xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision c5a3860f36902e1701817b0961bf99a5c96ede3a)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rxtx.h"
32 #include "mlx5_common_os.h"
33 #include "rte_pmd_mlx5.h"
34 
35 struct tunnel_default_miss_ctx {
36 	uint16_t *queue;
37 	__extension__
38 	union {
39 		struct rte_flow_action_rss action_rss;
40 		struct rte_flow_action_queue miss_queue;
41 		struct rte_flow_action_jump miss_jump;
42 		uint8_t raw[0];
43 	};
44 };
45 
46 static int
47 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
48 			     struct rte_flow *flow,
49 			     const struct rte_flow_attr *attr,
50 			     const struct rte_flow_action *app_actions,
51 			     uint32_t flow_idx,
52 			     struct tunnel_default_miss_ctx *ctx,
53 			     struct rte_flow_error *error);
54 static struct mlx5_flow_tunnel *
55 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
56 static void
57 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
58 static uint32_t
59 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
60 				const struct mlx5_flow_tunnel *tunnel,
61 				uint32_t group, uint32_t *table,
62 				struct rte_flow_error *error);
63 
64 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
65 static void mlx5_flow_pop_thread_workspace(void);
66 
67 
68 /** Device flow drivers. */
69 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
70 
71 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
72 
73 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
74 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
75 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
76 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
77 #endif
78 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
79 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
80 };
81 
82 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
83 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
84 	(const int []){ \
85 		__VA_ARGS__, 0, \
86 	}
87 
88 /** Node object of input graph for mlx5_flow_expand_rss(). */
89 struct mlx5_flow_expand_node {
90 	const int *const next;
91 	/**<
92 	 * List of next node indexes. Index 0 is interpreted as a terminator.
93 	 */
94 	const enum rte_flow_item_type type;
95 	/**< Pattern item type of current node. */
96 	uint64_t rss_types;
97 	/**<
98 	 * RSS types bit-field associated with this node
99 	 * (see ETH_RSS_* definitions).
100 	 */
101 };
102 
103 /** Object returned by mlx5_flow_expand_rss(). */
104 struct mlx5_flow_expand_rss {
105 	uint32_t entries;
106 	/**< Number of entries @p patterns and @p priorities. */
107 	struct {
108 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
109 		uint32_t priority; /**< Priority offset for each expansion. */
110 	} entry[];
111 };
112 
113 static enum rte_flow_item_type
114 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
115 {
116 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
117 	uint16_t ether_type = 0;
118 	uint16_t ether_type_m;
119 	uint8_t ip_next_proto = 0;
120 	uint8_t ip_next_proto_m;
121 
122 	if (item == NULL || item->spec == NULL)
123 		return ret;
124 	switch (item->type) {
125 	case RTE_FLOW_ITEM_TYPE_ETH:
126 		if (item->mask)
127 			ether_type_m = ((const struct rte_flow_item_eth *)
128 						(item->mask))->type;
129 		else
130 			ether_type_m = rte_flow_item_eth_mask.type;
131 		if (ether_type_m != RTE_BE16(0xFFFF))
132 			break;
133 		ether_type = ((const struct rte_flow_item_eth *)
134 				(item->spec))->type;
135 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
136 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
137 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
138 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
139 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
140 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
141 		else
142 			ret = RTE_FLOW_ITEM_TYPE_END;
143 		break;
144 	case RTE_FLOW_ITEM_TYPE_VLAN:
145 		if (item->mask)
146 			ether_type_m = ((const struct rte_flow_item_vlan *)
147 						(item->mask))->inner_type;
148 		else
149 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
150 		if (ether_type_m != RTE_BE16(0xFFFF))
151 			break;
152 		ether_type = ((const struct rte_flow_item_vlan *)
153 				(item->spec))->inner_type;
154 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
155 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
156 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
157 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
158 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
159 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
160 		else
161 			ret = RTE_FLOW_ITEM_TYPE_END;
162 		break;
163 	case RTE_FLOW_ITEM_TYPE_IPV4:
164 		if (item->mask)
165 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
166 					(item->mask))->hdr.next_proto_id;
167 		else
168 			ip_next_proto_m =
169 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
170 		if (ip_next_proto_m != 0xFF)
171 			break;
172 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
173 				(item->spec))->hdr.next_proto_id;
174 		if (ip_next_proto == IPPROTO_UDP)
175 			ret = RTE_FLOW_ITEM_TYPE_UDP;
176 		else if (ip_next_proto == IPPROTO_TCP)
177 			ret = RTE_FLOW_ITEM_TYPE_TCP;
178 		else if (ip_next_proto == IPPROTO_IP)
179 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
180 		else if (ip_next_proto == IPPROTO_IPV6)
181 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
182 		else
183 			ret = RTE_FLOW_ITEM_TYPE_END;
184 		break;
185 	case RTE_FLOW_ITEM_TYPE_IPV6:
186 		if (item->mask)
187 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
188 						(item->mask))->hdr.proto;
189 		else
190 			ip_next_proto_m =
191 				rte_flow_item_ipv6_mask.hdr.proto;
192 		if (ip_next_proto_m != 0xFF)
193 			break;
194 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
195 				(item->spec))->hdr.proto;
196 		if (ip_next_proto == IPPROTO_UDP)
197 			ret = RTE_FLOW_ITEM_TYPE_UDP;
198 		else if (ip_next_proto == IPPROTO_TCP)
199 			ret = RTE_FLOW_ITEM_TYPE_TCP;
200 		else if (ip_next_proto == IPPROTO_IP)
201 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
202 		else if (ip_next_proto == IPPROTO_IPV6)
203 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
204 		else
205 			ret = RTE_FLOW_ITEM_TYPE_END;
206 		break;
207 	default:
208 		ret = RTE_FLOW_ITEM_TYPE_VOID;
209 		break;
210 	}
211 	return ret;
212 }
213 
214 #define MLX5_RSS_EXP_ELT_N 8
215 
216 /**
217  * Expand RSS flows into several possible flows according to the RSS hash
218  * fields requested and the driver capabilities.
219  *
220  * @param[out] buf
221  *   Buffer to store the result expansion.
222  * @param[in] size
223  *   Buffer size in bytes. If 0, @p buf can be NULL.
224  * @param[in] pattern
225  *   User flow pattern.
226  * @param[in] types
227  *   RSS types to expand (see ETH_RSS_* definitions).
228  * @param[in] graph
229  *   Input graph to expand @p pattern according to @p types.
230  * @param[in] graph_root_index
231  *   Index of root node in @p graph, typically 0.
232  *
233  * @return
234  *   A positive value representing the size of @p buf in bytes regardless of
235  *   @p size on success, a negative errno value otherwise and rte_errno is
236  *   set, the following errors are defined:
237  *
238  *   -E2BIG: graph-depth @p graph is too deep.
239  */
240 static int
241 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
242 		     const struct rte_flow_item *pattern, uint64_t types,
243 		     const struct mlx5_flow_expand_node graph[],
244 		     int graph_root_index)
245 {
246 	const struct rte_flow_item *item;
247 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
248 	const int *next_node;
249 	const int *stack[MLX5_RSS_EXP_ELT_N];
250 	int stack_pos = 0;
251 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
252 	unsigned int i;
253 	size_t lsize;
254 	size_t user_pattern_size = 0;
255 	void *addr = NULL;
256 	const struct mlx5_flow_expand_node *next = NULL;
257 	struct rte_flow_item missed_item;
258 	int missed = 0;
259 	int elt = 0;
260 	const struct rte_flow_item *last_item = NULL;
261 
262 	memset(&missed_item, 0, sizeof(missed_item));
263 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
264 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
265 	if (lsize <= size) {
266 		buf->entry[0].priority = 0;
267 		buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
268 		buf->entries = 0;
269 		addr = buf->entry[0].pattern;
270 	}
271 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
272 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
273 			last_item = item;
274 		for (i = 0; node->next && node->next[i]; ++i) {
275 			next = &graph[node->next[i]];
276 			if (next->type == item->type)
277 				break;
278 		}
279 		if (next)
280 			node = next;
281 		user_pattern_size += sizeof(*item);
282 	}
283 	user_pattern_size += sizeof(*item); /* Handle END item. */
284 	lsize += user_pattern_size;
285 	/* Copy the user pattern in the first entry of the buffer. */
286 	if (lsize <= size) {
287 		rte_memcpy(addr, pattern, user_pattern_size);
288 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
289 		buf->entries = 1;
290 	}
291 	/* Start expanding. */
292 	memset(flow_items, 0, sizeof(flow_items));
293 	user_pattern_size -= sizeof(*item);
294 	/*
295 	 * Check if the last valid item has spec set, need complete pattern,
296 	 * and the pattern can be used for expansion.
297 	 */
298 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
299 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
300 		/* Item type END indicates expansion is not required. */
301 		return lsize;
302 	}
303 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
304 		next = NULL;
305 		missed = 1;
306 		for (i = 0; node->next && node->next[i]; ++i) {
307 			next = &graph[node->next[i]];
308 			if (next->type == missed_item.type) {
309 				flow_items[0].type = missed_item.type;
310 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
311 				break;
312 			}
313 			next = NULL;
314 		}
315 	}
316 	if (next && missed) {
317 		elt = 2; /* missed item + item end. */
318 		node = next;
319 		lsize += elt * sizeof(*item) + user_pattern_size;
320 		if ((node->rss_types & types) && lsize <= size) {
321 			buf->entry[buf->entries].priority = 1;
322 			buf->entry[buf->entries].pattern = addr;
323 			buf->entries++;
324 			rte_memcpy(addr, buf->entry[0].pattern,
325 				   user_pattern_size);
326 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
327 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
328 			addr = (void *)(((uintptr_t)addr) +
329 					elt * sizeof(*item));
330 		}
331 	}
332 	memset(flow_items, 0, sizeof(flow_items));
333 	next_node = node->next;
334 	stack[stack_pos] = next_node;
335 	node = next_node ? &graph[*next_node] : NULL;
336 	while (node) {
337 		flow_items[stack_pos].type = node->type;
338 		if (node->rss_types & types) {
339 			/*
340 			 * compute the number of items to copy from the
341 			 * expansion and copy it.
342 			 * When the stack_pos is 0, there are 1 element in it,
343 			 * plus the addition END item.
344 			 */
345 			elt = stack_pos + 2;
346 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
347 			lsize += elt * sizeof(*item) + user_pattern_size;
348 			if (lsize <= size) {
349 				size_t n = elt * sizeof(*item);
350 
351 				buf->entry[buf->entries].priority =
352 					stack_pos + 1 + missed;
353 				buf->entry[buf->entries].pattern = addr;
354 				buf->entries++;
355 				rte_memcpy(addr, buf->entry[0].pattern,
356 					   user_pattern_size);
357 				addr = (void *)(((uintptr_t)addr) +
358 						user_pattern_size);
359 				rte_memcpy(addr, &missed_item,
360 					   missed * sizeof(*item));
361 				addr = (void *)(((uintptr_t)addr) +
362 					missed * sizeof(*item));
363 				rte_memcpy(addr, flow_items, n);
364 				addr = (void *)(((uintptr_t)addr) + n);
365 			}
366 		}
367 		/* Go deeper. */
368 		if (node->next) {
369 			next_node = node->next;
370 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
371 				rte_errno = E2BIG;
372 				return -rte_errno;
373 			}
374 			stack[stack_pos] = next_node;
375 		} else if (*(next_node + 1)) {
376 			/* Follow up with the next possibility. */
377 			++next_node;
378 		} else {
379 			/* Move to the next path. */
380 			if (stack_pos)
381 				next_node = stack[--stack_pos];
382 			next_node++;
383 			stack[stack_pos] = next_node;
384 		}
385 		node = *next_node ? &graph[*next_node] : NULL;
386 	};
387 	return lsize;
388 }
389 
390 enum mlx5_expansion {
391 	MLX5_EXPANSION_ROOT,
392 	MLX5_EXPANSION_ROOT_OUTER,
393 	MLX5_EXPANSION_ROOT_ETH_VLAN,
394 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
395 	MLX5_EXPANSION_OUTER_ETH,
396 	MLX5_EXPANSION_OUTER_ETH_VLAN,
397 	MLX5_EXPANSION_OUTER_VLAN,
398 	MLX5_EXPANSION_OUTER_IPV4,
399 	MLX5_EXPANSION_OUTER_IPV4_UDP,
400 	MLX5_EXPANSION_OUTER_IPV4_TCP,
401 	MLX5_EXPANSION_OUTER_IPV6,
402 	MLX5_EXPANSION_OUTER_IPV6_UDP,
403 	MLX5_EXPANSION_OUTER_IPV6_TCP,
404 	MLX5_EXPANSION_VXLAN,
405 	MLX5_EXPANSION_VXLAN_GPE,
406 	MLX5_EXPANSION_GRE,
407 	MLX5_EXPANSION_MPLS,
408 	MLX5_EXPANSION_ETH,
409 	MLX5_EXPANSION_ETH_VLAN,
410 	MLX5_EXPANSION_VLAN,
411 	MLX5_EXPANSION_IPV4,
412 	MLX5_EXPANSION_IPV4_UDP,
413 	MLX5_EXPANSION_IPV4_TCP,
414 	MLX5_EXPANSION_IPV6,
415 	MLX5_EXPANSION_IPV6_UDP,
416 	MLX5_EXPANSION_IPV6_TCP,
417 };
418 
419 /** Supported expansion of items. */
420 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
421 	[MLX5_EXPANSION_ROOT] = {
422 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
423 						  MLX5_EXPANSION_IPV4,
424 						  MLX5_EXPANSION_IPV6),
425 		.type = RTE_FLOW_ITEM_TYPE_END,
426 	},
427 	[MLX5_EXPANSION_ROOT_OUTER] = {
428 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
429 						  MLX5_EXPANSION_OUTER_IPV4,
430 						  MLX5_EXPANSION_OUTER_IPV6),
431 		.type = RTE_FLOW_ITEM_TYPE_END,
432 	},
433 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
434 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
435 		.type = RTE_FLOW_ITEM_TYPE_END,
436 	},
437 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
438 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
439 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
440 		.type = RTE_FLOW_ITEM_TYPE_END,
441 	},
442 	[MLX5_EXPANSION_OUTER_ETH] = {
443 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
444 						  MLX5_EXPANSION_OUTER_IPV6,
445 						  MLX5_EXPANSION_MPLS),
446 		.type = RTE_FLOW_ITEM_TYPE_ETH,
447 		.rss_types = 0,
448 	},
449 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
450 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
451 		.type = RTE_FLOW_ITEM_TYPE_ETH,
452 		.rss_types = 0,
453 	},
454 	[MLX5_EXPANSION_OUTER_VLAN] = {
455 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
456 						  MLX5_EXPANSION_OUTER_IPV6),
457 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
458 	},
459 	[MLX5_EXPANSION_OUTER_IPV4] = {
460 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
461 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
462 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
463 			 MLX5_EXPANSION_GRE,
464 			 MLX5_EXPANSION_IPV4,
465 			 MLX5_EXPANSION_IPV6),
466 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
467 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
468 			ETH_RSS_NONFRAG_IPV4_OTHER,
469 	},
470 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
471 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
472 						  MLX5_EXPANSION_VXLAN_GPE),
473 		.type = RTE_FLOW_ITEM_TYPE_UDP,
474 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
475 	},
476 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
477 		.type = RTE_FLOW_ITEM_TYPE_TCP,
478 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
479 	},
480 	[MLX5_EXPANSION_OUTER_IPV6] = {
481 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
482 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
483 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
484 			 MLX5_EXPANSION_IPV4,
485 			 MLX5_EXPANSION_IPV6,
486 			 MLX5_EXPANSION_GRE),
487 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
488 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
489 			ETH_RSS_NONFRAG_IPV6_OTHER,
490 	},
491 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
492 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
493 						  MLX5_EXPANSION_VXLAN_GPE),
494 		.type = RTE_FLOW_ITEM_TYPE_UDP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
496 	},
497 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
498 		.type = RTE_FLOW_ITEM_TYPE_TCP,
499 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
500 	},
501 	[MLX5_EXPANSION_VXLAN] = {
502 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
503 						  MLX5_EXPANSION_IPV4,
504 						  MLX5_EXPANSION_IPV6),
505 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
506 	},
507 	[MLX5_EXPANSION_VXLAN_GPE] = {
508 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
509 						  MLX5_EXPANSION_IPV4,
510 						  MLX5_EXPANSION_IPV6),
511 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
512 	},
513 	[MLX5_EXPANSION_GRE] = {
514 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
515 						  MLX5_EXPANSION_IPV6),
516 		.type = RTE_FLOW_ITEM_TYPE_GRE,
517 	},
518 	[MLX5_EXPANSION_MPLS] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
522 	},
523 	[MLX5_EXPANSION_ETH] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
525 						  MLX5_EXPANSION_IPV6),
526 		.type = RTE_FLOW_ITEM_TYPE_ETH,
527 	},
528 	[MLX5_EXPANSION_ETH_VLAN] = {
529 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
530 		.type = RTE_FLOW_ITEM_TYPE_ETH,
531 	},
532 	[MLX5_EXPANSION_VLAN] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
534 						  MLX5_EXPANSION_IPV6),
535 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
536 	},
537 	[MLX5_EXPANSION_IPV4] = {
538 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
539 						  MLX5_EXPANSION_IPV4_TCP),
540 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
541 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
542 			ETH_RSS_NONFRAG_IPV4_OTHER,
543 	},
544 	[MLX5_EXPANSION_IPV4_UDP] = {
545 		.type = RTE_FLOW_ITEM_TYPE_UDP,
546 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
547 	},
548 	[MLX5_EXPANSION_IPV4_TCP] = {
549 		.type = RTE_FLOW_ITEM_TYPE_TCP,
550 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
551 	},
552 	[MLX5_EXPANSION_IPV6] = {
553 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
554 						  MLX5_EXPANSION_IPV6_TCP),
555 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
556 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
557 			ETH_RSS_NONFRAG_IPV6_OTHER,
558 	},
559 	[MLX5_EXPANSION_IPV6_UDP] = {
560 		.type = RTE_FLOW_ITEM_TYPE_UDP,
561 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
562 	},
563 	[MLX5_EXPANSION_IPV6_TCP] = {
564 		.type = RTE_FLOW_ITEM_TYPE_TCP,
565 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
566 	},
567 };
568 
569 static struct rte_flow_shared_action *
570 mlx5_shared_action_create(struct rte_eth_dev *dev,
571 			  const struct rte_flow_shared_action_conf *conf,
572 			  const struct rte_flow_action *action,
573 			  struct rte_flow_error *error);
574 static int mlx5_shared_action_destroy
575 				(struct rte_eth_dev *dev,
576 				 struct rte_flow_shared_action *shared_action,
577 				 struct rte_flow_error *error);
578 static int mlx5_shared_action_update
579 				(struct rte_eth_dev *dev,
580 				 struct rte_flow_shared_action *shared_action,
581 				 const struct rte_flow_action *action,
582 				 struct rte_flow_error *error);
583 static int mlx5_shared_action_query
584 				(struct rte_eth_dev *dev,
585 				 const struct rte_flow_shared_action *action,
586 				 void *data,
587 				 struct rte_flow_error *error);
588 static int
589 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
590 		    struct rte_flow_tunnel *app_tunnel,
591 		    struct rte_flow_action **actions,
592 		    uint32_t *num_of_actions,
593 		    struct rte_flow_error *error);
594 static int
595 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
596 		       struct rte_flow_tunnel *app_tunnel,
597 		       struct rte_flow_item **items,
598 		       uint32_t *num_of_items,
599 		       struct rte_flow_error *error);
600 static int
601 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
602 			      struct rte_flow_item *pmd_items,
603 			      uint32_t num_items, struct rte_flow_error *err);
604 static int
605 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
606 				struct rte_flow_action *pmd_actions,
607 				uint32_t num_actions,
608 				struct rte_flow_error *err);
609 static int
610 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
611 				  struct rte_mbuf *m,
612 				  struct rte_flow_restore_info *info,
613 				  struct rte_flow_error *err);
614 
615 static const struct rte_flow_ops mlx5_flow_ops = {
616 	.validate = mlx5_flow_validate,
617 	.create = mlx5_flow_create,
618 	.destroy = mlx5_flow_destroy,
619 	.flush = mlx5_flow_flush,
620 	.isolate = mlx5_flow_isolate,
621 	.query = mlx5_flow_query,
622 	.dev_dump = mlx5_flow_dev_dump,
623 	.get_aged_flows = mlx5_flow_get_aged_flows,
624 	.shared_action_create = mlx5_shared_action_create,
625 	.shared_action_destroy = mlx5_shared_action_destroy,
626 	.shared_action_update = mlx5_shared_action_update,
627 	.shared_action_query = mlx5_shared_action_query,
628 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
629 	.tunnel_match = mlx5_flow_tunnel_match,
630 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
631 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
632 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
633 };
634 
635 /* Tunnel information. */
636 struct mlx5_flow_tunnel_info {
637 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
638 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
639 };
640 
641 static struct mlx5_flow_tunnel_info tunnels_info[] = {
642 	{
643 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
644 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
645 	},
646 	{
647 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
648 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
649 	},
650 	{
651 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
652 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
653 	},
654 	{
655 		.tunnel = MLX5_FLOW_LAYER_GRE,
656 		.ptype = RTE_PTYPE_TUNNEL_GRE,
657 	},
658 	{
659 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
660 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
661 	},
662 	{
663 		.tunnel = MLX5_FLOW_LAYER_MPLS,
664 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
665 	},
666 	{
667 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
668 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
669 	},
670 	{
671 		.tunnel = MLX5_FLOW_LAYER_IPIP,
672 		.ptype = RTE_PTYPE_TUNNEL_IP,
673 	},
674 	{
675 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
676 		.ptype = RTE_PTYPE_TUNNEL_IP,
677 	},
678 	{
679 		.tunnel = MLX5_FLOW_LAYER_GTP,
680 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
681 	},
682 };
683 
684 
685 
686 /**
687  * Translate tag ID to register.
688  *
689  * @param[in] dev
690  *   Pointer to the Ethernet device structure.
691  * @param[in] feature
692  *   The feature that request the register.
693  * @param[in] id
694  *   The request register ID.
695  * @param[out] error
696  *   Error description in case of any.
697  *
698  * @return
699  *   The request register on success, a negative errno
700  *   value otherwise and rte_errno is set.
701  */
702 int
703 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
704 		     enum mlx5_feature_name feature,
705 		     uint32_t id,
706 		     struct rte_flow_error *error)
707 {
708 	struct mlx5_priv *priv = dev->data->dev_private;
709 	struct mlx5_dev_config *config = &priv->config;
710 	enum modify_reg start_reg;
711 	bool skip_mtr_reg = false;
712 
713 	switch (feature) {
714 	case MLX5_HAIRPIN_RX:
715 		return REG_B;
716 	case MLX5_HAIRPIN_TX:
717 		return REG_A;
718 	case MLX5_METADATA_RX:
719 		switch (config->dv_xmeta_en) {
720 		case MLX5_XMETA_MODE_LEGACY:
721 			return REG_B;
722 		case MLX5_XMETA_MODE_META16:
723 			return REG_C_0;
724 		case MLX5_XMETA_MODE_META32:
725 			return REG_C_1;
726 		}
727 		break;
728 	case MLX5_METADATA_TX:
729 		return REG_A;
730 	case MLX5_METADATA_FDB:
731 		switch (config->dv_xmeta_en) {
732 		case MLX5_XMETA_MODE_LEGACY:
733 			return REG_NON;
734 		case MLX5_XMETA_MODE_META16:
735 			return REG_C_0;
736 		case MLX5_XMETA_MODE_META32:
737 			return REG_C_1;
738 		}
739 		break;
740 	case MLX5_FLOW_MARK:
741 		switch (config->dv_xmeta_en) {
742 		case MLX5_XMETA_MODE_LEGACY:
743 			return REG_NON;
744 		case MLX5_XMETA_MODE_META16:
745 			return REG_C_1;
746 		case MLX5_XMETA_MODE_META32:
747 			return REG_C_0;
748 		}
749 		break;
750 	case MLX5_MTR_SFX:
751 		/*
752 		 * If meter color and flow match share one register, flow match
753 		 * should use the meter color register for match.
754 		 */
755 		if (priv->mtr_reg_share)
756 			return priv->mtr_color_reg;
757 		else
758 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
759 			       REG_C_3;
760 	case MLX5_MTR_COLOR:
761 	case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
762 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
763 		return priv->mtr_color_reg;
764 	case MLX5_COPY_MARK:
765 		/*
766 		 * Metadata COPY_MARK register using is in meter suffix sub
767 		 * flow while with meter. It's safe to share the same register.
768 		 */
769 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
770 	case MLX5_APP_TAG:
771 		/*
772 		 * If meter is enable, it will engage the register for color
773 		 * match and flow match. If meter color match is not using the
774 		 * REG_C_2, need to skip the REG_C_x be used by meter color
775 		 * match.
776 		 * If meter is disable, free to use all available registers.
777 		 */
778 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
779 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
780 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
781 		if (id > (uint32_t)(REG_C_7 - start_reg))
782 			return rte_flow_error_set(error, EINVAL,
783 						  RTE_FLOW_ERROR_TYPE_ITEM,
784 						  NULL, "invalid tag id");
785 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
786 			return rte_flow_error_set(error, ENOTSUP,
787 						  RTE_FLOW_ERROR_TYPE_ITEM,
788 						  NULL, "unsupported tag id");
789 		/*
790 		 * This case means meter is using the REG_C_x great than 2.
791 		 * Take care not to conflict with meter color REG_C_x.
792 		 * If the available index REG_C_y >= REG_C_x, skip the
793 		 * color register.
794 		 */
795 		if (skip_mtr_reg && config->flow_mreg_c
796 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
797 			if (id >= (uint32_t)(REG_C_7 - start_reg))
798 				return rte_flow_error_set(error, EINVAL,
799 						       RTE_FLOW_ERROR_TYPE_ITEM,
800 							NULL, "invalid tag id");
801 			if (config->flow_mreg_c
802 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
803 				return config->flow_mreg_c
804 					       [id + 1 + start_reg - REG_C_0];
805 			return rte_flow_error_set(error, ENOTSUP,
806 						  RTE_FLOW_ERROR_TYPE_ITEM,
807 						  NULL, "unsupported tag id");
808 		}
809 		return config->flow_mreg_c[id + start_reg - REG_C_0];
810 	}
811 	MLX5_ASSERT(false);
812 	return rte_flow_error_set(error, EINVAL,
813 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
814 				  NULL, "invalid feature name");
815 }
816 
817 /**
818  * Check extensive flow metadata register support.
819  *
820  * @param dev
821  *   Pointer to rte_eth_dev structure.
822  *
823  * @return
824  *   True if device supports extensive flow metadata register, otherwise false.
825  */
826 bool
827 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
828 {
829 	struct mlx5_priv *priv = dev->data->dev_private;
830 	struct mlx5_dev_config *config = &priv->config;
831 
832 	/*
833 	 * Having available reg_c can be regarded inclusively as supporting
834 	 * extensive flow metadata register, which could mean,
835 	 * - metadata register copy action by modify header.
836 	 * - 16 modify header actions is supported.
837 	 * - reg_c's are preserved across different domain (FDB and NIC) on
838 	 *   packet loopback by flow lookup miss.
839 	 */
840 	return config->flow_mreg_c[2] != REG_NON;
841 }
842 
843 /**
844  * Get the lowest priority.
845  *
846  * @param[in] dev
847  *   Pointer to the Ethernet device structure.
848  * @param[in] attributes
849  *   Pointer to device flow rule attributes.
850  *
851  * @return
852  *   The value of lowest priority of flow.
853  */
854 uint32_t
855 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
856 			  const struct rte_flow_attr *attr)
857 {
858 	struct mlx5_priv *priv = dev->data->dev_private;
859 
860 	if (!attr->group && !attr->transfer)
861 		return priv->config.flow_prio - 2;
862 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
863 }
864 
865 /**
866  * Calculate matcher priority of the flow.
867  *
868  * @param[in] dev
869  *   Pointer to the Ethernet device structure.
870  * @param[in] attr
871  *   Pointer to device flow rule attributes.
872  * @param[in] subpriority
873  *   The priority based on the items.
874  * @return
875  *   The matcher priority of the flow.
876  */
877 uint16_t
878 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
879 			  const struct rte_flow_attr *attr,
880 			  uint32_t subpriority)
881 {
882 	uint16_t priority = (uint16_t)attr->priority;
883 	struct mlx5_priv *priv = dev->data->dev_private;
884 
885 	if (!attr->group && !attr->transfer) {
886 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
887 			priority = priv->config.flow_prio - 1;
888 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
889 	}
890 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
891 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
892 	return priority * 3 + subpriority;
893 }
894 
895 /**
896  * Verify the @p item specifications (spec, last, mask) are compatible with the
897  * NIC capabilities.
898  *
899  * @param[in] item
900  *   Item specification.
901  * @param[in] mask
902  *   @p item->mask or flow default bit-masks.
903  * @param[in] nic_mask
904  *   Bit-masks covering supported fields by the NIC to compare with user mask.
905  * @param[in] size
906  *   Bit-masks size in bytes.
907  * @param[in] range_accepted
908  *   True if range of values is accepted for specific fields, false otherwise.
909  * @param[out] error
910  *   Pointer to error structure.
911  *
912  * @return
913  *   0 on success, a negative errno value otherwise and rte_errno is set.
914  */
915 int
916 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
917 			  const uint8_t *mask,
918 			  const uint8_t *nic_mask,
919 			  unsigned int size,
920 			  bool range_accepted,
921 			  struct rte_flow_error *error)
922 {
923 	unsigned int i;
924 
925 	MLX5_ASSERT(nic_mask);
926 	for (i = 0; i < size; ++i)
927 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
928 			return rte_flow_error_set(error, ENOTSUP,
929 						  RTE_FLOW_ERROR_TYPE_ITEM,
930 						  item,
931 						  "mask enables non supported"
932 						  " bits");
933 	if (!item->spec && (item->mask || item->last))
934 		return rte_flow_error_set(error, EINVAL,
935 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
936 					  "mask/last without a spec is not"
937 					  " supported");
938 	if (item->spec && item->last && !range_accepted) {
939 		uint8_t spec[size];
940 		uint8_t last[size];
941 		unsigned int i;
942 		int ret;
943 
944 		for (i = 0; i < size; ++i) {
945 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
946 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
947 		}
948 		ret = memcmp(spec, last, size);
949 		if (ret != 0)
950 			return rte_flow_error_set(error, EINVAL,
951 						  RTE_FLOW_ERROR_TYPE_ITEM,
952 						  item,
953 						  "range is not valid");
954 	}
955 	return 0;
956 }
957 
958 /**
959  * Adjust the hash fields according to the @p flow information.
960  *
961  * @param[in] dev_flow.
962  *   Pointer to the mlx5_flow.
963  * @param[in] tunnel
964  *   1 when the hash field is for a tunnel item.
965  * @param[in] layer_types
966  *   ETH_RSS_* types.
967  * @param[in] hash_fields
968  *   Item hash fields.
969  *
970  * @return
971  *   The hash fields that should be used.
972  */
973 uint64_t
974 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
975 			    int tunnel __rte_unused, uint64_t layer_types,
976 			    uint64_t hash_fields)
977 {
978 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
979 	int rss_request_inner = rss_desc->level >= 2;
980 
981 	/* Check RSS hash level for tunnel. */
982 	if (tunnel && rss_request_inner)
983 		hash_fields |= IBV_RX_HASH_INNER;
984 	else if (tunnel || rss_request_inner)
985 		return 0;
986 #endif
987 	/* Check if requested layer matches RSS hash fields. */
988 	if (!(rss_desc->types & layer_types))
989 		return 0;
990 	return hash_fields;
991 }
992 
993 /**
994  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
995  * if several tunnel rules are used on this queue, the tunnel ptype will be
996  * cleared.
997  *
998  * @param rxq_ctrl
999  *   Rx queue to update.
1000  */
1001 static void
1002 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1003 {
1004 	unsigned int i;
1005 	uint32_t tunnel_ptype = 0;
1006 
1007 	/* Look up for the ptype to use. */
1008 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1009 		if (!rxq_ctrl->flow_tunnels_n[i])
1010 			continue;
1011 		if (!tunnel_ptype) {
1012 			tunnel_ptype = tunnels_info[i].ptype;
1013 		} else {
1014 			tunnel_ptype = 0;
1015 			break;
1016 		}
1017 	}
1018 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1019 }
1020 
1021 /**
1022  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1023  * flow.
1024  *
1025  * @param[in] dev
1026  *   Pointer to the Ethernet device structure.
1027  * @param[in] dev_handle
1028  *   Pointer to device flow handle structure.
