xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 23f627e0ed28ceb7da97ab490fd032529793e097)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <rte_ethdev_driver.h>
16 #include <rte_eal_paging.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
21 #include <rte_ip.h>
22 
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
25 #include <mlx5_prm.h>
26 #include <mlx5_malloc.h>
27 
28 #include "mlx5_defs.h"
29 #include "mlx5.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 static struct mlx5_flow_tunnel *
37 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
38 static void
39 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
40 static const struct mlx5_flow_tbl_data_entry  *
41 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark);
42 static int
43 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
44 		     const struct rte_flow_tunnel *app_tunnel,
45 		     struct mlx5_flow_tunnel **tunnel);
46 
47 
48 /** Device flow drivers. */
49 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
50 
51 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
52 
53 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
54 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
55 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
56 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
57 #endif
58 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
59 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
60 };
61 
62 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
63 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
64 	(const int []){ \
65 		__VA_ARGS__, 0, \
66 	}
67 
68 /** Node object of input graph for mlx5_flow_expand_rss(). */
69 struct mlx5_flow_expand_node {
70 	const int *const next;
71 	/**<
72 	 * List of next node indexes. Index 0 is interpreted as a terminator.
73 	 */
74 	const enum rte_flow_item_type type;
75 	/**< Pattern item type of current node. */
76 	uint64_t rss_types;
77 	/**<
78 	 * RSS types bit-field associated with this node
79 	 * (see ETH_RSS_* definitions).
80 	 */
81 };
82 
83 /** Object returned by mlx5_flow_expand_rss(). */
84 struct mlx5_flow_expand_rss {
85 	uint32_t entries;
86 	/**< Number of entries @p patterns and @p priorities. */
87 	struct {
88 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
89 		uint32_t priority; /**< Priority offset for each expansion. */
90 	} entry[];
91 };
92 
93 static enum rte_flow_item_type
94 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
95 {
96 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
97 	uint16_t ether_type = 0;
98 	uint16_t ether_type_m;
99 	uint8_t ip_next_proto = 0;
100 	uint8_t ip_next_proto_m;
101 
102 	if (item == NULL || item->spec == NULL)
103 		return ret;
104 	switch (item->type) {
105 	case RTE_FLOW_ITEM_TYPE_ETH:
106 		if (item->mask)
107 			ether_type_m = ((const struct rte_flow_item_eth *)
108 						(item->mask))->type;
109 		else
110 			ether_type_m = rte_flow_item_eth_mask.type;
111 		if (ether_type_m != RTE_BE16(0xFFFF))
112 			break;
113 		ether_type = ((const struct rte_flow_item_eth *)
114 				(item->spec))->type;
115 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
116 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
117 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
118 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
119 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
120 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
121 		else
122 			ret = RTE_FLOW_ITEM_TYPE_END;
123 		break;
124 	case RTE_FLOW_ITEM_TYPE_VLAN:
125 		if (item->mask)
126 			ether_type_m = ((const struct rte_flow_item_vlan *)
127 						(item->mask))->inner_type;
128 		else
129 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
130 		if (ether_type_m != RTE_BE16(0xFFFF))
131 			break;
132 		ether_type = ((const struct rte_flow_item_vlan *)
133 				(item->spec))->inner_type;
134 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
135 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
136 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
139 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
140 		else
141 			ret = RTE_FLOW_ITEM_TYPE_END;
142 		break;
143 	case RTE_FLOW_ITEM_TYPE_IPV4:
144 		if (item->mask)
145 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
146 					(item->mask))->hdr.next_proto_id;
147 		else
148 			ip_next_proto_m =
149 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
150 		if (ip_next_proto_m != 0xFF)
151 			break;
152 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
153 				(item->spec))->hdr.next_proto_id;
154 		if (ip_next_proto == IPPROTO_UDP)
155 			ret = RTE_FLOW_ITEM_TYPE_UDP;
156 		else if (ip_next_proto == IPPROTO_TCP)
157 			ret = RTE_FLOW_ITEM_TYPE_TCP;
158 		else if (ip_next_proto == IPPROTO_IP)
159 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
160 		else if (ip_next_proto == IPPROTO_IPV6)
161 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
162 		else
163 			ret = RTE_FLOW_ITEM_TYPE_END;
164 		break;
165 	case RTE_FLOW_ITEM_TYPE_IPV6:
166 		if (item->mask)
167 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
168 						(item->mask))->hdr.proto;
169 		else
170 			ip_next_proto_m =
171 				rte_flow_item_ipv6_mask.hdr.proto;
172 		if (ip_next_proto_m != 0xFF)
173 			break;
174 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
175 				(item->spec))->hdr.proto;
176 		if (ip_next_proto == IPPROTO_UDP)
177 			ret = RTE_FLOW_ITEM_TYPE_UDP;
178 		else if (ip_next_proto == IPPROTO_TCP)
179 			ret = RTE_FLOW_ITEM_TYPE_TCP;
180 		else if (ip_next_proto == IPPROTO_IP)
181 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
182 		else if (ip_next_proto == IPPROTO_IPV6)
183 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
184 		else
185 			ret = RTE_FLOW_ITEM_TYPE_END;
186 		break;
187 	default:
188 		ret = RTE_FLOW_ITEM_TYPE_VOID;
189 		break;
190 	}
191 	return ret;
192 }
193 
194 /**
195  * Expand RSS flows into several possible flows according to the RSS hash
196  * fields requested and the driver capabilities.
197  *
198  * @param[out] buf
199  *   Buffer to store the result expansion.
200  * @param[in] size
201  *   Buffer size in bytes. If 0, @p buf can be NULL.
202  * @param[in] pattern
203  *   User flow pattern.
204  * @param[in] types
205  *   RSS types to expand (see ETH_RSS_* definitions).
206  * @param[in] graph
207  *   Input graph to expand @p pattern according to @p types.
208  * @param[in] graph_root_index
209  *   Index of root node in @p graph, typically 0.
210  *
211  * @return
212  *   A positive value representing the size of @p buf in bytes regardless of
213  *   @p size on success, a negative errno value otherwise and rte_errno is
214  *   set, the following errors are defined:
215  *
216  *   -E2BIG: graph-depth @p graph is too deep.
217  */
218 static int
219 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
220 		     const struct rte_flow_item *pattern, uint64_t types,
221 		     const struct mlx5_flow_expand_node graph[],
222 		     int graph_root_index)
223 {
224 	const int elt_n = 8;
225 	const struct rte_flow_item *item;
226 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
227 	const int *next_node;
228 	const int *stack[elt_n];
229 	int stack_pos = 0;
230 	struct rte_flow_item flow_items[elt_n];
231 	unsigned int i;
232 	size_t lsize;
233 	size_t user_pattern_size = 0;
234 	void *addr = NULL;
235 	const struct mlx5_flow_expand_node *next = NULL;
236 	struct rte_flow_item missed_item;
237 	int missed = 0;
238 	int elt = 0;
239 	const struct rte_flow_item *last_item = NULL;
240 
241 	memset(&missed_item, 0, sizeof(missed_item));
242 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
243 		elt_n * sizeof(buf->entry[0]);
244 	if (lsize <= size) {
245 		buf->entry[0].priority = 0;
246 		buf->entry[0].pattern = (void *)&buf->entry[elt_n];
247 		buf->entries = 0;
248 		addr = buf->entry[0].pattern;
249 	}
250 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
251 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
252 			last_item = item;
253 		for (i = 0; node->next && node->next[i]; ++i) {
254 			next = &graph[node->next[i]];
255 			if (next->type == item->type)
256 				break;
257 		}
258 		if (next)
259 			node = next;
260 		user_pattern_size += sizeof(*item);
261 	}
262 	user_pattern_size += sizeof(*item); /* Handle END item. */
263 	lsize += user_pattern_size;
264 	/* Copy the user pattern in the first entry of the buffer. */
265 	if (lsize <= size) {
266 		rte_memcpy(addr, pattern, user_pattern_size);
267 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
268 		buf->entries = 1;
269 	}
270 	/* Start expanding. */
271 	memset(flow_items, 0, sizeof(flow_items));
272 	user_pattern_size -= sizeof(*item);
273 	/*
274 	 * Check if the last valid item has spec set, need complete pattern,
275 	 * and the pattern can be used for expansion.
276 	 */
277 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
278 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
279 		/* Item type END indicates expansion is not required. */
280 		return lsize;
281 	}
282 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
283 		next = NULL;
284 		missed = 1;
285 		for (i = 0; node->next && node->next[i]; ++i) {
286 			next = &graph[node->next[i]];
287 			if (next->type == missed_item.type) {
288 				flow_items[0].type = missed_item.type;
289 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
290 				break;
291 			}
292 			next = NULL;
293 		}
294 	}
295 	if (next && missed) {
296 		elt = 2; /* missed item + item end. */
297 		node = next;
298 		lsize += elt * sizeof(*item) + user_pattern_size;
299 		if ((node->rss_types & types) && lsize <= size) {
300 			buf->entry[buf->entries].priority = 1;
301 			buf->entry[buf->entries].pattern = addr;
302 			buf->entries++;
303 			rte_memcpy(addr, buf->entry[0].pattern,
304 				   user_pattern_size);
305 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
306 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
307 			addr = (void *)(((uintptr_t)addr) +
308 					elt * sizeof(*item));
309 		}
310 	}
311 	memset(flow_items, 0, sizeof(flow_items));
312 	next_node = node->next;
313 	stack[stack_pos] = next_node;
314 	node = next_node ? &graph[*next_node] : NULL;
315 	while (node) {
316 		flow_items[stack_pos].type = node->type;
317 		if (node->rss_types & types) {
318 			/*
319 			 * compute the number of items to copy from the
320 			 * expansion and copy it.
321 			 * When the stack_pos is 0, there are 1 element in it,
322 			 * plus the addition END item.
323 			 */
324 			elt = stack_pos + 2;
325 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
326 			lsize += elt * sizeof(*item) + user_pattern_size;
327 			if (lsize <= size) {
328 				size_t n = elt * sizeof(*item);
329 
330 				buf->entry[buf->entries].priority =
331 					stack_pos + 1 + missed;
332 				buf->entry[buf->entries].pattern = addr;
333 				buf->entries++;
334 				rte_memcpy(addr, buf->entry[0].pattern,
335 					   user_pattern_size);
336 				addr = (void *)(((uintptr_t)addr) +
337 						user_pattern_size);
338 				rte_memcpy(addr, &missed_item,
339 					   missed * sizeof(*item));
340 				addr = (void *)(((uintptr_t)addr) +
341 					missed * sizeof(*item));
342 				rte_memcpy(addr, flow_items, n);
343 				addr = (void *)(((uintptr_t)addr) + n);
344 			}
345 		}
346 		/* Go deeper. */
347 		if (node->next) {
348 			next_node = node->next;
349 			if (stack_pos++ == elt_n) {
350 				rte_errno = E2BIG;
351 				return -rte_errno;
352 			}
353 			stack[stack_pos] = next_node;
354 		} else if (*(next_node + 1)) {
355 			/* Follow up with the next possibility. */
356 			++next_node;
357 		} else {
358 			/* Move to the next path. */
359 			if (stack_pos)
360 				next_node = stack[--stack_pos];
361 			next_node++;
362 			stack[stack_pos] = next_node;
363 		}
364 		node = *next_node ? &graph[*next_node] : NULL;
365 	};
366 	/* no expanded flows but we have missed item, create one rule for it */
367 	if (buf->entries == 1 && missed != 0) {
368 		elt = 2;
369 		lsize += elt * sizeof(*item) + user_pattern_size;
370 		if (lsize <= size) {
371 			buf->entry[buf->entries].priority = 1;
372 			buf->entry[buf->entries].pattern = addr;
373 			buf->entries++;
374 			flow_items[0].type = missed_item.type;
375 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
376 			rte_memcpy(addr, buf->entry[0].pattern,
377 				   user_pattern_size);
378 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
379 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
380 			addr = (void *)(((uintptr_t)addr) +
381 					elt * sizeof(*item));
382 		}
383 	}
384 	return lsize;
385 }
386 
387 enum mlx5_expansion {
388 	MLX5_EXPANSION_ROOT,
389 	MLX5_EXPANSION_ROOT_OUTER,
390 	MLX5_EXPANSION_ROOT_ETH_VLAN,
391 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
392 	MLX5_EXPANSION_OUTER_ETH,
393 	MLX5_EXPANSION_OUTER_ETH_VLAN,
394 	MLX5_EXPANSION_OUTER_VLAN,
395 	MLX5_EXPANSION_OUTER_IPV4,
396 	MLX5_EXPANSION_OUTER_IPV4_UDP,
397 	MLX5_EXPANSION_OUTER_IPV4_TCP,
398 	MLX5_EXPANSION_OUTER_IPV6,
399 	MLX5_EXPANSION_OUTER_IPV6_UDP,
400 	MLX5_EXPANSION_OUTER_IPV6_TCP,
401 	MLX5_EXPANSION_VXLAN,
402 	MLX5_EXPANSION_VXLAN_GPE,
403 	MLX5_EXPANSION_GRE,
404 	MLX5_EXPANSION_MPLS,
405 	MLX5_EXPANSION_ETH,
406 	MLX5_EXPANSION_ETH_VLAN,
407 	MLX5_EXPANSION_VLAN,
408 	MLX5_EXPANSION_IPV4,
409 	MLX5_EXPANSION_IPV4_UDP,
410 	MLX5_EXPANSION_IPV4_TCP,
411 	MLX5_EXPANSION_IPV6,
412 	MLX5_EXPANSION_IPV6_UDP,
413 	MLX5_EXPANSION_IPV6_TCP,
414 };
415 
416 /** Supported expansion of items. */
417 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
418 	[MLX5_EXPANSION_ROOT] = {
419 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
420 						  MLX5_EXPANSION_IPV4,
421 						  MLX5_EXPANSION_IPV6),
422 		.type = RTE_FLOW_ITEM_TYPE_END,
423 	},
424 	[MLX5_EXPANSION_ROOT_OUTER] = {
425 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
426 						  MLX5_EXPANSION_OUTER_IPV4,
427 						  MLX5_EXPANSION_OUTER_IPV6),
428 		.type = RTE_FLOW_ITEM_TYPE_END,
429 	},
430 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
431 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
432 		.type = RTE_FLOW_ITEM_TYPE_END,
433 	},
434 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
435 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
436 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
437 		.type = RTE_FLOW_ITEM_TYPE_END,
438 	},
439 	[MLX5_EXPANSION_OUTER_ETH] = {
440 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
441 						  MLX5_EXPANSION_OUTER_IPV6,
442 						  MLX5_EXPANSION_MPLS),
443 		.type = RTE_FLOW_ITEM_TYPE_ETH,
444 		.rss_types = 0,
445 	},
446 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
447 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
448 		.type = RTE_FLOW_ITEM_TYPE_ETH,
449 		.rss_types = 0,
450 	},
451 	[MLX5_EXPANSION_OUTER_VLAN] = {
452 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
453 						  MLX5_EXPANSION_OUTER_IPV6),
454 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
455 	},
456 	[MLX5_EXPANSION_OUTER_IPV4] = {
457 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
458 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
459 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
460 			 MLX5_EXPANSION_GRE,
461 			 MLX5_EXPANSION_IPV4,
462 			 MLX5_EXPANSION_IPV6),
463 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
464 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
465 			ETH_RSS_NONFRAG_IPV4_OTHER,
466 	},
467 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
468 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
469 						  MLX5_EXPANSION_VXLAN_GPE),
470 		.type = RTE_FLOW_ITEM_TYPE_UDP,
471 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
472 	},
473 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
474 		.type = RTE_FLOW_ITEM_TYPE_TCP,
475 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV6] = {
478 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
479 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
480 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
481 			 MLX5_EXPANSION_IPV4,
482 			 MLX5_EXPANSION_IPV6),
483 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
484 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
485 			ETH_RSS_NONFRAG_IPV6_OTHER,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 						  MLX5_EXPANSION_VXLAN_GPE),
490 		.type = RTE_FLOW_ITEM_TYPE_UDP,
491 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
492 	},
493 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
494 		.type = RTE_FLOW_ITEM_TYPE_TCP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
496 	},
497 	[MLX5_EXPANSION_VXLAN] = {
498 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
499 						  MLX5_EXPANSION_IPV4,
500 						  MLX5_EXPANSION_IPV6),
501 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
502 	},
503 	[MLX5_EXPANSION_VXLAN_GPE] = {
504 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
505 						  MLX5_EXPANSION_IPV4,
506 						  MLX5_EXPANSION_IPV6),
507 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
508 	},
509 	[MLX5_EXPANSION_GRE] = {
510 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
511 		.type = RTE_FLOW_ITEM_TYPE_GRE,
512 	},
513 	[MLX5_EXPANSION_MPLS] = {
514 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
515 						  MLX5_EXPANSION_IPV6),
516 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
517 	},
518 	[MLX5_EXPANSION_ETH] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_ETH,
522 	},
523 	[MLX5_EXPANSION_ETH_VLAN] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
