xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision c2341bb6713dcaa43113db6f8ee3dd40ae57aba7)
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 /**
3270  * Get ASO age action by index.
3271  *
3272  * @param[in] dev
3273  *   Pointer to the Ethernet device structure.
3274  * @param[in] age_idx
3275  *   Index to the ASO age action.
3276  *
3277  * @return
3278  *   The specified ASO age action.
3279  */
3280 struct mlx5_aso_age_action*
3281 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3282 {
3283 	uint16_t pool_idx = age_idx & UINT16_MAX;
3284 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3285 	struct mlx5_priv *priv = dev->data->dev_private;
3286 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3287 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3288 
3289 	return &pool->actions[offset - 1];
3290 }
3291 
3292 /* maps shared action to translated non shared in some actions array */
3293 struct mlx5_translated_shared_action {
3294 	struct rte_flow_shared_action *action; /**< Shared action */
3295 	int index; /**< Index in related array of rte_flow_action */
3296 };
3297 
3298 /**
3299  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3300  * non shared action if translation possible.
3301  * This functionality used to run same execution path for both shared & non
3302  * shared actions on flow create. All necessary preparations for shared
3303  * action handling should be preformed on *shared* actions list returned
3304  * from this call.
3305  *
3306  * @param[in] dev
3307  *   Pointer to Ethernet device.
3308  * @param[in] actions
3309  *   List of actions to translate.
3310  * @param[out] shared
3311  *   List to store translated shared actions.
3312  * @param[in, out] shared_n
3313  *   Size of *shared* array. On return should be updated with number of shared
3314  *   actions retrieved from the *actions* list.
3315  * @param[out] translated_actions
3316  *   List of actions where all shared actions were translated to non shared
3317  *   if possible. NULL if no translation took place.
3318  * @param[out] error
3319  *   Pointer to the error structure.
3320  *
3321  * @return
3322  *   0 on success, a negative errno value otherwise and rte_errno is set.
3323  */
3324 static int
3325 flow_shared_actions_translate(struct rte_eth_dev *dev,
3326 			      const struct rte_flow_action actions[],
3327 			      struct mlx5_translated_shared_action *shared,
3328 			      int *shared_n,
3329 			      struct rte_flow_action **translated_actions,
3330 			      struct rte_flow_error *error)
3331 {
3332 	struct mlx5_priv *priv = dev->data->dev_private;
3333 	struct rte_flow_action *translated = NULL;
3334 	size_t actions_size;
3335 	int n;
3336 	int copied_n = 0;
3337 	struct mlx5_translated_shared_action *shared_end = NULL;
3338 
3339 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3340 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3341 			continue;
3342 		if (copied_n == *shared_n) {
3343 			return rte_flow_error_set
3344 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3345 				 NULL, "too many shared actions");
3346 		}
3347 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3348 			   sizeof(actions[n].conf));
3349 		shared[copied_n].index = n;
3350 		copied_n++;
3351 	}
3352 	n++;
3353 	*shared_n = copied_n;
3354 	if (!copied_n)
3355 		return 0;
3356 	actions_size = sizeof(struct rte_flow_action) * n;
3357 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3358 	if (!translated) {
3359 		rte_errno = ENOMEM;
3360 		return -ENOMEM;
3361 	}
3362 	memcpy(translated, actions, actions_size);
3363 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3364 		struct mlx5_shared_action_rss *shared_rss;
3365 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3366 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3367 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3368 									   - 1);
3369 
3370 		switch (type) {
3371 		case MLX5_SHARED_ACTION_TYPE_RSS:
3372 			shared_rss = mlx5_ipool_get
3373 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3374 			translated[shared->index].type =
3375 				RTE_FLOW_ACTION_TYPE_RSS;
3376 			translated[shared->index].conf =
3377 				&shared_rss->origin;
3378 			break;
3379 		case MLX5_SHARED_ACTION_TYPE_AGE:
3380 			if (priv->sh->flow_hit_aso_en) {
3381 				translated[shared->index].type =
3382 					(enum rte_flow_action_type)
3383 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3384 				translated[shared->index].conf =
3385 							 (void *)(uintptr_t)idx;
3386 				break;
3387 			}
3388 			/* Fall-through */
3389 		default:
3390 			mlx5_free(translated);
3391 			return rte_flow_error_set
3392 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3393 				 NULL, "invalid shared action type");
3394 		}
3395 	}
3396 	*translated_actions = translated;
3397 	return 0;
3398 }
3399 
3400 /**
3401  * Get Shared RSS action from the action list.
3402  *
3403  * @param[in] dev
3404  *   Pointer to Ethernet device.
3405  * @param[in] shared
3406  *   Pointer to the list of actions.
3407  * @param[in] shared_n
3408  *   Actions list length.
3409  *
3410  * @return
3411  *   The MLX5 RSS action ID if exists, otherwise return 0.
3412  */
3413 static uint32_t
3414 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3415 			   struct mlx5_translated_shared_action *shared,
3416 			   int shared_n)
3417 {
3418 	struct mlx5_translated_shared_action *shared_end;
3419 	struct mlx5_priv *priv = dev->data->dev_private;
3420 	struct mlx5_shared_action_rss *shared_rss;
3421 
3422 
3423 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3424 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3425 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3426 		uint32_t idx = act_idx &
3427 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3428 		switch (type) {
3429 		case MLX5_SHARED_ACTION_TYPE_RSS:
3430 			shared_rss = mlx5_ipool_get
3431 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3432 									   idx);
3433 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3434 					   __ATOMIC_RELAXED);
3435 			return idx;
3436 		default:
3437 			break;
3438 		}
3439 	}
3440 	return 0;
3441 }
3442 
3443 static unsigned int
3444 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3445 {
3446 	const struct rte_flow_item *item;
3447 	unsigned int has_vlan = 0;
3448 
3449 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3450 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3451 			has_vlan = 1;
3452 			break;
3453 		}
3454 	}
3455 	if (has_vlan)
3456 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3457 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3458 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3459 			       MLX5_EXPANSION_ROOT_OUTER;
3460 }
3461 
3462 /**
3463  *  Get layer flags from the prefix flow.
3464  *
3465  *  Some flows may be split to several subflows, the prefix subflow gets the
3466  *  match items and the suffix sub flow gets the actions.
3467  *  Some actions need the user defined match item flags to get the detail for
3468  *  the action.
3469  *  This function helps the suffix flow to get the item layer flags from prefix
3470  *  subflow.
3471  *
3472  * @param[in] dev_flow
3473  *   Pointer the created preifx subflow.
3474  *
3475  * @return
3476  *   The layers get from prefix subflow.
3477  */
3478 static inline uint64_t
3479 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3480 {
3481 	uint64_t layers = 0;
3482 
3483 	/*
3484 	 * Layers bits could be localization, but usually the compiler will
3485 	 * help to do the optimization work for source code.
3486 	 * If no decap actions, use the layers directly.
3487 	 */
3488 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3489 		return dev_flow->handle->layers;
3490 	/* Convert L3 layers with decap action. */
3491 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3492 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3493 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3494 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3495 	/* Convert L4 layers with decap action.  */
3496 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3497 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3498 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3499 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3500 	return layers;
3501 }
3502 
3503 /**
3504  * Get metadata split action information.
3505  *
3506  * @param[in] actions
3507  *   Pointer to the list of actions.
3508  * @param[out] qrss
3509  *   Pointer to the return pointer.
3510  * @param[out] qrss_type
3511  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3512  *   if no QUEUE/RSS is found.
3513  * @param[out] encap_idx
3514  *   Pointer to the index of the encap action if exists, otherwise the last
3515  *   action index.
3516  *
3517  * @return
3518  *   Total number of actions.
3519  */
3520 static int
3521 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3522 				       const struct rte_flow_action **qrss,
3523 				       int *encap_idx)
3524 {
3525 	const struct rte_flow_action_raw_encap *raw_encap;
3526 	int actions_n = 0;
3527 	int raw_decap_idx = -1;
3528 
3529 	*encap_idx = -1;
3530 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3531 		switch (actions->type) {
3532 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3533 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3534 			*encap_idx = actions_n;
3535 			break;
3536 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3537 			raw_decap_idx = actions_n;
3538 			break;
3539 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3540 			raw_encap = actions->conf;
3541 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3542 				*encap_idx = raw_decap_idx != -1 ?
3543 						      raw_decap_idx : actions_n;
3544 			break;
3545 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3546 		case RTE_FLOW_ACTION_TYPE_RSS:
3547 			*qrss = actions;
3548 			break;
3549 		default:
3550 			break;
3551 		}
3552 		actions_n++;
3553 	}
3554 	if (*encap_idx == -1)
3555 		*encap_idx = actions_n;
3556 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3557 	return actions_n + 1;
3558 }
3559 
3560 /**
3561  * Check meter action from the action list.
3562  *
3563  * @param[in] actions
3564  *   Pointer to the list of actions.
3565  * @param[out] mtr
3566  *   Pointer to the meter exist flag.
3567  *
3568  * @return
3569  *   Total number of actions.
3570  */
3571 static int
3572 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3573 {
3574 	int actions_n = 0;
3575 
3576 	MLX5_ASSERT(mtr);
3577 	*mtr = 0;
3578 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3579 		switch (actions->type) {
3580 		case RTE_FLOW_ACTION_TYPE_METER:
3581 			*mtr = 1;
3582 			break;
3583 		default:
3584 			break;
3585 		}
3586 		actions_n++;
3587 	}
3588 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3589 	return actions_n + 1;
3590 }
3591 
3592 /**
3593  * Check if the flow should be split due to hairpin.
3594  * The reason for the split is that in current HW we can't
3595  * support encap and push-vlan on Rx, so if a flow contains
3596  * these actions we move it to Tx.
3597  *
3598  * @param dev
3599  *   Pointer to Ethernet device.
3600  * @param[in] attr
3601  *   Flow rule attributes.
3602  * @param[in] actions
3603  *   Associated actions (list terminated by the END action).
3604  *
3605  * @return
3606  *   > 0 the number of actions and the flow should be split,
3607  *   0 when no split required.
3608  */
3609 static int
3610 flow_check_hairpin_split(struct rte_eth_dev *dev,
3611 			 const struct rte_flow_attr *attr,
3612 			 const struct rte_flow_action actions[])
3613 {
3614 	int queue_action = 0;
3615 	int action_n = 0;
3616 	int split = 0;
3617 	const struct rte_flow_action_queue *queue;
3618 	const struct rte_flow_action_rss *rss;
3619 	const struct rte_flow_action_raw_encap *raw_encap;
3620 	const struct rte_eth_hairpin_conf *conf;
3621 
3622 	if (!attr->ingress)
3623 		return 0;
3624 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3625 		switch (actions->type) {
3626 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3627 			queue = actions->conf;
3628 			if (queue == NULL)
3629 				return 0;
3630 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3631 			if (conf != NULL && !!conf->tx_explicit)
3632 				return 0;
3633 			queue_action = 1;
3634 			action_n++;
3635 			break;
3636 		case RTE_FLOW_ACTION_TYPE_RSS:
3637 			rss = actions->conf;
3638 			if (rss == NULL || rss->queue_num == 0)
3639 				return 0;
3640 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3641 			if (conf != NULL && !!conf->tx_explicit)
3642 				return 0;
3643 			queue_action = 1;
3644 			action_n++;
3645 			break;
3646 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3647 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3648 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3649 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3650 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3651 			split++;
3652 			action_n++;
3653 			break;
3654 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3655 			raw_encap = actions->conf;
3656 			if (raw_encap->size >
3657 			    (sizeof(struct rte_flow_item_eth) +
3658 			     sizeof(struct rte_flow_item_ipv4)))
3659 				split++;
3660 			action_n++;
3661 			break;
3662 		default:
3663 			action_n++;
3664 			break;
3665 		}
3666 	}
3667 	if (split && queue_action)
3668 		return action_n;
3669 	return 0;
3670 }
3671 
3672 /* Declare flow create/destroy prototype in advance. */
3673 static uint32_t
3674 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3675 		 const struct rte_flow_attr *attr,
3676 		 const struct rte_flow_item items[],
3677 		 const struct rte_flow_action actions[],
3678 		 bool external, struct rte_flow_error *error);
3679 
3680 static void
3681 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3682 		  uint32_t flow_idx);
3683 
3684 struct mlx5_hlist_entry *
3685 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3686 		       void *cb_ctx)
3687 {
3688 	struct rte_eth_dev *dev = list->ctx;
3689 	struct mlx5_priv *priv = dev->data->dev_private;
3690 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3691 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3692 	struct rte_flow_error *error = ctx->error;
3693 	uint32_t idx = 0;
3694 	int ret;
3695 	uint32_t mark_id = key;
3696 	struct rte_flow_attr attr = {
3697 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3698 		.ingress = 1,
3699 	};
3700 	struct mlx5_rte_flow_item_tag tag_spec = {
3701 		.data = mark_id,
3702 	};
3703 	struct rte_flow_item items[] = {
3704 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3705 	};
3706 	struct rte_flow_action_mark ftag = {
3707 		.id = mark_id,
3708 	};
3709 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3710 		.dst = REG_B,
3711 		.src = REG_NON,
3712 	};
3713 	struct rte_flow_action_jump jump = {
3714 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3715 	};
3716 	struct rte_flow_action actions[] = {
3717 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3718 	};
3719 
3720 	/* Fill the register fileds in the flow. */
3721 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3722 	if (ret < 0)
3723 		return NULL;
3724 	tag_spec.id = ret;
3725 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3726 	if (ret < 0)
3727 		return NULL;
3728 	cp_mreg.src = ret;
3729 	/* Provide the full width of FLAG specific value. */
3730 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3731 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3732 	/* Build a new flow. */
3733 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3734 		items[0] = (struct rte_flow_item){
3735 			.type = (enum rte_flow_item_type)
3736 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3737 			.spec = &tag_spec,
3738 		};
3739 		items[1] = (struct rte_flow_item){
3740 			.type = RTE_FLOW_ITEM_TYPE_END,
3741 		};
3742 		actions[0] = (struct rte_flow_action){
3743 			.type = (enum rte_flow_action_type)
3744 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3745 			.conf = &ftag,
3746 		};
3747 		actions[1] = (struct rte_flow_action){
3748 			.type = (enum rte_flow_action_type)
3749 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3750 			.conf = &cp_mreg,
3751 		};
3752 		actions[2] = (struct rte_flow_action){
3753 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3754 			.conf = &jump,
3755 		};
3756 		actions[3] = (struct rte_flow_action){
3757 			.type = RTE_FLOW_ACTION_TYPE_END,
3758 		};
3759 	} else {
3760 		/* Default rule, wildcard match. */
3761 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3762 		items[0] = (struct rte_flow_item){
3763 			.type = RTE_FLOW_ITEM_TYPE_END,
3764 		};
3765 		actions[0] = (struct rte_flow_action){
3766 			.type = (enum rte_flow_action_type)
3767 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3768 			.conf = &cp_mreg,
3769 		};
3770 		actions[1] = (struct rte_flow_action){
3771 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3772 			.conf = &jump,
3773 		};
3774 		actions[2] = (struct rte_flow_action){
3775 			.type = RTE_FLOW_ACTION_TYPE_END,
3776 		};
3777 	}
3778 	/* Build a new entry. */
3779 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3780 	if (!mcp_res) {
3781 		rte_errno = ENOMEM;
3782 		return NULL;
3783 	}
3784 	mcp_res->idx = idx;
3785 	/*
3786 	 * The copy Flows are not included in any list. There
3787 	 * ones are referenced from other Flows and can not
3788 	 * be applied, removed, deleted in ardbitrary order
3789 	 * by list traversing.
3790 	 */
3791 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3792 					 actions, false, error);
3793 	if (!mcp_res->rix_flow) {
3794 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3795 		return NULL;
3796 	}
3797 	return &mcp_res->hlist_ent;
3798 }
3799 
3800 /**
3801  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3802  *
3803  * As mark_id is unique, if there's already a registered flow for the mark_id,
3804  * return by increasing the reference counter of the resource. Otherwise, create
3805  * the resource (mcp_res) and flow.
3806  *
3807  * Flow looks like,
3808  *   - If ingress port is ANY and reg_c[1] is mark_id,
3809  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3810  *
3811  * For default flow (zero mark_id), flow is like,
3812  *   - If ingress port is ANY,
3813  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3814  *
3815  * @param dev
3816  *   Pointer to Ethernet device.
3817  * @param mark_id
3818  *   ID of MARK action, zero means default flow for META.
3819  * @param[out] error
3820  *   Perform verbose error reporting if not NULL.
3821  *
3822  * @return
3823  *   Associated resource on success, NULL otherwise and rte_errno is set.
3824  */
3825 static struct mlx5_flow_mreg_copy_resource *
3826 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3827 			  struct rte_flow_error *error)
3828 {
3829 	struct mlx5_priv *priv = dev->data->dev_private;
3830 	struct mlx5_hlist_entry *entry;
3831 	struct mlx5_flow_cb_ctx ctx = {
3832 		.dev = dev,
3833 		.error = error,
3834 	};
3835 
3836 	/* Check if already registered. */
3837 	MLX5_ASSERT(priv->mreg_cp_tbl);
3838 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3839 	if (!entry)
3840 		return NULL;
3841 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3842 			    hlist_ent);
3843 }
3844 
3845 void
3846 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3847 {
3848 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3849 		container_of(entry, typeof(*mcp_res), hlist_ent);
3850 	struct rte_eth_dev *dev = list->ctx;
3851 	struct mlx5_priv *priv = dev->data->dev_private;
3852 
3853 	MLX5_ASSERT(mcp_res->rix_flow);
3854 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3855 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3856 }
3857 
3858 /**
3859  * Release flow in RX_CP_TBL.
3860  *
3861  * @param dev
3862  *   Pointer to Ethernet device.
3863  * @flow
3864  *   Parent flow for wich copying is provided.
