xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 3bb3ebb51b789d4ecb417cbdb1dce5c7211f6f18)
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 <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 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
77 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
78 #endif
79 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
80 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
81 };
82 
83 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
84 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
85 	(const int []){ \
86 		__VA_ARGS__, 0, \
87 	}
88 
89 /** Node object of input graph for mlx5_flow_expand_rss(). */
90 struct mlx5_flow_expand_node {
91 	const int *const next;
92 	/**<
93 	 * List of next node indexes. Index 0 is interpreted as a terminator.
94 	 */
95 	const enum rte_flow_item_type type;
96 	/**< Pattern item type of current node. */
97 	uint64_t rss_types;
98 	/**<
99 	 * RSS types bit-field associated with this node
100 	 * (see ETH_RSS_* definitions).
101 	 */
102 };
103 
104 /** Object returned by mlx5_flow_expand_rss(). */
105 struct mlx5_flow_expand_rss {
106 	uint32_t entries;
107 	/**< Number of entries @p patterns and @p priorities. */
108 	struct {
109 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
110 		uint32_t priority; /**< Priority offset for each expansion. */
111 	} entry[];
112 };
113 
114 static enum rte_flow_item_type
115 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
116 {
117 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
118 	uint16_t ether_type = 0;
119 	uint16_t ether_type_m;
120 	uint8_t ip_next_proto = 0;
121 	uint8_t ip_next_proto_m;
122 
123 	if (item == NULL || item->spec == NULL)
124 		return ret;
125 	switch (item->type) {
126 	case RTE_FLOW_ITEM_TYPE_ETH:
127 		if (item->mask)
128 			ether_type_m = ((const struct rte_flow_item_eth *)
129 						(item->mask))->type;
130 		else
131 			ether_type_m = rte_flow_item_eth_mask.type;
132 		if (ether_type_m != RTE_BE16(0xFFFF))
133 			break;
134 		ether_type = ((const struct rte_flow_item_eth *)
135 				(item->spec))->type;
136 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
139 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
140 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
141 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
142 		else
143 			ret = RTE_FLOW_ITEM_TYPE_END;
144 		break;
145 	case RTE_FLOW_ITEM_TYPE_VLAN:
146 		if (item->mask)
147 			ether_type_m = ((const struct rte_flow_item_vlan *)
148 						(item->mask))->inner_type;
149 		else
150 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
151 		if (ether_type_m != RTE_BE16(0xFFFF))
152 			break;
153 		ether_type = ((const struct rte_flow_item_vlan *)
154 				(item->spec))->inner_type;
155 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
156 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
157 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
158 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
159 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
160 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
161 		else
162 			ret = RTE_FLOW_ITEM_TYPE_END;
163 		break;
164 	case RTE_FLOW_ITEM_TYPE_IPV4:
165 		if (item->mask)
166 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
167 					(item->mask))->hdr.next_proto_id;
168 		else
169 			ip_next_proto_m =
170 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
171 		if (ip_next_proto_m != 0xFF)
172 			break;
173 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
174 				(item->spec))->hdr.next_proto_id;
175 		if (ip_next_proto == IPPROTO_UDP)
176 			ret = RTE_FLOW_ITEM_TYPE_UDP;
177 		else if (ip_next_proto == IPPROTO_TCP)
178 			ret = RTE_FLOW_ITEM_TYPE_TCP;
179 		else if (ip_next_proto == IPPROTO_IP)
180 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
181 		else if (ip_next_proto == IPPROTO_IPV6)
182 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
183 		else
184 			ret = RTE_FLOW_ITEM_TYPE_END;
185 		break;
186 	case RTE_FLOW_ITEM_TYPE_IPV6:
187 		if (item->mask)
188 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
189 						(item->mask))->hdr.proto;
190 		else
191 			ip_next_proto_m =
192 				rte_flow_item_ipv6_mask.hdr.proto;
193 		if (ip_next_proto_m != 0xFF)
194 			break;
195 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
196 				(item->spec))->hdr.proto;
197 		if (ip_next_proto == IPPROTO_UDP)
198 			ret = RTE_FLOW_ITEM_TYPE_UDP;
199 		else if (ip_next_proto == IPPROTO_TCP)
200 			ret = RTE_FLOW_ITEM_TYPE_TCP;
201 		else if (ip_next_proto == IPPROTO_IP)
202 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
203 		else if (ip_next_proto == IPPROTO_IPV6)
204 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
205 		else
206 			ret = RTE_FLOW_ITEM_TYPE_END;
207 		break;
208 	default:
209 		ret = RTE_FLOW_ITEM_TYPE_VOID;
210 		break;
211 	}
212 	return ret;
213 }
214 
215 #define MLX5_RSS_EXP_ELT_N 8
216 
217 /**
218  * Expand RSS flows into several possible flows according to the RSS hash
219  * fields requested and the driver capabilities.
220  *
221  * @param[out] buf
222  *   Buffer to store the result expansion.
223  * @param[in] size
224  *   Buffer size in bytes. If 0, @p buf can be NULL.
225  * @param[in] pattern
226  *   User flow pattern.
227  * @param[in] types
228  *   RSS types to expand (see ETH_RSS_* definitions).
229  * @param[in] graph
230  *   Input graph to expand @p pattern according to @p types.
231  * @param[in] graph_root_index
232  *   Index of root node in @p graph, typically 0.
233  *
234  * @return
235  *   A positive value representing the size of @p buf in bytes regardless of
236  *   @p size on success, a negative errno value otherwise and rte_errno is
237  *   set, the following errors are defined:
238  *
239  *   -E2BIG: graph-depth @p graph is too deep.
240  */
241 static int
242 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
243 		     const struct rte_flow_item *pattern, uint64_t types,
244 		     const struct mlx5_flow_expand_node graph[],
245 		     int graph_root_index)
246 {
247 	const struct rte_flow_item *item;
248 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
249 	const int *next_node;
250 	const int *stack[MLX5_RSS_EXP_ELT_N];
251 	int stack_pos = 0;
252 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
253 	unsigned int i;
254 	size_t lsize;
255 	size_t user_pattern_size = 0;
256 	void *addr = NULL;
257 	const struct mlx5_flow_expand_node *next = NULL;
258 	struct rte_flow_item missed_item;
259 	int missed = 0;
260 	int elt = 0;
261 	const struct rte_flow_item *last_item = NULL;
262 
263 	memset(&missed_item, 0, sizeof(missed_item));
264 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
265 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
266 	if (lsize <= size) {
267 		buf->entry[0].priority = 0;
268 		buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
269 		buf->entries = 0;
270 		addr = buf->entry[0].pattern;
271 	}
272 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
273 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
274 			last_item = item;
275 		for (i = 0; node->next && node->next[i]; ++i) {
276 			next = &graph[node->next[i]];
277 			if (next->type == item->type)
278 				break;
279 		}
280 		if (next)
281 			node = next;
282 		user_pattern_size += sizeof(*item);
283 	}
284 	user_pattern_size += sizeof(*item); /* Handle END item. */
285 	lsize += user_pattern_size;
286 	/* Copy the user pattern in the first entry of the buffer. */
287 	if (lsize <= size) {
288 		rte_memcpy(addr, pattern, user_pattern_size);
289 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
290 		buf->entries = 1;
291 	}
292 	/* Start expanding. */
293 	memset(flow_items, 0, sizeof(flow_items));
294 	user_pattern_size -= sizeof(*item);
295 	/*
296 	 * Check if the last valid item has spec set, need complete pattern,
297 	 * and the pattern can be used for expansion.
298 	 */
299 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
300 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
301 		/* Item type END indicates expansion is not required. */
302 		return lsize;
303 	}
304 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
305 		next = NULL;
306 		missed = 1;
307 		for (i = 0; node->next && node->next[i]; ++i) {
308 			next = &graph[node->next[i]];
309 			if (next->type == missed_item.type) {
310 				flow_items[0].type = missed_item.type;
311 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
312 				break;
313 			}
314 			next = NULL;
315 		}
316 	}
317 	if (next && missed) {
318 		elt = 2; /* missed item + item end. */
319 		node = next;
320 		lsize += elt * sizeof(*item) + user_pattern_size;
321 		if ((node->rss_types & types) && lsize <= size) {
322 			buf->entry[buf->entries].priority = 1;
323 			buf->entry[buf->entries].pattern = addr;
324 			buf->entries++;
325 			rte_memcpy(addr, buf->entry[0].pattern,
326 				   user_pattern_size);
327 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
328 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
329 			addr = (void *)(((uintptr_t)addr) +
330 					elt * sizeof(*item));
331 		}
332 	}
333 	memset(flow_items, 0, sizeof(flow_items));
334 	next_node = node->next;
335 	stack[stack_pos] = next_node;
336 	node = next_node ? &graph[*next_node] : NULL;
337 	while (node) {
338 		flow_items[stack_pos].type = node->type;
339 		if (node->rss_types & types) {
340 			/*
341 			 * compute the number of items to copy from the
342 			 * expansion and copy it.
343 			 * When the stack_pos is 0, there are 1 element in it,
344 			 * plus the addition END item.
345 			 */
346 			elt = stack_pos + 2;
347 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
348 			lsize += elt * sizeof(*item) + user_pattern_size;
349 			if (lsize <= size) {
350 				size_t n = elt * sizeof(*item);
351 
352 				buf->entry[buf->entries].priority =
353 					stack_pos + 1 + missed;
354 				buf->entry[buf->entries].pattern = addr;
355 				buf->entries++;
356 				rte_memcpy(addr, buf->entry[0].pattern,
357 					   user_pattern_size);
358 				addr = (void *)(((uintptr_t)addr) +
359 						user_pattern_size);
360 				rte_memcpy(addr, &missed_item,
361 					   missed * sizeof(*item));
362 				addr = (void *)(((uintptr_t)addr) +
363 					missed * sizeof(*item));
364 				rte_memcpy(addr, flow_items, n);
365 				addr = (void *)(((uintptr_t)addr) + n);
366 			}
367 		}
368 		/* Go deeper. */
369 		if (node->next) {
370 			next_node = node->next;
371 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
372 				rte_errno = E2BIG;
373 				return -rte_errno;
374 			}
375 			stack[stack_pos] = next_node;
376 		} else if (*(next_node + 1)) {
377 			/* Follow up with the next possibility. */
378 			++next_node;
379 		} else {
380 			/* Move to the next path. */
381 			if (stack_pos)
382 				next_node = stack[--stack_pos];
383 			next_node++;
384 			stack[stack_pos] = next_node;
385 		}
386 		node = *next_node ? &graph[*next_node] : NULL;
387 	};
388 	/* no expanded flows but we have missed item, create one rule for it */
389 	if (buf->entries == 1 && missed != 0) {
390 		elt = 2;
391 		lsize += elt * sizeof(*item) + user_pattern_size;
392 		if (lsize <= size) {
393 			buf->entry[buf->entries].priority = 1;
394 			buf->entry[buf->entries].pattern = addr;
395 			buf->entries++;
396 			flow_items[0].type = missed_item.type;
397 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
398 			rte_memcpy(addr, buf->entry[0].pattern,
399 				   user_pattern_size);
400 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
401 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
402 		}
403 	}
404 	return lsize;
405 }
406 
407 enum mlx5_expansion {
408 	MLX5_EXPANSION_ROOT,
409 	MLX5_EXPANSION_ROOT_OUTER,
410 	MLX5_EXPANSION_ROOT_ETH_VLAN,
411 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
412 	MLX5_EXPANSION_OUTER_ETH,
413 	MLX5_EXPANSION_OUTER_ETH_VLAN,
414 	MLX5_EXPANSION_OUTER_VLAN,
415 	MLX5_EXPANSION_OUTER_IPV4,
416 	MLX5_EXPANSION_OUTER_IPV4_UDP,
417 	MLX5_EXPANSION_OUTER_IPV4_TCP,
418 	MLX5_EXPANSION_OUTER_IPV6,
419 	MLX5_EXPANSION_OUTER_IPV6_UDP,
420 	MLX5_EXPANSION_OUTER_IPV6_TCP,
421 	MLX5_EXPANSION_VXLAN,
422 	MLX5_EXPANSION_VXLAN_GPE,
423 	MLX5_EXPANSION_GRE,
424 	MLX5_EXPANSION_MPLS,
425 	MLX5_EXPANSION_ETH,
426 	MLX5_EXPANSION_ETH_VLAN,
427 	MLX5_EXPANSION_VLAN,
428 	MLX5_EXPANSION_IPV4,
429 	MLX5_EXPANSION_IPV4_UDP,
430 	MLX5_EXPANSION_IPV4_TCP,
431 	MLX5_EXPANSION_IPV6,
432 	MLX5_EXPANSION_IPV6_UDP,
433 	MLX5_EXPANSION_IPV6_TCP,
434 };
435 
436 /** Supported expansion of items. */
437 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
438 	[MLX5_EXPANSION_ROOT] = {
439 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
440 						  MLX5_EXPANSION_IPV4,
441 						  MLX5_EXPANSION_IPV6),
442 		.type = RTE_FLOW_ITEM_TYPE_END,
443 	},
444 	[MLX5_EXPANSION_ROOT_OUTER] = {
445 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
446 						  MLX5_EXPANSION_OUTER_IPV4,
447 						  MLX5_EXPANSION_OUTER_IPV6),
448 		.type = RTE_FLOW_ITEM_TYPE_END,
449 	},
450 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
451 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
452 		.type = RTE_FLOW_ITEM_TYPE_END,
453 	},
454 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
455 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
456 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
457 		.type = RTE_FLOW_ITEM_TYPE_END,
458 	},
459 	[MLX5_EXPANSION_OUTER_ETH] = {
460 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
461 						  MLX5_EXPANSION_OUTER_IPV6,
462 						  MLX5_EXPANSION_MPLS),
463 		.type = RTE_FLOW_ITEM_TYPE_ETH,
464 		.rss_types = 0,
465 	},
466 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
467 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
468 		.type = RTE_FLOW_ITEM_TYPE_ETH,
469 		.rss_types = 0,
470 	},
471 	[MLX5_EXPANSION_OUTER_VLAN] = {
472 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
473 						  MLX5_EXPANSION_OUTER_IPV6),
474 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
475 	},
476 	[MLX5_EXPANSION_OUTER_IPV4] = {
477 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
478 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
479 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
480 			 MLX5_EXPANSION_GRE,
481 			 MLX5_EXPANSION_IPV4,
482 			 MLX5_EXPANSION_IPV6),
483 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
484 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
485 			ETH_RSS_NONFRAG_IPV4_OTHER,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 						  MLX5_EXPANSION_VXLAN_GPE),
490 		.type = RTE_FLOW_ITEM_TYPE_UDP,
491 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
492 	},
493 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
494 		.type = RTE_FLOW_ITEM_TYPE_TCP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
496 	},
497 	[MLX5_EXPANSION_OUTER_IPV6] = {
498 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
499 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
500 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
501 			 MLX5_EXPANSION_IPV4,
502 			 MLX5_EXPANSION_IPV6),
503 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
504 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
505 			ETH_RSS_NONFRAG_IPV6_OTHER,
506 	},
507 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
508 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
509 						  MLX5_EXPANSION_VXLAN_GPE),
510 		.type = RTE_FLOW_ITEM_TYPE_UDP,
511 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
512 	},
513 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
514 		.type = RTE_FLOW_ITEM_TYPE_TCP,
515 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
516 	},
517 	[MLX5_EXPANSION_VXLAN] = {
518 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
519 						  MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
522 	},
523 	[MLX5_EXPANSION_VXLAN_GPE] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
525 						  MLX5_EXPANSION_IPV4,
526 						  MLX5_EXPANSION_IPV6),
527 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
528 	},
529 	[MLX5_EXPANSION_GRE] = {
530 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
531 		.type = RTE_FLOW_ITEM_TYPE_GRE,
532 	},
533 	[MLX5_EXPANSION_MPLS] = {
534 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
535 						  MLX5_EXPANSION_IPV6),
536 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
537 	},
538 	[MLX5_EXPANSION_ETH] = {
539 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
540 						  MLX5_EXPANSION_IPV6),
541 		.type = RTE_FLOW_ITEM_TYPE_ETH,
542 	},
543 	[MLX5_EXPANSION_ETH_VLAN] = {
544 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
545 		.type = RTE_FLOW_ITEM_TYPE_ETH,
546 	},
547 	[MLX5_EXPANSION_VLAN] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
549 						  MLX5_EXPANSION_IPV6),
550 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
551 	},
552 	[MLX5_EXPANSION_IPV4] = {
553 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
554 						  MLX5_EXPANSION_IPV4_TCP),
555 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
556 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
557 			ETH_RSS_NONFRAG_IPV4_OTHER,
558 	},
559 	[MLX5_EXPANSION_IPV4_UDP] = {
560 		.type = RTE_FLOW_ITEM_TYPE_UDP,
561 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
562 	},
563 	[MLX5_EXPANSION_IPV4_TCP] = {
564 		.type = RTE_FLOW_ITEM_TYPE_TCP,
565 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
566 	},
567 	[MLX5_EXPANSION_IPV6] = {
568 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
569 						  MLX5_EXPANSION_IPV6_TCP),
570 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
571 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
572 			ETH_RSS_NONFRAG_IPV6_OTHER,
573 	},
574 	[MLX5_EXPANSION_IPV6_UDP] = {
575 		.type = RTE_FLOW_ITEM_TYPE_UDP,
576 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
577 	},
578 	[MLX5_EXPANSION_IPV6_TCP] = {
579 		.type = RTE_FLOW_ITEM_TYPE_TCP,
580 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
581 	},
582 };
583 
584 static struct rte_flow_shared_action *
585 mlx5_shared_action_create(struct rte_eth_dev *dev,
586 			  const struct rte_flow_shared_action_conf *conf,
587 			  const struct rte_flow_action *action,
588 			  struct rte_flow_error *error);
589 static int mlx5_shared_action_destroy
590 				(struct rte_eth_dev *dev,
591 				 struct rte_flow_shared_action *shared_action,
592 				 struct rte_flow_error *error);
593 static int mlx5_shared_action_update
594 				(struct rte_eth_dev *dev,
595 				 struct rte_flow_shared_action *shared_action,
596 				 const struct rte_flow_action *action,
597 				 struct rte_flow_error *error);
598 static int mlx5_shared_action_query
599 				(struct rte_eth_dev *dev,
600 				 const struct rte_flow_shared_action *action,
601 				 void *data,
602 				 struct rte_flow_error *error);
603 static int
604 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
605 		    struct rte_flow_tunnel *app_tunnel,
606 		    struct rte_flow_action **actions,
607 		    uint32_t *num_of_actions,
608 		    struct rte_flow_error *error);
609 static int
610 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
611 		       struct rte_flow_tunnel *app_tunnel,
612 		       struct rte_flow_item **items,
613 		       uint32_t *num_of_items,
614 		       struct rte_flow_error *error);
615 static int
616 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
617 			      struct rte_flow_item *pmd_items,
618 			      uint32_t num_items, struct rte_flow_error *err);
619 static int
620 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
621 				struct rte_flow_action *pmd_actions,
622 				uint32_t num_actions,
623 				struct rte_flow_error *err);
624 static int
625 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
626 				  struct rte_mbuf *m,
627 				  struct rte_flow_restore_info *info,
628 				  struct rte_flow_error *err);
629 
630 static const struct rte_flow_ops mlx5_flow_ops = {
631 	.validate = mlx5_flow_validate,
632 	.create = mlx5_flow_create,
633 	.destroy = mlx5_flow_destroy,
634 	.flush = mlx5_flow_flush,
635 	.isolate = mlx5_flow_isolate,
636 	.query = mlx5_flow_query,
637 	.dev_dump = mlx5_flow_dev_dump,
638 	.get_aged_flows = mlx5_flow_get_aged_flows,
639 	.shared_action_create = mlx5_shared_action_create,
640 	.shared_action_destroy = mlx5_shared_action_destroy,
641 	.shared_action_update = mlx5_shared_action_update,
642 	.shared_action_query = mlx5_shared_action_query,
643 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
644 	.tunnel_match = mlx5_flow_tunnel_match,
645 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
646 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
647 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
648 };
649 
650 /* Tunnel information. */
651 struct mlx5_flow_tunnel_info {
652 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
653 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
654 };
655 
656 static struct mlx5_flow_tunnel_info tunnels_info[] = {
657 	{
658 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
659 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
660 	},
661 	{
662 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
663 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
664 	},
665 	{
666 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
667 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
668 	},
669 	{
670 		.tunnel = MLX5_FLOW_LAYER_GRE,
671 		.ptype = RTE_PTYPE_TUNNEL_GRE,
672 	},
673 	{
674 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
675 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
676 	},
677 	{
678 		.tunnel = MLX5_FLOW_LAYER_MPLS,
679 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
680 	},
681 	{
682 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
683 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
684 	},
685 	{
686 		.tunnel = MLX5_FLOW_LAYER_IPIP,
687 		.ptype = RTE_PTYPE_TUNNEL_IP,
688 	},
689 	{
690 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
691 		.ptype = RTE_PTYPE_TUNNEL_IP,
692 	},
693 	{
694 		.tunnel = MLX5_FLOW_LAYER_GTP,
695 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
696 	},
697 };
698 
699 
700 
701 /**
702  * Translate tag ID to register.
703  *
704  * @param[in] dev
705  *   Pointer to the Ethernet device structure.
706  * @param[in] feature
707  *   The feature that request the register.
708  * @param[in] id
709  *   The request register ID.
710  * @param[out] error
711  *   Error description in case of any.
712  *
713  * @return
714  *   The request register on success, a negative errno
715  *   value otherwise and rte_errno is set.
716  */
717 int
718 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
719 		     enum mlx5_feature_name feature,
720 		     uint32_t id,
721 		     struct rte_flow_error *error)
722 {
723 	struct mlx5_priv *priv = dev->data->dev_private;
724 	struct mlx5_dev_config *config = &priv->config;
725 	enum modify_reg start_reg;
726 	bool skip_mtr_reg = false;
727 
728 	switch (feature) {
729 	case MLX5_HAIRPIN_RX:
730 		return REG_B;
731 	case MLX5_HAIRPIN_TX:
732 		return REG_A;
733 	case MLX5_METADATA_RX:
734 		switch (config->dv_xmeta_en) {
735 		case MLX5_XMETA_MODE_LEGACY:
736 			return REG_B;
737 		case MLX5_XMETA_MODE_META16:
738 			return REG_C_0;
739 		case MLX5_XMETA_MODE_META32:
740 			return REG_C_1;
741 		}
742 		break;
743 	case MLX5_METADATA_TX:
744 		return REG_A;
745 	case MLX5_METADATA_FDB:
746 		switch (config->dv_xmeta_en) {
747 		case MLX5_XMETA_MODE_LEGACY:
748 			return REG_NON;
749 		case MLX5_XMETA_MODE_META16:
750 			return REG_C_0;
751 		case MLX5_XMETA_MODE_META32:
752 			return REG_C_1;
753 		}
754 		break;
755 	case MLX5_FLOW_MARK:
756 		switch (config->dv_xmeta_en) {
757 		case MLX5_XMETA_MODE_LEGACY:
758 			return REG_NON;
759 		case MLX5_XMETA_MODE_META16:
760 			return REG_C_1;
761 		case MLX5_XMETA_MODE_META32:
762 			return REG_C_0;
763 		}
764 		break;
765 	case MLX5_MTR_SFX:
766 		/*
767 		 * If meter color and flow match share one register, flow match
768 		 * should use the meter color register for match.
769 		 */
770 		if (priv->mtr_reg_share)
771 			return priv->mtr_color_reg;
772 		else
773 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
774 			       REG_C_3;
775 	case MLX5_MTR_COLOR:
776 	case MLX5_ASO_FLOW_HIT: /* Both features use the same REG_C. */
777 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
778 		return priv->mtr_color_reg;
779 	case MLX5_COPY_MARK:
780 		/*
781 		 * Metadata COPY_MARK register using is in meter suffix sub
782 		 * flow while with meter. It's safe to share the same register.
783 		 */
784 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
785 	case MLX5_APP_TAG:
786 		/*
787 		 * If meter is enable, it will engage the register for color
788 		 * match and flow match. If meter color match is not using the
789 		 * REG_C_2, need to skip the REG_C_x be used by meter color
790 		 * match.
791 		 * If meter is disable, free to use all available registers.
792 		 */
793 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
794 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
795 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
796 		if (id > (uint32_t)(REG_C_7 - start_reg))
797 			return rte_flow_error_set(error, EINVAL,
798 						  RTE_FLOW_ERROR_TYPE_ITEM,
799 						  NULL, "invalid tag id");
800 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
801 			return rte_flow_error_set(error, ENOTSUP,
802 						  RTE_FLOW_ERROR_TYPE_ITEM,
803 						  NULL, "unsupported tag id");
804 		/*
805 		 * This case means meter is using the REG_C_x great than 2.
806 		 * Take care not to conflict with meter color REG_C_x.
807 		 * If the available index REG_C_y >= REG_C_x, skip the
808 		 * color register.
809 		 */
810 		if (skip_mtr_reg && config->flow_mreg_c
811 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
812 			if (id >= (uint32_t)(REG_C_7 - start_reg))
813 				return rte_flow_error_set(error, EINVAL,
814 						       RTE_FLOW_ERROR_TYPE_ITEM,
815 							NULL, "invalid tag id");
816 			if (config->flow_mreg_c
817 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
818 				return config->flow_mreg_c
819 					       [id + 1 + start_reg - REG_C_0];
820 			return rte_flow_error_set(error, ENOTSUP,
821 						  RTE_FLOW_ERROR_TYPE_ITEM,
822 						  NULL, "unsupported tag id");
823 		}
824 		return config->flow_mreg_c[id + start_reg - REG_C_0];
825 	}
826 	MLX5_ASSERT(false);
827 	return rte_flow_error_set(error, EINVAL,
828 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
829 				  NULL, "invalid feature name");
830 }
831 
832 /**
833  * Check extensive flow metadata register support.
834  *
835  * @param dev
836  *   Pointer to rte_eth_dev structure.
837  *
838  * @return
839  *   True if device supports extensive flow metadata register, otherwise false.
840  */
841 bool
842 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
843 {
844 	struct mlx5_priv *priv = dev->data->dev_private;
845 	struct mlx5_dev_config *config = &priv->config;
846 
847 	/*
848 	 * Having available reg_c can be regarded inclusively as supporting
849 	 * extensive flow metadata register, which could mean,
850 	 * - metadata register copy action by modify header.
851 	 * - 16 modify header actions is supported.
852 	 * - reg_c's are preserved across different domain (FDB and NIC) on
853 	 *   packet loopback by flow lookup miss.
854 	 */
855 	return config->flow_mreg_c[2] != REG_NON;
856 }
857 
858 /**
859  * Get the lowest priority.
860  *
861  * @param[in] dev
862  *   Pointer to the Ethernet device structure.
863  * @param[in] attributes
864  *   Pointer to device flow rule attributes.
865  *
866  * @return
867  *   The value of lowest priority of flow.
868  */
869 uint32_t
870 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
871 			  const struct rte_flow_attr *attr)
872 {
873 	struct mlx5_priv *priv = dev->data->dev_private;
874 
875 	if (!attr->group && !attr->transfer)
876 		return priv->config.flow_prio - 2;
877 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
878 }
879 
880 /**
881  * Calculate matcher priority of the flow.
882  *
883  * @param[in] dev
884  *   Pointer to the Ethernet device structure.
885  * @param[in] attr
886  *   Pointer to device flow rule attributes.
887  * @param[in] subpriority
888  *   The priority based on the items.
889  * @return
890  *   The matcher priority of the flow.
891  */
892 uint16_t
893 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
894 			  const struct rte_flow_attr *attr,
895 			  uint32_t subpriority)
896 {
897 	uint16_t priority = (uint16_t)attr->priority;
898 	struct mlx5_priv *priv = dev->data->dev_private;
899 
900 	if (!attr->group && !attr->transfer) {
901 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
902 			priority = priv->config.flow_prio - 1;
903 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
904 	}
905 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
906 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
907 	return priority * 3 + subpriority;
908 }
909 
910 /**
911  * Verify the @p item specifications (spec, last, mask) are compatible with the
912  * NIC capabilities.
913  *
914  * @param[in] item
915  *   Item specification.
916  * @param[in] mask
917  *   @p item->mask or flow default bit-masks.
918  * @param[in] nic_mask
919  *   Bit-masks covering supported fields by the NIC to compare with user mask.
920  * @param[in] size
921  *   Bit-masks size in bytes.
922  * @param[in] range_accepted
923  *   True if range of values is accepted for specific fields, false otherwise.
924  * @param[out] error
925  *   Pointer to error structure.
926  *
927  * @return
928  *   0 on success, a negative errno value otherwise and rte_errno is set.
929  */
930 int
931 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
932 			  const uint8_t *mask,
933 			  const uint8_t *nic_mask,
934 			  unsigned int size,
935 			  bool range_accepted,
936 			  struct rte_flow_error *error)
937 {
938 	unsigned int i;
939 
940 	MLX5_ASSERT(nic_mask);
941 	for (i = 0; i < size; ++i)
942 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
943 			return rte_flow_error_set(error, ENOTSUP,
944 						  RTE_FLOW_ERROR_TYPE_ITEM,
945 						  item,
946 						  "mask enables non supported"
947 						  " bits");
948 	if (!item->spec && (item->mask || item->last))
949 		return rte_flow_error_set(error, EINVAL,
950 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
951 					  "mask/last without a spec is not"
952 					  " supported");
953 	if (item->spec && item->last && !range_accepted) {
954 		uint8_t spec[size];
955 		uint8_t last[size];
956 		unsigned int i;
957 		int ret;
958 
959 		for (i = 0; i < size; ++i) {
960 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
961 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
962 		}
963 		ret = memcmp(spec, last, size);
964 		if (ret != 0)
965 			return rte_flow_error_set(error, EINVAL,
966 						  RTE_FLOW_ERROR_TYPE_ITEM,
967 						  item,
968 						  "range is not valid");
969 	}
970 	return 0;
971 }
972 
973 /**
974  * Adjust the hash fields according to the @p flow information.
975  *
976  * @param[in] dev_flow.
977  *   Pointer to the mlx5_flow.
978  * @param[in] tunnel
979  *   1 when the hash field is for a tunnel item.
980  * @param[in] layer_types
981  *   ETH_RSS_* types.
982  * @param[in] hash_fields
983  *   Item hash fields.
984  *
985  * @return
986  *   The hash fields that should be used.
987  */
988 uint64_t
989 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
990 			    int tunnel __rte_unused, uint64_t layer_types,
991 			    uint64_t hash_fields)
992 {
993 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
994 	int rss_request_inner = rss_desc->level >= 2;
995 
996 	/* Check RSS hash level for tunnel. */
997 	if (tunnel && rss_request_inner)
998 		hash_fields |= IBV_RX_HASH_INNER;
999 	else if (tunnel || rss_request_inner)
1000 		return 0;
1001 #endif
1002 	/* Check if requested layer matches RSS hash fields. */
1003 	if (!(rss_desc->types & layer_types))
1004 		return 0;
1005 	return hash_fields;
1006 }
1007 
1008 /**
1009  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1010  * if several tunnel rules are used on this queue, the tunnel ptype will be
1011  * cleared.
1012  *
1013  * @param rxq_ctrl
1014  *   Rx queue to update.
1015  */
1016 static void
1017 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1018 {
1019 	unsigned int i;
1020 	uint32_t tunnel_ptype = 0;
1021 
1022 	/* Look up for the ptype to use. */
1023 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1024 		if (!rxq_ctrl->flow_tunnels_n[i])
1025 			continue;
1026 		if (!tunnel_ptype) {
1027 			tunnel_ptype = tunnels_info[i].ptype;
1028 		} else {
1029 			tunnel_ptype = 0;
1030 			break;
1031 		}
1032 	}
1033 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1034 }
1035 
1036 /**
1037  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1038  * flow.
1039  *
1040  * @param[in] dev
1041  *   Pointer to the Ethernet device structure.
1042  * @param[in] dev_handle
1043  *   Pointer to device flow handle structure.
1044  */
1045 static void
1046 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1047 		       struct mlx5_flow_handle *dev_handle)
1048 {
1049 	struct mlx5_priv *priv = dev->data->dev_private;
1050 	const int mark = dev_handle->mark;
1051 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1052 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1053 	unsigned int i;
1054 
1055 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1056 		struct mlx5_hrxq *hrxq;
1057 
1058 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1059 			      dev_handle->rix_hrxq);
1060 		if (hrxq)
1061 			ind_tbl = hrxq->ind_table;
1062 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1063 		struct mlx5_shared_action_rss *shared_rss;
1064 
1065 		shared_rss = mlx5_ipool_get
1066 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1067 			 dev_handle->rix_srss);
1068 		if (shared_rss)
1069 			ind_tbl = shared_rss->ind_tbl;
1070 	}
1071 	if (!ind_tbl)
1072 		return;
1073 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1074 		int idx = ind_tbl->queues[i];
1075 		struct mlx5_rxq_ctrl *rxq_ctrl =
1076 			container_of((*priv->rxqs)[idx],
1077 				     struct mlx5_rxq_ctrl, rxq);
1078 
1079 		/*
1080 		 * To support metadata register copy on Tx loopback,
1081 		 * this must be always enabled (metadata may arive
1082 		 * from other port - not from local flows only.
