 	const struct rte_flow_item_ecpri *mask = item->mask;
2858 	const struct rte_flow_item_ecpri nic_mask = {
2859 		.hdr = {
2860 			.common = {
2861 				.u32 =
2862 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2863 					.type = 0xFF,
2864 					}).u32),
2865 			},
2866 			.dummy[0] = 0xFFFFFFFF,
2867 		},
2868 	};
2869 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2870 					MLX5_FLOW_LAYER_OUTER_VLAN);
2871 	struct rte_flow_item_ecpri mask_lo;
2872 
2873 	if ((last_item & outer_l2_vlan) && ether_type &&
2874 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2875 		return rte_flow_error_set(error, EINVAL,
2876 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2877 					  "eCPRI cannot follow L2/VLAN layer "
2878 					  "which ether type is not 0xAEFE.");
2879 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2880 		return rte_flow_error_set(error, EINVAL,
2881 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2882 					  "eCPRI with tunnel is not supported "
2883 					  "right now.");
2884 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2885 		return rte_flow_error_set(error, ENOTSUP,
2886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2887 					  "multiple L3 layers not supported");
2888 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2889 		return rte_flow_error_set(error, EINVAL,
2890 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2891 					  "eCPRI cannot follow a TCP layer.");
2892 	/* In specification, eCPRI could be over UDP layer. */
2893 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2894 		return rte_flow_error_set(error, EINVAL,
2895 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2896 					  "eCPRI over UDP layer is not yet "
2897 					  "supported right now.");
2898 	/* Mask for type field in common header could be zero. */
2899 	if (!mask)
2900 		mask = &rte_flow_item_ecpri_mask;
2901 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2902 	/* Input mask is in big-endian format. */
2903 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2904 		return rte_flow_error_set(error, EINVAL,
2905 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2906 					  "partial mask is not supported "
2907 					  "for protocol");
2908 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2909 		return rte_flow_error_set(error, EINVAL,
2910 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2911 					  "message header mask must be after "
2912 					  "a type mask");
2913 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2914 					 acc_mask ? (const uint8_t *)acc_mask
2915 						  : (const uint8_t *)&nic_mask,
2916 					 sizeof(struct rte_flow_item_ecpri),
2917 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2918 }
2919 
2920 /**
2921  * Release resource related QUEUE/RSS action split.
2922  *
2923  * @param dev
2924  *   Pointer to Ethernet device.
2925  * @param flow
2926  *   Flow to release id's from.
2927  */
2928 static void
2929 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2930 			     struct rte_flow *flow)
2931 {
2932 	struct mlx5_priv *priv = dev->data->dev_private;
2933 	uint32_t handle_idx;
2934 	struct mlx5_flow_handle *dev_handle;
2935 
2936 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2937 		       handle_idx, dev_handle, next)
2938 		if (dev_handle->split_flow_id)
2939 			mlx5_ipool_free(priv->sh->ipool
2940 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2941 					dev_handle->split_flow_id);
2942 }
2943 
2944 static int
2945 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2946 		   const struct rte_flow_attr *attr __rte_unused,
2947 		   const struct rte_flow_item items[] __rte_unused,
2948 		   const struct rte_flow_action actions[] __rte_unused,
2949 		   bool external __rte_unused,
2950 		   int hairpin __rte_unused,
2951 		   struct rte_flow_error *error)
2952 {
2953 	return rte_flow_error_set(error, ENOTSUP,
2954 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2955 }
2956 
2957 static struct mlx5_flow *
2958 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2959 		  const struct rte_flow_attr *attr __rte_unused,
2960 		  const struct rte_flow_item items[] __rte_unused,
2961 		  const struct rte_flow_action actions[] __rte_unused,
2962 		  struct rte_flow_error *error)
2963 {
2964 	rte_flow_error_set(error, ENOTSUP,
2965 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2966 	return NULL;
2967 }
2968 
2969 static int
2970 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2971 		    struct mlx5_flow *dev_flow __rte_unused,
2972 		    const struct rte_flow_attr *attr __rte_unused,
2973 		    const struct rte_flow_item items[] __rte_unused,
2974 		    const struct rte_flow_action actions[] __rte_unused,
2975 		    struct rte_flow_error *error)
2976 {
2977 	return rte_flow_error_set(error, ENOTSUP,
2978 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2979 }
2980 
2981 static int
2982 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2983 		struct rte_flow *flow __rte_unused,
2984 		struct rte_flow_error *error)
2985 {
2986 	return rte_flow_error_set(error, ENOTSUP,
2987 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2988 }
2989 
2990 static void
2991 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2992 		 struct rte_flow *flow __rte_unused)
2993 {
2994 }
2995 
2996 static void
2997 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2998 		  struct rte_flow *flow __rte_unused)
2999 {
3000 }
3001 
3002 static int
3003 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3004 		struct rte_flow *flow __rte_unused,
3005 		const struct rte_flow_action *actions __rte_unused,
3006 		void *data __rte_unused,
3007 		struct rte_flow_error *error)
3008 {
3009 	return rte_flow_error_set(error, ENOTSUP,
3010 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3011 }
3012 
3013 static int
3014 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3015 		      uint32_t domains __rte_unused,
3016 		      uint32_t flags __rte_unused)
3017 {
3018 	return 0;
3019 }
3020 
3021 /* Void driver to protect from null pointer reference. */
3022 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3023 	.validate = flow_null_validate,
3024 	.prepare = flow_null_prepare,
3025 	.translate = flow_null_translate,
3026 	.apply = flow_null_apply,
3027 	.remove = flow_null_remove,
3028 	.destroy = flow_null_destroy,
3029 	.query = flow_null_query,
3030 	.sync_domain = flow_null_sync_domain,
3031 };
3032 
3033 /**
3034  * Select flow driver type according to flow attributes and device
3035  * configuration.
3036  *
3037  * @param[in] dev
3038  *   Pointer to the dev structure.
3039  * @param[in] attr
3040  *   Pointer to the flow attributes.
3041  *
3042  * @return
3043  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3044  */
3045 static enum mlx5_flow_drv_type
3046 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3047 {
3048 	struct mlx5_priv *priv = dev->data->dev_private;
3049 	/* The OS can determine first a specific flow type (DV, VERBS) */
3050 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3051 
3052 	if (type != MLX5_FLOW_TYPE_MAX)
3053 		return type;
3054 	/* If no OS specific type - continue with DV/VERBS selection */
3055 	if (attr->transfer && priv->config.dv_esw_en)
3056 		type = MLX5_FLOW_TYPE_DV;
3057 	if (!attr->transfer)
3058 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3059 						 MLX5_FLOW_TYPE_VERBS;
3060 	return type;
3061 }
3062 
3063 #define flow_get_drv_ops(type) flow_drv_ops[type]
3064 
3065 /**
3066  * Flow driver validation API. This abstracts calling driver specific functions.
3067  * The type of flow driver is determined according to flow attributes.
3068  *
3069  * @param[in] dev
3070  *   Pointer to the dev structure.
3071  * @param[in] attr
3072  *   Pointer to the flow attributes.
3073  * @param[in] items
3074  *   Pointer to the list of items.
3075  * @param[in] actions
3076  *   Pointer to the list of actions.
3077  * @param[in] external
3078  *   This flow rule is created by request external to PMD.
3079  * @param[in] hairpin
3080  *   Number of hairpin TX actions, 0 means classic flow.
3081  * @param[out] error
3082  *   Pointer to the error structure.
3083  *
3084  * @return
3085  *   0 on success, a negative errno value otherwise and rte_errno is set.
3086  */
3087 static inline int
3088 flow_drv_validate(struct rte_eth_dev *dev,
3089 		  const struct rte_flow_attr *attr,
3090 		  const struct rte_flow_item items[],
3091 		  const struct rte_flow_action actions[],
3092 		  bool external, int hairpin, struct rte_flow_error *error)
3093 {
3094 	const struct mlx5_flow_driver_ops *fops;
3095 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3096 
3097 	fops = flow_get_drv_ops(type);
3098 	return fops->validate(dev, attr, items, actions, external,
3099 			      hairpin, error);
3100 }
3101 
3102 /**
3103  * Flow driver preparation API. This abstracts calling driver specific
3104  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3105  * calculates the size of memory required for device flow, allocates the memory,
3106  * initializes the device flow and returns the pointer.
3107  *
3108  * @note
3109  *   This function initializes device flow structure such as dv or verbs in
3110  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3111  *   rest. For example, adding returning device flow to flow->dev_flow list and
3112  *   setting backward reference to the flow should be done out of this function.
3113  *   layers field is not filled either.
3114  *
3115  * @param[in] dev
3116  *   Pointer to the dev structure.
3117  * @param[in] attr
3118  *   Pointer to the flow attributes.
3119  * @param[in] items
3120  *   Pointer to the list of items.
3121  * @param[in] actions
3122  *   Pointer to the list of actions.
3123  * @param[in] flow_idx
3124  *   This memory pool index to the flow.
3125  * @param[out] error
3126  *   Pointer to the error structure.
3127  *
3128  * @return
3129  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3130  */
3131 static inline struct mlx5_flow *
3132 flow_drv_prepare(struct rte_eth_dev *dev,
3133 		 const struct rte_flow *flow,
3134 		 const struct rte_flow_attr *attr,
3135 		 const struct rte_flow_item items[],
3136 		 const struct rte_flow_action actions[],
3137 		 uint32_t flow_idx,
3138 		 struct rte_flow_error *error)
3139 {
3140 	const struct mlx5_flow_driver_ops *fops;
3141 	enum mlx5_flow_drv_type type = flow->drv_type;
3142 	struct mlx5_flow *mlx5_flow = NULL;
3143 
3144 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3145 	fops = flow_get_drv_ops(type);
3146 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3147 	if (mlx5_flow)
3148 		mlx5_flow->flow_idx = flow_idx;
3149 	return mlx5_flow;
3150 }
3151 
3152 /**
3153  * Flow driver translation API. This abstracts calling driver specific
3154  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3155  * translates a generic flow into a driver flow. flow_drv_prepare() must
3156  * precede.
3157  *
3158  * @note
3159  *   dev_flow->layers could be filled as a result of parsing during translation
3160  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3161  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3162  *   flow->actions could be overwritten even though all the expanded dev_flows
3163  *   have the same actions.
3164  *
3165  * @param[in] dev
3166  *   Pointer to the rte dev structure.
3167  * @param[in, out] dev_flow
3168  *   Pointer to the mlx5 flow.
3169  * @param[in] attr
3170  *   Pointer to the flow attributes.
3171  * @param[in] items
3172  *   Pointer to the list of items.
3173  * @param[in] actions
3174  *   Pointer to the list of actions.
3175  * @param[out] error
3176  *   Pointer to the error structure.
3177  *
3178  * @return
3179  *   0 on success, a negative errno value otherwise and rte_errno is set.