3865  */
3866 static void
3867 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3868 			  struct rte_flow *flow)
3869 {
3870 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3871 	struct mlx5_priv *priv = dev->data->dev_private;
3872 
3873 	if (!flow->rix_mreg_copy)
3874 		return;
3875 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3876 				 flow->rix_mreg_copy);
3877 	if (!mcp_res || !priv->mreg_cp_tbl)
3878 		return;
3879 	MLX5_ASSERT(mcp_res->rix_flow);
3880 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3881 	flow->rix_mreg_copy = 0;
3882 }
3883 
3884 /**
3885  * Remove the default copy action from RX_CP_TBL.
3886  *
3887  * This functions is called in the mlx5_dev_start(). No thread safe
3888  * is guaranteed.
3889  *
3890  * @param dev
3891  *   Pointer to Ethernet device.
3892  */
3893 static void
3894 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3895 {
3896 	struct mlx5_hlist_entry *entry;
3897 	struct mlx5_priv *priv = dev->data->dev_private;
3898 
3899 	/* Check if default flow is registered. */
3900 	if (!priv->mreg_cp_tbl)
3901 		return;
3902 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3903 				  MLX5_DEFAULT_COPY_ID, NULL);
3904 	if (!entry)
3905 		return;
3906 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3907 }
3908 
3909 /**
3910  * Add the default copy action in in RX_CP_TBL.
3911  *
3912  * This functions is called in the mlx5_dev_start(). No thread safe
3913  * is guaranteed.
3914  *
3915  * @param dev
3916  *   Pointer to Ethernet device.
3917  * @param[out] error
3918  *   Perform verbose error reporting if not NULL.
3919  *
3920  * @return
3921  *   0 for success, negative value otherwise and rte_errno is set.
3922  */
3923 static int
3924 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3925 				  struct rte_flow_error *error)
3926 {
3927 	struct mlx5_priv *priv = dev->data->dev_private;
3928 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3929 
3930 	/* Check whether extensive metadata feature is engaged. */
3931 	if (!priv->config.dv_flow_en ||
3932 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3933 	    !mlx5_flow_ext_mreg_supported(dev) ||
3934 	    !priv->sh->dv_regc0_mask)
3935 		return 0;
3936 	/*
3937 	 * Add default mreg copy flow may be called multiple time, but
3938 	 * only be called once in stop. Avoid register it twice.
3939 	 */
3940 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
3941 		return 0;
3942 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3943 	if (!mcp_res)
3944 		return -rte_errno;
3945 	return 0;
3946 }
3947 
3948 /**
3949  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3950  *
3951  * All the flow having Q/RSS action should be split by
3952  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3953  * performs the following,
3954  *   - CQE->flow_tag := reg_c[1] (MARK)
3955  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3956  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3957  * but there should be a flow per each MARK ID set by MARK action.
3958  *
3959  * For the aforementioned reason, if there's a MARK action in flow's action
3960  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3961  * the MARK ID to CQE's flow_tag like,
3962  *   - If reg_c[1] is mark_id,
3963  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3964  *
3965  * For SET_META action which stores value in reg_c[0], as the destination is
3966  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3967  * MARK ID means the default flow. The default flow looks like,
3968  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3969  *
3970  * @param dev
3971  *   Pointer to Ethernet device.
3972  * @param flow
3973  *   Pointer to flow structure.
3974  * @param[in] actions
3975  *   Pointer to the list of actions.
3976  * @param[out] error
3977  *   Perform verbose error reporting if not NULL.
3978  *
3979  * @return
3980  *   0 on success, negative value otherwise and rte_errno is set.
3981  */
3982 static int
3983 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3984 			    struct rte_flow *flow,
3985 			    const struct rte_flow_action *actions,
3986 			    struct rte_flow_error *error)
3987 {
3988 	struct mlx5_priv *priv = dev->data->dev_private;
3989 	struct mlx5_dev_config *config = &priv->config;
3990 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3991 	const struct rte_flow_action_mark *mark;
3992 
3993 	/* Check whether extensive metadata feature is engaged. */
3994 	if (!config->dv_flow_en ||
3995 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3996 	    !mlx5_flow_ext_mreg_supported(dev) ||
3997 	    !priv->sh->dv_regc0_mask)
3998 		return 0;
3999 	/* Find MARK action. */
4000 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4001 		switch (actions->type) {
4002 		case RTE_FLOW_ACTION_TYPE_FLAG:
4003 			mcp_res = flow_mreg_add_copy_action
4004 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4005 			if (!mcp_res)
4006 				return -rte_errno;
4007 			flow->rix_mreg_copy = mcp_res->idx;
4008 			return 0;
4009 		case RTE_FLOW_ACTION_TYPE_MARK:
4010 			mark = (const struct rte_flow_action_mark *)
4011 				actions->conf;
4012 			mcp_res =
4013 				flow_mreg_add_copy_action(dev, mark->id, error);
4014 			if (!mcp_res)
4015 				return -rte_errno;
4016 			flow->rix_mreg_copy = mcp_res->idx;
4017 			return 0;
4018 		default:
4019 			break;
4020 		}
4021 	}
4022 	return 0;
4023 }
4024 
4025 #define MLX5_MAX_SPLIT_ACTIONS 24
4026 #define MLX5_MAX_SPLIT_ITEMS 24
4027 
4028 /**
4029  * Split the hairpin flow.
4030  * Since HW can't support encap and push-vlan on Rx, we move these
4031  * actions to Tx.
4032  * If the count action is after the encap then we also
4033  * move the count action. in this case the count will also measure
4034  * the outer bytes.
4035  *
4036  * @param dev
4037  *   Pointer to Ethernet device.
4038  * @param[in] actions
4039  *   Associated actions (list terminated by the END action).
4040  * @param[out] actions_rx
4041  *   Rx flow actions.
4042  * @param[out] actions_tx
4043  *   Tx flow actions..
4044  * @param[out] pattern_tx
4045  *   The pattern items for the Tx flow.
4046  * @param[out] flow_id
4047  *   The flow ID connected to this flow.
4048  *
4049  * @return
4050  *   0 on success.
4051  */
4052 static int
4053 flow_hairpin_split(struct rte_eth_dev *dev,
4054 		   const struct rte_flow_action actions[],
4055 		   struct rte_flow_action actions_rx[],
4056 		   struct rte_flow_action actions_tx[],
4057 		   struct rte_flow_item pattern_tx[],
4058 		   uint32_t flow_id)
4059 {
4060 	const struct rte_flow_action_raw_encap *raw_encap;
4061 	const struct rte_flow_action_raw_decap *raw_decap;
4062 	struct mlx5_rte_flow_action_set_tag *set_tag;
4063 	struct rte_flow_action *tag_action;
4064 	struct mlx5_rte_flow_item_tag *tag_item;
4065 	struct rte_flow_item *item;
4066 	char *addr;
4067 	int encap = 0;
4068 
4069 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4070 		switch (actions->type) {
4071 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4072 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4073 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4074 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4075 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4076 			rte_memcpy(actions_tx, actions,
4077 			       sizeof(struct rte_flow_action));
4078 			actions_tx++;
4079 			break;
4080 		case RTE_FLOW_ACTION_TYPE_COUNT:
4081 			if (encap) {
4082 				rte_memcpy(actions_tx, actions,
4083 					   sizeof(struct rte_flow_action));
4084 				actions_tx++;
4085 			} else {
4086 				rte_memcpy(actions_rx, actions,
4087 					   sizeof(struct rte_flow_action));
4088 				actions_rx++;
4089 			}
4090 			break;
4091 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4092 			raw_encap = actions->conf;
4093 			if (raw_encap->size >
4094 			    (sizeof(struct rte_flow_item_eth) +
4095 			     sizeof(struct rte_flow_item_ipv4))) {
4096 				memcpy(actions_tx, actions,
4097 				       sizeof(struct rte_flow_action));
4098 				actions_tx++;
4099 				encap = 1;
4100 			} else {
4101 				rte_memcpy(actions_rx, actions,
4102 					   sizeof(struct rte_flow_action));
4103 				actions_rx++;
4104 			}
4105 			break;
4106 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4107 			raw_decap = actions->conf;
4108 			if (raw_decap->size <
4109 			    (sizeof(struct rte_flow_item_eth) +
4110 			     sizeof(struct rte_flow_item_ipv4))) {
4111 				memcpy(actions_tx, actions,
4112 				       sizeof(struct rte_flow_action));
4113 				actions_tx++;
4114 			} else {
4115 				rte_memcpy(actions_rx, actions,
4116 					   sizeof(struct rte_flow_action));
4117 				actions_rx++;
4118 			}
4119 			break;
4120 		default:
4121 			rte_memcpy(actions_rx, actions,
4122 				   sizeof(struct rte_flow_action));
4123 			actions_rx++;
4124 			break;
4125 		}
4126 	}
4127 	/* Add set meta action and end action for the Rx flow. */
4128 	tag_action = actions_rx;
4129 	tag_action->type = (enum rte_flow_action_type)
4130 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4131 	actions_rx++;
4132 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4133 	actions_rx++;
4134 	set_tag = (void *)actions_rx;
4135 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4136 	MLX5_ASSERT(set_tag->id > REG_NON);
4137 	set_tag->data = flow_id;
4138 	tag_action->conf = set_tag;
4139 	/* Create Tx item list. */
4140 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4141 	addr = (void *)&pattern_tx[2];
4142 	item = pattern_tx;
4143 	item->type = (enum rte_flow_item_type)
4144 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4145 	tag_item = (void *)addr;
4146 	tag_item->data = flow_id;
4147 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4148 	MLX5_ASSERT(set_tag->id > REG_NON);
4149 	item->spec = tag_item;
4150 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4151 	tag_item = (void *)addr;
4152 	tag_item->data = UINT32_MAX;
4153 	tag_item->id = UINT16_MAX;
4154 	item->mask = tag_item;
4155 	item->last = NULL;
4156 	item++;
4157 	item->type = RTE_FLOW_ITEM_TYPE_END;
4158 	return 0;
4159 }
4160 
4161 __extension__
4162 union tunnel_offload_mark {
4163 	uint32_t val;
4164 	struct {
4165 		uint32_t app_reserve:8;
4166 		uint32_t table_id:15;
4167 		uint32_t transfer:1;
4168 		uint32_t _unused_:8;
4169 	};
4170 };
4171 
4172 struct tunnel_default_miss_ctx {
4173 	uint16_t *queue;
4174 	__extension__
4175 	union {
4176 		struct rte_flow_action_rss action_rss;
4177 		struct rte_flow_action_queue miss_queue;
4178 		struct rte_flow_action_jump miss_jump;
4179 		uint8_t raw[0];
4180 	};
4181 };
4182 
4183 static int
4184 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4185 			     struct rte_flow *flow,
4186 			     const struct rte_flow_attr *attr,
4187 			     const struct rte_flow_action *app_actions,
4188 			     uint32_t flow_idx,
4189 			     struct tunnel_default_miss_ctx *ctx,
4190 			     struct rte_flow_error *error)
4191 {
4192 	struct mlx5_priv *priv = dev->data->dev_private;
4193 	struct mlx5_flow *dev_flow;
4194 	struct rte_flow_attr miss_attr = *attr;
4195 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4196 	const struct rte_flow_item miss_items[2] = {
4197 		{
4198 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4199 			.spec = NULL,
4200 			.last = NULL,
4201 			.mask = NULL
4202 		},
4203 		{
4204 			.type = RTE_FLOW_ITEM_TYPE_END,
4205 			.spec = NULL,
4206 			.last = NULL,
4207 			.mask = NULL
4208 		}
4209 	};
4210 	union tunnel_offload_mark mark_id;
4211 	struct rte_flow_action_mark miss_mark;
4212 	struct rte_flow_action miss_actions[3] = {
4213 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4214 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
4215 	};
4216 	const struct rte_flow_action_jump *jump_data;
4217 	uint32_t i, flow_table = 0; /* prevent compilation warning */
4218 	struct flow_grp_info grp_info = {
4219 		.external = 1,
4220 		.transfer = attr->transfer,
4221 		.fdb_def_rule = !!priv->fdb_def_rule,
4222 		.std_tbl_fix = 0,
4223 	};
4224 	int ret;
4225 
4226 	if (!attr->transfer) {
4227 		uint32_t q_size;
4228 
4229 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4230 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4231 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4232 					 0, SOCKET_ID_ANY);
4233 		if (!ctx->queue)
4234 			return rte_flow_error_set
4235 				(error, ENOMEM,
4236 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4237 				NULL, "invalid default miss RSS");
4238 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4239 		ctx->action_rss.level = 0,
4240 		ctx->action_rss.types = priv->rss_conf.rss_hf,
4241 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4242 		ctx->action_rss.queue_num = priv->reta_idx_n,
4243 		ctx->action_rss.key = priv->rss_conf.rss_key,
4244 		ctx->action_rss.queue = ctx->queue;
4245 		if (!priv->reta_idx_n || !priv->rxqs_n)
4246 			return rte_flow_error_set
4247 				(error, EINVAL,
4248 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4249 				NULL, "invalid port configuration");
4250 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4251 			ctx->action_rss.types = 0;
4252 		for (i = 0; i != priv->reta_idx_n; ++i)
4253 			ctx->queue[i] = (*priv->reta_idx)[i];
4254 	} else {
4255 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4256 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4257 	}
4258 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4259 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4260 	jump_data = app_actions->conf;
4261 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4262 	miss_attr.group = jump_data->group;
4263 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4264 				       &flow_table, grp_info, error);
4265 	if (ret)
4266 		return rte_flow_error_set(error, EINVAL,
4267 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4268 					  NULL, "invalid tunnel id");
4269 	mark_id.app_reserve = 0;
4270 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4271 	mark_id.transfer = !!attr->transfer;
4272 	mark_id._unused_ = 0;
4273 	miss_mark.id = mark_id.val;
4274 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4275 				    miss_items, miss_actions, flow_idx, error);
4276 	if (!dev_flow)
4277 		return -rte_errno;
4278 	dev_flow->flow = flow;
4279 	dev_flow->external = true;
4280 	dev_flow->tunnel = tunnel;
4281 	/* Subflow object was created, we must include one in the list. */
4282 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4283 		      dev_flow->handle, next);
4284 	DRV_LOG(DEBUG,
4285 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4286 		dev->data->port_id, tunnel->app_tunnel.type,
4287 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4288 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4289 				  miss_actions, error);
4290 	if (!ret)
4291 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4292 						  error);
4293 
4294 	return ret;
4295 }
4296 
4297 /**
4298  * The last stage of splitting chain, just creates the subflow
4299  * without any modification.
4300  *
4301  * @param[in] dev
4302  *   Pointer to Ethernet device.
4303  * @param[in] flow
4304  *   Parent flow structure pointer.
4305  * @param[in, out] sub_flow
4306  *   Pointer to return the created subflow, may be NULL.
4307  * @param[in] prefix_layers
4308  *   Prefix subflow layers, may be 0.
4309  * @param[in] prefix_mark
4310  *   Prefix subflow mark flag, may be 0.
4311  * @param[in] attr
4312  *   Flow rule attributes.
4313  * @param[in] items
4314  *   Pattern specification (list terminated by the END pattern item).
4315  * @param[in] actions
4316  *   Associated actions (list terminated by the END action).
4317  * @param[in] external
4318  *   This flow rule is created by request external to PMD.
4319  * @param[in] flow_idx
4320  *   This memory pool index to the flow.
4321  * @param[out] error
4322  *   Perform verbose error reporting if not NULL.
4323  * @return
4324  *   0 on success, negative value otherwise
4325  */
4326 static int
4327 flow_create_split_inner(struct rte_eth_dev *dev,
4328 			struct rte_flow *flow,
4329 			struct mlx5_flow **sub_flow,
4330 			uint64_t prefix_layers,
4331 			uint32_t prefix_mark,
4332 			const struct rte_flow_attr *attr,
4333 			const struct rte_flow_item items[],
4334 			const struct rte_flow_action actions[],
4335 			bool external, uint32_t flow_idx,
4336 			struct rte_flow_error *error)
4337 {
4338 	struct mlx5_flow *dev_flow;
4339 
4340 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4341 		flow_idx, error);
4342 	if (!dev_flow)
4343 		return -rte_errno;
4344 	dev_flow->flow = flow;
4345 	dev_flow->external = external;
4346 	/* Subflow object was created, we must include one in the list. */
4347 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4348 		      dev_flow->handle, next);
4349 	/*
4350 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4351 	 * flow may need some user defined item layer flags, and pass the
4352 	 * Metadate rxq mark flag to suffix flow as well.
4353 	 */
4354 	if (prefix_layers)
4355 		dev_flow->handle->layers = prefix_layers;
4356 	if (prefix_mark)
4357 		dev_flow->handle->mark = 1;
4358 	if (sub_flow)
4359 		*sub_flow = dev_flow;
4360 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4361 }
4362 
4363 /**
4364  * Split the meter flow.
4365  *
4366  * As meter flow will split to three sub flow, other than meter
4367  * action, the other actions make sense to only meter accepts
4368  * the packet. If it need to be dropped, no other additional
4369  * actions should be take.
4370  *
4371  * One kind of special action which decapsulates the L3 tunnel
4372  * header will be in the prefix sub flow, as not to take the
4373  * L3 tunnel header into account.
4374  *
4375  * @param dev
4376  *   Pointer to Ethernet device.
4377  * @param[in] items
4378  *   Pattern specification (list terminated by the END pattern item).
4379  * @param[out] sfx_items
4380  *   Suffix flow match items (list terminated by the END pattern item).
4381  * @param[in] actions
4382  *   Associated actions (list terminated by the END action).
4383  * @param[out] actions_sfx
4384  *   Suffix flow actions.
4385  * @param[out] actions_pre
4386  *   Prefix flow actions.
4387  * @param[out] pattern_sfx
4388  *   The pattern items for the suffix flow.