525 		.type = RTE_FLOW_ITEM_TYPE_ETH,
526 	},
527 	[MLX5_EXPANSION_VLAN] = {
528 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 						  MLX5_EXPANSION_IPV6),
530 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
531 	},
532 	[MLX5_EXPANSION_IPV4] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
534 						  MLX5_EXPANSION_IPV4_TCP),
535 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
536 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
537 			ETH_RSS_NONFRAG_IPV4_OTHER,
538 	},
539 	[MLX5_EXPANSION_IPV4_UDP] = {
540 		.type = RTE_FLOW_ITEM_TYPE_UDP,
541 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
542 	},
543 	[MLX5_EXPANSION_IPV4_TCP] = {
544 		.type = RTE_FLOW_ITEM_TYPE_TCP,
545 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
546 	},
547 	[MLX5_EXPANSION_IPV6] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
549 						  MLX5_EXPANSION_IPV6_TCP),
550 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
551 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
552 			ETH_RSS_NONFRAG_IPV6_OTHER,
553 	},
554 	[MLX5_EXPANSION_IPV6_UDP] = {
555 		.type = RTE_FLOW_ITEM_TYPE_UDP,
556 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
557 	},
558 	[MLX5_EXPANSION_IPV6_TCP] = {
559 		.type = RTE_FLOW_ITEM_TYPE_TCP,
560 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
561 	},
562 };
563 
564 static struct rte_flow_shared_action *
565 mlx5_shared_action_create(struct rte_eth_dev *dev,
566 			  const struct rte_flow_shared_action_conf *conf,
567 			  const struct rte_flow_action *action,
568 			  struct rte_flow_error *error);
569 static int mlx5_shared_action_destroy
570 				(struct rte_eth_dev *dev,
571 				 struct rte_flow_shared_action *shared_action,
572 				 struct rte_flow_error *error);
573 static int mlx5_shared_action_update
574 				(struct rte_eth_dev *dev,
575 				 struct rte_flow_shared_action *shared_action,
576 				 const struct rte_flow_action *action,
577 				 struct rte_flow_error *error);
578 static int mlx5_shared_action_query
579 				(struct rte_eth_dev *dev,
580 				 const struct rte_flow_shared_action *action,
581 				 void *data,
582 				 struct rte_flow_error *error);
583 static inline bool
584 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
585 			  struct rte_flow_tunnel *tunnel,
586 			  const char *err_msg)
587 {
588 	err_msg = NULL;
589 	if (!is_tunnel_offload_active(dev)) {
590 		err_msg = "tunnel offload was not activated";
591 		goto out;
592 	} else if (!tunnel) {
593 		err_msg = "no application tunnel";
594 		goto out;
595 	}
596 
597 	switch (tunnel->type) {
598 	default:
599 		err_msg = "unsupported tunnel type";
600 		goto out;
601 	case RTE_FLOW_ITEM_TYPE_VXLAN:
602 		break;
603 	}
604 
605 out:
606 	return !err_msg;
607 }
608 
609 
610 static int
611 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
612 		    struct rte_flow_tunnel *app_tunnel,
613 		    struct rte_flow_action **actions,
614 		    uint32_t *num_of_actions,
615 		    struct rte_flow_error *error)
616 {
617 	int ret;
618 	struct mlx5_flow_tunnel *tunnel;
619 	const char *err_msg = NULL;
620 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
621 
622 	if (!verdict)
623 		return rte_flow_error_set(error, EINVAL,
624 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
625 					  err_msg);
626 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
627 	if (ret < 0) {
628 		return rte_flow_error_set(error, ret,
629 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
630 					  "failed to initialize pmd tunnel");
631 	}
632 	*actions = &tunnel->action;
633 	*num_of_actions = 1;
634 	return 0;
635 }
636 
637 static int
638 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
639 		       struct rte_flow_tunnel *app_tunnel,
640 		       struct rte_flow_item **items,
641 		       uint32_t *num_of_items,
642 		       struct rte_flow_error *error)
643 {
644 	int ret;
645 	struct mlx5_flow_tunnel *tunnel;
646 	const char *err_msg = NULL;
647 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
648 
649 	if (!verdict)
650 		return rte_flow_error_set(error, EINVAL,
651 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
652 					  err_msg);
653 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
654 	if (ret < 0) {
655 		return rte_flow_error_set(error, ret,
656 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
657 					  "failed to initialize pmd tunnel");
658 	}
659 	*items = &tunnel->item;
660 	*num_of_items = 1;
661 	return 0;
662 }
663 
664 static int
665 mlx5_flow_item_release(struct rte_eth_dev *dev,
666 		       struct rte_flow_item *pmd_items,
667 		       uint32_t num_items, struct rte_flow_error *err)
668 {
669 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
670 	struct mlx5_flow_tunnel *tun;
671 
672 	LIST_FOREACH(tun, &thub->tunnels, chain) {
673 		if (&tun->item == pmd_items)
674 			break;
675 	}
676 	if (!tun || num_items != 1)
677 		return rte_flow_error_set(err, EINVAL,
678 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
679 					  "invalid argument");
680 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
681 		mlx5_flow_tunnel_free(dev, tun);
682 	return 0;
683 }
684 
685 static int
686 mlx5_flow_action_release(struct rte_eth_dev *dev,
687 			 struct rte_flow_action *pmd_actions,
688 			 uint32_t num_actions, struct rte_flow_error *err)
689 {
690 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
691 	struct mlx5_flow_tunnel *tun;
692 
693 	LIST_FOREACH(tun, &thub->tunnels, chain) {
694 		if (&tun->action == pmd_actions)
695 			break;
696 	}
697 	if (!tun || num_actions != 1)
698 		return rte_flow_error_set(err, EINVAL,
699 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
700 					  "invalid argument");
701 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
702 		mlx5_flow_tunnel_free(dev, tun);
703 
704 	return 0;
705 }
706 
707 static int
708 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
709 				  struct rte_mbuf *m,
710 				  struct rte_flow_restore_info *info,
711 				  struct rte_flow_error *err)
712 {
713 	uint64_t ol_flags = m->ol_flags;
714 	const struct mlx5_flow_tbl_data_entry *tble;
715 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
716 
717 	if ((ol_flags & mask) != mask)
718 		goto err;
719 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
720 	if (!tble) {
721 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
722 			dev->data->port_id, m->hash.fdir.hi);
723 		goto err;
724 	}
725 	MLX5_ASSERT(tble->tunnel);
726 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
727 	info->group_id = tble->group_id;
728 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
729 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
730 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
731 
732 	return 0;
733 
734 err:
735 	return rte_flow_error_set(err, EINVAL,
736 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
737 				  "failed to get restore info");
738 }
739 
740 static const struct rte_flow_ops mlx5_flow_ops = {
741 	.validate = mlx5_flow_validate,
742 	.create = mlx5_flow_create,
743 	.destroy = mlx5_flow_destroy,
744 	.flush = mlx5_flow_flush,
745 	.isolate = mlx5_flow_isolate,
746 	.query = mlx5_flow_query,
747 	.dev_dump = mlx5_flow_dev_dump,
748 	.get_aged_flows = mlx5_flow_get_aged_flows,
749 	.shared_action_create = mlx5_shared_action_create,
750 	.shared_action_destroy = mlx5_shared_action_destroy,
751 	.shared_action_update = mlx5_shared_action_update,
752 	.shared_action_query = mlx5_shared_action_query,
753 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
754 	.tunnel_match = mlx5_flow_tunnel_match,
755 	.tunnel_action_decap_release = mlx5_flow_action_release,
756 	.tunnel_item_release = mlx5_flow_item_release,
757 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
758 };
759 
760 /* Convert FDIR request to Generic flow. */
761 struct mlx5_fdir {
762 	struct rte_flow_attr attr;
763 	struct rte_flow_item items[4];
764 	struct rte_flow_item_eth l2;
765 	struct rte_flow_item_eth l2_mask;
766 	union {
767 		struct rte_flow_item_ipv4 ipv4;
768 		struct rte_flow_item_ipv6 ipv6;
769 	} l3;
770 	union {
771 		struct rte_flow_item_ipv4 ipv4;
772 		struct rte_flow_item_ipv6 ipv6;
773 	} l3_mask;
774 	union {
775 		struct rte_flow_item_udp udp;
776 		struct rte_flow_item_tcp tcp;
777 	} l4;
778 	union {
779 		struct rte_flow_item_udp udp;
780 		struct rte_flow_item_tcp tcp;
781 	} l4_mask;
782 	struct rte_flow_action actions[2];
783 	struct rte_flow_action_queue queue;
784 };
785 
786 /* Tunnel information. */
787 struct mlx5_flow_tunnel_info {
788 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
789 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
790 };
791 
792 static struct mlx5_flow_tunnel_info tunnels_info[] = {
793 	{
794 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
795 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
796 	},
797 	{
798 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
799 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
800 	},
801 	{
802 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
803 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
804 	},
805 	{
806 		.tunnel = MLX5_FLOW_LAYER_GRE,
807 		.ptype = RTE_PTYPE_TUNNEL_GRE,
808 	},
809 	{
810 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
811 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
812 	},
813 	{
814 		.tunnel = MLX5_FLOW_LAYER_MPLS,
815 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
816 	},
817 	{
818 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
819 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
820 	},
821 	{
822 		.tunnel = MLX5_FLOW_LAYER_IPIP,
823 		.ptype = RTE_PTYPE_TUNNEL_IP,
824 	},
825 	{
826 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
827 		.ptype = RTE_PTYPE_TUNNEL_IP,
828 	},
829 	{
830 		.tunnel = MLX5_FLOW_LAYER_GTP,
831 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
832 	},
833 };
834 
835 /**
836  * Translate tag ID to register.
837  *
838  * @param[in] dev
839  *   Pointer to the Ethernet device structure.
840  * @param[in] feature
841  *   The feature that request the register.
842  * @param[in] id
843  *   The request register ID.
844  * @param[out] error
845  *   Error description in case of any.
846  *
847  * @return
848  *   The request register on success, a negative errno
849  *   value otherwise and rte_errno is set.
850  */
851 int
852 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
853 		     enum mlx5_feature_name feature,
854 		     uint32_t id,
855 		     struct rte_flow_error *error)
856 {
857 	struct mlx5_priv *priv = dev->data->dev_private;
858 	struct mlx5_dev_config *config = &priv->config;
859 	enum modify_reg start_reg;
860 	bool skip_mtr_reg = false;
861 
862 	switch (feature) {
863 	case MLX5_HAIRPIN_RX:
864 		return REG_B;
865 	case MLX5_HAIRPIN_TX:
866 		return REG_A;
867 	case MLX5_METADATA_RX:
868 		switch (config->dv_xmeta_en) {
869 		case MLX5_XMETA_MODE_LEGACY:
870 			return REG_B;
871 		case MLX5_XMETA_MODE_META16:
872 			return REG_C_0;
873 		case MLX5_XMETA_MODE_META32:
874 			return REG_C_1;
875 		}
876 		break;
877 	case MLX5_METADATA_TX:
878 		return REG_A;
879 	case MLX5_METADATA_FDB:
880 		switch (config->dv_xmeta_en) {
881 		case MLX5_XMETA_MODE_LEGACY:
882 			return REG_NON;
883 		case MLX5_XMETA_MODE_META16:
884 			return REG_C_0;
885 		case MLX5_XMETA_MODE_META32:
886 			return REG_C_1;
887 		}
888 		break;
889 	case MLX5_FLOW_MARK:
890 		switch (config->dv_xmeta_en) {
891 		case MLX5_XMETA_MODE_LEGACY:
892 			return REG_NON;
893 		case MLX5_XMETA_MODE_META16:
894 			return REG_C_1;
895 		case MLX5_XMETA_MODE_META32:
896 			return REG_C_0;
897 		}
898 		break;
899 	case MLX5_MTR_SFX:
900 		/*
901 		 * If meter color and flow match share one register, flow match
902 		 * should use the meter color register for match.
903 		 */
904 		if (priv->mtr_reg_share)
905 			return priv->mtr_color_reg;
906 		else
907 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
908 			       REG_C_3;
909 	case MLX5_MTR_COLOR:
910 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
911 		return priv->mtr_color_reg;
912 	case MLX5_COPY_MARK:
913 		/*
914 		 * Metadata COPY_MARK register using is in meter suffix sub
915 		 * flow while with meter. It's safe to share the same register.
916 		 */
917 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
918 	case MLX5_APP_TAG:
919 		/*
920 		 * If meter is enable, it will engage the register for color
921 		 * match and flow match. If meter color match is not using the
922 		 * REG_C_2, need to skip the REG_C_x be used by meter color
923 		 * match.
924 		 * If meter is disable, free to use all available registers.
925 		 */
926 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
927 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
928 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
929 		if (id > (REG_C_7 - start_reg))
930 			return rte_flow_error_set(error, EINVAL,
931 						  RTE_FLOW_ERROR_TYPE_ITEM,
932 						  NULL, "invalid tag id");
933 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
934 			return rte_flow_error_set(error, ENOTSUP,
935 						  RTE_FLOW_ERROR_TYPE_ITEM,
936 						  NULL, "unsupported tag id");
937 		/*
938 		 * This case means meter is using the REG_C_x great than 2.
939 		 * Take care not to conflict with meter color REG_C_x.
940 		 * If the available index REG_C_y >= REG_C_x, skip the
941 		 * color register.
942 		 */
943 		if (skip_mtr_reg && config->flow_mreg_c
944 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
945 			if (id >= (REG_C_7 - start_reg))
946 				return rte_flow_error_set(error, EINVAL,
947 						       RTE_FLOW_ERROR_TYPE_ITEM,
948 							NULL, "invalid tag id");
949 			if (config->flow_mreg_c
950 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
951 				return config->flow_mreg_c
952 					       [id + 1 + start_reg - REG_C_0];
953 			return rte_flow_error_set(error, ENOTSUP,
954 						  RTE_FLOW_ERROR_TYPE_ITEM,
955 						  NULL, "unsupported tag id");
956 		}
957 		return config->flow_mreg_c[id + start_reg - REG_C_0];
958 	}
959 	MLX5_ASSERT(false);
960 	return rte_flow_error_set(error, EINVAL,
961 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
962 				  NULL, "invalid feature name");
963 }
964 
965 /**
966  * Check extensive flow metadata register support.
967  *
968  * @param dev
969  *   Pointer to rte_eth_dev structure.
970  *
971  * @return
972  *   True if device supports extensive flow metadata register, otherwise false.
973  */
974 bool
975 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
976 {
977 	struct mlx5_priv *priv = dev->data->dev_private;
978 	struct mlx5_dev_config *config = &priv->config;
979 
980 	/*
981 	 * Having available reg_c can be regarded inclusively as supporting
982 	 * extensive flow metadata register, which could mean,
983 	 * - metadata register copy action by modify header.
984 	 * - 16 modify header actions is supported.
985 	 * - reg_c's are preserved across different domain (FDB and NIC) on
986 	 *   packet loopback by flow lookup miss.
987 	 */
988 	return config->flow_mreg_c[2] != REG_NON;
989 }
990 
991 /**
992  * Verify the @p item specifications (spec, last, mask) are compatible with the
993  * NIC capabilities.
994  *
995  * @param[in] item
996  *   Item specification.
997  * @param[in] mask
998  *   @p item->mask or flow default bit-masks.
999  * @param[in] nic_mask
1000  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1001  * @param[in] size
1002  *   Bit-masks size in bytes.
1003  * @param[in] range_accepted
1004  *   True if range of values is accepted for specific fields, false otherwise.
1005  * @param[out] error
1006  *   Pointer to error structure.
1007  *
1008  * @return
1009  *   0 on success, a negative errno value otherwise and rte_errno is set.
1010  */
1011 int
1012 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1013 			  const uint8_t *mask,
1014 			  const uint8_t *nic_mask,
1015 			  unsigned int size,
1016 			  bool range_accepted,
1017 			  struct rte_flow_error *error)
1018 {
1019 	unsigned int i;
1020 
1021 	MLX5_ASSERT(nic_mask);
1022 	for (i = 0; i < size; ++i)
1023 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1024 			return rte_flow_error_set(error, ENOTSUP,
1025 						  RTE_FLOW_ERROR_TYPE_ITEM,
1026 						  item,
1027 						  "mask enables non supported"
1028 						  " bits");
1029 	if (!item->spec && (item->mask || item->last))
1030 		return rte_flow_error_set(error, EINVAL,
1031 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1032 					  "mask/last without a spec is not"
1033 					  " supported");
1034 	if (item->spec && item->last && !range_accepted) {
1035 		uint8_t spec[size];
1036 		uint8_t last[size];
1037 		unsigned int i;
1038 		int ret;
1039 
1040 		for (i = 0; i < size; ++i) {
1041 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1042 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1043 		}
1044 		ret = memcmp(spec, last, size);
1045 		if (ret != 0)
1046 			return rte_flow_error_set(error, EINVAL,
1047 						  RTE_FLOW_ERROR_TYPE_ITEM,
1048 						  item,
1049 						  "range is not valid");
1050 	}
1051 	return 0;
1052 }
1053 
1054 /**
1055  * Adjust the hash fields according to the @p flow information.
1056  *
1057  * @param[in] dev_flow.
1058  *   Pointer to the mlx5_flow.
1059  * @param[in] tunnel
1060  *   1 when the hash field is for a tunnel item.
1061  * @param[in] layer_types
1062  *   ETH_RSS_* types.
1063  * @param[in] hash_fields
1064  *   Item hash fields.
1065  *
1066  * @return
1067  *   The hash fields that should be used.
