xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 4b61b8774be951c7caeaba2edde27c42f2f4c58a)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rx.h"
32 #include "mlx5_tx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 struct tunnel_default_miss_ctx {
37 	uint16_t *queue;
38 	__extension__
39 	union {
40 		struct rte_flow_action_rss action_rss;
41 		struct rte_flow_action_queue miss_queue;
42 		struct rte_flow_action_jump miss_jump;
43 		uint8_t raw[0];
44 	};
45 };
46 
47 static int
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 			     struct rte_flow *flow,
50 			     const struct rte_flow_attr *attr,
51 			     const struct rte_flow_action *app_actions,
52 			     uint32_t flow_idx,
53 			     struct tunnel_default_miss_ctx *ctx,
54 			     struct rte_flow_error *error);
55 static struct mlx5_flow_tunnel *
56 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
57 static void
58 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
59 static uint32_t
60 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
61 				const struct mlx5_flow_tunnel *tunnel,
62 				uint32_t group, uint32_t *table,
63 				struct rte_flow_error *error);
64 
65 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
66 static void mlx5_flow_pop_thread_workspace(void);
67 
68 
69 /** Device flow drivers. */
70 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
71 
72 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
73 
74 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
75 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
76 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
77 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
78 #endif
79 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
80 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
81 };
82 
83 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
84 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
85 	(const int []){ \
86 		__VA_ARGS__, 0, \
87 	}
88 
89 /** Node object of input graph for mlx5_flow_expand_rss(). */
90 struct mlx5_flow_expand_node {
91 	const int *const next;
92 	/**<
93 	 * List of next node indexes. Index 0 is interpreted as a terminator.
94 	 */
95 	const enum rte_flow_item_type type;
96 	/**< Pattern item type of current node. */
97 	uint64_t rss_types;
98 	/**<
99 	 * RSS types bit-field associated with this node
100 	 * (see ETH_RSS_* definitions).
101 	 */
102 };
103 
104 /** Object returned by mlx5_flow_expand_rss(). */
105 struct mlx5_flow_expand_rss {
106 	uint32_t entries;
107 	/**< Number of entries @p patterns and @p priorities. */
108 	struct {
109 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
110 		uint32_t priority; /**< Priority offset for each expansion. */
111 	} entry[];
112 };
113 
114 static enum rte_flow_item_type
115 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
116 {
117 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
118 	uint16_t ether_type = 0;
119 	uint16_t ether_type_m;
120 	uint8_t ip_next_proto = 0;
121 	uint8_t ip_next_proto_m;
122 
123 	if (item == NULL || item->spec == NULL)
124 		return ret;
125 	switch (item->type) {
126 	case RTE_FLOW_ITEM_TYPE_ETH:
127 		if (item->mask)
128 			ether_type_m = ((const struct rte_flow_item_eth *)
129 						(item->mask))->type;
130 		else
131 			ether_type_m = rte_flow_item_eth_mask.type;
132 		if (ether_type_m != RTE_BE16(0xFFFF))
133 			break;
134 		ether_type = ((const struct rte_flow_item_eth *)
135 				(item->spec))->type;
136 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
139 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
140 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
141 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
142 		else
143 			ret = RTE_FLOW_ITEM_TYPE_END;
144 		break;
145 	case RTE_FLOW_ITEM_TYPE_VLAN:
146 		if (item->mask)
147 			ether_type_m = ((const struct rte_flow_item_vlan *)
148 						(item->mask))->inner_type;
149 		else
150 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
151 		if (ether_type_m != RTE_BE16(0xFFFF))
152 			break;
153 		ether_type = ((const struct rte_flow_item_vlan *)
154 				(item->spec))->inner_type;
155 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
156 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
157 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
158 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
159 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
160 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
161 		else
162 			ret = RTE_FLOW_ITEM_TYPE_END;
163 		break;
164 	case RTE_FLOW_ITEM_TYPE_IPV4:
165 		if (item->mask)
166 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
167 					(item->mask))->hdr.next_proto_id;
168 		else
169 			ip_next_proto_m =
170 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
171 		if (ip_next_proto_m != 0xFF)
172 			break;
173 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
174 				(item->spec))->hdr.next_proto_id;
175 		if (ip_next_proto == IPPROTO_UDP)
176 			ret = RTE_FLOW_ITEM_TYPE_UDP;
177 		else if (ip_next_proto == IPPROTO_TCP)
178 			ret = RTE_FLOW_ITEM_TYPE_TCP;
179 		else if (ip_next_proto == IPPROTO_IP)
180 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
181 		else if (ip_next_proto == IPPROTO_IPV6)
182 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
183 		else
184 			ret = RTE_FLOW_ITEM_TYPE_END;
185 		break;
186 	case RTE_FLOW_ITEM_TYPE_IPV6:
187 		if (item->mask)
188 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
189 						(item->mask))->hdr.proto;
190 		else
191 			ip_next_proto_m =
192 				rte_flow_item_ipv6_mask.hdr.proto;
193 		if (ip_next_proto_m != 0xFF)
194 			break;
195 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
196 				(item->spec))->hdr.proto;
197 		if (ip_next_proto == IPPROTO_UDP)
198 			ret = RTE_FLOW_ITEM_TYPE_UDP;
199 		else if (ip_next_proto == IPPROTO_TCP)
200 			ret = RTE_FLOW_ITEM_TYPE_TCP;
201 		else if (ip_next_proto == IPPROTO_IP)
202 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
203 		else if (ip_next_proto == IPPROTO_IPV6)
204 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
205 		else
206 			ret = RTE_FLOW_ITEM_TYPE_END;
207 		break;
208 	default:
209 		ret = RTE_FLOW_ITEM_TYPE_VOID;
210 		break;
211 	}
212 	return ret;
213 }
214 
215 #define MLX5_RSS_EXP_ELT_N 8
216 
217 /**
218  * Expand RSS flows into several possible flows according to the RSS hash
219  * fields requested and the driver capabilities.
220  *
221  * @param[out] buf
222  *   Buffer to store the result expansion.
223  * @param[in] size
224  *   Buffer size in bytes. If 0, @p buf can be NULL.
225  * @param[in] pattern
226  *   User flow pattern.
227  * @param[in] types
228  *   RSS types to expand (see ETH_RSS_* definitions).
229  * @param[in] graph
230  *   Input graph to expand @p pattern according to @p types.
231  * @param[in] graph_root_index
232  *   Index of root node in @p graph, typically 0.
233  *
234  * @return
235  *   A positive value representing the size of @p buf in bytes regardless of
236  *   @p size on success, a negative errno value otherwise and rte_errno is
237  *   set, the following errors are defined:
238  *
239  *   -E2BIG: graph-depth @p graph is too deep.
240  */
241 static int
242 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
243 		     const struct rte_flow_item *pattern, uint64_t types,
244 		     const struct mlx5_flow_expand_node graph[],
245 		     int graph_root_index)
246 {
247 	const struct rte_flow_item *item;
248 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
249 	const int *next_node;
250 	const int *stack[MLX5_RSS_EXP_ELT_N];
251 	int stack_pos = 0;
252 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
253 	unsigned int i;
254 	size_t lsize;
255 	size_t user_pattern_size = 0;
256 	void *addr = NULL;
257 	const struct mlx5_flow_expand_node *next = NULL;
258 	struct rte_flow_item missed_item;
259 	int missed = 0;
260 	int elt = 0;
261 	const struct rte_flow_item *last_item = NULL;
262 
263 	memset(&missed_item, 0, sizeof(missed_item));
264 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
265 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
266 	if (lsize <= size) {
267 		buf->entry[0].priority = 0;
268 		buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
269 		buf->entries = 0;
270 		addr = buf->entry[0].pattern;
271 	}
272 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
273 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
274 			last_item = item;
275 		for (i = 0; node->next && node->next[i]; ++i) {
276 			next = &graph[node->next[i]];
277 			if (next->type == item->type)
278 				break;
279 		}
280 		if (next)
281 			node = next;
282 		user_pattern_size += sizeof(*item);
283 	}
284 	user_pattern_size += sizeof(*item); /* Handle END item. */
285 	lsize += user_pattern_size;
286 	/* Copy the user pattern in the first entry of the buffer. */
287 	if (lsize <= size) {
288 		rte_memcpy(addr, pattern, user_pattern_size);
289 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
290 		buf->entries = 1;
291 	}
292 	/* Start expanding. */
293 	memset(flow_items, 0, sizeof(flow_items));
294 	user_pattern_size -= sizeof(*item);
295 	/*
296 	 * Check if the last valid item has spec set, need complete pattern,
297 	 * and the pattern can be used for expansion.
298 	 */
299 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
300 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
301 		/* Item type END indicates expansion is not required. */
302 		return lsize;
303 	}
304 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
305 		next = NULL;
306 		missed = 1;
307 		for (i = 0; node->next && node->next[i]; ++i) {
308 			next = &graph[node->next[i]];
309 			if (next->type == missed_item.type) {
310 				flow_items[0].type = missed_item.type;
311 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
312 				break;
313 			}
314 			next = NULL;
315 		}
316 	}
317 	if (next && missed) {
318 		elt = 2; /* missed item + item end. */
319 		node = next;
320 		lsize += elt * sizeof(*item) + user_pattern_size;
321 		if ((node->rss_types & types) && lsize <= size) {
322 			buf->entry[buf->entries].priority = 1;
323 			buf->entry[buf->entries].pattern = addr;
324 			buf->entries++;
325 			rte_memcpy(addr, buf->entry[0].pattern,
326 				   user_pattern_size);
327 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
328 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
329 			addr = (void *)(((uintptr_t)addr) +
330 					elt * sizeof(*item));
331 		}
332 	}
333 	memset(flow_items, 0, sizeof(flow_items));
334 	next_node = node->next;
335 	stack[stack_pos] = next_node;
336 	node = next_node ? &graph[*next_node] : NULL;
337 	while (node) {
338 		flow_items[stack_pos].type = node->type;
339 		if (node->rss_types & types) {
340 			/*
341 			 * compute the number of items to copy from the
342 			 * expansion and copy it.
343 			 * When the stack_pos is 0, there are 1 element in it,
344 			 * plus the addition END item.
345 			 */
346 			elt = stack_pos + 2;
347 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
348 			lsize += elt * sizeof(*item) + user_pattern_size;
349 			if (lsize <= size) {
350 				size_t n = elt * sizeof(*item);
351 
352 				buf->entry[buf->entries].priority =
353 					stack_pos + 1 + missed;
354 				buf->entry[buf->entries].pattern = addr;
355 				buf->entries++;
356 				rte_memcpy(addr, buf->entry[0].pattern,
357 					   user_pattern_size);
358 				addr = (void *)(((uintptr_t)addr) +
359 						user_pattern_size);
360 				rte_memcpy(addr, &missed_item,
361 					   missed * sizeof(*item));
362 				addr = (void *)(((uintptr_t)addr) +
363 					missed * sizeof(*item));
364 				rte_memcpy(addr, flow_items, n);
365 				addr = (void *)(((uintptr_t)addr) + n);
366 			}
367 		}
368 		/* Go deeper. */
369 		if (node->next) {
370 			next_node = node->next;
371 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
372 				rte_errno = E2BIG;
373 				return -rte_errno;
374 			}
375 			stack[stack_pos] = next_node;
376 		} else if (*(next_node + 1)) {
377 			/* Follow up with the next possibility. */
378 			++next_node;
379 		} else {
380 			/* Move to the next path. */
381 			if (stack_pos)
382 				next_node = stack[--stack_pos];
383 			next_node++;
384 			stack[stack_pos] = next_node;
385 		}
386 		node = *next_node ? &graph[*next_node] : NULL;
387 	};
388 	return lsize;
389 }
390 
391 enum mlx5_expansion {
392 	MLX5_EXPANSION_ROOT,
393 	MLX5_EXPANSION_ROOT_OUTER,
394 	MLX5_EXPANSION_ROOT_ETH_VLAN,
395 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
396 	MLX5_EXPANSION_OUTER_ETH,
397 	MLX5_EXPANSION_OUTER_ETH_VLAN,
398 	MLX5_EXPANSION_OUTER_VLAN,
399 	MLX5_EXPANSION_OUTER_IPV4,
400 	MLX5_EXPANSION_OUTER_IPV4_UDP,
401 	MLX5_EXPANSION_OUTER_IPV4_TCP,
402 	MLX5_EXPANSION_OUTER_IPV6,
403 	MLX5_EXPANSION_OUTER_IPV6_UDP,
404 	MLX5_EXPANSION_OUTER_IPV6_TCP,
405 	MLX5_EXPANSION_VXLAN,
406 	MLX5_EXPANSION_VXLAN_GPE,
407 	MLX5_EXPANSION_GRE,
408 	MLX5_EXPANSION_MPLS,
409 	MLX5_EXPANSION_ETH,
410 	MLX5_EXPANSION_ETH_VLAN,
411 	MLX5_EXPANSION_VLAN,
412 	MLX5_EXPANSION_IPV4,
413 	MLX5_EXPANSION_IPV4_UDP,
414 	MLX5_EXPANSION_IPV4_TCP,
415 	MLX5_EXPANSION_IPV6,
416 	MLX5_EXPANSION_IPV6_UDP,
417 	MLX5_EXPANSION_IPV6_TCP,
418 };
419 
420 /** Supported expansion of items. */
421 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
422 	[MLX5_EXPANSION_ROOT] = {
423 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
424 						  MLX5_EXPANSION_IPV4,
425 						  MLX5_EXPANSION_IPV6),
426 		.type = RTE_FLOW_ITEM_TYPE_END,
427 	},
428 	[MLX5_EXPANSION_ROOT_OUTER] = {
429 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
430 						  MLX5_EXPANSION_OUTER_IPV4,
431 						  MLX5_EXPANSION_OUTER_IPV6),
432 		.type = RTE_FLOW_ITEM_TYPE_END,
433 	},
434 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
435 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
436 		.type = RTE_FLOW_ITEM_TYPE_END,
437 	},
438 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
439 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
440 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
441 		.type = RTE_FLOW_ITEM_TYPE_END,
442 	},
443 	[MLX5_EXPANSION_OUTER_ETH] = {
444 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
445 						  MLX5_EXPANSION_OUTER_IPV6,
446 						  MLX5_EXPANSION_MPLS),
447 		.type = RTE_FLOW_ITEM_TYPE_ETH,
448 		.rss_types = 0,
449 	},
450 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
451 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
452 		.type = RTE_FLOW_ITEM_TYPE_ETH,
453 		.rss_types = 0,
454 	},
455 	[MLX5_EXPANSION_OUTER_VLAN] = {
456 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
457 						  MLX5_EXPANSION_OUTER_IPV6),
458 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
459 	},
460 	[MLX5_EXPANSION_OUTER_IPV4] = {
461 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
462 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
463 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
464 			 MLX5_EXPANSION_GRE,
465 			 MLX5_EXPANSION_IPV4,
466 			 MLX5_EXPANSION_IPV6),
467 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
468 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
469 			ETH_RSS_NONFRAG_IPV4_OTHER,
470 	},
471 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
472 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
473 						  MLX5_EXPANSION_VXLAN_GPE),
474 		.type = RTE_FLOW_ITEM_TYPE_UDP,
475 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
478 		.type = RTE_FLOW_ITEM_TYPE_TCP,
479 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
480 	},
481 	[MLX5_EXPANSION_OUTER_IPV6] = {
482 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
483 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
484 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
485 			 MLX5_EXPANSION_IPV4,
486 			 MLX5_EXPANSION_IPV6,
487 			 MLX5_EXPANSION_GRE),
488 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
489 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
490 			ETH_RSS_NONFRAG_IPV6_OTHER,
491 	},
492 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
493 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
494 						  MLX5_EXPANSION_VXLAN_GPE),
495 		.type = RTE_FLOW_ITEM_TYPE_UDP,
496 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
497 	},
498 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
499 		.type = RTE_FLOW_ITEM_TYPE_TCP,
500 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
501 	},
502 	[MLX5_EXPANSION_VXLAN] = {
503 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
504 						  MLX5_EXPANSION_IPV4,
505 						  MLX5_EXPANSION_IPV6),
506 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
507 	},
508 	[MLX5_EXPANSION_VXLAN_GPE] = {
509 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
510 						  MLX5_EXPANSION_IPV4,
511 						  MLX5_EXPANSION_IPV6),
512 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
513 	},
514 	[MLX5_EXPANSION_GRE] = {
515 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
516 						  MLX5_EXPANSION_IPV6),
517 		.type = RTE_FLOW_ITEM_TYPE_GRE,
518 	},
519 	[MLX5_EXPANSION_MPLS] = {
520 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
521 						  MLX5_EXPANSION_IPV6),
522 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
523 	},
524 	[MLX5_EXPANSION_ETH] = {
525 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
526 						  MLX5_EXPANSION_IPV6),
527 		.type = RTE_FLOW_ITEM_TYPE_ETH,
528 	},
529 	[MLX5_EXPANSION_ETH_VLAN] = {
530 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
531 		.type = RTE_FLOW_ITEM_TYPE_ETH,
532 	},
533 	[MLX5_EXPANSION_VLAN] = {
534 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 						  MLX5_EXPANSION_IPV6),
536 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
537 	},
538 	[MLX5_EXPANSION_IPV4] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
540 						  MLX5_EXPANSION_IPV4_TCP),
541 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
542 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
543 			ETH_RSS_NONFRAG_IPV4_OTHER,
544 	},
545 	[MLX5_EXPANSION_IPV4_UDP] = {
546 		.type = RTE_FLOW_ITEM_TYPE_UDP,
547 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
548 	},
549 	[MLX5_EXPANSION_IPV4_TCP] = {
550 		.type = RTE_FLOW_ITEM_TYPE_TCP,
551 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
552 	},
553 	[MLX5_EXPANSION_IPV6] = {
554 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
555 						  MLX5_EXPANSION_IPV6_TCP),
556 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
557 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
558 			ETH_RSS_NONFRAG_IPV6_OTHER,
559 	},
560 	[MLX5_EXPANSION_IPV6_UDP] = {
561 		.type = RTE_FLOW_ITEM_TYPE_UDP,
562 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
563 	},
564 	[MLX5_EXPANSION_IPV6_TCP] = {
565 		.type = RTE_FLOW_ITEM_TYPE_TCP,
566 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
567 	},
568 };
569 
570 static struct rte_flow_action_handle *
571 mlx5_action_handle_create(struct rte_eth_dev *dev,
572 			  const struct rte_flow_indir_action_conf *conf,
573 			  const struct rte_flow_action *action,
574 			  struct rte_flow_error *error);
575 static int mlx5_action_handle_destroy
576 				(struct rte_eth_dev *dev,
577 				 struct rte_flow_action_handle *handle,
578 				 struct rte_flow_error *error);
579 static int mlx5_action_handle_update
580 				(struct rte_eth_dev *dev,
581 				 struct rte_flow_action_handle *handle,
582 				 const void *update,
583 				 struct rte_flow_error *error);
584 static int mlx5_action_handle_query
585 				(struct rte_eth_dev *dev,
586 				 const struct rte_flow_action_handle *handle,
587 				 void *data,
588 				 struct rte_flow_error *error);
589 static int
590 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
591 		    struct rte_flow_tunnel *app_tunnel,
592 		    struct rte_flow_action **actions,
593 		    uint32_t *num_of_actions,
594 		    struct rte_flow_error *error);
595 static int
596 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
597 		       struct rte_flow_tunnel *app_tunnel,
598 		       struct rte_flow_item **items,
599 		       uint32_t *num_of_items,
600 		       struct rte_flow_error *error);
601 static int
602 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
603 			      struct rte_flow_item *pmd_items,
604 			      uint32_t num_items, struct rte_flow_error *err);
605 static int
606 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
607 				struct rte_flow_action *pmd_actions,
608 				uint32_t num_actions,
609 				struct rte_flow_error *err);
610 static int
611 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
612 				  struct rte_mbuf *m,
613 				  struct rte_flow_restore_info *info,
614 				  struct rte_flow_error *err);
615 
616 static const struct rte_flow_ops mlx5_flow_ops = {
617 	.validate = mlx5_flow_validate,
618 	.create = mlx5_flow_create,
619 	.destroy = mlx5_flow_destroy,
620 	.flush = mlx5_flow_flush,
621 	.isolate = mlx5_flow_isolate,
622 	.query = mlx5_flow_query,
623 	.dev_dump = mlx5_flow_dev_dump,
624 	.get_aged_flows = mlx5_flow_get_aged_flows,
625 	.action_handle_create = mlx5_action_handle_create,
626 	.action_handle_destroy = mlx5_action_handle_destroy,
627 	.action_handle_update = mlx5_action_handle_update,
628 	.action_handle_query = mlx5_action_handle_query,
629 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
630 	.tunnel_match = mlx5_flow_tunnel_match,
631 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
632 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
633 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
634 };
635 
636 /* Tunnel information. */
637 struct mlx5_flow_tunnel_info {
638 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
639 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
640 };
641 
642 static struct mlx5_flow_tunnel_info tunnels_info[] = {
643 	{
644 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
645 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
646 	},
647 	{
648 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
649 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
650 	},
651 	{
652 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
653 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
654 	},
655 	{
656 		.tunnel = MLX5_FLOW_LAYER_GRE,
657 		.ptype = RTE_PTYPE_TUNNEL_GRE,
658 	},
659 	{
660 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
661 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
662 	},
663 	{
664 		.tunnel = MLX5_FLOW_LAYER_MPLS,
665 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
666 	},
667 	{
668 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
669 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
670 	},
671 	{
672 		.tunnel = MLX5_FLOW_LAYER_IPIP,
673 		.ptype = RTE_PTYPE_TUNNEL_IP,
674 	},
675 	{
676 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
677 		.ptype = RTE_PTYPE_TUNNEL_IP,
678 	},
679 	{
680 		.tunnel = MLX5_FLOW_LAYER_GTP,
681 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
682 	},
683 };
684 
685 
686 
687 /**
688  * Translate tag ID to register.
689  *
690  * @param[in] dev
691  *   Pointer to the Ethernet device structure.
692  * @param[in] feature
693  *   The feature that request the register.
694  * @param[in] id
695  *   The request register ID.
696  * @param[out] error
697  *   Error description in case of any.
698  *
699  * @return
700  *   The request register on success, a negative errno
701  *   value otherwise and rte_errno is set.
702  */
703 int
704 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
705 		     enum mlx5_feature_name feature,
706 		     uint32_t id,
707 		     struct rte_flow_error *error)
708 {
709 	struct mlx5_priv *priv = dev->data->dev_private;
710 	struct mlx5_dev_config *config = &priv->config;
711 	enum modify_reg start_reg;
712 	bool skip_mtr_reg = false;
713 
714 	switch (feature) {
715 	case MLX5_HAIRPIN_RX:
716 		return REG_B;
717 	case MLX5_HAIRPIN_TX:
718 		return REG_A;
719 	case MLX5_METADATA_RX:
720 		switch (config->dv_xmeta_en) {
721 		case MLX5_XMETA_MODE_LEGACY:
722 			return REG_B;
723 		case MLX5_XMETA_MODE_META16:
724 			return REG_C_0;
725 		case MLX5_XMETA_MODE_META32:
726 			return REG_C_1;
727 		}
728 		break;
729 	case MLX5_METADATA_TX:
730 		return REG_A;
731 	case MLX5_METADATA_FDB:
732 		switch (config->dv_xmeta_en) {
733 		case MLX5_XMETA_MODE_LEGACY:
734 			return REG_NON;
735 		case MLX5_XMETA_MODE_META16:
736 			return REG_C_0;
737 		case MLX5_XMETA_MODE_META32:
738 			return REG_C_1;
739 		}
740 		break;
741 	case MLX5_FLOW_MARK:
742 		switch (config->dv_xmeta_en) {
743 		case MLX5_XMETA_MODE_LEGACY:
744 			return REG_NON;
745 		case MLX5_XMETA_MODE_META16:
746 			return REG_C_1;
747 		case MLX5_XMETA_MODE_META32:
748 			return REG_C_0;
749 		}
750 		break;
751 	case MLX5_MTR_SFX:
752 		/*
753 		 * If meter color and flow match share one register, flow match
754 		 * should use the meter color register for match.
755 		 */
756 		if (priv->mtr_reg_share)
757 			return priv->mtr_color_reg;
758 		else
759 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
760 			       REG_C_3;
761 	case MLX5_MTR_COLOR:
762 	case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
763 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
764 		return priv->mtr_color_reg;
765 	case MLX5_COPY_MARK:
766 		/*
767 		 * Metadata COPY_MARK register using is in meter suffix sub
768 		 * flow while with meter. It's safe to share the same register.
769 		 */
770 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
771 	case MLX5_APP_TAG:
772 		/*
773 		 * If meter is enable, it will engage the register for color
774 		 * match and flow match. If meter color match is not using the
775 		 * REG_C_2, need to skip the REG_C_x be used by meter color
776 		 * match.
777 		 * If meter is disable, free to use all available registers.
778 		 */
779 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
780 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
781 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
782 		if (id > (uint32_t)(REG_C_7 - start_reg))
783 			return rte_flow_error_set(error, EINVAL,
784 						  RTE_FLOW_ERROR_TYPE_ITEM,
785 						  NULL, "invalid tag id");
786 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
787 			return rte_flow_error_set(error, ENOTSUP,
788 						  RTE_FLOW_ERROR_TYPE_ITEM,
789 						  NULL, "unsupported tag id");
790 		/*
791 		 * This case means meter is using the REG_C_x great than 2.
792 		 * Take care not to conflict with meter color REG_C_x.
793 		 * If the available index REG_C_y >= REG_C_x, skip the
794 		 * color register.
795 		 */
796 		if (skip_mtr_reg && config->flow_mreg_c
797 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
798 			if (id >= (uint32_t)(REG_C_7 - start_reg))
799 				return rte_flow_error_set(error, EINVAL,
800 						       RTE_FLOW_ERROR_TYPE_ITEM,
801 							NULL, "invalid tag id");
802 			if (config->flow_mreg_c
803 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
804 				return config->flow_mreg_c
805 					       [id + 1 + start_reg - REG_C_0];
806 			return rte_flow_error_set(error, ENOTSUP,
807 						  RTE_FLOW_ERROR_TYPE_ITEM,
808 						  NULL, "unsupported tag id");
809 		}
810 		return config->flow_mreg_c[id + start_reg - REG_C_0];
811 	}
812 	MLX5_ASSERT(false);
813 	return rte_flow_error_set(error, EINVAL,
814 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
815 				  NULL, "invalid feature name");
816 }
817 
818 /**
819  * Check extensive flow metadata register support.
820  *
821  * @param dev
822  *   Pointer to rte_eth_dev structure.
823  *
824  * @return
825  *   True if device supports extensive flow metadata register, otherwise false.
826  */
827 bool
828 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
829 {
830 	struct mlx5_priv *priv = dev->data->dev_private;
831 	struct mlx5_dev_config *config = &priv->config;
832 
833 	/*
834 	 * Having available reg_c can be regarded inclusively as supporting
835 	 * extensive flow metadata register, which could mean,
836 	 * - metadata register copy action by modify header.
837 	 * - 16 modify header actions is supported.
838 	 * - reg_c's are preserved across different domain (FDB and NIC) on
839 	 *   packet loopback by flow lookup miss.
840 	 */
841 	return config->flow_mreg_c[2] != REG_NON;
842 }
843 
844 /**
845  * Get the lowest priority.
846  *
847  * @param[in] dev
848  *   Pointer to the Ethernet device structure.
849  * @param[in] attributes
850  *   Pointer to device flow rule attributes.
851  *
852  * @return
853  *   The value of lowest priority of flow.
854  */
855 uint32_t
856 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
857 			  const struct rte_flow_attr *attr)
858 {
859 	struct mlx5_priv *priv = dev->data->dev_private;
860 
861 	if (!attr->group && !attr->transfer)
862 		return priv->config.flow_prio - 2;
863 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
864 }
865 
866 /**
867  * Calculate matcher priority of the flow.
868  *
869  * @param[in] dev
870  *   Pointer to the Ethernet device structure.
871  * @param[in] attr
872  *   Pointer to device flow rule attributes.
873  * @param[in] subpriority
874  *   The priority based on the items.
875  * @return
876  *   The matcher priority of the flow.
877  */
878 uint16_t
879 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
880 			  const struct rte_flow_attr *attr,
881 			  uint32_t subpriority)
882 {
883 	uint16_t priority = (uint16_t)attr->priority;
884 	struct mlx5_priv *priv = dev->data->dev_private;
885 
886 	if (!attr->group && !attr->transfer) {
887 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
888 			priority = priv->config.flow_prio - 1;
889 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
890 	}
891 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
892 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
893 	return priority * 3 + subpriority;
894 }
895 
896 /**
897  * Verify the @p item specifications (spec, last, mask) are compatible with the
898  * NIC capabilities.
899  *
900  * @param[in] item
901  *   Item specification.
902  * @param[in] mask
903  *   @p item->mask or flow default bit-masks.
904  * @param[in] nic_mask
905  *   Bit-masks covering supported fields by the NIC to compare with user mask.
906  * @param[in] size
907  *   Bit-masks size in bytes.
908  * @param[in] range_accepted
909  *   True if range of values is accepted for specific fields, false otherwise.
910  * @param[out] error
911  *   Pointer to error structure.
912  *
913  * @return
914  *   0 on success, a negative errno value otherwise and rte_errno is set.
915  */
916 int
917 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
918 			  const uint8_t *mask,
919 			  const uint8_t *nic_mask,
920 			  unsigned int size,
921 			  bool range_accepted,
922 			  struct rte_flow_error *error)
923 {
924 	unsigned int i;
925 
926 	MLX5_ASSERT(nic_mask);
927 	for (i = 0; i < size; ++i)
928 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
929 			return rte_flow_error_set(error, ENOTSUP,
930 						  RTE_FLOW_ERROR_TYPE_ITEM,
931 						  item,
932 						  "mask enables non supported"
933 						  " bits");
934 	if (!item->spec && (item->mask || item->last))
935 		return rte_flow_error_set(error, EINVAL,
936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
937 					  "mask/last without a spec is not"
938 					  " supported");
939 	if (item->spec && item->last && !range_accepted) {
940 		uint8_t spec[size];
941 		uint8_t last[size];
942 		unsigned int i;
943 		int ret;
944 
945 		for (i = 0; i < size; ++i) {
946 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
947 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
948 		}
949 		ret = memcmp(spec, last, size);
950 		if (ret != 0)
951 			return rte_flow_error_set(error, EINVAL,
952 						  RTE_FLOW_ERROR_TYPE_ITEM,
953 						  item,
954 						  "range is not valid");
955 	}
956 	return 0;
957 }
958 
959 /**
960  * Adjust the hash fields according to the @p flow information.
961  *
962  * @param[in] dev_flow.
963  *   Pointer to the mlx5_flow.
964  * @param[in] tunnel
965  *   1 when the hash field is for a tunnel item.
966  * @param[in] layer_types
967  *   ETH_RSS_* types.
968  * @param[in] hash_fields
969  *   Item hash fields.
970  *
971  * @return
972  *   The hash fields that should be used.
973  */
974 uint64_t
975 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
976 			    int tunnel __rte_unused, uint64_t layer_types,
977 			    uint64_t hash_fields)
978 {
979 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
980 	int rss_request_inner = rss_desc->level >= 2;
981 
982 	/* Check RSS hash level for tunnel. */
983 	if (tunnel && rss_request_inner)
984 		hash_fields |= IBV_RX_HASH_INNER;
985 	else if (tunnel || rss_request_inner)
986 		return 0;
987 #endif
988 	/* Check if requested layer matches RSS hash fields. */
989 	if (!(rss_desc->types & layer_types))
990 		return 0;
991 	return hash_fields;
992 }
993 
994 /**
995  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
996  * if several tunnel rules are used on this queue, the tunnel ptype will be
997  * cleared.
998  *
999  * @param rxq_ctrl
1000  *   Rx queue to update.
1001  */
1002 static void
1003 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1004 {
1005 	unsigned int i;
1006 	uint32_t tunnel_ptype = 0;
1007 
1008 	/* Look up for the ptype to use. */
1009 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1010 		if (!rxq_ctrl->flow_tunnels_n[i])
1011 			continue;
1012 		if (!tunnel_ptype) {
1013 			tunnel_ptype = tunnels_info[i].ptype;
1014 		} else {
1015 			tunnel_ptype = 0;
1016 			break;
1017 		}
1018 	}
1019 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1020 }
1021 
1022 /**
1023  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1024  * flow.
1025  *
1026  * @param[in] dev
1027  *   Pointer to the Ethernet device structure.
1028  * @param[in] dev_handle
1029  *   Pointer to device flow handle structure.