1029  */
1030 static void
1031 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1032 		       struct mlx5_flow_handle *dev_handle)
1033 {
1034 	struct mlx5_priv *priv = dev->data->dev_private;
1035 	const int mark = dev_handle->mark;
1036 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1037 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1038 	unsigned int i;
1039 
1040 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1041 		struct mlx5_hrxq *hrxq;
1042 
1043 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1044 			      dev_handle->rix_hrxq);
1045 		if (hrxq)
1046 			ind_tbl = hrxq->ind_table;
1047 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1048 		struct mlx5_shared_action_rss *shared_rss;
1049 
1050 		shared_rss = mlx5_ipool_get
1051 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1052 			 dev_handle->rix_srss);
1053 		if (shared_rss)
1054 			ind_tbl = shared_rss->ind_tbl;
1055 	}
1056 	if (!ind_tbl)
1057 		return;
1058 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1059 		int idx = ind_tbl->queues[i];
1060 		struct mlx5_rxq_ctrl *rxq_ctrl =
1061 			container_of((*priv->rxqs)[idx],
1062 				     struct mlx5_rxq_ctrl, rxq);
1063 
1064 		/*
1065 		 * To support metadata register copy on Tx loopback,
1066 		 * this must be always enabled (metadata may arive
1067 		 * from other port - not from local flows only.
1068 		 */
1069 		if (priv->config.dv_flow_en &&
1070 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1071 		    mlx5_flow_ext_mreg_supported(dev)) {
1072 			rxq_ctrl->rxq.mark = 1;
1073 			rxq_ctrl->flow_mark_n = 1;
1074 		} else if (mark) {
1075 			rxq_ctrl->rxq.mark = 1;
1076 			rxq_ctrl->flow_mark_n++;
1077 		}
1078 		if (tunnel) {
1079 			unsigned int j;
1080 
1081 			/* Increase the counter matching the flow. */
1082 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1083 				if ((tunnels_info[j].tunnel &
1084 				     dev_handle->layers) ==
1085 				    tunnels_info[j].tunnel) {
1086 					rxq_ctrl->flow_tunnels_n[j]++;
1087 					break;
1088 				}
1089 			}
1090 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1091 		}
1092 	}
1093 }
1094 
1095 /**
1096  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1097  *
1098  * @param[in] dev
1099  *   Pointer to the Ethernet device structure.
1100  * @param[in] flow
1101  *   Pointer to flow structure.
1102  */
1103 static void
1104 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1105 {
1106 	struct mlx5_priv *priv = dev->data->dev_private;
1107 	uint32_t handle_idx;
1108 	struct mlx5_flow_handle *dev_handle;
1109 
1110 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1111 		       handle_idx, dev_handle, next)
1112 		flow_drv_rxq_flags_set(dev, dev_handle);
1113 }
1114 
1115 /**
1116  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1117  * device flow if no other flow uses it with the same kind of request.
1118  *
1119  * @param dev
1120  *   Pointer to Ethernet device.
1121  * @param[in] dev_handle
1122  *   Pointer to the device flow handle structure.
1123  */
1124 static void
1125 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1126 			struct mlx5_flow_handle *dev_handle)
1127 {
1128 	struct mlx5_priv *priv = dev->data->dev_private;
1129 	const int mark = dev_handle->mark;
1130 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1131 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1132 	unsigned int i;
1133 
1134 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1135 		struct mlx5_hrxq *hrxq;
1136 
1137 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1138 			      dev_handle->rix_hrxq);
1139 		if (hrxq)
1140 			ind_tbl = hrxq->ind_table;
1141 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1142 		struct mlx5_shared_action_rss *shared_rss;
1143 
1144 		shared_rss = mlx5_ipool_get
1145 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1146 			 dev_handle->rix_srss);
1147 		if (shared_rss)
1148 			ind_tbl = shared_rss->ind_tbl;
1149 	}
1150 	if (!ind_tbl)
1151 		return;
1152 	MLX5_ASSERT(dev->data->dev_started);
1153 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1154 		int idx = ind_tbl->queues[i];
1155 		struct mlx5_rxq_ctrl *rxq_ctrl =
1156 			container_of((*priv->rxqs)[idx],
1157 				     struct mlx5_rxq_ctrl, rxq);
1158 
1159 		if (priv->config.dv_flow_en &&
1160 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1161 		    mlx5_flow_ext_mreg_supported(dev)) {
1162 			rxq_ctrl->rxq.mark = 1;
1163 			rxq_ctrl->flow_mark_n = 1;
1164 		} else if (mark) {
1165 			rxq_ctrl->flow_mark_n--;
1166 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1167 		}
1168 		if (tunnel) {
1169 			unsigned int j;
1170 
1171 			/* Decrease the counter matching the flow. */
1172 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1173 				if ((tunnels_info[j].tunnel &
1174 				     dev_handle->layers) ==
1175 				    tunnels_info[j].tunnel) {
1176 					rxq_ctrl->flow_tunnels_n[j]--;
1177 					break;
1178 				}
1179 			}
1180 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1181 		}
1182 	}
1183 }
1184 
1185 /**
1186  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1187  * @p flow if no other flow uses it with the same kind of request.
1188  *
1189  * @param dev
1190  *   Pointer to Ethernet device.
1191  * @param[in] flow
1192  *   Pointer to the flow.
1193  */
1194 static void
1195 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1196 {
1197 	struct mlx5_priv *priv = dev->data->dev_private;
1198 	uint32_t handle_idx;
1199 	struct mlx5_flow_handle *dev_handle;
1200 
1201 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1202 		       handle_idx, dev_handle, next)
1203 		flow_drv_rxq_flags_trim(dev, dev_handle);
1204 }
1205 
1206 /**
1207  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1208  *
1209  * @param dev
1210  *   Pointer to Ethernet device.
1211  */
1212 static void
1213 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1214 {
1215 	struct mlx5_priv *priv = dev->data->dev_private;
1216 	unsigned int i;
1217 
1218 	for (i = 0; i != priv->rxqs_n; ++i) {
1219 		struct mlx5_rxq_ctrl *rxq_ctrl;
1220 		unsigned int j;
1221 
1222 		if (!(*priv->rxqs)[i])
1223 			continue;
1224 		rxq_ctrl = container_of((*priv->rxqs)[i],
1225 					struct mlx5_rxq_ctrl, rxq);
1226 		rxq_ctrl->flow_mark_n = 0;
1227 		rxq_ctrl->rxq.mark = 0;
1228 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1229 			rxq_ctrl->flow_tunnels_n[j] = 0;
1230 		rxq_ctrl->rxq.tunnel = 0;
1231 	}
1232 }
1233 
1234 /**
1235  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1236  *
1237  * @param[in] dev
1238  *   Pointer to the Ethernet device structure.
1239  */
1240 void
1241 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1242 {
1243 	struct mlx5_priv *priv = dev->data->dev_private;
1244 	struct mlx5_rxq_data *data;
1245 	unsigned int i;
1246 
1247 	for (i = 0; i != priv->rxqs_n; ++i) {
1248 		if (!(*priv->rxqs)[i])
1249 			continue;
1250 		data = (*priv->rxqs)[i];
1251 		if (!rte_flow_dynf_metadata_avail()) {
1252 			data->dynf_meta = 0;
1253 			data->flow_meta_mask = 0;
1254 			data->flow_meta_offset = -1;
1255 			data->flow_meta_port_mask = 0;
1256 		} else {
1257 			data->dynf_meta = 1;
1258 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1259 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1260 			data->flow_meta_port_mask = (uint32_t)~0;
1261 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1262 				data->flow_meta_port_mask >>= 16;
1263 		}
1264 	}
1265 }
1266 
1267 /*
1268  * return a pointer to the desired action in the list of actions.
1269  *
1270  * @param[in] actions
1271  *   The list of actions to search the action in.
1272  * @param[in] action
1273  *   The action to find.
1274  *
1275  * @return
1276  *   Pointer to the action in the list, if found. NULL otherwise.
1277  */
1278 const struct rte_flow_action *
1279 mlx5_flow_find_action(const struct rte_flow_action *actions,
1280 		      enum rte_flow_action_type action)
1281 {
1282 	if (actions == NULL)
1283 		return NULL;
1284 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1285 		if (actions->type == action)
1286 			return actions;
1287 	return NULL;
1288 }
1289 
1290 /*
1291  * Validate the flag action.
1292  *
1293  * @param[in] action_flags
1294  *   Bit-fields that holds the actions detected until now.
1295  * @param[in] attr
1296  *   Attributes of flow that includes this action.
1297  * @param[out] error
1298  *   Pointer to error structure.
1299  *
1300  * @return
1301  *   0 on success, a negative errno value otherwise and rte_errno is set.
1302  */
1303 int
1304 mlx5_flow_validate_action_flag(uint64_t action_flags,
1305 			       const struct rte_flow_attr *attr,
1306 			       struct rte_flow_error *error)
1307 {
1308 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1309 		return rte_flow_error_set(error, EINVAL,
1310 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1311 					  "can't mark and flag in same flow");
1312 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1313 		return rte_flow_error_set(error, EINVAL,
1314 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1315 					  "can't have 2 flag"
1316 					  " actions in same flow");
1317 	if (attr->egress)
1318 		return rte_flow_error_set(error, ENOTSUP,
1319 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1320 					  "flag action not supported for "
1321 					  "egress");
1322 	return 0;
1323 }
1324 
1325 /*
1326  * Validate the mark action.
1327  *
1328  * @param[in] action
1329  *   Pointer to the queue action.
1330  * @param[in] action_flags
1331  *   Bit-fields that holds the actions detected until now.
1332  * @param[in] attr
1333  *   Attributes of flow that includes this action.
1334  * @param[out] error
1335  *   Pointer to error structure.
1336  *
1337  * @return
1338  *   0 on success, a negative errno value otherwise and rte_errno is set.
1339  */
1340 int
1341 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1342 			       uint64_t action_flags,
1343 			       const struct rte_flow_attr *attr,
1344 			       struct rte_flow_error *error)
1345 {
1346 	const struct rte_flow_action_mark *mark = action->conf;
1347 
1348 	if (!mark)
1349 		return rte_flow_error_set(error, EINVAL,
1350 					  RTE_FLOW_ERROR_TYPE_ACTION,
1351 					  action,
1352 					  "configuration cannot be null");
1353 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1354 		return rte_flow_error_set(error, EINVAL,
1355 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1356 					  &mark->id,
1357 					  "mark id must in 0 <= id < "
1358 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1359 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1360 		return rte_flow_error_set(error, EINVAL,
1361 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1362 					  "can't flag and mark in same flow");
1363 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1364 		return rte_flow_error_set(error, EINVAL,
1365 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1366 					  "can't have 2 mark actions in same"
1367 					  " flow");
1368 	if (attr->egress)
1369 		return rte_flow_error_set(error, ENOTSUP,
1370 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1371 					  "mark action not supported for "
1372 					  "egress");
1373 	return 0;
1374 }
1375 
1376 /*
1377  * Validate the drop action.
1378  *
1379  * @param[in] action_flags
1380  *   Bit-fields that holds the actions detected until now.
1381  * @param[in] attr
1382  *   Attributes of flow that includes this action.
1383  * @param[out] error
1384  *   Pointer to error structure.
1385  *
1386  * @return
1387  *   0 on success, a negative errno value otherwise and rte_errno is set.
1388  */
1389 int
1390 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1391 			       const struct rte_flow_attr *attr,
1392 			       struct rte_flow_error *error)
1393 {
1394 	if (attr->egress)
1395 		return rte_flow_error_set(error, ENOTSUP,
1396 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1397 					  "drop action not supported for "
1398 					  "egress");
1399 	return 0;
1400 }
1401 
1402 /*
1403  * Validate the queue action.
1404  *
1405  * @param[in] action
1406  *   Pointer to the queue action.
1407  * @param[in] action_flags
1408  *   Bit-fields that holds the actions detected until now.
1409  * @param[in] dev
1410  *   Pointer to the Ethernet device structure.
1411  * @param[in] attr
1412  *   Attributes of flow that includes this action.
1413  * @param[out] error
1414  *   Pointer to error structure.
1415  *
1416  * @return
1417  *   0 on success, a negative errno value otherwise and rte_errno is set.
1418  */
1419 int
1420 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1421 				uint64_t action_flags,
1422 				struct rte_eth_dev *dev,
1423 				const struct rte_flow_attr *attr,
1424 				struct rte_flow_error *error)
1425 {
1426 	struct mlx5_priv *priv = dev->data->dev_private;
1427 	const struct rte_flow_action_queue *queue = action->conf;
1428 
1429 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1430 		return rte_flow_error_set(error, EINVAL,
1431 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1432 					  "can't have 2 fate actions in"
1433 					  " same flow");
1434 	if (!priv->rxqs_n)
1435 		return rte_flow_error_set(error, EINVAL,
1436 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1437 					  NULL, "No Rx queues configured");
1438 	if (queue->index >= priv->rxqs_n)
1439 		return rte_flow_error_set(error, EINVAL,
1440 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1441 					  &queue->index,
1442 					  "queue index out of range");
1443 	if (!(*priv->rxqs)[queue->index])
1444 		return rte_flow_error_set(error, EINVAL,
1445 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1446 					  &queue->index,
1447 					  "queue is not configured");
1448 	if (attr->egress)
1449 		return rte_flow_error_set(error, ENOTSUP,
1450 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1451 					  "queue action not supported for "
1452 					  "egress");
1453 	return 0;
1454 }
1455 
1456 /*
1457  * Validate the rss action.
1458  *
1459  * @param[in] dev
1460  *   Pointer to the Ethernet device structure.
1461  * @param[in] action
1462  *   Pointer to the queue action.
1463  * @param[out] error
1464  *   Pointer to error structure.
1465  *
1466  * @return
1467  *   0 on success, a negative errno value otherwise and rte_errno is set.
1468  */
1469 int
1470 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1471 			 const struct rte_flow_action *action,
1472 			 struct rte_flow_error *error)
1473 {
1474 	struct mlx5_priv *priv = dev->data->dev_private;
1475 	const struct rte_flow_action_rss *rss = action->conf;
1476 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1477 	unsigned int i;
1478 
1479 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1480 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1481 		return rte_flow_error_set(error, ENOTSUP,
1482 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1483 					  &rss->func,
1484 					  "RSS hash function not supported");
1485 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1486 	if (rss->level > 2)
1487 #else
1488 	if (rss->level > 1)
1489 #endif
1490 		return rte_flow_error_set(error, ENOTSUP,
1491 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1492 					  &rss->level,
1493 					  "tunnel RSS is not supported");
1494 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1495 	if (rss->key_len == 0 && rss->key != NULL)
1496 		return rte_flow_error_set(error, ENOTSUP,
1497 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 					  &rss->key_len,
1499 					  "RSS hash key length 0");
1500 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1501 		return rte_flow_error_set(error, ENOTSUP,
1502 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1503 					  &rss->key_len,
1504 					  "RSS hash key too small");
1505 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1506 		return rte_flow_error_set(error, ENOTSUP,
1507 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1508 					  &rss->key_len,
1509 					  "RSS hash key too large");
1510 	if (rss->queue_num > priv->config.ind_table_max_size)
1511 		return rte_flow_error_set(error, ENOTSUP,
1512 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1513 					  &rss->queue_num,
1514 					  "number of queues too large");
1515 	if (rss->types & MLX5_RSS_HF_MASK)
1516 		return rte_flow_error_set(error, ENOTSUP,
1517 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1518 					  &rss->types,
1519 					  "some RSS protocols are not"
1520 					  " supported");
1521 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1522 	    !(rss->types & ETH_RSS_IP))
1523 		return rte_flow_error_set(error, EINVAL,
1524 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1525 					  "L3 partial RSS requested but L3 RSS"
1526 					  " type not specified");
1527 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1528 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1529 		return rte_flow_error_set(error, EINVAL,
1530 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1531 					  "L4 partial RSS requested but L4 RSS"
1532 					  " type not specified");
1533 	if (!priv->rxqs_n)
1534 		return rte_flow_error_set(error, EINVAL,
1535 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1536 					  NULL, "No Rx queues configured");
1537 	if (!rss->queue_num)
1538 		return rte_flow_error_set(error, EINVAL,
1539 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1540 					  NULL, "No queues configured");
1541 	for (i = 0; i != rss->queue_num; ++i) {
1542 		struct mlx5_rxq_ctrl *rxq_ctrl;
1543 
1544 		if (rss->queue[i] >= priv->rxqs_n)
1545 			return rte_flow_error_set
1546 				(error, EINVAL,
1547 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1548 				 &rss->queue[i], "queue index out of range");
1549 		if (!(*priv->rxqs)[rss->queue[i]])
1550 			return rte_flow_error_set
1551 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1552 				 &rss->queue[i], "queue is not configured");
1553 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1554 					struct mlx5_rxq_ctrl, rxq);
1555 		if (i == 0)
1556 			rxq_type = rxq_ctrl->type;
1557 		if (rxq_type != rxq_ctrl->type)
1558 			return rte_flow_error_set
1559 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1560 				 &rss->queue[i],
1561 				 "combining hairpin and regular RSS queues is not supported");
1562 	}
1563 	return 0;
1564 }
1565 
1566 /*
1567  * Validate the rss action.
1568  *
1569  * @param[in] action
1570  *   Pointer to the queue action.
1571  * @param[in] action_flags
1572  *   Bit-fields that holds the actions detected until now.
1573  * @param[in] dev
1574  *   Pointer to the Ethernet device structure.
1575  * @param[in] attr
1576  *   Attributes of flow that includes this action.
1577  * @param[in] item_flags
1578  *   Items that were detected.
1579  * @param[out] error
1580  *   Pointer to error structure.
1581  *
1582  * @return
1583  *   0 on success, a negative errno value otherwise and rte_errno is set.
1584  */
1585 int
1586 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1587 			      uint64_t action_flags,
1588 			      struct rte_eth_dev *dev,
1589 			      const struct rte_flow_attr *attr,
1590 			      uint64_t item_flags,
1591 			      struct rte_flow_error *error)
1592 {
1593 	const struct rte_flow_action_rss *rss = action->conf;
1594 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1595 	int ret;
1596 
1597 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1598 		return rte_flow_error_set(error, EINVAL,
1599 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1600 					  "can't have 2 fate actions"
1601 					  " in same flow");
1602 	ret = mlx5_validate_action_rss(dev, action, error);
1603 	if (ret)
1604 		return ret;
1605 	if (attr->egress)
1606 		return rte_flow_error_set(error, ENOTSUP,
1607 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1608 					  "rss action not supported for "
1609 					  "egress");
1610 	if (rss->level > 1 && !tunnel)
1611 		return rte_flow_error_set(error, EINVAL,
1612 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1613 					  "inner RSS is not supported for "
1614 					  "non-tunnel flows");
1615 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1616 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1617 		return rte_flow_error_set(error, EINVAL,
1618 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1619 					  "RSS on eCPRI is not supported now");
1620 	}
1621 	return 0;
1622 }
1623 
1624 /*
1625  * Validate the default miss action.
1626  *
1627  * @param[in] action_flags
1628  *   Bit-fields that holds the actions detected until now.
1629  * @param[out] error
1630  *   Pointer to error structure.
1631  *
1632  * @return
1633  *   0 on success, a negative errno value otherwise and rte_errno is set.
1634  */
1635 int
1636 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1637 				const struct rte_flow_attr *attr,
1638 				struct rte_flow_error *error)
1639 {
1640 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1641 		return rte_flow_error_set(error, EINVAL,
1642 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1643 					  "can't have 2 fate actions in"
1644 					  " same flow");
1645 	if (attr->egress)
1646 		return rte_flow_error_set(error, ENOTSUP,
1647 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1648 					  "default miss action not supported "
1649 					  "for egress");
1650 	if (attr->group)
1651 		return rte_flow_error_set(error, ENOTSUP,
1652 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1653 					  "only group 0 is supported");
1654 	if (attr->transfer)
1655 		return rte_flow_error_set(error, ENOTSUP,
1656 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1657 					  NULL, "transfer is not supported");
1658 	return 0;
1659 }
1660 
1661 /*
1662  * Validate the count action.
1663  *
1664  * @param[in] dev
1665  *   Pointer to the Ethernet device structure.
1666  * @param[in] attr
1667  *   Attributes of flow that includes this action.
1668  * @param[out] error
1669  *   Pointer to error structure.
1670  *
1671  * @return
1672  *   0 on success, a negative errno value otherwise and rte_errno is set.
1673  */
1674 int
1675 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1676 				const struct rte_flow_attr *attr,
1677 				struct rte_flow_error *error)
1678 {
1679 	if (attr->egress)
1680 		return rte_flow_error_set(error, ENOTSUP,
1681 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1682 					  "count action not supported for "
1683 					  "egress");
1684 	return 0;
1685 }
1686 
1687 /**
1688  * Verify the @p attributes will be correctly understood by the NIC and store
1689  * them in the @p flow if everything is correct.
1690  *
1691  * @param[in] dev
1692  *   Pointer to the Ethernet device structure.
1693  * @param[in] attributes
1694  *   Pointer to flow attributes
1695  * @param[out] error
1696  *   Pointer to error structure.
1697  *
1698  * @return
1699  *   0 on success, a negative errno value otherwise and rte_errno is set.
1700  */
1701 int
1702 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1703 			      const struct rte_flow_attr *attributes,
1704 			      struct rte_flow_error *error)
1705 {
1706 	struct mlx5_priv *priv = dev->data->dev_private;
1707 	uint32_t priority_max = priv->config.flow_prio - 1;
1708 
1709 	if (attributes->group)
1710 		return rte_flow_error_set(error, ENOTSUP,
1711 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1712 					  NULL, "groups is not supported");
1713 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1714 	    attributes->priority >= priority_max)
1715 		return rte_flow_error_set(error, ENOTSUP,
1716 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1717 					  NULL, "priority out of range");
1718 	if (attributes->egress)
1719 		return rte_flow_error_set(error, ENOTSUP,
1720 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1721 					  "egress is not supported");
1722 	if (attributes->transfer && !priv->config.dv_esw_en)
1723 		return rte_flow_error_set(error, ENOTSUP,
1724 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1725 					  NULL, "transfer is not supported");
1726 	if (!attributes->ingress)
1727 		return rte_flow_error_set(error, EINVAL,
1728 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1729 					  NULL,
1730 					  "ingress attribute is mandatory");
1731 	return 0;
1732 }
1733 
1734 /**
1735  * Validate ICMP6 item.
1736  *
1737  * @param[in] item
1738  *   Item specification.
1739  * @param[in] item_flags
1740  *   Bit-fields that holds the items detected until now.
1741  * @param[in] ext_vlan_sup
1742  *   Whether extended VLAN features are supported or not.
1743  * @param[out] error
1744  *   Pointer to error structure.
1745  *
1746  * @return
1747  *   0 on success, a negative errno value otherwise and rte_errno is set.
1748  */
1749 int
1750 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1751 			       uint64_t item_flags,
1752 			       uint8_t target_protocol,
1753 			       struct rte_flow_error *error)
1754 {
1755 	const struct rte_flow_item_icmp6 *mask = item->mask;
1756 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1757 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1758 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1759 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1760 				      MLX5_FLOW_LAYER_OUTER_L4;
1761 	int ret;
1762 
1763 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1764 		return rte_flow_error_set(error, EINVAL,
1765 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1766 					  "protocol filtering not compatible"
1767 					  " with ICMP6 layer");
1768 	if (!(item_flags & l3m))
1769 		return rte_flow_error_set(error, EINVAL,
1770 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1771 					  "IPv6 is mandatory to filter on"
1772 					  " ICMP6");
1773 	if (item_flags & l4m)
1774 		return rte_flow_error_set(error, EINVAL,
1775 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1776 					  "multiple L4 layers not supported");
1777 	if (!mask)
1778 		mask = &rte_flow_item_icmp6_mask;
1779 	ret = mlx5_flow_item_acceptable
1780 		(item, (const uint8_t *)mask,
1781 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1782 		 sizeof(struct rte_flow_item_icmp6),
1783 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1784 	if (ret < 0)
1785 		return ret;
1786 	return 0;
1787 }
1788 
1789 /**
1790  * Validate ICMP item.
1791  *
1792  * @param[in] item
1793  *   Item specification.
1794  * @param[in] item_flags
1795  *   Bit-fields that holds the items detected until now.
1796  * @param[out] error
1797  *   Pointer to error structure.
1798  *
1799  * @return
1800  *   0 on success, a negative errno value otherwise and rte_errno is set.
1801  */
1802 int
1803 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1804 			     uint64_t item_flags,
1805 			     uint8_t target_protocol,
1806 			     struct rte_flow_error *error)
1807 {
1808 	const struct rte_flow_item_icmp *mask = item->mask;
1809 	const struct rte_flow_item_icmp nic_mask = {
1810 		.hdr.icmp_type = 0xff,
1811 		.hdr.icmp_code = 0xff,
1812 		.hdr.icmp_ident = RTE_BE16(0xffff),
1813 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1814 	};
1815 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1816 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1817 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1818 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1819 				      MLX5_FLOW_LAYER_OUTER_L4;
1820 	int ret;
1821 
1822 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1823 		return rte_flow_error_set(error, EINVAL,
1824 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1825 					  "protocol filtering not compatible"
1826 					  " with ICMP layer");
1827 	if (!(item_flags & l3m))
1828 		return rte_flow_error_set(error, EINVAL,
1829 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1830 					  "IPv4 is mandatory to filter"
1831 					  " on ICMP");
1832 	if (item_flags & l4m)
1833 		return rte_flow_error_set(error, EINVAL,
1834 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1835 					  "multiple L4 layers not supported");
1836 	if (!mask)
1837 		mask = &nic_mask;
1838 	ret = mlx5_flow_item_acceptable
1839 		(item, (const uint8_t *)mask,
1840 		 (const uint8_t *)&nic_mask,
1841 		 sizeof(struct rte_flow_item_icmp),
1842 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1843 	if (ret < 0)
1844 		return ret;
1845 	return 0;
1846 }
1847 
1848 /**
1849  * Validate Ethernet item.
1850  *
1851  * @param[in] item
1852  *   Item specification.
1853  * @param[in] item_flags
1854  *   Bit-fields that holds the items detected until now.
1855  * @param[out] error
1856  *   Pointer to error structure.
1857  *
1858  * @return
1859  *   0 on success, a negative errno value otherwise and rte_errno is set.
1860  */
1861 int
1862 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1863 			    uint64_t item_flags, bool ext_vlan_sup,
1864 			    struct rte_flow_error *error)
1865 {
1866 	const struct rte_flow_item_eth *mask = item->mask;
1867 	const struct rte_flow_item_eth nic_mask = {
1868 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1869 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1870 		.type = RTE_BE16(0xffff),
1871 		.has_vlan = ext_vlan_sup ? 1 : 0,
1872 	};
1873 	int ret;
1874 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1875 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1876 				       MLX5_FLOW_LAYER_OUTER_L2;
1877 
1878 	if (item_flags & ethm)
1879 		return rte_flow_error_set(error, ENOTSUP,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "multiple L2 layers not supported");
1882 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1883 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1884 		return rte_flow_error_set(error, EINVAL,
1885 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1886 					  "L2 layer should not follow "
1887 					  "L3 layers");
1888 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1889 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1890 		return rte_flow_error_set(error, EINVAL,
1891 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1892 					  "L2 layer should not follow VLAN");
1893 	if (!mask)
1894 		mask = &rte_flow_item_eth_mask;
1895 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1896 					(const uint8_t *)&nic_mask,
1897 					sizeof(struct rte_flow_item_eth),
1898 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1899 	return ret;
1900 }
1901 
1902 /**
1903  * Validate VLAN item.
1904  *
1905  * @param[in] item
1906  *   Item specification.
1907  * @param[in] item_flags
1908  *   Bit-fields that holds the items detected until now.
1909  * @param[in] dev
1910  *   Ethernet device flow is being created on.
1911  * @param[out] error
1912  *   Pointer to error structure.
1913  *
1914  * @return
1915  *   0 on success, a negative errno value otherwise and rte_errno is set.
1916  */
1917 int
1918 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1919 			     uint64_t item_flags,
1920 			     struct rte_eth_dev *dev,
1921 			     struct rte_flow_error *error)
1922 {
1923 	const struct rte_flow_item_vlan *spec = item->spec;
1924 	const struct rte_flow_item_vlan *mask = item->mask;
1925 	const struct rte_flow_item_vlan nic_mask = {
1926 		.tci = RTE_BE16(UINT16_MAX),
1927 		.inner_type = RTE_BE16(UINT16_MAX),
1928 	};
1929 	uint16_t vlan_tag = 0;
1930 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1931 	int ret;
1932 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1933 					MLX5_FLOW_LAYER_INNER_L4) :
1934 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1935 					MLX5_FLOW_LAYER_OUTER_L4);
1936 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1937 					MLX5_FLOW_LAYER_OUTER_VLAN;
1938 
1939 	if (item_flags & vlanm)
1940 		return rte_flow_error_set(error, EINVAL,
1941 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1942 					  "multiple VLAN layers not supported");
1943 	else if ((item_flags & l34m) != 0)
1944 		return rte_flow_error_set(error, EINVAL,
1945 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1946 					  "VLAN cannot follow L3/L4 layer");
1947 	if (!mask)
1948 		mask = &rte_flow_item_vlan_mask;
1949 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1950 					(const uint8_t *)&nic_mask,
1951 					sizeof(struct rte_flow_item_vlan),
1952 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1953 	if (ret)
1954 		return ret;
1955 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1956 		struct mlx5_priv *priv = dev->data->dev_private;
1957 
1958 		if (priv->vmwa_context) {
1959 			/*
1960 			 * Non-NULL context means we have a virtual machine
1961 			 * and SR-IOV enabled, we have to create VLAN interface
1962 			 * to make hypervisor to setup E-Switch vport
1963 			 * context correctly. We avoid creating the multiple
1964 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1965 			 */
1966 			return rte_flow_error_set(error, EINVAL,
1967 						  RTE_FLOW_ERROR_TYPE_ITEM,
1968 						  item,
1969 						  "VLAN tag mask is not"
1970 						  " supported in virtual"
1971 						  " environment");
1972 		}
1973 	}
1974 	if (spec) {
1975 		vlan_tag = spec->tci;
1976 		vlan_tag &= mask->tci;
1977 	}
1978 	/*
1979 	 * From verbs perspective an empty VLAN is equivalent
1980 	 * to a packet without VLAN layer.
1981 	 */
1982 	if (!vlan_tag)
1983 		return rte_flow_error_set(error, EINVAL,
1984 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1985 					  item->spec,
1986 					  "VLAN cannot be empty");
1987 	return 0;
1988 }
1989 
1990 /**
1991  * Validate IPV4 item.
1992  *
1993  * @param[in] item
1994  *   Item specification.
1995  * @param[in] item_flags
1996  *   Bit-fields that holds the items detected until now.
1997  * @param[in] last_item
1998  *   Previous validated item in the pattern items.
1999  * @param[in] ether_type
2000  *   Type in the ethernet layer header (including dot1q).
2001  * @param[in] acc_mask
2002  *   Acceptable mask, if NULL default internal default mask
2003  *   will be used to check whether item fields are supported.
2004  * @param[in] range_accepted
2005  *   True if range of values is accepted for specific fields, false otherwise.
2006  * @param[out] error
2007  *   Pointer to error structure.
2008  *
2009  * @return
2010  *   0 on success, a negative errno value otherwise and rte_errno is set.
2011  */
2012 int
2013 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2014 			     uint64_t item_flags,
2015 			     uint64_t last_item,
2016 			     uint16_t ether_type,
2017 			     const struct rte_flow_item_ipv4 *acc_mask,
2018 			     bool range_accepted,
2019 			     struct rte_flow_error *error)
2020 {
2021 	const struct rte_flow_item_ipv4 *mask = item->mask;
2022 	const struct rte_flow_item_ipv4 *spec = item->spec;
2023 	const struct rte_flow_item_ipv4 nic_mask = {
2024 		.hdr = {
2025 			.src_addr = RTE_BE32(0xffffffff),
2026 			.dst_addr = RTE_BE32(0xffffffff),
2027 			.type_of_service = 0xff,
2028 			.next_proto_id = 0xff,
2029 		},
2030 	};
2031 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2032 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2033 				      MLX5_FLOW_LAYER_OUTER_L3;
2034 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2035 				      MLX5_FLOW_LAYER_OUTER_L4;
2036 	int ret;
2037 	uint8_t next_proto = 0xFF;
2038 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2039 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2040 				  MLX5_FLOW_LAYER_INNER_VLAN);
2041 
2042 	if ((last_item & l2_vlan) && ether_type &&
2043 	    ether_type != RTE_ETHER_TYPE_IPV4)
2044 		return rte_flow_error_set(error, EINVAL,
2045 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2046 					  "IPv4 cannot follow L2/VLAN layer "
2047 					  "which ether type is not IPv4");
2048 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2049 		if (mask && spec)
2050 			next_proto = mask->hdr.next_proto_id &
2051 				     spec->hdr.next_proto_id;
2052 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2053 			return rte_flow_error_set(error, EINVAL,
2054 						  RTE_FLOW_ERROR_TYPE_ITEM,
2055 						  item,
2056 						  "multiple tunnel "
2057 						  "not supported");
2058 	}
2059 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2060 		return rte_flow_error_set(error, EINVAL,
2061 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2062 					  "wrong tunnel type - IPv6 specified "
2063 					  "but IPv4 item provided");
2064 	if (item_flags & l3m)
2065 		return rte_flow_error_set(error, ENOTSUP,
2066 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2067 					  "multiple L3 layers not supported");
2068 	else if (item_flags & l4m)
2069 		return rte_flow_error_set(error, EINVAL,
2070 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2071 					  "L3 cannot follow an L4 layer.");
2072 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2073 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2074 		return rte_flow_error_set(error, EINVAL,
2075 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 					  "L3 cannot follow an NVGRE layer.");
2077 	if (!mask)
2078 		mask = &rte_flow_item_ipv4_mask;
2079 	else if (mask->hdr.next_proto_id != 0 &&
2080 		 mask->hdr.next_proto_id != 0xff)
2081 		return rte_flow_error_set(error, EINVAL,
2082 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2083 					  "partial mask is not supported"
2084 					  " for protocol");
2085 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2086 					acc_mask ? (const uint8_t *)acc_mask
2087 						 : (const uint8_t *)&nic_mask,
2088 					sizeof(struct rte_flow_item_ipv4),
2089 					range_accepted, error);
2090 	if (ret < 0)
2091 		return ret;
2092 	return 0;
2093 }
2094 
2095 /**
2096  * Validate IPV6 item.