3180  */
3181 static inline int
3182 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3183 		   const struct rte_flow_attr *attr,
3184 		   const struct rte_flow_item items[],
3185 		   const struct rte_flow_action actions[],
3186 		   struct rte_flow_error *error)
3187 {
3188 	const struct mlx5_flow_driver_ops *fops;
3189 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3190 
3191 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3192 	fops = flow_get_drv_ops(type);
3193 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3194 }
3195 
3196 /**
3197  * Flow driver apply API. This abstracts calling driver specific functions.
3198  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3199  * translated driver flows on to device. flow_drv_translate() must precede.
3200  *
3201  * @param[in] dev
3202  *   Pointer to Ethernet device structure.
3203  * @param[in, out] flow
3204  *   Pointer to flow structure.
3205  * @param[out] error
3206  *   Pointer to error structure.
3207  *
3208  * @return
3209  *   0 on success, a negative errno value otherwise and rte_errno is set.
3210  */
3211 static inline int
3212 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3213 	       struct rte_flow_error *error)
3214 {
3215 	const struct mlx5_flow_driver_ops *fops;
3216 	enum mlx5_flow_drv_type type = flow->drv_type;
3217 
3218 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3219 	fops = flow_get_drv_ops(type);
3220 	return fops->apply(dev, flow, error);
3221 }
3222 
3223 /**
3224  * Flow driver destroy API. This abstracts calling driver specific functions.
3225  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3226  * on device and releases resources of the flow.
3227  *
3228  * @param[in] dev
3229  *   Pointer to Ethernet device.
3230  * @param[in, out] flow
3231  *   Pointer to flow structure.
3232  */
3233 static inline void
3234 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3235 {
3236 	const struct mlx5_flow_driver_ops *fops;
3237 	enum mlx5_flow_drv_type type = flow->drv_type;
3238 
3239 	flow_mreg_split_qrss_release(dev, flow);
3240 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3241 	fops = flow_get_drv_ops(type);
3242 	fops->destroy(dev, flow);
3243 }
3244 
3245 /**
3246  * Get RSS action from the action list.
3247  *
3248  * @param[in] actions
3249  *   Pointer to the list of actions.
3250  *
3251  * @return
3252  *   Pointer to the RSS action if exist, else return NULL.
3253  */
3254 static const struct rte_flow_action_rss*
3255 flow_get_rss_action(const struct rte_flow_action actions[])
3256 {
3257 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3258 		switch (actions->type) {
3259 		case RTE_FLOW_ACTION_TYPE_RSS:
3260 			return (const struct rte_flow_action_rss *)
3261 			       actions->conf;
3262 		default:
3263 			break;
3264 		}
3265 	}
3266 	return NULL;
3267 }
3268 
3269 /* maps shared action to translated non shared in some actions array */
3270 struct mlx5_translated_shared_action {
3271 	struct rte_flow_shared_action *action; /**< Shared action */
3272 	int index; /**< Index in related array of rte_flow_action */
3273 };
3274 
3275 /**
3276  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3277  * non shared action if translation possible.
3278  * This functionality used to run same execution path for both shared & non
3279  * shared actions on flow create. All necessary preparations for shared
3280  * action handling should be preformed on *shared* actions list returned
3281  * from this call.
3282  *
3283  * @param[in] actions
3284  *   List of actions to translate.
3285  * @param[out] shared
3286  *   List to store translated shared actions.
3287  * @param[in, out] shared_n
3288  *   Size of *shared* array. On return should be updated with number of shared
3289  *   actions retrieved from the *actions* list.
3290  * @param[out] translated_actions
3291  *   List of actions where all shared actions were translated to non shared
3292  *   if possible. NULL if no translation took place.
3293  * @param[out] error
3294  *   Pointer to the error structure.
3295  *
3296  * @return
3297  *   0 on success, a negative errno value otherwise and rte_errno is set.
3298  */
3299 static int
3300 flow_shared_actions_translate(const struct rte_flow_action actions[],
3301 	struct mlx5_translated_shared_action *shared,
3302 	int *shared_n,
3303 	struct rte_flow_action **translated_actions,
3304 	struct rte_flow_error *error)
3305 {
3306 	struct rte_flow_action *translated = NULL;
3307 	size_t actions_size;
3308 	int n;
3309 	int copied_n = 0;
3310 	struct mlx5_translated_shared_action *shared_end = NULL;
3311 
3312 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3313 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3314 			continue;
3315 		if (copied_n == *shared_n) {
3316 			return rte_flow_error_set
3317 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3318 				 NULL, "too many shared actions");
3319 		}
3320 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3321 			   sizeof(actions[n].conf));
3322 		shared[copied_n].index = n;
3323 		copied_n++;
3324 	}
3325 	n++;
3326 	*shared_n = copied_n;
3327 	if (!copied_n)
3328 		return 0;
3329 	actions_size = sizeof(struct rte_flow_action) * n;
3330 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3331 	if (!translated) {
3332 		rte_errno = ENOMEM;
3333 		return -ENOMEM;
3334 	}
3335 	memcpy(translated, actions, actions_size);
3336 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3337 		const struct rte_flow_shared_action *shared_action;
3338 
3339 		shared_action = shared->action;
3340 		switch (shared_action->type) {
3341 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3342 			translated[shared->index].type =
3343 				RTE_FLOW_ACTION_TYPE_RSS;
3344 			translated[shared->index].conf =
3345 				&shared_action->rss.origin;
3346 			break;
3347 		default:
3348 			mlx5_free(translated);
3349 			return rte_flow_error_set
3350 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3351 				 NULL, "invalid shared action type");
3352 		}
3353 	}
3354 	*translated_actions = translated;
3355 	return 0;
3356 }
3357 
3358 /**
3359  * Get Shared RSS action from the action list.
3360  *
3361  * @param[in] shared
3362  *   Pointer to the list of actions.
3363  * @param[in] shared_n
3364  *   Actions list length.
3365  *
3366  * @return
3367  *   Pointer to the MLX5 RSS action if exists, otherwise return NULL.
3368  */
3369 static struct mlx5_shared_action_rss *
3370 flow_get_shared_rss_action(struct mlx5_translated_shared_action *shared,
3371 			   int shared_n)
3372 {
3373 	struct mlx5_translated_shared_action *shared_end;
3374 
3375 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3376 		struct rte_flow_shared_action *shared_action;
3377 
3378 		shared_action = shared->action;
3379 		switch (shared_action->type) {
3380 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3381 			__atomic_add_fetch(&shared_action->refcnt, 1,
3382 					   __ATOMIC_RELAXED);
3383 			return &shared_action->rss;
3384 		default:
3385 			break;
3386 		}
3387 	}
3388 	return NULL;
3389 }
3390 
3391 struct rte_flow_shared_action *
3392 mlx5_flow_get_shared_rss(struct rte_flow *flow)
3393 {
3394 	if (flow->shared_rss)
3395 		return container_of(flow->shared_rss,
3396 				    struct rte_flow_shared_action, rss);
3397 	else
3398 		return NULL;
3399 }
3400 
3401 static unsigned int
3402 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3403 {
3404 	const struct rte_flow_item *item;
3405 	unsigned int has_vlan = 0;
3406 
3407 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3408 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3409 			has_vlan = 1;
3410 			break;
3411 		}
3412 	}
3413 	if (has_vlan)
3414 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3415 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3416 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3417 			       MLX5_EXPANSION_ROOT_OUTER;
3418 }
3419 
3420 /**
3421  *  Get layer flags from the prefix flow.
3422  *
3423  *  Some flows may be split to several subflows, the prefix subflow gets the
3424  *  match items and the suffix sub flow gets the actions.
3425  *  Some actions need the user defined match item flags to get the detail for
3426  *  the action.
3427  *  This function helps the suffix flow to get the item layer flags from prefix
3428  *  subflow.
3429  *
3430  * @param[in] dev_flow
3431  *   Pointer the created preifx subflow.
3432  *
3433  * @return
3434  *   The layers get from prefix subflow.
3435  */
3436 static inline uint64_t
3437 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3438 {
3439 	uint64_t layers = 0;
3440 
3441 	/*
3442 	 * Layers bits could be localization, but usually the compiler will
3443 	 * help to do the optimization work for source code.
3444 	 * If no decap actions, use the layers directly.
3445 	 */
3446 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3447 		return dev_flow->handle->layers;
3448 	/* Convert L3 layers with decap action. */
3449 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3450 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3451 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3452 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3453 	/* Convert L4 layers with decap action.  */
3454 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3455 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3456 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3457 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3458 	return layers;
3459 }
3460 
3461 /**
3462  * Get metadata split action information.
3463  *
3464  * @param[in] actions
3465  *   Pointer to the list of actions.
3466  * @param[out] qrss
3467  *   Pointer to the return pointer.
3468  * @param[out] qrss_type
3469  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3470  *   if no QUEUE/RSS is found.
3471  * @param[out] encap_idx
3472  *   Pointer to the index of the encap action if exists, otherwise the last
3473  *   action index.
3474  *
3475  * @return
3476  *   Total number of actions.
3477  */
3478 static int
3479 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3480 				       const struct rte_flow_action **qrss,
3481 				       int *encap_idx)
3482 {
3483 	const struct rte_flow_action_raw_encap *raw_encap;
3484 	int actions_n = 0;
3485 	int raw_decap_idx = -1;
3486 
3487 	*encap_idx = -1;
3488 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3489 		switch (actions->type) {
3490 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3491 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3492 			*encap_idx = actions_n;
3493 			break;
3494 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3495 			raw_decap_idx = actions_n;
3496 			break;
3497 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3498 			raw_encap = actions->conf;
3499 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3500 				*encap_idx = raw_decap_idx != -1 ?
3501 						      raw_decap_idx : actions_n;
3502 			break;
3503 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3504 		case RTE_FLOW_ACTION_TYPE_RSS:
3505 			*qrss = actions;
3506 			break;
3507 		default:
3508 			break;
3509 		}
3510 		actions_n++;
3511 	}
3512 	if (*encap_idx == -1)
3513 		*encap_idx = actions_n;
3514 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3515 	return actions_n + 1;
3516 }
3517 
3518 /**
3519  * Check meter action from the action list.
3520  *
3521  * @param[in] actions
3522  *   Pointer to the list of actions.
3523  * @param[out] mtr
3524  *   Pointer to the meter exist flag.
3525  *
3526  * @return
3527  *   Total number of actions.
3528  */
3529 static int
3530 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3531 {
3532 	int actions_n = 0;
3533 
3534 	MLX5_ASSERT(mtr);
3535 	*mtr = 0;
3536 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3537 		switch (actions->type) {
3538 		case RTE_FLOW_ACTION_TYPE_METER:
3539 			*mtr = 1;
3540 			break;
3541 		default:
3542 			break;
3543 		}
3544 		actions_n++;
3545 	}
3546 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3547 	return actions_n + 1;
3548 }
3549 
3550 /**
3551  * Check if the flow should be split due to hairpin.
3552  * The reason for the split is that in current HW we can't
3553  * support encap and push-vlan on Rx, so if a flow contains
3554  * these actions we move it to Tx.