1068  */
1069 uint64_t
1070 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1071 			    int tunnel __rte_unused, uint64_t layer_types,
1072 			    uint64_t hash_fields)
1073 {
1074 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1075 	int rss_request_inner = rss_desc->level >= 2;
1076 
1077 	/* Check RSS hash level for tunnel. */
1078 	if (tunnel && rss_request_inner)
1079 		hash_fields |= IBV_RX_HASH_INNER;
1080 	else if (tunnel || rss_request_inner)
1081 		return 0;
1082 #endif
1083 	/* Check if requested layer matches RSS hash fields. */
1084 	if (!(rss_desc->types & layer_types))
1085 		return 0;
1086 	return hash_fields;
1087 }
1088 
1089 /**
1090  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1091  * if several tunnel rules are used on this queue, the tunnel ptype will be
1092  * cleared.
1093  *
1094  * @param rxq_ctrl
1095  *   Rx queue to update.
1096  */
1097 static void
1098 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1099 {
1100 	unsigned int i;
1101 	uint32_t tunnel_ptype = 0;
1102 
1103 	/* Look up for the ptype to use. */
1104 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1105 		if (!rxq_ctrl->flow_tunnels_n[i])
1106 			continue;
1107 		if (!tunnel_ptype) {
1108 			tunnel_ptype = tunnels_info[i].ptype;
1109 		} else {
1110 			tunnel_ptype = 0;
1111 			break;
1112 		}
1113 	}
1114 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1115 }
1116 
1117 /**
1118  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1119  * flow.
1120  *
1121  * @param[in] dev
1122  *   Pointer to the Ethernet device structure.
1123  * @param[in] dev_handle
1124  *   Pointer to device flow handle structure.
1125  */
1126 static void
1127 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1128 		       struct mlx5_flow_handle *dev_handle)
1129 {
1130 	struct mlx5_priv *priv = dev->data->dev_private;
1131 	const int mark = dev_handle->mark;
1132 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1133 	struct mlx5_hrxq *hrxq;
1134 	unsigned int i;
1135 
1136 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1137 		return;
1138 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1139 			      dev_handle->rix_hrxq);
1140 	if (!hrxq)
1141 		return;
1142 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1143 		int idx = hrxq->ind_table->queues[i];
1144 		struct mlx5_rxq_ctrl *rxq_ctrl =
1145 			container_of((*priv->rxqs)[idx],
1146 				     struct mlx5_rxq_ctrl, rxq);
1147 
1148 		/*
1149 		 * To support metadata register copy on Tx loopback,
1150 		 * this must be always enabled (metadata may arive
1151 		 * from other port - not from local flows only.
1152 		 */
1153 		if (priv->config.dv_flow_en &&
1154 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1155 		    mlx5_flow_ext_mreg_supported(dev)) {
1156 			rxq_ctrl->rxq.mark = 1;
1157 			rxq_ctrl->flow_mark_n = 1;
1158 		} else if (mark) {
1159 			rxq_ctrl->rxq.mark = 1;
1160 			rxq_ctrl->flow_mark_n++;
1161 		}
1162 		if (tunnel) {
1163 			unsigned int j;
1164 
1165 			/* Increase the counter matching the flow. */
1166 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1167 				if ((tunnels_info[j].tunnel &
1168 				     dev_handle->layers) ==
1169 				    tunnels_info[j].tunnel) {
1170 					rxq_ctrl->flow_tunnels_n[j]++;
1171 					break;
1172 				}
1173 			}
1174 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1175 		}
1176 	}
1177 }
1178 
1179 /**
1180  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1181  *
1182  * @param[in] dev
1183  *   Pointer to the Ethernet device structure.
1184  * @param[in] flow
1185  *   Pointer to flow structure.
1186  */
1187 static void
1188 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1189 {
1190 	struct mlx5_priv *priv = dev->data->dev_private;
1191 	uint32_t handle_idx;
1192 	struct mlx5_flow_handle *dev_handle;
1193 
1194 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1195 		       handle_idx, dev_handle, next)
1196 		flow_drv_rxq_flags_set(dev, dev_handle);
1197 }
1198 
1199 /**
1200  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1201  * device flow if no other flow uses it with the same kind of request.
1202  *
1203  * @param dev
1204  *   Pointer to Ethernet device.
1205  * @param[in] dev_handle
1206  *   Pointer to the device flow handle structure.
1207  */
1208 static void
1209 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1210 			struct mlx5_flow_handle *dev_handle)
1211 {
1212 	struct mlx5_priv *priv = dev->data->dev_private;
1213 	const int mark = dev_handle->mark;
1214 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1215 	struct mlx5_hrxq *hrxq;
1216 	unsigned int i;
1217 
1218 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1219 		return;
1220 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1221 			      dev_handle->rix_hrxq);
1222 	if (!hrxq)
1223 		return;
1224 	MLX5_ASSERT(dev->data->dev_started);
1225 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1226 		int idx = hrxq->ind_table->queues[i];
1227 		struct mlx5_rxq_ctrl *rxq_ctrl =
1228 			container_of((*priv->rxqs)[idx],
1229 				     struct mlx5_rxq_ctrl, rxq);
1230 
1231 		if (priv->config.dv_flow_en &&
1232 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1233 		    mlx5_flow_ext_mreg_supported(dev)) {
1234 			rxq_ctrl->rxq.mark = 1;
1235 			rxq_ctrl->flow_mark_n = 1;
1236 		} else if (mark) {
1237 			rxq_ctrl->flow_mark_n--;
1238 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1239 		}
1240 		if (tunnel) {
1241 			unsigned int j;
1242 
1243 			/* Decrease the counter matching the flow. */
1244 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1245 				if ((tunnels_info[j].tunnel &
1246 				     dev_handle->layers) ==
1247 				    tunnels_info[j].tunnel) {
1248 					rxq_ctrl->flow_tunnels_n[j]--;
1249 					break;
1250 				}
1251 			}
1252 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1253 		}
1254 	}
1255 }
1256 
1257 /**
1258  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1259  * @p flow if no other flow uses it with the same kind of request.
1260  *
1261  * @param dev
1262  *   Pointer to Ethernet device.
1263  * @param[in] flow
1264  *   Pointer to the flow.
1265  */
1266 static void
1267 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1268 {
1269 	struct mlx5_priv *priv = dev->data->dev_private;
1270 	uint32_t handle_idx;
1271 	struct mlx5_flow_handle *dev_handle;
1272 
1273 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1274 		       handle_idx, dev_handle, next)
1275 		flow_drv_rxq_flags_trim(dev, dev_handle);
1276 }
1277 
1278 /**
1279  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1280  *
1281  * @param dev
1282  *   Pointer to Ethernet device.
1283  */
1284 static void
1285 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1286 {
1287 	struct mlx5_priv *priv = dev->data->dev_private;
1288 	unsigned int i;
1289 
1290 	for (i = 0; i != priv->rxqs_n; ++i) {
1291 		struct mlx5_rxq_ctrl *rxq_ctrl;
1292 		unsigned int j;
1293 
1294 		if (!(*priv->rxqs)[i])
1295 			continue;
1296 		rxq_ctrl = container_of((*priv->rxqs)[i],
1297 					struct mlx5_rxq_ctrl, rxq);
1298 		rxq_ctrl->flow_mark_n = 0;
1299 		rxq_ctrl->rxq.mark = 0;
1300 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1301 			rxq_ctrl->flow_tunnels_n[j] = 0;
1302 		rxq_ctrl->rxq.tunnel = 0;
1303 	}
1304 }
1305 
1306 /**
1307  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1308  *
1309  * @param[in] dev
1310  *   Pointer to the Ethernet device structure.
1311  */
1312 void
1313 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1314 {
1315 	struct mlx5_priv *priv = dev->data->dev_private;
1316 	struct mlx5_rxq_data *data;
1317 	unsigned int i;
1318 
1319 	for (i = 0; i != priv->rxqs_n; ++i) {
1320 		if (!(*priv->rxqs)[i])
1321 			continue;
1322 		data = (*priv->rxqs)[i];
1323 		if (!rte_flow_dynf_metadata_avail()) {
1324 			data->dynf_meta = 0;
1325 			data->flow_meta_mask = 0;
1326 			data->flow_meta_offset = -1;
1327 		} else {
1328 			data->dynf_meta = 1;
1329 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1330 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1331 		}
1332 	}
1333 }
1334 
1335 /*
1336  * return a pointer to the desired action in the list of actions.
1337  *
1338  * @param[in] actions
1339  *   The list of actions to search the action in.
1340  * @param[in] action
1341  *   The action to find.
1342  *
1343  * @return
1344  *   Pointer to the action in the list, if found. NULL otherwise.
1345  */
1346 const struct rte_flow_action *
1347 mlx5_flow_find_action(const struct rte_flow_action *actions,
1348 		      enum rte_flow_action_type action)
1349 {
1350 	if (actions == NULL)
1351 		return NULL;
1352 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1353 		if (actions->type == action)
1354 			return actions;
1355 	return NULL;
1356 }
1357 
1358 /*
1359  * Validate the flag action.
1360  *
1361  * @param[in] action_flags
1362  *   Bit-fields that holds the actions detected until now.
1363  * @param[in] attr
1364  *   Attributes of flow that includes this action.
1365  * @param[out] error
1366  *   Pointer to error structure.
1367  *
1368  * @return
1369  *   0 on success, a negative errno value otherwise and rte_errno is set.
1370  */
1371 int
1372 mlx5_flow_validate_action_flag(uint64_t action_flags,
1373 			       const struct rte_flow_attr *attr,
1374 			       struct rte_flow_error *error)
1375 {
1376 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1377 		return rte_flow_error_set(error, EINVAL,
1378 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1379 					  "can't mark and flag in same flow");
1380 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1381 		return rte_flow_error_set(error, EINVAL,
1382 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1383 					  "can't have 2 flag"
1384 					  " actions in same flow");
1385 	if (attr->egress)
1386 		return rte_flow_error_set(error, ENOTSUP,
1387 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1388 					  "flag action not supported for "
1389 					  "egress");
1390 	return 0;
1391 }
1392 
1393 /*
1394  * Validate the mark action.
1395  *
1396  * @param[in] action
1397  *   Pointer to the queue action.
1398  * @param[in] action_flags
1399  *   Bit-fields that holds the actions detected until now.
1400  * @param[in] attr
1401  *   Attributes of flow that includes this action.
1402  * @param[out] error
1403  *   Pointer to error structure.
1404  *
1405  * @return
1406  *   0 on success, a negative errno value otherwise and rte_errno is set.
1407  */
1408 int
1409 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1410 			       uint64_t action_flags,
1411 			       const struct rte_flow_attr *attr,
1412 			       struct rte_flow_error *error)
1413 {
1414 	const struct rte_flow_action_mark *mark = action->conf;
1415 
1416 	if (!mark)
1417 		return rte_flow_error_set(error, EINVAL,
1418 					  RTE_FLOW_ERROR_TYPE_ACTION,
1419 					  action,
1420 					  "configuration cannot be null");
1421 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1422 		return rte_flow_error_set(error, EINVAL,
1423 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1424 					  &mark->id,
1425 					  "mark id must in 0 <= id < "
1426 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1427 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1428 		return rte_flow_error_set(error, EINVAL,
1429 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1430 					  "can't flag and mark in same flow");
1431 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1432 		return rte_flow_error_set(error, EINVAL,
1433 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1434 					  "can't have 2 mark actions in same"
1435 					  " flow");
1436 	if (attr->egress)
1437 		return rte_flow_error_set(error, ENOTSUP,
1438 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1439 					  "mark action not supported for "
1440 					  "egress");
1441 	return 0;
1442 }
1443 
1444 /*
1445  * Validate the drop action.
1446  *
1447  * @param[in] action_flags
1448  *   Bit-fields that holds the actions detected until now.
1449  * @param[in] attr
1450  *   Attributes of flow that includes this action.
1451  * @param[out] error
1452  *   Pointer to error structure.
1453  *
1454  * @return
1455  *   0 on success, a negative errno value otherwise and rte_errno is set.
1456  */
1457 int
1458 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1459 			       const struct rte_flow_attr *attr,
1460 			       struct rte_flow_error *error)
1461 {
1462 	if (attr->egress)
1463 		return rte_flow_error_set(error, ENOTSUP,
1464 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1465 					  "drop action not supported for "
1466 					  "egress");
1467 	return 0;
1468 }
1469 
1470 /*
1471  * Validate the queue action.
1472  *
1473  * @param[in] action
1474  *   Pointer to the queue action.
1475  * @param[in] action_flags
1476  *   Bit-fields that holds the actions detected until now.
1477  * @param[in] dev
1478  *   Pointer to the Ethernet device structure.
1479  * @param[in] attr
1480  *   Attributes of flow that includes this action.
1481  * @param[out] error
1482  *   Pointer to error structure.
1483  *
1484  * @return
1485  *   0 on success, a negative errno value otherwise and rte_errno is set.
1486  */
1487 int
1488 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1489 				uint64_t action_flags,
1490 				struct rte_eth_dev *dev,
1491 				const struct rte_flow_attr *attr,
1492 				struct rte_flow_error *error)
1493 {
1494 	struct mlx5_priv *priv = dev->data->dev_private;
1495 	const struct rte_flow_action_queue *queue = action->conf;
1496 
1497 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1498 		return rte_flow_error_set(error, EINVAL,
1499 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1500 					  "can't have 2 fate actions in"
1501 					  " same flow");
1502 	if (!priv->rxqs_n)
1503 		return rte_flow_error_set(error, EINVAL,
1504 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1505 					  NULL, "No Rx queues configured");
1506 	if (queue->index >= priv->rxqs_n)
1507 		return rte_flow_error_set(error, EINVAL,
1508 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1509 					  &queue->index,
1510 					  "queue index out of range");
1511 	if (!(*priv->rxqs)[queue->index])
1512 		return rte_flow_error_set(error, EINVAL,
1513 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1514 					  &queue->index,
1515 					  "queue is not configured");
1516 	if (attr->egress)
1517 		return rte_flow_error_set(error, ENOTSUP,
1518 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1519 					  "queue action not supported for "
1520 					  "egress");
1521 	return 0;
1522 }
1523 
1524 /*
1525  * Validate the rss action.
1526  *
1527  * @param[in] dev
1528  *   Pointer to the Ethernet device structure.
1529  * @param[in] action
1530  *   Pointer to the queue action.
1531  * @param[out] error
1532  *   Pointer to error structure.
1533  *
1534  * @return
1535  *   0 on success, a negative errno value otherwise and rte_errno is set.
1536  */
1537 int
1538 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1539 			 const struct rte_flow_action *action,
1540 			 struct rte_flow_error *error)
1541 {
1542 	struct mlx5_priv *priv = dev->data->dev_private;
1543 	const struct rte_flow_action_rss *rss = action->conf;
1544 	unsigned int i;
1545 
1546 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1547 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1548 		return rte_flow_error_set(error, ENOTSUP,
1549 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1550 					  &rss->func,
1551 					  "RSS hash function not supported");
1552 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1553 	if (rss->level > 2)
1554 #else
1555 	if (rss->level > 1)
1556 #endif
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1559 					  &rss->level,
1560 					  "tunnel RSS is not supported");
1561 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1562 	if (rss->key_len == 0 && rss->key != NULL)
1563 		return rte_flow_error_set(error, ENOTSUP,
1564 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1565 					  &rss->key_len,
1566 					  "RSS hash key length 0");
1567 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1568 		return rte_flow_error_set(error, ENOTSUP,
1569 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1570 					  &rss->key_len,
1571 					  "RSS hash key too small");
1572 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1573 		return rte_flow_error_set(error, ENOTSUP,
1574 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1575 					  &rss->key_len,
1576 					  "RSS hash key too large");
1577 	if (rss->queue_num > priv->config.ind_table_max_size)
1578 		return rte_flow_error_set(error, ENOTSUP,
1579 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1580 					  &rss->queue_num,
1581 					  "number of queues too large");
1582 	if (rss->types & MLX5_RSS_HF_MASK)
1583 		return rte_flow_error_set(error, ENOTSUP,
1584 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1585 					  &rss->types,
1586 					  "some RSS protocols are not"
1587 					  " supported");
1588 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1589 	    !(rss->types & ETH_RSS_IP))
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1592 					  "L3 partial RSS requested but L3 RSS"
1593 					  " type not specified");
1594 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1595 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1596 		return rte_flow_error_set(error, EINVAL,
1597 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1598 					  "L4 partial RSS requested but L4 RSS"
1599 					  " type not specified");
1600 	if (!priv->rxqs_n)
1601 		return rte_flow_error_set(error, EINVAL,
1602 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1603 					  NULL, "No Rx queues configured");
1604 	if (!rss->queue_num)
1605 		return rte_flow_error_set(error, EINVAL,
1606 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1607 					  NULL, "No queues configured");
1608 	for (i = 0; i != rss->queue_num; ++i) {
1609 		if (rss->queue[i] >= priv->rxqs_n)
1610 			return rte_flow_error_set
1611 				(error, EINVAL,
1612 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1613 				 &rss->queue[i], "queue index out of range");
1614 		if (!(*priv->rxqs)[rss->queue[i]])
1615 			return rte_flow_error_set
1616 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1617 				 &rss->queue[i], "queue is not configured");
1618 	}
1619 	return 0;
1620 }
1621 
1622 /*
1623  * Validate the rss action.