4389  * @param[out] tag_sfx
4390  *   Pointer to suffix flow tag.
4391  *
4392  * @return
4393  *   0 on success.
4394  */
4395 static int
4396 flow_meter_split_prep(struct rte_eth_dev *dev,
4397 		 const struct rte_flow_item items[],
4398 		 struct rte_flow_item sfx_items[],
4399 		 const struct rte_flow_action actions[],
4400 		 struct rte_flow_action actions_sfx[],
4401 		 struct rte_flow_action actions_pre[])
4402 {
4403 	struct mlx5_priv *priv = dev->data->dev_private;
4404 	struct rte_flow_action *tag_action = NULL;
4405 	struct rte_flow_item *tag_item;
4406 	struct mlx5_rte_flow_action_set_tag *set_tag;
4407 	struct rte_flow_error error;
4408 	const struct rte_flow_action_raw_encap *raw_encap;
4409 	const struct rte_flow_action_raw_decap *raw_decap;
4410 	struct mlx5_rte_flow_item_tag *tag_spec;
4411 	struct mlx5_rte_flow_item_tag *tag_mask;
4412 	uint32_t tag_id = 0;
4413 	bool copy_vlan = false;
4414 
4415 	/* Prepare the actions for prefix and suffix flow. */
4416 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4417 		struct rte_flow_action **action_cur = NULL;
4418 
4419 		switch (actions->type) {
4420 		case RTE_FLOW_ACTION_TYPE_METER:
4421 			/* Add the extra tag action first. */
4422 			tag_action = actions_pre;
4423 			tag_action->type = (enum rte_flow_action_type)
4424 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4425 			actions_pre++;
4426 			action_cur = &actions_pre;
4427 			break;
4428 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4429 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4430 			action_cur = &actions_pre;
4431 			break;
4432 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4433 			raw_encap = actions->conf;
4434 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4435 				action_cur = &actions_pre;
4436 			break;
4437 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4438 			raw_decap = actions->conf;
4439 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4440 				action_cur = &actions_pre;
4441 			break;
4442 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4443 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4444 			copy_vlan = true;
4445 			break;
4446 		default:
4447 			break;
4448 		}
4449 		if (!action_cur)
4450 			action_cur = &actions_sfx;
4451 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4452 		(*action_cur)++;
4453 	}
4454 	/* Add end action to the actions. */
4455 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4456 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4457 	actions_pre++;
4458 	/* Set the tag. */
4459 	set_tag = (void *)actions_pre;
4460 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4461 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4462 			  &tag_id);
4463 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4464 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4465 			dev->data->port_id);
4466 		mlx5_ipool_free(priv->sh->ipool
4467 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4468 		return 0;
4469 	} else if (!tag_id) {
4470 		return 0;
4471 	}
4472 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4473 	assert(tag_action);
4474 	tag_action->conf = set_tag;
4475 	/* Prepare the suffix subflow items. */
4476 	tag_item = sfx_items++;
4477 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4478 		int item_type = items->type;
4479 
4480 		switch (item_type) {
4481 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4482 			memcpy(sfx_items, items, sizeof(*sfx_items));
4483 			sfx_items++;
4484 			break;
4485 		case RTE_FLOW_ITEM_TYPE_VLAN:
4486 			if (copy_vlan) {
4487 				memcpy(sfx_items, items, sizeof(*sfx_items));
4488 				/*
4489 				 * Convert to internal match item, it is used
4490 				 * for vlan push and set vid.
4491 				 */
4492 				sfx_items->type = (enum rte_flow_item_type)
4493 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4494 				sfx_items++;
4495 			}
4496 			break;
4497 		default:
4498 			break;
4499 		}
4500 	}
4501 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4502 	sfx_items++;
4503 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4504 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4505 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4506 	tag_mask = tag_spec + 1;
4507 	tag_mask->data = 0xffffff00;
4508 	tag_item->type = (enum rte_flow_item_type)
4509 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4510 	tag_item->spec = tag_spec;
4511 	tag_item->last = NULL;
4512 	tag_item->mask = tag_mask;
4513 	return tag_id;
4514 }
4515 
4516 /**
4517  * Split action list having QUEUE/RSS for metadata register copy.
4518  *
4519  * Once Q/RSS action is detected in user's action list, the flow action
4520  * should be split in order to copy metadata registers, which will happen in
4521  * RX_CP_TBL like,
4522  *   - CQE->flow_tag := reg_c[1] (MARK)
4523  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4524  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4525  * This is because the last action of each flow must be a terminal action
4526  * (QUEUE, RSS or DROP).
4527  *
4528  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4529  * stored and kept in the mlx5_flow structure per each sub_flow.
4530  *
4531  * The Q/RSS action is replaced with,
4532  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4533  * And the following JUMP action is added at the end,
4534  *   - JUMP, to RX_CP_TBL.
4535  *
4536  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4537  * flow_create_split_metadata() routine. The flow will look like,
4538  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4539  *
4540  * @param dev
4541  *   Pointer to Ethernet device.
4542  * @param[out] split_actions
4543  *   Pointer to store split actions to jump to CP_TBL.
4544  * @param[in] actions
4545  *   Pointer to the list of original flow actions.
4546  * @param[in] qrss
4547  *   Pointer to the Q/RSS action.
4548  * @param[in] actions_n
4549  *   Number of original actions.
4550  * @param[out] error
4551  *   Perform verbose error reporting if not NULL.
4552  *
4553  * @return
4554  *   non-zero unique flow_id on success, otherwise 0 and
4555  *   error/rte_error are set.
4556  */
4557 static uint32_t
4558 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4559 			  struct rte_flow_action *split_actions,
4560 			  const struct rte_flow_action *actions,
4561 			  const struct rte_flow_action *qrss,
4562 			  int actions_n, struct rte_flow_error *error)
4563 {
4564 	struct mlx5_priv *priv = dev->data->dev_private;
4565 	struct mlx5_rte_flow_action_set_tag *set_tag;
4566 	struct rte_flow_action_jump *jump;
4567 	const int qrss_idx = qrss - actions;
4568 	uint32_t flow_id = 0;
4569 	int ret = 0;
4570 
4571 	/*
4572 	 * Given actions will be split
4573 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4574 	 * - Add jump to mreg CP_TBL.
4575 	 * As a result, there will be one more action.
4576 	 */
4577 	++actions_n;
4578 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4579 	set_tag = (void *)(split_actions + actions_n);
4580 	/*
4581 	 * If tag action is not set to void(it means we are not the meter
4582 	 * suffix flow), add the tag action. Since meter suffix flow already
4583 	 * has the tag added.
4584 	 */
4585 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4586 		/*
4587 		 * Allocate the new subflow ID. This one is unique within
4588 		 * device and not shared with representors. Otherwise,
4589 		 * we would have to resolve multi-thread access synch
4590 		 * issue. Each flow on the shared device is appended
4591 		 * with source vport identifier, so the resulting
4592 		 * flows will be unique in the shared (by master and
4593 		 * representors) domain even if they have coinciding
4594 		 * IDs.
4595 		 */
4596 		mlx5_ipool_malloc(priv->sh->ipool
4597 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4598 		if (!flow_id)
4599 			return rte_flow_error_set(error, ENOMEM,
4600 						  RTE_FLOW_ERROR_TYPE_ACTION,
4601 						  NULL, "can't allocate id "
4602 						  "for split Q/RSS subflow");
4603 		/* Internal SET_TAG action to set flow ID. */
4604 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4605 			.data = flow_id,
4606 		};
4607 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4608 		if (ret < 0)
4609 			return ret;
4610 		set_tag->id = ret;
4611 		/* Construct new actions array. */
4612 		/* Replace QUEUE/RSS action. */
4613 		split_actions[qrss_idx] = (struct rte_flow_action){
4614 			.type = (enum rte_flow_action_type)
4615 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4616 			.conf = set_tag,
4617 		};
4618 	}
4619 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4620 	jump = (void *)(set_tag + 1);
4621 	*jump = (struct rte_flow_action_jump){
4622 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4623 	};
4624 	split_actions[actions_n - 2] = (struct rte_flow_action){
4625 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4626 		.conf = jump,
4627 	};
4628 	split_actions[actions_n - 1] = (struct rte_flow_action){
4629 		.type = RTE_FLOW_ACTION_TYPE_END,
4630 	};
4631 	return flow_id;
4632 }
4633 
4634 /**
4635  * Extend the given action list for Tx metadata copy.
4636  *
4637  * Copy the given action list to the ext_actions and add flow metadata register
4638  * copy action in order to copy reg_a set by WQE to reg_c[0].
4639  *
4640  * @param[out] ext_actions
4641  *   Pointer to the extended action list.
4642  * @param[in] actions
4643  *   Pointer to the list of actions.
4644  * @param[in] actions_n
4645  *   Number of actions in the list.
4646  * @param[out] error
4647  *   Perform verbose error reporting if not NULL.
4648  * @param[in] encap_idx
4649  *   The encap action inndex.
4650  *
4651  * @return
4652  *   0 on success, negative value otherwise
4653  */
4654 static int
4655 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4656 		       struct rte_flow_action *ext_actions,
4657 		       const struct rte_flow_action *actions,
4658 		       int actions_n, struct rte_flow_error *error,
4659 		       int encap_idx)
4660 {
4661 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4662 		(struct mlx5_flow_action_copy_mreg *)
4663 			(ext_actions + actions_n + 1);
4664 	int ret;
4665 
4666 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4667 	if (ret < 0)
4668 		return ret;
4669 	cp_mreg->dst = ret;
4670 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4671 	if (ret < 0)
4672 		return ret;
4673 	cp_mreg->src = ret;
4674 	if (encap_idx != 0)
4675 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4676 	if (encap_idx == actions_n - 1) {
4677 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4678 			.type = (enum rte_flow_action_type)
4679 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4680 			.conf = cp_mreg,
4681 		};
4682 		ext_actions[actions_n] = (struct rte_flow_action){
4683 			.type = RTE_FLOW_ACTION_TYPE_END,
4684 		};
4685 	} else {
4686 		ext_actions[encap_idx] = (struct rte_flow_action){
4687 			.type = (enum rte_flow_action_type)
4688 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4689 			.conf = cp_mreg,
4690 		};
4691 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4692 				sizeof(*ext_actions) * (actions_n - encap_idx));
4693 	}
4694 	return 0;
4695 }
4696 
4697 /**
4698  * Check the match action from the action list.
4699  *
4700  * @param[in] actions
4701  *   Pointer to the list of actions.
4702  * @param[in] attr
4703  *   Flow rule attributes.
4704  * @param[in] action
4705  *   The action to be check if exist.
4706  * @param[out] match_action_pos
4707  *   Pointer to the position of the matched action if exists, otherwise is -1.
4708  * @param[out] qrss_action_pos
4709  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4710  *
4711  * @return
4712  *   > 0 the total number of actions.
4713  *   0 if not found match action in action list.
4714  */
4715 static int
4716 flow_check_match_action(const struct rte_flow_action actions[],
4717 			const struct rte_flow_attr *attr,
4718 			enum rte_flow_action_type action,
4719 			int *match_action_pos, int *qrss_action_pos)
4720 {
4721 	const struct rte_flow_action_sample *sample;
4722 	int actions_n = 0;
4723 	int jump_flag = 0;
4724 	uint32_t ratio = 0;
4725 	int sub_type = 0;
4726 	int flag = 0;
4727 
4728 	*match_action_pos = -1;
4729 	*qrss_action_pos = -1;
4730 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4731 		if (actions->type == action) {
4732 			flag = 1;
4733 			*match_action_pos = actions_n;
4734 		}
4735 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4736 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4737 			*qrss_action_pos = actions_n;
4738 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4739 			jump_flag = 1;
4740 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4741 			sample = actions->conf;
4742 			ratio = sample->ratio;
4743 			sub_type = ((const struct rte_flow_action *)
4744 					(sample->actions))->type;
4745 		}
4746 		actions_n++;
4747 	}
4748 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4749 		if (ratio == 1) {
4750 			/* JUMP Action not support for Mirroring;
4751 			 * Mirroring support multi-destination;
4752 			 */
4753 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4754 				flag = 0;
4755 		}
4756 	}
4757 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4758 	return flag ? actions_n + 1 : 0;
4759 }
4760 
4761 #define SAMPLE_SUFFIX_ITEM 2
4762 
4763 /**
4764  * Split the sample flow.
4765  *
4766  * As sample flow will split to two sub flow, sample flow with
4767  * sample action, the other actions will move to new suffix flow.
4768  *
4769  * Also add unique tag id with tag action in the sample flow,
4770  * the same tag id will be as match in the suffix flow.
4771  *
4772  * @param dev
4773  *   Pointer to Ethernet device.
4774  * @param[in] fdb_tx
4775  *   FDB egress flow flag.
4776  * @param[out] sfx_items
4777  *   Suffix flow match items (list terminated by the END pattern item).
4778  * @param[in] actions
4779  *   Associated actions (list terminated by the END action).
4780  * @param[out] actions_sfx
4781  *   Suffix flow actions.
4782  * @param[out] actions_pre
4783  *   Prefix flow actions.
4784  * @param[in] actions_n
4785  *  The total number of actions.
4786  * @param[in] sample_action_pos
4787  *   The sample action position.
4788  * @param[in] qrss_action_pos
4789  *   The Queue/RSS action position.
4790  * @param[out] error
4791  *   Perform verbose error reporting if not NULL.
4792  *
4793  * @return
4794  *   0 on success, or unique flow_id, a negative errno value
4795  *   otherwise and rte_errno is set.
4796  */
4797 static int
4798 flow_sample_split_prep(struct rte_eth_dev *dev,
4799 		       uint32_t fdb_tx,
4800 		       struct rte_flow_item sfx_items[],
4801 		       const struct rte_flow_action actions[],
4802 		       struct rte_flow_action actions_sfx[],
4803 		       struct rte_flow_action actions_pre[],
4804 		       int actions_n,
4805 		       int sample_action_pos,
4806 		       int qrss_action_pos,
4807 		       struct rte_flow_error *error)
4808 {
4809 	struct mlx5_priv *priv = dev->data->dev_private;
4810 	struct mlx5_rte_flow_action_set_tag *set_tag;
4811 	struct mlx5_rte_flow_item_tag *tag_spec;
4812 	struct mlx5_rte_flow_item_tag *tag_mask;
4813 	uint32_t tag_id = 0;
4814 	int index;
4815 	int ret;
4816 
4817 	if (sample_action_pos < 0)
4818 		return rte_flow_error_set(error, EINVAL,
4819 					  RTE_FLOW_ERROR_TYPE_ACTION,
4820 					  NULL, "invalid position of sample "
4821 					  "action in list");
4822 	if (!fdb_tx) {
4823 		/* Prepare the prefix tag action. */
4824 		set_tag = (void *)(actions_pre + actions_n + 1);
4825 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4826 		if (ret < 0)
4827 			return ret;
4828 		set_tag->id = ret;
4829 		mlx5_ipool_malloc(priv->sh->ipool
4830 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4831 		set_tag->data = tag_id;
4832 		/* Prepare the suffix subflow items. */
4833 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4834 		tag_spec->data = tag_id;
4835 		tag_spec->id = set_tag->id;
4836 		tag_mask = tag_spec + 1;
4837 		tag_mask->data = UINT32_MAX;
4838 		sfx_items[0] = (struct rte_flow_item){
4839 			.type = (enum rte_flow_item_type)
4840 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4841 			.spec = tag_spec,
4842 			.last = NULL,
4843 			.mask = tag_mask,
4844 		};
4845 		sfx_items[1] = (struct rte_flow_item){
4846 			.type = (enum rte_flow_item_type)
4847 				RTE_FLOW_ITEM_TYPE_END,
4848 		};
4849 	}
4850 	/* Prepare the actions for prefix and suffix flow. */
4851 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4852 		index = qrss_action_pos;
4853 		/* Put the preceding the Queue/RSS action into prefix flow. */
4854 		if (index != 0)
4855 			memcpy(actions_pre, actions,
4856 			       sizeof(struct rte_flow_action) * index);
4857 		/* Put others preceding the sample action into prefix flow. */
4858 		if (sample_action_pos > index + 1)
4859 			memcpy(actions_pre + index, actions + index + 1,
4860 			       sizeof(struct rte_flow_action) *
4861 			       (sample_action_pos - index - 1));
4862 		index = sample_action_pos - 1;
4863 		/* Put Queue/RSS action into Suffix flow. */
4864 		memcpy(actions_sfx, actions + qrss_action_pos,
4865 		       sizeof(struct rte_flow_action));
4866 		actions_sfx++;
4867 	} else {
4868 		index = sample_action_pos;
4869 		if (index != 0)
4870 			memcpy(actions_pre, actions,
4871 			       sizeof(struct rte_flow_action) * index);
4872 	}
4873 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4874 	if (!fdb_tx) {
4875 		actions_pre[index++] =
4876 			(struct rte_flow_action){
4877 			.type = (enum rte_flow_action_type)
4878 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4879 			.conf = set_tag,
4880 		};
4881 	}
4882 	memcpy(actions_pre + index, actions + sample_action_pos,
4883 	       sizeof(struct rte_flow_action));
4884 	index += 1;
4885 	actions_pre[index] = (struct rte_flow_action){
4886 		.type = (enum rte_flow_action_type)
4887 			RTE_FLOW_ACTION_TYPE_END,
4888 	};
4889 	/* Put the actions after sample into Suffix flow. */
4890 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4891 	       sizeof(struct rte_flow_action) *
4892 	       (actions_n - sample_action_pos - 1));
4893 	return tag_id;
4894 }
4895 
4896 /**
4897  * The splitting for metadata feature.
4898  *
4899  * - Q/RSS action on NIC Rx should be split in order to pass by
4900  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4901  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4902  *
4903  * - All the actions on NIC Tx should have a mreg copy action to
4904  *   copy reg_a from WQE to reg_c[0].
