xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 9cac7ded37479ec59ca2f52c043a5e4326c24a22)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <rte_ethdev_driver.h>
16 #include <rte_eal_paging.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
21 #include <rte_ip.h>
22 
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
25 #include <mlx5_prm.h>
26 #include <mlx5_malloc.h>
27 
28 #include "mlx5_defs.h"
29 #include "mlx5.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 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 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
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 /**
216  * Expand RSS flows into several possible flows according to the RSS hash
217  * fields requested and the driver capabilities.
218  *
219  * @param[out] buf
220  *   Buffer to store the result expansion.
221  * @param[in] size
222  *   Buffer size in bytes. If 0, @p buf can be NULL.
223  * @param[in] pattern
224  *   User flow pattern.
225  * @param[in] types
226  *   RSS types to expand (see ETH_RSS_* definitions).
227  * @param[in] graph
228  *   Input graph to expand @p pattern according to @p types.
229  * @param[in] graph_root_index
230  *   Index of root node in @p graph, typically 0.
231  *
232  * @return
233  *   A positive value representing the size of @p buf in bytes regardless of
234  *   @p size on success, a negative errno value otherwise and rte_errno is
235  *   set, the following errors are defined:
236  *
237  *   -E2BIG: graph-depth @p graph is too deep.
238  */
239 static int
240 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
241 		     const struct rte_flow_item *pattern, uint64_t types,
242 		     const struct mlx5_flow_expand_node graph[],
243 		     int graph_root_index)
244 {
245 	const int elt_n = 8;
246 	const struct rte_flow_item *item;
247 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
248 	const int *next_node;
249 	const int *stack[elt_n];
250 	int stack_pos = 0;
251 	struct rte_flow_item flow_items[elt_n];
252 	unsigned int i;
253 	size_t lsize;
254 	size_t user_pattern_size = 0;
255 	void *addr = NULL;
256 	const struct mlx5_flow_expand_node *next = NULL;
257 	struct rte_flow_item missed_item;
258 	int missed = 0;
259 	int elt = 0;
260 	const struct rte_flow_item *last_item = NULL;
261 
262 	memset(&missed_item, 0, sizeof(missed_item));
263 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
264 		elt_n * sizeof(buf->entry[0]);
265 	if (lsize <= size) {
266 		buf->entry[0].priority = 0;
267 		buf->entry[0].pattern = (void *)&buf->entry[elt_n];
268 		buf->entries = 0;
269 		addr = buf->entry[0].pattern;
270 	}
271 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
272 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
273 			last_item = item;
274 		for (i = 0; node->next && node->next[i]; ++i) {
275 			next = &graph[node->next[i]];
276 			if (next->type == item->type)
277 				break;
278 		}
279 		if (next)
280 			node = next;
281 		user_pattern_size += sizeof(*item);
282 	}
283 	user_pattern_size += sizeof(*item); /* Handle END item. */
284 	lsize += user_pattern_size;
285 	/* Copy the user pattern in the first entry of the buffer. */
286 	if (lsize <= size) {
287 		rte_memcpy(addr, pattern, user_pattern_size);
288 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
289 		buf->entries = 1;
290 	}
291 	/* Start expanding. */
292 	memset(flow_items, 0, sizeof(flow_items));
293 	user_pattern_size -= sizeof(*item);
294 	/*
295 	 * Check if the last valid item has spec set, need complete pattern,
296 	 * and the pattern can be used for expansion.
297 	 */
298 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
299 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
300 		/* Item type END indicates expansion is not required. */
301 		return lsize;
302 	}
303 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
304 		next = NULL;
305 		missed = 1;
306 		for (i = 0; node->next && node->next[i]; ++i) {
307 			next = &graph[node->next[i]];
308 			if (next->type == missed_item.type) {
309 				flow_items[0].type = missed_item.type;
310 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
311 				break;
312 			}
313 			next = NULL;
314 		}
315 	}
316 	if (next && missed) {
317 		elt = 2; /* missed item + item end. */
318 		node = next;
319 		lsize += elt * sizeof(*item) + user_pattern_size;
320 		if ((node->rss_types & types) && lsize <= size) {
321 			buf->entry[buf->entries].priority = 1;
322 			buf->entry[buf->entries].pattern = addr;
323 			buf->entries++;
324 			rte_memcpy(addr, buf->entry[0].pattern,
325 				   user_pattern_size);
326 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
327 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
328 			addr = (void *)(((uintptr_t)addr) +
329 					elt * sizeof(*item));
330 		}
331 	}
332 	memset(flow_items, 0, sizeof(flow_items));
333 	next_node = node->next;
334 	stack[stack_pos] = next_node;
335 	node = next_node ? &graph[*next_node] : NULL;
336 	while (node) {
337 		flow_items[stack_pos].type = node->type;
338 		if (node->rss_types & types) {
339 			/*
340 			 * compute the number of items to copy from the
341 			 * expansion and copy it.
342 			 * When the stack_pos is 0, there are 1 element in it,
343 			 * plus the addition END item.
344 			 */
345 			elt = stack_pos + 2;
346 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
347 			lsize += elt * sizeof(*item) + user_pattern_size;
348 			if (lsize <= size) {
349 				size_t n = elt * sizeof(*item);
350 
351 				buf->entry[buf->entries].priority =
352 					stack_pos + 1 + missed;
353 				buf->entry[buf->entries].pattern = addr;
354 				buf->entries++;
355 				rte_memcpy(addr, buf->entry[0].pattern,
356 					   user_pattern_size);
357 				addr = (void *)(((uintptr_t)addr) +
358 						user_pattern_size);
359 				rte_memcpy(addr, &missed_item,
360 					   missed * sizeof(*item));
361 				addr = (void *)(((uintptr_t)addr) +
362 					missed * sizeof(*item));
363 				rte_memcpy(addr, flow_items, n);
364 				addr = (void *)(((uintptr_t)addr) + n);
365 			}
366 		}
367 		/* Go deeper. */
368 		if (node->next) {
369 			next_node = node->next;
370 			if (stack_pos++ == elt_n) {
371 				rte_errno = E2BIG;
372 				return -rte_errno;
373 			}
374 			stack[stack_pos] = next_node;
375 		} else if (*(next_node + 1)) {
376 			/* Follow up with the next possibility. */
377 			++next_node;
378 		} else {
379 			/* Move to the next path. */
380 			if (stack_pos)
381 				next_node = stack[--stack_pos];
382 			next_node++;
383 			stack[stack_pos] = next_node;
384 		}
385 		node = *next_node ? &graph[*next_node] : NULL;
386 	};
387 	/* no expanded flows but we have missed item, create one rule for it */
388 	if (buf->entries == 1 && missed != 0) {
389 		elt = 2;
390 		lsize += elt * sizeof(*item) + user_pattern_size;
391 		if (lsize <= size) {
392 			buf->entry[buf->entries].priority = 1;
393 			buf->entry[buf->entries].pattern = addr;
394 			buf->entries++;
395 			flow_items[0].type = missed_item.type;
396 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
397 			rte_memcpy(addr, buf->entry[0].pattern,
398 				   user_pattern_size);
399 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
400 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
401 		}
402 	}
403 	return lsize;
404 }
405 
406 enum mlx5_expansion {
407 	MLX5_EXPANSION_ROOT,
408 	MLX5_EXPANSION_ROOT_OUTER,
409 	MLX5_EXPANSION_ROOT_ETH_VLAN,
410 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
411 	MLX5_EXPANSION_OUTER_ETH,
412 	MLX5_EXPANSION_OUTER_ETH_VLAN,
413 	MLX5_EXPANSION_OUTER_VLAN,
414 	MLX5_EXPANSION_OUTER_IPV4,
415 	MLX5_EXPANSION_OUTER_IPV4_UDP,
416 	MLX5_EXPANSION_OUTER_IPV4_TCP,
417 	MLX5_EXPANSION_OUTER_IPV6,
418 	MLX5_EXPANSION_OUTER_IPV6_UDP,
419 	MLX5_EXPANSION_OUTER_IPV6_TCP,
420 	MLX5_EXPANSION_VXLAN,
421 	MLX5_EXPANSION_VXLAN_GPE,
422 	MLX5_EXPANSION_GRE,
423 	MLX5_EXPANSION_MPLS,
424 	MLX5_EXPANSION_ETH,
425 	MLX5_EXPANSION_ETH_VLAN,
426 	MLX5_EXPANSION_VLAN,
427 	MLX5_EXPANSION_IPV4,
428 	MLX5_EXPANSION_IPV4_UDP,
429 	MLX5_EXPANSION_IPV4_TCP,
430 	MLX5_EXPANSION_IPV6,
431 	MLX5_EXPANSION_IPV6_UDP,
432 	MLX5_EXPANSION_IPV6_TCP,
433 };
434 
435 /** Supported expansion of items. */
436 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
437 	[MLX5_EXPANSION_ROOT] = {
438 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
439 						  MLX5_EXPANSION_IPV4,
440 						  MLX5_EXPANSION_IPV6),
441 		.type = RTE_FLOW_ITEM_TYPE_END,
442 	},
443 	[MLX5_EXPANSION_ROOT_OUTER] = {
444 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
445 						  MLX5_EXPANSION_OUTER_IPV4,
446 						  MLX5_EXPANSION_OUTER_IPV6),
447 		.type = RTE_FLOW_ITEM_TYPE_END,
448 	},
449 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
450 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
451 		.type = RTE_FLOW_ITEM_TYPE_END,
452 	},
453 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
454 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
455 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
456 		.type = RTE_FLOW_ITEM_TYPE_END,
457 	},
458 	[MLX5_EXPANSION_OUTER_ETH] = {
459 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
460 						  MLX5_EXPANSION_OUTER_IPV6,
461 						  MLX5_EXPANSION_MPLS),
462 		.type = RTE_FLOW_ITEM_TYPE_ETH,
463 		.rss_types = 0,
464 	},
465 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
466 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
467 		.type = RTE_FLOW_ITEM_TYPE_ETH,
468 		.rss_types = 0,
469 	},
470 	[MLX5_EXPANSION_OUTER_VLAN] = {
471 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
472 						  MLX5_EXPANSION_OUTER_IPV6),
473 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
474 	},
475 	[MLX5_EXPANSION_OUTER_IPV4] = {
476 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
477 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
478 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
479 			 MLX5_EXPANSION_GRE,
480 			 MLX5_EXPANSION_IPV4,
481 			 MLX5_EXPANSION_IPV6),
482 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
483 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
484 			ETH_RSS_NONFRAG_IPV4_OTHER,
485 	},
486 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
487 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
488 						  MLX5_EXPANSION_VXLAN_GPE),
489 		.type = RTE_FLOW_ITEM_TYPE_UDP,
490 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
491 	},
492 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
493 		.type = RTE_FLOW_ITEM_TYPE_TCP,
494 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
495 	},
496 	[MLX5_EXPANSION_OUTER_IPV6] = {
497 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
498 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
499 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
500 			 MLX5_EXPANSION_IPV4,
501 			 MLX5_EXPANSION_IPV6),
502 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
503 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
504 			ETH_RSS_NONFRAG_IPV6_OTHER,
505 	},
506 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
507 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
508 						  MLX5_EXPANSION_VXLAN_GPE),
509 		.type = RTE_FLOW_ITEM_TYPE_UDP,
510 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
511 	},
512 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
513 		.type = RTE_FLOW_ITEM_TYPE_TCP,
514 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
515 	},
516 	[MLX5_EXPANSION_VXLAN] = {
517 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
518 						  MLX5_EXPANSION_IPV4,
519 						  MLX5_EXPANSION_IPV6),
520 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
521 	},
522 	[MLX5_EXPANSION_VXLAN_GPE] = {
523 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
524 						  MLX5_EXPANSION_IPV4,
525 						  MLX5_EXPANSION_IPV6),
526 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
527 	},
528 	[MLX5_EXPANSION_GRE] = {
529 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
530 		.type = RTE_FLOW_ITEM_TYPE_GRE,
531 	},
532 	[MLX5_EXPANSION_MPLS] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
534 						  MLX5_EXPANSION_IPV6),
535 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
536 	},
537 	[MLX5_EXPANSION_ETH] = {
538 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
539 						  MLX5_EXPANSION_IPV6),
540 		.type = RTE_FLOW_ITEM_TYPE_ETH,
541 	},
542 	[MLX5_EXPANSION_ETH_VLAN] = {
543 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
544 		.type = RTE_FLOW_ITEM_TYPE_ETH,
545 	},
546 	[MLX5_EXPANSION_VLAN] = {
547 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
548 						  MLX5_EXPANSION_IPV6),
549 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
550 	},
551 	[MLX5_EXPANSION_IPV4] = {
552 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
553 						  MLX5_EXPANSION_IPV4_TCP),
554 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
555 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
556 			ETH_RSS_NONFRAG_IPV4_OTHER,
557 	},
558 	[MLX5_EXPANSION_IPV4_UDP] = {
559 		.type = RTE_FLOW_ITEM_TYPE_UDP,
560 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
561 	},
562 	[MLX5_EXPANSION_IPV4_TCP] = {
563 		.type = RTE_FLOW_ITEM_TYPE_TCP,
564 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
565 	},
566 	[MLX5_EXPANSION_IPV6] = {
567 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
568 						  MLX5_EXPANSION_IPV6_TCP),
569 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
570 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
571 			ETH_RSS_NONFRAG_IPV6_OTHER,
572 	},
573 	[MLX5_EXPANSION_IPV6_UDP] = {
574 		.type = RTE_FLOW_ITEM_TYPE_UDP,
575 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
576 	},
577 	[MLX5_EXPANSION_IPV6_TCP] = {
578 		.type = RTE_FLOW_ITEM_TYPE_TCP,
579 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
580 	},
581 };
582 
583 static struct rte_flow_shared_action *
584 mlx5_shared_action_create(struct rte_eth_dev *dev,
585 			  const struct rte_flow_shared_action_conf *conf,
586 			  const struct rte_flow_action *action,
587 			  struct rte_flow_error *error);
588 static int mlx5_shared_action_destroy
589 				(struct rte_eth_dev *dev,
590 				 struct rte_flow_shared_action *shared_action,
591 				 struct rte_flow_error *error);
592 static int mlx5_shared_action_update
593 				(struct rte_eth_dev *dev,
594 				 struct rte_flow_shared_action *shared_action,
595 				 const struct rte_flow_action *action,
596 				 struct rte_flow_error *error);
597 static int mlx5_shared_action_query
598 				(struct rte_eth_dev *dev,
599 				 const struct rte_flow_shared_action *action,
600 				 void *data,
601 				 struct rte_flow_error *error);
602 static int
603 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
604 		    struct rte_flow_tunnel *app_tunnel,
605 		    struct rte_flow_action **actions,
606 		    uint32_t *num_of_actions,
607 		    struct rte_flow_error *error);
608 static int
609 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
610 		       struct rte_flow_tunnel *app_tunnel,
611 		       struct rte_flow_item **items,
612 		       uint32_t *num_of_items,
613 		       struct rte_flow_error *error);
614 static int
615 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
616 			      struct rte_flow_item *pmd_items,
617 			      uint32_t num_items, struct rte_flow_error *err);
618 static int
619 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
620 				struct rte_flow_action *pmd_actions,
621 				uint32_t num_actions,
622 				struct rte_flow_error *err);
623 static int
624 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
625 				  struct rte_mbuf *m,
626 				  struct rte_flow_restore_info *info,
627 				  struct rte_flow_error *err);
628 
629 static const struct rte_flow_ops mlx5_flow_ops = {
630 	.validate = mlx5_flow_validate,
631 	.create = mlx5_flow_create,
632 	.destroy = mlx5_flow_destroy,
633 	.flush = mlx5_flow_flush,
634 	.isolate = mlx5_flow_isolate,
635 	.query = mlx5_flow_query,
636 	.dev_dump = mlx5_flow_dev_dump,
637 	.get_aged_flows = mlx5_flow_get_aged_flows,
638 	.shared_action_create = mlx5_shared_action_create,
639 	.shared_action_destroy = mlx5_shared_action_destroy,
640 	.shared_action_update = mlx5_shared_action_update,
641 	.shared_action_query = mlx5_shared_action_query,
642 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
643 	.tunnel_match = mlx5_flow_tunnel_match,
644 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
645 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
646 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
647 };
648 
649 /* Tunnel information. */
650 struct mlx5_flow_tunnel_info {
651 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
652 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
653 };
654 
655 static struct mlx5_flow_tunnel_info tunnels_info[] = {
656 	{
657 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
658 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
659 	},
660 	{
661 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
662 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
663 	},
664 	{
665 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
666 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
667 	},
668 	{
669 		.tunnel = MLX5_FLOW_LAYER_GRE,
670 		.ptype = RTE_PTYPE_TUNNEL_GRE,
671 	},
672 	{
673 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
674 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
675 	},
676 	{
677 		.tunnel = MLX5_FLOW_LAYER_MPLS,
678 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
679 	},
680 	{
681 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
682 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
683 	},
684 	{
685 		.tunnel = MLX5_FLOW_LAYER_IPIP,
686 		.ptype = RTE_PTYPE_TUNNEL_IP,
687 	},
688 	{
689 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
690 		.ptype = RTE_PTYPE_TUNNEL_IP,
691 	},
692 	{
693 		.tunnel = MLX5_FLOW_LAYER_GTP,
694 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
695 	},
696 };
697 
698 /* Key of thread specific flow workspace data. */
699 static pthread_key_t key_workspace;
700 
701 /* Thread specific flow workspace data once initialization data. */
702 static pthread_once_t key_workspace_init;
703 
704 
705 /**
706  * Translate tag ID to register.
707  *
708  * @param[in] dev
709  *   Pointer to the Ethernet device structure.
710  * @param[in] feature
711  *   The feature that request the register.
712  * @param[in] id
713  *   The request register ID.
714  * @param[out] error
715  *   Error description in case of any.
716  *
717  * @return
718  *   The request register on success, a negative errno
719  *   value otherwise and rte_errno is set.
720  */
721 int
722 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
723 		     enum mlx5_feature_name feature,
724 		     uint32_t id,
725 		     struct rte_flow_error *error)
726 {
727 	struct mlx5_priv *priv = dev->data->dev_private;
728 	struct mlx5_dev_config *config = &priv->config;
729 	enum modify_reg start_reg;
730 	bool skip_mtr_reg = false;
731 
732 	switch (feature) {
733 	case MLX5_HAIRPIN_RX:
734 		return REG_B;
735 	case MLX5_HAIRPIN_TX:
736 		return REG_A;
737 	case MLX5_METADATA_RX:
738 		switch (config->dv_xmeta_en) {
739 		case MLX5_XMETA_MODE_LEGACY:
740 			return REG_B;
741 		case MLX5_XMETA_MODE_META16:
742 			return REG_C_0;
743 		case MLX5_XMETA_MODE_META32:
744 			return REG_C_1;
745 		}
746 		break;
747 	case MLX5_METADATA_TX:
748 		return REG_A;
749 	case MLX5_METADATA_FDB:
750 		switch (config->dv_xmeta_en) {
751 		case MLX5_XMETA_MODE_LEGACY:
752 			return REG_NON;
753 		case MLX5_XMETA_MODE_META16:
754 			return REG_C_0;
755 		case MLX5_XMETA_MODE_META32:
756 			return REG_C_1;
757 		}
758 		break;
759 	case MLX5_FLOW_MARK:
760 		switch (config->dv_xmeta_en) {
761 		case MLX5_XMETA_MODE_LEGACY:
762 			return REG_NON;
763 		case MLX5_XMETA_MODE_META16:
764 			return REG_C_1;
765 		case MLX5_XMETA_MODE_META32:
766 			return REG_C_0;
767 		}
768 		break;
769 	case MLX5_MTR_SFX:
770 		/*
771 		 * If meter color and flow match share one register, flow match
772 		 * should use the meter color register for match.
773 		 */
774 		if (priv->mtr_reg_share)
775 			return priv->mtr_color_reg;
776 		else
777 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
778 			       REG_C_3;
779 	case MLX5_MTR_COLOR:
780 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
781 		return priv->mtr_color_reg;
782 	case MLX5_COPY_MARK:
783 		/*
784 		 * Metadata COPY_MARK register using is in meter suffix sub
785 		 * flow while with meter. It's safe to share the same register.
786 		 */
787 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
788 	case MLX5_APP_TAG:
789 		/*
790 		 * If meter is enable, it will engage the register for color
791 		 * match and flow match. If meter color match is not using the
792 		 * REG_C_2, need to skip the REG_C_x be used by meter color
793 		 * match.
794 		 * If meter is disable, free to use all available registers.
795 		 */
796 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
797 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
798 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
799 		if (id > (REG_C_7 - start_reg))
800 			return rte_flow_error_set(error, EINVAL,
801 						  RTE_FLOW_ERROR_TYPE_ITEM,
802 						  NULL, "invalid tag id");
803 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
804 			return rte_flow_error_set(error, ENOTSUP,
805 						  RTE_FLOW_ERROR_TYPE_ITEM,
806 						  NULL, "unsupported tag id");
807 		/*
808 		 * This case means meter is using the REG_C_x great than 2.
809 		 * Take care not to conflict with meter color REG_C_x.
810 		 * If the available index REG_C_y >= REG_C_x, skip the
811 		 * color register.
812 		 */
813 		if (skip_mtr_reg && config->flow_mreg_c
814 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
815 			if (id >= (REG_C_7 - start_reg))
816 				return rte_flow_error_set(error, EINVAL,
817 						       RTE_FLOW_ERROR_TYPE_ITEM,
818 							NULL, "invalid tag id");
819 			if (config->flow_mreg_c
820 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
821 				return config->flow_mreg_c
822 					       [id + 1 + start_reg - REG_C_0];
823 			return rte_flow_error_set(error, ENOTSUP,
824 						  RTE_FLOW_ERROR_TYPE_ITEM,
825 						  NULL, "unsupported tag id");
826 		}
827 		return config->flow_mreg_c[id + start_reg - REG_C_0];
828 	}
829 	MLX5_ASSERT(false);
830 	return rte_flow_error_set(error, EINVAL,
831 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
832 				  NULL, "invalid feature name");
833 }
834 
835 /**
836  * Check extensive flow metadata register support.
837  *
838  * @param dev
839  *   Pointer to rte_eth_dev structure.
840  *
841  * @return
842  *   True if device supports extensive flow metadata register, otherwise false.
843  */
844 bool
845 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
846 {
847 	struct mlx5_priv *priv = dev->data->dev_private;
848 	struct mlx5_dev_config *config = &priv->config;
849 
850 	/*
851 	 * Having available reg_c can be regarded inclusively as supporting
852 	 * extensive flow metadata register, which could mean,
853 	 * - metadata register copy action by modify header.
854 	 * - 16 modify header actions is supported.
855 	 * - reg_c's are preserved across different domain (FDB and NIC) on
856 	 *   packet loopback by flow lookup miss.
857 	 */
858 	return config->flow_mreg_c[2] != REG_NON;
859 }
860 
861 /**
862  * Verify the @p item specifications (spec, last, mask) are compatible with the
863  * NIC capabilities.
864  *
865  * @param[in] item
866  *   Item specification.
867  * @param[in] mask
868  *   @p item->mask or flow default bit-masks.
869  * @param[in] nic_mask
870  *   Bit-masks covering supported fields by the NIC to compare with user mask.
871  * @param[in] size
872  *   Bit-masks size in bytes.
873  * @param[in] range_accepted
874  *   True if range of values is accepted for specific fields, false otherwise.
875  * @param[out] error
876  *   Pointer to error structure.
877  *
878  * @return
879  *   0 on success, a negative errno value otherwise and rte_errno is set.
880  */
881 int
882 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
883 			  const uint8_t *mask,
884 			  const uint8_t *nic_mask,
885 			  unsigned int size,
886 			  bool range_accepted,
887 			  struct rte_flow_error *error)
888 {
889 	unsigned int i;
890 
891 	MLX5_ASSERT(nic_mask);
892 	for (i = 0; i < size; ++i)
893 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
894 			return rte_flow_error_set(error, ENOTSUP,
895 						  RTE_FLOW_ERROR_TYPE_ITEM,
896 						  item,
897 						  "mask enables non supported"
898 						  " bits");
899 	if (!item->spec && (item->mask || item->last))
900 		return rte_flow_error_set(error, EINVAL,
901 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
902 					  "mask/last without a spec is not"
903 					  " supported");
904 	if (item->spec && item->last && !range_accepted) {
905 		uint8_t spec[size];
906 		uint8_t last[size];
907 		unsigned int i;
908 		int ret;
909 
910 		for (i = 0; i < size; ++i) {
911 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
912 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
913 		}
914 		ret = memcmp(spec, last, size);
915 		if (ret != 0)
916 			return rte_flow_error_set(error, EINVAL,
917 						  RTE_FLOW_ERROR_TYPE_ITEM,
918 						  item,
919 						  "range is not valid");
920 	}
921 	return 0;
922 }
923 
924 /**
925  * Adjust the hash fields according to the @p flow information.
926  *
927  * @param[in] dev_flow.
928  *   Pointer to the mlx5_flow.
929  * @param[in] tunnel
930  *   1 when the hash field is for a tunnel item.
931  * @param[in] layer_types
932  *   ETH_RSS_* types.
933  * @param[in] hash_fields
934  *   Item hash fields.
935  *
936  * @return
937  *   The hash fields that should be used.
938  */
939 uint64_t
940 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
941 			    int tunnel __rte_unused, uint64_t layer_types,
942 			    uint64_t hash_fields)
943 {
944 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
945 	int rss_request_inner = rss_desc->level >= 2;
946 
947 	/* Check RSS hash level for tunnel. */
948 	if (tunnel && rss_request_inner)
949 		hash_fields |= IBV_RX_HASH_INNER;
950 	else if (tunnel || rss_request_inner)
951 		return 0;
952 #endif
953 	/* Check if requested layer matches RSS hash fields. */
954 	if (!(rss_desc->types & layer_types))
955 		return 0;
956 	return hash_fields;
957 }
958 
959 /**
960  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
961  * if several tunnel rules are used on this queue, the tunnel ptype will be
962  * cleared.
963  *
964  * @param rxq_ctrl
965  *   Rx queue to update.
966  */
967 static void
968 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
969 {
970 	unsigned int i;
971 	uint32_t tunnel_ptype = 0;
972 
973 	/* Look up for the ptype to use. */
974 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
975 		if (!rxq_ctrl->flow_tunnels_n[i])
976 			continue;
977 		if (!tunnel_ptype) {
978 			tunnel_ptype = tunnels_info[i].ptype;
979 		} else {
980 			tunnel_ptype = 0;
981 			break;
982 		}
983 	}
984 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
985 }
986 
987 /**
988  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
989  * flow.
990  *
991  * @param[in] dev
992  *   Pointer to the Ethernet device structure.
993  * @param[in] dev_handle
994  *   Pointer to device flow handle structure.
995  */
996 static void
997 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
998 		       struct mlx5_flow_handle *dev_handle)
999 {
1000 	struct mlx5_priv *priv = dev->data->dev_private;
1001 	const int mark = dev_handle->mark;
1002 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1003 	struct mlx5_hrxq *hrxq;
1004 	unsigned int i;
1005 
1006 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1007 		return;
1008 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1009 			      dev_handle->rix_hrxq);
1010 	if (!hrxq)
1011 		return;
1012 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1013 		int idx = hrxq->ind_table->queues[i];
1014 		struct mlx5_rxq_ctrl *rxq_ctrl =
1015 			container_of((*priv->rxqs)[idx],
1016 				     struct mlx5_rxq_ctrl, rxq);
1017 
1018 		/*
1019 		 * To support metadata register copy on Tx loopback,
1020 		 * this must be always enabled (metadata may arive
1021 		 * from other port - not from local flows only.
1022 		 */
1023 		if (priv->config.dv_flow_en &&
1024 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1025 		    mlx5_flow_ext_mreg_supported(dev)) {
1026 			rxq_ctrl->rxq.mark = 1;
1027 			rxq_ctrl->flow_mark_n = 1;
1028 		} else if (mark) {
1029 			rxq_ctrl->rxq.mark = 1;
1030 			rxq_ctrl->flow_mark_n++;
1031 		}
1032 		if (tunnel) {
1033 			unsigned int j;
1034 
1035 			/* Increase the counter matching the flow. */
1036 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1037 				if ((tunnels_info[j].tunnel &
1038 				     dev_handle->layers) ==
1039 				    tunnels_info[j].tunnel) {
1040 					rxq_ctrl->flow_tunnels_n[j]++;
1041 					break;
1042 				}
1043 			}
1044 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1045 		}
1046 	}
1047 }
1048 
1049 /**
1050  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1051  *
1052  * @param[in] dev
1053  *   Pointer to the Ethernet device structure.
1054  * @param[in] flow
1055  *   Pointer to flow structure.
1056  */
1057 static void
1058 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1059 {
1060 	struct mlx5_priv *priv = dev->data->dev_private;
1061 	uint32_t handle_idx;
1062 	struct mlx5_flow_handle *dev_handle;
1063 
1064 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1065 		       handle_idx, dev_handle, next)
1066 		flow_drv_rxq_flags_set(dev, dev_handle);
1067 }
1068 
1069 /**
1070  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1071  * device flow if no other flow uses it with the same kind of request.
1072  *
1073  * @param dev
1074  *   Pointer to Ethernet device.
1075  * @param[in] dev_handle
1076  *   Pointer to the device flow handle structure.
1077  */
1078 static void
1079 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1080 			struct mlx5_flow_handle *dev_handle)
1081 {
1082 	struct mlx5_priv *priv = dev->data->dev_private;
1083 	const int mark = dev_handle->mark;
1084 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1085 	struct mlx5_hrxq *hrxq;
1086 	unsigned int i;
1087 
1088 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1089 		return;
1090 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1091 			      dev_handle->rix_hrxq);
1092 	if (!hrxq)
1093 		return;
1094 	MLX5_ASSERT(dev->data->dev_started);
1095 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1096 		int idx = hrxq->ind_table->queues[i];
1097 		struct mlx5_rxq_ctrl *rxq_ctrl =
1098 			container_of((*priv->rxqs)[idx],
1099 				     struct mlx5_rxq_ctrl, rxq);
1100 
1101 		if (priv->config.dv_flow_en &&
1102 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1103 		    mlx5_flow_ext_mreg_supported(dev)) {
1104 			rxq_ctrl->rxq.mark = 1;
1105 			rxq_ctrl->flow_mark_n = 1;
1106 		} else if (mark) {
1107 			rxq_ctrl->flow_mark_n--;
1108 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1109 		}
1110 		if (tunnel) {
1111 			unsigned int j;
1112 
1113 			/* Decrease the counter matching the flow. */
1114 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1115 				if ((tunnels_info[j].tunnel &
1116 				     dev_handle->layers) ==
1117 				    tunnels_info[j].tunnel) {
1118 					rxq_ctrl->flow_tunnels_n[j]--;
1119 					break;
1120 				}
1121 			}
1122 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1123 		}
1124 	}
1125 }
1126 
1127 /**
1128  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1129  * @p flow if no other flow uses it with the same kind of request.