1624  *
1625  * @param[in] action
1626  *   Pointer to the queue action.
1627  * @param[in] action_flags
1628  *   Bit-fields that holds the actions detected until now.
1629  * @param[in] dev
1630  *   Pointer to the Ethernet device structure.
1631  * @param[in] attr
1632  *   Attributes of flow that includes this action.
1633  * @param[in] item_flags
1634  *   Items that were detected.
1635  * @param[out] error
1636  *   Pointer to error structure.
1637  *
1638  * @return
1639  *   0 on success, a negative errno value otherwise and rte_errno is set.
1640  */
1641 int
1642 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1643 			      uint64_t action_flags,
1644 			      struct rte_eth_dev *dev,
1645 			      const struct rte_flow_attr *attr,
1646 			      uint64_t item_flags,
1647 			      struct rte_flow_error *error)
1648 {
1649 	const struct rte_flow_action_rss *rss = action->conf;
1650 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1651 	int ret;
1652 
1653 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1654 		return rte_flow_error_set(error, EINVAL,
1655 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1656 					  "can't have 2 fate actions"
1657 					  " in same flow");
1658 	ret = mlx5_validate_action_rss(dev, action, error);
1659 	if (ret)
1660 		return ret;
1661 	if (attr->egress)
1662 		return rte_flow_error_set(error, ENOTSUP,
1663 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1664 					  "rss action not supported for "
1665 					  "egress");
1666 	if (rss->level > 1 && !tunnel)
1667 		return rte_flow_error_set(error, EINVAL,
1668 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1669 					  "inner RSS is not supported for "
1670 					  "non-tunnel flows");
1671 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1672 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1673 		return rte_flow_error_set(error, EINVAL,
1674 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1675 					  "RSS on eCPRI is not supported now");
1676 	}
1677 	return 0;
1678 }
1679 
1680 /*
1681  * Validate the default miss action.
1682  *
1683  * @param[in] action_flags
1684  *   Bit-fields that holds the actions detected until now.
1685  * @param[out] error
1686  *   Pointer to error structure.
1687  *
1688  * @return
1689  *   0 on success, a negative errno value otherwise and rte_errno is set.
1690  */
1691 int
1692 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1693 				const struct rte_flow_attr *attr,
1694 				struct rte_flow_error *error)
1695 {
1696 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1697 		return rte_flow_error_set(error, EINVAL,
1698 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1699 					  "can't have 2 fate actions in"
1700 					  " same flow");
1701 	if (attr->egress)
1702 		return rte_flow_error_set(error, ENOTSUP,
1703 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1704 					  "default miss action not supported "
1705 					  "for egress");
1706 	if (attr->group)
1707 		return rte_flow_error_set(error, ENOTSUP,
1708 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1709 					  "only group 0 is supported");
1710 	if (attr->transfer)
1711 		return rte_flow_error_set(error, ENOTSUP,
1712 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1713 					  NULL, "transfer is not supported");
1714 	return 0;
1715 }
1716 
1717 /*
1718  * Validate the count action.
1719  *
1720  * @param[in] dev
1721  *   Pointer to the Ethernet device structure.
1722  * @param[in] attr
1723  *   Attributes of flow that includes this action.
1724  * @param[out] error
1725  *   Pointer to error structure.
1726  *
1727  * @return
1728  *   0 on success, a negative errno value otherwise and rte_errno is set.
1729  */
1730 int
1731 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1732 				const struct rte_flow_attr *attr,
1733 				struct rte_flow_error *error)
1734 {
1735 	if (attr->egress)
1736 		return rte_flow_error_set(error, ENOTSUP,
1737 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1738 					  "count action not supported for "
1739 					  "egress");
1740 	return 0;
1741 }
1742 
1743 /**
1744  * Verify the @p attributes will be correctly understood by the NIC and store
1745  * them in the @p flow if everything is correct.
1746  *
1747  * @param[in] dev
1748  *   Pointer to the Ethernet device structure.
1749  * @param[in] attributes
1750  *   Pointer to flow attributes
1751  * @param[out] error
1752  *   Pointer to error structure.
1753  *
1754  * @return
1755  *   0 on success, a negative errno value otherwise and rte_errno is set.
1756  */
1757 int
1758 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1759 			      const struct rte_flow_attr *attributes,
1760 			      struct rte_flow_error *error)
1761 {
1762 	struct mlx5_priv *priv = dev->data->dev_private;
1763 	uint32_t priority_max = priv->config.flow_prio - 1;
1764 
1765 	if (attributes->group)
1766 		return rte_flow_error_set(error, ENOTSUP,
1767 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1768 					  NULL, "groups is not supported");
1769 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1770 	    attributes->priority >= priority_max)
1771 		return rte_flow_error_set(error, ENOTSUP,
1772 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1773 					  NULL, "priority out of range");
1774 	if (attributes->egress)
1775 		return rte_flow_error_set(error, ENOTSUP,
1776 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1777 					  "egress is not supported");
1778 	if (attributes->transfer && !priv->config.dv_esw_en)
1779 		return rte_flow_error_set(error, ENOTSUP,
1780 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1781 					  NULL, "transfer is not supported");
1782 	if (!attributes->ingress)
1783 		return rte_flow_error_set(error, EINVAL,
1784 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1785 					  NULL,
1786 					  "ingress attribute is mandatory");
1787 	return 0;
1788 }
1789 
1790 /**
1791  * Validate ICMP6 item.
1792  *
1793  * @param[in] item
1794  *   Item specification.
1795  * @param[in] item_flags
1796  *   Bit-fields that holds the items detected until now.
1797  * @param[in] ext_vlan_sup
1798  *   Whether extended VLAN features are supported or not.
1799  * @param[out] error
1800  *   Pointer to error structure.
1801  *
1802  * @return
1803  *   0 on success, a negative errno value otherwise and rte_errno is set.
1804  */
1805 int
1806 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1807 			       uint64_t item_flags,
1808 			       uint8_t target_protocol,
1809 			       struct rte_flow_error *error)
1810 {
1811 	const struct rte_flow_item_icmp6 *mask = item->mask;
1812 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1813 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1814 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1815 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1816 				      MLX5_FLOW_LAYER_OUTER_L4;
1817 	int ret;
1818 
1819 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1820 		return rte_flow_error_set(error, EINVAL,
1821 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1822 					  "protocol filtering not compatible"
1823 					  " with ICMP6 layer");
1824 	if (!(item_flags & l3m))
1825 		return rte_flow_error_set(error, EINVAL,
1826 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1827 					  "IPv6 is mandatory to filter on"
1828 					  " ICMP6");
1829 	if (item_flags & l4m)
1830 		return rte_flow_error_set(error, EINVAL,
1831 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1832 					  "multiple L4 layers not supported");
1833 	if (!mask)
1834 		mask = &rte_flow_item_icmp6_mask;
1835 	ret = mlx5_flow_item_acceptable
1836 		(item, (const uint8_t *)mask,
1837 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1838 		 sizeof(struct rte_flow_item_icmp6),
1839 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1840 	if (ret < 0)
1841 		return ret;
1842 	return 0;
1843 }
1844 
1845 /**
1846  * Validate ICMP item.
1847  *
1848  * @param[in] item
1849  *   Item specification.
1850  * @param[in] item_flags
1851  *   Bit-fields that holds the items detected until now.
1852  * @param[out] error
1853  *   Pointer to error structure.
1854  *
1855  * @return
1856  *   0 on success, a negative errno value otherwise and rte_errno is set.
1857  */
1858 int
1859 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1860 			     uint64_t item_flags,
1861 			     uint8_t target_protocol,
1862 			     struct rte_flow_error *error)
1863 {
1864 	const struct rte_flow_item_icmp *mask = item->mask;
1865 	const struct rte_flow_item_icmp nic_mask = {
1866 		.hdr.icmp_type = 0xff,
1867 		.hdr.icmp_code = 0xff,
1868 		.hdr.icmp_ident = RTE_BE16(0xffff),
1869 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1870 	};
1871 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1872 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1873 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1874 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1875 				      MLX5_FLOW_LAYER_OUTER_L4;
1876 	int ret;
1877 
1878 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1879 		return rte_flow_error_set(error, EINVAL,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "protocol filtering not compatible"
1882 					  " with ICMP layer");
1883 	if (!(item_flags & l3m))
1884 		return rte_flow_error_set(error, EINVAL,
1885 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1886 					  "IPv4 is mandatory to filter"
1887 					  " on ICMP");
1888 	if (item_flags & l4m)
1889 		return rte_flow_error_set(error, EINVAL,
1890 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1891 					  "multiple L4 layers not supported");
1892 	if (!mask)
1893 		mask = &nic_mask;
1894 	ret = mlx5_flow_item_acceptable
1895 		(item, (const uint8_t *)mask,
1896 		 (const uint8_t *)&nic_mask,
1897 		 sizeof(struct rte_flow_item_icmp),
1898 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1899 	if (ret < 0)
1900 		return ret;
1901 	return 0;
1902 }
1903 
1904 /**
1905  * Validate Ethernet item.
1906  *
1907  * @param[in] item
1908  *   Item specification.
1909  * @param[in] item_flags
1910  *   Bit-fields that holds the items detected until now.
1911  * @param[out] error
1912  *   Pointer to error structure.
1913  *
1914  * @return
1915  *   0 on success, a negative errno value otherwise and rte_errno is set.
1916  */
1917 int
1918 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1919 			    uint64_t item_flags, bool ext_vlan_sup,
1920 			    struct rte_flow_error *error)
1921 {
1922 	const struct rte_flow_item_eth *mask = item->mask;
1923 	const struct rte_flow_item_eth nic_mask = {
1924 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1925 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1926 		.type = RTE_BE16(0xffff),
1927 		.has_vlan = ext_vlan_sup ? 1 : 0,
1928 	};
1929 	int ret;
1930 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1931 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1932 				       MLX5_FLOW_LAYER_OUTER_L2;
1933 
1934 	if (item_flags & ethm)
1935 		return rte_flow_error_set(error, ENOTSUP,
1936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1937 					  "multiple L2 layers not supported");
1938 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1939 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1940 		return rte_flow_error_set(error, EINVAL,
1941 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1942 					  "L2 layer should not follow "
1943 					  "L3 layers");
1944 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1945 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1946 		return rte_flow_error_set(error, EINVAL,
1947 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1948 					  "L2 layer should not follow VLAN");
1949 	if (!mask)
1950 		mask = &rte_flow_item_eth_mask;
1951 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1952 					(const uint8_t *)&nic_mask,
1953 					sizeof(struct rte_flow_item_eth),
1954 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1955 	return ret;
1956 }
1957 
1958 /**
1959  * Validate VLAN item.
1960  *
1961  * @param[in] item
1962  *   Item specification.
1963  * @param[in] item_flags
1964  *   Bit-fields that holds the items detected until now.
1965  * @param[in] dev
1966  *   Ethernet device flow is being created on.
1967  * @param[out] error
1968  *   Pointer to error structure.
1969  *
1970  * @return
1971  *   0 on success, a negative errno value otherwise and rte_errno is set.
1972  */
1973 int
1974 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1975 			     uint64_t item_flags,
1976 			     struct rte_eth_dev *dev,
1977 			     struct rte_flow_error *error)
1978 {
1979 	const struct rte_flow_item_vlan *spec = item->spec;
1980 	const struct rte_flow_item_vlan *mask = item->mask;
1981 	const struct rte_flow_item_vlan nic_mask = {
1982 		.tci = RTE_BE16(UINT16_MAX),
1983 		.inner_type = RTE_BE16(UINT16_MAX),
1984 	};
1985 	uint16_t vlan_tag = 0;
1986 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1987 	int ret;
1988 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1989 					MLX5_FLOW_LAYER_INNER_L4) :
1990 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1991 					MLX5_FLOW_LAYER_OUTER_L4);
1992 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1993 					MLX5_FLOW_LAYER_OUTER_VLAN;
1994 
1995 	if (item_flags & vlanm)
1996 		return rte_flow_error_set(error, EINVAL,
1997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 					  "multiple VLAN layers not supported");
1999 	else if ((item_flags & l34m) != 0)
2000 		return rte_flow_error_set(error, EINVAL,
2001 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2002 					  "VLAN cannot follow L3/L4 layer");
2003 	if (!mask)
2004 		mask = &rte_flow_item_vlan_mask;
2005 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2006 					(const uint8_t *)&nic_mask,
2007 					sizeof(struct rte_flow_item_vlan),
2008 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2009 	if (ret)
2010 		return ret;
2011 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2012 		struct mlx5_priv *priv = dev->data->dev_private;
2013 
2014 		if (priv->vmwa_context) {
2015 			/*
2016 			 * Non-NULL context means we have a virtual machine
2017 			 * and SR-IOV enabled, we have to create VLAN interface
2018 			 * to make hypervisor to setup E-Switch vport
2019 			 * context correctly. We avoid creating the multiple
2020 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2021 			 */
2022 			return rte_flow_error_set(error, EINVAL,
2023 						  RTE_FLOW_ERROR_TYPE_ITEM,
2024 						  item,
2025 						  "VLAN tag mask is not"
2026 						  " supported in virtual"
2027 						  " environment");
2028 		}
2029 	}
2030 	if (spec) {
2031 		vlan_tag = spec->tci;
2032 		vlan_tag &= mask->tci;
2033 	}
2034 	/*
2035 	 * From verbs perspective an empty VLAN is equivalent
2036 	 * to a packet without VLAN layer.
2037 	 */
2038 	if (!vlan_tag)
2039 		return rte_flow_error_set(error, EINVAL,
2040 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2041 					  item->spec,
2042 					  "VLAN cannot be empty");
2043 	return 0;
2044 }
2045 
2046 /**
2047  * Validate IPV4 item.
2048  *
2049  * @param[in] item
2050  *   Item specification.
2051  * @param[in] item_flags
2052  *   Bit-fields that holds the items detected until now.
2053  * @param[in] last_item
2054  *   Previous validated item in the pattern items.
2055  * @param[in] ether_type
2056  *   Type in the ethernet layer header (including dot1q).
2057  * @param[in] acc_mask
2058  *   Acceptable mask, if NULL default internal default mask
2059  *   will be used to check whether item fields are supported.
2060  * @param[in] range_accepted
2061  *   True if range of values is accepted for specific fields, false otherwise.
2062  * @param[out] error
2063  *   Pointer to error structure.
2064  *
2065  * @return
2066  *   0 on success, a negative errno value otherwise and rte_errno is set.
2067  */
2068 int
2069 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2070 			     uint64_t item_flags,
2071 			     uint64_t last_item,
2072 			     uint16_t ether_type,
2073 			     const struct rte_flow_item_ipv4 *acc_mask,
2074 			     bool range_accepted,
2075 			     struct rte_flow_error *error)
2076 {
2077 	const struct rte_flow_item_ipv4 *mask = item->mask;
2078 	const struct rte_flow_item_ipv4 *spec = item->spec;
2079 	const struct rte_flow_item_ipv4 nic_mask = {
2080 		.hdr = {
2081 			.src_addr = RTE_BE32(0xffffffff),
2082 			.dst_addr = RTE_BE32(0xffffffff),
2083 			.type_of_service = 0xff,
2084 			.next_proto_id = 0xff,
2085 		},
2086 	};
2087 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2088 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2089 				      MLX5_FLOW_LAYER_OUTER_L3;
2090 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2091 				      MLX5_FLOW_LAYER_OUTER_L4;
2092 	int ret;
2093 	uint8_t next_proto = 0xFF;
2094 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2095 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2096 				  MLX5_FLOW_LAYER_INNER_VLAN);
2097 
2098 	if ((last_item & l2_vlan) && ether_type &&
2099 	    ether_type != RTE_ETHER_TYPE_IPV4)
2100 		return rte_flow_error_set(error, EINVAL,
2101 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2102 					  "IPv4 cannot follow L2/VLAN layer "
2103 					  "which ether type is not IPv4");
2104 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2105 		if (mask && spec)
2106 			next_proto = mask->hdr.next_proto_id &
2107 				     spec->hdr.next_proto_id;
2108 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2109 			return rte_flow_error_set(error, EINVAL,
2110 						  RTE_FLOW_ERROR_TYPE_ITEM,
2111 						  item,
2112 						  "multiple tunnel "
2113 						  "not supported");
2114 	}
2115 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2116 		return rte_flow_error_set(error, EINVAL,
2117 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2118 					  "wrong tunnel type - IPv6 specified "
2119 					  "but IPv4 item provided");
2120 	if (item_flags & l3m)
2121 		return rte_flow_error_set(error, ENOTSUP,
2122 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2123 					  "multiple L3 layers not supported");
2124 	else if (item_flags & l4m)
2125 		return rte_flow_error_set(error, EINVAL,
2126 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2127 					  "L3 cannot follow an L4 layer.");
2128 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2129 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2130 		return rte_flow_error_set(error, EINVAL,
2131 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2132 					  "L3 cannot follow an NVGRE layer.");
2133 	if (!mask)
2134 		mask = &rte_flow_item_ipv4_mask;
2135 	else if (mask->hdr.next_proto_id != 0 &&
2136 		 mask->hdr.next_proto_id != 0xff)
2137 		return rte_flow_error_set(error, EINVAL,
2138 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2139 					  "partial mask is not supported"
2140 					  " for protocol");
2141 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2142 					acc_mask ? (const uint8_t *)acc_mask
2143 						 : (const uint8_t *)&nic_mask,
2144 					sizeof(struct rte_flow_item_ipv4),
2145 					range_accepted, error);
2146 	if (ret < 0)
2147 		return ret;
2148 	return 0;
2149 }
2150 
2151 /**
2152  * Validate IPV6 item.