4905  *
4906  * @param dev
4907  *   Pointer to Ethernet device.
4908  * @param[in] flow
4909  *   Parent flow structure pointer.
4910  * @param[in] prefix_layers
4911  *   Prefix flow layer flags.
4912  * @param[in] prefix_mark
4913  *   Prefix subflow mark flag, may be 0.
4914  * @param[in] attr
4915  *   Flow rule attributes.
4916  * @param[in] items
4917  *   Pattern specification (list terminated by the END pattern item).
4918  * @param[in] actions
4919  *   Associated actions (list terminated by the END action).
4920  * @param[in] external
4921  *   This flow rule is created by request external to PMD.
4922  * @param[in] flow_idx
4923  *   This memory pool index to the flow.
4924  * @param[out] error
4925  *   Perform verbose error reporting if not NULL.
4926  * @return
4927  *   0 on success, negative value otherwise
4928  */
4929 static int
4930 flow_create_split_metadata(struct rte_eth_dev *dev,
4931 			   struct rte_flow *flow,
4932 			   uint64_t prefix_layers,
4933 			   uint32_t prefix_mark,
4934 			   const struct rte_flow_attr *attr,
4935 			   const struct rte_flow_item items[],
4936 			   const struct rte_flow_action actions[],
4937 			   bool external, uint32_t flow_idx,
4938 			   struct rte_flow_error *error)
4939 {
4940 	struct mlx5_priv *priv = dev->data->dev_private;
4941 	struct mlx5_dev_config *config = &priv->config;
4942 	const struct rte_flow_action *qrss = NULL;
4943 	struct rte_flow_action *ext_actions = NULL;
4944 	struct mlx5_flow *dev_flow = NULL;
4945 	uint32_t qrss_id = 0;
4946 	int mtr_sfx = 0;
4947 	size_t act_size;
4948 	int actions_n;
4949 	int encap_idx;
4950 	int ret;
4951 
4952 	/* Check whether extensive metadata feature is engaged. */
4953 	if (!config->dv_flow_en ||
4954 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4955 	    !mlx5_flow_ext_mreg_supported(dev))
4956 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
4957 					       prefix_mark, attr, items,
4958 					       actions, external, flow_idx,
4959 					       error);
4960 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4961 							   &encap_idx);
4962 	if (qrss) {
4963 		/* Exclude hairpin flows from splitting. */
4964 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4965 			const struct rte_flow_action_queue *queue;
4966 
4967 			queue = qrss->conf;
4968 			if (mlx5_rxq_get_type(dev, queue->index) ==
4969 			    MLX5_RXQ_TYPE_HAIRPIN)
4970 				qrss = NULL;
4971 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4972 			const struct rte_flow_action_rss *rss;
4973 
4974 			rss = qrss->conf;
4975 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4976 			    MLX5_RXQ_TYPE_HAIRPIN)
4977 				qrss = NULL;
4978 		}
4979 	}
4980 	if (qrss) {
4981 		/* Check if it is in meter suffix table. */
4982 		mtr_sfx = attr->group == (attr->transfer ?
4983 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4984 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4985 		/*
4986 		 * Q/RSS action on NIC Rx should be split in order to pass by
4987 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4988 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4989 		 */
4990 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4991 			   sizeof(struct rte_flow_action_set_tag) +
4992 			   sizeof(struct rte_flow_action_jump);
4993 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4994 					  SOCKET_ID_ANY);
4995 		if (!ext_actions)
4996 			return rte_flow_error_set(error, ENOMEM,
4997 						  RTE_FLOW_ERROR_TYPE_ACTION,
4998 						  NULL, "no memory to split "
4999 						  "metadata flow");
5000 		/*
5001 		 * If we are the suffix flow of meter, tag already exist.
5002 		 * Set the tag action to void.
5003 		 */
5004 		if (mtr_sfx)
5005 			ext_actions[qrss - actions].type =
5006 						RTE_FLOW_ACTION_TYPE_VOID;
5007 		else
5008 			ext_actions[qrss - actions].type =
5009 						(enum rte_flow_action_type)
5010 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5011 		/*
5012 		 * Create the new actions list with removed Q/RSS action
5013 		 * and appended set tag and jump to register copy table
5014 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5015 		 * in advance, because it is needed for set tag action.
5016 		 */
5017 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5018 						    qrss, actions_n, error);
5019 		if (!mtr_sfx && !qrss_id) {
5020 			ret = -rte_errno;
5021 			goto exit;
5022 		}
5023 	} else if (attr->egress && !attr->transfer) {
5024 		/*
5025 		 * All the actions on NIC Tx should have a metadata register
5026 		 * copy action to copy reg_a from WQE to reg_c[meta]
5027 		 */
5028 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5029 			   sizeof(struct mlx5_flow_action_copy_mreg);
5030 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5031 					  SOCKET_ID_ANY);
5032 		if (!ext_actions)
5033 			return rte_flow_error_set(error, ENOMEM,
5034 						  RTE_FLOW_ERROR_TYPE_ACTION,
5035 						  NULL, "no memory to split "
5036 						  "metadata flow");
5037 		/* Create the action list appended with copy register. */
5038 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5039 					     actions_n, error, encap_idx);
5040 		if (ret < 0)
5041 			goto exit;
5042 	}
5043 	/* Add the unmodified original or prefix subflow. */
5044 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers,
5045 				      prefix_mark, attr,
5046 				      items, ext_actions ? ext_actions :
5047 				      actions, external, flow_idx, error);
5048 	if (ret < 0)
5049 		goto exit;
5050 	MLX5_ASSERT(dev_flow);
5051 	if (qrss) {
5052 		const struct rte_flow_attr q_attr = {
5053 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5054 			.ingress = 1,
5055 		};
5056 		/* Internal PMD action to set register. */
5057 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5058 			.data = qrss_id,
5059 			.id = REG_NON,
5060 		};
5061 		struct rte_flow_item q_items[] = {
5062 			{
5063 				.type = (enum rte_flow_item_type)
5064 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5065 				.spec = &q_tag_spec,
5066 				.last = NULL,
5067 				.mask = NULL,
5068 			},
5069 			{
5070 				.type = RTE_FLOW_ITEM_TYPE_END,
5071 			},
5072 		};
5073 		struct rte_flow_action q_actions[] = {
5074 			{
5075 				.type = qrss->type,
5076 				.conf = qrss->conf,
5077 			},
5078 			{
5079 				.type = RTE_FLOW_ACTION_TYPE_END,
5080 			},
5081 		};
5082 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5083 
5084 		/*
5085 		 * Configure the tag item only if there is no meter subflow.
5086 		 * Since tag is already marked in the meter suffix subflow
5087 		 * we can just use the meter suffix items as is.
5088 		 */
5089 		if (qrss_id) {
5090 			/* Not meter subflow. */
5091 			MLX5_ASSERT(!mtr_sfx);
5092 			/*
5093 			 * Put unique id in prefix flow due to it is destroyed
5094 			 * after suffix flow and id will be freed after there
5095 			 * is no actual flows with this id and identifier
5096 			 * reallocation becomes possible (for example, for
5097 			 * other flows in other threads).
5098 			 */
5099 			dev_flow->handle->split_flow_id = qrss_id;
5100 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5101 						   error);
5102 			if (ret < 0)
5103 				goto exit;
5104 			q_tag_spec.id = ret;
5105 		}
5106 		dev_flow = NULL;
5107 		/* Add suffix subflow to execute Q/RSS. */
5108 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers, 0,
5109 					      &q_attr, mtr_sfx ? items :
5110 					      q_items, q_actions,
5111 					      external, flow_idx, error);
5112 		if (ret < 0)
5113 			goto exit;
5114 		/* qrss ID should be freed if failed. */
5115 		qrss_id = 0;
5116 		MLX5_ASSERT(dev_flow);
5117 	}
5118 
5119 exit:
5120 	/*
5121 	 * We do not destroy the partially created sub_flows in case of error.
5122 	 * These ones are included into parent flow list and will be destroyed
5123 	 * by flow_drv_destroy.
5124 	 */
5125 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5126 			qrss_id);
5127 	mlx5_free(ext_actions);
5128 	return ret;
5129 }
5130 
5131 /**
5132  * The splitting for meter feature.
5133  *
5134  * - The meter flow will be split to two flows as prefix and
5135  *   suffix flow. The packets make sense only it pass the prefix
5136  *   meter action.
5137  *
5138  * - Reg_C_5 is used for the packet to match betweend prefix and
5139  *   suffix flow.
5140  *
5141  * @param dev
5142  *   Pointer to Ethernet device.
5143  * @param[in] flow
5144  *   Parent flow structure pointer.
5145  * @param[in] prefix_layers
5146  *   Prefix subflow layers, may be 0.
5147  * @param[in] prefix_mark
5148  *   Prefix subflow mark flag, may be 0.
5149  * @param[in] attr
5150  *   Flow rule attributes.
5151  * @param[in] items
5152  *   Pattern specification (list terminated by the END pattern item).
5153  * @param[in] actions
5154  *   Associated actions (list terminated by the END action).
5155  * @param[in] external
5156  *   This flow rule is created by request external to PMD.
5157  * @param[in] flow_idx
5158  *   This memory pool index to the flow.
5159  * @param[out] error
5160  *   Perform verbose error reporting if not NULL.
5161  * @return
5162  *   0 on success, negative value otherwise
5163  */
5164 static int
5165 flow_create_split_meter(struct rte_eth_dev *dev,
5166 			struct rte_flow *flow,
5167 			uint64_t prefix_layers,
5168 			uint32_t prefix_mark,
5169 			const struct rte_flow_attr *attr,
5170 			const struct rte_flow_item items[],
5171 			const struct rte_flow_action actions[],
5172 			bool external, uint32_t flow_idx,
5173 			struct rte_flow_error *error)
5174 {
5175 	struct mlx5_priv *priv = dev->data->dev_private;
5176 	struct rte_flow_action *sfx_actions = NULL;
5177 	struct rte_flow_action *pre_actions = NULL;
5178 	struct rte_flow_item *sfx_items = NULL;
5179 	struct mlx5_flow *dev_flow = NULL;
5180 	struct rte_flow_attr sfx_attr = *attr;
5181 	uint32_t mtr = 0;
5182 	uint32_t mtr_tag_id = 0;
5183 	size_t act_size;
5184 	size_t item_size;
5185 	int actions_n = 0;
5186 	int ret;
5187 
5188 	if (priv->mtr_en)
5189 		actions_n = flow_check_meter_action(actions, &mtr);
5190 	if (mtr) {
5191 		/* The five prefix actions: meter, decap, encap, tag, end. */
5192 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5193 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5194 		/* tag, vlan, port id, end. */
5195 #define METER_SUFFIX_ITEM 4
5196 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5197 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5198 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5199 					  0, SOCKET_ID_ANY);
5200 		if (!sfx_actions)
5201 			return rte_flow_error_set(error, ENOMEM,
5202 						  RTE_FLOW_ERROR_TYPE_ACTION,
5203 						  NULL, "no memory to split "
5204 						  "meter flow");
5205 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5206 			     act_size);
5207 		pre_actions = sfx_actions + actions_n;
5208 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5209 						   actions, sfx_actions,
5210 						   pre_actions);
5211 		if (!mtr_tag_id) {
5212 			ret = -rte_errno;
5213 			goto exit;
5214 		}
5215 		/* Add the prefix subflow. */
5216 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5217 					      prefix_layers, 0,
5218 					      attr, items,
5219 					      pre_actions, external,
5220 					      flow_idx, error);
5221 		if (ret) {
5222 			ret = -rte_errno;
5223 			goto exit;
5224 		}
5225 		dev_flow->handle->split_flow_id = mtr_tag_id;
5226 		/* Setting the sfx group atrr. */
5227 		sfx_attr.group = sfx_attr.transfer ?
5228 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5229 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5230 	}
5231 	/* Add the prefix subflow. */
5232 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
5233 					 flow_get_prefix_layer_flags(dev_flow) :
5234 					 prefix_layers, dev_flow ?
5235 					 dev_flow->handle->mark : prefix_mark,
5236 					 &sfx_attr, sfx_items ?
5237 					 sfx_items : items,
5238 					 sfx_actions ? sfx_actions : actions,
5239 					 external, flow_idx, error);
5240 exit:
5241 	if (sfx_actions)
5242 		mlx5_free(sfx_actions);
5243 	return ret;
5244 }
5245 
5246 /**
5247  * The splitting for sample feature.
5248  *
5249  * Once Sample action is detected in the action list, the flow actions should
5250  * be split into prefix sub flow and suffix sub flow.
5251  *
5252  * The original items remain in the prefix sub flow, all actions preceding the
5253  * sample action and the sample action itself will be copied to the prefix
5254  * sub flow, the actions following the sample action will be copied to the
5255  * suffix sub flow, Queue action always be located in the suffix sub flow.
5256  *
5257  * In order to make the packet from prefix sub flow matches with suffix sub
5258  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5259  * flow uses tag item with the unique flow id.
5260  *
5261  * @param dev
5262  *   Pointer to Ethernet device.
5263  * @param[in] flow
5264  *   Parent flow structure pointer.
5265  * @param[in] attr
5266  *   Flow rule attributes.
5267  * @param[in] items
5268  *   Pattern specification (list terminated by the END pattern item).
5269  * @param[in] actions
5270  *   Associated actions (list terminated by the END action).
5271  * @param[in] external
5272  *   This flow rule is created by request external to PMD.
5273  * @param[in] flow_idx
5274  *   This memory pool index to the flow.
5275  * @param[out] error
5276  *   Perform verbose error reporting if not NULL.
5277  * @return
5278  *   0 on success, negative value otherwise
5279  */
5280 static int
5281 flow_create_split_sample(struct rte_eth_dev *dev,
5282 			 struct rte_flow *flow,
5283 			 const struct rte_flow_attr *attr,
5284 			 const struct rte_flow_item items[],
5285 			 const struct rte_flow_action actions[],
5286 			 bool external, uint32_t flow_idx,
5287 			 struct rte_flow_error *error)
5288 {
5289 	struct mlx5_priv *priv = dev->data->dev_private;
5290 	struct rte_flow_action *sfx_actions = NULL;
5291 	struct rte_flow_action *pre_actions = NULL;
5292 	struct rte_flow_item *sfx_items = NULL;
5293 	struct mlx5_flow *dev_flow = NULL;
5294 	struct rte_flow_attr sfx_attr = *attr;
5295 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5296 	struct mlx5_flow_dv_sample_resource *sample_res;
5297 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5298 	struct mlx5_flow_tbl_resource *sfx_tbl;
5299 	union mlx5_flow_tbl_key sfx_table_key;
5300 #endif
5301 	size_t act_size;
5302 	size_t item_size;
5303 	uint32_t fdb_tx = 0;
5304 	int32_t tag_id = 0;
5305 	int actions_n = 0;
5306 	int sample_action_pos;
5307 	int qrss_action_pos;
5308 	int ret = 0;
5309 
5310 	if (priv->sampler_en)
5311 		actions_n = flow_check_match_action(actions, attr,
5312 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5313 					&sample_action_pos, &qrss_action_pos);
5314 	if (actions_n) {
5315 		/* The prefix actions must includes sample, tag, end. */
5316 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5317 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5318 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5319 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5320 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5321 					  item_size), 0, SOCKET_ID_ANY);
5322 		if (!sfx_actions)
5323 			return rte_flow_error_set(error, ENOMEM,
5324 						  RTE_FLOW_ERROR_TYPE_ACTION,
5325 						  NULL, "no memory to split "
5326 						  "sample flow");
5327 		/* The representor_id is -1 for uplink. */
5328 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5329 		if (!fdb_tx)
5330 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5331 					+ act_size);
5332 		pre_actions = sfx_actions + actions_n;
5333 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5334 						actions, sfx_actions,
5335 						pre_actions, actions_n,
5336 						sample_action_pos,
5337 						qrss_action_pos, error);
5338 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5339 			ret = -rte_errno;
5340 			goto exit;
5341 		}
5342 		/* Add the prefix subflow. */
5343 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, 0, attr,
5344 					      items, pre_actions, external,
5345 					      flow_idx, error);
5346 		if (ret) {
5347 			ret = -rte_errno;
5348 			goto exit;
5349 		}
5350 		dev_flow->handle->split_flow_id = tag_id;
5351 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5352 		/* Set the sfx group attr. */
5353 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5354 					dev_flow->dv.sample_res;
5355 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5356 					sample_res->normal_path_tbl;
5357 		sfx_tbl_data = container_of(sfx_tbl,
5358 					struct mlx5_flow_tbl_data_entry, tbl);
5359 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5360 		sfx_attr.group = sfx_attr.transfer ?
5361 					(sfx_table_key.table_id - 1) :
5362 					 sfx_table_key.table_id;
5363 #endif
5364 	}
5365 	/* Add the suffix subflow. */
5366 	ret = flow_create_split_meter(dev, flow, dev_flow ?
5367 				 flow_get_prefix_layer_flags(dev_flow) : 0,
5368 				 dev_flow ? dev_flow->handle->mark : 0,
5369 				 &sfx_attr, sfx_items ? sfx_items : items,
5370 				 sfx_actions ? sfx_actions : actions,
5371 				 external, flow_idx, error);
5372 exit:
5373 	if (sfx_actions)
5374 		mlx5_free(sfx_actions);
5375 	return ret;
5376 }
5377 
5378 /**
5379  * Split the flow to subflow set. The splitters might be linked
5380  * in the chain, like this:
5381  * flow_create_split_outer() calls:
5382  *   flow_create_split_meter() calls:
5383  *     flow_create_split_metadata(meter_subflow_0) calls:
5384  *       flow_create_split_inner(metadata_subflow_0)
5385  *       flow_create_split_inner(metadata_subflow_1)
5386  *       flow_create_split_inner(metadata_subflow_2)
5387  *     flow_create_split_metadata(meter_subflow_1) calls:
5388  *       flow_create_split_inner(metadata_subflow_0)
5389  *       flow_create_split_inner(metadata_subflow_1)
5390  *       flow_create_split_inner(metadata_subflow_2)
5391  *
5392  * This provide flexible way to add new levels of flow splitting.