1030  */
1031 static void
1032 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1033 		       struct mlx5_flow_handle *dev_handle)
1034 {
1035 	struct mlx5_priv *priv = dev->data->dev_private;
1036 	const int mark = dev_handle->mark;
1037 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1038 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1039 	unsigned int i;
1040 
1041 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1042 		struct mlx5_hrxq *hrxq;
1043 
1044 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1045 			      dev_handle->rix_hrxq);
1046 		if (hrxq)
1047 			ind_tbl = hrxq->ind_table;
1048 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1049 		struct mlx5_shared_action_rss *shared_rss;
1050 
1051 		shared_rss = mlx5_ipool_get
1052 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1053 			 dev_handle->rix_srss);
1054 		if (shared_rss)
1055 			ind_tbl = shared_rss->ind_tbl;
1056 	}
1057 	if (!ind_tbl)
1058 		return;
1059 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1060 		int idx = ind_tbl->queues[i];
1061 		struct mlx5_rxq_ctrl *rxq_ctrl =
1062 			container_of((*priv->rxqs)[idx],
1063 				     struct mlx5_rxq_ctrl, rxq);
1064 
1065 		/*
1066 		 * To support metadata register copy on Tx loopback,
1067 		 * this must be always enabled (metadata may arive
1068 		 * from other port - not from local flows only.
1069 		 */
1070 		if (priv->config.dv_flow_en &&
1071 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1072 		    mlx5_flow_ext_mreg_supported(dev)) {
1073 			rxq_ctrl->rxq.mark = 1;
1074 			rxq_ctrl->flow_mark_n = 1;
1075 		} else if (mark) {
1076 			rxq_ctrl->rxq.mark = 1;
1077 			rxq_ctrl->flow_mark_n++;
1078 		}
1079 		if (tunnel) {
1080 			unsigned int j;
1081 
1082 			/* Increase the counter matching the flow. */
1083 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1084 				if ((tunnels_info[j].tunnel &
1085 				     dev_handle->layers) ==
1086 				    tunnels_info[j].tunnel) {
1087 					rxq_ctrl->flow_tunnels_n[j]++;
1088 					break;
1089 				}
1090 			}
1091 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1092 		}
1093 	}
1094 }
1095 
1096 /**
1097  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1098  *
1099  * @param[in] dev
1100  *   Pointer to the Ethernet device structure.
1101  * @param[in] flow
1102  *   Pointer to flow structure.
1103  */
1104 static void
1105 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1106 {
1107 	struct mlx5_priv *priv = dev->data->dev_private;
1108 	uint32_t handle_idx;
1109 	struct mlx5_flow_handle *dev_handle;
1110 
1111 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1112 		       handle_idx, dev_handle, next)
1113 		flow_drv_rxq_flags_set(dev, dev_handle);
1114 }
1115 
1116 /**
1117  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1118  * device flow if no other flow uses it with the same kind of request.
1119  *
1120  * @param dev
1121  *   Pointer to Ethernet device.
1122  * @param[in] dev_handle
1123  *   Pointer to the device flow handle structure.
1124  */
1125 static void
1126 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1127 			struct mlx5_flow_handle *dev_handle)
1128 {
1129 	struct mlx5_priv *priv = dev->data->dev_private;
1130 	const int mark = dev_handle->mark;
1131 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1132 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1133 	unsigned int i;
1134 
1135 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1136 		struct mlx5_hrxq *hrxq;
1137 
1138 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1139 			      dev_handle->rix_hrxq);
1140 		if (hrxq)
1141 			ind_tbl = hrxq->ind_table;
1142 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1143 		struct mlx5_shared_action_rss *shared_rss;
1144 
1145 		shared_rss = mlx5_ipool_get
1146 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1147 			 dev_handle->rix_srss);
1148 		if (shared_rss)
1149 			ind_tbl = shared_rss->ind_tbl;
1150 	}
1151 	if (!ind_tbl)
1152 		return;
1153 	MLX5_ASSERT(dev->data->dev_started);
1154 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1155 		int idx = ind_tbl->queues[i];
1156 		struct mlx5_rxq_ctrl *rxq_ctrl =
1157 			container_of((*priv->rxqs)[idx],
1158 				     struct mlx5_rxq_ctrl, rxq);
1159 
1160 		if (priv->config.dv_flow_en &&
1161 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1162 		    mlx5_flow_ext_mreg_supported(dev)) {
1163 			rxq_ctrl->rxq.mark = 1;
1164 			rxq_ctrl->flow_mark_n = 1;
1165 		} else if (mark) {
1166 			rxq_ctrl->flow_mark_n--;
1167 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1168 		}
1169 		if (tunnel) {
1170 			unsigned int j;
1171 
1172 			/* Decrease the counter matching the flow. */
1173 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1174 				if ((tunnels_info[j].tunnel &
1175 				     dev_handle->layers) ==
1176 				    tunnels_info[j].tunnel) {
1177 					rxq_ctrl->flow_tunnels_n[j]--;
1178 					break;
1179 				}
1180 			}
1181 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1182 		}
1183 	}
1184 }
1185 
1186 /**
1187  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1188  * @p flow if no other flow uses it with the same kind of request.
1189  *
1190  * @param dev
1191  *   Pointer to Ethernet device.
1192  * @param[in] flow
1193  *   Pointer to the flow.
1194  */
1195 static void
1196 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1197 {
1198 	struct mlx5_priv *priv = dev->data->dev_private;
1199 	uint32_t handle_idx;
1200 	struct mlx5_flow_handle *dev_handle;
1201 
1202 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1203 		       handle_idx, dev_handle, next)
1204 		flow_drv_rxq_flags_trim(dev, dev_handle);
1205 }
1206 
1207 /**
1208  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1209  *
1210  * @param dev
1211  *   Pointer to Ethernet device.
1212  */
1213 static void
1214 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1215 {
1216 	struct mlx5_priv *priv = dev->data->dev_private;
1217 	unsigned int i;
1218 
1219 	for (i = 0; i != priv->rxqs_n; ++i) {
1220 		struct mlx5_rxq_ctrl *rxq_ctrl;
1221 		unsigned int j;
1222 
1223 		if (!(*priv->rxqs)[i])
1224 			continue;
1225 		rxq_ctrl = container_of((*priv->rxqs)[i],
1226 					struct mlx5_rxq_ctrl, rxq);
1227 		rxq_ctrl->flow_mark_n = 0;
1228 		rxq_ctrl->rxq.mark = 0;
1229 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1230 			rxq_ctrl->flow_tunnels_n[j] = 0;
1231 		rxq_ctrl->rxq.tunnel = 0;
1232 	}
1233 }
1234 
1235 /**
1236  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1237  *
1238  * @param[in] dev
1239  *   Pointer to the Ethernet device structure.
1240  */
1241 void
1242 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1243 {
1244 	struct mlx5_priv *priv = dev->data->dev_private;
1245 	struct mlx5_rxq_data *data;
1246 	unsigned int i;
1247 
1248 	for (i = 0; i != priv->rxqs_n; ++i) {
1249 		if (!(*priv->rxqs)[i])
1250 			continue;
1251 		data = (*priv->rxqs)[i];
1252 		if (!rte_flow_dynf_metadata_avail()) {
1253 			data->dynf_meta = 0;
1254 			data->flow_meta_mask = 0;
1255 			data->flow_meta_offset = -1;
1256 			data->flow_meta_port_mask = 0;
1257 		} else {
1258 			data->dynf_meta = 1;
1259 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1260 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1261 			data->flow_meta_port_mask = (uint32_t)~0;
1262 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1263 				data->flow_meta_port_mask >>= 16;
1264 		}
1265 	}
1266 }
1267 
1268 /*
1269  * return a pointer to the desired action in the list of actions.
1270  *
1271  * @param[in] actions
1272  *   The list of actions to search the action in.
1273  * @param[in] action
1274  *   The action to find.
1275  *
1276  * @return
1277  *   Pointer to the action in the list, if found. NULL otherwise.
1278  */
1279 const struct rte_flow_action *
1280 mlx5_flow_find_action(const struct rte_flow_action *actions,
1281 		      enum rte_flow_action_type action)
1282 {
1283 	if (actions == NULL)
1284 		return NULL;
1285 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1286 		if (actions->type == action)
1287 			return actions;
1288 	return NULL;
1289 }
1290 
1291 /*
1292  * Validate the flag action.
1293  *
1294  * @param[in] action_flags
1295  *   Bit-fields that holds the actions detected until now.
1296  * @param[in] attr
1297  *   Attributes of flow that includes this action.
1298  * @param[out] error
1299  *   Pointer to error structure.
1300  *
1301  * @return
1302  *   0 on success, a negative errno value otherwise and rte_errno is set.
1303  */
1304 int
1305 mlx5_flow_validate_action_flag(uint64_t action_flags,
1306 			       const struct rte_flow_attr *attr,
1307 			       struct rte_flow_error *error)
1308 {
1309 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1310 		return rte_flow_error_set(error, EINVAL,
1311 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1312 					  "can't mark and flag in same flow");
1313 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1314 		return rte_flow_error_set(error, EINVAL,
1315 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1316 					  "can't have 2 flag"
1317 					  " actions in same flow");
1318 	if (attr->egress)
1319 		return rte_flow_error_set(error, ENOTSUP,
1320 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1321 					  "flag action not supported for "
1322 					  "egress");
1323 	return 0;
1324 }
1325 
1326 /*
1327  * Validate the mark action.
1328  *
1329  * @param[in] action
1330  *   Pointer to the queue action.
1331  * @param[in] action_flags
1332  *   Bit-fields that holds the actions detected until now.
1333  * @param[in] attr
1334  *   Attributes of flow that includes this action.
1335  * @param[out] error
1336  *   Pointer to error structure.
1337  *
1338  * @return
1339  *   0 on success, a negative errno value otherwise and rte_errno is set.
1340  */
1341 int
1342 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1343 			       uint64_t action_flags,
1344 			       const struct rte_flow_attr *attr,
1345 			       struct rte_flow_error *error)
1346 {
1347 	const struct rte_flow_action_mark *mark = action->conf;
1348 
1349 	if (!mark)
1350 		return rte_flow_error_set(error, EINVAL,
1351 					  RTE_FLOW_ERROR_TYPE_ACTION,
1352 					  action,
1353 					  "configuration cannot be null");
1354 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1355 		return rte_flow_error_set(error, EINVAL,
1356 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1357 					  &mark->id,
1358 					  "mark id must in 0 <= id < "
1359 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1360 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1361 		return rte_flow_error_set(error, EINVAL,
1362 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1363 					  "can't flag and mark in same flow");
1364 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1365 		return rte_flow_error_set(error, EINVAL,
1366 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1367 					  "can't have 2 mark actions in same"
1368 					  " flow");
1369 	if (attr->egress)
1370 		return rte_flow_error_set(error, ENOTSUP,
1371 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1372 					  "mark action not supported for "
1373 					  "egress");
1374 	return 0;
1375 }
1376 
1377 /*
1378  * Validate the drop action.
1379  *
1380  * @param[in] action_flags
1381  *   Bit-fields that holds the actions detected until now.
1382  * @param[in] attr
1383  *   Attributes of flow that includes this action.
1384  * @param[out] error
1385  *   Pointer to error structure.
1386  *
1387  * @return
1388  *   0 on success, a negative errno value otherwise and rte_errno is set.
1389  */
1390 int
1391 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1392 			       const struct rte_flow_attr *attr,
1393 			       struct rte_flow_error *error)
1394 {
1395 	if (attr->egress)
1396 		return rte_flow_error_set(error, ENOTSUP,
1397 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1398 					  "drop action not supported for "
1399 					  "egress");
1400 	return 0;
1401 }
1402 
1403 /*
1404  * Validate the queue action.
1405  *
1406  * @param[in] action
1407  *   Pointer to the queue action.
1408  * @param[in] action_flags
1409  *   Bit-fields that holds the actions detected until now.
1410  * @param[in] dev
1411  *   Pointer to the Ethernet device structure.
1412  * @param[in] attr
1413  *   Attributes of flow that includes this action.
1414  * @param[out] error
1415  *   Pointer to error structure.
1416  *
1417  * @return
1418  *   0 on success, a negative errno value otherwise and rte_errno is set.
1419  */
1420 int
1421 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1422 				uint64_t action_flags,
1423 				struct rte_eth_dev *dev,
1424 				const struct rte_flow_attr *attr,
1425 				struct rte_flow_error *error)
1426 {
1427 	struct mlx5_priv *priv = dev->data->dev_private;
1428 	const struct rte_flow_action_queue *queue = action->conf;
1429 
1430 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1431 		return rte_flow_error_set(error, EINVAL,
1432 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1433 					  "can't have 2 fate actions in"
1434 					  " same flow");
1435 	if (!priv->rxqs_n)
1436 		return rte_flow_error_set(error, EINVAL,
1437 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1438 					  NULL, "No Rx queues configured");
1439 	if (queue->index >= priv->rxqs_n)
1440 		return rte_flow_error_set(error, EINVAL,
1441 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1442 					  &queue->index,
1443 					  "queue index out of range");
1444 	if (!(*priv->rxqs)[queue->index])
1445 		return rte_flow_error_set(error, EINVAL,
1446 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1447 					  &queue->index,
1448 					  "queue is not configured");
1449 	if (attr->egress)
1450 		return rte_flow_error_set(error, ENOTSUP,
1451 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1452 					  "queue action not supported for "
1453 					  "egress");
1454 	return 0;
1455 }
1456 
1457 /*
1458  * Validate the rss action.
1459  *
1460  * @param[in] dev
1461  *   Pointer to the Ethernet device structure.
1462  * @param[in] action
1463  *   Pointer to the queue action.
1464  * @param[out] error
1465  *   Pointer to error structure.
1466  *
1467  * @return
1468  *   0 on success, a negative errno value otherwise and rte_errno is set.
1469  */
1470 int
1471 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1472 			 const struct rte_flow_action *action,
1473 			 struct rte_flow_error *error)
1474 {
1475 	struct mlx5_priv *priv = dev->data->dev_private;
1476 	const struct rte_flow_action_rss *rss = action->conf;
1477 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1478 	unsigned int i;
1479 
1480 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1481 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1482 		return rte_flow_error_set(error, ENOTSUP,
1483 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1484 					  &rss->func,
1485 					  "RSS hash function not supported");
1486 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1487 	if (rss->level > 2)
1488 #else
1489 	if (rss->level > 1)
1490 #endif
1491 		return rte_flow_error_set(error, ENOTSUP,
1492 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1493 					  &rss->level,
1494 					  "tunnel RSS is not supported");
1495 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1496 	if (rss->key_len == 0 && rss->key != NULL)
1497 		return rte_flow_error_set(error, ENOTSUP,
1498 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499 					  &rss->key_len,
1500 					  "RSS hash key length 0");
1501 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1502 		return rte_flow_error_set(error, ENOTSUP,
1503 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1504 					  &rss->key_len,
1505 					  "RSS hash key too small");
1506 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1507 		return rte_flow_error_set(error, ENOTSUP,
1508 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1509 					  &rss->key_len,
1510 					  "RSS hash key too large");
1511 	if (rss->queue_num > priv->config.ind_table_max_size)
1512 		return rte_flow_error_set(error, ENOTSUP,
1513 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1514 					  &rss->queue_num,
1515 					  "number of queues too large");
1516 	if (rss->types & MLX5_RSS_HF_MASK)
1517 		return rte_flow_error_set(error, ENOTSUP,
1518 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1519 					  &rss->types,
1520 					  "some RSS protocols are not"
1521 					  " supported");
1522 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1523 	    !(rss->types & ETH_RSS_IP))
1524 		return rte_flow_error_set(error, EINVAL,
1525 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1526 					  "L3 partial RSS requested but L3 RSS"
1527 					  " type not specified");
1528 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1529 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1530 		return rte_flow_error_set(error, EINVAL,
1531 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1532 					  "L4 partial RSS requested but L4 RSS"
1533 					  " type not specified");
1534 	if (!priv->rxqs_n)
1535 		return rte_flow_error_set(error, EINVAL,
1536 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1537 					  NULL, "No Rx queues configured");
1538 	if (!rss->queue_num)
1539 		return rte_flow_error_set(error, EINVAL,
1540 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1541 					  NULL, "No queues configured");
1542 	for (i = 0; i != rss->queue_num; ++i) {
1543 		struct mlx5_rxq_ctrl *rxq_ctrl;
1544 
1545 		if (rss->queue[i] >= priv->rxqs_n)
1546 			return rte_flow_error_set
1547 				(error, EINVAL,
1548 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1549 				 &rss->queue[i], "queue index out of range");
1550 		if (!(*priv->rxqs)[rss->queue[i]])
1551 			return rte_flow_error_set
1552 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1553 				 &rss->queue[i], "queue is not configured");
1554 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1555 					struct mlx5_rxq_ctrl, rxq);
1556 		if (i == 0)
1557 			rxq_type = rxq_ctrl->type;
1558 		if (rxq_type != rxq_ctrl->type)
1559 			return rte_flow_error_set
1560 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1561 				 &rss->queue[i],
1562 				 "combining hairpin and regular RSS queues is not supported");
1563 	}
1564 	return 0;
1565 }
1566 
1567 /*
1568  * Validate the rss action.
1569  *
1570  * @param[in] action
1571  *   Pointer to the queue action.
1572  * @param[in] action_flags
1573  *   Bit-fields that holds the actions detected until now.
1574  * @param[in] dev
1575  *   Pointer to the Ethernet device structure.
1576  * @param[in] attr
1577  *   Attributes of flow that includes this action.
1578  * @param[in] item_flags
1579  *   Items that were detected.
1580  * @param[out] error
1581  *   Pointer to error structure.
1582  *
1583  * @return
1584  *   0 on success, a negative errno value otherwise and rte_errno is set.
1585  */
1586 int
1587 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1588 			      uint64_t action_flags,
1589 			      struct rte_eth_dev *dev,
1590 			      const struct rte_flow_attr *attr,
1591 			      uint64_t item_flags,
1592 			      struct rte_flow_error *error)
1593 {
1594 	const struct rte_flow_action_rss *rss = action->conf;
1595 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1596 	int ret;
1597 
1598 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1599 		return rte_flow_error_set(error, EINVAL,
1600 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1601 					  "can't have 2 fate actions"
1602 					  " in same flow");
1603 	ret = mlx5_validate_action_rss(dev, action, error);
1604 	if (ret)
1605 		return ret;
1606 	if (attr->egress)
1607 		return rte_flow_error_set(error, ENOTSUP,
1608 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1609 					  "rss action not supported for "
1610 					  "egress");
1611 	if (rss->level > 1 && !tunnel)
1612 		return rte_flow_error_set(error, EINVAL,
1613 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1614 					  "inner RSS is not supported for "
1615 					  "non-tunnel flows");
1616 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1617 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1618 		return rte_flow_error_set(error, EINVAL,
1619 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1620 					  "RSS on eCPRI is not supported now");
1621 	}
1622 	return 0;
1623 }
1624 
1625 /*
1626  * Validate the default miss action.
1627  *
1628  * @param[in] action_flags
1629  *   Bit-fields that holds the actions detected until now.
1630  * @param[out] error
1631  *   Pointer to error structure.
1632  *
1633  * @return
1634  *   0 on success, a negative errno value otherwise and rte_errno is set.
1635  */
1636 int
1637 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1638 				const struct rte_flow_attr *attr,
1639 				struct rte_flow_error *error)
1640 {
1641 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1642 		return rte_flow_error_set(error, EINVAL,
1643 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1644 					  "can't have 2 fate actions in"
1645 					  " same flow");
1646 	if (attr->egress)
1647 		return rte_flow_error_set(error, ENOTSUP,
1648 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1649 					  "default miss action not supported "
1650 					  "for egress");
1651 	if (attr->group)
1652 		return rte_flow_error_set(error, ENOTSUP,
1653 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1654 					  "only group 0 is supported");
1655 	if (attr->transfer)
1656 		return rte_flow_error_set(error, ENOTSUP,
1657 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1658 					  NULL, "transfer is not supported");
1659 	return 0;
1660 }
1661 
1662 /*
1663  * Validate the count action.
1664  *
1665  * @param[in] dev
1666  *   Pointer to the Ethernet device structure.
1667  * @param[in] attr
1668  *   Attributes of flow that includes this action.
1669  * @param[out] error
1670  *   Pointer to error structure.
1671  *
1672  * @return
1673  *   0 on success, a negative errno value otherwise and rte_errno is set.
1674  */
1675 int
1676 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1677 				const struct rte_flow_attr *attr,
1678 				struct rte_flow_error *error)
1679 {
1680 	if (attr->egress)
1681 		return rte_flow_error_set(error, ENOTSUP,
1682 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1683 					  "count action not supported for "
1684 					  "egress");
1685 	return 0;
1686 }
1687 
1688 /**
1689  * Verify the @p attributes will be correctly understood by the NIC and store
1690  * them in the @p flow if everything is correct.
1691  *
1692  * @param[in] dev
1693  *   Pointer to the Ethernet device structure.
1694  * @param[in] attributes
1695  *   Pointer to flow attributes
1696  * @param[out] error
1697  *   Pointer to error structure.
1698  *
1699  * @return
1700  *   0 on success, a negative errno value otherwise and rte_errno is set.
1701  */
1702 int
1703 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1704 			      const struct rte_flow_attr *attributes,
1705 			      struct rte_flow_error *error)
1706 {
1707 	struct mlx5_priv *priv = dev->data->dev_private;
1708 	uint32_t priority_max = priv->config.flow_prio - 1;
1709 
1710 	if (attributes->group)
1711 		return rte_flow_error_set(error, ENOTSUP,
1712 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1713 					  NULL, "groups is not supported");
1714 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1715 	    attributes->priority >= priority_max)
1716 		return rte_flow_error_set(error, ENOTSUP,
1717 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1718 					  NULL, "priority out of range");
1719 	if (attributes->egress)
1720 		return rte_flow_error_set(error, ENOTSUP,
1721 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1722 					  "egress is not supported");
1723 	if (attributes->transfer && !priv->config.dv_esw_en)
1724 		return rte_flow_error_set(error, ENOTSUP,
1725 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1726 					  NULL, "transfer is not supported");
1727 	if (!attributes->ingress)
1728 		return rte_flow_error_set(error, EINVAL,
1729 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1730 					  NULL,
1731 					  "ingress attribute is mandatory");
1732 	return 0;
1733 }
1734 
1735 /**
1736  * Validate ICMP6 item.
1737  *
1738  * @param[in] item
1739  *   Item specification.
1740  * @param[in] item_flags
1741  *   Bit-fields that holds the items detected until now.
1742  * @param[in] ext_vlan_sup
1743  *   Whether extended VLAN features are supported or not.
1744  * @param[out] error
1745  *   Pointer to error structure.
1746  *
1747  * @return
1748  *   0 on success, a negative errno value otherwise and rte_errno is set.
1749  */
1750 int
1751 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1752 			       uint64_t item_flags,
1753 			       uint8_t target_protocol,
1754 			       struct rte_flow_error *error)
1755 {
1756 	const struct rte_flow_item_icmp6 *mask = item->mask;
1757 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1758 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1759 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1760 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1761 				      MLX5_FLOW_LAYER_OUTER_L4;
1762 	int ret;
1763 
1764 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1765 		return rte_flow_error_set(error, EINVAL,
1766 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1767 					  "protocol filtering not compatible"
1768 					  " with ICMP6 layer");
1769 	if (!(item_flags & l3m))
1770 		return rte_flow_error_set(error, EINVAL,
1771 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1772 					  "IPv6 is mandatory to filter on"
1773 					  " ICMP6");
1774 	if (item_flags & l4m)
1775 		return rte_flow_error_set(error, EINVAL,
1776 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1777 					  "multiple L4 layers not supported");
1778 	if (!mask)
1779 		mask = &rte_flow_item_icmp6_mask;
1780 	ret = mlx5_flow_item_acceptable
1781 		(item, (const uint8_t *)mask,
1782 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1783 		 sizeof(struct rte_flow_item_icmp6),
1784 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1785 	if (ret < 0)
1786 		return ret;
1787 	return 0;
1788 }
1789 
1790 /**
1791  * Validate ICMP item.
1792  *
1793  * @param[in] item
1794  *   Item specification.
1795  * @param[in] item_flags
1796  *   Bit-fields that holds the items detected until now.
1797  * @param[out] error
1798  *   Pointer to error structure.
1799  *
1800  * @return
1801  *   0 on success, a negative errno value otherwise and rte_errno is set.
1802  */
1803 int
1804 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1805 			     uint64_t item_flags,
1806 			     uint8_t target_protocol,
1807 			     struct rte_flow_error *error)
1808 {
1809 	const struct rte_flow_item_icmp *mask = item->mask;
1810 	const struct rte_flow_item_icmp nic_mask = {
1811 		.hdr.icmp_type = 0xff,
1812 		.hdr.icmp_code = 0xff,
1813 		.hdr.icmp_ident = RTE_BE16(0xffff),
1814 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1815 	};
1816 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1817 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1818 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1819 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1820 				      MLX5_FLOW_LAYER_OUTER_L4;
1821 	int ret;
1822 
1823 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1824 		return rte_flow_error_set(error, EINVAL,
1825 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1826 					  "protocol filtering not compatible"
1827 					  " with ICMP layer");
1828 	if (!(item_flags & l3m))
1829 		return rte_flow_error_set(error, EINVAL,
1830 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1831 					  "IPv4 is mandatory to filter"
1832 					  " on ICMP");
1833 	if (item_flags & l4m)
1834 		return rte_flow_error_set(error, EINVAL,
1835 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1836 					  "multiple L4 layers not supported");
1837 	if (!mask)
1838 		mask = &nic_mask;
1839 	ret = mlx5_flow_item_acceptable
1840 		(item, (const uint8_t *)mask,
1841 		 (const uint8_t *)&nic_mask,
1842 		 sizeof(struct rte_flow_item_icmp),
1843 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1844 	if (ret < 0)
1845 		return ret;
1846 	return 0;
1847 }
1848 
1849 /**
1850  * Validate Ethernet item.
1851  *
1852  * @param[in] item
1853  *   Item specification.
1854  * @param[in] item_flags
1855  *   Bit-fields that holds the items detected until now.
1856  * @param[out] error
1857  *   Pointer to error structure.
1858  *
1859  * @return
1860  *   0 on success, a negative errno value otherwise and rte_errno is set.
1861  */
1862 int
1863 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1864 			    uint64_t item_flags, bool ext_vlan_sup,
1865 			    struct rte_flow_error *error)
1866 {
1867 	const struct rte_flow_item_eth *mask = item->mask;
1868 	const struct rte_flow_item_eth nic_mask = {
1869 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1870 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1871 		.type = RTE_BE16(0xffff),
1872 		.has_vlan = ext_vlan_sup ? 1 : 0,
1873 	};
1874 	int ret;
1875 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1876 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1877 				       MLX5_FLOW_LAYER_OUTER_L2;
1878 
1879 	if (item_flags & ethm)
1880 		return rte_flow_error_set(error, ENOTSUP,
1881 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1882 					  "multiple L2 layers not supported");
1883 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1884 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1885 		return rte_flow_error_set(error, EINVAL,
1886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1887 					  "L2 layer should not follow "
1888 					  "L3 layers");
1889 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1890 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1891 		return rte_flow_error_set(error, EINVAL,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "L2 layer should not follow VLAN");
1894 	if (!mask)
1895 		mask = &rte_flow_item_eth_mask;
1896 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1897 					(const uint8_t *)&nic_mask,
1898 					sizeof(struct rte_flow_item_eth),
1899 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1900 	return ret;
1901 }
1902 
1903 /**
1904  * Validate VLAN item.
1905  *
1906  * @param[in] item
1907  *   Item specification.
1908  * @param[in] item_flags
1909  *   Bit-fields that holds the items detected until now.
1910  * @param[in] dev
1911  *   Ethernet device flow is being created on.
1912  * @param[out] error
1913  *   Pointer to error structure.
1914  *
1915  * @return
1916  *   0 on success, a negative errno value otherwise and rte_errno is set.
1917  */
1918 int
1919 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1920 			     uint64_t item_flags,
1921 			     struct rte_eth_dev *dev,
1922 			     struct rte_flow_error *error)
1923 {
1924 	const struct rte_flow_item_vlan *spec = item->spec;
1925 	const struct rte_flow_item_vlan *mask = item->mask;
1926 	const struct rte_flow_item_vlan nic_mask = {
1927 		.tci = RTE_BE16(UINT16_MAX),
1928 		.inner_type = RTE_BE16(UINT16_MAX),
1929 	};
1930 	uint16_t vlan_tag = 0;
1931 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1932 	int ret;
1933 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1934 					MLX5_FLOW_LAYER_INNER_L4) :
1935 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1936 					MLX5_FLOW_LAYER_OUTER_L4);
1937 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1938 					MLX5_FLOW_LAYER_OUTER_VLAN;
1939 
1940 	if (item_flags & vlanm)
1941 		return rte_flow_error_set(error, EINVAL,
1942 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1943 					  "multiple VLAN layers not supported");
1944 	else if ((item_flags & l34m) != 0)
1945 		return rte_flow_error_set(error, EINVAL,
1946 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1947 					  "VLAN cannot follow L3/L4 layer");
1948 	if (!mask)
1949 		mask = &rte_flow_item_vlan_mask;
1950 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1951 					(const uint8_t *)&nic_mask,
1952 					sizeof(struct rte_flow_item_vlan),
1953 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1954 	if (ret)
1955 		return ret;
1956 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1957 		struct mlx5_priv *priv = dev->data->dev_private;
1958 
1959 		if (priv->vmwa_context) {
1960 			/*
1961 			 * Non-NULL context means we have a virtual machine
1962 			 * and SR-IOV enabled, we have to create VLAN interface
1963 			 * to make hypervisor to setup E-Switch vport
1964 			 * context correctly. We avoid creating the multiple
1965 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1966 			 */
1967 			return rte_flow_error_set(error, EINVAL,
1968 						  RTE_FLOW_ERROR_TYPE_ITEM,
1969 						  item,
1970 						  "VLAN tag mask is not"
1971 						  " supported in virtual"
1972 						  " environment");
1973 		}
1974 	}
1975 	if (spec) {
1976 		vlan_tag = spec->tci;
1977 		vlan_tag &= mask->tci;
1978 	}
1979 	/*
1980 	 * From verbs perspective an empty VLAN is equivalent
1981 	 * to a packet without VLAN layer.
1982 	 */
1983 	if (!vlan_tag)
1984 		return rte_flow_error_set(error, EINVAL,
1985 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1986 					  item->spec,
1987 					  "VLAN cannot be empty");
1988 	return 0;
1989 }
1990 
1991 /**
1992  * Validate IPV4 item.
1993  *
1994  * @param[in] item
1995  *   Item specification.
1996  * @param[in] item_flags
1997  *   Bit-fields that holds the items detected until now.
1998  * @param[in] last_item
1999  *   Previous validated item in the pattern items.
2000  * @param[in] ether_type
2001  *   Type in the ethernet layer header (including dot1q).
2002  * @param[in] acc_mask
2003  *   Acceptable mask, if NULL default internal default mask
2004  *   will be used to check whether item fields are supported.
2005  * @param[in] range_accepted
2006  *   True if range of values is accepted for specific fields, false otherwise.
2007  * @param[out] error
2008  *   Pointer to error structure.
2009  *
2010  * @return
2011  *   0 on success, a negative errno value otherwise and rte_errno is set.
2012  */
2013 int
2014 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2015 			     uint64_t item_flags,
2016 			     uint64_t last_item,
2017 			     uint16_t ether_type,
2018 			     const struct rte_flow_item_ipv4 *acc_mask,
2019 			     bool range_accepted,
2020 			     struct rte_flow_error *error)
2021 {
2022 	const struct rte_flow_item_ipv4 *mask = item->mask;
2023 	const struct rte_flow_item_ipv4 *spec = item->spec;
2024 	const struct rte_flow_item_ipv4 nic_mask = {
2025 		.hdr = {
2026 			.src_addr = RTE_BE32(0xffffffff),
2027 			.dst_addr = RTE_BE32(0xffffffff),
2028 			.type_of_service = 0xff,
2029 			.next_proto_id = 0xff,
2030 		},
2031 	};
2032 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2034 				      MLX5_FLOW_LAYER_OUTER_L3;
2035 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2036 				      MLX5_FLOW_LAYER_OUTER_L4;
2037 	int ret;
2038 	uint8_t next_proto = 0xFF;
2039 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2040 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2041 				  MLX5_FLOW_LAYER_INNER_VLAN);
2042 
2043 	if ((last_item & l2_vlan) && ether_type &&
2044 	    ether_type != RTE_ETHER_TYPE_IPV4)
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "IPv4 cannot follow L2/VLAN layer "
2048 					  "which ether type is not IPv4");
2049 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2050 		if (mask && spec)
2051 			next_proto = mask->hdr.next_proto_id &
2052 				     spec->hdr.next_proto_id;
2053 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2054 			return rte_flow_error_set(error, EINVAL,
2055 						  RTE_FLOW_ERROR_TYPE_ITEM,
2056 						  item,
2057 						  "multiple tunnel "
2058 						  "not supported");
2059 	}
2060 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2061 		return rte_flow_error_set(error, EINVAL,
2062 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2063 					  "wrong tunnel type - IPv6 specified "
2064 					  "but IPv4 item provided");
2065 	if (item_flags & l3m)
2066 		return rte_flow_error_set(error, ENOTSUP,
2067 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2068 					  "multiple L3 layers not supported");
2069 	else if (item_flags & l4m)
2070 		return rte_flow_error_set(error, EINVAL,
2071 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2072 					  "L3 cannot follow an L4 layer.");
2073 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2074 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2075 		return rte_flow_error_set(error, EINVAL,
2076 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2077 					  "L3 cannot follow an NVGRE layer.");
2078 	if (!mask)
2079 		mask = &rte_flow_item_ipv4_mask;
2080 	else if (mask->hdr.next_proto_id != 0 &&
2081 		 mask->hdr.next_proto_id != 0xff)
2082 		return rte_flow_error_set(error, EINVAL,
2083 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2084 					  "partial mask is not supported"
2085 					  " for protocol");
2086 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2087 					acc_mask ? (const uint8_t *)acc_mask
2088 						 : (const uint8_t *)&nic_mask,
2089 					sizeof(struct rte_flow_item_ipv4),
2090 					range_accepted, error);
2091 	if (ret < 0)
2092 		return ret;
2093 	return 0;
2094 }
2095 
2096 /**
2097  * Validate IPV6 item.