2153  *
2154  * @param[in] item
2155  *   Item specification.
2156  * @param[in] item_flags
2157  *   Bit-fields that holds the items detected until now.
2158  * @param[in] last_item
2159  *   Previous validated item in the pattern items.
2160  * @param[in] ether_type
2161  *   Type in the ethernet layer header (including dot1q).
2162  * @param[in] acc_mask
2163  *   Acceptable mask, if NULL default internal default mask
2164  *   will be used to check whether item fields are supported.
2165  * @param[out] error
2166  *   Pointer to error structure.
2167  *
2168  * @return
2169  *   0 on success, a negative errno value otherwise and rte_errno is set.
2170  */
2171 int
2172 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2173 			     uint64_t item_flags,
2174 			     uint64_t last_item,
2175 			     uint16_t ether_type,
2176 			     const struct rte_flow_item_ipv6 *acc_mask,
2177 			     struct rte_flow_error *error)
2178 {
2179 	const struct rte_flow_item_ipv6 *mask = item->mask;
2180 	const struct rte_flow_item_ipv6 *spec = item->spec;
2181 	const struct rte_flow_item_ipv6 nic_mask = {
2182 		.hdr = {
2183 			.src_addr =
2184 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2185 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2186 			.dst_addr =
2187 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2188 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2189 			.vtc_flow = RTE_BE32(0xffffffff),
2190 			.proto = 0xff,
2191 		},
2192 	};
2193 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2194 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2195 				      MLX5_FLOW_LAYER_OUTER_L3;
2196 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2197 				      MLX5_FLOW_LAYER_OUTER_L4;
2198 	int ret;
2199 	uint8_t next_proto = 0xFF;
2200 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2201 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2202 				  MLX5_FLOW_LAYER_INNER_VLAN);
2203 
2204 	if ((last_item & l2_vlan) && ether_type &&
2205 	    ether_type != RTE_ETHER_TYPE_IPV6)
2206 		return rte_flow_error_set(error, EINVAL,
2207 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2208 					  "IPv6 cannot follow L2/VLAN layer "
2209 					  "which ether type is not IPv6");
2210 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2211 		next_proto = spec->hdr.proto;
2212 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2213 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2214 			return rte_flow_error_set(error, EINVAL,
2215 						  RTE_FLOW_ERROR_TYPE_ITEM,
2216 						  item,
2217 						  "multiple tunnel "
2218 						  "not supported");
2219 	}
2220 	if (next_proto == IPPROTO_HOPOPTS  ||
2221 	    next_proto == IPPROTO_ROUTING  ||
2222 	    next_proto == IPPROTO_FRAGMENT ||
2223 	    next_proto == IPPROTO_ESP	   ||
2224 	    next_proto == IPPROTO_AH	   ||
2225 	    next_proto == IPPROTO_DSTOPTS)
2226 		return rte_flow_error_set(error, EINVAL,
2227 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2228 					  "IPv6 proto (next header) should "
2229 					  "not be set as extension header");
2230 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2231 		return rte_flow_error_set(error, EINVAL,
2232 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 					  "wrong tunnel type - IPv4 specified "
2234 					  "but IPv6 item provided");
2235 	if (item_flags & l3m)
2236 		return rte_flow_error_set(error, ENOTSUP,
2237 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2238 					  "multiple L3 layers not supported");
2239 	else if (item_flags & l4m)
2240 		return rte_flow_error_set(error, EINVAL,
2241 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2242 					  "L3 cannot follow an L4 layer.");
2243 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2244 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2245 		return rte_flow_error_set(error, EINVAL,
2246 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2247 					  "L3 cannot follow an NVGRE layer.");
2248 	if (!mask)
2249 		mask = &rte_flow_item_ipv6_mask;
2250 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2251 					acc_mask ? (const uint8_t *)acc_mask
2252 						 : (const uint8_t *)&nic_mask,
2253 					sizeof(struct rte_flow_item_ipv6),
2254 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2255 	if (ret < 0)
2256 		return ret;
2257 	return 0;
2258 }
2259 
2260 /**
2261  * Validate UDP item.
2262  *
2263  * @param[in] item
2264  *   Item specification.
2265  * @param[in] item_flags
2266  *   Bit-fields that holds the items detected until now.
2267  * @param[in] target_protocol
2268  *   The next protocol in the previous item.
2269  * @param[in] flow_mask
2270  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2271  * @param[out] error
2272  *   Pointer to error structure.
2273  *
2274  * @return
2275  *   0 on success, a negative errno value otherwise and rte_errno is set.
2276  */
2277 int
2278 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2279 			    uint64_t item_flags,
2280 			    uint8_t target_protocol,
2281 			    struct rte_flow_error *error)
2282 {
2283 	const struct rte_flow_item_udp *mask = item->mask;
2284 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2285 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2286 				      MLX5_FLOW_LAYER_OUTER_L3;
2287 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2288 				      MLX5_FLOW_LAYER_OUTER_L4;
2289 	int ret;
2290 
2291 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2292 		return rte_flow_error_set(error, EINVAL,
2293 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2294 					  "protocol filtering not compatible"
2295 					  " with UDP layer");
2296 	if (!(item_flags & l3m))
2297 		return rte_flow_error_set(error, EINVAL,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "L3 is mandatory to filter on L4");
2300 	if (item_flags & l4m)
2301 		return rte_flow_error_set(error, EINVAL,
2302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 					  "multiple L4 layers not supported");
2304 	if (!mask)
2305 		mask = &rte_flow_item_udp_mask;
2306 	ret = mlx5_flow_item_acceptable
2307 		(item, (const uint8_t *)mask,
2308 		 (const uint8_t *)&rte_flow_item_udp_mask,
2309 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2310 		 error);
2311 	if (ret < 0)
2312 		return ret;
2313 	return 0;
2314 }
2315 
2316 /**
2317  * Validate TCP item.
2318  *
2319  * @param[in] item
2320  *   Item specification.
2321  * @param[in] item_flags
2322  *   Bit-fields that holds the items detected until now.
2323  * @param[in] target_protocol
2324  *   The next protocol in the previous item.
2325  * @param[out] error
2326  *   Pointer to error structure.
2327  *
2328  * @return
2329  *   0 on success, a negative errno value otherwise and rte_errno is set.
2330  */
2331 int
2332 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2333 			    uint64_t item_flags,
2334 			    uint8_t target_protocol,
2335 			    const struct rte_flow_item_tcp *flow_mask,
2336 			    struct rte_flow_error *error)
2337 {
2338 	const struct rte_flow_item_tcp *mask = item->mask;
2339 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2340 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2341 				      MLX5_FLOW_LAYER_OUTER_L3;
2342 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2343 				      MLX5_FLOW_LAYER_OUTER_L4;
2344 	int ret;
2345 
2346 	MLX5_ASSERT(flow_mask);
2347 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2348 		return rte_flow_error_set(error, EINVAL,
2349 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2350 					  "protocol filtering not compatible"
2351 					  " with TCP layer");
2352 	if (!(item_flags & l3m))
2353 		return rte_flow_error_set(error, EINVAL,
2354 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 					  "L3 is mandatory to filter on L4");
2356 	if (item_flags & l4m)
2357 		return rte_flow_error_set(error, EINVAL,
2358 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2359 					  "multiple L4 layers not supported");
2360 	if (!mask)
2361 		mask = &rte_flow_item_tcp_mask;
2362 	ret = mlx5_flow_item_acceptable
2363 		(item, (const uint8_t *)mask,
2364 		 (const uint8_t *)flow_mask,
2365 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2366 		 error);
2367 	if (ret < 0)
2368 		return ret;
2369 	return 0;
2370 }
2371 
2372 /**
2373  * Validate VXLAN item.
2374  *
2375  * @param[in] item
2376  *   Item specification.
2377  * @param[in] item_flags
2378  *   Bit-fields that holds the items detected until now.
2379  * @param[in] target_protocol
2380  *   The next protocol in the previous item.
2381  * @param[out] error
2382  *   Pointer to error structure.
2383  *
2384  * @return
2385  *   0 on success, a negative errno value otherwise and rte_errno is set.
2386  */
2387 int
2388 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2389 			      uint64_t item_flags,
2390 			      struct rte_flow_error *error)
2391 {
2392 	const struct rte_flow_item_vxlan *spec = item->spec;
2393 	const struct rte_flow_item_vxlan *mask = item->mask;
2394 	int ret;
2395 	union vni {
2396 		uint32_t vlan_id;
2397 		uint8_t vni[4];
2398 	} id = { .vlan_id = 0, };
2399 
2400 
2401 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2402 		return rte_flow_error_set(error, ENOTSUP,
2403 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2404 					  "multiple tunnel layers not"
2405 					  " supported");
2406 	/*
2407 	 * Verify only UDPv4 is present as defined in
2408 	 * https://tools.ietf.org/html/rfc7348
2409 	 */
2410 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2411 		return rte_flow_error_set(error, EINVAL,
2412 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2413 					  "no outer UDP layer found");
2414 	if (!mask)
2415 		mask = &rte_flow_item_vxlan_mask;
2416 	ret = mlx5_flow_item_acceptable
2417 		(item, (const uint8_t *)mask,
2418 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2419 		 sizeof(struct rte_flow_item_vxlan),
2420 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2421 	if (ret < 0)
2422 		return ret;
2423 	if (spec) {
2424 		memcpy(&id.vni[1], spec->vni, 3);
2425 		memcpy(&id.vni[1], mask->vni, 3);
2426 	}
2427 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2428 		return rte_flow_error_set(error, ENOTSUP,
2429 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2430 					  "VXLAN tunnel must be fully defined");
2431 	return 0;
2432 }
2433 
2434 /**
2435  * Validate VXLAN_GPE item.
2436  *
2437  * @param[in] item
2438  *   Item specification.
2439  * @param[in] item_flags
2440  *   Bit-fields that holds the items detected until now.
2441  * @param[in] priv
2442  *   Pointer to the private data structure.
2443  * @param[in] target_protocol
2444  *   The next protocol in the previous item.
2445  * @param[out] error
2446  *   Pointer to error structure.
2447  *
2448  * @return
2449  *   0 on success, a negative errno value otherwise and rte_errno is set.
2450  */
2451 int
2452 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2453 				  uint64_t item_flags,
2454 				  struct rte_eth_dev *dev,
2455 				  struct rte_flow_error *error)
2456 {
2457 	struct mlx5_priv *priv = dev->data->dev_private;
2458 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2459 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2460 	int ret;
2461 	union vni {
2462 		uint32_t vlan_id;
2463 		uint8_t vni[4];
2464 	} id = { .vlan_id = 0, };
2465 
2466 	if (!priv->config.l3_vxlan_en)
2467 		return rte_flow_error_set(error, ENOTSUP,
2468 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2469 					  "L3 VXLAN is not enabled by device"
2470 					  " parameter and/or not configured in"
2471 					  " firmware");
2472 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2473 		return rte_flow_error_set(error, ENOTSUP,
2474 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2475 					  "multiple tunnel layers not"
2476 					  " supported");
2477 	/*
2478 	 * Verify only UDPv4 is present as defined in
2479 	 * https://tools.ietf.org/html/rfc7348
2480 	 */
2481 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2482 		return rte_flow_error_set(error, EINVAL,
2483 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2484 					  "no outer UDP layer found");
2485 	if (!mask)
2486 		mask = &rte_flow_item_vxlan_gpe_mask;
2487 	ret = mlx5_flow_item_acceptable
2488 		(item, (const uint8_t *)mask,
2489 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2490 		 sizeof(struct rte_flow_item_vxlan_gpe),
2491 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2492 	if (ret < 0)
2493 		return ret;
2494 	if (spec) {
2495 		if (spec->protocol)
2496 			return rte_flow_error_set(error, ENOTSUP,
2497 						  RTE_FLOW_ERROR_TYPE_ITEM,
2498 						  item,
2499 						  "VxLAN-GPE protocol"
2500 						  " not supported");
2501 		memcpy(&id.vni[1], spec->vni, 3);
2502 		memcpy(&id.vni[1], mask->vni, 3);
2503 	}
2504 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2505 		return rte_flow_error_set(error, ENOTSUP,
2506 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2507 					  "VXLAN-GPE tunnel must be fully"
2508 					  " defined");
2509 	return 0;
2510 }
2511 /**
2512  * Validate GRE Key item.
2513  *
2514  * @param[in] item
2515  *   Item specification.
2516  * @param[in] item_flags
2517  *   Bit flags to mark detected items.
2518  * @param[in] gre_item
2519  *   Pointer to gre_item
2520  * @param[out] error
2521  *   Pointer to error structure.
2522  *
2523  * @return
2524  *   0 on success, a negative errno value otherwise and rte_errno is set.
2525  */
2526 int
2527 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2528 				uint64_t item_flags,
2529 				const struct rte_flow_item *gre_item,
2530 				struct rte_flow_error *error)
2531 {
2532 	const rte_be32_t *mask = item->mask;
2533 	int ret = 0;
2534 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2535 	const struct rte_flow_item_gre *gre_spec;
2536 	const struct rte_flow_item_gre *gre_mask;
2537 
2538 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2539 		return rte_flow_error_set(error, ENOTSUP,
2540 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2541 					  "Multiple GRE key not support");
2542 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2543 		return rte_flow_error_set(error, ENOTSUP,
2544 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2545 					  "No preceding GRE header");
2546 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2547 		return rte_flow_error_set(error, ENOTSUP,
2548 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2549 					  "GRE key following a wrong item");
2550 	gre_mask = gre_item->mask;
2551 	if (!gre_mask)
2552 		gre_mask = &rte_flow_item_gre_mask;
2553 	gre_spec = gre_item->spec;
2554 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2555 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2556 		return rte_flow_error_set(error, EINVAL,
2557 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2558 					  "Key bit must be on");
2559 
2560 	if (!mask)
2561 		mask = &gre_key_default_mask;
2562 	ret = mlx5_flow_item_acceptable
2563 		(item, (const uint8_t *)mask,
2564 		 (const uint8_t *)&gre_key_default_mask,
2565 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2566 	return ret;
2567 }
2568 
2569 /**
2570  * Validate GRE item.
2571  *
2572  * @param[in] item
2573  *   Item specification.
2574  * @param[in] item_flags
2575  *   Bit flags to mark detected items.
2576  * @param[in] target_protocol
2577  *   The next protocol in the previous item.
2578  * @param[out] error
2579  *   Pointer to error structure.
2580  *
2581  * @return
2582  *   0 on success, a negative errno value otherwise and rte_errno is set.
2583  */
2584 int
2585 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2586 			    uint64_t item_flags,
2587 			    uint8_t target_protocol,
2588 			    struct rte_flow_error *error)
2589 {
2590 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2591 	const struct rte_flow_item_gre *mask = item->mask;
2592 	int ret;
2593 	const struct rte_flow_item_gre nic_mask = {
2594 		.c_rsvd0_ver = RTE_BE16(0xB000),
2595 		.protocol = RTE_BE16(UINT16_MAX),
2596 	};
2597 
2598 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2599 		return rte_flow_error_set(error, EINVAL,
2600 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2601 					  "protocol filtering not compatible"
2602 					  " with this GRE layer");
2603 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2604 		return rte_flow_error_set(error, ENOTSUP,
2605 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2606 					  "multiple tunnel layers not"
2607 					  " supported");
2608 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2609 		return rte_flow_error_set(error, ENOTSUP,
2610 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2611 					  "L3 Layer is missing");
2612 	if (!mask)
2613 		mask = &rte_flow_item_gre_mask;
2614 	ret = mlx5_flow_item_acceptable
2615 		(item, (const uint8_t *)mask,
2616 		 (const uint8_t *)&nic_mask,
2617 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2618 		 error);
2619 	if (ret < 0)
2620 		return ret;
2621 #ifndef HAVE_MLX5DV_DR
2622 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2623 	if (spec && (spec->protocol & mask->protocol))
2624 		return rte_flow_error_set(error, ENOTSUP,
2625 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2626 					  "without MPLS support the"
2627 					  " specification cannot be used for"
2628 					  " filtering");
2629 #endif
2630 #endif
2631 	return 0;
2632 }
2633 
2634 /**
2635  * Validate Geneve item.
2636  *
2637  * @param[in] item
2638  *   Item specification.
2639  * @param[in] itemFlags
2640  *   Bit-fields that holds the items detected until now.
2641  * @param[in] enPriv
2642  *   Pointer to the private data structure.
2643  * @param[out] error
2644  *   Pointer to error structure.
2645  *
2646  * @return
2647  *   0 on success, a negative errno value otherwise and rte_errno is set.