2097  *
2098  * @param[in] item
2099  *   Item specification.
2100  * @param[in] item_flags
2101  *   Bit-fields that holds the items detected until now.
2102  * @param[in] last_item
2103  *   Previous validated item in the pattern items.
2104  * @param[in] ether_type
2105  *   Type in the ethernet layer header (including dot1q).
2106  * @param[in] acc_mask
2107  *   Acceptable mask, if NULL default internal default mask
2108  *   will be used to check whether item fields are supported.
2109  * @param[out] error
2110  *   Pointer to error structure.
2111  *
2112  * @return
2113  *   0 on success, a negative errno value otherwise and rte_errno is set.
2114  */
2115 int
2116 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2117 			     uint64_t item_flags,
2118 			     uint64_t last_item,
2119 			     uint16_t ether_type,
2120 			     const struct rte_flow_item_ipv6 *acc_mask,
2121 			     struct rte_flow_error *error)
2122 {
2123 	const struct rte_flow_item_ipv6 *mask = item->mask;
2124 	const struct rte_flow_item_ipv6 *spec = item->spec;
2125 	const struct rte_flow_item_ipv6 nic_mask = {
2126 		.hdr = {
2127 			.src_addr =
2128 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2129 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2130 			.dst_addr =
2131 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2132 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2133 			.vtc_flow = RTE_BE32(0xffffffff),
2134 			.proto = 0xff,
2135 		},
2136 	};
2137 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2138 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2139 				      MLX5_FLOW_LAYER_OUTER_L3;
2140 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2141 				      MLX5_FLOW_LAYER_OUTER_L4;
2142 	int ret;
2143 	uint8_t next_proto = 0xFF;
2144 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2145 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2146 				  MLX5_FLOW_LAYER_INNER_VLAN);
2147 
2148 	if ((last_item & l2_vlan) && ether_type &&
2149 	    ether_type != RTE_ETHER_TYPE_IPV6)
2150 		return rte_flow_error_set(error, EINVAL,
2151 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2152 					  "IPv6 cannot follow L2/VLAN layer "
2153 					  "which ether type is not IPv6");
2154 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2155 		next_proto = spec->hdr.proto;
2156 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2157 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2158 			return rte_flow_error_set(error, EINVAL,
2159 						  RTE_FLOW_ERROR_TYPE_ITEM,
2160 						  item,
2161 						  "multiple tunnel "
2162 						  "not supported");
2163 	}
2164 	if (next_proto == IPPROTO_HOPOPTS  ||
2165 	    next_proto == IPPROTO_ROUTING  ||
2166 	    next_proto == IPPROTO_FRAGMENT ||
2167 	    next_proto == IPPROTO_ESP	   ||
2168 	    next_proto == IPPROTO_AH	   ||
2169 	    next_proto == IPPROTO_DSTOPTS)
2170 		return rte_flow_error_set(error, EINVAL,
2171 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2172 					  "IPv6 proto (next header) should "
2173 					  "not be set as extension header");
2174 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2175 		return rte_flow_error_set(error, EINVAL,
2176 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2177 					  "wrong tunnel type - IPv4 specified "
2178 					  "but IPv6 item provided");
2179 	if (item_flags & l3m)
2180 		return rte_flow_error_set(error, ENOTSUP,
2181 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2182 					  "multiple L3 layers not supported");
2183 	else if (item_flags & l4m)
2184 		return rte_flow_error_set(error, EINVAL,
2185 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2186 					  "L3 cannot follow an L4 layer.");
2187 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2188 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2189 		return rte_flow_error_set(error, EINVAL,
2190 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2191 					  "L3 cannot follow an NVGRE layer.");
2192 	if (!mask)
2193 		mask = &rte_flow_item_ipv6_mask;
2194 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2195 					acc_mask ? (const uint8_t *)acc_mask
2196 						 : (const uint8_t *)&nic_mask,
2197 					sizeof(struct rte_flow_item_ipv6),
2198 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2199 	if (ret < 0)
2200 		return ret;
2201 	return 0;
2202 }
2203 
2204 /**
2205  * Validate UDP item.
2206  *
2207  * @param[in] item
2208  *   Item specification.
2209  * @param[in] item_flags
2210  *   Bit-fields that holds the items detected until now.
2211  * @param[in] target_protocol
2212  *   The next protocol in the previous item.
2213  * @param[in] flow_mask
2214  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2215  * @param[out] error
2216  *   Pointer to error structure.
2217  *
2218  * @return
2219  *   0 on success, a negative errno value otherwise and rte_errno is set.
2220  */
2221 int
2222 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2223 			    uint64_t item_flags,
2224 			    uint8_t target_protocol,
2225 			    struct rte_flow_error *error)
2226 {
2227 	const struct rte_flow_item_udp *mask = item->mask;
2228 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2229 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2230 				      MLX5_FLOW_LAYER_OUTER_L3;
2231 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2232 				      MLX5_FLOW_LAYER_OUTER_L4;
2233 	int ret;
2234 
2235 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2236 		return rte_flow_error_set(error, EINVAL,
2237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 					  "protocol filtering not compatible"
2239 					  " with UDP layer");
2240 	if (!(item_flags & l3m))
2241 		return rte_flow_error_set(error, EINVAL,
2242 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2243 					  "L3 is mandatory to filter on L4");
2244 	if (item_flags & l4m)
2245 		return rte_flow_error_set(error, EINVAL,
2246 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2247 					  "multiple L4 layers not supported");
2248 	if (!mask)
2249 		mask = &rte_flow_item_udp_mask;
2250 	ret = mlx5_flow_item_acceptable
2251 		(item, (const uint8_t *)mask,
2252 		 (const uint8_t *)&rte_flow_item_udp_mask,
2253 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2254 		 error);
2255 	if (ret < 0)
2256 		return ret;
2257 	return 0;
2258 }
2259 
2260 /**
2261  * Validate TCP item.
2262  *
2263  * @param[in] item
2264  *   Item specification.
2265  * @param[in] item_flags
2266  *   Bit-fields that holds the items detected until now.
2267  * @param[in] target_protocol
2268  *   The next protocol in the previous item.
2269  * @param[out] error
2270  *   Pointer to error structure.
2271  *
2272  * @return
2273  *   0 on success, a negative errno value otherwise and rte_errno is set.
2274  */
2275 int
2276 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2277 			    uint64_t item_flags,
2278 			    uint8_t target_protocol,
2279 			    const struct rte_flow_item_tcp *flow_mask,
2280 			    struct rte_flow_error *error)
2281 {
2282 	const struct rte_flow_item_tcp *mask = item->mask;
2283 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2284 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2285 				      MLX5_FLOW_LAYER_OUTER_L3;
2286 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2287 				      MLX5_FLOW_LAYER_OUTER_L4;
2288 	int ret;
2289 
2290 	MLX5_ASSERT(flow_mask);
2291 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2292 		return rte_flow_error_set(error, EINVAL,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "protocol filtering not compatible"
2295 					  " with TCP layer");
2296 	if (!(item_flags & l3m))
2297 		return rte_flow_error_set(error, EINVAL,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "L3 is mandatory to filter on L4");
2300 	if (item_flags & l4m)
2301 		return rte_flow_error_set(error, EINVAL,
2302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 					  "multiple L4 layers not supported");
2304 	if (!mask)
2305 		mask = &rte_flow_item_tcp_mask;
2306 	ret = mlx5_flow_item_acceptable
2307 		(item, (const uint8_t *)mask,
2308 		 (const uint8_t *)flow_mask,
2309 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2310 		 error);
2311 	if (ret < 0)
2312 		return ret;
2313 	return 0;
2314 }
2315 
2316 /**
2317  * Validate VXLAN item.
2318  *
2319  * @param[in] item
2320  *   Item specification.
2321  * @param[in] item_flags
2322  *   Bit-fields that holds the items detected until now.
2323  * @param[in] target_protocol
2324  *   The next protocol in the previous item.
2325  * @param[out] error
2326  *   Pointer to error structure.
2327  *
2328  * @return
2329  *   0 on success, a negative errno value otherwise and rte_errno is set.
2330  */
2331 int
2332 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2333 			      uint64_t item_flags,
2334 			      struct rte_flow_error *error)
2335 {
2336 	const struct rte_flow_item_vxlan *spec = item->spec;
2337 	const struct rte_flow_item_vxlan *mask = item->mask;
2338 	int ret;
2339 	union vni {
2340 		uint32_t vlan_id;
2341 		uint8_t vni[4];
2342 	} id = { .vlan_id = 0, };
2343 
2344 
2345 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2346 		return rte_flow_error_set(error, ENOTSUP,
2347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2348 					  "multiple tunnel layers not"
2349 					  " supported");
2350 	/*
2351 	 * Verify only UDPv4 is present as defined in
2352 	 * https://tools.ietf.org/html/rfc7348
2353 	 */
2354 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2355 		return rte_flow_error_set(error, EINVAL,
2356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2357 					  "no outer UDP layer found");
2358 	if (!mask)
2359 		mask = &rte_flow_item_vxlan_mask;
2360 	ret = mlx5_flow_item_acceptable
2361 		(item, (const uint8_t *)mask,
2362 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2363 		 sizeof(struct rte_flow_item_vxlan),
2364 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2365 	if (ret < 0)
2366 		return ret;
2367 	if (spec) {
2368 		memcpy(&id.vni[1], spec->vni, 3);
2369 		memcpy(&id.vni[1], mask->vni, 3);
2370 	}
2371 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2372 		return rte_flow_error_set(error, ENOTSUP,
2373 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2374 					  "VXLAN tunnel must be fully defined");
2375 	return 0;
2376 }
2377 
2378 /**
2379  * Validate VXLAN_GPE item.
2380  *
2381  * @param[in] item
2382  *   Item specification.
2383  * @param[in] item_flags
2384  *   Bit-fields that holds the items detected until now.
2385  * @param[in] priv
2386  *   Pointer to the private data structure.
2387  * @param[in] target_protocol
2388  *   The next protocol in the previous item.
2389  * @param[out] error
2390  *   Pointer to error structure.
2391  *
2392  * @return
2393  *   0 on success, a negative errno value otherwise and rte_errno is set.
2394  */
2395 int
2396 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2397 				  uint64_t item_flags,
2398 				  struct rte_eth_dev *dev,
2399 				  struct rte_flow_error *error)
2400 {
2401 	struct mlx5_priv *priv = dev->data->dev_private;
2402 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2403 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2404 	int ret;
2405 	union vni {
2406 		uint32_t vlan_id;
2407 		uint8_t vni[4];
2408 	} id = { .vlan_id = 0, };
2409 
2410 	if (!priv->config.l3_vxlan_en)
2411 		return rte_flow_error_set(error, ENOTSUP,
2412 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2413 					  "L3 VXLAN is not enabled by device"
2414 					  " parameter and/or not configured in"
2415 					  " firmware");
2416 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2417 		return rte_flow_error_set(error, ENOTSUP,
2418 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 					  "multiple tunnel layers not"
2420 					  " supported");
2421 	/*
2422 	 * Verify only UDPv4 is present as defined in
2423 	 * https://tools.ietf.org/html/rfc7348
2424 	 */
2425 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2426 		return rte_flow_error_set(error, EINVAL,
2427 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2428 					  "no outer UDP layer found");
2429 	if (!mask)
2430 		mask = &rte_flow_item_vxlan_gpe_mask;
2431 	ret = mlx5_flow_item_acceptable
2432 		(item, (const uint8_t *)mask,
2433 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2434 		 sizeof(struct rte_flow_item_vxlan_gpe),
2435 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2436 	if (ret < 0)
2437 		return ret;
2438 	if (spec) {
2439 		if (spec->protocol)
2440 			return rte_flow_error_set(error, ENOTSUP,
2441 						  RTE_FLOW_ERROR_TYPE_ITEM,
2442 						  item,
2443 						  "VxLAN-GPE protocol"
2444 						  " not supported");
2445 		memcpy(&id.vni[1], spec->vni, 3);
2446 		memcpy(&id.vni[1], mask->vni, 3);
2447 	}
2448 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2449 		return rte_flow_error_set(error, ENOTSUP,
2450 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2451 					  "VXLAN-GPE tunnel must be fully"
2452 					  " defined");
2453 	return 0;
2454 }
2455 /**
2456  * Validate GRE Key item.
2457  *
2458  * @param[in] item
2459  *   Item specification.
2460  * @param[in] item_flags
2461  *   Bit flags to mark detected items.
2462  * @param[in] gre_item
2463  *   Pointer to gre_item
2464  * @param[out] error
2465  *   Pointer to error structure.
2466  *
2467  * @return
2468  *   0 on success, a negative errno value otherwise and rte_errno is set.
2469  */
2470 int
2471 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2472 				uint64_t item_flags,
2473 				const struct rte_flow_item *gre_item,
2474 				struct rte_flow_error *error)
2475 {
2476 	const rte_be32_t *mask = item->mask;
2477 	int ret = 0;
2478 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2479 	const struct rte_flow_item_gre *gre_spec;
2480 	const struct rte_flow_item_gre *gre_mask;
2481 
2482 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2483 		return rte_flow_error_set(error, ENOTSUP,
2484 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2485 					  "Multiple GRE key not support");
2486 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2487 		return rte_flow_error_set(error, ENOTSUP,
2488 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2489 					  "No preceding GRE header");
2490 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2491 		return rte_flow_error_set(error, ENOTSUP,
2492 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2493 					  "GRE key following a wrong item");
2494 	gre_mask = gre_item->mask;
2495 	if (!gre_mask)
2496 		gre_mask = &rte_flow_item_gre_mask;
2497 	gre_spec = gre_item->spec;
2498 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2499 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2500 		return rte_flow_error_set(error, EINVAL,
2501 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2502 					  "Key bit must be on");
2503 
2504 	if (!mask)
2505 		mask = &gre_key_default_mask;
2506 	ret = mlx5_flow_item_acceptable
2507 		(item, (const uint8_t *)mask,
2508 		 (const uint8_t *)&gre_key_default_mask,
2509 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2510 	return ret;
2511 }
2512 
2513 /**
2514  * Validate GRE item.
2515  *
2516  * @param[in] item
2517  *   Item specification.
2518  * @param[in] item_flags
2519  *   Bit flags to mark detected items.
2520  * @param[in] target_protocol
2521  *   The next protocol in the previous item.
2522  * @param[out] error
2523  *   Pointer to error structure.
2524  *
2525  * @return
2526  *   0 on success, a negative errno value otherwise and rte_errno is set.
2527  */
2528 int
2529 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2530 			    uint64_t item_flags,
2531 			    uint8_t target_protocol,
2532 			    struct rte_flow_error *error)
2533 {
2534 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2535 	const struct rte_flow_item_gre *mask = item->mask;
2536 	int ret;
2537 	const struct rte_flow_item_gre nic_mask = {
2538 		.c_rsvd0_ver = RTE_BE16(0xB000),
2539 		.protocol = RTE_BE16(UINT16_MAX),
2540 	};
2541 
2542 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2543 		return rte_flow_error_set(error, EINVAL,
2544 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2545 					  "protocol filtering not compatible"
2546 					  " with this GRE layer");
2547 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2548 		return rte_flow_error_set(error, ENOTSUP,
2549 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2550 					  "multiple tunnel layers not"
2551 					  " supported");
2552 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2553 		return rte_flow_error_set(error, ENOTSUP,
2554 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2555 					  "L3 Layer is missing");
2556 	if (!mask)
2557 		mask = &rte_flow_item_gre_mask;
2558 	ret = mlx5_flow_item_acceptable
2559 		(item, (const uint8_t *)mask,
2560 		 (const uint8_t *)&nic_mask,
2561 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2562 		 error);
2563 	if (ret < 0)
2564 		return ret;
2565 #ifndef HAVE_MLX5DV_DR
2566 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2567 	if (spec && (spec->protocol & mask->protocol))
2568 		return rte_flow_error_set(error, ENOTSUP,
2569 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2570 					  "without MPLS support the"
2571 					  " specification cannot be used for"
2572 					  " filtering");
2573 #endif
2574 #endif
2575 	return 0;
2576 }
2577 
2578 /**
2579  * Validate Geneve item.
2580  *
2581  * @param[in] item
2582  *   Item specification.
2583  * @param[in] itemFlags
2584  *   Bit-fields that holds the items detected until now.
2585  * @param[in] enPriv
2586  *   Pointer to the private data structure.
2587  * @param[out] error
2588  *   Pointer to error structure.
2589  *
2590  * @return
2591  *   0 on success, a negative errno value otherwise and rte_errno is set.
2592  */
2593 
2594 int
2595 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2596 			       uint64_t item_flags,
2597 			       struct rte_eth_dev *dev,
2598 			       struct rte_flow_error *error)
2599 {
2600 	struct mlx5_priv *priv = dev->data->dev_private;
2601 	const struct rte_flow_item_geneve *spec = item->spec;
2602 	const struct rte_flow_item_geneve *mask = item->mask;
2603 	int ret;
2604 	uint16_t gbhdr;
2605 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2606 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2607 	const struct rte_flow_item_geneve nic_mask = {
2608 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2609 		.vni = "\xff\xff\xff",
2610 		.protocol = RTE_BE16(UINT16_MAX),
2611 	};
2612 
2613 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2614 		return rte_flow_error_set(error, ENOTSUP,
2615 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2616 					  "L3 Geneve is not enabled by device"
2617 					  " parameter and/or not configured in"
2618 					  " firmware");
2619 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2620 		return rte_flow_error_set(error, ENOTSUP,
2621 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2622 					  "multiple tunnel layers not"
2623 					  " supported");
2624 	/*
2625 	 * Verify only UDPv4 is present as defined in
2626 	 * https://tools.ietf.org/html/rfc7348
2627 	 */
2628 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2629 		return rte_flow_error_set(error, EINVAL,
2630 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2631 					  "no outer UDP layer found");
2632 	if (!mask)
2633 		mask = &rte_flow_item_geneve_mask;
2634 	ret = mlx5_flow_item_acceptable
2635 				  (item, (const uint8_t *)mask,
2636 				   (const uint8_t *)&nic_mask,
2637 				   sizeof(struct rte_flow_item_geneve),
2638 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2639 	if (ret)
2640 		return ret;
2641 	if (spec) {
2642 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2643 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2644 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2645 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2646 			return rte_flow_error_set(error, ENOTSUP,
2647 						  RTE_FLOW_ERROR_TYPE_ITEM,
2648 						  item,
2649 						  "Geneve protocol unsupported"
2650 						  " fields are being used");
2651 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2652 			return rte_flow_error_set
2653 					(error, ENOTSUP,
2654 					 RTE_FLOW_ERROR_TYPE_ITEM,
2655 					 item,
2656 					 "Unsupported Geneve options length");
2657 	}
2658 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2659 		return rte_flow_error_set
2660 				    (error, ENOTSUP,
2661 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2662 				     "Geneve tunnel must be fully defined");
2663 	return 0;
2664 }
2665 
2666 /**
2667  * Validate Geneve TLV option item.
2668  *
2669  * @param[in] item
2670  *   Item specification.
2671  * @param[in] last_item
2672  *   Previous validated item in the pattern items.
2673  * @param[in] geneve_item
2674  *   Previous GENEVE item specification.
2675  * @param[in] dev
2676  *   Pointer to the rte_eth_dev structure.
2677  * @param[out] error
2678  *   Pointer to error structure.
2679  *
2680  * @return
2681  *   0 on success, a negative errno value otherwise and rte_errno is set.
2682  */
2683 int
2684 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2685 				   uint64_t last_item,
2686 				   const struct rte_flow_item *geneve_item,
2687 				   struct rte_eth_dev *dev,
2688 				   struct rte_flow_error *error)
2689 {
2690 	struct mlx5_priv *priv = dev->data->dev_private;
2691 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2692 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2693 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2694 	uint8_t data_max_supported =
2695 			hca_attr->max_geneve_tlv_option_data_len * 4;
2696 	struct mlx5_dev_config *config = &priv->config;
2697 	const struct rte_flow_item_geneve *geneve_spec;
2698 	const struct rte_flow_item_geneve *geneve_mask;
2699 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2700 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2701 	unsigned int i;
2702 	unsigned int data_len;
2703 	uint8_t tlv_option_len;
2704 	uint16_t optlen_m, optlen_v;
2705 	const struct rte_flow_item_geneve_opt full_mask = {
2706 		.option_class = RTE_BE16(0xffff),
2707 		.option_type = 0xff,
2708 		.option_len = 0x1f,
2709 	};
2710 
2711 	if (!mask)
2712 		mask = &rte_flow_item_geneve_opt_mask;
2713 	if (!spec)
2714 		return rte_flow_error_set
2715 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2716 			"Geneve TLV opt class/type/length must be specified");
2717 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2718 		return rte_flow_error_set
2719 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2720 			"Geneve TLV opt length exceeeds the limit (31)");
2721 	/* Check if class type and length masks are full. */
2722 	if (full_mask.option_class != mask->option_class ||
2723 	    full_mask.option_type != mask->option_type ||
2724 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2725 		return rte_flow_error_set
2726 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2727 			"Geneve TLV opt class/type/length masks must be full");
2728 	/* Check if length is supported */
2729 	if ((uint32_t)spec->option_len >
2730 			config->hca_attr.max_geneve_tlv_option_data_len)
2731 		return rte_flow_error_set
2732 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2733 			"Geneve TLV opt length not supported");
2734 	if (config->hca_attr.max_geneve_tlv_options > 1)
2735 		DRV_LOG(DEBUG,
2736 			"max_geneve_tlv_options supports more than 1 option");
2737 	/* Check GENEVE item preceding. */
2738 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2739 		return rte_flow_error_set
2740 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2741 			"Geneve opt item must be preceded with Geneve item");
2742 	geneve_spec = geneve_item->spec;
2743 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2744 					  &rte_flow_item_geneve_mask;
2745 	/* Check if GENEVE TLV option size doesn't exceed option length */
2746 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2747 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2748 		tlv_option_len = spec->option_len & mask->option_len;
2749 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2750 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2751 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2752 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2753 		if ((optlen_v & optlen_m) <= tlv_option_len)
2754 			return rte_flow_error_set
2755 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2756 				 "GENEVE TLV option length exceeds optlen");
2757 	}
2758 	/* Check if length is 0 or data is 0. */
2759 	if (spec->data == NULL || spec->option_len == 0)
2760 		return rte_flow_error_set
2761 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2762 			"Geneve TLV opt with zero data/length not supported");
2763 	/* Check not all data & mask are 0. */
2764 	data_len = spec->option_len * 4;
2765 	if (mask->data == NULL) {
2766 		for (i = 0; i < data_len; i++)
2767 			if (spec->data[i])
2768 				break;
2769 		if (i == data_len)
2770 			return rte_flow_error_set(error, ENOTSUP,
2771 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 				"Can't match on Geneve option data 0");
2773 	} else {
2774 		for (i = 0; i < data_len; i++)
2775 			if (spec->data[i] & mask->data[i])
2776 				break;
2777 		if (i == data_len)
2778 			return rte_flow_error_set(error, ENOTSUP,
2779 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2780 				"Can't match on Geneve option data and mask 0");
2781 		/* Check data mask supported. */
2782 		for (i = data_max_supported; i < data_len ; i++)
2783 			if (mask->data[i])
2784 				return rte_flow_error_set(error, ENOTSUP,
2785 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2786 					"Data mask is of unsupported size");
2787 	}
2788 	/* Check GENEVE option is supported in NIC. */
2789 	if (!config->hca_attr.geneve_tlv_opt)
2790 		return rte_flow_error_set
2791 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2792 			"Geneve TLV opt not supported");
2793 	/* Check if we already have geneve option with different type/class. */
2794 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2795 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2796 	if (geneve_opt_resource != NULL)
2797 		if (geneve_opt_resource->option_class != spec->option_class ||
2798 		    geneve_opt_resource->option_type != spec->option_type ||
2799 		    geneve_opt_resource->length != spec->option_len) {
2800 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2801 			return rte_flow_error_set(error, ENOTSUP,
2802 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2803 				"Only one Geneve TLV option supported");
2804 		}
2805 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2806 	return 0;
2807 }
2808 
2809 /**
2810  * Validate MPLS item.
2811  *
2812  * @param[in] dev
2813  *   Pointer to the rte_eth_dev structure.
2814  * @param[in] item
2815  *   Item specification.
2816  * @param[in] item_flags
2817  *   Bit-fields that holds the items detected until now.
2818  * @param[in] prev_layer
2819  *   The protocol layer indicated in previous item.
2820  * @param[out] error
2821  *   Pointer to error structure.
2822  *
2823  * @return
2824  *   0 on success, a negative errno value otherwise and rte_errno is set.
2825  */
2826 int
2827 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2828 			     const struct rte_flow_item *item __rte_unused,
2829 			     uint64_t item_flags __rte_unused,
2830 			     uint64_t prev_layer __rte_unused,
2831 			     struct rte_flow_error *error)
2832 {
2833 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2834 	const struct rte_flow_item_mpls *mask = item->mask;
2835 	struct mlx5_priv *priv = dev->data->dev_private;
2836 	int ret;
2837 
2838 	if (!priv->config.mpls_en)
2839 		return rte_flow_error_set(error, ENOTSUP,
2840 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2841 					  "MPLS not supported or"
2842 					  " disabled in firmware"
2843 					  " configuration.");
2844 	/* MPLS over IP, UDP, GRE is allowed */
2845 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2846 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2847 			    MLX5_FLOW_LAYER_GRE |
2848 			    MLX5_FLOW_LAYER_GRE_KEY)))
2849 		return rte_flow_error_set(error, EINVAL,
2850 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2851 					  "protocol filtering not compatible"
2852 					  " with MPLS layer");
2853 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2854 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2855 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2856 		return rte_flow_error_set(error, ENOTSUP,
2857 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2858 					  "multiple tunnel layers not"
2859 					  " supported");
2860 	if (!mask)
2861 		mask = &rte_flow_item_mpls_mask;
2862 	ret = mlx5_flow_item_acceptable
2863 		(item, (const uint8_t *)mask,
2864 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2865 		 sizeof(struct rte_flow_item_mpls),
2866 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2867 	if (ret < 0)
2868 		return ret;
2869 	return 0;
2870 #else
2871 	return rte_flow_error_set(error, ENOTSUP,
2872 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2873 				  "MPLS is not supported by Verbs, please"
2874 				  " update.");
2875 #endif
2876 }
2877 
2878 /**
2879  * Validate NVGRE item.
2880  *
2881  * @param[in] item
2882  *   Item specification.
2883  * @param[in] item_flags
2884  *   Bit flags to mark detected items.
2885  * @param[in] target_protocol
2886  *   The next protocol in the previous item.
2887  * @param[out] error
2888  *   Pointer to error structure.
2889  *
2890  * @return
2891  *   0 on success, a negative errno value otherwise and rte_errno is set.
2892  */
2893 int
2894 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2895 			      uint64_t item_flags,
2896 			      uint8_t target_protocol,
2897 			      struct rte_flow_error *error)
2898 {
2899 	const struct rte_flow_item_nvgre *mask = item->mask;
2900 	int ret;
2901 
2902 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2903 		return rte_flow_error_set(error, EINVAL,
2904 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2905 					  "protocol filtering not compatible"
2906 					  " with this GRE layer");
2907 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2908 		return rte_flow_error_set(error, ENOTSUP,
2909 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2910 					  "multiple tunnel layers not"
2911 					  " supported");
2912 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2913 		return rte_flow_error_set(error, ENOTSUP,
2914 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2915 					  "L3 Layer is missing");
2916 	if (!mask)
2917 		mask = &rte_flow_item_nvgre_mask;
2918 	ret = mlx5_flow_item_acceptable
2919 		(item, (const uint8_t *)mask,
2920 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2921 		 sizeof(struct rte_flow_item_nvgre),
2922 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2923 	if (ret < 0)
2924 		return ret;
2925 	return 0;
2926 }
2927 
2928 /**
2929  * Validate eCPRI item.
2930  *
2931  * @param[in] item
2932  *   Item specification.
2933  * @param[in] item_flags
2934  *   Bit-fields that holds the items detected until now.
2935  * @param[in] last_item
2936  *   Previous validated item in the pattern items.
2937  * @param[in] ether_type
2938  *   Type in the ethernet layer header (including dot1q).
2939  * @param[in] acc_mask
2940  *   Acceptable mask, if NULL default internal default mask
2941  *   will be used to check whether item fields are supported.
2942  * @param[out] error
2943  *   Pointer to error structure.
2944  *
2945  * @return
2946  *   0 on success, a negative errno value otherwise and rte_errno is set.
2947  */
2948 int
2949 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2950 			      uint64_t item_flags,
2951 			      uint64_t last_item,
2952 			      uint16_t ether_type,
2953 			      const struct rte_flow_item_ecpri *acc_mask,
2954 			      struct rte_flow_error *error)
2955 {
2956 	const struct rte_flow_item_ecpri *mask = item->mask;
2957 	const struct rte_flow_item_ecpri nic_mask = {
2958 		.hdr = {
2959 			.common = {
2960 				.u32 =
2961 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2962 					.type = 0xFF,
2963 					}).u32),
2964 			},
2965 			.dummy[0] = 0xFFFFFFFF,
2966 		},
2967 	};
2968 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2969 					MLX5_FLOW_LAYER_OUTER_VLAN);
2970 	struct rte_flow_item_ecpri mask_lo;
2971 
2972 	if (!(last_item & outer_l2_vlan) &&
2973 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2974 		return rte_flow_error_set(error, EINVAL,
2975 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2976 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2977 	if ((last_item & outer_l2_vlan) && ether_type &&
2978 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2979 		return rte_flow_error_set(error, EINVAL,
2980 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2981 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2982 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2983 		return rte_flow_error_set(error, EINVAL,
2984 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2985 					  "eCPRI with tunnel is not supported right now");
2986 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2987 		return rte_flow_error_set(error, ENOTSUP,
2988 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2989 					  "multiple L3 layers not supported");
2990 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2991 		return rte_flow_error_set(error, EINVAL,
2992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2993 					  "eCPRI cannot coexist with a TCP layer");
2994 	/* In specification, eCPRI could be over UDP layer. */
2995 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2996 		return rte_flow_error_set(error, EINVAL,
2997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2998 					  "eCPRI over UDP layer is not yet supported right now");
2999 	/* Mask for type field in common header could be zero. */
3000 	if (!mask)
3001 		mask = &rte_flow_item_ecpri_mask;
3002 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3003 	/* Input mask is in big-endian format. */
3004 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3005 		return rte_flow_error_set(error, EINVAL,
3006 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3007 					  "partial mask is not supported for protocol");
3008 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3009 		return rte_flow_error_set(error, EINVAL,
3010 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3011 					  "message header mask must be after a type mask");
3012 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3013 					 acc_mask ? (const uint8_t *)acc_mask
3014 						  : (const uint8_t *)&nic_mask,
3015 					 sizeof(struct rte_flow_item_ecpri),
3016 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3017 }
3018 
3019 /**
3020  * Release resource related QUEUE/RSS action split.
3021  *
3022  * @param dev
3023  *   Pointer to Ethernet device.
3024  * @param flow
3025  *   Flow to release id's from.