5393  * The all of successfully created subflows are included to the
5394  * parent flow dev_flow list.
5395  *
5396  * @param dev
5397  *   Pointer to Ethernet device.
5398  * @param[in] flow
5399  *   Parent flow structure pointer.
5400  * @param[in] attr
5401  *   Flow rule attributes.
5402  * @param[in] items
5403  *   Pattern specification (list terminated by the END pattern item).
5404  * @param[in] actions
5405  *   Associated actions (list terminated by the END action).
5406  * @param[in] external
5407  *   This flow rule is created by request external to PMD.
5408  * @param[in] flow_idx
5409  *   This memory pool index to the flow.
5410  * @param[out] error
5411  *   Perform verbose error reporting if not NULL.
5412  * @return
5413  *   0 on success, negative value otherwise
5414  */
5415 static int
5416 flow_create_split_outer(struct rte_eth_dev *dev,
5417 			struct rte_flow *flow,
5418 			const struct rte_flow_attr *attr,
5419 			const struct rte_flow_item items[],
5420 			const struct rte_flow_action actions[],
5421 			bool external, uint32_t flow_idx,
5422 			struct rte_flow_error *error)
5423 {
5424 	int ret;
5425 
5426 	ret = flow_create_split_sample(dev, flow, attr, items,
5427 				       actions, external, flow_idx, error);
5428 	MLX5_ASSERT(ret <= 0);
5429 	return ret;
5430 }
5431 
5432 static struct mlx5_flow_tunnel *
5433 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5434 		      const struct rte_flow_attr *attr,
5435 		      const struct rte_flow_item items[],
5436 		      const struct rte_flow_action actions[])
5437 {
5438 	struct mlx5_flow_tunnel *tunnel;
5439 
5440 #pragma GCC diagnostic push
5441 #pragma GCC diagnostic ignored "-Wcast-qual"
5442 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5443 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5444 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5445 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5446 	else
5447 		tunnel = NULL;
5448 #pragma GCC diagnostic pop
5449 
5450 	return tunnel;
5451 }
5452 
5453 /**
5454  * Adjust flow RSS workspace if needed.
5455  *
5456  * @param wks
5457  *   Pointer to thread flow work space.
5458  * @param rss_desc
5459  *   Pointer to RSS descriptor.
5460  * @param[in] nrssq_num
5461  *   New RSS queue number.
5462  *
5463  * @return
5464  *   0 on success, -1 otherwise and rte_errno is set.
5465  */
5466 static int
5467 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5468 			  struct mlx5_flow_rss_desc *rss_desc,
5469 			  uint32_t nrssq_num)
5470 {
5471 	bool fidx = !!wks->flow_idx;
5472 
5473 	if (likely(nrssq_num <= wks->rssq_num[fidx]))
5474 		return 0;
5475 	rss_desc->queue = realloc(rss_desc->queue,
5476 			  sizeof(rss_desc->queue[0]) * RTE_ALIGN(nrssq_num, 2));
5477 	if (!rss_desc->queue) {
5478 		rte_errno = ENOMEM;
5479 		return -1;
5480 	}
5481 	wks->rssq_num[fidx] = RTE_ALIGN(nrssq_num, 2);
5482 	return 0;
5483 }
5484 
5485 /**
5486  * Create a flow and add it to @p list.
5487  *
5488  * @param dev
5489  *   Pointer to Ethernet device.
5490  * @param list
5491  *   Pointer to a TAILQ flow list. If this parameter NULL,
5492  *   no list insertion occurred, flow is just created,
5493  *   this is caller's responsibility to track the
5494  *   created flow.
5495  * @param[in] attr
5496  *   Flow rule attributes.
5497  * @param[in] items
5498  *   Pattern specification (list terminated by the END pattern item).
5499  * @param[in] actions
5500  *   Associated actions (list terminated by the END action).
5501  * @param[in] external
5502  *   This flow rule is created by request external to PMD.
5503  * @param[out] error
5504  *   Perform verbose error reporting if not NULL.
5505  *
5506  * @return
5507  *   A flow index on success, 0 otherwise and rte_errno is set.
5508  */
5509 static uint32_t
5510 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5511 		 const struct rte_flow_attr *attr,
5512 		 const struct rte_flow_item items[],
5513 		 const struct rte_flow_action original_actions[],
5514 		 bool external, struct rte_flow_error *error)
5515 {
5516 	struct mlx5_priv *priv = dev->data->dev_private;
5517 	struct rte_flow *flow = NULL;
5518 	struct mlx5_flow *dev_flow;
5519 	const struct rte_flow_action_rss *rss;
5520 	struct mlx5_translated_shared_action
5521 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5522 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5523 	union {
5524 		struct mlx5_flow_expand_rss buf;
5525 		uint8_t buffer[2048];
5526 	} expand_buffer;
5527 	union {
5528 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5529 		uint8_t buffer[2048];
5530 	} actions_rx;
5531 	union {
5532 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5533 		uint8_t buffer[2048];
5534 	} actions_hairpin_tx;
5535 	union {
5536 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5537 		uint8_t buffer[2048];
5538 	} items_tx;
5539 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5540 	struct mlx5_flow_rss_desc *rss_desc;
5541 	const struct rte_flow_action *p_actions_rx;
5542 	uint32_t i;
5543 	uint32_t idx = 0;
5544 	int hairpin_flow;
5545 	struct rte_flow_attr attr_tx = { .priority = 0 };
5546 	struct rte_flow_attr attr_factor = {0};
5547 	const struct rte_flow_action *actions;
5548 	struct rte_flow_action *translated_actions = NULL;
5549 	struct mlx5_flow_tunnel *tunnel;
5550 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5551 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5552 	bool fidx = !!wks->flow_idx;
5553 	int ret;
5554 
5555 	MLX5_ASSERT(wks);
5556 	rss_desc = &wks->rss_desc[fidx];
5557 	ret = flow_shared_actions_translate(dev, original_actions,
5558 					    shared_actions,
5559 					    &shared_actions_n,
5560 					    &translated_actions, error);
5561 	if (ret < 0) {
5562 		MLX5_ASSERT(translated_actions == NULL);
5563 		return 0;
5564 	}
5565 	actions = translated_actions ? translated_actions : original_actions;
5566 	memcpy((void *)&attr_factor, (const void *)attr, sizeof(*attr));
5567 	p_actions_rx = actions;
5568 	hairpin_flow = flow_check_hairpin_split(dev, &attr_factor, actions);
5569 	ret = flow_drv_validate(dev, &attr_factor, items, p_actions_rx,
5570 				external, hairpin_flow, error);
5571 	if (ret < 0)
5572 		goto error_before_hairpin_split;
5573 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5574 	if (!flow) {
5575 		rte_errno = ENOMEM;
5576 		goto error_before_hairpin_split;
5577 	}
5578 	if (hairpin_flow > 0) {
5579 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5580 			rte_errno = EINVAL;
5581 			goto error_before_hairpin_split;
5582 		}
5583 		flow_hairpin_split(dev, actions, actions_rx.actions,
5584 				   actions_hairpin_tx.actions, items_tx.items,
5585 				   idx);
5586 		p_actions_rx = actions_rx.actions;
5587 	}
5588 	flow->drv_type = flow_get_drv_type(dev, &attr_factor);
5589 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5590 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5591 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5592 	rss = flow_get_rss_action(p_actions_rx);
5593 	if (rss) {
5594 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5595 			return 0;
5596 		/*
5597 		 * The following information is required by
5598 		 * mlx5_flow_hashfields_adjust() in advance.
5599 		 */
5600 		rss_desc->level = rss->level;
5601 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5602 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5603 	}
5604 	flow->dev_handles = 0;
5605 	if (rss && rss->types) {
5606 		unsigned int graph_root;
5607 
5608 		graph_root = find_graph_root(items, rss->level);
5609 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5610 					   items, rss->types,
5611 					   mlx5_support_expansion, graph_root);
5612 		MLX5_ASSERT(ret > 0 &&
5613 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5614 	} else {
5615 		buf->entries = 1;
5616 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5617 	}
5618 	flow->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5619 						      shared_actions_n);
5620 	/*
5621 	 * Record the start index when there is a nested call. All sub-flows
5622 	 * need to be translated before another calling.
5623 	 * No need to use ping-pong buffer to save memory here.
5624 	 */
5625 	if (fidx) {
5626 		MLX5_ASSERT(!wks->flow_nested_idx);
5627 		wks->flow_nested_idx = fidx;
5628 	}
5629 	for (i = 0; i < buf->entries; ++i) {
5630 		/*
5631 		 * The splitter may create multiple dev_flows,
5632 		 * depending on configuration. In the simplest
5633 		 * case it just creates unmodified original flow.
5634 		 */
5635 		ret = flow_create_split_outer(dev, flow, &attr_factor,
5636 					      buf->entry[i].pattern,
5637 					      p_actions_rx, external, idx,
5638 					      error);
5639 		if (ret < 0)
5640 			goto error;
5641 		if (is_flow_tunnel_steer_rule(dev, attr,
5642 					      buf->entry[i].pattern,
5643 					      p_actions_rx)) {
5644 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5645 							   p_actions_rx,
5646 							   idx,
5647 							   &default_miss_ctx,
5648 							   error);
5649 			if (ret < 0) {
5650 				mlx5_free(default_miss_ctx.queue);
5651 				goto error;
5652 			}
5653 		}
5654 	}
5655 	/* Create the tx flow. */
5656 	if (hairpin_flow) {
5657 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5658 		attr_tx.ingress = 0;
5659 		attr_tx.egress = 1;
5660 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5661 					 actions_hairpin_tx.actions,
5662 					 idx, error);
5663 		if (!dev_flow)
5664 			goto error;
5665 		dev_flow->flow = flow;
5666 		dev_flow->external = 0;
5667 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5668 			      dev_flow->handle, next);
5669 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5670 					 items_tx.items,
5671 					 actions_hairpin_tx.actions, error);
5672 		if (ret < 0)
5673 			goto error;
5674 	}
5675 	/*
5676 	 * Update the metadata register copy table. If extensive
5677 	 * metadata feature is enabled and registers are supported
5678 	 * we might create the extra rte_flow for each unique
5679 	 * MARK/FLAG action ID.
5680 	 *
5681 	 * The table is updated for ingress Flows only, because
5682 	 * the egress Flows belong to the different device and
5683 	 * copy table should be updated in peer NIC Rx domain.
5684 	 */
5685 	if (attr_factor.ingress &&
5686 	    (external || attr_factor.group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5687 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5688 		if (ret)
5689 			goto error;
5690 	}
5691 	/*
5692 	 * If the flow is external (from application) OR device is started, then
5693 	 * the flow will be applied immediately.
5694 	 */
5695 	if (external || dev->data->dev_started) {
5696 		ret = flow_drv_apply(dev, flow, error);
5697 		if (ret < 0)
5698 			goto error;
5699 	}
5700 	if (list) {
5701 		rte_spinlock_lock(&priv->flow_list_lock);
5702 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5703 			     flow, next);
5704 		rte_spinlock_unlock(&priv->flow_list_lock);
5705 	}
5706 	flow_rxq_flags_set(dev, flow);
5707 	rte_free(translated_actions);
5708 	/* Nested flow creation index recovery. */
5709 	wks->flow_idx = wks->flow_nested_idx;
5710 	if (wks->flow_nested_idx)
5711 		wks->flow_nested_idx = 0;
5712 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5713 	if (tunnel) {
5714 		flow->tunnel = 1;
5715 		flow->tunnel_id = tunnel->tunnel_id;
5716 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5717 		mlx5_free(default_miss_ctx.queue);
5718 	}
5719 	return idx;
5720 error:
5721 	MLX5_ASSERT(flow);
5722 	ret = rte_errno; /* Save rte_errno before cleanup. */
5723 	flow_mreg_del_copy_action(dev, flow);
5724 	flow_drv_destroy(dev, flow);
5725 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5726 	rte_errno = ret; /* Restore rte_errno. */
5727 	ret = rte_errno;
5728 	rte_errno = ret;
5729 	wks->flow_idx = wks->flow_nested_idx;
5730 	if (wks->flow_nested_idx)
5731 		wks->flow_nested_idx = 0;
5732 error_before_hairpin_split:
5733 	rte_free(translated_actions);
5734 	return 0;
5735 }
5736 
5737 /**
5738  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5739  * incoming packets to table 1.
5740  *
5741  * Other flow rules, requested for group n, will be created in
5742  * e-switch table n+1.
5743  * Jump action to e-switch group n will be created to group n+1.
5744  *
5745  * Used when working in switchdev mode, to utilise advantages of table 1
5746  * and above.
5747  *
5748  * @param dev
5749  *   Pointer to Ethernet device.
5750  *
5751  * @return
5752  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5753  */
5754 struct rte_flow *
5755 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5756 {
5757 	const struct rte_flow_attr attr = {
5758 		.group = 0,
5759 		.priority = 0,
5760 		.ingress = 1,
5761 		.egress = 0,
5762 		.transfer = 1,
5763 	};
5764 	const struct rte_flow_item pattern = {
5765 		.type = RTE_FLOW_ITEM_TYPE_END,
5766 	};
5767 	struct rte_flow_action_jump jump = {
5768 		.group = 1,
5769 	};
5770 	const struct rte_flow_action actions[] = {
5771 		{
5772 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5773 			.conf = &jump,
5774 		},
5775 		{
5776 			.type = RTE_FLOW_ACTION_TYPE_END,
5777 		},
5778 	};
5779 	struct mlx5_priv *priv = dev->data->dev_private;
5780 	struct rte_flow_error error;
5781 
5782 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5783 						   &attr, &pattern,
5784 						   actions, false, &error);
5785 }
5786 
5787 /**
5788  * Validate a flow supported by the NIC.
5789  *
5790  * @see rte_flow_validate()
5791  * @see rte_flow_ops
5792  */
5793 int
5794 mlx5_flow_validate(struct rte_eth_dev *dev,
5795 		   const struct rte_flow_attr *attr,
5796 		   const struct rte_flow_item items[],
5797 		   const struct rte_flow_action original_actions[],
5798 		   struct rte_flow_error *error)
5799 {
5800 	int hairpin_flow;
5801 	struct mlx5_translated_shared_action
5802 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5803 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5804 	const struct rte_flow_action *actions;
5805 	struct rte_flow_action *translated_actions = NULL;
5806 	int ret = flow_shared_actions_translate(dev, original_actions,
5807 						shared_actions,
5808 						&shared_actions_n,
5809 						&translated_actions, error);
5810 
5811 	if (ret)
5812 		return ret;
5813 	actions = translated_actions ? translated_actions : original_actions;
5814 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5815 	ret = flow_drv_validate(dev, attr, items, actions,
5816 				true, hairpin_flow, error);
5817 	rte_free(translated_actions);
5818 	return ret;
5819 }
5820 
5821 /**
5822  * Create a flow.
5823  *
5824  * @see rte_flow_create()
5825  * @see rte_flow_ops
5826  */
5827 struct rte_flow *
5828 mlx5_flow_create(struct rte_eth_dev *dev,
5829 		 const struct rte_flow_attr *attr,
5830 		 const struct rte_flow_item items[],
5831 		 const struct rte_flow_action actions[],
5832 		 struct rte_flow_error *error)
5833 {
5834 	struct mlx5_priv *priv = dev->data->dev_private;
5835 
5836 	/*
5837 	 * If the device is not started yet, it is not allowed to created a
5838 	 * flow from application. PMD default flows and traffic control flows
5839 	 * are not affected.
5840 	 */
5841 	if (unlikely(!dev->data->dev_started)) {
5842 		DRV_LOG(DEBUG, "port %u is not started when "
5843 			"inserting a flow", dev->data->port_id);
5844 		rte_flow_error_set(error, ENODEV,
5845 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5846 				   NULL,
5847 				   "port not started");
5848 		return NULL;
5849 	}
5850 
5851 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5852 				  attr, items, actions, true, error);
5853 }
5854 
5855 /**
5856  * Destroy a flow in a list.
5857  *
5858  * @param dev
5859  *   Pointer to Ethernet device.
5860  * @param list
5861  *   Pointer to the Indexed flow list. If this parameter NULL,
5862  *   there is no flow removal from the list. Be noted that as
5863  *   flow is add to the indexed list, memory of the indexed
5864  *   list points to maybe changed as flow destroyed.
5865  * @param[in] flow_idx
5866  *   Index of flow to destroy.
5867  */
5868 static void
5869 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5870 		  uint32_t flow_idx)
5871 {
5872 	struct mlx5_priv *priv = dev->data->dev_private;
5873 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5874 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5875 
5876 	if (!flow)
5877 		return;
5878 	/*
5879 	 * Update RX queue flags only if port is started, otherwise it is
5880 	 * already clean.
5881 	 */
5882 	if (dev->data->dev_started)
5883 		flow_rxq_flags_trim(dev, flow);
5884 	flow_drv_destroy(dev, flow);
5885 	if (list) {
5886 		rte_spinlock_lock(&priv->flow_list_lock);
5887 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5888 			     flow_idx, flow, next);
5889 		rte_spinlock_unlock(&priv->flow_list_lock);
5890 	}
5891 	if (flow->tunnel) {
5892 		struct mlx5_flow_tunnel *tunnel;
5893 
5894 		rte_spinlock_lock(&mlx5_tunnel_hub(dev)->sl);
5895 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5896 		RTE_VERIFY(tunnel);
5897 		LIST_REMOVE(tunnel, chain);
5898 		rte_spinlock_unlock(&mlx5_tunnel_hub(dev)->sl);
5899 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5900 			mlx5_flow_tunnel_free(dev, tunnel);
5901 	}
5902 	flow_mreg_del_copy_action(dev, flow);
5903 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5904 }
5905 
5906 /**
5907  * Destroy all flows.