1130  *
1131  * @param dev
1132  *   Pointer to Ethernet device.
1133  * @param[in] flow
1134  *   Pointer to the flow.
1135  */
1136 static void
1137 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1138 {
1139 	struct mlx5_priv *priv = dev->data->dev_private;
1140 	uint32_t handle_idx;
1141 	struct mlx5_flow_handle *dev_handle;
1142 
1143 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1144 		       handle_idx, dev_handle, next)
1145 		flow_drv_rxq_flags_trim(dev, dev_handle);
1146 }
1147 
1148 /**
1149  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1150  *
1151  * @param dev
1152  *   Pointer to Ethernet device.
1153  */
1154 static void
1155 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1156 {
1157 	struct mlx5_priv *priv = dev->data->dev_private;
1158 	unsigned int i;
1159 
1160 	for (i = 0; i != priv->rxqs_n; ++i) {
1161 		struct mlx5_rxq_ctrl *rxq_ctrl;
1162 		unsigned int j;
1163 
1164 		if (!(*priv->rxqs)[i])
1165 			continue;
1166 		rxq_ctrl = container_of((*priv->rxqs)[i],
1167 					struct mlx5_rxq_ctrl, rxq);
1168 		rxq_ctrl->flow_mark_n = 0;
1169 		rxq_ctrl->rxq.mark = 0;
1170 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1171 			rxq_ctrl->flow_tunnels_n[j] = 0;
1172 		rxq_ctrl->rxq.tunnel = 0;
1173 	}
1174 }
1175 
1176 /**
1177  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1178  *
1179  * @param[in] dev
1180  *   Pointer to the Ethernet device structure.
1181  */
1182 void
1183 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1184 {
1185 	struct mlx5_priv *priv = dev->data->dev_private;
1186 	struct mlx5_rxq_data *data;
1187 	unsigned int i;
1188 
1189 	for (i = 0; i != priv->rxqs_n; ++i) {
1190 		if (!(*priv->rxqs)[i])
1191 			continue;
1192 		data = (*priv->rxqs)[i];
1193 		if (!rte_flow_dynf_metadata_avail()) {
1194 			data->dynf_meta = 0;
1195 			data->flow_meta_mask = 0;
1196 			data->flow_meta_offset = -1;
1197 		} else {
1198 			data->dynf_meta = 1;
1199 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1200 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1201 		}
1202 	}
1203 }
1204 
1205 /*
1206  * return a pointer to the desired action in the list of actions.
1207  *
1208  * @param[in] actions
1209  *   The list of actions to search the action in.
1210  * @param[in] action
1211  *   The action to find.
1212  *
1213  * @return
1214  *   Pointer to the action in the list, if found. NULL otherwise.
1215  */
1216 const struct rte_flow_action *
1217 mlx5_flow_find_action(const struct rte_flow_action *actions,
1218 		      enum rte_flow_action_type action)
1219 {
1220 	if (actions == NULL)
1221 		return NULL;
1222 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1223 		if (actions->type == action)
1224 			return actions;
1225 	return NULL;
1226 }
1227 
1228 /*
1229  * Validate the flag action.
1230  *
1231  * @param[in] action_flags
1232  *   Bit-fields that holds the actions detected until now.
1233  * @param[in] attr
1234  *   Attributes of flow that includes this action.
1235  * @param[out] error
1236  *   Pointer to error structure.
1237  *
1238  * @return
1239  *   0 on success, a negative errno value otherwise and rte_errno is set.
1240  */
1241 int
1242 mlx5_flow_validate_action_flag(uint64_t action_flags,
1243 			       const struct rte_flow_attr *attr,
1244 			       struct rte_flow_error *error)
1245 {
1246 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1247 		return rte_flow_error_set(error, EINVAL,
1248 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1249 					  "can't mark and flag in same flow");
1250 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1251 		return rte_flow_error_set(error, EINVAL,
1252 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1253 					  "can't have 2 flag"
1254 					  " actions in same flow");
1255 	if (attr->egress)
1256 		return rte_flow_error_set(error, ENOTSUP,
1257 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1258 					  "flag action not supported for "
1259 					  "egress");
1260 	return 0;
1261 }
1262 
1263 /*
1264  * Validate the mark action.
1265  *
1266  * @param[in] action
1267  *   Pointer to the queue action.
1268  * @param[in] action_flags
1269  *   Bit-fields that holds the actions detected until now.
1270  * @param[in] attr
1271  *   Attributes of flow that includes this action.
1272  * @param[out] error
1273  *   Pointer to error structure.
1274  *
1275  * @return
1276  *   0 on success, a negative errno value otherwise and rte_errno is set.
1277  */
1278 int
1279 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1280 			       uint64_t action_flags,
1281 			       const struct rte_flow_attr *attr,
1282 			       struct rte_flow_error *error)
1283 {
1284 	const struct rte_flow_action_mark *mark = action->conf;
1285 
1286 	if (!mark)
1287 		return rte_flow_error_set(error, EINVAL,
1288 					  RTE_FLOW_ERROR_TYPE_ACTION,
1289 					  action,
1290 					  "configuration cannot be null");
1291 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1292 		return rte_flow_error_set(error, EINVAL,
1293 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1294 					  &mark->id,
1295 					  "mark id must in 0 <= id < "
1296 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1297 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1298 		return rte_flow_error_set(error, EINVAL,
1299 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1300 					  "can't flag and mark in same flow");
1301 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1302 		return rte_flow_error_set(error, EINVAL,
1303 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1304 					  "can't have 2 mark actions in same"
1305 					  " flow");
1306 	if (attr->egress)
1307 		return rte_flow_error_set(error, ENOTSUP,
1308 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1309 					  "mark action not supported for "
1310 					  "egress");
1311 	return 0;
1312 }
1313 
1314 /*
1315  * Validate the drop action.
1316  *
1317  * @param[in] action_flags
1318  *   Bit-fields that holds the actions detected until now.
1319  * @param[in] attr
1320  *   Attributes of flow that includes this action.
1321  * @param[out] error
1322  *   Pointer to error structure.
1323  *
1324  * @return
1325  *   0 on success, a negative errno value otherwise and rte_errno is set.
1326  */
1327 int
1328 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1329 			       const struct rte_flow_attr *attr,
1330 			       struct rte_flow_error *error)
1331 {
1332 	if (attr->egress)
1333 		return rte_flow_error_set(error, ENOTSUP,
1334 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1335 					  "drop action not supported for "
1336 					  "egress");
1337 	return 0;
1338 }
1339 
1340 /*
1341  * Validate the queue action.
1342  *
1343  * @param[in] action
1344  *   Pointer to the queue action.
1345  * @param[in] action_flags
1346  *   Bit-fields that holds the actions detected until now.
1347  * @param[in] dev
1348  *   Pointer to the Ethernet device structure.
1349  * @param[in] attr
1350  *   Attributes of flow that includes this action.
1351  * @param[out] error
1352  *   Pointer to error structure.
1353  *
1354  * @return
1355  *   0 on success, a negative errno value otherwise and rte_errno is set.
1356  */
1357 int
1358 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1359 				uint64_t action_flags,
1360 				struct rte_eth_dev *dev,
1361 				const struct rte_flow_attr *attr,
1362 				struct rte_flow_error *error)
1363 {
1364 	struct mlx5_priv *priv = dev->data->dev_private;
1365 	const struct rte_flow_action_queue *queue = action->conf;
1366 
1367 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1368 		return rte_flow_error_set(error, EINVAL,
1369 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1370 					  "can't have 2 fate actions in"
1371 					  " same flow");
1372 	if (!priv->rxqs_n)
1373 		return rte_flow_error_set(error, EINVAL,
1374 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1375 					  NULL, "No Rx queues configured");
1376 	if (queue->index >= priv->rxqs_n)
1377 		return rte_flow_error_set(error, EINVAL,
1378 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1379 					  &queue->index,
1380 					  "queue index out of range");
1381 	if (!(*priv->rxqs)[queue->index])
1382 		return rte_flow_error_set(error, EINVAL,
1383 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1384 					  &queue->index,
1385 					  "queue is not configured");
1386 	if (attr->egress)
1387 		return rte_flow_error_set(error, ENOTSUP,
1388 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1389 					  "queue action not supported for "
1390 					  "egress");
1391 	return 0;
1392 }
1393 
1394 /*
1395  * Validate the rss action.
1396  *
1397  * @param[in] dev
1398  *   Pointer to the Ethernet device structure.
1399  * @param[in] action
1400  *   Pointer to the queue action.
1401  * @param[out] error
1402  *   Pointer to error structure.
1403  *
1404  * @return
1405  *   0 on success, a negative errno value otherwise and rte_errno is set.
1406  */
1407 int
1408 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1409 			 const struct rte_flow_action *action,
1410 			 struct rte_flow_error *error)
1411 {
1412 	struct mlx5_priv *priv = dev->data->dev_private;
1413 	const struct rte_flow_action_rss *rss = action->conf;
1414 	unsigned int i;
1415 
1416 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1417 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1418 		return rte_flow_error_set(error, ENOTSUP,
1419 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1420 					  &rss->func,
1421 					  "RSS hash function not supported");
1422 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1423 	if (rss->level > 2)
1424 #else
1425 	if (rss->level > 1)
1426 #endif
1427 		return rte_flow_error_set(error, ENOTSUP,
1428 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1429 					  &rss->level,
1430 					  "tunnel RSS is not supported");
1431 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1432 	if (rss->key_len == 0 && rss->key != NULL)
1433 		return rte_flow_error_set(error, ENOTSUP,
1434 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1435 					  &rss->key_len,
1436 					  "RSS hash key length 0");
1437 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1438 		return rte_flow_error_set(error, ENOTSUP,
1439 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1440 					  &rss->key_len,
1441 					  "RSS hash key too small");
1442 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1443 		return rte_flow_error_set(error, ENOTSUP,
1444 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1445 					  &rss->key_len,
1446 					  "RSS hash key too large");
1447 	if (rss->queue_num > priv->config.ind_table_max_size)
1448 		return rte_flow_error_set(error, ENOTSUP,
1449 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1450 					  &rss->queue_num,
1451 					  "number of queues too large");
1452 	if (rss->types & MLX5_RSS_HF_MASK)
1453 		return rte_flow_error_set(error, ENOTSUP,
1454 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1455 					  &rss->types,
1456 					  "some RSS protocols are not"
1457 					  " supported");
1458 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1459 	    !(rss->types & ETH_RSS_IP))
1460 		return rte_flow_error_set(error, EINVAL,
1461 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1462 					  "L3 partial RSS requested but L3 RSS"
1463 					  " type not specified");
1464 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1465 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1466 		return rte_flow_error_set(error, EINVAL,
1467 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1468 					  "L4 partial RSS requested but L4 RSS"
1469 					  " type not specified");
1470 	if (!priv->rxqs_n)
1471 		return rte_flow_error_set(error, EINVAL,
1472 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1473 					  NULL, "No Rx queues configured");
1474 	if (!rss->queue_num)
1475 		return rte_flow_error_set(error, EINVAL,
1476 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1477 					  NULL, "No queues configured");
1478 	for (i = 0; i != rss->queue_num; ++i) {
1479 		if (rss->queue[i] >= priv->rxqs_n)
1480 			return rte_flow_error_set
1481 				(error, EINVAL,
1482 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1483 				 &rss->queue[i], "queue index out of range");
1484 		if (!(*priv->rxqs)[rss->queue[i]])
1485 			return rte_flow_error_set
1486 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1487 				 &rss->queue[i], "queue is not configured");
1488 	}
1489 	return 0;
1490 }
1491 
1492 /*
1493  * Validate the rss action.
1494  *
1495  * @param[in] action
1496  *   Pointer to the queue action.
1497  * @param[in] action_flags
1498  *   Bit-fields that holds the actions detected until now.
1499  * @param[in] dev
1500  *   Pointer to the Ethernet device structure.
1501  * @param[in] attr
1502  *   Attributes of flow that includes this action.
1503  * @param[in] item_flags
1504  *   Items that were detected.
1505  * @param[out] error
1506  *   Pointer to error structure.
1507  *
1508  * @return
1509  *   0 on success, a negative errno value otherwise and rte_errno is set.
1510  */
1511 int
1512 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1513 			      uint64_t action_flags,
1514 			      struct rte_eth_dev *dev,
1515 			      const struct rte_flow_attr *attr,
1516 			      uint64_t item_flags,
1517 			      struct rte_flow_error *error)
1518 {
1519 	const struct rte_flow_action_rss *rss = action->conf;
1520 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1521 	int ret;
1522 
1523 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1524 		return rte_flow_error_set(error, EINVAL,
1525 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1526 					  "can't have 2 fate actions"
1527 					  " in same flow");
1528 	ret = mlx5_validate_action_rss(dev, action, error);
1529 	if (ret)
1530 		return ret;
1531 	if (attr->egress)
1532 		return rte_flow_error_set(error, ENOTSUP,
1533 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1534 					  "rss action not supported for "
1535 					  "egress");
1536 	if (rss->level > 1 && !tunnel)
1537 		return rte_flow_error_set(error, EINVAL,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1539 					  "inner RSS is not supported for "
1540 					  "non-tunnel flows");
1541 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1542 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1543 		return rte_flow_error_set(error, EINVAL,
1544 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1545 					  "RSS on eCPRI is not supported now");
1546 	}
1547 	return 0;
1548 }
1549 
1550 /*
1551  * Validate the default miss action.
1552  *
1553  * @param[in] action_flags
1554  *   Bit-fields that holds the actions detected until now.
1555  * @param[out] error
1556  *   Pointer to error structure.
1557  *
1558  * @return
1559  *   0 on success, a negative errno value otherwise and rte_errno is set.
1560  */
1561 int
1562 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1563 				const struct rte_flow_attr *attr,
1564 				struct rte_flow_error *error)
1565 {
1566 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1567 		return rte_flow_error_set(error, EINVAL,
1568 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1569 					  "can't have 2 fate actions in"
1570 					  " same flow");
1571 	if (attr->egress)
1572 		return rte_flow_error_set(error, ENOTSUP,
1573 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1574 					  "default miss action not supported "
1575 					  "for egress");
1576 	if (attr->group)
1577 		return rte_flow_error_set(error, ENOTSUP,
1578 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1579 					  "only group 0 is supported");
1580 	if (attr->transfer)
1581 		return rte_flow_error_set(error, ENOTSUP,
1582 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1583 					  NULL, "transfer is not supported");
1584 	return 0;
1585 }
1586 
1587 /*
1588  * Validate the count action.
1589  *
1590  * @param[in] dev
1591  *   Pointer to the Ethernet device structure.
1592  * @param[in] attr
1593  *   Attributes of flow that includes this action.
1594  * @param[out] error
1595  *   Pointer to error structure.
1596  *
1597  * @return
1598  *   0 on success, a negative errno value otherwise and rte_errno is set.
1599  */
1600 int
1601 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1602 				const struct rte_flow_attr *attr,
1603 				struct rte_flow_error *error)
1604 {
1605 	if (attr->egress)
1606 		return rte_flow_error_set(error, ENOTSUP,
1607 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1608 					  "count action not supported for "
1609 					  "egress");
1610 	return 0;
1611 }
1612 
1613 /**
1614  * Verify the @p attributes will be correctly understood by the NIC and store
1615  * them in the @p flow if everything is correct.
1616  *
1617  * @param[in] dev
1618  *   Pointer to the Ethernet device structure.
1619  * @param[in] attributes
1620  *   Pointer to flow attributes
1621  * @param[out] error
1622  *   Pointer to error structure.
1623  *
1624  * @return
1625  *   0 on success, a negative errno value otherwise and rte_errno is set.
1626  */
1627 int
1628 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1629 			      const struct rte_flow_attr *attributes,
1630 			      struct rte_flow_error *error)
1631 {
1632 	struct mlx5_priv *priv = dev->data->dev_private;
1633 	uint32_t priority_max = priv->config.flow_prio - 1;
1634 
1635 	if (attributes->group)
1636 		return rte_flow_error_set(error, ENOTSUP,
1637 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1638 					  NULL, "groups is not supported");
1639 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1640 	    attributes->priority >= priority_max)
1641 		return rte_flow_error_set(error, ENOTSUP,
1642 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1643 					  NULL, "priority out of range");
1644 	if (attributes->egress)
1645 		return rte_flow_error_set(error, ENOTSUP,
1646 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1647 					  "egress is not supported");
1648 	if (attributes->transfer && !priv->config.dv_esw_en)
1649 		return rte_flow_error_set(error, ENOTSUP,
1650 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1651 					  NULL, "transfer is not supported");
1652 	if (!attributes->ingress)
1653 		return rte_flow_error_set(error, EINVAL,
1654 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1655 					  NULL,
1656 					  "ingress attribute is mandatory");
1657 	return 0;
1658 }
1659 
1660 /**
1661  * Validate ICMP6 item.
1662  *
1663  * @param[in] item
1664  *   Item specification.
1665  * @param[in] item_flags
1666  *   Bit-fields that holds the items detected until now.
1667  * @param[in] ext_vlan_sup
1668  *   Whether extended VLAN features are supported or not.
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_item_icmp6(const struct rte_flow_item *item,
1677 			       uint64_t item_flags,
1678 			       uint8_t target_protocol,
1679 			       struct rte_flow_error *error)
1680 {
1681 	const struct rte_flow_item_icmp6 *mask = item->mask;
1682 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1683 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1684 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1685 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1686 				      MLX5_FLOW_LAYER_OUTER_L4;
1687 	int ret;
1688 
1689 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1690 		return rte_flow_error_set(error, EINVAL,
1691 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1692 					  "protocol filtering not compatible"
1693 					  " with ICMP6 layer");
1694 	if (!(item_flags & l3m))
1695 		return rte_flow_error_set(error, EINVAL,
1696 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1697 					  "IPv6 is mandatory to filter on"
1698 					  " ICMP6");
1699 	if (item_flags & l4m)
1700 		return rte_flow_error_set(error, EINVAL,
1701 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1702 					  "multiple L4 layers not supported");
1703 	if (!mask)
1704 		mask = &rte_flow_item_icmp6_mask;
1705 	ret = mlx5_flow_item_acceptable
1706 		(item, (const uint8_t *)mask,
1707 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1708 		 sizeof(struct rte_flow_item_icmp6),
1709 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1710 	if (ret < 0)
1711 		return ret;
1712 	return 0;
1713 }
1714 
1715 /**
1716  * Validate ICMP item.
1717  *
1718  * @param[in] item
1719  *   Item specification.
1720  * @param[in] item_flags
1721  *   Bit-fields that holds the items detected until now.
1722  * @param[out] error
1723  *   Pointer to error structure.
1724  *
1725  * @return
1726  *   0 on success, a negative errno value otherwise and rte_errno is set.
1727  */
1728 int
1729 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1730 			     uint64_t item_flags,
1731 			     uint8_t target_protocol,
1732 			     struct rte_flow_error *error)
1733 {
1734 	const struct rte_flow_item_icmp *mask = item->mask;
1735 	const struct rte_flow_item_icmp nic_mask = {
1736 		.hdr.icmp_type = 0xff,
1737 		.hdr.icmp_code = 0xff,
1738 		.hdr.icmp_ident = RTE_BE16(0xffff),
1739 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1740 	};
1741 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1742 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1743 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1744 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1745 				      MLX5_FLOW_LAYER_OUTER_L4;
1746 	int ret;
1747 
1748 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1749 		return rte_flow_error_set(error, EINVAL,
1750 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1751 					  "protocol filtering not compatible"
1752 					  " with ICMP layer");
1753 	if (!(item_flags & l3m))
1754 		return rte_flow_error_set(error, EINVAL,
1755 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1756 					  "IPv4 is mandatory to filter"
1757 					  " on ICMP");
1758 	if (item_flags & l4m)
1759 		return rte_flow_error_set(error, EINVAL,
1760 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1761 					  "multiple L4 layers not supported");
1762 	if (!mask)
1763 		mask = &nic_mask;
1764 	ret = mlx5_flow_item_acceptable
1765 		(item, (const uint8_t *)mask,
1766 		 (const uint8_t *)&nic_mask,
1767 		 sizeof(struct rte_flow_item_icmp),
1768 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1769 	if (ret < 0)
1770 		return ret;
1771 	return 0;
1772 }
1773 
1774 /**
1775  * Validate Ethernet item.
1776  *
1777  * @param[in] item
1778  *   Item specification.
1779  * @param[in] item_flags
1780  *   Bit-fields that holds the items detected until now.
1781  * @param[out] error
1782  *   Pointer to error structure.
1783  *
1784  * @return
1785  *   0 on success, a negative errno value otherwise and rte_errno is set.
1786  */
1787 int
1788 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1789 			    uint64_t item_flags, bool ext_vlan_sup,
1790 			    struct rte_flow_error *error)
1791 {
1792 	const struct rte_flow_item_eth *mask = item->mask;
1793 	const struct rte_flow_item_eth nic_mask = {
1794 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1795 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1796 		.type = RTE_BE16(0xffff),
1797 		.has_vlan = ext_vlan_sup ? 1 : 0,
1798 	};
1799 	int ret;
1800 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1801 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1802 				       MLX5_FLOW_LAYER_OUTER_L2;
1803 
1804 	if (item_flags & ethm)
1805 		return rte_flow_error_set(error, ENOTSUP,
1806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1807 					  "multiple L2 layers not supported");
1808 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1809 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1810 		return rte_flow_error_set(error, EINVAL,
1811 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1812 					  "L2 layer should not follow "
1813 					  "L3 layers");
1814 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1815 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1816 		return rte_flow_error_set(error, EINVAL,
1817 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1818 					  "L2 layer should not follow VLAN");
1819 	if (!mask)
1820 		mask = &rte_flow_item_eth_mask;
1821 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1822 					(const uint8_t *)&nic_mask,
1823 					sizeof(struct rte_flow_item_eth),
1824 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1825 	return ret;
1826 }
1827 
1828 /**
1829  * Validate VLAN item.
1830  *
1831  * @param[in] item
1832  *   Item specification.
1833  * @param[in] item_flags
1834  *   Bit-fields that holds the items detected until now.
1835  * @param[in] dev
1836  *   Ethernet device flow is being created on.
1837  * @param[out] error
1838  *   Pointer to error structure.
1839  *
1840  * @return
1841  *   0 on success, a negative errno value otherwise and rte_errno is set.
1842  */
1843 int
1844 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1845 			     uint64_t item_flags,
1846 			     struct rte_eth_dev *dev,
1847 			     struct rte_flow_error *error)
1848 {
1849 	const struct rte_flow_item_vlan *spec = item->spec;
1850 	const struct rte_flow_item_vlan *mask = item->mask;
1851 	const struct rte_flow_item_vlan nic_mask = {
1852 		.tci = RTE_BE16(UINT16_MAX),
1853 		.inner_type = RTE_BE16(UINT16_MAX),
1854 	};
1855 	uint16_t vlan_tag = 0;
1856 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1857 	int ret;
1858 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1859 					MLX5_FLOW_LAYER_INNER_L4) :
1860 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1861 					MLX5_FLOW_LAYER_OUTER_L4);
1862 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1863 					MLX5_FLOW_LAYER_OUTER_VLAN;
1864 
1865 	if (item_flags & vlanm)
1866 		return rte_flow_error_set(error, EINVAL,
1867 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1868 					  "multiple VLAN layers not supported");
1869 	else if ((item_flags & l34m) != 0)
1870 		return rte_flow_error_set(error, EINVAL,
1871 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1872 					  "VLAN cannot follow L3/L4 layer");
1873 	if (!mask)
1874 		mask = &rte_flow_item_vlan_mask;
1875 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1876 					(const uint8_t *)&nic_mask,
1877 					sizeof(struct rte_flow_item_vlan),
1878 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1879 	if (ret)
1880 		return ret;
1881 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1882 		struct mlx5_priv *priv = dev->data->dev_private;
1883 
1884 		if (priv->vmwa_context) {
1885 			/*
1886 			 * Non-NULL context means we have a virtual machine
1887 			 * and SR-IOV enabled, we have to create VLAN interface
1888 			 * to make hypervisor to setup E-Switch vport
1889 			 * context correctly. We avoid creating the multiple
1890 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1891 			 */
1892 			return rte_flow_error_set(error, EINVAL,
1893 						  RTE_FLOW_ERROR_TYPE_ITEM,
1894 						  item,
1895 						  "VLAN tag mask is not"
1896 						  " supported in virtual"
1897 						  " environment");
1898 		}
1899 	}
1900 	if (spec) {
1901 		vlan_tag = spec->tci;
1902 		vlan_tag &= mask->tci;
1903 	}
1904 	/*
1905 	 * From verbs perspective an empty VLAN is equivalent
1906 	 * to a packet without VLAN layer.
1907 	 */
1908 	if (!vlan_tag)
1909 		return rte_flow_error_set(error, EINVAL,
1910 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1911 					  item->spec,
1912 					  "VLAN cannot be empty");
1913 	return 0;
1914 }
1915 
1916 /**
1917  * Validate IPV4 item.
1918  *
1919  * @param[in] item
1920  *   Item specification.
1921  * @param[in] item_flags
1922  *   Bit-fields that holds the items detected until now.
1923  * @param[in] last_item
1924  *   Previous validated item in the pattern items.
1925  * @param[in] ether_type
1926  *   Type in the ethernet layer header (including dot1q).
1927  * @param[in] acc_mask
1928  *   Acceptable mask, if NULL default internal default mask
1929  *   will be used to check whether item fields are supported.
1930  * @param[in] range_accepted
1931  *   True if range of values is accepted for specific fields, false otherwise.
1932  * @param[out] error
1933  *   Pointer to error structure.
1934  *
1935  * @return
1936  *   0 on success, a negative errno value otherwise and rte_errno is set.
1937  */
1938 int
1939 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1940 			     uint64_t item_flags,
1941 			     uint64_t last_item,
1942 			     uint16_t ether_type,
1943 			     const struct rte_flow_item_ipv4 *acc_mask,
1944 			     bool range_accepted,
1945 			     struct rte_flow_error *error)
1946 {
1947 	const struct rte_flow_item_ipv4 *mask = item->mask;
1948 	const struct rte_flow_item_ipv4 *spec = item->spec;
1949 	const struct rte_flow_item_ipv4 nic_mask = {
1950 		.hdr = {
1951 			.src_addr = RTE_BE32(0xffffffff),
1952 			.dst_addr = RTE_BE32(0xffffffff),
1953 			.type_of_service = 0xff,
1954 			.next_proto_id = 0xff,
1955 		},
1956 	};
1957 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1958 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1959 				      MLX5_FLOW_LAYER_OUTER_L3;
1960 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1961 				      MLX5_FLOW_LAYER_OUTER_L4;
1962 	int ret;
1963 	uint8_t next_proto = 0xFF;
1964 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1965 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1966 				  MLX5_FLOW_LAYER_INNER_VLAN);
1967 
1968 	if ((last_item & l2_vlan) && ether_type &&
1969 	    ether_type != RTE_ETHER_TYPE_IPV4)
1970 		return rte_flow_error_set(error, EINVAL,
1971 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1972 					  "IPv4 cannot follow L2/VLAN layer "
1973 					  "which ether type is not IPv4");
1974 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1975 		if (mask && spec)
1976 			next_proto = mask->hdr.next_proto_id &
1977 				     spec->hdr.next_proto_id;
1978 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1979 			return rte_flow_error_set(error, EINVAL,
1980 						  RTE_FLOW_ERROR_TYPE_ITEM,
1981 						  item,
1982 						  "multiple tunnel "
1983 						  "not supported");
1984 	}
1985 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1988 					  "wrong tunnel type - IPv6 specified "
1989 					  "but IPv4 item provided");
1990 	if (item_flags & l3m)
1991 		return rte_flow_error_set(error, ENOTSUP,
1992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 					  "multiple L3 layers not supported");
1994 	else if (item_flags & l4m)
1995 		return rte_flow_error_set(error, EINVAL,
1996 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1997 					  "L3 cannot follow an L4 layer.");
1998 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1999 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2000 		return rte_flow_error_set(error, EINVAL,
2001 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2002 					  "L3 cannot follow an NVGRE layer.");
2003 	if (!mask)
2004 		mask = &rte_flow_item_ipv4_mask;
2005 	else if (mask->hdr.next_proto_id != 0 &&
2006 		 mask->hdr.next_proto_id != 0xff)
2007 		return rte_flow_error_set(error, EINVAL,
2008 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2009 					  "partial mask is not supported"
2010 					  " for protocol");
2011 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2012 					acc_mask ? (const uint8_t *)acc_mask
2013 						 : (const uint8_t *)&nic_mask,
2014 					sizeof(struct rte_flow_item_ipv4),
2015 					range_accepted, error);
2016 	if (ret < 0)
2017 		return ret;
2018 	return 0;
2019 }
2020 
2021 /**
2022  * Validate IPV6 item.
2023  *
2024  * @param[in] item
2025  *   Item specification.
2026  * @param[in] item_flags
2027  *   Bit-fields that holds the items detected until now.
2028  * @param[in] last_item
2029  *   Previous validated item in the pattern items.
2030  * @param[in] ether_type
2031  *   Type in the ethernet layer header (including dot1q).
2032  * @param[in] acc_mask
2033  *   Acceptable mask, if NULL default internal default mask
2034  *   will be used to check whether item fields are supported.
2035  * @param[out] error
2036  *   Pointer to error structure.
2037  *
2038  * @return
2039  *   0 on success, a negative errno value otherwise and rte_errno is set.