3555  *
3556  * @param dev
3557  *   Pointer to Ethernet device.
3558  * @param[in] attr
3559  *   Flow rule attributes.
3560  * @param[in] actions
3561  *   Associated actions (list terminated by the END action).
3562  *
3563  * @return
3564  *   > 0 the number of actions and the flow should be split,
3565  *   0 when no split required.
3566  */
3567 static int
3568 flow_check_hairpin_split(struct rte_eth_dev *dev,
3569 			 const struct rte_flow_attr *attr,
3570 			 const struct rte_flow_action actions[])
3571 {
3572 	int queue_action = 0;
3573 	int action_n = 0;
3574 	int split = 0;
3575 	const struct rte_flow_action_queue *queue;
3576 	const struct rte_flow_action_rss *rss;
3577 	const struct rte_flow_action_raw_encap *raw_encap;
3578 	const struct rte_eth_hairpin_conf *conf;
3579 
3580 	if (!attr->ingress)
3581 		return 0;
3582 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3583 		switch (actions->type) {
3584 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3585 			queue = actions->conf;
3586 			if (queue == NULL)
3587 				return 0;
3588 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3589 			if (conf != NULL && !!conf->tx_explicit)
3590 				return 0;
3591 			queue_action = 1;
3592 			action_n++;
3593 			break;
3594 		case RTE_FLOW_ACTION_TYPE_RSS:
3595 			rss = actions->conf;
3596 			if (rss == NULL || rss->queue_num == 0)
3597 				return 0;
3598 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3599 			if (conf != NULL && !!conf->tx_explicit)
3600 				return 0;
3601 			queue_action = 1;
3602 			action_n++;
3603 			break;
3604 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3605 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3606 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3607 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3608 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3609 			split++;
3610 			action_n++;
3611 			break;
3612 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3613 			raw_encap = actions->conf;
3614 			if (raw_encap->size >
3615 			    (sizeof(struct rte_flow_item_eth) +
3616 			     sizeof(struct rte_flow_item_ipv4)))
3617 				split++;
3618 			action_n++;
3619 			break;
3620 		default:
3621 			action_n++;
3622 			break;
3623 		}
3624 	}
3625 	if (split && queue_action)
3626 		return action_n;
3627 	return 0;
3628 }
3629 
3630 /* Declare flow create/destroy prototype in advance. */
3631 static uint32_t
3632 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3633 		 const struct rte_flow_attr *attr,
3634 		 const struct rte_flow_item items[],
3635 		 const struct rte_flow_action actions[],
3636 		 bool external, struct rte_flow_error *error);
3637 
3638 static void
3639 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3640 		  uint32_t flow_idx);
3641 
3642 struct mlx5_hlist_entry *
3643 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3644 		       void *cb_ctx)
3645 {
3646 	struct rte_eth_dev *dev = list->ctx;
3647 	struct mlx5_priv *priv = dev->data->dev_private;
3648 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3649 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3650 	struct rte_flow_error *error = ctx->error;
3651 	uint32_t idx = 0;
3652 	int ret;
3653 	uint32_t mark_id = key;
3654 	struct rte_flow_attr attr = {
3655 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3656 		.ingress = 1,
3657 	};
3658 	struct mlx5_rte_flow_item_tag tag_spec = {
3659 		.data = mark_id,
3660 	};
3661 	struct rte_flow_item items[] = {
3662 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3663 	};
3664 	struct rte_flow_action_mark ftag = {
3665 		.id = mark_id,
3666 	};
3667 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3668 		.dst = REG_B,
3669 		.src = REG_NON,
3670 	};
3671 	struct rte_flow_action_jump jump = {
3672 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3673 	};
3674 	struct rte_flow_action actions[] = {
3675 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3676 	};
3677 
3678 	/* Fill the register fileds in the flow. */
3679 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3680 	if (ret < 0)
3681 		return NULL;
3682 	tag_spec.id = ret;
3683 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3684 	if (ret < 0)
3685 		return NULL;
3686 	cp_mreg.src = ret;
3687 	/* Provide the full width of FLAG specific value. */
3688 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3689 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3690 	/* Build a new flow. */
3691 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3692 		items[0] = (struct rte_flow_item){
3693 			.type = (enum rte_flow_item_type)
3694 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3695 			.spec = &tag_spec,
3696 		};
3697 		items[1] = (struct rte_flow_item){
3698 			.type = RTE_FLOW_ITEM_TYPE_END,
3699 		};
3700 		actions[0] = (struct rte_flow_action){
3701 			.type = (enum rte_flow_action_type)
3702 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3703 			.conf = &ftag,
3704 		};
3705 		actions[1] = (struct rte_flow_action){
3706 			.type = (enum rte_flow_action_type)
3707 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3708 			.conf = &cp_mreg,
3709 		};
3710 		actions[2] = (struct rte_flow_action){
3711 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3712 			.conf = &jump,
3713 		};
3714 		actions[3] = (struct rte_flow_action){
3715 			.type = RTE_FLOW_ACTION_TYPE_END,
3716 		};
3717 	} else {
3718 		/* Default rule, wildcard match. */
3719 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3720 		items[0] = (struct rte_flow_item){
3721 			.type = RTE_FLOW_ITEM_TYPE_END,
3722 		};
3723 		actions[0] = (struct rte_flow_action){
3724 			.type = (enum rte_flow_action_type)
3725 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3726 			.conf = &cp_mreg,
3727 		};
3728 		actions[1] = (struct rte_flow_action){
3729 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3730 			.conf = &jump,
3731 		};
3732 		actions[2] = (struct rte_flow_action){
3733 			.type = RTE_FLOW_ACTION_TYPE_END,
3734 		};
3735 	}
3736 	/* Build a new entry. */
3737 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3738 	if (!mcp_res) {
3739 		rte_errno = ENOMEM;
3740 		return NULL;
3741 	}
3742 	mcp_res->idx = idx;
3743 	/*
3744 	 * The copy Flows are not included in any list. There
3745 	 * ones are referenced from other Flows and can not
3746 	 * be applied, removed, deleted in ardbitrary order
3747 	 * by list traversing.
3748 	 */
3749 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3750 					 actions, false, error);
3751 	if (!mcp_res->rix_flow) {
3752 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3753 		return NULL;
3754 	}
3755 	return &mcp_res->hlist_ent;
3756 }
3757 
3758 /**
3759  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3760  *
3761  * As mark_id is unique, if there's already a registered flow for the mark_id,
3762  * return by increasing the reference counter of the resource. Otherwise, create
3763  * the resource (mcp_res) and flow.
3764  *
3765  * Flow looks like,
3766  *   - If ingress port is ANY and reg_c[1] is mark_id,
3767  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3768  *
3769  * For default flow (zero mark_id), flow is like,
3770  *   - If ingress port is ANY,
3771  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3772  *
3773  * @param dev
3774  *   Pointer to Ethernet device.
3775  * @param mark_id
3776  *   ID of MARK action, zero means default flow for META.
3777  * @param[out] error
3778  *   Perform verbose error reporting if not NULL.
3779  *
3780  * @return
3781  *   Associated resource on success, NULL otherwise and rte_errno is set.
3782  */
3783 static struct mlx5_flow_mreg_copy_resource *
3784 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3785 			  struct rte_flow_error *error)
3786 {
3787 	struct mlx5_priv *priv = dev->data->dev_private;
3788 	struct mlx5_hlist_entry *entry;
3789 	struct mlx5_flow_cb_ctx ctx = {
3790 		.dev = dev,
3791 		.error = error,
3792 	};
3793 
3794 	/* Check if already registered. */
3795 	MLX5_ASSERT(priv->mreg_cp_tbl);
3796 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3797 	if (!entry)
3798 		return NULL;
3799 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3800 			    hlist_ent);
3801 }
3802 
3803 void
3804 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3805 {
3806 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3807 		container_of(entry, typeof(*mcp_res), hlist_ent);
3808 	struct rte_eth_dev *dev = list->ctx;
3809 	struct mlx5_priv *priv = dev->data->dev_private;
3810 
3811 	MLX5_ASSERT(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 }
3815 
3816 /**
3817  * Release flow in RX_CP_TBL.
3818  *
3819  * @param dev
3820  *   Pointer to Ethernet device.
3821  * @flow
3822  *   Parent flow for wich copying is provided.
3823  */
3824 static void
3825 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3826 			  struct rte_flow *flow)
3827 {
3828 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3829 	struct mlx5_priv *priv = dev->data->dev_private;
3830 
3831 	if (!flow->rix_mreg_copy)
3832 		return;
3833 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3834 				 flow->rix_mreg_copy);
3835 	if (!mcp_res || !priv->mreg_cp_tbl)
3836 		return;
3837 	MLX5_ASSERT(mcp_res->rix_flow);
3838 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3839 	flow->rix_mreg_copy = 0;
3840 }
3841 
3842 /**
3843  * Remove the default copy action from RX_CP_TBL.
3844  *
3845  * This functions is called in the mlx5_dev_start(). No thread safe
3846  * is guaranteed.
3847  *
3848  * @param dev
3849  *   Pointer to Ethernet device.
3850  */
3851 static void
3852 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3853 {
3854 	struct mlx5_hlist_entry *entry;
3855 	struct mlx5_priv *priv = dev->data->dev_private;
3856 
3857 	/* Check if default flow is registered. */
3858 	if (!priv->mreg_cp_tbl)
3859 		return;
3860 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3861 				  MLX5_DEFAULT_COPY_ID, NULL);
3862 	if (!entry)
3863 		return;
3864 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3865 }
3866 
3867 /**
3868  * Add the default copy action in in RX_CP_TBL.
3869  *
3870  * This functions is called in the mlx5_dev_start(). No thread safe
3871  * is guaranteed.
3872  *
3873  * @param dev
3874  *   Pointer to Ethernet device.
3875  * @param[out] error
3876  *   Perform verbose error reporting if not NULL.
3877  *
3878  * @return
3879  *   0 for success, negative value otherwise and rte_errno is set.
3880  */
3881 static int
3882 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3883 				  struct rte_flow_error *error)
3884 {
3885 	struct mlx5_priv *priv = dev->data->dev_private;
3886 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3887 
3888 	/* Check whether extensive metadata feature is engaged. */
3889 	if (!priv->config.dv_flow_en ||
3890 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3891 	    !mlx5_flow_ext_mreg_supported(dev) ||
3892 	    !priv->sh->dv_regc0_mask)
3893 		return 0;
3894 	/*
3895 	 * Add default mreg copy flow may be called multiple time, but
3896 	 * only be called once in stop. Avoid register it twice.
3897 	 */
3898 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
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 rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5832 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5833 
5834 	if (!flow)
5835 		return;
5836 	/*
5837 	 * Update RX queue flags only if port is started, otherwise it is
5838 	 * already clean.