3026  */
3027 static void
3028 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3029 			     struct rte_flow *flow)
3030 {
3031 	struct mlx5_priv *priv = dev->data->dev_private;
3032 	uint32_t handle_idx;
3033 	struct mlx5_flow_handle *dev_handle;
3034 
3035 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3036 		       handle_idx, dev_handle, next)
3037 		if (dev_handle->split_flow_id)
3038 			mlx5_ipool_free(priv->sh->ipool
3039 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3040 					dev_handle->split_flow_id);
3041 }
3042 
3043 static int
3044 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3045 		   const struct rte_flow_attr *attr __rte_unused,
3046 		   const struct rte_flow_item items[] __rte_unused,
3047 		   const struct rte_flow_action actions[] __rte_unused,
3048 		   bool external __rte_unused,
3049 		   int hairpin __rte_unused,
3050 		   struct rte_flow_error *error)
3051 {
3052 	return rte_flow_error_set(error, ENOTSUP,
3053 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3054 }
3055 
3056 static struct mlx5_flow *
3057 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3058 		  const struct rte_flow_attr *attr __rte_unused,
3059 		  const struct rte_flow_item items[] __rte_unused,
3060 		  const struct rte_flow_action actions[] __rte_unused,
3061 		  struct rte_flow_error *error)
3062 {
3063 	rte_flow_error_set(error, ENOTSUP,
3064 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3065 	return NULL;
3066 }
3067 
3068 static int
3069 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3070 		    struct mlx5_flow *dev_flow __rte_unused,
3071 		    const struct rte_flow_attr *attr __rte_unused,
3072 		    const struct rte_flow_item items[] __rte_unused,
3073 		    const struct rte_flow_action actions[] __rte_unused,
3074 		    struct rte_flow_error *error)
3075 {
3076 	return rte_flow_error_set(error, ENOTSUP,
3077 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3078 }
3079 
3080 static int
3081 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3082 		struct rte_flow *flow __rte_unused,
3083 		struct rte_flow_error *error)
3084 {
3085 	return rte_flow_error_set(error, ENOTSUP,
3086 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3087 }
3088 
3089 static void
3090 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3091 		 struct rte_flow *flow __rte_unused)
3092 {
3093 }
3094 
3095 static void
3096 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3097 		  struct rte_flow *flow __rte_unused)
3098 {
3099 }
3100 
3101 static int
3102 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3103 		struct rte_flow *flow __rte_unused,
3104 		const struct rte_flow_action *actions __rte_unused,
3105 		void *data __rte_unused,
3106 		struct rte_flow_error *error)
3107 {
3108 	return rte_flow_error_set(error, ENOTSUP,
3109 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3110 }
3111 
3112 static int
3113 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3114 		      uint32_t domains __rte_unused,
3115 		      uint32_t flags __rte_unused)
3116 {
3117 	return 0;
3118 }
3119 
3120 /* Void driver to protect from null pointer reference. */
3121 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3122 	.validate = flow_null_validate,
3123 	.prepare = flow_null_prepare,
3124 	.translate = flow_null_translate,
3125 	.apply = flow_null_apply,
3126 	.remove = flow_null_remove,
3127 	.destroy = flow_null_destroy,
3128 	.query = flow_null_query,
3129 	.sync_domain = flow_null_sync_domain,
3130 };
3131 
3132 /**
3133  * Select flow driver type according to flow attributes and device
3134  * configuration.
3135  *
3136  * @param[in] dev
3137  *   Pointer to the dev structure.
3138  * @param[in] attr
3139  *   Pointer to the flow attributes.
3140  *
3141  * @return
3142  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3143  */
3144 static enum mlx5_flow_drv_type
3145 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3146 {
3147 	struct mlx5_priv *priv = dev->data->dev_private;
3148 	/* The OS can determine first a specific flow type (DV, VERBS) */
3149 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3150 
3151 	if (type != MLX5_FLOW_TYPE_MAX)
3152 		return type;
3153 	/* If no OS specific type - continue with DV/VERBS selection */
3154 	if (attr->transfer && priv->config.dv_esw_en)
3155 		type = MLX5_FLOW_TYPE_DV;
3156 	if (!attr->transfer)
3157 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3158 						 MLX5_FLOW_TYPE_VERBS;
3159 	return type;
3160 }
3161 
3162 #define flow_get_drv_ops(type) flow_drv_ops[type]
3163 
3164 /**
3165  * Flow driver validation API. This abstracts calling driver specific functions.
3166  * The type of flow driver is determined according to flow attributes.
3167  *
3168  * @param[in] dev
3169  *   Pointer to the dev structure.
3170  * @param[in] attr
3171  *   Pointer to the flow attributes.
3172  * @param[in] items
3173  *   Pointer to the list of items.
3174  * @param[in] actions
3175  *   Pointer to the list of actions.
3176  * @param[in] external
3177  *   This flow rule is created by request external to PMD.
3178  * @param[in] hairpin
3179  *   Number of hairpin TX actions, 0 means classic flow.
3180  * @param[out] error
3181  *   Pointer to the error structure.
3182  *
3183  * @return
3184  *   0 on success, a negative errno value otherwise and rte_errno is set.
3185  */
3186 static inline int
3187 flow_drv_validate(struct rte_eth_dev *dev,
3188 		  const struct rte_flow_attr *attr,
3189 		  const struct rte_flow_item items[],
3190 		  const struct rte_flow_action actions[],
3191 		  bool external, int hairpin, struct rte_flow_error *error)
3192 {
3193 	const struct mlx5_flow_driver_ops *fops;
3194 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3195 
3196 	fops = flow_get_drv_ops(type);
3197 	return fops->validate(dev, attr, items, actions, external,
3198 			      hairpin, error);
3199 }
3200 
3201 /**
3202  * Flow driver preparation API. This abstracts calling driver specific
3203  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3204  * calculates the size of memory required for device flow, allocates the memory,
3205  * initializes the device flow and returns the pointer.
3206  *
3207  * @note
3208  *   This function initializes device flow structure such as dv or verbs in
3209  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3210  *   rest. For example, adding returning device flow to flow->dev_flow list and
3211  *   setting backward reference to the flow should be done out of this function.
3212  *   layers field is not filled either.
3213  *
3214  * @param[in] dev
3215  *   Pointer to the dev structure.
3216  * @param[in] attr
3217  *   Pointer to the flow attributes.
3218  * @param[in] items
3219  *   Pointer to the list of items.
3220  * @param[in] actions
3221  *   Pointer to the list of actions.
3222  * @param[in] flow_idx
3223  *   This memory pool index to the flow.
3224  * @param[out] error
3225  *   Pointer to the error structure.
3226  *
3227  * @return
3228  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3229  */
3230 static inline struct mlx5_flow *
3231 flow_drv_prepare(struct rte_eth_dev *dev,
3232 		 const struct rte_flow *flow,
3233 		 const struct rte_flow_attr *attr,
3234 		 const struct rte_flow_item items[],
3235 		 const struct rte_flow_action actions[],
3236 		 uint32_t flow_idx,
3237 		 struct rte_flow_error *error)
3238 {
3239 	const struct mlx5_flow_driver_ops *fops;
3240 	enum mlx5_flow_drv_type type = flow->drv_type;
3241 	struct mlx5_flow *mlx5_flow = NULL;
3242 
3243 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3244 	fops = flow_get_drv_ops(type);
3245 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3246 	if (mlx5_flow)
3247 		mlx5_flow->flow_idx = flow_idx;
3248 	return mlx5_flow;
3249 }
3250 
3251 /**
3252  * Flow driver translation API. This abstracts calling driver specific
3253  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3254  * translates a generic flow into a driver flow. flow_drv_prepare() must
3255  * precede.
3256  *
3257  * @note
3258  *   dev_flow->layers could be filled as a result of parsing during translation
3259  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3260  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3261  *   flow->actions could be overwritten even though all the expanded dev_flows
3262  *   have the same actions.
3263  *
3264  * @param[in] dev
3265  *   Pointer to the rte dev structure.
3266  * @param[in, out] dev_flow
3267  *   Pointer to the mlx5 flow.
3268  * @param[in] attr
3269  *   Pointer to the flow attributes.
3270  * @param[in] items
3271  *   Pointer to the list of items.
3272  * @param[in] actions
3273  *   Pointer to the list of actions.
3274  * @param[out] error
3275  *   Pointer to the error structure.
3276  *
3277  * @return
3278  *   0 on success, a negative errno value otherwise and rte_errno is set.
3279  */
3280 static inline int
3281 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3282 		   const struct rte_flow_attr *attr,
3283 		   const struct rte_flow_item items[],
3284 		   const struct rte_flow_action actions[],
3285 		   struct rte_flow_error *error)
3286 {
3287 	const struct mlx5_flow_driver_ops *fops;
3288 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3289 
3290 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3291 	fops = flow_get_drv_ops(type);
3292 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3293 }
3294 
3295 /**
3296  * Flow driver apply API. This abstracts calling driver specific functions.
3297  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3298  * translated driver flows on to device. flow_drv_translate() must precede.
3299  *
3300  * @param[in] dev
3301  *   Pointer to Ethernet device structure.
3302  * @param[in, out] flow
3303  *   Pointer to flow structure.
3304  * @param[out] error
3305  *   Pointer to error structure.
3306  *
3307  * @return
3308  *   0 on success, a negative errno value otherwise and rte_errno is set.
3309  */
3310 static inline int
3311 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3312 	       struct rte_flow_error *error)
3313 {
3314 	const struct mlx5_flow_driver_ops *fops;
3315 	enum mlx5_flow_drv_type type = flow->drv_type;
3316 
3317 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3318 	fops = flow_get_drv_ops(type);
3319 	return fops->apply(dev, flow, error);
3320 }
3321 
3322 /**
3323  * Flow driver destroy API. This abstracts calling driver specific functions.
3324  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3325  * on device and releases resources of the flow.
3326  *
3327  * @param[in] dev
3328  *   Pointer to Ethernet device.
3329  * @param[in, out] flow
3330  *   Pointer to flow structure.
3331  */
3332 static inline void
3333 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3334 {
3335 	const struct mlx5_flow_driver_ops *fops;
3336 	enum mlx5_flow_drv_type type = flow->drv_type;
3337 
3338 	flow_mreg_split_qrss_release(dev, flow);
3339 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3340 	fops = flow_get_drv_ops(type);
3341 	fops->destroy(dev, flow);
3342 }
3343 
3344 /**
3345  * Get RSS action from the action list.
3346  *
3347  * @param[in] actions
3348  *   Pointer to the list of actions.
3349  *
3350  * @return
3351  *   Pointer to the RSS action if exist, else return NULL.
3352  */
3353 static const struct rte_flow_action_rss*
3354 flow_get_rss_action(const struct rte_flow_action actions[])
3355 {
3356 	const struct rte_flow_action_rss *rss = NULL;
3357 
3358 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3359 		switch (actions->type) {
3360 		case RTE_FLOW_ACTION_TYPE_RSS:
3361 			rss = actions->conf;
3362 			break;
3363 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3364 		{
3365 			const struct rte_flow_action_sample *sample =
3366 								actions->conf;
3367 			const struct rte_flow_action *act = sample->actions;
3368 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3369 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3370 					rss = act->conf;
3371 			break;
3372 		}
3373 		default:
3374 			break;
3375 		}
3376 	}
3377 	return rss;
3378 }
3379 
3380 /**
3381  * Get ASO age action by index.
3382  *
3383  * @param[in] dev
3384  *   Pointer to the Ethernet device structure.
3385  * @param[in] age_idx
3386  *   Index to the ASO age action.
3387  *
3388  * @return
3389  *   The specified ASO age action.
3390  */
3391 struct mlx5_aso_age_action*
3392 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3393 {
3394 	uint16_t pool_idx = age_idx & UINT16_MAX;
3395 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3396 	struct mlx5_priv *priv = dev->data->dev_private;
3397 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3398 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3399 
3400 	return &pool->actions[offset - 1];
3401 }
3402 
3403 /* maps shared action to translated non shared in some actions array */
3404 struct mlx5_translated_shared_action {
3405 	struct rte_flow_shared_action *action; /**< Shared action */
3406 	int index; /**< Index in related array of rte_flow_action */
3407 };
3408 
3409 /**
3410  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3411  * non shared action if translation possible.
3412  * This functionality used to run same execution path for both shared & non
3413  * shared actions on flow create. All necessary preparations for shared
3414  * action handling should be preformed on *shared* actions list returned
3415  * from this call.
3416  *
3417  * @param[in] dev
3418  *   Pointer to Ethernet device.
3419  * @param[in] actions
3420  *   List of actions to translate.
3421  * @param[out] shared
3422  *   List to store translated shared actions.
3423  * @param[in, out] shared_n
3424  *   Size of *shared* array. On return should be updated with number of shared
3425  *   actions retrieved from the *actions* list.
3426  * @param[out] translated_actions
3427  *   List of actions where all shared actions were translated to non shared
3428  *   if possible. NULL if no translation took place.
3429  * @param[out] error
3430  *   Pointer to the error structure.
3431  *
3432  * @return
3433  *   0 on success, a negative errno value otherwise and rte_errno is set.
3434  */
3435 static int
3436 flow_shared_actions_translate(struct rte_eth_dev *dev,
3437 			      const struct rte_flow_action actions[],
3438 			      struct mlx5_translated_shared_action *shared,
3439 			      int *shared_n,
3440 			      struct rte_flow_action **translated_actions,
3441 			      struct rte_flow_error *error)
3442 {
3443 	struct mlx5_priv *priv = dev->data->dev_private;
3444 	struct rte_flow_action *translated = NULL;
3445 	size_t actions_size;
3446 	int n;
3447 	int copied_n = 0;
3448 	struct mlx5_translated_shared_action *shared_end = NULL;
3449 
3450 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3451 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3452 			continue;
3453 		if (copied_n == *shared_n) {
3454 			return rte_flow_error_set
3455 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3456 				 NULL, "too many shared actions");
3457 		}
3458 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3459 			   sizeof(actions[n].conf));
3460 		shared[copied_n].index = n;
3461 		copied_n++;
3462 	}
3463 	n++;
3464 	*shared_n = copied_n;
3465 	if (!copied_n)
3466 		return 0;
3467 	actions_size = sizeof(struct rte_flow_action) * n;
3468 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3469 	if (!translated) {
3470 		rte_errno = ENOMEM;
3471 		return -ENOMEM;
3472 	}
3473 	memcpy(translated, actions, actions_size);
3474 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3475 		struct mlx5_shared_action_rss *shared_rss;
3476 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3477 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3478 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3479 									   - 1);
3480 
3481 		switch (type) {
3482 		case MLX5_SHARED_ACTION_TYPE_RSS:
3483 			shared_rss = mlx5_ipool_get
3484 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3485 			translated[shared->index].type =
3486 				RTE_FLOW_ACTION_TYPE_RSS;
3487 			translated[shared->index].conf =
3488 				&shared_rss->origin;
3489 			break;
3490 		case MLX5_SHARED_ACTION_TYPE_AGE:
3491 			if (priv->sh->flow_hit_aso_en) {
3492 				translated[shared->index].type =
3493 					(enum rte_flow_action_type)
3494 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3495 				translated[shared->index].conf =
3496 							 (void *)(uintptr_t)idx;
3497 				break;
3498 			}
3499 			/* Fall-through */
3500 		default:
3501 			mlx5_free(translated);
3502 			return rte_flow_error_set
3503 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3504 				 NULL, "invalid shared action type");
3505 		}
3506 	}
3507 	*translated_actions = translated;
3508 	return 0;
3509 }
3510 
3511 /**
3512  * Get Shared RSS action from the action list.
3513  *
3514  * @param[in] dev
3515  *   Pointer to Ethernet device.
3516  * @param[in] shared
3517  *   Pointer to the list of actions.
3518  * @param[in] shared_n
3519  *   Actions list length.
3520  *
3521  * @return
3522  *   The MLX5 RSS action ID if exists, otherwise return 0.
3523  */
3524 static uint32_t
3525 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3526 			   struct mlx5_translated_shared_action *shared,
3527 			   int shared_n)
3528 {
3529 	struct mlx5_translated_shared_action *shared_end;
3530 	struct mlx5_priv *priv = dev->data->dev_private;
3531 	struct mlx5_shared_action_rss *shared_rss;
3532 
3533 
3534 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3535 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3536 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3537 		uint32_t idx = act_idx &
3538 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3539 		switch (type) {
3540 		case MLX5_SHARED_ACTION_TYPE_RSS:
3541 			shared_rss = mlx5_ipool_get
3542 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3543 									   idx);
3544 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3545 					   __ATOMIC_RELAXED);
3546 			return idx;
3547 		default:
3548 			break;
3549 		}
3550 	}
3551 	return 0;
3552 }
3553 
3554 static unsigned int
3555 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3556 {
3557 	const struct rte_flow_item *item;
3558 	unsigned int has_vlan = 0;
3559 
3560 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3561 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3562 			has_vlan = 1;
3563 			break;
3564 		}
3565 	}
3566 	if (has_vlan)
3567 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3568 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3569 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3570 			       MLX5_EXPANSION_ROOT_OUTER;
3571 }
3572 
3573 /**
3574  *  Get layer flags from the prefix flow.
3575  *
3576  *  Some flows may be split to several subflows, the prefix subflow gets the
3577  *  match items and the suffix sub flow gets the actions.
3578  *  Some actions need the user defined match item flags to get the detail for
3579  *  the action.
3580  *  This function helps the suffix flow to get the item layer flags from prefix
3581  *  subflow.
3582  *
3583  * @param[in] dev_flow
3584  *   Pointer the created preifx subflow.
3585  *
3586  * @return
3587  *   The layers get from prefix subflow.
3588  */
3589 static inline uint64_t
3590 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3591 {
3592 	uint64_t layers = 0;
3593 
3594 	/*
3595 	 * Layers bits could be localization, but usually the compiler will
3596 	 * help to do the optimization work for source code.
3597 	 * If no decap actions, use the layers directly.
3598 	 */
3599 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3600 		return dev_flow->handle->layers;
3601 	/* Convert L3 layers with decap action. */
3602 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3603 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3604 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3605 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3606 	/* Convert L4 layers with decap action.  */
3607 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3608 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3609 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3610 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3611 	return layers;
3612 }
3613 
3614 /**
3615  * Get metadata split action information.
3616  *
3617  * @param[in] actions
3618  *   Pointer to the list of actions.
3619  * @param[out] qrss
3620  *   Pointer to the return pointer.
3621  * @param[out] qrss_type
3622  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3623  *   if no QUEUE/RSS is found.
3624  * @param[out] encap_idx
3625  *   Pointer to the index of the encap action if exists, otherwise the last
3626  *   action index.
3627  *
3628  * @return
3629  *   Total number of actions.
3630  */
3631 static int
3632 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3633 				       const struct rte_flow_action **qrss,
3634 				       int *encap_idx)
3635 {
3636 	const struct rte_flow_action_raw_encap *raw_encap;
3637 	int actions_n = 0;
3638 	int raw_decap_idx = -1;
3639 
3640 	*encap_idx = -1;
3641 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3642 		switch (actions->type) {
3643 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3644 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3645 			*encap_idx = actions_n;
3646 			break;
3647 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3648 			raw_decap_idx = actions_n;
3649 			break;
3650 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3651 			raw_encap = actions->conf;
3652 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3653 				*encap_idx = raw_decap_idx != -1 ?
3654 						      raw_decap_idx : actions_n;
3655 			break;
3656 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3657 		case RTE_FLOW_ACTION_TYPE_RSS:
3658 			*qrss = actions;
3659 			break;
3660 		default:
3661 			break;
3662 		}
3663 		actions_n++;
3664 	}
3665 	if (*encap_idx == -1)
3666 		*encap_idx = actions_n;
3667 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3668 	return actions_n + 1;
3669 }
3670 
3671 /**
3672  * Check meter action from the action list.
3673  *
3674  * @param[in] actions
3675  *   Pointer to the list of actions.
3676  * @param[out] mtr
3677  *   Pointer to the meter exist flag.
3678  *
3679  * @return
3680  *   Total number of actions.
3681  */
3682 static int
3683 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3684 {
3685 	int actions_n = 0;
3686 
3687 	MLX5_ASSERT(mtr);
3688 	*mtr = 0;
3689 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3690 		switch (actions->type) {
3691 		case RTE_FLOW_ACTION_TYPE_METER:
3692 			*mtr = 1;
3693 			break;
3694 		default:
3695 			break;
3696 		}
3697 		actions_n++;
3698 	}
3699 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3700 	return actions_n + 1;
3701 }
3702 
3703 /**
3704  * Check if the flow should be split due to hairpin.
3705  * The reason for the split is that in current HW we can't
3706  * support encap and push-vlan on Rx, so if a flow contains
3707  * these actions we move it to Tx.
3708  *
3709  * @param dev
3710  *   Pointer to Ethernet device.
3711  * @param[in] attr
3712  *   Flow rule attributes.
3713  * @param[in] actions
3714  *   Associated actions (list terminated by the END action).
3715  *
3716  * @return
3717  *   > 0 the number of actions and the flow should be split,
3718  *   0 when no split required.
3719  */
3720 static int
3721 flow_check_hairpin_split(struct rte_eth_dev *dev,
3722 			 const struct rte_flow_attr *attr,
3723 			 const struct rte_flow_action actions[])
3724 {
3725 	int queue_action = 0;
3726 	int action_n = 0;
3727 	int split = 0;
3728 	const struct rte_flow_action_queue *queue;
3729 	const struct rte_flow_action_rss *rss;
3730 	const struct rte_flow_action_raw_encap *raw_encap;
3731 	const struct rte_eth_hairpin_conf *conf;
3732 
3733 	if (!attr->ingress)
3734 		return 0;
3735 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3736 		switch (actions->type) {
3737 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3738 			queue = actions->conf;
3739 			if (queue == NULL)
3740 				return 0;
3741 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3742 			if (conf == NULL || conf->tx_explicit != 0)
3743 				return 0;
3744 			queue_action = 1;
3745 			action_n++;
3746 			break;
3747 		case RTE_FLOW_ACTION_TYPE_RSS:
3748 			rss = actions->conf;
3749 			if (rss == NULL || rss->queue_num == 0)
3750 				return 0;
3751 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3752 			if (conf == NULL || conf->tx_explicit != 0)
3753 				return 0;
3754 			queue_action = 1;
3755 			action_n++;
3756 			break;
3757 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3758 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3759 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3760 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3761 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3762 			split++;
3763 			action_n++;
3764 			break;
3765 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3766 			raw_encap = actions->conf;
3767 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3768 				split++;
3769 			action_n++;
3770 			break;
3771 		default:
3772 			action_n++;
3773 			break;
3774 		}
3775 	}
3776 	if (split && queue_action)
3777 		return action_n;
3778 	return 0;
3779 }
3780 
3781 /* Declare flow create/destroy prototype in advance. */
3782 static uint32_t
3783 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3784 		 const struct rte_flow_attr *attr,
3785 		 const struct rte_flow_item items[],
3786 		 const struct rte_flow_action actions[],
3787 		 bool external, struct rte_flow_error *error);
3788 
3789 static void
3790 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3791 		  uint32_t flow_idx);
3792 
3793 int
3794 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3795 		      struct mlx5_hlist_entry *entry,
3796 		      uint64_t key, void *cb_ctx __rte_unused)
3797 {
3798 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3799 		container_of(entry, typeof(*mcp_res), hlist_ent);
3800 
3801 	return mcp_res->mark_id != key;
3802 }
3803 
3804 struct mlx5_hlist_entry *
3805 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3806 		       void *cb_ctx)
3807 {
3808 	struct rte_eth_dev *dev = list->ctx;
3809 	struct mlx5_priv *priv = dev->data->dev_private;
3810 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3811 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3812 	struct rte_flow_error *error = ctx->error;
3813 	uint32_t idx = 0;
3814 	int ret;
3815 	uint32_t mark_id = key;
3816 	struct rte_flow_attr attr = {
3817 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3818 		.ingress = 1,
3819 	};
3820 	struct mlx5_rte_flow_item_tag tag_spec = {
3821 		.data = mark_id,
3822 	};
3823 	struct rte_flow_item items[] = {
3824 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3825 	};
3826 	struct rte_flow_action_mark ftag = {
3827 		.id = mark_id,
3828 	};
3829 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3830 		.dst = REG_B,
3831 		.src = REG_NON,
3832 	};
3833 	struct rte_flow_action_jump jump = {
3834 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3835 	};
3836 	struct rte_flow_action actions[] = {
3837 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3838 	};
3839 
3840 	/* Fill the register fileds in the flow. */
3841 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3842 	if (ret < 0)
3843 		return NULL;
3844 	tag_spec.id = ret;
3845 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3846 	if (ret < 0)
3847 		return NULL;
3848 	cp_mreg.src = ret;
3849 	/* Provide the full width of FLAG specific value. */
3850 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3851 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3852 	/* Build a new flow. */
3853 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3854 		items[0] = (struct rte_flow_item){
3855 			.type = (enum rte_flow_item_type)
3856 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3857 			.spec = &tag_spec,
3858 		};
3859 		items[1] = (struct rte_flow_item){
3860 			.type = RTE_FLOW_ITEM_TYPE_END,
3861 		};
3862 		actions[0] = (struct rte_flow_action){
3863 			.type = (enum rte_flow_action_type)
3864 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3865 			.conf = &ftag,
3866 		};
3867 		actions[1] = (struct rte_flow_action){
3868 			.type = (enum rte_flow_action_type)
3869 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3870 			.conf = &cp_mreg,
3871 		};
3872 		actions[2] = (struct rte_flow_action){
3873 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3874 			.conf = &jump,
3875 		};
3876 		actions[3] = (struct rte_flow_action){
3877 			.type = RTE_FLOW_ACTION_TYPE_END,
3878 		};
3879 	} else {
3880 		/* Default rule, wildcard match. */
3881 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
3882 		items[0] = (struct rte_flow_item){
3883 			.type = RTE_FLOW_ITEM_TYPE_END,
3884 		};
3885 		actions[0] = (struct rte_flow_action){
3886 			.type = (enum rte_flow_action_type)
3887 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3888 			.conf = &cp_mreg,
3889 		};
3890 		actions[1] = (struct rte_flow_action){
3891 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3892 			.conf = &jump,
3893 		};
3894 		actions[2] = (struct rte_flow_action){
3895 			.type = RTE_FLOW_ACTION_TYPE_END,
3896 		};
3897 	}
3898 	/* Build a new entry. */
3899 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3900 	if (!mcp_res) {
3901 		rte_errno = ENOMEM;
3902 		return NULL;
3903 	}
3904 	mcp_res->idx = idx;
3905 	mcp_res->mark_id = mark_id;
3906 	/*
3907 	 * The copy Flows are not included in any list. There
3908 	 * ones are referenced from other Flows and can not
3909 	 * be applied, removed, deleted in ardbitrary order
3910 	 * by list traversing.
3911 	 */
3912 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3913 					 actions, false, error);
3914 	if (!mcp_res->rix_flow) {
3915 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3916 		return NULL;
3917 	}
3918 	return &mcp_res->hlist_ent;
3919 }
3920 
3921 /**
3922  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3923  *
3924  * As mark_id is unique, if there's already a registered flow for the mark_id,
3925  * return by increasing the reference counter of the resource. Otherwise, create
3926  * the resource (mcp_res) and flow.
3927  *
3928  * Flow looks like,
3929  *   - If ingress port is ANY and reg_c[1] is mark_id,
3930  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3931  *
3932  * For default flow (zero mark_id), flow is like,
3933  *   - If ingress port is ANY,
3934  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3935  *
3936  * @param dev
3937  *   Pointer to Ethernet device.
3938  * @param mark_id
3939  *   ID of MARK action, zero means default flow for META.
3940  * @param[out] error
3941  *   Perform verbose error reporting if not NULL.
3942  *
3943  * @return
3944  *   Associated resource on success, NULL otherwise and rte_errno is set.
3945  */
3946 static struct mlx5_flow_mreg_copy_resource *
3947 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3948 			  struct rte_flow_error *error)
3949 {
3950 	struct mlx5_priv *priv = dev->data->dev_private;
3951 	struct mlx5_hlist_entry *entry;
3952 	struct mlx5_flow_cb_ctx ctx = {
3953 		.dev = dev,
3954 		.error = error,
3955 	};
3956 
3957 	/* Check if already registered. */
3958 	MLX5_ASSERT(priv->mreg_cp_tbl);
3959 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3960 	if (!entry)
3961 		return NULL;
3962 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3963 			    hlist_ent);
3964 }
3965 
3966 void
3967 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3968 {
3969 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3970 		container_of(entry, typeof(*mcp_res), hlist_ent);
3971 	struct rte_eth_dev *dev = list->ctx;
3972 	struct mlx5_priv *priv = dev->data->dev_private;
3973 
3974 	MLX5_ASSERT(mcp_res->rix_flow);
3975 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3976 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3977 }
3978 
3979 /**
3980  * Release flow in RX_CP_TBL.
3981  *
3982  * @param dev
3983  *   Pointer to Ethernet device.
3984  * @flow
3985  *   Parent flow for wich copying is provided.
3986  */
3987 static void
3988 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3989 			  struct rte_flow *flow)
3990 {
3991 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3992 	struct mlx5_priv *priv = dev->data->dev_private;
3993 
3994 	if (!flow->rix_mreg_copy)
3995 		return;
3996 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3997 				 flow->rix_mreg_copy);
3998 	if (!mcp_res || !priv->mreg_cp_tbl)
3999 		return;
4000 	MLX5_ASSERT(mcp_res->rix_flow);
4001 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4002 	flow->rix_mreg_copy = 0;
4003 }
4004 
4005 /**
4006  * Remove the default copy action from RX_CP_TBL.
4007  *
4008  * This functions is called in the mlx5_dev_start(). No thread safe
4009  * is guaranteed.
4010  *
4011  * @param dev
4012  *   Pointer to Ethernet device.
4013  */
4014 static void
4015 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4016 {
4017 	struct mlx5_hlist_entry *entry;
4018 	struct mlx5_priv *priv = dev->data->dev_private;
4019 
4020 	/* Check if default flow is registered. */
4021 	if (!priv->mreg_cp_tbl)
4022 		return;
4023 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4024 				  MLX5_DEFAULT_COPY_ID, NULL);
4025 	if (!entry)
4026 		return;
4027 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4028 }
4029 
4030 /**
4031  * Add the default copy action in in RX_CP_TBL.
4032  *
4033  * This functions is called in the mlx5_dev_start(). No thread safe
4034  * is guaranteed.
4035  *
4036  * @param dev
4037  *   Pointer to Ethernet device.
4038  * @param[out] error
4039  *   Perform verbose error reporting if not NULL.
4040  *
4041  * @return
4042  *   0 for success, negative value otherwise and rte_errno is set.
4043  */
4044 static int
4045 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4046 				  struct rte_flow_error *error)
4047 {
4048 	struct mlx5_priv *priv = dev->data->dev_private;
4049 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4050 
4051 	/* Check whether extensive metadata feature is engaged. */
4052 	if (!priv->config.dv_flow_en ||
4053 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4054 	    !mlx5_flow_ext_mreg_supported(dev) ||
4055 	    !priv->sh->dv_regc0_mask)
4056 		return 0;
4057 	/*
4058 	 * Add default mreg copy flow may be called multiple time, but
4059 	 * only be called once in stop. Avoid register it twice.
4060 	 */
4061 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4062 		return 0;
4063 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4064 	if (!mcp_res)
4065 		return -rte_errno;
4066 	return 0;
4067 }
4068 
4069 /**
4070  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4071  *
4072  * All the flow having Q/RSS action should be split by
4073  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4074  * performs the following,
4075  *   - CQE->flow_tag := reg_c[1] (MARK)
4076  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4077  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4078  * but there should be a flow per each MARK ID set by MARK action.
4079  *
4080  * For the aforementioned reason, if there's a MARK action in flow's action
4081  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4082  * the MARK ID to CQE's flow_tag like,
4083  *   - If reg_c[1] is mark_id,
4084  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4085  *
4086  * For SET_META action which stores value in reg_c[0], as the destination is
4087  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4088  * MARK ID means the default flow. The default flow looks like,
4089  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4090  *
4091  * @param dev
4092  *   Pointer to Ethernet device.
4093  * @param flow
4094  *   Pointer to flow structure.
4095  * @param[in] actions
4096  *   Pointer to the list of actions.
4097  * @param[out] error
4098  *   Perform verbose error reporting if not NULL.
4099  *
4100  * @return
4101  *   0 on success, negative value otherwise and rte_errno is set.
4102  */
4103 static int
4104 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4105 			    struct rte_flow *flow,
4106 			    const struct rte_flow_action *actions,
4107 			    struct rte_flow_error *error)
4108 {
4109 	struct mlx5_priv *priv = dev->data->dev_private;
4110 	struct mlx5_dev_config *config = &priv->config;
4111 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4112 	const struct rte_flow_action_mark *mark;
4113 
4114 	/* Check whether extensive metadata feature is engaged. */
4115 	if (!config->dv_flow_en ||
4116 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4117 	    !mlx5_flow_ext_mreg_supported(dev) ||
4118 	    !priv->sh->dv_regc0_mask)
4119 		return 0;
4120 	/* Find MARK action. */
4121 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4122 		switch (actions->type) {
4123 		case RTE_FLOW_ACTION_TYPE_FLAG:
4124 			mcp_res = flow_mreg_add_copy_action
4125 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4126 			if (!mcp_res)
4127 				return -rte_errno;
4128 			flow->rix_mreg_copy = mcp_res->idx;
4129 			return 0;
4130 		case RTE_FLOW_ACTION_TYPE_MARK:
4131 			mark = (const struct rte_flow_action_mark *)
4132 				actions->conf;
4133 			mcp_res =
4134 				flow_mreg_add_copy_action(dev, mark->id, error);
4135 			if (!mcp_res)
4136 				return -rte_errno;
4137 			flow->rix_mreg_copy = mcp_res->idx;
4138 			return 0;
4139 		default:
4140 			break;
4141 		}
4142 	}
4143 	return 0;
4144 }
4145 
4146 #define MLX5_MAX_SPLIT_ACTIONS 24
4147 #define MLX5_MAX_SPLIT_ITEMS 24
4148 
4149 /**
4150  * Split the hairpin flow.