2040  */
2041 int
2042 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2043 			     uint64_t item_flags,
2044 			     uint64_t last_item,
2045 			     uint16_t ether_type,
2046 			     const struct rte_flow_item_ipv6 *acc_mask,
2047 			     struct rte_flow_error *error)
2048 {
2049 	const struct rte_flow_item_ipv6 *mask = item->mask;
2050 	const struct rte_flow_item_ipv6 *spec = item->spec;
2051 	const struct rte_flow_item_ipv6 nic_mask = {
2052 		.hdr = {
2053 			.src_addr =
2054 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2055 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2056 			.dst_addr =
2057 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2058 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2059 			.vtc_flow = RTE_BE32(0xffffffff),
2060 			.proto = 0xff,
2061 		},
2062 	};
2063 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2064 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2065 				      MLX5_FLOW_LAYER_OUTER_L3;
2066 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2067 				      MLX5_FLOW_LAYER_OUTER_L4;
2068 	int ret;
2069 	uint8_t next_proto = 0xFF;
2070 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2071 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2072 				  MLX5_FLOW_LAYER_INNER_VLAN);
2073 
2074 	if ((last_item & l2_vlan) && ether_type &&
2075 	    ether_type != RTE_ETHER_TYPE_IPV6)
2076 		return rte_flow_error_set(error, EINVAL,
2077 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2078 					  "IPv6 cannot follow L2/VLAN layer "
2079 					  "which ether type is not IPv6");
2080 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2081 		next_proto = spec->hdr.proto;
2082 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2083 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2084 			return rte_flow_error_set(error, EINVAL,
2085 						  RTE_FLOW_ERROR_TYPE_ITEM,
2086 						  item,
2087 						  "multiple tunnel "
2088 						  "not supported");
2089 	}
2090 	if (next_proto == IPPROTO_HOPOPTS  ||
2091 	    next_proto == IPPROTO_ROUTING  ||
2092 	    next_proto == IPPROTO_FRAGMENT ||
2093 	    next_proto == IPPROTO_ESP	   ||
2094 	    next_proto == IPPROTO_AH	   ||
2095 	    next_proto == IPPROTO_DSTOPTS)
2096 		return rte_flow_error_set(error, EINVAL,
2097 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2098 					  "IPv6 proto (next header) should "
2099 					  "not be set as extension header");
2100 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2101 		return rte_flow_error_set(error, EINVAL,
2102 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2103 					  "wrong tunnel type - IPv4 specified "
2104 					  "but IPv6 item provided");
2105 	if (item_flags & l3m)
2106 		return rte_flow_error_set(error, ENOTSUP,
2107 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2108 					  "multiple L3 layers not supported");
2109 	else if (item_flags & l4m)
2110 		return rte_flow_error_set(error, EINVAL,
2111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2112 					  "L3 cannot follow an L4 layer.");
2113 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2114 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2115 		return rte_flow_error_set(error, EINVAL,
2116 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2117 					  "L3 cannot follow an NVGRE layer.");
2118 	if (!mask)
2119 		mask = &rte_flow_item_ipv6_mask;
2120 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2121 					acc_mask ? (const uint8_t *)acc_mask
2122 						 : (const uint8_t *)&nic_mask,
2123 					sizeof(struct rte_flow_item_ipv6),
2124 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2125 	if (ret < 0)
2126 		return ret;
2127 	return 0;
2128 }
2129 
2130 /**
2131  * Validate UDP item.
2132  *
2133  * @param[in] item
2134  *   Item specification.
2135  * @param[in] item_flags
2136  *   Bit-fields that holds the items detected until now.
2137  * @param[in] target_protocol
2138  *   The next protocol in the previous item.
2139  * @param[in] flow_mask
2140  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2141  * @param[out] error
2142  *   Pointer to error structure.
2143  *
2144  * @return
2145  *   0 on success, a negative errno value otherwise and rte_errno is set.
2146  */
2147 int
2148 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2149 			    uint64_t item_flags,
2150 			    uint8_t target_protocol,
2151 			    struct rte_flow_error *error)
2152 {
2153 	const struct rte_flow_item_udp *mask = item->mask;
2154 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2155 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2156 				      MLX5_FLOW_LAYER_OUTER_L3;
2157 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2158 				      MLX5_FLOW_LAYER_OUTER_L4;
2159 	int ret;
2160 
2161 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2162 		return rte_flow_error_set(error, EINVAL,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "protocol filtering not compatible"
2165 					  " with UDP layer");
2166 	if (!(item_flags & l3m))
2167 		return rte_flow_error_set(error, EINVAL,
2168 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2169 					  "L3 is mandatory to filter on L4");
2170 	if (item_flags & l4m)
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "multiple L4 layers not supported");
2174 	if (!mask)
2175 		mask = &rte_flow_item_udp_mask;
2176 	ret = mlx5_flow_item_acceptable
2177 		(item, (const uint8_t *)mask,
2178 		 (const uint8_t *)&rte_flow_item_udp_mask,
2179 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2180 		 error);
2181 	if (ret < 0)
2182 		return ret;
2183 	return 0;
2184 }
2185 
2186 /**
2187  * Validate TCP item.
2188  *
2189  * @param[in] item
2190  *   Item specification.
2191  * @param[in] item_flags
2192  *   Bit-fields that holds the items detected until now.
2193  * @param[in] target_protocol
2194  *   The next protocol in the previous item.
2195  * @param[out] error
2196  *   Pointer to error structure.
2197  *
2198  * @return
2199  *   0 on success, a negative errno value otherwise and rte_errno is set.
2200  */
2201 int
2202 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2203 			    uint64_t item_flags,
2204 			    uint8_t target_protocol,
2205 			    const struct rte_flow_item_tcp *flow_mask,
2206 			    struct rte_flow_error *error)
2207 {
2208 	const struct rte_flow_item_tcp *mask = item->mask;
2209 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2210 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2211 				      MLX5_FLOW_LAYER_OUTER_L3;
2212 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2213 				      MLX5_FLOW_LAYER_OUTER_L4;
2214 	int ret;
2215 
2216 	MLX5_ASSERT(flow_mask);
2217 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2218 		return rte_flow_error_set(error, EINVAL,
2219 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2220 					  "protocol filtering not compatible"
2221 					  " with TCP layer");
2222 	if (!(item_flags & l3m))
2223 		return rte_flow_error_set(error, EINVAL,
2224 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2225 					  "L3 is mandatory to filter on L4");
2226 	if (item_flags & l4m)
2227 		return rte_flow_error_set(error, EINVAL,
2228 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 					  "multiple L4 layers not supported");
2230 	if (!mask)
2231 		mask = &rte_flow_item_tcp_mask;
2232 	ret = mlx5_flow_item_acceptable
2233 		(item, (const uint8_t *)mask,
2234 		 (const uint8_t *)flow_mask,
2235 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2236 		 error);
2237 	if (ret < 0)
2238 		return ret;
2239 	return 0;
2240 }
2241 
2242 /**
2243  * Validate VXLAN item.
2244  *
2245  * @param[in] item
2246  *   Item specification.
2247  * @param[in] item_flags
2248  *   Bit-fields that holds the items detected until now.
2249  * @param[in] target_protocol
2250  *   The next protocol in the previous item.
2251  * @param[out] error
2252  *   Pointer to error structure.
2253  *
2254  * @return
2255  *   0 on success, a negative errno value otherwise and rte_errno is set.
2256  */
2257 int
2258 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2259 			      uint64_t item_flags,
2260 			      struct rte_flow_error *error)
2261 {
2262 	const struct rte_flow_item_vxlan *spec = item->spec;
2263 	const struct rte_flow_item_vxlan *mask = item->mask;
2264 	int ret;
2265 	union vni {
2266 		uint32_t vlan_id;
2267 		uint8_t vni[4];
2268 	} id = { .vlan_id = 0, };
2269 
2270 
2271 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2272 		return rte_flow_error_set(error, ENOTSUP,
2273 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2274 					  "multiple tunnel layers not"
2275 					  " supported");
2276 	/*
2277 	 * Verify only UDPv4 is present as defined in
2278 	 * https://tools.ietf.org/html/rfc7348
2279 	 */
2280 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2281 		return rte_flow_error_set(error, EINVAL,
2282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2283 					  "no outer UDP layer found");
2284 	if (!mask)
2285 		mask = &rte_flow_item_vxlan_mask;
2286 	ret = mlx5_flow_item_acceptable
2287 		(item, (const uint8_t *)mask,
2288 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2289 		 sizeof(struct rte_flow_item_vxlan),
2290 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2291 	if (ret < 0)
2292 		return ret;
2293 	if (spec) {
2294 		memcpy(&id.vni[1], spec->vni, 3);
2295 		memcpy(&id.vni[1], mask->vni, 3);
2296 	}
2297 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2298 		return rte_flow_error_set(error, ENOTSUP,
2299 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2300 					  "VXLAN tunnel must be fully defined");
2301 	return 0;
2302 }
2303 
2304 /**
2305  * Validate VXLAN_GPE item.
2306  *
2307  * @param[in] item
2308  *   Item specification.
2309  * @param[in] item_flags
2310  *   Bit-fields that holds the items detected until now.
2311  * @param[in] priv
2312  *   Pointer to the private data structure.
2313  * @param[in] target_protocol
2314  *   The next protocol in the previous item.
2315  * @param[out] error
2316  *   Pointer to error structure.
2317  *
2318  * @return
2319  *   0 on success, a negative errno value otherwise and rte_errno is set.
2320  */
2321 int
2322 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2323 				  uint64_t item_flags,
2324 				  struct rte_eth_dev *dev,
2325 				  struct rte_flow_error *error)
2326 {
2327 	struct mlx5_priv *priv = dev->data->dev_private;
2328 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2329 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2330 	int ret;
2331 	union vni {
2332 		uint32_t vlan_id;
2333 		uint8_t vni[4];
2334 	} id = { .vlan_id = 0, };
2335 
2336 	if (!priv->config.l3_vxlan_en)
2337 		return rte_flow_error_set(error, ENOTSUP,
2338 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2339 					  "L3 VXLAN is not enabled by device"
2340 					  " parameter and/or not configured in"
2341 					  " firmware");
2342 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2343 		return rte_flow_error_set(error, ENOTSUP,
2344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2345 					  "multiple tunnel layers not"
2346 					  " supported");
2347 	/*
2348 	 * Verify only UDPv4 is present as defined in
2349 	 * https://tools.ietf.org/html/rfc7348
2350 	 */
2351 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2352 		return rte_flow_error_set(error, EINVAL,
2353 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2354 					  "no outer UDP layer found");
2355 	if (!mask)
2356 		mask = &rte_flow_item_vxlan_gpe_mask;
2357 	ret = mlx5_flow_item_acceptable
2358 		(item, (const uint8_t *)mask,
2359 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2360 		 sizeof(struct rte_flow_item_vxlan_gpe),
2361 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2362 	if (ret < 0)
2363 		return ret;
2364 	if (spec) {
2365 		if (spec->protocol)
2366 			return rte_flow_error_set(error, ENOTSUP,
2367 						  RTE_FLOW_ERROR_TYPE_ITEM,
2368 						  item,
2369 						  "VxLAN-GPE protocol"
2370 						  " not supported");
2371 		memcpy(&id.vni[1], spec->vni, 3);
2372 		memcpy(&id.vni[1], mask->vni, 3);
2373 	}
2374 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2375 		return rte_flow_error_set(error, ENOTSUP,
2376 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2377 					  "VXLAN-GPE tunnel must be fully"
2378 					  " defined");
2379 	return 0;
2380 }
2381 /**
2382  * Validate GRE Key item.
2383  *
2384  * @param[in] item
2385  *   Item specification.
2386  * @param[in] item_flags
2387  *   Bit flags to mark detected items.
2388  * @param[in] gre_item
2389  *   Pointer to gre_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_gre_key(const struct rte_flow_item *item,
2398 				uint64_t item_flags,
2399 				const struct rte_flow_item *gre_item,
2400 				struct rte_flow_error *error)
2401 {
2402 	const rte_be32_t *mask = item->mask;
2403 	int ret = 0;
2404 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2405 	const struct rte_flow_item_gre *gre_spec;
2406 	const struct rte_flow_item_gre *gre_mask;
2407 
2408 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2409 		return rte_flow_error_set(error, ENOTSUP,
2410 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2411 					  "Multiple GRE key not support");
2412 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2413 		return rte_flow_error_set(error, ENOTSUP,
2414 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2415 					  "No preceding GRE header");
2416 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2417 		return rte_flow_error_set(error, ENOTSUP,
2418 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 					  "GRE key following a wrong item");
2420 	gre_mask = gre_item->mask;
2421 	if (!gre_mask)
2422 		gre_mask = &rte_flow_item_gre_mask;
2423 	gre_spec = gre_item->spec;
2424 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2425 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2426 		return rte_flow_error_set(error, EINVAL,
2427 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2428 					  "Key bit must be on");
2429 
2430 	if (!mask)
2431 		mask = &gre_key_default_mask;
2432 	ret = mlx5_flow_item_acceptable
2433 		(item, (const uint8_t *)mask,
2434 		 (const uint8_t *)&gre_key_default_mask,
2435 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2436 	return ret;
2437 }
2438 
2439 /**
2440  * Validate GRE item.
2441  *
2442  * @param[in] item
2443  *   Item specification.
2444  * @param[in] item_flags
2445  *   Bit flags to mark detected items.
2446  * @param[in] target_protocol
2447  *   The next protocol in the previous item.
2448  * @param[out] error
2449  *   Pointer to error structure.
2450  *
2451  * @return
2452  *   0 on success, a negative errno value otherwise and rte_errno is set.
2453  */
2454 int
2455 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2456 			    uint64_t item_flags,
2457 			    uint8_t target_protocol,
2458 			    struct rte_flow_error *error)
2459 {
2460 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2461 	const struct rte_flow_item_gre *mask = item->mask;
2462 	int ret;
2463 	const struct rte_flow_item_gre nic_mask = {
2464 		.c_rsvd0_ver = RTE_BE16(0xB000),
2465 		.protocol = RTE_BE16(UINT16_MAX),
2466 	};
2467 
2468 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2469 		return rte_flow_error_set(error, EINVAL,
2470 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 					  "protocol filtering not compatible"
2472 					  " with this GRE layer");
2473 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2474 		return rte_flow_error_set(error, ENOTSUP,
2475 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2476 					  "multiple tunnel layers not"
2477 					  " supported");
2478 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2479 		return rte_flow_error_set(error, ENOTSUP,
2480 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2481 					  "L3 Layer is missing");
2482 	if (!mask)
2483 		mask = &rte_flow_item_gre_mask;
2484 	ret = mlx5_flow_item_acceptable
2485 		(item, (const uint8_t *)mask,
2486 		 (const uint8_t *)&nic_mask,
2487 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2488 		 error);
2489 	if (ret < 0)
2490 		return ret;
2491 #ifndef HAVE_MLX5DV_DR
2492 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2493 	if (spec && (spec->protocol & mask->protocol))
2494 		return rte_flow_error_set(error, ENOTSUP,
2495 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2496 					  "without MPLS support the"
2497 					  " specification cannot be used for"
2498 					  " filtering");
2499 #endif
2500 #endif
2501 	return 0;
2502 }
2503 
2504 /**
2505  * Validate Geneve item.
2506  *
2507  * @param[in] item
2508  *   Item specification.
2509  * @param[in] itemFlags
2510  *   Bit-fields that holds the items detected until now.
2511  * @param[in] enPriv
2512  *   Pointer to the private data structure.
2513  * @param[out] error
2514  *   Pointer to error structure.
2515  *
2516  * @return
2517  *   0 on success, a negative errno value otherwise and rte_errno is set.
2518  */
2519 
2520 int
2521 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2522 			       uint64_t item_flags,
2523 			       struct rte_eth_dev *dev,
2524 			       struct rte_flow_error *error)
2525 {
2526 	struct mlx5_priv *priv = dev->data->dev_private;
2527 	const struct rte_flow_item_geneve *spec = item->spec;
2528 	const struct rte_flow_item_geneve *mask = item->mask;
2529 	int ret;
2530 	uint16_t gbhdr;
2531 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2532 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2533 	const struct rte_flow_item_geneve nic_mask = {
2534 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2535 		.vni = "\xff\xff\xff",
2536 		.protocol = RTE_BE16(UINT16_MAX),
2537 	};
2538 
2539 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2540 		return rte_flow_error_set(error, ENOTSUP,
2541 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2542 					  "L3 Geneve is not enabled by device"
2543 					  " parameter and/or not configured in"
2544 					  " firmware");
2545 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2546 		return rte_flow_error_set(error, ENOTSUP,
2547 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2548 					  "multiple tunnel layers not"
2549 					  " supported");
2550 	/*
2551 	 * Verify only UDPv4 is present as defined in
2552 	 * https://tools.ietf.org/html/rfc7348
2553 	 */
2554 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2555 		return rte_flow_error_set(error, EINVAL,
2556 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2557 					  "no outer UDP layer found");
2558 	if (!mask)
2559 		mask = &rte_flow_item_geneve_mask;
2560 	ret = mlx5_flow_item_acceptable
2561 				  (item, (const uint8_t *)mask,
2562 				   (const uint8_t *)&nic_mask,
2563 				   sizeof(struct rte_flow_item_geneve),
2564 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2565 	if (ret)
2566 		return ret;
2567 	if (spec) {
2568 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2569 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2570 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2571 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2572 			return rte_flow_error_set(error, ENOTSUP,
2573 						  RTE_FLOW_ERROR_TYPE_ITEM,
2574 						  item,
2575 						  "Geneve protocol unsupported"
2576 						  " fields are being used");
2577 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2578 			return rte_flow_error_set
2579 					(error, ENOTSUP,
2580 					 RTE_FLOW_ERROR_TYPE_ITEM,
2581 					 item,
2582 					 "Unsupported Geneve options length");
2583 	}
2584 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2585 		return rte_flow_error_set
2586 				    (error, ENOTSUP,
2587 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2588 				     "Geneve tunnel must be fully defined");
2589 	return 0;
2590 }
2591 
2592 /**
2593  * Validate MPLS item.
2594  *
2595  * @param[in] dev
2596  *   Pointer to the rte_eth_dev structure.
2597  * @param[in] item
2598  *   Item specification.
2599  * @param[in] item_flags
2600  *   Bit-fields that holds the items detected until now.
2601  * @param[in] prev_layer
2602  *   The protocol layer indicated in previous item.
2603  * @param[out] error
2604  *   Pointer to error structure.
2605  *
2606  * @return
2607  *   0 on success, a negative errno value otherwise and rte_errno is set.
2608  */
2609 int
2610 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2611 			     const struct rte_flow_item *item __rte_unused,
2612 			     uint64_t item_flags __rte_unused,
2613 			     uint64_t prev_layer __rte_unused,
2614 			     struct rte_flow_error *error)
2615 {
2616 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2617 	const struct rte_flow_item_mpls *mask = item->mask;
2618 	struct mlx5_priv *priv = dev->data->dev_private;
2619 	int ret;
2620 
2621 	if (!priv->config.mpls_en)
2622 		return rte_flow_error_set(error, ENOTSUP,
2623 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2624 					  "MPLS not supported or"
2625 					  " disabled in firmware"
2626 					  " configuration.");
2627 	/* MPLS over IP, UDP, GRE is allowed */
2628 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2629 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2630 			    MLX5_FLOW_LAYER_GRE |
2631 			    MLX5_FLOW_LAYER_GRE_KEY)))
2632 		return rte_flow_error_set(error, EINVAL,
2633 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2634 					  "protocol filtering not compatible"
2635 					  " with MPLS layer");
2636 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2637 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2638 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2639 		return rte_flow_error_set(error, ENOTSUP,
2640 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2641 					  "multiple tunnel layers not"
2642 					  " supported");
2643 	if (!mask)
2644 		mask = &rte_flow_item_mpls_mask;
2645 	ret = mlx5_flow_item_acceptable
2646 		(item, (const uint8_t *)mask,
2647 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2648 		 sizeof(struct rte_flow_item_mpls),
2649 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2650 	if (ret < 0)
2651 		return ret;
2652 	return 0;
2653 #else
2654 	return rte_flow_error_set(error, ENOTSUP,
2655 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2656 				  "MPLS is not supported by Verbs, please"
2657 				  " update.");
2658 #endif
2659 }
2660 
2661 /**
2662  * Validate NVGRE item.
2663  *
2664  * @param[in] item
2665  *   Item specification.
2666  * @param[in] item_flags
2667  *   Bit flags to mark detected items.
2668  * @param[in] target_protocol
2669  *   The next protocol in the previous item.
2670  * @param[out] error
2671  *   Pointer to error structure.
2672  *
2673  * @return
2674  *   0 on success, a negative errno value otherwise and rte_errno is set.
2675  */
2676 int
2677 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2678 			      uint64_t item_flags,
2679 			      uint8_t target_protocol,
2680 			      struct rte_flow_error *error)
2681 {
2682 	const struct rte_flow_item_nvgre *mask = item->mask;
2683 	int ret;
2684 
2685 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2686 		return rte_flow_error_set(error, EINVAL,
2687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2688 					  "protocol filtering not compatible"
2689 					  " with this GRE layer");
2690 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2691 		return rte_flow_error_set(error, ENOTSUP,
2692 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2693 					  "multiple tunnel layers not"
2694 					  " supported");
2695 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2696 		return rte_flow_error_set(error, ENOTSUP,
2697 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2698 					  "L3 Layer is missing");
2699 	if (!mask)
2700 		mask = &rte_flow_item_nvgre_mask;
2701 	ret = mlx5_flow_item_acceptable
2702 		(item, (const uint8_t *)mask,
2703 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2704 		 sizeof(struct rte_flow_item_nvgre),
2705 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2706 	if (ret < 0)
2707 		return ret;
2708 	return 0;
2709 }
2710 
2711 /**
2712  * Validate eCPRI item.
2713  *
2714  * @param[in] item
2715  *   Item specification.
2716  * @param[in] item_flags
2717  *   Bit-fields that holds the items detected until now.
2718  * @param[in] last_item
2719  *   Previous validated item in the pattern items.
2720  * @param[in] ether_type
2721  *   Type in the ethernet layer header (including dot1q).
2722  * @param[in] acc_mask
2723  *   Acceptable mask, if NULL default internal default mask
2724  *   will be used to check whether item fields are supported.
2725  * @param[out] error
2726  *   Pointer to error structure.
2727  *
2728  * @return
2729  *   0 on success, a negative errno value otherwise and rte_errno is set.
2730  */
2731 int
2732 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2733 			      uint64_t item_flags,
2734 			      uint64_t last_item,
2735 			      uint16_t ether_type,
2736 			      const struct rte_flow_item_ecpri *acc_mask,
2737 			      struct rte_flow_error *error)
2738 {
2739 	const struct rte_flow_item_ecpri *mask = item->mask;
2740 	const struct rte_flow_item_ecpri nic_mask = {
2741 		.hdr = {
2742 			.common = {
2743 				.u32 =
2744 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2745 					.type = 0xFF,
2746 					}).u32),
2747 			},
2748 			.dummy[0] = 0xFFFFFFFF,
2749 		},
2750 	};
2751 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2752 					MLX5_FLOW_LAYER_OUTER_VLAN);
2753 	struct rte_flow_item_ecpri mask_lo;
2754 
2755 	if (!(last_item & outer_l2_vlan) &&
2756 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2757 		return rte_flow_error_set(error, EINVAL,
2758 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2759 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2760 	if ((last_item & outer_l2_vlan) && ether_type &&
2761 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2762 		return rte_flow_error_set(error, EINVAL,
2763 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2764 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2765 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2766 		return rte_flow_error_set(error, EINVAL,
2767 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2768 					  "eCPRI with tunnel is not supported right now");
2769 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2770 		return rte_flow_error_set(error, ENOTSUP,
2771 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2772 					  "multiple L3 layers not supported");
2773 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2774 		return rte_flow_error_set(error, EINVAL,
2775 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2776 					  "eCPRI cannot coexist with a TCP layer");
2777 	/* In specification, eCPRI could be over UDP layer. */
2778 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2779 		return rte_flow_error_set(error, EINVAL,
2780 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2781 					  "eCPRI over UDP layer is not yet supported right now");
2782 	/* Mask for type field in common header could be zero. */
2783 	if (!mask)
2784 		mask = &rte_flow_item_ecpri_mask;
2785 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2786 	/* Input mask is in big-endian format. */
2787 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2788 		return rte_flow_error_set(error, EINVAL,
2789 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2790 					  "partial mask is not supported for protocol");
2791 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2792 		return rte_flow_error_set(error, EINVAL,
2793 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2794 					  "message header mask must be after a type mask");
2795 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2796 					 acc_mask ? (const uint8_t *)acc_mask
2797 						  : (const uint8_t *)&nic_mask,
2798 					 sizeof(struct rte_flow_item_ecpri),
2799 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2800 }
2801 
2802 /**
2803  * Release resource related QUEUE/RSS action split.
2804  *
2805  * @param dev
2806  *   Pointer to Ethernet device.
2807  * @param flow
2808  *   Flow to release id's from.
2809  */
2810 static void
2811 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2812 			     struct rte_flow *flow)
2813 {
2814 	struct mlx5_priv *priv = dev->data->dev_private;
2815 	uint32_t handle_idx;
2816 	struct mlx5_flow_handle *dev_handle;
2817 
2818 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2819 		       handle_idx, dev_handle, next)
2820 		if (dev_handle->split_flow_id)
2821 			mlx5_ipool_free(priv->sh->ipool
2822 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2823 					dev_handle->split_flow_id);
2824 }
2825 
2826 static int
2827 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2828 		   const struct rte_flow_attr *attr __rte_unused,
2829 		   const struct rte_flow_item items[] __rte_unused,
2830 		   const struct rte_flow_action actions[] __rte_unused,
2831 		   bool external __rte_unused,
2832 		   int hairpin __rte_unused,
2833 		   struct rte_flow_error *error)
2834 {
2835 	return rte_flow_error_set(error, ENOTSUP,
2836 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2837 }
2838 
2839 static struct mlx5_flow *
2840 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2841 		  const struct rte_flow_attr *attr __rte_unused,
2842 		  const struct rte_flow_item items[] __rte_unused,
2843 		  const struct rte_flow_action actions[] __rte_unused,
2844 		  struct rte_flow_error *error)
2845 {
2846 	rte_flow_error_set(error, ENOTSUP,
2847 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2848 	return NULL;
2849 }
2850 
2851 static int
2852 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2853 		    struct mlx5_flow *dev_flow __rte_unused,
2854 		    const struct rte_flow_attr *attr __rte_unused,
2855 		    const struct rte_flow_item items[] __rte_unused,
2856 		    const struct rte_flow_action actions[] __rte_unused,
2857 		    struct rte_flow_error *error)
2858 {
2859 	return rte_flow_error_set(error, ENOTSUP,
2860 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2861 }
2862 
2863 static int
2864 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2865 		struct rte_flow *flow __rte_unused,
2866 		struct rte_flow_error *error)
2867 {
2868 	return rte_flow_error_set(error, ENOTSUP,
2869 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2870 }
2871 
2872 static void
2873 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2874 		 struct rte_flow *flow __rte_unused)
2875 {
2876 }
2877 
2878 static void
2879 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2880 		  struct rte_flow *flow __rte_unused)
2881 {
2882 }
2883 
2884 static int
2885 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2886 		struct rte_flow *flow __rte_unused,
2887 		const struct rte_flow_action *actions __rte_unused,
2888 		void *data __rte_unused,
2889 		struct rte_flow_error *error)
2890 {
2891 	return rte_flow_error_set(error, ENOTSUP,
2892 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2893 }
2894 
2895 static int
2896 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
2897 		      uint32_t domains __rte_unused,
2898 		      uint32_t flags __rte_unused)
2899 {
2900 	return 0;
2901 }
2902 
2903 /* Void driver to protect from null pointer reference. */
2904 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2905 	.validate = flow_null_validate,
2906 	.prepare = flow_null_prepare,
2907 	.translate = flow_null_translate,
2908 	.apply = flow_null_apply,
2909 	.remove = flow_null_remove,
2910 	.destroy = flow_null_destroy,
2911 	.query = flow_null_query,
2912 	.sync_domain = flow_null_sync_domain,
2913 };
2914 
2915 /**
2916  * Select flow driver type according to flow attributes and device
2917  * configuration.
2918  *
2919  * @param[in] dev
2920  *   Pointer to the dev structure.
2921  * @param[in] attr
2922  *   Pointer to the flow attributes.
2923  *
2924  * @return
2925  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2926  */
2927 static enum mlx5_flow_drv_type
2928 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2929 {
2930 	struct mlx5_priv *priv = dev->data->dev_private;
2931 	/* The OS can determine first a specific flow type (DV, VERBS) */
2932 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2933 
2934 	if (type != MLX5_FLOW_TYPE_MAX)
2935 		return type;
2936 	/* If no OS specific type - continue with DV/VERBS selection */
2937 	if (attr->transfer && priv->config.dv_esw_en)
2938 		type = MLX5_FLOW_TYPE_DV;
2939 	if (!attr->transfer)
2940 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2941 						 MLX5_FLOW_TYPE_VERBS;
2942 	return type;
2943 }
2944 
2945 #define flow_get_drv_ops(type) flow_drv_ops[type]
2946 
2947 /**
2948  * Flow driver validation API. This abstracts calling driver specific functions.
2949  * The type of flow driver is determined according to flow attributes.
2950  *
2951  * @param[in] dev
2952  *   Pointer to the dev structure.
2953  * @param[in] attr
2954  *   Pointer to the flow attributes.
2955  * @param[in] items
2956  *   Pointer to the list of items.
2957  * @param[in] actions
2958  *   Pointer to the list of actions.
2959  * @param[in] external
2960  *   This flow rule is created by request external to PMD.
2961  * @param[in] hairpin
2962  *   Number of hairpin TX actions, 0 means classic flow.
2963  * @param[out] error
2964  *   Pointer to the error structure.
2965  *
2966  * @return
2967  *   0 on success, a negative errno value otherwise and rte_errno is set.
2968  */
2969 static inline int
2970 flow_drv_validate(struct rte_eth_dev *dev,
2971 		  const struct rte_flow_attr *attr,
2972 		  const struct rte_flow_item items[],
2973 		  const struct rte_flow_action actions[],
2974 		  bool external, int hairpin, struct rte_flow_error *error)
2975 {
2976 	const struct mlx5_flow_driver_ops *fops;
2977 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2978 
2979 	fops = flow_get_drv_ops(type);
2980 	return fops->validate(dev, attr, items, actions, external,
2981 			      hairpin, error);
2982 }
2983 
2984 /**
2985  * Flow driver preparation API. This abstracts calling driver specific
2986  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2987  * calculates the size of memory required for device flow, allocates the memory,
2988  * initializes the device flow and returns the pointer.
2989  *
2990  * @note
2991  *   This function initializes device flow structure such as dv or verbs in
2992  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2993  *   rest. For example, adding returning device flow to flow->dev_flow list and
2994  *   setting backward reference to the flow should be done out of this function.
2995  *   layers field is not filled either.
2996  *
2997  * @param[in] dev
2998  *   Pointer to the dev structure.
2999  * @param[in] attr
3000  *   Pointer to the flow attributes.
3001  * @param[in] items
3002  *   Pointer to the list of items.
3003  * @param[in] actions
3004  *   Pointer to the list of actions.
3005  * @param[in] flow_idx
3006  *   This memory pool index to the flow.