1083 		 */
1084 		if (priv->config.dv_flow_en &&
1085 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1086 		    mlx5_flow_ext_mreg_supported(dev)) {
1087 			rxq_ctrl->rxq.mark = 1;
1088 			rxq_ctrl->flow_mark_n = 1;
1089 		} else if (mark) {
1090 			rxq_ctrl->rxq.mark = 1;
1091 			rxq_ctrl->flow_mark_n++;
1092 		}
1093 		if (tunnel) {
1094 			unsigned int j;
1095 
1096 			/* Increase the counter matching the flow. */
1097 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1098 				if ((tunnels_info[j].tunnel &
1099 				     dev_handle->layers) ==
1100 				    tunnels_info[j].tunnel) {
1101 					rxq_ctrl->flow_tunnels_n[j]++;
1102 					break;
1103 				}
1104 			}
1105 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1106 		}
1107 	}
1108 }
1109 
1110 /**
1111  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1112  *
1113  * @param[in] dev
1114  *   Pointer to the Ethernet device structure.
1115  * @param[in] flow
1116  *   Pointer to flow structure.
1117  */
1118 static void
1119 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1120 {
1121 	struct mlx5_priv *priv = dev->data->dev_private;
1122 	uint32_t handle_idx;
1123 	struct mlx5_flow_handle *dev_handle;
1124 
1125 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1126 		       handle_idx, dev_handle, next)
1127 		flow_drv_rxq_flags_set(dev, dev_handle);
1128 }
1129 
1130 /**
1131  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1132  * device flow if no other flow uses it with the same kind of request.
1133  *
1134  * @param dev
1135  *   Pointer to Ethernet device.
1136  * @param[in] dev_handle
1137  *   Pointer to the device flow handle structure.
1138  */
1139 static void
1140 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1141 			struct mlx5_flow_handle *dev_handle)
1142 {
1143 	struct mlx5_priv *priv = dev->data->dev_private;
1144 	const int mark = dev_handle->mark;
1145 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1146 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1147 	unsigned int i;
1148 
1149 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1150 		struct mlx5_hrxq *hrxq;
1151 
1152 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1153 			      dev_handle->rix_hrxq);
1154 		if (hrxq)
1155 			ind_tbl = hrxq->ind_table;
1156 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1157 		struct mlx5_shared_action_rss *shared_rss;
1158 
1159 		shared_rss = mlx5_ipool_get
1160 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1161 			 dev_handle->rix_srss);
1162 		if (shared_rss)
1163 			ind_tbl = shared_rss->ind_tbl;
1164 	}
1165 	if (!ind_tbl)
1166 		return;
1167 	MLX5_ASSERT(dev->data->dev_started);
1168 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1169 		int idx = ind_tbl->queues[i];
1170 		struct mlx5_rxq_ctrl *rxq_ctrl =
1171 			container_of((*priv->rxqs)[idx],
1172 				     struct mlx5_rxq_ctrl, rxq);
1173 
1174 		if (priv->config.dv_flow_en &&
1175 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1176 		    mlx5_flow_ext_mreg_supported(dev)) {
1177 			rxq_ctrl->rxq.mark = 1;
1178 			rxq_ctrl->flow_mark_n = 1;
1179 		} else if (mark) {
1180 			rxq_ctrl->flow_mark_n--;
1181 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1182 		}
1183 		if (tunnel) {
1184 			unsigned int j;
1185 
1186 			/* Decrease the counter matching the flow. */
1187 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1188 				if ((tunnels_info[j].tunnel &
1189 				     dev_handle->layers) ==
1190 				    tunnels_info[j].tunnel) {
1191 					rxq_ctrl->flow_tunnels_n[j]--;
1192 					break;
1193 				}
1194 			}
1195 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1196 		}
1197 	}
1198 }
1199 
1200 /**
1201  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1202  * @p flow if no other flow uses it with the same kind of request.
1203  *
1204  * @param dev
1205  *   Pointer to Ethernet device.
1206  * @param[in] flow
1207  *   Pointer to the flow.
1208  */
1209 static void
1210 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1211 {
1212 	struct mlx5_priv *priv = dev->data->dev_private;
1213 	uint32_t handle_idx;
1214 	struct mlx5_flow_handle *dev_handle;
1215 
1216 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1217 		       handle_idx, dev_handle, next)
1218 		flow_drv_rxq_flags_trim(dev, dev_handle);
1219 }
1220 
1221 /**
1222  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1223  *
1224  * @param dev
1225  *   Pointer to Ethernet device.
1226  */
1227 static void
1228 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1229 {
1230 	struct mlx5_priv *priv = dev->data->dev_private;
1231 	unsigned int i;
1232 
1233 	for (i = 0; i != priv->rxqs_n; ++i) {
1234 		struct mlx5_rxq_ctrl *rxq_ctrl;
1235 		unsigned int j;
1236 
1237 		if (!(*priv->rxqs)[i])
1238 			continue;
1239 		rxq_ctrl = container_of((*priv->rxqs)[i],
1240 					struct mlx5_rxq_ctrl, rxq);
1241 		rxq_ctrl->flow_mark_n = 0;
1242 		rxq_ctrl->rxq.mark = 0;
1243 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1244 			rxq_ctrl->flow_tunnels_n[j] = 0;
1245 		rxq_ctrl->rxq.tunnel = 0;
1246 	}
1247 }
1248 
1249 /**
1250  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1251  *
1252  * @param[in] dev
1253  *   Pointer to the Ethernet device structure.
1254  */
1255 void
1256 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1257 {
1258 	struct mlx5_priv *priv = dev->data->dev_private;
1259 	struct mlx5_rxq_data *data;
1260 	unsigned int i;
1261 
1262 	for (i = 0; i != priv->rxqs_n; ++i) {
1263 		if (!(*priv->rxqs)[i])
1264 			continue;
1265 		data = (*priv->rxqs)[i];
1266 		if (!rte_flow_dynf_metadata_avail()) {
1267 			data->dynf_meta = 0;
1268 			data->flow_meta_mask = 0;
1269 			data->flow_meta_offset = -1;
1270 			data->flow_meta_port_mask = 0;
1271 		} else {
1272 			data->dynf_meta = 1;
1273 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1274 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1275 			data->flow_meta_port_mask = (uint32_t)~0;
1276 			if (priv->config.dv_xmeta_en == MLX5_XMETA_MODE_META16)
1277 				data->flow_meta_port_mask >>= 16;
1278 		}
1279 	}
1280 }
1281 
1282 /*
1283  * return a pointer to the desired action in the list of actions.
1284  *
1285  * @param[in] actions
1286  *   The list of actions to search the action in.
1287  * @param[in] action
1288  *   The action to find.
1289  *
1290  * @return
1291  *   Pointer to the action in the list, if found. NULL otherwise.
1292  */
1293 const struct rte_flow_action *
1294 mlx5_flow_find_action(const struct rte_flow_action *actions,
1295 		      enum rte_flow_action_type action)
1296 {
1297 	if (actions == NULL)
1298 		return NULL;
1299 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1300 		if (actions->type == action)
1301 			return actions;
1302 	return NULL;
1303 }
1304 
1305 /*
1306  * Validate the flag action.
1307  *
1308  * @param[in] action_flags
1309  *   Bit-fields that holds the actions detected until now.
1310  * @param[in] attr
1311  *   Attributes of flow that includes this action.
1312  * @param[out] error
1313  *   Pointer to error structure.
1314  *
1315  * @return
1316  *   0 on success, a negative errno value otherwise and rte_errno is set.
1317  */
1318 int
1319 mlx5_flow_validate_action_flag(uint64_t action_flags,
1320 			       const struct rte_flow_attr *attr,
1321 			       struct rte_flow_error *error)
1322 {
1323 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1324 		return rte_flow_error_set(error, EINVAL,
1325 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1326 					  "can't mark and flag in same flow");
1327 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1328 		return rte_flow_error_set(error, EINVAL,
1329 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1330 					  "can't have 2 flag"
1331 					  " actions in same flow");
1332 	if (attr->egress)
1333 		return rte_flow_error_set(error, ENOTSUP,
1334 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1335 					  "flag action not supported for "
1336 					  "egress");
1337 	return 0;
1338 }
1339 
1340 /*
1341  * Validate the mark 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] attr
1348  *   Attributes of flow that includes this action.
1349  * @param[out] error
1350  *   Pointer to error structure.
1351  *
1352  * @return
1353  *   0 on success, a negative errno value otherwise and rte_errno is set.
1354  */
1355 int
1356 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1357 			       uint64_t action_flags,
1358 			       const struct rte_flow_attr *attr,
1359 			       struct rte_flow_error *error)
1360 {
1361 	const struct rte_flow_action_mark *mark = action->conf;
1362 
1363 	if (!mark)
1364 		return rte_flow_error_set(error, EINVAL,
1365 					  RTE_FLOW_ERROR_TYPE_ACTION,
1366 					  action,
1367 					  "configuration cannot be null");
1368 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1369 		return rte_flow_error_set(error, EINVAL,
1370 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1371 					  &mark->id,
1372 					  "mark id must in 0 <= id < "
1373 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1374 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1375 		return rte_flow_error_set(error, EINVAL,
1376 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1377 					  "can't flag and mark in same flow");
1378 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1379 		return rte_flow_error_set(error, EINVAL,
1380 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1381 					  "can't have 2 mark actions in same"
1382 					  " flow");
1383 	if (attr->egress)
1384 		return rte_flow_error_set(error, ENOTSUP,
1385 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1386 					  "mark action not supported for "
1387 					  "egress");
1388 	return 0;
1389 }
1390 
1391 /*
1392  * Validate the drop action.
1393  *
1394  * @param[in] action_flags
1395  *   Bit-fields that holds the actions detected until now.
1396  * @param[in] attr
1397  *   Attributes of flow that includes this action.
1398  * @param[out] error
1399  *   Pointer to error structure.
1400  *
1401  * @return
1402  *   0 on success, a negative errno value otherwise and rte_errno is set.
1403  */
1404 int
1405 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1406 			       const struct rte_flow_attr *attr,
1407 			       struct rte_flow_error *error)
1408 {
1409 	if (attr->egress)
1410 		return rte_flow_error_set(error, ENOTSUP,
1411 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1412 					  "drop action not supported for "
1413 					  "egress");
1414 	return 0;
1415 }
1416 
1417 /*
1418  * Validate the queue action.
1419  *
1420  * @param[in] action
1421  *   Pointer to the queue action.
1422  * @param[in] action_flags
1423  *   Bit-fields that holds the actions detected until now.
1424  * @param[in] dev
1425  *   Pointer to the Ethernet device structure.
1426  * @param[in] attr
1427  *   Attributes of flow that includes this action.
1428  * @param[out] error
1429  *   Pointer to error structure.
1430  *
1431  * @return
1432  *   0 on success, a negative errno value otherwise and rte_errno is set.
1433  */
1434 int
1435 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1436 				uint64_t action_flags,
1437 				struct rte_eth_dev *dev,
1438 				const struct rte_flow_attr *attr,
1439 				struct rte_flow_error *error)
1440 {
1441 	struct mlx5_priv *priv = dev->data->dev_private;
1442 	const struct rte_flow_action_queue *queue = action->conf;
1443 
1444 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1445 		return rte_flow_error_set(error, EINVAL,
1446 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1447 					  "can't have 2 fate actions in"
1448 					  " same flow");
1449 	if (!priv->rxqs_n)
1450 		return rte_flow_error_set(error, EINVAL,
1451 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1452 					  NULL, "No Rx queues configured");
1453 	if (queue->index >= priv->rxqs_n)
1454 		return rte_flow_error_set(error, EINVAL,
1455 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1456 					  &queue->index,
1457 					  "queue index out of range");
1458 	if (!(*priv->rxqs)[queue->index])
1459 		return rte_flow_error_set(error, EINVAL,
1460 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1461 					  &queue->index,
1462 					  "queue is not configured");
1463 	if (attr->egress)
1464 		return rte_flow_error_set(error, ENOTSUP,
1465 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1466 					  "queue action not supported for "
1467 					  "egress");
1468 	return 0;
1469 }
1470 
1471 /*
1472  * Validate the rss action.
1473  *
1474  * @param[in] dev
1475  *   Pointer to the Ethernet device structure.
1476  * @param[in] action
1477  *   Pointer to the queue action.
1478  * @param[out] error
1479  *   Pointer to error structure.
1480  *
1481  * @return
1482  *   0 on success, a negative errno value otherwise and rte_errno is set.
1483  */
1484 int
1485 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1486 			 const struct rte_flow_action *action,
1487 			 struct rte_flow_error *error)
1488 {
1489 	struct mlx5_priv *priv = dev->data->dev_private;
1490 	const struct rte_flow_action_rss *rss = action->conf;
1491 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1492 	unsigned int i;
1493 
1494 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1495 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1496 		return rte_flow_error_set(error, ENOTSUP,
1497 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 					  &rss->func,
1499 					  "RSS hash function not supported");
1500 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1501 	if (rss->level > 2)
1502 #else
1503 	if (rss->level > 1)
1504 #endif
1505 		return rte_flow_error_set(error, ENOTSUP,
1506 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1507 					  &rss->level,
1508 					  "tunnel RSS is not supported");
1509 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1510 	if (rss->key_len == 0 && rss->key != NULL)
1511 		return rte_flow_error_set(error, ENOTSUP,
1512 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1513 					  &rss->key_len,
1514 					  "RSS hash key length 0");
1515 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1516 		return rte_flow_error_set(error, ENOTSUP,
1517 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1518 					  &rss->key_len,
1519 					  "RSS hash key too small");
1520 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1521 		return rte_flow_error_set(error, ENOTSUP,
1522 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1523 					  &rss->key_len,
1524 					  "RSS hash key too large");
1525 	if (rss->queue_num > priv->config.ind_table_max_size)
1526 		return rte_flow_error_set(error, ENOTSUP,
1527 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1528 					  &rss->queue_num,
1529 					  "number of queues too large");
1530 	if (rss->types & MLX5_RSS_HF_MASK)
1531 		return rte_flow_error_set(error, ENOTSUP,
1532 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1533 					  &rss->types,
1534 					  "some RSS protocols are not"
1535 					  " supported");
1536 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1537 	    !(rss->types & ETH_RSS_IP))
1538 		return rte_flow_error_set(error, EINVAL,
1539 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1540 					  "L3 partial RSS requested but L3 RSS"
1541 					  " type not specified");
1542 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1543 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1544 		return rte_flow_error_set(error, EINVAL,
1545 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1546 					  "L4 partial RSS requested but L4 RSS"
1547 					  " type not specified");
1548 	if (!priv->rxqs_n)
1549 		return rte_flow_error_set(error, EINVAL,
1550 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1551 					  NULL, "No Rx queues configured");
1552 	if (!rss->queue_num)
1553 		return rte_flow_error_set(error, EINVAL,
1554 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1555 					  NULL, "No queues configured");
1556 	for (i = 0; i != rss->queue_num; ++i) {
1557 		struct mlx5_rxq_ctrl *rxq_ctrl;
1558 
1559 		if (rss->queue[i] >= priv->rxqs_n)
1560 			return rte_flow_error_set
1561 				(error, EINVAL,
1562 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1563 				 &rss->queue[i], "queue index out of range");
1564 		if (!(*priv->rxqs)[rss->queue[i]])
1565 			return rte_flow_error_set
1566 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1567 				 &rss->queue[i], "queue is not configured");
1568 		rxq_ctrl = container_of((*priv->rxqs)[rss->queue[i]],
1569 					struct mlx5_rxq_ctrl, rxq);
1570 		if (i == 0)
1571 			rxq_type = rxq_ctrl->type;
1572 		if (rxq_type != rxq_ctrl->type)
1573 			return rte_flow_error_set
1574 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1575 				 &rss->queue[i],
1576 				 "combining hairpin and regular RSS queues is not supported");
1577 	}
1578 	return 0;
1579 }
1580 
1581 /*
1582  * Validate the rss action.
1583  *
1584  * @param[in] action
1585  *   Pointer to the queue action.
1586  * @param[in] action_flags
1587  *   Bit-fields that holds the actions detected until now.
1588  * @param[in] dev
1589  *   Pointer to the Ethernet device structure.
1590  * @param[in] attr
1591  *   Attributes of flow that includes this action.
1592  * @param[in] item_flags
1593  *   Items that were detected.
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_rss(const struct rte_flow_action *action,
1602 			      uint64_t action_flags,
1603 			      struct rte_eth_dev *dev,
1604 			      const struct rte_flow_attr *attr,
1605 			      uint64_t item_flags,
1606 			      struct rte_flow_error *error)
1607 {
1608 	const struct rte_flow_action_rss *rss = action->conf;
1609 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1610 	int ret;
1611 
1612 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1613 		return rte_flow_error_set(error, EINVAL,
1614 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1615 					  "can't have 2 fate actions"
1616 					  " in same flow");
1617 	ret = mlx5_validate_action_rss(dev, action, error);
1618 	if (ret)
1619 		return ret;
1620 	if (attr->egress)
1621 		return rte_flow_error_set(error, ENOTSUP,
1622 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1623 					  "rss action not supported for "
1624 					  "egress");
1625 	if (rss->level > 1 && !tunnel)
1626 		return rte_flow_error_set(error, EINVAL,
1627 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1628 					  "inner RSS is not supported for "
1629 					  "non-tunnel flows");
1630 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1631 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1632 		return rte_flow_error_set(error, EINVAL,
1633 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1634 					  "RSS on eCPRI is not supported now");
1635 	}
1636 	return 0;
1637 }
1638 
1639 /*
1640  * Validate the default miss action.
1641  *
1642  * @param[in] action_flags
1643  *   Bit-fields that holds the actions detected until now.
1644  * @param[out] error
1645  *   Pointer to error structure.
1646  *
1647  * @return
1648  *   0 on success, a negative errno value otherwise and rte_errno is set.
1649  */
1650 int
1651 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1652 				const struct rte_flow_attr *attr,
1653 				struct rte_flow_error *error)
1654 {
1655 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1656 		return rte_flow_error_set(error, EINVAL,
1657 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1658 					  "can't have 2 fate actions in"
1659 					  " same flow");
1660 	if (attr->egress)
1661 		return rte_flow_error_set(error, ENOTSUP,
1662 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1663 					  "default miss action not supported "
1664 					  "for egress");
1665 	if (attr->group)
1666 		return rte_flow_error_set(error, ENOTSUP,
1667 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1668 					  "only group 0 is supported");
1669 	if (attr->transfer)
1670 		return rte_flow_error_set(error, ENOTSUP,
1671 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1672 					  NULL, "transfer is not supported");
1673 	return 0;
1674 }
1675 
1676 /*
1677  * Validate the count action.
1678  *
1679  * @param[in] dev
1680  *   Pointer to the Ethernet device structure.
1681  * @param[in] attr
1682  *   Attributes of flow that includes this action.
1683  * @param[out] error
1684  *   Pointer to error structure.
1685  *
1686  * @return
1687  *   0 on success, a negative errno value otherwise and rte_errno is set.
1688  */
1689 int
1690 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1691 				const struct rte_flow_attr *attr,
1692 				struct rte_flow_error *error)
1693 {
1694 	if (attr->egress)
1695 		return rte_flow_error_set(error, ENOTSUP,
1696 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1697 					  "count action not supported for "
1698 					  "egress");
1699 	return 0;
1700 }
1701 
1702 /**
1703  * Verify the @p attributes will be correctly understood by the NIC and store
1704  * them in the @p flow if everything is correct.
1705  *
1706  * @param[in] dev
1707  *   Pointer to the Ethernet device structure.
1708  * @param[in] attributes
1709  *   Pointer to flow attributes
1710  * @param[out] error
1711  *   Pointer to error structure.
1712  *
1713  * @return
1714  *   0 on success, a negative errno value otherwise and rte_errno is set.
1715  */
1716 int
1717 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1718 			      const struct rte_flow_attr *attributes,
1719 			      struct rte_flow_error *error)
1720 {
1721 	struct mlx5_priv *priv = dev->data->dev_private;
1722 	uint32_t priority_max = priv->config.flow_prio - 1;
1723 
1724 	if (attributes->group)
1725 		return rte_flow_error_set(error, ENOTSUP,
1726 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1727 					  NULL, "groups is not supported");
1728 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1729 	    attributes->priority >= priority_max)
1730 		return rte_flow_error_set(error, ENOTSUP,
1731 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1732 					  NULL, "priority out of range");
1733 	if (attributes->egress)
1734 		return rte_flow_error_set(error, ENOTSUP,
1735 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1736 					  "egress is not supported");
1737 	if (attributes->transfer && !priv->config.dv_esw_en)
1738 		return rte_flow_error_set(error, ENOTSUP,
1739 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1740 					  NULL, "transfer is not supported");
1741 	if (!attributes->ingress)
1742 		return rte_flow_error_set(error, EINVAL,
1743 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1744 					  NULL,
1745 					  "ingress attribute is mandatory");
1746 	return 0;
1747 }
1748 
1749 /**
1750  * Validate ICMP6 item.
1751  *
1752  * @param[in] item
1753  *   Item specification.
1754  * @param[in] item_flags
1755  *   Bit-fields that holds the items detected until now.
1756  * @param[in] ext_vlan_sup
1757  *   Whether extended VLAN features are supported or not.
1758  * @param[out] error
1759  *   Pointer to error structure.
1760  *
1761  * @return
1762  *   0 on success, a negative errno value otherwise and rte_errno is set.
1763  */
1764 int
1765 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1766 			       uint64_t item_flags,
1767 			       uint8_t target_protocol,
1768 			       struct rte_flow_error *error)
1769 {
1770 	const struct rte_flow_item_icmp6 *mask = item->mask;
1771 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1772 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1773 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1774 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1775 				      MLX5_FLOW_LAYER_OUTER_L4;
1776 	int ret;
1777 
1778 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1779 		return rte_flow_error_set(error, EINVAL,
1780 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1781 					  "protocol filtering not compatible"
1782 					  " with ICMP6 layer");
1783 	if (!(item_flags & l3m))
1784 		return rte_flow_error_set(error, EINVAL,
1785 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1786 					  "IPv6 is mandatory to filter on"
1787 					  " ICMP6");
1788 	if (item_flags & l4m)
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1791 					  "multiple L4 layers not supported");
1792 	if (!mask)
1793 		mask = &rte_flow_item_icmp6_mask;
1794 	ret = mlx5_flow_item_acceptable
1795 		(item, (const uint8_t *)mask,
1796 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1797 		 sizeof(struct rte_flow_item_icmp6),
1798 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1799 	if (ret < 0)
1800 		return ret;
1801 	return 0;
1802 }
1803 
1804 /**
1805  * Validate ICMP item.
1806  *
1807  * @param[in] item
1808  *   Item specification.
1809  * @param[in] item_flags
1810  *   Bit-fields that holds the items detected until now.
1811  * @param[out] error
1812  *   Pointer to error structure.
1813  *
1814  * @return
1815  *   0 on success, a negative errno value otherwise and rte_errno is set.
1816  */
1817 int
1818 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1819 			     uint64_t item_flags,
1820 			     uint8_t target_protocol,
1821 			     struct rte_flow_error *error)
1822 {
1823 	const struct rte_flow_item_icmp *mask = item->mask;
1824 	const struct rte_flow_item_icmp nic_mask = {
1825 		.hdr.icmp_type = 0xff,
1826 		.hdr.icmp_code = 0xff,
1827 		.hdr.icmp_ident = RTE_BE16(0xffff),
1828 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1829 	};
1830 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1831 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1832 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1833 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1834 				      MLX5_FLOW_LAYER_OUTER_L4;
1835 	int ret;
1836 
1837 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1838 		return rte_flow_error_set(error, EINVAL,
1839 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1840 					  "protocol filtering not compatible"
1841 					  " with ICMP layer");
1842 	if (!(item_flags & l3m))
1843 		return rte_flow_error_set(error, EINVAL,
1844 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1845 					  "IPv4 is mandatory to filter"
1846 					  " on ICMP");
1847 	if (item_flags & l4m)
1848 		return rte_flow_error_set(error, EINVAL,
1849 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1850 					  "multiple L4 layers not supported");
1851 	if (!mask)
1852 		mask = &nic_mask;
1853 	ret = mlx5_flow_item_acceptable
1854 		(item, (const uint8_t *)mask,
1855 		 (const uint8_t *)&nic_mask,
1856 		 sizeof(struct rte_flow_item_icmp),
1857 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1858 	if (ret < 0)
1859 		return ret;
1860 	return 0;
1861 }
1862 
1863 /**
1864  * Validate Ethernet item.
1865  *
1866  * @param[in] item
1867  *   Item specification.
1868  * @param[in] item_flags
1869  *   Bit-fields that holds the items detected until now.
1870  * @param[out] error
1871  *   Pointer to error structure.
1872  *
1873  * @return
1874  *   0 on success, a negative errno value otherwise and rte_errno is set.
1875  */
1876 int
1877 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1878 			    uint64_t item_flags, bool ext_vlan_sup,
1879 			    struct rte_flow_error *error)
1880 {
1881 	const struct rte_flow_item_eth *mask = item->mask;
1882 	const struct rte_flow_item_eth nic_mask = {
1883 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1884 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1885 		.type = RTE_BE16(0xffff),
1886 		.has_vlan = ext_vlan_sup ? 1 : 0,
1887 	};
1888 	int ret;
1889 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1890 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1891 				       MLX5_FLOW_LAYER_OUTER_L2;
1892 
1893 	if (item_flags & ethm)
1894 		return rte_flow_error_set(error, ENOTSUP,
1895 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1896 					  "multiple L2 layers not supported");
1897 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1898 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1899 		return rte_flow_error_set(error, EINVAL,
1900 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1901 					  "L2 layer should not follow "
1902 					  "L3 layers");
1903 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1904 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1905 		return rte_flow_error_set(error, EINVAL,
1906 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1907 					  "L2 layer should not follow VLAN");
1908 	if (!mask)
1909 		mask = &rte_flow_item_eth_mask;
1910 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1911 					(const uint8_t *)&nic_mask,
1912 					sizeof(struct rte_flow_item_eth),
1913 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1914 	return ret;
1915 }
1916 
1917 /**
1918  * Validate VLAN item.
1919  *
1920  * @param[in] item
1921  *   Item specification.
1922  * @param[in] item_flags
1923  *   Bit-fields that holds the items detected until now.
1924  * @param[in] dev
1925  *   Ethernet device flow is being created on.
1926  * @param[out] error
1927  *   Pointer to error structure.
1928  *
1929  * @return
1930  *   0 on success, a negative errno value otherwise and rte_errno is set.
1931  */
1932 int
1933 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1934 			     uint64_t item_flags,
1935 			     struct rte_eth_dev *dev,
1936 			     struct rte_flow_error *error)
1937 {
1938 	const struct rte_flow_item_vlan *spec = item->spec;
1939 	const struct rte_flow_item_vlan *mask = item->mask;
1940 	const struct rte_flow_item_vlan nic_mask = {
1941 		.tci = RTE_BE16(UINT16_MAX),
1942 		.inner_type = RTE_BE16(UINT16_MAX),
1943 	};
1944 	uint16_t vlan_tag = 0;
1945 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1946 	int ret;
1947 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1948 					MLX5_FLOW_LAYER_INNER_L4) :
1949 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1950 					MLX5_FLOW_LAYER_OUTER_L4);
1951 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1952 					MLX5_FLOW_LAYER_OUTER_VLAN;
1953 
1954 	if (item_flags & vlanm)
1955 		return rte_flow_error_set(error, EINVAL,
1956 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1957 					  "multiple VLAN layers not supported");
1958 	else if ((item_flags & l34m) != 0)
1959 		return rte_flow_error_set(error, EINVAL,
1960 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1961 					  "VLAN cannot follow L3/L4 layer");
1962 	if (!mask)
1963 		mask = &rte_flow_item_vlan_mask;
1964 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1965 					(const uint8_t *)&nic_mask,
1966 					sizeof(struct rte_flow_item_vlan),
1967 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1968 	if (ret)
1969 		return ret;
1970 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1971 		struct mlx5_priv *priv = dev->data->dev_private;
1972 
1973 		if (priv->vmwa_context) {
1974 			/*
1975 			 * Non-NULL context means we have a virtual machine
1976 			 * and SR-IOV enabled, we have to create VLAN interface
1977 			 * to make hypervisor to setup E-Switch vport
1978 			 * context correctly. We avoid creating the multiple
1979 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1980 			 */
1981 			return rte_flow_error_set(error, EINVAL,
1982 						  RTE_FLOW_ERROR_TYPE_ITEM,
1983 						  item,
1984 						  "VLAN tag mask is not"
1985 						  " supported in virtual"
1986 						  " environment");
1987 		}
1988 	}
1989 	if (spec) {
1990 		vlan_tag = spec->tci;
1991 		vlan_tag &= mask->tci;
1992 	}
1993 	/*
1994 	 * From verbs perspective an empty VLAN is equivalent
1995 	 * to a packet without VLAN layer.
1996 	 */
1997 	if (!vlan_tag)
1998 		return rte_flow_error_set(error, EINVAL,
1999 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2000 					  item->spec,
2001 					  "VLAN cannot be empty");
2002 	return 0;
2003 }
2004 
2005 /**
2006  * Validate IPV4 item.
2007  *
2008  * @param[in] item
2009  *   Item specification.
2010  * @param[in] item_flags
2011  *   Bit-fields that holds the items detected until now.
2012  * @param[in] last_item
2013  *   Previous validated item in the pattern items.
2014  * @param[in] ether_type
2015  *   Type in the ethernet layer header (including dot1q).
2016  * @param[in] acc_mask
2017  *   Acceptable mask, if NULL default internal default mask
2018  *   will be used to check whether item fields are supported.
2019  * @param[in] range_accepted
2020  *   True if range of values is accepted for specific fields, false otherwise.
2021  * @param[out] error
2022  *   Pointer to error structure.
2023  *
2024  * @return
2025  *   0 on success, a negative errno value otherwise and rte_errno is set.
2026  */
2027 int
2028 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2029 			     uint64_t item_flags,
2030 			     uint64_t last_item,
2031 			     uint16_t ether_type,
2032 			     const struct rte_flow_item_ipv4 *acc_mask,
2033 			     bool range_accepted,
2034 			     struct rte_flow_error *error)
2035 {
2036 	const struct rte_flow_item_ipv4 *mask = item->mask;
2037 	const struct rte_flow_item_ipv4 *spec = item->spec;
2038 	const struct rte_flow_item_ipv4 nic_mask = {
2039 		.hdr = {
2040 			.src_addr = RTE_BE32(0xffffffff),
2041 			.dst_addr = RTE_BE32(0xffffffff),
2042 			.type_of_service = 0xff,
2043 			.next_proto_id = 0xff,
2044 		},
2045 	};
2046 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2047 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2048 				      MLX5_FLOW_LAYER_OUTER_L3;
2049 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2050 				      MLX5_FLOW_LAYER_OUTER_L4;
2051 	int ret;
2052 	uint8_t next_proto = 0xFF;
2053 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2054 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2055 				  MLX5_FLOW_LAYER_INNER_VLAN);
2056 
2057 	if ((last_item & l2_vlan) && ether_type &&
2058 	    ether_type != RTE_ETHER_TYPE_IPV4)
2059 		return rte_flow_error_set(error, EINVAL,
2060 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2061 					  "IPv4 cannot follow L2/VLAN layer "
2062 					  "which ether type is not IPv4");
2063 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2064 		if (mask && spec)
2065 			next_proto = mask->hdr.next_proto_id &
2066 				     spec->hdr.next_proto_id;
2067 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2068 			return rte_flow_error_set(error, EINVAL,
2069 						  RTE_FLOW_ERROR_TYPE_ITEM,
2070 						  item,
2071 						  "multiple tunnel "
2072 						  "not supported");
2073 	}
2074 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2075 		return rte_flow_error_set(error, EINVAL,
2076 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2077 					  "wrong tunnel type - IPv6 specified "
2078 					  "but IPv4 item provided");
2079 	if (item_flags & l3m)
2080 		return rte_flow_error_set(error, ENOTSUP,
2081 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2082 					  "multiple L3 layers not supported");
2083 	else if (item_flags & l4m)
2084 		return rte_flow_error_set(error, EINVAL,
2085 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2086 					  "L3 cannot follow an L4 layer.");
2087 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2088 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2089 		return rte_flow_error_set(error, EINVAL,
2090 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2091 					  "L3 cannot follow an NVGRE layer.");
2092 	if (!mask)
2093 		mask = &rte_flow_item_ipv4_mask;
2094 	else if (mask->hdr.next_proto_id != 0 &&
2095 		 mask->hdr.next_proto_id != 0xff)
2096 		return rte_flow_error_set(error, EINVAL,
2097 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2098 					  "partial mask is not supported"
2099 					  " for protocol");
2100 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2101 					acc_mask ? (const uint8_t *)acc_mask
2102 						 : (const uint8_t *)&nic_mask,
2103 					sizeof(struct rte_flow_item_ipv4),
2104 					range_accepted, error);
2105 	if (ret < 0)
2106 		return ret;
2107 	return 0;
2108 }
2109 
2110 /**
2111  * Validate IPV6 item.