5839 	 */
5840 	if (dev->data->dev_started)
5841 		flow_rxq_flags_trim(dev, flow);
5842 	flow_drv_destroy(dev, flow);
5843 	if (list) {
5844 		rte_spinlock_lock(&priv->flow_list_lock);
5845 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5846 			     flow_idx, flow, next);
5847 		rte_spinlock_unlock(&priv->flow_list_lock);
5848 	}
5849 	flow_mreg_del_copy_action(dev, flow);
5850 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5851 	if (flow->tunnel) {
5852 		struct mlx5_flow_tunnel *tunnel;
5853 
5854 		rte_spinlock_lock(&mlx5_tunnel_hub(dev)->sl);
5855 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5856 		RTE_VERIFY(tunnel);
5857 		LIST_REMOVE(tunnel, chain);
5858 		rte_spinlock_unlock(&mlx5_tunnel_hub(dev)->sl);
5859 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5860 			mlx5_flow_tunnel_free(dev, tunnel);
5861 	}
5862 }
5863 
5864 /**
5865  * Destroy all flows.
5866  *
5867  * @param dev
5868  *   Pointer to Ethernet device.
5869  * @param list
5870  *   Pointer to the Indexed flow list.
5871  * @param active
5872  *   If flushing is called avtively.
5873  */
5874 void
5875 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5876 {
5877 	uint32_t num_flushed = 0;
5878 
5879 	while (*list) {
5880 		flow_list_destroy(dev, list, *list);
5881 		num_flushed++;
5882 	}
5883 	if (active) {
5884 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5885 			dev->data->port_id, num_flushed);
5886 	}
5887 }
5888 
5889 /**
5890  * Stop all default actions for flows.
5891  *
5892  * @param dev
5893  *   Pointer to Ethernet device.
5894  */
5895 void
5896 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5897 {
5898 	flow_mreg_del_default_copy_action(dev);
5899 	flow_rxq_flags_clear(dev);
5900 }
5901 
5902 /**
5903  * Start all default actions for flows.
5904  *
5905  * @param dev
5906  *   Pointer to Ethernet device.
5907  * @return
5908  *   0 on success, a negative errno value otherwise and rte_errno is set.
5909  */
5910 int
5911 mlx5_flow_start_default(struct rte_eth_dev *dev)
5912 {
5913 	struct rte_flow_error error;
5914 
5915 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5916 	return flow_mreg_add_default_copy_action(dev, &error);
5917 }
5918 
5919 /**
5920  * Release key of thread specific flow workspace data.
5921  */
5922 static void
5923 flow_release_workspace(void *data)
5924 {
5925 	struct mlx5_flow_workspace *wks = data;
5926 
5927 	if (!wks)
5928 		return;
5929 	free(wks->rss_desc[0].queue);
5930 	free(wks->rss_desc[1].queue);
5931 	free(wks);
5932 }
5933 
5934 /**
5935  * Initialize key of thread specific flow workspace data.
5936  */
5937 static void
5938 flow_alloc_workspace(void)
5939 {
5940 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5941 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5942 }
5943 
5944 /**
5945  * Get thread specific flow workspace.
5946  *
5947  * @return pointer to thread specific flowworkspace data, NULL on error.
5948  */
5949 struct mlx5_flow_workspace*
5950 mlx5_flow_get_thread_workspace(void)
5951 {
5952 	struct mlx5_flow_workspace *data;
5953 
5954 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
5955 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
5956 		return NULL;
5957 	}
5958 	data = pthread_getspecific(key_workspace);
5959 	if (!data) {
5960 		data = calloc(1, sizeof(*data));
5961 		if (!data) {
5962 			DRV_LOG(ERR, "Failed to allocate flow workspace "
5963 				"memory.");
5964 			return NULL;
5965 		}
5966 		data->rss_desc[0].queue = calloc(1,
5967 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5968 		if (!data->rss_desc[0].queue)
5969 			goto err;
5970 		data->rss_desc[1].queue = calloc(1,
5971 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5972 		if (!data->rss_desc[1].queue)
5973 			goto err;
5974 		data->rssq_num[0] = MLX5_RSSQ_DEFAULT_NUM;
5975 		data->rssq_num[1] = MLX5_RSSQ_DEFAULT_NUM;
5976 		if (pthread_setspecific(key_workspace, data)) {
5977 			DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5978 			goto err;
5979 		}
5980 	}
5981 	return data;
5982 err:
5983 	if (data->rss_desc[0].queue)
5984 		free(data->rss_desc[0].queue);
5985 	if (data->rss_desc[1].queue)
5986 		free(data->rss_desc[1].queue);
5987 	free(data);
5988 	return NULL;
5989 }
5990 
5991 /**
5992  * Verify the flow list is empty
5993  *
5994  * @param dev
5995  *  Pointer to Ethernet device.
5996  *
5997  * @return the number of flows not released.
5998  */
5999 int
6000 mlx5_flow_verify(struct rte_eth_dev *dev)
6001 {
6002 	struct mlx5_priv *priv = dev->data->dev_private;
6003 	struct rte_flow *flow;
6004 	uint32_t idx;
6005 	int ret = 0;
6006 
6007 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6008 		      flow, next) {
6009 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6010 			dev->data->port_id, (void *)flow);
6011 		++ret;
6012 	}
6013 	return ret;
6014 }
6015 
6016 /**
6017  * Enable default hairpin egress flow.
6018  *
6019  * @param dev
6020  *   Pointer to Ethernet device.
6021  * @param queue
6022  *   The queue index.
6023  *
6024  * @return
6025  *   0 on success, a negative errno value otherwise and rte_errno is set.
6026  */
6027 int
6028 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6029 			    uint32_t queue)
6030 {
6031 	struct mlx5_priv *priv = dev->data->dev_private;
6032 	const struct rte_flow_attr attr = {
6033 		.egress = 1,
6034 		.priority = 0,
6035 	};
6036 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6037 		.queue = queue,
6038 	};
6039 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6040 		.queue = UINT32_MAX,
6041 	};
6042 	struct rte_flow_item items[] = {
6043 		{
6044 			.type = (enum rte_flow_item_type)
6045 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6046 			.spec = &queue_spec,
6047 			.last = NULL,
6048 			.mask = &queue_mask,
6049 		},
6050 		{
6051 			.type = RTE_FLOW_ITEM_TYPE_END,
6052 		},
6053 	};
6054 	struct rte_flow_action_jump jump = {
6055 		.group = MLX5_HAIRPIN_TX_TABLE,
6056 	};
6057 	struct rte_flow_action actions[2];
6058 	uint32_t flow_idx;
6059 	struct rte_flow_error error;
6060 
6061 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6062 	actions[0].conf = &jump;
6063 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6064 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6065 				&attr, items, actions, false, &error);
6066 	if (!flow_idx) {
6067 		DRV_LOG(DEBUG,
6068 			"Failed to create ctrl flow: rte_errno(%d),"
6069 			" type(%d), message(%s)",
6070 			rte_errno, error.type,
6071 			error.message ? error.message : " (no stated reason)");
6072 		return -rte_errno;
6073 	}
6074 	return 0;
6075 }
6076 
6077 /**
6078  * Enable a control flow configured from the control plane.
6079  *
6080  * @param dev
6081  *   Pointer to Ethernet device.
6082  * @param eth_spec
6083  *   An Ethernet flow spec to apply.
6084  * @param eth_mask
6085  *   An Ethernet flow mask to apply.
6086  * @param vlan_spec
6087  *   A VLAN flow spec to apply.
6088  * @param vlan_mask
6089  *   A VLAN flow mask to apply.
6090  *
6091  * @return
6092  *   0 on success, a negative errno value otherwise and rte_errno is set.
6093  */
6094 int
6095 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6096 		    struct rte_flow_item_eth *eth_spec,
6097 		    struct rte_flow_item_eth *eth_mask,
6098 		    struct rte_flow_item_vlan *vlan_spec,
6099 		    struct rte_flow_item_vlan *vlan_mask)
6100 {
6101 	struct mlx5_priv *priv = dev->data->dev_private;
6102 	const struct rte_flow_attr attr = {
6103 		.ingress = 1,
6104 		.priority = MLX5_FLOW_PRIO_RSVD,
6105 	};
6106 	struct rte_flow_item items[] = {
6107 		{
6108 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6109 			.spec = eth_spec,
6110 			.last = NULL,
6111 			.mask = eth_mask,
6112 		},
6113 		{
6114 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6115 					      RTE_FLOW_ITEM_TYPE_END,
6116 			.spec = vlan_spec,
6117 			.last = NULL,
6118 			.mask = vlan_mask,
6119 		},
6120 		{
6121 			.type = RTE_FLOW_ITEM_TYPE_END,
6122 		},
6123 	};
6124 	uint16_t queue[priv->reta_idx_n];
6125 	struct rte_flow_action_rss action_rss = {
6126 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6127 		.level = 0,
6128 		.types = priv->rss_conf.rss_hf,
6129 		.key_len = priv->rss_conf.rss_key_len,
6130 		.queue_num = priv->reta_idx_n,
6131 		.key = priv->rss_conf.rss_key,
6132 		.queue = queue,
6133 	};
6134 	struct rte_flow_action actions[] = {
6135 		{
6136 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6137 			.conf = &action_rss,
6138 		},
6139 		{
6140 			.type = RTE_FLOW_ACTION_TYPE_END,
6141 		},
6142 	};
6143 	uint32_t flow_idx;
6144 	struct rte_flow_error error;
6145 	unsigned int i;
6146 
6147 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6148 		return 0;
6149 	}
6150 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6151 		action_rss.types = 0;
6152 	for (i = 0; i != priv->reta_idx_n; ++i)
6153 		queue[i] = (*priv->reta_idx)[i];
6154 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6155 				&attr, items, actions, false, &error);
6156 	if (!flow_idx)
6157 		return -rte_errno;
6158 	return 0;
6159 }
6160 
6161 /**
6162  * Enable a flow control configured from the control plane.
6163  *
6164  * @param dev
6165  *   Pointer to Ethernet device.
6166  * @param eth_spec
6167  *   An Ethernet flow spec to apply.
6168  * @param eth_mask
6169  *   An Ethernet flow mask to apply.
6170  *
6171  * @return
6172  *   0 on success, a negative errno value otherwise and rte_errno is set.
6173  */
6174 int
6175 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6176 	       struct rte_flow_item_eth *eth_spec,
6177 	       struct rte_flow_item_eth *eth_mask)
6178 {
6179 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6180 }
6181 
6182 /**
6183  * Create default miss flow rule matching lacp traffic
6184  *
6185  * @param dev
6186  *   Pointer to Ethernet device.
6187  * @param eth_spec
6188  *   An Ethernet flow spec to apply.
6189  *
6190  * @return
6191  *   0 on success, a negative errno value otherwise and rte_errno is set.
6192  */
6193 int
6194 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6195 {
6196 	struct mlx5_priv *priv = dev->data->dev_private;
6197 	/*
6198 	 * The LACP matching is done by only using ether type since using
6199 	 * a multicast dst mac causes kernel to give low priority to this flow.