2098  *
2099  * @param[in] item
2100  *   Item specification.
2101  * @param[in] item_flags
2102  *   Bit-fields that holds the items detected until now.
2103  * @param[in] last_item
2104  *   Previous validated item in the pattern items.
2105  * @param[in] ether_type
2106  *   Type in the ethernet layer header (including dot1q).
2107  * @param[in] acc_mask
2108  *   Acceptable mask, if NULL default internal default mask
2109  *   will be used to check whether item fields are supported.
2110  * @param[out] error
2111  *   Pointer to error structure.
2112  *
2113  * @return
2114  *   0 on success, a negative errno value otherwise and rte_errno is set.
2115  */
2116 int
2117 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2118 			     uint64_t item_flags,
2119 			     uint64_t last_item,
2120 			     uint16_t ether_type,
2121 			     const struct rte_flow_item_ipv6 *acc_mask,
2122 			     struct rte_flow_error *error)
2123 {
2124 	const struct rte_flow_item_ipv6 *mask = item->mask;
2125 	const struct rte_flow_item_ipv6 *spec = item->spec;
2126 	const struct rte_flow_item_ipv6 nic_mask = {
2127 		.hdr = {
2128 			.src_addr =
2129 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2130 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2131 			.dst_addr =
2132 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2133 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2134 			.vtc_flow = RTE_BE32(0xffffffff),
2135 			.proto = 0xff,
2136 		},
2137 	};
2138 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2139 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2140 				      MLX5_FLOW_LAYER_OUTER_L3;
2141 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2142 				      MLX5_FLOW_LAYER_OUTER_L4;
2143 	int ret;
2144 	uint8_t next_proto = 0xFF;
2145 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2146 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2147 				  MLX5_FLOW_LAYER_INNER_VLAN);
2148 
2149 	if ((last_item & l2_vlan) && ether_type &&
2150 	    ether_type != RTE_ETHER_TYPE_IPV6)
2151 		return rte_flow_error_set(error, EINVAL,
2152 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2153 					  "IPv6 cannot follow L2/VLAN layer "
2154 					  "which ether type is not IPv6");
2155 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2156 		next_proto = spec->hdr.proto;
2157 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2158 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2159 			return rte_flow_error_set(error, EINVAL,
2160 						  RTE_FLOW_ERROR_TYPE_ITEM,
2161 						  item,
2162 						  "multiple tunnel "
2163 						  "not supported");
2164 	}
2165 	if (next_proto == IPPROTO_HOPOPTS  ||
2166 	    next_proto == IPPROTO_ROUTING  ||
2167 	    next_proto == IPPROTO_FRAGMENT ||
2168 	    next_proto == IPPROTO_ESP	   ||
2169 	    next_proto == IPPROTO_AH	   ||
2170 	    next_proto == IPPROTO_DSTOPTS)
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "IPv6 proto (next header) should "
2174 					  "not be set as extension header");
2175 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2176 		return rte_flow_error_set(error, EINVAL,
2177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2178 					  "wrong tunnel type - IPv4 specified "
2179 					  "but IPv6 item provided");
2180 	if (item_flags & l3m)
2181 		return rte_flow_error_set(error, ENOTSUP,
2182 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2183 					  "multiple L3 layers not supported");
2184 	else if (item_flags & l4m)
2185 		return rte_flow_error_set(error, EINVAL,
2186 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2187 					  "L3 cannot follow an L4 layer.");
2188 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2189 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2190 		return rte_flow_error_set(error, EINVAL,
2191 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2192 					  "L3 cannot follow an NVGRE layer.");
2193 	if (!mask)
2194 		mask = &rte_flow_item_ipv6_mask;
2195 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2196 					acc_mask ? (const uint8_t *)acc_mask
2197 						 : (const uint8_t *)&nic_mask,
2198 					sizeof(struct rte_flow_item_ipv6),
2199 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2200 	if (ret < 0)
2201 		return ret;
2202 	return 0;
2203 }
2204 
2205 /**
2206  * Validate UDP item.
2207  *
2208  * @param[in] item
2209  *   Item specification.
2210  * @param[in] item_flags
2211  *   Bit-fields that holds the items detected until now.
2212  * @param[in] target_protocol
2213  *   The next protocol in the previous item.
2214  * @param[in] flow_mask
2215  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2216  * @param[out] error
2217  *   Pointer to error structure.
2218  *
2219  * @return
2220  *   0 on success, a negative errno value otherwise and rte_errno is set.
2221  */
2222 int
2223 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2224 			    uint64_t item_flags,
2225 			    uint8_t target_protocol,
2226 			    struct rte_flow_error *error)
2227 {
2228 	const struct rte_flow_item_udp *mask = item->mask;
2229 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2230 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2231 				      MLX5_FLOW_LAYER_OUTER_L3;
2232 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2233 				      MLX5_FLOW_LAYER_OUTER_L4;
2234 	int ret;
2235 
2236 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2237 		return rte_flow_error_set(error, EINVAL,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "protocol filtering not compatible"
2240 					  " with UDP layer");
2241 	if (!(item_flags & l3m))
2242 		return rte_flow_error_set(error, EINVAL,
2243 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2244 					  "L3 is mandatory to filter on L4");
2245 	if (item_flags & l4m)
2246 		return rte_flow_error_set(error, EINVAL,
2247 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2248 					  "multiple L4 layers not supported");
2249 	if (!mask)
2250 		mask = &rte_flow_item_udp_mask;
2251 	ret = mlx5_flow_item_acceptable
2252 		(item, (const uint8_t *)mask,
2253 		 (const uint8_t *)&rte_flow_item_udp_mask,
2254 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2255 		 error);
2256 	if (ret < 0)
2257 		return ret;
2258 	return 0;
2259 }
2260 
2261 /**
2262  * Validate TCP item.
2263  *
2264  * @param[in] item
2265  *   Item specification.
2266  * @param[in] item_flags
2267  *   Bit-fields that holds the items detected until now.
2268  * @param[in] target_protocol
2269  *   The next protocol in the previous item.
2270  * @param[out] error
2271  *   Pointer to error structure.
2272  *
2273  * @return
2274  *   0 on success, a negative errno value otherwise and rte_errno is set.
2275  */
2276 int
2277 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2278 			    uint64_t item_flags,
2279 			    uint8_t target_protocol,
2280 			    const struct rte_flow_item_tcp *flow_mask,
2281 			    struct rte_flow_error *error)
2282 {
2283 	const struct rte_flow_item_tcp *mask = item->mask;
2284 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2285 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2286 				      MLX5_FLOW_LAYER_OUTER_L3;
2287 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2288 				      MLX5_FLOW_LAYER_OUTER_L4;
2289 	int ret;
2290 
2291 	MLX5_ASSERT(flow_mask);
2292 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2293 		return rte_flow_error_set(error, EINVAL,
2294 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 					  "protocol filtering not compatible"
2296 					  " with TCP layer");
2297 	if (!(item_flags & l3m))
2298 		return rte_flow_error_set(error, EINVAL,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "L3 is mandatory to filter on L4");
2301 	if (item_flags & l4m)
2302 		return rte_flow_error_set(error, EINVAL,
2303 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2304 					  "multiple L4 layers not supported");
2305 	if (!mask)
2306 		mask = &rte_flow_item_tcp_mask;
2307 	ret = mlx5_flow_item_acceptable
2308 		(item, (const uint8_t *)mask,
2309 		 (const uint8_t *)flow_mask,
2310 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2311 		 error);
2312 	if (ret < 0)
2313 		return ret;
2314 	return 0;
2315 }
2316 
2317 /**
2318  * Validate VXLAN item.
2319  *
2320  * @param[in] item
2321  *   Item specification.
2322  * @param[in] item_flags
2323  *   Bit-fields that holds the items detected until now.
2324  * @param[in] target_protocol
2325  *   The next protocol in the previous item.
2326  * @param[out] error
2327  *   Pointer to error structure.
2328  *
2329  * @return
2330  *   0 on success, a negative errno value otherwise and rte_errno is set.
2331  */
2332 int
2333 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2334 			      uint64_t item_flags,
2335 			      struct rte_flow_error *error)
2336 {
2337 	const struct rte_flow_item_vxlan *spec = item->spec;
2338 	const struct rte_flow_item_vxlan *mask = item->mask;
2339 	int ret;
2340 	union vni {
2341 		uint32_t vlan_id;
2342 		uint8_t vni[4];
2343 	} id = { .vlan_id = 0, };
2344 
2345 
2346 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2347 		return rte_flow_error_set(error, ENOTSUP,
2348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2349 					  "multiple tunnel layers not"
2350 					  " supported");
2351 	/*
2352 	 * Verify only UDPv4 is present as defined in
2353 	 * https://tools.ietf.org/html/rfc7348
2354 	 */
2355 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2356 		return rte_flow_error_set(error, EINVAL,
2357 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2358 					  "no outer UDP layer found");
2359 	if (!mask)
2360 		mask = &rte_flow_item_vxlan_mask;
2361 	ret = mlx5_flow_item_acceptable
2362 		(item, (const uint8_t *)mask,
2363 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2364 		 sizeof(struct rte_flow_item_vxlan),
2365 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2366 	if (ret < 0)
2367 		return ret;
2368 	if (spec) {
2369 		memcpy(&id.vni[1], spec->vni, 3);
2370 		memcpy(&id.vni[1], mask->vni, 3);
2371 	}
2372 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2373 		return rte_flow_error_set(error, ENOTSUP,
2374 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2375 					  "VXLAN tunnel must be fully defined");
2376 	return 0;
2377 }
2378 
2379 /**
2380  * Validate VXLAN_GPE item.
2381  *
2382  * @param[in] item
2383  *   Item specification.
2384  * @param[in] item_flags
2385  *   Bit-fields that holds the items detected until now.
2386  * @param[in] priv
2387  *   Pointer to the private data structure.
2388  * @param[in] target_protocol
2389  *   The next protocol in the previous item.
2390  * @param[out] error
2391  *   Pointer to error structure.
2392  *
2393  * @return
2394  *   0 on success, a negative errno value otherwise and rte_errno is set.
2395  */
2396 int
2397 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2398 				  uint64_t item_flags,
2399 				  struct rte_eth_dev *dev,
2400 				  struct rte_flow_error *error)
2401 {
2402 	struct mlx5_priv *priv = dev->data->dev_private;
2403 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2404 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2405 	int ret;
2406 	union vni {
2407 		uint32_t vlan_id;
2408 		uint8_t vni[4];
2409 	} id = { .vlan_id = 0, };
2410 
2411 	if (!priv->config.l3_vxlan_en)
2412 		return rte_flow_error_set(error, ENOTSUP,
2413 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2414 					  "L3 VXLAN is not enabled by device"
2415 					  " parameter and/or not configured in"
2416 					  " firmware");
2417 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2418 		return rte_flow_error_set(error, ENOTSUP,
2419 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2420 					  "multiple tunnel layers not"
2421 					  " supported");
2422 	/*
2423 	 * Verify only UDPv4 is present as defined in
2424 	 * https://tools.ietf.org/html/rfc7348
2425 	 */
2426 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2427 		return rte_flow_error_set(error, EINVAL,
2428 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2429 					  "no outer UDP layer found");
2430 	if (!mask)
2431 		mask = &rte_flow_item_vxlan_gpe_mask;
2432 	ret = mlx5_flow_item_acceptable
2433 		(item, (const uint8_t *)mask,
2434 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2435 		 sizeof(struct rte_flow_item_vxlan_gpe),
2436 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2437 	if (ret < 0)
2438 		return ret;
2439 	if (spec) {
2440 		if (spec->protocol)
2441 			return rte_flow_error_set(error, ENOTSUP,
2442 						  RTE_FLOW_ERROR_TYPE_ITEM,
2443 						  item,
2444 						  "VxLAN-GPE protocol"
2445 						  " not supported");
2446 		memcpy(&id.vni[1], spec->vni, 3);
2447 		memcpy(&id.vni[1], mask->vni, 3);
2448 	}
2449 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2450 		return rte_flow_error_set(error, ENOTSUP,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 					  "VXLAN-GPE tunnel must be fully"
2453 					  " defined");
2454 	return 0;
2455 }
2456 /**
2457  * Validate GRE Key item.
2458  *
2459  * @param[in] item
2460  *   Item specification.
2461  * @param[in] item_flags
2462  *   Bit flags to mark detected items.
2463  * @param[in] gre_item
2464  *   Pointer to gre_item
2465  * @param[out] error
2466  *   Pointer to error structure.
2467  *
2468  * @return
2469  *   0 on success, a negative errno value otherwise and rte_errno is set.
2470  */
2471 int
2472 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2473 				uint64_t item_flags,
2474 				const struct rte_flow_item *gre_item,
2475 				struct rte_flow_error *error)
2476 {
2477 	const rte_be32_t *mask = item->mask;
2478 	int ret = 0;
2479 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2480 	const struct rte_flow_item_gre *gre_spec;
2481 	const struct rte_flow_item_gre *gre_mask;
2482 
2483 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2484 		return rte_flow_error_set(error, ENOTSUP,
2485 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2486 					  "Multiple GRE key not support");
2487 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2488 		return rte_flow_error_set(error, ENOTSUP,
2489 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2490 					  "No preceding GRE header");
2491 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2492 		return rte_flow_error_set(error, ENOTSUP,
2493 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2494 					  "GRE key following a wrong item");
2495 	gre_mask = gre_item->mask;
2496 	if (!gre_mask)
2497 		gre_mask = &rte_flow_item_gre_mask;
2498 	gre_spec = gre_item->spec;
2499 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2500 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2501 		return rte_flow_error_set(error, EINVAL,
2502 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2503 					  "Key bit must be on");
2504 
2505 	if (!mask)
2506 		mask = &gre_key_default_mask;
2507 	ret = mlx5_flow_item_acceptable
2508 		(item, (const uint8_t *)mask,
2509 		 (const uint8_t *)&gre_key_default_mask,
2510 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2511 	return ret;
2512 }
2513 
2514 /**
2515  * Validate GRE item.
2516  *
2517  * @param[in] item
2518  *   Item specification.
2519  * @param[in] item_flags
2520  *   Bit flags to mark detected items.
2521  * @param[in] target_protocol
2522  *   The next protocol in the previous item.
2523  * @param[out] error
2524  *   Pointer to error structure.
2525  *
2526  * @return
2527  *   0 on success, a negative errno value otherwise and rte_errno is set.
2528  */
2529 int
2530 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2531 			    uint64_t item_flags,
2532 			    uint8_t target_protocol,
2533 			    struct rte_flow_error *error)
2534 {
2535 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2536 	const struct rte_flow_item_gre *mask = item->mask;
2537 	int ret;
2538 	const struct rte_flow_item_gre nic_mask = {
2539 		.c_rsvd0_ver = RTE_BE16(0xB000),
2540 		.protocol = RTE_BE16(UINT16_MAX),
2541 	};
2542 
2543 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2544 		return rte_flow_error_set(error, EINVAL,
2545 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2546 					  "protocol filtering not compatible"
2547 					  " with this GRE layer");
2548 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2549 		return rte_flow_error_set(error, ENOTSUP,
2550 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2551 					  "multiple tunnel layers not"
2552 					  " supported");
2553 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2554 		return rte_flow_error_set(error, ENOTSUP,
2555 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2556 					  "L3 Layer is missing");
2557 	if (!mask)
2558 		mask = &rte_flow_item_gre_mask;
2559 	ret = mlx5_flow_item_acceptable
2560 		(item, (const uint8_t *)mask,
2561 		 (const uint8_t *)&nic_mask,
2562 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2563 		 error);
2564 	if (ret < 0)
2565 		return ret;
2566 #ifndef HAVE_MLX5DV_DR
2567 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2568 	if (spec && (spec->protocol & mask->protocol))
2569 		return rte_flow_error_set(error, ENOTSUP,
2570 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2571 					  "without MPLS support the"
2572 					  " specification cannot be used for"
2573 					  " filtering");
2574 #endif
2575 #endif
2576 	return 0;
2577 }
2578 
2579 /**
2580  * Validate Geneve item.
2581  *
2582  * @param[in] item
2583  *   Item specification.
2584  * @param[in] itemFlags
2585  *   Bit-fields that holds the items detected until now.
2586  * @param[in] enPriv
2587  *   Pointer to the private data structure.
2588  * @param[out] error
2589  *   Pointer to error structure.
2590  *
2591  * @return
2592  *   0 on success, a negative errno value otherwise and rte_errno is set.
2593  */
2594 
2595 int
2596 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2597 			       uint64_t item_flags,
2598 			       struct rte_eth_dev *dev,
2599 			       struct rte_flow_error *error)
2600 {
2601 	struct mlx5_priv *priv = dev->data->dev_private;
2602 	const struct rte_flow_item_geneve *spec = item->spec;
2603 	const struct rte_flow_item_geneve *mask = item->mask;
2604 	int ret;
2605 	uint16_t gbhdr;
2606 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2607 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2608 	const struct rte_flow_item_geneve nic_mask = {
2609 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2610 		.vni = "\xff\xff\xff",
2611 		.protocol = RTE_BE16(UINT16_MAX),
2612 	};
2613 
2614 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2615 		return rte_flow_error_set(error, ENOTSUP,
2616 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2617 					  "L3 Geneve is not enabled by device"
2618 					  " parameter and/or not configured in"
2619 					  " firmware");
2620 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2621 		return rte_flow_error_set(error, ENOTSUP,
2622 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2623 					  "multiple tunnel layers not"
2624 					  " supported");
2625 	/*
2626 	 * Verify only UDPv4 is present as defined in
2627 	 * https://tools.ietf.org/html/rfc7348
2628 	 */
2629 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2630 		return rte_flow_error_set(error, EINVAL,
2631 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2632 					  "no outer UDP layer found");
2633 	if (!mask)
2634 		mask = &rte_flow_item_geneve_mask;
2635 	ret = mlx5_flow_item_acceptable
2636 				  (item, (const uint8_t *)mask,
2637 				   (const uint8_t *)&nic_mask,
2638 				   sizeof(struct rte_flow_item_geneve),
2639 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2640 	if (ret)
2641 		return ret;
2642 	if (spec) {
2643 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2644 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2645 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2646 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2647 			return rte_flow_error_set(error, ENOTSUP,
2648 						  RTE_FLOW_ERROR_TYPE_ITEM,
2649 						  item,
2650 						  "Geneve protocol unsupported"
2651 						  " fields are being used");
2652 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2653 			return rte_flow_error_set
2654 					(error, ENOTSUP,
2655 					 RTE_FLOW_ERROR_TYPE_ITEM,
2656 					 item,
2657 					 "Unsupported Geneve options length");
2658 	}
2659 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2660 		return rte_flow_error_set
2661 				    (error, ENOTSUP,
2662 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2663 				     "Geneve tunnel must be fully defined");
2664 	return 0;
2665 }
2666 
2667 /**
2668  * Validate Geneve TLV option item.
2669  *
2670  * @param[in] item
2671  *   Item specification.
2672  * @param[in] last_item
2673  *   Previous validated item in the pattern items.
2674  * @param[in] geneve_item
2675  *   Previous GENEVE item specification.
2676  * @param[in] dev
2677  *   Pointer to the rte_eth_dev structure.
2678  * @param[out] error
2679  *   Pointer to error structure.
2680  *
2681  * @return
2682  *   0 on success, a negative errno value otherwise and rte_errno is set.
2683  */
2684 int
2685 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2686 				   uint64_t last_item,
2687 				   const struct rte_flow_item *geneve_item,
2688 				   struct rte_eth_dev *dev,
2689 				   struct rte_flow_error *error)
2690 {
2691 	struct mlx5_priv *priv = dev->data->dev_private;
2692 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2693 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2694 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2695 	uint8_t data_max_supported =
2696 			hca_attr->max_geneve_tlv_option_data_len * 4;
2697 	struct mlx5_dev_config *config = &priv->config;
2698 	const struct rte_flow_item_geneve *geneve_spec;
2699 	const struct rte_flow_item_geneve *geneve_mask;
2700 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2701 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2702 	unsigned int i;
2703 	unsigned int data_len;
2704 	uint8_t tlv_option_len;
2705 	uint16_t optlen_m, optlen_v;
2706 	const struct rte_flow_item_geneve_opt full_mask = {
2707 		.option_class = RTE_BE16(0xffff),
2708 		.option_type = 0xff,
2709 		.option_len = 0x1f,
2710 	};
2711 
2712 	if (!mask)
2713 		mask = &rte_flow_item_geneve_opt_mask;
2714 	if (!spec)
2715 		return rte_flow_error_set
2716 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2717 			"Geneve TLV opt class/type/length must be specified");
2718 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2719 		return rte_flow_error_set
2720 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2721 			"Geneve TLV opt length exceeeds the limit (31)");
2722 	/* Check if class type and length masks are full. */
2723 	if (full_mask.option_class != mask->option_class ||
2724 	    full_mask.option_type != mask->option_type ||
2725 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2726 		return rte_flow_error_set
2727 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2728 			"Geneve TLV opt class/type/length masks must be full");
2729 	/* Check if length is supported */
2730 	if ((uint32_t)spec->option_len >
2731 			config->hca_attr.max_geneve_tlv_option_data_len)
2732 		return rte_flow_error_set
2733 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2734 			"Geneve TLV opt length not supported");
2735 	if (config->hca_attr.max_geneve_tlv_options > 1)
2736 		DRV_LOG(DEBUG,
2737 			"max_geneve_tlv_options supports more than 1 option");
2738 	/* Check GENEVE item preceding. */
2739 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2740 		return rte_flow_error_set
2741 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 			"Geneve opt item must be preceded with Geneve item");
2743 	geneve_spec = geneve_item->spec;
2744 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2745 					  &rte_flow_item_geneve_mask;
2746 	/* Check if GENEVE TLV option size doesn't exceed option length */
2747 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2748 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2749 		tlv_option_len = spec->option_len & mask->option_len;
2750 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2751 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2752 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2753 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2754 		if ((optlen_v & optlen_m) <= tlv_option_len)
2755 			return rte_flow_error_set
2756 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2757 				 "GENEVE TLV option length exceeds optlen");
2758 	}
2759 	/* Check if length is 0 or data is 0. */
2760 	if (spec->data == NULL || spec->option_len == 0)
2761 		return rte_flow_error_set
2762 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2763 			"Geneve TLV opt with zero data/length not supported");
2764 	/* Check not all data & mask are 0. */
2765 	data_len = spec->option_len * 4;
2766 	if (mask->data == NULL) {
2767 		for (i = 0; i < data_len; i++)
2768 			if (spec->data[i])
2769 				break;
2770 		if (i == data_len)
2771 			return rte_flow_error_set(error, ENOTSUP,
2772 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2773 				"Can't match on Geneve option data 0");
2774 	} else {
2775 		for (i = 0; i < data_len; i++)
2776 			if (spec->data[i] & mask->data[i])
2777 				break;
2778 		if (i == data_len)
2779 			return rte_flow_error_set(error, ENOTSUP,
2780 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2781 				"Can't match on Geneve option data and mask 0");
2782 		/* Check data mask supported. */
2783 		for (i = data_max_supported; i < data_len ; i++)
2784 			if (mask->data[i])
2785 				return rte_flow_error_set(error, ENOTSUP,
2786 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2787 					"Data mask is of unsupported size");
2788 	}
2789 	/* Check GENEVE option is supported in NIC. */
2790 	if (!config->hca_attr.geneve_tlv_opt)
2791 		return rte_flow_error_set
2792 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2793 			"Geneve TLV opt not supported");
2794 	/* Check if we already have geneve option with different type/class. */
2795 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2796 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2797 	if (geneve_opt_resource != NULL)
2798 		if (geneve_opt_resource->option_class != spec->option_class ||
2799 		    geneve_opt_resource->option_type != spec->option_type ||
2800 		    geneve_opt_resource->length != spec->option_len) {
2801 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2802 			return rte_flow_error_set(error, ENOTSUP,
2803 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2804 				"Only one Geneve TLV option supported");
2805 		}
2806 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2807 	return 0;
2808 }
2809 
2810 /**
2811  * Validate MPLS item.
2812  *
2813  * @param[in] dev
2814  *   Pointer to the rte_eth_dev structure.
2815  * @param[in] item
2816  *   Item specification.
2817  * @param[in] item_flags
2818  *   Bit-fields that holds the items detected until now.
2819  * @param[in] prev_layer
2820  *   The protocol layer indicated in previous item.
2821  * @param[out] error
2822  *   Pointer to error structure.
2823  *
2824  * @return
2825  *   0 on success, a negative errno value otherwise and rte_errno is set.
2826  */
2827 int
2828 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2829 			     const struct rte_flow_item *item __rte_unused,
2830 			     uint64_t item_flags __rte_unused,
2831 			     uint64_t prev_layer __rte_unused,
2832 			     struct rte_flow_error *error)
2833 {
2834 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2835 	const struct rte_flow_item_mpls *mask = item->mask;
2836 	struct mlx5_priv *priv = dev->data->dev_private;
2837 	int ret;
2838 
2839 	if (!priv->config.mpls_en)
2840 		return rte_flow_error_set(error, ENOTSUP,
2841 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2842 					  "MPLS not supported or"
2843 					  " disabled in firmware"
2844 					  " configuration.");
2845 	/* MPLS over IP, UDP, GRE is allowed */
2846 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2847 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2848 			    MLX5_FLOW_LAYER_GRE |
2849 			    MLX5_FLOW_LAYER_GRE_KEY)))
2850 		return rte_flow_error_set(error, EINVAL,
2851 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2852 					  "protocol filtering not compatible"
2853 					  " with MPLS layer");
2854 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2855 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2856 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2857 		return rte_flow_error_set(error, ENOTSUP,
2858 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2859 					  "multiple tunnel layers not"
2860 					  " supported");
2861 	if (!mask)
2862 		mask = &rte_flow_item_mpls_mask;
2863 	ret = mlx5_flow_item_acceptable
2864 		(item, (const uint8_t *)mask,
2865 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2866 		 sizeof(struct rte_flow_item_mpls),
2867 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2868 	if (ret < 0)
2869 		return ret;
2870 	return 0;
2871 #else
2872 	return rte_flow_error_set(error, ENOTSUP,
2873 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2874 				  "MPLS is not supported by Verbs, please"
2875 				  " update.");
2876 #endif
2877 }
2878 
2879 /**
2880  * Validate NVGRE item.
2881  *
2882  * @param[in] item
2883  *   Item specification.
2884  * @param[in] item_flags
2885  *   Bit flags to mark detected items.
2886  * @param[in] target_protocol
2887  *   The next protocol in the previous item.
2888  * @param[out] error
2889  *   Pointer to error structure.
2890  *
2891  * @return
2892  *   0 on success, a negative errno value otherwise and rte_errno is set.
2893  */
2894 int
2895 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2896 			      uint64_t item_flags,
2897 			      uint8_t target_protocol,
2898 			      struct rte_flow_error *error)
2899 {
2900 	const struct rte_flow_item_nvgre *mask = item->mask;
2901 	int ret;
2902 
2903 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2904 		return rte_flow_error_set(error, EINVAL,
2905 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2906 					  "protocol filtering not compatible"
2907 					  " with this GRE layer");
2908 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2909 		return rte_flow_error_set(error, ENOTSUP,
2910 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2911 					  "multiple tunnel layers not"
2912 					  " supported");
2913 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2914 		return rte_flow_error_set(error, ENOTSUP,
2915 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2916 					  "L3 Layer is missing");
2917 	if (!mask)
2918 		mask = &rte_flow_item_nvgre_mask;
2919 	ret = mlx5_flow_item_acceptable
2920 		(item, (const uint8_t *)mask,
2921 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2922 		 sizeof(struct rte_flow_item_nvgre),
2923 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2924 	if (ret < 0)
2925 		return ret;
2926 	return 0;
2927 }
2928 
2929 /**
2930  * Validate eCPRI item.
2931  *
2932  * @param[in] item
2933  *   Item specification.
2934  * @param[in] item_flags
2935  *   Bit-fields that holds the items detected until now.
2936  * @param[in] last_item
2937  *   Previous validated item in the pattern items.
2938  * @param[in] ether_type
2939  *   Type in the ethernet layer header (including dot1q).
2940  * @param[in] acc_mask
2941  *   Acceptable mask, if NULL default internal default mask
2942  *   will be used to check whether item fields are supported.
2943  * @param[out] error
2944  *   Pointer to error structure.
2945  *
2946  * @return
2947  *   0 on success, a negative errno value otherwise and rte_errno is set.
2948  */
2949 int
2950 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2951 			      uint64_t item_flags,
2952 			      uint64_t last_item,
2953 			      uint16_t ether_type,
2954 			      const struct rte_flow_item_ecpri *acc_mask,
2955 			      struct rte_flow_error *error)
2956 {
2957 	const struct rte_flow_item_ecpri *mask = item->mask;
2958 	const struct rte_flow_item_ecpri nic_mask = {
2959 		.hdr = {
2960 			.common = {
2961 				.u32 =
2962 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2963 					.type = 0xFF,
2964 					}).u32),
2965 			},
2966 			.dummy[0] = 0xFFFFFFFF,
2967 		},
2968 	};
2969 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2970 					MLX5_FLOW_LAYER_OUTER_VLAN);
2971 	struct rte_flow_item_ecpri mask_lo;
2972 
2973 	if (!(last_item & outer_l2_vlan) &&
2974 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2975 		return rte_flow_error_set(error, EINVAL,
2976 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2977 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2978 	if ((last_item & outer_l2_vlan) && ether_type &&
2979 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2980 		return rte_flow_error_set(error, EINVAL,
2981 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2982 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2983 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2984 		return rte_flow_error_set(error, EINVAL,
2985 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2986 					  "eCPRI with tunnel is not supported right now");
2987 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2988 		return rte_flow_error_set(error, ENOTSUP,
2989 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2990 					  "multiple L3 layers not supported");
2991 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2992 		return rte_flow_error_set(error, EINVAL,
2993 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2994 					  "eCPRI cannot coexist with a TCP layer");
2995 	/* In specification, eCPRI could be over UDP layer. */
2996 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2997 		return rte_flow_error_set(error, EINVAL,
2998 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2999 					  "eCPRI over UDP layer is not yet supported right now");
3000 	/* Mask for type field in common header could be zero. */
3001 	if (!mask)
3002 		mask = &rte_flow_item_ecpri_mask;
3003 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3004 	/* Input mask is in big-endian format. */
3005 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3006 		return rte_flow_error_set(error, EINVAL,
3007 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3008 					  "partial mask is not supported for protocol");
3009 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3010 		return rte_flow_error_set(error, EINVAL,
3011 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3012 					  "message header mask must be after a type mask");
3013 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3014 					 acc_mask ? (const uint8_t *)acc_mask
3015 						  : (const uint8_t *)&nic_mask,
3016 					 sizeof(struct rte_flow_item_ecpri),
3017 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3018 }
3019 
3020 /**
3021  * Release resource related QUEUE/RSS action split.
3022  *
3023  * @param dev
3024  *   Pointer to Ethernet device.
3025  * @param flow
3026  *   Flow to release id's from.