2112  *
2113  * @param[in] item
2114  *   Item specification.
2115  * @param[in] item_flags
2116  *   Bit-fields that holds the items detected until now.
2117  * @param[in] last_item
2118  *   Previous validated item in the pattern items.
2119  * @param[in] ether_type
2120  *   Type in the ethernet layer header (including dot1q).
2121  * @param[in] acc_mask
2122  *   Acceptable mask, if NULL default internal default mask
2123  *   will be used to check whether item fields are supported.
2124  * @param[out] error
2125  *   Pointer to error structure.
2126  *
2127  * @return
2128  *   0 on success, a negative errno value otherwise and rte_errno is set.
2129  */
2130 int
2131 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2132 			     uint64_t item_flags,
2133 			     uint64_t last_item,
2134 			     uint16_t ether_type,
2135 			     const struct rte_flow_item_ipv6 *acc_mask,
2136 			     struct rte_flow_error *error)
2137 {
2138 	const struct rte_flow_item_ipv6 *mask = item->mask;
2139 	const struct rte_flow_item_ipv6 *spec = item->spec;
2140 	const struct rte_flow_item_ipv6 nic_mask = {
2141 		.hdr = {
2142 			.src_addr =
2143 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2144 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2145 			.dst_addr =
2146 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2147 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2148 			.vtc_flow = RTE_BE32(0xffffffff),
2149 			.proto = 0xff,
2150 		},
2151 	};
2152 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2153 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2154 				      MLX5_FLOW_LAYER_OUTER_L3;
2155 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2156 				      MLX5_FLOW_LAYER_OUTER_L4;
2157 	int ret;
2158 	uint8_t next_proto = 0xFF;
2159 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2160 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2161 				  MLX5_FLOW_LAYER_INNER_VLAN);
2162 
2163 	if ((last_item & l2_vlan) && ether_type &&
2164 	    ether_type != RTE_ETHER_TYPE_IPV6)
2165 		return rte_flow_error_set(error, EINVAL,
2166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 					  "IPv6 cannot follow L2/VLAN layer "
2168 					  "which ether type is not IPv6");
2169 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2170 		next_proto = spec->hdr.proto;
2171 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2172 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2173 			return rte_flow_error_set(error, EINVAL,
2174 						  RTE_FLOW_ERROR_TYPE_ITEM,
2175 						  item,
2176 						  "multiple tunnel "
2177 						  "not supported");
2178 	}
2179 	if (next_proto == IPPROTO_HOPOPTS  ||
2180 	    next_proto == IPPROTO_ROUTING  ||
2181 	    next_proto == IPPROTO_FRAGMENT ||
2182 	    next_proto == IPPROTO_ESP	   ||
2183 	    next_proto == IPPROTO_AH	   ||
2184 	    next_proto == IPPROTO_DSTOPTS)
2185 		return rte_flow_error_set(error, EINVAL,
2186 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2187 					  "IPv6 proto (next header) should "
2188 					  "not be set as extension header");
2189 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2190 		return rte_flow_error_set(error, EINVAL,
2191 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2192 					  "wrong tunnel type - IPv4 specified "
2193 					  "but IPv6 item provided");
2194 	if (item_flags & l3m)
2195 		return rte_flow_error_set(error, ENOTSUP,
2196 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 					  "multiple L3 layers not supported");
2198 	else if (item_flags & l4m)
2199 		return rte_flow_error_set(error, EINVAL,
2200 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2201 					  "L3 cannot follow an L4 layer.");
2202 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2203 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2204 		return rte_flow_error_set(error, EINVAL,
2205 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2206 					  "L3 cannot follow an NVGRE layer.");
2207 	if (!mask)
2208 		mask = &rte_flow_item_ipv6_mask;
2209 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2210 					acc_mask ? (const uint8_t *)acc_mask
2211 						 : (const uint8_t *)&nic_mask,
2212 					sizeof(struct rte_flow_item_ipv6),
2213 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2214 	if (ret < 0)
2215 		return ret;
2216 	return 0;
2217 }
2218 
2219 /**
2220  * Validate UDP item.
2221  *
2222  * @param[in] item
2223  *   Item specification.
2224  * @param[in] item_flags
2225  *   Bit-fields that holds the items detected until now.
2226  * @param[in] target_protocol
2227  *   The next protocol in the previous item.
2228  * @param[in] flow_mask
2229  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2230  * @param[out] error
2231  *   Pointer to error structure.
2232  *
2233  * @return
2234  *   0 on success, a negative errno value otherwise and rte_errno is set.
2235  */
2236 int
2237 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2238 			    uint64_t item_flags,
2239 			    uint8_t target_protocol,
2240 			    struct rte_flow_error *error)
2241 {
2242 	const struct rte_flow_item_udp *mask = item->mask;
2243 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2244 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2245 				      MLX5_FLOW_LAYER_OUTER_L3;
2246 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2247 				      MLX5_FLOW_LAYER_OUTER_L4;
2248 	int ret;
2249 
2250 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2251 		return rte_flow_error_set(error, EINVAL,
2252 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2253 					  "protocol filtering not compatible"
2254 					  " with UDP layer");
2255 	if (!(item_flags & l3m))
2256 		return rte_flow_error_set(error, EINVAL,
2257 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2258 					  "L3 is mandatory to filter on L4");
2259 	if (item_flags & l4m)
2260 		return rte_flow_error_set(error, EINVAL,
2261 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2262 					  "multiple L4 layers not supported");
2263 	if (!mask)
2264 		mask = &rte_flow_item_udp_mask;
2265 	ret = mlx5_flow_item_acceptable
2266 		(item, (const uint8_t *)mask,
2267 		 (const uint8_t *)&rte_flow_item_udp_mask,
2268 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2269 		 error);
2270 	if (ret < 0)
2271 		return ret;
2272 	return 0;
2273 }
2274 
2275 /**
2276  * Validate TCP item.
2277  *
2278  * @param[in] item
2279  *   Item specification.
2280  * @param[in] item_flags
2281  *   Bit-fields that holds the items detected until now.
2282  * @param[in] target_protocol
2283  *   The next protocol in the previous item.
2284  * @param[out] error
2285  *   Pointer to error structure.
2286  *
2287  * @return
2288  *   0 on success, a negative errno value otherwise and rte_errno is set.
2289  */
2290 int
2291 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2292 			    uint64_t item_flags,
2293 			    uint8_t target_protocol,
2294 			    const struct rte_flow_item_tcp *flow_mask,
2295 			    struct rte_flow_error *error)
2296 {
2297 	const struct rte_flow_item_tcp *mask = item->mask;
2298 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2299 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2300 				      MLX5_FLOW_LAYER_OUTER_L3;
2301 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2302 				      MLX5_FLOW_LAYER_OUTER_L4;
2303 	int ret;
2304 
2305 	MLX5_ASSERT(flow_mask);
2306 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2307 		return rte_flow_error_set(error, EINVAL,
2308 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2309 					  "protocol filtering not compatible"
2310 					  " with TCP layer");
2311 	if (!(item_flags & l3m))
2312 		return rte_flow_error_set(error, EINVAL,
2313 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2314 					  "L3 is mandatory to filter on L4");
2315 	if (item_flags & l4m)
2316 		return rte_flow_error_set(error, EINVAL,
2317 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2318 					  "multiple L4 layers not supported");
2319 	if (!mask)
2320 		mask = &rte_flow_item_tcp_mask;
2321 	ret = mlx5_flow_item_acceptable
2322 		(item, (const uint8_t *)mask,
2323 		 (const uint8_t *)flow_mask,
2324 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2325 		 error);
2326 	if (ret < 0)
2327 		return ret;
2328 	return 0;
2329 }
2330 
2331 /**
2332  * Validate VXLAN item.
2333  *
2334  * @param[in] item
2335  *   Item specification.
2336  * @param[in] item_flags
2337  *   Bit-fields that holds the items detected until now.
2338  * @param[in] target_protocol
2339  *   The next protocol in the previous item.
2340  * @param[out] error
2341  *   Pointer to error structure.
2342  *
2343  * @return
2344  *   0 on success, a negative errno value otherwise and rte_errno is set.
2345  */
2346 int
2347 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2348 			      uint64_t item_flags,
2349 			      struct rte_flow_error *error)
2350 {
2351 	const struct rte_flow_item_vxlan *spec = item->spec;
2352 	const struct rte_flow_item_vxlan *mask = item->mask;
2353 	int ret;
2354 	union vni {
2355 		uint32_t vlan_id;
2356 		uint8_t vni[4];
2357 	} id = { .vlan_id = 0, };
2358 
2359 
2360 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2361 		return rte_flow_error_set(error, ENOTSUP,
2362 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2363 					  "multiple tunnel layers not"
2364 					  " supported");
2365 	/*
2366 	 * Verify only UDPv4 is present as defined in
2367 	 * https://tools.ietf.org/html/rfc7348
2368 	 */
2369 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2370 		return rte_flow_error_set(error, EINVAL,
2371 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2372 					  "no outer UDP layer found");
2373 	if (!mask)
2374 		mask = &rte_flow_item_vxlan_mask;
2375 	ret = mlx5_flow_item_acceptable
2376 		(item, (const uint8_t *)mask,
2377 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2378 		 sizeof(struct rte_flow_item_vxlan),
2379 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2380 	if (ret < 0)
2381 		return ret;
2382 	if (spec) {
2383 		memcpy(&id.vni[1], spec->vni, 3);
2384 		memcpy(&id.vni[1], mask->vni, 3);
2385 	}
2386 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2387 		return rte_flow_error_set(error, ENOTSUP,
2388 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2389 					  "VXLAN tunnel must be fully defined");
2390 	return 0;
2391 }
2392 
2393 /**
2394  * Validate VXLAN_GPE item.
2395  *
2396  * @param[in] item
2397  *   Item specification.
2398  * @param[in] item_flags
2399  *   Bit-fields that holds the items detected until now.
2400  * @param[in] priv
2401  *   Pointer to the private data structure.
2402  * @param[in] target_protocol
2403  *   The next protocol in the previous item.
2404  * @param[out] error
2405  *   Pointer to error structure.
2406  *
2407  * @return
2408  *   0 on success, a negative errno value otherwise and rte_errno is set.
2409  */
2410 int
2411 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2412 				  uint64_t item_flags,
2413 				  struct rte_eth_dev *dev,
2414 				  struct rte_flow_error *error)
2415 {
2416 	struct mlx5_priv *priv = dev->data->dev_private;
2417 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2418 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2419 	int ret;
2420 	union vni {
2421 		uint32_t vlan_id;
2422 		uint8_t vni[4];
2423 	} id = { .vlan_id = 0, };
2424 
2425 	if (!priv->config.l3_vxlan_en)
2426 		return rte_flow_error_set(error, ENOTSUP,
2427 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2428 					  "L3 VXLAN is not enabled by device"
2429 					  " parameter and/or not configured in"
2430 					  " firmware");
2431 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2432 		return rte_flow_error_set(error, ENOTSUP,
2433 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2434 					  "multiple tunnel layers not"
2435 					  " supported");
2436 	/*
2437 	 * Verify only UDPv4 is present as defined in
2438 	 * https://tools.ietf.org/html/rfc7348
2439 	 */
2440 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2441 		return rte_flow_error_set(error, EINVAL,
2442 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2443 					  "no outer UDP layer found");
2444 	if (!mask)
2445 		mask = &rte_flow_item_vxlan_gpe_mask;
2446 	ret = mlx5_flow_item_acceptable
2447 		(item, (const uint8_t *)mask,
2448 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2449 		 sizeof(struct rte_flow_item_vxlan_gpe),
2450 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2451 	if (ret < 0)
2452 		return ret;
2453 	if (spec) {
2454 		if (spec->protocol)
2455 			return rte_flow_error_set(error, ENOTSUP,
2456 						  RTE_FLOW_ERROR_TYPE_ITEM,
2457 						  item,
2458 						  "VxLAN-GPE protocol"
2459 						  " not supported");
2460 		memcpy(&id.vni[1], spec->vni, 3);
2461 		memcpy(&id.vni[1], mask->vni, 3);
2462 	}
2463 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2464 		return rte_flow_error_set(error, ENOTSUP,
2465 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2466 					  "VXLAN-GPE tunnel must be fully"
2467 					  " defined");
2468 	return 0;
2469 }
2470 /**
2471  * Validate GRE Key item.
2472  *
2473  * @param[in] item
2474  *   Item specification.
2475  * @param[in] item_flags
2476  *   Bit flags to mark detected items.
2477  * @param[in] gre_item
2478  *   Pointer to gre_item
2479  * @param[out] error
2480  *   Pointer to error structure.
2481  *
2482  * @return
2483  *   0 on success, a negative errno value otherwise and rte_errno is set.
2484  */
2485 int
2486 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2487 				uint64_t item_flags,
2488 				const struct rte_flow_item *gre_item,
2489 				struct rte_flow_error *error)
2490 {
2491 	const rte_be32_t *mask = item->mask;
2492 	int ret = 0;
2493 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2494 	const struct rte_flow_item_gre *gre_spec;
2495 	const struct rte_flow_item_gre *gre_mask;
2496 
2497 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2498 		return rte_flow_error_set(error, ENOTSUP,
2499 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2500 					  "Multiple GRE key not support");
2501 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2502 		return rte_flow_error_set(error, ENOTSUP,
2503 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2504 					  "No preceding GRE header");
2505 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2506 		return rte_flow_error_set(error, ENOTSUP,
2507 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2508 					  "GRE key following a wrong item");
2509 	gre_mask = gre_item->mask;
2510 	if (!gre_mask)
2511 		gre_mask = &rte_flow_item_gre_mask;
2512 	gre_spec = gre_item->spec;
2513 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2514 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2515 		return rte_flow_error_set(error, EINVAL,
2516 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2517 					  "Key bit must be on");
2518 
2519 	if (!mask)
2520 		mask = &gre_key_default_mask;
2521 	ret = mlx5_flow_item_acceptable
2522 		(item, (const uint8_t *)mask,
2523 		 (const uint8_t *)&gre_key_default_mask,
2524 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2525 	return ret;
2526 }
2527 
2528 /**
2529  * Validate GRE item.
2530  *
2531  * @param[in] item
2532  *   Item specification.
2533  * @param[in] item_flags
2534  *   Bit flags to mark detected items.
2535  * @param[in] target_protocol
2536  *   The next protocol in the previous item.
2537  * @param[out] error
2538  *   Pointer to error structure.
2539  *
2540  * @return
2541  *   0 on success, a negative errno value otherwise and rte_errno is set.
2542  */
2543 int
2544 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2545 			    uint64_t item_flags,
2546 			    uint8_t target_protocol,
2547 			    struct rte_flow_error *error)
2548 {
2549 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2550 	const struct rte_flow_item_gre *mask = item->mask;
2551 	int ret;
2552 	const struct rte_flow_item_gre nic_mask = {
2553 		.c_rsvd0_ver = RTE_BE16(0xB000),
2554 		.protocol = RTE_BE16(UINT16_MAX),
2555 	};
2556 
2557 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2558 		return rte_flow_error_set(error, EINVAL,
2559 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2560 					  "protocol filtering not compatible"
2561 					  " with this GRE layer");
2562 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2563 		return rte_flow_error_set(error, ENOTSUP,
2564 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2565 					  "multiple tunnel layers not"
2566 					  " supported");
2567 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2568 		return rte_flow_error_set(error, ENOTSUP,
2569 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2570 					  "L3 Layer is missing");
2571 	if (!mask)
2572 		mask = &rte_flow_item_gre_mask;
2573 	ret = mlx5_flow_item_acceptable
2574 		(item, (const uint8_t *)mask,
2575 		 (const uint8_t *)&nic_mask,
2576 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2577 		 error);
2578 	if (ret < 0)
2579 		return ret;
2580 #ifndef HAVE_MLX5DV_DR
2581 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2582 	if (spec && (spec->protocol & mask->protocol))
2583 		return rte_flow_error_set(error, ENOTSUP,
2584 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2585 					  "without MPLS support the"
2586 					  " specification cannot be used for"
2587 					  " filtering");
2588 #endif
2589 #endif
2590 	return 0;
2591 }
2592 
2593 /**
2594  * Validate Geneve item.
2595  *
2596  * @param[in] item
2597  *   Item specification.
2598  * @param[in] itemFlags
2599  *   Bit-fields that holds the items detected until now.
2600  * @param[in] enPriv
2601  *   Pointer to the private data structure.
2602  * @param[out] error
2603  *   Pointer to error structure.
2604  *
2605  * @return
2606  *   0 on success, a negative errno value otherwise and rte_errno is set.
2607  */
2608 
2609 int
2610 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2611 			       uint64_t item_flags,
2612 			       struct rte_eth_dev *dev,
2613 			       struct rte_flow_error *error)
2614 {
2615 	struct mlx5_priv *priv = dev->data->dev_private;
2616 	const struct rte_flow_item_geneve *spec = item->spec;
2617 	const struct rte_flow_item_geneve *mask = item->mask;
2618 	int ret;
2619 	uint16_t gbhdr;
2620 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2621 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2622 	const struct rte_flow_item_geneve nic_mask = {
2623 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2624 		.vni = "\xff\xff\xff",
2625 		.protocol = RTE_BE16(UINT16_MAX),
2626 	};
2627 
2628 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2629 		return rte_flow_error_set(error, ENOTSUP,
2630 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2631 					  "L3 Geneve is not enabled by device"
2632 					  " parameter and/or not configured in"
2633 					  " firmware");
2634 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2635 		return rte_flow_error_set(error, ENOTSUP,
2636 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2637 					  "multiple tunnel layers not"
2638 					  " supported");
2639 	/*
2640 	 * Verify only UDPv4 is present as defined in
2641 	 * https://tools.ietf.org/html/rfc7348
2642 	 */
2643 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2644 		return rte_flow_error_set(error, EINVAL,
2645 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2646 					  "no outer UDP layer found");
2647 	if (!mask)
2648 		mask = &rte_flow_item_geneve_mask;
2649 	ret = mlx5_flow_item_acceptable
2650 				  (item, (const uint8_t *)mask,
2651 				   (const uint8_t *)&nic_mask,
2652 				   sizeof(struct rte_flow_item_geneve),
2653 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2654 	if (ret)
2655 		return ret;
2656 	if (spec) {
2657 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2658 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2659 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2660 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2661 			return rte_flow_error_set(error, ENOTSUP,
2662 						  RTE_FLOW_ERROR_TYPE_ITEM,
2663 						  item,
2664 						  "Geneve protocol unsupported"
2665 						  " fields are being used");
2666 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2667 			return rte_flow_error_set
2668 					(error, ENOTSUP,
2669 					 RTE_FLOW_ERROR_TYPE_ITEM,
2670 					 item,
2671 					 "Unsupported Geneve options length");
2672 	}
2673 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2674 		return rte_flow_error_set
2675 				    (error, ENOTSUP,
2676 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2677 				     "Geneve tunnel must be fully defined");
2678 	return 0;
2679 }
2680 
2681 /**
2682  * Validate Geneve TLV option item.
2683  *
2684  * @param[in] item
2685  *   Item specification.
2686  * @param[in] last_item
2687  *   Previous validated item in the pattern items.
2688  * @param[in] geneve_item
2689  *   Previous GENEVE item specification.
2690  * @param[in] dev
2691  *   Pointer to the rte_eth_dev structure.
2692  * @param[out] error
2693  *   Pointer to error structure.
2694  *
2695  * @return
2696  *   0 on success, a negative errno value otherwise and rte_errno is set.
2697  */
2698 int
2699 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2700 				   uint64_t last_item,
2701 				   const struct rte_flow_item *geneve_item,
2702 				   struct rte_eth_dev *dev,
2703 				   struct rte_flow_error *error)
2704 {
2705 	struct mlx5_priv *priv = dev->data->dev_private;
2706 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2707 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2708 	struct mlx5_hca_attr *hca_attr = &priv->config.hca_attr;
2709 	uint8_t data_max_supported =
2710 			hca_attr->max_geneve_tlv_option_data_len * 4;
2711 	struct mlx5_dev_config *config = &priv->config;
2712 	const struct rte_flow_item_geneve *geneve_spec;
2713 	const struct rte_flow_item_geneve *geneve_mask;
2714 	const struct rte_flow_item_geneve_opt *spec = item->spec;
2715 	const struct rte_flow_item_geneve_opt *mask = item->mask;
2716 	unsigned int i;
2717 	unsigned int data_len;
2718 	uint8_t tlv_option_len;
2719 	uint16_t optlen_m, optlen_v;
2720 	const struct rte_flow_item_geneve_opt full_mask = {
2721 		.option_class = RTE_BE16(0xffff),
2722 		.option_type = 0xff,
2723 		.option_len = 0x1f,
2724 	};
2725 
2726 	if (!mask)
2727 		mask = &rte_flow_item_geneve_opt_mask;
2728 	if (!spec)
2729 		return rte_flow_error_set
2730 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2731 			"Geneve TLV opt class/type/length must be specified");
2732 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
2733 		return rte_flow_error_set
2734 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2735 			"Geneve TLV opt length exceeeds the limit (31)");
2736 	/* Check if class type and length masks are full. */
2737 	if (full_mask.option_class != mask->option_class ||
2738 	    full_mask.option_type != mask->option_type ||
2739 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
2740 		return rte_flow_error_set
2741 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2742 			"Geneve TLV opt class/type/length masks must be full");
2743 	/* Check if length is supported */
2744 	if ((uint32_t)spec->option_len >
2745 			config->hca_attr.max_geneve_tlv_option_data_len)
2746 		return rte_flow_error_set
2747 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2748 			"Geneve TLV opt length not supported");
2749 	if (config->hca_attr.max_geneve_tlv_options > 1)
2750 		DRV_LOG(DEBUG,
2751 			"max_geneve_tlv_options supports more than 1 option");
2752 	/* Check GENEVE item preceding. */
2753 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
2754 		return rte_flow_error_set
2755 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2756 			"Geneve opt item must be preceded with Geneve item");
2757 	geneve_spec = geneve_item->spec;
2758 	geneve_mask = geneve_item->mask ? geneve_item->mask :
2759 					  &rte_flow_item_geneve_mask;
2760 	/* Check if GENEVE TLV option size doesn't exceed option length */
2761 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
2762 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
2763 		tlv_option_len = spec->option_len & mask->option_len;
2764 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
2765 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
2766 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
2767 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
2768 		if ((optlen_v & optlen_m) <= tlv_option_len)
2769 			return rte_flow_error_set
2770 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2771 				 "GENEVE TLV option length exceeds optlen");
2772 	}
2773 	/* Check if length is 0 or data is 0. */
2774 	if (spec->data == NULL || spec->option_len == 0)
2775 		return rte_flow_error_set
2776 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2777 			"Geneve TLV opt with zero data/length not supported");
2778 	/* Check not all data & mask are 0. */
2779 	data_len = spec->option_len * 4;
2780 	if (mask->data == NULL) {
2781 		for (i = 0; i < data_len; i++)
2782 			if (spec->data[i])
2783 				break;
2784 		if (i == data_len)
2785 			return rte_flow_error_set(error, ENOTSUP,
2786 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2787 				"Can't match on Geneve option data 0");
2788 	} else {
2789 		for (i = 0; i < data_len; i++)
2790 			if (spec->data[i] & mask->data[i])
2791 				break;
2792 		if (i == data_len)
2793 			return rte_flow_error_set(error, ENOTSUP,
2794 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2795 				"Can't match on Geneve option data and mask 0");
2796 		/* Check data mask supported. */
2797 		for (i = data_max_supported; i < data_len ; i++)
2798 			if (mask->data[i])
2799 				return rte_flow_error_set(error, ENOTSUP,
2800 					RTE_FLOW_ERROR_TYPE_ITEM, item,
2801 					"Data mask is of unsupported size");
2802 	}
2803 	/* Check GENEVE option is supported in NIC. */
2804 	if (!config->hca_attr.geneve_tlv_opt)
2805 		return rte_flow_error_set
2806 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 			"Geneve TLV opt not supported");
2808 	/* Check if we already have geneve option with different type/class. */
2809 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
2810 	geneve_opt_resource = sh->geneve_tlv_option_resource;
2811 	if (geneve_opt_resource != NULL)
2812 		if (geneve_opt_resource->option_class != spec->option_class ||
2813 		    geneve_opt_resource->option_type != spec->option_type ||
2814 		    geneve_opt_resource->length != spec->option_len) {
2815 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2816 			return rte_flow_error_set(error, ENOTSUP,
2817 				RTE_FLOW_ERROR_TYPE_ITEM, item,
2818 				"Only one Geneve TLV option supported");
2819 		}
2820 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
2821 	return 0;
2822 }
2823 
2824 /**
2825  * Validate MPLS item.
2826  *
2827  * @param[in] dev
2828  *   Pointer to the rte_eth_dev structure.
2829  * @param[in] item
2830  *   Item specification.
2831  * @param[in] item_flags
2832  *   Bit-fields that holds the items detected until now.
2833  * @param[in] prev_layer
2834  *   The protocol layer indicated in previous item.
2835  * @param[out] error
2836  *   Pointer to error structure.
2837  *
2838  * @return
2839  *   0 on success, a negative errno value otherwise and rte_errno is set.
2840  */
2841 int
2842 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2843 			     const struct rte_flow_item *item __rte_unused,
2844 			     uint64_t item_flags __rte_unused,
2845 			     uint64_t prev_layer __rte_unused,
2846 			     struct rte_flow_error *error)
2847 {
2848 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2849 	const struct rte_flow_item_mpls *mask = item->mask;
2850 	struct mlx5_priv *priv = dev->data->dev_private;
2851 	int ret;
2852 
2853 	if (!priv->config.mpls_en)
2854 		return rte_flow_error_set(error, ENOTSUP,
2855 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2856 					  "MPLS not supported or"
2857 					  " disabled in firmware"
2858 					  " configuration.");
2859 	/* MPLS over IP, UDP, GRE is allowed */
2860 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2861 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2862 			    MLX5_FLOW_LAYER_GRE |
2863 			    MLX5_FLOW_LAYER_GRE_KEY)))
2864 		return rte_flow_error_set(error, EINVAL,
2865 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2866 					  "protocol filtering not compatible"
2867 					  " with MPLS layer");
2868 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2869 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2870 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2871 		return rte_flow_error_set(error, ENOTSUP,
2872 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2873 					  "multiple tunnel layers not"
2874 					  " supported");
2875 	if (!mask)
2876 		mask = &rte_flow_item_mpls_mask;
2877 	ret = mlx5_flow_item_acceptable
2878 		(item, (const uint8_t *)mask,
2879 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2880 		 sizeof(struct rte_flow_item_mpls),
2881 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2882 	if (ret < 0)
2883 		return ret;
2884 	return 0;
2885 #else
2886 	return rte_flow_error_set(error, ENOTSUP,
2887 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2888 				  "MPLS is not supported by Verbs, please"
2889 				  " update.");
2890 #endif
2891 }
2892 
2893 /**
2894  * Validate NVGRE item.
2895  *
2896  * @param[in] item
2897  *   Item specification.
2898  * @param[in] item_flags
2899  *   Bit flags to mark detected items.
2900  * @param[in] target_protocol
2901  *   The next protocol in the previous item.
2902  * @param[out] error
2903  *   Pointer to error structure.
2904  *
2905  * @return
2906  *   0 on success, a negative errno value otherwise and rte_errno is set.
2907  */
2908 int
2909 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2910 			      uint64_t item_flags,
2911 			      uint8_t target_protocol,
2912 			      struct rte_flow_error *error)
2913 {
2914 	const struct rte_flow_item_nvgre *mask = item->mask;
2915 	int ret;
2916 
2917 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2918 		return rte_flow_error_set(error, EINVAL,
2919 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2920 					  "protocol filtering not compatible"
2921 					  " with this GRE layer");
2922 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2923 		return rte_flow_error_set(error, ENOTSUP,
2924 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2925 					  "multiple tunnel layers not"
2926 					  " supported");
2927 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2928 		return rte_flow_error_set(error, ENOTSUP,
2929 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2930 					  "L3 Layer is missing");
2931 	if (!mask)
2932 		mask = &rte_flow_item_nvgre_mask;
2933 	ret = mlx5_flow_item_acceptable
2934 		(item, (const uint8_t *)mask,
2935 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2936 		 sizeof(struct rte_flow_item_nvgre),
2937 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2938 	if (ret < 0)
2939 		return ret;
2940 	return 0;
2941 }
2942 
2943 /**
2944  * Validate eCPRI item.
2945  *
2946  * @param[in] item
2947  *   Item specification.
2948  * @param[in] item_flags
2949  *   Bit-fields that holds the items detected until now.
2950  * @param[in] last_item
2951  *   Previous validated item in the pattern items.
2952  * @param[in] ether_type
2953  *   Type in the ethernet layer header (including dot1q).
2954  * @param[in] acc_mask
2955  *   Acceptable mask, if NULL default internal default mask
2956  *   will be used to check whether item fields are supported.
2957  * @param[out] error
2958  *   Pointer to error structure.
2959  *
2960  * @return
2961  *   0 on success, a negative errno value otherwise and rte_errno is set.
2962  */
2963 int
2964 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2965 			      uint64_t item_flags,
2966 			      uint64_t last_item,
2967 			      uint16_t ether_type,
2968 			      const struct rte_flow_item_ecpri *acc_mask,
2969 			      struct rte_flow_error *error)
2970 {
2971 	const struct rte_flow_item_ecpri *mask = item->mask;
2972 	const struct rte_flow_item_ecpri nic_mask = {
2973 		.hdr = {
2974 			.common = {
2975 				.u32 =
2976 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2977 					.type = 0xFF,
2978 					}).u32),
2979 			},
2980 			.dummy[0] = 0xFFFFFFFF,
2981 		},
2982 	};
2983 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2984 					MLX5_FLOW_LAYER_OUTER_VLAN);
2985 	struct rte_flow_item_ecpri mask_lo;
2986 
2987 	if (!(last_item & outer_l2_vlan) &&
2988 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2989 		return rte_flow_error_set(error, EINVAL,
2990 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2991 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2992 	if ((last_item & outer_l2_vlan) && ether_type &&
2993 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2994 		return rte_flow_error_set(error, EINVAL,
2995 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2996 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2997 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2998 		return rte_flow_error_set(error, EINVAL,
2999 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3000 					  "eCPRI with tunnel is not supported right now");
3001 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3002 		return rte_flow_error_set(error, ENOTSUP,
3003 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3004 					  "multiple L3 layers not supported");
3005 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3006 		return rte_flow_error_set(error, EINVAL,
3007 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3008 					  "eCPRI cannot coexist with a TCP layer");
3009 	/* In specification, eCPRI could be over UDP layer. */
3010 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3011 		return rte_flow_error_set(error, EINVAL,
3012 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3013 					  "eCPRI over UDP layer is not yet supported right now");
3014 	/* Mask for type field in common header could be zero. */
3015 	if (!mask)
3016 		mask = &rte_flow_item_ecpri_mask;
3017 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3018 	/* Input mask is in big-endian format. */
3019 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3020 		return rte_flow_error_set(error, EINVAL,
3021 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3022 					  "partial mask is not supported for protocol");
3023 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3024 		return rte_flow_error_set(error, EINVAL,
3025 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3026 					  "message header mask must be after a type mask");
3027 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3028 					 acc_mask ? (const uint8_t *)acc_mask
3029 						  : (const uint8_t *)&nic_mask,
3030 					 sizeof(struct rte_flow_item_ecpri),
3031 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3032 }
3033 
3034 /**
3035  * Release resource related QUEUE/RSS action split.
3036  *
3037  * @param dev
3038  *   Pointer to Ethernet device.
3039  * @param flow
3040  *   Flow to release id's from.