4151  * Since HW can't support encap and push-vlan on Rx, we move these
4152  * actions to Tx.
4153  * If the count action is after the encap then we also
4154  * move the count action. in this case the count will also measure
4155  * the outer bytes.
4156  *
4157  * @param dev
4158  *   Pointer to Ethernet device.
4159  * @param[in] actions
4160  *   Associated actions (list terminated by the END action).
4161  * @param[out] actions_rx
4162  *   Rx flow actions.
4163  * @param[out] actions_tx
4164  *   Tx flow actions..
4165  * @param[out] pattern_tx
4166  *   The pattern items for the Tx flow.
4167  * @param[out] flow_id
4168  *   The flow ID connected to this flow.
4169  *
4170  * @return
4171  *   0 on success.
4172  */
4173 static int
4174 flow_hairpin_split(struct rte_eth_dev *dev,
4175 		   const struct rte_flow_action actions[],
4176 		   struct rte_flow_action actions_rx[],
4177 		   struct rte_flow_action actions_tx[],
4178 		   struct rte_flow_item pattern_tx[],
4179 		   uint32_t flow_id)
4180 {
4181 	const struct rte_flow_action_raw_encap *raw_encap;
4182 	const struct rte_flow_action_raw_decap *raw_decap;
4183 	struct mlx5_rte_flow_action_set_tag *set_tag;
4184 	struct rte_flow_action *tag_action;
4185 	struct mlx5_rte_flow_item_tag *tag_item;
4186 	struct rte_flow_item *item;
4187 	char *addr;
4188 	int encap = 0;
4189 
4190 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4191 		switch (actions->type) {
4192 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4193 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4194 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4195 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4196 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4197 			rte_memcpy(actions_tx, actions,
4198 			       sizeof(struct rte_flow_action));
4199 			actions_tx++;
4200 			break;
4201 		case RTE_FLOW_ACTION_TYPE_COUNT:
4202 			if (encap) {
4203 				rte_memcpy(actions_tx, actions,
4204 					   sizeof(struct rte_flow_action));
4205 				actions_tx++;
4206 			} else {
4207 				rte_memcpy(actions_rx, actions,
4208 					   sizeof(struct rte_flow_action));
4209 				actions_rx++;
4210 			}
4211 			break;
4212 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4213 			raw_encap = actions->conf;
4214 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4215 				memcpy(actions_tx, actions,
4216 				       sizeof(struct rte_flow_action));
4217 				actions_tx++;
4218 				encap = 1;
4219 			} else {
4220 				rte_memcpy(actions_rx, actions,
4221 					   sizeof(struct rte_flow_action));
4222 				actions_rx++;
4223 			}
4224 			break;
4225 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4226 			raw_decap = actions->conf;
4227 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4228 				memcpy(actions_tx, actions,
4229 				       sizeof(struct rte_flow_action));
4230 				actions_tx++;
4231 			} else {
4232 				rte_memcpy(actions_rx, actions,
4233 					   sizeof(struct rte_flow_action));
4234 				actions_rx++;
4235 			}
4236 			break;
4237 		default:
4238 			rte_memcpy(actions_rx, actions,
4239 				   sizeof(struct rte_flow_action));
4240 			actions_rx++;
4241 			break;
4242 		}
4243 	}
4244 	/* Add set meta action and end action for the Rx flow. */
4245 	tag_action = actions_rx;
4246 	tag_action->type = (enum rte_flow_action_type)
4247 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4248 	actions_rx++;
4249 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4250 	actions_rx++;
4251 	set_tag = (void *)actions_rx;
4252 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4253 	MLX5_ASSERT(set_tag->id > REG_NON);
4254 	set_tag->data = flow_id;
4255 	tag_action->conf = set_tag;
4256 	/* Create Tx item list. */
4257 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4258 	addr = (void *)&pattern_tx[2];
4259 	item = pattern_tx;
4260 	item->type = (enum rte_flow_item_type)
4261 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4262 	tag_item = (void *)addr;
4263 	tag_item->data = flow_id;
4264 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4265 	MLX5_ASSERT(set_tag->id > REG_NON);
4266 	item->spec = tag_item;
4267 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4268 	tag_item = (void *)addr;
4269 	tag_item->data = UINT32_MAX;
4270 	tag_item->id = UINT16_MAX;
4271 	item->mask = tag_item;
4272 	item->last = NULL;
4273 	item++;
4274 	item->type = RTE_FLOW_ITEM_TYPE_END;
4275 	return 0;
4276 }
4277 
4278 /**
4279  * The last stage of splitting chain, just creates the subflow
4280  * without any modification.
4281  *
4282  * @param[in] dev
4283  *   Pointer to Ethernet device.
4284  * @param[in] flow
4285  *   Parent flow structure pointer.
4286  * @param[in, out] sub_flow
4287  *   Pointer to return the created subflow, may be NULL.
4288  * @param[in] attr
4289  *   Flow rule attributes.
4290  * @param[in] items
4291  *   Pattern specification (list terminated by the END pattern item).
4292  * @param[in] actions
4293  *   Associated actions (list terminated by the END action).
4294  * @param[in] flow_split_info
4295  *   Pointer to flow split info structure.
4296  * @param[out] error
4297  *   Perform verbose error reporting if not NULL.
4298  * @return
4299  *   0 on success, negative value otherwise
4300  */
4301 static int
4302 flow_create_split_inner(struct rte_eth_dev *dev,
4303 			struct rte_flow *flow,
4304 			struct mlx5_flow **sub_flow,
4305 			const struct rte_flow_attr *attr,
4306 			const struct rte_flow_item items[],
4307 			const struct rte_flow_action actions[],
4308 			struct mlx5_flow_split_info *flow_split_info,
4309 			struct rte_flow_error *error)
4310 {
4311 	struct mlx5_flow *dev_flow;
4312 
4313 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4314 				    flow_split_info->flow_idx, error);
4315 	if (!dev_flow)
4316 		return -rte_errno;
4317 	dev_flow->flow = flow;
4318 	dev_flow->external = flow_split_info->external;
4319 	dev_flow->skip_scale = flow_split_info->skip_scale;
4320 	/* Subflow object was created, we must include one in the list. */
4321 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4322 		      dev_flow->handle, next);
4323 	/*
4324 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4325 	 * flow may need some user defined item layer flags, and pass the
4326 	 * Metadate rxq mark flag to suffix flow as well.
4327 	 */
4328 	if (flow_split_info->prefix_layers)
4329 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4330 	if (flow_split_info->prefix_mark)
4331 		dev_flow->handle->mark = 1;
4332 	if (sub_flow)
4333 		*sub_flow = dev_flow;
4334 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4335 }
4336 
4337 /**
4338  * Split the meter flow.
4339  *
4340  * As meter flow will split to three sub flow, other than meter
4341  * action, the other actions make sense to only meter accepts
4342  * the packet. If it need to be dropped, no other additional
4343  * actions should be take.
4344  *
4345  * One kind of special action which decapsulates the L3 tunnel
4346  * header will be in the prefix sub flow, as not to take the
4347  * L3 tunnel header into account.
4348  *
4349  * @param dev
4350  *   Pointer to Ethernet device.
4351  * @param[in] items
4352  *   Pattern specification (list terminated by the END pattern item).
4353  * @param[out] sfx_items
4354  *   Suffix flow match items (list terminated by the END pattern item).
4355  * @param[in] actions
4356  *   Associated actions (list terminated by the END action).
4357  * @param[out] actions_sfx
4358  *   Suffix flow actions.
4359  * @param[out] actions_pre
4360  *   Prefix flow actions.
4361  * @param[out] pattern_sfx
4362  *   The pattern items for the suffix flow.
4363  * @param[out] tag_sfx
4364  *   Pointer to suffix flow tag.
4365  *
4366  * @return
4367  *   0 on success.
4368  */
4369 static int
4370 flow_meter_split_prep(struct rte_eth_dev *dev,
4371 		 const struct rte_flow_item items[],
4372 		 struct rte_flow_item sfx_items[],
4373 		 const struct rte_flow_action actions[],
4374 		 struct rte_flow_action actions_sfx[],
4375 		 struct rte_flow_action actions_pre[])
4376 {
4377 	struct mlx5_priv *priv = dev->data->dev_private;
4378 	struct rte_flow_action *tag_action = NULL;
4379 	struct rte_flow_item *tag_item;
4380 	struct mlx5_rte_flow_action_set_tag *set_tag;
4381 	struct rte_flow_error error;
4382 	const struct rte_flow_action_raw_encap *raw_encap;
4383 	const struct rte_flow_action_raw_decap *raw_decap;
4384 	struct mlx5_rte_flow_item_tag *tag_spec;
4385 	struct mlx5_rte_flow_item_tag *tag_mask;
4386 	uint32_t tag_id = 0;
4387 	bool copy_vlan = false;
4388 
4389 	/* Prepare the actions for prefix and suffix flow. */
4390 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4391 		struct rte_flow_action **action_cur = NULL;
4392 
4393 		switch (actions->type) {
4394 		case RTE_FLOW_ACTION_TYPE_METER:
4395 			/* Add the extra tag action first. */
4396 			tag_action = actions_pre;
4397 			tag_action->type = (enum rte_flow_action_type)
4398 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4399 			actions_pre++;
4400 			action_cur = &actions_pre;
4401 			break;
4402 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4403 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4404 			action_cur = &actions_pre;
4405 			break;
4406 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4407 			raw_encap = actions->conf;
4408 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4409 				action_cur = &actions_pre;
4410 			break;
4411 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4412 			raw_decap = actions->conf;
4413 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4414 				action_cur = &actions_pre;
4415 			break;
4416 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4417 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4418 			copy_vlan = true;
4419 			break;
4420 		default:
4421 			break;
4422 		}
4423 		if (!action_cur)
4424 			action_cur = &actions_sfx;
4425 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4426 		(*action_cur)++;
4427 	}
4428 	/* Add end action to the actions. */
4429 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4430 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4431 	actions_pre++;
4432 	/* Set the tag. */
4433 	set_tag = (void *)actions_pre;
4434 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4435 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4436 			  &tag_id);
4437 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4438 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4439 			dev->data->port_id);
4440 		mlx5_ipool_free(priv->sh->ipool
4441 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4442 		return 0;
4443 	} else if (!tag_id) {
4444 		return 0;
4445 	}
4446 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4447 	assert(tag_action);
4448 	tag_action->conf = set_tag;
4449 	/* Prepare the suffix subflow items. */
4450 	tag_item = sfx_items++;
4451 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4452 		int item_type = items->type;
4453 
4454 		switch (item_type) {
4455 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4456 			memcpy(sfx_items, items, sizeof(*sfx_items));
4457 			sfx_items++;
4458 			break;
4459 		case RTE_FLOW_ITEM_TYPE_VLAN:
4460 			if (copy_vlan) {
4461 				memcpy(sfx_items, items, sizeof(*sfx_items));
4462 				/*
4463 				 * Convert to internal match item, it is used
4464 				 * for vlan push and set vid.
4465 				 */
4466 				sfx_items->type = (enum rte_flow_item_type)
4467 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4468 				sfx_items++;
4469 			}
4470 			break;
4471 		default:
4472 			break;
4473 		}
4474 	}
4475 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4476 	sfx_items++;
4477 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4478 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4479 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4480 	tag_mask = tag_spec + 1;
4481 	tag_mask->data = 0xffffff00;
4482 	tag_item->type = (enum rte_flow_item_type)
4483 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4484 	tag_item->spec = tag_spec;
4485 	tag_item->last = NULL;
4486 	tag_item->mask = tag_mask;
4487 	return tag_id;
4488 }
4489 
4490 /**
4491  * Split action list having QUEUE/RSS for metadata register copy.
4492  *
4493  * Once Q/RSS action is detected in user's action list, the flow action
4494  * should be split in order to copy metadata registers, which will happen in
4495  * RX_CP_TBL like,
4496  *   - CQE->flow_tag := reg_c[1] (MARK)
4497  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4498  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4499  * This is because the last action of each flow must be a terminal action
4500  * (QUEUE, RSS or DROP).
4501  *
4502  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4503  * stored and kept in the mlx5_flow structure per each sub_flow.
4504  *
4505  * The Q/RSS action is replaced with,
4506  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4507  * And the following JUMP action is added at the end,
4508  *   - JUMP, to RX_CP_TBL.
4509  *
4510  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4511  * flow_create_split_metadata() routine. The flow will look like,
4512  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4513  *
4514  * @param dev
4515  *   Pointer to Ethernet device.
4516  * @param[out] split_actions
4517  *   Pointer to store split actions to jump to CP_TBL.
4518  * @param[in] actions
4519  *   Pointer to the list of original flow actions.
4520  * @param[in] qrss
4521  *   Pointer to the Q/RSS action.
4522  * @param[in] actions_n
4523  *   Number of original actions.
4524  * @param[out] error
4525  *   Perform verbose error reporting if not NULL.
4526  *
4527  * @return
4528  *   non-zero unique flow_id on success, otherwise 0 and
4529  *   error/rte_error are set.
4530  */
4531 static uint32_t
4532 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4533 			  struct rte_flow_action *split_actions,
4534 			  const struct rte_flow_action *actions,
4535 			  const struct rte_flow_action *qrss,
4536 			  int actions_n, struct rte_flow_error *error)
4537 {
4538 	struct mlx5_priv *priv = dev->data->dev_private;
4539 	struct mlx5_rte_flow_action_set_tag *set_tag;
4540 	struct rte_flow_action_jump *jump;
4541 	const int qrss_idx = qrss - actions;
4542 	uint32_t flow_id = 0;
4543 	int ret = 0;
4544 
4545 	/*
4546 	 * Given actions will be split
4547 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4548 	 * - Add jump to mreg CP_TBL.
4549 	 * As a result, there will be one more action.
4550 	 */
4551 	++actions_n;
4552 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4553 	set_tag = (void *)(split_actions + actions_n);
4554 	/*
4555 	 * If tag action is not set to void(it means we are not the meter
4556 	 * suffix flow), add the tag action. Since meter suffix flow already
4557 	 * has the tag added.
4558 	 */
4559 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4560 		/*
4561 		 * Allocate the new subflow ID. This one is unique within
4562 		 * device and not shared with representors. Otherwise,
4563 		 * we would have to resolve multi-thread access synch
4564 		 * issue. Each flow on the shared device is appended
4565 		 * with source vport identifier, so the resulting
4566 		 * flows will be unique in the shared (by master and
4567 		 * representors) domain even if they have coinciding
4568 		 * IDs.
4569 		 */
4570 		mlx5_ipool_malloc(priv->sh->ipool
4571 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4572 		if (!flow_id)
4573 			return rte_flow_error_set(error, ENOMEM,
4574 						  RTE_FLOW_ERROR_TYPE_ACTION,
4575 						  NULL, "can't allocate id "
4576 						  "for split Q/RSS subflow");
4577 		/* Internal SET_TAG action to set flow ID. */
4578 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4579 			.data = flow_id,
4580 		};
4581 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4582 		if (ret < 0)
4583 			return ret;
4584 		set_tag->id = ret;
4585 		/* Construct new actions array. */
4586 		/* Replace QUEUE/RSS action. */
4587 		split_actions[qrss_idx] = (struct rte_flow_action){
4588 			.type = (enum rte_flow_action_type)
4589 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4590 			.conf = set_tag,
4591 		};
4592 	}
4593 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4594 	jump = (void *)(set_tag + 1);
4595 	*jump = (struct rte_flow_action_jump){
4596 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4597 	};
4598 	split_actions[actions_n - 2] = (struct rte_flow_action){
4599 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4600 		.conf = jump,
4601 	};
4602 	split_actions[actions_n - 1] = (struct rte_flow_action){
4603 		.type = RTE_FLOW_ACTION_TYPE_END,
4604 	};
4605 	return flow_id;
4606 }
4607 
4608 /**
4609  * Extend the given action list for Tx metadata copy.
4610  *
4611  * Copy the given action list to the ext_actions and add flow metadata register
4612  * copy action in order to copy reg_a set by WQE to reg_c[0].
4613  *
4614  * @param[out] ext_actions
4615  *   Pointer to the extended action list.
4616  * @param[in] actions
4617  *   Pointer to the list of actions.
4618  * @param[in] actions_n
4619  *   Number of actions in the list.
4620  * @param[out] error
4621  *   Perform verbose error reporting if not NULL.
4622  * @param[in] encap_idx
4623  *   The encap action inndex.
4624  *
4625  * @return
4626  *   0 on success, negative value otherwise
4627  */
4628 static int
4629 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4630 		       struct rte_flow_action *ext_actions,
4631 		       const struct rte_flow_action *actions,
4632 		       int actions_n, struct rte_flow_error *error,
4633 		       int encap_idx)
4634 {
4635 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4636 		(struct mlx5_flow_action_copy_mreg *)
4637 			(ext_actions + actions_n + 1);
4638 	int ret;
4639 
4640 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4641 	if (ret < 0)
4642 		return ret;
4643 	cp_mreg->dst = ret;
4644 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4645 	if (ret < 0)
4646 		return ret;
4647 	cp_mreg->src = ret;
4648 	if (encap_idx != 0)
4649 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4650 	if (encap_idx == actions_n - 1) {
4651 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4652 			.type = (enum rte_flow_action_type)
4653 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4654 			.conf = cp_mreg,
4655 		};
4656 		ext_actions[actions_n] = (struct rte_flow_action){
4657 			.type = RTE_FLOW_ACTION_TYPE_END,
4658 		};
4659 	} else {
4660 		ext_actions[encap_idx] = (struct rte_flow_action){
4661 			.type = (enum rte_flow_action_type)
4662 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4663 			.conf = cp_mreg,
4664 		};
4665 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4666 				sizeof(*ext_actions) * (actions_n - encap_idx));
4667 	}
4668 	return 0;
4669 }
4670 
4671 /**
4672  * Check the match action from the action list.
4673  *
4674  * @param[in] actions
4675  *   Pointer to the list of actions.
4676  * @param[in] attr
4677  *   Flow rule attributes.
4678  * @param[in] action
4679  *   The action to be check if exist.
4680  * @param[out] match_action_pos
4681  *   Pointer to the position of the matched action if exists, otherwise is -1.
4682  * @param[out] qrss_action_pos
4683  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4684  * @param[out] modify_after_mirror
4685  *   Pointer to the flag of modify action after FDB mirroring.
4686  *
4687  * @return
4688  *   > 0 the total number of actions.
4689  *   0 if not found match action in action list.
4690  */
4691 static int
4692 flow_check_match_action(const struct rte_flow_action actions[],
4693 			const struct rte_flow_attr *attr,
4694 			enum rte_flow_action_type action,
4695 			int *match_action_pos, int *qrss_action_pos,
4696 			int *modify_after_mirror)
4697 {
4698 	const struct rte_flow_action_sample *sample;
4699 	int actions_n = 0;
4700 	uint32_t ratio = 0;
4701 	int sub_type = 0;
4702 	int flag = 0;
4703 	int fdb_mirror = 0;
4704 
4705 	*match_action_pos = -1;
4706 	*qrss_action_pos = -1;
4707 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4708 		if (actions->type == action) {
4709 			flag = 1;
4710 			*match_action_pos = actions_n;
4711 		}
4712 		switch (actions->type) {
4713 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4714 		case RTE_FLOW_ACTION_TYPE_RSS:
4715 			*qrss_action_pos = actions_n;
4716 			break;
4717 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
4718 			sample = actions->conf;
4719 			ratio = sample->ratio;
4720 			sub_type = ((const struct rte_flow_action *)
4721 					(sample->actions))->type;
4722 			if (ratio == 1 && attr->transfer)
4723 				fdb_mirror = 1;
4724 			break;
4725 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4726 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4727 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4728 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4729 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4730 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4731 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4732 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4733 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4734 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
4735 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4736 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4737 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4738 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4739 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4740 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4741 		case RTE_FLOW_ACTION_TYPE_FLAG:
4742 		case RTE_FLOW_ACTION_TYPE_MARK:
4743 		case RTE_FLOW_ACTION_TYPE_SET_META:
4744 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
4745 			if (fdb_mirror)
4746 				*modify_after_mirror = 1;
4747 			break;
4748 		default:
4749 			break;
4750 		}
4751 		actions_n++;
4752 	}
4753 	if (flag && fdb_mirror && !*modify_after_mirror) {
4754 		/* FDB mirroring uses the destination array to implement
4755 		 * instead of FLOW_SAMPLER object.
4756 		 */
4757 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4758 			flag = 0;
4759 	}
4760 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4761 	return flag ? actions_n + 1 : 0;
4762 }
4763 
4764 #define SAMPLE_SUFFIX_ITEM 2
4765 
4766 /**
4767  * Split the sample flow.
4768  *
4769  * As sample flow will split to two sub flow, sample flow with
4770  * sample action, the other actions will move to new suffix flow.
4771  *
4772  * Also add unique tag id with tag action in the sample flow,
4773  * the same tag id will be as match in the suffix flow.
4774  *
4775  * @param dev
4776  *   Pointer to Ethernet device.
4777  * @param[in] add_tag
4778  *   Add extra tag action flag.
4779  * @param[out] sfx_items
4780  *   Suffix flow match items (list terminated by the END pattern item).
4781  * @param[in] actions
4782  *   Associated actions (list terminated by the END action).
4783  * @param[out] actions_sfx
4784  *   Suffix flow actions.
4785  * @param[out] actions_pre
4786  *   Prefix flow actions.
4787  * @param[in] actions_n
4788  *  The total number of actions.
4789  * @param[in] sample_action_pos
4790  *   The sample action position.
4791  * @param[in] qrss_action_pos
4792  *   The Queue/RSS action position.
4793  * @param[in] jump_table
4794  *   Add extra jump action flag.
4795  * @param[out] error
4796  *   Perform verbose error reporting if not NULL.
4797  *
4798  * @return
4799  *   0 on success, or unique flow_id, a negative errno value
4800  *   otherwise and rte_errno is set.
4801  */
4802 static int
4803 flow_sample_split_prep(struct rte_eth_dev *dev,
4804 		       int add_tag,
4805 		       struct rte_flow_item sfx_items[],
4806 		       const struct rte_flow_action actions[],
4807 		       struct rte_flow_action actions_sfx[],
4808 		       struct rte_flow_action actions_pre[],
4809 		       int actions_n,
4810 		       int sample_action_pos,
4811 		       int qrss_action_pos,
4812 		       int jump_table,
4813 		       struct rte_flow_error *error)
4814 {
4815 	struct mlx5_priv *priv = dev->data->dev_private;
4816 	struct mlx5_rte_flow_action_set_tag *set_tag;
4817 	struct mlx5_rte_flow_item_tag *tag_spec;
4818 	struct mlx5_rte_flow_item_tag *tag_mask;
4819 	struct rte_flow_action_jump *jump_action;
4820 	uint32_t tag_id = 0;
4821 	int index;
4822 	int append_index = 0;
4823 	int ret;
4824 
4825 	if (sample_action_pos < 0)
4826 		return rte_flow_error_set(error, EINVAL,
4827 					  RTE_FLOW_ERROR_TYPE_ACTION,
4828 					  NULL, "invalid position of sample "
4829 					  "action in list");
4830 	/* Prepare the actions for prefix and suffix flow. */
4831 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4832 		index = qrss_action_pos;
4833 		/* Put the preceding the Queue/RSS action into prefix flow. */
4834 		if (index != 0)
4835 			memcpy(actions_pre, actions,
4836 			       sizeof(struct rte_flow_action) * index);
4837 		/* Put others preceding the sample action into prefix flow. */
4838 		if (sample_action_pos > index + 1)
4839 			memcpy(actions_pre + index, actions + index + 1,
4840 			       sizeof(struct rte_flow_action) *
4841 			       (sample_action_pos - index - 1));
4842 		index = sample_action_pos - 1;
4843 		/* Put Queue/RSS action into Suffix flow. */
4844 		memcpy(actions_sfx, actions + qrss_action_pos,
4845 		       sizeof(struct rte_flow_action));
4846 		actions_sfx++;
4847 	} else {
4848 		index = sample_action_pos;
4849 		if (index != 0)
4850 			memcpy(actions_pre, actions,
4851 			       sizeof(struct rte_flow_action) * index);
4852 	}
4853 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4854 	 * For CX6DX and above, metadata registers Cx preserve their value,
4855 	 * add an extra tag action for NIC-RX and E-Switch Domain.
4856 	 */
4857 	if (add_tag) {
4858 		/* Prepare the prefix tag action. */
4859 		append_index++;
4860 		set_tag = (void *)(actions_pre + actions_n + append_index);
4861 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4862 		if (ret < 0)
4863 			return ret;
4864 		set_tag->id = ret;
4865 		mlx5_ipool_malloc(priv->sh->ipool
4866 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4867 		set_tag->data = tag_id;
4868 		/* Prepare the suffix subflow items. */
4869 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4870 		tag_spec->data = tag_id;
4871 		tag_spec->id = set_tag->id;
4872 		tag_mask = tag_spec + 1;
4873 		tag_mask->data = UINT32_MAX;
4874 		sfx_items[0] = (struct rte_flow_item){
4875 			.type = (enum rte_flow_item_type)
4876 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4877 			.spec = tag_spec,
4878 			.last = NULL,
4879 			.mask = tag_mask,
4880 		};
4881 		sfx_items[1] = (struct rte_flow_item){
4882 			.type = (enum rte_flow_item_type)
4883 				RTE_FLOW_ITEM_TYPE_END,
4884 		};
4885 		/* Prepare the tag action in prefix subflow. */
4886 		actions_pre[index++] =
4887 			(struct rte_flow_action){
4888 			.type = (enum rte_flow_action_type)
4889 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4890 			.conf = set_tag,
4891 		};
4892 	}
4893 	memcpy(actions_pre + index, actions + sample_action_pos,
4894 	       sizeof(struct rte_flow_action));
4895 	index += 1;
4896 	/* For the modify action after the sample action in E-Switch mirroring,
4897 	 * Add the extra jump action in prefix subflow and jump into the next
4898 	 * table, then do the modify action in the new table.
4899 	 */
4900 	if (jump_table) {
4901 		/* Prepare the prefix jump action. */
4902 		append_index++;
4903 		jump_action = (void *)(actions_pre + actions_n + append_index);
4904 		jump_action->group = jump_table;
4905 		actions_pre[index++] =
4906 			(struct rte_flow_action){
4907 			.type = (enum rte_flow_action_type)
4908 				RTE_FLOW_ACTION_TYPE_JUMP,
4909 			.conf = jump_action,
4910 		};
4911 	}
4912 	actions_pre[index] = (struct rte_flow_action){
4913 		.type = (enum rte_flow_action_type)
4914 			RTE_FLOW_ACTION_TYPE_END,
4915 	};
4916 	/* Put the actions after sample into Suffix flow. */
4917 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4918 	       sizeof(struct rte_flow_action) *
4919 	       (actions_n - sample_action_pos - 1));
4920 	return tag_id;
4921 }
4922 
4923 /**
4924  * The splitting for metadata feature.
4925  *
4926  * - Q/RSS action on NIC Rx should be split in order to pass by
4927  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4928  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4929  *
4930  * - All the actions on NIC Tx should have a mreg copy action to
4931  *   copy reg_a from WQE to reg_c[0].
4932  *
4933  * @param dev
4934  *   Pointer to Ethernet device.
4935  * @param[in] flow
4936  *   Parent flow structure pointer.
4937  * @param[in] attr
4938  *   Flow rule attributes.
4939  * @param[in] items
4940  *   Pattern specification (list terminated by the END pattern item).
4941  * @param[in] actions
4942  *   Associated actions (list terminated by the END action).
4943  * @param[in] flow_split_info
4944  *   Pointer to flow split info structure.
4945  * @param[out] error
4946  *   Perform verbose error reporting if not NULL.
4947  * @return
4948  *   0 on success, negative value otherwise
4949  */
4950 static int
4951 flow_create_split_metadata(struct rte_eth_dev *dev,
4952 			   struct rte_flow *flow,
4953 			   const struct rte_flow_attr *attr,
4954 			   const struct rte_flow_item items[],
4955 			   const struct rte_flow_action actions[],
4956 			   struct mlx5_flow_split_info *flow_split_info,
4957 			   struct rte_flow_error *error)
4958 {
4959 	struct mlx5_priv *priv = dev->data->dev_private;
4960 	struct mlx5_dev_config *config = &priv->config;
4961 	const struct rte_flow_action *qrss = NULL;
4962 	struct rte_flow_action *ext_actions = NULL;
4963 	struct mlx5_flow *dev_flow = NULL;
4964 	uint32_t qrss_id = 0;
4965 	int mtr_sfx = 0;
4966 	size_t act_size;
4967 	int actions_n;
4968 	int encap_idx;
4969 	int ret;
4970 
4971 	/* Check whether extensive metadata feature is engaged. */
4972 	if (!config->dv_flow_en ||
4973 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4974 	    !mlx5_flow_ext_mreg_supported(dev))
4975 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4976 					       actions, flow_split_info, error);
4977 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4978 							   &encap_idx);
4979 	if (qrss) {
4980 		/* Exclude hairpin flows from splitting. */
4981 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4982 			const struct rte_flow_action_queue *queue;
4983 
4984 			queue = qrss->conf;
4985 			if (mlx5_rxq_get_type(dev, queue->index) ==
4986 			    MLX5_RXQ_TYPE_HAIRPIN)
4987 				qrss = NULL;
4988 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4989 			const struct rte_flow_action_rss *rss;
4990 
4991 			rss = qrss->conf;
4992 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4993 			    MLX5_RXQ_TYPE_HAIRPIN)
4994 				qrss = NULL;
4995 		}
4996 	}
4997 	if (qrss) {
4998 		/* Check if it is in meter suffix table. */
4999 		mtr_sfx = attr->group == (attr->transfer ?
5000 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5001 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5002 		/*
5003 		 * Q/RSS action on NIC Rx should be split in order to pass by
5004 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5005 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5006 		 */
5007 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5008 			   sizeof(struct rte_flow_action_set_tag) +
5009 			   sizeof(struct rte_flow_action_jump);
5010 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5011 					  SOCKET_ID_ANY);
5012 		if (!ext_actions)
5013 			return rte_flow_error_set(error, ENOMEM,
5014 						  RTE_FLOW_ERROR_TYPE_ACTION,
5015 						  NULL, "no memory to split "
5016 						  "metadata flow");
5017 		/*
5018 		 * If we are the suffix flow of meter, tag already exist.
5019 		 * Set the tag action to void.
5020 		 */
5021 		if (mtr_sfx)
5022 			ext_actions[qrss - actions].type =
5023 						RTE_FLOW_ACTION_TYPE_VOID;
5024 		else
5025 			ext_actions[qrss - actions].type =
5026 						(enum rte_flow_action_type)
5027 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5028 		/*
5029 		 * Create the new actions list with removed Q/RSS action
5030 		 * and appended set tag and jump to register copy table
5031 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5032 		 * in advance, because it is needed for set tag action.
5033 		 */
5034 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5035 						    qrss, actions_n, error);
5036 		if (!mtr_sfx && !qrss_id) {
5037 			ret = -rte_errno;
5038 			goto exit;
5039 		}
5040 	} else if (attr->egress && !attr->transfer) {
5041 		/*
5042 		 * All the actions on NIC Tx should have a metadata register
5043 		 * copy action to copy reg_a from WQE to reg_c[meta]
5044 		 */
5045 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5046 			   sizeof(struct mlx5_flow_action_copy_mreg);
5047 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5048 					  SOCKET_ID_ANY);
5049 		if (!ext_actions)
5050 			return rte_flow_error_set(error, ENOMEM,
5051 						  RTE_FLOW_ERROR_TYPE_ACTION,
5052 						  NULL, "no memory to split "
5053 						  "metadata flow");
5054 		/* Create the action list appended with copy register. */
5055 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5056 					     actions_n, error, encap_idx);
5057 		if (ret < 0)
5058 			goto exit;
5059 	}
5060 	/* Add the unmodified original or prefix subflow. */
5061 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5062 				      items, ext_actions ? ext_actions :
5063 				      actions, flow_split_info, error);
5064 	if (ret < 0)
5065 		goto exit;
5066 	MLX5_ASSERT(dev_flow);
5067 	if (qrss) {
5068 		const struct rte_flow_attr q_attr = {
5069 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5070 			.ingress = 1,
5071 		};
5072 		/* Internal PMD action to set register. */
5073 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5074 			.data = qrss_id,
5075 			.id = REG_NON,
5076 		};
5077 		struct rte_flow_item q_items[] = {
5078 			{
5079 				.type = (enum rte_flow_item_type)
5080 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5081 				.spec = &q_tag_spec,
5082 				.last = NULL,
5083 				.mask = NULL,
5084 			},
5085 			{
5086 				.type = RTE_FLOW_ITEM_TYPE_END,
5087 			},
5088 		};
5089 		struct rte_flow_action q_actions[] = {
5090 			{
5091 				.type = qrss->type,
5092 				.conf = qrss->conf,
5093 			},
5094 			{
5095 				.type = RTE_FLOW_ACTION_TYPE_END,
5096 			},
5097 		};
5098 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5099 
5100 		/*
5101 		 * Configure the tag item only if there is no meter subflow.