6200 	 */
6201 	static const struct rte_flow_item_eth lacp_spec = {
6202 		.type = RTE_BE16(0x8809),
6203 	};
6204 	static const struct rte_flow_item_eth lacp_mask = {
6205 		.type = 0xffff,
6206 	};
6207 	const struct rte_flow_attr attr = {
6208 		.ingress = 1,
6209 	};
6210 	struct rte_flow_item items[] = {
6211 		{
6212 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6213 			.spec = &lacp_spec,
6214 			.mask = &lacp_mask,
6215 		},
6216 		{
6217 			.type = RTE_FLOW_ITEM_TYPE_END,
6218 		},
6219 	};
6220 	struct rte_flow_action actions[] = {
6221 		{
6222 			.type = (enum rte_flow_action_type)
6223 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6224 		},
6225 		{
6226 			.type = RTE_FLOW_ACTION_TYPE_END,
6227 		},
6228 	};
6229 	struct rte_flow_error error;
6230 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6231 				&attr, items, actions, false, &error);
6232 
6233 	if (!flow_idx)
6234 		return -rte_errno;
6235 	return 0;
6236 }
6237 
6238 /**
6239  * Destroy a flow.
6240  *
6241  * @see rte_flow_destroy()
6242  * @see rte_flow_ops
6243  */
6244 int
6245 mlx5_flow_destroy(struct rte_eth_dev *dev,
6246 		  struct rte_flow *flow,
6247 		  struct rte_flow_error *error __rte_unused)
6248 {
6249 	struct mlx5_priv *priv = dev->data->dev_private;
6250 
6251 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6252 	return 0;
6253 }
6254 
6255 /**
6256  * Destroy all flows.
6257  *
6258  * @see rte_flow_flush()
6259  * @see rte_flow_ops
6260  */
6261 int
6262 mlx5_flow_flush(struct rte_eth_dev *dev,
6263 		struct rte_flow_error *error __rte_unused)
6264 {
6265 	struct mlx5_priv *priv = dev->data->dev_private;
6266 
6267 	mlx5_flow_list_flush(dev, &priv->flows, false);
6268 	return 0;
6269 }
6270 
6271 /**
6272  * Isolated mode.
6273  *
6274  * @see rte_flow_isolate()
6275  * @see rte_flow_ops
6276  */
6277 int
6278 mlx5_flow_isolate(struct rte_eth_dev *dev,
6279 		  int enable,
6280 		  struct rte_flow_error *error)
6281 {
6282 	struct mlx5_priv *priv = dev->data->dev_private;
6283 
6284 	if (dev->data->dev_started) {
6285 		rte_flow_error_set(error, EBUSY,
6286 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6287 				   NULL,
6288 				   "port must be stopped first");
6289 		return -rte_errno;
6290 	}
6291 	priv->isolated = !!enable;
6292 	if (enable)
6293 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6294 	else
6295 		dev->dev_ops = &mlx5_os_dev_ops;
6296 
6297 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6298 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6299 
6300 	return 0;
6301 }
6302 
6303 /**
6304  * Query a flow.
6305  *
6306  * @see rte_flow_query()
6307  * @see rte_flow_ops
6308  */
6309 static int
6310 flow_drv_query(struct rte_eth_dev *dev,
6311 	       uint32_t flow_idx,
6312 	       const struct rte_flow_action *actions,
6313 	       void *data,
6314 	       struct rte_flow_error *error)
6315 {
6316 	struct mlx5_priv *priv = dev->data->dev_private;
6317 	const struct mlx5_flow_driver_ops *fops;
6318 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6319 					       [MLX5_IPOOL_RTE_FLOW],
6320 					       flow_idx);
6321 	enum mlx5_flow_drv_type ftype;
6322 
6323 	if (!flow) {
6324 		return rte_flow_error_set(error, ENOENT,
6325 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6326 			  NULL,
6327 			  "invalid flow handle");
6328 	}
6329 	ftype = flow->drv_type;
6330 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6331 	fops = flow_get_drv_ops(ftype);
6332 
6333 	return fops->query(dev, flow, actions, data, error);
6334 }
6335 
6336 /**
6337  * Query a flow.
6338  *
6339  * @see rte_flow_query()
6340  * @see rte_flow_ops
6341  */
6342 int
6343 mlx5_flow_query(struct rte_eth_dev *dev,
6344 		struct rte_flow *flow,
6345 		const struct rte_flow_action *actions,
6346 		void *data,
6347 		struct rte_flow_error *error)
6348 {
6349 	int ret;
6350 
6351 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6352 			     error);
6353 	if (ret < 0)
6354 		return ret;
6355 	return 0;
6356 }
6357 
6358 /**
6359  * Manage filter operations.
6360  *
6361  * @param dev
6362  *   Pointer to Ethernet device structure.
6363  * @param filter_type
6364  *   Filter type.
6365  * @param filter_op
6366  *   Operation to perform.
6367  * @param arg
6368  *   Pointer to operation-specific structure.
6369  *
6370  * @return
6371  *   0 on success, a negative errno value otherwise and rte_errno is set.
6372  */
6373 int
6374 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6375 		     enum rte_filter_type filter_type,
6376 		     enum rte_filter_op filter_op,
6377 		     void *arg)
6378 {
6379 	switch (filter_type) {
6380 	case RTE_ETH_FILTER_GENERIC:
6381 		if (filter_op != RTE_ETH_FILTER_GET) {
6382 			rte_errno = EINVAL;
6383 			return -rte_errno;
6384 		}
6385 		*(const void **)arg = &mlx5_flow_ops;
6386 		return 0;
6387 	default:
6388 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6389 			dev->data->port_id, filter_type);
6390 		rte_errno = ENOTSUP;
6391 		return -rte_errno;
6392 	}
6393 	return 0;
6394 }
6395 
6396 /**
6397  * Create the needed meter and suffix tables.
6398  *
6399  * @param[in] dev
6400  *   Pointer to Ethernet device.
6401  * @param[in] fm
6402  *   Pointer to the flow meter.
6403  *
6404  * @return
6405  *   Pointer to table set on success, NULL otherwise.
6406  */
6407 struct mlx5_meter_domains_infos *
6408 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6409 			  const struct mlx5_flow_meter *fm)
6410 {
6411 	const struct mlx5_flow_driver_ops *fops;
6412 
6413 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6414 	return fops->create_mtr_tbls(dev, fm);
6415 }
6416 
6417 /**
6418  * Destroy the meter table set.
6419  *
6420  * @param[in] dev
6421  *   Pointer to Ethernet device.
6422  * @param[in] tbl
6423  *   Pointer to the meter table set.
6424  *
6425  * @return
6426  *   0 on success.
6427  */
6428 int
6429 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6430 			   struct mlx5_meter_domains_infos *tbls)
6431 {
6432 	const struct mlx5_flow_driver_ops *fops;
6433 
6434 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6435 	return fops->destroy_mtr_tbls(dev, tbls);
6436 }
6437 
6438 /**
6439  * Create policer rules.
6440  *
6441  * @param[in] dev
6442  *   Pointer to Ethernet device.
6443  * @param[in] fm
6444  *   Pointer to flow meter structure.
6445  * @param[in] attr
6446  *   Pointer to flow attributes.
6447  *
6448  * @return
6449  *   0 on success, -1 otherwise.
6450  */
6451 int
6452 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6453 			       struct mlx5_flow_meter *fm,
6454 			       const struct rte_flow_attr *attr)
6455 {
6456 	const struct mlx5_flow_driver_ops *fops;
6457 
6458 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6459 	return fops->create_policer_rules(dev, fm, attr);
6460 }
6461 
6462 /**
6463  * Destroy policer rules.
6464  *
6465  * @param[in] fm
6466  *   Pointer to flow meter structure.
6467  * @param[in] attr
6468  *   Pointer to flow attributes.
6469  *
6470  * @return
6471  *   0 on success, -1 otherwise.
6472  */
6473 int
6474 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6475 				struct mlx5_flow_meter *fm,
6476 				const struct rte_flow_attr *attr)
6477 {
6478 	const struct mlx5_flow_driver_ops *fops;
6479 
6480 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6481 	return fops->destroy_policer_rules(dev, fm, attr);
6482 }
6483 
6484 /**
6485  * Allocate a counter.
6486  *
6487  * @param[in] dev
6488  *   Pointer to Ethernet device structure.
6489  *
6490  * @return
6491  *   Index to allocated counter  on success, 0 otherwise.
6492  */
6493 uint32_t
6494 mlx5_counter_alloc(struct rte_eth_dev *dev)
6495 {
6496 	const struct mlx5_flow_driver_ops *fops;
6497 	struct rte_flow_attr attr = { .transfer = 0 };
6498 
6499 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6500 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6501 		return fops->counter_alloc(dev);
6502 	}
6503 	DRV_LOG(ERR,
6504 		"port %u counter allocate is not supported.",
6505 		 dev->data->port_id);
6506 	return 0;
6507 }
6508 
6509 /**
6510  * Free a counter.
6511  *
6512  * @param[in] dev
6513  *   Pointer to Ethernet device structure.
6514  * @param[in] cnt
6515  *   Index to counter to be free.
6516  */
6517 void
6518 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6519 {
6520 	const struct mlx5_flow_driver_ops *fops;
6521 	struct rte_flow_attr attr = { .transfer = 0 };
6522 
6523 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6524 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6525 		fops->counter_free(dev, cnt);
6526 		return;
6527 	}
6528 	DRV_LOG(ERR,
6529 		"port %u counter free is not supported.",
6530 		 dev->data->port_id);
6531 }
6532 
6533 /**
6534  * Query counter statistics.
6535  *
6536  * @param[in] dev
6537  *   Pointer to Ethernet device structure.
6538  * @param[in] cnt
6539  *   Index to counter to query.
6540  * @param[in] clear
6541  *   Set to clear counter statistics.
6542  * @param[out] pkts
6543  *   The counter hits packets number to save.
6544  * @param[out] bytes
6545  *   The counter hits bytes number to save.
6546  *
6547  * @return
6548  *   0 on success, a negative errno value otherwise.
6549  */
6550 int
6551 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6552 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6553 {
6554 	const struct mlx5_flow_driver_ops *fops;
6555 	struct rte_flow_attr attr = { .transfer = 0 };
6556 
6557 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6558 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6559 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6560 	}
6561 	DRV_LOG(ERR,
6562 		"port %u counter query is not supported.",
6563 		 dev->data->port_id);
6564 	return -ENOTSUP;
6565 }
6566 
6567 /**
6568  * Allocate a new memory for the counter values wrapped by all the needed
6569  * management.
6570  *
6571  * @param[in] sh
6572  *   Pointer to mlx5_dev_ctx_shared object.
6573  *
6574  * @return
6575  *   0 on success, a negative errno value otherwise.