3041  */
3042 static void
3043 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
3044 			     struct rte_flow *flow)
3045 {
3046 	struct mlx5_priv *priv = dev->data->dev_private;
3047 	uint32_t handle_idx;
3048 	struct mlx5_flow_handle *dev_handle;
3049 
3050 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
3051 		       handle_idx, dev_handle, next)
3052 		if (dev_handle->split_flow_id)
3053 			mlx5_ipool_free(priv->sh->ipool
3054 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
3055 					dev_handle->split_flow_id);
3056 }
3057 
3058 static int
3059 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3060 		   const struct rte_flow_attr *attr __rte_unused,
3061 		   const struct rte_flow_item items[] __rte_unused,
3062 		   const struct rte_flow_action actions[] __rte_unused,
3063 		   bool external __rte_unused,
3064 		   int hairpin __rte_unused,
3065 		   struct rte_flow_error *error)
3066 {
3067 	return rte_flow_error_set(error, ENOTSUP,
3068 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3069 }
3070 
3071 static struct mlx5_flow *
3072 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3073 		  const struct rte_flow_attr *attr __rte_unused,
3074 		  const struct rte_flow_item items[] __rte_unused,
3075 		  const struct rte_flow_action actions[] __rte_unused,
3076 		  struct rte_flow_error *error)
3077 {
3078 	rte_flow_error_set(error, ENOTSUP,
3079 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3080 	return NULL;
3081 }
3082 
3083 static int
3084 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3085 		    struct mlx5_flow *dev_flow __rte_unused,
3086 		    const struct rte_flow_attr *attr __rte_unused,
3087 		    const struct rte_flow_item items[] __rte_unused,
3088 		    const struct rte_flow_action actions[] __rte_unused,
3089 		    struct rte_flow_error *error)
3090 {
3091 	return rte_flow_error_set(error, ENOTSUP,
3092 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3093 }
3094 
3095 static int
3096 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3097 		struct rte_flow *flow __rte_unused,
3098 		struct rte_flow_error *error)
3099 {
3100 	return rte_flow_error_set(error, ENOTSUP,
3101 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3102 }
3103 
3104 static void
3105 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3106 		 struct rte_flow *flow __rte_unused)
3107 {
3108 }
3109 
3110 static void
3111 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3112 		  struct rte_flow *flow __rte_unused)
3113 {
3114 }
3115 
3116 static int
3117 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3118 		struct rte_flow *flow __rte_unused,
3119 		const struct rte_flow_action *actions __rte_unused,
3120 		void *data __rte_unused,
3121 		struct rte_flow_error *error)
3122 {
3123 	return rte_flow_error_set(error, ENOTSUP,
3124 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3125 }
3126 
3127 static int
3128 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3129 		      uint32_t domains __rte_unused,
3130 		      uint32_t flags __rte_unused)
3131 {
3132 	return 0;
3133 }
3134 
3135 /* Void driver to protect from null pointer reference. */
3136 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3137 	.validate = flow_null_validate,
3138 	.prepare = flow_null_prepare,
3139 	.translate = flow_null_translate,
3140 	.apply = flow_null_apply,
3141 	.remove = flow_null_remove,
3142 	.destroy = flow_null_destroy,
3143 	.query = flow_null_query,
3144 	.sync_domain = flow_null_sync_domain,
3145 };
3146 
3147 /**
3148  * Select flow driver type according to flow attributes and device
3149  * configuration.
3150  *
3151  * @param[in] dev
3152  *   Pointer to the dev structure.
3153  * @param[in] attr
3154  *   Pointer to the flow attributes.
3155  *
3156  * @return
3157  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3158  */
3159 static enum mlx5_flow_drv_type
3160 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3161 {
3162 	struct mlx5_priv *priv = dev->data->dev_private;
3163 	/* The OS can determine first a specific flow type (DV, VERBS) */
3164 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3165 
3166 	if (type != MLX5_FLOW_TYPE_MAX)
3167 		return type;
3168 	/* If no OS specific type - continue with DV/VERBS selection */
3169 	if (attr->transfer && priv->config.dv_esw_en)
3170 		type = MLX5_FLOW_TYPE_DV;
3171 	if (!attr->transfer)
3172 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3173 						 MLX5_FLOW_TYPE_VERBS;
3174 	return type;
3175 }
3176 
3177 #define flow_get_drv_ops(type) flow_drv_ops[type]
3178 
3179 /**
3180  * Flow driver validation API. This abstracts calling driver specific functions.
3181  * The type of flow driver is determined according to flow attributes.
3182  *
3183  * @param[in] dev
3184  *   Pointer to the dev structure.
3185  * @param[in] attr
3186  *   Pointer to the flow attributes.
3187  * @param[in] items
3188  *   Pointer to the list of items.
3189  * @param[in] actions
3190  *   Pointer to the list of actions.
3191  * @param[in] external
3192  *   This flow rule is created by request external to PMD.
3193  * @param[in] hairpin
3194  *   Number of hairpin TX actions, 0 means classic flow.
3195  * @param[out] error
3196  *   Pointer to the error structure.
3197  *
3198  * @return
3199  *   0 on success, a negative errno value otherwise and rte_errno is set.
3200  */
3201 static inline int
3202 flow_drv_validate(struct rte_eth_dev *dev,
3203 		  const struct rte_flow_attr *attr,
3204 		  const struct rte_flow_item items[],
3205 		  const struct rte_flow_action actions[],
3206 		  bool external, int hairpin, struct rte_flow_error *error)
3207 {
3208 	const struct mlx5_flow_driver_ops *fops;
3209 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3210 
3211 	fops = flow_get_drv_ops(type);
3212 	return fops->validate(dev, attr, items, actions, external,
3213 			      hairpin, error);
3214 }
3215 
3216 /**
3217  * Flow driver preparation API. This abstracts calling driver specific
3218  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3219  * calculates the size of memory required for device flow, allocates the memory,
3220  * initializes the device flow and returns the pointer.
3221  *
3222  * @note
3223  *   This function initializes device flow structure such as dv or verbs in
3224  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3225  *   rest. For example, adding returning device flow to flow->dev_flow list and
3226  *   setting backward reference to the flow should be done out of this function.
3227  *   layers field is not filled either.
3228  *
3229  * @param[in] dev
3230  *   Pointer to the dev structure.
3231  * @param[in] attr
3232  *   Pointer to the flow attributes.
3233  * @param[in] items
3234  *   Pointer to the list of items.
3235  * @param[in] actions
3236  *   Pointer to the list of actions.
3237  * @param[in] flow_idx
3238  *   This memory pool index to the flow.
3239  * @param[out] error
3240  *   Pointer to the error structure.
3241  *
3242  * @return
3243  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3244  */
3245 static inline struct mlx5_flow *
3246 flow_drv_prepare(struct rte_eth_dev *dev,
3247 		 const struct rte_flow *flow,
3248 		 const struct rte_flow_attr *attr,
3249 		 const struct rte_flow_item items[],
3250 		 const struct rte_flow_action actions[],
3251 		 uint32_t flow_idx,
3252 		 struct rte_flow_error *error)
3253 {
3254 	const struct mlx5_flow_driver_ops *fops;
3255 	enum mlx5_flow_drv_type type = flow->drv_type;
3256 	struct mlx5_flow *mlx5_flow = NULL;
3257 
3258 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3259 	fops = flow_get_drv_ops(type);
3260 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3261 	if (mlx5_flow)
3262 		mlx5_flow->flow_idx = flow_idx;
3263 	return mlx5_flow;
3264 }
3265 
3266 /**
3267  * Flow driver translation API. This abstracts calling driver specific
3268  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3269  * translates a generic flow into a driver flow. flow_drv_prepare() must
3270  * precede.
3271  *
3272  * @note
3273  *   dev_flow->layers could be filled as a result of parsing during translation
3274  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3275  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3276  *   flow->actions could be overwritten even though all the expanded dev_flows
3277  *   have the same actions.
3278  *
3279  * @param[in] dev
3280  *   Pointer to the rte dev structure.
3281  * @param[in, out] dev_flow
3282  *   Pointer to the mlx5 flow.
3283  * @param[in] attr
3284  *   Pointer to the flow attributes.
3285  * @param[in] items
3286  *   Pointer to the list of items.
3287  * @param[in] actions
3288  *   Pointer to the list of actions.
3289  * @param[out] error
3290  *   Pointer to the error structure.
3291  *
3292  * @return
3293  *   0 on success, a negative errno value otherwise and rte_errno is set.
3294  */
3295 static inline int
3296 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3297 		   const struct rte_flow_attr *attr,
3298 		   const struct rte_flow_item items[],
3299 		   const struct rte_flow_action actions[],
3300 		   struct rte_flow_error *error)
3301 {
3302 	const struct mlx5_flow_driver_ops *fops;
3303 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3304 
3305 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3306 	fops = flow_get_drv_ops(type);
3307 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3308 }
3309 
3310 /**
3311  * Flow driver apply API. This abstracts calling driver specific functions.
3312  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3313  * translated driver flows on to device. flow_drv_translate() must precede.
3314  *
3315  * @param[in] dev
3316  *   Pointer to Ethernet device structure.
3317  * @param[in, out] flow
3318  *   Pointer to flow structure.
3319  * @param[out] error
3320  *   Pointer to error structure.
3321  *
3322  * @return
3323  *   0 on success, a negative errno value otherwise and rte_errno is set.
3324  */
3325 static inline int
3326 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3327 	       struct rte_flow_error *error)
3328 {
3329 	const struct mlx5_flow_driver_ops *fops;
3330 	enum mlx5_flow_drv_type type = flow->drv_type;
3331 
3332 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3333 	fops = flow_get_drv_ops(type);
3334 	return fops->apply(dev, flow, error);
3335 }
3336 
3337 /**
3338  * Flow driver destroy API. This abstracts calling driver specific functions.
3339  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3340  * on device and releases resources of the flow.
3341  *
3342  * @param[in] dev
3343  *   Pointer to Ethernet device.
3344  * @param[in, out] flow
3345  *   Pointer to flow structure.
3346  */
3347 static inline void
3348 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3349 {
3350 	const struct mlx5_flow_driver_ops *fops;
3351 	enum mlx5_flow_drv_type type = flow->drv_type;
3352 
3353 	flow_mreg_split_qrss_release(dev, flow);
3354 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3355 	fops = flow_get_drv_ops(type);
3356 	fops->destroy(dev, flow);
3357 }
3358 
3359 /**
3360  * Get RSS action from the action list.
3361  *
3362  * @param[in] actions
3363  *   Pointer to the list of actions.
3364  *
3365  * @return
3366  *   Pointer to the RSS action if exist, else return NULL.
3367  */
3368 static const struct rte_flow_action_rss*
3369 flow_get_rss_action(const struct rte_flow_action actions[])
3370 {
3371 	const struct rte_flow_action_rss *rss = NULL;
3372 
3373 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3374 		switch (actions->type) {
3375 		case RTE_FLOW_ACTION_TYPE_RSS:
3376 			rss = actions->conf;
3377 			break;
3378 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3379 		{
3380 			const struct rte_flow_action_sample *sample =
3381 								actions->conf;
3382 			const struct rte_flow_action *act = sample->actions;
3383 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3384 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3385 					rss = act->conf;
3386 			break;
3387 		}
3388 		default:
3389 			break;
3390 		}
3391 	}
3392 	return rss;
3393 }
3394 
3395 /**
3396  * Get ASO age action by index.
3397  *
3398  * @param[in] dev
3399  *   Pointer to the Ethernet device structure.
3400  * @param[in] age_idx
3401  *   Index to the ASO age action.
3402  *
3403  * @return
3404  *   The specified ASO age action.
3405  */
3406 struct mlx5_aso_age_action*
3407 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3408 {
3409 	uint16_t pool_idx = age_idx & UINT16_MAX;
3410 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3411 	struct mlx5_priv *priv = dev->data->dev_private;
3412 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3413 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3414 
3415 	return &pool->actions[offset - 1];
3416 }
3417 
3418 /* maps shared action to translated non shared in some actions array */
3419 struct mlx5_translated_shared_action {
3420 	struct rte_flow_shared_action *action; /**< Shared action */
3421 	int index; /**< Index in related array of rte_flow_action */
3422 };
3423 
3424 /**
3425  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3426  * non shared action if translation possible.
3427  * This functionality used to run same execution path for both shared & non
3428  * shared actions on flow create. All necessary preparations for shared
3429  * action handling should be preformed on *shared* actions list returned
3430  * from this call.
3431  *
3432  * @param[in] dev
3433  *   Pointer to Ethernet device.
3434  * @param[in] actions
3435  *   List of actions to translate.
3436  * @param[out] shared
3437  *   List to store translated shared actions.
3438  * @param[in, out] shared_n
3439  *   Size of *shared* array. On return should be updated with number of shared
3440  *   actions retrieved from the *actions* list.
3441  * @param[out] translated_actions
3442  *   List of actions where all shared actions were translated to non shared
3443  *   if possible. NULL if no translation took place.
3444  * @param[out] error
3445  *   Pointer to the error structure.
3446  *
3447  * @return
3448  *   0 on success, a negative errno value otherwise and rte_errno is set.
3449  */
3450 static int
3451 flow_shared_actions_translate(struct rte_eth_dev *dev,
3452 			      const struct rte_flow_action actions[],
3453 			      struct mlx5_translated_shared_action *shared,
3454 			      int *shared_n,
3455 			      struct rte_flow_action **translated_actions,
3456 			      struct rte_flow_error *error)
3457 {
3458 	struct mlx5_priv *priv = dev->data->dev_private;
3459 	struct rte_flow_action *translated = NULL;
3460 	size_t actions_size;
3461 	int n;
3462 	int copied_n = 0;
3463 	struct mlx5_translated_shared_action *shared_end = NULL;
3464 
3465 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3466 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3467 			continue;
3468 		if (copied_n == *shared_n) {
3469 			return rte_flow_error_set
3470 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3471 				 NULL, "too many shared actions");
3472 		}
3473 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3474 			   sizeof(actions[n].conf));
3475 		shared[copied_n].index = n;
3476 		copied_n++;
3477 	}
3478 	n++;
3479 	*shared_n = copied_n;
3480 	if (!copied_n)
3481 		return 0;
3482 	actions_size = sizeof(struct rte_flow_action) * n;
3483 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3484 	if (!translated) {
3485 		rte_errno = ENOMEM;
3486 		return -ENOMEM;
3487 	}
3488 	memcpy(translated, actions, actions_size);
3489 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3490 		struct mlx5_shared_action_rss *shared_rss;
3491 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3492 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3493 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3494 									   - 1);
3495 
3496 		switch (type) {
3497 		case MLX5_SHARED_ACTION_TYPE_RSS:
3498 			shared_rss = mlx5_ipool_get
3499 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3500 			translated[shared->index].type =
3501 				RTE_FLOW_ACTION_TYPE_RSS;
3502 			translated[shared->index].conf =
3503 				&shared_rss->origin;
3504 			break;
3505 		case MLX5_SHARED_ACTION_TYPE_AGE:
3506 			if (priv->sh->flow_hit_aso_en) {
3507 				translated[shared->index].type =
3508 					(enum rte_flow_action_type)
3509 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3510 				translated[shared->index].conf =
3511 							 (void *)(uintptr_t)idx;
3512 				break;
3513 			}
3514 			/* Fall-through */
3515 		default:
3516 			mlx5_free(translated);
3517 			return rte_flow_error_set
3518 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3519 				 NULL, "invalid shared action type");
3520 		}
3521 	}
3522 	*translated_actions = translated;
3523 	return 0;
3524 }
3525 
3526 /**
3527  * Get Shared RSS action from the action list.
3528  *
3529  * @param[in] dev
3530  *   Pointer to Ethernet device.
3531  * @param[in] shared
3532  *   Pointer to the list of actions.
3533  * @param[in] shared_n
3534  *   Actions list length.
3535  *
3536  * @return
3537  *   The MLX5 RSS action ID if exists, otherwise return 0.
3538  */
3539 static uint32_t
3540 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3541 			   struct mlx5_translated_shared_action *shared,
3542 			   int shared_n)
3543 {
3544 	struct mlx5_translated_shared_action *shared_end;
3545 	struct mlx5_priv *priv = dev->data->dev_private;
3546 	struct mlx5_shared_action_rss *shared_rss;
3547 
3548 
3549 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3550 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3551 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3552 		uint32_t idx = act_idx &
3553 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3554 		switch (type) {
3555 		case MLX5_SHARED_ACTION_TYPE_RSS:
3556 			shared_rss = mlx5_ipool_get
3557 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3558 									   idx);
3559 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3560 					   __ATOMIC_RELAXED);
3561 			return idx;
3562 		default:
3563 			break;
3564 		}
3565 	}
3566 	return 0;
3567 }
3568 
3569 static unsigned int
3570 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3571 {
3572 	const struct rte_flow_item *item;
3573 	unsigned int has_vlan = 0;
3574 
3575 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3576 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3577 			has_vlan = 1;
3578 			break;
3579 		}
3580 	}
3581 	if (has_vlan)
3582 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3583 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3584 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3585 			       MLX5_EXPANSION_ROOT_OUTER;
3586 }
3587 
3588 /**
3589  *  Get layer flags from the prefix flow.
3590  *
3591  *  Some flows may be split to several subflows, the prefix subflow gets the
3592  *  match items and the suffix sub flow gets the actions.
3593  *  Some actions need the user defined match item flags to get the detail for
3594  *  the action.
3595  *  This function helps the suffix flow to get the item layer flags from prefix
3596  *  subflow.
3597  *
3598  * @param[in] dev_flow
3599  *   Pointer the created preifx subflow.
3600  *
3601  * @return
3602  *   The layers get from prefix subflow.
3603  */
3604 static inline uint64_t
3605 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3606 {
3607 	uint64_t layers = 0;
3608 
3609 	/*
3610 	 * Layers bits could be localization, but usually the compiler will
3611 	 * help to do the optimization work for source code.
3612 	 * If no decap actions, use the layers directly.
3613 	 */
3614 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3615 		return dev_flow->handle->layers;
3616 	/* Convert L3 layers with decap action. */
3617 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3618 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3619 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3620 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3621 	/* Convert L4 layers with decap action.  */
3622 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3623 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3624 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3625 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3626 	return layers;
3627 }
3628 
3629 /**
3630  * Get metadata split action information.
3631  *
3632  * @param[in] actions
3633  *   Pointer to the list of actions.
3634  * @param[out] qrss
3635  *   Pointer to the return pointer.
3636  * @param[out] qrss_type
3637  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3638  *   if no QUEUE/RSS is found.
3639  * @param[out] encap_idx
3640  *   Pointer to the index of the encap action if exists, otherwise the last
3641  *   action index.
3642  *
3643  * @return
3644  *   Total number of actions.
3645  */
3646 static int
3647 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3648 				       const struct rte_flow_action **qrss,
3649 				       int *encap_idx)
3650 {
3651 	const struct rte_flow_action_raw_encap *raw_encap;
3652 	int actions_n = 0;
3653 	int raw_decap_idx = -1;
3654 
3655 	*encap_idx = -1;
3656 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3657 		switch (actions->type) {
3658 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3659 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3660 			*encap_idx = actions_n;
3661 			break;
3662 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3663 			raw_decap_idx = actions_n;
3664 			break;
3665 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3666 			raw_encap = actions->conf;
3667 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3668 				*encap_idx = raw_decap_idx != -1 ?
3669 						      raw_decap_idx : actions_n;
3670 			break;
3671 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3672 		case RTE_FLOW_ACTION_TYPE_RSS:
3673 			*qrss = actions;
3674 			break;
3675 		default:
3676 			break;
3677 		}
3678 		actions_n++;
3679 	}
3680 	if (*encap_idx == -1)
3681 		*encap_idx = actions_n;
3682 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3683 	return actions_n + 1;
3684 }
3685 
3686 /**
3687  * Check meter action from the action list.
3688  *
3689  * @param[in] actions
3690  *   Pointer to the list of actions.
3691  * @param[out] mtr
3692  *   Pointer to the meter exist flag.
3693  *
3694  * @return
3695  *   Total number of actions.
3696  */
3697 static int
3698 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3699 {
3700 	int actions_n = 0;
3701 
3702 	MLX5_ASSERT(mtr);
3703 	*mtr = 0;
3704 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3705 		switch (actions->type) {
3706 		case RTE_FLOW_ACTION_TYPE_METER:
3707 			*mtr = 1;
3708 			break;
3709 		default:
3710 			break;
3711 		}
3712 		actions_n++;
3713 	}
3714 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3715 	return actions_n + 1;
3716 }
3717 
3718 /**
3719  * Check if the flow should be split due to hairpin.
3720  * The reason for the split is that in current HW we can't
3721  * support encap and push-vlan on Rx, so if a flow contains
3722  * these actions we move it to Tx.
3723  *
3724  * @param dev
3725  *   Pointer to Ethernet device.
3726  * @param[in] attr
3727  *   Flow rule attributes.
3728  * @param[in] actions
3729  *   Associated actions (list terminated by the END action).
3730  *
3731  * @return
3732  *   > 0 the number of actions and the flow should be split,
3733  *   0 when no split required.
3734  */
3735 static int
3736 flow_check_hairpin_split(struct rte_eth_dev *dev,
3737 			 const struct rte_flow_attr *attr,
3738 			 const struct rte_flow_action actions[])
3739 {
3740 	int queue_action = 0;
3741 	int action_n = 0;
3742 	int split = 0;
3743 	const struct rte_flow_action_queue *queue;
3744 	const struct rte_flow_action_rss *rss;
3745 	const struct rte_flow_action_raw_encap *raw_encap;
3746 	const struct rte_eth_hairpin_conf *conf;
3747 
3748 	if (!attr->ingress)
3749 		return 0;
3750 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3751 		switch (actions->type) {
3752 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3753 			queue = actions->conf;
3754 			if (queue == NULL)
3755 				return 0;
3756 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3757 			if (conf == NULL || conf->tx_explicit != 0)
3758 				return 0;
3759 			queue_action = 1;
3760 			action_n++;
3761 			break;
3762 		case RTE_FLOW_ACTION_TYPE_RSS:
3763 			rss = actions->conf;
3764 			if (rss == NULL || rss->queue_num == 0)
3765 				return 0;
3766 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3767 			if (conf == NULL || conf->tx_explicit != 0)
3768 				return 0;
3769 			queue_action = 1;
3770 			action_n++;
3771 			break;
3772 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3773 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3774 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3775 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3776 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3777 			split++;
3778 			action_n++;
3779 			break;
3780 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3781 			raw_encap = actions->conf;
3782 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3783 				split++;
3784 			action_n++;
3785 			break;
3786 		default:
3787 			action_n++;
3788 			break;
3789 		}
3790 	}
3791 	if (split && queue_action)
3792 		return action_n;
3793 	return 0;
3794 }
3795 
3796 /* Declare flow create/destroy prototype in advance. */
3797 static uint32_t
3798 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3799 		 const struct rte_flow_attr *attr,
3800 		 const struct rte_flow_item items[],
3801 		 const struct rte_flow_action actions[],
3802 		 bool external, struct rte_flow_error *error);
3803 
3804 static void
3805 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3806 		  uint32_t flow_idx);
3807 
3808 int
3809 flow_dv_mreg_match_cb(struct mlx5_hlist *list __rte_unused,
3810 		      struct mlx5_hlist_entry *entry,
3811 		      uint64_t key, void *cb_ctx __rte_unused)
3812 {
3813 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3814 		container_of(entry, typeof(*mcp_res), hlist_ent);
3815 
3816 	return mcp_res->mark_id != key;
3817 }
3818 
3819 struct mlx5_hlist_entry *
3820 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3821 		       void *cb_ctx)
3822 {
3823 	struct rte_eth_dev *dev = list->ctx;
3824 	struct mlx5_priv *priv = dev->data->dev_private;
3825 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3826 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3827 	struct rte_flow_error *error = ctx->error;
3828 	uint32_t idx = 0;
3829 	int ret;
3830 	uint32_t mark_id = key;
3831 	struct rte_flow_attr attr = {
3832 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3833 		.ingress = 1,
3834 	};
3835 	struct mlx5_rte_flow_item_tag tag_spec = {
3836 		.data = mark_id,
3837 	};
3838 	struct rte_flow_item items[] = {
3839 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3840 	};
3841 	struct rte_flow_action_mark ftag = {
3842 		.id = mark_id,
3843 	};
3844 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3845 		.dst = REG_B,
3846 		.src = REG_NON,
3847 	};
3848 	struct rte_flow_action_jump jump = {
3849 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3850 	};
3851 	struct rte_flow_action actions[] = {
3852 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3853 	};
3854 
3855 	/* Fill the register fileds in the flow. */
3856 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3857 	if (ret < 0)
3858 		return NULL;
3859 	tag_spec.id = ret;
3860 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3861 	if (ret < 0)
3862 		return NULL;
3863 	cp_mreg.src = ret;
3864 	/* Provide the full width of FLAG specific value. */
3865 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3866 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3867 	/* Build a new flow. */
3868 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3869 		items[0] = (struct rte_flow_item){
3870 			.type = (enum rte_flow_item_type)
3871 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3872 			.spec = &tag_spec,
3873 		};
3874 		items[1] = (struct rte_flow_item){
3875 			.type = RTE_FLOW_ITEM_TYPE_END,
3876 		};
3877 		actions[0] = (struct rte_flow_action){
3878 			.type = (enum rte_flow_action_type)
3879 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3880 			.conf = &ftag,
3881 		};
3882 		actions[1] = (struct rte_flow_action){
3883 			.type = (enum rte_flow_action_type)
3884 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3885 			.conf = &cp_mreg,
3886 		};
3887 		actions[2] = (struct rte_flow_action){
3888 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3889 			.conf = &jump,
3890 		};
3891 		actions[3] = (struct rte_flow_action){
3892 			.type = RTE_FLOW_ACTION_TYPE_END,
3893 		};
3894 	} else {
3895 		/* Default rule, wildcard match. */
3896 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
3897 		items[0] = (struct rte_flow_item){
3898 			.type = RTE_FLOW_ITEM_TYPE_END,
3899 		};
3900 		actions[0] = (struct rte_flow_action){
3901 			.type = (enum rte_flow_action_type)
3902 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3903 			.conf = &cp_mreg,
3904 		};
3905 		actions[1] = (struct rte_flow_action){
3906 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3907 			.conf = &jump,
3908 		};
3909 		actions[2] = (struct rte_flow_action){
3910 			.type = RTE_FLOW_ACTION_TYPE_END,
3911 		};
3912 	}
3913 	/* Build a new entry. */
3914 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3915 	if (!mcp_res) {
3916 		rte_errno = ENOMEM;
3917 		return NULL;
3918 	}
3919 	mcp_res->idx = idx;
3920 	mcp_res->mark_id = mark_id;
3921 	/*
3922 	 * The copy Flows are not included in any list. There
3923 	 * ones are referenced from other Flows and can not
3924 	 * be applied, removed, deleted in ardbitrary order
3925 	 * by list traversing.
3926 	 */
3927 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3928 					 actions, false, error);
3929 	if (!mcp_res->rix_flow) {
3930 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3931 		return NULL;
3932 	}
3933 	return &mcp_res->hlist_ent;
3934 }
3935 
3936 /**
3937  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3938  *
3939  * As mark_id is unique, if there's already a registered flow for the mark_id,
3940  * return by increasing the reference counter of the resource. Otherwise, create
3941  * the resource (mcp_res) and flow.
3942  *
3943  * Flow looks like,
3944  *   - If ingress port is ANY and reg_c[1] is mark_id,
3945  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3946  *
3947  * For default flow (zero mark_id), flow is like,
3948  *   - If ingress port is ANY,
3949  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3950  *
3951  * @param dev
3952  *   Pointer to Ethernet device.
3953  * @param mark_id
3954  *   ID of MARK action, zero means default flow for META.
3955  * @param[out] error
3956  *   Perform verbose error reporting if not NULL.
3957  *
3958  * @return
3959  *   Associated resource on success, NULL otherwise and rte_errno is set.
3960  */
3961 static struct mlx5_flow_mreg_copy_resource *
3962 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3963 			  struct rte_flow_error *error)
3964 {
3965 	struct mlx5_priv *priv = dev->data->dev_private;
3966 	struct mlx5_hlist_entry *entry;
3967 	struct mlx5_flow_cb_ctx ctx = {
3968 		.dev = dev,
3969 		.error = error,
3970 	};
3971 
3972 	/* Check if already registered. */
3973 	MLX5_ASSERT(priv->mreg_cp_tbl);
3974 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3975 	if (!entry)
3976 		return NULL;
3977 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3978 			    hlist_ent);
3979 }
3980 
3981 void
3982 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3983 {
3984 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3985 		container_of(entry, typeof(*mcp_res), hlist_ent);
3986 	struct rte_eth_dev *dev = list->ctx;
3987 	struct mlx5_priv *priv = dev->data->dev_private;
3988 
3989 	MLX5_ASSERT(mcp_res->rix_flow);
3990 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3991 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3992 }
3993 
3994 /**
3995  * Release flow in RX_CP_TBL.
3996  *
3997  * @param dev
3998  *   Pointer to Ethernet device.
3999  * @flow
4000  *   Parent flow for wich copying is provided.
4001  */
4002 static void
4003 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4004 			  struct rte_flow *flow)
4005 {
4006 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4007 	struct mlx5_priv *priv = dev->data->dev_private;
4008 
4009 	if (!flow->rix_mreg_copy)
4010 		return;
4011 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4012 				 flow->rix_mreg_copy);
4013 	if (!mcp_res || !priv->mreg_cp_tbl)
4014 		return;
4015 	MLX5_ASSERT(mcp_res->rix_flow);
4016 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4017 	flow->rix_mreg_copy = 0;
4018 }
4019 
4020 /**
4021  * Remove the default copy action from RX_CP_TBL.
4022  *
4023  * This functions is called in the mlx5_dev_start(). No thread safe
4024  * is guaranteed.
4025  *
4026  * @param dev
4027  *   Pointer to Ethernet device.
4028  */
4029 static void
4030 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4031 {
4032 	struct mlx5_hlist_entry *entry;
4033 	struct mlx5_priv *priv = dev->data->dev_private;
4034 
4035 	/* Check if default flow is registered. */
4036 	if (!priv->mreg_cp_tbl)
4037 		return;
4038 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
4039 				  MLX5_DEFAULT_COPY_ID, NULL);
4040 	if (!entry)
4041 		return;
4042 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4043 }
4044 
4045 /**
4046  * Add the default copy action in in RX_CP_TBL.
4047  *
4048  * This functions is called in the mlx5_dev_start(). No thread safe
4049  * is guaranteed.
4050  *
4051  * @param dev
4052  *   Pointer to Ethernet device.
4053  * @param[out] error
4054  *   Perform verbose error reporting if not NULL.
4055  *
4056  * @return
4057  *   0 for success, negative value otherwise and rte_errno is set.
4058  */
4059 static int
4060 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4061 				  struct rte_flow_error *error)
4062 {
4063 	struct mlx5_priv *priv = dev->data->dev_private;
4064 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4065 
4066 	/* Check whether extensive metadata feature is engaged. */
4067 	if (!priv->config.dv_flow_en ||
4068 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4069 	    !mlx5_flow_ext_mreg_supported(dev) ||
4070 	    !priv->sh->dv_regc0_mask)
4071 		return 0;
4072 	/*
4073 	 * Add default mreg copy flow may be called multiple time, but
4074 	 * only be called once in stop. Avoid register it twice.
4075 	 */
4076 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
4077 		return 0;
4078 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4079 	if (!mcp_res)
4080 		return -rte_errno;
4081 	return 0;
4082 }
4083 
4084 /**
4085  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4086  *
4087  * All the flow having Q/RSS action should be split by
4088  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4089  * performs the following,
4090  *   - CQE->flow_tag := reg_c[1] (MARK)
4091  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4092  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4093  * but there should be a flow per each MARK ID set by MARK action.
4094  *
4095  * For the aforementioned reason, if there's a MARK action in flow's action
4096  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4097  * the MARK ID to CQE's flow_tag like,
4098  *   - If reg_c[1] is mark_id,
4099  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4100  *
4101  * For SET_META action which stores value in reg_c[0], as the destination is
4102  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4103  * MARK ID means the default flow. The default flow looks like,
4104  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4105  *
4106  * @param dev
4107  *   Pointer to Ethernet device.
4108  * @param flow
4109  *   Pointer to flow structure.
4110  * @param[in] actions
4111  *   Pointer to the list of actions.
4112  * @param[out] error
4113  *   Perform verbose error reporting if not NULL.
4114  *
4115  * @return
4116  *   0 on success, negative value otherwise and rte_errno is set.
4117  */
4118 static int
4119 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4120 			    struct rte_flow *flow,
4121 			    const struct rte_flow_action *actions,
4122 			    struct rte_flow_error *error)
4123 {
4124 	struct mlx5_priv *priv = dev->data->dev_private;
4125 	struct mlx5_dev_config *config = &priv->config;
4126 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4127 	const struct rte_flow_action_mark *mark;
4128 
4129 	/* Check whether extensive metadata feature is engaged. */
4130 	if (!config->dv_flow_en ||
4131 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4132 	    !mlx5_flow_ext_mreg_supported(dev) ||
4133 	    !priv->sh->dv_regc0_mask)
4134 		return 0;
4135 	/* Find MARK action. */
4136 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4137 		switch (actions->type) {
4138 		case RTE_FLOW_ACTION_TYPE_FLAG:
4139 			mcp_res = flow_mreg_add_copy_action
4140 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4141 			if (!mcp_res)
4142 				return -rte_errno;
4143 			flow->rix_mreg_copy = mcp_res->idx;
4144 			return 0;
4145 		case RTE_FLOW_ACTION_TYPE_MARK:
4146 			mark = (const struct rte_flow_action_mark *)
4147 				actions->conf;
4148 			mcp_res =
4149 				flow_mreg_add_copy_action(dev, mark->id, error);
4150 			if (!mcp_res)
4151 				return -rte_errno;
4152 			flow->rix_mreg_copy = mcp_res->idx;
4153 			return 0;
4154 		default:
4155 			break;
4156 		}
4157 	}
4158 	return 0;
4159 }
4160 
4161 #define MLX5_MAX_SPLIT_ACTIONS 24
4162 #define MLX5_MAX_SPLIT_ITEMS 24
4163 
4164 /**
4165  * Split the hairpin flow.
4166  * Since HW can't support encap and push-vlan on Rx, we move these
4167  * actions to Tx.
4168  * If the count action is after the encap then we also
4169  * move the count action. in this case the count will also measure
4170  * the outer bytes.
4171  *
4172  * @param dev
4173  *   Pointer to Ethernet device.
4174  * @param[in] actions
4175  *   Associated actions (list terminated by the END action).