3007  * @param[out] error
3008  *   Pointer to the error structure.
3009  *
3010  * @return
3011  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3012  */
3013 static inline struct mlx5_flow *
3014 flow_drv_prepare(struct rte_eth_dev *dev,
3015 		 const struct rte_flow *flow,
3016 		 const struct rte_flow_attr *attr,
3017 		 const struct rte_flow_item items[],
3018 		 const struct rte_flow_action actions[],
3019 		 uint32_t flow_idx,
3020 		 struct rte_flow_error *error)
3021 {
3022 	const struct mlx5_flow_driver_ops *fops;
3023 	enum mlx5_flow_drv_type type = flow->drv_type;
3024 	struct mlx5_flow *mlx5_flow = NULL;
3025 
3026 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3027 	fops = flow_get_drv_ops(type);
3028 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3029 	if (mlx5_flow)
3030 		mlx5_flow->flow_idx = flow_idx;
3031 	return mlx5_flow;
3032 }
3033 
3034 /**
3035  * Flow driver translation API. This abstracts calling driver specific
3036  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3037  * translates a generic flow into a driver flow. flow_drv_prepare() must
3038  * precede.
3039  *
3040  * @note
3041  *   dev_flow->layers could be filled as a result of parsing during translation
3042  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3043  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3044  *   flow->actions could be overwritten even though all the expanded dev_flows
3045  *   have the same actions.
3046  *
3047  * @param[in] dev
3048  *   Pointer to the rte dev structure.
3049  * @param[in, out] dev_flow
3050  *   Pointer to the mlx5 flow.
3051  * @param[in] attr
3052  *   Pointer to the flow attributes.
3053  * @param[in] items
3054  *   Pointer to the list of items.
3055  * @param[in] actions
3056  *   Pointer to the list of actions.
3057  * @param[out] error
3058  *   Pointer to the error structure.
3059  *
3060  * @return
3061  *   0 on success, a negative errno value otherwise and rte_errno is set.
3062  */
3063 static inline int
3064 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3065 		   const struct rte_flow_attr *attr,
3066 		   const struct rte_flow_item items[],
3067 		   const struct rte_flow_action actions[],
3068 		   struct rte_flow_error *error)
3069 {
3070 	const struct mlx5_flow_driver_ops *fops;
3071 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3072 
3073 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3074 	fops = flow_get_drv_ops(type);
3075 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3076 }
3077 
3078 /**
3079  * Flow driver apply API. This abstracts calling driver specific functions.
3080  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3081  * translated driver flows on to device. flow_drv_translate() must precede.
3082  *
3083  * @param[in] dev
3084  *   Pointer to Ethernet device structure.
3085  * @param[in, out] flow
3086  *   Pointer to flow structure.
3087  * @param[out] error
3088  *   Pointer to error structure.
3089  *
3090  * @return
3091  *   0 on success, a negative errno value otherwise and rte_errno is set.
3092  */
3093 static inline int
3094 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3095 	       struct rte_flow_error *error)
3096 {
3097 	const struct mlx5_flow_driver_ops *fops;
3098 	enum mlx5_flow_drv_type type = flow->drv_type;
3099 
3100 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3101 	fops = flow_get_drv_ops(type);
3102 	return fops->apply(dev, flow, error);
3103 }
3104 
3105 /**
3106  * Flow driver destroy API. This abstracts calling driver specific functions.
3107  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3108  * on device and releases resources of the flow.
3109  *
3110  * @param[in] dev
3111  *   Pointer to Ethernet device.
3112  * @param[in, out] flow
3113  *   Pointer to flow structure.
3114  */
3115 static inline void
3116 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3117 {
3118 	const struct mlx5_flow_driver_ops *fops;
3119 	enum mlx5_flow_drv_type type = flow->drv_type;
3120 
3121 	flow_mreg_split_qrss_release(dev, flow);
3122 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3123 	fops = flow_get_drv_ops(type);
3124 	fops->destroy(dev, flow);
3125 }
3126 
3127 /**
3128  * Get RSS action from the action list.
3129  *
3130  * @param[in] actions
3131  *   Pointer to the list of actions.
3132  *
3133  * @return
3134  *   Pointer to the RSS action if exist, else return NULL.
3135  */
3136 static const struct rte_flow_action_rss*
3137 flow_get_rss_action(const struct rte_flow_action actions[])
3138 {
3139 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3140 		switch (actions->type) {
3141 		case RTE_FLOW_ACTION_TYPE_RSS:
3142 			return (const struct rte_flow_action_rss *)
3143 			       actions->conf;
3144 		default:
3145 			break;
3146 		}
3147 	}
3148 	return NULL;
3149 }
3150 
3151 /**
3152  * Get ASO age action by index.
3153  *
3154  * @param[in] dev
3155  *   Pointer to the Ethernet device structure.
3156  * @param[in] age_idx
3157  *   Index to the ASO age action.
3158  *
3159  * @return
3160  *   The specified ASO age action.
3161  */
3162 struct mlx5_aso_age_action*
3163 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3164 {
3165 	uint16_t pool_idx = age_idx & UINT16_MAX;
3166 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3167 	struct mlx5_priv *priv = dev->data->dev_private;
3168 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3169 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3170 
3171 	return &pool->actions[offset - 1];
3172 }
3173 
3174 /* maps shared action to translated non shared in some actions array */
3175 struct mlx5_translated_shared_action {
3176 	struct rte_flow_shared_action *action; /**< Shared action */
3177 	int index; /**< Index in related array of rte_flow_action */
3178 };
3179 
3180 /**
3181  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3182  * non shared action if translation possible.
3183  * This functionality used to run same execution path for both shared & non
3184  * shared actions on flow create. All necessary preparations for shared
3185  * action handling should be preformed on *shared* actions list returned
3186  * from this call.
3187  *
3188  * @param[in] dev
3189  *   Pointer to Ethernet device.
3190  * @param[in] actions
3191  *   List of actions to translate.
3192  * @param[out] shared
3193  *   List to store translated shared actions.
3194  * @param[in, out] shared_n
3195  *   Size of *shared* array. On return should be updated with number of shared
3196  *   actions retrieved from the *actions* list.
3197  * @param[out] translated_actions
3198  *   List of actions where all shared actions were translated to non shared
3199  *   if possible. NULL if no translation took place.
3200  * @param[out] error
3201  *   Pointer to the error structure.
3202  *
3203  * @return
3204  *   0 on success, a negative errno value otherwise and rte_errno is set.
3205  */
3206 static int
3207 flow_shared_actions_translate(struct rte_eth_dev *dev,
3208 			      const struct rte_flow_action actions[],
3209 			      struct mlx5_translated_shared_action *shared,
3210 			      int *shared_n,
3211 			      struct rte_flow_action **translated_actions,
3212 			      struct rte_flow_error *error)
3213 {
3214 	struct mlx5_priv *priv = dev->data->dev_private;
3215 	struct rte_flow_action *translated = NULL;
3216 	size_t actions_size;
3217 	int n;
3218 	int copied_n = 0;
3219 	struct mlx5_translated_shared_action *shared_end = NULL;
3220 
3221 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3222 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3223 			continue;
3224 		if (copied_n == *shared_n) {
3225 			return rte_flow_error_set
3226 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3227 				 NULL, "too many shared actions");
3228 		}
3229 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3230 			   sizeof(actions[n].conf));
3231 		shared[copied_n].index = n;
3232 		copied_n++;
3233 	}
3234 	n++;
3235 	*shared_n = copied_n;
3236 	if (!copied_n)
3237 		return 0;
3238 	actions_size = sizeof(struct rte_flow_action) * n;
3239 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3240 	if (!translated) {
3241 		rte_errno = ENOMEM;
3242 		return -ENOMEM;
3243 	}
3244 	memcpy(translated, actions, actions_size);
3245 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3246 		struct mlx5_shared_action_rss *shared_rss;
3247 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3248 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3249 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3250 									   - 1);
3251 
3252 		switch (type) {
3253 		case MLX5_SHARED_ACTION_TYPE_RSS:
3254 			shared_rss = mlx5_ipool_get
3255 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3256 			translated[shared->index].type =
3257 				RTE_FLOW_ACTION_TYPE_RSS;
3258 			translated[shared->index].conf =
3259 				&shared_rss->origin;
3260 			break;
3261 		case MLX5_SHARED_ACTION_TYPE_AGE:
3262 			if (priv->sh->flow_hit_aso_en) {
3263 				translated[shared->index].type =
3264 					(enum rte_flow_action_type)
3265 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3266 				translated[shared->index].conf =
3267 							 (void *)(uintptr_t)idx;
3268 				break;
3269 			}
3270 			/* Fall-through */
3271 		default:
3272 			mlx5_free(translated);
3273 			return rte_flow_error_set
3274 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3275 				 NULL, "invalid shared action type");
3276 		}
3277 	}
3278 	*translated_actions = translated;
3279 	return 0;
3280 }
3281 
3282 /**
3283  * Get Shared RSS action from the action list.
3284  *
3285  * @param[in] dev
3286  *   Pointer to Ethernet device.
3287  * @param[in] shared
3288  *   Pointer to the list of actions.
3289  * @param[in] shared_n
3290  *   Actions list length.
3291  *
3292  * @return
3293  *   The MLX5 RSS action ID if exists, otherwise return 0.
3294  */
3295 static uint32_t
3296 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3297 			   struct mlx5_translated_shared_action *shared,
3298 			   int shared_n)
3299 {
3300 	struct mlx5_translated_shared_action *shared_end;
3301 	struct mlx5_priv *priv = dev->data->dev_private;
3302 	struct mlx5_shared_action_rss *shared_rss;
3303 
3304 
3305 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3306 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3307 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3308 		uint32_t idx = act_idx &
3309 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3310 		switch (type) {
3311 		case MLX5_SHARED_ACTION_TYPE_RSS:
3312 			shared_rss = mlx5_ipool_get
3313 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3314 									   idx);
3315 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3316 					   __ATOMIC_RELAXED);
3317 			return idx;
3318 		default:
3319 			break;
3320 		}
3321 	}
3322 	return 0;
3323 }
3324 
3325 static unsigned int
3326 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3327 {
3328 	const struct rte_flow_item *item;
3329 	unsigned int has_vlan = 0;
3330 
3331 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3332 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3333 			has_vlan = 1;
3334 			break;
3335 		}
3336 	}
3337 	if (has_vlan)
3338 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3339 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3340 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3341 			       MLX5_EXPANSION_ROOT_OUTER;
3342 }
3343 
3344 /**
3345  *  Get layer flags from the prefix flow.
3346  *
3347  *  Some flows may be split to several subflows, the prefix subflow gets the
3348  *  match items and the suffix sub flow gets the actions.
3349  *  Some actions need the user defined match item flags to get the detail for
3350  *  the action.
3351  *  This function helps the suffix flow to get the item layer flags from prefix
3352  *  subflow.
3353  *
3354  * @param[in] dev_flow
3355  *   Pointer the created preifx subflow.
3356  *
3357  * @return
3358  *   The layers get from prefix subflow.
3359  */
3360 static inline uint64_t
3361 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3362 {
3363 	uint64_t layers = 0;
3364 
3365 	/*
3366 	 * Layers bits could be localization, but usually the compiler will
3367 	 * help to do the optimization work for source code.
3368 	 * If no decap actions, use the layers directly.
3369 	 */
3370 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3371 		return dev_flow->handle->layers;
3372 	/* Convert L3 layers with decap action. */
3373 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3374 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3375 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3376 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3377 	/* Convert L4 layers with decap action.  */
3378 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3379 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3380 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3381 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3382 	return layers;
3383 }
3384 
3385 /**
3386  * Get metadata split action information.
3387  *
3388  * @param[in] actions
3389  *   Pointer to the list of actions.
3390  * @param[out] qrss
3391  *   Pointer to the return pointer.
3392  * @param[out] qrss_type
3393  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3394  *   if no QUEUE/RSS is found.
3395  * @param[out] encap_idx
3396  *   Pointer to the index of the encap action if exists, otherwise the last
3397  *   action index.
3398  *
3399  * @return
3400  *   Total number of actions.
3401  */
3402 static int
3403 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3404 				       const struct rte_flow_action **qrss,
3405 				       int *encap_idx)
3406 {
3407 	const struct rte_flow_action_raw_encap *raw_encap;
3408 	int actions_n = 0;
3409 	int raw_decap_idx = -1;
3410 
3411 	*encap_idx = -1;
3412 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3413 		switch (actions->type) {
3414 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3415 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3416 			*encap_idx = actions_n;
3417 			break;
3418 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3419 			raw_decap_idx = actions_n;
3420 			break;
3421 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3422 			raw_encap = actions->conf;
3423 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3424 				*encap_idx = raw_decap_idx != -1 ?
3425 						      raw_decap_idx : actions_n;
3426 			break;
3427 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3428 		case RTE_FLOW_ACTION_TYPE_RSS:
3429 			*qrss = actions;
3430 			break;
3431 		default:
3432 			break;
3433 		}
3434 		actions_n++;
3435 	}
3436 	if (*encap_idx == -1)
3437 		*encap_idx = actions_n;
3438 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3439 	return actions_n + 1;
3440 }
3441 
3442 /**
3443  * Check meter action from the action list.
3444  *
3445  * @param[in] actions
3446  *   Pointer to the list of actions.
3447  * @param[out] mtr
3448  *   Pointer to the meter exist flag.
3449  *
3450  * @return
3451  *   Total number of actions.
3452  */
3453 static int
3454 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3455 {
3456 	int actions_n = 0;
3457 
3458 	MLX5_ASSERT(mtr);
3459 	*mtr = 0;
3460 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3461 		switch (actions->type) {
3462 		case RTE_FLOW_ACTION_TYPE_METER:
3463 			*mtr = 1;
3464 			break;
3465 		default:
3466 			break;
3467 		}
3468 		actions_n++;
3469 	}
3470 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3471 	return actions_n + 1;
3472 }
3473 
3474 /**
3475  * Check if the flow should be split due to hairpin.
3476  * The reason for the split is that in current HW we can't
3477  * support encap and push-vlan on Rx, so if a flow contains
3478  * these actions we move it to Tx.
3479  *
3480  * @param dev
3481  *   Pointer to Ethernet device.
3482  * @param[in] attr
3483  *   Flow rule attributes.
3484  * @param[in] actions
3485  *   Associated actions (list terminated by the END action).
3486  *
3487  * @return
3488  *   > 0 the number of actions and the flow should be split,
3489  *   0 when no split required.
3490  */
3491 static int
3492 flow_check_hairpin_split(struct rte_eth_dev *dev,
3493 			 const struct rte_flow_attr *attr,
3494 			 const struct rte_flow_action actions[])
3495 {
3496 	int queue_action = 0;
3497 	int action_n = 0;
3498 	int split = 0;
3499 	const struct rte_flow_action_queue *queue;
3500 	const struct rte_flow_action_rss *rss;
3501 	const struct rte_flow_action_raw_encap *raw_encap;
3502 	const struct rte_eth_hairpin_conf *conf;
3503 
3504 	if (!attr->ingress)
3505 		return 0;
3506 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3507 		switch (actions->type) {
3508 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3509 			queue = actions->conf;
3510 			if (queue == NULL)
3511 				return 0;
3512 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3513 			if (conf != NULL && !!conf->tx_explicit)
3514 				return 0;
3515 			queue_action = 1;
3516 			action_n++;
3517 			break;
3518 		case RTE_FLOW_ACTION_TYPE_RSS:
3519 			rss = actions->conf;
3520 			if (rss == NULL || rss->queue_num == 0)
3521 				return 0;
3522 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3523 			if (conf != NULL && !!conf->tx_explicit)
3524 				return 0;
3525 			queue_action = 1;
3526 			action_n++;
3527 			break;
3528 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3529 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3530 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3531 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3532 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3533 			split++;
3534 			action_n++;
3535 			break;
3536 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3537 			raw_encap = actions->conf;
3538 			if (raw_encap->size >
3539 			    (sizeof(struct rte_flow_item_eth) +
3540 			     sizeof(struct rte_flow_item_ipv4)))
3541 				split++;
3542 			action_n++;
3543 			break;
3544 		default:
3545 			action_n++;
3546 			break;
3547 		}
3548 	}
3549 	if (split && queue_action)
3550 		return action_n;
3551 	return 0;
3552 }
3553 
3554 /* Declare flow create/destroy prototype in advance. */
3555 static uint32_t
3556 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3557 		 const struct rte_flow_attr *attr,
3558 		 const struct rte_flow_item items[],
3559 		 const struct rte_flow_action actions[],
3560 		 bool external, struct rte_flow_error *error);
3561 
3562 static void
3563 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3564 		  uint32_t flow_idx);
3565 
3566 struct mlx5_hlist_entry *
3567 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3568 		       void *cb_ctx)
3569 {
3570 	struct rte_eth_dev *dev = list->ctx;
3571 	struct mlx5_priv *priv = dev->data->dev_private;
3572 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3573 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3574 	struct rte_flow_error *error = ctx->error;
3575 	uint32_t idx = 0;
3576 	int ret;
3577 	uint32_t mark_id = key;
3578 	struct rte_flow_attr attr = {
3579 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3580 		.ingress = 1,
3581 	};
3582 	struct mlx5_rte_flow_item_tag tag_spec = {
3583 		.data = mark_id,
3584 	};
3585 	struct rte_flow_item items[] = {
3586 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3587 	};
3588 	struct rte_flow_action_mark ftag = {
3589 		.id = mark_id,
3590 	};
3591 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3592 		.dst = REG_B,
3593 		.src = REG_NON,
3594 	};
3595 	struct rte_flow_action_jump jump = {
3596 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3597 	};
3598 	struct rte_flow_action actions[] = {
3599 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3600 	};
3601 
3602 	/* Fill the register fileds in the flow. */
3603 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3604 	if (ret < 0)
3605 		return NULL;
3606 	tag_spec.id = ret;
3607 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3608 	if (ret < 0)
3609 		return NULL;
3610 	cp_mreg.src = ret;
3611 	/* Provide the full width of FLAG specific value. */
3612 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3613 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3614 	/* Build a new flow. */
3615 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3616 		items[0] = (struct rte_flow_item){
3617 			.type = (enum rte_flow_item_type)
3618 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3619 			.spec = &tag_spec,
3620 		};
3621 		items[1] = (struct rte_flow_item){
3622 			.type = RTE_FLOW_ITEM_TYPE_END,
3623 		};
3624 		actions[0] = (struct rte_flow_action){
3625 			.type = (enum rte_flow_action_type)
3626 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3627 			.conf = &ftag,
3628 		};
3629 		actions[1] = (struct rte_flow_action){
3630 			.type = (enum rte_flow_action_type)
3631 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3632 			.conf = &cp_mreg,
3633 		};
3634 		actions[2] = (struct rte_flow_action){
3635 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3636 			.conf = &jump,
3637 		};
3638 		actions[3] = (struct rte_flow_action){
3639 			.type = RTE_FLOW_ACTION_TYPE_END,
3640 		};
3641 	} else {
3642 		/* Default rule, wildcard match. */
3643 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3644 		items[0] = (struct rte_flow_item){
3645 			.type = RTE_FLOW_ITEM_TYPE_END,
3646 		};
3647 		actions[0] = (struct rte_flow_action){
3648 			.type = (enum rte_flow_action_type)
3649 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3650 			.conf = &cp_mreg,
3651 		};
3652 		actions[1] = (struct rte_flow_action){
3653 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3654 			.conf = &jump,
3655 		};
3656 		actions[2] = (struct rte_flow_action){
3657 			.type = RTE_FLOW_ACTION_TYPE_END,
3658 		};
3659 	}
3660 	/* Build a new entry. */
3661 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3662 	if (!mcp_res) {
3663 		rte_errno = ENOMEM;
3664 		return NULL;
3665 	}
3666 	mcp_res->idx = idx;
3667 	/*
3668 	 * The copy Flows are not included in any list. There
3669 	 * ones are referenced from other Flows and can not
3670 	 * be applied, removed, deleted in ardbitrary order
3671 	 * by list traversing.
3672 	 */
3673 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3674 					 actions, false, error);
3675 	if (!mcp_res->rix_flow) {
3676 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3677 		return NULL;
3678 	}
3679 	return &mcp_res->hlist_ent;
3680 }
3681 
3682 /**
3683  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3684  *
3685  * As mark_id is unique, if there's already a registered flow for the mark_id,
3686  * return by increasing the reference counter of the resource. Otherwise, create
3687  * the resource (mcp_res) and flow.
3688  *
3689  * Flow looks like,
3690  *   - If ingress port is ANY and reg_c[1] is mark_id,
3691  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3692  *
3693  * For default flow (zero mark_id), flow is like,
3694  *   - If ingress port is ANY,
3695  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3696  *
3697  * @param dev
3698  *   Pointer to Ethernet device.
3699  * @param mark_id
3700  *   ID of MARK action, zero means default flow for META.
3701  * @param[out] error
3702  *   Perform verbose error reporting if not NULL.
3703  *
3704  * @return
3705  *   Associated resource on success, NULL otherwise and rte_errno is set.
3706  */
3707 static struct mlx5_flow_mreg_copy_resource *
3708 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3709 			  struct rte_flow_error *error)
3710 {
3711 	struct mlx5_priv *priv = dev->data->dev_private;
3712 	struct mlx5_hlist_entry *entry;
3713 	struct mlx5_flow_cb_ctx ctx = {
3714 		.dev = dev,
3715 		.error = error,
3716 	};
3717 
3718 	/* Check if already registered. */
3719 	MLX5_ASSERT(priv->mreg_cp_tbl);
3720 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3721 	if (!entry)
3722 		return NULL;
3723 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3724 			    hlist_ent);
3725 }
3726 
3727 void
3728 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3729 {
3730 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3731 		container_of(entry, typeof(*mcp_res), hlist_ent);
3732 	struct rte_eth_dev *dev = list->ctx;
3733 	struct mlx5_priv *priv = dev->data->dev_private;
3734 
3735 	MLX5_ASSERT(mcp_res->rix_flow);
3736 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3737 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3738 }
3739 
3740 /**
3741  * Release flow in RX_CP_TBL.
3742  *
3743  * @param dev
3744  *   Pointer to Ethernet device.
3745  * @flow
3746  *   Parent flow for wich copying is provided.
3747  */
3748 static void
3749 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3750 			  struct rte_flow *flow)
3751 {
3752 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3753 	struct mlx5_priv *priv = dev->data->dev_private;
3754 
3755 	if (!flow->rix_mreg_copy)
3756 		return;
3757 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3758 				 flow->rix_mreg_copy);
3759 	if (!mcp_res || !priv->mreg_cp_tbl)
3760 		return;
3761 	MLX5_ASSERT(mcp_res->rix_flow);
3762 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3763 	flow->rix_mreg_copy = 0;
3764 }
3765 
3766 /**
3767  * Remove the default copy action from RX_CP_TBL.
3768  *
3769  * This functions is called in the mlx5_dev_start(). No thread safe
3770  * is guaranteed.
3771  *
3772  * @param dev
3773  *   Pointer to Ethernet device.
3774  */
3775 static void
3776 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3777 {
3778 	struct mlx5_hlist_entry *entry;
3779 	struct mlx5_priv *priv = dev->data->dev_private;
3780 
3781 	/* Check if default flow is registered. */
3782 	if (!priv->mreg_cp_tbl)
3783 		return;
3784 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3785 				  MLX5_DEFAULT_COPY_ID, NULL);
3786 	if (!entry)
3787 		return;
3788 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3789 }
3790 
3791 /**
3792  * Add the default copy action in in RX_CP_TBL.
3793  *
3794  * This functions is called in the mlx5_dev_start(). No thread safe
3795  * is guaranteed.
3796  *
3797  * @param dev
3798  *   Pointer to Ethernet device.
3799  * @param[out] error
3800  *   Perform verbose error reporting if not NULL.
3801  *
3802  * @return
3803  *   0 for success, negative value otherwise and rte_errno is set.
3804  */
3805 static int
3806 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3807 				  struct rte_flow_error *error)
3808 {
3809 	struct mlx5_priv *priv = dev->data->dev_private;
3810 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3811 
3812 	/* Check whether extensive metadata feature is engaged. */
3813 	if (!priv->config.dv_flow_en ||
3814 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3815 	    !mlx5_flow_ext_mreg_supported(dev) ||
3816 	    !priv->sh->dv_regc0_mask)
3817 		return 0;
3818 	/*
3819 	 * Add default mreg copy flow may be called multiple time, but
3820 	 * only be called once in stop. Avoid register it twice.
3821 	 */
3822 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
3823 		return 0;
3824 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3825 	if (!mcp_res)
3826 		return -rte_errno;
3827 	return 0;
3828 }
3829 
3830 /**
3831  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3832  *
3833  * All the flow having Q/RSS action should be split by
3834  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3835  * performs the following,
3836  *   - CQE->flow_tag := reg_c[1] (MARK)
3837  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3838  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3839  * but there should be a flow per each MARK ID set by MARK action.
3840  *
3841  * For the aforementioned reason, if there's a MARK action in flow's action
3842  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3843  * the MARK ID to CQE's flow_tag like,
3844  *   - If reg_c[1] is mark_id,
3845  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3846  *
3847  * For SET_META action which stores value in reg_c[0], as the destination is
3848  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3849  * MARK ID means the default flow. The default flow looks like,
3850  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3851  *
3852  * @param dev
3853  *   Pointer to Ethernet device.
3854  * @param flow
3855  *   Pointer to flow structure.
3856  * @param[in] actions
3857  *   Pointer to the list of actions.
3858  * @param[out] error
3859  *   Perform verbose error reporting if not NULL.
3860  *
3861  * @return
3862  *   0 on success, negative value otherwise and rte_errno is set.
3863  */
3864 static int
3865 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3866 			    struct rte_flow *flow,
3867 			    const struct rte_flow_action *actions,
3868 			    struct rte_flow_error *error)
3869 {
3870 	struct mlx5_priv *priv = dev->data->dev_private;
3871 	struct mlx5_dev_config *config = &priv->config;
3872 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3873 	const struct rte_flow_action_mark *mark;
3874 
3875 	/* Check whether extensive metadata feature is engaged. */
3876 	if (!config->dv_flow_en ||
3877 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3878 	    !mlx5_flow_ext_mreg_supported(dev) ||
3879 	    !priv->sh->dv_regc0_mask)
3880 		return 0;
3881 	/* Find MARK action. */
3882 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3883 		switch (actions->type) {
3884 		case RTE_FLOW_ACTION_TYPE_FLAG:
3885 			mcp_res = flow_mreg_add_copy_action
3886 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3887 			if (!mcp_res)
3888 				return -rte_errno;
3889 			flow->rix_mreg_copy = mcp_res->idx;
3890 			return 0;
3891 		case RTE_FLOW_ACTION_TYPE_MARK:
3892 			mark = (const struct rte_flow_action_mark *)
3893 				actions->conf;
3894 			mcp_res =
3895 				flow_mreg_add_copy_action(dev, mark->id, error);
3896 			if (!mcp_res)
3897 				return -rte_errno;
3898 			flow->rix_mreg_copy = mcp_res->idx;
3899 			return 0;
3900 		default:
3901 			break;
3902 		}
3903 	}
3904 	return 0;
3905 }
3906 
3907 #define MLX5_MAX_SPLIT_ACTIONS 24
3908 #define MLX5_MAX_SPLIT_ITEMS 24
3909 
3910 /**
3911  * Split the hairpin flow.
3912  * Since HW can't support encap and push-vlan on Rx, we move these
3913  * actions to Tx.
3914  * If the count action is after the encap then we also
3915  * move the count action. in this case the count will also measure
3916  * the outer bytes.
3917  *
3918  * @param dev
3919  *   Pointer to Ethernet device.
3920  * @param[in] actions
3921  *   Associated actions (list terminated by the END action).
3922  * @param[out] actions_rx
3923  *   Rx flow actions.
3924  * @param[out] actions_tx
3925  *   Tx flow actions..
3926  * @param[out] pattern_tx
3927  *   The pattern items for the Tx flow.
3928  * @param[out] flow_id
3929  *   The flow ID connected to this flow.
3930  *
3931  * @return
3932  *   0 on success.
3933  */
3934 static int
3935 flow_hairpin_split(struct rte_eth_dev *dev,
3936 		   const struct rte_flow_action actions[],
3937 		   struct rte_flow_action actions_rx[],
3938 		   struct rte_flow_action actions_tx[],
3939 		   struct rte_flow_item pattern_tx[],
3940 		   uint32_t flow_id)
3941 {
3942 	const struct rte_flow_action_raw_encap *raw_encap;
3943 	const struct rte_flow_action_raw_decap *raw_decap;
3944 	struct mlx5_rte_flow_action_set_tag *set_tag;
3945 	struct rte_flow_action *tag_action;
3946 	struct mlx5_rte_flow_item_tag *tag_item;
3947 	struct rte_flow_item *item;
3948 	char *addr;
3949 	int encap = 0;
3950 
3951 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3952 		switch (actions->type) {
3953 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3954 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3955 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3956 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3957 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3958 			rte_memcpy(actions_tx, actions,
3959 			       sizeof(struct rte_flow_action));
3960 			actions_tx++;
3961 			break;
3962 		case RTE_FLOW_ACTION_TYPE_COUNT:
3963 			if (encap) {
3964 				rte_memcpy(actions_tx, actions,
3965 					   sizeof(struct rte_flow_action));
3966 				actions_tx++;
3967 			} else {
3968 				rte_memcpy(actions_rx, actions,
3969 					   sizeof(struct rte_flow_action));
3970 				actions_rx++;
3971 			}
3972 			break;
3973 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3974 			raw_encap = actions->conf;
3975 			if (raw_encap->size >
3976 			    (sizeof(struct rte_flow_item_eth) +
3977 			     sizeof(struct rte_flow_item_ipv4))) {
3978 				memcpy(actions_tx, actions,
3979 				       sizeof(struct rte_flow_action));
3980 				actions_tx++;
3981 				encap = 1;
3982 			} else {
3983 				rte_memcpy(actions_rx, actions,
3984 					   sizeof(struct rte_flow_action));
3985 				actions_rx++;
3986 			}
3987 			break;
3988 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3989 			raw_decap = actions->conf;
3990 			if (raw_decap->size <
3991 			    (sizeof(struct rte_flow_item_eth) +
3992 			     sizeof(struct rte_flow_item_ipv4))) {
3993 				memcpy(actions_tx, actions,
3994 				       sizeof(struct rte_flow_action));
3995 				actions_tx++;
3996 			} else {
3997 				rte_memcpy(actions_rx, actions,
3998 					   sizeof(struct rte_flow_action));
3999 				actions_rx++;
4000 			}
4001 			break;
4002 		default:
4003 			rte_memcpy(actions_rx, actions,
4004 				   sizeof(struct rte_flow_action));
4005 			actions_rx++;
4006 			break;
4007 		}
4008 	}
4009 	/* Add set meta action and end action for the Rx flow. */
4010 	tag_action = actions_rx;
4011 	tag_action->type = (enum rte_flow_action_type)
4012 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4013 	actions_rx++;
4014 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4015 	actions_rx++;
4016 	set_tag = (void *)actions_rx;
4017 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4018 	MLX5_ASSERT(set_tag->id > REG_NON);
4019 	set_tag->data = flow_id;
4020 	tag_action->conf = set_tag;
4021 	/* Create Tx item list. */
4022 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4023 	addr = (void *)&pattern_tx[2];
4024 	item = pattern_tx;
4025 	item->type = (enum rte_flow_item_type)
4026 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4027 	tag_item = (void *)addr;
4028 	tag_item->data = flow_id;
4029 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4030 	MLX5_ASSERT(set_tag->id > REG_NON);
4031 	item->spec = tag_item;
4032 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4033 	tag_item = (void *)addr;
4034 	tag_item->data = UINT32_MAX;
4035 	tag_item->id = UINT16_MAX;
4036 	item->mask = tag_item;
4037 	item->last = NULL;
4038 	item++;
4039 	item->type = RTE_FLOW_ITEM_TYPE_END;
4040 	return 0;
4041 }
4042 
4043 /**
4044  * The last stage of splitting chain, just creates the subflow
4045  * without any modification.