5908  *
5909  * @param dev
5910  *   Pointer to Ethernet device.
5911  * @param list
5912  *   Pointer to the Indexed flow list.
5913  * @param active
5914  *   If flushing is called avtively.
5915  */
5916 void
5917 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5918 {
5919 	uint32_t num_flushed = 0;
5920 
5921 	while (*list) {
5922 		flow_list_destroy(dev, list, *list);
5923 		num_flushed++;
5924 	}
5925 	if (active) {
5926 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5927 			dev->data->port_id, num_flushed);
5928 	}
5929 }
5930 
5931 /**
5932  * Stop all default actions for flows.
5933  *
5934  * @param dev
5935  *   Pointer to Ethernet device.
5936  */
5937 void
5938 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5939 {
5940 	flow_mreg_del_default_copy_action(dev);
5941 	flow_rxq_flags_clear(dev);
5942 }
5943 
5944 /**
5945  * Start all default actions for flows.
5946  *
5947  * @param dev
5948  *   Pointer to Ethernet device.
5949  * @return
5950  *   0 on success, a negative errno value otherwise and rte_errno is set.
5951  */
5952 int
5953 mlx5_flow_start_default(struct rte_eth_dev *dev)
5954 {
5955 	struct rte_flow_error error;
5956 
5957 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5958 	return flow_mreg_add_default_copy_action(dev, &error);
5959 }
5960 
5961 /**
5962  * Release key of thread specific flow workspace data.
5963  */
5964 static void
5965 flow_release_workspace(void *data)
5966 {
5967 	struct mlx5_flow_workspace *wks = data;
5968 
5969 	if (!wks)
5970 		return;
5971 	free(wks->rss_desc[0].queue);
5972 	free(wks->rss_desc[1].queue);
5973 	free(wks);
5974 }
5975 
5976 /**
5977  * Initialize key of thread specific flow workspace data.
5978  */
5979 static void
5980 flow_alloc_workspace(void)
5981 {
5982 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5983 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5984 }
5985 
5986 /**
5987  * Get thread specific flow workspace.
5988  *
5989  * @return pointer to thread specific flowworkspace data, NULL on error.
5990  */
5991 struct mlx5_flow_workspace*
5992 mlx5_flow_get_thread_workspace(void)
5993 {
5994 	struct mlx5_flow_workspace *data;
5995 
5996 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
5997 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
5998 		return NULL;
5999 	}
6000 	data = pthread_getspecific(key_workspace);
6001 	if (!data) {
6002 		data = calloc(1, sizeof(*data));
6003 		if (!data) {
6004 			DRV_LOG(ERR, "Failed to allocate flow workspace "
6005 				"memory.");
6006 			return NULL;
6007 		}
6008 		data->rss_desc[0].queue = calloc(1,
6009 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6010 		if (!data->rss_desc[0].queue)
6011 			goto err;
6012 		data->rss_desc[1].queue = calloc(1,
6013 				sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6014 		if (!data->rss_desc[1].queue)
6015 			goto err;
6016 		data->rssq_num[0] = MLX5_RSSQ_DEFAULT_NUM;
6017 		data->rssq_num[1] = MLX5_RSSQ_DEFAULT_NUM;
6018 		if (pthread_setspecific(key_workspace, data)) {
6019 			DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6020 			goto err;
6021 		}
6022 	}
6023 	return data;
6024 err:
6025 	if (data->rss_desc[0].queue)
6026 		free(data->rss_desc[0].queue);
6027 	if (data->rss_desc[1].queue)
6028 		free(data->rss_desc[1].queue);
6029 	free(data);
6030 	return NULL;
6031 }
6032 
6033 /**
6034  * Verify the flow list is empty
6035  *
6036  * @param dev
6037  *  Pointer to Ethernet device.
6038  *
6039  * @return the number of flows not released.
6040  */
6041 int
6042 mlx5_flow_verify(struct rte_eth_dev *dev)
6043 {
6044 	struct mlx5_priv *priv = dev->data->dev_private;
6045 	struct rte_flow *flow;
6046 	uint32_t idx;
6047 	int ret = 0;
6048 
6049 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6050 		      flow, next) {
6051 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6052 			dev->data->port_id, (void *)flow);
6053 		++ret;
6054 	}
6055 	return ret;
6056 }
6057 
6058 /**
6059  * Enable default hairpin egress flow.
6060  *
6061  * @param dev
6062  *   Pointer to Ethernet device.
6063  * @param queue
6064  *   The queue index.
6065  *
6066  * @return
6067  *   0 on success, a negative errno value otherwise and rte_errno is set.
6068  */
6069 int
6070 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6071 			    uint32_t queue)
6072 {
6073 	struct mlx5_priv *priv = dev->data->dev_private;
6074 	const struct rte_flow_attr attr = {
6075 		.egress = 1,
6076 		.priority = 0,
6077 	};
6078 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6079 		.queue = queue,
6080 	};
6081 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6082 		.queue = UINT32_MAX,
6083 	};
6084 	struct rte_flow_item items[] = {
6085 		{
6086 			.type = (enum rte_flow_item_type)
6087 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6088 			.spec = &queue_spec,
6089 			.last = NULL,
6090 			.mask = &queue_mask,
6091 		},
6092 		{
6093 			.type = RTE_FLOW_ITEM_TYPE_END,
6094 		},
6095 	};
6096 	struct rte_flow_action_jump jump = {
6097 		.group = MLX5_HAIRPIN_TX_TABLE,
6098 	};
6099 	struct rte_flow_action actions[2];
6100 	uint32_t flow_idx;
6101 	struct rte_flow_error error;
6102 
6103 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6104 	actions[0].conf = &jump;
6105 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6106 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6107 				&attr, items, actions, false, &error);
6108 	if (!flow_idx) {
6109 		DRV_LOG(DEBUG,
6110 			"Failed to create ctrl flow: rte_errno(%d),"
6111 			" type(%d), message(%s)",
6112 			rte_errno, error.type,
6113 			error.message ? error.message : " (no stated reason)");
6114 		return -rte_errno;
6115 	}
6116 	return 0;
6117 }
6118 
6119 /**
6120  * Enable a control flow configured from the control plane.
6121  *
6122  * @param dev
6123  *   Pointer to Ethernet device.
6124  * @param eth_spec
6125  *   An Ethernet flow spec to apply.
6126  * @param eth_mask
6127  *   An Ethernet flow mask to apply.
6128  * @param vlan_spec
6129  *   A VLAN flow spec to apply.
6130  * @param vlan_mask
6131  *   A VLAN flow mask to apply.
6132  *
6133  * @return
6134  *   0 on success, a negative errno value otherwise and rte_errno is set.
6135  */
6136 int
6137 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6138 		    struct rte_flow_item_eth *eth_spec,
6139 		    struct rte_flow_item_eth *eth_mask,
6140 		    struct rte_flow_item_vlan *vlan_spec,
6141 		    struct rte_flow_item_vlan *vlan_mask)
6142 {
6143 	struct mlx5_priv *priv = dev->data->dev_private;
6144 	const struct rte_flow_attr attr = {
6145 		.ingress = 1,
6146 		.priority = MLX5_FLOW_PRIO_RSVD,
6147 	};
6148 	struct rte_flow_item items[] = {
6149 		{
6150 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6151 			.spec = eth_spec,
6152 			.last = NULL,
6153 			.mask = eth_mask,
6154 		},
6155 		{
6156 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6157 					      RTE_FLOW_ITEM_TYPE_END,
6158 			.spec = vlan_spec,
6159 			.last = NULL,
6160 			.mask = vlan_mask,
6161 		},
6162 		{
6163 			.type = RTE_FLOW_ITEM_TYPE_END,
6164 		},
6165 	};
6166 	uint16_t queue[priv->reta_idx_n];
6167 	struct rte_flow_action_rss action_rss = {
6168 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6169 		.level = 0,
6170 		.types = priv->rss_conf.rss_hf,
6171 		.key_len = priv->rss_conf.rss_key_len,
6172 		.queue_num = priv->reta_idx_n,
6173 		.key = priv->rss_conf.rss_key,
6174 		.queue = queue,
6175 	};
6176 	struct rte_flow_action actions[] = {
6177 		{
6178 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6179 			.conf = &action_rss,
6180 		},
6181 		{
6182 			.type = RTE_FLOW_ACTION_TYPE_END,
6183 		},
6184 	};
6185 	uint32_t flow_idx;
6186 	struct rte_flow_error error;
6187 	unsigned int i;
6188 
6189 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6190 		return 0;
6191 	}
6192 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6193 		action_rss.types = 0;
6194 	for (i = 0; i != priv->reta_idx_n; ++i)
6195 		queue[i] = (*priv->reta_idx)[i];
6196 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6197 				&attr, items, actions, false, &error);
6198 	if (!flow_idx)
6199 		return -rte_errno;
6200 	return 0;
6201 }
6202 
6203 /**
6204  * Enable a flow control configured from the control plane.
6205  *
6206  * @param dev
6207  *   Pointer to Ethernet device.
6208  * @param eth_spec
6209  *   An Ethernet flow spec to apply.
6210  * @param eth_mask
6211  *   An Ethernet flow mask to apply.
6212  *
6213  * @return
6214  *   0 on success, a negative errno value otherwise and rte_errno is set.
6215  */
6216 int
6217 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6218 	       struct rte_flow_item_eth *eth_spec,
6219 	       struct rte_flow_item_eth *eth_mask)
6220 {
6221 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6222 }
6223 
6224 /**
6225  * Create default miss flow rule matching lacp traffic
6226  *
6227  * @param dev
6228  *   Pointer to Ethernet device.
6229  * @param eth_spec
6230  *   An Ethernet flow spec to apply.
6231  *
6232  * @return
6233  *   0 on success, a negative errno value otherwise and rte_errno is set.
6234  */
6235 int
6236 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6237 {
6238 	struct mlx5_priv *priv = dev->data->dev_private;
6239 	/*
6240 	 * The LACP matching is done by only using ether type since using
6241 	 * a multicast dst mac causes kernel to give low priority to this flow.
6242 	 */
6243 	static const struct rte_flow_item_eth lacp_spec = {
6244 		.type = RTE_BE16(0x8809),
6245 	};
6246 	static const struct rte_flow_item_eth lacp_mask = {
6247 		.type = 0xffff,
6248 	};
6249 	const struct rte_flow_attr attr = {
6250 		.ingress = 1,
6251 	};
6252 	struct rte_flow_item items[] = {
6253 		{
6254 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6255 			.spec = &lacp_spec,
6256 			.mask = &lacp_mask,
6257 		},
6258 		{
6259 			.type = RTE_FLOW_ITEM_TYPE_END,
6260 		},
6261 	};
6262 	struct rte_flow_action actions[] = {
6263 		{
6264 			.type = (enum rte_flow_action_type)
6265 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6266 		},
6267 		{
6268 			.type = RTE_FLOW_ACTION_TYPE_END,
6269 		},
6270 	};
6271 	struct rte_flow_error error;
6272 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6273 				&attr, items, actions, false, &error);
6274 
6275 	if (!flow_idx)
6276 		return -rte_errno;
6277 	return 0;
6278 }
6279 
6280 /**
6281  * Destroy a flow.
6282  *
6283  * @see rte_flow_destroy()
6284  * @see rte_flow_ops
6285  */
6286 int
6287 mlx5_flow_destroy(struct rte_eth_dev *dev,
6288 		  struct rte_flow *flow,
6289 		  struct rte_flow_error *error __rte_unused)
6290 {
6291 	struct mlx5_priv *priv = dev->data->dev_private;
6292 
6293 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6294 	return 0;
6295 }
6296 
6297 /**
6298  * Destroy all flows.
6299  *
6300  * @see rte_flow_flush()
6301  * @see rte_flow_ops
6302  */
6303 int
6304 mlx5_flow_flush(struct rte_eth_dev *dev,
6305 		struct rte_flow_error *error __rte_unused)
6306 {
6307 	struct mlx5_priv *priv = dev->data->dev_private;
6308 
6309 	mlx5_flow_list_flush(dev, &priv->flows, false);
6310 	return 0;
6311 }
6312 
6313 /**
6314  * Isolated mode.
6315  *
6316  * @see rte_flow_isolate()
6317  * @see rte_flow_ops
6318  */
6319 int
6320 mlx5_flow_isolate(struct rte_eth_dev *dev,
6321 		  int enable,
6322 		  struct rte_flow_error *error)
6323 {
6324 	struct mlx5_priv *priv = dev->data->dev_private;
6325 
6326 	if (dev->data->dev_started) {
6327 		rte_flow_error_set(error, EBUSY,
6328 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6329 				   NULL,
6330 				   "port must be stopped first");
6331 		return -rte_errno;
6332 	}
6333 	priv->isolated = !!enable;
6334 	if (enable)
6335 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6336 	else
6337 		dev->dev_ops = &mlx5_os_dev_ops;
6338 
6339 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6340 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6341 
6342 	return 0;
6343 }
6344 
6345 /**
6346  * Query a flow.
6347  *
6348  * @see rte_flow_query()
6349  * @see rte_flow_ops
6350  */
6351 static int
6352 flow_drv_query(struct rte_eth_dev *dev,
6353 	       uint32_t flow_idx,
6354 	       const struct rte_flow_action *actions,
6355 	       void *data,
6356 	       struct rte_flow_error *error)
6357 {
6358 	struct mlx5_priv *priv = dev->data->dev_private;
6359 	const struct mlx5_flow_driver_ops *fops;
6360 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6361 					       [MLX5_IPOOL_RTE_FLOW],
6362 					       flow_idx);
6363 	enum mlx5_flow_drv_type ftype;
6364 
6365 	if (!flow) {
6366 		return rte_flow_error_set(error, ENOENT,
6367 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6368 			  NULL,
6369 			  "invalid flow handle");
6370 	}
6371 	ftype = flow->drv_type;
6372 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6373 	fops = flow_get_drv_ops(ftype);
6374 
6375 	return fops->query(dev, flow, actions, data, error);
6376 }
6377 
6378 /**
6379  * Query a flow.
6380  *
6381  * @see rte_flow_query()
6382  * @see rte_flow_ops
6383  */
6384 int
6385 mlx5_flow_query(struct rte_eth_dev *dev,
6386 		struct rte_flow *flow,
6387 		const struct rte_flow_action *actions,
6388 		void *data,
6389 		struct rte_flow_error *error)
6390 {
6391 	int ret;
6392 
6393 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6394 			     error);
6395 	if (ret < 0)
6396 		return ret;
6397 	return 0;
6398 }
6399 
6400 /**
6401  * Manage filter operations.
6402  *
6403  * @param dev
6404  *   Pointer to Ethernet device structure.
6405  * @param filter_type
6406  *   Filter type.
6407  * @param filter_op
6408  *   Operation to perform.
6409  * @param arg
6410  *   Pointer to operation-specific structure.
6411  *
6412  * @return
6413  *   0 on success, a negative errno value otherwise and rte_errno is set.
6414  */
6415 int
6416 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6417 		     enum rte_filter_type filter_type,
6418 		     enum rte_filter_op filter_op,
6419 		     void *arg)
6420 {
6421 	switch (filter_type) {
6422 	case RTE_ETH_FILTER_GENERIC:
6423 		if (filter_op != RTE_ETH_FILTER_GET) {
6424 			rte_errno = EINVAL;
6425 			return -rte_errno;
6426 		}
6427 		*(const void **)arg = &mlx5_flow_ops;
6428 		return 0;
6429 	default:
6430 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6431 			dev->data->port_id, filter_type);
6432 		rte_errno = ENOTSUP;
6433 		return -rte_errno;
6434 	}
6435 	return 0;
6436 }
6437 
6438 /**
6439  * Create the needed meter and suffix tables.
6440  *
6441  * @param[in] dev
6442  *   Pointer to Ethernet device.
6443  * @param[in] fm
6444  *   Pointer to the flow meter.
6445  *
6446  * @return
6447  *   Pointer to table set on success, NULL otherwise.
6448  */
6449 struct mlx5_meter_domains_infos *
6450 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6451 			  const struct mlx5_flow_meter *fm)
6452 {
6453 	const struct mlx5_flow_driver_ops *fops;
6454 
6455 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6456 	return fops->create_mtr_tbls(dev, fm);
6457 }
6458 
6459 /**
6460  * Destroy the meter table set.
6461  *
6462  * @param[in] dev
6463  *   Pointer to Ethernet device.
6464  * @param[in] tbl
6465  *   Pointer to the meter table set.
6466  *
6467  * @return
6468  *   0 on success.
6469  */
6470 int
6471 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6472 			   struct mlx5_meter_domains_infos *tbls)
6473 {
6474 	const struct mlx5_flow_driver_ops *fops;
6475 
6476 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6477 	return fops->destroy_mtr_tbls(dev, tbls);
6478 }
6479 
6480 /**
6481  * Create policer rules.
6482  *
6483  * @param[in] dev
6484  *   Pointer to Ethernet device.
6485  * @param[in] fm
6486  *   Pointer to flow meter structure.
6487  * @param[in] attr
6488  *   Pointer to flow attributes.
6489  *
6490  * @return
6491  *   0 on success, -1 otherwise.
6492  */
6493 int
6494 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6495 			       struct mlx5_flow_meter *fm,
6496 			       const struct rte_flow_attr *attr)
6497 {
6498 	const struct mlx5_flow_driver_ops *fops;
6499 
6500 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6501 	return fops->create_policer_rules(dev, fm, attr);
6502 }
6503 
6504 /**
6505  * Destroy policer rules.
6506  *
6507  * @param[in] fm
6508  *   Pointer to flow meter structure.
6509  * @param[in] attr
6510  *   Pointer to flow attributes.