6576  */
6577 static int
6578 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6579 {
6580 	struct mlx5_devx_mkey_attr mkey_attr;
6581 	struct mlx5_counter_stats_mem_mng *mem_mng;
6582 	volatile struct flow_counter_stats *raw_data;
6583 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6584 	int size = (sizeof(struct flow_counter_stats) *
6585 			MLX5_COUNTERS_PER_POOL +
6586 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6587 			sizeof(struct mlx5_counter_stats_mem_mng);
6588 	size_t pgsize = rte_mem_page_size();
6589 	uint8_t *mem;
6590 	int i;
6591 
6592 	if (pgsize == (size_t)-1) {
6593 		DRV_LOG(ERR, "Failed to get mem page size");
6594 		rte_errno = ENOMEM;
6595 		return -ENOMEM;
6596 	}
6597 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6598 	if (!mem) {
6599 		rte_errno = ENOMEM;
6600 		return -ENOMEM;
6601 	}
6602 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6603 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6604 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
6605 						 IBV_ACCESS_LOCAL_WRITE);
6606 	if (!mem_mng->umem) {
6607 		rte_errno = errno;
6608 		mlx5_free(mem);
6609 		return -rte_errno;
6610 	}
6611 	mkey_attr.addr = (uintptr_t)mem;
6612 	mkey_attr.size = size;
6613 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6614 	mkey_attr.pd = sh->pdn;
6615 	mkey_attr.log_entity_size = 0;
6616 	mkey_attr.pg_access = 0;
6617 	mkey_attr.klm_array = NULL;
6618 	mkey_attr.klm_num = 0;
6619 	mkey_attr.relaxed_ordering = sh->cmng.relaxed_ordering;
6620 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6621 	if (!mem_mng->dm) {
6622 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
6623 		rte_errno = errno;
6624 		mlx5_free(mem);
6625 		return -rte_errno;
6626 	}
6627 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6628 	raw_data = (volatile struct flow_counter_stats *)mem;
6629 	for (i = 0; i < raws_n; ++i) {
6630 		mem_mng->raws[i].mem_mng = mem_mng;
6631 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6632 	}
6633 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6634 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6635 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6636 				 next);
6637 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6638 	sh->cmng.mem_mng = mem_mng;
6639 	return 0;
6640 }
6641 
6642 /**
6643  * Set the statistic memory to the new counter pool.
6644  *
6645  * @param[in] sh
6646  *   Pointer to mlx5_dev_ctx_shared object.
6647  * @param[in] pool
6648  *   Pointer to the pool to set the statistic memory.
6649  *
6650  * @return
6651  *   0 on success, a negative errno value otherwise.
6652  */
6653 static int
6654 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6655 			       struct mlx5_flow_counter_pool *pool)
6656 {
6657 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6658 	/* Resize statistic memory once used out. */
6659 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6660 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6661 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6662 		return -1;
6663 	}
6664 	rte_spinlock_lock(&pool->sl);
6665 	pool->raw = cmng->mem_mng->raws + pool->index %
6666 		    MLX5_CNT_CONTAINER_RESIZE;
6667 	rte_spinlock_unlock(&pool->sl);
6668 	pool->raw_hw = NULL;
6669 	return 0;
6670 }
6671 
6672 #define MLX5_POOL_QUERY_FREQ_US 1000000
6673 
6674 /**
6675  * Set the periodic procedure for triggering asynchronous batch queries for all
6676  * the counter pools.
6677  *
6678  * @param[in] sh
6679  *   Pointer to mlx5_dev_ctx_shared object.
6680  */
6681 void
6682 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6683 {
6684 	uint32_t pools_n, us;
6685 
6686 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6687 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6688 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6689 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6690 		sh->cmng.query_thread_on = 0;
6691 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6692 	} else {
6693 		sh->cmng.query_thread_on = 1;
6694 	}
6695 }
6696 
6697 /**
6698  * The periodic procedure for triggering asynchronous batch queries for all the
6699  * counter pools. This function is probably called by the host thread.
6700  *
6701  * @param[in] arg
6702  *   The parameter for the alarm process.
6703  */
6704 void
6705 mlx5_flow_query_alarm(void *arg)
6706 {
6707 	struct mlx5_dev_ctx_shared *sh = arg;
6708 	int ret;
6709 	uint16_t pool_index = sh->cmng.pool_index;
6710 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6711 	struct mlx5_flow_counter_pool *pool;
6712 	uint16_t n_valid;
6713 
6714 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6715 		goto set_alarm;
6716 	rte_spinlock_lock(&cmng->pool_update_sl);
6717 	pool = cmng->pools[pool_index];
6718 	n_valid = cmng->n_valid;
6719 	rte_spinlock_unlock(&cmng->pool_update_sl);
6720 	/* Set the statistic memory to the new created pool. */
6721 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6722 		goto set_alarm;
6723 	if (pool->raw_hw)
6724 		/* There is a pool query in progress. */
6725 		goto set_alarm;
6726 	pool->raw_hw =
6727 		LIST_FIRST(&sh->cmng.free_stat_raws);
6728 	if (!pool->raw_hw)
6729 		/* No free counter statistics raw memory. */
6730 		goto set_alarm;
6731 	/*
6732 	 * Identify the counters released between query trigger and query
6733 	 * handle more efficiently. The counter released in this gap period
6734 	 * should wait for a new round of query as the new arrived packets
6735 	 * will not be taken into account.
6736 	 */
6737 	pool->query_gen++;
6738 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6739 					       MLX5_COUNTERS_PER_POOL,
6740 					       NULL, NULL,
6741 					       pool->raw_hw->mem_mng->dm->id,
6742 					       (void *)(uintptr_t)
6743 					       pool->raw_hw->data,
6744 					       sh->devx_comp,
6745 					       (uint64_t)(uintptr_t)pool);
6746 	if (ret) {
6747 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6748 			" %d", pool->min_dcs->id);
6749 		pool->raw_hw = NULL;
6750 		goto set_alarm;
6751 	}
6752 	LIST_REMOVE(pool->raw_hw, next);
6753 	sh->cmng.pending_queries++;
6754 	pool_index++;
6755 	if (pool_index >= n_valid)
6756 		pool_index = 0;
6757 set_alarm:
6758 	sh->cmng.pool_index = pool_index;
6759 	mlx5_set_query_alarm(sh);
6760 }
6761 
6762 /**
6763  * Check and callback event for new aged flow in the counter pool
6764  *
6765  * @param[in] sh
6766  *   Pointer to mlx5_dev_ctx_shared object.
6767  * @param[in] pool
6768  *   Pointer to Current counter pool.
6769  */
6770 static void
6771 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6772 		   struct mlx5_flow_counter_pool *pool)
6773 {
6774 	struct mlx5_priv *priv;
6775 	struct mlx5_flow_counter *cnt;
6776 	struct mlx5_age_info *age_info;
6777 	struct mlx5_age_param *age_param;
6778 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6779 	struct mlx5_counter_stats_raw *prev = pool->raw;
6780 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6781 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6782 	uint16_t expected = AGE_CANDIDATE;
6783 	uint32_t i;
6784 
6785 	pool->time_of_last_age_check = curr_time;
6786 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6787 		cnt = MLX5_POOL_GET_CNT(pool, i);
6788 		age_param = MLX5_CNT_TO_AGE(cnt);
6789 		if (__atomic_load_n(&age_param->state,
6790 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6791 			continue;
6792 		if (cur->data[i].hits != prev->data[i].hits) {
6793 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6794 					 __ATOMIC_RELAXED);
6795 			continue;
6796 		}
6797 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6798 				       time_delta,
6799 				       __ATOMIC_RELAXED) <= age_param->timeout)
6800 			continue;
6801 		/**
6802 		 * Hold the lock first, or if between the
6803 		 * state AGE_TMOUT and tailq operation the
6804 		 * release happened, the release procedure
6805 		 * may delete a non-existent tailq node.
6806 		 */
6807 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6808 		age_info = GET_PORT_AGE_INFO(priv);
6809 		rte_spinlock_lock(&age_info->aged_sl);
6810 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6811 						AGE_TMOUT, false,
6812 						__ATOMIC_RELAXED,
6813 						__ATOMIC_RELAXED)) {
6814 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6815 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6816 		}
6817 		rte_spinlock_unlock(&age_info->aged_sl);
6818 	}
6819 	mlx5_age_event_prepare(sh);
6820 }
6821 
6822 /**
6823  * Handler for the HW respond about ready values from an asynchronous batch
6824  * query. This function is probably called by the host thread.
6825  *
6826  * @param[in] sh
6827  *   The pointer to the shared device context.
6828  * @param[in] async_id
6829  *   The Devx async ID.
6830  * @param[in] status
6831  *   The status of the completion.
6832  */
6833 void
6834 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6835 				  uint64_t async_id, int status)
6836 {
6837 	struct mlx5_flow_counter_pool *pool =
6838 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6839 	struct mlx5_counter_stats_raw *raw_to_free;
6840 	uint8_t query_gen = pool->query_gen ^ 1;
6841 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6842 	enum mlx5_counter_type cnt_type =
6843 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6844 				MLX5_COUNTER_TYPE_ORIGIN;
6845 
6846 	if (unlikely(status)) {
6847 		raw_to_free = pool->raw_hw;
6848 	} else {
6849 		raw_to_free = pool->raw;
6850 		if (pool->is_aged)
6851 			mlx5_flow_aging_check(sh, pool);
6852 		rte_spinlock_lock(&pool->sl);
6853 		pool->raw = pool->raw_hw;
6854 		rte_spinlock_unlock(&pool->sl);
6855 		/* Be sure the new raw counters data is updated in memory. */
6856 		rte_io_wmb();
6857 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6858 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6859 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6860 				     &pool->counters[query_gen], next);
6861 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6862 		}
6863 	}
6864 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6865 	pool->raw_hw = NULL;
6866 	sh->cmng.pending_queries--;
6867 }
6868 
6869 static const struct mlx5_flow_tbl_data_entry  *
6870 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
6871 {
6872 	struct mlx5_priv *priv = dev->data->dev_private;
6873 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6874 	struct mlx5_hlist_entry *he;
6875 	union tunnel_offload_mark mbits = { .val = mark };
6876 	union mlx5_flow_tbl_key table_key = {
6877 		{
6878 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
6879 			.dummy = 0,
6880 			.domain = !!mbits.transfer,
6881 			.direction = 0,
6882 		}
6883 	};
6884 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
6885 	return he ?