4046  *
4047  * @param[in] dev
4048  *   Pointer to Ethernet device.
4049  * @param[in] flow
4050  *   Parent flow structure pointer.
4051  * @param[in, out] sub_flow
4052  *   Pointer to return the created subflow, may be NULL.
4053  * @param[in] attr
4054  *   Flow rule attributes.
4055  * @param[in] items
4056  *   Pattern specification (list terminated by the END pattern item).
4057  * @param[in] actions
4058  *   Associated actions (list terminated by the END action).
4059  * @param[in] flow_split_info
4060  *   Pointer to flow split info structure.
4061  * @param[out] error
4062  *   Perform verbose error reporting if not NULL.
4063  * @return
4064  *   0 on success, negative value otherwise
4065  */
4066 static int
4067 flow_create_split_inner(struct rte_eth_dev *dev,
4068 			struct rte_flow *flow,
4069 			struct mlx5_flow **sub_flow,
4070 			const struct rte_flow_attr *attr,
4071 			const struct rte_flow_item items[],
4072 			const struct rte_flow_action actions[],
4073 			struct mlx5_flow_split_info *flow_split_info,
4074 			struct rte_flow_error *error)
4075 {
4076 	struct mlx5_flow *dev_flow;
4077 
4078 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4079 				    flow_split_info->flow_idx, error);
4080 	if (!dev_flow)
4081 		return -rte_errno;
4082 	dev_flow->flow = flow;
4083 	dev_flow->external = flow_split_info->external;
4084 	dev_flow->skip_scale = flow_split_info->skip_scale;
4085 	/* Subflow object was created, we must include one in the list. */
4086 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4087 		      dev_flow->handle, next);
4088 	/*
4089 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4090 	 * flow may need some user defined item layer flags, and pass the
4091 	 * Metadate rxq mark flag to suffix flow as well.
4092 	 */
4093 	if (flow_split_info->prefix_layers)
4094 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4095 	if (flow_split_info->prefix_mark)
4096 		dev_flow->handle->mark = 1;
4097 	if (sub_flow)
4098 		*sub_flow = dev_flow;
4099 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4100 }
4101 
4102 /**
4103  * Split the meter flow.
4104  *
4105  * As meter flow will split to three sub flow, other than meter
4106  * action, the other actions make sense to only meter accepts
4107  * the packet. If it need to be dropped, no other additional
4108  * actions should be take.
4109  *
4110  * One kind of special action which decapsulates the L3 tunnel
4111  * header will be in the prefix sub flow, as not to take the
4112  * L3 tunnel header into account.
4113  *
4114  * @param dev
4115  *   Pointer to Ethernet device.
4116  * @param[in] items
4117  *   Pattern specification (list terminated by the END pattern item).
4118  * @param[out] sfx_items
4119  *   Suffix flow match items (list terminated by the END pattern item).
4120  * @param[in] actions
4121  *   Associated actions (list terminated by the END action).
4122  * @param[out] actions_sfx
4123  *   Suffix flow actions.
4124  * @param[out] actions_pre
4125  *   Prefix flow actions.
4126  * @param[out] pattern_sfx
4127  *   The pattern items for the suffix flow.
4128  * @param[out] tag_sfx
4129  *   Pointer to suffix flow tag.
4130  *
4131  * @return
4132  *   0 on success.
4133  */
4134 static int
4135 flow_meter_split_prep(struct rte_eth_dev *dev,
4136 		 const struct rte_flow_item items[],
4137 		 struct rte_flow_item sfx_items[],
4138 		 const struct rte_flow_action actions[],
4139 		 struct rte_flow_action actions_sfx[],
4140 		 struct rte_flow_action actions_pre[])
4141 {
4142 	struct mlx5_priv *priv = dev->data->dev_private;
4143 	struct rte_flow_action *tag_action = NULL;
4144 	struct rte_flow_item *tag_item;
4145 	struct mlx5_rte_flow_action_set_tag *set_tag;
4146 	struct rte_flow_error error;
4147 	const struct rte_flow_action_raw_encap *raw_encap;
4148 	const struct rte_flow_action_raw_decap *raw_decap;
4149 	struct mlx5_rte_flow_item_tag *tag_spec;
4150 	struct mlx5_rte_flow_item_tag *tag_mask;
4151 	uint32_t tag_id = 0;
4152 	bool copy_vlan = false;
4153 
4154 	/* Prepare the actions for prefix and suffix flow. */
4155 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4156 		struct rte_flow_action **action_cur = NULL;
4157 
4158 		switch (actions->type) {
4159 		case RTE_FLOW_ACTION_TYPE_METER:
4160 			/* Add the extra tag action first. */
4161 			tag_action = actions_pre;
4162 			tag_action->type = (enum rte_flow_action_type)
4163 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4164 			actions_pre++;
4165 			action_cur = &actions_pre;
4166 			break;
4167 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4168 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4169 			action_cur = &actions_pre;
4170 			break;
4171 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4172 			raw_encap = actions->conf;
4173 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4174 				action_cur = &actions_pre;
4175 			break;
4176 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4177 			raw_decap = actions->conf;
4178 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4179 				action_cur = &actions_pre;
4180 			break;
4181 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4182 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4183 			copy_vlan = true;
4184 			break;
4185 		default:
4186 			break;
4187 		}
4188 		if (!action_cur)
4189 			action_cur = &actions_sfx;
4190 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4191 		(*action_cur)++;
4192 	}
4193 	/* Add end action to the actions. */
4194 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4195 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4196 	actions_pre++;
4197 	/* Set the tag. */
4198 	set_tag = (void *)actions_pre;
4199 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4200 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4201 			  &tag_id);
4202 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4203 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4204 			dev->data->port_id);
4205 		mlx5_ipool_free(priv->sh->ipool
4206 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4207 		return 0;
4208 	} else if (!tag_id) {
4209 		return 0;
4210 	}
4211 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4212 	assert(tag_action);
4213 	tag_action->conf = set_tag;
4214 	/* Prepare the suffix subflow items. */
4215 	tag_item = sfx_items++;
4216 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4217 		int item_type = items->type;
4218 
4219 		switch (item_type) {
4220 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4221 			memcpy(sfx_items, items, sizeof(*sfx_items));
4222 			sfx_items++;
4223 			break;
4224 		case RTE_FLOW_ITEM_TYPE_VLAN:
4225 			if (copy_vlan) {
4226 				memcpy(sfx_items, items, sizeof(*sfx_items));
4227 				/*
4228 				 * Convert to internal match item, it is used
4229 				 * for vlan push and set vid.
4230 				 */
4231 				sfx_items->type = (enum rte_flow_item_type)
4232 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4233 				sfx_items++;
4234 			}
4235 			break;
4236 		default:
4237 			break;
4238 		}
4239 	}
4240 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4241 	sfx_items++;
4242 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4243 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4244 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4245 	tag_mask = tag_spec + 1;
4246 	tag_mask->data = 0xffffff00;
4247 	tag_item->type = (enum rte_flow_item_type)
4248 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4249 	tag_item->spec = tag_spec;
4250 	tag_item->last = NULL;
4251 	tag_item->mask = tag_mask;
4252 	return tag_id;
4253 }
4254 
4255 /**
4256  * Split action list having QUEUE/RSS for metadata register copy.
4257  *
4258  * Once Q/RSS action is detected in user's action list, the flow action
4259  * should be split in order to copy metadata registers, which will happen in
4260  * RX_CP_TBL like,
4261  *   - CQE->flow_tag := reg_c[1] (MARK)
4262  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4263  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4264  * This is because the last action of each flow must be a terminal action
4265  * (QUEUE, RSS or DROP).
4266  *
4267  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4268  * stored and kept in the mlx5_flow structure per each sub_flow.
4269  *
4270  * The Q/RSS action is replaced with,
4271  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4272  * And the following JUMP action is added at the end,
4273  *   - JUMP, to RX_CP_TBL.
4274  *
4275  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4276  * flow_create_split_metadata() routine. The flow will look like,
4277  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4278  *
4279  * @param dev
4280  *   Pointer to Ethernet device.
4281  * @param[out] split_actions
4282  *   Pointer to store split actions to jump to CP_TBL.
4283  * @param[in] actions
4284  *   Pointer to the list of original flow actions.
4285  * @param[in] qrss
4286  *   Pointer to the Q/RSS action.
4287  * @param[in] actions_n
4288  *   Number of original actions.
4289  * @param[out] error
4290  *   Perform verbose error reporting if not NULL.
4291  *
4292  * @return
4293  *   non-zero unique flow_id on success, otherwise 0 and
4294  *   error/rte_error are set.
4295  */
4296 static uint32_t
4297 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4298 			  struct rte_flow_action *split_actions,
4299 			  const struct rte_flow_action *actions,
4300 			  const struct rte_flow_action *qrss,
4301 			  int actions_n, struct rte_flow_error *error)
4302 {
4303 	struct mlx5_priv *priv = dev->data->dev_private;
4304 	struct mlx5_rte_flow_action_set_tag *set_tag;
4305 	struct rte_flow_action_jump *jump;
4306 	const int qrss_idx = qrss - actions;
4307 	uint32_t flow_id = 0;
4308 	int ret = 0;
4309 
4310 	/*
4311 	 * Given actions will be split
4312 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4313 	 * - Add jump to mreg CP_TBL.
4314 	 * As a result, there will be one more action.
4315 	 */
4316 	++actions_n;
4317 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4318 	set_tag = (void *)(split_actions + actions_n);
4319 	/*
4320 	 * If tag action is not set to void(it means we are not the meter
4321 	 * suffix flow), add the tag action. Since meter suffix flow already
4322 	 * has the tag added.
4323 	 */
4324 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4325 		/*
4326 		 * Allocate the new subflow ID. This one is unique within
4327 		 * device and not shared with representors. Otherwise,
4328 		 * we would have to resolve multi-thread access synch
4329 		 * issue. Each flow on the shared device is appended
4330 		 * with source vport identifier, so the resulting
4331 		 * flows will be unique in the shared (by master and
4332 		 * representors) domain even if they have coinciding
4333 		 * IDs.
4334 		 */
4335 		mlx5_ipool_malloc(priv->sh->ipool
4336 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4337 		if (!flow_id)
4338 			return rte_flow_error_set(error, ENOMEM,
4339 						  RTE_FLOW_ERROR_TYPE_ACTION,
4340 						  NULL, "can't allocate id "
4341 						  "for split Q/RSS subflow");
4342 		/* Internal SET_TAG action to set flow ID. */
4343 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4344 			.data = flow_id,
4345 		};
4346 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4347 		if (ret < 0)
4348 			return ret;
4349 		set_tag->id = ret;
4350 		/* Construct new actions array. */
4351 		/* Replace QUEUE/RSS action. */
4352 		split_actions[qrss_idx] = (struct rte_flow_action){
4353 			.type = (enum rte_flow_action_type)
4354 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4355 			.conf = set_tag,
4356 		};
4357 	}
4358 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4359 	jump = (void *)(set_tag + 1);
4360 	*jump = (struct rte_flow_action_jump){
4361 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4362 	};
4363 	split_actions[actions_n - 2] = (struct rte_flow_action){
4364 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4365 		.conf = jump,
4366 	};
4367 	split_actions[actions_n - 1] = (struct rte_flow_action){
4368 		.type = RTE_FLOW_ACTION_TYPE_END,
4369 	};
4370 	return flow_id;
4371 }
4372 
4373 /**
4374  * Extend the given action list for Tx metadata copy.
4375  *
4376  * Copy the given action list to the ext_actions and add flow metadata register
4377  * copy action in order to copy reg_a set by WQE to reg_c[0].
4378  *
4379  * @param[out] ext_actions
4380  *   Pointer to the extended action list.
4381  * @param[in] actions
4382  *   Pointer to the list of actions.
4383  * @param[in] actions_n
4384  *   Number of actions in the list.
4385  * @param[out] error
4386  *   Perform verbose error reporting if not NULL.
4387  * @param[in] encap_idx
4388  *   The encap action inndex.
4389  *
4390  * @return
4391  *   0 on success, negative value otherwise
4392  */
4393 static int
4394 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4395 		       struct rte_flow_action *ext_actions,
4396 		       const struct rte_flow_action *actions,
4397 		       int actions_n, struct rte_flow_error *error,
4398 		       int encap_idx)
4399 {
4400 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4401 		(struct mlx5_flow_action_copy_mreg *)
4402 			(ext_actions + actions_n + 1);
4403 	int ret;
4404 
4405 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4406 	if (ret < 0)
4407 		return ret;
4408 	cp_mreg->dst = ret;
4409 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4410 	if (ret < 0)
4411 		return ret;
4412 	cp_mreg->src = ret;
4413 	if (encap_idx != 0)
4414 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4415 	if (encap_idx == actions_n - 1) {
4416 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4417 			.type = (enum rte_flow_action_type)
4418 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4419 			.conf = cp_mreg,
4420 		};
4421 		ext_actions[actions_n] = (struct rte_flow_action){
4422 			.type = RTE_FLOW_ACTION_TYPE_END,
4423 		};
4424 	} else {
4425 		ext_actions[encap_idx] = (struct rte_flow_action){
4426 			.type = (enum rte_flow_action_type)
4427 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4428 			.conf = cp_mreg,
4429 		};
4430 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4431 				sizeof(*ext_actions) * (actions_n - encap_idx));
4432 	}
4433 	return 0;
4434 }
4435 
4436 /**
4437  * Check the match action from the action list.
4438  *
4439  * @param[in] actions
4440  *   Pointer to the list of actions.
4441  * @param[in] attr
4442  *   Flow rule attributes.
4443  * @param[in] action
4444  *   The action to be check if exist.
4445  * @param[out] match_action_pos
4446  *   Pointer to the position of the matched action if exists, otherwise is -1.
4447  * @param[out] qrss_action_pos
4448  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4449  *
4450  * @return
4451  *   > 0 the total number of actions.
4452  *   0 if not found match action in action list.
4453  */
4454 static int
4455 flow_check_match_action(const struct rte_flow_action actions[],
4456 			const struct rte_flow_attr *attr,
4457 			enum rte_flow_action_type action,
4458 			int *match_action_pos, int *qrss_action_pos)
4459 {
4460 	const struct rte_flow_action_sample *sample;
4461 	int actions_n = 0;
4462 	int jump_flag = 0;
4463 	uint32_t ratio = 0;
4464 	int sub_type = 0;
4465 	int flag = 0;
4466 
4467 	*match_action_pos = -1;
4468 	*qrss_action_pos = -1;
4469 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4470 		if (actions->type == action) {
4471 			flag = 1;
4472 			*match_action_pos = actions_n;
4473 		}
4474 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4475 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4476 			*qrss_action_pos = actions_n;
4477 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4478 			jump_flag = 1;
4479 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4480 			sample = actions->conf;
4481 			ratio = sample->ratio;
4482 			sub_type = ((const struct rte_flow_action *)
4483 					(sample->actions))->type;
4484 		}
4485 		actions_n++;
4486 	}
4487 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4488 		if (ratio == 1) {
4489 			/* JUMP Action not support for Mirroring;
4490 			 * Mirroring support multi-destination;
4491 			 */
4492 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4493 				flag = 0;
4494 		}
4495 	}
4496 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4497 	return flag ? actions_n + 1 : 0;
4498 }
4499 
4500 #define SAMPLE_SUFFIX_ITEM 2
4501 
4502 /**
4503  * Split the sample flow.
4504  *
4505  * As sample flow will split to two sub flow, sample flow with
4506  * sample action, the other actions will move to new suffix flow.
4507  *
4508  * Also add unique tag id with tag action in the sample flow,
4509  * the same tag id will be as match in the suffix flow.
4510  *
4511  * @param dev
4512  *   Pointer to Ethernet device.
4513  * @param[in] fdb_tx
4514  *   FDB egress flow flag.
4515  * @param[out] sfx_items
4516  *   Suffix flow match items (list terminated by the END pattern item).
4517  * @param[in] actions
4518  *   Associated actions (list terminated by the END action).
4519  * @param[out] actions_sfx
4520  *   Suffix flow actions.
4521  * @param[out] actions_pre
4522  *   Prefix flow actions.
4523  * @param[in] actions_n
4524  *  The total number of actions.
4525  * @param[in] sample_action_pos
4526  *   The sample action position.
4527  * @param[in] qrss_action_pos
4528  *   The Queue/RSS action position.
4529  * @param[out] error
4530  *   Perform verbose error reporting if not NULL.
4531  *
4532  * @return
4533  *   0 on success, or unique flow_id, a negative errno value
4534  *   otherwise and rte_errno is set.
4535  */
4536 static int
4537 flow_sample_split_prep(struct rte_eth_dev *dev,
4538 		       uint32_t fdb_tx,
4539 		       struct rte_flow_item sfx_items[],
4540 		       const struct rte_flow_action actions[],
4541 		       struct rte_flow_action actions_sfx[],
4542 		       struct rte_flow_action actions_pre[],
4543 		       int actions_n,
4544 		       int sample_action_pos,
4545 		       int qrss_action_pos,
4546 		       struct rte_flow_error *error)
4547 {
4548 	struct mlx5_priv *priv = dev->data->dev_private;
4549 	struct mlx5_rte_flow_action_set_tag *set_tag;
4550 	struct mlx5_rte_flow_item_tag *tag_spec;
4551 	struct mlx5_rte_flow_item_tag *tag_mask;
4552 	uint32_t tag_id = 0;
4553 	int index;
4554 	int ret;
4555 
4556 	if (sample_action_pos < 0)
4557 		return rte_flow_error_set(error, EINVAL,
4558 					  RTE_FLOW_ERROR_TYPE_ACTION,
4559 					  NULL, "invalid position of sample "
4560 					  "action in list");
4561 	if (!fdb_tx) {
4562 		/* Prepare the prefix tag action. */
4563 		set_tag = (void *)(actions_pre + actions_n + 1);
4564 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4565 		if (ret < 0)
4566 			return ret;
4567 		set_tag->id = ret;
4568 		mlx5_ipool_malloc(priv->sh->ipool
4569 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4570 		set_tag->data = tag_id;
4571 		/* Prepare the suffix subflow items. */
4572 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4573 		tag_spec->data = tag_id;
4574 		tag_spec->id = set_tag->id;
4575 		tag_mask = tag_spec + 1;
4576 		tag_mask->data = UINT32_MAX;
4577 		sfx_items[0] = (struct rte_flow_item){
4578 			.type = (enum rte_flow_item_type)
4579 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4580 			.spec = tag_spec,
4581 			.last = NULL,
4582 			.mask = tag_mask,
4583 		};
4584 		sfx_items[1] = (struct rte_flow_item){
4585 			.type = (enum rte_flow_item_type)
4586 				RTE_FLOW_ITEM_TYPE_END,
4587 		};
4588 	}
4589 	/* Prepare the actions for prefix and suffix flow. */
4590 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4591 		index = qrss_action_pos;
4592 		/* Put the preceding the Queue/RSS action into prefix flow. */
4593 		if (index != 0)
4594 			memcpy(actions_pre, actions,
4595 			       sizeof(struct rte_flow_action) * index);
4596 		/* Put others preceding the sample action into prefix flow. */
4597 		if (sample_action_pos > index + 1)
4598 			memcpy(actions_pre + index, actions + index + 1,
4599 			       sizeof(struct rte_flow_action) *
4600 			       (sample_action_pos - index - 1));
4601 		index = sample_action_pos - 1;
4602 		/* Put Queue/RSS action into Suffix flow. */
4603 		memcpy(actions_sfx, actions + qrss_action_pos,
4604 		       sizeof(struct rte_flow_action));
4605 		actions_sfx++;
4606 	} else {
4607 		index = sample_action_pos;
4608 		if (index != 0)
4609 			memcpy(actions_pre, actions,
4610 			       sizeof(struct rte_flow_action) * index);
4611 	}
4612 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4613 	if (!fdb_tx) {
4614 		actions_pre[index++] =
4615 			(struct rte_flow_action){
4616 			.type = (enum rte_flow_action_type)
4617 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4618 			.conf = set_tag,
4619 		};
4620 	}
4621 	memcpy(actions_pre + index, actions + sample_action_pos,
4622 	       sizeof(struct rte_flow_action));
4623 	index += 1;
4624 	actions_pre[index] = (struct rte_flow_action){
4625 		.type = (enum rte_flow_action_type)
4626 			RTE_FLOW_ACTION_TYPE_END,
4627 	};
4628 	/* Put the actions after sample into Suffix flow. */
4629 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4630 	       sizeof(struct rte_flow_action) *
4631 	       (actions_n - sample_action_pos - 1));
4632 	return tag_id;
4633 }
4634 
4635 /**
4636  * The splitting for metadata feature.
4637  *
4638  * - Q/RSS action on NIC Rx should be split in order to pass by
4639  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4640  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4641  *
4642  * - All the actions on NIC Tx should have a mreg copy action to
4643  *   copy reg_a from WQE to reg_c[0].
4644  *
4645  * @param dev
4646  *   Pointer to Ethernet device.
4647  * @param[in] flow
4648  *   Parent flow structure pointer.
4649  * @param[in] attr
4650  *   Flow rule attributes.
4651  * @param[in] items
4652  *   Pattern specification (list terminated by the END pattern item).
4653  * @param[in] actions
4654  *   Associated actions (list terminated by the END action).
4655  * @param[in] flow_split_info
4656  *   Pointer to flow split info structure.
4657  * @param[out] error
4658  *   Perform verbose error reporting if not NULL.
4659  * @return
4660  *   0 on success, negative value otherwise
4661  */
4662 static int
4663 flow_create_split_metadata(struct rte_eth_dev *dev,
4664 			   struct rte_flow *flow,
4665 			   const struct rte_flow_attr *attr,
4666 			   const struct rte_flow_item items[],
4667 			   const struct rte_flow_action actions[],
4668 			   struct mlx5_flow_split_info *flow_split_info,
4669 			   struct rte_flow_error *error)
4670 {
4671 	struct mlx5_priv *priv = dev->data->dev_private;
4672 	struct mlx5_dev_config *config = &priv->config;
4673 	const struct rte_flow_action *qrss = NULL;
4674 	struct rte_flow_action *ext_actions = NULL;
4675 	struct mlx5_flow *dev_flow = NULL;
4676 	uint32_t qrss_id = 0;
4677 	int mtr_sfx = 0;
4678 	size_t act_size;
4679 	int actions_n;
4680 	int encap_idx;
4681 	int ret;
4682 
4683 	/* Check whether extensive metadata feature is engaged. */
4684 	if (!config->dv_flow_en ||
4685 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4686 	    !mlx5_flow_ext_mreg_supported(dev))
4687 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4688 					       actions, flow_split_info, error);
4689 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4690 							   &encap_idx);
4691 	if (qrss) {
4692 		/* Exclude hairpin flows from splitting. */
4693 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4694 			const struct rte_flow_action_queue *queue;
4695 
4696 			queue = qrss->conf;
4697 			if (mlx5_rxq_get_type(dev, queue->index) ==
4698 			    MLX5_RXQ_TYPE_HAIRPIN)
4699 				qrss = NULL;
4700 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4701 			const struct rte_flow_action_rss *rss;
4702 
4703 			rss = qrss->conf;
4704 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4705 			    MLX5_RXQ_TYPE_HAIRPIN)
4706 				qrss = NULL;
4707 		}
4708 	}
4709 	if (qrss) {
4710 		/* Check if it is in meter suffix table. */
4711 		mtr_sfx = attr->group == (attr->transfer ?
4712 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4713 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4714 		/*
4715 		 * Q/RSS action on NIC Rx should be split in order to pass by
4716 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4717 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4718 		 */
4719 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4720 			   sizeof(struct rte_flow_action_set_tag) +
4721 			   sizeof(struct rte_flow_action_jump);
4722 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4723 					  SOCKET_ID_ANY);
4724 		if (!ext_actions)
4725 			return rte_flow_error_set(error, ENOMEM,
4726 						  RTE_FLOW_ERROR_TYPE_ACTION,
4727 						  NULL, "no memory to split "
4728 						  "metadata flow");
4729 		/*
4730 		 * If we are the suffix flow of meter, tag already exist.
4731 		 * Set the tag action to void.
4732 		 */
4733 		if (mtr_sfx)
4734 			ext_actions[qrss - actions].type =
4735 						RTE_FLOW_ACTION_TYPE_VOID;
4736 		else
4737 			ext_actions[qrss - actions].type =
4738 						(enum rte_flow_action_type)
4739 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4740 		/*
4741 		 * Create the new actions list with removed Q/RSS action
4742 		 * and appended set tag and jump to register copy table
4743 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4744 		 * in advance, because it is needed for set tag action.
4745 		 */
4746 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4747 						    qrss, actions_n, error);
4748 		if (!mtr_sfx && !qrss_id) {
4749 			ret = -rte_errno;
4750 			goto exit;
4751 		}
4752 	} else if (attr->egress && !attr->transfer) {
4753 		/*
4754 		 * All the actions on NIC Tx should have a metadata register
4755 		 * copy action to copy reg_a from WQE to reg_c[meta]
4756 		 */
4757 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4758 			   sizeof(struct mlx5_flow_action_copy_mreg);
4759 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4760 					  SOCKET_ID_ANY);
4761 		if (!ext_actions)
4762 			return rte_flow_error_set(error, ENOMEM,
4763 						  RTE_FLOW_ERROR_TYPE_ACTION,
4764 						  NULL, "no memory to split "
4765 						  "metadata flow");
4766 		/* Create the action list appended with copy register. */
4767 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4768 					     actions_n, error, encap_idx);
4769 		if (ret < 0)
4770 			goto exit;
4771 	}
4772 	/* Add the unmodified original or prefix subflow. */
4773 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
4774 				      items, ext_actions ? ext_actions :
4775 				      actions, flow_split_info, error);
4776 	if (ret < 0)
4777 		goto exit;
4778 	MLX5_ASSERT(dev_flow);
4779 	if (qrss) {
4780 		const struct rte_flow_attr q_attr = {
4781 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4782 			.ingress = 1,
4783 		};
4784 		/* Internal PMD action to set register. */
4785 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4786 			.data = qrss_id,
4787 			.id = REG_NON,
4788 		};
4789 		struct rte_flow_item q_items[] = {
4790 			{
4791 				.type = (enum rte_flow_item_type)
4792 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4793 				.spec = &q_tag_spec,
4794 				.last = NULL,
4795 				.mask = NULL,
4796 			},
4797 			{
4798 				.type = RTE_FLOW_ITEM_TYPE_END,
4799 			},
4800 		};
4801 		struct rte_flow_action q_actions[] = {
4802 			{
4803 				.type = qrss->type,
4804 				.conf = qrss->conf,
4805 			},
4806 			{
4807 				.type = RTE_FLOW_ACTION_TYPE_END,
4808 			},
4809 		};
4810 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4811 
4812 		/*
4813 		 * Configure the tag item only if there is no meter subflow.
4814 		 * Since tag is already marked in the meter suffix subflow
4815 		 * we can just use the meter suffix items as is.
4816 		 */
4817 		if (qrss_id) {
4818 			/* Not meter subflow. */
4819 			MLX5_ASSERT(!mtr_sfx);
4820 			/*
4821 			 * Put unique id in prefix flow due to it is destroyed
4822 			 * after suffix flow and id will be freed after there
4823 			 * is no actual flows with this id and identifier
4824 			 * reallocation becomes possible (for example, for
4825 			 * other flows in other threads).
4826 			 */
4827 			dev_flow->handle->split_flow_id = qrss_id;
4828 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4829 						   error);
4830 			if (ret < 0)
4831 				goto exit;
4832 			q_tag_spec.id = ret;
4833 		}
4834 		dev_flow = NULL;
4835 		/* Add suffix subflow to execute Q/RSS. */
4836 		flow_split_info->prefix_layers = layers;
4837 		flow_split_info->prefix_mark = 0;
4838 		ret = flow_create_split_inner(dev, flow, &dev_flow,
4839 					      &q_attr, mtr_sfx ? items :
4840 					      q_items, q_actions,
4841 					      flow_split_info, error);
4842 		if (ret < 0)
4843 			goto exit;
4844 		/* qrss ID should be freed if failed. */
4845 		qrss_id = 0;
4846 		MLX5_ASSERT(dev_flow);
4847 	}
4848 
4849 exit:
4850 	/*
4851 	 * We do not destroy the partially created sub_flows in case of error.
4852 	 * These ones are included into parent flow list and will be destroyed
4853 	 * by flow_drv_destroy.
4854 	 */
4855 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4856 			qrss_id);
4857 	mlx5_free(ext_actions);
4858 	return ret;
4859 }
4860 
4861 /**
4862  * The splitting for meter feature.
4863  *
4864  * - The meter flow will be split to two flows as prefix and
4865  *   suffix flow. The packets make sense only it pass the prefix
4866  *   meter action.
4867  *
4868  * - Reg_C_5 is used for the packet to match betweend prefix and
4869  *   suffix flow.
4870  *
4871  * @param dev
4872  *   Pointer to Ethernet device.
4873  * @param[in] flow
4874  *   Parent flow structure pointer.
4875  * @param[in] attr
4876  *   Flow rule attributes.
4877  * @param[in] items
4878  *   Pattern specification (list terminated by the END pattern item).