2648  */
2649 
2650 int
2651 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2652 			       uint64_t item_flags,
2653 			       struct rte_eth_dev *dev,
2654 			       struct rte_flow_error *error)
2655 {
2656 	struct mlx5_priv *priv = dev->data->dev_private;
2657 	const struct rte_flow_item_geneve *spec = item->spec;
2658 	const struct rte_flow_item_geneve *mask = item->mask;
2659 	int ret;
2660 	uint16_t gbhdr;
2661 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2662 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2663 	const struct rte_flow_item_geneve nic_mask = {
2664 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2665 		.vni = "\xff\xff\xff",
2666 		.protocol = RTE_BE16(UINT16_MAX),
2667 	};
2668 
2669 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2670 		return rte_flow_error_set(error, ENOTSUP,
2671 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2672 					  "L3 Geneve is not enabled by device"
2673 					  " parameter and/or not configured in"
2674 					  " firmware");
2675 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2676 		return rte_flow_error_set(error, ENOTSUP,
2677 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2678 					  "multiple tunnel layers not"
2679 					  " supported");
2680 	/*
2681 	 * Verify only UDPv4 is present as defined in
2682 	 * https://tools.ietf.org/html/rfc7348
2683 	 */
2684 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2685 		return rte_flow_error_set(error, EINVAL,
2686 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2687 					  "no outer UDP layer found");
2688 	if (!mask)
2689 		mask = &rte_flow_item_geneve_mask;
2690 	ret = mlx5_flow_item_acceptable
2691 				  (item, (const uint8_t *)mask,
2692 				   (const uint8_t *)&nic_mask,
2693 				   sizeof(struct rte_flow_item_geneve),
2694 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2695 	if (ret)
2696 		return ret;
2697 	if (spec) {
2698 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2699 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2700 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2701 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2702 			return rte_flow_error_set(error, ENOTSUP,
2703 						  RTE_FLOW_ERROR_TYPE_ITEM,
2704 						  item,
2705 						  "Geneve protocol unsupported"
2706 						  " fields are being used");
2707 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2708 			return rte_flow_error_set
2709 					(error, ENOTSUP,
2710 					 RTE_FLOW_ERROR_TYPE_ITEM,
2711 					 item,
2712 					 "Unsupported Geneve options length");
2713 	}
2714 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2715 		return rte_flow_error_set
2716 				    (error, ENOTSUP,
2717 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2718 				     "Geneve tunnel must be fully defined");
2719 	return 0;
2720 }
2721 
2722 /**
2723  * Validate MPLS item.
2724  *
2725  * @param[in] dev
2726  *   Pointer to the rte_eth_dev structure.
2727  * @param[in] item
2728  *   Item specification.
2729  * @param[in] item_flags
2730  *   Bit-fields that holds the items detected until now.
2731  * @param[in] prev_layer
2732  *   The protocol layer indicated in previous item.
2733  * @param[out] error
2734  *   Pointer to error structure.
2735  *
2736  * @return
2737  *   0 on success, a negative errno value otherwise and rte_errno is set.
2738  */
2739 int
2740 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2741 			     const struct rte_flow_item *item __rte_unused,
2742 			     uint64_t item_flags __rte_unused,
2743 			     uint64_t prev_layer __rte_unused,
2744 			     struct rte_flow_error *error)
2745 {
2746 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2747 	const struct rte_flow_item_mpls *mask = item->mask;
2748 	struct mlx5_priv *priv = dev->data->dev_private;
2749 	int ret;
2750 
2751 	if (!priv->config.mpls_en)
2752 		return rte_flow_error_set(error, ENOTSUP,
2753 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2754 					  "MPLS not supported or"
2755 					  " disabled in firmware"
2756 					  " configuration.");
2757 	/* MPLS over IP, UDP, GRE is allowed */
2758 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2759 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2760 			    MLX5_FLOW_LAYER_GRE)))
2761 		return rte_flow_error_set(error, EINVAL,
2762 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2763 					  "protocol filtering not compatible"
2764 					  " with MPLS layer");
2765 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2766 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2767 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2768 		return rte_flow_error_set(error, ENOTSUP,
2769 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2770 					  "multiple tunnel layers not"
2771 					  " supported");
2772 	if (!mask)
2773 		mask = &rte_flow_item_mpls_mask;
2774 	ret = mlx5_flow_item_acceptable
2775 		(item, (const uint8_t *)mask,
2776 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2777 		 sizeof(struct rte_flow_item_mpls),
2778 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2779 	if (ret < 0)
2780 		return ret;
2781 	return 0;
2782 #else
2783 	return rte_flow_error_set(error, ENOTSUP,
2784 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2785 				  "MPLS is not supported by Verbs, please"
2786 				  " update.");
2787 #endif
2788 }
2789 
2790 /**
2791  * Validate NVGRE item.
2792  *
2793  * @param[in] item
2794  *   Item specification.
2795  * @param[in] item_flags
2796  *   Bit flags to mark detected items.
2797  * @param[in] target_protocol
2798  *   The next protocol in the previous item.
2799  * @param[out] error
2800  *   Pointer to error structure.
2801  *
2802  * @return
2803  *   0 on success, a negative errno value otherwise and rte_errno is set.
2804  */
2805 int
2806 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2807 			      uint64_t item_flags,
2808 			      uint8_t target_protocol,
2809 			      struct rte_flow_error *error)
2810 {
2811 	const struct rte_flow_item_nvgre *mask = item->mask;
2812 	int ret;
2813 
2814 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2815 		return rte_flow_error_set(error, EINVAL,
2816 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2817 					  "protocol filtering not compatible"
2818 					  " with this GRE layer");
2819 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2820 		return rte_flow_error_set(error, ENOTSUP,
2821 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2822 					  "multiple tunnel layers not"
2823 					  " supported");
2824 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2825 		return rte_flow_error_set(error, ENOTSUP,
2826 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2827 					  "L3 Layer is missing");
2828 	if (!mask)
2829 		mask = &rte_flow_item_nvgre_mask;
2830 	ret = mlx5_flow_item_acceptable
2831 		(item, (const uint8_t *)mask,
2832 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2833 		 sizeof(struct rte_flow_item_nvgre),
2834 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2835 	if (ret < 0)
2836 		return ret;
2837 	return 0;
2838 }
2839 
2840 /**
2841  * Validate eCPRI item.
2842  *
2843  * @param[in] item
2844  *   Item specification.
2845  * @param[in] item_flags
2846  *   Bit-fields that holds the items detected until now.
2847  * @param[in] last_item
2848  *   Previous validated item in the pattern items.
2849  * @param[in] ether_type
2850  *   Type in the ethernet layer header (including dot1q).
2851  * @param[in] acc_mask
2852  *   Acceptable mask, if NULL default internal default mask
2853  *   will be used to check whether item fields are supported.
2854  * @param[out] error
2855  *   Pointer to error structure.
2856  *
2857  * @return
2858  *   0 on success, a negative errno value otherwise and rte_errno is set.
2859  */
2860 int
2861 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2862 			      uint64_t item_flags,
2863 			      uint64_t last_item,
2864 			      uint16_t ether_type,
2865 			      const struct rte_flow_item_ecpri *acc_mask,
2866 			      struct rte_flow_error *error)
2867 {
2868 	const struct rte_flow_item_ecpri *mask = item->mask;
2869 	const struct rte_flow_item_ecpri nic_mask = {
2870 		.hdr = {
2871 			.common = {
2872 				.u32 =
2873 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2874 					.type = 0xFF,
2875 					}).u32),
2876 			},
2877 			.dummy[0] = 0xFFFFFFFF,
2878 		},
2879 	};
2880 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2881 					MLX5_FLOW_LAYER_OUTER_VLAN);
2882 	struct rte_flow_item_ecpri mask_lo;
2883 
2884 	if ((last_item & outer_l2_vlan) && ether_type &&
2885 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2886 		return rte_flow_error_set(error, EINVAL,
2887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2888 					  "eCPRI cannot follow L2/VLAN layer "
2889 					  "which ether type is not 0xAEFE.");
2890 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2891 		return rte_flow_error_set(error, EINVAL,
2892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2893 					  "eCPRI with tunnel is not supported "
2894 					  "right now.");
2895 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2896 		return rte_flow_error_set(error, ENOTSUP,
2897 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2898 					  "multiple L3 layers not supported");
2899 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2900 		return rte_flow_error_set(error, EINVAL,
2901 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2902 					  "eCPRI cannot follow a TCP layer.");
2903 	/* In specification, eCPRI could be over UDP layer. */
2904 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2905 		return rte_flow_error_set(error, EINVAL,
2906 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2907 					  "eCPRI over UDP layer is not yet "
2908 					  "supported right now.");
2909 	/* Mask for type field in common header could be zero. */
2910 	if (!mask)
2911 		mask = &rte_flow_item_ecpri_mask;
2912 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2913 	/* Input mask is in big-endian format. */
2914 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2915 		return rte_flow_error_set(error, EINVAL,
2916 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2917 					  "partial mask is not supported "
2918 					  "for protocol");
2919 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2920 		return rte_flow_error_set(error, EINVAL,
2921 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2922 					  "message header mask must be after "
2923 					  "a type mask");
2924 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2925 					 acc_mask ? (const uint8_t *)acc_mask
2926 						  : (const uint8_t *)&nic_mask,
2927 					 sizeof(struct rte_flow_item_ecpri),
2928 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2929 }
2930 
2931 /* Allocate unique ID for the split Q/RSS subflows. */
2932 static uint32_t
2933 flow_qrss_get_id(struct rte_eth_dev *dev)
2934 {
2935 	struct mlx5_priv *priv = dev->data->dev_private;
2936 	uint32_t qrss_id, ret;
2937 
2938 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2939 	if (ret)
2940 		return 0;
2941 	MLX5_ASSERT(qrss_id);
2942 	return qrss_id;
2943 }
2944 
2945 /* Free unique ID for the split Q/RSS subflows. */
2946 static void
2947 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2948 {
2949 	struct mlx5_priv *priv = dev->data->dev_private;
2950 
2951 	if (qrss_id)
2952 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2953 }
2954 
2955 /**
2956  * Release resource related QUEUE/RSS action split.
2957  *
2958  * @param dev
2959  *   Pointer to Ethernet device.
2960  * @param flow
2961  *   Flow to release id's from.
2962  */
2963 static void
2964 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2965 			     struct rte_flow *flow)
2966 {
2967 	struct mlx5_priv *priv = dev->data->dev_private;
2968 	uint32_t handle_idx;
2969 	struct mlx5_flow_handle *dev_handle;
2970 
2971 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2972 		       handle_idx, dev_handle, next)
2973 		if (dev_handle->split_flow_id)
2974 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2975 }
2976 
2977 static int
2978 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2979 		   const struct rte_flow_attr *attr __rte_unused,
2980 		   const struct rte_flow_item items[] __rte_unused,
2981 		   const struct rte_flow_action actions[] __rte_unused,
2982 		   bool external __rte_unused,
2983 		   int hairpin __rte_unused,
2984 		   struct rte_flow_error *error)
2985 {
2986 	return rte_flow_error_set(error, ENOTSUP,
2987 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2988 }
2989 
2990 static struct mlx5_flow *
2991 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2992 		  const struct rte_flow_attr *attr __rte_unused,
2993 		  const struct rte_flow_item items[] __rte_unused,
2994 		  const struct rte_flow_action actions[] __rte_unused,
2995 		  struct rte_flow_error *error)
2996 {
2997 	rte_flow_error_set(error, ENOTSUP,
2998 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2999 	return NULL;
3000 }
3001 
3002 static int
3003 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3004 		    struct mlx5_flow *dev_flow __rte_unused,
3005 		    const struct rte_flow_attr *attr __rte_unused,
3006 		    const struct rte_flow_item items[] __rte_unused,
3007 		    const struct rte_flow_action actions[] __rte_unused,
3008 		    struct rte_flow_error *error)
3009 {
3010 	return rte_flow_error_set(error, ENOTSUP,
3011 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3012 }
3013 
3014 static int
3015 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3016 		struct rte_flow *flow __rte_unused,
3017 		struct rte_flow_error *error)
3018 {
3019 	return rte_flow_error_set(error, ENOTSUP,
3020 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3021 }
3022 
3023 static void
3024 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3025 		 struct rte_flow *flow __rte_unused)
3026 {
3027 }
3028 
3029 static void
3030 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3031 		  struct rte_flow *flow __rte_unused)
3032 {
3033 }
3034 
3035 static int
3036 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3037 		struct rte_flow *flow __rte_unused,
3038 		const struct rte_flow_action *actions __rte_unused,
3039 		void *data __rte_unused,
3040 		struct rte_flow_error *error)
3041 {
3042 	return rte_flow_error_set(error, ENOTSUP,
3043 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3044 }
3045 
3046 static int
3047 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3048 		      uint32_t domains __rte_unused,
3049 		      uint32_t flags __rte_unused)
3050 {
3051 	return 0;
3052 }
3053 
3054 /* Void driver to protect from null pointer reference. */
3055 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3056 	.validate = flow_null_validate,
3057 	.prepare = flow_null_prepare,
3058 	.translate = flow_null_translate,
3059 	.apply = flow_null_apply,
3060 	.remove = flow_null_remove,
3061 	.destroy = flow_null_destroy,
3062 	.query = flow_null_query,
3063 	.sync_domain = flow_null_sync_domain,
3064 };
3065 
3066 /**
3067  * Select flow driver type according to flow attributes and device
3068  * configuration.
3069  *
3070  * @param[in] dev
3071  *   Pointer to the dev structure.
3072  * @param[in] attr
3073  *   Pointer to the flow attributes.
3074  *
3075  * @return
3076  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3077  */
3078 static enum mlx5_flow_drv_type
3079 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3080 {
3081 	struct mlx5_priv *priv = dev->data->dev_private;
3082 	/* The OS can determine first a specific flow type (DV, VERBS) */
3083 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3084 
3085 	if (type != MLX5_FLOW_TYPE_MAX)
3086 		return type;
3087 	/* If no OS specific type - continue with DV/VERBS selection */
3088 	if (attr->transfer && priv->config.dv_esw_en)
3089 		type = MLX5_FLOW_TYPE_DV;
3090 	if (!attr->transfer)
3091 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3092 						 MLX5_FLOW_TYPE_VERBS;
3093 	return type;
3094 }
3095 
3096 #define flow_get_drv_ops(type) flow_drv_ops[type]
3097 
3098 /**
3099  * Flow driver validation API. This abstracts calling driver specific functions.
3100  * The type of flow driver is determined according to flow attributes.
3101  *
3102  * @param[in] dev
3103  *   Pointer to the dev structure.
3104  * @param[in] attr
3105  *   Pointer to the flow attributes.
3106  * @param[in] items
3107  *   Pointer to the list of items.
3108  * @param[in] actions
3109  *   Pointer to the list of actions.
3110  * @param[in] external
3111  *   This flow rule is created by request external to PMD.
3112  * @param[in] hairpin
3113  *   Number of hairpin TX actions, 0 means classic flow.
3114  * @param[out] error
3115  *   Pointer to the error structure.
3116  *
3117  * @return
3118  *   0 on success, a negative errno value otherwise and rte_errno is set.
3119  */
3120 static inline int
3121 flow_drv_validate(struct rte_eth_dev *dev,
3122 		  const struct rte_flow_attr *attr,
3123 		  const struct rte_flow_item items[],
3124 		  const struct rte_flow_action actions[],
3125 		  bool external, int hairpin, struct rte_flow_error *error)
3126 {
3127 	const struct mlx5_flow_driver_ops *fops;
3128 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3129 
3130 	fops = flow_get_drv_ops(type);
3131 	return fops->validate(dev, attr, items, actions, external,
3132 			      hairpin, error);
3133 }
3134 
3135 /**
3136  * Flow driver preparation API. This abstracts calling driver specific
3137  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3138  * calculates the size of memory required for device flow, allocates the memory,
3139  * initializes the device flow and returns the pointer.
3140  *
3141  * @note
3142  *   This function initializes device flow structure such as dv or verbs in
3143  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3144  *   rest. For example, adding returning device flow to flow->dev_flow list and
3145  *   setting backward reference to the flow should be done out of this function.
3146  *   layers field is not filled either.
3147  *
3148  * @param[in] dev
3149  *   Pointer to the dev structure.
3150  * @param[in] attr
3151  *   Pointer to the flow attributes.
3152  * @param[in] items
3153  *   Pointer to the list of items.
3154  * @param[in] actions
3155  *   Pointer to the list of actions.
3156  * @param[in] flow_idx
3157  *   This memory pool index to the flow.
3158  * @param[out] error
3159  *   Pointer to the error structure.
3160  *
3161  * @return
3162  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3163  */
3164 static inline struct mlx5_flow *
3165 flow_drv_prepare(struct rte_eth_dev *dev,
3166 		 const struct rte_flow *flow,
3167 		 const struct rte_flow_attr *attr,
3168 		 const struct rte_flow_item items[],
3169 		 const struct rte_flow_action actions[],
3170 		 uint32_t flow_idx,
3171 		 struct rte_flow_error *error)
3172 {
3173 	const struct mlx5_flow_driver_ops *fops;
3174 	enum mlx5_flow_drv_type type = flow->drv_type;
3175 	struct mlx5_flow *mlx5_flow = NULL;
3176 
3177 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3178 	fops = flow_get_drv_ops(type);
3179 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3180 	if (mlx5_flow)
3181 		mlx5_flow->flow_idx = flow_idx;
3182 	return mlx5_flow;
3183 }
3184 
3185 /**
3186  * Flow driver translation API. This abstracts calling driver specific
3187  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3188  * translates a generic flow into a driver flow. flow_drv_prepare() must
3189  * precede.