6511  *
6512  * @return
6513  *   0 on success, -1 otherwise.
6514  */
6515 int
6516 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6517 				struct mlx5_flow_meter *fm,
6518 				const struct rte_flow_attr *attr)
6519 {
6520 	const struct mlx5_flow_driver_ops *fops;
6521 
6522 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6523 	return fops->destroy_policer_rules(dev, fm, attr);
6524 }
6525 
6526 /**
6527  * Allocate a counter.
6528  *
6529  * @param[in] dev
6530  *   Pointer to Ethernet device structure.
6531  *
6532  * @return
6533  *   Index to allocated counter  on success, 0 otherwise.
6534  */
6535 uint32_t
6536 mlx5_counter_alloc(struct rte_eth_dev *dev)
6537 {
6538 	const struct mlx5_flow_driver_ops *fops;
6539 	struct rte_flow_attr attr = { .transfer = 0 };
6540 
6541 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6542 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6543 		return fops->counter_alloc(dev);
6544 	}
6545 	DRV_LOG(ERR,
6546 		"port %u counter allocate is not supported.",
6547 		 dev->data->port_id);
6548 	return 0;
6549 }
6550 
6551 /**
6552  * Free a counter.
6553  *
6554  * @param[in] dev
6555  *   Pointer to Ethernet device structure.
6556  * @param[in] cnt
6557  *   Index to counter to be free.
6558  */
6559 void
6560 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6561 {
6562 	const struct mlx5_flow_driver_ops *fops;
6563 	struct rte_flow_attr attr = { .transfer = 0 };
6564 
6565 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6566 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6567 		fops->counter_free(dev, cnt);
6568 		return;
6569 	}
6570 	DRV_LOG(ERR,
6571 		"port %u counter free is not supported.",
6572 		 dev->data->port_id);
6573 }
6574 
6575 /**
6576  * Query counter statistics.
6577  *
6578  * @param[in] dev
6579  *   Pointer to Ethernet device structure.
6580  * @param[in] cnt
6581  *   Index to counter to query.
6582  * @param[in] clear
6583  *   Set to clear counter statistics.
6584  * @param[out] pkts
6585  *   The counter hits packets number to save.
6586  * @param[out] bytes
6587  *   The counter hits bytes number to save.
6588  *
6589  * @return
6590  *   0 on success, a negative errno value otherwise.
6591  */
6592 int
6593 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6594 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6595 {
6596 	const struct mlx5_flow_driver_ops *fops;
6597 	struct rte_flow_attr attr = { .transfer = 0 };
6598 
6599 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6600 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6601 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6602 	}
6603 	DRV_LOG(ERR,
6604 		"port %u counter query is not supported.",
6605 		 dev->data->port_id);
6606 	return -ENOTSUP;
6607 }
6608 
6609 /**
6610  * Allocate a new memory for the counter values wrapped by all the needed
6611  * management.
6612  *
6613  * @param[in] sh
6614  *   Pointer to mlx5_dev_ctx_shared object.
6615  *
6616  * @return
6617  *   0 on success, a negative errno value otherwise.
6618  */
6619 static int
6620 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6621 {
6622 	struct mlx5_devx_mkey_attr mkey_attr;
6623 	struct mlx5_counter_stats_mem_mng *mem_mng;
6624 	volatile struct flow_counter_stats *raw_data;
6625 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6626 	int size = (sizeof(struct flow_counter_stats) *
6627 			MLX5_COUNTERS_PER_POOL +
6628 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6629 			sizeof(struct mlx5_counter_stats_mem_mng);
6630 	size_t pgsize = rte_mem_page_size();
6631 	uint8_t *mem;
6632 	int i;
6633 
6634 	if (pgsize == (size_t)-1) {
6635 		DRV_LOG(ERR, "Failed to get mem page size");
6636 		rte_errno = ENOMEM;
6637 		return -ENOMEM;
6638 	}
6639 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6640 	if (!mem) {
6641 		rte_errno = ENOMEM;
6642 		return -ENOMEM;
6643 	}
6644 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6645 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6646 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
6647 						 IBV_ACCESS_LOCAL_WRITE);
6648 	if (!mem_mng->umem) {
6649 		rte_errno = errno;
6650 		mlx5_free(mem);
6651 		return -rte_errno;
6652 	}
6653 	mkey_attr.addr = (uintptr_t)mem;
6654 	mkey_attr.size = size;
6655 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6656 	mkey_attr.pd = sh->pdn;
6657 	mkey_attr.log_entity_size = 0;
6658 	mkey_attr.pg_access = 0;
6659 	mkey_attr.klm_array = NULL;
6660 	mkey_attr.klm_num = 0;
6661 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6662 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6663 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6664 	if (!mem_mng->dm) {
6665 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
6666 		rte_errno = errno;
6667 		mlx5_free(mem);
6668 		return -rte_errno;
6669 	}
6670 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6671 	raw_data = (volatile struct flow_counter_stats *)mem;
6672 	for (i = 0; i < raws_n; ++i) {
6673 		mem_mng->raws[i].mem_mng = mem_mng;
6674 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6675 	}
6676 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6677 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6678 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6679 				 next);
6680 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6681 	sh->cmng.mem_mng = mem_mng;
6682 	return 0;
6683 }
6684 
6685 /**
6686  * Set the statistic memory to the new counter pool.
6687  *
6688  * @param[in] sh
6689  *   Pointer to mlx5_dev_ctx_shared object.
6690  * @param[in] pool
6691  *   Pointer to the pool to set the statistic memory.
6692  *
6693  * @return
6694  *   0 on success, a negative errno value otherwise.
6695  */
6696 static int
6697 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6698 			       struct mlx5_flow_counter_pool *pool)
6699 {
6700 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6701 	/* Resize statistic memory once used out. */
6702 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6703 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6704 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6705 		return -1;
6706 	}
6707 	rte_spinlock_lock(&pool->sl);
6708 	pool->raw = cmng->mem_mng->raws + pool->index %
6709 		    MLX5_CNT_CONTAINER_RESIZE;
6710 	rte_spinlock_unlock(&pool->sl);
6711 	pool->raw_hw = NULL;
6712 	return 0;
6713 }
6714 
6715 #define MLX5_POOL_QUERY_FREQ_US 1000000
6716 
6717 /**
6718  * Set the periodic procedure for triggering asynchronous batch queries for all
6719  * the counter pools.
6720  *
6721  * @param[in] sh
6722  *   Pointer to mlx5_dev_ctx_shared object.
6723  */
6724 void
6725 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6726 {
6727 	uint32_t pools_n, us;
6728 
6729 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6730 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6731 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6732 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6733 		sh->cmng.query_thread_on = 0;
6734 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6735 	} else {
6736 		sh->cmng.query_thread_on = 1;
6737 	}
6738 }
6739 
6740 /**
6741  * The periodic procedure for triggering asynchronous batch queries for all the
6742  * counter pools. This function is probably called by the host thread.
6743  *
6744  * @param[in] arg
6745  *   The parameter for the alarm process.
6746  */
6747 void
6748 mlx5_flow_query_alarm(void *arg)
6749 {
6750 	struct mlx5_dev_ctx_shared *sh = arg;
6751 	int ret;
6752 	uint16_t pool_index = sh->cmng.pool_index;
6753 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6754 	struct mlx5_flow_counter_pool *pool;
6755 	uint16_t n_valid;
6756 
6757 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6758 		goto set_alarm;
6759 	rte_spinlock_lock(&cmng->pool_update_sl);
6760 	pool = cmng->pools[pool_index];
6761 	n_valid = cmng->n_valid;
6762 	rte_spinlock_unlock(&cmng->pool_update_sl);
6763 	/* Set the statistic memory to the new created pool. */
6764 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6765 		goto set_alarm;
6766 	if (pool->raw_hw)
6767 		/* There is a pool query in progress. */
6768 		goto set_alarm;
6769 	pool->raw_hw =
6770 		LIST_FIRST(&sh->cmng.free_stat_raws);
6771 	if (!pool->raw_hw)
6772 		/* No free counter statistics raw memory. */
6773 		goto set_alarm;
6774 	/*
6775 	 * Identify the counters released between query trigger and query
6776 	 * handle more efficiently. The counter released in this gap period
6777 	 * should wait for a new round of query as the new arrived packets
6778 	 * will not be taken into account.
6779 	 */
6780 	pool->query_gen++;
6781 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6782 					       MLX5_COUNTERS_PER_POOL,
6783 					       NULL, NULL,
6784 					       pool->raw_hw->mem_mng->dm->id,
6785 					       (void *)(uintptr_t)
6786 					       pool->raw_hw->data,
6787 					       sh->devx_comp,
6788 					       (uint64_t)(uintptr_t)pool);
6789 	if (ret) {
6790 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6791 			" %d", pool->min_dcs->id);
6792 		pool->raw_hw = NULL;
6793 		goto set_alarm;
6794 	}
6795 	LIST_REMOVE(pool->raw_hw, next);
6796 	sh->cmng.pending_queries++;
6797 	pool_index++;
6798 	if (pool_index >= n_valid)
6799 		pool_index = 0;
6800 set_alarm:
6801 	sh->cmng.pool_index = pool_index;
6802 	mlx5_set_query_alarm(sh);
6803 }
6804 
6805 /**
6806  * Check and callback event for new aged flow in the counter pool
6807  *
6808  * @param[in] sh
6809  *   Pointer to mlx5_dev_ctx_shared object.
6810  * @param[in] pool
6811  *   Pointer to Current counter pool.
6812  */
6813 static void
6814 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6815 		   struct mlx5_flow_counter_pool *pool)
6816 {
6817 	struct mlx5_priv *priv;
6818 	struct mlx5_flow_counter *cnt;
6819 	struct mlx5_age_info *age_info;
6820 	struct mlx5_age_param *age_param;
6821 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6822 	struct mlx5_counter_stats_raw *prev = pool->raw;
6823 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6824 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6825 	uint16_t expected = AGE_CANDIDATE;
6826 	uint32_t i;
6827 
6828 	pool->time_of_last_age_check = curr_time;
6829 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6830 		cnt = MLX5_POOL_GET_CNT(pool, i);
6831 		age_param = MLX5_CNT_TO_AGE(cnt);
6832 		if (__atomic_load_n(&age_param->state,
6833 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6834 			continue;
6835 		if (cur->data[i].hits != prev->data[i].hits) {
6836 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6837 					 __ATOMIC_RELAXED);
6838 			continue;
6839 		}
6840 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6841 				       time_delta,
6842 				       __ATOMIC_RELAXED) <= age_param->timeout)
6843 			continue;
6844 		/**
6845 		 * Hold the lock first, or if between the
6846 		 * state AGE_TMOUT and tailq operation the
6847 		 * release happened, the release procedure
6848 		 * may delete a non-existent tailq node.
6849 		 */
6850 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6851 		age_info = GET_PORT_AGE_INFO(priv);
6852 		rte_spinlock_lock(&age_info->aged_sl);
6853 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6854 						AGE_TMOUT, false,
6855 						__ATOMIC_RELAXED,
6856 						__ATOMIC_RELAXED)) {
6857 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6858 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6859 		}
6860 		rte_spinlock_unlock(&age_info->aged_sl);
6861 	}
6862 	mlx5_age_event_prepare(sh);
6863 }
6864 
6865 /**
6866  * Handler for the HW respond about ready values from an asynchronous batch
6867  * query. This function is probably called by the host thread.
6868  *
6869  * @param[in] sh
6870  *   The pointer to the shared device context.
6871  * @param[in] async_id
6872  *   The Devx async ID.
6873  * @param[in] status
6874  *   The status of the completion.
6875  */
6876 void
6877 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6878 				  uint64_t async_id, int status)
6879 {
6880 	struct mlx5_flow_counter_pool *pool =
6881 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6882 	struct mlx5_counter_stats_raw *raw_to_free;
6883 	uint8_t query_gen = pool->query_gen ^ 1;
6884 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6885 	enum mlx5_counter_type cnt_type =
6886 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6887 				MLX5_COUNTER_TYPE_ORIGIN;
6888 
6889 	if (unlikely(status)) {
6890 		raw_to_free = pool->raw_hw;
6891 	} else {
6892 		raw_to_free = pool->raw;
6893 		if (pool->is_aged)
6894 			mlx5_flow_aging_check(sh, pool);
6895 		rte_spinlock_lock(&pool->sl);
6896 		pool->raw = pool->raw_hw;
6897 		rte_spinlock_unlock(&pool->sl);
6898 		/* Be sure the new raw counters data is updated in memory. */
6899 		rte_io_wmb();
6900 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6901 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6902 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6903 				     &pool->counters[query_gen], next);
6904 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6905 		}
6906 	}
6907 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6908 	pool->raw_hw = NULL;
6909 	sh->cmng.pending_queries--;
6910 }
6911 
6912 static const struct mlx5_flow_tbl_data_entry  *
6913 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
6914 {
6915 	struct mlx5_priv *priv = dev->data->dev_private;
6916 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6917 	struct mlx5_hlist_entry *he;
6918 	union tunnel_offload_mark mbits = { .val = mark };
6919 	union mlx5_flow_tbl_key table_key = {
6920 		{
6921 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
6922 			.dummy = 0,
6923 			.domain = !!mbits.transfer,
6924 			.direction = 0,
6925 		}
6926 	};
6927 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
6928 	return he ?
6929 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
6930 }
6931 
6932 static void
6933 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
6934 				   struct mlx5_hlist_entry *entry)
6935 {
6936 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6937 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
6938 
6939 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6940 			tunnel_flow_tbl_to_id(tte->flow_table));
6941 	mlx5_free(tte);
6942 }
6943 
6944 static struct mlx5_hlist_entry *
6945 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list,
6946 				   uint64_t key __rte_unused,
6947 				   void *ctx __rte_unused)
6948 {
6949 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6950 	struct tunnel_tbl_entry *tte;
6951 
6952 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
6953 			  sizeof(*tte), 0,
6954 			  SOCKET_ID_ANY);
6955 	if (!tte)
6956 		goto err;
6957 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6958 			  &tte->flow_table);
6959 	if (tte->flow_table >= MLX5_MAX_TABLES) {
6960 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
6961 			tte->flow_table);
6962 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6963 				tte->flow_table);
6964 		goto err;
6965 	} else if (!tte->flow_table) {
6966 		goto err;
6967 	}
6968 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
6969 	return &tte->hash;
6970 err:
6971 	if (tte)
6972 		mlx5_free(tte);
6973 	return NULL;
6974 }
6975 
6976 static uint32_t
6977 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
6978 				const struct mlx5_flow_tunnel *tunnel,
6979 				uint32_t group, uint32_t *table,
6980 				struct rte_flow_error *error)
6981 {
6982 	struct mlx5_hlist_entry *he;
6983 	struct tunnel_tbl_entry *tte;
6984 	union tunnel_tbl_key key = {
6985 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
6986 		.group = group
6987 	};
6988 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
6989 	struct mlx5_hlist *group_hash;
6990 
6991 	group_hash = tunnel ? tunnel->groups : thub->groups;
6992 	he = mlx5_hlist_register(group_hash, key.val, NULL);
6993 	if (!he)
6994 		return rte_flow_error_set(error, EINVAL,
6995 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6996 					  NULL,
6997 					  "tunnel group index not supported");
6998 	tte = container_of(he, typeof(*tte), hash);
6999 	*table = tte->flow_table;
7000 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7001 		dev->data->port_id, key.tunnel_id, group, *table);
7002 	return 0;
7003 }
7004 
7005 static int
7006 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7007 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
7008 {
7009 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7010 		if (group == UINT32_MAX)
7011 			return rte_flow_error_set
7012 						(error, EINVAL,
7013 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7014 						 NULL,
7015 						 "group index not supported");
7016 		*table = group + 1;
7017 	} else {
7018 		*table = group;
7019 	}
7020 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7021 	return 0;
7022 }
7023 
7024 /**
7025  * Translate the rte_flow group index to HW table value.
7026  *
7027  * If tunnel offload is disabled, all group ids converted to flow table
7028  * id using the standard method.
7029  * If tunnel offload is enabled, group id can be converted using the
7030  * standard or tunnel conversion method. Group conversion method
7031  * selection depends on flags in `grp_info` parameter:
7032  * - Internal (grp_info.external == 0) groups conversion uses the
7033  *   standard method.
7034  * - Group ids in JUMP action converted with the tunnel conversion.
7035  * - Group id in rule attribute conversion depends on a rule type and
7036  *   group id value:
7037  *   ** non zero group attributes converted with the tunnel method
7038  *   ** zero group attribute in non-tunnel rule is converted using the
7039  *      standard method - there's only one root table
7040  *   ** zero group attribute in steer tunnel rule is converted with the
7041  *      standard method - single root table
7042  *   ** zero group attribute in match tunnel rule is a special OvS
7043  *      case: that value is used for portability reasons. That group
7044  *      id is converted with the tunnel conversion method.
7045  *
7046  * @param[in] dev
7047  *   Port device
7048  * @param[in] tunnel
7049  *   PMD tunnel offload object
7050  * @param[in] group
7051  *   rte_flow group index value.
7052  * @param[out] table
7053  *   HW table value.
7054  * @param[in] grp_info
7055  *   flags used for conversion
7056  * @param[out] error
7057  *   Pointer to error structure.
7058  *
7059  * @return
7060  *   0 on success, a negative errno value otherwise and rte_errno is set.