4879  * @param[in] actions
4880  *   Associated actions (list terminated by the END action).
4881  * @param[in] flow_split_info
4882  *   Pointer to flow split info structure.
4883  * @param[out] error
4884  *   Perform verbose error reporting if not NULL.
4885  * @return
4886  *   0 on success, negative value otherwise
4887  */
4888 static int
4889 flow_create_split_meter(struct rte_eth_dev *dev,
4890 			struct rte_flow *flow,
4891 			const struct rte_flow_attr *attr,
4892 			const struct rte_flow_item items[],
4893 			const struct rte_flow_action actions[],
4894 			struct mlx5_flow_split_info *flow_split_info,
4895 			struct rte_flow_error *error)
4896 {
4897 	struct mlx5_priv *priv = dev->data->dev_private;
4898 	struct rte_flow_action *sfx_actions = NULL;
4899 	struct rte_flow_action *pre_actions = NULL;
4900 	struct rte_flow_item *sfx_items = NULL;
4901 	struct mlx5_flow *dev_flow = NULL;
4902 	struct rte_flow_attr sfx_attr = *attr;
4903 	uint32_t mtr = 0;
4904 	uint32_t mtr_tag_id = 0;
4905 	size_t act_size;
4906 	size_t item_size;
4907 	int actions_n = 0;
4908 	int ret;
4909 
4910 	if (priv->mtr_en)
4911 		actions_n = flow_check_meter_action(actions, &mtr);
4912 	if (mtr) {
4913 		/* The five prefix actions: meter, decap, encap, tag, end. */
4914 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4915 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4916 		/* tag, vlan, port id, end. */
4917 #define METER_SUFFIX_ITEM 4
4918 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4919 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4920 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4921 					  0, SOCKET_ID_ANY);
4922 		if (!sfx_actions)
4923 			return rte_flow_error_set(error, ENOMEM,
4924 						  RTE_FLOW_ERROR_TYPE_ACTION,
4925 						  NULL, "no memory to split "
4926 						  "meter flow");
4927 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4928 			     act_size);
4929 		pre_actions = sfx_actions + actions_n;
4930 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4931 						   actions, sfx_actions,
4932 						   pre_actions);
4933 		if (!mtr_tag_id) {
4934 			ret = -rte_errno;
4935 			goto exit;
4936 		}
4937 		/* Add the prefix subflow. */
4938 		flow_split_info->prefix_mark = 0;
4939 		ret = flow_create_split_inner(dev, flow, &dev_flow,
4940 					      attr, items, pre_actions,
4941 					      flow_split_info, error);
4942 		if (ret) {
4943 			ret = -rte_errno;
4944 			goto exit;
4945 		}
4946 		dev_flow->handle->split_flow_id = mtr_tag_id;
4947 		/* Setting the sfx group atrr. */
4948 		sfx_attr.group = sfx_attr.transfer ?
4949 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4950 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4951 		flow_split_info->prefix_layers =
4952 				flow_get_prefix_layer_flags(dev_flow);
4953 		flow_split_info->prefix_mark = dev_flow->handle->mark;
4954 	}
4955 	/* Add the prefix subflow. */
4956 	ret = flow_create_split_metadata(dev, flow,
4957 					 &sfx_attr, sfx_items ?
4958 					 sfx_items : items,
4959 					 sfx_actions ? sfx_actions : actions,
4960 					 flow_split_info, error);
4961 exit:
4962 	if (sfx_actions)
4963 		mlx5_free(sfx_actions);
4964 	return ret;
4965 }
4966 
4967 /**
4968  * The splitting for sample feature.
4969  *
4970  * Once Sample action is detected in the action list, the flow actions should
4971  * be split into prefix sub flow and suffix sub flow.
4972  *
4973  * The original items remain in the prefix sub flow, all actions preceding the
4974  * sample action and the sample action itself will be copied to the prefix
4975  * sub flow, the actions following the sample action will be copied to the
4976  * suffix sub flow, Queue action always be located in the suffix sub flow.
4977  *
4978  * In order to make the packet from prefix sub flow matches with suffix sub
4979  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
4980  * flow uses tag item with the unique flow id.
4981  *
4982  * @param dev
4983  *   Pointer to Ethernet device.
4984  * @param[in] flow
4985  *   Parent flow structure pointer.
4986  * @param[in] attr
4987  *   Flow rule attributes.
4988  * @param[in] items
4989  *   Pattern specification (list terminated by the END pattern item).
4990  * @param[in] actions
4991  *   Associated actions (list terminated by the END action).
4992  * @param[in] flow_split_info
4993  *   Pointer to flow split info structure.
4994  * @param[out] error
4995  *   Perform verbose error reporting if not NULL.
4996  * @return
4997  *   0 on success, negative value otherwise
4998  */
4999 static int
5000 flow_create_split_sample(struct rte_eth_dev *dev,
5001 			 struct rte_flow *flow,
5002 			 const struct rte_flow_attr *attr,
5003 			 const struct rte_flow_item items[],
5004 			 const struct rte_flow_action actions[],
5005 			 struct mlx5_flow_split_info *flow_split_info,
5006 			 struct rte_flow_error *error)
5007 {
5008 	struct mlx5_priv *priv = dev->data->dev_private;
5009 	struct rte_flow_action *sfx_actions = NULL;
5010 	struct rte_flow_action *pre_actions = NULL;
5011 	struct rte_flow_item *sfx_items = NULL;
5012 	struct mlx5_flow *dev_flow = NULL;
5013 	struct rte_flow_attr sfx_attr = *attr;
5014 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5015 	struct mlx5_flow_dv_sample_resource *sample_res;
5016 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5017 	struct mlx5_flow_tbl_resource *sfx_tbl;
5018 	union mlx5_flow_tbl_key sfx_table_key;
5019 #endif
5020 	size_t act_size;
5021 	size_t item_size;
5022 	uint32_t fdb_tx = 0;
5023 	int32_t tag_id = 0;
5024 	int actions_n = 0;
5025 	int sample_action_pos;
5026 	int qrss_action_pos;
5027 	int ret = 0;
5028 
5029 	if (priv->sampler_en)
5030 		actions_n = flow_check_match_action(actions, attr,
5031 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5032 					&sample_action_pos, &qrss_action_pos);
5033 	if (actions_n) {
5034 		/* The prefix actions must includes sample, tag, end. */
5035 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5036 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5037 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5038 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5039 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5040 					  item_size), 0, SOCKET_ID_ANY);
5041 		if (!sfx_actions)
5042 			return rte_flow_error_set(error, ENOMEM,
5043 						  RTE_FLOW_ERROR_TYPE_ACTION,
5044 						  NULL, "no memory to split "
5045 						  "sample flow");
5046 		/* The representor_id is -1 for uplink. */
5047 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5048 		if (!fdb_tx)
5049 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5050 					+ act_size);
5051 		pre_actions = sfx_actions + actions_n;
5052 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5053 						actions, sfx_actions,
5054 						pre_actions, actions_n,
5055 						sample_action_pos,
5056 						qrss_action_pos, error);
5057 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5058 			ret = -rte_errno;
5059 			goto exit;
5060 		}
5061 		/* Add the prefix subflow. */
5062 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5063 					      items, pre_actions,
5064 					      flow_split_info, error);
5065 		if (ret) {
5066 			ret = -rte_errno;
5067 			goto exit;
5068 		}
5069 		dev_flow->handle->split_flow_id = tag_id;
5070 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5071 		/* Set the sfx group attr. */
5072 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5073 					dev_flow->dv.sample_res;
5074 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5075 					sample_res->normal_path_tbl;
5076 		sfx_tbl_data = container_of(sfx_tbl,
5077 					struct mlx5_flow_tbl_data_entry, tbl);
5078 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5079 		sfx_attr.group = sfx_attr.transfer ?
5080 					(sfx_table_key.table_id - 1) :
5081 					 sfx_table_key.table_id;
5082 		flow_split_info->prefix_layers =
5083 				flow_get_prefix_layer_flags(dev_flow);
5084 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5085 		/* Suffix group level already be scaled with factor, set
5086 		 * skip_scale to 1 to avoid scale again in translation.
5087 		 */
5088 		flow_split_info->skip_scale = 1;
5089 #endif
5090 	}
5091 	/* Add the suffix subflow. */
5092 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5093 				      sfx_items ? sfx_items : items,
5094 				      sfx_actions ? sfx_actions : actions,
5095 				      flow_split_info, error);
5096 exit:
5097 	if (sfx_actions)
5098 		mlx5_free(sfx_actions);
5099 	return ret;
5100 }
5101 
5102 /**
5103  * Split the flow to subflow set. The splitters might be linked
5104  * in the chain, like this:
5105  * flow_create_split_outer() calls:
5106  *   flow_create_split_meter() calls:
5107  *     flow_create_split_metadata(meter_subflow_0) calls:
5108  *       flow_create_split_inner(metadata_subflow_0)
5109  *       flow_create_split_inner(metadata_subflow_1)
5110  *       flow_create_split_inner(metadata_subflow_2)
5111  *     flow_create_split_metadata(meter_subflow_1) calls:
5112  *       flow_create_split_inner(metadata_subflow_0)
5113  *       flow_create_split_inner(metadata_subflow_1)
5114  *       flow_create_split_inner(metadata_subflow_2)
5115  *
5116  * This provide flexible way to add new levels of flow splitting.
5117  * The all of successfully created subflows are included to the
5118  * parent flow dev_flow list.
5119  *
5120  * @param dev
5121  *   Pointer to Ethernet device.
5122  * @param[in] flow
5123  *   Parent flow structure pointer.
5124  * @param[in] attr
5125  *   Flow rule attributes.
5126  * @param[in] items
5127  *   Pattern specification (list terminated by the END pattern item).
5128  * @param[in] actions
5129  *   Associated actions (list terminated by the END action).
5130  * @param[in] flow_split_info
5131  *   Pointer to flow split info structure.
5132  * @param[out] error
5133  *   Perform verbose error reporting if not NULL.
5134  * @return
5135  *   0 on success, negative value otherwise
5136  */
5137 static int
5138 flow_create_split_outer(struct rte_eth_dev *dev,
5139 			struct rte_flow *flow,
5140 			const struct rte_flow_attr *attr,
5141 			const struct rte_flow_item items[],
5142 			const struct rte_flow_action actions[],
5143 			struct mlx5_flow_split_info *flow_split_info,
5144 			struct rte_flow_error *error)
5145 {
5146 	int ret;
5147 
5148 	ret = flow_create_split_sample(dev, flow, attr, items,
5149 				       actions, flow_split_info, error);
5150 	MLX5_ASSERT(ret <= 0);
5151 	return ret;
5152 }
5153 
5154 static struct mlx5_flow_tunnel *
5155 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5156 		      const struct rte_flow_attr *attr,
5157 		      const struct rte_flow_item items[],
5158 		      const struct rte_flow_action actions[])
5159 {
5160 	struct mlx5_flow_tunnel *tunnel;
5161 
5162 #pragma GCC diagnostic push
5163 #pragma GCC diagnostic ignored "-Wcast-qual"
5164 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5165 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5166 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5167 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5168 	else
5169 		tunnel = NULL;
5170 #pragma GCC diagnostic pop
5171 
5172 	return tunnel;
5173 }
5174 
5175 /**
5176  * Adjust flow RSS workspace if needed.
5177  *
5178  * @param wks
5179  *   Pointer to thread flow work space.
5180  * @param rss_desc
5181  *   Pointer to RSS descriptor.
5182  * @param[in] nrssq_num
5183  *   New RSS queue number.
5184  *
5185  * @return
5186  *   0 on success, -1 otherwise and rte_errno is set.
5187  */
5188 static int
5189 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5190 			  struct mlx5_flow_rss_desc *rss_desc,
5191 			  uint32_t nrssq_num)
5192 {
5193 	if (likely(nrssq_num <= wks->rssq_num))
5194 		return 0;
5195 	rss_desc->queue = realloc(rss_desc->queue,
5196 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5197 	if (!rss_desc->queue) {
5198 		rte_errno = ENOMEM;
5199 		return -1;
5200 	}
5201 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5202 	return 0;
5203 }
5204 
5205 /**
5206  * Create a flow and add it to @p list.
5207  *
5208  * @param dev
5209  *   Pointer to Ethernet device.
5210  * @param list
5211  *   Pointer to a TAILQ flow list. If this parameter NULL,
5212  *   no list insertion occurred, flow is just created,
5213  *   this is caller's responsibility to track the
5214  *   created flow.
5215  * @param[in] attr
5216  *   Flow rule attributes.
5217  * @param[in] items
5218  *   Pattern specification (list terminated by the END pattern item).
5219  * @param[in] actions
5220  *   Associated actions (list terminated by the END action).
5221  * @param[in] external
5222  *   This flow rule is created by request external to PMD.
5223  * @param[out] error
5224  *   Perform verbose error reporting if not NULL.
5225  *
5226  * @return
5227  *   A flow index on success, 0 otherwise and rte_errno is set.
5228  */
5229 static uint32_t
5230 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5231 		 const struct rte_flow_attr *attr,
5232 		 const struct rte_flow_item items[],
5233 		 const struct rte_flow_action original_actions[],
5234 		 bool external, struct rte_flow_error *error)
5235 {
5236 	struct mlx5_priv *priv = dev->data->dev_private;
5237 	struct rte_flow *flow = NULL;
5238 	struct mlx5_flow *dev_flow;
5239 	const struct rte_flow_action_rss *rss;
5240 	struct mlx5_translated_shared_action
5241 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5242 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5243 	union {
5244 		struct mlx5_flow_expand_rss buf;
5245 		uint8_t buffer[2048];
5246 	} expand_buffer;
5247 	union {
5248 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5249 		uint8_t buffer[2048];
5250 	} actions_rx;
5251 	union {
5252 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5253 		uint8_t buffer[2048];
5254 	} actions_hairpin_tx;
5255 	union {
5256 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5257 		uint8_t buffer[2048];
5258 	} items_tx;
5259 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5260 	struct mlx5_flow_rss_desc *rss_desc;
5261 	const struct rte_flow_action *p_actions_rx;
5262 	uint32_t i;
5263 	uint32_t idx = 0;
5264 	int hairpin_flow;
5265 	struct rte_flow_attr attr_tx = { .priority = 0 };
5266 	const struct rte_flow_action *actions;
5267 	struct rte_flow_action *translated_actions = NULL;
5268 	struct mlx5_flow_tunnel *tunnel;
5269 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5270 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5271 	struct mlx5_flow_split_info flow_split_info = {
5272 		.external = !!external,
5273 		.skip_scale = 0,
5274 		.flow_idx = 0,
5275 		.prefix_mark = 0,
5276 		.prefix_layers = 0
5277 	};
5278 	int ret;
5279 
5280 	MLX5_ASSERT(wks);
5281 	rss_desc = &wks->rss_desc;
5282 	ret = flow_shared_actions_translate(dev, original_actions,
5283 					    shared_actions,
5284 					    &shared_actions_n,
5285 					    &translated_actions, error);
5286 	if (ret < 0) {
5287 		MLX5_ASSERT(translated_actions == NULL);
5288 		return 0;
5289 	}
5290 	actions = translated_actions ? translated_actions : original_actions;
5291 	p_actions_rx = actions;
5292 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5293 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5294 				external, hairpin_flow, error);
5295 	if (ret < 0)
5296 		goto error_before_hairpin_split;
5297 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5298 	if (!flow) {
5299 		rte_errno = ENOMEM;
5300 		goto error_before_hairpin_split;
5301 	}
5302 	if (hairpin_flow > 0) {
5303 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5304 			rte_errno = EINVAL;
5305 			goto error_before_hairpin_split;
5306 		}
5307 		flow_hairpin_split(dev, actions, actions_rx.actions,
5308 				   actions_hairpin_tx.actions, items_tx.items,
5309 				   idx);
5310 		p_actions_rx = actions_rx.actions;
5311 	}
5312 	flow_split_info.flow_idx = idx;
5313 	flow->drv_type = flow_get_drv_type(dev, attr);
5314 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5315 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5316 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5317 	rss = flow_get_rss_action(p_actions_rx);
5318 	if (rss) {
5319 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5320 			return 0;
5321 		/*
5322 		 * The following information is required by
5323 		 * mlx5_flow_hashfields_adjust() in advance.
5324 		 */
5325 		rss_desc->level = rss->level;
5326 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5327 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5328 	}
5329 	flow->dev_handles = 0;
5330 	if (rss && rss->types) {
5331 		unsigned int graph_root;
5332 
5333 		graph_root = find_graph_root(items, rss->level);
5334 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5335 					   items, rss->types,
5336 					   mlx5_support_expansion, graph_root);
5337 		MLX5_ASSERT(ret > 0 &&
5338 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5339 	} else {
5340 		buf->entries = 1;
5341 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5342 	}
5343 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5344 						      shared_actions_n);
5345 	for (i = 0; i < buf->entries; ++i) {
5346 		/* Initialize flow split data. */
5347 		flow_split_info.prefix_layers = 0;
5348 		flow_split_info.prefix_mark = 0;
5349 		flow_split_info.skip_scale = 0;
5350 		/*
5351 		 * The splitter may create multiple dev_flows,
5352 		 * depending on configuration. In the simplest
5353 		 * case it just creates unmodified original flow.
5354 		 */
5355 		ret = flow_create_split_outer(dev, flow, attr,
5356 					      buf->entry[i].pattern,
5357 					      p_actions_rx, &flow_split_info,
5358 					      error);
5359 		if (ret < 0)
5360 			goto error;
5361 		if (is_flow_tunnel_steer_rule(dev, attr,
5362 					      buf->entry[i].pattern,
5363 					      p_actions_rx)) {
5364 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5365 							   p_actions_rx,
5366 							   idx,
5367 							   &default_miss_ctx,
5368 							   error);
5369 			if (ret < 0) {
5370 				mlx5_free(default_miss_ctx.queue);
5371 				goto error;
5372 			}
5373 		}
5374 	}
5375 	/* Create the tx flow. */
5376 	if (hairpin_flow) {
5377 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5378 		attr_tx.ingress = 0;
5379 		attr_tx.egress = 1;
5380 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5381 					 actions_hairpin_tx.actions,
5382 					 idx, error);
5383 		if (!dev_flow)
5384 			goto error;
5385 		dev_flow->flow = flow;
5386 		dev_flow->external = 0;
5387 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5388 			      dev_flow->handle, next);
5389 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5390 					 items_tx.items,
5391 					 actions_hairpin_tx.actions, error);
5392 		if (ret < 0)
5393 			goto error;
5394 	}
5395 	/*
5396 	 * Update the metadata register copy table. If extensive
5397 	 * metadata feature is enabled and registers are supported
5398 	 * we might create the extra rte_flow for each unique
5399 	 * MARK/FLAG action ID.
5400 	 *
5401 	 * The table is updated for ingress Flows only, because
5402 	 * the egress Flows belong to the different device and
5403 	 * copy table should be updated in peer NIC Rx domain.
5404 	 */
5405 	if (attr->ingress &&
5406 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5407 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5408 		if (ret)
5409 			goto error;
5410 	}
5411 	/*
5412 	 * If the flow is external (from application) OR device is started,
5413 	 * OR mreg discover, then apply immediately.
5414 	 */
5415 	if (external || dev->data->dev_started ||
5416 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5417 	     attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5418 		ret = flow_drv_apply(dev, flow, error);
5419 		if (ret < 0)
5420 			goto error;
5421 	}
5422 	if (list) {
5423 		rte_spinlock_lock(&priv->flow_list_lock);
5424 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5425 			     flow, next);
5426 		rte_spinlock_unlock(&priv->flow_list_lock);
5427 	}
5428 	flow_rxq_flags_set(dev, flow);
5429 	rte_free(translated_actions);
5430 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5431 	if (tunnel) {
5432 		flow->tunnel = 1;
5433 		flow->tunnel_id = tunnel->tunnel_id;
5434 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5435 		mlx5_free(default_miss_ctx.queue);
5436 	}
5437 	mlx5_flow_pop_thread_workspace();
5438 	return idx;
5439 error:
5440 	MLX5_ASSERT(flow);
5441 	ret = rte_errno; /* Save rte_errno before cleanup. */
5442 	flow_mreg_del_copy_action(dev, flow);
5443 	flow_drv_destroy(dev, flow);
5444 	if (rss_desc->shared_rss)
5445 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5446 			mlx5_ipool_get
5447 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5448 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5449 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5450 	rte_errno = ret; /* Restore rte_errno. */
5451 	ret = rte_errno;
5452 	rte_errno = ret;
5453 	mlx5_flow_pop_thread_workspace();
5454 error_before_hairpin_split:
5455 	rte_free(translated_actions);
5456 	return 0;
5457 }
5458 
5459 /**
5460  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5461  * incoming packets to table 1.
5462  *
5463  * Other flow rules, requested for group n, will be created in
5464  * e-switch table n+1.
5465  * Jump action to e-switch group n will be created to group n+1.
5466  *
5467  * Used when working in switchdev mode, to utilise advantages of table 1
5468  * and above.
5469  *
5470  * @param dev
5471  *   Pointer to Ethernet device.
5472  *
5473  * @return
5474  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5475  */
5476 struct rte_flow *
5477 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5478 {
5479 	const struct rte_flow_attr attr = {
5480 		.group = 0,
5481 		.priority = 0,
5482 		.ingress = 1,
5483 		.egress = 0,
5484 		.transfer = 1,
5485 	};
5486 	const struct rte_flow_item pattern = {
5487 		.type = RTE_FLOW_ITEM_TYPE_END,
5488 	};
5489 	struct rte_flow_action_jump jump = {
5490 		.group = 1,
5491 	};
5492 	const struct rte_flow_action actions[] = {
5493 		{
5494 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5495 			.conf = &jump,
5496 		},
5497 		{
5498 			.type = RTE_FLOW_ACTION_TYPE_END,
5499 		},
5500 	};
5501 	struct mlx5_priv *priv = dev->data->dev_private;
5502 	struct rte_flow_error error;
5503 
5504 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5505 						   &attr, &pattern,
5506 						   actions, false, &error);
5507 }
5508 
5509 /**
5510  * Validate a flow supported by the NIC.
5511  *
5512  * @see rte_flow_validate()
5513  * @see rte_flow_ops
5514  */
5515 int
5516 mlx5_flow_validate(struct rte_eth_dev *dev,
5517 		   const struct rte_flow_attr *attr,
5518 		   const struct rte_flow_item items[],
5519 		   const struct rte_flow_action original_actions[],
5520 		   struct rte_flow_error *error)
5521 {
5522 	int hairpin_flow;
5523 	struct mlx5_translated_shared_action
5524 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5525 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5526 	const struct rte_flow_action *actions;
5527 	struct rte_flow_action *translated_actions = NULL;
5528 	int ret = flow_shared_actions_translate(dev, original_actions,
5529 						shared_actions,
5530 						&shared_actions_n,
5531 						&translated_actions, error);
5532 
5533 	if (ret)
5534 		return ret;
5535 	actions = translated_actions ? translated_actions : original_actions;
5536 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5537 	ret = flow_drv_validate(dev, attr, items, actions,
5538 				true, hairpin_flow, error);
5539 	rte_free(translated_actions);
5540 	return ret;
5541 }
5542 
5543 /**
5544  * Create a flow.
5545  *
5546  * @see rte_flow_create()
5547  * @see rte_flow_ops
5548  */
5549 struct rte_flow *
5550 mlx5_flow_create(struct rte_eth_dev *dev,
5551 		 const struct rte_flow_attr *attr,
5552 		 const struct rte_flow_item items[],
5553 		 const struct rte_flow_action actions[],
5554 		 struct rte_flow_error *error)
5555 {
5556 	struct mlx5_priv *priv = dev->data->dev_private;
5557 
5558 	/*
5559 	 * If the device is not started yet, it is not allowed to created a
5560 	 * flow from application. PMD default flows and traffic control flows
5561 	 * are not affected.
5562 	 */
5563 	if (unlikely(!dev->data->dev_started)) {
5564 		DRV_LOG(DEBUG, "port %u is not started when "
5565 			"inserting a flow", dev->data->port_id);
5566 		rte_flow_error_set(error, ENODEV,
5567 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5568 				   NULL,
5569 				   "port not started");
5570 		return NULL;
5571 	}
5572 
5573 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5574 				  attr, items, actions, true, error);
5575 }
5576 
5577 /**
5578  * Destroy a flow in a list.
5579  *
5580  * @param dev
5581  *   Pointer to Ethernet device.
5582  * @param list
5583  *   Pointer to the Indexed flow list. If this parameter NULL,
5584  *   there is no flow removal from the list. Be noted that as
5585  *   flow is add to the indexed list, memory of the indexed
5586  *   list points to maybe changed as flow destroyed.
5587  * @param[in] flow_idx
5588  *   Index of flow to destroy.
5589  */
5590 static void
5591 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5592 		  uint32_t flow_idx)
5593 {
5594 	struct mlx5_priv *priv = dev->data->dev_private;
5595 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5596 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5597 
5598 	if (!flow)
5599 		return;
5600 	/*
5601 	 * Update RX queue flags only if port is started, otherwise it is
5602 	 * already clean.
5603 	 */
5604 	if (dev->data->dev_started)
5605 		flow_rxq_flags_trim(dev, flow);
5606 	flow_drv_destroy(dev, flow);
5607 	if (list) {
5608 		rte_spinlock_lock(&priv->flow_list_lock);
5609 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5610 			     flow_idx, flow, next);
5611 		rte_spinlock_unlock(&priv->flow_list_lock);
5612 	}
5613 	if (flow->tunnel) {
5614 		struct mlx5_flow_tunnel *tunnel;
5615 
5616 		rte_spinlock_lock(&mlx5_tunnel_hub(dev)->sl);
5617 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5618 		RTE_VERIFY(tunnel);
5619 		LIST_REMOVE(tunnel, chain);
5620 		rte_spinlock_unlock(&mlx5_tunnel_hub(dev)->sl);
5621 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5622 			mlx5_flow_tunnel_free(dev, tunnel);
5623 	}
5624 	flow_mreg_del_copy_action(dev, flow);
5625 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5626 }
5627 
5628 /**
5629  * Destroy all flows.
5630  *
5631  * @param dev
5632  *   Pointer to Ethernet device.
5633  * @param list
5634  *   Pointer to the Indexed flow list.
5635  * @param active
5636  *   If flushing is called avtively.
5637  */
5638 void
5639 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5640 {
5641 	uint32_t num_flushed = 0;
5642 
5643 	while (*list) {
5644 		flow_list_destroy(dev, list, *list);
5645 		num_flushed++;
5646 	}
5647 	if (active) {
5648 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5649 			dev->data->port_id, num_flushed);
5650 	}
5651 }
5652 
5653 /**
5654  * Stop all default actions for flows.
5655  *
5656  * @param dev
5657  *   Pointer to Ethernet device.
5658  */
5659 void
5660 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5661 {
5662 	flow_mreg_del_default_copy_action(dev);
5663 	flow_rxq_flags_clear(dev);
5664 }
5665 
5666 /**
5667  * Start all default actions for flows.
5668  *
5669  * @param dev
5670  *   Pointer to Ethernet device.
5671  * @return
5672  *   0 on success, a negative errno value otherwise and rte_errno is set.
5673  */
5674 int
5675 mlx5_flow_start_default(struct rte_eth_dev *dev)
5676 {
5677 	struct rte_flow_error error;
5678 
5679 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5680 	return flow_mreg_add_default_copy_action(dev, &error);
5681 }
5682 
5683 /**
5684  * Release key of thread specific flow workspace data.
5685  */
5686 static void
5687 flow_release_workspace(void *data)
5688 {
5689 	struct mlx5_flow_workspace *wks = data;
5690 	struct mlx5_flow_workspace *next;
5691 
5692 	while (wks) {
5693 		next = wks->next;
5694 		free(wks->rss_desc.queue);
5695 		free(wks);
5696 		wks = next;
5697 	}
5698 }
5699 
5700 /**
5701  * Initialize key of thread specific flow workspace data.
5702  */
5703 static void
5704 flow_alloc_workspace(void)
5705 {
5706 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5707 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5708 }
5709 
5710 /**
5711  * Get thread specific current flow workspace.
5712  *
5713  * @return pointer to thread specific flow workspace data, NULL on error.
5714  */
5715 struct mlx5_flow_workspace*
5716 mlx5_flow_get_thread_workspace(void)
5717 {
5718 	struct mlx5_flow_workspace *data;
5719 
5720 	data = pthread_getspecific(key_workspace);
5721 	MLX5_ASSERT(data && data->inuse);
5722 	if (!data || !data->inuse)
5723 		DRV_LOG(ERR, "flow workspace not initialized.");
5724 	return data;
5725 }
5726 
5727 /**
5728  * Allocate and init new flow workspace.
5729  *
5730  * @return pointer to flow workspace data, NULL on error.
5731  */
5732 static struct mlx5_flow_workspace*
5733 flow_alloc_thread_workspace(void)
5734 {
5735 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5736 
5737 	if (!data) {
5738 		DRV_LOG(ERR, "Failed to allocate flow workspace "
5739 			"memory.");
5740 		return NULL;
5741 	}
5742 	data->rss_desc.queue = calloc(1,
5743 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5744 	if (!data->rss_desc.queue)
5745 		goto err;
5746 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
5747 	return data;
5748 err:
5749 	if (data->rss_desc.queue)
5750 		free(data->rss_desc.queue);
5751 	free(data);
5752 	return NULL;
5753 }
5754 
5755 /**
5756  * Get new thread specific flow workspace.
5757  *
5758  * If current workspace inuse, create new one and set as current.
5759  *
5760  * @return pointer to thread specific flow workspace data, NULL on error.
5761  */
5762 static struct mlx5_flow_workspace*
5763 mlx5_flow_push_thread_workspace(void)
5764 {
5765 	struct mlx5_flow_workspace *curr;
5766 	struct mlx5_flow_workspace *data;
5767 
5768 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
5769 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
5770 		return NULL;
5771 	}
5772 	curr = pthread_getspecific(key_workspace);
5773 	if (!curr) {
5774 		data = flow_alloc_thread_workspace();
5775 		if (!data)
5776 			return NULL;
5777 	} else if (!curr->inuse) {
5778 		data = curr;
5779 	} else if (curr->next) {
5780 		data = curr->next;
5781 	} else {
5782 		data = flow_alloc_thread_workspace();
5783 		if (!data)
5784 			return NULL;
5785 		curr->next = data;
5786 		data->prev = curr;
5787 	}
5788 	data->inuse = 1;
5789 	data->flow_idx = 0;
5790 	/* Set as current workspace */
5791 	if (pthread_setspecific(key_workspace, data))
5792 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5793 	return data;
5794 }
5795 
5796 /**
5797  * Close current thread specific flow workspace.