3027  */
3028 static void
3029 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3030 			     struct rte_flow *flow)
3031 {
3032 	struct mlx5_priv *priv = dev->data->dev_private;
3033 	uint32_t handle_idx;
3034 	struct mlx5_flow_handle *dev_handle;
3035 
3036 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3037 		       handle_idx, dev_handle, next)
3038 		if (dev_handle->split_flow_id)
3039 			mlx5_ipool_free(priv->sh->ipool
3040 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3041 					dev_handle->split_flow_id);
3042 }
3043 
3044 static int
3045 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3046 		   const struct rte_flow_attr *attr __rte_unused,
3047 		   const struct rte_flow_item items[] __rte_unused,
3048 		   const struct rte_flow_action actions[] __rte_unused,
3049 		   bool external __rte_unused,
3050 		   int hairpin __rte_unused,
3051 		   struct rte_flow_error *error)
3052 {
3053 	return rte_flow_error_set(error, ENOTSUP,
3054 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3055 }
3056 
3057 static struct mlx5_flow *
3058 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3059 		  const struct rte_flow_attr *attr __rte_unused,
3060 		  const struct rte_flow_item items[] __rte_unused,
3061 		  const struct rte_flow_action actions[] __rte_unused,
3062 		  struct rte_flow_error *error)
3063 {
3064 	rte_flow_error_set(error, ENOTSUP,
3065 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3066 	return NULL;
3067 }
3068 
3069 static int
3070 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3071 		    struct mlx5_flow *dev_flow __rte_unused,
3072 		    const struct rte_flow_attr *attr __rte_unused,
3073 		    const struct rte_flow_item items[] __rte_unused,
3074 		    const struct rte_flow_action actions[] __rte_unused,
3075 		    struct rte_flow_error *error)
3076 {
3077 	return rte_flow_error_set(error, ENOTSUP,
3078 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3079 }
3080 
3081 static int
3082 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3083 		struct rte_flow *flow __rte_unused,
3084 		struct rte_flow_error *error)
3085 {
3086 	return rte_flow_error_set(error, ENOTSUP,
3087 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3088 }
3089 
3090 static void
3091 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3092 		 struct rte_flow *flow __rte_unused)
3093 {
3094 }
3095 
3096 static void
3097 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3098 		  struct rte_flow *flow __rte_unused)
3099 {
3100 }
3101 
3102 static int
3103 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3104 		struct rte_flow *flow __rte_unused,
3105 		const struct rte_flow_action *actions __rte_unused,
3106 		void *data __rte_unused,
3107 		struct rte_flow_error *error)
3108 {
3109 	return rte_flow_error_set(error, ENOTSUP,
3110 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3111 }
3112 
3113 static int
3114 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3115 		      uint32_t domains __rte_unused,
3116 		      uint32_t flags __rte_unused)
3117 {
3118 	return 0;
3119 }
3120 
3121 /* Void driver to protect from null pointer reference. */
3122 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3123 	.validate = flow_null_validate,
3124 	.prepare = flow_null_prepare,
3125 	.translate = flow_null_translate,
3126 	.apply = flow_null_apply,
3127 	.remove = flow_null_remove,
3128 	.destroy = flow_null_destroy,
3129 	.query = flow_null_query,
3130 	.sync_domain = flow_null_sync_domain,
3131 };
3132 
3133 /**
3134  * Select flow driver type according to flow attributes and device
3135  * configuration.
3136  *
3137  * @param[in] dev
3138  *   Pointer to the dev structure.
3139  * @param[in] attr
3140  *   Pointer to the flow attributes.
3141  *
3142  * @return
3143  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3144  */
3145 static enum mlx5_flow_drv_type
3146 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3147 {
3148 	struct mlx5_priv *priv = dev->data->dev_private;
3149 	/* The OS can determine first a specific flow type (DV, VERBS) */
3150 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3151 
3152 	if (type != MLX5_FLOW_TYPE_MAX)
3153 		return type;
3154 	/* If no OS specific type - continue with DV/VERBS selection */
3155 	if (attr->transfer && priv->config.dv_esw_en)
3156 		type = MLX5_FLOW_TYPE_DV;
3157 	if (!attr->transfer)
3158 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3159 						 MLX5_FLOW_TYPE_VERBS;
3160 	return type;
3161 }
3162 
3163 #define flow_get_drv_ops(type) flow_drv_ops[type]
3164 
3165 /**
3166  * Flow driver validation API. This abstracts calling driver specific functions.
3167  * The type of flow driver is determined according to flow attributes.
3168  *
3169  * @param[in] dev
3170  *   Pointer to the dev structure.
3171  * @param[in] attr
3172  *   Pointer to the flow attributes.
3173  * @param[in] items
3174  *   Pointer to the list of items.
3175  * @param[in] actions
3176  *   Pointer to the list of actions.
3177  * @param[in] external
3178  *   This flow rule is created by request external to PMD.
3179  * @param[in] hairpin
3180  *   Number of hairpin TX actions, 0 means classic flow.
3181  * @param[out] error
3182  *   Pointer to the error structure.
3183  *
3184  * @return
3185  *   0 on success, a negative errno value otherwise and rte_errno is set.
3186  */
3187 static inline int
3188 flow_drv_validate(struct rte_eth_dev *dev,
3189 		  const struct rte_flow_attr *attr,
3190 		  const struct rte_flow_item items[],
3191 		  const struct rte_flow_action actions[],
3192 		  bool external, int hairpin, struct rte_flow_error *error)
3193 {
3194 	const struct mlx5_flow_driver_ops *fops;
3195 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3196 
3197 	fops = flow_get_drv_ops(type);
3198 	return fops->validate(dev, attr, items, actions, external,
3199 			      hairpin, error);
3200 }
3201 
3202 /**
3203  * Flow driver preparation API. This abstracts calling driver specific
3204  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3205  * calculates the size of memory required for device flow, allocates the memory,
3206  * initializes the device flow and returns the pointer.
3207  *
3208  * @note
3209  *   This function initializes device flow structure such as dv or verbs in
3210  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3211  *   rest. For example, adding returning device flow to flow->dev_flow list and
3212  *   setting backward reference to the flow should be done out of this function.
3213  *   layers field is not filled either.
3214  *
3215  * @param[in] dev
3216  *   Pointer to the dev structure.
3217  * @param[in] attr
3218  *   Pointer to the flow attributes.
3219  * @param[in] items
3220  *   Pointer to the list of items.
3221  * @param[in] actions
3222  *   Pointer to the list of actions.
3223  * @param[in] flow_idx
3224  *   This memory pool index to the flow.
3225  * @param[out] error
3226  *   Pointer to the error structure.
3227  *
3228  * @return
3229  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3230  */
3231 static inline struct mlx5_flow *
3232 flow_drv_prepare(struct rte_eth_dev *dev,
3233 		 const struct rte_flow *flow,
3234 		 const struct rte_flow_attr *attr,
3235 		 const struct rte_flow_item items[],
3236 		 const struct rte_flow_action actions[],
3237 		 uint32_t flow_idx,
3238 		 struct rte_flow_error *error)
3239 {
3240 	const struct mlx5_flow_driver_ops *fops;
3241 	enum mlx5_flow_drv_type type = flow->drv_type;
3242 	struct mlx5_flow *mlx5_flow = NULL;
3243 
3244 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3245 	fops = flow_get_drv_ops(type);
3246 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3247 	if (mlx5_flow)
3248 		mlx5_flow->flow_idx = flow_idx;
3249 	return mlx5_flow;
3250 }
3251 
3252 /**
3253  * Flow driver translation API. This abstracts calling driver specific
3254  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3255  * translates a generic flow into a driver flow. flow_drv_prepare() must
3256  * precede.
3257  *
3258  * @note
3259  *   dev_flow->layers could be filled as a result of parsing during translation
3260  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3261  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3262  *   flow->actions could be overwritten even though all the expanded dev_flows
3263  *   have the same actions.
3264  *
3265  * @param[in] dev
3266  *   Pointer to the rte dev structure.
3267  * @param[in, out] dev_flow
3268  *   Pointer to the mlx5 flow.
3269  * @param[in] attr
3270  *   Pointer to the flow attributes.
3271  * @param[in] items
3272  *   Pointer to the list of items.
3273  * @param[in] actions
3274  *   Pointer to the list of actions.
3275  * @param[out] error
3276  *   Pointer to the error structure.
3277  *
3278  * @return
3279  *   0 on success, a negative errno value otherwise and rte_errno is set.
3280  */
3281 static inline int
3282 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3283 		   const struct rte_flow_attr *attr,
3284 		   const struct rte_flow_item items[],
3285 		   const struct rte_flow_action actions[],
3286 		   struct rte_flow_error *error)
3287 {
3288 	const struct mlx5_flow_driver_ops *fops;
3289 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3290 
3291 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3292 	fops = flow_get_drv_ops(type);
3293 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3294 }
3295 
3296 /**
3297  * Flow driver apply API. This abstracts calling driver specific functions.
3298  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3299  * translated driver flows on to device. flow_drv_translate() must precede.
3300  *
3301  * @param[in] dev
3302  *   Pointer to Ethernet device structure.
3303  * @param[in, out] flow
3304  *   Pointer to flow structure.
3305  * @param[out] error
3306  *   Pointer to error structure.
3307  *
3308  * @return
3309  *   0 on success, a negative errno value otherwise and rte_errno is set.
3310  */
3311 static inline int
3312 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3313 	       struct rte_flow_error *error)
3314 {
3315 	const struct mlx5_flow_driver_ops *fops;
3316 	enum mlx5_flow_drv_type type = flow->drv_type;
3317 
3318 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3319 	fops = flow_get_drv_ops(type);
3320 	return fops->apply(dev, flow, error);
3321 }
3322 
3323 /**
3324  * Flow driver destroy API. This abstracts calling driver specific functions.
3325  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3326  * on device and releases resources of the flow.
3327  *
3328  * @param[in] dev
3329  *   Pointer to Ethernet device.
3330  * @param[in, out] flow
3331  *   Pointer to flow structure.
3332  */
3333 static inline void
3334 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3335 {
3336 	const struct mlx5_flow_driver_ops *fops;
3337 	enum mlx5_flow_drv_type type = flow->drv_type;
3338 
3339 	flow_mreg_split_qrss_release(dev, flow);
3340 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3341 	fops = flow_get_drv_ops(type);
3342 	fops->destroy(dev, flow);
3343 }
3344 
3345 /**
3346  * Get RSS action from the action list.
3347  *
3348  * @param[in] actions
3349  *   Pointer to the list of actions.
3350  *
3351  * @return
3352  *   Pointer to the RSS action if exist, else return NULL.
3353  */
3354 static const struct rte_flow_action_rss*
3355 flow_get_rss_action(const struct rte_flow_action actions[])
3356 {
3357 	const struct rte_flow_action_rss *rss = NULL;
3358 
3359 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3360 		switch (actions->type) {
3361 		case RTE_FLOW_ACTION_TYPE_RSS:
3362 			rss = actions->conf;
3363 			break;
3364 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3365 		{
3366 			const struct rte_flow_action_sample *sample =
3367 								actions->conf;
3368 			const struct rte_flow_action *act = sample->actions;
3369 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3370 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3371 					rss = act->conf;
3372 			break;
3373 		}
3374 		default:
3375 			break;
3376 		}
3377 	}
3378 	return rss;
3379 }
3380 
3381 /**
3382  * Get ASO age action by index.
3383  *
3384  * @param[in] dev
3385  *   Pointer to the Ethernet device structure.
3386  * @param[in] age_idx
3387  *   Index to the ASO age action.
3388  *
3389  * @return
3390  *   The specified ASO age action.
3391  */
3392 struct mlx5_aso_age_action*
3393 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3394 {
3395 	uint16_t pool_idx = age_idx & UINT16_MAX;
3396 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3397 	struct mlx5_priv *priv = dev->data->dev_private;
3398 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3399 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3400 
3401 	return &pool->actions[offset - 1];
3402 }
3403 
3404 /* maps indirect action to translated direct in some actions array */
3405 struct mlx5_translated_action_handle {
3406 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3407 	int index; /**< Index in related array of rte_flow_action. */
3408 };
3409 
3410 /**
3411  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3412  * direct action if translation possible.
3413  * This functionality used to run same execution path for both direct and
3414  * indirect actions on flow create. All necessary preparations for indirect
3415  * action handling should be performed on *handle* actions list returned
3416  * from this call.
3417  *
3418  * @param[in] dev
3419  *   Pointer to Ethernet device.
3420  * @param[in] actions
3421  *   List of actions to translate.
3422  * @param[out] handle
3423  *   List to store translated indirect action object handles.
3424  * @param[in, out] indir_n
3425  *   Size of *handle* array. On return should be updated with number of
3426  *   indirect actions retrieved from the *actions* list.
3427  * @param[out] translated_actions
3428  *   List of actions where all indirect actions were translated to direct
3429  *   if possible. NULL if no translation took place.
3430  * @param[out] error
3431  *   Pointer to the error structure.
3432  *
3433  * @return
3434  *   0 on success, a negative errno value otherwise and rte_errno is set.
3435  */
3436 static int
3437 flow_action_handles_translate(struct rte_eth_dev *dev,
3438 			      const struct rte_flow_action actions[],
3439 			      struct mlx5_translated_action_handle *handle,
3440 			      int *indir_n,
3441 			      struct rte_flow_action **translated_actions,
3442 			      struct rte_flow_error *error)
3443 {
3444 	struct mlx5_priv *priv = dev->data->dev_private;
3445 	struct rte_flow_action *translated = NULL;
3446 	size_t actions_size;
3447 	int n;
3448 	int copied_n = 0;
3449 	struct mlx5_translated_action_handle *handle_end = NULL;
3450 
3451 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3452 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3453 			continue;
3454 		if (copied_n == *indir_n) {
3455 			return rte_flow_error_set
3456 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3457 				 NULL, "too many shared actions");
3458 		}
3459 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3460 			   sizeof(actions[n].conf));
3461 		handle[copied_n].index = n;
3462 		copied_n++;
3463 	}
3464 	n++;
3465 	*indir_n = copied_n;
3466 	if (!copied_n)
3467 		return 0;
3468 	actions_size = sizeof(struct rte_flow_action) * n;
3469 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3470 	if (!translated) {
3471 		rte_errno = ENOMEM;
3472 		return -ENOMEM;
3473 	}
3474 	memcpy(translated, actions, actions_size);
3475 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3476 		struct mlx5_shared_action_rss *shared_rss;
3477 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3478 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3479 		uint32_t idx = act_idx &
3480 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3481 
3482 		switch (type) {
3483 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3484 			shared_rss = mlx5_ipool_get
3485 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3486 			translated[handle->index].type =
3487 				RTE_FLOW_ACTION_TYPE_RSS;
3488 			translated[handle->index].conf =
3489 				&shared_rss->origin;
3490 			break;
3491 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3492 			if (priv->sh->flow_hit_aso_en) {
3493 				translated[handle->index].type =
3494 					(enum rte_flow_action_type)
3495 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3496 				translated[handle->index].conf =
3497 							 (void *)(uintptr_t)idx;
3498 				break;
3499 			}
3500 			/* Fall-through */
3501 		default:
3502 			mlx5_free(translated);
3503 			return rte_flow_error_set
3504 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3505 				 NULL, "invalid indirect action type");
3506 		}
3507 	}
3508 	*translated_actions = translated;
3509 	return 0;
3510 }
3511 
3512 /**
3513  * Get Shared RSS action from the action list.
3514  *
3515  * @param[in] dev
3516  *   Pointer to Ethernet device.
3517  * @param[in] shared
3518  *   Pointer to the list of actions.
3519  * @param[in] shared_n
3520  *   Actions list length.
3521  *
3522  * @return
3523  *   The MLX5 RSS action ID if exists, otherwise return 0.
3524  */
3525 static uint32_t
3526 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3527 			   struct mlx5_translated_action_handle *handle,
3528 			   int shared_n)
3529 {
3530 	struct mlx5_translated_action_handle *handle_end;
3531 	struct mlx5_priv *priv = dev->data->dev_private;
3532 	struct mlx5_shared_action_rss *shared_rss;
3533 
3534 
3535 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3536 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3537 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3538 		uint32_t idx = act_idx &
3539 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3540 		switch (type) {
3541 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3542 			shared_rss = mlx5_ipool_get
3543 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3544 									   idx);
3545 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3546 					   __ATOMIC_RELAXED);
3547 			return idx;
3548 		default:
3549 			break;
3550 		}
3551 	}
3552 	return 0;
3553 }
3554 
3555 static unsigned int
3556 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3557 {
3558 	const struct rte_flow_item *item;
3559 	unsigned int has_vlan = 0;
3560 
3561 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3562 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3563 			has_vlan = 1;
3564 			break;
3565 		}
3566 	}
3567 	if (has_vlan)
3568 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3569 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3570 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3571 			       MLX5_EXPANSION_ROOT_OUTER;
3572 }
3573 
3574 /**
3575  *  Get layer flags from the prefix flow.
3576  *
3577  *  Some flows may be split to several subflows, the prefix subflow gets the
3578  *  match items and the suffix sub flow gets the actions.
3579  *  Some actions need the user defined match item flags to get the detail for
3580  *  the action.
3581  *  This function helps the suffix flow to get the item layer flags from prefix
3582  *  subflow.
3583  *
3584  * @param[in] dev_flow
3585  *   Pointer the created preifx subflow.
3586  *
3587  * @return
3588  *   The layers get from prefix subflow.
3589  */
3590 static inline uint64_t
3591 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3592 {
3593 	uint64_t layers = 0;
3594 
3595 	/*
3596 	 * Layers bits could be localization, but usually the compiler will
3597 	 * help to do the optimization work for source code.
3598 	 * If no decap actions, use the layers directly.
3599 	 */
3600 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3601 		return dev_flow->handle->layers;
3602 	/* Convert L3 layers with decap action. */
3603 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3604 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3605 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3606 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3607 	/* Convert L4 layers with decap action.  */
3608 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3609 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3610 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3611 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3612 	return layers;
3613 }
3614 
3615 /**
3616  * Get metadata split action information.
3617  *
3618  * @param[in] actions
3619  *   Pointer to the list of actions.
3620  * @param[out] qrss
3621  *   Pointer to the return pointer.
3622  * @param[out] qrss_type
3623  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3624  *   if no QUEUE/RSS is found.
3625  * @param[out] encap_idx
3626  *   Pointer to the index of the encap action if exists, otherwise the last
3627  *   action index.
3628  *
3629  * @return
3630  *   Total number of actions.
3631  */
3632 static int
3633 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3634 				       const struct rte_flow_action **qrss,
3635 				       int *encap_idx)
3636 {
3637 	const struct rte_flow_action_raw_encap *raw_encap;
3638 	int actions_n = 0;
3639 	int raw_decap_idx = -1;
3640 
3641 	*encap_idx = -1;
3642 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3643 		switch (actions->type) {
3644 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3645 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3646 			*encap_idx = actions_n;
3647 			break;
3648 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3649 			raw_decap_idx = actions_n;
3650 			break;
3651 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3652 			raw_encap = actions->conf;
3653 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3654 				*encap_idx = raw_decap_idx != -1 ?
3655 						      raw_decap_idx : actions_n;
3656 			break;
3657 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3658 		case RTE_FLOW_ACTION_TYPE_RSS:
3659 			*qrss = actions;
3660 			break;
3661 		default:
3662 			break;
3663 		}
3664 		actions_n++;
3665 	}
3666 	if (*encap_idx == -1)
3667 		*encap_idx = actions_n;
3668 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3669 	return actions_n + 1;
3670 }
3671 
3672 /**
3673  * Check meter action from the action list.
3674  *
3675  * @param[in] actions
3676  *   Pointer to the list of actions.
3677  * @param[out] mtr
3678  *   Pointer to the meter exist flag.
3679  *
3680  * @return
3681  *   Total number of actions.
3682  */
3683 static int
3684 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3685 {
3686 	int actions_n = 0;
3687 
3688 	MLX5_ASSERT(mtr);
3689 	*mtr = 0;
3690 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3691 		switch (actions->type) {
3692 		case RTE_FLOW_ACTION_TYPE_METER:
3693 			*mtr = 1;
3694 			break;
3695 		default:
3696 			break;
3697 		}
3698 		actions_n++;
3699 	}
3700 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3701 	return actions_n + 1;
3702 }
3703 
3704 /**
3705  * Check if the flow should be split due to hairpin.
3706  * The reason for the split is that in current HW we can't
3707  * support encap and push-vlan on Rx, so if a flow contains
3708  * these actions we move it to Tx.
3709  *
3710  * @param dev
3711  *   Pointer to Ethernet device.
3712  * @param[in] attr
3713  *   Flow rule attributes.
3714  * @param[in] actions
3715  *   Associated actions (list terminated by the END action).
3716  *
3717  * @return
3718  *   > 0 the number of actions and the flow should be split,
3719  *   0 when no split required.
3720  */
3721 static int
3722 flow_check_hairpin_split(struct rte_eth_dev *dev,
3723 			 const struct rte_flow_attr *attr,
3724 			 const struct rte_flow_action actions[])
3725 {
3726 	int queue_action = 0;
3727 	int action_n = 0;
3728 	int split = 0;
3729 	const struct rte_flow_action_queue *queue;
3730 	const struct rte_flow_action_rss *rss;
3731 	const struct rte_flow_action_raw_encap *raw_encap;
3732 	const struct rte_eth_hairpin_conf *conf;
3733 
3734 	if (!attr->ingress)
3735 		return 0;
3736 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3737 		switch (actions->type) {
3738 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3739 			queue = actions->conf;
3740 			if (queue == NULL)
3741 				return 0;
3742 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3743 			if (conf == NULL || conf->tx_explicit != 0)
3744 				return 0;
3745 			queue_action = 1;
3746 			action_n++;
3747 			break;
3748 		case RTE_FLOW_ACTION_TYPE_RSS:
3749 			rss = actions->conf;
3750 			if (rss == NULL || rss->queue_num == 0)
3751 				return 0;
3752 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3753 			if (conf == NULL || conf->tx_explicit != 0)
3754 				return 0;
3755 			queue_action = 1;
3756 			action_n++;
3757 			break;
3758 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3759 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3760 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3761 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3762 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3763 			split++;
3764 			action_n++;
3765 			break;
3766 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3767 			raw_encap = actions->conf;
3768 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3769 				split++;
3770 			action_n++;
3771 			break;
3772 		default:
3773 			action_n++;
3774 			break;
3775 		}
3776 	}
3777 	if (split && queue_action)
3778 		return action_n;
3779 	return 0;
3780 }
3781 
3782 /* Declare flow create/destroy prototype in advance. */
3783 static uint32_t
3784 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3785 		 const struct rte_flow_attr *attr,
3786 		 const struct rte_flow_item items[],
3787 		 const struct rte_flow_action actions[],
3788 		 bool external, struct rte_flow_error *error);
3789 
3790 static void
3791 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3792 		  uint32_t flow_idx);
3793 
3794 int
3795 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3796 		      struct mlx5_hlist_entry *entry,
3797 		      uint64_t key, void *cb_ctx __rte_unused)
3798 {
3799 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3800 		container_of(entry, typeof(*mcp_res), hlist_ent);
3801 
3802 	return mcp_res->mark_id != key;
3803 }
3804 
3805 struct mlx5_hlist_entry *
3806 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3807 		       void *cb_ctx)
3808 {
3809 	struct rte_eth_dev *dev = list->ctx;
3810 	struct mlx5_priv *priv = dev->data->dev_private;
3811 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3812 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3813 	struct rte_flow_error *error = ctx->error;
3814 	uint32_t idx = 0;
3815 	int ret;
3816 	uint32_t mark_id = key;
3817 	struct rte_flow_attr attr = {
3818 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3819 		.ingress = 1,
3820 	};
3821 	struct mlx5_rte_flow_item_tag tag_spec = {
3822 		.data = mark_id,
3823 	};
3824 	struct rte_flow_item items[] = {
3825 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3826 	};
3827 	struct rte_flow_action_mark ftag = {
3828 		.id = mark_id,
3829 	};
3830 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3831 		.dst = REG_B,
3832 		.src = REG_NON,
3833 	};
3834 	struct rte_flow_action_jump jump = {
3835 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3836 	};
3837 	struct rte_flow_action actions[] = {
3838 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3839 	};
3840 
3841 	/* Fill the register fileds in the flow. */
3842 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3843 	if (ret < 0)
3844 		return NULL;
3845 	tag_spec.id = ret;
3846 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3847 	if (ret < 0)
3848 		return NULL;
3849 	cp_mreg.src = ret;
3850 	/* Provide the full width of FLAG specific value. */
3851 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3852 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3853 	/* Build a new flow. */
3854 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3855 		items[0] = (struct rte_flow_item){
3856 			.type = (enum rte_flow_item_type)
3857 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3858 			.spec = &tag_spec,
3859 		};
3860 		items[1] = (struct rte_flow_item){
3861 			.type = RTE_FLOW_ITEM_TYPE_END,
3862 		};
3863 		actions[0] = (struct rte_flow_action){
3864 			.type = (enum rte_flow_action_type)
3865 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3866 			.conf = &ftag,
3867 		};
3868 		actions[1] = (struct rte_flow_action){
3869 			.type = (enum rte_flow_action_type)
3870 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3871 			.conf = &cp_mreg,
3872 		};
3873 		actions[2] = (struct rte_flow_action){
3874 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3875 			.conf = &jump,
3876 		};
3877 		actions[3] = (struct rte_flow_action){
3878 			.type = RTE_FLOW_ACTION_TYPE_END,
3879 		};
3880 	} else {
3881 		/* Default rule, wildcard match. */
3882 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
3883 		items[0] = (struct rte_flow_item){
3884 			.type = RTE_FLOW_ITEM_TYPE_END,
3885 		};
3886 		actions[0] = (struct rte_flow_action){
3887 			.type = (enum rte_flow_action_type)
3888 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3889 			.conf = &cp_mreg,
3890 		};
3891 		actions[1] = (struct rte_flow_action){
3892 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3893 			.conf = &jump,
3894 		};
3895 		actions[2] = (struct rte_flow_action){
3896 			.type = RTE_FLOW_ACTION_TYPE_END,
3897 		};
3898 	}
3899 	/* Build a new entry. */
3900 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3901 	if (!mcp_res) {
3902 		rte_errno = ENOMEM;
3903 		return NULL;
3904 	}
3905 	mcp_res->idx = idx;
3906 	mcp_res->mark_id = mark_id;
3907 	/*
3908 	 * The copy Flows are not included in any list. There
3909 	 * ones are referenced from other Flows and can not
3910 	 * be applied, removed, deleted in ardbitrary order
3911 	 * by list traversing.
3912 	 */
3913 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3914 					 actions, false, error);
3915 	if (!mcp_res->rix_flow) {
3916 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3917 		return NULL;
3918 	}
3919 	return &mcp_res->hlist_ent;
3920 }
3921 
3922 /**
3923  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3924  *
3925  * As mark_id is unique, if there's already a registered flow for the mark_id,
3926  * return by increasing the reference counter of the resource. Otherwise, create
3927  * the resource (mcp_res) and flow.
3928  *
3929  * Flow looks like,
3930  *   - If ingress port is ANY and reg_c[1] is mark_id,
3931  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3932  *
3933  * For default flow (zero mark_id), flow is like,
3934  *   - If ingress port is ANY,
3935  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3936  *
3937  * @param dev
3938  *   Pointer to Ethernet device.
3939  * @param mark_id
3940  *   ID of MARK action, zero means default flow for META.
3941  * @param[out] error
3942  *   Perform verbose error reporting if not NULL.
3943  *
3944  * @return
3945  *   Associated resource on success, NULL otherwise and rte_errno is set.
3946  */
3947 static struct mlx5_flow_mreg_copy_resource *
3948 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3949 			  struct rte_flow_error *error)
3950 {
3951 	struct mlx5_priv *priv = dev->data->dev_private;
3952 	struct mlx5_hlist_entry *entry;
3953 	struct mlx5_flow_cb_ctx ctx = {
3954 		.dev = dev,
3955 		.error = error,
3956 	};
3957 
3958 	/* Check if already registered. */
3959 	MLX5_ASSERT(priv->mreg_cp_tbl);
3960 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3961 	if (!entry)
3962 		return NULL;
3963 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3964 			    hlist_ent);
3965 }
3966 
3967 void
3968 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3969 {
3970 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3971 		container_of(entry, typeof(*mcp_res), hlist_ent);
3972 	struct rte_eth_dev *dev = list->ctx;
3973 	struct mlx5_priv *priv = dev->data->dev_private;
3974 
3975 	MLX5_ASSERT(mcp_res->rix_flow);
3976 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3977 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3978 }
3979 
3980 /**
3981  * Release flow in RX_CP_TBL.
3982  *
3983  * @param dev
3984  *   Pointer to Ethernet device.
3985  * @flow
3986  *   Parent flow for wich copying is provided.
3987  */
3988 static void
3989 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3990 			  struct rte_flow *flow)
3991 {
3992 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3993 	struct mlx5_priv *priv = dev->data->dev_private;
3994 
3995 	if (!flow->rix_mreg_copy)
3996 		return;
3997 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3998 				 flow->rix_mreg_copy);
3999 	if (!mcp_res || !priv->mreg_cp_tbl)
4000 		return;
4001 	MLX5_ASSERT(mcp_res->rix_flow);
4002 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4003 	flow->rix_mreg_copy = 0;
4004 }
4005 
4006 /**
4007  * Remove the default copy action from RX_CP_TBL.
4008  *
4009  * This functions is called in the mlx5_dev_start(). No thread safe
4010  * is guaranteed.
4011  *
4012  * @param dev
4013  *   Pointer to Ethernet device.
4014  */
4015 static void
4016 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4017 {
4018 	struct mlx5_hlist_entry *entry;
4019 	struct mlx5_priv *priv = dev->data->dev_private;
4020 
4021 	/* Check if default flow is registered. */
4022 	if (!priv->mreg_cp_tbl)
4023 		return;
4024 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4025 				  MLX5_DEFAULT_COPY_ID, NULL);
4026 	if (!entry)
4027 		return;
4028 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4029 }
4030 
4031 /**
4032  * Add the default copy action in in RX_CP_TBL.
4033  *
4034  * This functions is called in the mlx5_dev_start(). No thread safe
4035  * is guaranteed.
4036  *
4037  * @param dev
4038  *   Pointer to Ethernet device.
4039  * @param[out] error
4040  *   Perform verbose error reporting if not NULL.
4041  *
4042  * @return
4043  *   0 for success, negative value otherwise and rte_errno is set.
4044  */
4045 static int
4046 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4047 				  struct rte_flow_error *error)
4048 {
4049 	struct mlx5_priv *priv = dev->data->dev_private;
4050 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4051 
4052 	/* Check whether extensive metadata feature is engaged. */
4053 	if (!priv->config.dv_flow_en ||
4054 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4055 	    !mlx5_flow_ext_mreg_supported(dev) ||
4056 	    !priv->sh->dv_regc0_mask)
4057 		return 0;
4058 	/*
4059 	 * Add default mreg copy flow may be called multiple time, but
4060 	 * only be called once in stop. Avoid register it twice.
4061 	 */
4062 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4063 		return 0;
4064 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4065 	if (!mcp_res)
4066 		return -rte_errno;
4067 	return 0;
4068 }
4069 
4070 /**
4071  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4072  *
4073  * All the flow having Q/RSS action should be split by
4074  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4075  * performs the following,
4076  *   - CQE->flow_tag := reg_c[1] (MARK)
4077  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4078  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4079  * but there should be a flow per each MARK ID set by MARK action.
4080  *
4081  * For the aforementioned reason, if there's a MARK action in flow's action
4082  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4083  * the MARK ID to CQE's flow_tag like,
4084  *   - If reg_c[1] is mark_id,
4085  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4086  *
4087  * For SET_META action which stores value in reg_c[0], as the destination is
4088  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4089  * MARK ID means the default flow. The default flow looks like,
4090  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4091  *
4092  * @param dev
4093  *   Pointer to Ethernet device.
4094  * @param flow
4095  *   Pointer to flow structure.
4096  * @param[in] actions
4097  *   Pointer to the list of actions.
4098  * @param[out] error
4099  *   Perform verbose error reporting if not NULL.
4100  *
4101  * @return
4102  *   0 on success, negative value otherwise and rte_errno is set.
4103  */
4104 static int
4105 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4106 			    struct rte_flow *flow,
4107 			    const struct rte_flow_action *actions,
4108 			    struct rte_flow_error *error)
4109 {
4110 	struct mlx5_priv *priv = dev->data->dev_private;
4111 	struct mlx5_dev_config *config = &priv->config;
4112 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4113 	const struct rte_flow_action_mark *mark;
4114 
4115 	/* Check whether extensive metadata feature is engaged. */
4116 	if (!config->dv_flow_en ||
4117 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4118 	    !mlx5_flow_ext_mreg_supported(dev) ||
4119 	    !priv->sh->dv_regc0_mask)
4120 		return 0;
4121 	/* Find MARK action. */
4122 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4123 		switch (actions->type) {
4124 		case RTE_FLOW_ACTION_TYPE_FLAG:
4125 			mcp_res = flow_mreg_add_copy_action
4126 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4127 			if (!mcp_res)
4128 				return -rte_errno;
4129 			flow->rix_mreg_copy = mcp_res->idx;
4130 			return 0;
4131 		case RTE_FLOW_ACTION_TYPE_MARK:
4132 			mark = (const struct rte_flow_action_mark *)
4133 				actions->conf;
4134 			mcp_res =
4135 				flow_mreg_add_copy_action(dev, mark->id, error);
4136 			if (!mcp_res)
4137 				return -rte_errno;
4138 			flow->rix_mreg_copy = mcp_res->idx;
4139 			return 0;
4140 		default:
4141 			break;
4142 		}
4143 	}
4144 	return 0;
4145 }
4146 
4147 #define MLX5_MAX_SPLIT_ACTIONS 24
4148 #define MLX5_MAX_SPLIT_ITEMS 24
4149 
4150 /**
4151  * Split the hairpin flow.