5102 		 * Since tag is already marked in the meter suffix subflow
5103 		 * we can just use the meter suffix items as is.
5104 		 */
5105 		if (qrss_id) {
5106 			/* Not meter subflow. */
5107 			MLX5_ASSERT(!mtr_sfx);
5108 			/*
5109 			 * Put unique id in prefix flow due to it is destroyed
5110 			 * after suffix flow and id will be freed after there
5111 			 * is no actual flows with this id and identifier
5112 			 * reallocation becomes possible (for example, for
5113 			 * other flows in other threads).
5114 			 */
5115 			dev_flow->handle->split_flow_id = qrss_id;
5116 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5117 						   error);
5118 			if (ret < 0)
5119 				goto exit;
5120 			q_tag_spec.id = ret;
5121 		}
5122 		dev_flow = NULL;
5123 		/* Add suffix subflow to execute Q/RSS. */
5124 		flow_split_info->prefix_layers = layers;
5125 		flow_split_info->prefix_mark = 0;
5126 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5127 					      &q_attr, mtr_sfx ? items :
5128 					      q_items, q_actions,
5129 					      flow_split_info, error);
5130 		if (ret < 0)
5131 			goto exit;
5132 		/* qrss ID should be freed if failed. */
5133 		qrss_id = 0;
5134 		MLX5_ASSERT(dev_flow);
5135 	}
5136 
5137 exit:
5138 	/*
5139 	 * We do not destroy the partially created sub_flows in case of error.
5140 	 * These ones are included into parent flow list and will be destroyed
5141 	 * by flow_drv_destroy.
5142 	 */
5143 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5144 			qrss_id);
5145 	mlx5_free(ext_actions);
5146 	return ret;
5147 }
5148 
5149 /**
5150  * The splitting for meter feature.
5151  *
5152  * - The meter flow will be split to two flows as prefix and
5153  *   suffix flow. The packets make sense only it pass the prefix
5154  *   meter action.
5155  *
5156  * - Reg_C_5 is used for the packet to match betweend prefix and
5157  *   suffix flow.
5158  *
5159  * @param dev
5160  *   Pointer to Ethernet device.
5161  * @param[in] flow
5162  *   Parent flow structure pointer.
5163  * @param[in] attr
5164  *   Flow rule attributes.
5165  * @param[in] items
5166  *   Pattern specification (list terminated by the END pattern item).
5167  * @param[in] actions
5168  *   Associated actions (list terminated by the END action).
5169  * @param[in] flow_split_info
5170  *   Pointer to flow split info structure.
5171  * @param[out] error
5172  *   Perform verbose error reporting if not NULL.
5173  * @return
5174  *   0 on success, negative value otherwise
5175  */
5176 static int
5177 flow_create_split_meter(struct rte_eth_dev *dev,
5178 			struct rte_flow *flow,
5179 			const struct rte_flow_attr *attr,
5180 			const struct rte_flow_item items[],
5181 			const struct rte_flow_action actions[],
5182 			struct mlx5_flow_split_info *flow_split_info,
5183 			struct rte_flow_error *error)
5184 {
5185 	struct mlx5_priv *priv = dev->data->dev_private;
5186 	struct rte_flow_action *sfx_actions = NULL;
5187 	struct rte_flow_action *pre_actions = NULL;
5188 	struct rte_flow_item *sfx_items = NULL;
5189 	struct mlx5_flow *dev_flow = NULL;
5190 	struct rte_flow_attr sfx_attr = *attr;
5191 	uint32_t mtr = 0;
5192 	uint32_t mtr_tag_id = 0;
5193 	size_t act_size;
5194 	size_t item_size;
5195 	int actions_n = 0;
5196 	int ret;
5197 
5198 	if (priv->mtr_en)
5199 		actions_n = flow_check_meter_action(actions, &mtr);
5200 	if (mtr) {
5201 		/* The five prefix actions: meter, decap, encap, tag, end. */
5202 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5203 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5204 		/* tag, vlan, port id, end. */
5205 #define METER_SUFFIX_ITEM 4
5206 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5207 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5208 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5209 					  0, SOCKET_ID_ANY);
5210 		if (!sfx_actions)
5211 			return rte_flow_error_set(error, ENOMEM,
5212 						  RTE_FLOW_ERROR_TYPE_ACTION,
5213 						  NULL, "no memory to split "
5214 						  "meter flow");
5215 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5216 			     act_size);
5217 		pre_actions = sfx_actions + actions_n;
5218 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5219 						   actions, sfx_actions,
5220 						   pre_actions);
5221 		if (!mtr_tag_id) {
5222 			ret = -rte_errno;
5223 			goto exit;
5224 		}
5225 		/* Add the prefix subflow. */
5226 		flow_split_info->prefix_mark = 0;
5227 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5228 					      attr, items, pre_actions,
5229 					      flow_split_info, error);
5230 		if (ret) {
5231 			ret = -rte_errno;
5232 			goto exit;
5233 		}
5234 		dev_flow->handle->split_flow_id = mtr_tag_id;
5235 		/* Setting the sfx group atrr. */
5236 		sfx_attr.group = sfx_attr.transfer ?
5237 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5238 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5239 		flow_split_info->prefix_layers =
5240 				flow_get_prefix_layer_flags(dev_flow);
5241 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5242 	}
5243 	/* Add the prefix subflow. */
5244 	ret = flow_create_split_metadata(dev, flow,
5245 					 &sfx_attr, sfx_items ?
5246 					 sfx_items : items,
5247 					 sfx_actions ? sfx_actions : actions,
5248 					 flow_split_info, error);
5249 exit:
5250 	if (sfx_actions)
5251 		mlx5_free(sfx_actions);
5252 	return ret;
5253 }
5254 
5255 /**
5256  * The splitting for sample feature.
5257  *
5258  * Once Sample action is detected in the action list, the flow actions should
5259  * be split into prefix sub flow and suffix sub flow.
5260  *
5261  * The original items remain in the prefix sub flow, all actions preceding the
5262  * sample action and the sample action itself will be copied to the prefix
5263  * sub flow, the actions following the sample action will be copied to the
5264  * suffix sub flow, Queue action always be located in the suffix sub flow.
5265  *
5266  * In order to make the packet from prefix sub flow matches with suffix sub
5267  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5268  * flow uses tag item with the unique flow id.
5269  *
5270  * @param dev
5271  *   Pointer to Ethernet device.
5272  * @param[in] flow
5273  *   Parent flow structure pointer.
5274  * @param[in] attr
5275  *   Flow rule attributes.
5276  * @param[in] items
5277  *   Pattern specification (list terminated by the END pattern item).
5278  * @param[in] actions
5279  *   Associated actions (list terminated by the END action).
5280  * @param[in] flow_split_info
5281  *   Pointer to flow split info structure.
5282  * @param[out] error
5283  *   Perform verbose error reporting if not NULL.
5284  * @return
5285  *   0 on success, negative value otherwise
5286  */
5287 static int
5288 flow_create_split_sample(struct rte_eth_dev *dev,
5289 			 struct rte_flow *flow,
5290 			 const struct rte_flow_attr *attr,
5291 			 const struct rte_flow_item items[],
5292 			 const struct rte_flow_action actions[],
5293 			 struct mlx5_flow_split_info *flow_split_info,
5294 			 struct rte_flow_error *error)
5295 {
5296 	struct mlx5_priv *priv = dev->data->dev_private;
5297 	struct rte_flow_action *sfx_actions = NULL;
5298 	struct rte_flow_action *pre_actions = NULL;
5299 	struct rte_flow_item *sfx_items = NULL;
5300 	struct mlx5_flow *dev_flow = NULL;
5301 	struct rte_flow_attr sfx_attr = *attr;
5302 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5303 	struct mlx5_flow_dv_sample_resource *sample_res;
5304 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5305 	struct mlx5_flow_tbl_resource *sfx_tbl;
5306 #endif
5307 	size_t act_size;
5308 	size_t item_size;
5309 	uint32_t fdb_tx = 0;
5310 	int32_t tag_id = 0;
5311 	int actions_n = 0;
5312 	int sample_action_pos;
5313 	int qrss_action_pos;
5314 	int add_tag = 0;
5315 	int modify_after_mirror = 0;
5316 	uint16_t jump_table = 0;
5317 	const uint32_t next_ft_step = 1;
5318 	int ret = 0;
5319 
5320 	if (priv->sampler_en)
5321 		actions_n = flow_check_match_action(actions, attr,
5322 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5323 					&sample_action_pos, &qrss_action_pos,
5324 					&modify_after_mirror);
5325 	if (actions_n) {
5326 		/* The prefix actions must includes sample, tag, end. */
5327 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5328 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5329 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5330 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5331 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5332 					  item_size), 0, SOCKET_ID_ANY);
5333 		if (!sfx_actions)
5334 			return rte_flow_error_set(error, ENOMEM,
5335 						  RTE_FLOW_ERROR_TYPE_ACTION,
5336 						  NULL, "no memory to split "
5337 						  "sample flow");
5338 		/* The representor_id is -1 for uplink. */
5339 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5340 		/*
5341 		 * When reg_c_preserve is set, metadata registers Cx preserve
5342 		 * their value even through packet duplication.
5343 		 */
5344 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5345 		if (add_tag)
5346 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5347 					+ act_size);
5348 		if (modify_after_mirror)
5349 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5350 				     next_ft_step;
5351 		pre_actions = sfx_actions + actions_n;
5352 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5353 						actions, sfx_actions,
5354 						pre_actions, actions_n,
5355 						sample_action_pos,
5356 						qrss_action_pos, jump_table,
5357 						error);
5358 		if (tag_id < 0 || (add_tag && !tag_id)) {
5359 			ret = -rte_errno;
5360 			goto exit;
5361 		}
5362 		if (modify_after_mirror)
5363 			flow_split_info->skip_scale =
5364 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5365 		/* Add the prefix subflow. */
5366 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5367 					      items, pre_actions,
5368 					      flow_split_info, error);
5369 		if (ret) {
5370 			ret = -rte_errno;
5371 			goto exit;
5372 		}
5373 		dev_flow->handle->split_flow_id = tag_id;
5374 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5375 		if (!modify_after_mirror) {
5376 			/* Set the sfx group attr. */
5377 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5378 						dev_flow->dv.sample_res;
5379 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5380 						sample_res->normal_path_tbl;
5381 			sfx_tbl_data = container_of(sfx_tbl,
5382 						struct mlx5_flow_tbl_data_entry,
5383 						tbl);
5384 			sfx_attr.group = sfx_attr.transfer ?
5385 						(sfx_tbl_data->table_id - 1) :
5386 						sfx_tbl_data->table_id;
5387 		} else {
5388 			MLX5_ASSERT(attr->transfer);
5389 			sfx_attr.group = jump_table;
5390 		}
5391 		flow_split_info->prefix_layers =
5392 				flow_get_prefix_layer_flags(dev_flow);
5393 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5394 		/* Suffix group level already be scaled with factor, set
5395 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5396 		 * again in translation.
5397 		 */
5398 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5399 #endif
5400 	}
5401 	/* Add the suffix subflow. */
5402 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5403 				      sfx_items ? sfx_items : items,
5404 				      sfx_actions ? sfx_actions : actions,
5405 				      flow_split_info, error);
5406 exit:
5407 	if (sfx_actions)
5408 		mlx5_free(sfx_actions);
5409 	return ret;
5410 }
5411 
5412 /**
5413  * Split the flow to subflow set. The splitters might be linked
5414  * in the chain, like this:
5415  * flow_create_split_outer() calls:
5416  *   flow_create_split_meter() calls:
5417  *     flow_create_split_metadata(meter_subflow_0) calls:
5418  *       flow_create_split_inner(metadata_subflow_0)
5419  *       flow_create_split_inner(metadata_subflow_1)
5420  *       flow_create_split_inner(metadata_subflow_2)
5421  *     flow_create_split_metadata(meter_subflow_1) calls:
5422  *       flow_create_split_inner(metadata_subflow_0)
5423  *       flow_create_split_inner(metadata_subflow_1)
5424  *       flow_create_split_inner(metadata_subflow_2)
5425  *
5426  * This provide flexible way to add new levels of flow splitting.
5427  * The all of successfully created subflows are included to the
5428  * parent flow dev_flow list.
5429  *
5430  * @param dev
5431  *   Pointer to Ethernet device.
5432  * @param[in] flow
5433  *   Parent flow structure pointer.
5434  * @param[in] attr
5435  *   Flow rule attributes.
5436  * @param[in] items
5437  *   Pattern specification (list terminated by the END pattern item).
5438  * @param[in] actions
5439  *   Associated actions (list terminated by the END action).
5440  * @param[in] flow_split_info
5441  *   Pointer to flow split info structure.
5442  * @param[out] error
5443  *   Perform verbose error reporting if not NULL.
5444  * @return
5445  *   0 on success, negative value otherwise
5446  */
5447 static int
5448 flow_create_split_outer(struct rte_eth_dev *dev,
5449 			struct rte_flow *flow,
5450 			const struct rte_flow_attr *attr,
5451 			const struct rte_flow_item items[],
5452 			const struct rte_flow_action actions[],
5453 			struct mlx5_flow_split_info *flow_split_info,
5454 			struct rte_flow_error *error)
5455 {
5456 	int ret;
5457 
5458 	ret = flow_create_split_sample(dev, flow, attr, items,
5459 				       actions, flow_split_info, error);
5460 	MLX5_ASSERT(ret <= 0);
5461 	return ret;
5462 }
5463 
5464 static struct mlx5_flow_tunnel *
5465 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5466 		      const struct rte_flow_attr *attr,
5467 		      const struct rte_flow_item items[],
5468 		      const struct rte_flow_action actions[])
5469 {
5470 	struct mlx5_flow_tunnel *tunnel;
5471 
5472 #pragma GCC diagnostic push
5473 #pragma GCC diagnostic ignored "-Wcast-qual"
5474 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5475 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5476 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5477 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5478 	else
5479 		tunnel = NULL;
5480 #pragma GCC diagnostic pop
5481 
5482 	return tunnel;
5483 }
5484 
5485 /**
5486  * Adjust flow RSS workspace if needed.
5487  *
5488  * @param wks
5489  *   Pointer to thread flow work space.
5490  * @param rss_desc
5491  *   Pointer to RSS descriptor.
5492  * @param[in] nrssq_num
5493  *   New RSS queue number.
5494  *
5495  * @return
5496  *   0 on success, -1 otherwise and rte_errno is set.
5497  */
5498 static int
5499 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5500 			  struct mlx5_flow_rss_desc *rss_desc,
5501 			  uint32_t nrssq_num)
5502 {
5503 	if (likely(nrssq_num <= wks->rssq_num))
5504 		return 0;
5505 	rss_desc->queue = realloc(rss_desc->queue,
5506 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5507 	if (!rss_desc->queue) {
5508 		rte_errno = ENOMEM;
5509 		return -1;
5510 	}
5511 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5512 	return 0;
5513 }
5514 
5515 /**
5516  * Create a flow and add it to @p list.
5517  *
5518  * @param dev
5519  *   Pointer to Ethernet device.
5520  * @param list
5521  *   Pointer to a TAILQ flow list. If this parameter NULL,
5522  *   no list insertion occurred, flow is just created,
5523  *   this is caller's responsibility to track the
5524  *   created flow.
5525  * @param[in] attr
5526  *   Flow rule attributes.
5527  * @param[in] items
5528  *   Pattern specification (list terminated by the END pattern item).
5529  * @param[in] actions
5530  *   Associated actions (list terminated by the END action).
5531  * @param[in] external
5532  *   This flow rule is created by request external to PMD.
5533  * @param[out] error
5534  *   Perform verbose error reporting if not NULL.
5535  *
5536  * @return
5537  *   A flow index on success, 0 otherwise and rte_errno is set.
5538  */
5539 static uint32_t
5540 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5541 		 const struct rte_flow_attr *attr,
5542 		 const struct rte_flow_item items[],
5543 		 const struct rte_flow_action original_actions[],
5544 		 bool external, struct rte_flow_error *error)
5545 {
5546 	struct mlx5_priv *priv = dev->data->dev_private;
5547 	struct rte_flow *flow = NULL;
5548 	struct mlx5_flow *dev_flow;
5549 	const struct rte_flow_action_rss *rss = NULL;
5550 	struct mlx5_translated_shared_action
5551 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5552 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5553 	union {
5554 		struct mlx5_flow_expand_rss buf;
5555 		uint8_t buffer[2048];
5556 	} expand_buffer;
5557 	union {
5558 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5559 		uint8_t buffer[2048];
5560 	} actions_rx;
5561 	union {
5562 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5563 		uint8_t buffer[2048];
5564 	} actions_hairpin_tx;
5565 	union {
5566 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5567 		uint8_t buffer[2048];
5568 	} items_tx;
5569 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5570 	struct mlx5_flow_rss_desc *rss_desc;
5571 	const struct rte_flow_action *p_actions_rx;
5572 	uint32_t i;
5573 	uint32_t idx = 0;
5574 	int hairpin_flow;
5575 	struct rte_flow_attr attr_tx = { .priority = 0 };
5576 	const struct rte_flow_action *actions;
5577 	struct rte_flow_action *translated_actions = NULL;
5578 	struct mlx5_flow_tunnel *tunnel;
5579 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5580 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5581 	struct mlx5_flow_split_info flow_split_info = {
5582 		.external = !!external,
5583 		.skip_scale = 0,
5584 		.flow_idx = 0,
5585 		.prefix_mark = 0,
5586 		.prefix_layers = 0
5587 	};
5588 	int ret;
5589 
5590 	MLX5_ASSERT(wks);
5591 	rss_desc = &wks->rss_desc;
5592 	ret = flow_shared_actions_translate(dev, original_actions,
5593 					    shared_actions,
5594 					    &shared_actions_n,
5595 					    &translated_actions, error);
5596 	if (ret < 0) {
5597 		MLX5_ASSERT(translated_actions == NULL);
5598 		return 0;
5599 	}
5600 	actions = translated_actions ? translated_actions : original_actions;
5601 	p_actions_rx = actions;
5602 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5603 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5604 				external, hairpin_flow, error);
5605 	if (ret < 0)
5606 		goto error_before_hairpin_split;
5607 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5608 	if (!flow) {
5609 		rte_errno = ENOMEM;
5610 		goto error_before_hairpin_split;
5611 	}
5612 	if (hairpin_flow > 0) {
5613 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5614 			rte_errno = EINVAL;
5615 			goto error_before_hairpin_split;
5616 		}
5617 		flow_hairpin_split(dev, actions, actions_rx.actions,
5618 				   actions_hairpin_tx.actions, items_tx.items,
5619 				   idx);
5620 		p_actions_rx = actions_rx.actions;
5621 	}
5622 	flow_split_info.flow_idx = idx;
5623 	flow->drv_type = flow_get_drv_type(dev, attr);
5624 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5625 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5626 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5627 	/* RSS Action only works on NIC RX domain */
5628 	if (attr->ingress && !attr->transfer)
5629 		rss = flow_get_rss_action(p_actions_rx);
5630 	if (rss) {
5631 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5632 			return 0;
5633 		/*
5634 		 * The following information is required by
5635 		 * mlx5_flow_hashfields_adjust() in advance.
5636 		 */
5637 		rss_desc->level = rss->level;
5638 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5639 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5640 	}
5641 	flow->dev_handles = 0;
5642 	if (rss && rss->types) {
5643 		unsigned int graph_root;
5644 
5645 		graph_root = find_graph_root(items, rss->level);
5646 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5647 					   items, rss->types,
5648 					   mlx5_support_expansion, graph_root);
5649 		MLX5_ASSERT(ret > 0 &&
5650 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5651 	} else {
5652 		buf->entries = 1;
5653 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5654 	}
5655 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5656 						      shared_actions_n);
5657 	for (i = 0; i < buf->entries; ++i) {
5658 		/* Initialize flow split data. */
5659 		flow_split_info.prefix_layers = 0;
5660 		flow_split_info.prefix_mark = 0;
5661 		flow_split_info.skip_scale = 0;
5662 		/*
5663 		 * The splitter may create multiple dev_flows,
5664 		 * depending on configuration. In the simplest
5665 		 * case it just creates unmodified original flow.
5666 		 */
5667 		ret = flow_create_split_outer(dev, flow, attr,
5668 					      buf->entry[i].pattern,
5669 					      p_actions_rx, &flow_split_info,
5670 					      error);
5671 		if (ret < 0)
5672 			goto error;
5673 		if (is_flow_tunnel_steer_rule(dev, attr,
5674 					      buf->entry[i].pattern,
5675 					      p_actions_rx)) {
5676 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5677 							   p_actions_rx,
5678 							   idx,
5679 							   &default_miss_ctx,
5680 							   error);
5681 			if (ret < 0) {
5682 				mlx5_free(default_miss_ctx.queue);
5683 				goto error;
5684 			}
5685 		}
5686 	}
5687 	/* Create the tx flow. */
5688 	if (hairpin_flow) {
5689 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5690 		attr_tx.ingress = 0;
5691 		attr_tx.egress = 1;
5692 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5693 					 actions_hairpin_tx.actions,
5694 					 idx, error);
5695 		if (!dev_flow)
5696 			goto error;
5697 		dev_flow->flow = flow;
5698 		dev_flow->external = 0;
5699 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5700 			      dev_flow->handle, next);
5701 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5702 					 items_tx.items,
5703 					 actions_hairpin_tx.actions, error);
5704 		if (ret < 0)
5705 			goto error;
5706 	}
5707 	/*
5708 	 * Update the metadata register copy table. If extensive
5709 	 * metadata feature is enabled and registers are supported
5710 	 * we might create the extra rte_flow for each unique
5711 	 * MARK/FLAG action ID.
5712 	 *
5713 	 * The table is updated for ingress Flows only, because
5714 	 * the egress Flows belong to the different device and
5715 	 * copy table should be updated in peer NIC Rx domain.
5716 	 */
5717 	if (attr->ingress &&
5718 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5719 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5720 		if (ret)
5721 			goto error;
5722 	}
5723 	/*
5724 	 * If the flow is external (from application) OR device is started,
5725 	 * OR mreg discover, then apply immediately.
5726 	 */
5727 	if (external || dev->data->dev_started ||
5728 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5729 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
5730 		ret = flow_drv_apply(dev, flow, error);
5731 		if (ret < 0)
5732 			goto error;
5733 	}
5734 	if (list) {
5735 		rte_spinlock_lock(&priv->flow_list_lock);
5736 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5737 			     flow, next);
5738 		rte_spinlock_unlock(&priv->flow_list_lock);
5739 	}
5740 	flow_rxq_flags_set(dev, flow);
5741 	rte_free(translated_actions);
5742 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5743 	if (tunnel) {
5744 		flow->tunnel = 1;
5745 		flow->tunnel_id = tunnel->tunnel_id;
5746 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5747 		mlx5_free(default_miss_ctx.queue);
5748 	}
5749 	mlx5_flow_pop_thread_workspace();
5750 	return idx;
5751 error:
5752 	MLX5_ASSERT(flow);
5753 	ret = rte_errno; /* Save rte_errno before cleanup. */
5754 	flow_mreg_del_copy_action(dev, flow);
5755 	flow_drv_destroy(dev, flow);
5756 	if (rss_desc->shared_rss)
5757 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5758 			mlx5_ipool_get
5759 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5760 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5761 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5762 	rte_errno = ret; /* Restore rte_errno. */
5763 	ret = rte_errno;
5764 	rte_errno = ret;
5765 	mlx5_flow_pop_thread_workspace();
5766 error_before_hairpin_split:
5767 	rte_free(translated_actions);
5768 	return 0;
5769 }
5770 
5771 /**
5772  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5773  * incoming packets to table 1.
5774  *
5775  * Other flow rules, requested for group n, will be created in
5776  * e-switch table n+1.
5777  * Jump action to e-switch group n will be created to group n+1.
5778  *
5779  * Used when working in switchdev mode, to utilise advantages of table 1
5780  * and above.
5781  *
5782  * @param dev
5783  *   Pointer to Ethernet device.
5784  *
5785  * @return
5786  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5787  */
5788 struct rte_flow *
5789 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5790 {
5791 	const struct rte_flow_attr attr = {
5792 		.group = 0,
5793 		.priority = 0,
5794 		.ingress = 1,
5795 		.egress = 0,
5796 		.transfer = 1,
5797 	};
5798 	const struct rte_flow_item pattern = {
5799 		.type = RTE_FLOW_ITEM_TYPE_END,
5800 	};
5801 	struct rte_flow_action_jump jump = {
5802 		.group = 1,
5803 	};
5804 	const struct rte_flow_action actions[] = {
5805 		{
5806 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5807 			.conf = &jump,
5808 		},
5809 		{
5810 			.type = RTE_FLOW_ACTION_TYPE_END,
5811 		},
5812 	};
5813 	struct mlx5_priv *priv = dev->data->dev_private;
5814 	struct rte_flow_error error;
5815 
5816 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5817 						   &attr, &pattern,
5818 						   actions, false, &error);
5819 }
5820 
5821 /**
5822  * Validate a flow supported by the NIC.
5823  *
5824  * @see rte_flow_validate()
5825  * @see rte_flow_ops
5826  */
5827 int
5828 mlx5_flow_validate(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 original_actions[],
5832 		   struct rte_flow_error *error)
5833 {
5834 	int hairpin_flow;
5835 	struct mlx5_translated_shared_action
5836 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5837 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5838 	const struct rte_flow_action *actions;
5839 	struct rte_flow_action *translated_actions = NULL;
5840 	int ret = flow_shared_actions_translate(dev, original_actions,
5841 						shared_actions,
5842 						&shared_actions_n,
5843 						&translated_actions, error);
5844 
5845 	if (ret)
5846 		return ret;
5847 	actions = translated_actions ? translated_actions : original_actions;
5848 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5849 	ret = flow_drv_validate(dev, attr, items, actions,
5850 				true, hairpin_flow, error);
5851 	rte_free(translated_actions);
5852 	return ret;
5853 }
5854 
5855 /**
5856  * Create a flow.
5857  *
5858  * @see rte_flow_create()
5859  * @see rte_flow_ops
5860  */
5861 struct rte_flow *
5862 mlx5_flow_create(struct rte_eth_dev *dev,
5863 		 const struct rte_flow_attr *attr,
5864 		 const struct rte_flow_item items[],
5865 		 const struct rte_flow_action actions[],
5866 		 struct rte_flow_error *error)
5867 {
5868 	struct mlx5_priv *priv = dev->data->dev_private;
5869 
5870 	/*
5871 	 * If the device is not started yet, it is not allowed to created a
5872 	 * flow from application. PMD default flows and traffic control flows
5873 	 * are not affected.
5874 	 */
5875 	if (unlikely(!dev->data->dev_started)) {
5876 		DRV_LOG(DEBUG, "port %u is not started when "
5877 			"inserting a flow", dev->data->port_id);
5878 		rte_flow_error_set(error, ENODEV,
5879 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5880 				   NULL,
5881 				   "port not started");
5882 		return NULL;
5883 	}
5884 
5885 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5886 				  attr, items, actions, true, error);
5887 }
5888 
5889 /**
5890  * Destroy a flow in a list.
5891  *
5892  * @param dev
5893  *   Pointer to Ethernet device.
5894  * @param list
5895  *   Pointer to the Indexed flow list. If this parameter NULL,
5896  *   there is no flow removal from the list. Be noted that as
5897  *   flow is add to the indexed list, memory of the indexed
5898  *   list points to maybe changed as flow destroyed.
5899  * @param[in] flow_idx
5900  *   Index of flow to destroy.
5901  */
5902 static void
5903 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5904 		  uint32_t flow_idx)
5905 {
5906 	struct mlx5_priv *priv = dev->data->dev_private;
5907 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5908 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5909 
5910 	if (!flow)
5911 		return;
5912 	/*
5913 	 * Update RX queue flags only if port is started, otherwise it is
5914 	 * already clean.
5915 	 */
5916 	if (dev->data->dev_started)
5917 		flow_rxq_flags_trim(dev, flow);
5918 	flow_drv_destroy(dev, flow);
5919 	if (list) {
5920 		rte_spinlock_lock(&priv->flow_list_lock);
5921 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5922 			     flow_idx, flow, next);
5923 		rte_spinlock_unlock(&priv->flow_list_lock);
5924 	}
5925 	if (flow->tunnel) {
5926 		struct mlx5_flow_tunnel *tunnel;
5927 
5928 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5929 		RTE_VERIFY(tunnel);
5930 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5931 			mlx5_flow_tunnel_free(dev, tunnel);
5932 	}
5933 	flow_mreg_del_copy_action(dev, flow);
5934 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5935 }
5936 
5937 /**
5938  * Destroy all flows.
5939  *
5940  * @param dev
5941  *   Pointer to Ethernet device.
5942  * @param list
5943  *   Pointer to the Indexed flow list.
5944  * @param active
5945  *   If flushing is called avtively.
5946  */
5947 void
5948 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5949 {
5950 	uint32_t num_flushed = 0;
5951 
5952 	while (*list) {
5953 		flow_list_destroy(dev, list, *list);
5954 		num_flushed++;
5955 	}
5956 	if (active) {
5957 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5958 			dev->data->port_id, num_flushed);
5959 	}
5960 }
5961 
5962 /**
5963  * Stop all default actions for flows.
5964  *
5965  * @param dev
5966  *   Pointer to Ethernet device.
5967  */
5968 void
5969 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5970 {
5971 	flow_mreg_del_default_copy_action(dev);
5972 	flow_rxq_flags_clear(dev);
5973 }
5974 
5975 /**
5976  * Start all default actions for flows.
5977  *
5978  * @param dev
5979  *   Pointer to Ethernet device.
5980  * @return
5981  *   0 on success, a negative errno value otherwise and rte_errno is set.
5982  */
5983 int
5984 mlx5_flow_start_default(struct rte_eth_dev *dev)
5985 {
5986 	struct rte_flow_error error;
5987 
5988 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5989 	return flow_mreg_add_default_copy_action(dev, &error);
5990 }
5991 
5992 /**
5993  * Release key of thread specific flow workspace data.
5994  */
5995 void
5996 flow_release_workspace(void *data)
5997 {
5998 	struct mlx5_flow_workspace *wks = data;
5999 	struct mlx5_flow_workspace *next;
6000 
6001 	while (wks) {
6002 		next = wks->next;
6003 		free(wks->rss_desc.queue);
6004 		free(wks);
6005 		wks = next;
6006 	}
6007 }
6008 
6009 /**
6010  * Get thread specific current flow workspace.
6011  *
6012  * @return pointer to thread specific flow workspace data, NULL on error.
6013  */
6014 struct mlx5_flow_workspace*
6015 mlx5_flow_get_thread_workspace(void)
6016 {
6017 	struct mlx5_flow_workspace *data;
6018 
6019 	data = mlx5_flow_os_get_specific_workspace();
6020 	MLX5_ASSERT(data && data->inuse);
6021 	if (!data || !data->inuse)
6022 		DRV_LOG(ERR, "flow workspace not initialized.");
6023 	return data;
6024 }
6025 
6026 /**
6027  * Allocate and init new flow workspace.
6028  *
6029  * @return pointer to flow workspace data, NULL on error.
6030  */
6031 static struct mlx5_flow_workspace*
6032 flow_alloc_thread_workspace(void)
6033 {
6034 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6035 
6036 	if (!data) {
6037 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6038 			"memory.");
6039 		return NULL;
6040 	}
6041 	data->rss_desc.queue = calloc(1,
6042 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6043 	if (!data->rss_desc.queue)
6044 		goto err;
6045 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6046 	return data;
6047 err:
6048 	if (data->rss_desc.queue)
6049 		free(data->rss_desc.queue);
6050 	free(data);
6051 	return NULL;
6052 }
6053 
6054 /**
6055  * Get new thread specific flow workspace.
6056  *
6057  * If current workspace inuse, create new one and set as current.
6058  *
6059  * @return pointer to thread specific flow workspace data, NULL on error.