6886 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
6887 }
6888 
6889 static void
6890 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
6891 				   struct mlx5_hlist_entry *entry)
6892 {
6893 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6894 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
6895 
6896 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6897 			tunnel_flow_tbl_to_id(tte->flow_table));
6898 	mlx5_free(tte);
6899 }
6900 
6901 static struct mlx5_hlist_entry *
6902 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list,
6903 				   uint64_t key __rte_unused,
6904 				   void *ctx __rte_unused)
6905 {
6906 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6907 	struct tunnel_tbl_entry *tte;
6908 
6909 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
6910 			  sizeof(*tte), 0,
6911 			  SOCKET_ID_ANY);
6912 	if (!tte)
6913 		goto err;
6914 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6915 			  &tte->flow_table);
6916 	if (tte->flow_table >= MLX5_MAX_TABLES) {
6917 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
6918 			tte->flow_table);
6919 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6920 				tte->flow_table);
6921 		goto err;
6922 	} else if (!tte->flow_table) {
6923 		goto err;
6924 	}
6925 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
6926 	return &tte->hash;
6927 err:
6928 	if (tte)
6929 		mlx5_free(tte);
6930 	return NULL;
6931 }
6932 
6933 static uint32_t
6934 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
6935 				const struct mlx5_flow_tunnel *tunnel,
6936 				uint32_t group, uint32_t *table,
6937 				struct rte_flow_error *error)
6938 {
6939 	struct mlx5_hlist_entry *he;
6940 	struct tunnel_tbl_entry *tte;
6941 	union tunnel_tbl_key key = {
6942 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
6943 		.group = group
6944 	};
6945 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
6946 	struct mlx5_hlist *group_hash;
6947 
6948 	group_hash = tunnel ? tunnel->groups : thub->groups;
6949 	he = mlx5_hlist_register(group_hash, key.val, NULL);
6950 	if (!he)
6951 		return rte_flow_error_set(error, EINVAL,
6952 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6953 					  NULL,
6954 					  "tunnel group index not supported");
6955 	tte = container_of(he, typeof(*tte), hash);
6956 	*table = tte->flow_table;
6957 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
6958 		dev->data->port_id, key.tunnel_id, group, *table);
6959 	return 0;
6960 }
6961 
6962 static int
6963 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6964 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
6965 {
6966 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
6967 		if (group == UINT32_MAX)
6968 			return rte_flow_error_set
6969 						(error, EINVAL,
6970 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6971 						 NULL,
6972 						 "group index not supported");
6973 		*table = group + 1;
6974 	} else {
6975 		*table = group;
6976 	}
6977 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6978 	return 0;
6979 }
6980 
6981 /**
6982  * Translate the rte_flow group index to HW table value.
6983  *
6984  * If tunnel offload is disabled, all group ids converted to flow table
6985  * id using the standard method.
6986  * If tunnel offload is enabled, group id can be converted using the
6987  * standard or tunnel conversion method. Group conversion method
6988  * selection depends on flags in `grp_info` parameter:
6989  * - Internal (grp_info.external == 0) groups conversion uses the
6990  *   standard method.
6991  * - Group ids in JUMP action converted with the tunnel conversion.
6992  * - Group id in rule attribute conversion depends on a rule type and
6993  *   group id value:
6994  *   ** non zero group attributes converted with the tunnel method
6995  *   ** zero group attribute in non-tunnel rule is converted using the
6996  *      standard method - there's only one root table
6997  *   ** zero group attribute in steer tunnel rule is converted with the
6998  *      standard method - single root table
6999  *   ** zero group attribute in match tunnel rule is a special OvS
7000  *      case: that value is used for portability reasons. That group
7001  *      id is converted with the tunnel conversion method.
7002  *
7003  * @param[in] dev
7004  *   Port device
7005  * @param[in] tunnel
7006  *   PMD tunnel offload object
7007  * @param[in] group
7008  *   rte_flow group index value.
7009  * @param[out] table
7010  *   HW table value.
7011  * @param[in] grp_info
7012  *   flags used for conversion
7013  * @param[out] error
7014  *   Pointer to error structure.
7015  *
7016  * @return
7017  *   0 on success, a negative errno value otherwise and rte_errno is set.
7018  */
7019 int
7020 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7021 			 const struct mlx5_flow_tunnel *tunnel,
7022 			 uint32_t group, uint32_t *table,
7023 			 struct flow_grp_info grp_info,
7024 			 struct rte_flow_error *error)
7025 {
7026 	int ret;
7027 	bool standard_translation;
7028 
7029 	if (grp_info.external && group < MLX5_MAX_TABLES_EXTERNAL)
7030 		group *= MLX5_FLOW_TABLE_FACTOR;
7031 	if (is_tunnel_offload_active(dev)) {
7032 		standard_translation = !grp_info.external ||
7033 					grp_info.std_tbl_fix;
7034 	} else {
7035 		standard_translation = true;
7036 	}
7037 	DRV_LOG(DEBUG,
7038 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7039 		dev->data->port_id, group, grp_info.transfer,
7040 		grp_info.external, grp_info.fdb_def_rule,
7041 		standard_translation ? "STANDARD" : "TUNNEL");
7042 	if (standard_translation)
7043 		ret = flow_group_to_table(dev->data->port_id, group, table,
7044 					  grp_info, error);
7045 	else
7046 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7047 						      table, error);
7048 
7049 	return ret;
7050 }
7051 
7052 /**
7053  * Discover availability of metadata reg_c's.
7054  *
7055  * Iteratively use test flows to check availability.
7056  *
7057  * @param[in] dev
7058  *   Pointer to the Ethernet device structure.
7059  *
7060  * @return
7061  *   0 on success, a negative errno value otherwise and rte_errno is set.
7062  */
7063 int
7064 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7065 {
7066 	struct mlx5_priv *priv = dev->data->dev_private;
7067 	struct mlx5_dev_config *config = &priv->config;
7068 	enum modify_reg idx;
7069 	int n = 0;
7070 
7071 	/* reg_c[0] and reg_c[1] are reserved. */
7072 	config->flow_mreg_c[n++] = REG_C_0;
7073 	config->flow_mreg_c[n++] = REG_C_1;
7074 	/* Discover availability of other reg_c's. */
7075 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7076 		struct rte_flow_attr attr = {
7077 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7078 			.priority = MLX5_FLOW_PRIO_RSVD,
7079 			.ingress = 1,
7080 		};
7081 		struct rte_flow_item items[] = {
7082 			[0] = {
7083 				.type = RTE_FLOW_ITEM_TYPE_END,
7084 			},
7085 		};
7086 		struct rte_flow_action actions[] = {
7087 			[0] = {
7088 				.type = (enum rte_flow_action_type)
7089 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7090 				.conf = &(struct mlx5_flow_action_copy_mreg){
7091 					.src = REG_C_1,
7092 					.dst = idx,
7093 				},
7094 			},
7095 			[1] = {
7096 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7097 				.conf = &(struct rte_flow_action_jump){
7098 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7099 				},
7100 			},
7101 			[2] = {
7102 				.type = RTE_FLOW_ACTION_TYPE_END,
7103 			},
7104 		};
7105 		uint32_t flow_idx;
7106 		struct rte_flow *flow;
7107 		struct rte_flow_error error;
7108 
7109 		if (!config->dv_flow_en)
7110 			break;
7111 		/* Create internal flow, validation skips copy action. */
7112 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7113 					    actions, false, &error);
7114 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7115 				      flow_idx);
7116 		if (!flow)
7117 			continue;
7118 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
7119 			config->flow_mreg_c[n++] = idx;
7120 		flow_list_destroy(dev, NULL, flow_idx);
7121 	}
7122 	for (; n < MLX5_MREG_C_NUM; ++n)
7123 		config->flow_mreg_c[n] = REG_NON;
7124 	return 0;
7125 }
7126 
7127 /**
7128  * Dump flow raw hw data to file
7129  *
7130  * @param[in] dev
7131  *    The pointer to Ethernet device.
7132  * @param[in] file
7133  *   A pointer to a file for output.
7134  * @param[out] error
7135  *   Perform verbose error reporting if not NULL. PMDs initialize this
7136  *   structure in case of error only.
7137  * @return
7138  *   0 on success, a nagative value otherwise.
7139  */
7140 int
7141 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7142 		   FILE *file,
7143 		   struct rte_flow_error *error __rte_unused)
7144 {
7145 	struct mlx5_priv *priv = dev->data->dev_private;
7146 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7147 
7148 	if (!priv->config.dv_flow_en) {
7149 		if (fputs("device dv flow disabled\n", file) <= 0)
7150 			return -errno;
7151 		return -ENOTSUP;
7152 	}
7153 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7154 				       sh->tx_domain, file);
7155 }
7156 
7157 /**
7158  * Get aged-out flows.
7159  *
7160  * @param[in] dev
7161  *   Pointer to the Ethernet device structure.
7162  * @param[in] context
7163  *   The address of an array of pointers to the aged-out flows contexts.
7164  * @param[in] nb_countexts
7165  *   The length of context array pointers.
7166  * @param[out] error
7167  *   Perform verbose error reporting if not NULL. Initialized in case of
7168  *   error only.
7169  *
7170  * @return
7171  *   how many contexts get in success, otherwise negative errno value.
7172  *   if nb_contexts is 0, return the amount of all aged contexts.
7173  *   if nb_contexts is not 0 , return the amount of aged flows reported
7174  *   in the context array.
7175  */
7176 int
7177 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7178 			uint32_t nb_contexts, struct rte_flow_error *error)
7179 {
7180 	const struct mlx5_flow_driver_ops *fops;
7181 	struct rte_flow_attr attr = { .transfer = 0 };
7182 
7183 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7184 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7185 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7186 						    error);
7187 	}
7188 	DRV_LOG(ERR,
7189 		"port %u get aged flows is not supported.",
7190 		 dev->data->port_id);
7191 	return -ENOTSUP;
7192 }
7193 
7194 /* Wrapper for driver action_validate op callback */
7195 static int
7196 flow_drv_action_validate(struct rte_eth_dev *dev,
7197 			 const struct rte_flow_shared_action_conf *conf,
7198 			 const struct rte_flow_action *action,
7199 			 const struct mlx5_flow_driver_ops *fops,
7200 			 struct rte_flow_error *error)
7201 {
7202 	static const char err_msg[] = "shared action validation unsupported";
7203 
7204 	if (!fops->action_validate) {
7205 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7206 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7207 				   NULL, err_msg);
7208 		return -rte_errno;
7209 	}
7210 	return fops->action_validate(dev, conf, action, error);
7211 }
7212 
7213 /**
7214  * Destroys the shared action by handle.
7215  *
7216  * @param dev
7217  *   Pointer to Ethernet device structure.
7218  * @param[in] action
7219  *   Handle for the shared action to be destroyed.
7220  * @param[out] error
7221  *   Perform verbose error reporting if not NULL. PMDs initialize this
7222  *   structure in case of error only.
7223  *
7224  * @return
7225  *   0 on success, a negative errno value otherwise and rte_errno is set.
7226  *
7227  * @note: wrapper for driver action_create op callback.
7228  */
7229 static int
7230 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7231 			   struct rte_flow_shared_action *action,
7232 			   struct rte_flow_error *error)
7233 {
7234 	static const char err_msg[] = "shared action destruction unsupported";
7235 	struct rte_flow_attr attr = { .transfer = 0 };
7236 	const struct mlx5_flow_driver_ops *fops =
7237 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7238 
7239 	if (!fops->action_destroy) {
7240 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7241 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7242 				   NULL, err_msg);
7243 		return -rte_errno;
7244 	}
7245 	return fops->action_destroy(dev, action, error);
7246 }
7247 
7248 /* Wrapper for driver action_destroy op callback */
7249 static int
7250 flow_drv_action_update(struct rte_eth_dev *dev,
7251 		       struct rte_flow_shared_action *action,
7252 		       const void *action_conf,
7253 		       const struct mlx5_flow_driver_ops *fops,
7254 		       struct rte_flow_error *error)
7255 {
7256 	static const char err_msg[] = "shared action update unsupported";
7257 
7258 	if (!fops->action_update) {
7259 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7260 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7261 				   NULL, err_msg);
7262 		return -rte_errno;
7263 	}
7264 	return fops->action_update(dev, action, action_conf, error);
7265 }
7266 
7267 /**
7268  * Create shared action for reuse in multiple flow rules.