4152  * Since HW can't support encap and push-vlan on Rx, we move these
4153  * actions to Tx.
4154  * If the count action is after the encap then we also
4155  * move the count action. in this case the count will also measure
4156  * the outer bytes.
4157  *
4158  * @param dev
4159  *   Pointer to Ethernet device.
4160  * @param[in] actions
4161  *   Associated actions (list terminated by the END action).
4162  * @param[out] actions_rx
4163  *   Rx flow actions.
4164  * @param[out] actions_tx
4165  *   Tx flow actions..
4166  * @param[out] pattern_tx
4167  *   The pattern items for the Tx flow.
4168  * @param[out] flow_id
4169  *   The flow ID connected to this flow.
4170  *
4171  * @return
4172  *   0 on success.
4173  */
4174 static int
4175 flow_hairpin_split(struct rte_eth_dev *dev,
4176 		   const struct rte_flow_action actions[],
4177 		   struct rte_flow_action actions_rx[],
4178 		   struct rte_flow_action actions_tx[],
4179 		   struct rte_flow_item pattern_tx[],
4180 		   uint32_t flow_id)
4181 {
4182 	const struct rte_flow_action_raw_encap *raw_encap;
4183 	const struct rte_flow_action_raw_decap *raw_decap;
4184 	struct mlx5_rte_flow_action_set_tag *set_tag;
4185 	struct rte_flow_action *tag_action;
4186 	struct mlx5_rte_flow_item_tag *tag_item;
4187 	struct rte_flow_item *item;
4188 	char *addr;
4189 	int encap = 0;
4190 
4191 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4192 		switch (actions->type) {
4193 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4194 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4195 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4196 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4197 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4198 			rte_memcpy(actions_tx, actions,
4199 			       sizeof(struct rte_flow_action));
4200 			actions_tx++;
4201 			break;
4202 		case RTE_FLOW_ACTION_TYPE_COUNT:
4203 			if (encap) {
4204 				rte_memcpy(actions_tx, actions,
4205 					   sizeof(struct rte_flow_action));
4206 				actions_tx++;
4207 			} else {
4208 				rte_memcpy(actions_rx, actions,
4209 					   sizeof(struct rte_flow_action));
4210 				actions_rx++;
4211 			}
4212 			break;
4213 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4214 			raw_encap = actions->conf;
4215 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4216 				memcpy(actions_tx, actions,
4217 				       sizeof(struct rte_flow_action));
4218 				actions_tx++;
4219 				encap = 1;
4220 			} else {
4221 				rte_memcpy(actions_rx, actions,
4222 					   sizeof(struct rte_flow_action));
4223 				actions_rx++;
4224 			}
4225 			break;
4226 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4227 			raw_decap = actions->conf;
4228 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4229 				memcpy(actions_tx, actions,
4230 				       sizeof(struct rte_flow_action));
4231 				actions_tx++;
4232 			} else {
4233 				rte_memcpy(actions_rx, actions,
4234 					   sizeof(struct rte_flow_action));
4235 				actions_rx++;
4236 			}
4237 			break;
4238 		default:
4239 			rte_memcpy(actions_rx, actions,
4240 				   sizeof(struct rte_flow_action));
4241 			actions_rx++;
4242 			break;
4243 		}
4244 	}
4245 	/* Add set meta action and end action for the Rx flow. */
4246 	tag_action = actions_rx;
4247 	tag_action->type = (enum rte_flow_action_type)
4248 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4249 	actions_rx++;
4250 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4251 	actions_rx++;
4252 	set_tag = (void *)actions_rx;
4253 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4254 	MLX5_ASSERT(set_tag->id > REG_NON);
4255 	set_tag->data = flow_id;
4256 	tag_action->conf = set_tag;
4257 	/* Create Tx item list. */
4258 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4259 	addr = (void *)&pattern_tx[2];
4260 	item = pattern_tx;
4261 	item->type = (enum rte_flow_item_type)
4262 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4263 	tag_item = (void *)addr;
4264 	tag_item->data = flow_id;
4265 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4266 	MLX5_ASSERT(set_tag->id > REG_NON);
4267 	item->spec = tag_item;
4268 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4269 	tag_item = (void *)addr;
4270 	tag_item->data = UINT32_MAX;
4271 	tag_item->id = UINT16_MAX;
4272 	item->mask = tag_item;
4273 	item->last = NULL;
4274 	item++;
4275 	item->type = RTE_FLOW_ITEM_TYPE_END;
4276 	return 0;
4277 }
4278 
4279 /**
4280  * The last stage of splitting chain, just creates the subflow
4281  * without any modification.
4282  *
4283  * @param[in] dev
4284  *   Pointer to Ethernet device.
4285  * @param[in] flow
4286  *   Parent flow structure pointer.
4287  * @param[in, out] sub_flow
4288  *   Pointer to return the created subflow, may be NULL.
4289  * @param[in] attr
4290  *   Flow rule attributes.
4291  * @param[in] items
4292  *   Pattern specification (list terminated by the END pattern item).
4293  * @param[in] actions
4294  *   Associated actions (list terminated by the END action).
4295  * @param[in] flow_split_info
4296  *   Pointer to flow split info structure.
4297  * @param[out] error
4298  *   Perform verbose error reporting if not NULL.
4299  * @return
4300  *   0 on success, negative value otherwise
4301  */
4302 static int
4303 flow_create_split_inner(struct rte_eth_dev *dev,
4304 			struct rte_flow *flow,
4305 			struct mlx5_flow **sub_flow,
4306 			const struct rte_flow_attr *attr,
4307 			const struct rte_flow_item items[],
4308 			const struct rte_flow_action actions[],
4309 			struct mlx5_flow_split_info *flow_split_info,
4310 			struct rte_flow_error *error)
4311 {
4312 	struct mlx5_flow *dev_flow;
4313 
4314 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4315 				    flow_split_info->flow_idx, error);
4316 	if (!dev_flow)
4317 		return -rte_errno;
4318 	dev_flow->flow = flow;
4319 	dev_flow->external = flow_split_info->external;
4320 	dev_flow->skip_scale = flow_split_info->skip_scale;
4321 	/* Subflow object was created, we must include one in the list. */
4322 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4323 		      dev_flow->handle, next);
4324 	/*
4325 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4326 	 * flow may need some user defined item layer flags, and pass the
4327 	 * Metadate rxq mark flag to suffix flow as well.
4328 	 */
4329 	if (flow_split_info->prefix_layers)
4330 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4331 	if (flow_split_info->prefix_mark)
4332 		dev_flow->handle->mark = 1;
4333 	if (sub_flow)
4334 		*sub_flow = dev_flow;
4335 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4336 }
4337 
4338 /**
4339  * Split the meter flow.
4340  *
4341  * As meter flow will split to three sub flow, other than meter
4342  * action, the other actions make sense to only meter accepts
4343  * the packet. If it need to be dropped, no other additional
4344  * actions should be take.
4345  *
4346  * One kind of special action which decapsulates the L3 tunnel
4347  * header will be in the prefix sub flow, as not to take the
4348  * L3 tunnel header into account.
4349  *
4350  * @param dev
4351  *   Pointer to Ethernet device.
4352  * @param[in] items
4353  *   Pattern specification (list terminated by the END pattern item).
4354  * @param[out] sfx_items
4355  *   Suffix flow match items (list terminated by the END pattern item).
4356  * @param[in] actions
4357  *   Associated actions (list terminated by the END action).
4358  * @param[out] actions_sfx
4359  *   Suffix flow actions.
4360  * @param[out] actions_pre
4361  *   Prefix flow actions.
4362  * @param[out] pattern_sfx
4363  *   The pattern items for the suffix flow.
4364  * @param[out] tag_sfx
4365  *   Pointer to suffix flow tag.
4366  *
4367  * @return
4368  *   0 on success.
4369  */
4370 static int
4371 flow_meter_split_prep(struct rte_eth_dev *dev,
4372 		 const struct rte_flow_item items[],
4373 		 struct rte_flow_item sfx_items[],
4374 		 const struct rte_flow_action actions[],
4375 		 struct rte_flow_action actions_sfx[],
4376 		 struct rte_flow_action actions_pre[])
4377 {
4378 	struct mlx5_priv *priv = dev->data->dev_private;
4379 	struct rte_flow_action *tag_action = NULL;
4380 	struct rte_flow_item *tag_item;
4381 	struct mlx5_rte_flow_action_set_tag *set_tag;
4382 	struct rte_flow_error error;
4383 	const struct rte_flow_action_raw_encap *raw_encap;
4384 	const struct rte_flow_action_raw_decap *raw_decap;
4385 	struct mlx5_rte_flow_item_tag *tag_spec;
4386 	struct mlx5_rte_flow_item_tag *tag_mask;
4387 	uint32_t tag_id = 0;
4388 	bool copy_vlan = false;
4389 
4390 	/* Prepare the actions for prefix and suffix flow. */
4391 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4392 		struct rte_flow_action **action_cur = NULL;
4393 
4394 		switch (actions->type) {
4395 		case RTE_FLOW_ACTION_TYPE_METER:
4396 			/* Add the extra tag action first. */
4397 			tag_action = actions_pre;
4398 			tag_action->type = (enum rte_flow_action_type)
4399 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4400 			actions_pre++;
4401 			action_cur = &actions_pre;
4402 			break;
4403 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4404 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4405 			action_cur = &actions_pre;
4406 			break;
4407 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4408 			raw_encap = actions->conf;
4409 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4410 				action_cur = &actions_pre;
4411 			break;
4412 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4413 			raw_decap = actions->conf;
4414 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4415 				action_cur = &actions_pre;
4416 			break;
4417 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4418 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4419 			copy_vlan = true;
4420 			break;
4421 		default:
4422 			break;
4423 		}
4424 		if (!action_cur)
4425 			action_cur = &actions_sfx;
4426 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4427 		(*action_cur)++;
4428 	}
4429 	/* Add end action to the actions. */
4430 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4431 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4432 	actions_pre++;
4433 	/* Set the tag. */
4434 	set_tag = (void *)actions_pre;
4435 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4436 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4437 			  &tag_id);
4438 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4439 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4440 			dev->data->port_id);
4441 		mlx5_ipool_free(priv->sh->ipool
4442 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4443 		return 0;
4444 	} else if (!tag_id) {
4445 		return 0;
4446 	}
4447 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4448 	assert(tag_action);
4449 	tag_action->conf = set_tag;
4450 	/* Prepare the suffix subflow items. */
4451 	tag_item = sfx_items++;
4452 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4453 		int item_type = items->type;
4454 
4455 		switch (item_type) {
4456 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4457 			memcpy(sfx_items, items, sizeof(*sfx_items));
4458 			sfx_items++;
4459 			break;
4460 		case RTE_FLOW_ITEM_TYPE_VLAN:
4461 			if (copy_vlan) {
4462 				memcpy(sfx_items, items, sizeof(*sfx_items));
4463 				/*
4464 				 * Convert to internal match item, it is used
4465 				 * for vlan push and set vid.
4466 				 */
4467 				sfx_items->type = (enum rte_flow_item_type)
4468 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4469 				sfx_items++;
4470 			}
4471 			break;
4472 		default:
4473 			break;
4474 		}
4475 	}
4476 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4477 	sfx_items++;
4478 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4479 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4480 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4481 	tag_mask = tag_spec + 1;
4482 	tag_mask->data = 0xffffff00;
4483 	tag_item->type = (enum rte_flow_item_type)
4484 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4485 	tag_item->spec = tag_spec;
4486 	tag_item->last = NULL;
4487 	tag_item->mask = tag_mask;
4488 	return tag_id;
4489 }
4490 
4491 /**
4492  * Split action list having QUEUE/RSS for metadata register copy.
4493  *
4494  * Once Q/RSS action is detected in user's action list, the flow action
4495  * should be split in order to copy metadata registers, which will happen in
4496  * RX_CP_TBL like,
4497  *   - CQE->flow_tag := reg_c[1] (MARK)
4498  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4499  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4500  * This is because the last action of each flow must be a terminal action
4501  * (QUEUE, RSS or DROP).
4502  *
4503  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4504  * stored and kept in the mlx5_flow structure per each sub_flow.
4505  *
4506  * The Q/RSS action is replaced with,
4507  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4508  * And the following JUMP action is added at the end,
4509  *   - JUMP, to RX_CP_TBL.
4510  *
4511  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4512  * flow_create_split_metadata() routine. The flow will look like,
4513  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4514  *
4515  * @param dev
4516  *   Pointer to Ethernet device.
4517  * @param[out] split_actions
4518  *   Pointer to store split actions to jump to CP_TBL.
4519  * @param[in] actions
4520  *   Pointer to the list of original flow actions.
4521  * @param[in] qrss
4522  *   Pointer to the Q/RSS action.
4523  * @param[in] actions_n
4524  *   Number of original actions.
4525  * @param[out] error
4526  *   Perform verbose error reporting if not NULL.
4527  *
4528  * @return
4529  *   non-zero unique flow_id on success, otherwise 0 and
4530  *   error/rte_error are set.
4531  */
4532 static uint32_t
4533 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4534 			  struct rte_flow_action *split_actions,
4535 			  const struct rte_flow_action *actions,
4536 			  const struct rte_flow_action *qrss,
4537 			  int actions_n, struct rte_flow_error *error)
4538 {
4539 	struct mlx5_priv *priv = dev->data->dev_private;
4540 	struct mlx5_rte_flow_action_set_tag *set_tag;
4541 	struct rte_flow_action_jump *jump;
4542 	const int qrss_idx = qrss - actions;
4543 	uint32_t flow_id = 0;
4544 	int ret = 0;
4545 
4546 	/*
4547 	 * Given actions will be split
4548 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4549 	 * - Add jump to mreg CP_TBL.
4550 	 * As a result, there will be one more action.
4551 	 */
4552 	++actions_n;
4553 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4554 	set_tag = (void *)(split_actions + actions_n);
4555 	/*
4556 	 * If tag action is not set to void(it means we are not the meter
4557 	 * suffix flow), add the tag action. Since meter suffix flow already
4558 	 * has the tag added.
4559 	 */
4560 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4561 		/*
4562 		 * Allocate the new subflow ID. This one is unique within
4563 		 * device and not shared with representors. Otherwise,
4564 		 * we would have to resolve multi-thread access synch
4565 		 * issue. Each flow on the shared device is appended
4566 		 * with source vport identifier, so the resulting
4567 		 * flows will be unique in the shared (by master and
4568 		 * representors) domain even if they have coinciding
4569 		 * IDs.
4570 		 */
4571 		mlx5_ipool_malloc(priv->sh->ipool
4572 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4573 		if (!flow_id)
4574 			return rte_flow_error_set(error, ENOMEM,
4575 						  RTE_FLOW_ERROR_TYPE_ACTION,
4576 						  NULL, "can't allocate id "
4577 						  "for split Q/RSS subflow");
4578 		/* Internal SET_TAG action to set flow ID. */
4579 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4580 			.data = flow_id,
4581 		};
4582 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4583 		if (ret < 0)
4584 			return ret;
4585 		set_tag->id = ret;
4586 		/* Construct new actions array. */
4587 		/* Replace QUEUE/RSS action. */
4588 		split_actions[qrss_idx] = (struct rte_flow_action){
4589 			.type = (enum rte_flow_action_type)
4590 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4591 			.conf = set_tag,
4592 		};
4593 	}
4594 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4595 	jump = (void *)(set_tag + 1);
4596 	*jump = (struct rte_flow_action_jump){
4597 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4598 	};
4599 	split_actions[actions_n - 2] = (struct rte_flow_action){
4600 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4601 		.conf = jump,
4602 	};
4603 	split_actions[actions_n - 1] = (struct rte_flow_action){
4604 		.type = RTE_FLOW_ACTION_TYPE_END,
4605 	};
4606 	return flow_id;
4607 }
4608 
4609 /**
4610  * Extend the given action list for Tx metadata copy.
4611  *
4612  * Copy the given action list to the ext_actions and add flow metadata register
4613  * copy action in order to copy reg_a set by WQE to reg_c[0].
4614  *
4615  * @param[out] ext_actions
4616  *   Pointer to the extended action list.
4617  * @param[in] actions
4618  *   Pointer to the list of actions.
4619  * @param[in] actions_n
4620  *   Number of actions in the list.
4621  * @param[out] error
4622  *   Perform verbose error reporting if not NULL.
4623  * @param[in] encap_idx
4624  *   The encap action inndex.
4625  *
4626  * @return
4627  *   0 on success, negative value otherwise
4628  */
4629 static int
4630 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4631 		       struct rte_flow_action *ext_actions,
4632 		       const struct rte_flow_action *actions,
4633 		       int actions_n, struct rte_flow_error *error,
4634 		       int encap_idx)
4635 {
4636 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4637 		(struct mlx5_flow_action_copy_mreg *)
4638 			(ext_actions + actions_n + 1);
4639 	int ret;
4640 
4641 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4642 	if (ret < 0)
4643 		return ret;
4644 	cp_mreg->dst = ret;
4645 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4646 	if (ret < 0)
4647 		return ret;
4648 	cp_mreg->src = ret;
4649 	if (encap_idx != 0)
4650 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4651 	if (encap_idx == actions_n - 1) {
4652 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4653 			.type = (enum rte_flow_action_type)
4654 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4655 			.conf = cp_mreg,
4656 		};
4657 		ext_actions[actions_n] = (struct rte_flow_action){
4658 			.type = RTE_FLOW_ACTION_TYPE_END,
4659 		};
4660 	} else {
4661 		ext_actions[encap_idx] = (struct rte_flow_action){
4662 			.type = (enum rte_flow_action_type)
4663 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4664 			.conf = cp_mreg,
4665 		};
4666 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4667 				sizeof(*ext_actions) * (actions_n - encap_idx));
4668 	}
4669 	return 0;
4670 }
4671 
4672 /**
4673  * Check the match action from the action list.
4674  *
4675  * @param[in] actions
4676  *   Pointer to the list of actions.
4677  * @param[in] attr
4678  *   Flow rule attributes.
4679  * @param[in] action
4680  *   The action to be check if exist.
4681  * @param[out] match_action_pos
4682  *   Pointer to the position of the matched action if exists, otherwise is -1.
4683  * @param[out] qrss_action_pos
4684  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4685  * @param[out] modify_after_mirror
4686  *   Pointer to the flag of modify action after FDB mirroring.
4687  *
4688  * @return
4689  *   > 0 the total number of actions.
4690  *   0 if not found match action in action list.
4691  */
4692 static int
4693 flow_check_match_action(const struct rte_flow_action actions[],
4694 			const struct rte_flow_attr *attr,
4695 			enum rte_flow_action_type action,
4696 			int *match_action_pos, int *qrss_action_pos,
4697 			int *modify_after_mirror)
4698 {
4699 	const struct rte_flow_action_sample *sample;
4700 	int actions_n = 0;
4701 	uint32_t ratio = 0;
4702 	int sub_type = 0;
4703 	int flag = 0;
4704 	int fdb_mirror = 0;
4705 
4706 	*match_action_pos = -1;
4707 	*qrss_action_pos = -1;
4708 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4709 		if (actions->type == action) {
4710 			flag = 1;
4711 			*match_action_pos = actions_n;
4712 		}
4713 		switch (actions->type) {
4714 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4715 		case RTE_FLOW_ACTION_TYPE_RSS:
4716 			*qrss_action_pos = actions_n;
4717 			break;
4718 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
4719 			sample = actions->conf;
4720 			ratio = sample->ratio;
4721 			sub_type = ((const struct rte_flow_action *)
4722 					(sample->actions))->type;
4723 			if (ratio == 1 && attr->transfer)
4724 				fdb_mirror = 1;
4725 			break;
4726 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4727 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4728 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4729 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4730 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4731 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4732 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4733 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4734 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4735 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
4736 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4737 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4738 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4739 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4740 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4741 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4742 		case RTE_FLOW_ACTION_TYPE_FLAG:
4743 		case RTE_FLOW_ACTION_TYPE_MARK:
4744 		case RTE_FLOW_ACTION_TYPE_SET_META:
4745 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
4746 			if (fdb_mirror)
4747 				*modify_after_mirror = 1;
4748 			break;
4749 		default:
4750 			break;
4751 		}
4752 		actions_n++;
4753 	}
4754 	if (flag && fdb_mirror && !*modify_after_mirror) {
4755 		/* FDB mirroring uses the destination array to implement
4756 		 * instead of FLOW_SAMPLER object.
4757 		 */
4758 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4759 			flag = 0;
4760 	}
4761 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4762 	return flag ? actions_n + 1 : 0;
4763 }
4764 
4765 #define SAMPLE_SUFFIX_ITEM 2
4766 
4767 /**
4768  * Split the sample flow.
4769  *
4770  * As sample flow will split to two sub flow, sample flow with
4771  * sample action, the other actions will move to new suffix flow.
4772  *
4773  * Also add unique tag id with tag action in the sample flow,
4774  * the same tag id will be as match in the suffix flow.
4775  *
4776  * @param dev
4777  *   Pointer to Ethernet device.
4778  * @param[in] add_tag
4779  *   Add extra tag action flag.
4780  * @param[out] sfx_items
4781  *   Suffix flow match items (list terminated by the END pattern item).
4782  * @param[in] actions
4783  *   Associated actions (list terminated by the END action).
4784  * @param[out] actions_sfx
4785  *   Suffix flow actions.
4786  * @param[out] actions_pre
4787  *   Prefix flow actions.
4788  * @param[in] actions_n
4789  *  The total number of actions.
4790  * @param[in] sample_action_pos
4791  *   The sample action position.
4792  * @param[in] qrss_action_pos
4793  *   The Queue/RSS action position.
4794  * @param[in] jump_table
4795  *   Add extra jump action flag.
4796  * @param[out] error
4797  *   Perform verbose error reporting if not NULL.
4798  *
4799  * @return
4800  *   0 on success, or unique flow_id, a negative errno value
4801  *   otherwise and rte_errno is set.
4802  */
4803 static int
4804 flow_sample_split_prep(struct rte_eth_dev *dev,
4805 		       int add_tag,
4806 		       struct rte_flow_item sfx_items[],
4807 		       const struct rte_flow_action actions[],
4808 		       struct rte_flow_action actions_sfx[],
4809 		       struct rte_flow_action actions_pre[],
4810 		       int actions_n,
4811 		       int sample_action_pos,
4812 		       int qrss_action_pos,
4813 		       int jump_table,
4814 		       struct rte_flow_error *error)
4815 {
4816 	struct mlx5_priv *priv = dev->data->dev_private;
4817 	struct mlx5_rte_flow_action_set_tag *set_tag;
4818 	struct mlx5_rte_flow_item_tag *tag_spec;
4819 	struct mlx5_rte_flow_item_tag *tag_mask;
4820 	struct rte_flow_action_jump *jump_action;
4821 	uint32_t tag_id = 0;
4822 	int index;
4823 	int append_index = 0;
4824 	int ret;
4825 
4826 	if (sample_action_pos < 0)
4827 		return rte_flow_error_set(error, EINVAL,
4828 					  RTE_FLOW_ERROR_TYPE_ACTION,
4829 					  NULL, "invalid position of sample "
4830 					  "action in list");
4831 	/* Prepare the actions for prefix and suffix flow. */
4832 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4833 		index = qrss_action_pos;
4834 		/* Put the preceding the Queue/RSS action into prefix flow. */
4835 		if (index != 0)
4836 			memcpy(actions_pre, actions,
4837 			       sizeof(struct rte_flow_action) * index);
4838 		/* Put others preceding the sample action into prefix flow. */
4839 		if (sample_action_pos > index + 1)
4840 			memcpy(actions_pre + index, actions + index + 1,
4841 			       sizeof(struct rte_flow_action) *
4842 			       (sample_action_pos - index - 1));
4843 		index = sample_action_pos - 1;
4844 		/* Put Queue/RSS action into Suffix flow. */
4845 		memcpy(actions_sfx, actions + qrss_action_pos,
4846 		       sizeof(struct rte_flow_action));
4847 		actions_sfx++;
4848 	} else {
4849 		index = sample_action_pos;
4850 		if (index != 0)
4851 			memcpy(actions_pre, actions,
4852 			       sizeof(struct rte_flow_action) * index);
4853 	}
4854 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4855 	 * For CX6DX and above, metadata registers Cx preserve their value,
4856 	 * add an extra tag action for NIC-RX and E-Switch Domain.
4857 	 */
4858 	if (add_tag) {
4859 		/* Prepare the prefix tag action. */
4860 		append_index++;
4861 		set_tag = (void *)(actions_pre + actions_n + append_index);
4862 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4863 		if (ret < 0)
4864 			return ret;
4865 		set_tag->id = ret;
4866 		mlx5_ipool_malloc(priv->sh->ipool
4867 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4868 		set_tag->data = tag_id;
4869 		/* Prepare the suffix subflow items. */
4870 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4871 		tag_spec->data = tag_id;
4872 		tag_spec->id = set_tag->id;
4873 		tag_mask = tag_spec + 1;
4874 		tag_mask->data = UINT32_MAX;
4875 		sfx_items[0] = (struct rte_flow_item){
4876 			.type = (enum rte_flow_item_type)
4877 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4878 			.spec = tag_spec,
4879 			.last = NULL,
4880 			.mask = tag_mask,
4881 		};
4882 		sfx_items[1] = (struct rte_flow_item){
4883 			.type = (enum rte_flow_item_type)
4884 				RTE_FLOW_ITEM_TYPE_END,
4885 		};
4886 		/* Prepare the tag action in prefix subflow. */
4887 		actions_pre[index++] =
4888 			(struct rte_flow_action){
4889 			.type = (enum rte_flow_action_type)
4890 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4891 			.conf = set_tag,
4892 		};
4893 	}
4894 	memcpy(actions_pre + index, actions + sample_action_pos,
4895 	       sizeof(struct rte_flow_action));
4896 	index += 1;
4897 	/* For the modify action after the sample action in E-Switch mirroring,
4898 	 * Add the extra jump action in prefix subflow and jump into the next
4899 	 * table, then do the modify action in the new table.
4900 	 */
4901 	if (jump_table) {
4902 		/* Prepare the prefix jump action. */
4903 		append_index++;
4904 		jump_action = (void *)(actions_pre + actions_n + append_index);
4905 		jump_action->group = jump_table;
4906 		actions_pre[index++] =
4907 			(struct rte_flow_action){
4908 			.type = (enum rte_flow_action_type)
4909 				RTE_FLOW_ACTION_TYPE_JUMP,
4910 			.conf = jump_action,
4911 		};
4912 	}
4913 	actions_pre[index] = (struct rte_flow_action){
4914 		.type = (enum rte_flow_action_type)
4915 			RTE_FLOW_ACTION_TYPE_END,
4916 	};
4917 	/* Put the actions after sample into Suffix flow. */
4918 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4919 	       sizeof(struct rte_flow_action) *
4920 	       (actions_n - sample_action_pos - 1));
4921 	return tag_id;
4922 }
4923 
4924 /**
4925  * The splitting for metadata feature.
4926  *
4927  * - Q/RSS action on NIC Rx should be split in order to pass by
4928  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4929  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4930  *
4931  * - All the actions on NIC Tx should have a mreg copy action to
4932  *   copy reg_a from WQE to reg_c[0].
4933  *
4934  * @param dev
4935  *   Pointer to Ethernet device.
4936  * @param[in] flow
4937  *   Parent flow structure pointer.
4938  * @param[in] attr
4939  *   Flow rule attributes.
4940  * @param[in] items
4941  *   Pattern specification (list terminated by the END pattern item).
4942  * @param[in] actions
4943  *   Associated actions (list terminated by the END action).
4944  * @param[in] flow_split_info
4945  *   Pointer to flow split info structure.
4946  * @param[out] error
4947  *   Perform verbose error reporting if not NULL.
4948  * @return
4949  *   0 on success, negative value otherwise
4950  */
4951 static int
4952 flow_create_split_metadata(struct rte_eth_dev *dev,
4953 			   struct rte_flow *flow,
4954 			   const struct rte_flow_attr *attr,
4955 			   const struct rte_flow_item items[],
4956 			   const struct rte_flow_action actions[],
4957 			   struct mlx5_flow_split_info *flow_split_info,
4958 			   struct rte_flow_error *error)
4959 {
4960 	struct mlx5_priv *priv = dev->data->dev_private;
4961 	struct mlx5_dev_config *config = &priv->config;
4962 	const struct rte_flow_action *qrss = NULL;
4963 	struct rte_flow_action *ext_actions = NULL;
4964 	struct mlx5_flow *dev_flow = NULL;
4965 	uint32_t qrss_id = 0;
4966 	int mtr_sfx = 0;
4967 	size_t act_size;
4968 	int actions_n;
4969 	int encap_idx;
4970 	int ret;
4971 
4972 	/* Check whether extensive metadata feature is engaged. */
4973 	if (!config->dv_flow_en ||
4974 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4975 	    !mlx5_flow_ext_mreg_supported(dev))
4976 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4977 					       actions, flow_split_info, error);
4978 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4979 							   &encap_idx);
4980 	if (qrss) {
4981 		/* Exclude hairpin flows from splitting. */
4982 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4983 			const struct rte_flow_action_queue *queue;
4984 
4985 			queue = qrss->conf;
4986 			if (mlx5_rxq_get_type(dev, queue->index) ==
4987 			    MLX5_RXQ_TYPE_HAIRPIN)
4988 				qrss = NULL;
4989 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4990 			const struct rte_flow_action_rss *rss;
4991 
4992 			rss = qrss->conf;
4993 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4994 			    MLX5_RXQ_TYPE_HAIRPIN)
4995 				qrss = NULL;
4996 		}
4997 	}
4998 	if (qrss) {
4999 		/* Check if it is in meter suffix table. */
5000 		mtr_sfx = attr->group == (attr->transfer ?
5001 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5002 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5003 		/*
5004 		 * Q/RSS action on NIC Rx should be split in order to pass by
5005 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5006 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5007 		 */
5008 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5009 			   sizeof(struct rte_flow_action_set_tag) +
5010 			   sizeof(struct rte_flow_action_jump);
5011 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5012 					  SOCKET_ID_ANY);
5013 		if (!ext_actions)
5014 			return rte_flow_error_set(error, ENOMEM,
5015 						  RTE_FLOW_ERROR_TYPE_ACTION,
5016 						  NULL, "no memory to split "
5017 						  "metadata flow");
5018 		/*
5019 		 * If we are the suffix flow of meter, tag already exist.
5020 		 * Set the tag action to void.
5021 		 */
5022 		if (mtr_sfx)
5023 			ext_actions[qrss - actions].type =
5024 						RTE_FLOW_ACTION_TYPE_VOID;
5025 		else
5026 			ext_actions[qrss - actions].type =
5027 						(enum rte_flow_action_type)
5028 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5029 		/*
5030 		 * Create the new actions list with removed Q/RSS action
5031 		 * and appended set tag and jump to register copy table
5032 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5033 		 * in advance, because it is needed for set tag action.
5034 		 */
5035 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5036 						    qrss, actions_n, error);
5037 		if (!mtr_sfx && !qrss_id) {
5038 			ret = -rte_errno;
5039 			goto exit;
5040 		}
5041 	} else if (attr->egress && !attr->transfer) {
5042 		/*
5043 		 * All the actions on NIC Tx should have a metadata register
5044 		 * copy action to copy reg_a from WQE to reg_c[meta]
5045 		 */
5046 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5047 			   sizeof(struct mlx5_flow_action_copy_mreg);
5048 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5049 					  SOCKET_ID_ANY);
5050 		if (!ext_actions)
5051 			return rte_flow_error_set(error, ENOMEM,
5052 						  RTE_FLOW_ERROR_TYPE_ACTION,
5053 						  NULL, "no memory to split "
5054 						  "metadata flow");
5055 		/* Create the action list appended with copy register. */
5056 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5057 					     actions_n, error, encap_idx);
5058 		if (ret < 0)
5059 			goto exit;
5060 	}
5061 	/* Add the unmodified original or prefix subflow. */
5062 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5063 				      items, ext_actions ? ext_actions :
5064 				      actions, flow_split_info, error);
5065 	if (ret < 0)
5066 		goto exit;
5067 	MLX5_ASSERT(dev_flow);
5068 	if (qrss) {
5069 		const struct rte_flow_attr q_attr = {
5070 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5071 			.ingress = 1,
5072 		};
5073 		/* Internal PMD action to set register. */
5074 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5075 			.data = qrss_id,
5076 			.id = REG_NON,
5077 		};
5078 		struct rte_flow_item q_items[] = {
5079 			{
5080 				.type = (enum rte_flow_item_type)
5081 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5082 				.spec = &q_tag_spec,
5083 				.last = NULL,
5084 				.mask = NULL,
5085 			},
5086 			{
5087 				.type = RTE_FLOW_ITEM_TYPE_END,
5088 			},
5089 		};
5090 		struct rte_flow_action q_actions[] = {
5091 			{
5092 				.type = qrss->type,
5093 				.conf = qrss->conf,
5094 			},
5095 			{
5096 				.type = RTE_FLOW_ACTION_TYPE_END,
5097 			},
5098 		};
5099 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5100 
5101 		/*
5102 		 * Configure the tag item only if there is no meter subflow.