6060  */
6061 static struct mlx5_flow_workspace*
6062 mlx5_flow_push_thread_workspace(void)
6063 {
6064 	struct mlx5_flow_workspace *curr;
6065 	struct mlx5_flow_workspace *data;
6066 
6067 	curr = mlx5_flow_os_get_specific_workspace();
6068 	if (!curr) {
6069 		data = flow_alloc_thread_workspace();
6070 		if (!data)
6071 			return NULL;
6072 	} else if (!curr->inuse) {
6073 		data = curr;
6074 	} else if (curr->next) {
6075 		data = curr->next;
6076 	} else {
6077 		data = flow_alloc_thread_workspace();
6078 		if (!data)
6079 			return NULL;
6080 		curr->next = data;
6081 		data->prev = curr;
6082 	}
6083 	data->inuse = 1;
6084 	data->flow_idx = 0;
6085 	/* Set as current workspace */
6086 	if (mlx5_flow_os_set_specific_workspace(data))
6087 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6088 	return data;
6089 }
6090 
6091 /**
6092  * Close current thread specific flow workspace.
6093  *
6094  * If previous workspace available, set it as current.
6095  *
6096  * @return pointer to thread specific flow workspace data, NULL on error.
6097  */
6098 static void
6099 mlx5_flow_pop_thread_workspace(void)
6100 {
6101 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6102 
6103 	if (!data)
6104 		return;
6105 	if (!data->inuse) {
6106 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6107 		return;
6108 	}
6109 	data->inuse = 0;
6110 	if (!data->prev)
6111 		return;
6112 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6113 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6114 }
6115 
6116 /**
6117  * Verify the flow list is empty
6118  *
6119  * @param dev
6120  *  Pointer to Ethernet device.
6121  *
6122  * @return the number of flows not released.
6123  */
6124 int
6125 mlx5_flow_verify(struct rte_eth_dev *dev)
6126 {
6127 	struct mlx5_priv *priv = dev->data->dev_private;
6128 	struct rte_flow *flow;
6129 	uint32_t idx;
6130 	int ret = 0;
6131 
6132 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6133 		      flow, next) {
6134 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6135 			dev->data->port_id, (void *)flow);
6136 		++ret;
6137 	}
6138 	return ret;
6139 }
6140 
6141 /**
6142  * Enable default hairpin egress flow.
6143  *
6144  * @param dev
6145  *   Pointer to Ethernet device.
6146  * @param queue
6147  *   The queue index.
6148  *
6149  * @return
6150  *   0 on success, a negative errno value otherwise and rte_errno is set.
6151  */
6152 int
6153 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6154 			    uint32_t queue)
6155 {
6156 	struct mlx5_priv *priv = dev->data->dev_private;
6157 	const struct rte_flow_attr attr = {
6158 		.egress = 1,
6159 		.priority = 0,
6160 	};
6161 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6162 		.queue = queue,
6163 	};
6164 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6165 		.queue = UINT32_MAX,
6166 	};
6167 	struct rte_flow_item items[] = {
6168 		{
6169 			.type = (enum rte_flow_item_type)
6170 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6171 			.spec = &queue_spec,
6172 			.last = NULL,
6173 			.mask = &queue_mask,
6174 		},
6175 		{
6176 			.type = RTE_FLOW_ITEM_TYPE_END,
6177 		},
6178 	};
6179 	struct rte_flow_action_jump jump = {
6180 		.group = MLX5_HAIRPIN_TX_TABLE,
6181 	};
6182 	struct rte_flow_action actions[2];
6183 	uint32_t flow_idx;
6184 	struct rte_flow_error error;
6185 
6186 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6187 	actions[0].conf = &jump;
6188 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6189 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6190 				&attr, items, actions, false, &error);
6191 	if (!flow_idx) {
6192 		DRV_LOG(DEBUG,
6193 			"Failed to create ctrl flow: rte_errno(%d),"
6194 			" type(%d), message(%s)",
6195 			rte_errno, error.type,
6196 			error.message ? error.message : " (no stated reason)");
6197 		return -rte_errno;
6198 	}
6199 	return 0;
6200 }
6201 
6202 /**
6203  * Enable a control flow configured from the control plane.
6204  *
6205  * @param dev
6206  *   Pointer to Ethernet device.
6207  * @param eth_spec
6208  *   An Ethernet flow spec to apply.
6209  * @param eth_mask
6210  *   An Ethernet flow mask to apply.
6211  * @param vlan_spec
6212  *   A VLAN flow spec to apply.
6213  * @param vlan_mask
6214  *   A VLAN flow mask to apply.
6215  *
6216  * @return
6217  *   0 on success, a negative errno value otherwise and rte_errno is set.
6218  */
6219 int
6220 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6221 		    struct rte_flow_item_eth *eth_spec,
6222 		    struct rte_flow_item_eth *eth_mask,
6223 		    struct rte_flow_item_vlan *vlan_spec,
6224 		    struct rte_flow_item_vlan *vlan_mask)
6225 {
6226 	struct mlx5_priv *priv = dev->data->dev_private;
6227 	const struct rte_flow_attr attr = {
6228 		.ingress = 1,
6229 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6230 	};
6231 	struct rte_flow_item items[] = {
6232 		{
6233 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6234 			.spec = eth_spec,
6235 			.last = NULL,
6236 			.mask = eth_mask,
6237 		},
6238 		{
6239 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6240 					      RTE_FLOW_ITEM_TYPE_END,
6241 			.spec = vlan_spec,
6242 			.last = NULL,
6243 			.mask = vlan_mask,
6244 		},
6245 		{
6246 			.type = RTE_FLOW_ITEM_TYPE_END,
6247 		},
6248 	};
6249 	uint16_t queue[priv->reta_idx_n];
6250 	struct rte_flow_action_rss action_rss = {
6251 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6252 		.level = 0,
6253 		.types = priv->rss_conf.rss_hf,
6254 		.key_len = priv->rss_conf.rss_key_len,
6255 		.queue_num = priv->reta_idx_n,
6256 		.key = priv->rss_conf.rss_key,
6257 		.queue = queue,
6258 	};
6259 	struct rte_flow_action actions[] = {
6260 		{
6261 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6262 			.conf = &action_rss,
6263 		},
6264 		{
6265 			.type = RTE_FLOW_ACTION_TYPE_END,
6266 		},
6267 	};
6268 	uint32_t flow_idx;
6269 	struct rte_flow_error error;
6270 	unsigned int i;
6271 
6272 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6273 		return 0;
6274 	}
6275 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6276 		action_rss.types = 0;
6277 	for (i = 0; i != priv->reta_idx_n; ++i)
6278 		queue[i] = (*priv->reta_idx)[i];
6279 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6280 				&attr, items, actions, false, &error);
6281 	if (!flow_idx)
6282 		return -rte_errno;
6283 	return 0;
6284 }
6285 
6286 /**
6287  * Enable a flow control configured from the control plane.
6288  *
6289  * @param dev
6290  *   Pointer to Ethernet device.
6291  * @param eth_spec
6292  *   An Ethernet flow spec to apply.
6293  * @param eth_mask
6294  *   An Ethernet flow mask to apply.
6295  *
6296  * @return
6297  *   0 on success, a negative errno value otherwise and rte_errno is set.
6298  */
6299 int
6300 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6301 	       struct rte_flow_item_eth *eth_spec,
6302 	       struct rte_flow_item_eth *eth_mask)
6303 {
6304 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6305 }
6306 
6307 /**
6308  * Create default miss flow rule matching lacp traffic
6309  *
6310  * @param dev
6311  *   Pointer to Ethernet device.
6312  * @param eth_spec
6313  *   An Ethernet flow spec to apply.
6314  *
6315  * @return
6316  *   0 on success, a negative errno value otherwise and rte_errno is set.
6317  */
6318 int
6319 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6320 {
6321 	struct mlx5_priv *priv = dev->data->dev_private;
6322 	/*
6323 	 * The LACP matching is done by only using ether type since using
6324 	 * a multicast dst mac causes kernel to give low priority to this flow.
6325 	 */
6326 	static const struct rte_flow_item_eth lacp_spec = {
6327 		.type = RTE_BE16(0x8809),
6328 	};
6329 	static const struct rte_flow_item_eth lacp_mask = {
6330 		.type = 0xffff,
6331 	};
6332 	const struct rte_flow_attr attr = {
6333 		.ingress = 1,
6334 	};
6335 	struct rte_flow_item items[] = {
6336 		{
6337 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6338 			.spec = &lacp_spec,
6339 			.mask = &lacp_mask,
6340 		},
6341 		{
6342 			.type = RTE_FLOW_ITEM_TYPE_END,
6343 		},
6344 	};
6345 	struct rte_flow_action actions[] = {
6346 		{
6347 			.type = (enum rte_flow_action_type)
6348 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6349 		},
6350 		{
6351 			.type = RTE_FLOW_ACTION_TYPE_END,
6352 		},
6353 	};
6354 	struct rte_flow_error error;
6355 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6356 				&attr, items, actions, false, &error);
6357 
6358 	if (!flow_idx)
6359 		return -rte_errno;
6360 	return 0;
6361 }
6362 
6363 /**
6364  * Destroy a flow.
6365  *
6366  * @see rte_flow_destroy()
6367  * @see rte_flow_ops
6368  */
6369 int
6370 mlx5_flow_destroy(struct rte_eth_dev *dev,
6371 		  struct rte_flow *flow,
6372 		  struct rte_flow_error *error __rte_unused)
6373 {
6374 	struct mlx5_priv *priv = dev->data->dev_private;
6375 
6376 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6377 	return 0;
6378 }
6379 
6380 /**
6381  * Destroy all flows.
6382  *
6383  * @see rte_flow_flush()
6384  * @see rte_flow_ops
6385  */
6386 int
6387 mlx5_flow_flush(struct rte_eth_dev *dev,
6388 		struct rte_flow_error *error __rte_unused)
6389 {
6390 	struct mlx5_priv *priv = dev->data->dev_private;
6391 
6392 	mlx5_flow_list_flush(dev, &priv->flows, false);
6393 	return 0;
6394 }
6395 
6396 /**
6397  * Isolated mode.
6398  *
6399  * @see rte_flow_isolate()
6400  * @see rte_flow_ops
6401  */
6402 int
6403 mlx5_flow_isolate(struct rte_eth_dev *dev,
6404 		  int enable,
6405 		  struct rte_flow_error *error)
6406 {
6407 	struct mlx5_priv *priv = dev->data->dev_private;
6408 
6409 	if (dev->data->dev_started) {
6410 		rte_flow_error_set(error, EBUSY,
6411 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6412 				   NULL,
6413 				   "port must be stopped first");
6414 		return -rte_errno;
6415 	}
6416 	priv->isolated = !!enable;
6417 	if (enable)
6418 		dev->dev_ops = &mlx5_dev_ops_isolate;
6419 	else
6420 		dev->dev_ops = &mlx5_dev_ops;
6421 
6422 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6423 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6424 
6425 	return 0;
6426 }
6427 
6428 /**
6429  * Query a flow.
6430  *
6431  * @see rte_flow_query()
6432  * @see rte_flow_ops
6433  */
6434 static int
6435 flow_drv_query(struct rte_eth_dev *dev,
6436 	       uint32_t flow_idx,
6437 	       const struct rte_flow_action *actions,
6438 	       void *data,
6439 	       struct rte_flow_error *error)
6440 {
6441 	struct mlx5_priv *priv = dev->data->dev_private;
6442 	const struct mlx5_flow_driver_ops *fops;
6443 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6444 					       [MLX5_IPOOL_RTE_FLOW],
6445 					       flow_idx);
6446 	enum mlx5_flow_drv_type ftype;
6447 
6448 	if (!flow) {
6449 		return rte_flow_error_set(error, ENOENT,
6450 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6451 			  NULL,
6452 			  "invalid flow handle");
6453 	}
6454 	ftype = flow->drv_type;
6455 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6456 	fops = flow_get_drv_ops(ftype);
6457 
6458 	return fops->query(dev, flow, actions, data, error);
6459 }
6460 
6461 /**
6462  * Query a flow.
6463  *
6464  * @see rte_flow_query()
6465  * @see rte_flow_ops
6466  */
6467 int
6468 mlx5_flow_query(struct rte_eth_dev *dev,
6469 		struct rte_flow *flow,
6470 		const struct rte_flow_action *actions,
6471 		void *data,
6472 		struct rte_flow_error *error)
6473 {
6474 	int ret;
6475 
6476 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6477 			     error);
6478 	if (ret < 0)
6479 		return ret;
6480 	return 0;
6481 }
6482 
6483 /**
6484  * Get rte_flow callbacks.
6485  *
6486  * @param dev
6487  *   Pointer to Ethernet device structure.
6488  * @param ops
6489  *   Pointer to operation-specific structure.
6490  *
6491  * @return 0
6492  */
6493 int
6494 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
6495 		  const struct rte_flow_ops **ops)
6496 {
6497 	*ops = &mlx5_flow_ops;
6498 	return 0;
6499 }
6500 
6501 /**
6502  * Create the needed meter and suffix tables.
6503  *
6504  * @param[in] dev
6505  *   Pointer to Ethernet device.
6506  * @param[in] fm
6507  *   Pointer to the flow meter.
6508  *
6509  * @return
6510  *   Pointer to table set on success, NULL otherwise.
6511  */
6512 struct mlx5_meter_domains_infos *
6513 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6514 			  const struct mlx5_flow_meter *fm)
6515 {
6516 	const struct mlx5_flow_driver_ops *fops;
6517 
6518 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6519 	return fops->create_mtr_tbls(dev, fm);
6520 }
6521 
6522 /**
6523  * Destroy the meter table set.
6524  *
6525  * @param[in] dev
6526  *   Pointer to Ethernet device.
6527  * @param[in] tbl
6528  *   Pointer to the meter table set.
6529  *
6530  * @return
6531  *   0 on success.
6532  */
6533 int
6534 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6535 			   struct mlx5_meter_domains_infos *tbls)
6536 {
6537 	const struct mlx5_flow_driver_ops *fops;
6538 
6539 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6540 	return fops->destroy_mtr_tbls(dev, tbls);
6541 }
6542 
6543 /**
6544  * Create policer rules.
6545  *
6546  * @param[in] dev
6547  *   Pointer to Ethernet device.
6548  * @param[in] fm
6549  *   Pointer to flow meter structure.
6550  * @param[in] attr
6551  *   Pointer to flow attributes.
6552  *
6553  * @return
6554  *   0 on success, -1 otherwise.
6555  */
6556 int
6557 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6558 			       struct mlx5_flow_meter *fm,
6559 			       const struct rte_flow_attr *attr)
6560 {
6561 	const struct mlx5_flow_driver_ops *fops;
6562 
6563 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6564 	return fops->create_policer_rules(dev, fm, attr);
6565 }
6566 
6567 /**
6568  * Destroy policer rules.
6569  *
6570  * @param[in] fm
6571  *   Pointer to flow meter structure.
6572  * @param[in] attr
6573  *   Pointer to flow attributes.
6574  *
6575  * @return
6576  *   0 on success, -1 otherwise.
6577  */
6578 int
6579 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6580 				struct mlx5_flow_meter *fm,
6581 				const struct rte_flow_attr *attr)
6582 {
6583 	const struct mlx5_flow_driver_ops *fops;
6584 
6585 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6586 	return fops->destroy_policer_rules(dev, fm, attr);
6587 }
6588 
6589 /**
6590  * Allocate a counter.
6591  *
6592  * @param[in] dev
6593  *   Pointer to Ethernet device structure.
6594  *
6595  * @return
6596  *   Index to allocated counter  on success, 0 otherwise.
6597  */
6598 uint32_t
6599 mlx5_counter_alloc(struct rte_eth_dev *dev)
6600 {
6601 	const struct mlx5_flow_driver_ops *fops;
6602 	struct rte_flow_attr attr = { .transfer = 0 };
6603 
6604 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6605 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6606 		return fops->counter_alloc(dev);
6607 	}
6608 	DRV_LOG(ERR,
6609 		"port %u counter allocate is not supported.",
6610 		 dev->data->port_id);
6611 	return 0;
6612 }
6613 
6614 /**
6615  * Free a counter.
6616  *
6617  * @param[in] dev
6618  *   Pointer to Ethernet device structure.
6619  * @param[in] cnt
6620  *   Index to counter to be free.
6621  */
6622 void
6623 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6624 {
6625 	const struct mlx5_flow_driver_ops *fops;
6626 	struct rte_flow_attr attr = { .transfer = 0 };
6627 
6628 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6629 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6630 		fops->counter_free(dev, cnt);
6631 		return;
6632 	}
6633 	DRV_LOG(ERR,
6634 		"port %u counter free is not supported.",
6635 		 dev->data->port_id);
6636 }
6637 
6638 /**
6639  * Query counter statistics.
6640  *
6641  * @param[in] dev
6642  *   Pointer to Ethernet device structure.
6643  * @param[in] cnt
6644  *   Index to counter to query.
6645  * @param[in] clear
6646  *   Set to clear counter statistics.
6647  * @param[out] pkts
6648  *   The counter hits packets number to save.
6649  * @param[out] bytes
6650  *   The counter hits bytes number to save.
6651  *
6652  * @return
6653  *   0 on success, a negative errno value otherwise.
6654  */
6655 int
6656 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6657 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6658 {
6659 	const struct mlx5_flow_driver_ops *fops;
6660 	struct rte_flow_attr attr = { .transfer = 0 };
6661 
6662 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6663 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6664 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6665 	}
6666 	DRV_LOG(ERR,
6667 		"port %u counter query is not supported.",
6668 		 dev->data->port_id);
6669 	return -ENOTSUP;
6670 }
6671 
6672 /**
6673  * Allocate a new memory for the counter values wrapped by all the needed
6674  * management.
6675  *
6676  * @param[in] sh
6677  *   Pointer to mlx5_dev_ctx_shared object.
6678  *
6679  * @return
6680  *   0 on success, a negative errno value otherwise.
6681  */
6682 static int
6683 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6684 {
6685 	struct mlx5_devx_mkey_attr mkey_attr;
6686 	struct mlx5_counter_stats_mem_mng *mem_mng;
6687 	volatile struct flow_counter_stats *raw_data;
6688 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6689 	int size = (sizeof(struct flow_counter_stats) *
6690 			MLX5_COUNTERS_PER_POOL +
6691 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6692 			sizeof(struct mlx5_counter_stats_mem_mng);
6693 	size_t pgsize = rte_mem_page_size();
6694 	uint8_t *mem;
6695 	int i;
6696 
6697 	if (pgsize == (size_t)-1) {
6698 		DRV_LOG(ERR, "Failed to get mem page size");
6699 		rte_errno = ENOMEM;
6700 		return -ENOMEM;
6701 	}
6702 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6703 	if (!mem) {
6704 		rte_errno = ENOMEM;
6705 		return -ENOMEM;
6706 	}
6707 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6708 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6709 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6710 						 IBV_ACCESS_LOCAL_WRITE);
6711 	if (!mem_mng->umem) {
6712 		rte_errno = errno;
6713 		mlx5_free(mem);
6714 		return -rte_errno;
6715 	}
6716 	mkey_attr.addr = (uintptr_t)mem;
6717 	mkey_attr.size = size;
6718 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6719 	mkey_attr.pd = sh->pdn;
6720 	mkey_attr.log_entity_size = 0;
6721 	mkey_attr.pg_access = 0;
6722 	mkey_attr.klm_array = NULL;
6723 	mkey_attr.klm_num = 0;
6724 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6725 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6726 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6727 	if (!mem_mng->dm) {
6728 		mlx5_os_umem_dereg(mem_mng->umem);
6729 		rte_errno = errno;
6730 		mlx5_free(mem);
6731 		return -rte_errno;
6732 	}
6733 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6734 	raw_data = (volatile struct flow_counter_stats *)mem;
6735 	for (i = 0; i < raws_n; ++i) {
6736 		mem_mng->raws[i].mem_mng = mem_mng;
6737 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6738 	}
6739 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6740 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6741 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6742 				 next);
6743 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6744 	sh->cmng.mem_mng = mem_mng;
6745 	return 0;
6746 }
6747 
6748 /**
6749  * Set the statistic memory to the new counter pool.
6750  *
6751  * @param[in] sh
6752  *   Pointer to mlx5_dev_ctx_shared object.
6753  * @param[in] pool
6754  *   Pointer to the pool to set the statistic memory.
6755  *
6756  * @return
6757  *   0 on success, a negative errno value otherwise.
6758  */
6759 static int
6760 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6761 			       struct mlx5_flow_counter_pool *pool)
6762 {
6763 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6764 	/* Resize statistic memory once used out. */
6765 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6766 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6767 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6768 		return -1;
6769 	}
6770 	rte_spinlock_lock(&pool->sl);
6771 	pool->raw = cmng->mem_mng->raws + pool->index %
6772 		    MLX5_CNT_CONTAINER_RESIZE;
6773 	rte_spinlock_unlock(&pool->sl);
6774 	pool->raw_hw = NULL;
6775 	return 0;
6776 }
6777 
6778 #define MLX5_POOL_QUERY_FREQ_US 1000000
6779 
6780 /**
6781  * Set the periodic procedure for triggering asynchronous batch queries for all
6782  * the counter pools.
6783  *
6784  * @param[in] sh
6785  *   Pointer to mlx5_dev_ctx_shared object.
6786  */
6787 void
6788 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6789 {
6790 	uint32_t pools_n, us;
6791 
6792 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6793 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6794 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6795 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6796 		sh->cmng.query_thread_on = 0;
6797 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6798 	} else {
6799 		sh->cmng.query_thread_on = 1;
6800 	}
6801 }
6802 
6803 /**
6804  * The periodic procedure for triggering asynchronous batch queries for all the
6805  * counter pools. This function is probably called by the host thread.
6806  *
6807  * @param[in] arg
6808  *   The parameter for the alarm process.
6809  */
6810 void
6811 mlx5_flow_query_alarm(void *arg)
6812 {
6813 	struct mlx5_dev_ctx_shared *sh = arg;
6814 	int ret;
6815 	uint16_t pool_index = sh->cmng.pool_index;
6816 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6817 	struct mlx5_flow_counter_pool *pool;
6818 	uint16_t n_valid;
6819 
6820 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6821 		goto set_alarm;
6822 	rte_spinlock_lock(&cmng->pool_update_sl);
6823 	pool = cmng->pools[pool_index];
6824 	n_valid = cmng->n_valid;
6825 	rte_spinlock_unlock(&cmng->pool_update_sl);
6826 	/* Set the statistic memory to the new created pool. */
6827 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6828 		goto set_alarm;
6829 	if (pool->raw_hw)
6830 		/* There is a pool query in progress. */
6831 		goto set_alarm;
6832 	pool->raw_hw =
6833 		LIST_FIRST(&sh->cmng.free_stat_raws);
6834 	if (!pool->raw_hw)
6835 		/* No free counter statistics raw memory. */
6836 		goto set_alarm;
6837 	/*
6838 	 * Identify the counters released between query trigger and query
6839 	 * handle more efficiently. The counter released in this gap period
6840 	 * should wait for a new round of query as the new arrived packets
6841 	 * will not be taken into account.
6842 	 */
6843 	pool->query_gen++;
6844 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6845 					       MLX5_COUNTERS_PER_POOL,
6846 					       NULL, NULL,
6847 					       pool->raw_hw->mem_mng->dm->id,
6848 					       (void *)(uintptr_t)
6849 					       pool->raw_hw->data,
6850 					       sh->devx_comp,
6851 					       (uint64_t)(uintptr_t)pool);
6852 	if (ret) {
6853 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6854 			" %d", pool->min_dcs->id);
6855 		pool->raw_hw = NULL;
6856 		goto set_alarm;
6857 	}
6858 	LIST_REMOVE(pool->raw_hw, next);
6859 	sh->cmng.pending_queries++;
6860 	pool_index++;
6861 	if (pool_index >= n_valid)
6862 		pool_index = 0;
6863 set_alarm:
6864 	sh->cmng.pool_index = pool_index;
6865 	mlx5_set_query_alarm(sh);
6866 }
6867 
6868 /**
6869  * Check and callback event for new aged flow in the counter pool
6870  *
6871  * @param[in] sh
6872  *   Pointer to mlx5_dev_ctx_shared object.
6873  * @param[in] pool
6874  *   Pointer to Current counter pool.
6875  */
6876 static void
6877 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6878 		   struct mlx5_flow_counter_pool *pool)
6879 {
6880 	struct mlx5_priv *priv;
6881 	struct mlx5_flow_counter *cnt;
6882 	struct mlx5_age_info *age_info;
6883 	struct mlx5_age_param *age_param;
6884 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6885 	struct mlx5_counter_stats_raw *prev = pool->raw;
6886 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6887 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6888 	uint16_t expected = AGE_CANDIDATE;
6889 	uint32_t i;
6890 
6891 	pool->time_of_last_age_check = curr_time;
6892 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6893 		cnt = MLX5_POOL_GET_CNT(pool, i);
6894 		age_param = MLX5_CNT_TO_AGE(cnt);
6895 		if (__atomic_load_n(&age_param->state,
6896 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6897 			continue;
6898 		if (cur->data[i].hits != prev->data[i].hits) {
6899 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6900 					 __ATOMIC_RELAXED);
6901 			continue;
6902 		}
6903 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6904 				       time_delta,
6905 				       __ATOMIC_RELAXED) <= age_param->timeout)
6906 			continue;
6907 		/**
6908 		 * Hold the lock first, or if between the
6909 		 * state AGE_TMOUT and tailq operation the
6910 		 * release happened, the release procedure
6911 		 * may delete a non-existent tailq node.
6912 		 */
6913 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6914 		age_info = GET_PORT_AGE_INFO(priv);
6915 		rte_spinlock_lock(&age_info->aged_sl);
6916 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6917 						AGE_TMOUT, false,
6918 						__ATOMIC_RELAXED,
6919 						__ATOMIC_RELAXED)) {
6920 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6921 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6922 		}
6923 		rte_spinlock_unlock(&age_info->aged_sl);
6924 	}
6925 	mlx5_age_event_prepare(sh);
6926 }
6927 
6928 /**
6929  * Handler for the HW respond about ready values from an asynchronous batch
6930  * query. This function is probably called by the host thread.
6931  *
6932  * @param[in] sh
6933  *   The pointer to the shared device context.
6934  * @param[in] async_id
6935  *   The Devx async ID.
6936  * @param[in] status
6937  *   The status of the completion.
6938  */
6939 void
6940 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6941 				  uint64_t async_id, int status)
6942 {
6943 	struct mlx5_flow_counter_pool *pool =
6944 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6945 	struct mlx5_counter_stats_raw *raw_to_free;
6946 	uint8_t query_gen = pool->query_gen ^ 1;
6947 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6948 	enum mlx5_counter_type cnt_type =
6949 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6950 				MLX5_COUNTER_TYPE_ORIGIN;
6951 
6952 	if (unlikely(status)) {
6953 		raw_to_free = pool->raw_hw;
6954 	} else {
6955 		raw_to_free = pool->raw;
6956 		if (pool->is_aged)
6957 			mlx5_flow_aging_check(sh, pool);
6958 		rte_spinlock_lock(&pool->sl);
6959 		pool->raw = pool->raw_hw;
6960 		rte_spinlock_unlock(&pool->sl);
6961 		/* Be sure the new raw counters data is updated in memory. */
6962 		rte_io_wmb();
6963 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6964 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6965 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6966 				     &pool->counters[query_gen], next);
6967 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6968 		}
6969 	}
6970 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6971 	pool->raw_hw = NULL;
6972 	sh->cmng.pending_queries--;
6973 }
6974 
6975 static int
6976 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6977 		    const struct flow_grp_info *grp_info,
6978 		    struct rte_flow_error *error)
6979 {
6980 	if (grp_info->transfer && grp_info->external &&
6981 	    grp_info->fdb_def_rule) {
6982 		if (group == UINT32_MAX)
6983 			return rte_flow_error_set
6984 						(error, EINVAL,
6985 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6986 						 NULL,
6987 						 "group index not supported");
6988 		*table = group + 1;
6989 	} else {
6990 		*table = group;
6991 	}
6992 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6993 	return 0;
6994 }
6995 
6996 /**
6997  * Translate the rte_flow group index to HW table value.
6998  *
6999  * If tunnel offload is disabled, all group ids converted to flow table
7000  * id using the standard method.
7001  * If tunnel offload is enabled, group id can be converted using the
7002  * standard or tunnel conversion method. Group conversion method
7003  * selection depends on flags in `grp_info` parameter:
7004  * - Internal (grp_info.external == 0) groups conversion uses the
7005  *   standard method.
7006  * - Group ids in JUMP action converted with the tunnel conversion.
7007  * - Group id in rule attribute conversion depends on a rule type and
7008  *   group id value:
7009  *   ** non zero group attributes converted with the tunnel method
7010  *   ** zero group attribute in non-tunnel rule is converted using the
7011  *      standard method - there's only one root table
7012  *   ** zero group attribute in steer tunnel rule is converted with the
7013  *      standard method - single root table
7014  *   ** zero group attribute in match tunnel rule is a special OvS
7015  *      case: that value is used for portability reasons. That group
7016  *      id is converted with the tunnel conversion method.
7017  *
7018  * @param[in] dev
7019  *   Port device
7020  * @param[in] tunnel
7021  *   PMD tunnel offload object
7022  * @param[in] group
7023  *   rte_flow group index value.
7024  * @param[out] table
7025  *   HW table value.
7026  * @param[in] grp_info
7027  *   flags used for conversion
7028  * @param[out] error
7029  *   Pointer to error structure.
7030  *
7031  * @return
7032  *   0 on success, a negative errno value otherwise and rte_errno is set.
7033  */
7034 int
7035 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7036 			 const struct mlx5_flow_tunnel *tunnel,
7037 			 uint32_t group, uint32_t *table,
7038 			 const struct flow_grp_info *grp_info,
7039 			 struct rte_flow_error *error)
7040 {
7041 	int ret;
7042 	bool standard_translation;
7043 
7044 	if (!grp_info->skip_scale && grp_info->external &&
7045 	    group < MLX5_MAX_TABLES_EXTERNAL)
7046 		group *= MLX5_FLOW_TABLE_FACTOR;
7047 	if (is_tunnel_offload_active(dev)) {
7048 		standard_translation = !grp_info->external ||
7049 					grp_info->std_tbl_fix;
7050 	} else {
7051 		standard_translation = true;
7052 	}
7053 	DRV_LOG(DEBUG,
7054 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7055 		dev->data->port_id, group, grp_info->transfer,
7056 		grp_info->external, grp_info->fdb_def_rule,
7057 		standard_translation ? "STANDARD" : "TUNNEL");
7058 	if (standard_translation)
7059 		ret = flow_group_to_table(dev->data->port_id, group, table,
7060 					  grp_info, error);
7061 	else
7062 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7063 						      table, error);
7064 
7065 	return ret;
7066 }
7067 
7068 /**
7069  * Discover availability of metadata reg_c's.
7070  *
7071  * Iteratively use test flows to check availability.
7072  *
7073  * @param[in] dev
7074  *   Pointer to the Ethernet device structure.
7075  *
7076  * @return
7077  *   0 on success, a negative errno value otherwise and rte_errno is set.
7078  */
7079 int
7080 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7081 {
7082 	struct mlx5_priv *priv = dev->data->dev_private;
7083 	struct mlx5_dev_config *config = &priv->config;
7084 	enum modify_reg idx;
7085 	int n = 0;
7086 
7087 	/* reg_c[0] and reg_c[1] are reserved. */
7088 	config->flow_mreg_c[n++] = REG_C_0;
7089 	config->flow_mreg_c[n++] = REG_C_1;
7090 	/* Discover availability of other reg_c's. */
7091 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7092 		struct rte_flow_attr attr = {
7093 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7094 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7095 			.ingress = 1,
7096 		};
7097 		struct rte_flow_item items[] = {
7098 			[0] = {
7099 				.type = RTE_FLOW_ITEM_TYPE_END,
7100 			},
7101 		};
7102 		struct rte_flow_action actions[] = {
7103 			[0] = {
7104 				.type = (enum rte_flow_action_type)
7105 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7106 				.conf = &(struct mlx5_flow_action_copy_mreg){
7107 					.src = REG_C_1,
7108 					.dst = idx,
7109 				},
7110 			},
7111 			[1] = {
7112 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7113 				.conf = &(struct rte_flow_action_jump){
7114 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7115 				},
7116 			},
7117 			[2] = {
7118 				.type = RTE_FLOW_ACTION_TYPE_END,
7119 			},
7120 		};
7121 		uint32_t flow_idx;
7122 		struct rte_flow *flow;
7123 		struct rte_flow_error error;
7124 
7125 		if (!config->dv_flow_en)
7126 			break;
7127 		/* Create internal flow, validation skips copy action. */
7128 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7129 					    actions, false, &error);
7130 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7131 				      flow_idx);
7132 		if (!flow)
7133 			continue;
7134 		config->flow_mreg_c[n++] = idx;
7135 		flow_list_destroy(dev, NULL, flow_idx);
7136 	}
7137 	for (; n < MLX5_MREG_C_NUM; ++n)
7138 		config->flow_mreg_c[n] = REG_NON;
7139 	return 0;
7140 }
7141 
7142 /**
7143  * Dump flow raw hw data to file
7144  *
7145  * @param[in] dev
7146  *    The pointer to Ethernet device.