3190  *
3191  * @note
3192  *   dev_flow->layers could be filled as a result of parsing during translation
3193  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3194  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3195  *   flow->actions could be overwritten even though all the expanded dev_flows
3196  *   have the same actions.
3197  *
3198  * @param[in] dev
3199  *   Pointer to the rte dev structure.
3200  * @param[in, out] dev_flow
3201  *   Pointer to the mlx5 flow.
3202  * @param[in] attr
3203  *   Pointer to the flow attributes.
3204  * @param[in] items
3205  *   Pointer to the list of items.
3206  * @param[in] actions
3207  *   Pointer to the list of actions.
3208  * @param[out] error
3209  *   Pointer to the error structure.
3210  *
3211  * @return
3212  *   0 on success, a negative errno value otherwise and rte_errno is set.
3213  */
3214 static inline int
3215 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3216 		   const struct rte_flow_attr *attr,
3217 		   const struct rte_flow_item items[],
3218 		   const struct rte_flow_action actions[],
3219 		   struct rte_flow_error *error)
3220 {
3221 	const struct mlx5_flow_driver_ops *fops;
3222 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3223 
3224 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3225 	fops = flow_get_drv_ops(type);
3226 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3227 }
3228 
3229 /**
3230  * Flow driver apply API. This abstracts calling driver specific functions.
3231  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3232  * translated driver flows on to device. flow_drv_translate() must precede.
3233  *
3234  * @param[in] dev
3235  *   Pointer to Ethernet device structure.
3236  * @param[in, out] flow
3237  *   Pointer to flow structure.
3238  * @param[out] error
3239  *   Pointer to error structure.
3240  *
3241  * @return
3242  *   0 on success, a negative errno value otherwise and rte_errno is set.
3243  */
3244 static inline int
3245 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3246 	       struct rte_flow_error *error)
3247 {
3248 	const struct mlx5_flow_driver_ops *fops;
3249 	enum mlx5_flow_drv_type type = flow->drv_type;
3250 
3251 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3252 	fops = flow_get_drv_ops(type);
3253 	return fops->apply(dev, flow, error);
3254 }
3255 
3256 /**
3257  * Flow driver remove API. This abstracts calling driver specific functions.
3258  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3259  * on device. All the resources of the flow should be freed by calling
3260  * flow_drv_destroy().
3261  *
3262  * @param[in] dev
3263  *   Pointer to Ethernet device.
3264  * @param[in, out] flow
3265  *   Pointer to flow structure.
3266  */
3267 static inline void
3268 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
3269 {
3270 	const struct mlx5_flow_driver_ops *fops;
3271 	enum mlx5_flow_drv_type type = flow->drv_type;
3272 
3273 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3274 	fops = flow_get_drv_ops(type);
3275 	fops->remove(dev, flow);
3276 }
3277 
3278 /**
3279  * Flow driver destroy API. This abstracts calling driver specific functions.
3280  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3281  * on device and releases resources of the flow.
3282  *
3283  * @param[in] dev
3284  *   Pointer to Ethernet device.
3285  * @param[in, out] flow
3286  *   Pointer to flow structure.
3287  */
3288 static inline void
3289 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3290 {
3291 	const struct mlx5_flow_driver_ops *fops;
3292 	enum mlx5_flow_drv_type type = flow->drv_type;
3293 
3294 	flow_mreg_split_qrss_release(dev, flow);
3295 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3296 	fops = flow_get_drv_ops(type);
3297 	fops->destroy(dev, flow);
3298 }
3299 
3300 /**
3301  * Get RSS action from the action list.
3302  *
3303  * @param[in] actions
3304  *   Pointer to the list of actions.
3305  *
3306  * @return
3307  *   Pointer to the RSS action if exist, else return NULL.
3308  */
3309 static const struct rte_flow_action_rss*
3310 flow_get_rss_action(const struct rte_flow_action actions[])
3311 {
3312 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3313 		switch (actions->type) {
3314 		case RTE_FLOW_ACTION_TYPE_RSS:
3315 			return (const struct rte_flow_action_rss *)
3316 			       actions->conf;
3317 		default:
3318 			break;
3319 		}
3320 	}
3321 	return NULL;
3322 }
3323 
3324 /* maps shared action to translated non shared in some actions array */
3325 struct mlx5_translated_shared_action {
3326 	struct rte_flow_shared_action *action; /**< Shared action */
3327 	int index; /**< Index in related array of rte_flow_action */
3328 };
3329 
3330 /**
3331  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3332  * non shared action if translation possible.
3333  * This functionality used to run same execution path for both shared & non
3334  * shared actions on flow create. All necessary preparations for shared
3335  * action handling should be preformed on *shared* actions list returned
3336  * from this call.
3337  *
3338  * @param[in] actions
3339  *   List of actions to translate.
3340  * @param[out] shared
3341  *   List to store translated shared actions.
3342  * @param[in, out] shared_n
3343  *   Size of *shared* array. On return should be updated with number of shared
3344  *   actions retrieved from the *actions* list.
3345  * @param[out] translated_actions
3346  *   List of actions where all shared actions were translated to non shared
3347  *   if possible. NULL if no translation took place.
3348  * @param[out] error
3349  *   Pointer to the error structure.
3350  *
3351  * @return
3352  *   0 on success, a negative errno value otherwise and rte_errno is set.
3353  */
3354 static int
3355 flow_shared_actions_translate(const struct rte_flow_action actions[],
3356 	struct mlx5_translated_shared_action *shared,
3357 	int *shared_n,
3358 	struct rte_flow_action **translated_actions,
3359 	struct rte_flow_error *error)
3360 {
3361 	struct rte_flow_action *translated = NULL;
3362 	size_t actions_size;
3363 	int n;
3364 	int copied_n = 0;
3365 	struct mlx5_translated_shared_action *shared_end = NULL;
3366 
3367 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3368 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3369 			continue;
3370 		if (copied_n == *shared_n) {
3371 			return rte_flow_error_set
3372 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3373 				 NULL, "too many shared actions");
3374 		}
3375 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3376 			   sizeof(actions[n].conf));
3377 		shared[copied_n].index = n;
3378 		copied_n++;
3379 	}
3380 	n++;
3381 	*shared_n = copied_n;
3382 	if (!copied_n)
3383 		return 0;
3384 	actions_size = sizeof(struct rte_flow_action) * n;
3385 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3386 	if (!translated) {
3387 		rte_errno = ENOMEM;
3388 		return -ENOMEM;
3389 	}
3390 	memcpy(translated, actions, actions_size);
3391 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3392 		const struct rte_flow_shared_action *shared_action;
3393 
3394 		shared_action = shared->action;
3395 		switch (shared_action->type) {
3396 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3397 			translated[shared->index].type =
3398 				RTE_FLOW_ACTION_TYPE_RSS;
3399 			translated[shared->index].conf =
3400 				&shared_action->rss.origin;
3401 			break;
3402 		default:
3403 			mlx5_free(translated);
3404 			return rte_flow_error_set
3405 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3406 				 NULL, "invalid shared action type");
3407 		}
3408 	}
3409 	*translated_actions = translated;
3410 	return 0;
3411 }
3412 
3413 /**
3414  * Get Shared RSS action from the action list.
3415  *
3416  * @param[in] shared
3417  *   Pointer to the list of actions.
3418  * @param[in] shared_n
3419  *   Actions list length.
3420  *
3421  * @return
3422  *   Pointer to the MLX5 RSS action if exists, otherwise return NULL.
3423  */
3424 static struct mlx5_shared_action_rss *
3425 flow_get_shared_rss_action(struct mlx5_translated_shared_action *shared,
3426 			   int shared_n)
3427 {
3428 	struct mlx5_translated_shared_action *shared_end;
3429 
3430 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3431 		struct rte_flow_shared_action *shared_action;
3432 
3433 		shared_action = shared->action;
3434 		switch (shared_action->type) {
3435 		case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
3436 			__atomic_add_fetch(&shared_action->refcnt, 1,
3437 					   __ATOMIC_RELAXED);
3438 			return &shared_action->rss;
3439 		default:
3440 			break;
3441 		}
3442 	}
3443 	return NULL;
3444 }
3445 
3446 struct rte_flow_shared_action *
3447 mlx5_flow_get_shared_rss(struct rte_flow *flow)
3448 {
3449 	if (flow->shared_rss)
3450 		return container_of(flow->shared_rss,
3451 				    struct rte_flow_shared_action, rss);
3452 	else
3453 		return NULL;
3454 }
3455 
3456 static unsigned int
3457 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3458 {
3459 	const struct rte_flow_item *item;
3460 	unsigned int has_vlan = 0;
3461 
3462 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3463 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3464 			has_vlan = 1;
3465 			break;
3466 		}
3467 	}
3468 	if (has_vlan)
3469 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3470 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3471 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3472 			       MLX5_EXPANSION_ROOT_OUTER;
3473 }
3474 
3475 /**
3476  *  Get layer flags from the prefix flow.
3477  *
3478  *  Some flows may be split to several subflows, the prefix subflow gets the
3479  *  match items and the suffix sub flow gets the actions.
3480  *  Some actions need the user defined match item flags to get the detail for
3481  *  the action.
3482  *  This function helps the suffix flow to get the item layer flags from prefix
3483  *  subflow.
3484  *
3485  * @param[in] dev_flow
3486  *   Pointer the created preifx subflow.
3487  *
3488  * @return
3489  *   The layers get from prefix subflow.
3490  */
3491 static inline uint64_t
3492 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3493 {
3494 	uint64_t layers = 0;
3495 
3496 	/*
3497 	 * Layers bits could be localization, but usually the compiler will
3498 	 * help to do the optimization work for source code.
3499 	 * If no decap actions, use the layers directly.
3500 	 */
3501 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3502 		return dev_flow->handle->layers;
3503 	/* Convert L3 layers with decap action. */
3504 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3505 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3506 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3507 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3508 	/* Convert L4 layers with decap action.  */
3509 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3510 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3511 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3512 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3513 	return layers;
3514 }
3515 
3516 /**
3517  * Get metadata split action information.
3518  *
3519  * @param[in] actions
3520  *   Pointer to the list of actions.
3521  * @param[out] qrss
3522  *   Pointer to the return pointer.
3523  * @param[out] qrss_type
3524  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3525  *   if no QUEUE/RSS is found.
3526  * @param[out] encap_idx
3527  *   Pointer to the index of the encap action if exists, otherwise the last
3528  *   action index.
3529  *
3530  * @return
3531  *   Total number of actions.
3532  */
3533 static int
3534 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3535 				       const struct rte_flow_action **qrss,
3536 				       int *encap_idx)
3537 {
3538 	const struct rte_flow_action_raw_encap *raw_encap;
3539 	int actions_n = 0;
3540 	int raw_decap_idx = -1;
3541 
3542 	*encap_idx = -1;
3543 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3544 		switch (actions->type) {
3545 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3546 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3547 			*encap_idx = actions_n;
3548 			break;
3549 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3550 			raw_decap_idx = actions_n;
3551 			break;
3552 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3553 			raw_encap = actions->conf;
3554 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3555 				*encap_idx = raw_decap_idx != -1 ?
3556 						      raw_decap_idx : actions_n;
3557 			break;
3558 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3559 		case RTE_FLOW_ACTION_TYPE_RSS:
3560 			*qrss = actions;
3561 			break;
3562 		default:
3563 			break;
3564 		}
3565 		actions_n++;
3566 	}
3567 	if (*encap_idx == -1)
3568 		*encap_idx = actions_n;
3569 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3570 	return actions_n + 1;
3571 }
3572 
3573 /**
3574  * Check meter action from the action list.
3575  *
3576  * @param[in] actions
3577  *   Pointer to the list of actions.
3578  * @param[out] mtr
3579  *   Pointer to the meter exist flag.
3580  *
3581  * @return
3582  *   Total number of actions.
3583  */
3584 static int
3585 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3586 {
3587 	int actions_n = 0;
3588 
3589 	MLX5_ASSERT(mtr);
3590 	*mtr = 0;
3591 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3592 		switch (actions->type) {
3593 		case RTE_FLOW_ACTION_TYPE_METER:
3594 			*mtr = 1;
3595 			break;
3596 		default:
3597 			break;
3598 		}
3599 		actions_n++;
3600 	}
3601 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3602 	return actions_n + 1;
3603 }
3604 
3605 /**
3606  * Check if the flow should be split due to hairpin.
3607  * The reason for the split is that in current HW we can't
3608  * support encap and push-vlan on Rx, so if a flow contains
3609  * these actions we move it to Tx.
3610  *
3611  * @param dev
3612  *   Pointer to Ethernet device.
3613  * @param[in] attr
3614  *   Flow rule attributes.
3615  * @param[in] actions
3616  *   Associated actions (list terminated by the END action).
3617  *
3618  * @return
3619  *   > 0 the number of actions and the flow should be split,
3620  *   0 when no split required.
3621  */
3622 static int
3623 flow_check_hairpin_split(struct rte_eth_dev *dev,
3624 			 const struct rte_flow_attr *attr,
3625 			 const struct rte_flow_action actions[])
3626 {
3627 	int queue_action = 0;
3628 	int action_n = 0;
3629 	int split = 0;
3630 	const struct rte_flow_action_queue *queue;
3631 	const struct rte_flow_action_rss *rss;
3632 	const struct rte_flow_action_raw_encap *raw_encap;
3633 	const struct rte_eth_hairpin_conf *conf;
3634 
3635 	if (!attr->ingress)
3636 		return 0;
3637 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3638 		switch (actions->type) {
3639 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3640 			queue = actions->conf;
3641 			if (queue == NULL)
3642 				return 0;
3643 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3644 			if (conf != NULL && !!conf->tx_explicit)
3645 				return 0;
3646 			queue_action = 1;
3647 			action_n++;
3648 			break;
3649 		case RTE_FLOW_ACTION_TYPE_RSS:
3650 			rss = actions->conf;
3651 			if (rss == NULL || rss->queue_num == 0)
3652 				return 0;
3653 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3654 			if (conf != NULL && !!conf->tx_explicit)
3655 				return 0;
3656 			queue_action = 1;
3657 			action_n++;
3658 			break;
3659 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3660 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3661 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3662 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3663 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3664 			split++;
3665 			action_n++;
3666 			break;
3667 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3668 			raw_encap = actions->conf;
3669 			if (raw_encap->size >
3670 			    (sizeof(struct rte_flow_item_eth) +
3671 			     sizeof(struct rte_flow_item_ipv4)))
3672 				split++;
3673 			action_n++;
3674 			break;
3675 		default:
3676 			action_n++;
3677 			break;
3678 		}
3679 	}
3680 	if (split && queue_action)
3681 		return action_n;
3682 	return 0;
3683 }
3684 
3685 /* Declare flow create/destroy prototype in advance. */
3686 static uint32_t
3687 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3688 		 const struct rte_flow_attr *attr,
3689 		 const struct rte_flow_item items[],
3690 		 const struct rte_flow_action actions[],
3691 		 bool external, struct rte_flow_error *error);
3692 
3693 static void
3694 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3695 		  uint32_t flow_idx);
3696 
3697 /**
3698  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3699  *
3700  * As mark_id is unique, if there's already a registered flow for the mark_id,
3701  * return by increasing the reference counter of the resource. Otherwise, create
3702  * the resource (mcp_res) and flow.
3703  *
3704  * Flow looks like,
3705  *   - If ingress port is ANY and reg_c[1] is mark_id,
3706  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3707  *
3708  * For default flow (zero mark_id), flow is like,
3709  *   - If ingress port is ANY,
3710  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3711  *
3712  * @param dev
3713  *   Pointer to Ethernet device.
3714  * @param mark_id
3715  *   ID of MARK action, zero means default flow for META.
3716  * @param[out] error
3717  *   Perform verbose error reporting if not NULL.
3718  *
3719  * @return
3720  *   Associated resource on success, NULL otherwise and rte_errno is set.