4176  * @param[out] actions_rx
4177  *   Rx flow actions.
4178  * @param[out] actions_tx
4179  *   Tx flow actions..
4180  * @param[out] pattern_tx
4181  *   The pattern items for the Tx flow.
4182  * @param[out] flow_id
4183  *   The flow ID connected to this flow.
4184  *
4185  * @return
4186  *   0 on success.
4187  */
4188 static int
4189 flow_hairpin_split(struct rte_eth_dev *dev,
4190 		   const struct rte_flow_action actions[],
4191 		   struct rte_flow_action actions_rx[],
4192 		   struct rte_flow_action actions_tx[],
4193 		   struct rte_flow_item pattern_tx[],
4194 		   uint32_t flow_id)
4195 {
4196 	const struct rte_flow_action_raw_encap *raw_encap;
4197 	const struct rte_flow_action_raw_decap *raw_decap;
4198 	struct mlx5_rte_flow_action_set_tag *set_tag;
4199 	struct rte_flow_action *tag_action;
4200 	struct mlx5_rte_flow_item_tag *tag_item;
4201 	struct rte_flow_item *item;
4202 	char *addr;
4203 	int encap = 0;
4204 
4205 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4206 		switch (actions->type) {
4207 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4208 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4209 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4210 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4211 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4212 			rte_memcpy(actions_tx, actions,
4213 			       sizeof(struct rte_flow_action));
4214 			actions_tx++;
4215 			break;
4216 		case RTE_FLOW_ACTION_TYPE_COUNT:
4217 			if (encap) {
4218 				rte_memcpy(actions_tx, actions,
4219 					   sizeof(struct rte_flow_action));
4220 				actions_tx++;
4221 			} else {
4222 				rte_memcpy(actions_rx, actions,
4223 					   sizeof(struct rte_flow_action));
4224 				actions_rx++;
4225 			}
4226 			break;
4227 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4228 			raw_encap = actions->conf;
4229 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4230 				memcpy(actions_tx, actions,
4231 				       sizeof(struct rte_flow_action));
4232 				actions_tx++;
4233 				encap = 1;
4234 			} else {
4235 				rte_memcpy(actions_rx, actions,
4236 					   sizeof(struct rte_flow_action));
4237 				actions_rx++;
4238 			}
4239 			break;
4240 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4241 			raw_decap = actions->conf;
4242 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4243 				memcpy(actions_tx, actions,
4244 				       sizeof(struct rte_flow_action));
4245 				actions_tx++;
4246 			} else {
4247 				rte_memcpy(actions_rx, actions,
4248 					   sizeof(struct rte_flow_action));
4249 				actions_rx++;
4250 			}
4251 			break;
4252 		default:
4253 			rte_memcpy(actions_rx, actions,
4254 				   sizeof(struct rte_flow_action));
4255 			actions_rx++;
4256 			break;
4257 		}
4258 	}
4259 	/* Add set meta action and end action for the Rx flow. */
4260 	tag_action = actions_rx;
4261 	tag_action->type = (enum rte_flow_action_type)
4262 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4263 	actions_rx++;
4264 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4265 	actions_rx++;
4266 	set_tag = (void *)actions_rx;
4267 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4268 	MLX5_ASSERT(set_tag->id > REG_NON);
4269 	set_tag->data = flow_id;
4270 	tag_action->conf = set_tag;
4271 	/* Create Tx item list. */
4272 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4273 	addr = (void *)&pattern_tx[2];
4274 	item = pattern_tx;
4275 	item->type = (enum rte_flow_item_type)
4276 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4277 	tag_item = (void *)addr;
4278 	tag_item->data = flow_id;
4279 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4280 	MLX5_ASSERT(set_tag->id > REG_NON);
4281 	item->spec = tag_item;
4282 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4283 	tag_item = (void *)addr;
4284 	tag_item->data = UINT32_MAX;
4285 	tag_item->id = UINT16_MAX;
4286 	item->mask = tag_item;
4287 	item->last = NULL;
4288 	item++;
4289 	item->type = RTE_FLOW_ITEM_TYPE_END;
4290 	return 0;
4291 }
4292 
4293 /**
4294  * The last stage of splitting chain, just creates the subflow
4295  * without any modification.
4296  *
4297  * @param[in] dev
4298  *   Pointer to Ethernet device.
4299  * @param[in] flow
4300  *   Parent flow structure pointer.
4301  * @param[in, out] sub_flow
4302  *   Pointer to return the created subflow, may be NULL.
4303  * @param[in] attr
4304  *   Flow rule attributes.
4305  * @param[in] items
4306  *   Pattern specification (list terminated by the END pattern item).
4307  * @param[in] actions
4308  *   Associated actions (list terminated by the END action).
4309  * @param[in] flow_split_info
4310  *   Pointer to flow split info structure.
4311  * @param[out] error
4312  *   Perform verbose error reporting if not NULL.
4313  * @return
4314  *   0 on success, negative value otherwise
4315  */
4316 static int
4317 flow_create_split_inner(struct rte_eth_dev *dev,
4318 			struct rte_flow *flow,
4319 			struct mlx5_flow **sub_flow,
4320 			const struct rte_flow_attr *attr,
4321 			const struct rte_flow_item items[],
4322 			const struct rte_flow_action actions[],
4323 			struct mlx5_flow_split_info *flow_split_info,
4324 			struct rte_flow_error *error)
4325 {
4326 	struct mlx5_flow *dev_flow;
4327 
4328 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4329 				    flow_split_info->flow_idx, error);
4330 	if (!dev_flow)
4331 		return -rte_errno;
4332 	dev_flow->flow = flow;
4333 	dev_flow->external = flow_split_info->external;
4334 	dev_flow->skip_scale = flow_split_info->skip_scale;
4335 	/* Subflow object was created, we must include one in the list. */
4336 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4337 		      dev_flow->handle, next);
4338 	/*
4339 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4340 	 * flow may need some user defined item layer flags, and pass the
4341 	 * Metadate rxq mark flag to suffix flow as well.
4342 	 */
4343 	if (flow_split_info->prefix_layers)
4344 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4345 	if (flow_split_info->prefix_mark)
4346 		dev_flow->handle->mark = 1;
4347 	if (sub_flow)
4348 		*sub_flow = dev_flow;
4349 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4350 }
4351 
4352 /**
4353  * Split the meter flow.
4354  *
4355  * As meter flow will split to three sub flow, other than meter
4356  * action, the other actions make sense to only meter accepts
4357  * the packet. If it need to be dropped, no other additional
4358  * actions should be take.
4359  *
4360  * One kind of special action which decapsulates the L3 tunnel
4361  * header will be in the prefix sub flow, as not to take the
4362  * L3 tunnel header into account.
4363  *
4364  * @param dev
4365  *   Pointer to Ethernet device.
4366  * @param[in] items
4367  *   Pattern specification (list terminated by the END pattern item).
4368  * @param[out] sfx_items
4369  *   Suffix flow match items (list terminated by the END pattern item).
4370  * @param[in] actions
4371  *   Associated actions (list terminated by the END action).
4372  * @param[out] actions_sfx
4373  *   Suffix flow actions.
4374  * @param[out] actions_pre
4375  *   Prefix flow actions.
4376  * @param[out] pattern_sfx
4377  *   The pattern items for the suffix flow.
4378  * @param[out] tag_sfx
4379  *   Pointer to suffix flow tag.
4380  *
4381  * @return
4382  *   0 on success.
4383  */
4384 static int
4385 flow_meter_split_prep(struct rte_eth_dev *dev,
4386 		 const struct rte_flow_item items[],
4387 		 struct rte_flow_item sfx_items[],
4388 		 const struct rte_flow_action actions[],
4389 		 struct rte_flow_action actions_sfx[],
4390 		 struct rte_flow_action actions_pre[])
4391 {
4392 	struct mlx5_priv *priv = dev->data->dev_private;
4393 	struct rte_flow_action *tag_action = NULL;
4394 	struct rte_flow_item *tag_item;
4395 	struct mlx5_rte_flow_action_set_tag *set_tag;
4396 	struct rte_flow_error error;
4397 	const struct rte_flow_action_raw_encap *raw_encap;
4398 	const struct rte_flow_action_raw_decap *raw_decap;
4399 	struct mlx5_rte_flow_item_tag *tag_spec;
4400 	struct mlx5_rte_flow_item_tag *tag_mask;
4401 	uint32_t tag_id = 0;
4402 	bool copy_vlan = false;
4403 
4404 	/* Prepare the actions for prefix and suffix flow. */
4405 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4406 		struct rte_flow_action **action_cur = NULL;
4407 
4408 		switch (actions->type) {
4409 		case RTE_FLOW_ACTION_TYPE_METER:
4410 			/* Add the extra tag action first. */
4411 			tag_action = actions_pre;
4412 			tag_action->type = (enum rte_flow_action_type)
4413 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4414 			actions_pre++;
4415 			action_cur = &actions_pre;
4416 			break;
4417 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4418 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4419 			action_cur = &actions_pre;
4420 			break;
4421 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4422 			raw_encap = actions->conf;
4423 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4424 				action_cur = &actions_pre;
4425 			break;
4426 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4427 			raw_decap = actions->conf;
4428 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4429 				action_cur = &actions_pre;
4430 			break;
4431 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4432 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4433 			copy_vlan = true;
4434 			break;
4435 		default:
4436 			break;
4437 		}
4438 		if (!action_cur)
4439 			action_cur = &actions_sfx;
4440 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4441 		(*action_cur)++;
4442 	}
4443 	/* Add end action to the actions. */
4444 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4445 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4446 	actions_pre++;
4447 	/* Set the tag. */
4448 	set_tag = (void *)actions_pre;
4449 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4450 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4451 			  &tag_id);
4452 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4453 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4454 			dev->data->port_id);
4455 		mlx5_ipool_free(priv->sh->ipool
4456 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4457 		return 0;
4458 	} else if (!tag_id) {
4459 		return 0;
4460 	}
4461 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4462 	assert(tag_action);
4463 	tag_action->conf = set_tag;
4464 	/* Prepare the suffix subflow items. */
4465 	tag_item = sfx_items++;
4466 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4467 		int item_type = items->type;
4468 
4469 		switch (item_type) {
4470 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4471 			memcpy(sfx_items, items, sizeof(*sfx_items));
4472 			sfx_items++;
4473 			break;
4474 		case RTE_FLOW_ITEM_TYPE_VLAN:
4475 			if (copy_vlan) {
4476 				memcpy(sfx_items, items, sizeof(*sfx_items));
4477 				/*
4478 				 * Convert to internal match item, it is used
4479 				 * for vlan push and set vid.
4480 				 */
4481 				sfx_items->type = (enum rte_flow_item_type)
4482 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4483 				sfx_items++;
4484 			}
4485 			break;
4486 		default:
4487 			break;
4488 		}
4489 	}
4490 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4491 	sfx_items++;
4492 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4493 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4494 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4495 	tag_mask = tag_spec + 1;
4496 	tag_mask->data = 0xffffff00;
4497 	tag_item->type = (enum rte_flow_item_type)
4498 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4499 	tag_item->spec = tag_spec;
4500 	tag_item->last = NULL;
4501 	tag_item->mask = tag_mask;
4502 	return tag_id;
4503 }
4504 
4505 /**
4506  * Split action list having QUEUE/RSS for metadata register copy.
4507  *
4508  * Once Q/RSS action is detected in user's action list, the flow action
4509  * should be split in order to copy metadata registers, which will happen in
4510  * RX_CP_TBL like,
4511  *   - CQE->flow_tag := reg_c[1] (MARK)
4512  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4513  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4514  * This is because the last action of each flow must be a terminal action
4515  * (QUEUE, RSS or DROP).
4516  *
4517  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4518  * stored and kept in the mlx5_flow structure per each sub_flow.
4519  *
4520  * The Q/RSS action is replaced with,
4521  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4522  * And the following JUMP action is added at the end,
4523  *   - JUMP, to RX_CP_TBL.
4524  *
4525  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4526  * flow_create_split_metadata() routine. The flow will look like,
4527  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4528  *
4529  * @param dev
4530  *   Pointer to Ethernet device.
4531  * @param[out] split_actions
4532  *   Pointer to store split actions to jump to CP_TBL.
4533  * @param[in] actions
4534  *   Pointer to the list of original flow actions.
4535  * @param[in] qrss
4536  *   Pointer to the Q/RSS action.
4537  * @param[in] actions_n
4538  *   Number of original actions.
4539  * @param[out] error
4540  *   Perform verbose error reporting if not NULL.
4541  *
4542  * @return
4543  *   non-zero unique flow_id on success, otherwise 0 and
4544  *   error/rte_error are set.
4545  */
4546 static uint32_t
4547 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4548 			  struct rte_flow_action *split_actions,
4549 			  const struct rte_flow_action *actions,
4550 			  const struct rte_flow_action *qrss,
4551 			  int actions_n, struct rte_flow_error *error)
4552 {
4553 	struct mlx5_priv *priv = dev->data->dev_private;
4554 	struct mlx5_rte_flow_action_set_tag *set_tag;
4555 	struct rte_flow_action_jump *jump;
4556 	const int qrss_idx = qrss - actions;
4557 	uint32_t flow_id = 0;
4558 	int ret = 0;
4559 
4560 	/*
4561 	 * Given actions will be split
4562 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4563 	 * - Add jump to mreg CP_TBL.
4564 	 * As a result, there will be one more action.
4565 	 */
4566 	++actions_n;
4567 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4568 	set_tag = (void *)(split_actions + actions_n);
4569 	/*
4570 	 * If tag action is not set to void(it means we are not the meter
4571 	 * suffix flow), add the tag action. Since meter suffix flow already
4572 	 * has the tag added.
4573 	 */
4574 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4575 		/*
4576 		 * Allocate the new subflow ID. This one is unique within
4577 		 * device and not shared with representors. Otherwise,
4578 		 * we would have to resolve multi-thread access synch
4579 		 * issue. Each flow on the shared device is appended
4580 		 * with source vport identifier, so the resulting
4581 		 * flows will be unique in the shared (by master and
4582 		 * representors) domain even if they have coinciding
4583 		 * IDs.
4584 		 */
4585 		mlx5_ipool_malloc(priv->sh->ipool
4586 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4587 		if (!flow_id)
4588 			return rte_flow_error_set(error, ENOMEM,
4589 						  RTE_FLOW_ERROR_TYPE_ACTION,
4590 						  NULL, "can't allocate id "
4591 						  "for split Q/RSS subflow");
4592 		/* Internal SET_TAG action to set flow ID. */
4593 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4594 			.data = flow_id,
4595 		};
4596 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4597 		if (ret < 0)
4598 			return ret;
4599 		set_tag->id = ret;
4600 		/* Construct new actions array. */
4601 		/* Replace QUEUE/RSS action. */
4602 		split_actions[qrss_idx] = (struct rte_flow_action){
4603 			.type = (enum rte_flow_action_type)
4604 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4605 			.conf = set_tag,
4606 		};
4607 	}
4608 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4609 	jump = (void *)(set_tag + 1);
4610 	*jump = (struct rte_flow_action_jump){
4611 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4612 	};
4613 	split_actions[actions_n - 2] = (struct rte_flow_action){
4614 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4615 		.conf = jump,
4616 	};
4617 	split_actions[actions_n - 1] = (struct rte_flow_action){
4618 		.type = RTE_FLOW_ACTION_TYPE_END,
4619 	};
4620 	return flow_id;
4621 }
4622 
4623 /**
4624  * Extend the given action list for Tx metadata copy.
4625  *
4626  * Copy the given action list to the ext_actions and add flow metadata register
4627  * copy action in order to copy reg_a set by WQE to reg_c[0].
4628  *
4629  * @param[out] ext_actions
4630  *   Pointer to the extended action list.
4631  * @param[in] actions
4632  *   Pointer to the list of actions.
4633  * @param[in] actions_n
4634  *   Number of actions in the list.
4635  * @param[out] error
4636  *   Perform verbose error reporting if not NULL.
4637  * @param[in] encap_idx
4638  *   The encap action inndex.
4639  *
4640  * @return
4641  *   0 on success, negative value otherwise
4642  */
4643 static int
4644 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4645 		       struct rte_flow_action *ext_actions,
4646 		       const struct rte_flow_action *actions,
4647 		       int actions_n, struct rte_flow_error *error,
4648 		       int encap_idx)
4649 {
4650 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4651 		(struct mlx5_flow_action_copy_mreg *)
4652 			(ext_actions + actions_n + 1);
4653 	int ret;
4654 
4655 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4656 	if (ret < 0)
4657 		return ret;
4658 	cp_mreg->dst = ret;
4659 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4660 	if (ret < 0)
4661 		return ret;
4662 	cp_mreg->src = ret;
4663 	if (encap_idx != 0)
4664 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4665 	if (encap_idx == actions_n - 1) {
4666 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4667 			.type = (enum rte_flow_action_type)
4668 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4669 			.conf = cp_mreg,
4670 		};
4671 		ext_actions[actions_n] = (struct rte_flow_action){
4672 			.type = RTE_FLOW_ACTION_TYPE_END,
4673 		};
4674 	} else {
4675 		ext_actions[encap_idx] = (struct rte_flow_action){
4676 			.type = (enum rte_flow_action_type)
4677 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4678 			.conf = cp_mreg,
4679 		};
4680 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4681 				sizeof(*ext_actions) * (actions_n - encap_idx));
4682 	}
4683 	return 0;
4684 }
4685 
4686 /**
4687  * Check the match action from the action list.
4688  *
4689  * @param[in] actions
4690  *   Pointer to the list of actions.
4691  * @param[in] attr
4692  *   Flow rule attributes.
4693  * @param[in] action
4694  *   The action to be check if exist.
4695  * @param[out] match_action_pos
4696  *   Pointer to the position of the matched action if exists, otherwise is -1.
4697  * @param[out] qrss_action_pos
4698  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4699  * @param[out] modify_after_mirror
4700  *   Pointer to the flag of modify action after FDB mirroring.
4701  *
4702  * @return
4703  *   > 0 the total number of actions.
4704  *   0 if not found match action in action list.
4705  */
4706 static int
4707 flow_check_match_action(const struct rte_flow_action actions[],
4708 			const struct rte_flow_attr *attr,
4709 			enum rte_flow_action_type action,
4710 			int *match_action_pos, int *qrss_action_pos,
4711 			int *modify_after_mirror)
4712 {
4713 	const struct rte_flow_action_sample *sample;
4714 	int actions_n = 0;
4715 	uint32_t ratio = 0;
4716 	int sub_type = 0;
4717 	int flag = 0;
4718 	int fdb_mirror = 0;
4719 
4720 	*match_action_pos = -1;
4721 	*qrss_action_pos = -1;
4722 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4723 		if (actions->type == action) {
4724 			flag = 1;
4725 			*match_action_pos = actions_n;
4726 		}
4727 		switch (actions->type) {
4728 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4729 		case RTE_FLOW_ACTION_TYPE_RSS:
4730 			*qrss_action_pos = actions_n;
4731 			break;
4732 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
4733 			sample = actions->conf;
4734 			ratio = sample->ratio;
4735 			sub_type = ((const struct rte_flow_action *)
4736 					(sample->actions))->type;
4737 			if (ratio == 1 && attr->transfer)
4738 				fdb_mirror = 1;
4739 			break;
4740 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4741 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4742 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4743 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4744 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4745 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4746 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4747 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4748 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4749 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
4750 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4751 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4752 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4753 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4754 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4755 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4756 		case RTE_FLOW_ACTION_TYPE_FLAG:
4757 		case RTE_FLOW_ACTION_TYPE_MARK:
4758 		case RTE_FLOW_ACTION_TYPE_SET_META:
4759 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
4760 			if (fdb_mirror)
4761 				*modify_after_mirror = 1;
4762 			break;
4763 		default:
4764 			break;
4765 		}
4766 		actions_n++;
4767 	}
4768 	if (flag && fdb_mirror && !*modify_after_mirror) {
4769 		/* FDB mirroring uses the destination array to implement
4770 		 * instead of FLOW_SAMPLER object.
4771 		 */
4772 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
4773 			flag = 0;
4774 	}
4775 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4776 	return flag ? actions_n + 1 : 0;
4777 }
4778 
4779 #define SAMPLE_SUFFIX_ITEM 2
4780 
4781 /**
4782  * Split the sample flow.
4783  *
4784  * As sample flow will split to two sub flow, sample flow with
4785  * sample action, the other actions will move to new suffix flow.
4786  *
4787  * Also add unique tag id with tag action in the sample flow,
4788  * the same tag id will be as match in the suffix flow.
4789  *
4790  * @param dev
4791  *   Pointer to Ethernet device.
4792  * @param[in] add_tag
4793  *   Add extra tag action flag.
4794  * @param[out] sfx_items
4795  *   Suffix flow match items (list terminated by the END pattern item).
4796  * @param[in] actions
4797  *   Associated actions (list terminated by the END action).
4798  * @param[out] actions_sfx
4799  *   Suffix flow actions.
4800  * @param[out] actions_pre
4801  *   Prefix flow actions.
4802  * @param[in] actions_n
4803  *  The total number of actions.
4804  * @param[in] sample_action_pos
4805  *   The sample action position.
4806  * @param[in] qrss_action_pos
4807  *   The Queue/RSS action position.
4808  * @param[in] jump_table
4809  *   Add extra jump action flag.
4810  * @param[out] error
4811  *   Perform verbose error reporting if not NULL.
4812  *
4813  * @return
4814  *   0 on success, or unique flow_id, a negative errno value
4815  *   otherwise and rte_errno is set.
4816  */
4817 static int
4818 flow_sample_split_prep(struct rte_eth_dev *dev,
4819 		       int add_tag,
4820 		       struct rte_flow_item sfx_items[],
4821 		       const struct rte_flow_action actions[],
4822 		       struct rte_flow_action actions_sfx[],
4823 		       struct rte_flow_action actions_pre[],
4824 		       int actions_n,
4825 		       int sample_action_pos,
4826 		       int qrss_action_pos,
4827 		       int jump_table,
4828 		       struct rte_flow_error *error)
4829 {
4830 	struct mlx5_priv *priv = dev->data->dev_private;
4831 	struct mlx5_rte_flow_action_set_tag *set_tag;
4832 	struct mlx5_rte_flow_item_tag *tag_spec;
4833 	struct mlx5_rte_flow_item_tag *tag_mask;
4834 	struct rte_flow_action_jump *jump_action;
4835 	uint32_t tag_id = 0;
4836 	int index;
4837 	int append_index = 0;
4838 	int ret;
4839 
4840 	if (sample_action_pos < 0)
4841 		return rte_flow_error_set(error, EINVAL,
4842 					  RTE_FLOW_ERROR_TYPE_ACTION,
4843 					  NULL, "invalid position of sample "
4844 					  "action in list");
4845 	/* Prepare the actions for prefix and suffix flow. */
4846 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4847 		index = qrss_action_pos;
4848 		/* Put the preceding the Queue/RSS action into prefix flow. */
4849 		if (index != 0)
4850 			memcpy(actions_pre, actions,
4851 			       sizeof(struct rte_flow_action) * index);
4852 		/* Put others preceding the sample action into prefix flow. */
4853 		if (sample_action_pos > index + 1)
4854 			memcpy(actions_pre + index, actions + index + 1,
4855 			       sizeof(struct rte_flow_action) *
4856 			       (sample_action_pos - index - 1));
4857 		index = sample_action_pos - 1;
4858 		/* Put Queue/RSS action into Suffix flow. */
4859 		memcpy(actions_sfx, actions + qrss_action_pos,
4860 		       sizeof(struct rte_flow_action));
4861 		actions_sfx++;
4862 	} else {
4863 		index = sample_action_pos;
4864 		if (index != 0)
4865 			memcpy(actions_pre, actions,
4866 			       sizeof(struct rte_flow_action) * index);
4867 	}
4868 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
4869 	 * For CX6DX and above, metadata registers Cx preserve their value,
4870 	 * add an extra tag action for NIC-RX and E-Switch Domain.
4871 	 */
4872 	if (add_tag) {
4873 		/* Prepare the prefix tag action. */
4874 		append_index++;
4875 		set_tag = (void *)(actions_pre + actions_n + append_index);
4876 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4877 		if (ret < 0)
4878 			return ret;
4879 		set_tag->id = ret;
4880 		mlx5_ipool_malloc(priv->sh->ipool
4881 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4882 		set_tag->data = tag_id;
4883 		/* Prepare the suffix subflow items. */
4884 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4885 		tag_spec->data = tag_id;
4886 		tag_spec->id = set_tag->id;
4887 		tag_mask = tag_spec + 1;
4888 		tag_mask->data = UINT32_MAX;
4889 		sfx_items[0] = (struct rte_flow_item){
4890 			.type = (enum rte_flow_item_type)
4891 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4892 			.spec = tag_spec,
4893 			.last = NULL,
4894 			.mask = tag_mask,
4895 		};
4896 		sfx_items[1] = (struct rte_flow_item){
4897 			.type = (enum rte_flow_item_type)
4898 				RTE_FLOW_ITEM_TYPE_END,
4899 		};
4900 		/* Prepare the tag action in prefix subflow. */
4901 		actions_pre[index++] =
4902 			(struct rte_flow_action){
4903 			.type = (enum rte_flow_action_type)
4904 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4905 			.conf = set_tag,
4906 		};
4907 	}
4908 	memcpy(actions_pre + index, actions + sample_action_pos,
4909 	       sizeof(struct rte_flow_action));
4910 	index += 1;
4911 	/* For the modify action after the sample action in E-Switch mirroring,
4912 	 * Add the extra jump action in prefix subflow and jump into the next
4913 	 * table, then do the modify action in the new table.
4914 	 */
4915 	if (jump_table) {
4916 		/* Prepare the prefix jump action. */
4917 		append_index++;
4918 		jump_action = (void *)(actions_pre + actions_n + append_index);
4919 		jump_action->group = jump_table;
4920 		actions_pre[index++] =
4921 			(struct rte_flow_action){
4922 			.type = (enum rte_flow_action_type)
4923 				RTE_FLOW_ACTION_TYPE_JUMP,
4924 			.conf = jump_action,
4925 		};
4926 	}
4927 	actions_pre[index] = (struct rte_flow_action){
4928 		.type = (enum rte_flow_action_type)
4929 			RTE_FLOW_ACTION_TYPE_END,
4930 	};
4931 	/* Put the actions after sample into Suffix flow. */
4932 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4933 	       sizeof(struct rte_flow_action) *
4934 	       (actions_n - sample_action_pos - 1));
4935 	return tag_id;
4936 }
4937 
4938 /**
4939  * The splitting for metadata feature.
4940  *
4941  * - Q/RSS action on NIC Rx should be split in order to pass by
4942  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4943  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4944  *
4945  * - All the actions on NIC Tx should have a mreg copy action to
4946  *   copy reg_a from WQE to reg_c[0].
4947  *
4948  * @param dev
4949  *   Pointer to Ethernet device.
4950  * @param[in] flow
4951  *   Parent flow structure pointer.
4952  * @param[in] attr
4953  *   Flow rule attributes.
4954  * @param[in] items
4955  *   Pattern specification (list terminated by the END pattern item).
4956  * @param[in] actions
4957  *   Associated actions (list terminated by the END action).
4958  * @param[in] flow_split_info
4959  *   Pointer to flow split info structure.
4960  * @param[out] error
4961  *   Perform verbose error reporting if not NULL.
4962  * @return
4963  *   0 on success, negative value otherwise
4964  */
4965 static int
4966 flow_create_split_metadata(struct rte_eth_dev *dev,
4967 			   struct rte_flow *flow,
4968 			   const struct rte_flow_attr *attr,
4969 			   const struct rte_flow_item items[],
4970 			   const struct rte_flow_action actions[],
4971 			   struct mlx5_flow_split_info *flow_split_info,
4972 			   struct rte_flow_error *error)
4973 {
4974 	struct mlx5_priv *priv = dev->data->dev_private;
4975 	struct mlx5_dev_config *config = &priv->config;
4976 	const struct rte_flow_action *qrss = NULL;
4977 	struct rte_flow_action *ext_actions = NULL;
4978 	struct mlx5_flow *dev_flow = NULL;
4979 	uint32_t qrss_id = 0;
4980 	int mtr_sfx = 0;
4981 	size_t act_size;
4982 	int actions_n;
4983 	int encap_idx;
4984 	int ret;
4985 
4986 	/* Check whether extensive metadata feature is engaged. */
4987 	if (!config->dv_flow_en ||
4988 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4989 	    !mlx5_flow_ext_mreg_supported(dev))
4990 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4991 					       actions, flow_split_info, error);
4992 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4993 							   &encap_idx);
4994 	if (qrss) {
4995 		/* Exclude hairpin flows from splitting. */
4996 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4997 			const struct rte_flow_action_queue *queue;
4998 
4999 			queue = qrss->conf;
5000 			if (mlx5_rxq_get_type(dev, queue->index) ==
5001 			    MLX5_RXQ_TYPE_HAIRPIN)
5002 				qrss = NULL;
5003 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5004 			const struct rte_flow_action_rss *rss;
5005 
5006 			rss = qrss->conf;
5007 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5008 			    MLX5_RXQ_TYPE_HAIRPIN)
5009 				qrss = NULL;
5010 		}
5011 	}
5012 	if (qrss) {
5013 		/* Check if it is in meter suffix table. */
5014 		mtr_sfx = attr->group == (attr->transfer ?
5015 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5016 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5017 		/*
5018 		 * Q/RSS action on NIC Rx should be split in order to pass by
5019 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5020 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5021 		 */
5022 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5023 			   sizeof(struct rte_flow_action_set_tag) +
5024 			   sizeof(struct rte_flow_action_jump);
5025 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5026 					  SOCKET_ID_ANY);
5027 		if (!ext_actions)
5028 			return rte_flow_error_set(error, ENOMEM,
5029 						  RTE_FLOW_ERROR_TYPE_ACTION,
5030 						  NULL, "no memory to split "
5031 						  "metadata flow");
5032 		/*
5033 		 * If we are the suffix flow of meter, tag already exist.
5034 		 * Set the tag action to void.
5035 		 */
5036 		if (mtr_sfx)
5037 			ext_actions[qrss - actions].type =
5038 						RTE_FLOW_ACTION_TYPE_VOID;
5039 		else
5040 			ext_actions[qrss - actions].type =
5041 						(enum rte_flow_action_type)
5042 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5043 		/*
5044 		 * Create the new actions list with removed Q/RSS action
5045 		 * and appended set tag and jump to register copy table
5046 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5047 		 * in advance, because it is needed for set tag action.
5048 		 */
5049 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5050 						    qrss, actions_n, error);
5051 		if (!mtr_sfx && !qrss_id) {
5052 			ret = -rte_errno;
5053 			goto exit;
5054 		}
5055 	} else if (attr->egress && !attr->transfer) {
5056 		/*
5057 		 * All the actions on NIC Tx should have a metadata register
5058 		 * copy action to copy reg_a from WQE to reg_c[meta]
5059 		 */
5060 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5061 			   sizeof(struct mlx5_flow_action_copy_mreg);
5062 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5063 					  SOCKET_ID_ANY);
5064 		if (!ext_actions)
5065 			return rte_flow_error_set(error, ENOMEM,
5066 						  RTE_FLOW_ERROR_TYPE_ACTION,
5067 						  NULL, "no memory to split "
5068 						  "metadata flow");
5069 		/* Create the action list appended with copy register. */
5070 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5071 					     actions_n, error, encap_idx);
5072 		if (ret < 0)
5073 			goto exit;
5074 	}
5075 	/* Add the unmodified original or prefix subflow. */
5076 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5077 				      items, ext_actions ? ext_actions :
5078 				      actions, flow_split_info, error);
5079 	if (ret < 0)
5080 		goto exit;
5081 	MLX5_ASSERT(dev_flow);
5082 	if (qrss) {
5083 		const struct rte_flow_attr q_attr = {
5084 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5085 			.ingress = 1,
5086 		};
5087 		/* Internal PMD action to set register. */
5088 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5089 			.data = qrss_id,
5090 			.id = REG_NON,
5091 		};
5092 		struct rte_flow_item q_items[] = {
5093 			{
5094 				.type = (enum rte_flow_item_type)
5095 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5096 				.spec = &q_tag_spec,
5097 				.last = NULL,
5098 				.mask = NULL,
5099 			},
5100 			{
5101 				.type = RTE_FLOW_ITEM_TYPE_END,
5102 			},
5103 		};
5104 		struct rte_flow_action q_actions[] = {
5105 			{
5106 				.type = qrss->type,
5107 				.conf = qrss->conf,
5108 			},
5109 			{
5110 				.type = RTE_FLOW_ACTION_TYPE_END,
5111 			},
5112 		};
5113 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5114 
5115 		/*
5116 		 * Configure the tag item only if there is no meter subflow.
5117 		 * Since tag is already marked in the meter suffix subflow
5118 		 * we can just use the meter suffix items as is.
5119 		 */
5120 		if (qrss_id) {
5121 			/* Not meter subflow. */
5122 			MLX5_ASSERT(!mtr_sfx);
5123 			/*
5124 			 * Put unique id in prefix flow due to it is destroyed
5125 			 * after suffix flow and id will be freed after there
5126 			 * is no actual flows with this id and identifier
5127 			 * reallocation becomes possible (for example, for
5128 			 * other flows in other threads).