5103 		 * Since tag is already marked in the meter suffix subflow
5104 		 * we can just use the meter suffix items as is.
5105 		 */
5106 		if (qrss_id) {
5107 			/* Not meter subflow. */
5108 			MLX5_ASSERT(!mtr_sfx);
5109 			/*
5110 			 * Put unique id in prefix flow due to it is destroyed
5111 			 * after suffix flow and id will be freed after there
5112 			 * is no actual flows with this id and identifier
5113 			 * reallocation becomes possible (for example, for
5114 			 * other flows in other threads).
5115 			 */
5116 			dev_flow->handle->split_flow_id = qrss_id;
5117 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5118 						   error);
5119 			if (ret < 0)
5120 				goto exit;
5121 			q_tag_spec.id = ret;
5122 		}
5123 		dev_flow = NULL;
5124 		/* Add suffix subflow to execute Q/RSS. */
5125 		flow_split_info->prefix_layers = layers;
5126 		flow_split_info->prefix_mark = 0;
5127 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5128 					      &q_attr, mtr_sfx ? items :
5129 					      q_items, q_actions,
5130 					      flow_split_info, error);
5131 		if (ret < 0)
5132 			goto exit;
5133 		/* qrss ID should be freed if failed. */
5134 		qrss_id = 0;
5135 		MLX5_ASSERT(dev_flow);
5136 	}
5137 
5138 exit:
5139 	/*
5140 	 * We do not destroy the partially created sub_flows in case of error.
5141 	 * These ones are included into parent flow list and will be destroyed
5142 	 * by flow_drv_destroy.
5143 	 */
5144 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5145 			qrss_id);
5146 	mlx5_free(ext_actions);
5147 	return ret;
5148 }
5149 
5150 /**
5151  * The splitting for meter feature.
5152  *
5153  * - The meter flow will be split to two flows as prefix and
5154  *   suffix flow. The packets make sense only it pass the prefix
5155  *   meter action.
5156  *
5157  * - Reg_C_5 is used for the packet to match betweend prefix and
5158  *   suffix flow.
5159  *
5160  * @param dev
5161  *   Pointer to Ethernet device.
5162  * @param[in] flow
5163  *   Parent flow structure pointer.
5164  * @param[in] attr
5165  *   Flow rule attributes.
5166  * @param[in] items
5167  *   Pattern specification (list terminated by the END pattern item).
5168  * @param[in] actions
5169  *   Associated actions (list terminated by the END action).
5170  * @param[in] flow_split_info
5171  *   Pointer to flow split info structure.
5172  * @param[out] error
5173  *   Perform verbose error reporting if not NULL.
5174  * @return
5175  *   0 on success, negative value otherwise
5176  */
5177 static int
5178 flow_create_split_meter(struct rte_eth_dev *dev,
5179 			struct rte_flow *flow,
5180 			const struct rte_flow_attr *attr,
5181 			const struct rte_flow_item items[],
5182 			const struct rte_flow_action actions[],
5183 			struct mlx5_flow_split_info *flow_split_info,
5184 			struct rte_flow_error *error)
5185 {
5186 	struct mlx5_priv *priv = dev->data->dev_private;
5187 	struct rte_flow_action *sfx_actions = NULL;
5188 	struct rte_flow_action *pre_actions = NULL;
5189 	struct rte_flow_item *sfx_items = NULL;
5190 	struct mlx5_flow *dev_flow = NULL;
5191 	struct rte_flow_attr sfx_attr = *attr;
5192 	uint32_t mtr = 0;
5193 	uint32_t mtr_tag_id = 0;
5194 	size_t act_size;
5195 	size_t item_size;
5196 	int actions_n = 0;
5197 	int ret;
5198 
5199 	if (priv->mtr_en)
5200 		actions_n = flow_check_meter_action(actions, &mtr);
5201 	if (mtr) {
5202 		/* The five prefix actions: meter, decap, encap, tag, end. */
5203 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5204 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5205 		/* tag, vlan, port id, end. */
5206 #define METER_SUFFIX_ITEM 4
5207 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5208 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5209 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5210 					  0, SOCKET_ID_ANY);
5211 		if (!sfx_actions)
5212 			return rte_flow_error_set(error, ENOMEM,
5213 						  RTE_FLOW_ERROR_TYPE_ACTION,
5214 						  NULL, "no memory to split "
5215 						  "meter flow");
5216 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5217 			     act_size);
5218 		pre_actions = sfx_actions + actions_n;
5219 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5220 						   actions, sfx_actions,
5221 						   pre_actions);
5222 		if (!mtr_tag_id) {
5223 			ret = -rte_errno;
5224 			goto exit;
5225 		}
5226 		/* Add the prefix subflow. */
5227 		flow_split_info->prefix_mark = 0;
5228 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5229 					      attr, items, pre_actions,
5230 					      flow_split_info, error);
5231 		if (ret) {
5232 			ret = -rte_errno;
5233 			goto exit;
5234 		}
5235 		dev_flow->handle->split_flow_id = mtr_tag_id;
5236 		/* Setting the sfx group atrr. */
5237 		sfx_attr.group = sfx_attr.transfer ?
5238 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5239 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5240 		flow_split_info->prefix_layers =
5241 				flow_get_prefix_layer_flags(dev_flow);
5242 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5243 	}
5244 	/* Add the prefix subflow. */
5245 	ret = flow_create_split_metadata(dev, flow,
5246 					 &sfx_attr, sfx_items ?
5247 					 sfx_items : items,
5248 					 sfx_actions ? sfx_actions : actions,
5249 					 flow_split_info, error);
5250 exit:
5251 	if (sfx_actions)
5252 		mlx5_free(sfx_actions);
5253 	return ret;
5254 }
5255 
5256 /**
5257  * The splitting for sample feature.
5258  *
5259  * Once Sample action is detected in the action list, the flow actions should
5260  * be split into prefix sub flow and suffix sub flow.
5261  *
5262  * The original items remain in the prefix sub flow, all actions preceding the
5263  * sample action and the sample action itself will be copied to the prefix
5264  * sub flow, the actions following the sample action will be copied to the
5265  * suffix sub flow, Queue action always be located in the suffix sub flow.
5266  *
5267  * In order to make the packet from prefix sub flow matches with suffix sub
5268  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5269  * flow uses tag item with the unique flow id.
5270  *
5271  * @param dev
5272  *   Pointer to Ethernet device.
5273  * @param[in] flow
5274  *   Parent flow structure pointer.
5275  * @param[in] attr
5276  *   Flow rule attributes.
5277  * @param[in] items
5278  *   Pattern specification (list terminated by the END pattern item).
5279  * @param[in] actions
5280  *   Associated actions (list terminated by the END action).
5281  * @param[in] flow_split_info
5282  *   Pointer to flow split info structure.
5283  * @param[out] error
5284  *   Perform verbose error reporting if not NULL.
5285  * @return
5286  *   0 on success, negative value otherwise
5287  */
5288 static int
5289 flow_create_split_sample(struct rte_eth_dev *dev,
5290 			 struct rte_flow *flow,
5291 			 const struct rte_flow_attr *attr,
5292 			 const struct rte_flow_item items[],
5293 			 const struct rte_flow_action actions[],
5294 			 struct mlx5_flow_split_info *flow_split_info,
5295 			 struct rte_flow_error *error)
5296 {
5297 	struct mlx5_priv *priv = dev->data->dev_private;
5298 	struct rte_flow_action *sfx_actions = NULL;
5299 	struct rte_flow_action *pre_actions = NULL;
5300 	struct rte_flow_item *sfx_items = NULL;
5301 	struct mlx5_flow *dev_flow = NULL;
5302 	struct rte_flow_attr sfx_attr = *attr;
5303 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5304 	struct mlx5_flow_dv_sample_resource *sample_res;
5305 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5306 	struct mlx5_flow_tbl_resource *sfx_tbl;
5307 #endif
5308 	size_t act_size;
5309 	size_t item_size;
5310 	uint32_t fdb_tx = 0;
5311 	int32_t tag_id = 0;
5312 	int actions_n = 0;
5313 	int sample_action_pos;
5314 	int qrss_action_pos;
5315 	int add_tag = 0;
5316 	int modify_after_mirror = 0;
5317 	uint16_t jump_table = 0;
5318 	const uint32_t next_ft_step = 1;
5319 	int ret = 0;
5320 
5321 	if (priv->sampler_en)
5322 		actions_n = flow_check_match_action(actions, attr,
5323 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5324 					&sample_action_pos, &qrss_action_pos,
5325 					&modify_after_mirror);
5326 	if (actions_n) {
5327 		/* The prefix actions must includes sample, tag, end. */
5328 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5329 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5330 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5331 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5332 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5333 					  item_size), 0, SOCKET_ID_ANY);
5334 		if (!sfx_actions)
5335 			return rte_flow_error_set(error, ENOMEM,
5336 						  RTE_FLOW_ERROR_TYPE_ACTION,
5337 						  NULL, "no memory to split "
5338 						  "sample flow");
5339 		/* The representor_id is -1 for uplink. */
5340 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5341 		/*
5342 		 * When reg_c_preserve is set, metadata registers Cx preserve
5343 		 * their value even through packet duplication.
5344 		 */
5345 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5346 		if (add_tag)
5347 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5348 					+ act_size);
5349 		if (modify_after_mirror)
5350 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5351 				     next_ft_step;
5352 		pre_actions = sfx_actions + actions_n;
5353 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5354 						actions, sfx_actions,
5355 						pre_actions, actions_n,
5356 						sample_action_pos,
5357 						qrss_action_pos, jump_table,
5358 						error);
5359 		if (tag_id < 0 || (add_tag && !tag_id)) {
5360 			ret = -rte_errno;
5361 			goto exit;
5362 		}
5363 		if (modify_after_mirror)
5364 			flow_split_info->skip_scale =
5365 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5366 		/* Add the prefix subflow. */
5367 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5368 					      items, pre_actions,
5369 					      flow_split_info, error);
5370 		if (ret) {
5371 			ret = -rte_errno;
5372 			goto exit;
5373 		}
5374 		dev_flow->handle->split_flow_id = tag_id;
5375 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5376 		if (!modify_after_mirror) {
5377 			/* Set the sfx group attr. */
5378 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5379 						dev_flow->dv.sample_res;
5380 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5381 						sample_res->normal_path_tbl;
5382 			sfx_tbl_data = container_of(sfx_tbl,
5383 						struct mlx5_flow_tbl_data_entry,
5384 						tbl);
5385 			sfx_attr.group = sfx_attr.transfer ?
5386 						(sfx_tbl_data->table_id - 1) :
5387 						sfx_tbl_data->table_id;
5388 		} else {
5389 			MLX5_ASSERT(attr->transfer);
5390 			sfx_attr.group = jump_table;
5391 		}
5392 		flow_split_info->prefix_layers =
5393 				flow_get_prefix_layer_flags(dev_flow);
5394 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5395 		/* Suffix group level already be scaled with factor, set
5396 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5397 		 * again in translation.
5398 		 */
5399 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5400 #endif
5401 	}
5402 	/* Add the suffix subflow. */
5403 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5404 				      sfx_items ? sfx_items : items,
5405 				      sfx_actions ? sfx_actions : actions,
5406 				      flow_split_info, error);
5407 exit:
5408 	if (sfx_actions)
5409 		mlx5_free(sfx_actions);
5410 	return ret;
5411 }
5412 
5413 /**
5414  * Split the flow to subflow set. The splitters might be linked
5415  * in the chain, like this:
5416  * flow_create_split_outer() calls:
5417  *   flow_create_split_meter() calls:
5418  *     flow_create_split_metadata(meter_subflow_0) calls:
5419  *       flow_create_split_inner(metadata_subflow_0)
5420  *       flow_create_split_inner(metadata_subflow_1)
5421  *       flow_create_split_inner(metadata_subflow_2)
5422  *     flow_create_split_metadata(meter_subflow_1) calls:
5423  *       flow_create_split_inner(metadata_subflow_0)
5424  *       flow_create_split_inner(metadata_subflow_1)
5425  *       flow_create_split_inner(metadata_subflow_2)
5426  *
5427  * This provide flexible way to add new levels of flow splitting.
5428  * The all of successfully created subflows are included to the
5429  * parent flow dev_flow list.
5430  *
5431  * @param dev
5432  *   Pointer to Ethernet device.
5433  * @param[in] flow
5434  *   Parent flow structure pointer.
5435  * @param[in] attr
5436  *   Flow rule attributes.
5437  * @param[in] items
5438  *   Pattern specification (list terminated by the END pattern item).
5439  * @param[in] actions
5440  *   Associated actions (list terminated by the END action).
5441  * @param[in] flow_split_info
5442  *   Pointer to flow split info structure.
5443  * @param[out] error
5444  *   Perform verbose error reporting if not NULL.
5445  * @return
5446  *   0 on success, negative value otherwise
5447  */
5448 static int
5449 flow_create_split_outer(struct rte_eth_dev *dev,
5450 			struct rte_flow *flow,
5451 			const struct rte_flow_attr *attr,
5452 			const struct rte_flow_item items[],
5453 			const struct rte_flow_action actions[],
5454 			struct mlx5_flow_split_info *flow_split_info,
5455 			struct rte_flow_error *error)
5456 {
5457 	int ret;
5458 
5459 	ret = flow_create_split_sample(dev, flow, attr, items,
5460 				       actions, flow_split_info, error);
5461 	MLX5_ASSERT(ret <= 0);
5462 	return ret;
5463 }
5464 
5465 static struct mlx5_flow_tunnel *
5466 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5467 		      const struct rte_flow_attr *attr,
5468 		      const struct rte_flow_item items[],
5469 		      const struct rte_flow_action actions[])
5470 {
5471 	struct mlx5_flow_tunnel *tunnel;
5472 
5473 #pragma GCC diagnostic push
5474 #pragma GCC diagnostic ignored "-Wcast-qual"
5475 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5476 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5477 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5478 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5479 	else
5480 		tunnel = NULL;
5481 #pragma GCC diagnostic pop
5482 
5483 	return tunnel;
5484 }
5485 
5486 /**
5487  * Adjust flow RSS workspace if needed.
5488  *
5489  * @param wks
5490  *   Pointer to thread flow work space.
5491  * @param rss_desc
5492  *   Pointer to RSS descriptor.
5493  * @param[in] nrssq_num
5494  *   New RSS queue number.
5495  *
5496  * @return
5497  *   0 on success, -1 otherwise and rte_errno is set.
5498  */
5499 static int
5500 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5501 			  struct mlx5_flow_rss_desc *rss_desc,
5502 			  uint32_t nrssq_num)
5503 {
5504 	if (likely(nrssq_num <= wks->rssq_num))
5505 		return 0;
5506 	rss_desc->queue = realloc(rss_desc->queue,
5507 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5508 	if (!rss_desc->queue) {
5509 		rte_errno = ENOMEM;
5510 		return -1;
5511 	}
5512 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5513 	return 0;
5514 }
5515 
5516 /**
5517  * Create a flow and add it to @p list.
5518  *
5519  * @param dev
5520  *   Pointer to Ethernet device.
5521  * @param list
5522  *   Pointer to a TAILQ flow list. If this parameter NULL,
5523  *   no list insertion occurred, flow is just created,
5524  *   this is caller's responsibility to track the
5525  *   created flow.
5526  * @param[in] attr
5527  *   Flow rule attributes.
5528  * @param[in] items
5529  *   Pattern specification (list terminated by the END pattern item).
5530  * @param[in] actions
5531  *   Associated actions (list terminated by the END action).
5532  * @param[in] external
5533  *   This flow rule is created by request external to PMD.
5534  * @param[out] error
5535  *   Perform verbose error reporting if not NULL.
5536  *
5537  * @return
5538  *   A flow index on success, 0 otherwise and rte_errno is set.
5539  */
5540 static uint32_t
5541 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5542 		 const struct rte_flow_attr *attr,
5543 		 const struct rte_flow_item items[],
5544 		 const struct rte_flow_action original_actions[],
5545 		 bool external, struct rte_flow_error *error)
5546 {
5547 	struct mlx5_priv *priv = dev->data->dev_private;
5548 	struct rte_flow *flow = NULL;
5549 	struct mlx5_flow *dev_flow;
5550 	const struct rte_flow_action_rss *rss = NULL;
5551 	struct mlx5_translated_action_handle
5552 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
5553 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
5554 	union {
5555 		struct mlx5_flow_expand_rss buf;
5556 		uint8_t buffer[2048];
5557 	} expand_buffer;
5558 	union {
5559 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5560 		uint8_t buffer[2048];
5561 	} actions_rx;
5562 	union {
5563 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5564 		uint8_t buffer[2048];
5565 	} actions_hairpin_tx;
5566 	union {
5567 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5568 		uint8_t buffer[2048];
5569 	} items_tx;
5570 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5571 	struct mlx5_flow_rss_desc *rss_desc;
5572 	const struct rte_flow_action *p_actions_rx;
5573 	uint32_t i;
5574 	uint32_t idx = 0;
5575 	int hairpin_flow;
5576 	struct rte_flow_attr attr_tx = { .priority = 0 };
5577 	const struct rte_flow_action *actions;
5578 	struct rte_flow_action *translated_actions = NULL;
5579 	struct mlx5_flow_tunnel *tunnel;
5580 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5581 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5582 	struct mlx5_flow_split_info flow_split_info = {
5583 		.external = !!external,
5584 		.skip_scale = 0,
5585 		.flow_idx = 0,
5586 		.prefix_mark = 0,
5587 		.prefix_layers = 0
5588 	};
5589 	int ret;
5590 
5591 	MLX5_ASSERT(wks);
5592 	rss_desc = &wks->rss_desc;
5593 	ret = flow_action_handles_translate(dev, original_actions,
5594 					    indir_actions,
5595 					    &indir_actions_n,
5596 					    &translated_actions, error);
5597 	if (ret < 0) {
5598 		MLX5_ASSERT(translated_actions == NULL);
5599 		return 0;
5600 	}
5601 	actions = translated_actions ? translated_actions : original_actions;
5602 	p_actions_rx = actions;
5603 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5604 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5605 				external, hairpin_flow, error);
5606 	if (ret < 0)
5607 		goto error_before_hairpin_split;
5608 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5609 	if (!flow) {
5610 		rte_errno = ENOMEM;
5611 		goto error_before_hairpin_split;
5612 	}
5613 	if (hairpin_flow > 0) {
5614 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5615 			rte_errno = EINVAL;
5616 			goto error_before_hairpin_split;
5617 		}
5618 		flow_hairpin_split(dev, actions, actions_rx.actions,
5619 				   actions_hairpin_tx.actions, items_tx.items,
5620 				   idx);
5621 		p_actions_rx = actions_rx.actions;
5622 	}
5623 	flow_split_info.flow_idx = idx;
5624 	flow->drv_type = flow_get_drv_type(dev, attr);
5625 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5626 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5627 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5628 	/* RSS Action only works on NIC RX domain */
5629 	if (attr->ingress && !attr->transfer)
5630 		rss = flow_get_rss_action(p_actions_rx);
5631 	if (rss) {
5632 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5633 			return 0;
5634 		/*
5635 		 * The following information is required by
5636 		 * mlx5_flow_hashfields_adjust() in advance.
5637 		 */
5638 		rss_desc->level = rss->level;
5639 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5640 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5641 	}
5642 	flow->dev_handles = 0;
5643 	if (rss && rss->types) {
5644 		unsigned int graph_root;
5645 
5646 		graph_root = find_graph_root(items, rss->level);
5647 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5648 					   items, rss->types,
5649 					   mlx5_support_expansion, graph_root);
5650 		MLX5_ASSERT(ret > 0 &&
5651 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5652 	} else {
5653 		buf->entries = 1;
5654 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5655 	}
5656 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
5657 						      indir_actions_n);
5658 	for (i = 0; i < buf->entries; ++i) {
5659 		/* Initialize flow split data. */
5660 		flow_split_info.prefix_layers = 0;
5661 		flow_split_info.prefix_mark = 0;
5662 		flow_split_info.skip_scale = 0;
5663 		/*
5664 		 * The splitter may create multiple dev_flows,
5665 		 * depending on configuration. In the simplest
5666 		 * case it just creates unmodified original flow.
5667 		 */
5668 		ret = flow_create_split_outer(dev, flow, attr,
5669 					      buf->entry[i].pattern,
5670 					      p_actions_rx, &flow_split_info,
5671 					      error);
5672 		if (ret < 0)
5673 			goto error;
5674 		if (is_flow_tunnel_steer_rule(dev, attr,
5675 					      buf->entry[i].pattern,
5676 					      p_actions_rx)) {
5677 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5678 							   p_actions_rx,
5679 							   idx,
5680 							   &default_miss_ctx,
5681 							   error);
5682 			if (ret < 0) {
5683 				mlx5_free(default_miss_ctx.queue);
5684 				goto error;
5685 			}
5686 		}
5687 	}
5688 	/* Create the tx flow. */
5689 	if (hairpin_flow) {
5690 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5691 		attr_tx.ingress = 0;
5692 		attr_tx.egress = 1;
5693 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5694 					 actions_hairpin_tx.actions,
5695 					 idx, error);
5696 		if (!dev_flow)
5697 			goto error;
5698 		dev_flow->flow = flow;
5699 		dev_flow->external = 0;
5700 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5701 			      dev_flow->handle, next);
5702 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5703 					 items_tx.items,
5704 					 actions_hairpin_tx.actions, error);
5705 		if (ret < 0)
5706 			goto error;
5707 	}
5708 	/*
5709 	 * Update the metadata register copy table. If extensive
5710 	 * metadata feature is enabled and registers are supported
5711 	 * we might create the extra rte_flow for each unique
5712 	 * MARK/FLAG action ID.
5713 	 *
5714 	 * The table is updated for ingress Flows only, because
5715 	 * the egress Flows belong to the different device and
5716 	 * copy table should be updated in peer NIC Rx domain.
5717 	 */
5718 	if (attr->ingress &&
5719 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5720 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5721 		if (ret)
5722 			goto error;
5723 	}
5724 	/*
5725 	 * If the flow is external (from application) OR device is started,
5726 	 * OR mreg discover, then apply immediately.
5727 	 */
5728 	if (external || dev->data->dev_started ||
5729 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5730 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
5731 		ret = flow_drv_apply(dev, flow, error);
5732 		if (ret < 0)
5733 			goto error;
5734 	}
5735 	if (list) {
5736 		rte_spinlock_lock(&priv->flow_list_lock);
5737 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5738 			     flow, next);
5739 		rte_spinlock_unlock(&priv->flow_list_lock);
5740 	}
5741 	flow_rxq_flags_set(dev, flow);
5742 	rte_free(translated_actions);
5743 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5744 	if (tunnel) {
5745 		flow->tunnel = 1;
5746 		flow->tunnel_id = tunnel->tunnel_id;
5747 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5748 		mlx5_free(default_miss_ctx.queue);
5749 	}
5750 	mlx5_flow_pop_thread_workspace();
5751 	return idx;
5752 error:
5753 	MLX5_ASSERT(flow);
5754 	ret = rte_errno; /* Save rte_errno before cleanup. */
5755 	flow_mreg_del_copy_action(dev, flow);
5756 	flow_drv_destroy(dev, flow);
5757 	if (rss_desc->shared_rss)
5758 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5759 			mlx5_ipool_get
5760 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5761 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5762 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5763 	rte_errno = ret; /* Restore rte_errno. */
5764 	ret = rte_errno;
5765 	rte_errno = ret;
5766 	mlx5_flow_pop_thread_workspace();
5767 error_before_hairpin_split:
5768 	rte_free(translated_actions);
5769 	return 0;
5770 }
5771 
5772 /**
5773  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5774  * incoming packets to table 1.
5775  *
5776  * Other flow rules, requested for group n, will be created in
5777  * e-switch table n+1.
5778  * Jump action to e-switch group n will be created to group n+1.
5779  *
5780  * Used when working in switchdev mode, to utilise advantages of table 1
5781  * and above.
5782  *
5783  * @param dev
5784  *   Pointer to Ethernet device.
5785  *
5786  * @return
5787  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5788  */
5789 struct rte_flow *
5790 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5791 {
5792 	const struct rte_flow_attr attr = {
5793 		.group = 0,
5794 		.priority = 0,
5795 		.ingress = 1,
5796 		.egress = 0,
5797 		.transfer = 1,
5798 	};
5799 	const struct rte_flow_item pattern = {
5800 		.type = RTE_FLOW_ITEM_TYPE_END,
5801 	};
5802 	struct rte_flow_action_jump jump = {
5803 		.group = 1,
5804 	};
5805 	const struct rte_flow_action actions[] = {
5806 		{
5807 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5808 			.conf = &jump,
5809 		},
5810 		{
5811 			.type = RTE_FLOW_ACTION_TYPE_END,
5812 		},
5813 	};
5814 	struct mlx5_priv *priv = dev->data->dev_private;
5815 	struct rte_flow_error error;
5816 
5817 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5818 						   &attr, &pattern,
5819 						   actions, false, &error);
5820 }
5821 
5822 /**
5823  * Validate a flow supported by the NIC.
5824  *
5825  * @see rte_flow_validate()
5826  * @see rte_flow_ops
5827  */
5828 int
5829 mlx5_flow_validate(struct rte_eth_dev *dev,
5830 		   const struct rte_flow_attr *attr,
5831 		   const struct rte_flow_item items[],
5832 		   const struct rte_flow_action original_actions[],
5833 		   struct rte_flow_error *error)
5834 {
5835 	int hairpin_flow;
5836 	struct mlx5_translated_action_handle
5837 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
5838 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
5839 	const struct rte_flow_action *actions;
5840 	struct rte_flow_action *translated_actions = NULL;
5841 	int ret = flow_action_handles_translate(dev, original_actions,
5842 						indir_actions,
5843 						&indir_actions_n,
5844 						&translated_actions, error);
5845 
5846 	if (ret)
5847 		return ret;
5848 	actions = translated_actions ? translated_actions : original_actions;
5849 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5850 	ret = flow_drv_validate(dev, attr, items, actions,
5851 				true, hairpin_flow, error);
5852 	rte_free(translated_actions);
5853 	return ret;
5854 }
5855 
5856 /**
5857  * Create a flow.
5858  *
5859  * @see rte_flow_create()
5860  * @see rte_flow_ops
5861  */
5862 struct rte_flow *
5863 mlx5_flow_create(struct rte_eth_dev *dev,
5864 		 const struct rte_flow_attr *attr,
5865 		 const struct rte_flow_item items[],
5866 		 const struct rte_flow_action actions[],
5867 		 struct rte_flow_error *error)
5868 {
5869 	struct mlx5_priv *priv = dev->data->dev_private;
5870 
5871 	/*
5872 	 * If the device is not started yet, it is not allowed to created a
5873 	 * flow from application. PMD default flows and traffic control flows
5874 	 * are not affected.
5875 	 */
5876 	if (unlikely(!dev->data->dev_started)) {
5877 		DRV_LOG(DEBUG, "port %u is not started when "
5878 			"inserting a flow", dev->data->port_id);
5879 		rte_flow_error_set(error, ENODEV,
5880 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5881 				   NULL,
5882 				   "port not started");
5883 		return NULL;
5884 	}
5885 
5886 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5887 				  attr, items, actions, true, error);
5888 }
5889 
5890 /**
5891  * Destroy a flow in a list.
5892  *
5893  * @param dev
5894  *   Pointer to Ethernet device.
5895  * @param list
5896  *   Pointer to the Indexed flow list. If this parameter NULL,
5897  *   there is no flow removal from the list. Be noted that as
5898  *   flow is add to the indexed list, memory of the indexed
5899  *   list points to maybe changed as flow destroyed.
5900  * @param[in] flow_idx
5901  *   Index of flow to destroy.
5902  */
5903 static void
5904 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5905 		  uint32_t flow_idx)
5906 {
5907 	struct mlx5_priv *priv = dev->data->dev_private;
5908 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5909 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5910 
5911 	if (!flow)
5912 		return;
5913 	/*
5914 	 * Update RX queue flags only if port is started, otherwise it is
5915 	 * already clean.
5916 	 */
5917 	if (dev->data->dev_started)
5918 		flow_rxq_flags_trim(dev, flow);
5919 	flow_drv_destroy(dev, flow);
5920 	if (list) {
5921 		rte_spinlock_lock(&priv->flow_list_lock);
5922 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5923 			     flow_idx, flow, next);
5924 		rte_spinlock_unlock(&priv->flow_list_lock);
5925 	}
5926 	if (flow->tunnel) {
5927 		struct mlx5_flow_tunnel *tunnel;
5928 
5929 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5930 		RTE_VERIFY(tunnel);
5931 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5932 			mlx5_flow_tunnel_free(dev, tunnel);
5933 	}
5934 	flow_mreg_del_copy_action(dev, flow);
5935 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5936 }
5937 
5938 /**
5939  * Destroy all flows.
5940  *
5941  * @param dev
5942  *   Pointer to Ethernet device.
5943  * @param list
5944  *   Pointer to the Indexed flow list.
5945  * @param active
5946  *   If flushing is called avtively.
5947  */
5948 void
5949 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5950 {
5951 	uint32_t num_flushed = 0;
5952 
5953 	while (*list) {
5954 		flow_list_destroy(dev, list, *list);
5955 		num_flushed++;
5956 	}
5957 	if (active) {
5958 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5959 			dev->data->port_id, num_flushed);
5960 	}
5961 }
5962 
5963 /**
5964  * Stop all default actions for flows.
5965  *
5966  * @param dev
5967  *   Pointer to Ethernet device.
5968  */
5969 void
5970 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5971 {
5972 	flow_mreg_del_default_copy_action(dev);
5973 	flow_rxq_flags_clear(dev);
5974 }
5975 
5976 /**
5977  * Start all default actions for flows.
5978  *
5979  * @param dev
5980  *   Pointer to Ethernet device.
5981  * @return
5982  *   0 on success, a negative errno value otherwise and rte_errno is set.
5983  */
5984 int
5985 mlx5_flow_start_default(struct rte_eth_dev *dev)
5986 {
5987 	struct rte_flow_error error;
5988 
5989 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5990 	return flow_mreg_add_default_copy_action(dev, &error);
5991 }
5992 
5993 /**
5994  * Release key of thread specific flow workspace data.
5995  */
5996 void
5997 flow_release_workspace(void *data)
5998 {
5999 	struct mlx5_flow_workspace *wks = data;
6000 	struct mlx5_flow_workspace *next;
6001 
6002 	while (wks) {
6003 		next = wks->next;
6004 		free(wks->rss_desc.queue);
6005 		free(wks);
6006 		wks = next;
6007 	}
6008 }
6009 
6010 /**
6011  * Get thread specific current flow workspace.
6012  *
6013  * @return pointer to thread specific flow workspace data, NULL on error.
6014  */
6015 struct mlx5_flow_workspace*
6016 mlx5_flow_get_thread_workspace(void)
6017 {
6018 	struct mlx5_flow_workspace *data;
6019 
6020 	data = mlx5_flow_os_get_specific_workspace();
6021 	MLX5_ASSERT(data && data->inuse);
6022 	if (!data || !data->inuse)
6023 		DRV_LOG(ERR, "flow workspace not initialized.");
6024 	return data;
6025 }
6026 
6027 /**
6028  * Allocate and init new flow workspace.
6029  *
6030  * @return pointer to flow workspace data, NULL on error.
6031  */
6032 static struct mlx5_flow_workspace*
6033 flow_alloc_thread_workspace(void)
6034 {
6035 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6036 
6037 	if (!data) {
6038 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6039 			"memory.");
6040 		return NULL;
6041 	}
6042 	data->rss_desc.queue = calloc(1,
6043 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6044 	if (!data->rss_desc.queue)
6045 		goto err;
6046 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6047 	return data;
6048 err:
6049 	if (data->rss_desc.queue)
6050 		free(data->rss_desc.queue);
6051 	free(data);
6052 	return NULL;
6053 }
6054 
6055 /**
6056  * Get new thread specific flow workspace.
6057  *
6058  * If current workspace inuse, create new one and set as current.
6059  *
6060  * @return pointer to thread specific flow workspace data, NULL on error.
6061  */
6062 static struct mlx5_flow_workspace*
6063 mlx5_flow_push_thread_workspace(void)
6064 {
6065 	struct mlx5_flow_workspace *curr;
6066 	struct mlx5_flow_workspace *data;
6067 
6068 	curr = mlx5_flow_os_get_specific_workspace();
6069 	if (!curr) {
6070 		data = flow_alloc_thread_workspace();
6071 		if (!data)
6072 			return NULL;
6073 	} else if (!curr->inuse) {
6074 		data = curr;
6075 	} else if (curr->next) {
6076 		data = curr->next;
6077 	} else {
6078 		data = flow_alloc_thread_workspace();
6079 		if (!data)
6080 			return NULL;
6081 		curr->next = data;
6082 		data->prev = curr;
6083 	}
6084 	data->inuse = 1;
6085 	data->flow_idx = 0;
6086 	/* Set as current workspace */
6087 	if (mlx5_flow_os_set_specific_workspace(data))
6088 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6089 	return data;
6090 }
6091 
6092 /**
6093  * Close current thread specific flow workspace.