5798  *
5799  * If previous workspace available, set it as current.
5800  *
5801  * @return pointer to thread specific flow workspace data, NULL on error.
5802  */
5803 static void
5804 mlx5_flow_pop_thread_workspace(void)
5805 {
5806 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
5807 
5808 	if (!data)
5809 		return;
5810 	if (!data->inuse) {
5811 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
5812 		return;
5813 	}
5814 	data->inuse = 0;
5815 	if (!data->prev)
5816 		return;
5817 	if (pthread_setspecific(key_workspace, data->prev))
5818 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
5819 }
5820 
5821 /**
5822  * Verify the flow list is empty
5823  *
5824  * @param dev
5825  *  Pointer to Ethernet device.
5826  *
5827  * @return the number of flows not released.
5828  */
5829 int
5830 mlx5_flow_verify(struct rte_eth_dev *dev)
5831 {
5832 	struct mlx5_priv *priv = dev->data->dev_private;
5833 	struct rte_flow *flow;
5834 	uint32_t idx;
5835 	int ret = 0;
5836 
5837 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
5838 		      flow, next) {
5839 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
5840 			dev->data->port_id, (void *)flow);
5841 		++ret;
5842 	}
5843 	return ret;
5844 }
5845 
5846 /**
5847  * Enable default hairpin egress flow.
5848  *
5849  * @param dev
5850  *   Pointer to Ethernet device.
5851  * @param queue
5852  *   The queue index.
5853  *
5854  * @return
5855  *   0 on success, a negative errno value otherwise and rte_errno is set.
5856  */
5857 int
5858 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
5859 			    uint32_t queue)
5860 {
5861 	struct mlx5_priv *priv = dev->data->dev_private;
5862 	const struct rte_flow_attr attr = {
5863 		.egress = 1,
5864 		.priority = 0,
5865 	};
5866 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
5867 		.queue = queue,
5868 	};
5869 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
5870 		.queue = UINT32_MAX,
5871 	};
5872 	struct rte_flow_item items[] = {
5873 		{
5874 			.type = (enum rte_flow_item_type)
5875 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
5876 			.spec = &queue_spec,
5877 			.last = NULL,
5878 			.mask = &queue_mask,
5879 		},
5880 		{
5881 			.type = RTE_FLOW_ITEM_TYPE_END,
5882 		},
5883 	};
5884 	struct rte_flow_action_jump jump = {
5885 		.group = MLX5_HAIRPIN_TX_TABLE,
5886 	};
5887 	struct rte_flow_action actions[2];
5888 	uint32_t flow_idx;
5889 	struct rte_flow_error error;
5890 
5891 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
5892 	actions[0].conf = &jump;
5893 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
5894 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5895 				&attr, items, actions, false, &error);
5896 	if (!flow_idx) {
5897 		DRV_LOG(DEBUG,
5898 			"Failed to create ctrl flow: rte_errno(%d),"
5899 			" type(%d), message(%s)",
5900 			rte_errno, error.type,
5901 			error.message ? error.message : " (no stated reason)");
5902 		return -rte_errno;
5903 	}
5904 	return 0;
5905 }
5906 
5907 /**
5908  * Enable a control flow configured from the control plane.
5909  *
5910  * @param dev
5911  *   Pointer to Ethernet device.
5912  * @param eth_spec
5913  *   An Ethernet flow spec to apply.
5914  * @param eth_mask
5915  *   An Ethernet flow mask to apply.
5916  * @param vlan_spec
5917  *   A VLAN flow spec to apply.
5918  * @param vlan_mask
5919  *   A VLAN flow mask to apply.
5920  *
5921  * @return
5922  *   0 on success, a negative errno value otherwise and rte_errno is set.
5923  */
5924 int
5925 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
5926 		    struct rte_flow_item_eth *eth_spec,
5927 		    struct rte_flow_item_eth *eth_mask,
5928 		    struct rte_flow_item_vlan *vlan_spec,
5929 		    struct rte_flow_item_vlan *vlan_mask)
5930 {
5931 	struct mlx5_priv *priv = dev->data->dev_private;
5932 	const struct rte_flow_attr attr = {
5933 		.ingress = 1,
5934 		.priority = MLX5_FLOW_PRIO_RSVD,
5935 	};
5936 	struct rte_flow_item items[] = {
5937 		{
5938 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5939 			.spec = eth_spec,
5940 			.last = NULL,
5941 			.mask = eth_mask,
5942 		},
5943 		{
5944 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
5945 					      RTE_FLOW_ITEM_TYPE_END,
5946 			.spec = vlan_spec,
5947 			.last = NULL,
5948 			.mask = vlan_mask,
5949 		},
5950 		{
5951 			.type = RTE_FLOW_ITEM_TYPE_END,
5952 		},
5953 	};
5954 	uint16_t queue[priv->reta_idx_n];
5955 	struct rte_flow_action_rss action_rss = {
5956 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
5957 		.level = 0,
5958 		.types = priv->rss_conf.rss_hf,
5959 		.key_len = priv->rss_conf.rss_key_len,
5960 		.queue_num = priv->reta_idx_n,
5961 		.key = priv->rss_conf.rss_key,
5962 		.queue = queue,
5963 	};
5964 	struct rte_flow_action actions[] = {
5965 		{
5966 			.type = RTE_FLOW_ACTION_TYPE_RSS,
5967 			.conf = &action_rss,
5968 		},
5969 		{
5970 			.type = RTE_FLOW_ACTION_TYPE_END,
5971 		},
5972 	};
5973 	uint32_t flow_idx;
5974 	struct rte_flow_error error;
5975 	unsigned int i;
5976 
5977 	if (!priv->reta_idx_n || !priv->rxqs_n) {
5978 		return 0;
5979 	}
5980 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
5981 		action_rss.types = 0;
5982 	for (i = 0; i != priv->reta_idx_n; ++i)
5983 		queue[i] = (*priv->reta_idx)[i];
5984 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5985 				&attr, items, actions, false, &error);
5986 	if (!flow_idx)
5987 		return -rte_errno;
5988 	return 0;
5989 }
5990 
5991 /**
5992  * Enable a flow control configured from the control plane.
5993  *
5994  * @param dev
5995  *   Pointer to Ethernet device.
5996  * @param eth_spec
5997  *   An Ethernet flow spec to apply.
5998  * @param eth_mask
5999  *   An Ethernet flow mask to apply.
6000  *
6001  * @return
6002  *   0 on success, a negative errno value otherwise and rte_errno is set.
6003  */
6004 int
6005 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6006 	       struct rte_flow_item_eth *eth_spec,
6007 	       struct rte_flow_item_eth *eth_mask)
6008 {
6009 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6010 }
6011 
6012 /**
6013  * Create default miss flow rule matching lacp traffic
6014  *
6015  * @param dev
6016  *   Pointer to Ethernet device.
6017  * @param eth_spec
6018  *   An Ethernet flow spec to apply.
6019  *
6020  * @return
6021  *   0 on success, a negative errno value otherwise and rte_errno is set.
6022  */
6023 int
6024 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6025 {
6026 	struct mlx5_priv *priv = dev->data->dev_private;
6027 	/*
6028 	 * The LACP matching is done by only using ether type since using
6029 	 * a multicast dst mac causes kernel to give low priority to this flow.
6030 	 */
6031 	static const struct rte_flow_item_eth lacp_spec = {
6032 		.type = RTE_BE16(0x8809),
6033 	};
6034 	static const struct rte_flow_item_eth lacp_mask = {
6035 		.type = 0xffff,
6036 	};
6037 	const struct rte_flow_attr attr = {
6038 		.ingress = 1,
6039 	};
6040 	struct rte_flow_item items[] = {
6041 		{
6042 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6043 			.spec = &lacp_spec,
6044 			.mask = &lacp_mask,
6045 		},
6046 		{
6047 			.type = RTE_FLOW_ITEM_TYPE_END,
6048 		},
6049 	};
6050 	struct rte_flow_action actions[] = {
6051 		{
6052 			.type = (enum rte_flow_action_type)
6053 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6054 		},
6055 		{
6056 			.type = RTE_FLOW_ACTION_TYPE_END,
6057 		},
6058 	};
6059 	struct rte_flow_error error;
6060 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6061 				&attr, items, actions, false, &error);
6062 
6063 	if (!flow_idx)
6064 		return -rte_errno;
6065 	return 0;
6066 }
6067 
6068 /**
6069  * Destroy a flow.
6070  *
6071  * @see rte_flow_destroy()
6072  * @see rte_flow_ops
6073  */
6074 int
6075 mlx5_flow_destroy(struct rte_eth_dev *dev,
6076 		  struct rte_flow *flow,
6077 		  struct rte_flow_error *error __rte_unused)
6078 {
6079 	struct mlx5_priv *priv = dev->data->dev_private;
6080 
6081 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6082 	return 0;
6083 }
6084 
6085 /**
6086  * Destroy all flows.
6087  *
6088  * @see rte_flow_flush()
6089  * @see rte_flow_ops
6090  */
6091 int
6092 mlx5_flow_flush(struct rte_eth_dev *dev,
6093 		struct rte_flow_error *error __rte_unused)
6094 {
6095 	struct mlx5_priv *priv = dev->data->dev_private;
6096 
6097 	mlx5_flow_list_flush(dev, &priv->flows, false);
6098 	return 0;
6099 }
6100 
6101 /**
6102  * Isolated mode.
6103  *
6104  * @see rte_flow_isolate()
6105  * @see rte_flow_ops
6106  */
6107 int
6108 mlx5_flow_isolate(struct rte_eth_dev *dev,
6109 		  int enable,
6110 		  struct rte_flow_error *error)
6111 {
6112 	struct mlx5_priv *priv = dev->data->dev_private;
6113 
6114 	if (dev->data->dev_started) {
6115 		rte_flow_error_set(error, EBUSY,
6116 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6117 				   NULL,
6118 				   "port must be stopped first");
6119 		return -rte_errno;
6120 	}
6121 	priv->isolated = !!enable;
6122 	if (enable)
6123 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6124 	else
6125 		dev->dev_ops = &mlx5_os_dev_ops;
6126 
6127 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6128 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6129 
6130 	return 0;
6131 }
6132 
6133 /**
6134  * Query a flow.
6135  *
6136  * @see rte_flow_query()
6137  * @see rte_flow_ops
6138  */
6139 static int
6140 flow_drv_query(struct rte_eth_dev *dev,
6141 	       uint32_t flow_idx,
6142 	       const struct rte_flow_action *actions,
6143 	       void *data,
6144 	       struct rte_flow_error *error)
6145 {
6146 	struct mlx5_priv *priv = dev->data->dev_private;
6147 	const struct mlx5_flow_driver_ops *fops;
6148 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6149 					       [MLX5_IPOOL_RTE_FLOW],
6150 					       flow_idx);
6151 	enum mlx5_flow_drv_type ftype;
6152 
6153 	if (!flow) {
6154 		return rte_flow_error_set(error, ENOENT,
6155 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6156 			  NULL,
6157 			  "invalid flow handle");
6158 	}
6159 	ftype = flow->drv_type;
6160 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6161 	fops = flow_get_drv_ops(ftype);
6162 
6163 	return fops->query(dev, flow, actions, data, error);
6164 }
6165 
6166 /**
6167  * Query a flow.
6168  *
6169  * @see rte_flow_query()
6170  * @see rte_flow_ops
6171  */
6172 int
6173 mlx5_flow_query(struct rte_eth_dev *dev,
6174 		struct rte_flow *flow,
6175 		const struct rte_flow_action *actions,
6176 		void *data,
6177 		struct rte_flow_error *error)
6178 {
6179 	int ret;
6180 
6181 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6182 			     error);
6183 	if (ret < 0)
6184 		return ret;
6185 	return 0;
6186 }
6187 
6188 /**
6189  * Manage filter operations.
6190  *
6191  * @param dev
6192  *   Pointer to Ethernet device structure.
6193  * @param filter_type
6194  *   Filter type.
6195  * @param filter_op
6196  *   Operation to perform.
6197  * @param arg
6198  *   Pointer to operation-specific structure.
6199  *
6200  * @return
6201  *   0 on success, a negative errno value otherwise and rte_errno is set.
6202  */
6203 int
6204 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6205 		     enum rte_filter_type filter_type,
6206 		     enum rte_filter_op filter_op,
6207 		     void *arg)
6208 {
6209 	switch (filter_type) {
6210 	case RTE_ETH_FILTER_GENERIC:
6211 		if (filter_op != RTE_ETH_FILTER_GET) {
6212 			rte_errno = EINVAL;
6213 			return -rte_errno;
6214 		}
6215 		*(const void **)arg = &mlx5_flow_ops;
6216 		return 0;
6217 	default:
6218 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6219 			dev->data->port_id, filter_type);
6220 		rte_errno = ENOTSUP;
6221 		return -rte_errno;
6222 	}
6223 	return 0;
6224 }
6225 
6226 /**
6227  * Create the needed meter and suffix tables.
6228  *
6229  * @param[in] dev
6230  *   Pointer to Ethernet device.
6231  * @param[in] fm
6232  *   Pointer to the flow meter.
6233  *
6234  * @return
6235  *   Pointer to table set on success, NULL otherwise.
6236  */
6237 struct mlx5_meter_domains_infos *
6238 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6239 			  const struct mlx5_flow_meter *fm)
6240 {
6241 	const struct mlx5_flow_driver_ops *fops;
6242 
6243 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6244 	return fops->create_mtr_tbls(dev, fm);
6245 }
6246 
6247 /**
6248  * Destroy the meter table set.
6249  *
6250  * @param[in] dev
6251  *   Pointer to Ethernet device.
6252  * @param[in] tbl
6253  *   Pointer to the meter table set.
6254  *
6255  * @return
6256  *   0 on success.
6257  */
6258 int
6259 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6260 			   struct mlx5_meter_domains_infos *tbls)
6261 {
6262 	const struct mlx5_flow_driver_ops *fops;
6263 
6264 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6265 	return fops->destroy_mtr_tbls(dev, tbls);
6266 }
6267 
6268 /**
6269  * Create policer rules.
6270  *
6271  * @param[in] dev
6272  *   Pointer to Ethernet device.
6273  * @param[in] fm
6274  *   Pointer to flow meter structure.
6275  * @param[in] attr
6276  *   Pointer to flow attributes.
6277  *
6278  * @return
6279  *   0 on success, -1 otherwise.
6280  */
6281 int
6282 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6283 			       struct mlx5_flow_meter *fm,
6284 			       const struct rte_flow_attr *attr)
6285 {
6286 	const struct mlx5_flow_driver_ops *fops;
6287 
6288 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6289 	return fops->create_policer_rules(dev, fm, attr);
6290 }
6291 
6292 /**
6293  * Destroy policer rules.
6294  *
6295  * @param[in] fm
6296  *   Pointer to flow meter structure.
6297  * @param[in] attr
6298  *   Pointer to flow attributes.
6299  *
6300  * @return
6301  *   0 on success, -1 otherwise.
6302  */
6303 int
6304 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6305 				struct mlx5_flow_meter *fm,
6306 				const struct rte_flow_attr *attr)
6307 {
6308 	const struct mlx5_flow_driver_ops *fops;
6309 
6310 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6311 	return fops->destroy_policer_rules(dev, fm, attr);
6312 }
6313 
6314 /**
6315  * Allocate a counter.
6316  *
6317  * @param[in] dev
6318  *   Pointer to Ethernet device structure.
6319  *
6320  * @return
6321  *   Index to allocated counter  on success, 0 otherwise.
6322  */
6323 uint32_t
6324 mlx5_counter_alloc(struct rte_eth_dev *dev)
6325 {
6326 	const struct mlx5_flow_driver_ops *fops;
6327 	struct rte_flow_attr attr = { .transfer = 0 };
6328 
6329 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6330 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6331 		return fops->counter_alloc(dev);
6332 	}
6333 	DRV_LOG(ERR,
6334 		"port %u counter allocate is not supported.",
6335 		 dev->data->port_id);
6336 	return 0;
6337 }
6338 
6339 /**
6340  * Free a counter.
6341  *
6342  * @param[in] dev
6343  *   Pointer to Ethernet device structure.
6344  * @param[in] cnt
6345  *   Index to counter to be free.
6346  */
6347 void
6348 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6349 {
6350 	const struct mlx5_flow_driver_ops *fops;
6351 	struct rte_flow_attr attr = { .transfer = 0 };
6352 
6353 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6354 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6355 		fops->counter_free(dev, cnt);
6356 		return;
6357 	}
6358 	DRV_LOG(ERR,
6359 		"port %u counter free is not supported.",
6360 		 dev->data->port_id);
6361 }
6362 
6363 /**
6364  * Query counter statistics.
6365  *
6366  * @param[in] dev
6367  *   Pointer to Ethernet device structure.
6368  * @param[in] cnt
6369  *   Index to counter to query.
6370  * @param[in] clear
6371  *   Set to clear counter statistics.
6372  * @param[out] pkts
6373  *   The counter hits packets number to save.
6374  * @param[out] bytes
6375  *   The counter hits bytes number to save.
6376  *
6377  * @return
6378  *   0 on success, a negative errno value otherwise.
6379  */
6380 int
6381 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6382 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6383 {
6384 	const struct mlx5_flow_driver_ops *fops;
6385 	struct rte_flow_attr attr = { .transfer = 0 };
6386 
6387 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6388 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6389 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6390 	}
6391 	DRV_LOG(ERR,
6392 		"port %u counter query is not supported.",
6393 		 dev->data->port_id);
6394 	return -ENOTSUP;
6395 }
6396 
6397 /**
6398  * Allocate a new memory for the counter values wrapped by all the needed
6399  * management.
6400  *
6401  * @param[in] sh
6402  *   Pointer to mlx5_dev_ctx_shared object.
6403  *
6404  * @return
6405  *   0 on success, a negative errno value otherwise.
6406  */
6407 static int
6408 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6409 {
6410 	struct mlx5_devx_mkey_attr mkey_attr;
6411 	struct mlx5_counter_stats_mem_mng *mem_mng;
6412 	volatile struct flow_counter_stats *raw_data;
6413 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6414 	int size = (sizeof(struct flow_counter_stats) *
6415 			MLX5_COUNTERS_PER_POOL +
6416 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6417 			sizeof(struct mlx5_counter_stats_mem_mng);
6418 	size_t pgsize = rte_mem_page_size();
6419 	uint8_t *mem;
6420 	int i;
6421 
6422 	if (pgsize == (size_t)-1) {
6423 		DRV_LOG(ERR, "Failed to get mem page size");
6424 		rte_errno = ENOMEM;
6425 		return -ENOMEM;
6426 	}
6427 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6428 	if (!mem) {
6429 		rte_errno = ENOMEM;
6430 		return -ENOMEM;
6431 	}
6432 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6433 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6434 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
6435 						 IBV_ACCESS_LOCAL_WRITE);
6436 	if (!mem_mng->umem) {
6437 		rte_errno = errno;
6438 		mlx5_free(mem);
6439 		return -rte_errno;
6440 	}
6441 	mkey_attr.addr = (uintptr_t)mem;
6442 	mkey_attr.size = size;
6443 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6444 	mkey_attr.pd = sh->pdn;
6445 	mkey_attr.log_entity_size = 0;
6446 	mkey_attr.pg_access = 0;
6447 	mkey_attr.klm_array = NULL;
6448 	mkey_attr.klm_num = 0;
6449 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6450 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6451 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6452 	if (!mem_mng->dm) {
6453 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
6454 		rte_errno = errno;
6455 		mlx5_free(mem);
6456 		return -rte_errno;
6457 	}
6458 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6459 	raw_data = (volatile struct flow_counter_stats *)mem;
6460 	for (i = 0; i < raws_n; ++i) {
6461 		mem_mng->raws[i].mem_mng = mem_mng;
6462 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6463 	}
6464 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6465 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6466 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6467 				 next);
6468 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6469 	sh->cmng.mem_mng = mem_mng;
6470 	return 0;
6471 }
6472 
6473 /**
6474  * Set the statistic memory to the new counter pool.
6475  *
6476  * @param[in] sh
6477  *   Pointer to mlx5_dev_ctx_shared object.
6478  * @param[in] pool
6479  *   Pointer to the pool to set the statistic memory.
6480  *
6481  * @return
6482  *   0 on success, a negative errno value otherwise.
6483  */
6484 static int
6485 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6486 			       struct mlx5_flow_counter_pool *pool)
6487 {
6488 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6489 	/* Resize statistic memory once used out. */
6490 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6491 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6492 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6493 		return -1;
6494 	}
6495 	rte_spinlock_lock(&pool->sl);
6496 	pool->raw = cmng->mem_mng->raws + pool->index %
6497 		    MLX5_CNT_CONTAINER_RESIZE;
6498 	rte_spinlock_unlock(&pool->sl);
6499 	pool->raw_hw = NULL;
6500 	return 0;
6501 }
6502 
6503 #define MLX5_POOL_QUERY_FREQ_US 1000000
6504 
6505 /**
6506  * Set the periodic procedure for triggering asynchronous batch queries for all
6507  * the counter pools.
6508  *
6509  * @param[in] sh
6510  *   Pointer to mlx5_dev_ctx_shared object.
6511  */
6512 void
6513 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6514 {
6515 	uint32_t pools_n, us;
6516 
6517 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6518 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6519 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6520 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6521 		sh->cmng.query_thread_on = 0;
6522 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6523 	} else {
6524 		sh->cmng.query_thread_on = 1;
6525 	}
6526 }
6527 
6528 /**
6529  * The periodic procedure for triggering asynchronous batch queries for all the
6530  * counter pools. This function is probably called by the host thread.
6531  *
6532  * @param[in] arg
6533  *   The parameter for the alarm process.
6534  */
6535 void
6536 mlx5_flow_query_alarm(void *arg)
6537 {
6538 	struct mlx5_dev_ctx_shared *sh = arg;
6539 	int ret;
6540 	uint16_t pool_index = sh->cmng.pool_index;
6541 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6542 	struct mlx5_flow_counter_pool *pool;
6543 	uint16_t n_valid;
6544 
6545 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6546 		goto set_alarm;
6547 	rte_spinlock_lock(&cmng->pool_update_sl);
6548 	pool = cmng->pools[pool_index];
6549 	n_valid = cmng->n_valid;
6550 	rte_spinlock_unlock(&cmng->pool_update_sl);
6551 	/* Set the statistic memory to the new created pool. */
6552 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6553 		goto set_alarm;
6554 	if (pool->raw_hw)
6555 		/* There is a pool query in progress. */
6556 		goto set_alarm;
6557 	pool->raw_hw =
6558 		LIST_FIRST(&sh->cmng.free_stat_raws);
6559 	if (!pool->raw_hw)
6560 		/* No free counter statistics raw memory. */
6561 		goto set_alarm;
6562 	/*
6563 	 * Identify the counters released between query trigger and query
6564 	 * handle more efficiently. The counter released in this gap period
6565 	 * should wait for a new round of query as the new arrived packets
6566 	 * will not be taken into account.
6567 	 */
6568 	pool->query_gen++;
6569 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6570 					       MLX5_COUNTERS_PER_POOL,
6571 					       NULL, NULL,
6572 					       pool->raw_hw->mem_mng->dm->id,
6573 					       (void *)(uintptr_t)
6574 					       pool->raw_hw->data,
6575 					       sh->devx_comp,
6576 					       (uint64_t)(uintptr_t)pool);
6577 	if (ret) {
6578 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6579 			" %d", pool->min_dcs->id);
6580 		pool->raw_hw = NULL;
6581 		goto set_alarm;
6582 	}
6583 	LIST_REMOVE(pool->raw_hw, next);
6584 	sh->cmng.pending_queries++;
6585 	pool_index++;
6586 	if (pool_index >= n_valid)
6587 		pool_index = 0;
6588 set_alarm:
6589 	sh->cmng.pool_index = pool_index;
6590 	mlx5_set_query_alarm(sh);
6591 }
6592 
6593 /**
6594  * Check and callback event for new aged flow in the counter pool
6595  *
6596  * @param[in] sh
6597  *   Pointer to mlx5_dev_ctx_shared object.
6598  * @param[in] pool
6599  *   Pointer to Current counter pool.
6600  */
6601 static void
6602 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6603 		   struct mlx5_flow_counter_pool *pool)
6604 {
6605 	struct mlx5_priv *priv;
6606 	struct mlx5_flow_counter *cnt;
6607 	struct mlx5_age_info *age_info;
6608 	struct mlx5_age_param *age_param;
6609 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6610 	struct mlx5_counter_stats_raw *prev = pool->raw;
6611 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6612 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6613 	uint16_t expected = AGE_CANDIDATE;
6614 	uint32_t i;
6615 
6616 	pool->time_of_last_age_check = curr_time;
6617 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6618 		cnt = MLX5_POOL_GET_CNT(pool, i);
6619 		age_param = MLX5_CNT_TO_AGE(cnt);
6620 		if (__atomic_load_n(&age_param->state,
6621 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6622 			continue;
6623 		if (cur->data[i].hits != prev->data[i].hits) {
6624 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6625 					 __ATOMIC_RELAXED);
6626 			continue;
6627 		}
6628 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6629 				       time_delta,
6630 				       __ATOMIC_RELAXED) <= age_param->timeout)
6631 			continue;
6632 		/**
6633 		 * Hold the lock first, or if between the
6634 		 * state AGE_TMOUT and tailq operation the
6635 		 * release happened, the release procedure
6636 		 * may delete a non-existent tailq node.
6637 		 */
6638 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6639 		age_info = GET_PORT_AGE_INFO(priv);
6640 		rte_spinlock_lock(&age_info->aged_sl);
6641 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6642 						AGE_TMOUT, false,
6643 						__ATOMIC_RELAXED,
6644 						__ATOMIC_RELAXED)) {
6645 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6646 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6647 		}
6648 		rte_spinlock_unlock(&age_info->aged_sl);
6649 	}
6650 	mlx5_age_event_prepare(sh);
6651 }
6652 
6653 /**
6654  * Handler for the HW respond about ready values from an asynchronous batch
6655  * query. This function is probably called by the host thread.
6656  *
6657  * @param[in] sh
6658  *   The pointer to the shared device context.
6659  * @param[in] async_id
6660  *   The Devx async ID.
6661  * @param[in] status
6662  *   The status of the completion.
6663  */
6664 void
6665 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6666 				  uint64_t async_id, int status)
6667 {
6668 	struct mlx5_flow_counter_pool *pool =
6669 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6670 	struct mlx5_counter_stats_raw *raw_to_free;
6671 	uint8_t query_gen = pool->query_gen ^ 1;
6672 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6673 	enum mlx5_counter_type cnt_type =
6674 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6675 				MLX5_COUNTER_TYPE_ORIGIN;
6676 
6677 	if (unlikely(status)) {
6678 		raw_to_free = pool->raw_hw;
6679 	} else {
6680 		raw_to_free = pool->raw;
6681 		if (pool->is_aged)
6682 			mlx5_flow_aging_check(sh, pool);
6683 		rte_spinlock_lock(&pool->sl);
6684 		pool->raw = pool->raw_hw;
6685 		rte_spinlock_unlock(&pool->sl);
6686 		/* Be sure the new raw counters data is updated in memory. */
6687 		rte_io_wmb();
6688 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6689 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6690 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6691 				     &pool->counters[query_gen], next);
6692 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6693 		}
6694 	}
6695 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6696 	pool->raw_hw = NULL;
6697 	sh->cmng.pending_queries--;
6698 }
6699 
6700 static int
6701 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
6702 		    const struct flow_grp_info *grp_info,
6703 		    struct rte_flow_error *error)
6704 {
6705 	if (grp_info->transfer && grp_info->external &&
6706 	    grp_info->fdb_def_rule) {
6707 		if (group == UINT32_MAX)
6708 			return rte_flow_error_set
6709 						(error, EINVAL,
6710 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6711 						 NULL,
6712 						 "group index not supported");
6713 		*table = group + 1;
6714 	} else {
6715 		*table = group;
6716 	}
6717 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
6718 	return 0;
6719 }
6720 
6721 /**
6722  * Translate the rte_flow group index to HW table value.
6723  *
6724  * If tunnel offload is disabled, all group ids converted to flow table
6725  * id using the standard method.
6726  * If tunnel offload is enabled, group id can be converted using the
6727  * standard or tunnel conversion method. Group conversion method
6728  * selection depends on flags in `grp_info` parameter:
6729  * - Internal (grp_info.external == 0) groups conversion uses the
6730  *   standard method.
6731  * - Group ids in JUMP action converted with the tunnel conversion.
6732  * - Group id in rule attribute conversion depends on a rule type and
6733  *   group id value:
6734  *   ** non zero group attributes converted with the tunnel method
6735  *   ** zero group attribute in non-tunnel rule is converted using the
6736  *      standard method - there's only one root table
6737  *   ** zero group attribute in steer tunnel rule is converted with the
6738  *      standard method - single root table
6739  *   ** zero group attribute in match tunnel rule is a special OvS
6740  *      case: that value is used for portability reasons. That group
6741  *      id is converted with the tunnel conversion method.
6742  *
6743  * @param[in] dev
6744  *   Port device
6745  * @param[in] tunnel
6746  *   PMD tunnel offload object
6747  * @param[in] group
6748  *   rte_flow group index value.
6749  * @param[out] table
6750  *   HW table value.
6751  * @param[in] grp_info
6752  *   flags used for conversion
6753  * @param[out] error
6754  *   Pointer to error structure.
6755  *
6756  * @return
6757  *   0 on success, a negative errno value otherwise and rte_errno is set.
6758  */
6759 int
6760 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
6761 			 const struct mlx5_flow_tunnel *tunnel,
6762 			 uint32_t group, uint32_t *table,
6763 			 const struct flow_grp_info *grp_info,
6764 			 struct rte_flow_error *error)
6765 {
6766 	int ret;
6767 	bool standard_translation;
6768 
6769 	if (!grp_info->skip_scale && grp_info->external &&
6770 	    group < MLX5_MAX_TABLES_EXTERNAL)
6771 		group *= MLX5_FLOW_TABLE_FACTOR;
6772 	if (is_tunnel_offload_active(dev)) {
6773 		standard_translation = !grp_info->external ||
6774 					grp_info->std_tbl_fix;
6775 	} else {
6776 		standard_translation = true;
6777 	}
6778 	DRV_LOG(DEBUG,
6779 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
6780 		dev->data->port_id, group, grp_info->transfer,
6781 		grp_info->external, grp_info->fdb_def_rule,
6782 		standard_translation ? "STANDARD" : "TUNNEL");
6783 	if (standard_translation)
6784 		ret = flow_group_to_table(dev->data->port_id, group, table,
6785 					  grp_info, error);
6786 	else
6787 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
6788 						      table, error);
6789 
6790 	return ret;
6791 }
6792 
6793 /**
6794  * Discover availability of metadata reg_c's.