5129 			 */
5130 			dev_flow->handle->split_flow_id = qrss_id;
5131 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5132 						   error);
5133 			if (ret < 0)
5134 				goto exit;
5135 			q_tag_spec.id = ret;
5136 		}
5137 		dev_flow = NULL;
5138 		/* Add suffix subflow to execute Q/RSS. */
5139 		flow_split_info->prefix_layers = layers;
5140 		flow_split_info->prefix_mark = 0;
5141 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5142 					      &q_attr, mtr_sfx ? items :
5143 					      q_items, q_actions,
5144 					      flow_split_info, error);
5145 		if (ret < 0)
5146 			goto exit;
5147 		/* qrss ID should be freed if failed. */
5148 		qrss_id = 0;
5149 		MLX5_ASSERT(dev_flow);
5150 	}
5151 
5152 exit:
5153 	/*
5154 	 * We do not destroy the partially created sub_flows in case of error.
5155 	 * These ones are included into parent flow list and will be destroyed
5156 	 * by flow_drv_destroy.
5157 	 */
5158 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5159 			qrss_id);
5160 	mlx5_free(ext_actions);
5161 	return ret;
5162 }
5163 
5164 /**
5165  * The splitting for meter feature.
5166  *
5167  * - The meter flow will be split to two flows as prefix and
5168  *   suffix flow. The packets make sense only it pass the prefix
5169  *   meter action.
5170  *
5171  * - Reg_C_5 is used for the packet to match betweend prefix and
5172  *   suffix flow.
5173  *
5174  * @param dev
5175  *   Pointer to Ethernet device.
5176  * @param[in] flow
5177  *   Parent flow structure pointer.
5178  * @param[in] attr
5179  *   Flow rule attributes.
5180  * @param[in] items
5181  *   Pattern specification (list terminated by the END pattern item).
5182  * @param[in] actions
5183  *   Associated actions (list terminated by the END action).
5184  * @param[in] flow_split_info
5185  *   Pointer to flow split info structure.
5186  * @param[out] error
5187  *   Perform verbose error reporting if not NULL.
5188  * @return
5189  *   0 on success, negative value otherwise
5190  */
5191 static int
5192 flow_create_split_meter(struct rte_eth_dev *dev,
5193 			struct rte_flow *flow,
5194 			const struct rte_flow_attr *attr,
5195 			const struct rte_flow_item items[],
5196 			const struct rte_flow_action actions[],
5197 			struct mlx5_flow_split_info *flow_split_info,
5198 			struct rte_flow_error *error)
5199 {
5200 	struct mlx5_priv *priv = dev->data->dev_private;
5201 	struct rte_flow_action *sfx_actions = NULL;
5202 	struct rte_flow_action *pre_actions = NULL;
5203 	struct rte_flow_item *sfx_items = NULL;
5204 	struct mlx5_flow *dev_flow = NULL;
5205 	struct rte_flow_attr sfx_attr = *attr;
5206 	uint32_t mtr = 0;
5207 	uint32_t mtr_tag_id = 0;
5208 	size_t act_size;
5209 	size_t item_size;
5210 	int actions_n = 0;
5211 	int ret;
5212 
5213 	if (priv->mtr_en)
5214 		actions_n = flow_check_meter_action(actions, &mtr);
5215 	if (mtr) {
5216 		/* The five prefix actions: meter, decap, encap, tag, end. */
5217 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5218 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5219 		/* tag, vlan, port id, end. */
5220 #define METER_SUFFIX_ITEM 4
5221 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5222 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5223 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5224 					  0, SOCKET_ID_ANY);
5225 		if (!sfx_actions)
5226 			return rte_flow_error_set(error, ENOMEM,
5227 						  RTE_FLOW_ERROR_TYPE_ACTION,
5228 						  NULL, "no memory to split "
5229 						  "meter flow");
5230 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5231 			     act_size);
5232 		pre_actions = sfx_actions + actions_n;
5233 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5234 						   actions, sfx_actions,
5235 						   pre_actions);
5236 		if (!mtr_tag_id) {
5237 			ret = -rte_errno;
5238 			goto exit;
5239 		}
5240 		/* Add the prefix subflow. */
5241 		flow_split_info->prefix_mark = 0;
5242 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5243 					      attr, items, pre_actions,
5244 					      flow_split_info, error);
5245 		if (ret) {
5246 			ret = -rte_errno;
5247 			goto exit;
5248 		}
5249 		dev_flow->handle->split_flow_id = mtr_tag_id;
5250 		/* Setting the sfx group atrr. */
5251 		sfx_attr.group = sfx_attr.transfer ?
5252 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5253 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5254 		flow_split_info->prefix_layers =
5255 				flow_get_prefix_layer_flags(dev_flow);
5256 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5257 	}
5258 	/* Add the prefix subflow. */
5259 	ret = flow_create_split_metadata(dev, flow,
5260 					 &sfx_attr, sfx_items ?
5261 					 sfx_items : items,
5262 					 sfx_actions ? sfx_actions : actions,
5263 					 flow_split_info, error);
5264 exit:
5265 	if (sfx_actions)
5266 		mlx5_free(sfx_actions);
5267 	return ret;
5268 }
5269 
5270 /**
5271  * The splitting for sample feature.
5272  *
5273  * Once Sample action is detected in the action list, the flow actions should
5274  * be split into prefix sub flow and suffix sub flow.
5275  *
5276  * The original items remain in the prefix sub flow, all actions preceding the
5277  * sample action and the sample action itself will be copied to the prefix
5278  * sub flow, the actions following the sample action will be copied to the
5279  * suffix sub flow, Queue action always be located in the suffix sub flow.
5280  *
5281  * In order to make the packet from prefix sub flow matches with suffix sub
5282  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5283  * flow uses tag item with the unique flow id.
5284  *
5285  * @param dev
5286  *   Pointer to Ethernet device.
5287  * @param[in] flow
5288  *   Parent flow structure pointer.
5289  * @param[in] attr
5290  *   Flow rule attributes.
5291  * @param[in] items
5292  *   Pattern specification (list terminated by the END pattern item).
5293  * @param[in] actions
5294  *   Associated actions (list terminated by the END action).
5295  * @param[in] flow_split_info
5296  *   Pointer to flow split info structure.
5297  * @param[out] error
5298  *   Perform verbose error reporting if not NULL.
5299  * @return
5300  *   0 on success, negative value otherwise
5301  */
5302 static int
5303 flow_create_split_sample(struct rte_eth_dev *dev,
5304 			 struct rte_flow *flow,
5305 			 const struct rte_flow_attr *attr,
5306 			 const struct rte_flow_item items[],
5307 			 const struct rte_flow_action actions[],
5308 			 struct mlx5_flow_split_info *flow_split_info,
5309 			 struct rte_flow_error *error)
5310 {
5311 	struct mlx5_priv *priv = dev->data->dev_private;
5312 	struct rte_flow_action *sfx_actions = NULL;
5313 	struct rte_flow_action *pre_actions = NULL;
5314 	struct rte_flow_item *sfx_items = NULL;
5315 	struct mlx5_flow *dev_flow = NULL;
5316 	struct rte_flow_attr sfx_attr = *attr;
5317 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5318 	struct mlx5_flow_dv_sample_resource *sample_res;
5319 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5320 	struct mlx5_flow_tbl_resource *sfx_tbl;
5321 #endif
5322 	size_t act_size;
5323 	size_t item_size;
5324 	uint32_t fdb_tx = 0;
5325 	int32_t tag_id = 0;
5326 	int actions_n = 0;
5327 	int sample_action_pos;
5328 	int qrss_action_pos;
5329 	int add_tag = 0;
5330 	int modify_after_mirror = 0;
5331 	uint16_t jump_table = 0;
5332 	const uint32_t next_ft_step = 1;
5333 	int ret = 0;
5334 
5335 	if (priv->sampler_en)
5336 		actions_n = flow_check_match_action(actions, attr,
5337 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5338 					&sample_action_pos, &qrss_action_pos,
5339 					&modify_after_mirror);
5340 	if (actions_n) {
5341 		/* The prefix actions must includes sample, tag, end. */
5342 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5343 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5344 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5345 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5346 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5347 					  item_size), 0, SOCKET_ID_ANY);
5348 		if (!sfx_actions)
5349 			return rte_flow_error_set(error, ENOMEM,
5350 						  RTE_FLOW_ERROR_TYPE_ACTION,
5351 						  NULL, "no memory to split "
5352 						  "sample flow");
5353 		/* The representor_id is -1 for uplink. */
5354 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5355 		/*
5356 		 * When reg_c_preserve is set, metadata registers Cx preserve
5357 		 * their value even through packet duplication.
5358 		 */
5359 		add_tag = (!fdb_tx || priv->config.hca_attr.reg_c_preserve);
5360 		if (add_tag)
5361 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5362 					+ act_size);
5363 		if (modify_after_mirror)
5364 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
5365 				     next_ft_step;
5366 		pre_actions = sfx_actions + actions_n;
5367 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
5368 						actions, sfx_actions,
5369 						pre_actions, actions_n,
5370 						sample_action_pos,
5371 						qrss_action_pos, jump_table,
5372 						error);
5373 		if (tag_id < 0 || (add_tag && !tag_id)) {
5374 			ret = -rte_errno;
5375 			goto exit;
5376 		}
5377 		if (modify_after_mirror)
5378 			flow_split_info->skip_scale =
5379 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
5380 		/* Add the prefix subflow. */
5381 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5382 					      items, pre_actions,
5383 					      flow_split_info, error);
5384 		if (ret) {
5385 			ret = -rte_errno;
5386 			goto exit;
5387 		}
5388 		dev_flow->handle->split_flow_id = tag_id;
5389 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5390 		if (!modify_after_mirror) {
5391 			/* Set the sfx group attr. */
5392 			sample_res = (struct mlx5_flow_dv_sample_resource *)
5393 						dev_flow->dv.sample_res;
5394 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
5395 						sample_res->normal_path_tbl;
5396 			sfx_tbl_data = container_of(sfx_tbl,
5397 						struct mlx5_flow_tbl_data_entry,
5398 						tbl);
5399 			sfx_attr.group = sfx_attr.transfer ?
5400 						(sfx_tbl_data->table_id - 1) :
5401 						sfx_tbl_data->table_id;
5402 		} else {
5403 			MLX5_ASSERT(attr->transfer);
5404 			sfx_attr.group = jump_table;
5405 		}
5406 		flow_split_info->prefix_layers =
5407 				flow_get_prefix_layer_flags(dev_flow);
5408 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5409 		/* Suffix group level already be scaled with factor, set
5410 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
5411 		 * again in translation.
5412 		 */
5413 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5414 #endif
5415 	}
5416 	/* Add the suffix subflow. */
5417 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5418 				      sfx_items ? sfx_items : items,
5419 				      sfx_actions ? sfx_actions : actions,
5420 				      flow_split_info, error);
5421 exit:
5422 	if (sfx_actions)
5423 		mlx5_free(sfx_actions);
5424 	return ret;
5425 }
5426 
5427 /**
5428  * Split the flow to subflow set. The splitters might be linked
5429  * in the chain, like this:
5430  * flow_create_split_outer() calls:
5431  *   flow_create_split_meter() calls:
5432  *     flow_create_split_metadata(meter_subflow_0) calls:
5433  *       flow_create_split_inner(metadata_subflow_0)
5434  *       flow_create_split_inner(metadata_subflow_1)
5435  *       flow_create_split_inner(metadata_subflow_2)
5436  *     flow_create_split_metadata(meter_subflow_1) calls:
5437  *       flow_create_split_inner(metadata_subflow_0)
5438  *       flow_create_split_inner(metadata_subflow_1)
5439  *       flow_create_split_inner(metadata_subflow_2)
5440  *
5441  * This provide flexible way to add new levels of flow splitting.
5442  * The all of successfully created subflows are included to the
5443  * parent flow dev_flow list.
5444  *
5445  * @param dev
5446  *   Pointer to Ethernet device.
5447  * @param[in] flow
5448  *   Parent flow structure pointer.
5449  * @param[in] attr
5450  *   Flow rule attributes.
5451  * @param[in] items
5452  *   Pattern specification (list terminated by the END pattern item).
5453  * @param[in] actions
5454  *   Associated actions (list terminated by the END action).
5455  * @param[in] flow_split_info
5456  *   Pointer to flow split info structure.
5457  * @param[out] error
5458  *   Perform verbose error reporting if not NULL.
5459  * @return
5460  *   0 on success, negative value otherwise
5461  */
5462 static int
5463 flow_create_split_outer(struct rte_eth_dev *dev,
5464 			struct rte_flow *flow,
5465 			const struct rte_flow_attr *attr,
5466 			const struct rte_flow_item items[],
5467 			const struct rte_flow_action actions[],
5468 			struct mlx5_flow_split_info *flow_split_info,
5469 			struct rte_flow_error *error)
5470 {
5471 	int ret;
5472 
5473 	ret = flow_create_split_sample(dev, flow, attr, items,
5474 				       actions, flow_split_info, error);
5475 	MLX5_ASSERT(ret <= 0);
5476 	return ret;
5477 }
5478 
5479 static struct mlx5_flow_tunnel *
5480 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5481 		      const struct rte_flow_attr *attr,
5482 		      const struct rte_flow_item items[],
5483 		      const struct rte_flow_action actions[])
5484 {
5485 	struct mlx5_flow_tunnel *tunnel;
5486 
5487 #pragma GCC diagnostic push
5488 #pragma GCC diagnostic ignored "-Wcast-qual"
5489 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5490 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5491 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5492 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5493 	else
5494 		tunnel = NULL;
5495 #pragma GCC diagnostic pop
5496 
5497 	return tunnel;
5498 }
5499 
5500 /**
5501  * Adjust flow RSS workspace if needed.
5502  *
5503  * @param wks
5504  *   Pointer to thread flow work space.
5505  * @param rss_desc
5506  *   Pointer to RSS descriptor.
5507  * @param[in] nrssq_num
5508  *   New RSS queue number.
5509  *
5510  * @return
5511  *   0 on success, -1 otherwise and rte_errno is set.
5512  */
5513 static int
5514 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5515 			  struct mlx5_flow_rss_desc *rss_desc,
5516 			  uint32_t nrssq_num)
5517 {
5518 	if (likely(nrssq_num <= wks->rssq_num))
5519 		return 0;
5520 	rss_desc->queue = realloc(rss_desc->queue,
5521 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5522 	if (!rss_desc->queue) {
5523 		rte_errno = ENOMEM;
5524 		return -1;
5525 	}
5526 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5527 	return 0;
5528 }
5529 
5530 /**
5531  * Create a flow and add it to @p list.
5532  *
5533  * @param dev
5534  *   Pointer to Ethernet device.
5535  * @param list
5536  *   Pointer to a TAILQ flow list. If this parameter NULL,
5537  *   no list insertion occurred, flow is just created,
5538  *   this is caller's responsibility to track the
5539  *   created flow.
5540  * @param[in] attr
5541  *   Flow rule attributes.
5542  * @param[in] items
5543  *   Pattern specification (list terminated by the END pattern item).
5544  * @param[in] actions
5545  *   Associated actions (list terminated by the END action).
5546  * @param[in] external
5547  *   This flow rule is created by request external to PMD.
5548  * @param[out] error
5549  *   Perform verbose error reporting if not NULL.
5550  *
5551  * @return
5552  *   A flow index on success, 0 otherwise and rte_errno is set.
5553  */
5554 static uint32_t
5555 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5556 		 const struct rte_flow_attr *attr,
5557 		 const struct rte_flow_item items[],
5558 		 const struct rte_flow_action original_actions[],
5559 		 bool external, struct rte_flow_error *error)
5560 {
5561 	struct mlx5_priv *priv = dev->data->dev_private;
5562 	struct rte_flow *flow = NULL;
5563 	struct mlx5_flow *dev_flow;
5564 	const struct rte_flow_action_rss *rss = NULL;
5565 	struct mlx5_translated_shared_action
5566 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5567 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5568 	union {
5569 		struct mlx5_flow_expand_rss buf;
5570 		uint8_t buffer[2048];
5571 	} expand_buffer;
5572 	union {
5573 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5574 		uint8_t buffer[2048];
5575 	} actions_rx;
5576 	union {
5577 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5578 		uint8_t buffer[2048];
5579 	} actions_hairpin_tx;
5580 	union {
5581 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5582 		uint8_t buffer[2048];
5583 	} items_tx;
5584 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5585 	struct mlx5_flow_rss_desc *rss_desc;
5586 	const struct rte_flow_action *p_actions_rx;
5587 	uint32_t i;
5588 	uint32_t idx = 0;
5589 	int hairpin_flow;
5590 	struct rte_flow_attr attr_tx = { .priority = 0 };
5591 	const struct rte_flow_action *actions;
5592 	struct rte_flow_action *translated_actions = NULL;
5593 	struct mlx5_flow_tunnel *tunnel;
5594 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5595 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5596 	struct mlx5_flow_split_info flow_split_info = {
5597 		.external = !!external,
5598 		.skip_scale = 0,
5599 		.flow_idx = 0,
5600 		.prefix_mark = 0,
5601 		.prefix_layers = 0
5602 	};
5603 	int ret;
5604 
5605 	MLX5_ASSERT(wks);
5606 	rss_desc = &wks->rss_desc;
5607 	ret = flow_shared_actions_translate(dev, original_actions,
5608 					    shared_actions,
5609 					    &shared_actions_n,
5610 					    &translated_actions, error);
5611 	if (ret < 0) {
5612 		MLX5_ASSERT(translated_actions == NULL);
5613 		return 0;
5614 	}
5615 	actions = translated_actions ? translated_actions : original_actions;
5616 	p_actions_rx = actions;
5617 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5618 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5619 				external, hairpin_flow, error);
5620 	if (ret < 0)
5621 		goto error_before_hairpin_split;
5622 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5623 	if (!flow) {
5624 		rte_errno = ENOMEM;
5625 		goto error_before_hairpin_split;
5626 	}
5627 	if (hairpin_flow > 0) {
5628 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5629 			rte_errno = EINVAL;
5630 			goto error_before_hairpin_split;
5631 		}
5632 		flow_hairpin_split(dev, actions, actions_rx.actions,
5633 				   actions_hairpin_tx.actions, items_tx.items,
5634 				   idx);
5635 		p_actions_rx = actions_rx.actions;
5636 	}
5637 	flow_split_info.flow_idx = idx;
5638 	flow->drv_type = flow_get_drv_type(dev, attr);
5639 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5640 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5641 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5642 	/* RSS Action only works on NIC RX domain */
5643 	if (attr->ingress && !attr->transfer)
5644 		rss = flow_get_rss_action(p_actions_rx);
5645 	if (rss) {
5646 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5647 			return 0;
5648 		/*
5649 		 * The following information is required by
5650 		 * mlx5_flow_hashfields_adjust() in advance.
5651 		 */
5652 		rss_desc->level = rss->level;
5653 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5654 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5655 	}
5656 	flow->dev_handles = 0;
5657 	if (rss && rss->types) {
5658 		unsigned int graph_root;
5659 
5660 		graph_root = find_graph_root(items, rss->level);
5661 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5662 					   items, rss->types,
5663 					   mlx5_support_expansion, graph_root);
5664 		MLX5_ASSERT(ret > 0 &&
5665 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5666 	} else {
5667 		buf->entries = 1;
5668 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5669 	}
5670 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5671 						      shared_actions_n);
5672 	for (i = 0; i < buf->entries; ++i) {
5673 		/* Initialize flow split data. */
5674 		flow_split_info.prefix_layers = 0;
5675 		flow_split_info.prefix_mark = 0;
5676 		flow_split_info.skip_scale = 0;
5677 		/*
5678 		 * The splitter may create multiple dev_flows,
5679 		 * depending on configuration. In the simplest
5680 		 * case it just creates unmodified original flow.
5681 		 */
5682 		ret = flow_create_split_outer(dev, flow, attr,
5683 					      buf->entry[i].pattern,
5684 					      p_actions_rx, &flow_split_info,
5685 					      error);
5686 		if (ret < 0)
5687 			goto error;
5688 		if (is_flow_tunnel_steer_rule(dev, attr,
5689 					      buf->entry[i].pattern,
5690 					      p_actions_rx)) {
5691 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5692 							   p_actions_rx,
5693 							   idx,
5694 							   &default_miss_ctx,
5695 							   error);
5696 			if (ret < 0) {
5697 				mlx5_free(default_miss_ctx.queue);
5698 				goto error;
5699 			}
5700 		}
5701 	}
5702 	/* Create the tx flow. */
5703 	if (hairpin_flow) {
5704 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5705 		attr_tx.ingress = 0;
5706 		attr_tx.egress = 1;
5707 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5708 					 actions_hairpin_tx.actions,
5709 					 idx, error);
5710 		if (!dev_flow)
5711 			goto error;
5712 		dev_flow->flow = flow;
5713 		dev_flow->external = 0;
5714 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5715 			      dev_flow->handle, next);
5716 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5717 					 items_tx.items,
5718 					 actions_hairpin_tx.actions, error);
5719 		if (ret < 0)
5720 			goto error;
5721 	}
5722 	/*
5723 	 * Update the metadata register copy table. If extensive
5724 	 * metadata feature is enabled and registers are supported
5725 	 * we might create the extra rte_flow for each unique
5726 	 * MARK/FLAG action ID.
5727 	 *
5728 	 * The table is updated for ingress Flows only, because
5729 	 * the egress Flows belong to the different device and
5730 	 * copy table should be updated in peer NIC Rx domain.
5731 	 */
5732 	if (attr->ingress &&
5733 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5734 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5735 		if (ret)
5736 			goto error;
5737 	}
5738 	/*
5739 	 * If the flow is external (from application) OR device is started,
5740 	 * OR mreg discover, then apply immediately.
5741 	 */
5742 	if (external || dev->data->dev_started ||
5743 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5744 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
5745 		ret = flow_drv_apply(dev, flow, error);
5746 		if (ret < 0)
5747 			goto error;
5748 	}
5749 	if (list) {
5750 		rte_spinlock_lock(&priv->flow_list_lock);
5751 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5752 			     flow, next);
5753 		rte_spinlock_unlock(&priv->flow_list_lock);
5754 	}
5755 	flow_rxq_flags_set(dev, flow);
5756 	rte_free(translated_actions);
5757 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5758 	if (tunnel) {
5759 		flow->tunnel = 1;
5760 		flow->tunnel_id = tunnel->tunnel_id;
5761 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5762 		mlx5_free(default_miss_ctx.queue);
5763 	}
5764 	mlx5_flow_pop_thread_workspace();
5765 	return idx;
5766 error:
5767 	MLX5_ASSERT(flow);
5768 	ret = rte_errno; /* Save rte_errno before cleanup. */
5769 	flow_mreg_del_copy_action(dev, flow);
5770 	flow_drv_destroy(dev, flow);
5771 	if (rss_desc->shared_rss)
5772 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5773 			mlx5_ipool_get
5774 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5775 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5776 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5777 	rte_errno = ret; /* Restore rte_errno. */
5778 	ret = rte_errno;
5779 	rte_errno = ret;
5780 	mlx5_flow_pop_thread_workspace();
5781 error_before_hairpin_split:
5782 	rte_free(translated_actions);
5783 	return 0;
5784 }
5785 
5786 /**
5787  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5788  * incoming packets to table 1.
5789  *
5790  * Other flow rules, requested for group n, will be created in
5791  * e-switch table n+1.
5792  * Jump action to e-switch group n will be created to group n+1.
5793  *
5794  * Used when working in switchdev mode, to utilise advantages of table 1
5795  * and above.
5796  *
5797  * @param dev
5798  *   Pointer to Ethernet device.
5799  *
5800  * @return
5801  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5802  */
5803 struct rte_flow *
5804 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5805 {
5806 	const struct rte_flow_attr attr = {
5807 		.group = 0,
5808 		.priority = 0,
5809 		.ingress = 1,
5810 		.egress = 0,
5811 		.transfer = 1,
5812 	};
5813 	const struct rte_flow_item pattern = {
5814 		.type = RTE_FLOW_ITEM_TYPE_END,
5815 	};
5816 	struct rte_flow_action_jump jump = {
5817 		.group = 1,
5818 	};
5819 	const struct rte_flow_action actions[] = {
5820 		{
5821 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5822 			.conf = &jump,
5823 		},
5824 		{
5825 			.type = RTE_FLOW_ACTION_TYPE_END,
5826 		},
5827 	};
5828 	struct mlx5_priv *priv = dev->data->dev_private;
5829 	struct rte_flow_error error;
5830 
5831 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5832 						   &attr, &pattern,
5833 						   actions, false, &error);
5834 }
5835 
5836 /**
5837  * Validate a flow supported by the NIC.
5838  *
5839  * @see rte_flow_validate()
5840  * @see rte_flow_ops
5841  */
5842 int
5843 mlx5_flow_validate(struct rte_eth_dev *dev,
5844 		   const struct rte_flow_attr *attr,
5845 		   const struct rte_flow_item items[],
5846 		   const struct rte_flow_action original_actions[],
5847 		   struct rte_flow_error *error)
5848 {
5849 	int hairpin_flow;
5850 	struct mlx5_translated_shared_action
5851 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5852 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5853 	const struct rte_flow_action *actions;
5854 	struct rte_flow_action *translated_actions = NULL;
5855 	int ret = flow_shared_actions_translate(dev, original_actions,
5856 						shared_actions,
5857 						&shared_actions_n,
5858 						&translated_actions, error);
5859 
5860 	if (ret)
5861 		return ret;
5862 	actions = translated_actions ? translated_actions : original_actions;
5863 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5864 	ret = flow_drv_validate(dev, attr, items, actions,
5865 				true, hairpin_flow, error);
5866 	rte_free(translated_actions);
5867 	return ret;
5868 }
5869 
5870 /**
5871  * Create a flow.
5872  *
5873  * @see rte_flow_create()
5874  * @see rte_flow_ops
5875  */
5876 struct rte_flow *
5877 mlx5_flow_create(struct rte_eth_dev *dev,
5878 		 const struct rte_flow_attr *attr,
5879 		 const struct rte_flow_item items[],
5880 		 const struct rte_flow_action actions[],
5881 		 struct rte_flow_error *error)
5882 {
5883 	struct mlx5_priv *priv = dev->data->dev_private;
5884 
5885 	/*
5886 	 * If the device is not started yet, it is not allowed to created a
5887 	 * flow from application. PMD default flows and traffic control flows
5888 	 * are not affected.
5889 	 */
5890 	if (unlikely(!dev->data->dev_started)) {
5891 		DRV_LOG(DEBUG, "port %u is not started when "
5892 			"inserting a flow", dev->data->port_id);
5893 		rte_flow_error_set(error, ENODEV,
5894 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5895 				   NULL,
5896 				   "port not started");
5897 		return NULL;
5898 	}
5899 
5900 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5901 				  attr, items, actions, true, error);
5902 }
5903 
5904 /**
5905  * Destroy a flow in a list.
5906  *
5907  * @param dev
5908  *   Pointer to Ethernet device.
5909  * @param list
5910  *   Pointer to the Indexed flow list. If this parameter NULL,
5911  *   there is no flow removal from the list. Be noted that as
5912  *   flow is add to the indexed list, memory of the indexed
5913  *   list points to maybe changed as flow destroyed.
5914  * @param[in] flow_idx
5915  *   Index of flow to destroy.
5916  */
5917 static void
5918 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5919 		  uint32_t flow_idx)
5920 {
5921 	struct mlx5_priv *priv = dev->data->dev_private;
5922 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5923 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5924 
5925 	if (!flow)
5926 		return;
5927 	/*
5928 	 * Update RX queue flags only if port is started, otherwise it is
5929 	 * already clean.
5930 	 */
5931 	if (dev->data->dev_started)
5932 		flow_rxq_flags_trim(dev, flow);
5933 	flow_drv_destroy(dev, flow);
5934 	if (list) {
5935 		rte_spinlock_lock(&priv->flow_list_lock);
5936 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5937 			     flow_idx, flow, next);
5938 		rte_spinlock_unlock(&priv->flow_list_lock);
5939 	}
5940 	if (flow->tunnel) {
5941 		struct mlx5_flow_tunnel *tunnel;
5942 
5943 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5944 		RTE_VERIFY(tunnel);
5945 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5946 			mlx5_flow_tunnel_free(dev, tunnel);
5947 	}
5948 	flow_mreg_del_copy_action(dev, flow);
5949 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5950 }
5951 
5952 /**
5953  * Destroy all flows.
5954  *
5955  * @param dev
5956  *   Pointer to Ethernet device.
5957  * @param list
5958  *   Pointer to the Indexed flow list.
5959  * @param active
5960  *   If flushing is called avtively.
5961  */
5962 void
5963 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5964 {
5965 	uint32_t num_flushed = 0;
5966 
5967 	while (*list) {
5968 		flow_list_destroy(dev, list, *list);
5969 		num_flushed++;
5970 	}
5971 	if (active) {
5972 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5973 			dev->data->port_id, num_flushed);
5974 	}
5975 }
5976 
5977 /**
5978  * Stop all default actions for flows.
5979  *
5980  * @param dev
5981  *   Pointer to Ethernet device.
5982  */
5983 void
5984 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5985 {
5986 	flow_mreg_del_default_copy_action(dev);
5987 	flow_rxq_flags_clear(dev);
5988 }
5989 
5990 /**
5991  * Start all default actions for flows.
5992  *
5993  * @param dev
5994  *   Pointer to Ethernet device.
5995  * @return
5996  *   0 on success, a negative errno value otherwise and rte_errno is set.
5997  */
5998 int
5999 mlx5_flow_start_default(struct rte_eth_dev *dev)
6000 {
6001 	struct rte_flow_error error;
6002 
6003 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6004 	return flow_mreg_add_default_copy_action(dev, &error);
6005 }
6006 
6007 /**
6008  * Release key of thread specific flow workspace data.
6009  */
6010 void
6011 flow_release_workspace(void *data)
6012 {
6013 	struct mlx5_flow_workspace *wks = data;
6014 	struct mlx5_flow_workspace *next;
6015 
6016 	while (wks) {
6017 		next = wks->next;
6018 		free(wks->rss_desc.queue);
6019 		free(wks);
6020 		wks = next;
6021 	}
6022 }
6023 
6024 /**
6025  * Get thread specific current flow workspace.
6026  *
6027  * @return pointer to thread specific flow workspace data, NULL on error.
6028  */
6029 struct mlx5_flow_workspace*
6030 mlx5_flow_get_thread_workspace(void)
6031 {
6032 	struct mlx5_flow_workspace *data;
6033 
6034 	data = mlx5_flow_os_get_specific_workspace();
6035 	MLX5_ASSERT(data && data->inuse);
6036 	if (!data || !data->inuse)
6037 		DRV_LOG(ERR, "flow workspace not initialized.");
6038 	return data;
6039 }
6040 
6041 /**
6042  * Allocate and init new flow workspace.
6043  *
6044  * @return pointer to flow workspace data, NULL on error.
6045  */
6046 static struct mlx5_flow_workspace*
6047 flow_alloc_thread_workspace(void)
6048 {
6049 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
6050 
6051 	if (!data) {
6052 		DRV_LOG(ERR, "Failed to allocate flow workspace "
6053 			"memory.");
6054 		return NULL;
6055 	}
6056 	data->rss_desc.queue = calloc(1,
6057 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6058 	if (!data->rss_desc.queue)
6059 		goto err;
6060 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6061 	return data;
6062 err:
6063 	if (data->rss_desc.queue)
6064 		free(data->rss_desc.queue);
6065 	free(data);
6066 	return NULL;
6067 }
6068 
6069 /**
6070  * Get new thread specific flow workspace.
6071  *
6072  * If current workspace inuse, create new one and set as current.
6073  *
6074  * @return pointer to thread specific flow workspace data, NULL on error.
6075  */
6076 static struct mlx5_flow_workspace*
6077 mlx5_flow_push_thread_workspace(void)
6078 {
6079 	struct mlx5_flow_workspace *curr;
6080 	struct mlx5_flow_workspace *data;
6081 
6082 	curr = mlx5_flow_os_get_specific_workspace();
6083 	if (!curr) {
6084 		data = flow_alloc_thread_workspace();
6085 		if (!data)
6086 			return NULL;
6087 	} else if (!curr->inuse) {
6088 		data = curr;
6089 	} else if (curr->next) {
6090 		data = curr->next;
6091 	} else {
6092 		data = flow_alloc_thread_workspace();
6093 		if (!data)
6094 			return NULL;
6095 		curr->next = data;
6096 		data->prev = curr;
6097 	}
6098 	data->inuse = 1;
6099 	data->flow_idx = 0;
6100 	/* Set as current workspace */
6101 	if (mlx5_flow_os_set_specific_workspace(data))
6102 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6103 	return data;
6104 }
6105 
6106 /**
6107  * Close current thread specific flow workspace.
6108  *
6109  * If previous workspace available, set it as current.
6110  *
6111  * @return pointer to thread specific flow workspace data, NULL on error.