6094  *
6095  * If previous workspace available, set it as current.
6096  *
6097  * @return pointer to thread specific flow workspace data, NULL on error.
6098  */
6099 static void
6100 mlx5_flow_pop_thread_workspace(void)
6101 {
6102 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6103 
6104 	if (!data)
6105 		return;
6106 	if (!data->inuse) {
6107 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6108 		return;
6109 	}
6110 	data->inuse = 0;
6111 	if (!data->prev)
6112 		return;
6113 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6114 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6115 }
6116 
6117 /**
6118  * Verify the flow list is empty
6119  *
6120  * @param dev
6121  *  Pointer to Ethernet device.
6122  *
6123  * @return the number of flows not released.
6124  */
6125 int
6126 mlx5_flow_verify(struct rte_eth_dev *dev)
6127 {
6128 	struct mlx5_priv *priv = dev->data->dev_private;
6129 	struct rte_flow *flow;
6130 	uint32_t idx;
6131 	int ret = 0;
6132 
6133 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6134 		      flow, next) {
6135 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6136 			dev->data->port_id, (void *)flow);
6137 		++ret;
6138 	}
6139 	return ret;
6140 }
6141 
6142 /**
6143  * Enable default hairpin egress flow.
6144  *
6145  * @param dev
6146  *   Pointer to Ethernet device.
6147  * @param queue
6148  *   The queue index.
6149  *
6150  * @return
6151  *   0 on success, a negative errno value otherwise and rte_errno is set.
6152  */
6153 int
6154 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6155 			    uint32_t queue)
6156 {
6157 	struct mlx5_priv *priv = dev->data->dev_private;
6158 	const struct rte_flow_attr attr = {
6159 		.egress = 1,
6160 		.priority = 0,
6161 	};
6162 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6163 		.queue = queue,
6164 	};
6165 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6166 		.queue = UINT32_MAX,
6167 	};
6168 	struct rte_flow_item items[] = {
6169 		{
6170 			.type = (enum rte_flow_item_type)
6171 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6172 			.spec = &queue_spec,
6173 			.last = NULL,
6174 			.mask = &queue_mask,
6175 		},
6176 		{
6177 			.type = RTE_FLOW_ITEM_TYPE_END,
6178 		},
6179 	};
6180 	struct rte_flow_action_jump jump = {
6181 		.group = MLX5_HAIRPIN_TX_TABLE,
6182 	};
6183 	struct rte_flow_action actions[2];
6184 	uint32_t flow_idx;
6185 	struct rte_flow_error error;
6186 
6187 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6188 	actions[0].conf = &jump;
6189 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6190 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6191 				&attr, items, actions, false, &error);
6192 	if (!flow_idx) {
6193 		DRV_LOG(DEBUG,
6194 			"Failed to create ctrl flow: rte_errno(%d),"
6195 			" type(%d), message(%s)",
6196 			rte_errno, error.type,
6197 			error.message ? error.message : " (no stated reason)");
6198 		return -rte_errno;
6199 	}
6200 	return 0;
6201 }
6202 
6203 /**
6204  * Enable a control flow configured from the control plane.
6205  *
6206  * @param dev
6207  *   Pointer to Ethernet device.
6208  * @param eth_spec
6209  *   An Ethernet flow spec to apply.
6210  * @param eth_mask
6211  *   An Ethernet flow mask to apply.
6212  * @param vlan_spec
6213  *   A VLAN flow spec to apply.
6214  * @param vlan_mask
6215  *   A VLAN flow mask to apply.
6216  *
6217  * @return
6218  *   0 on success, a negative errno value otherwise and rte_errno is set.
6219  */
6220 int
6221 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6222 		    struct rte_flow_item_eth *eth_spec,
6223 		    struct rte_flow_item_eth *eth_mask,
6224 		    struct rte_flow_item_vlan *vlan_spec,
6225 		    struct rte_flow_item_vlan *vlan_mask)
6226 {
6227 	struct mlx5_priv *priv = dev->data->dev_private;
6228 	const struct rte_flow_attr attr = {
6229 		.ingress = 1,
6230 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6231 	};
6232 	struct rte_flow_item items[] = {
6233 		{
6234 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6235 			.spec = eth_spec,
6236 			.last = NULL,
6237 			.mask = eth_mask,
6238 		},
6239 		{
6240 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6241 					      RTE_FLOW_ITEM_TYPE_END,
6242 			.spec = vlan_spec,
6243 			.last = NULL,
6244 			.mask = vlan_mask,
6245 		},
6246 		{
6247 			.type = RTE_FLOW_ITEM_TYPE_END,
6248 		},
6249 	};
6250 	uint16_t queue[priv->reta_idx_n];
6251 	struct rte_flow_action_rss action_rss = {
6252 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6253 		.level = 0,
6254 		.types = priv->rss_conf.rss_hf,
6255 		.key_len = priv->rss_conf.rss_key_len,
6256 		.queue_num = priv->reta_idx_n,
6257 		.key = priv->rss_conf.rss_key,
6258 		.queue = queue,
6259 	};
6260 	struct rte_flow_action actions[] = {
6261 		{
6262 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6263 			.conf = &action_rss,
6264 		},
6265 		{
6266 			.type = RTE_FLOW_ACTION_TYPE_END,
6267 		},
6268 	};
6269 	uint32_t flow_idx;
6270 	struct rte_flow_error error;
6271 	unsigned int i;
6272 
6273 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6274 		return 0;
6275 	}
6276 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6277 		action_rss.types = 0;
6278 	for (i = 0; i != priv->reta_idx_n; ++i)
6279 		queue[i] = (*priv->reta_idx)[i];
6280 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6281 				&attr, items, actions, false, &error);
6282 	if (!flow_idx)
6283 		return -rte_errno;
6284 	return 0;
6285 }
6286 
6287 /**
6288  * Enable a flow control configured from the control plane.
6289  *
6290  * @param dev
6291  *   Pointer to Ethernet device.
6292  * @param eth_spec
6293  *   An Ethernet flow spec to apply.
6294  * @param eth_mask
6295  *   An Ethernet flow mask to apply.
6296  *
6297  * @return
6298  *   0 on success, a negative errno value otherwise and rte_errno is set.
6299  */
6300 int
6301 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6302 	       struct rte_flow_item_eth *eth_spec,
6303 	       struct rte_flow_item_eth *eth_mask)
6304 {
6305 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6306 }
6307 
6308 /**
6309  * Create default miss flow rule matching lacp traffic
6310  *
6311  * @param dev
6312  *   Pointer to Ethernet device.
6313  * @param eth_spec
6314  *   An Ethernet flow spec to apply.
6315  *
6316  * @return
6317  *   0 on success, a negative errno value otherwise and rte_errno is set.
6318  */
6319 int
6320 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6321 {
6322 	struct mlx5_priv *priv = dev->data->dev_private;
6323 	/*
6324 	 * The LACP matching is done by only using ether type since using
6325 	 * a multicast dst mac causes kernel to give low priority to this flow.
6326 	 */
6327 	static const struct rte_flow_item_eth lacp_spec = {
6328 		.type = RTE_BE16(0x8809),
6329 	};
6330 	static const struct rte_flow_item_eth lacp_mask = {
6331 		.type = 0xffff,
6332 	};
6333 	const struct rte_flow_attr attr = {
6334 		.ingress = 1,
6335 	};
6336 	struct rte_flow_item items[] = {
6337 		{
6338 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6339 			.spec = &lacp_spec,
6340 			.mask = &lacp_mask,
6341 		},
6342 		{
6343 			.type = RTE_FLOW_ITEM_TYPE_END,
6344 		},
6345 	};
6346 	struct rte_flow_action actions[] = {
6347 		{
6348 			.type = (enum rte_flow_action_type)
6349 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6350 		},
6351 		{
6352 			.type = RTE_FLOW_ACTION_TYPE_END,
6353 		},
6354 	};
6355 	struct rte_flow_error error;
6356 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6357 				&attr, items, actions, false, &error);
6358 
6359 	if (!flow_idx)
6360 		return -rte_errno;
6361 	return 0;
6362 }
6363 
6364 /**
6365  * Destroy a flow.
6366  *
6367  * @see rte_flow_destroy()
6368  * @see rte_flow_ops
6369  */
6370 int
6371 mlx5_flow_destroy(struct rte_eth_dev *dev,
6372 		  struct rte_flow *flow,
6373 		  struct rte_flow_error *error __rte_unused)
6374 {
6375 	struct mlx5_priv *priv = dev->data->dev_private;
6376 
6377 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6378 	return 0;
6379 }
6380 
6381 /**
6382  * Destroy all flows.
6383  *
6384  * @see rte_flow_flush()
6385  * @see rte_flow_ops
6386  */
6387 int
6388 mlx5_flow_flush(struct rte_eth_dev *dev,
6389 		struct rte_flow_error *error __rte_unused)
6390 {
6391 	struct mlx5_priv *priv = dev->data->dev_private;
6392 
6393 	mlx5_flow_list_flush(dev, &priv->flows, false);
6394 	return 0;
6395 }
6396 
6397 /**
6398  * Isolated mode.
6399  *
6400  * @see rte_flow_isolate()
6401  * @see rte_flow_ops
6402  */
6403 int
6404 mlx5_flow_isolate(struct rte_eth_dev *dev,
6405 		  int enable,
6406 		  struct rte_flow_error *error)
6407 {
6408 	struct mlx5_priv *priv = dev->data->dev_private;
6409 
6410 	if (dev->data->dev_started) {
6411 		rte_flow_error_set(error, EBUSY,
6412 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6413 				   NULL,
6414 				   "port must be stopped first");
6415 		return -rte_errno;
6416 	}
6417 	priv->isolated = !!enable;
6418 	if (enable)
6419 		dev->dev_ops = &mlx5_dev_ops_isolate;
6420 	else
6421 		dev->dev_ops = &mlx5_dev_ops;
6422 
6423 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6424 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6425 
6426 	return 0;
6427 }
6428 
6429 /**
6430  * Query a flow.
6431  *
6432  * @see rte_flow_query()
6433  * @see rte_flow_ops
6434  */
6435 static int
6436 flow_drv_query(struct rte_eth_dev *dev,
6437 	       uint32_t flow_idx,
6438 	       const struct rte_flow_action *actions,
6439 	       void *data,
6440 	       struct rte_flow_error *error)
6441 {
6442 	struct mlx5_priv *priv = dev->data->dev_private;
6443 	const struct mlx5_flow_driver_ops *fops;
6444 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6445 					       [MLX5_IPOOL_RTE_FLOW],
6446 					       flow_idx);
6447 	enum mlx5_flow_drv_type ftype;
6448 
6449 	if (!flow) {
6450 		return rte_flow_error_set(error, ENOENT,
6451 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6452 			  NULL,
6453 			  "invalid flow handle");
6454 	}
6455 	ftype = flow->drv_type;
6456 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6457 	fops = flow_get_drv_ops(ftype);
6458 
6459 	return fops->query(dev, flow, actions, data, error);
6460 }
6461 
6462 /**
6463  * Query a flow.
6464  *
6465  * @see rte_flow_query()
6466  * @see rte_flow_ops
6467  */
6468 int
6469 mlx5_flow_query(struct rte_eth_dev *dev,
6470 		struct rte_flow *flow,
6471 		const struct rte_flow_action *actions,
6472 		void *data,
6473 		struct rte_flow_error *error)
6474 {
6475 	int ret;
6476 
6477 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6478 			     error);
6479 	if (ret < 0)
6480 		return ret;
6481 	return 0;
6482 }
6483 
6484 /**
6485  * Get rte_flow callbacks.
6486  *
6487  * @param dev
6488  *   Pointer to Ethernet device structure.
6489  * @param ops
6490  *   Pointer to operation-specific structure.
6491  *
6492  * @return 0
6493  */
6494 int
6495 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
6496 		  const struct rte_flow_ops **ops)
6497 {
6498 	*ops = &mlx5_flow_ops;
6499 	return 0;
6500 }
6501 
6502 /**
6503  * Create the needed meter and suffix tables.
6504  *
6505  * @param[in] dev
6506  *   Pointer to Ethernet device.
6507  * @param[in] fm
6508  *   Pointer to the flow meter.
6509  *
6510  * @return
6511  *   Pointer to table set on success, NULL otherwise.
6512  */
6513 struct mlx5_meter_domains_infos *
6514 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6515 			  const struct mlx5_flow_meter *fm)
6516 {
6517 	const struct mlx5_flow_driver_ops *fops;
6518 
6519 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6520 	return fops->create_mtr_tbls(dev, fm);
6521 }
6522 
6523 /**
6524  * Destroy the meter table set.
6525  *
6526  * @param[in] dev
6527  *   Pointer to Ethernet device.
6528  * @param[in] tbl
6529  *   Pointer to the meter table set.
6530  *
6531  * @return
6532  *   0 on success.
6533  */
6534 int
6535 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6536 			   struct mlx5_meter_domains_infos *tbls)
6537 {
6538 	const struct mlx5_flow_driver_ops *fops;
6539 
6540 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6541 	return fops->destroy_mtr_tbls(dev, tbls);
6542 }
6543 
6544 /**
6545  * Create policer rules.
6546  *
6547  * @param[in] dev
6548  *   Pointer to Ethernet device.
6549  * @param[in] fm
6550  *   Pointer to flow meter structure.
6551  * @param[in] attr
6552  *   Pointer to flow attributes.
6553  *
6554  * @return
6555  *   0 on success, -1 otherwise.
6556  */
6557 int
6558 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6559 			       struct mlx5_flow_meter *fm,
6560 			       const struct rte_flow_attr *attr)
6561 {
6562 	const struct mlx5_flow_driver_ops *fops;
6563 
6564 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6565 	return fops->create_policer_rules(dev, fm, attr);
6566 }
6567 
6568 /**
6569  * Destroy policer rules.
6570  *
6571  * @param[in] fm
6572  *   Pointer to flow meter structure.
6573  * @param[in] attr
6574  *   Pointer to flow attributes.
6575  *
6576  * @return
6577  *   0 on success, -1 otherwise.
6578  */
6579 int
6580 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6581 				struct mlx5_flow_meter *fm,
6582 				const struct rte_flow_attr *attr)
6583 {
6584 	const struct mlx5_flow_driver_ops *fops;
6585 
6586 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6587 	return fops->destroy_policer_rules(dev, fm, attr);
6588 }
6589 
6590 /**
6591  * Allocate a counter.
6592  *
6593  * @param[in] dev
6594  *   Pointer to Ethernet device structure.
6595  *
6596  * @return
6597  *   Index to allocated counter  on success, 0 otherwise.
6598  */
6599 uint32_t
6600 mlx5_counter_alloc(struct rte_eth_dev *dev)
6601 {
6602 	const struct mlx5_flow_driver_ops *fops;
6603 	struct rte_flow_attr attr = { .transfer = 0 };
6604 
6605 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6606 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6607 		return fops->counter_alloc(dev);
6608 	}
6609 	DRV_LOG(ERR,
6610 		"port %u counter allocate is not supported.",
6611 		 dev->data->port_id);
6612 	return 0;
6613 }
6614 
6615 /**
6616  * Free a counter.
6617  *
6618  * @param[in] dev
6619  *   Pointer to Ethernet device structure.
6620  * @param[in] cnt
6621  *   Index to counter to be free.
6622  */
6623 void
6624 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6625 {
6626 	const struct mlx5_flow_driver_ops *fops;
6627 	struct rte_flow_attr attr = { .transfer = 0 };
6628 
6629 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6630 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6631 		fops->counter_free(dev, cnt);
6632 		return;
6633 	}
6634 	DRV_LOG(ERR,
6635 		"port %u counter free is not supported.",
6636 		 dev->data->port_id);
6637 }
6638 
6639 /**
6640  * Query counter statistics.
6641  *
6642  * @param[in] dev
6643  *   Pointer to Ethernet device structure.
6644  * @param[in] cnt
6645  *   Index to counter to query.
6646  * @param[in] clear
6647  *   Set to clear counter statistics.
6648  * @param[out] pkts
6649  *   The counter hits packets number to save.
6650  * @param[out] bytes
6651  *   The counter hits bytes number to save.
6652  *
6653  * @return
6654  *   0 on success, a negative errno value otherwise.
6655  */
6656 int
6657 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6658 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6659 {
6660 	const struct mlx5_flow_driver_ops *fops;
6661 	struct rte_flow_attr attr = { .transfer = 0 };
6662 
6663 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6664 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6665 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6666 	}
6667 	DRV_LOG(ERR,
6668 		"port %u counter query is not supported.",
6669 		 dev->data->port_id);
6670 	return -ENOTSUP;
6671 }
6672 
6673 /**
6674  * Allocate a new memory for the counter values wrapped by all the needed
6675  * management.
6676  *
6677  * @param[in] sh
6678  *   Pointer to mlx5_dev_ctx_shared object.
6679  *
6680  * @return
6681  *   0 on success, a negative errno value otherwise.
6682  */
6683 static int
6684 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6685 {
6686 	struct mlx5_devx_mkey_attr mkey_attr;
6687 	struct mlx5_counter_stats_mem_mng *mem_mng;
6688 	volatile struct flow_counter_stats *raw_data;
6689 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6690 	int size = (sizeof(struct flow_counter_stats) *
6691 			MLX5_COUNTERS_PER_POOL +
6692 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6693 			sizeof(struct mlx5_counter_stats_mem_mng);
6694 	size_t pgsize = rte_mem_page_size();
6695 	uint8_t *mem;
6696 	int i;
6697 
6698 	if (pgsize == (size_t)-1) {
6699 		DRV_LOG(ERR, "Failed to get mem page size");
6700 		rte_errno = ENOMEM;
6701 		return -ENOMEM;
6702 	}
6703 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6704 	if (!mem) {
6705 		rte_errno = ENOMEM;
6706 		return -ENOMEM;
6707 	}
6708 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6709 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6710 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6711 						 IBV_ACCESS_LOCAL_WRITE);
6712 	if (!mem_mng->umem) {
6713 		rte_errno = errno;
6714 		mlx5_free(mem);
6715 		return -rte_errno;
6716 	}
6717 	mkey_attr.addr = (uintptr_t)mem;
6718 	mkey_attr.size = size;
6719 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6720 	mkey_attr.pd = sh->pdn;
6721 	mkey_attr.log_entity_size = 0;
6722 	mkey_attr.pg_access = 0;
6723 	mkey_attr.klm_array = NULL;
6724 	mkey_attr.klm_num = 0;
6725 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6726 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6727 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6728 	if (!mem_mng->dm) {
6729 		mlx5_os_umem_dereg(mem_mng->umem);
6730 		rte_errno = errno;
6731 		mlx5_free(mem);
6732 		return -rte_errno;
6733 	}
6734 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6735 	raw_data = (volatile struct flow_counter_stats *)mem;
6736 	for (i = 0; i < raws_n; ++i) {
6737 		mem_mng->raws[i].mem_mng = mem_mng;
6738 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6739 	}
6740 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6741 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6742 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6743 				 next);
6744 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6745 	sh->cmng.mem_mng = mem_mng;
6746 	return 0;
6747 }
6748 
6749 /**
6750  * Set the statistic memory to the new counter pool.
6751  *
6752  * @param[in] sh
6753  *   Pointer to mlx5_dev_ctx_shared object.
6754  * @param[in] pool
6755  *   Pointer to the pool to set the statistic memory.
6756  *
6757  * @return
6758  *   0 on success, a negative errno value otherwise.
6759  */
6760 static int
6761 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6762 			       struct mlx5_flow_counter_pool *pool)
6763 {
6764 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6765 	/* Resize statistic memory once used out. */
6766 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6767 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6768 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6769 		return -1;
6770 	}
6771 	rte_spinlock_lock(&pool->sl);
6772 	pool->raw = cmng->mem_mng->raws + pool->index %
6773 		    MLX5_CNT_CONTAINER_RESIZE;
6774 	rte_spinlock_unlock(&pool->sl);
6775 	pool->raw_hw = NULL;
6776 	return 0;
6777 }
6778 
6779 #define MLX5_POOL_QUERY_FREQ_US 1000000
6780 
6781 /**
6782  * Set the periodic procedure for triggering asynchronous batch queries for all
6783  * the counter pools.
6784  *
6785  * @param[in] sh
6786  *   Pointer to mlx5_dev_ctx_shared object.
6787  */
6788 void
6789 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6790 {
6791 	uint32_t pools_n, us;
6792 
6793 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6794 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6795 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6796 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6797 		sh->cmng.query_thread_on = 0;
6798 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6799 	} else {
6800 		sh->cmng.query_thread_on = 1;
6801 	}
6802 }
6803 
6804 /**
6805  * The periodic procedure for triggering asynchronous batch queries for all the
6806  * counter pools. This function is probably called by the host thread.
6807  *
6808  * @param[in] arg
6809  *   The parameter for the alarm process.
6810  */
6811 void
6812 mlx5_flow_query_alarm(void *arg)
6813 {
6814 	struct mlx5_dev_ctx_shared *sh = arg;
6815 	int ret;
6816 	uint16_t pool_index = sh->cmng.pool_index;
6817 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6818 	struct mlx5_flow_counter_pool *pool;
6819 	uint16_t n_valid;
6820 
6821 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6822 		goto set_alarm;
6823 	rte_spinlock_lock(&cmng->pool_update_sl);
6824 	pool = cmng->pools[pool_index];
6825 	n_valid = cmng->n_valid;
6826 	rte_spinlock_unlock(&cmng->pool_update_sl);
6827 	/* Set the statistic memory to the new created pool. */
6828 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6829 		goto set_alarm;
6830 	if (pool->raw_hw)
6831 		/* There is a pool query in progress. */
6832 		goto set_alarm;
6833 	pool->raw_hw =
6834 		LIST_FIRST(&sh->cmng.free_stat_raws);
6835 	if (!pool->raw_hw)
6836 		/* No free counter statistics raw memory. */
6837 		goto set_alarm;
6838 	/*
6839 	 * Identify the counters released between query trigger and query
6840 	 * handle more efficiently. The counter released in this gap period
6841 	 * should wait for a new round of query as the new arrived packets
6842 	 * will not be taken into account.
6843 	 */
6844 	pool->query_gen++;
6845 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6846 					       MLX5_COUNTERS_PER_POOL,
6847 					       NULL, NULL,
6848 					       pool->raw_hw->mem_mng->dm->id,
6849 					       (void *)(uintptr_t)
6850 					       pool->raw_hw->data,
6851 					       sh->devx_comp,
6852 					       (uint64_t)(uintptr_t)pool);
6853 	if (ret) {
6854 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6855 			" %d", pool->min_dcs->id);
6856 		pool->raw_hw = NULL;
6857 		goto set_alarm;
6858 	}
6859 	LIST_REMOVE(pool->raw_hw, next);
6860 	sh->cmng.pending_queries++;
6861 	pool_index++;
6862 	if (pool_index >= n_valid)
6863 		pool_index = 0;
6864 set_alarm:
6865 	sh->cmng.pool_index = pool_index;
6866 	mlx5_set_query_alarm(sh);
6867 }
6868 
6869 /**
6870  * Check and callback event for new aged flow in the counter pool
6871  *
6872  * @param[in] sh
6873  *   Pointer to mlx5_dev_ctx_shared object.
6874  * @param[in] pool
6875  *   Pointer to Current counter pool.
6876  */
6877 static void
6878 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6879 		   struct mlx5_flow_counter_pool *pool)
6880 {
6881 	struct mlx5_priv *priv;
6882 	struct mlx5_flow_counter *cnt;
6883 	struct mlx5_age_info *age_info;
6884 	struct mlx5_age_param *age_param;
6885 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6886 	struct mlx5_counter_stats_raw *prev = pool->raw;
6887 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6888 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6889 	uint16_t expected = AGE_CANDIDATE;
6890 	uint32_t i;
6891 
6892 	pool->time_of_last_age_check = curr_time;
6893 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6894 		cnt = MLX5_POOL_GET_CNT(pool, i);
6895 		age_param = MLX5_CNT_TO_AGE(cnt);
6896 		if (__atomic_load_n(&age_param->state,
6897 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6898 			continue;
6899 		if (cur->data[i].hits != prev->data[i].hits) {
6900 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6901 					 __ATOMIC_RELAXED);
6902 			continue;
6903 		}
6904 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6905 				       time_delta,
6906 				       __ATOMIC_RELAXED) <= age_param->timeout)
6907 			continue;
6908 		/**
6909 		 * Hold the lock first, or if between the
6910 		 * state AGE_TMOUT and tailq operation the
6911 		 * release happened, the release procedure
6912 		 * may delete a non-existent tailq node.
6913 		 */
6914 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6915 		age_info = GET_PORT_AGE_INFO(priv);
6916 		rte_spinlock_lock(&age_info->aged_sl);
6917 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6918 						AGE_TMOUT, false,
6919 						__ATOMIC_RELAXED,
6920 						__ATOMIC_RELAXED)) {
6921 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6922 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6923 		}
6924 		rte_spinlock_unlock(&age_info->aged_sl);
6925 	}
6926 	mlx5_age_event_prepare(sh);
6927 }
6928 
6929 /**
6930  * Handler for the HW respond about ready values from an asynchronous batch
6931  * query. This function is probably called by the host thread.
6932  *
6933  * @param[in] sh
6934  *   The pointer to the shared device context.
6935  * @param[in] async_id
6936  *   The Devx async ID.
6937  * @param[in] status
6938  *   The status of the completion.
6939  */
6940 void
6941 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6942 				  uint64_t async_id, int status)
6943 {
6944 	struct mlx5_flow_counter_pool *pool =
6945 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6946 	struct mlx5_counter_stats_raw *raw_to_free;
6947 	uint8_t query_gen = pool->query_gen ^ 1;
6948 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6949 	enum mlx5_counter_type cnt_type =
6950 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6951 				MLX5_COUNTER_TYPE_ORIGIN;
6952 
6953 	if (unlikely(status)) {
6954 		raw_to_free = pool->raw_hw;
6955 	} else {
6956 		raw_to_free = pool->raw;
6957 		if (pool->is_aged)
6958 			mlx5_flow_aging_check(sh, pool);
6959 		rte_spinlock_lock(&pool->sl);
6960 		pool->raw = pool->raw_hw;
6961 		rte_spinlock_unlock(&pool->sl);
6962 		/* Be sure the new raw counters data is updated in memory. */
6963 		rte_io_wmb();
6964 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6965 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6966 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6967 				     &pool->counters[query_gen], next);
6968 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6969 		}
6970 	}
6971 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6972 	pool->raw_hw = NULL;
6973 	sh->cmng.pending_queries--;
6974 }
6975 
6976 static int
6977 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6978 		    const struct flow_grp_info *grp_info,
6979 		    struct rte_flow_error *error)
6980 {
6981 	if (grp_info->transfer && grp_info->external &&
6982 	    grp_info->fdb_def_rule) {
6983 		if (group == UINT32_MAX)
6984 			return rte_flow_error_set
6985 						(error, EINVAL,
6986 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6987 						 NULL,
6988 						 "group index not supported");
6989 		*table = group + 1;
6990 	} else {
6991 		*table = group;
6992 	}
6993 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6994 	return 0;
6995 }
6996 
6997 /**
6998  * Translate the rte_flow group index to HW table value.
6999  *
7000  * If tunnel offload is disabled, all group ids converted to flow table
7001  * id using the standard method.
7002  * If tunnel offload is enabled, group id can be converted using the
7003  * standard or tunnel conversion method. Group conversion method
7004  * selection depends on flags in `grp_info` parameter:
7005  * - Internal (grp_info.external == 0) groups conversion uses the
7006  *   standard method.
7007  * - Group ids in JUMP action converted with the tunnel conversion.
7008  * - Group id in rule attribute conversion depends on a rule type and
7009  *   group id value:
7010  *   ** non zero group attributes converted with the tunnel method
7011  *   ** zero group attribute in non-tunnel rule is converted using the
7012  *      standard method - there's only one root table
7013  *   ** zero group attribute in steer tunnel rule is converted with the
7014  *      standard method - single root table
7015  *   ** zero group attribute in match tunnel rule is a special OvS
7016  *      case: that value is used for portability reasons. That group
7017  *      id is converted with the tunnel conversion method.
7018  *
7019  * @param[in] dev
7020  *   Port device
7021  * @param[in] tunnel
7022  *   PMD tunnel offload object
7023  * @param[in] group
7024  *   rte_flow group index value.
7025  * @param[out] table
7026  *   HW table value.
7027  * @param[in] grp_info
7028  *   flags used for conversion
7029  * @param[out] error
7030  *   Pointer to error structure.
7031  *
7032  * @return
7033  *   0 on success, a negative errno value otherwise and rte_errno is set.
7034  */
7035 int
7036 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7037 			 const struct mlx5_flow_tunnel *tunnel,
7038 			 uint32_t group, uint32_t *table,
7039 			 const struct flow_grp_info *grp_info,
7040 			 struct rte_flow_error *error)
7041 {
7042 	int ret;
7043 	bool standard_translation;
7044 
7045 	if (!grp_info->skip_scale && grp_info->external &&
7046 	    group < MLX5_MAX_TABLES_EXTERNAL)
7047 		group *= MLX5_FLOW_TABLE_FACTOR;
7048 	if (is_tunnel_offload_active(dev)) {
7049 		standard_translation = !grp_info->external ||
7050 					grp_info->std_tbl_fix;
7051 	} else {
7052 		standard_translation = true;
7053 	}
7054 	DRV_LOG(DEBUG,
7055 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7056 		dev->data->port_id, group, grp_info->transfer,
7057 		grp_info->external, grp_info->fdb_def_rule,
7058 		standard_translation ? "STANDARD" : "TUNNEL");
7059 	if (standard_translation)
7060 		ret = flow_group_to_table(dev->data->port_id, group, table,
7061 					  grp_info, error);
7062 	else
7063 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7064 						      table, error);
7065 
7066 	return ret;
7067 }
7068 
7069 /**
7070  * Discover availability of metadata reg_c's.
7071  *
7072  * Iteratively use test flows to check availability.
7073  *
7074  * @param[in] dev
7075  *   Pointer to the Ethernet device structure.
7076  *
7077  * @return
7078  *   0 on success, a negative errno value otherwise and rte_errno is set.
7079  */
7080 int
7081 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7082 {
7083 	struct mlx5_priv *priv = dev->data->dev_private;
7084 	struct mlx5_dev_config *config = &priv->config;
7085 	enum modify_reg idx;
7086 	int n = 0;
7087 
7088 	/* reg_c[0] and reg_c[1] are reserved. */
7089 	config->flow_mreg_c[n++] = REG_C_0;
7090 	config->flow_mreg_c[n++] = REG_C_1;
7091 	/* Discover availability of other reg_c's. */
7092 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7093 		struct rte_flow_attr attr = {
7094 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7095 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7096 			.ingress = 1,
7097 		};
7098 		struct rte_flow_item items[] = {
7099 			[0] = {
7100 				.type = RTE_FLOW_ITEM_TYPE_END,
7101 			},
7102 		};
7103 		struct rte_flow_action actions[] = {
7104 			[0] = {
7105 				.type = (enum rte_flow_action_type)
7106 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7107 				.conf = &(struct mlx5_flow_action_copy_mreg){
7108 					.src = REG_C_1,
7109 					.dst = idx,
7110 				},
7111 			},
7112 			[1] = {
7113 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7114 				.conf = &(struct rte_flow_action_jump){
7115 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7116 				},
7117 			},
7118 			[2] = {
7119 				.type = RTE_FLOW_ACTION_TYPE_END,
7120 			},
7121 		};
7122 		uint32_t flow_idx;
7123 		struct rte_flow *flow;
7124 		struct rte_flow_error error;
7125 
7126 		if (!config->dv_flow_en)
7127 			break;
7128 		/* Create internal flow, validation skips copy action. */
7129 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7130 					    actions, false, &error);
7131 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7132 				      flow_idx);
7133 		if (!flow)
7134 			continue;
7135 		config->flow_mreg_c[n++] = idx;
7136 		flow_list_destroy(dev, NULL, flow_idx);
7137 	}
7138 	for (; n < MLX5_MREG_C_NUM; ++n)
7139 		config->flow_mreg_c[n] = REG_NON;
7140 	return 0;
7141 }
7142 
7143 /**
7144  * Dump flow raw hw data to file
7145  *
7146  * @param[in] dev
7147  *    The pointer to Ethernet device.
7148  * @param[in] file
7149  *   A pointer to a file for output.
7150  * @param[out] error
7151  *   Perform verbose error reporting if not NULL. PMDs initialize this
7152  *   structure in case of error only.
7153  * @return
7154  *   0 on success, a nagative value otherwise.