7147  * @param[in] file
7148  *   A pointer to a file for output.
7149  * @param[out] error
7150  *   Perform verbose error reporting if not NULL. PMDs initialize this
7151  *   structure in case of error only.
7152  * @return
7153  *   0 on success, a nagative value otherwise.
7154  */
7155 int
7156 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7157 		   FILE *file,
7158 		   struct rte_flow_error *error __rte_unused)
7159 {
7160 	struct mlx5_priv *priv = dev->data->dev_private;
7161 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7162 
7163 	if (!priv->config.dv_flow_en) {
7164 		if (fputs("device dv flow disabled\n", file) <= 0)
7165 			return -errno;
7166 		return -ENOTSUP;
7167 	}
7168 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7169 				       sh->tx_domain, file);
7170 }
7171 
7172 /**
7173  * Get aged-out flows.
7174  *
7175  * @param[in] dev
7176  *   Pointer to the Ethernet device structure.
7177  * @param[in] context
7178  *   The address of an array of pointers to the aged-out flows contexts.
7179  * @param[in] nb_countexts
7180  *   The length of context array pointers.
7181  * @param[out] error
7182  *   Perform verbose error reporting if not NULL. Initialized in case of
7183  *   error only.
7184  *
7185  * @return
7186  *   how many contexts get in success, otherwise negative errno value.
7187  *   if nb_contexts is 0, return the amount of all aged contexts.
7188  *   if nb_contexts is not 0 , return the amount of aged flows reported
7189  *   in the context array.
7190  */
7191 int
7192 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7193 			uint32_t nb_contexts, struct rte_flow_error *error)
7194 {
7195 	const struct mlx5_flow_driver_ops *fops;
7196 	struct rte_flow_attr attr = { .transfer = 0 };
7197 
7198 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7199 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7200 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7201 						    error);
7202 	}
7203 	DRV_LOG(ERR,
7204 		"port %u get aged flows is not supported.",
7205 		 dev->data->port_id);
7206 	return -ENOTSUP;
7207 }
7208 
7209 /* Wrapper for driver action_validate op callback */
7210 static int
7211 flow_drv_action_validate(struct rte_eth_dev *dev,
7212 			 const struct rte_flow_shared_action_conf *conf,
7213 			 const struct rte_flow_action *action,
7214 			 const struct mlx5_flow_driver_ops *fops,
7215 			 struct rte_flow_error *error)
7216 {
7217 	static const char err_msg[] = "shared action validation unsupported";
7218 
7219 	if (!fops->action_validate) {
7220 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7221 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7222 				   NULL, err_msg);
7223 		return -rte_errno;
7224 	}
7225 	return fops->action_validate(dev, conf, action, error);
7226 }
7227 
7228 /**
7229  * Destroys the shared action by handle.
7230  *
7231  * @param dev
7232  *   Pointer to Ethernet device structure.
7233  * @param[in] action
7234  *   Handle for the shared action to be destroyed.
7235  * @param[out] error
7236  *   Perform verbose error reporting if not NULL. PMDs initialize this
7237  *   structure in case of error only.
7238  *
7239  * @return
7240  *   0 on success, a negative errno value otherwise and rte_errno is set.
7241  *
7242  * @note: wrapper for driver action_create op callback.
7243  */
7244 static int
7245 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7246 			   struct rte_flow_shared_action *action,
7247 			   struct rte_flow_error *error)
7248 {
7249 	static const char err_msg[] = "shared action destruction unsupported";
7250 	struct rte_flow_attr attr = { .transfer = 0 };
7251 	const struct mlx5_flow_driver_ops *fops =
7252 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7253 
7254 	if (!fops->action_destroy) {
7255 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7256 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7257 				   NULL, err_msg);
7258 		return -rte_errno;
7259 	}
7260 	return fops->action_destroy(dev, action, error);
7261 }
7262 
7263 /* Wrapper for driver action_destroy op callback */
7264 static int
7265 flow_drv_action_update(struct rte_eth_dev *dev,
7266 		       struct rte_flow_shared_action *action,
7267 		       const void *action_conf,
7268 		       const struct mlx5_flow_driver_ops *fops,
7269 		       struct rte_flow_error *error)
7270 {
7271 	static const char err_msg[] = "shared action update unsupported";
7272 
7273 	if (!fops->action_update) {
7274 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7275 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7276 				   NULL, err_msg);
7277 		return -rte_errno;
7278 	}
7279 	return fops->action_update(dev, action, action_conf, error);
7280 }
7281 
7282 /* Wrapper for driver action_destroy op callback */
7283 static int
7284 flow_drv_action_query(struct rte_eth_dev *dev,
7285 		      const struct rte_flow_shared_action *action,
7286 		      void *data,
7287 		      const struct mlx5_flow_driver_ops *fops,
7288 		      struct rte_flow_error *error)
7289 {
7290 	static const char err_msg[] = "shared action query unsupported";
7291 
7292 	if (!fops->action_query) {
7293 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7294 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7295 				   NULL, err_msg);
7296 		return -rte_errno;
7297 	}
7298 	return fops->action_query(dev, action, data, error);
7299 }
7300 
7301 /**
7302  * Create shared action for reuse in multiple flow rules.
7303  *
7304  * @param dev
7305  *   Pointer to Ethernet device structure.
7306  * @param[in] action
7307  *   Action configuration for shared action creation.
7308  * @param[out] error
7309  *   Perform verbose error reporting if not NULL. PMDs initialize this
7310  *   structure in case of error only.
7311  * @return
7312  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7313  */
7314 static struct rte_flow_shared_action *
7315 mlx5_shared_action_create(struct rte_eth_dev *dev,
7316 			  const struct rte_flow_shared_action_conf *conf,
7317 			  const struct rte_flow_action *action,
7318 			  struct rte_flow_error *error)
7319 {
7320 	static const char err_msg[] = "shared action creation unsupported";
7321 	struct rte_flow_attr attr = { .transfer = 0 };
7322 	const struct mlx5_flow_driver_ops *fops =
7323 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7324 
7325 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7326 		return NULL;
7327 	if (!fops->action_create) {
7328 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7329 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7330 				   NULL, err_msg);
7331 		return NULL;
7332 	}
7333 	return fops->action_create(dev, conf, action, error);
7334 }
7335 
7336 /**
7337  * Updates inplace the shared action configuration pointed by *action* handle
7338  * with the configuration provided as *action* argument.
7339  * The update of the shared action configuration effects all flow rules reusing
7340  * the action via handle.
7341  *
7342  * @param dev
7343  *   Pointer to Ethernet device structure.
7344  * @param[in] shared_action
7345  *   Handle for the shared action to be updated.
7346  * @param[in] action
7347  *   Action specification used to modify the action pointed by handle.
7348  *   *action* should be of same type with the action pointed by the *action*
7349  *   handle argument, otherwise considered as invalid.
7350  * @param[out] error
7351  *   Perform verbose error reporting if not NULL. PMDs initialize this
7352  *   structure in case of error only.
7353  *
7354  * @return
7355  *   0 on success, a negative errno value otherwise and rte_errno is set.
7356  */
7357 static int
7358 mlx5_shared_action_update(struct rte_eth_dev *dev,
7359 		struct rte_flow_shared_action *shared_action,
7360 		const struct rte_flow_action *action,
7361 		struct rte_flow_error *error)
7362 {
7363 	struct rte_flow_attr attr = { .transfer = 0 };
7364 	const struct mlx5_flow_driver_ops *fops =
7365 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7366 	int ret;
7367 
7368 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7369 	if (ret)
7370 		return ret;
7371 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7372 				      error);
7373 }
7374 
7375 /**
7376  * Query the shared action by handle.
7377  *
7378  * This function allows retrieving action-specific data such as counters.
7379  * Data is gathered by special action which may be present/referenced in
7380  * more than one flow rule definition.
7381  *
7382  * \see RTE_FLOW_ACTION_TYPE_COUNT
7383  *
7384  * @param dev
7385  *   Pointer to Ethernet device structure.
7386  * @param[in] action
7387  *   Handle for the shared action to query.
7388  * @param[in, out] data
7389  *   Pointer to storage for the associated query data type.
7390  * @param[out] error
7391  *   Perform verbose error reporting if not NULL. PMDs initialize this
7392  *   structure in case of error only.
7393  *
7394  * @return
7395  *   0 on success, a negative errno value otherwise and rte_errno is set.
7396  */
7397 static int
7398 mlx5_shared_action_query(struct rte_eth_dev *dev,
7399 			 const struct rte_flow_shared_action *action,
7400 			 void *data,
7401 			 struct rte_flow_error *error)
7402 {
7403 	struct rte_flow_attr attr = { .transfer = 0 };
7404 	const struct mlx5_flow_driver_ops *fops =
7405 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7406 
7407 	return flow_drv_action_query(dev, action, data, fops, error);
7408 }
7409 
7410 /**
7411  * Destroy all shared actions.
7412  *
7413  * @param dev
7414  *   Pointer to Ethernet device.
7415  *
7416  * @return
7417  *   0 on success, a negative errno value otherwise and rte_errno is set.
7418  */
7419 int
7420 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7421 {
7422 	struct rte_flow_error error;
7423 	struct mlx5_priv *priv = dev->data->dev_private;
7424 	struct mlx5_shared_action_rss *shared_rss;
7425 	int ret = 0;
7426 	uint32_t idx;
7427 
7428 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7429 		      priv->rss_shared_actions, idx, shared_rss, next) {
7430 		ret |= mlx5_shared_action_destroy(dev,
7431 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7432 	}
7433 	return ret;
7434 }
7435 
7436 #ifndef HAVE_MLX5DV_DR
7437 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7438 #else
7439 #define MLX5_DOMAIN_SYNC_FLOW \
7440 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7441 #endif
7442 
7443 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7444 {
7445 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7446 	const struct mlx5_flow_driver_ops *fops;
7447 	int ret;
7448 	struct rte_flow_attr attr = { .transfer = 0 };
7449 
7450 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7451 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7452 	if (ret > 0)
7453 		ret = -ret;
7454 	return ret;
7455 }
7456 
7457 /**
7458  * tunnel offload functionalilty is defined for DV environment only
7459  */
7460 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7461 __extension__
7462 union tunnel_offload_mark {
7463 	uint32_t val;
7464 	struct {
7465 		uint32_t app_reserve:8;
7466 		uint32_t table_id:15;
7467 		uint32_t transfer:1;
7468 		uint32_t _unused_:8;
7469 	};
7470 };
7471 
7472 static bool
7473 mlx5_access_tunnel_offload_db
7474 	(struct rte_eth_dev *dev,
7475 	 bool (*match)(struct rte_eth_dev *,
7476 		       struct mlx5_flow_tunnel *, const void *),
7477 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7478 	 void (*miss)(struct rte_eth_dev *, void *),
7479 	 void *ctx, bool lock_op);
7480 
7481 static int
7482 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7483 			     struct rte_flow *flow,
7484 			     const struct rte_flow_attr *attr,
7485 			     const struct rte_flow_action *app_actions,
7486 			     uint32_t flow_idx,
7487 			     struct tunnel_default_miss_ctx *ctx,
7488 			     struct rte_flow_error *error)
7489 {
7490 	struct mlx5_priv *priv = dev->data->dev_private;
7491 	struct mlx5_flow *dev_flow;
7492 	struct rte_flow_attr miss_attr = *attr;
7493 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7494 	const struct rte_flow_item miss_items[2] = {
7495 		{
7496 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7497 			.spec = NULL,
7498 			.last = NULL,
7499 			.mask = NULL
7500 		},
7501 		{
7502 			.type = RTE_FLOW_ITEM_TYPE_END,
7503 			.spec = NULL,
7504 			.last = NULL,
7505 			.mask = NULL
7506 		}
7507 	};
7508 	union tunnel_offload_mark mark_id;
7509 	struct rte_flow_action_mark miss_mark;
7510 	struct rte_flow_action miss_actions[3] = {
7511 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7512 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7513 	};
7514 	const struct rte_flow_action_jump *jump_data;
7515 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7516 	struct flow_grp_info grp_info = {
7517 		.external = 1,
7518 		.transfer = attr->transfer,
7519 		.fdb_def_rule = !!priv->fdb_def_rule,
7520 		.std_tbl_fix = 0,
7521 	};
7522 	int ret;
7523 
7524 	if (!attr->transfer) {
7525 		uint32_t q_size;
7526 
7527 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7528 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7529 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7530 					 0, SOCKET_ID_ANY);
7531 		if (!ctx->queue)
7532 			return rte_flow_error_set
7533 				(error, ENOMEM,
7534 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7535 				NULL, "invalid default miss RSS");
7536 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7537 		ctx->action_rss.level = 0,
7538 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7539 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7540 		ctx->action_rss.queue_num = priv->reta_idx_n,
7541 		ctx->action_rss.key = priv->rss_conf.rss_key,
7542 		ctx->action_rss.queue = ctx->queue;
7543 		if (!priv->reta_idx_n || !priv->rxqs_n)
7544 			return rte_flow_error_set
7545 				(error, EINVAL,
7546 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7547 				NULL, "invalid port configuration");
7548 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7549 			ctx->action_rss.types = 0;
7550 		for (i = 0; i != priv->reta_idx_n; ++i)
7551 			ctx->queue[i] = (*priv->reta_idx)[i];
7552 	} else {
7553 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7554 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7555 	}
7556 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7557 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7558 	jump_data = app_actions->conf;
7559 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7560 	miss_attr.group = jump_data->group;
7561 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7562 				       &flow_table, &grp_info, error);
7563 	if (ret)
7564 		return rte_flow_error_set(error, EINVAL,
7565 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7566 					  NULL, "invalid tunnel id");
7567 	mark_id.app_reserve = 0;
7568 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7569 	mark_id.transfer = !!attr->transfer;
7570 	mark_id._unused_ = 0;
7571 	miss_mark.id = mark_id.val;
7572 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7573 				    miss_items, miss_actions, flow_idx, error);
7574 	if (!dev_flow)
7575 		return -rte_errno;
7576 	dev_flow->flow = flow;
7577 	dev_flow->external = true;
7578 	dev_flow->tunnel = tunnel;
7579 	/* Subflow object was created, we must include one in the list. */
7580 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7581 		      dev_flow->handle, next);
7582 	DRV_LOG(DEBUG,
7583 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7584 		dev->data->port_id, tunnel->app_tunnel.type,
7585 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7586 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7587 				  miss_actions, error);
7588 	if (!ret)
7589 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7590 						  error);
7591 
7592 	return ret;
7593 }
7594 
7595 static const struct mlx5_flow_tbl_data_entry  *
7596 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7597 {
7598 	struct mlx5_priv *priv = dev->data->dev_private;
7599 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7600 	struct mlx5_hlist_entry *he;
7601 	union tunnel_offload_mark mbits = { .val = mark };
7602 	union mlx5_flow_tbl_key table_key = {
7603 		{
7604 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7605 			.dummy = 0,
7606 			.domain = !!mbits.transfer,
7607 			.direction = 0,
7608 		}
7609 	};
7610 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7611 	return he ?
7612 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7613 }
7614 
7615 static void
7616 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7617 				   struct mlx5_hlist_entry *entry)
7618 {
7619 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7620 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7621 
7622 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7623 			tunnel_flow_tbl_to_id(tte->flow_table));
7624 	mlx5_free(tte);
7625 }
7626 
7627 static int
7628 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7629 				  struct mlx5_hlist_entry *entry,
7630 				  uint64_t key, void *cb_ctx __rte_unused)
7631 {
7632 	union tunnel_tbl_key tbl = {
7633 		.val = key,
7634 	};
7635 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7636 
7637 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7638 }
7639 
7640 static struct mlx5_hlist_entry *
7641 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7642 				   void *ctx __rte_unused)
7643 {
7644 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7645 	struct tunnel_tbl_entry *tte;
7646 	union tunnel_tbl_key tbl = {
7647 		.val = key,
7648 	};
7649 
7650 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7651 			  sizeof(*tte), 0,
7652 			  SOCKET_ID_ANY);
7653 	if (!tte)
7654 		goto err;
7655 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7656 			  &tte->flow_table);
7657 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7658 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7659 			tte->flow_table);
7660 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7661 				tte->flow_table);
7662 		goto err;
7663 	} else if (!tte->flow_table) {
7664 		goto err;
7665 	}
7666 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7667 	tte->tunnel_id = tbl.tunnel_id;
7668 	tte->group = tbl.group;
7669 	return &tte->hash;
7670 err:
7671 	if (tte)
7672 		mlx5_free(tte);
7673 	return NULL;
7674 }
7675 
7676 static uint32_t
7677 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7678 				const struct mlx5_flow_tunnel *tunnel,
7679 				uint32_t group, uint32_t *table,
7680 				struct rte_flow_error *error)
7681 {
7682 	struct mlx5_hlist_entry *he;
7683 	struct tunnel_tbl_entry *tte;
7684 	union tunnel_tbl_key key = {
7685 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7686 		.group = group
7687 	};
7688 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7689 	struct mlx5_hlist *group_hash;
7690 
7691 	group_hash = tunnel ? tunnel->groups : thub->groups;
7692 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7693 	if (!he)
7694 		return rte_flow_error_set(error, EINVAL,
7695 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7696 					  NULL,
7697 					  "tunnel group index not supported");
7698 	tte = container_of(he, typeof(*tte), hash);
7699 	*table = tte->flow_table;
7700 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7701 		dev->data->port_id, key.tunnel_id, group, *table);
7702 	return 0;
7703 }
7704 
7705 static void
7706 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7707 		      struct mlx5_flow_tunnel *tunnel)
7708 {
7709 	struct mlx5_priv *priv = dev->data->dev_private;
7710 	struct mlx5_indexed_pool *ipool;
7711 
7712 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7713 		dev->data->port_id, tunnel->tunnel_id);
7714 	LIST_REMOVE(tunnel, chain);
7715 	mlx5_hlist_destroy(tunnel->groups);
7716 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7717 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7718 }
7719 
7720 static bool
7721 mlx5_access_tunnel_offload_db
7722 	(struct rte_eth_dev *dev,
7723 	 bool (*match)(struct rte_eth_dev *,
7724 		       struct mlx5_flow_tunnel *, const void *),
7725 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7726 	 void (*miss)(struct rte_eth_dev *, void *),
7727 	 void *ctx, bool lock_op)
7728 {
7729 	bool verdict = false;
7730 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7731 	struct mlx5_flow_tunnel *tunnel;
7732 
7733 	rte_spinlock_lock(&thub->sl);
7734 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7735 		verdict = match(dev, tunnel, (const void *)ctx);
7736 		if (verdict)
7737 			break;
7738 	}
7739 	if (!lock_op)
7740 		rte_spinlock_unlock(&thub->sl);
7741 	if (verdict && hit)
7742 		hit(dev, tunnel, ctx);
7743 	if (!verdict && miss)
7744 		miss(dev, ctx);
7745 	if (lock_op)
7746 		rte_spinlock_unlock(&thub->sl);
7747 
7748 	return verdict;
7749 }
7750 
7751 struct tunnel_db_find_tunnel_id_ctx {
7752 	uint32_t tunnel_id;
7753 	struct mlx5_flow_tunnel *tunnel;
7754 };
7755 
7756 static bool
7757 find_tunnel_id_match(struct rte_eth_dev *dev,
7758 		     struct mlx5_flow_tunnel *tunnel, const void *x)
7759 {
7760 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7761 
7762 	RTE_SET_USED(dev);
7763 	return tunnel->tunnel_id == ctx->tunnel_id;
7764 }
7765 
7766 static void
7767 find_tunnel_id_hit(struct rte_eth_dev *dev,
7768 		   struct mlx5_flow_tunnel *tunnel, void *x)
7769 {
7770 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7771 	RTE_SET_USED(dev);
7772 	ctx->tunnel = tunnel;
7773 }
7774 
7775 static struct mlx5_flow_tunnel *
7776 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7777 {
7778 	struct tunnel_db_find_tunnel_id_ctx ctx = {
7779 		.tunnel_id = id,
7780 	};
7781 
7782 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7783 				      find_tunnel_id_hit, NULL, &ctx, true);
7784 
7785 	return ctx.tunnel;
7786 }
7787 
7788 static struct mlx5_flow_tunnel *
7789 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7790 			  const struct rte_flow_tunnel *app_tunnel)
7791 {
7792 	struct mlx5_priv *priv = dev->data->dev_private;
7793 	struct mlx5_indexed_pool *ipool;
7794 	struct mlx5_flow_tunnel *tunnel;
7795 	uint32_t id;
7796 
7797 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7798 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7799 	if (!tunnel)
7800 		return NULL;
7801 	if (id >= MLX5_MAX_TUNNELS) {
7802 		mlx5_ipool_free(ipool, id);
7803 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7804 		return NULL;
7805 	}
7806 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7807 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7808 					   mlx5_flow_tunnel_grp2tbl_match_cb,
7809 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7810 	if (!tunnel->groups) {
7811 		mlx5_ipool_free(ipool, id);
7812 		return NULL;
7813 	}
7814 	tunnel->groups->ctx = priv->sh;
7815 	/* initiate new PMD tunnel */
7816 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7817 	tunnel->tunnel_id = id;
7818 	tunnel->action.type = (typeof(tunnel->action.type))
7819 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7820 	tunnel->action.conf = tunnel;
7821 	tunnel->item.type = (typeof(tunnel->item.type))
7822 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7823 	tunnel->item.spec = tunnel;
7824 	tunnel->item.last = NULL;
7825 	tunnel->item.mask = NULL;
7826 
7827 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7828 		dev->data->port_id, tunnel->tunnel_id);
7829 
7830 	return tunnel;
7831 }
7832 
7833 struct tunnel_db_get_tunnel_ctx {
7834 	const struct rte_flow_tunnel *app_tunnel;
7835 	struct mlx5_flow_tunnel *tunnel;
7836 };
7837 
7838 static bool get_tunnel_match(struct rte_eth_dev *dev,
7839 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7840 {
7841 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
7842 
7843 	RTE_SET_USED(dev);
7844 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7845 		       sizeof(*ctx->app_tunnel));
7846 }
7847 
7848 static void get_tunnel_hit(struct rte_eth_dev *dev,
7849 			   struct mlx5_flow_tunnel *tunnel, void *x)
7850 {
7851 	/* called under tunnel spinlock protection */
7852 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7853 
7854 	RTE_SET_USED(dev);
7855 	tunnel->refctn++;
7856 	ctx->tunnel = tunnel;
7857 }
7858 
7859 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7860 {
7861 	/* called under tunnel spinlock protection */
7862 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7863 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7864 
7865 	rte_spinlock_unlock(&thub->sl);
7866 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7867 	rte_spinlock_lock(&thub->sl);
7868 	if (ctx->tunnel) {
7869 		ctx->tunnel->refctn = 1;
7870 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7871 	}
7872 }
7873 
7874 
7875 static int
7876 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7877 		     const struct rte_flow_tunnel *app_tunnel,
7878 		     struct mlx5_flow_tunnel **tunnel)
7879 {
7880 	struct tunnel_db_get_tunnel_ctx ctx = {
7881 		.app_tunnel = app_tunnel,
7882 	};
7883 
7884 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7885 				      get_tunnel_miss, &ctx, true);
7886 	*tunnel = ctx.tunnel;
7887 	return ctx.tunnel ? 0 : -ENOMEM;
7888 }
7889 
7890 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7891 {
7892 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7893 
7894 	if (!thub)
7895 		return;
7896 	if (!LIST_EMPTY(&thub->tunnels))
7897 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
7898 	mlx5_hlist_destroy(thub->groups);
7899 	mlx5_free(thub);
7900 }
7901 
7902 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7903 {
7904 	int err;
7905 	struct mlx5_flow_tunnel_hub *thub;
7906 
7907 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7908 			   0, SOCKET_ID_ANY);
7909 	if (!thub)
7910 		return -ENOMEM;
7911 	LIST_INIT(&thub->tunnels);
7912 	rte_spinlock_init(&thub->sl);
7913 	thub->groups = mlx5_hlist_create("flow groups",
7914 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
7915 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7916 					 mlx5_flow_tunnel_grp2tbl_match_cb,
7917 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7918 	if (!thub->groups) {
7919 		err = -rte_errno;
7920 		goto err;
7921 	}
7922 	thub->groups->ctx = sh;
7923 	sh->tunnel_hub = thub;
7924 
7925 	return 0;
7926 
7927 err:
7928 	if (thub->groups)
7929 		mlx5_hlist_destroy(thub->groups);
7930 	if (thub)
7931 		mlx5_free(thub);
7932 	return err;
7933 }
7934 
7935 static inline bool
7936 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7937 			  struct rte_flow_tunnel *tunnel,
7938 			  const char *err_msg)
7939 {
7940 	err_msg = NULL;
7941 	if (!is_tunnel_offload_active(dev)) {
7942 		err_msg = "tunnel offload was not activated";
7943 		goto out;
7944 	} else if (!tunnel) {
7945 		err_msg = "no application tunnel";
7946 		goto out;
7947 	}
7948 
7949 	switch (tunnel->type) {
7950 	default:
7951 		err_msg = "unsupported tunnel type";
7952 		goto out;
7953 	case RTE_FLOW_ITEM_TYPE_VXLAN:
7954 		break;
7955 	}
7956 
7957 out:
7958 	return !err_msg;
7959 }
7960 
7961 static int
7962 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7963 		    struct rte_flow_tunnel *app_tunnel,
7964 		    struct rte_flow_action **actions,
7965 		    uint32_t *num_of_actions,
7966 		    struct rte_flow_error *error)
7967 {
7968 	int ret;
7969 	struct mlx5_flow_tunnel *tunnel;
7970 	const char *err_msg = NULL;
7971 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7972 
7973 	if (!verdict)
7974 		return rte_flow_error_set(error, EINVAL,
7975 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7976 					  err_msg);
7977 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7978 	if (ret < 0) {
7979 		return rte_flow_error_set(error, ret,
7980 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7981 					  "failed to initialize pmd tunnel");
7982 	}
7983 	*actions = &tunnel->action;
7984 	*num_of_actions = 1;
7985 	return 0;
7986 }
7987 
7988 static int
7989 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7990 		       struct rte_flow_tunnel *app_tunnel,
7991 		       struct rte_flow_item **items,
7992 		       uint32_t *num_of_items,
7993 		       struct rte_flow_error *error)
7994 {
7995 	int ret;
7996 	struct mlx5_flow_tunnel *tunnel;
7997 	const char *err_msg = NULL;
7998 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7999 
8000 	if (!verdict)
8001 		return rte_flow_error_set(error, EINVAL,
8002 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8003 					  err_msg);
8004 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8005 	if (ret < 0) {
8006 		return rte_flow_error_set(error, ret,
8007 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8008 					  "failed to initialize pmd tunnel");
8009 	}
8010 	*items = &tunnel->item;
8011 	*num_of_items = 1;
8012 	return 0;
8013 }
8014 
8015 struct tunnel_db_element_release_ctx {
8016 	struct rte_flow_item *items;
8017 	struct rte_flow_action *actions;
8018 	uint32_t num_elements;
8019 	struct rte_flow_error *error;
8020 	int ret;
8021 };
8022 
8023 static bool
8024 tunnel_element_release_match(struct rte_eth_dev *dev,
8025 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8026 {
8027 	const struct tunnel_db_element_release_ctx *ctx = x;
8028 
8029 	RTE_SET_USED(dev);
8030 	if (ctx->num_elements != 1)
8031 		return false;
8032 	else if (ctx->items)
8033 		return ctx->items == &tunnel->item;
8034 	else if (ctx->actions)
8035 		return ctx->actions == &tunnel->action;
8036 
8037 	return false;
8038 }
8039 
8040 static void
8041 tunnel_element_release_hit(struct rte_eth_dev *dev,
8042 			   struct mlx5_flow_tunnel *tunnel, void *x)
8043 {
8044 	struct tunnel_db_element_release_ctx *ctx = x;
8045 	ctx->ret = 0;
8046 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8047 		mlx5_flow_tunnel_free(dev, tunnel);
8048 }
8049 
8050 static void
8051 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8052 {
8053 	struct tunnel_db_element_release_ctx *ctx = x;
8054 	RTE_SET_USED(dev);
8055 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8056 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8057 				      "invalid argument");
8058 }
8059 
8060 static int
8061 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8062 		       struct rte_flow_item *pmd_items,
8063 		       uint32_t num_items, struct rte_flow_error *err)
8064 {
8065 	struct tunnel_db_element_release_ctx ctx = {
8066 		.items = pmd_items,
8067 		.actions = NULL,
8068 		.num_elements = num_items,
8069 		.error = err,
8070 	};
8071 
8072 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8073 				      tunnel_element_release_hit,
8074 				      tunnel_element_release_miss, &ctx, false);
8075 
8076 	return ctx.ret;
8077 }
8078 
8079 static int
8080 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8081 			 struct rte_flow_action *pmd_actions,
8082 			 uint32_t num_actions, struct rte_flow_error *err)
8083 {
8084 	struct tunnel_db_element_release_ctx ctx = {
8085 		.items = NULL,
8086 		.actions = pmd_actions,
8087 		.num_elements = num_actions,
8088 		.error = err,
8089 	};
8090 
8091 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8092 				      tunnel_element_release_hit,
8093 				      tunnel_element_release_miss, &ctx, false);
8094 
8095 	return ctx.ret;
8096 }
8097 
8098 static int
8099 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8100 				  struct rte_mbuf *m,
8101 				  struct rte_flow_restore_info *info,
8102 				  struct rte_flow_error *err)
8103 {
8104 	uint64_t ol_flags = m->ol_flags;
8105 	const struct mlx5_flow_tbl_data_entry *tble;
8106 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8107 
8108 	if (!is_tunnel_offload_active(dev)) {
8109 		info->flags = 0;
8110 		return 0;
8111 	}
8112 
8113 	if ((ol_flags & mask) != mask)
8114 		goto err;
8115 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8116 	if (!tble) {
8117 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8118 			dev->data->port_id, m->hash.fdir.hi);
8119 		goto err;
8120 	}
8121 	MLX5_ASSERT(tble->tunnel);
8122 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8123 	info->group_id = tble->group_id;
8124 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8125 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8126 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8127 
8128 	return 0;
8129 
8130 err:
8131 	return rte_flow_error_set(err, EINVAL,
8132 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8133 				  "failed to get restore info");
8134 }
8135 
8136 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8137 static int
8138 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8139 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8140 			   __rte_unused struct rte_flow_action **actions,
8141 			   __rte_unused uint32_t *num_of_actions,
8142 			   __rte_unused struct rte_flow_error *error)
8143 {
8144 	return -ENOTSUP;
8145 }
8146 
8147 static int
8148 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8149 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8150 		       __rte_unused struct rte_flow_item **items,
8151 		       __rte_unused uint32_t *num_of_items,
8152 		       __rte_unused struct rte_flow_error *error)
8153 {
8154 	return -ENOTSUP;
8155 }
8156 
8157 static int
8158 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8159 			      __rte_unused struct rte_flow_item *pmd_items,
8160 			      __rte_unused uint32_t num_items,
8161 			      __rte_unused struct rte_flow_error *err)
8162 {
8163 	return -ENOTSUP;
8164 }
8165 
8166 static int
8167 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8168 				__rte_unused struct rte_flow_action *pmd_action,
8169 				__rte_unused uint32_t num_actions,
8170 				__rte_unused struct rte_flow_error *err)
8171 {
8172 	return -ENOTSUP;
8173 }
8174 
8175 static int
8176 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8177 				  __rte_unused struct rte_mbuf *m,
8178 				  __rte_unused struct rte_flow_restore_info *i,
8179 				  __rte_unused struct rte_flow_error *err)
8180 {
8181 	return -ENOTSUP;
8182 }
8183 
8184 static int
8185 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8186 			     __rte_unused struct rte_flow *flow,
8187 			     __rte_unused const struct rte_flow_attr *attr,
8188 			     __rte_unused const struct rte_flow_action *actions,
8189 			     __rte_unused uint32_t flow_idx,
8190 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8191 			     __rte_unused struct rte_flow_error *error)
8192 {
8193 	return -ENOTSUP;
8194 }
8195 
8196 static struct mlx5_flow_tunnel *
8197 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8198 		    __rte_unused uint32_t id)
8199 {
8200 	return NULL;
8201 }
8202 
8203 static void
8204 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8205 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8206 {
8207 }
8208 
8209 static uint32_t
8210 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8211 				__rte_unused const struct mlx5_flow_tunnel *t,
8212 				__rte_unused uint32_t group,
8213 				__rte_unused uint32_t *table,
8214 				struct rte_flow_error *error)
8215 {
8216 	return rte_flow_error_set(error, ENOTSUP,
8217 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8218 				  "tunnel offload requires DV support");
8219 }
8220 
8221 void
8222 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8223 			__rte_unused  uint16_t port_id)
8224 {
8225 }
8226 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8227