6112  */
6113 static void
6114 mlx5_flow_pop_thread_workspace(void)
6115 {
6116 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6117 
6118 	if (!data)
6119 		return;
6120 	if (!data->inuse) {
6121 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6122 		return;
6123 	}
6124 	data->inuse = 0;
6125 	if (!data->prev)
6126 		return;
6127 	if (mlx5_flow_os_set_specific_workspace(data->prev))
6128 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6129 }
6130 
6131 /**
6132  * Verify the flow list is empty
6133  *
6134  * @param dev
6135  *  Pointer to Ethernet device.
6136  *
6137  * @return the number of flows not released.
6138  */
6139 int
6140 mlx5_flow_verify(struct rte_eth_dev *dev)
6141 {
6142 	struct mlx5_priv *priv = dev->data->dev_private;
6143 	struct rte_flow *flow;
6144 	uint32_t idx;
6145 	int ret = 0;
6146 
6147 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6148 		      flow, next) {
6149 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6150 			dev->data->port_id, (void *)flow);
6151 		++ret;
6152 	}
6153 	return ret;
6154 }
6155 
6156 /**
6157  * Enable default hairpin egress flow.
6158  *
6159  * @param dev
6160  *   Pointer to Ethernet device.
6161  * @param queue
6162  *   The queue index.
6163  *
6164  * @return
6165  *   0 on success, a negative errno value otherwise and rte_errno is set.
6166  */
6167 int
6168 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6169 			    uint32_t queue)
6170 {
6171 	struct mlx5_priv *priv = dev->data->dev_private;
6172 	const struct rte_flow_attr attr = {
6173 		.egress = 1,
6174 		.priority = 0,
6175 	};
6176 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6177 		.queue = queue,
6178 	};
6179 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6180 		.queue = UINT32_MAX,
6181 	};
6182 	struct rte_flow_item items[] = {
6183 		{
6184 			.type = (enum rte_flow_item_type)
6185 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6186 			.spec = &queue_spec,
6187 			.last = NULL,
6188 			.mask = &queue_mask,
6189 		},
6190 		{
6191 			.type = RTE_FLOW_ITEM_TYPE_END,
6192 		},
6193 	};
6194 	struct rte_flow_action_jump jump = {
6195 		.group = MLX5_HAIRPIN_TX_TABLE,
6196 	};
6197 	struct rte_flow_action actions[2];
6198 	uint32_t flow_idx;
6199 	struct rte_flow_error error;
6200 
6201 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6202 	actions[0].conf = &jump;
6203 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6204 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6205 				&attr, items, actions, false, &error);
6206 	if (!flow_idx) {
6207 		DRV_LOG(DEBUG,
6208 			"Failed to create ctrl flow: rte_errno(%d),"
6209 			" type(%d), message(%s)",
6210 			rte_errno, error.type,
6211 			error.message ? error.message : " (no stated reason)");
6212 		return -rte_errno;
6213 	}
6214 	return 0;
6215 }
6216 
6217 /**
6218  * Enable a control flow configured from the control plane.
6219  *
6220  * @param dev
6221  *   Pointer to Ethernet device.
6222  * @param eth_spec
6223  *   An Ethernet flow spec to apply.
6224  * @param eth_mask
6225  *   An Ethernet flow mask to apply.
6226  * @param vlan_spec
6227  *   A VLAN flow spec to apply.
6228  * @param vlan_mask
6229  *   A VLAN flow mask to apply.
6230  *
6231  * @return
6232  *   0 on success, a negative errno value otherwise and rte_errno is set.
6233  */
6234 int
6235 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6236 		    struct rte_flow_item_eth *eth_spec,
6237 		    struct rte_flow_item_eth *eth_mask,
6238 		    struct rte_flow_item_vlan *vlan_spec,
6239 		    struct rte_flow_item_vlan *vlan_mask)
6240 {
6241 	struct mlx5_priv *priv = dev->data->dev_private;
6242 	const struct rte_flow_attr attr = {
6243 		.ingress = 1,
6244 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6245 	};
6246 	struct rte_flow_item items[] = {
6247 		{
6248 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6249 			.spec = eth_spec,
6250 			.last = NULL,
6251 			.mask = eth_mask,
6252 		},
6253 		{
6254 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6255 					      RTE_FLOW_ITEM_TYPE_END,
6256 			.spec = vlan_spec,
6257 			.last = NULL,
6258 			.mask = vlan_mask,
6259 		},
6260 		{
6261 			.type = RTE_FLOW_ITEM_TYPE_END,
6262 		},
6263 	};
6264 	uint16_t queue[priv->reta_idx_n];
6265 	struct rte_flow_action_rss action_rss = {
6266 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6267 		.level = 0,
6268 		.types = priv->rss_conf.rss_hf,
6269 		.key_len = priv->rss_conf.rss_key_len,
6270 		.queue_num = priv->reta_idx_n,
6271 		.key = priv->rss_conf.rss_key,
6272 		.queue = queue,
6273 	};
6274 	struct rte_flow_action actions[] = {
6275 		{
6276 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6277 			.conf = &action_rss,
6278 		},
6279 		{
6280 			.type = RTE_FLOW_ACTION_TYPE_END,
6281 		},
6282 	};
6283 	uint32_t flow_idx;
6284 	struct rte_flow_error error;
6285 	unsigned int i;
6286 
6287 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6288 		return 0;
6289 	}
6290 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6291 		action_rss.types = 0;
6292 	for (i = 0; i != priv->reta_idx_n; ++i)
6293 		queue[i] = (*priv->reta_idx)[i];
6294 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6295 				&attr, items, actions, false, &error);
6296 	if (!flow_idx)
6297 		return -rte_errno;
6298 	return 0;
6299 }
6300 
6301 /**
6302  * Enable a flow control configured from the control plane.
6303  *
6304  * @param dev
6305  *   Pointer to Ethernet device.
6306  * @param eth_spec
6307  *   An Ethernet flow spec to apply.
6308  * @param eth_mask
6309  *   An Ethernet flow mask to apply.
6310  *
6311  * @return
6312  *   0 on success, a negative errno value otherwise and rte_errno is set.
6313  */
6314 int
6315 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6316 	       struct rte_flow_item_eth *eth_spec,
6317 	       struct rte_flow_item_eth *eth_mask)
6318 {
6319 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6320 }
6321 
6322 /**
6323  * Create default miss flow rule matching lacp traffic
6324  *
6325  * @param dev
6326  *   Pointer to Ethernet device.
6327  * @param eth_spec
6328  *   An Ethernet flow spec to apply.
6329  *
6330  * @return
6331  *   0 on success, a negative errno value otherwise and rte_errno is set.
6332  */
6333 int
6334 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6335 {
6336 	struct mlx5_priv *priv = dev->data->dev_private;
6337 	/*
6338 	 * The LACP matching is done by only using ether type since using
6339 	 * a multicast dst mac causes kernel to give low priority to this flow.
6340 	 */
6341 	static const struct rte_flow_item_eth lacp_spec = {
6342 		.type = RTE_BE16(0x8809),
6343 	};
6344 	static const struct rte_flow_item_eth lacp_mask = {
6345 		.type = 0xffff,
6346 	};
6347 	const struct rte_flow_attr attr = {
6348 		.ingress = 1,
6349 	};
6350 	struct rte_flow_item items[] = {
6351 		{
6352 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6353 			.spec = &lacp_spec,
6354 			.mask = &lacp_mask,
6355 		},
6356 		{
6357 			.type = RTE_FLOW_ITEM_TYPE_END,
6358 		},
6359 	};
6360 	struct rte_flow_action actions[] = {
6361 		{
6362 			.type = (enum rte_flow_action_type)
6363 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6364 		},
6365 		{
6366 			.type = RTE_FLOW_ACTION_TYPE_END,
6367 		},
6368 	};
6369 	struct rte_flow_error error;
6370 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6371 				&attr, items, actions, false, &error);
6372 
6373 	if (!flow_idx)
6374 		return -rte_errno;
6375 	return 0;
6376 }
6377 
6378 /**
6379  * Destroy a flow.
6380  *
6381  * @see rte_flow_destroy()
6382  * @see rte_flow_ops
6383  */
6384 int
6385 mlx5_flow_destroy(struct rte_eth_dev *dev,
6386 		  struct rte_flow *flow,
6387 		  struct rte_flow_error *error __rte_unused)
6388 {
6389 	struct mlx5_priv *priv = dev->data->dev_private;
6390 
6391 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6392 	return 0;
6393 }
6394 
6395 /**
6396  * Destroy all flows.
6397  *
6398  * @see rte_flow_flush()
6399  * @see rte_flow_ops
6400  */
6401 int
6402 mlx5_flow_flush(struct rte_eth_dev *dev,
6403 		struct rte_flow_error *error __rte_unused)
6404 {
6405 	struct mlx5_priv *priv = dev->data->dev_private;
6406 
6407 	mlx5_flow_list_flush(dev, &priv->flows, false);
6408 	return 0;
6409 }
6410 
6411 /**
6412  * Isolated mode.
6413  *
6414  * @see rte_flow_isolate()
6415  * @see rte_flow_ops
6416  */
6417 int
6418 mlx5_flow_isolate(struct rte_eth_dev *dev,
6419 		  int enable,
6420 		  struct rte_flow_error *error)
6421 {
6422 	struct mlx5_priv *priv = dev->data->dev_private;
6423 
6424 	if (dev->data->dev_started) {
6425 		rte_flow_error_set(error, EBUSY,
6426 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6427 				   NULL,
6428 				   "port must be stopped first");
6429 		return -rte_errno;
6430 	}
6431 	priv->isolated = !!enable;
6432 	if (enable)
6433 		dev->dev_ops = &mlx5_dev_ops_isolate;
6434 	else
6435 		dev->dev_ops = &mlx5_dev_ops;
6436 
6437 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6438 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6439 
6440 	return 0;
6441 }
6442 
6443 /**
6444  * Query a flow.
6445  *
6446  * @see rte_flow_query()
6447  * @see rte_flow_ops
6448  */
6449 static int
6450 flow_drv_query(struct rte_eth_dev *dev,
6451 	       uint32_t flow_idx,
6452 	       const struct rte_flow_action *actions,
6453 	       void *data,
6454 	       struct rte_flow_error *error)
6455 {
6456 	struct mlx5_priv *priv = dev->data->dev_private;
6457 	const struct mlx5_flow_driver_ops *fops;
6458 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6459 					       [MLX5_IPOOL_RTE_FLOW],
6460 					       flow_idx);
6461 	enum mlx5_flow_drv_type ftype;
6462 
6463 	if (!flow) {
6464 		return rte_flow_error_set(error, ENOENT,
6465 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6466 			  NULL,
6467 			  "invalid flow handle");
6468 	}
6469 	ftype = flow->drv_type;
6470 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6471 	fops = flow_get_drv_ops(ftype);
6472 
6473 	return fops->query(dev, flow, actions, data, error);
6474 }
6475 
6476 /**
6477  * Query a flow.
6478  *
6479  * @see rte_flow_query()
6480  * @see rte_flow_ops
6481  */
6482 int
6483 mlx5_flow_query(struct rte_eth_dev *dev,
6484 		struct rte_flow *flow,
6485 		const struct rte_flow_action *actions,
6486 		void *data,
6487 		struct rte_flow_error *error)
6488 {
6489 	int ret;
6490 
6491 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6492 			     error);
6493 	if (ret < 0)
6494 		return ret;
6495 	return 0;
6496 }
6497 
6498 /**
6499  * Manage filter operations.
6500  *
6501  * @param dev
6502  *   Pointer to Ethernet device structure.
6503  * @param filter_type
6504  *   Filter type.
6505  * @param filter_op
6506  *   Operation to perform.
6507  * @param arg
6508  *   Pointer to operation-specific structure.
6509  *
6510  * @return
6511  *   0 on success, a negative errno value otherwise and rte_errno is set.
6512  */
6513 int
6514 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6515 		     enum rte_filter_type filter_type,
6516 		     enum rte_filter_op filter_op,
6517 		     void *arg)
6518 {
6519 	switch (filter_type) {
6520 	case RTE_ETH_FILTER_GENERIC:
6521 		if (filter_op != RTE_ETH_FILTER_GET) {
6522 			rte_errno = EINVAL;
6523 			return -rte_errno;
6524 		}
6525 		*(const void **)arg = &mlx5_flow_ops;
6526 		return 0;
6527 	default:
6528 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6529 			dev->data->port_id, filter_type);
6530 		rte_errno = ENOTSUP;
6531 		return -rte_errno;
6532 	}
6533 	return 0;
6534 }
6535 
6536 /**
6537  * Create the needed meter and suffix tables.
6538  *
6539  * @param[in] dev
6540  *   Pointer to Ethernet device.
6541  * @param[in] fm
6542  *   Pointer to the flow meter.
6543  *
6544  * @return
6545  *   Pointer to table set on success, NULL otherwise.
6546  */
6547 struct mlx5_meter_domains_infos *
6548 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6549 			  const struct mlx5_flow_meter *fm)
6550 {
6551 	const struct mlx5_flow_driver_ops *fops;
6552 
6553 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6554 	return fops->create_mtr_tbls(dev, fm);
6555 }
6556 
6557 /**
6558  * Destroy the meter table set.
6559  *
6560  * @param[in] dev
6561  *   Pointer to Ethernet device.
6562  * @param[in] tbl
6563  *   Pointer to the meter table set.
6564  *
6565  * @return
6566  *   0 on success.
6567  */
6568 int
6569 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6570 			   struct mlx5_meter_domains_infos *tbls)
6571 {
6572 	const struct mlx5_flow_driver_ops *fops;
6573 
6574 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6575 	return fops->destroy_mtr_tbls(dev, tbls);
6576 }
6577 
6578 /**
6579  * Create policer rules.
6580  *
6581  * @param[in] dev
6582  *   Pointer to Ethernet device.
6583  * @param[in] fm
6584  *   Pointer to flow meter structure.
6585  * @param[in] attr
6586  *   Pointer to flow attributes.
6587  *
6588  * @return
6589  *   0 on success, -1 otherwise.
6590  */
6591 int
6592 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6593 			       struct mlx5_flow_meter *fm,
6594 			       const struct rte_flow_attr *attr)
6595 {
6596 	const struct mlx5_flow_driver_ops *fops;
6597 
6598 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6599 	return fops->create_policer_rules(dev, fm, attr);
6600 }
6601 
6602 /**
6603  * Destroy policer rules.
6604  *
6605  * @param[in] fm
6606  *   Pointer to flow meter structure.
6607  * @param[in] attr
6608  *   Pointer to flow attributes.
6609  *
6610  * @return
6611  *   0 on success, -1 otherwise.
6612  */
6613 int
6614 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6615 				struct mlx5_flow_meter *fm,
6616 				const struct rte_flow_attr *attr)
6617 {
6618 	const struct mlx5_flow_driver_ops *fops;
6619 
6620 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6621 	return fops->destroy_policer_rules(dev, fm, attr);
6622 }
6623 
6624 /**
6625  * Allocate a counter.
6626  *
6627  * @param[in] dev
6628  *   Pointer to Ethernet device structure.
6629  *
6630  * @return
6631  *   Index to allocated counter  on success, 0 otherwise.
6632  */
6633 uint32_t
6634 mlx5_counter_alloc(struct rte_eth_dev *dev)
6635 {
6636 	const struct mlx5_flow_driver_ops *fops;
6637 	struct rte_flow_attr attr = { .transfer = 0 };
6638 
6639 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6640 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6641 		return fops->counter_alloc(dev);
6642 	}
6643 	DRV_LOG(ERR,
6644 		"port %u counter allocate is not supported.",
6645 		 dev->data->port_id);
6646 	return 0;
6647 }
6648 
6649 /**
6650  * Free a counter.
6651  *
6652  * @param[in] dev
6653  *   Pointer to Ethernet device structure.
6654  * @param[in] cnt
6655  *   Index to counter to be free.
6656  */
6657 void
6658 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6659 {
6660 	const struct mlx5_flow_driver_ops *fops;
6661 	struct rte_flow_attr attr = { .transfer = 0 };
6662 
6663 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6664 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6665 		fops->counter_free(dev, cnt);
6666 		return;
6667 	}
6668 	DRV_LOG(ERR,
6669 		"port %u counter free is not supported.",
6670 		 dev->data->port_id);
6671 }
6672 
6673 /**
6674  * Query counter statistics.
6675  *
6676  * @param[in] dev
6677  *   Pointer to Ethernet device structure.
6678  * @param[in] cnt
6679  *   Index to counter to query.
6680  * @param[in] clear
6681  *   Set to clear counter statistics.
6682  * @param[out] pkts
6683  *   The counter hits packets number to save.
6684  * @param[out] bytes
6685  *   The counter hits bytes number to save.
6686  *
6687  * @return
6688  *   0 on success, a negative errno value otherwise.
6689  */
6690 int
6691 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6692 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6693 {
6694 	const struct mlx5_flow_driver_ops *fops;
6695 	struct rte_flow_attr attr = { .transfer = 0 };
6696 
6697 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6698 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6699 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6700 	}
6701 	DRV_LOG(ERR,
6702 		"port %u counter query is not supported.",
6703 		 dev->data->port_id);
6704 	return -ENOTSUP;
6705 }
6706 
6707 /**
6708  * Allocate a new memory for the counter values wrapped by all the needed
6709  * management.
6710  *
6711  * @param[in] sh
6712  *   Pointer to mlx5_dev_ctx_shared object.
6713  *
6714  * @return
6715  *   0 on success, a negative errno value otherwise.
6716  */
6717 static int
6718 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6719 {
6720 	struct mlx5_devx_mkey_attr mkey_attr;
6721 	struct mlx5_counter_stats_mem_mng *mem_mng;
6722 	volatile struct flow_counter_stats *raw_data;
6723 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6724 	int size = (sizeof(struct flow_counter_stats) *
6725 			MLX5_COUNTERS_PER_POOL +
6726 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6727 			sizeof(struct mlx5_counter_stats_mem_mng);
6728 	size_t pgsize = rte_mem_page_size();
6729 	uint8_t *mem;
6730 	int i;
6731 
6732 	if (pgsize == (size_t)-1) {
6733 		DRV_LOG(ERR, "Failed to get mem page size");
6734 		rte_errno = ENOMEM;
6735 		return -ENOMEM;
6736 	}
6737 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6738 	if (!mem) {
6739 		rte_errno = ENOMEM;
6740 		return -ENOMEM;
6741 	}
6742 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6743 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6744 	mem_mng->umem = mlx5_os_umem_reg(sh->ctx, mem, size,
6745 						 IBV_ACCESS_LOCAL_WRITE);
6746 	if (!mem_mng->umem) {
6747 		rte_errno = errno;
6748 		mlx5_free(mem);
6749 		return -rte_errno;
6750 	}
6751 	mkey_attr.addr = (uintptr_t)mem;
6752 	mkey_attr.size = size;
6753 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6754 	mkey_attr.pd = sh->pdn;
6755 	mkey_attr.log_entity_size = 0;
6756 	mkey_attr.pg_access = 0;
6757 	mkey_attr.klm_array = NULL;
6758 	mkey_attr.klm_num = 0;
6759 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6760 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6761 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6762 	if (!mem_mng->dm) {
6763 		mlx5_os_umem_dereg(mem_mng->umem);
6764 		rte_errno = errno;
6765 		mlx5_free(mem);
6766 		return -rte_errno;
6767 	}
6768 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6769 	raw_data = (volatile struct flow_counter_stats *)mem;
6770 	for (i = 0; i < raws_n; ++i) {
6771 		mem_mng->raws[i].mem_mng = mem_mng;
6772 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6773 	}
6774 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6775 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6776 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6777 				 next);
6778 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6779 	sh->cmng.mem_mng = mem_mng;
6780 	return 0;
6781 }
6782 
6783 /**
6784  * Set the statistic memory to the new counter pool.
6785  *
6786  * @param[in] sh
6787  *   Pointer to mlx5_dev_ctx_shared object.
6788  * @param[in] pool
6789  *   Pointer to the pool to set the statistic memory.
6790  *
6791  * @return
6792  *   0 on success, a negative errno value otherwise.
6793  */
6794 static int
6795 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6796 			       struct mlx5_flow_counter_pool *pool)
6797 {
6798 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6799 	/* Resize statistic memory once used out. */
6800 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6801 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6802 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6803 		return -1;
6804 	}
6805 	rte_spinlock_lock(&pool->sl);
6806 	pool->raw = cmng->mem_mng->raws + pool->index %
6807 		    MLX5_CNT_CONTAINER_RESIZE;
6808 	rte_spinlock_unlock(&pool->sl);
6809 	pool->raw_hw = NULL;
6810 	return 0;
6811 }
6812 
6813 #define MLX5_POOL_QUERY_FREQ_US 1000000
6814 
6815 /**
6816  * Set the periodic procedure for triggering asynchronous batch queries for all
6817  * the counter pools.
6818  *
6819  * @param[in] sh
6820  *   Pointer to mlx5_dev_ctx_shared object.
6821  */
6822 void
6823 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6824 {
6825 	uint32_t pools_n, us;
6826 
6827 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6828 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6829 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6830 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6831 		sh->cmng.query_thread_on = 0;
6832 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6833 	} else {
6834 		sh->cmng.query_thread_on = 1;
6835 	}
6836 }
6837 
6838 /**
6839  * The periodic procedure for triggering asynchronous batch queries for all the
6840  * counter pools. This function is probably called by the host thread.
6841  *
6842  * @param[in] arg
6843  *   The parameter for the alarm process.
6844  */
6845 void
6846 mlx5_flow_query_alarm(void *arg)
6847 {
6848 	struct mlx5_dev_ctx_shared *sh = arg;
6849 	int ret;
6850 	uint16_t pool_index = sh->cmng.pool_index;
6851 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6852 	struct mlx5_flow_counter_pool *pool;
6853 	uint16_t n_valid;
6854 
6855 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6856 		goto set_alarm;
6857 	rte_spinlock_lock(&cmng->pool_update_sl);
6858 	pool = cmng->pools[pool_index];
6859 	n_valid = cmng->n_valid;
6860 	rte_spinlock_unlock(&cmng->pool_update_sl);
6861 	/* Set the statistic memory to the new created pool. */
6862 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6863 		goto set_alarm;
6864 	if (pool->raw_hw)
6865 		/* There is a pool query in progress. */
6866 		goto set_alarm;
6867 	pool->raw_hw =
6868 		LIST_FIRST(&sh->cmng.free_stat_raws);
6869 	if (!pool->raw_hw)
6870 		/* No free counter statistics raw memory. */
6871 		goto set_alarm;
6872 	/*
6873 	 * Identify the counters released between query trigger and query
6874 	 * handle more efficiently. The counter released in this gap period
6875 	 * should wait for a new round of query as the new arrived packets
6876 	 * will not be taken into account.
6877 	 */
6878 	pool->query_gen++;
6879 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6880 					       MLX5_COUNTERS_PER_POOL,
6881 					       NULL, NULL,
6882 					       pool->raw_hw->mem_mng->dm->id,
6883 					       (void *)(uintptr_t)
6884 					       pool->raw_hw->data,
6885 					       sh->devx_comp,
6886 					       (uint64_t)(uintptr_t)pool);
6887 	if (ret) {
6888 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6889 			" %d", pool->min_dcs->id);
6890 		pool->raw_hw = NULL;
6891 		goto set_alarm;
6892 	}
6893 	LIST_REMOVE(pool->raw_hw, next);
6894 	sh->cmng.pending_queries++;
6895 	pool_index++;
6896 	if (pool_index >= n_valid)
6897 		pool_index = 0;
6898 set_alarm:
6899 	sh->cmng.pool_index = pool_index;
6900 	mlx5_set_query_alarm(sh);
6901 }
6902 
6903 /**
6904  * Check and callback event for new aged flow in the counter pool
6905  *
6906  * @param[in] sh
6907  *   Pointer to mlx5_dev_ctx_shared object.
6908  * @param[in] pool
6909  *   Pointer to Current counter pool.
6910  */
6911 static void
6912 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6913 		   struct mlx5_flow_counter_pool *pool)
6914 {
6915 	struct mlx5_priv *priv;
6916 	struct mlx5_flow_counter *cnt;
6917 	struct mlx5_age_info *age_info;
6918 	struct mlx5_age_param *age_param;
6919 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6920 	struct mlx5_counter_stats_raw *prev = pool->raw;
6921 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6922 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6923 	uint16_t expected = AGE_CANDIDATE;
6924 	uint32_t i;
6925 
6926 	pool->time_of_last_age_check = curr_time;
6927 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6928 		cnt = MLX5_POOL_GET_CNT(pool, i);
6929 		age_param = MLX5_CNT_TO_AGE(cnt);
6930 		if (__atomic_load_n(&age_param->state,
6931 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6932 			continue;
6933 		if (cur->data[i].hits != prev->data[i].hits) {
6934 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6935 					 __ATOMIC_RELAXED);
6936 			continue;
6937 		}
6938 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6939 				       time_delta,
6940 				       __ATOMIC_RELAXED) <= age_param->timeout)
6941 			continue;
6942 		/**
6943 		 * Hold the lock first, or if between the
6944 		 * state AGE_TMOUT and tailq operation the
6945 		 * release happened, the release procedure
6946 		 * may delete a non-existent tailq node.
6947 		 */
6948 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6949 		age_info = GET_PORT_AGE_INFO(priv);
6950 		rte_spinlock_lock(&age_info->aged_sl);
6951 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6952 						AGE_TMOUT, false,
6953 						__ATOMIC_RELAXED,
6954 						__ATOMIC_RELAXED)) {
6955 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6956 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6957 		}
6958 		rte_spinlock_unlock(&age_info->aged_sl);
6959 	}
6960 	mlx5_age_event_prepare(sh);
6961 }
6962 
6963 /**
6964  * Handler for the HW respond about ready values from an asynchronous batch
6965  * query. This function is probably called by the host thread.
6966  *
6967  * @param[in] sh
6968  *   The pointer to the shared device context.
6969  * @param[in] async_id
6970  *   The Devx async ID.
6971  * @param[in] status
6972  *   The status of the completion.
6973  */
6974 void
6975 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6976 				  uint64_t async_id, int status)
6977 {
6978 	struct mlx5_flow_counter_pool *pool =
6979 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6980 	struct mlx5_counter_stats_raw *raw_to_free;
6981 	uint8_t query_gen = pool->query_gen ^ 1;
6982 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6983 	enum mlx5_counter_type cnt_type =
6984 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6985 				MLX5_COUNTER_TYPE_ORIGIN;
6986 
6987 	if (unlikely(status)) {
6988 		raw_to_free = pool->raw_hw;
6989 	} else {
6990 		raw_to_free = pool->raw;
6991 		if (pool->is_aged)
6992 			mlx5_flow_aging_check(sh, pool);
6993 		rte_spinlock_lock(&pool->sl);
6994 		pool->raw = pool->raw_hw;
6995 		rte_spinlock_unlock(&pool->sl);
6996 		/* Be sure the new raw counters data is updated in memory. */
6997 		rte_io_wmb();
6998 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6999 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7000 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7001 				     &pool->counters[query_gen], next);
7002 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7003 		}
7004 	}
7005 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7006 	pool->raw_hw = NULL;
7007 	sh->cmng.pending_queries--;
7008 }
7009 
7010 static int
7011 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7012 		    const struct flow_grp_info *grp_info,
7013 		    struct rte_flow_error *error)
7014 {
7015 	if (grp_info->transfer && grp_info->external &&
7016 	    grp_info->fdb_def_rule) {
7017 		if (group == UINT32_MAX)
7018 			return rte_flow_error_set
7019 						(error, EINVAL,
7020 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7021 						 NULL,
7022 						 "group index not supported");
7023 		*table = group + 1;
7024 	} else {
7025 		*table = group;
7026 	}
7027 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7028 	return 0;
7029 }
7030 
7031 /**
7032  * Translate the rte_flow group index to HW table value.
7033  *
7034  * If tunnel offload is disabled, all group ids converted to flow table
7035  * id using the standard method.
7036  * If tunnel offload is enabled, group id can be converted using the
7037  * standard or tunnel conversion method. Group conversion method
7038  * selection depends on flags in `grp_info` parameter:
7039  * - Internal (grp_info.external == 0) groups conversion uses the
7040  *   standard method.
7041  * - Group ids in JUMP action converted with the tunnel conversion.
7042  * - Group id in rule attribute conversion depends on a rule type and
7043  *   group id value:
7044  *   ** non zero group attributes converted with the tunnel method
7045  *   ** zero group attribute in non-tunnel rule is converted using the
7046  *      standard method - there's only one root table
7047  *   ** zero group attribute in steer tunnel rule is converted with the
7048  *      standard method - single root table
7049  *   ** zero group attribute in match tunnel rule is a special OvS
7050  *      case: that value is used for portability reasons. That group
7051  *      id is converted with the tunnel conversion method.
7052  *
7053  * @param[in] dev
7054  *   Port device
7055  * @param[in] tunnel
7056  *   PMD tunnel offload object
7057  * @param[in] group
7058  *   rte_flow group index value.
7059  * @param[out] table
7060  *   HW table value.
7061  * @param[in] grp_info
7062  *   flags used for conversion
7063  * @param[out] error
7064  *   Pointer to error structure.
7065  *
7066  * @return
7067  *   0 on success, a negative errno value otherwise and rte_errno is set.
7068  */
7069 int
7070 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7071 			 const struct mlx5_flow_tunnel *tunnel,
7072 			 uint32_t group, uint32_t *table,
7073 			 const struct flow_grp_info *grp_info,
7074 			 struct rte_flow_error *error)
7075 {
7076 	int ret;
7077 	bool standard_translation;
7078 
7079 	if (!grp_info->skip_scale && grp_info->external &&
7080 	    group < MLX5_MAX_TABLES_EXTERNAL)
7081 		group *= MLX5_FLOW_TABLE_FACTOR;
7082 	if (is_tunnel_offload_active(dev)) {
7083 		standard_translation = !grp_info->external ||
7084 					grp_info->std_tbl_fix;
7085 	} else {
7086 		standard_translation = true;
7087 	}
7088 	DRV_LOG(DEBUG,
7089 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
7090 		dev->data->port_id, group, grp_info->transfer,
7091 		grp_info->external, grp_info->fdb_def_rule,
7092 		standard_translation ? "STANDARD" : "TUNNEL");
7093 	if (standard_translation)
7094 		ret = flow_group_to_table(dev->data->port_id, group, table,
7095 					  grp_info, error);
7096 	else
7097 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7098 						      table, error);
7099 
7100 	return ret;
7101 }
7102 
7103 /**
7104  * Discover availability of metadata reg_c's.
7105  *
7106  * Iteratively use test flows to check availability.
7107  *
7108  * @param[in] dev
7109  *   Pointer to the Ethernet device structure.
7110  *
7111  * @return
7112  *   0 on success, a negative errno value otherwise and rte_errno is set.
7113  */
7114 int
7115 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7116 {
7117 	struct mlx5_priv *priv = dev->data->dev_private;
7118 	struct mlx5_dev_config *config = &priv->config;
7119 	enum modify_reg idx;
7120 	int n = 0;
7121 
7122 	/* reg_c[0] and reg_c[1] are reserved. */
7123 	config->flow_mreg_c[n++] = REG_C_0;
7124 	config->flow_mreg_c[n++] = REG_C_1;
7125 	/* Discover availability of other reg_c's. */
7126 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7127 		struct rte_flow_attr attr = {
7128 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7129 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7130 			.ingress = 1,
7131 		};
7132 		struct rte_flow_item items[] = {
7133 			[0] = {
7134 				.type = RTE_FLOW_ITEM_TYPE_END,
7135 			},
7136 		};
7137 		struct rte_flow_action actions[] = {
7138 			[0] = {
7139 				.type = (enum rte_flow_action_type)
7140 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7141 				.conf = &(struct mlx5_flow_action_copy_mreg){
7142 					.src = REG_C_1,
7143 					.dst = idx,
7144 				},
7145 			},
7146 			[1] = {
7147 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7148 				.conf = &(struct rte_flow_action_jump){
7149 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7150 				},
7151 			},
7152 			[2] = {
7153 				.type = RTE_FLOW_ACTION_TYPE_END,
7154 			},
7155 		};
7156 		uint32_t flow_idx;
7157 		struct rte_flow *flow;
7158 		struct rte_flow_error error;
7159 
7160 		if (!config->dv_flow_en)
7161 			break;
7162 		/* Create internal flow, validation skips copy action. */
7163 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7164 					    actions, false, &error);
7165 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7166 				      flow_idx);
7167 		if (!flow)
7168 			continue;
7169 		config->flow_mreg_c[n++] = idx;
7170 		flow_list_destroy(dev, NULL, flow_idx);
7171 	}
7172 	for (; n < MLX5_MREG_C_NUM; ++n)
7173 		config->flow_mreg_c[n] = REG_NON;
7174 	return 0;
7175 }
7176 
7177 /**
7178  * Dump flow raw hw data to file
7179  *
7180  * @param[in] dev
7181  *    The pointer to Ethernet device.
7182  * @param[in] file
7183  *   A pointer to a file for output.
7184  * @param[out] error
7185  *   Perform verbose error reporting if not NULL. PMDs initialize this
7186  *   structure in case of error only.
7187  * @return
7188  *   0 on success, a nagative value otherwise.