7155  */
7156 int
7157 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
7158 		   FILE *file,
7159 		   struct rte_flow_error *error __rte_unused)
7160 {
7161 	struct mlx5_priv *priv = dev->data->dev_private;
7162 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7163 
7164 	if (!priv->config.dv_flow_en) {
7165 		if (fputs("device dv flow disabled\n", file) <= 0)
7166 			return -errno;
7167 		return -ENOTSUP;
7168 	}
7169 
7170 	if (!flow_idx)
7171 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
7172 				sh->rx_domain, sh->tx_domain, file);
7173 	return -ENOTSUP;
7174 }
7175 
7176 /**
7177  * Get aged-out flows.
7178  *
7179  * @param[in] dev
7180  *   Pointer to the Ethernet device structure.
7181  * @param[in] context
7182  *   The address of an array of pointers to the aged-out flows contexts.
7183  * @param[in] nb_countexts
7184  *   The length of context array pointers.
7185  * @param[out] error
7186  *   Perform verbose error reporting if not NULL. Initialized in case of
7187  *   error only.
7188  *
7189  * @return
7190  *   how many contexts get in success, otherwise negative errno value.
7191  *   if nb_contexts is 0, return the amount of all aged contexts.
7192  *   if nb_contexts is not 0 , return the amount of aged flows reported
7193  *   in the context array.
7194  */
7195 int
7196 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7197 			uint32_t nb_contexts, struct rte_flow_error *error)
7198 {
7199 	const struct mlx5_flow_driver_ops *fops;
7200 	struct rte_flow_attr attr = { .transfer = 0 };
7201 
7202 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7203 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7204 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7205 						    error);
7206 	}
7207 	DRV_LOG(ERR,
7208 		"port %u get aged flows is not supported.",
7209 		 dev->data->port_id);
7210 	return -ENOTSUP;
7211 }
7212 
7213 /* Wrapper for driver action_validate op callback */
7214 static int
7215 flow_drv_action_validate(struct rte_eth_dev *dev,
7216 			 const struct rte_flow_indir_action_conf *conf,
7217 			 const struct rte_flow_action *action,
7218 			 const struct mlx5_flow_driver_ops *fops,
7219 			 struct rte_flow_error *error)
7220 {
7221 	static const char err_msg[] = "indirect action validation unsupported";
7222 
7223 	if (!fops->action_validate) {
7224 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7225 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7226 				   NULL, err_msg);
7227 		return -rte_errno;
7228 	}
7229 	return fops->action_validate(dev, conf, action, error);
7230 }
7231 
7232 /**
7233  * Destroys the shared action by handle.
7234  *
7235  * @param dev
7236  *   Pointer to Ethernet device structure.
7237  * @param[in] handle
7238  *   Handle for the indirect action object to be destroyed.
7239  * @param[out] error
7240  *   Perform verbose error reporting if not NULL. PMDs initialize this
7241  *   structure in case of error only.
7242  *
7243  * @return
7244  *   0 on success, a negative errno value otherwise and rte_errno is set.
7245  *
7246  * @note: wrapper for driver action_create op callback.
7247  */
7248 static int
7249 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
7250 			   struct rte_flow_action_handle *handle,
7251 			   struct rte_flow_error *error)
7252 {
7253 	static const char err_msg[] = "indirect action destruction unsupported";
7254 	struct rte_flow_attr attr = { .transfer = 0 };
7255 	const struct mlx5_flow_driver_ops *fops =
7256 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7257 
7258 	if (!fops->action_destroy) {
7259 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7260 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7261 				   NULL, err_msg);
7262 		return -rte_errno;
7263 	}
7264 	return fops->action_destroy(dev, handle, error);
7265 }
7266 
7267 /* Wrapper for driver action_destroy op callback */
7268 static int
7269 flow_drv_action_update(struct rte_eth_dev *dev,
7270 		       struct rte_flow_action_handle *handle,
7271 		       const void *update,
7272 		       const struct mlx5_flow_driver_ops *fops,
7273 		       struct rte_flow_error *error)
7274 {
7275 	static const char err_msg[] = "indirect action update unsupported";
7276 
7277 	if (!fops->action_update) {
7278 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7279 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7280 				   NULL, err_msg);
7281 		return -rte_errno;
7282 	}
7283 	return fops->action_update(dev, handle, update, error);
7284 }
7285 
7286 /* Wrapper for driver action_destroy op callback */
7287 static int
7288 flow_drv_action_query(struct rte_eth_dev *dev,
7289 		      const struct rte_flow_action_handle *handle,
7290 		      void *data,
7291 		      const struct mlx5_flow_driver_ops *fops,
7292 		      struct rte_flow_error *error)
7293 {
7294 	static const char err_msg[] = "indirect action query unsupported";
7295 
7296 	if (!fops->action_query) {
7297 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7298 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7299 				   NULL, err_msg);
7300 		return -rte_errno;
7301 	}
7302 	return fops->action_query(dev, handle, data, error);
7303 }
7304 
7305 /**
7306  * Create indirect action for reuse in multiple flow rules.
7307  *
7308  * @param dev
7309  *   Pointer to Ethernet device structure.
7310  * @param conf
7311  *   Pointer to indirect action object configuration.
7312  * @param[in] action
7313  *   Action configuration for indirect action object creation.
7314  * @param[out] error
7315  *   Perform verbose error reporting if not NULL. PMDs initialize this
7316  *   structure in case of error only.
7317  * @return
7318  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7319  */
7320 static struct rte_flow_action_handle *
7321 mlx5_action_handle_create(struct rte_eth_dev *dev,
7322 			  const struct rte_flow_indir_action_conf *conf,
7323 			  const struct rte_flow_action *action,
7324 			  struct rte_flow_error *error)
7325 {
7326 	static const char err_msg[] = "indirect action creation unsupported";
7327 	struct rte_flow_attr attr = { .transfer = 0 };
7328 	const struct mlx5_flow_driver_ops *fops =
7329 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7330 
7331 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7332 		return NULL;
7333 	if (!fops->action_create) {
7334 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7335 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7336 				   NULL, err_msg);
7337 		return NULL;
7338 	}
7339 	return fops->action_create(dev, conf, action, error);
7340 }
7341 
7342 /**
7343  * Updates inplace the indirect action configuration pointed by *handle*
7344  * with the configuration provided as *update* argument.
7345  * The update of the indirect action configuration effects all flow rules
7346  * reusing the action via handle.
7347  *
7348  * @param dev
7349  *   Pointer to Ethernet device structure.
7350  * @param[in] handle
7351  *   Handle for the indirect action to be updated.
7352  * @param[in] update
7353  *   Action specification used to modify the action pointed by handle.
7354  *   *update* could be of same type with the action pointed by the *handle*
7355  *   handle argument, or some other structures like a wrapper, depending on
7356  *   the indirect action type.
7357  * @param[out] error
7358  *   Perform verbose error reporting if not NULL. PMDs initialize this
7359  *   structure in case of error only.
7360  *
7361  * @return
7362  *   0 on success, a negative errno value otherwise and rte_errno is set.
7363  */
7364 static int
7365 mlx5_action_handle_update(struct rte_eth_dev *dev,
7366 		struct rte_flow_action_handle *handle,
7367 		const void *update,
7368 		struct rte_flow_error *error)
7369 {
7370 	struct rte_flow_attr attr = { .transfer = 0 };
7371 	const struct mlx5_flow_driver_ops *fops =
7372 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7373 	int ret;
7374 
7375 	ret = flow_drv_action_validate(dev, NULL,
7376 			(const struct rte_flow_action *)update, fops, error);
7377 	if (ret)
7378 		return ret;
7379 	return flow_drv_action_update(dev, handle, update, fops,
7380 				      error);
7381 }
7382 
7383 /**
7384  * Query the indirect action by handle.
7385  *
7386  * This function allows retrieving action-specific data such as counters.
7387  * Data is gathered by special action which may be present/referenced in
7388  * more than one flow rule definition.
7389  *
7390  * see @RTE_FLOW_ACTION_TYPE_COUNT
7391  *
7392  * @param dev
7393  *   Pointer to Ethernet device structure.
7394  * @param[in] handle
7395  *   Handle for the indirect action to query.
7396  * @param[in, out] data
7397  *   Pointer to storage for the associated query data type.
7398  * @param[out] error
7399  *   Perform verbose error reporting if not NULL. PMDs initialize this
7400  *   structure in case of error only.
7401  *
7402  * @return
7403  *   0 on success, a negative errno value otherwise and rte_errno is set.
7404  */
7405 static int
7406 mlx5_action_handle_query(struct rte_eth_dev *dev,
7407 			 const struct rte_flow_action_handle *handle,
7408 			 void *data,
7409 			 struct rte_flow_error *error)
7410 {
7411 	struct rte_flow_attr attr = { .transfer = 0 };
7412 	const struct mlx5_flow_driver_ops *fops =
7413 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7414 
7415 	return flow_drv_action_query(dev, handle, data, fops, error);
7416 }
7417 
7418 /**
7419  * Destroy all indirect actions (shared RSS).
7420  *
7421  * @param dev
7422  *   Pointer to Ethernet device.
7423  *
7424  * @return
7425  *   0 on success, a negative errno value otherwise and rte_errno is set.
7426  */
7427 int
7428 mlx5_action_handle_flush(struct rte_eth_dev *dev)
7429 {
7430 	struct rte_flow_error error;
7431 	struct mlx5_priv *priv = dev->data->dev_private;
7432 	struct mlx5_shared_action_rss *shared_rss;
7433 	int ret = 0;
7434 	uint32_t idx;
7435 
7436 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7437 		      priv->rss_shared_actions, idx, shared_rss, next) {
7438 		ret |= mlx5_action_handle_destroy(dev,
7439 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
7440 	}
7441 	return ret;
7442 }
7443 
7444 #ifndef HAVE_MLX5DV_DR
7445 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7446 #else
7447 #define MLX5_DOMAIN_SYNC_FLOW \
7448 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7449 #endif
7450 
7451 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7452 {
7453 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7454 	const struct mlx5_flow_driver_ops *fops;
7455 	int ret;
7456 	struct rte_flow_attr attr = { .transfer = 0 };
7457 
7458 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7459 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7460 	if (ret > 0)
7461 		ret = -ret;
7462 	return ret;
7463 }
7464 
7465 /**
7466  * tunnel offload functionalilty is defined for DV environment only
7467  */
7468 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7469 __extension__
7470 union tunnel_offload_mark {
7471 	uint32_t val;
7472 	struct {
7473 		uint32_t app_reserve:8;
7474 		uint32_t table_id:15;
7475 		uint32_t transfer:1;
7476 		uint32_t _unused_:8;
7477 	};
7478 };
7479 
7480 static bool
7481 mlx5_access_tunnel_offload_db
7482 	(struct rte_eth_dev *dev,
7483 	 bool (*match)(struct rte_eth_dev *,
7484 		       struct mlx5_flow_tunnel *, const void *),
7485 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7486 	 void (*miss)(struct rte_eth_dev *, void *),
7487 	 void *ctx, bool lock_op);
7488 
7489 static int
7490 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7491 			     struct rte_flow *flow,
7492 			     const struct rte_flow_attr *attr,
7493 			     const struct rte_flow_action *app_actions,
7494 			     uint32_t flow_idx,
7495 			     struct tunnel_default_miss_ctx *ctx,
7496 			     struct rte_flow_error *error)
7497 {
7498 	struct mlx5_priv *priv = dev->data->dev_private;
7499 	struct mlx5_flow *dev_flow;
7500 	struct rte_flow_attr miss_attr = *attr;
7501 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7502 	const struct rte_flow_item miss_items[2] = {
7503 		{
7504 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7505 			.spec = NULL,
7506 			.last = NULL,
7507 			.mask = NULL
7508 		},
7509 		{
7510 			.type = RTE_FLOW_ITEM_TYPE_END,
7511 			.spec = NULL,
7512 			.last = NULL,
7513 			.mask = NULL
7514 		}
7515 	};
7516 	union tunnel_offload_mark mark_id;
7517 	struct rte_flow_action_mark miss_mark;
7518 	struct rte_flow_action miss_actions[3] = {
7519 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7520 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7521 	};
7522 	const struct rte_flow_action_jump *jump_data;
7523 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7524 	struct flow_grp_info grp_info = {
7525 		.external = 1,
7526 		.transfer = attr->transfer,
7527 		.fdb_def_rule = !!priv->fdb_def_rule,
7528 		.std_tbl_fix = 0,
7529 	};
7530 	int ret;
7531 
7532 	if (!attr->transfer) {
7533 		uint32_t q_size;
7534 
7535 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7536 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7537 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7538 					 0, SOCKET_ID_ANY);
7539 		if (!ctx->queue)
7540 			return rte_flow_error_set
7541 				(error, ENOMEM,
7542 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7543 				NULL, "invalid default miss RSS");
7544 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7545 		ctx->action_rss.level = 0,
7546 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7547 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7548 		ctx->action_rss.queue_num = priv->reta_idx_n,
7549 		ctx->action_rss.key = priv->rss_conf.rss_key,
7550 		ctx->action_rss.queue = ctx->queue;
7551 		if (!priv->reta_idx_n || !priv->rxqs_n)
7552 			return rte_flow_error_set
7553 				(error, EINVAL,
7554 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7555 				NULL, "invalid port configuration");
7556 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7557 			ctx->action_rss.types = 0;
7558 		for (i = 0; i != priv->reta_idx_n; ++i)
7559 			ctx->queue[i] = (*priv->reta_idx)[i];
7560 	} else {
7561 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7562 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7563 	}
7564 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7565 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7566 	jump_data = app_actions->conf;
7567 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7568 	miss_attr.group = jump_data->group;
7569 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7570 				       &flow_table, &grp_info, error);
7571 	if (ret)
7572 		return rte_flow_error_set(error, EINVAL,
7573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7574 					  NULL, "invalid tunnel id");
7575 	mark_id.app_reserve = 0;
7576 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7577 	mark_id.transfer = !!attr->transfer;
7578 	mark_id._unused_ = 0;
7579 	miss_mark.id = mark_id.val;
7580 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7581 				    miss_items, miss_actions, flow_idx, error);
7582 	if (!dev_flow)
7583 		return -rte_errno;
7584 	dev_flow->flow = flow;
7585 	dev_flow->external = true;
7586 	dev_flow->tunnel = tunnel;
7587 	/* Subflow object was created, we must include one in the list. */
7588 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7589 		      dev_flow->handle, next);
7590 	DRV_LOG(DEBUG,
7591 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7592 		dev->data->port_id, tunnel->app_tunnel.type,
7593 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7594 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7595 				  miss_actions, error);
7596 	if (!ret)
7597 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7598 						  error);
7599 
7600 	return ret;
7601 }
7602 
7603 static const struct mlx5_flow_tbl_data_entry  *
7604 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7605 {
7606 	struct mlx5_priv *priv = dev->data->dev_private;
7607 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7608 	struct mlx5_hlist_entry *he;
7609 	union tunnel_offload_mark mbits = { .val = mark };
7610 	union mlx5_flow_tbl_key table_key = {
7611 		{
7612 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7613 			.dummy = 0,
7614 			.domain = !!mbits.transfer,
7615 			.direction = 0,
7616 		}
7617 	};
7618 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7619 	return he ?
7620 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7621 }
7622 
7623 static void
7624 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7625 				   struct mlx5_hlist_entry *entry)
7626 {
7627 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7628 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7629 
7630 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7631 			tunnel_flow_tbl_to_id(tte->flow_table));
7632 	mlx5_free(tte);
7633 }
7634 
7635 static int
7636 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7637 				  struct mlx5_hlist_entry *entry,
7638 				  uint64_t key, void *cb_ctx __rte_unused)
7639 {
7640 	union tunnel_tbl_key tbl = {
7641 		.val = key,
7642 	};
7643 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7644 
7645 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7646 }
7647 
7648 static struct mlx5_hlist_entry *
7649 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7650 				   void *ctx __rte_unused)
7651 {
7652 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7653 	struct tunnel_tbl_entry *tte;
7654 	union tunnel_tbl_key tbl = {
7655 		.val = key,
7656 	};
7657 
7658 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7659 			  sizeof(*tte), 0,
7660 			  SOCKET_ID_ANY);
7661 	if (!tte)
7662 		goto err;
7663 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7664 			  &tte->flow_table);
7665 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7666 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7667 			tte->flow_table);
7668 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7669 				tte->flow_table);
7670 		goto err;
7671 	} else if (!tte->flow_table) {
7672 		goto err;
7673 	}
7674 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7675 	tte->tunnel_id = tbl.tunnel_id;
7676 	tte->group = tbl.group;
7677 	return &tte->hash;
7678 err:
7679 	if (tte)
7680 		mlx5_free(tte);
7681 	return NULL;
7682 }
7683 
7684 static uint32_t
7685 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7686 				const struct mlx5_flow_tunnel *tunnel,
7687 				uint32_t group, uint32_t *table,
7688 				struct rte_flow_error *error)
7689 {
7690 	struct mlx5_hlist_entry *he;
7691 	struct tunnel_tbl_entry *tte;
7692 	union tunnel_tbl_key key = {
7693 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7694 		.group = group
7695 	};
7696 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7697 	struct mlx5_hlist *group_hash;
7698 
7699 	group_hash = tunnel ? tunnel->groups : thub->groups;
7700 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7701 	if (!he)
7702 		return rte_flow_error_set(error, EINVAL,
7703 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7704 					  NULL,
7705 					  "tunnel group index not supported");
7706 	tte = container_of(he, typeof(*tte), hash);
7707 	*table = tte->flow_table;
7708 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7709 		dev->data->port_id, key.tunnel_id, group, *table);
7710 	return 0;
7711 }
7712 
7713 static void
7714 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7715 		      struct mlx5_flow_tunnel *tunnel)
7716 {
7717 	struct mlx5_priv *priv = dev->data->dev_private;
7718 	struct mlx5_indexed_pool *ipool;
7719 
7720 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7721 		dev->data->port_id, tunnel->tunnel_id);
7722 	LIST_REMOVE(tunnel, chain);
7723 	mlx5_hlist_destroy(tunnel->groups);
7724 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7725 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7726 }
7727 
7728 static bool
7729 mlx5_access_tunnel_offload_db
7730 	(struct rte_eth_dev *dev,
7731 	 bool (*match)(struct rte_eth_dev *,
7732 		       struct mlx5_flow_tunnel *, const void *),
7733 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7734 	 void (*miss)(struct rte_eth_dev *, void *),
7735 	 void *ctx, bool lock_op)
7736 {
7737 	bool verdict = false;
7738 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7739 	struct mlx5_flow_tunnel *tunnel;
7740 
7741 	rte_spinlock_lock(&thub->sl);
7742 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7743 		verdict = match(dev, tunnel, (const void *)ctx);
7744 		if (verdict)
7745 			break;
7746 	}
7747 	if (!lock_op)
7748 		rte_spinlock_unlock(&thub->sl);
7749 	if (verdict && hit)
7750 		hit(dev, tunnel, ctx);
7751 	if (!verdict && miss)
7752 		miss(dev, ctx);
7753 	if (lock_op)
7754 		rte_spinlock_unlock(&thub->sl);
7755 
7756 	return verdict;
7757 }
7758 
7759 struct tunnel_db_find_tunnel_id_ctx {
7760 	uint32_t tunnel_id;
7761 	struct mlx5_flow_tunnel *tunnel;
7762 };
7763 
7764 static bool
7765 find_tunnel_id_match(struct rte_eth_dev *dev,
7766 		     struct mlx5_flow_tunnel *tunnel, const void *x)
7767 {
7768 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7769 
7770 	RTE_SET_USED(dev);
7771 	return tunnel->tunnel_id == ctx->tunnel_id;
7772 }
7773 
7774 static void
7775 find_tunnel_id_hit(struct rte_eth_dev *dev,
7776 		   struct mlx5_flow_tunnel *tunnel, void *x)
7777 {
7778 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7779 	RTE_SET_USED(dev);
7780 	ctx->tunnel = tunnel;
7781 }
7782 
7783 static struct mlx5_flow_tunnel *
7784 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7785 {
7786 	struct tunnel_db_find_tunnel_id_ctx ctx = {
7787 		.tunnel_id = id,
7788 	};
7789 
7790 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7791 				      find_tunnel_id_hit, NULL, &ctx, true);
7792 
7793 	return ctx.tunnel;
7794 }
7795 
7796 static struct mlx5_flow_tunnel *
7797 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7798 			  const struct rte_flow_tunnel *app_tunnel)
7799 {
7800 	struct mlx5_priv *priv = dev->data->dev_private;
7801 	struct mlx5_indexed_pool *ipool;
7802 	struct mlx5_flow_tunnel *tunnel;
7803 	uint32_t id;
7804 
7805 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7806 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7807 	if (!tunnel)
7808 		return NULL;
7809 	if (id >= MLX5_MAX_TUNNELS) {
7810 		mlx5_ipool_free(ipool, id);
7811 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7812 		return NULL;
7813 	}
7814 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7815 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7816 					   mlx5_flow_tunnel_grp2tbl_match_cb,
7817 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7818 	if (!tunnel->groups) {
7819 		mlx5_ipool_free(ipool, id);
7820 		return NULL;
7821 	}
7822 	tunnel->groups->ctx = priv->sh;
7823 	/* initiate new PMD tunnel */
7824 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7825 	tunnel->tunnel_id = id;
7826 	tunnel->action.type = (typeof(tunnel->action.type))
7827 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7828 	tunnel->action.conf = tunnel;
7829 	tunnel->item.type = (typeof(tunnel->item.type))
7830 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7831 	tunnel->item.spec = tunnel;
7832 	tunnel->item.last = NULL;
7833 	tunnel->item.mask = NULL;
7834 
7835 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7836 		dev->data->port_id, tunnel->tunnel_id);
7837 
7838 	return tunnel;
7839 }
7840 
7841 struct tunnel_db_get_tunnel_ctx {
7842 	const struct rte_flow_tunnel *app_tunnel;
7843 	struct mlx5_flow_tunnel *tunnel;
7844 };
7845 
7846 static bool get_tunnel_match(struct rte_eth_dev *dev,
7847 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7848 {
7849 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
7850 
7851 	RTE_SET_USED(dev);
7852 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7853 		       sizeof(*ctx->app_tunnel));
7854 }
7855 
7856 static void get_tunnel_hit(struct rte_eth_dev *dev,
7857 			   struct mlx5_flow_tunnel *tunnel, void *x)
7858 {
7859 	/* called under tunnel spinlock protection */
7860 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7861 
7862 	RTE_SET_USED(dev);
7863 	tunnel->refctn++;
7864 	ctx->tunnel = tunnel;
7865 }
7866 
7867 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7868 {
7869 	/* called under tunnel spinlock protection */
7870 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7871 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7872 
7873 	rte_spinlock_unlock(&thub->sl);
7874 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7875 	rte_spinlock_lock(&thub->sl);
7876 	if (ctx->tunnel) {
7877 		ctx->tunnel->refctn = 1;
7878 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7879 	}
7880 }
7881 
7882 
7883 static int
7884 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7885 		     const struct rte_flow_tunnel *app_tunnel,
7886 		     struct mlx5_flow_tunnel **tunnel)
7887 {
7888 	struct tunnel_db_get_tunnel_ctx ctx = {
7889 		.app_tunnel = app_tunnel,
7890 	};
7891 
7892 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7893 				      get_tunnel_miss, &ctx, true);
7894 	*tunnel = ctx.tunnel;
7895 	return ctx.tunnel ? 0 : -ENOMEM;
7896 }
7897 
7898 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7899 {
7900 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7901 
7902 	if (!thub)
7903 		return;
7904 	if (!LIST_EMPTY(&thub->tunnels))
7905 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
7906 	mlx5_hlist_destroy(thub->groups);
7907 	mlx5_free(thub);
7908 }
7909 
7910 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7911 {
7912 	int err;
7913 	struct mlx5_flow_tunnel_hub *thub;
7914 
7915 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7916 			   0, SOCKET_ID_ANY);
7917 	if (!thub)
7918 		return -ENOMEM;
7919 	LIST_INIT(&thub->tunnels);
7920 	rte_spinlock_init(&thub->sl);
7921 	thub->groups = mlx5_hlist_create("flow groups",
7922 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
7923 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7924 					 mlx5_flow_tunnel_grp2tbl_match_cb,
7925 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7926 	if (!thub->groups) {
7927 		err = -rte_errno;
7928 		goto err;
7929 	}
7930 	thub->groups->ctx = sh;
7931 	sh->tunnel_hub = thub;
7932 
7933 	return 0;
7934 
7935 err:
7936 	if (thub->groups)
7937 		mlx5_hlist_destroy(thub->groups);
7938 	if (thub)
7939 		mlx5_free(thub);
7940 	return err;
7941 }
7942 
7943 static inline bool
7944 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7945 			  struct rte_flow_tunnel *tunnel,
7946 			  const char *err_msg)
7947 {
7948 	err_msg = NULL;
7949 	if (!is_tunnel_offload_active(dev)) {
7950 		err_msg = "tunnel offload was not activated";
7951 		goto out;
7952 	} else if (!tunnel) {
7953 		err_msg = "no application tunnel";
7954 		goto out;
7955 	}
7956 
7957 	switch (tunnel->type) {
7958 	default:
7959 		err_msg = "unsupported tunnel type";
7960 		goto out;
7961 	case RTE_FLOW_ITEM_TYPE_VXLAN:
7962 		break;
7963 	}
7964 
7965 out:
7966 	return !err_msg;
7967 }
7968 
7969 static int
7970 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7971 		    struct rte_flow_tunnel *app_tunnel,
7972 		    struct rte_flow_action **actions,
7973 		    uint32_t *num_of_actions,
7974 		    struct rte_flow_error *error)
7975 {
7976 	int ret;
7977 	struct mlx5_flow_tunnel *tunnel;
7978 	const char *err_msg = NULL;
7979 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7980 
7981 	if (!verdict)
7982 		return rte_flow_error_set(error, EINVAL,
7983 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7984 					  err_msg);
7985 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7986 	if (ret < 0) {
7987 		return rte_flow_error_set(error, ret,
7988 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7989 					  "failed to initialize pmd tunnel");
7990 	}
7991 	*actions = &tunnel->action;
7992 	*num_of_actions = 1;
7993 	return 0;
7994 }
7995 
7996 static int
7997 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7998 		       struct rte_flow_tunnel *app_tunnel,
7999 		       struct rte_flow_item **items,
8000 		       uint32_t *num_of_items,
8001 		       struct rte_flow_error *error)
8002 {
8003 	int ret;
8004 	struct mlx5_flow_tunnel *tunnel;
8005 	const char *err_msg = NULL;
8006 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8007 
8008 	if (!verdict)
8009 		return rte_flow_error_set(error, EINVAL,
8010 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8011 					  err_msg);
8012 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8013 	if (ret < 0) {
8014 		return rte_flow_error_set(error, ret,
8015 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8016 					  "failed to initialize pmd tunnel");
8017 	}
8018 	*items = &tunnel->item;
8019 	*num_of_items = 1;
8020 	return 0;
8021 }
8022 
8023 struct tunnel_db_element_release_ctx {
8024 	struct rte_flow_item *items;
8025 	struct rte_flow_action *actions;
8026 	uint32_t num_elements;
8027 	struct rte_flow_error *error;
8028 	int ret;
8029 };
8030 
8031 static bool
8032 tunnel_element_release_match(struct rte_eth_dev *dev,
8033 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8034 {
8035 	const struct tunnel_db_element_release_ctx *ctx = x;
8036 
8037 	RTE_SET_USED(dev);
8038 	if (ctx->num_elements != 1)
8039 		return false;
8040 	else if (ctx->items)
8041 		return ctx->items == &tunnel->item;
8042 	else if (ctx->actions)
8043 		return ctx->actions == &tunnel->action;
8044 
8045 	return false;
8046 }
8047 
8048 static void
8049 tunnel_element_release_hit(struct rte_eth_dev *dev,
8050 			   struct mlx5_flow_tunnel *tunnel, void *x)
8051 {
8052 	struct tunnel_db_element_release_ctx *ctx = x;
8053 	ctx->ret = 0;
8054 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8055 		mlx5_flow_tunnel_free(dev, tunnel);
8056 }
8057 
8058 static void
8059 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8060 {
8061 	struct tunnel_db_element_release_ctx *ctx = x;
8062 	RTE_SET_USED(dev);
8063 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8064 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8065 				      "invalid argument");
8066 }
8067 
8068 static int
8069 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8070 		       struct rte_flow_item *pmd_items,
8071 		       uint32_t num_items, struct rte_flow_error *err)
8072 {
8073 	struct tunnel_db_element_release_ctx ctx = {
8074 		.items = pmd_items,
8075 		.actions = NULL,
8076 		.num_elements = num_items,
8077 		.error = err,
8078 	};
8079 
8080 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8081 				      tunnel_element_release_hit,
8082 				      tunnel_element_release_miss, &ctx, false);
8083 
8084 	return ctx.ret;
8085 }
8086 
8087 static int
8088 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8089 			 struct rte_flow_action *pmd_actions,
8090 			 uint32_t num_actions, struct rte_flow_error *err)
8091 {
8092 	struct tunnel_db_element_release_ctx ctx = {
8093 		.items = NULL,
8094 		.actions = pmd_actions,
8095 		.num_elements = num_actions,
8096 		.error = err,
8097 	};
8098 
8099 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8100 				      tunnel_element_release_hit,
8101 				      tunnel_element_release_miss, &ctx, false);
8102 
8103 	return ctx.ret;
8104 }
8105 
8106 static int
8107 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8108 				  struct rte_mbuf *m,
8109 				  struct rte_flow_restore_info *info,
8110 				  struct rte_flow_error *err)
8111 {
8112 	uint64_t ol_flags = m->ol_flags;
8113 	const struct mlx5_flow_tbl_data_entry *tble;
8114 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8115 
8116 	if (!is_tunnel_offload_active(dev)) {
8117 		info->flags = 0;
8118 		return 0;
8119 	}
8120 
8121 	if ((ol_flags & mask) != mask)
8122 		goto err;
8123 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8124 	if (!tble) {
8125 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8126 			dev->data->port_id, m->hash.fdir.hi);
8127 		goto err;
8128 	}
8129 	MLX5_ASSERT(tble->tunnel);
8130 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8131 	info->group_id = tble->group_id;
8132 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8133 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8134 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8135 
8136 	return 0;
8137 
8138 err:
8139 	return rte_flow_error_set(err, EINVAL,
8140 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8141 				  "failed to get restore info");
8142 }
8143 
8144 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8145 static int
8146 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8147 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8148 			   __rte_unused struct rte_flow_action **actions,
8149 			   __rte_unused uint32_t *num_of_actions,
8150 			   __rte_unused struct rte_flow_error *error)
8151 {
8152 	return -ENOTSUP;
8153 }
8154 
8155 static int
8156 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8157 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8158 		       __rte_unused struct rte_flow_item **items,
8159 		       __rte_unused uint32_t *num_of_items,
8160 		       __rte_unused struct rte_flow_error *error)
8161 {
8162 	return -ENOTSUP;
8163 }
8164 
8165 static int
8166 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8167 			      __rte_unused struct rte_flow_item *pmd_items,
8168 			      __rte_unused uint32_t num_items,
8169 			      __rte_unused struct rte_flow_error *err)
8170 {
8171 	return -ENOTSUP;
8172 }
8173 
8174 static int
8175 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8176 				__rte_unused struct rte_flow_action *pmd_action,
8177 				__rte_unused uint32_t num_actions,
8178 				__rte_unused struct rte_flow_error *err)
8179 {
8180 	return -ENOTSUP;
8181 }
8182 
8183 static int
8184 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8185 				  __rte_unused struct rte_mbuf *m,
8186 				  __rte_unused struct rte_flow_restore_info *i,
8187 				  __rte_unused struct rte_flow_error *err)
8188 {
8189 	return -ENOTSUP;
8190 }
8191 
8192 static int
8193 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8194 			     __rte_unused struct rte_flow *flow,
8195 			     __rte_unused const struct rte_flow_attr *attr,
8196 			     __rte_unused const struct rte_flow_action *actions,
8197 			     __rte_unused uint32_t flow_idx,
8198 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8199 			     __rte_unused struct rte_flow_error *error)
8200 {
8201 	return -ENOTSUP;
8202 }
8203 
8204 static struct mlx5_flow_tunnel *
8205 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8206 		    __rte_unused uint32_t id)
8207 {
8208 	return NULL;
8209 }
8210 
8211 static void
8212 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8213 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8214 {
8215 }
8216 
8217 static uint32_t
8218 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8219 				__rte_unused const struct mlx5_flow_tunnel *t,
8220 				__rte_unused uint32_t group,
8221 				__rte_unused uint32_t *table,
8222 				struct rte_flow_error *error)
8223 {
8224 	return rte_flow_error_set(error, ENOTSUP,
8225 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8226 				  "tunnel offload requires DV support");
8227 }
8228 
8229 void
8230 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8231 			__rte_unused  uint16_t port_id)
8232 {
8233 }
8234 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8235