7269  *
7270  * @param dev
7271  *   Pointer to Ethernet device structure.
7272  * @param[in] action
7273  *   Action configuration for shared action creation.
7274  * @param[out] error
7275  *   Perform verbose error reporting if not NULL. PMDs initialize this
7276  *   structure in case of error only.
7277  * @return
7278  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7279  */
7280 static struct rte_flow_shared_action *
7281 mlx5_shared_action_create(struct rte_eth_dev *dev,
7282 			  const struct rte_flow_shared_action_conf *conf,
7283 			  const struct rte_flow_action *action,
7284 			  struct rte_flow_error *error)
7285 {
7286 	static const char err_msg[] = "shared action creation unsupported";
7287 	struct rte_flow_attr attr = { .transfer = 0 };
7288 	const struct mlx5_flow_driver_ops *fops =
7289 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7290 
7291 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7292 		return NULL;
7293 	if (!fops->action_create) {
7294 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7295 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7296 				   NULL, err_msg);
7297 		return NULL;
7298 	}
7299 	return fops->action_create(dev, conf, action, error);
7300 }
7301 
7302 /**
7303  * Updates inplace the shared action configuration pointed by *action* handle
7304  * with the configuration provided as *action* argument.
7305  * The update of the shared action configuration effects all flow rules reusing
7306  * the action via handle.
7307  *
7308  * @param dev
7309  *   Pointer to Ethernet device structure.
7310  * @param[in] shared_action
7311  *   Handle for the shared action to be updated.
7312  * @param[in] action
7313  *   Action specification used to modify the action pointed by handle.
7314  *   *action* should be of same type with the action pointed by the *action*
7315  *   handle argument, otherwise considered as invalid.
7316  * @param[out] error
7317  *   Perform verbose error reporting if not NULL. PMDs initialize this
7318  *   structure in case of error only.
7319  *
7320  * @return
7321  *   0 on success, a negative errno value otherwise and rte_errno is set.
7322  */
7323 static int
7324 mlx5_shared_action_update(struct rte_eth_dev *dev,
7325 		struct rte_flow_shared_action *shared_action,
7326 		const struct rte_flow_action *action,
7327 		struct rte_flow_error *error)
7328 {
7329 	struct rte_flow_attr attr = { .transfer = 0 };
7330 	const struct mlx5_flow_driver_ops *fops =
7331 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7332 	int ret;
7333 
7334 	switch (shared_action->type) {
7335 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7336 		if (action->type != RTE_FLOW_ACTION_TYPE_RSS) {
7337 			return rte_flow_error_set(error, EINVAL,
7338 						  RTE_FLOW_ERROR_TYPE_ACTION,
7339 						  NULL,
7340 						  "update action type invalid");
7341 		}
7342 		ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7343 		if (ret)
7344 			return ret;
7345 		return flow_drv_action_update(dev, shared_action, action->conf,
7346 					      fops, error);
7347 	default:
7348 		return rte_flow_error_set(error, ENOTSUP,
7349 					  RTE_FLOW_ERROR_TYPE_ACTION,
7350 					  NULL,
7351 					  "action type not supported");
7352 	}
7353 }
7354 
7355 /**
7356  * Query the shared action by handle.
7357  *
7358  * This function allows retrieving action-specific data such as counters.
7359  * Data is gathered by special action which may be present/referenced in
7360  * more than one flow rule definition.
7361  *
7362  * \see RTE_FLOW_ACTION_TYPE_COUNT
7363  *
7364  * @param dev
7365  *   Pointer to Ethernet device structure.
7366  * @param[in] action
7367  *   Handle for the shared action to query.
7368  * @param[in, out] data
7369  *   Pointer to storage for the associated query data type.
7370  * @param[out] error
7371  *   Perform verbose error reporting if not NULL. PMDs initialize this
7372  *   structure in case of error only.
7373  *
7374  * @return
7375  *   0 on success, a negative errno value otherwise and rte_errno is set.
7376  */
7377 static int
7378 mlx5_shared_action_query(struct rte_eth_dev *dev,
7379 			 const struct rte_flow_shared_action *action,
7380 			 void *data,
7381 			 struct rte_flow_error *error)
7382 {
7383 	(void)dev;
7384 	switch (action->type) {
7385 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7386 		__atomic_load(&action->refcnt, (uint32_t *)data,
7387 			      __ATOMIC_RELAXED);
7388 		return 0;
7389 	default:
7390 		return rte_flow_error_set(error, ENOTSUP,
7391 					  RTE_FLOW_ERROR_TYPE_ACTION,
7392 					  NULL,
7393 					  "action type not supported");
7394 	}
7395 }
7396 
7397 /**
7398  * Destroy all shared actions.
7399  *
7400  * @param dev
7401  *   Pointer to Ethernet device.
7402  *
7403  * @return
7404  *   0 on success, a negative errno value otherwise and rte_errno is set.
7405  */
7406 int
7407 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7408 {
7409 	struct rte_flow_error error;
7410 	struct mlx5_priv *priv = dev->data->dev_private;
7411 	struct rte_flow_shared_action *action;
7412 	int ret = 0;
7413 
7414 	while (!LIST_EMPTY(&priv->shared_actions)) {
7415 		action = LIST_FIRST(&priv->shared_actions);
7416 		ret = mlx5_shared_action_destroy(dev, action, &error);
7417 	}
7418 	return ret;
7419 }
7420 
7421 static void
7422 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7423 		      struct mlx5_flow_tunnel *tunnel)
7424 {
7425 	struct mlx5_priv *priv = dev->data->dev_private;
7426 
7427 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7428 		dev->data->port_id, tunnel->tunnel_id);
7429 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
7430 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
7431 			tunnel->tunnel_id);
7432 	mlx5_hlist_destroy(tunnel->groups);
7433 	mlx5_free(tunnel);
7434 }
7435 
7436 static struct mlx5_flow_tunnel *
7437 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7438 {
7439 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7440 	struct mlx5_flow_tunnel *tun;
7441 
7442 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7443 		if (tun->tunnel_id == id)
7444 			break;
7445 	}
7446 
7447 	return tun;
7448 }
7449 
7450 static struct mlx5_flow_tunnel *
7451 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7452 			  const struct rte_flow_tunnel *app_tunnel)
7453 {
7454 	struct mlx5_priv *priv = dev->data->dev_private;
7455 	struct mlx5_flow_tunnel *tunnel;
7456 	uint32_t id;
7457 
7458 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
7459 			  &id);
7460 	if (id >= MLX5_MAX_TUNNELS) {
7461 		mlx5_ipool_free(priv->sh->ipool
7462 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7463 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7464 		return NULL;
7465 	} else if (!id) {
7466 		return NULL;
7467 	}
7468 	/**
7469 	 * mlx5 flow tunnel is an auxlilary data structure
7470 	 * It's not part of IO. No need to allocate it from
7471 	 * huge pages pools dedicated for IO
7472 	 */
7473 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
7474 			     0, SOCKET_ID_ANY);
7475 	if (!tunnel) {
7476 		mlx5_ipool_free(priv->sh->ipool
7477 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7478 		return NULL;
7479 	}
7480 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7481 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7482 					   NULL,
7483 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7484 	if (!tunnel->groups) {
7485 		mlx5_ipool_free(priv->sh->ipool
7486 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7487 		mlx5_free(tunnel);
7488 		return NULL;
7489 	}
7490 	tunnel->groups->ctx = priv->sh;
7491 	/* initiate new PMD tunnel */
7492 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7493 	tunnel->tunnel_id = id;
7494 	tunnel->action.type = (typeof(tunnel->action.type))
7495 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7496 	tunnel->action.conf = tunnel;
7497 	tunnel->item.type = (typeof(tunnel->item.type))
7498 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7499 	tunnel->item.spec = tunnel;
7500 	tunnel->item.last = NULL;
7501 	tunnel->item.mask = NULL;
7502 
7503 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7504 		dev->data->port_id, tunnel->tunnel_id);
7505 
7506 	return tunnel;
7507 }
7508 
7509 static int
7510 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7511 		     const struct rte_flow_tunnel *app_tunnel,
7512 		     struct mlx5_flow_tunnel **tunnel)
7513 {
7514 	int ret;
7515 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7516 	struct mlx5_flow_tunnel *tun;
7517 
7518 	rte_spinlock_lock(&thub->sl);
7519 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7520 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7521 			    sizeof(*app_tunnel))) {
7522 			*tunnel = tun;
7523 			ret = 0;
7524 			break;
7525 		}
7526 	}
7527 	if (!tun) {
7528 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
7529 		if (tun) {
7530 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
7531 			*tunnel = tun;
7532 		} else {
7533 			ret = -ENOMEM;
7534 		}
7535 	}
7536 	rte_spinlock_unlock(&thub->sl);
7537 	if (tun)
7538 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
7539 
7540 	return ret;
7541 }
7542 
7543 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7544 {
7545 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7546 
7547 	if (!thub)
7548 		return;
7549 	if (!LIST_EMPTY(&thub->tunnels))
7550 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7551 	mlx5_hlist_destroy(thub->groups);
7552 	mlx5_free(thub);
7553 }
7554 
7555 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7556 {
7557 	int err;
7558 	struct mlx5_flow_tunnel_hub *thub;
7559 
7560 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7561 			   0, SOCKET_ID_ANY);
7562 	if (!thub)
7563 		return -ENOMEM;
7564 	LIST_INIT(&thub->tunnels);
7565 	rte_spinlock_init(&thub->sl);
7566 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7567 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7568 					 NULL,
7569 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7570 	if (!thub->groups) {
7571 		err = -rte_errno;
7572 		goto err;
7573 	}
7574 	thub->groups->ctx = sh;
7575 	sh->tunnel_hub = thub;
7576 
7577 	return 0;
7578 
7579 err:
7580 	if (thub->groups)
7581 		mlx5_hlist_destroy(thub->groups);
7582 	if (thub)
7583 		mlx5_free(thub);
7584 	return err;
7585 }
7586 
7587 #ifndef HAVE_MLX5DV_DR
7588 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7589 #else
7590 #define MLX5_DOMAIN_SYNC_FLOW \
7591 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7592 #endif
7593 
7594 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7595 {
7596 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7597 	const struct mlx5_flow_driver_ops *fops;
7598 	int ret;
7599 	struct rte_flow_attr attr = { .transfer = 0 };
7600 
7601 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7602 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7603 	if (ret > 0)
7604 		ret = -ret;
7605 	return ret;
7606 }
7607