7189  */
7190 int
7191 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7192 		   FILE *file,
7193 		   struct rte_flow_error *error __rte_unused)
7194 {
7195 	struct mlx5_priv *priv = dev->data->dev_private;
7196 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7197 
7198 	if (!priv->config.dv_flow_en) {
7199 		if (fputs("device dv flow disabled\n", file) <= 0)
7200 			return -errno;
7201 		return -ENOTSUP;
7202 	}
7203 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7204 				       sh->tx_domain, file);
7205 }
7206 
7207 /**
7208  * Get aged-out flows.
7209  *
7210  * @param[in] dev
7211  *   Pointer to the Ethernet device structure.
7212  * @param[in] context
7213  *   The address of an array of pointers to the aged-out flows contexts.
7214  * @param[in] nb_countexts
7215  *   The length of context array pointers.
7216  * @param[out] error
7217  *   Perform verbose error reporting if not NULL. Initialized in case of
7218  *   error only.
7219  *
7220  * @return
7221  *   how many contexts get in success, otherwise negative errno value.
7222  *   if nb_contexts is 0, return the amount of all aged contexts.
7223  *   if nb_contexts is not 0 , return the amount of aged flows reported
7224  *   in the context array.
7225  */
7226 int
7227 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7228 			uint32_t nb_contexts, struct rte_flow_error *error)
7229 {
7230 	const struct mlx5_flow_driver_ops *fops;
7231 	struct rte_flow_attr attr = { .transfer = 0 };
7232 
7233 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7234 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7235 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7236 						    error);
7237 	}
7238 	DRV_LOG(ERR,
7239 		"port %u get aged flows is not supported.",
7240 		 dev->data->port_id);
7241 	return -ENOTSUP;
7242 }
7243 
7244 /* Wrapper for driver action_validate op callback */
7245 static int
7246 flow_drv_action_validate(struct rte_eth_dev *dev,
7247 			 const struct rte_flow_shared_action_conf *conf,
7248 			 const struct rte_flow_action *action,
7249 			 const struct mlx5_flow_driver_ops *fops,
7250 			 struct rte_flow_error *error)
7251 {
7252 	static const char err_msg[] = "shared action validation unsupported";
7253 
7254 	if (!fops->action_validate) {
7255 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7256 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7257 				   NULL, err_msg);
7258 		return -rte_errno;
7259 	}
7260 	return fops->action_validate(dev, conf, action, error);
7261 }
7262 
7263 /**
7264  * Destroys the shared action by handle.
7265  *
7266  * @param dev
7267  *   Pointer to Ethernet device structure.
7268  * @param[in] action
7269  *   Handle for the shared action to be destroyed.
7270  * @param[out] error
7271  *   Perform verbose error reporting if not NULL. PMDs initialize this
7272  *   structure in case of error only.
7273  *
7274  * @return
7275  *   0 on success, a negative errno value otherwise and rte_errno is set.
7276  *
7277  * @note: wrapper for driver action_create op callback.
7278  */
7279 static int
7280 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7281 			   struct rte_flow_shared_action *action,
7282 			   struct rte_flow_error *error)
7283 {
7284 	static const char err_msg[] = "shared action destruction unsupported";
7285 	struct rte_flow_attr attr = { .transfer = 0 };
7286 	const struct mlx5_flow_driver_ops *fops =
7287 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7288 
7289 	if (!fops->action_destroy) {
7290 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7291 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7292 				   NULL, err_msg);
7293 		return -rte_errno;
7294 	}
7295 	return fops->action_destroy(dev, action, error);
7296 }
7297 
7298 /* Wrapper for driver action_destroy op callback */
7299 static int
7300 flow_drv_action_update(struct rte_eth_dev *dev,
7301 		       struct rte_flow_shared_action *action,
7302 		       const void *action_conf,
7303 		       const struct mlx5_flow_driver_ops *fops,
7304 		       struct rte_flow_error *error)
7305 {
7306 	static const char err_msg[] = "shared action update unsupported";
7307 
7308 	if (!fops->action_update) {
7309 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7310 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7311 				   NULL, err_msg);
7312 		return -rte_errno;
7313 	}
7314 	return fops->action_update(dev, action, action_conf, error);
7315 }
7316 
7317 /* Wrapper for driver action_destroy op callback */
7318 static int
7319 flow_drv_action_query(struct rte_eth_dev *dev,
7320 		      const struct rte_flow_shared_action *action,
7321 		      void *data,
7322 		      const struct mlx5_flow_driver_ops *fops,
7323 		      struct rte_flow_error *error)
7324 {
7325 	static const char err_msg[] = "shared action query unsupported";
7326 
7327 	if (!fops->action_query) {
7328 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7329 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7330 				   NULL, err_msg);
7331 		return -rte_errno;
7332 	}
7333 	return fops->action_query(dev, action, data, error);
7334 }
7335 
7336 /**
7337  * Create shared action for reuse in multiple flow rules.
7338  *
7339  * @param dev
7340  *   Pointer to Ethernet device structure.
7341  * @param[in] action
7342  *   Action configuration for shared action creation.
7343  * @param[out] error
7344  *   Perform verbose error reporting if not NULL. PMDs initialize this
7345  *   structure in case of error only.
7346  * @return
7347  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7348  */
7349 static struct rte_flow_shared_action *
7350 mlx5_shared_action_create(struct rte_eth_dev *dev,
7351 			  const struct rte_flow_shared_action_conf *conf,
7352 			  const struct rte_flow_action *action,
7353 			  struct rte_flow_error *error)
7354 {
7355 	static const char err_msg[] = "shared action creation unsupported";
7356 	struct rte_flow_attr attr = { .transfer = 0 };
7357 	const struct mlx5_flow_driver_ops *fops =
7358 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7359 
7360 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7361 		return NULL;
7362 	if (!fops->action_create) {
7363 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7364 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7365 				   NULL, err_msg);
7366 		return NULL;
7367 	}
7368 	return fops->action_create(dev, conf, action, error);
7369 }
7370 
7371 /**
7372  * Updates inplace the shared action configuration pointed by *action* handle
7373  * with the configuration provided as *action* argument.
7374  * The update of the shared action configuration effects all flow rules reusing
7375  * the action via handle.
7376  *
7377  * @param dev
7378  *   Pointer to Ethernet device structure.
7379  * @param[in] shared_action
7380  *   Handle for the shared action to be updated.
7381  * @param[in] action
7382  *   Action specification used to modify the action pointed by handle.
7383  *   *action* should be of same type with the action pointed by the *action*
7384  *   handle argument, otherwise considered as invalid.
7385  * @param[out] error
7386  *   Perform verbose error reporting if not NULL. PMDs initialize this
7387  *   structure in case of error only.
7388  *
7389  * @return
7390  *   0 on success, a negative errno value otherwise and rte_errno is set.
7391  */
7392 static int
7393 mlx5_shared_action_update(struct rte_eth_dev *dev,
7394 		struct rte_flow_shared_action *shared_action,
7395 		const struct rte_flow_action *action,
7396 		struct rte_flow_error *error)
7397 {
7398 	struct rte_flow_attr attr = { .transfer = 0 };
7399 	const struct mlx5_flow_driver_ops *fops =
7400 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7401 	int ret;
7402 
7403 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7404 	if (ret)
7405 		return ret;
7406 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7407 				      error);
7408 }
7409 
7410 /**
7411  * Query the shared action by handle.
7412  *
7413  * This function allows retrieving action-specific data such as counters.
7414  * Data is gathered by special action which may be present/referenced in
7415  * more than one flow rule definition.
7416  *
7417  * \see RTE_FLOW_ACTION_TYPE_COUNT
7418  *
7419  * @param dev
7420  *   Pointer to Ethernet device structure.
7421  * @param[in] action
7422  *   Handle for the shared action to query.
7423  * @param[in, out] data
7424  *   Pointer to storage for the associated query data type.
7425  * @param[out] error
7426  *   Perform verbose error reporting if not NULL. PMDs initialize this
7427  *   structure in case of error only.
7428  *
7429  * @return
7430  *   0 on success, a negative errno value otherwise and rte_errno is set.
7431  */
7432 static int
7433 mlx5_shared_action_query(struct rte_eth_dev *dev,
7434 			 const struct rte_flow_shared_action *action,
7435 			 void *data,
7436 			 struct rte_flow_error *error)
7437 {
7438 	struct rte_flow_attr attr = { .transfer = 0 };
7439 	const struct mlx5_flow_driver_ops *fops =
7440 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7441 
7442 	return flow_drv_action_query(dev, action, data, fops, error);
7443 }
7444 
7445 /**
7446  * Destroy all shared actions.
7447  *
7448  * @param dev
7449  *   Pointer to Ethernet device.
7450  *
7451  * @return
7452  *   0 on success, a negative errno value otherwise and rte_errno is set.
7453  */
7454 int
7455 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7456 {
7457 	struct rte_flow_error error;
7458 	struct mlx5_priv *priv = dev->data->dev_private;
7459 	struct mlx5_shared_action_rss *shared_rss;
7460 	int ret = 0;
7461 	uint32_t idx;
7462 
7463 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7464 		      priv->rss_shared_actions, idx, shared_rss, next) {
7465 		ret |= mlx5_shared_action_destroy(dev,
7466 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7467 	}
7468 	return ret;
7469 }
7470 
7471 #ifndef HAVE_MLX5DV_DR
7472 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7473 #else
7474 #define MLX5_DOMAIN_SYNC_FLOW \
7475 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7476 #endif
7477 
7478 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7479 {
7480 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7481 	const struct mlx5_flow_driver_ops *fops;
7482 	int ret;
7483 	struct rte_flow_attr attr = { .transfer = 0 };
7484 
7485 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7486 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7487 	if (ret > 0)
7488 		ret = -ret;
7489 	return ret;
7490 }
7491 
7492 /**
7493  * tunnel offload functionalilty is defined for DV environment only
7494  */
7495 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7496 __extension__
7497 union tunnel_offload_mark {
7498 	uint32_t val;
7499 	struct {
7500 		uint32_t app_reserve:8;
7501 		uint32_t table_id:15;
7502 		uint32_t transfer:1;
7503 		uint32_t _unused_:8;
7504 	};
7505 };
7506 
7507 static bool
7508 mlx5_access_tunnel_offload_db
7509 	(struct rte_eth_dev *dev,
7510 	 bool (*match)(struct rte_eth_dev *,
7511 		       struct mlx5_flow_tunnel *, const void *),
7512 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7513 	 void (*miss)(struct rte_eth_dev *, void *),
7514 	 void *ctx, bool lock_op);
7515 
7516 static int
7517 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
7518 			     struct rte_flow *flow,
7519 			     const struct rte_flow_attr *attr,
7520 			     const struct rte_flow_action *app_actions,
7521 			     uint32_t flow_idx,
7522 			     struct tunnel_default_miss_ctx *ctx,
7523 			     struct rte_flow_error *error)
7524 {
7525 	struct mlx5_priv *priv = dev->data->dev_private;
7526 	struct mlx5_flow *dev_flow;
7527 	struct rte_flow_attr miss_attr = *attr;
7528 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
7529 	const struct rte_flow_item miss_items[2] = {
7530 		{
7531 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7532 			.spec = NULL,
7533 			.last = NULL,
7534 			.mask = NULL
7535 		},
7536 		{
7537 			.type = RTE_FLOW_ITEM_TYPE_END,
7538 			.spec = NULL,
7539 			.last = NULL,
7540 			.mask = NULL
7541 		}
7542 	};
7543 	union tunnel_offload_mark mark_id;
7544 	struct rte_flow_action_mark miss_mark;
7545 	struct rte_flow_action miss_actions[3] = {
7546 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
7547 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
7548 	};
7549 	const struct rte_flow_action_jump *jump_data;
7550 	uint32_t i, flow_table = 0; /* prevent compilation warning */
7551 	struct flow_grp_info grp_info = {
7552 		.external = 1,
7553 		.transfer = attr->transfer,
7554 		.fdb_def_rule = !!priv->fdb_def_rule,
7555 		.std_tbl_fix = 0,
7556 	};
7557 	int ret;
7558 
7559 	if (!attr->transfer) {
7560 		uint32_t q_size;
7561 
7562 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
7563 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
7564 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
7565 					 0, SOCKET_ID_ANY);
7566 		if (!ctx->queue)
7567 			return rte_flow_error_set
7568 				(error, ENOMEM,
7569 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7570 				NULL, "invalid default miss RSS");
7571 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7572 		ctx->action_rss.level = 0,
7573 		ctx->action_rss.types = priv->rss_conf.rss_hf,
7574 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
7575 		ctx->action_rss.queue_num = priv->reta_idx_n,
7576 		ctx->action_rss.key = priv->rss_conf.rss_key,
7577 		ctx->action_rss.queue = ctx->queue;
7578 		if (!priv->reta_idx_n || !priv->rxqs_n)
7579 			return rte_flow_error_set
7580 				(error, EINVAL,
7581 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7582 				NULL, "invalid port configuration");
7583 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
7584 			ctx->action_rss.types = 0;
7585 		for (i = 0; i != priv->reta_idx_n; ++i)
7586 			ctx->queue[i] = (*priv->reta_idx)[i];
7587 	} else {
7588 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
7589 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
7590 	}
7591 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
7592 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
7593 	jump_data = app_actions->conf;
7594 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
7595 	miss_attr.group = jump_data->group;
7596 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
7597 				       &flow_table, &grp_info, error);
7598 	if (ret)
7599 		return rte_flow_error_set(error, EINVAL,
7600 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
7601 					  NULL, "invalid tunnel id");
7602 	mark_id.app_reserve = 0;
7603 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
7604 	mark_id.transfer = !!attr->transfer;
7605 	mark_id._unused_ = 0;
7606 	miss_mark.id = mark_id.val;
7607 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
7608 				    miss_items, miss_actions, flow_idx, error);
7609 	if (!dev_flow)
7610 		return -rte_errno;
7611 	dev_flow->flow = flow;
7612 	dev_flow->external = true;
7613 	dev_flow->tunnel = tunnel;
7614 	/* Subflow object was created, we must include one in the list. */
7615 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
7616 		      dev_flow->handle, next);
7617 	DRV_LOG(DEBUG,
7618 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
7619 		dev->data->port_id, tunnel->app_tunnel.type,
7620 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
7621 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
7622 				  miss_actions, error);
7623 	if (!ret)
7624 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
7625 						  error);
7626 
7627 	return ret;
7628 }
7629 
7630 static const struct mlx5_flow_tbl_data_entry  *
7631 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7632 {
7633 	struct mlx5_priv *priv = dev->data->dev_private;
7634 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7635 	struct mlx5_hlist_entry *he;
7636 	union tunnel_offload_mark mbits = { .val = mark };
7637 	union mlx5_flow_tbl_key table_key = {
7638 		{
7639 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7640 			.dummy = 0,
7641 			.domain = !!mbits.transfer,
7642 			.direction = 0,
7643 		}
7644 	};
7645 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
7646 	return he ?
7647 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7648 }
7649 
7650 static void
7651 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
7652 				   struct mlx5_hlist_entry *entry)
7653 {
7654 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7655 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7656 
7657 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7658 			tunnel_flow_tbl_to_id(tte->flow_table));
7659 	mlx5_free(tte);
7660 }
7661 
7662 static int
7663 mlx5_flow_tunnel_grp2tbl_match_cb(struct mlx5_hlist *list __rte_unused,
7664 				  struct mlx5_hlist_entry *entry,
7665 				  uint64_t key, void *cb_ctx __rte_unused)
7666 {
7667 	union tunnel_tbl_key tbl = {
7668 		.val = key,
7669 	};
7670 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
7671 
7672 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
7673 }
7674 
7675 static struct mlx5_hlist_entry *
7676 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list, uint64_t key,
7677 				   void *ctx __rte_unused)
7678 {
7679 	struct mlx5_dev_ctx_shared *sh = list->ctx;
7680 	struct tunnel_tbl_entry *tte;
7681 	union tunnel_tbl_key tbl = {
7682 		.val = key,
7683 	};
7684 
7685 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7686 			  sizeof(*tte), 0,
7687 			  SOCKET_ID_ANY);
7688 	if (!tte)
7689 		goto err;
7690 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7691 			  &tte->flow_table);
7692 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7693 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7694 			tte->flow_table);
7695 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7696 				tte->flow_table);
7697 		goto err;
7698 	} else if (!tte->flow_table) {
7699 		goto err;
7700 	}
7701 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7702 	tte->tunnel_id = tbl.tunnel_id;
7703 	tte->group = tbl.group;
7704 	return &tte->hash;
7705 err:
7706 	if (tte)
7707 		mlx5_free(tte);
7708 	return NULL;
7709 }
7710 
7711 static uint32_t
7712 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7713 				const struct mlx5_flow_tunnel *tunnel,
7714 				uint32_t group, uint32_t *table,
7715 				struct rte_flow_error *error)
7716 {
7717 	struct mlx5_hlist_entry *he;
7718 	struct tunnel_tbl_entry *tte;
7719 	union tunnel_tbl_key key = {
7720 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7721 		.group = group
7722 	};
7723 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7724 	struct mlx5_hlist *group_hash;
7725 
7726 	group_hash = tunnel ? tunnel->groups : thub->groups;
7727 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7728 	if (!he)
7729 		return rte_flow_error_set(error, EINVAL,
7730 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7731 					  NULL,
7732 					  "tunnel group index not supported");
7733 	tte = container_of(he, typeof(*tte), hash);
7734 	*table = tte->flow_table;
7735 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7736 		dev->data->port_id, key.tunnel_id, group, *table);
7737 	return 0;
7738 }
7739 
7740 static void
7741 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7742 		      struct mlx5_flow_tunnel *tunnel)
7743 {
7744 	struct mlx5_priv *priv = dev->data->dev_private;
7745 	struct mlx5_indexed_pool *ipool;
7746 
7747 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7748 		dev->data->port_id, tunnel->tunnel_id);
7749 	LIST_REMOVE(tunnel, chain);
7750 	mlx5_hlist_destroy(tunnel->groups);
7751 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7752 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
7753 }
7754 
7755 static bool
7756 mlx5_access_tunnel_offload_db
7757 	(struct rte_eth_dev *dev,
7758 	 bool (*match)(struct rte_eth_dev *,
7759 		       struct mlx5_flow_tunnel *, const void *),
7760 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
7761 	 void (*miss)(struct rte_eth_dev *, void *),
7762 	 void *ctx, bool lock_op)
7763 {
7764 	bool verdict = false;
7765 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7766 	struct mlx5_flow_tunnel *tunnel;
7767 
7768 	rte_spinlock_lock(&thub->sl);
7769 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
7770 		verdict = match(dev, tunnel, (const void *)ctx);
7771 		if (verdict)
7772 			break;
7773 	}
7774 	if (!lock_op)
7775 		rte_spinlock_unlock(&thub->sl);
7776 	if (verdict && hit)
7777 		hit(dev, tunnel, ctx);
7778 	if (!verdict && miss)
7779 		miss(dev, ctx);
7780 	if (lock_op)
7781 		rte_spinlock_unlock(&thub->sl);
7782 
7783 	return verdict;
7784 }
7785 
7786 struct tunnel_db_find_tunnel_id_ctx {
7787 	uint32_t tunnel_id;
7788 	struct mlx5_flow_tunnel *tunnel;
7789 };
7790 
7791 static bool
7792 find_tunnel_id_match(struct rte_eth_dev *dev,
7793 		     struct mlx5_flow_tunnel *tunnel, const void *x)
7794 {
7795 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7796 
7797 	RTE_SET_USED(dev);
7798 	return tunnel->tunnel_id == ctx->tunnel_id;
7799 }
7800 
7801 static void
7802 find_tunnel_id_hit(struct rte_eth_dev *dev,
7803 		   struct mlx5_flow_tunnel *tunnel, void *x)
7804 {
7805 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
7806 	RTE_SET_USED(dev);
7807 	ctx->tunnel = tunnel;
7808 }
7809 
7810 static struct mlx5_flow_tunnel *
7811 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7812 {
7813 	struct tunnel_db_find_tunnel_id_ctx ctx = {
7814 		.tunnel_id = id,
7815 	};
7816 
7817 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
7818 				      find_tunnel_id_hit, NULL, &ctx, true);
7819 
7820 	return ctx.tunnel;
7821 }
7822 
7823 static struct mlx5_flow_tunnel *
7824 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7825 			  const struct rte_flow_tunnel *app_tunnel)
7826 {
7827 	struct mlx5_priv *priv = dev->data->dev_private;
7828 	struct mlx5_indexed_pool *ipool;
7829 	struct mlx5_flow_tunnel *tunnel;
7830 	uint32_t id;
7831 
7832 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
7833 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
7834 	if (!tunnel)
7835 		return NULL;
7836 	if (id >= MLX5_MAX_TUNNELS) {
7837 		mlx5_ipool_free(ipool, id);
7838 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7839 		return NULL;
7840 	}
7841 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7842 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7843 					   mlx5_flow_tunnel_grp2tbl_match_cb,
7844 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7845 	if (!tunnel->groups) {
7846 		mlx5_ipool_free(ipool, id);
7847 		return NULL;
7848 	}
7849 	tunnel->groups->ctx = priv->sh;
7850 	/* initiate new PMD tunnel */
7851 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7852 	tunnel->tunnel_id = id;
7853 	tunnel->action.type = (typeof(tunnel->action.type))
7854 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7855 	tunnel->action.conf = tunnel;
7856 	tunnel->item.type = (typeof(tunnel->item.type))
7857 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7858 	tunnel->item.spec = tunnel;
7859 	tunnel->item.last = NULL;
7860 	tunnel->item.mask = NULL;
7861 
7862 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7863 		dev->data->port_id, tunnel->tunnel_id);
7864 
7865 	return tunnel;
7866 }
7867 
7868 struct tunnel_db_get_tunnel_ctx {
7869 	const struct rte_flow_tunnel *app_tunnel;
7870 	struct mlx5_flow_tunnel *tunnel;
7871 };
7872 
7873 static bool get_tunnel_match(struct rte_eth_dev *dev,
7874 			     struct mlx5_flow_tunnel *tunnel, const void *x)
7875 {
7876 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
7877 
7878 	RTE_SET_USED(dev);
7879 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
7880 		       sizeof(*ctx->app_tunnel));
7881 }
7882 
7883 static void get_tunnel_hit(struct rte_eth_dev *dev,
7884 			   struct mlx5_flow_tunnel *tunnel, void *x)
7885 {
7886 	/* called under tunnel spinlock protection */
7887 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7888 
7889 	RTE_SET_USED(dev);
7890 	tunnel->refctn++;
7891 	ctx->tunnel = tunnel;
7892 }
7893 
7894 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
7895 {
7896 	/* called under tunnel spinlock protection */
7897 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7898 	struct tunnel_db_get_tunnel_ctx *ctx = x;
7899 
7900 	rte_spinlock_unlock(&thub->sl);
7901 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
7902 	ctx->tunnel->refctn = 1;
7903 	rte_spinlock_lock(&thub->sl);
7904 	if (ctx->tunnel)
7905 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
7906 }
7907 
7908 
7909 static int
7910 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7911 		     const struct rte_flow_tunnel *app_tunnel,
7912 		     struct mlx5_flow_tunnel **tunnel)
7913 {
7914 	struct tunnel_db_get_tunnel_ctx ctx = {
7915 		.app_tunnel = app_tunnel,
7916 	};
7917 
7918 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
7919 				      get_tunnel_miss, &ctx, true);
7920 	*tunnel = ctx.tunnel;
7921 	return ctx.tunnel ? 0 : -ENOMEM;
7922 }
7923 
7924 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7925 {
7926 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7927 
7928 	if (!thub)
7929 		return;
7930 	if (!LIST_EMPTY(&thub->tunnels))
7931 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
7932 	mlx5_hlist_destroy(thub->groups);
7933 	mlx5_free(thub);
7934 }
7935 
7936 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7937 {
7938 	int err;
7939 	struct mlx5_flow_tunnel_hub *thub;
7940 
7941 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7942 			   0, SOCKET_ID_ANY);
7943 	if (!thub)
7944 		return -ENOMEM;
7945 	LIST_INIT(&thub->tunnels);
7946 	rte_spinlock_init(&thub->sl);
7947 	thub->groups = mlx5_hlist_create("flow groups",
7948 					 rte_align32pow2(MLX5_MAX_TABLES), 0,
7949 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7950 					 mlx5_flow_tunnel_grp2tbl_match_cb,
7951 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7952 	if (!thub->groups) {
7953 		err = -rte_errno;
7954 		goto err;
7955 	}
7956 	thub->groups->ctx = sh;
7957 	sh->tunnel_hub = thub;
7958 
7959 	return 0;
7960 
7961 err:
7962 	if (thub->groups)
7963 		mlx5_hlist_destroy(thub->groups);
7964 	if (thub)
7965 		mlx5_free(thub);
7966 	return err;
7967 }
7968 
7969 static inline bool
7970 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
7971 			  struct rte_flow_tunnel *tunnel,
7972 			  const char *err_msg)
7973 {
7974 	err_msg = NULL;
7975 	if (!is_tunnel_offload_active(dev)) {
7976 		err_msg = "tunnel offload was not activated";
7977 		goto out;
7978 	} else if (!tunnel) {
7979 		err_msg = "no application tunnel";
7980 		goto out;
7981 	}
7982 
7983 	switch (tunnel->type) {
7984 	default:
7985 		err_msg = "unsupported tunnel type";
7986 		goto out;
7987 	case RTE_FLOW_ITEM_TYPE_VXLAN:
7988 		break;
7989 	}
7990 
7991 out:
7992 	return !err_msg;
7993 }
7994 
7995 static int
7996 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
7997 		    struct rte_flow_tunnel *app_tunnel,
7998 		    struct rte_flow_action **actions,
7999 		    uint32_t *num_of_actions,
8000 		    struct rte_flow_error *error)
8001 {
8002 	int ret;
8003 	struct mlx5_flow_tunnel *tunnel;
8004 	const char *err_msg = NULL;
8005 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8006 
8007 	if (!verdict)
8008 		return rte_flow_error_set(error, EINVAL,
8009 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8010 					  err_msg);
8011 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8012 	if (ret < 0) {
8013 		return rte_flow_error_set(error, ret,
8014 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
8015 					  "failed to initialize pmd tunnel");
8016 	}
8017 	*actions = &tunnel->action;
8018 	*num_of_actions = 1;
8019 	return 0;
8020 }
8021 
8022 static int
8023 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
8024 		       struct rte_flow_tunnel *app_tunnel,
8025 		       struct rte_flow_item **items,
8026 		       uint32_t *num_of_items,
8027 		       struct rte_flow_error *error)
8028 {
8029 	int ret;
8030 	struct mlx5_flow_tunnel *tunnel;
8031 	const char *err_msg = NULL;
8032 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
8033 
8034 	if (!verdict)
8035 		return rte_flow_error_set(error, EINVAL,
8036 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8037 					  err_msg);
8038 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
8039 	if (ret < 0) {
8040 		return rte_flow_error_set(error, ret,
8041 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8042 					  "failed to initialize pmd tunnel");
8043 	}
8044 	*items = &tunnel->item;
8045 	*num_of_items = 1;
8046 	return 0;
8047 }
8048 
8049 struct tunnel_db_element_release_ctx {
8050 	struct rte_flow_item *items;
8051 	struct rte_flow_action *actions;
8052 	uint32_t num_elements;
8053 	struct rte_flow_error *error;
8054 	int ret;
8055 };
8056 
8057 static bool
8058 tunnel_element_release_match(struct rte_eth_dev *dev,
8059 			     struct mlx5_flow_tunnel *tunnel, const void *x)
8060 {
8061 	const struct tunnel_db_element_release_ctx *ctx = x;
8062 
8063 	RTE_SET_USED(dev);
8064 	if (ctx->num_elements != 1)
8065 		return false;
8066 	else if (ctx->items)
8067 		return ctx->items == &tunnel->item;
8068 	else if (ctx->actions)
8069 		return ctx->actions == &tunnel->action;
8070 
8071 	return false;
8072 }
8073 
8074 static void
8075 tunnel_element_release_hit(struct rte_eth_dev *dev,
8076 			   struct mlx5_flow_tunnel *tunnel, void *x)
8077 {
8078 	struct tunnel_db_element_release_ctx *ctx = x;
8079 	ctx->ret = 0;
8080 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
8081 		mlx5_flow_tunnel_free(dev, tunnel);
8082 }
8083 
8084 static void
8085 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
8086 {
8087 	struct tunnel_db_element_release_ctx *ctx = x;
8088 	RTE_SET_USED(dev);
8089 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
8090 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
8091 				      "invalid argument");
8092 }
8093 
8094 static int
8095 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
8096 		       struct rte_flow_item *pmd_items,
8097 		       uint32_t num_items, struct rte_flow_error *err)
8098 {
8099 	struct tunnel_db_element_release_ctx ctx = {
8100 		.items = pmd_items,
8101 		.actions = NULL,
8102 		.num_elements = num_items,
8103 		.error = err,
8104 	};
8105 
8106 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8107 				      tunnel_element_release_hit,
8108 				      tunnel_element_release_miss, &ctx, false);
8109 
8110 	return ctx.ret;
8111 }
8112 
8113 static int
8114 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
8115 			 struct rte_flow_action *pmd_actions,
8116 			 uint32_t num_actions, struct rte_flow_error *err)
8117 {
8118 	struct tunnel_db_element_release_ctx ctx = {
8119 		.items = NULL,
8120 		.actions = pmd_actions,
8121 		.num_elements = num_actions,
8122 		.error = err,
8123 	};
8124 
8125 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
8126 				      tunnel_element_release_hit,
8127 				      tunnel_element_release_miss, &ctx, false);
8128 
8129 	return ctx.ret;
8130 }
8131 
8132 static int
8133 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
8134 				  struct rte_mbuf *m,
8135 				  struct rte_flow_restore_info *info,
8136 				  struct rte_flow_error *err)
8137 {
8138 	uint64_t ol_flags = m->ol_flags;
8139 	const struct mlx5_flow_tbl_data_entry *tble;
8140 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
8141 
8142 	if (!is_tunnel_offload_active(dev)) {
8143 		info->flags = 0;
8144 		return 0;
8145 	}
8146 
8147 	if ((ol_flags & mask) != mask)
8148 		goto err;
8149 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
8150 	if (!tble) {
8151 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
8152 			dev->data->port_id, m->hash.fdir.hi);
8153 		goto err;
8154 	}
8155 	MLX5_ASSERT(tble->tunnel);
8156 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
8157 	info->group_id = tble->group_id;
8158 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
8159 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
8160 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
8161 
8162 	return 0;
8163 
8164 err:
8165 	return rte_flow_error_set(err, EINVAL,
8166 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8167 				  "failed to get restore info");
8168 }
8169 
8170 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
8171 static int
8172 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
8173 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
8174 			   __rte_unused struct rte_flow_action **actions,
8175 			   __rte_unused uint32_t *num_of_actions,
8176 			   __rte_unused struct rte_flow_error *error)
8177 {
8178 	return -ENOTSUP;
8179 }
8180 
8181 static int
8182 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
8183 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
8184 		       __rte_unused struct rte_flow_item **items,
8185 		       __rte_unused uint32_t *num_of_items,
8186 		       __rte_unused struct rte_flow_error *error)
8187 {
8188 	return -ENOTSUP;
8189 }
8190 
8191 static int
8192 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
8193 			      __rte_unused struct rte_flow_item *pmd_items,
8194 			      __rte_unused uint32_t num_items,
8195 			      __rte_unused struct rte_flow_error *err)
8196 {
8197 	return -ENOTSUP;
8198 }
8199 
8200 static int
8201 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
8202 				__rte_unused struct rte_flow_action *pmd_action,
8203 				__rte_unused uint32_t num_actions,
8204 				__rte_unused struct rte_flow_error *err)
8205 {
8206 	return -ENOTSUP;
8207 }
8208 
8209 static int
8210 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
8211 				  __rte_unused struct rte_mbuf *m,
8212 				  __rte_unused struct rte_flow_restore_info *i,
8213 				  __rte_unused struct rte_flow_error *err)
8214 {
8215 	return -ENOTSUP;
8216 }
8217 
8218 static int
8219 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
8220 			     __rte_unused struct rte_flow *flow,
8221 			     __rte_unused const struct rte_flow_attr *attr,
8222 			     __rte_unused const struct rte_flow_action *actions,
8223 			     __rte_unused uint32_t flow_idx,
8224 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
8225 			     __rte_unused struct rte_flow_error *error)
8226 {
8227 	return -ENOTSUP;
8228 }
8229 
8230 static struct mlx5_flow_tunnel *
8231 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
8232 		    __rte_unused uint32_t id)
8233 {
8234 	return NULL;
8235 }
8236 
8237 static void
8238 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
8239 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
8240 {
8241 }
8242 
8243 static uint32_t
8244 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
8245 				__rte_unused const struct mlx5_flow_tunnel *t,
8246 				__rte_unused uint32_t group,
8247 				__rte_unused uint32_t *table,
8248 				struct rte_flow_error *error)
8249 {
8250 	return rte_flow_error_set(error, ENOTSUP,
8251 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8252 				  "tunnel offload requires DV support");
8253 }
8254 
8255 void
8256 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
8257 			__rte_unused  uint16_t port_id)
8258 {
8259 }
8260 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
8261 
8262