6795  *
6796  * Iteratively use test flows to check availability.
6797  *
6798  * @param[in] dev
6799  *   Pointer to the Ethernet device structure.
6800  *
6801  * @return
6802  *   0 on success, a negative errno value otherwise and rte_errno is set.
6803  */
6804 int
6805 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6806 {
6807 	struct mlx5_priv *priv = dev->data->dev_private;
6808 	struct mlx5_dev_config *config = &priv->config;
6809 	enum modify_reg idx;
6810 	int n = 0;
6811 
6812 	/* reg_c[0] and reg_c[1] are reserved. */
6813 	config->flow_mreg_c[n++] = REG_C_0;
6814 	config->flow_mreg_c[n++] = REG_C_1;
6815 	/* Discover availability of other reg_c's. */
6816 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6817 		struct rte_flow_attr attr = {
6818 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6819 			.priority = MLX5_FLOW_PRIO_RSVD,
6820 			.ingress = 1,
6821 		};
6822 		struct rte_flow_item items[] = {
6823 			[0] = {
6824 				.type = RTE_FLOW_ITEM_TYPE_END,
6825 			},
6826 		};
6827 		struct rte_flow_action actions[] = {
6828 			[0] = {
6829 				.type = (enum rte_flow_action_type)
6830 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6831 				.conf = &(struct mlx5_flow_action_copy_mreg){
6832 					.src = REG_C_1,
6833 					.dst = idx,
6834 				},
6835 			},
6836 			[1] = {
6837 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6838 				.conf = &(struct rte_flow_action_jump){
6839 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6840 				},
6841 			},
6842 			[2] = {
6843 				.type = RTE_FLOW_ACTION_TYPE_END,
6844 			},
6845 		};
6846 		uint32_t flow_idx;
6847 		struct rte_flow *flow;
6848 		struct rte_flow_error error;
6849 
6850 		if (!config->dv_flow_en)
6851 			break;
6852 		/* Create internal flow, validation skips copy action. */
6853 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6854 					    actions, false, &error);
6855 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6856 				      flow_idx);
6857 		if (!flow)
6858 			continue;
6859 		config->flow_mreg_c[n++] = idx;
6860 		flow_list_destroy(dev, NULL, flow_idx);
6861 	}
6862 	for (; n < MLX5_MREG_C_NUM; ++n)
6863 		config->flow_mreg_c[n] = REG_NON;
6864 	return 0;
6865 }
6866 
6867 /**
6868  * Dump flow raw hw data to file
6869  *
6870  * @param[in] dev
6871  *    The pointer to Ethernet device.
6872  * @param[in] file
6873  *   A pointer to a file for output.
6874  * @param[out] error
6875  *   Perform verbose error reporting if not NULL. PMDs initialize this
6876  *   structure in case of error only.
6877  * @return
6878  *   0 on success, a nagative value otherwise.
6879  */
6880 int
6881 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6882 		   FILE *file,
6883 		   struct rte_flow_error *error __rte_unused)
6884 {
6885 	struct mlx5_priv *priv = dev->data->dev_private;
6886 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6887 
6888 	if (!priv->config.dv_flow_en) {
6889 		if (fputs("device dv flow disabled\n", file) <= 0)
6890 			return -errno;
6891 		return -ENOTSUP;
6892 	}
6893 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6894 				       sh->tx_domain, file);
6895 }
6896 
6897 /**
6898  * Get aged-out flows.
6899  *
6900  * @param[in] dev
6901  *   Pointer to the Ethernet device structure.
6902  * @param[in] context
6903  *   The address of an array of pointers to the aged-out flows contexts.
6904  * @param[in] nb_countexts
6905  *   The length of context array pointers.
6906  * @param[out] error
6907  *   Perform verbose error reporting if not NULL. Initialized in case of
6908  *   error only.
6909  *
6910  * @return
6911  *   how many contexts get in success, otherwise negative errno value.
6912  *   if nb_contexts is 0, return the amount of all aged contexts.
6913  *   if nb_contexts is not 0 , return the amount of aged flows reported
6914  *   in the context array.
6915  */
6916 int
6917 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6918 			uint32_t nb_contexts, struct rte_flow_error *error)
6919 {
6920 	const struct mlx5_flow_driver_ops *fops;
6921 	struct rte_flow_attr attr = { .transfer = 0 };
6922 
6923 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6924 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6925 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6926 						    error);
6927 	}
6928 	DRV_LOG(ERR,
6929 		"port %u get aged flows is not supported.",
6930 		 dev->data->port_id);
6931 	return -ENOTSUP;
6932 }
6933 
6934 /* Wrapper for driver action_validate op callback */
6935 static int
6936 flow_drv_action_validate(struct rte_eth_dev *dev,
6937 			 const struct rte_flow_shared_action_conf *conf,
6938 			 const struct rte_flow_action *action,
6939 			 const struct mlx5_flow_driver_ops *fops,
6940 			 struct rte_flow_error *error)
6941 {
6942 	static const char err_msg[] = "shared action validation unsupported";
6943 
6944 	if (!fops->action_validate) {
6945 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
6946 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
6947 				   NULL, err_msg);
6948 		return -rte_errno;
6949 	}
6950 	return fops->action_validate(dev, conf, action, error);
6951 }
6952 
6953 /**
6954  * Destroys the shared action by handle.
6955  *
6956  * @param dev
6957  *   Pointer to Ethernet device structure.
6958  * @param[in] action
6959  *   Handle for the shared action to be destroyed.
6960  * @param[out] error
6961  *   Perform verbose error reporting if not NULL. PMDs initialize this
6962  *   structure in case of error only.
6963  *
6964  * @return
6965  *   0 on success, a negative errno value otherwise and rte_errno is set.
6966  *
6967  * @note: wrapper for driver action_create op callback.
6968  */
6969 static int
6970 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
6971 			   struct rte_flow_shared_action *action,
6972 			   struct rte_flow_error *error)
6973 {
6974 	static const char err_msg[] = "shared action destruction unsupported";
6975 	struct rte_flow_attr attr = { .transfer = 0 };
6976 	const struct mlx5_flow_driver_ops *fops =
6977 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
6978 
6979 	if (!fops->action_destroy) {
6980 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
6981 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
6982 				   NULL, err_msg);
6983 		return -rte_errno;
6984 	}
6985 	return fops->action_destroy(dev, action, error);
6986 }
6987 
6988 /* Wrapper for driver action_destroy op callback */
6989 static int
6990 flow_drv_action_update(struct rte_eth_dev *dev,
6991 		       struct rte_flow_shared_action *action,
6992 		       const void *action_conf,
6993 		       const struct mlx5_flow_driver_ops *fops,
6994 		       struct rte_flow_error *error)
6995 {
6996 	static const char err_msg[] = "shared action update unsupported";
6997 
6998 	if (!fops->action_update) {
6999 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7000 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7001 				   NULL, err_msg);
7002 		return -rte_errno;
7003 	}
7004 	return fops->action_update(dev, action, action_conf, error);
7005 }
7006 
7007 /* Wrapper for driver action_destroy op callback */
7008 static int
7009 flow_drv_action_query(struct rte_eth_dev *dev,
7010 		      const struct rte_flow_shared_action *action,
7011 		      void *data,
7012 		      const struct mlx5_flow_driver_ops *fops,
7013 		      struct rte_flow_error *error)
7014 {
7015 	static const char err_msg[] = "shared action query unsupported";
7016 
7017 	if (!fops->action_query) {
7018 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7019 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7020 				   NULL, err_msg);
7021 		return -rte_errno;
7022 	}
7023 	return fops->action_query(dev, action, data, error);
7024 }
7025 
7026 /**
7027  * Create shared action for reuse in multiple flow rules.
7028  *
7029  * @param dev
7030  *   Pointer to Ethernet device structure.
7031  * @param[in] action
7032  *   Action configuration for shared action creation.
7033  * @param[out] error
7034  *   Perform verbose error reporting if not NULL. PMDs initialize this
7035  *   structure in case of error only.
7036  * @return
7037  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7038  */
7039 static struct rte_flow_shared_action *
7040 mlx5_shared_action_create(struct rte_eth_dev *dev,
7041 			  const struct rte_flow_shared_action_conf *conf,
7042 			  const struct rte_flow_action *action,
7043 			  struct rte_flow_error *error)
7044 {
7045 	static const char err_msg[] = "shared action creation unsupported";
7046 	struct rte_flow_attr attr = { .transfer = 0 };
7047 	const struct mlx5_flow_driver_ops *fops =
7048 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7049 
7050 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7051 		return NULL;
7052 	if (!fops->action_create) {
7053 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7054 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7055 				   NULL, err_msg);
7056 		return NULL;
7057 	}
7058 	return fops->action_create(dev, conf, action, error);
7059 }
7060 
7061 /**
7062  * Updates inplace the shared action configuration pointed by *action* handle
7063  * with the configuration provided as *action* argument.
7064  * The update of the shared action configuration effects all flow rules reusing
7065  * the action via handle.
7066  *
7067  * @param dev
7068  *   Pointer to Ethernet device structure.
7069  * @param[in] shared_action
7070  *   Handle for the shared action to be updated.
7071  * @param[in] action
7072  *   Action specification used to modify the action pointed by handle.
7073  *   *action* should be of same type with the action pointed by the *action*
7074  *   handle argument, otherwise considered as invalid.
7075  * @param[out] error
7076  *   Perform verbose error reporting if not NULL. PMDs initialize this
7077  *   structure in case of error only.
7078  *
7079  * @return
7080  *   0 on success, a negative errno value otherwise and rte_errno is set.
7081  */
7082 static int
7083 mlx5_shared_action_update(struct rte_eth_dev *dev,
7084 		struct rte_flow_shared_action *shared_action,
7085 		const struct rte_flow_action *action,
7086 		struct rte_flow_error *error)
7087 {
7088 	struct rte_flow_attr attr = { .transfer = 0 };
7089 	const struct mlx5_flow_driver_ops *fops =
7090 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7091 	int ret;
7092 
7093 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7094 	if (ret)
7095 		return ret;
7096 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7097 				      error);
7098 }
7099 
7100 /**
7101  * Query the shared action by handle.
7102  *
7103  * This function allows retrieving action-specific data such as counters.
7104  * Data is gathered by special action which may be present/referenced in
7105  * more than one flow rule definition.
7106  *
7107  * \see RTE_FLOW_ACTION_TYPE_COUNT
7108  *
7109  * @param dev
7110  *   Pointer to Ethernet device structure.
7111  * @param[in] action
7112  *   Handle for the shared action to query.
7113  * @param[in, out] data
7114  *   Pointer to storage for the associated query data type.
7115  * @param[out] error
7116  *   Perform verbose error reporting if not NULL. PMDs initialize this
7117  *   structure in case of error only.
7118  *
7119  * @return
7120  *   0 on success, a negative errno value otherwise and rte_errno is set.
7121  */
7122 static int
7123 mlx5_shared_action_query(struct rte_eth_dev *dev,
7124 			 const struct rte_flow_shared_action *action,
7125 			 void *data,
7126 			 struct rte_flow_error *error)
7127 {
7128 	struct rte_flow_attr attr = { .transfer = 0 };
7129 	const struct mlx5_flow_driver_ops *fops =
7130 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7131 
7132 	return flow_drv_action_query(dev, action, data, fops, error);
7133 }
7134 
7135 /**
7136  * Destroy all shared actions.
7137  *
7138  * @param dev
7139  *   Pointer to Ethernet device.
7140  *
7141  * @return
7142  *   0 on success, a negative errno value otherwise and rte_errno is set.
7143  */
7144 int
7145 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7146 {
7147 	struct rte_flow_error error;
7148 	struct mlx5_priv *priv = dev->data->dev_private;
7149 	struct mlx5_shared_action_rss *action;
7150 	int ret = 0;
7151 	uint32_t idx;
7152 
7153 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7154 		      priv->rss_shared_actions, idx, action, next) {
7155 		ret |= mlx5_shared_action_destroy(dev,
7156 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7157 	}
7158 	return ret;
7159 }
7160 
7161 #ifndef HAVE_MLX5DV_DR
7162 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7163 #else
7164 #define MLX5_DOMAIN_SYNC_FLOW \
7165 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7166 #endif
7167 
7168 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7169 {
7170 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7171 	const struct mlx5_flow_driver_ops *fops;
7172 	int ret;
7173 	struct rte_flow_attr attr = { .transfer = 0 };
7174 
7175 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7176 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7177 	if (ret > 0)
7178 		ret = -ret;
7179 	return ret;
7180 }
7181 
7182 /**
7183  * tunnel offload functionalilty is defined for DV environment only
7184  */
7185 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7186 __extension__
7187 union tunnel_offload_mark {
7188 	uint32_t val;
7189 	struct {
7190 		uint32_t app_reserve:8;
7191 		uint32_t table_id:15;
7192 		uint32_t transfer:1;
7193 		uint32_t _unused_:8;
7194 	};
7195 };
7196 
7197 static int
7198 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7199 			     struct rte_flow *flow,
7200 			     const struct rte_flow_attr *attr,
7201 			     const struct rte_flow_action *app_actions,
7202 			     uint32_t flow_idx,
7203 			     struct tunnel_default_miss_ctx *ctx,
7204 			     struct rte_flow_error *error)
7205 {
7206 	struct mlx5_priv *priv = dev->data->dev_private;
7207 	struct mlx5_flow *dev_flow;
7208 	struct rte_flow_attr miss_attr = *attr;
7209 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7210 	const struct rte_flow_item miss_items[2] = {
7211 		{
7212 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7213 			.spec = NULL,
7214 			.last = NULL,
7215 			.mask = NULL
7216 		},
7217 		{
7218 			.type = RTE_FLOW_ITEM_TYPE_END,
7219 			.spec = NULL,
7220 			.last = NULL,
7221 			.mask = NULL
7222 		}
7223 	};
7224 	union tunnel_offload_mark mark_id;
7225 	struct rte_flow_action_mark miss_mark;
7226 	struct rte_flow_action miss_actions[3] = {
7227 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7228 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7229 	};
7230 	const struct rte_flow_action_jump *jump_data;
7231 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7232 	struct flow_grp_info grp_info = {
7233 		.external = 1,
7234 		.transfer = attr->transfer,
7235 		.fdb_def_rule = !!priv->fdb_def_rule,
7236 		.std_tbl_fix = 0,
7237 	};
7238 	int ret;
7239 
7240 	if (!attr->transfer) {
7241 		uint32_t q_size;
7242 
7243 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7244 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7245 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7246 					 0, SOCKET_ID_ANY);
7247 		if (!ctx->queue)
7248 			return rte_flow_error_set
7249 				(error, ENOMEM,
7250 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7251 				NULL, "invalid default miss RSS");
7252 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7253 		ctx->action_rss.level = 0,
7254 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7255 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7256 		ctx->action_rss.queue_num = priv->reta_idx_n,
7257 		ctx->action_rss.key = priv->rss_conf.rss_key,
7258 		ctx->action_rss.queue = ctx->queue;
7259 		if (!priv->reta_idx_n || !priv->rxqs_n)
7260 			return rte_flow_error_set
7261 				(error, EINVAL,
7262 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7263 				NULL, "invalid port configuration");
7264 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7265 			ctx->action_rss.types = 0;
7266 		for (i = 0; i != priv->reta_idx_n; ++i)
7267 			ctx->queue[i] = (*priv->reta_idx)[i];
7268 	} else {
7269 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7270 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7271 	}
7272 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7273 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7274 	jump_data = app_actions->conf;
7275 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7276 	miss_attr.group = jump_data->group;
7277 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7278 				       &flow_table, &grp_info, error);
7279 	if (ret)
7280 		return rte_flow_error_set(error, EINVAL,
7281 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7282 					  NULL, "invalid tunnel id");
7283 	mark_id.app_reserve = 0;
7284 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7285 	mark_id.transfer = !!attr->transfer;
7286 	mark_id._unused_ = 0;
7287 	miss_mark.id = mark_id.val;
7288 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7289 				    miss_items, miss_actions, flow_idx, error);
7290 	if (!dev_flow)
7291 		return -rte_errno;
7292 	dev_flow->flow = flow;
7293 	dev_flow->external = true;
7294 	dev_flow->tunnel = tunnel;
7295 	/* Subflow object was created, we must include one in the list. */
7296 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7297 		      dev_flow->handle, next);
7298 	DRV_LOG(DEBUG,
7299 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7300 		dev->data->port_id, tunnel->app_tunnel.type,
7301 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7302 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7303 				  miss_actions, error);
7304 	if (!ret)
7305 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7306 						  error);
7307 
7308 	return ret;
7309 }
7310 
7311 static const struct mlx5_flow_tbl_data_entry  *
7312 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7313 {
7314 	struct mlx5_priv *priv = dev->data->dev_private;
7315 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7316 	struct mlx5_hlist_entry *he;
7317 	union tunnel_offload_mark mbits = { .val = mark };
7318 	union mlx5_flow_tbl_key table_key = {
7319 		{
7320 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7321 			.dummy = 0,
7322 			.domain = !!mbits.transfer,
7323 			.direction = 0,
7324 		}
7325 	};
7326 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7327 	return he ?
7328 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7329 }
7330 
7331 static void
7332 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7333 				   struct mlx5_hlist_entry *entry)
7334 {
7335 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7336 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7337 
7338 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7339 			tunnel_flow_tbl_to_id(tte->flow_table));
7340 	mlx5_free(tte);
7341 }
7342 
7343 static struct mlx5_hlist_entry *
7344 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list,
7345 				   uint64_t key __rte_unused,
7346 				   void *ctx __rte_unused)
7347 {
7348 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7349 	struct tunnel_tbl_entry *tte;
7350 
7351 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7352 			  sizeof(*tte), 0,
7353 			  SOCKET_ID_ANY);
7354 	if (!tte)
7355 		goto err;
7356 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7357 			  &tte->flow_table);
7358 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7359 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7360 			tte->flow_table);
7361 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7362 				tte->flow_table);
7363 		goto err;
7364 	} else if (!tte->flow_table) {
7365 		goto err;
7366 	}
7367 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7368 	return &tte->hash;
7369 err:
7370 	if (tte)
7371 		mlx5_free(tte);
7372 	return NULL;
7373 }
7374 
7375 static uint32_t
7376 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7377 				const struct mlx5_flow_tunnel *tunnel,
7378 				uint32_t group, uint32_t *table,
7379 				struct rte_flow_error *error)
7380 {
7381 	struct mlx5_hlist_entry *he;
7382 	struct tunnel_tbl_entry *tte;
7383 	union tunnel_tbl_key key = {
7384 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7385 		.group = group
7386 	};
7387 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7388 	struct mlx5_hlist *group_hash;
7389 
7390 	group_hash = tunnel ? tunnel->groups : thub->groups;
7391 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7392 	if (!he)
7393 		return rte_flow_error_set(error, EINVAL,
7394 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7395 					  NULL,
7396 					  "tunnel group index not supported");
7397 	tte = container_of(he, typeof(*tte), hash);
7398 	*table = tte->flow_table;
7399 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7400 		dev->data->port_id, key.tunnel_id, group, *table);
7401 	return 0;
7402 }
7403 
7404 static void
7405 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7406 		      struct mlx5_flow_tunnel *tunnel)
7407 {
7408 	struct mlx5_priv *priv = dev->data->dev_private;
7409 	struct mlx5_indexed_pool *ipool;
7410 
7411 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7412 		dev->data->port_id, tunnel->tunnel_id);
7413 	mlx5_hlist_destroy(tunnel->groups);
7414 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7415 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7416 }
7417 
7418 static struct mlx5_flow_tunnel *
7419 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7420 {
7421 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7422 	struct mlx5_flow_tunnel *tun;
7423 
7424 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7425 		if (tun->tunnel_id == id)
7426 			break;
7427 	}
7428 
7429 	return tun;
7430 }
7431 
7432 static struct mlx5_flow_tunnel *
7433 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7434 			  const struct rte_flow_tunnel *app_tunnel)
7435 {
7436 	struct mlx5_priv *priv = dev->data->dev_private;
7437 	struct mlx5_indexed_pool *ipool;
7438 	struct mlx5_flow_tunnel *tunnel;
7439 	uint32_t id;
7440 
7441 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7442 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7443 	if (!tunnel)
7444 		return NULL;
7445 	if (id >= MLX5_MAX_TUNNELS) {
7446 		mlx5_ipool_free(ipool, id);
7447 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7448 		return NULL;
7449 	}
7450 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7451 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7452 					   NULL,
7453 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7454 	if (!tunnel->groups) {
7455 		mlx5_ipool_free(ipool, id);
7456 		return NULL;
7457 	}
7458 	tunnel->groups->ctx = priv->sh;
7459 	/* initiate new PMD tunnel */
7460 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7461 	tunnel->tunnel_id = id;
7462 	tunnel->action.type = (typeof(tunnel->action.type))
7463 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7464 	tunnel->action.conf = tunnel;
7465 	tunnel->item.type = (typeof(tunnel->item.type))
7466 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7467 	tunnel->item.spec = tunnel;
7468 	tunnel->item.last = NULL;
7469 	tunnel->item.mask = NULL;
7470 
7471 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7472 		dev->data->port_id, tunnel->tunnel_id);
7473 
7474 	return tunnel;
7475 }
7476 
7477 static int
7478 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7479 		     const struct rte_flow_tunnel *app_tunnel,
7480 		     struct mlx5_flow_tunnel **tunnel)
7481 {
7482 	int ret;
7483 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7484 	struct mlx5_flow_tunnel *tun;
7485 
7486 	rte_spinlock_lock(&thub->sl);
7487 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7488 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7489 			    sizeof(*app_tunnel))) {
7490 			*tunnel = tun;
7491 			ret = 0;
7492 			break;
7493 		}
7494 	}
7495 	if (!tun) {
7496 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
7497 		if (tun) {
7498 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
7499 			*tunnel = tun;
7500 		} else {
7501 			ret = -ENOMEM;
7502 		}
7503 	}
7504 	rte_spinlock_unlock(&thub->sl);
7505 	if (tun)
7506 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
7507 
7508 	return ret;
7509 }
7510 
7511 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7512 {
7513 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7514 
7515 	if (!thub)
7516 		return;
7517 	if (!LIST_EMPTY(&thub->tunnels))
7518 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7519 	mlx5_hlist_destroy(thub->groups);
7520 	mlx5_free(thub);
7521 }
7522 
7523 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7524 {
7525 	int err;
7526 	struct mlx5_flow_tunnel_hub *thub;
7527 
7528 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7529 			   0, SOCKET_ID_ANY);
7530 	if (!thub)
7531 		return -ENOMEM;
7532 	LIST_INIT(&thub->tunnels);
7533 	rte_spinlock_init(&thub->sl);
7534 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7535 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7536 					 NULL,
7537 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7538 	if (!thub->groups) {
7539 		err = -rte_errno;
7540 		goto err;
7541 	}
7542 	thub->groups->ctx = sh;
7543 	sh->tunnel_hub = thub;
7544 
7545 	return 0;
7546 
7547 err:
7548 	if (thub->groups)
7549 		mlx5_hlist_destroy(thub->groups);
7550 	if (thub)
7551 		mlx5_free(thub);
7552 	return err;
7553 }
7554 
7555 static inline bool
7556 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7557 			  struct rte_flow_tunnel *tunnel,
7558 			  const char *err_msg)
7559 {
7560 	err_msg = NULL;
7561 	if (!is_tunnel_offload_active(dev)) {
7562 		err_msg = "tunnel offload was not activated";
7563 		goto out;
7564 	} else if (!tunnel) {
7565 		err_msg = "no application tunnel";
7566 		goto out;
7567 	}
7568 
7569 	switch (tunnel->type) {
7570 	default:
7571 		err_msg = "unsupported tunnel type";
7572 		goto out;
7573 	case RTE_FLOW_ITEM_TYPE_VXLAN:
7574 		break;
7575 	}
7576 
7577 out:
7578 	return !err_msg;
7579 }
7580 
7581 static int
7582 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7583 		    struct rte_flow_tunnel *app_tunnel,
7584 		    struct rte_flow_action **actions,
7585 		    uint32_t *num_of_actions,
7586 		    struct rte_flow_error *error)
7587 {
7588 	int ret;
7589 	struct mlx5_flow_tunnel *tunnel;
7590 	const char *err_msg = NULL;
7591 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7592 
7593 	if (!verdict)
7594 		return rte_flow_error_set(error, EINVAL,
7595 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7596 					  err_msg);
7597 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7598 	if (ret < 0) {
7599 		return rte_flow_error_set(error, ret,
7600 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
7601 					  "failed to initialize pmd tunnel");
7602 	}
7603 	*actions = &tunnel->action;
7604 	*num_of_actions = 1;
7605 	return 0;
7606 }
7607 
7608 static int
7609 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
7610 		       struct rte_flow_tunnel *app_tunnel,
7611 		       struct rte_flow_item **items,
7612 		       uint32_t *num_of_items,
7613 		       struct rte_flow_error *error)
7614 {
7615 	int ret;
7616 	struct mlx5_flow_tunnel *tunnel;
7617 	const char *err_msg = NULL;
7618 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
7619 
7620 	if (!verdict)
7621 		return rte_flow_error_set(error, EINVAL,
7622 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7623 					  err_msg);
7624 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
7625 	if (ret < 0) {
7626 		return rte_flow_error_set(error, ret,
7627 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7628 					  "failed to initialize pmd tunnel");
7629 	}
7630 	*items = &tunnel->item;
7631 	*num_of_items = 1;
7632 	return 0;
7633 }
7634 static int
7635 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
7636 			      struct rte_flow_item *pmd_items,
7637 			      uint32_t num_items, struct rte_flow_error *err)
7638 {
7639 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7640 	struct mlx5_flow_tunnel *tun;
7641 
7642 	rte_spinlock_lock(&thub->sl);
7643 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7644 		if (&tun->item == pmd_items) {
7645 			LIST_REMOVE(tun, chain);
7646 			break;
7647 		}
7648 	}
7649 	rte_spinlock_unlock(&thub->sl);
7650 	if (!tun || num_items != 1)
7651 		return rte_flow_error_set(err, EINVAL,
7652 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7653 					  "invalid argument");
7654 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
7655 		mlx5_flow_tunnel_free(dev, tun);
7656 	return 0;
7657 }
7658 
7659 static int
7660 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
7661 				struct rte_flow_action *pmd_actions,
7662 				uint32_t num_actions,
7663 				struct rte_flow_error *err)
7664 {
7665 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7666 	struct mlx5_flow_tunnel *tun;
7667 
7668 	rte_spinlock_lock(&thub->sl);
7669 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7670 		if (&tun->action == pmd_actions) {
7671 			LIST_REMOVE(tun, chain);
7672 			break;
7673 		}
7674 	}
7675 	rte_spinlock_unlock(&thub->sl);
7676 	if (!tun || num_actions != 1)
7677 		return rte_flow_error_set(err, EINVAL,
7678 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
7679 					  "invalid argument");
7680 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
7681 		mlx5_flow_tunnel_free(dev, tun);
7682 
7683 	return 0;
7684 }
7685 
7686 static int
7687 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
7688 				  struct rte_mbuf *m,
7689 				  struct rte_flow_restore_info *info,
7690 				  struct rte_flow_error *err)
7691 {
7692 	uint64_t ol_flags = m->ol_flags;
7693 	const struct mlx5_flow_tbl_data_entry *tble;
7694 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
7695 
7696 	if ((ol_flags & mask) != mask)
7697 		goto err;
7698 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
7699 	if (!tble) {
7700 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
7701 			dev->data->port_id, m->hash.fdir.hi);
7702 		goto err;
7703 	}
7704 	MLX5_ASSERT(tble->tunnel);
7705 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
7706 	info->group_id = tble->group_id;
7707 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
7708 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
7709 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
7710 
7711 	return 0;
7712 
7713 err:
7714 	return rte_flow_error_set(err, EINVAL,
7715 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7716 				  "failed to get restore info");
7717 }
7718 
7719 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
7720 static int
7721 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
7722 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
7723 			   __rte_unused struct rte_flow_action **actions,
7724 			   __rte_unused uint32_t *num_of_actions,
7725 			   __rte_unused struct rte_flow_error *error)
7726 {
7727 	return -ENOTSUP;
7728 }
7729 
7730 static int
7731 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
7732 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
7733 		       __rte_unused struct rte_flow_item **items,
7734 		       __rte_unused uint32_t *num_of_items,
7735 		       __rte_unused struct rte_flow_error *error)
7736 {
7737 	return -ENOTSUP;
7738 }
7739 
7740 static int
7741 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
7742 			      __rte_unused struct rte_flow_item *pmd_items,
7743 			      __rte_unused uint32_t num_items,
7744 			      __rte_unused struct rte_flow_error *err)
7745 {
7746 	return -ENOTSUP;
7747 }
7748 
7749 static int
7750 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
7751 				__rte_unused struct rte_flow_action *pmd_action,
7752 				__rte_unused uint32_t num_actions,
7753 				__rte_unused struct rte_flow_error *err)
7754 {
7755 	return -ENOTSUP;
7756 }
7757 
7758 static int
7759 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
7760 				  __rte_unused struct rte_mbuf *m,
7761 				  __rte_unused struct rte_flow_restore_info *i,
7762 				  __rte_unused struct rte_flow_error *err)
7763 {
7764 	return -ENOTSUP;
7765 }
7766 
7767 static int
7768 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
7769 			     __rte_unused struct rte_flow *flow,
7770 			     __rte_unused const struct rte_flow_attr *attr,
7771 			     __rte_unused const struct rte_flow_action *actions,
7772 			     __rte_unused uint32_t flow_idx,
7773 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
7774 			     __rte_unused struct rte_flow_error *error)
7775 {
7776 	return -ENOTSUP;
7777 }
7778 
7779 static struct mlx5_flow_tunnel *
7780 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
7781 		    __rte_unused uint32_t id)
7782 {
7783 	return NULL;
7784 }
7785 
7786 static void
7787 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
7788 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
7789 {
7790 }
7791 
7792 static uint32_t
7793 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
7794 				__rte_unused const struct mlx5_flow_tunnel *t,
7795 				__rte_unused uint32_t group,
7796 				__rte_unused uint32_t *table,
7797 				struct rte_flow_error *error)
7798 {
7799 	return rte_flow_error_set(error, ENOTSUP,
7800 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7801 				  "tunnel offload requires DV support");
7802 }
7803 
7804 void
7805 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
7806 			__rte_unused  uint16_t port_id)
7807 {
7808 }
7809 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
7810 
7811