7061  */
7062 int
7063 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7064 			 const struct mlx5_flow_tunnel *tunnel,
7065 			 uint32_t group, uint32_t *table,
7066 			 struct flow_grp_info grp_info,
7067 			 struct rte_flow_error *error)
7068 {
7069 	int ret;
7070 	bool standard_translation;
7071 
7072 	if (grp_info.external && group < MLX5_MAX_TABLES_EXTERNAL)
7073 		group *= MLX5_FLOW_TABLE_FACTOR;
7074 	if (is_tunnel_offload_active(dev)) {
7075 		standard_translation = !grp_info.external ||
7076 					grp_info.std_tbl_fix;
7077 	} else {
7078 		standard_translation = true;
7079 	}
7080 	DRV_LOG(DEBUG,
7081 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7082 		dev->data->port_id, group, grp_info.transfer,
7083 		grp_info.external, grp_info.fdb_def_rule,
7084 		standard_translation ? "STANDARD" : "TUNNEL");
7085 	if (standard_translation)
7086 		ret = flow_group_to_table(dev->data->port_id, group, table,
7087 					  grp_info, error);
7088 	else
7089 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7090 						      table, error);
7091 
7092 	return ret;
7093 }
7094 
7095 /**
7096  * Discover availability of metadata reg_c's.
7097  *
7098  * Iteratively use test flows to check availability.
7099  *
7100  * @param[in] dev
7101  *   Pointer to the Ethernet device structure.
7102  *
7103  * @return
7104  *   0 on success, a negative errno value otherwise and rte_errno is set.
7105  */
7106 int
7107 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7108 {
7109 	struct mlx5_priv *priv = dev->data->dev_private;
7110 	struct mlx5_dev_config *config = &priv->config;
7111 	enum modify_reg idx;
7112 	int n = 0;
7113 
7114 	/* reg_c[0] and reg_c[1] are reserved. */
7115 	config->flow_mreg_c[n++] = REG_C_0;
7116 	config->flow_mreg_c[n++] = REG_C_1;
7117 	/* Discover availability of other reg_c's. */
7118 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7119 		struct rte_flow_attr attr = {
7120 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7121 			.priority = MLX5_FLOW_PRIO_RSVD,
7122 			.ingress = 1,
7123 		};
7124 		struct rte_flow_item items[] = {
7125 			[0] = {
7126 				.type = RTE_FLOW_ITEM_TYPE_END,
7127 			},
7128 		};
7129 		struct rte_flow_action actions[] = {
7130 			[0] = {
7131 				.type = (enum rte_flow_action_type)
7132 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7133 				.conf = &(struct mlx5_flow_action_copy_mreg){
7134 					.src = REG_C_1,
7135 					.dst = idx,
7136 				},
7137 			},
7138 			[1] = {
7139 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7140 				.conf = &(struct rte_flow_action_jump){
7141 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7142 				},
7143 			},
7144 			[2] = {
7145 				.type = RTE_FLOW_ACTION_TYPE_END,
7146 			},
7147 		};
7148 		uint32_t flow_idx;
7149 		struct rte_flow *flow;
7150 		struct rte_flow_error error;
7151 
7152 		if (!config->dv_flow_en)
7153 			break;
7154 		/* Create internal flow, validation skips copy action. */
7155 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7156 					    actions, false, &error);
7157 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7158 				      flow_idx);
7159 		if (!flow)
7160 			continue;
7161 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
7162 			config->flow_mreg_c[n++] = idx;
7163 		flow_list_destroy(dev, NULL, flow_idx);
7164 	}
7165 	for (; n < MLX5_MREG_C_NUM; ++n)
7166 		config->flow_mreg_c[n] = REG_NON;
7167 	return 0;
7168 }
7169 
7170 /**
7171  * Dump flow raw hw data to file
7172  *
7173  * @param[in] dev
7174  *    The pointer to Ethernet device.
7175  * @param[in] file
7176  *   A pointer to a file for output.
7177  * @param[out] error
7178  *   Perform verbose error reporting if not NULL. PMDs initialize this
7179  *   structure in case of error only.
7180  * @return
7181  *   0 on success, a nagative value otherwise.
7182  */
7183 int
7184 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7185 		   FILE *file,
7186 		   struct rte_flow_error *error __rte_unused)
7187 {
7188 	struct mlx5_priv *priv = dev->data->dev_private;
7189 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7190 
7191 	if (!priv->config.dv_flow_en) {
7192 		if (fputs("device dv flow disabled\n", file) <= 0)
7193 			return -errno;
7194 		return -ENOTSUP;
7195 	}
7196 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7197 				       sh->tx_domain, file);
7198 }
7199 
7200 /**
7201  * Get aged-out flows.
7202  *
7203  * @param[in] dev
7204  *   Pointer to the Ethernet device structure.
7205  * @param[in] context
7206  *   The address of an array of pointers to the aged-out flows contexts.
7207  * @param[in] nb_countexts
7208  *   The length of context array pointers.
7209  * @param[out] error
7210  *   Perform verbose error reporting if not NULL. Initialized in case of
7211  *   error only.
7212  *
7213  * @return
7214  *   how many contexts get in success, otherwise negative errno value.
7215  *   if nb_contexts is 0, return the amount of all aged contexts.
7216  *   if nb_contexts is not 0 , return the amount of aged flows reported
7217  *   in the context array.
7218  */
7219 int
7220 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7221 			uint32_t nb_contexts, struct rte_flow_error *error)
7222 {
7223 	const struct mlx5_flow_driver_ops *fops;
7224 	struct rte_flow_attr attr = { .transfer = 0 };
7225 
7226 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7227 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7228 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7229 						    error);
7230 	}
7231 	DRV_LOG(ERR,
7232 		"port %u get aged flows is not supported.",
7233 		 dev->data->port_id);
7234 	return -ENOTSUP;
7235 }
7236 
7237 /* Wrapper for driver action_validate op callback */
7238 static int
7239 flow_drv_action_validate(struct rte_eth_dev *dev,
7240 			 const struct rte_flow_shared_action_conf *conf,
7241 			 const struct rte_flow_action *action,
7242 			 const struct mlx5_flow_driver_ops *fops,
7243 			 struct rte_flow_error *error)
7244 {
7245 	static const char err_msg[] = "shared action validation unsupported";
7246 
7247 	if (!fops->action_validate) {
7248 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7249 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7250 				   NULL, err_msg);
7251 		return -rte_errno;
7252 	}
7253 	return fops->action_validate(dev, conf, action, error);
7254 }
7255 
7256 /**
7257  * Destroys the shared action by handle.
7258  *
7259  * @param dev
7260  *   Pointer to Ethernet device structure.
7261  * @param[in] action
7262  *   Handle for the shared action to be destroyed.
7263  * @param[out] error
7264  *   Perform verbose error reporting if not NULL. PMDs initialize this
7265  *   structure in case of error only.
7266  *
7267  * @return
7268  *   0 on success, a negative errno value otherwise and rte_errno is set.
7269  *
7270  * @note: wrapper for driver action_create op callback.
7271  */
7272 static int
7273 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7274 			   struct rte_flow_shared_action *action,
7275 			   struct rte_flow_error *error)
7276 {
7277 	static const char err_msg[] = "shared action destruction unsupported";
7278 	struct rte_flow_attr attr = { .transfer = 0 };
7279 	const struct mlx5_flow_driver_ops *fops =
7280 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7281 
7282 	if (!fops->action_destroy) {
7283 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7284 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7285 				   NULL, err_msg);
7286 		return -rte_errno;
7287 	}
7288 	return fops->action_destroy(dev, action, error);
7289 }
7290 
7291 /* Wrapper for driver action_destroy op callback */
7292 static int
7293 flow_drv_action_update(struct rte_eth_dev *dev,
7294 		       struct rte_flow_shared_action *action,
7295 		       const void *action_conf,
7296 		       const struct mlx5_flow_driver_ops *fops,
7297 		       struct rte_flow_error *error)
7298 {
7299 	static const char err_msg[] = "shared action update unsupported";
7300 
7301 	if (!fops->action_update) {
7302 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7303 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7304 				   NULL, err_msg);
7305 		return -rte_errno;
7306 	}
7307 	return fops->action_update(dev, action, action_conf, error);
7308 }
7309 
7310 /* Wrapper for driver action_destroy op callback */
7311 static int
7312 flow_drv_action_query(struct rte_eth_dev *dev,
7313 		      const struct rte_flow_shared_action *action,
7314 		      void *data,
7315 		      const struct mlx5_flow_driver_ops *fops,
7316 		      struct rte_flow_error *error)
7317 {
7318 	static const char err_msg[] = "shared action query unsupported";
7319 
7320 	if (!fops->action_query) {
7321 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7322 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7323 				   NULL, err_msg);
7324 		return -rte_errno;
7325 	}
7326 	return fops->action_query(dev, action, data, error);
7327 }
7328 
7329 /**
7330  * Create shared action for reuse in multiple flow rules.
7331  *
7332  * @param dev
7333  *   Pointer to Ethernet device structure.
7334  * @param[in] action
7335  *   Action configuration for shared action creation.
7336  * @param[out] error
7337  *   Perform verbose error reporting if not NULL. PMDs initialize this
7338  *   structure in case of error only.
7339  * @return
7340  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7341  */
7342 static struct rte_flow_shared_action *
7343 mlx5_shared_action_create(struct rte_eth_dev *dev,
7344 			  const struct rte_flow_shared_action_conf *conf,
7345 			  const struct rte_flow_action *action,
7346 			  struct rte_flow_error *error)
7347 {
7348 	static const char err_msg[] = "shared action creation unsupported";
7349 	struct rte_flow_attr attr = { .transfer = 0 };
7350 	const struct mlx5_flow_driver_ops *fops =
7351 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7352 
7353 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7354 		return NULL;
7355 	if (!fops->action_create) {
7356 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7357 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7358 				   NULL, err_msg);
7359 		return NULL;
7360 	}
7361 	return fops->action_create(dev, conf, action, error);
7362 }
7363 
7364 /**
7365  * Updates inplace the shared action configuration pointed by *action* handle
7366  * with the configuration provided as *action* argument.
7367  * The update of the shared action configuration effects all flow rules reusing
7368  * the action via handle.
7369  *
7370  * @param dev
7371  *   Pointer to Ethernet device structure.
7372  * @param[in] shared_action
7373  *   Handle for the shared action to be updated.
7374  * @param[in] action
7375  *   Action specification used to modify the action pointed by handle.
7376  *   *action* should be of same type with the action pointed by the *action*
7377  *   handle argument, otherwise considered as invalid.
7378  * @param[out] error
7379  *   Perform verbose error reporting if not NULL. PMDs initialize this
7380  *   structure in case of error only.
7381  *
7382  * @return
7383  *   0 on success, a negative errno value otherwise and rte_errno is set.
7384  */
7385 static int
7386 mlx5_shared_action_update(struct rte_eth_dev *dev,
7387 		struct rte_flow_shared_action *shared_action,
7388 		const struct rte_flow_action *action,
7389 		struct rte_flow_error *error)
7390 {
7391 	struct rte_flow_attr attr = { .transfer = 0 };
7392 	const struct mlx5_flow_driver_ops *fops =
7393 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7394 	int ret;
7395 
7396 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7397 	if (ret)
7398 		return ret;
7399 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7400 				      error);
7401 }
7402 
7403 /**
7404  * Query the shared action by handle.
7405  *
7406  * This function allows retrieving action-specific data such as counters.
7407  * Data is gathered by special action which may be present/referenced in
7408  * more than one flow rule definition.
7409  *
7410  * \see RTE_FLOW_ACTION_TYPE_COUNT
7411  *
7412  * @param dev
7413  *   Pointer to Ethernet device structure.
7414  * @param[in] action
7415  *   Handle for the shared action to query.
7416  * @param[in, out] data
7417  *   Pointer to storage for the associated query data type.
7418  * @param[out] error
7419  *   Perform verbose error reporting if not NULL. PMDs initialize this
7420  *   structure in case of error only.
7421  *
7422  * @return
7423  *   0 on success, a negative errno value otherwise and rte_errno is set.
7424  */
7425 static int
7426 mlx5_shared_action_query(struct rte_eth_dev *dev,
7427 			 const struct rte_flow_shared_action *action,
7428 			 void *data,
7429 			 struct rte_flow_error *error)
7430 {
7431 	struct rte_flow_attr attr = { .transfer = 0 };
7432 	const struct mlx5_flow_driver_ops *fops =
7433 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7434 
7435 	return flow_drv_action_query(dev, action, data, fops, error);
7436 }
7437 
7438 /**
7439  * Destroy all shared actions.
7440  *
7441  * @param dev
7442  *   Pointer to Ethernet device.
7443  *
7444  * @return
7445  *   0 on success, a negative errno value otherwise and rte_errno is set.
7446  */
7447 int
7448 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7449 {
7450 	struct rte_flow_error error;
7451 	struct mlx5_priv *priv = dev->data->dev_private;
7452 	struct mlx5_shared_action_rss *action;
7453 	int ret = 0;
7454 	uint32_t idx;
7455 
7456 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7457 		      priv->rss_shared_actions, idx, action, next) {
7458 		ret |= mlx5_shared_action_destroy(dev,
7459 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7460 	}
7461 	return ret;
7462 }
7463 
7464 static void
7465 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7466 		      struct mlx5_flow_tunnel *tunnel)
7467 {
7468 	struct mlx5_priv *priv = dev->data->dev_private;
7469 
7470 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7471 		dev->data->port_id, tunnel->tunnel_id);
7472 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
7473 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
7474 			tunnel->tunnel_id);
7475 	mlx5_hlist_destroy(tunnel->groups);
7476 	mlx5_free(tunnel);
7477 }
7478 
7479 static struct mlx5_flow_tunnel *
7480 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7481 {
7482 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7483 	struct mlx5_flow_tunnel *tun;
7484 
7485 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7486 		if (tun->tunnel_id == id)
7487 			break;
7488 	}
7489 
7490 	return tun;
7491 }
7492 
7493 static struct mlx5_flow_tunnel *
7494 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7495 			  const struct rte_flow_tunnel *app_tunnel)
7496 {
7497 	struct mlx5_priv *priv = dev->data->dev_private;
7498 	struct mlx5_flow_tunnel *tunnel;
7499 	uint32_t id;
7500 
7501 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
7502 			  &id);
7503 	if (id >= MLX5_MAX_TUNNELS) {
7504 		mlx5_ipool_free(priv->sh->ipool
7505 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7506 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7507 		return NULL;
7508 	} else if (!id) {
7509 		return NULL;
7510 	}
7511 	/**
7512 	 * mlx5 flow tunnel is an auxlilary data structure
7513 	 * It's not part of IO. No need to allocate it from
7514 	 * huge pages pools dedicated for IO
7515 	 */
7516 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
7517 			     0, SOCKET_ID_ANY);
7518 	if (!tunnel) {
7519 		mlx5_ipool_free(priv->sh->ipool
7520 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7521 		return NULL;
7522 	}
7523 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7524 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7525 					   NULL,
7526 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7527 	if (!tunnel->groups) {
7528 		mlx5_ipool_free(priv->sh->ipool
7529 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7530 		mlx5_free(tunnel);
7531 		return NULL;
7532 	}
7533 	tunnel->groups->ctx = priv->sh;
7534 	/* initiate new PMD tunnel */
7535 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7536 	tunnel->tunnel_id = id;
7537 	tunnel->action.type = (typeof(tunnel->action.type))
7538 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7539 	tunnel->action.conf = tunnel;
7540 	tunnel->item.type = (typeof(tunnel->item.type))
7541 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7542 	tunnel->item.spec = tunnel;
7543 	tunnel->item.last = NULL;
7544 	tunnel->item.mask = NULL;
7545 
7546 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7547 		dev->data->port_id, tunnel->tunnel_id);
7548 
7549 	return tunnel;
7550 }
7551 
7552 static int
7553 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7554 		     const struct rte_flow_tunnel *app_tunnel,
7555 		     struct mlx5_flow_tunnel **tunnel)
7556 {
7557 	int ret;
7558 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7559 	struct mlx5_flow_tunnel *tun;
7560 
7561 	rte_spinlock_lock(&thub->sl);
7562 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7563 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7564 			    sizeof(*app_tunnel))) {
7565 			*tunnel = tun;
7566 			ret = 0;
7567 			break;
7568 		}
7569 	}
7570 	if (!tun) {
7571 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
7572 		if (tun) {
7573 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
7574 			*tunnel = tun;
7575 		} else {
7576 			ret = -ENOMEM;
7577 		}
7578 	}
7579 	rte_spinlock_unlock(&thub->sl);
7580 	if (tun)
7581 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
7582 
7583 	return ret;
7584 }
7585 
7586 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7587 {
7588 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7589 
7590 	if (!thub)
7591 		return;
7592 	if (!LIST_EMPTY(&thub->tunnels))
7593 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7594 	mlx5_hlist_destroy(thub->groups);
7595 	mlx5_free(thub);
7596 }
7597 
7598 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7599 {
7600 	int err;
7601 	struct mlx5_flow_tunnel_hub *thub;
7602 
7603 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7604 			   0, SOCKET_ID_ANY);
7605 	if (!thub)
7606 		return -ENOMEM;
7607 	LIST_INIT(&thub->tunnels);
7608 	rte_spinlock_init(&thub->sl);
7609 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7610 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7611 					 NULL,
7612 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7613 	if (!thub->groups) {
7614 		err = -rte_errno;
7615 		goto err;
7616 	}
7617 	thub->groups->ctx = sh;
7618 	sh->tunnel_hub = thub;
7619 
7620 	return 0;
7621 
7622 err:
7623 	if (thub->groups)
7624 		mlx5_hlist_destroy(thub->groups);
7625 	if (thub)
7626 		mlx5_free(thub);
7627 	return err;
7628 }
7629 
7630 #ifndef HAVE_MLX5DV_DR
7631 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7632 #else
7633 #define MLX5_DOMAIN_SYNC_FLOW \
7634 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7635 #endif
7636 
7637 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7638 {
7639 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7640 	const struct mlx5_flow_driver_ops *fops;
7641 	int ret;
7642 	struct rte_flow_attr attr = { .transfer = 0 };
7643 
7644 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7645 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7646 	if (ret > 0)
7647 		ret = -ret;
7648 	return ret;
7649 }
7650