3721  */
3722 static struct mlx5_flow_mreg_copy_resource *
3723 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3724 			  struct rte_flow_error *error)
3725 {
3726 	struct mlx5_priv *priv = dev->data->dev_private;
3727 	struct rte_flow_attr attr = {
3728 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3729 		.ingress = 1,
3730 	};
3731 	struct mlx5_rte_flow_item_tag tag_spec = {
3732 		.data = mark_id,
3733 	};
3734 	struct rte_flow_item items[] = {
3735 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3736 	};
3737 	struct rte_flow_action_mark ftag = {
3738 		.id = mark_id,
3739 	};
3740 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3741 		.dst = REG_B,
3742 		.src = REG_NON,
3743 	};
3744 	struct rte_flow_action_jump jump = {
3745 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3746 	};
3747 	struct rte_flow_action actions[] = {
3748 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3749 	};
3750 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3751 	uint32_t idx = 0;
3752 	int ret;
3753 
3754 	/* Fill the register fileds in the flow. */
3755 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3756 	if (ret < 0)
3757 		return NULL;
3758 	tag_spec.id = ret;
3759 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3760 	if (ret < 0)
3761 		return NULL;
3762 	cp_mreg.src = ret;
3763 	/* Check if already registered. */
3764 	MLX5_ASSERT(priv->mreg_cp_tbl);
3765 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3766 	if (mcp_res) {
3767 		/* For non-default rule. */
3768 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3769 			mcp_res->refcnt++;
3770 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3771 			    mcp_res->refcnt == 1);
3772 		return mcp_res;
3773 	}
3774 	/* Provide the full width of FLAG specific value. */
3775 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3776 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3777 	/* Build a new flow. */
3778 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3779 		items[0] = (struct rte_flow_item){
3780 			.type = (enum rte_flow_item_type)
3781 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3782 			.spec = &tag_spec,
3783 		};
3784 		items[1] = (struct rte_flow_item){
3785 			.type = RTE_FLOW_ITEM_TYPE_END,
3786 		};
3787 		actions[0] = (struct rte_flow_action){
3788 			.type = (enum rte_flow_action_type)
3789 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3790 			.conf = &ftag,
3791 		};
3792 		actions[1] = (struct rte_flow_action){
3793 			.type = (enum rte_flow_action_type)
3794 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3795 			.conf = &cp_mreg,
3796 		};
3797 		actions[2] = (struct rte_flow_action){
3798 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3799 			.conf = &jump,
3800 		};
3801 		actions[3] = (struct rte_flow_action){
3802 			.type = RTE_FLOW_ACTION_TYPE_END,
3803 		};
3804 	} else {
3805 		/* Default rule, wildcard match. */
3806 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3807 		items[0] = (struct rte_flow_item){
3808 			.type = RTE_FLOW_ITEM_TYPE_END,
3809 		};
3810 		actions[0] = (struct rte_flow_action){
3811 			.type = (enum rte_flow_action_type)
3812 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3813 			.conf = &cp_mreg,
3814 		};
3815 		actions[1] = (struct rte_flow_action){
3816 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3817 			.conf = &jump,
3818 		};
3819 		actions[2] = (struct rte_flow_action){
3820 			.type = RTE_FLOW_ACTION_TYPE_END,
3821 		};
3822 	}
3823 	/* Build a new entry. */
3824 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3825 	if (!mcp_res) {
3826 		rte_errno = ENOMEM;
3827 		return NULL;
3828 	}
3829 	mcp_res->idx = idx;
3830 	/*
3831 	 * The copy Flows are not included in any list. There
3832 	 * ones are referenced from other Flows and can not
3833 	 * be applied, removed, deleted in ardbitrary order
3834 	 * by list traversing.
3835 	 */
3836 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3837 					 actions, false, error);
3838 	if (!mcp_res->rix_flow)
3839 		goto error;
3840 	mcp_res->refcnt++;
3841 	mcp_res->hlist_ent.key = mark_id;
3842 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3843 				&mcp_res->hlist_ent);
3844 	MLX5_ASSERT(!ret);
3845 	if (ret)
3846 		goto error;
3847 	return mcp_res;
3848 error:
3849 	if (mcp_res->rix_flow)
3850 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3851 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3852 	return NULL;
3853 }
3854 
3855 /**
3856  * Release flow in RX_CP_TBL.
3857  *
3858  * @param dev
3859  *   Pointer to Ethernet device.
3860  * @flow
3861  *   Parent flow for wich copying is provided.
3862  */
3863 static void
3864 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3865 			  struct rte_flow *flow)
3866 {
3867 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3868 	struct mlx5_priv *priv = dev->data->dev_private;
3869 
3870 	if (!flow->rix_mreg_copy)
3871 		return;
3872 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3873 				 flow->rix_mreg_copy);
3874 	if (!mcp_res || !priv->mreg_cp_tbl)
3875 		return;
3876 	if (flow->copy_applied) {
3877 		MLX5_ASSERT(mcp_res->appcnt);
3878 		flow->copy_applied = 0;
3879 		--mcp_res->appcnt;
3880 		if (!mcp_res->appcnt) {
3881 			struct rte_flow *mcp_flow = mlx5_ipool_get
3882 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3883 					mcp_res->rix_flow);
3884 
3885 			if (mcp_flow)
3886 				flow_drv_remove(dev, mcp_flow);
3887 		}
3888 	}
3889 	/*
3890 	 * We do not check availability of metadata registers here,
3891 	 * because copy resources are not allocated in this case.
3892 	 */
3893 	if (--mcp_res->refcnt)
3894 		return;
3895 	MLX5_ASSERT(mcp_res->rix_flow);
3896 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3897 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3898 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3899 	flow->rix_mreg_copy = 0;
3900 }
3901 
3902 /**
3903  * Start flow in RX_CP_TBL.
3904  *
3905  * @param dev
3906  *   Pointer to Ethernet device.
3907  * @flow
3908  *   Parent flow for wich copying is provided.
3909  *
3910  * @return
3911  *   0 on success, a negative errno value otherwise and rte_errno is set.
3912  */
3913 static int
3914 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3915 			    struct rte_flow *flow)
3916 {
3917 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3918 	struct mlx5_priv *priv = dev->data->dev_private;
3919 	int ret;
3920 
3921 	if (!flow->rix_mreg_copy || flow->copy_applied)
3922 		return 0;
3923 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3924 				 flow->rix_mreg_copy);
3925 	if (!mcp_res)
3926 		return 0;
3927 	if (!mcp_res->appcnt) {
3928 		struct rte_flow *mcp_flow = mlx5_ipool_get
3929 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3930 				mcp_res->rix_flow);
3931 
3932 		if (mcp_flow) {
3933 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3934 			if (ret)
3935 				return ret;
3936 		}
3937 	}
3938 	++mcp_res->appcnt;
3939 	flow->copy_applied = 1;
3940 	return 0;
3941 }
3942 
3943 /**
3944  * Stop flow in RX_CP_TBL.
3945  *
3946  * @param dev
3947  *   Pointer to Ethernet device.
3948  * @flow
3949  *   Parent flow for wich copying is provided.
3950  */
3951 static void
3952 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3953 			   struct rte_flow *flow)
3954 {
3955 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3956 	struct mlx5_priv *priv = dev->data->dev_private;
3957 
3958 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3959 		return;
3960 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3961 				 flow->rix_mreg_copy);
3962 	if (!mcp_res)
3963 		return;
3964 	MLX5_ASSERT(mcp_res->appcnt);
3965 	--mcp_res->appcnt;
3966 	flow->copy_applied = 0;
3967 	if (!mcp_res->appcnt) {
3968 		struct rte_flow *mcp_flow = mlx5_ipool_get
3969 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3970 				mcp_res->rix_flow);
3971 
3972 		if (mcp_flow)
3973 			flow_drv_remove(dev, mcp_flow);
3974 	}
3975 }
3976 
3977 /**
3978  * Remove the default copy action from RX_CP_TBL.
3979  *
3980  * @param dev
3981  *   Pointer to Ethernet device.
3982  */
3983 static void
3984 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3985 {
3986 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3987 	struct mlx5_priv *priv = dev->data->dev_private;
3988 
3989 	/* Check if default flow is registered. */
3990 	if (!priv->mreg_cp_tbl)
3991 		return;
3992 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3993 					    MLX5_DEFAULT_COPY_ID);
3994 	if (!mcp_res)
3995 		return;
3996 	MLX5_ASSERT(mcp_res->rix_flow);
3997 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3998 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3999 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4000 }
4001 
4002 /**
4003  * Add the default copy action in in RX_CP_TBL.
4004  *
4005  * @param dev
4006  *   Pointer to Ethernet device.
4007  * @param[out] error
4008  *   Perform verbose error reporting if not NULL.
4009  *
4010  * @return
4011  *   0 for success, negative value otherwise and rte_errno is set.
4012  */
4013 static int
4014 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4015 				  struct rte_flow_error *error)
4016 {
4017 	struct mlx5_priv *priv = dev->data->dev_private;
4018 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4019 
4020 	/* Check whether extensive metadata feature is engaged. */
4021 	if (!priv->config.dv_flow_en ||
4022 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4023 	    !mlx5_flow_ext_mreg_supported(dev) ||
4024 	    !priv->sh->dv_regc0_mask)
4025 		return 0;
4026 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
4027 	if (!mcp_res)
4028 		return -rte_errno;
4029 	return 0;
4030 }
4031 
4032 /**
4033  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4034  *
4035  * All the flow having Q/RSS action should be split by
4036  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4037  * performs the following,
4038  *   - CQE->flow_tag := reg_c[1] (MARK)
4039  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4040  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4041  * but there should be a flow per each MARK ID set by MARK action.
4042  *
4043  * For the aforementioned reason, if there's a MARK action in flow's action
4044  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4045  * the MARK ID to CQE's flow_tag like,
4046  *   - If reg_c[1] is mark_id,
4047  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4048  *
4049  * For SET_META action which stores value in reg_c[0], as the destination is
4050  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4051  * MARK ID means the default flow. The default flow looks like,
4052  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4053  *
4054  * @param dev
4055  *   Pointer to Ethernet device.
4056  * @param flow
4057  *   Pointer to flow structure.
4058  * @param[in] actions
4059  *   Pointer to the list of actions.
4060  * @param[out] error
4061  *   Perform verbose error reporting if not NULL.
4062  *
4063  * @return
4064  *   0 on success, negative value otherwise and rte_errno is set.
4065  */
4066 static int
4067 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4068 			    struct rte_flow *flow,
4069 			    const struct rte_flow_action *actions,
4070 			    struct rte_flow_error *error)
4071 {
4072 	struct mlx5_priv *priv = dev->data->dev_private;
4073 	struct mlx5_dev_config *config = &priv->config;
4074 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4075 	const struct rte_flow_action_mark *mark;
4076 
4077 	/* Check whether extensive metadata feature is engaged. */
4078 	if (!config->dv_flow_en ||
4079 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4080 	    !mlx5_flow_ext_mreg_supported(dev) ||
4081 	    !priv->sh->dv_regc0_mask)
4082 		return 0;
4083 	/* Find MARK action. */
4084 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4085 		switch (actions->type) {
4086 		case RTE_FLOW_ACTION_TYPE_FLAG:
4087 			mcp_res = flow_mreg_add_copy_action
4088 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4089 			if (!mcp_res)
4090 				return -rte_errno;
4091 			flow->rix_mreg_copy = mcp_res->idx;
4092 			if (dev->data->dev_started) {
4093 				mcp_res->appcnt++;
4094 				flow->copy_applied = 1;
4095 			}
4096 			return 0;
4097 		case RTE_FLOW_ACTION_TYPE_MARK:
4098 			mark = (const struct rte_flow_action_mark *)
4099 				actions->conf;
4100 			mcp_res =
4101 				flow_mreg_add_copy_action(dev, mark->id, error);
4102 			if (!mcp_res)
4103 				return -rte_errno;
4104 			flow->rix_mreg_copy = mcp_res->idx;
4105 			if (dev->data->dev_started) {
4106 				mcp_res->appcnt++;
4107 				flow->copy_applied = 1;
4108 			}
4109 			return 0;
4110 		default:
4111 			break;
4112 		}
4113 	}
4114 	return 0;
4115 }
4116 
4117 #define MLX5_MAX_SPLIT_ACTIONS 24
4118 #define MLX5_MAX_SPLIT_ITEMS 24
4119 
4120 /**
4121  * Split the hairpin flow.
4122  * Since HW can't support encap and push-vlan on Rx, we move these
4123  * actions to Tx.
4124  * If the count action is after the encap then we also
4125  * move the count action. in this case the count will also measure
4126  * the outer bytes.
4127  *
4128  * @param dev
4129  *   Pointer to Ethernet device.
4130  * @param[in] actions
4131  *   Associated actions (list terminated by the END action).
4132  * @param[out] actions_rx
4133  *   Rx flow actions.
4134  * @param[out] actions_tx
4135  *   Tx flow actions..
4136  * @param[out] pattern_tx
4137  *   The pattern items for the Tx flow.
4138  * @param[out] flow_id
4139  *   The flow ID connected to this flow.
4140  *
4141  * @return
4142  *   0 on success.
4143  */
4144 static int
4145 flow_hairpin_split(struct rte_eth_dev *dev,
4146 		   const struct rte_flow_action actions[],
4147 		   struct rte_flow_action actions_rx[],
4148 		   struct rte_flow_action actions_tx[],
4149 		   struct rte_flow_item pattern_tx[],
4150 		   uint32_t *flow_id)
4151 {
4152 	struct mlx5_priv *priv = dev->data->dev_private;
4153 	const struct rte_flow_action_raw_encap *raw_encap;
4154 	const struct rte_flow_action_raw_decap *raw_decap;
4155 	struct mlx5_rte_flow_action_set_tag *set_tag;
4156 	struct rte_flow_action *tag_action;
4157 	struct mlx5_rte_flow_item_tag *tag_item;
4158 	struct rte_flow_item *item;
4159 	char *addr;
4160 	int encap = 0;
4161 
4162 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
4163 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4164 		switch (actions->type) {
4165 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4166 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4167 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4168 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4169 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4170 			rte_memcpy(actions_tx, actions,
4171 			       sizeof(struct rte_flow_action));
4172 			actions_tx++;
4173 			break;
4174 		case RTE_FLOW_ACTION_TYPE_COUNT:
4175 			if (encap) {
4176 				rte_memcpy(actions_tx, actions,
4177 					   sizeof(struct rte_flow_action));
4178 				actions_tx++;
4179 			} else {
4180 				rte_memcpy(actions_rx, actions,
4181 					   sizeof(struct rte_flow_action));
4182 				actions_rx++;
4183 			}
4184 			break;
4185 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4186 			raw_encap = actions->conf;
4187 			if (raw_encap->size >
4188 			    (sizeof(struct rte_flow_item_eth) +
4189 			     sizeof(struct rte_flow_item_ipv4))) {
4190 				memcpy(actions_tx, actions,
4191 				       sizeof(struct rte_flow_action));
4192 				actions_tx++;
4193 				encap = 1;
4194 			} else {
4195 				rte_memcpy(actions_rx, actions,
4196 					   sizeof(struct rte_flow_action));
4197 				actions_rx++;
4198 			}
4199 			break;
4200 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4201 			raw_decap = actions->conf;
4202 			if (raw_decap->size <
4203 			    (sizeof(struct rte_flow_item_eth) +
4204 			     sizeof(struct rte_flow_item_ipv4))) {
4205 				memcpy(actions_tx, actions,
4206 				       sizeof(struct rte_flow_action));
4207 				actions_tx++;
4208 			} else {
4209 				rte_memcpy(actions_rx, actions,
4210 					   sizeof(struct rte_flow_action));
4211 				actions_rx++;
4212 			}
4213 			break;
4214 		default:
4215 			rte_memcpy(actions_rx, actions,
4216 				   sizeof(struct rte_flow_action));
4217 			actions_rx++;
4218 			break;
4219 		}
4220 	}
4221 	/* Add set meta action and end action for the Rx flow. */
4222 	tag_action = actions_rx;
4223 	tag_action->type = (enum rte_flow_action_type)
4224 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4225 	actions_rx++;
4226 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4227 	actions_rx++;
4228 	set_tag = (void *)actions_rx;
4229 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4230 	MLX5_ASSERT(set_tag->id > REG_NON);
4231 	set_tag->data = *flow_id;
4232 	tag_action->conf = set_tag;
4233 	/* Create Tx item list. */
4234 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4235 	addr = (void *)&pattern_tx[2];
4236 	item = pattern_tx;
4237 	item->type = (enum rte_flow_item_type)
4238 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4239 	tag_item = (void *)addr;
4240 	tag_item->data = *flow_id;
4241 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4242 	MLX5_ASSERT(set_tag->id > REG_NON);
4243 	item->spec = tag_item;
4244 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4245 	tag_item = (void *)addr;
4246 	tag_item->data = UINT32_MAX;
4247 	tag_item->id = UINT16_MAX;
4248 	item->mask = tag_item;
4249 	item->last = NULL;
4250 	item++;
4251 	item->type = RTE_FLOW_ITEM_TYPE_END;
4252 	return 0;
4253 }
4254 
4255 __extension__
4256 union tunnel_offload_mark {
4257 	uint32_t val;
4258 	struct {
4259 		uint32_t app_reserve:8;
4260 		uint32_t table_id:15;
4261 		uint32_t transfer:1;
4262 		uint32_t _unused_:8;
4263 	};
4264 };
4265 
4266 struct tunnel_default_miss_ctx {
4267 	uint16_t *queue;
4268 	__extension__
4269 	union {
4270 		struct rte_flow_action_rss action_rss;
4271 		struct rte_flow_action_queue miss_queue;
4272 		struct rte_flow_action_jump miss_jump;
4273 		uint8_t raw[0];
4274 	};
4275 };
4276 
4277 static int
4278 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4279 			     struct rte_flow *flow,
4280 			     const struct rte_flow_attr *attr,
4281 			     const struct rte_flow_action *app_actions,
4282 			     uint32_t flow_idx,
4283 			     struct tunnel_default_miss_ctx *ctx,
4284 			     struct rte_flow_error *error)
4285 {
4286 	struct mlx5_priv *priv = dev->data->dev_private;
4287 	struct mlx5_flow *dev_flow;
4288 	struct rte_flow_attr miss_attr = *attr;
4289 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4290 	const struct rte_flow_item miss_items[2] = {
4291 		{
4292 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4293 			.spec = NULL,
4294 			.last = NULL,
4295 			.mask = NULL
4296 		},
4297 		{
4298 			.type = RTE_FLOW_ITEM_TYPE_END,
4299 			.spec = NULL,
4300 			.last = NULL,
4301 			.mask = NULL
4302 		}
4303 	};
4304 	union tunnel_offload_mark mark_id;
4305 	struct rte_flow_action_mark miss_mark;
4306 	struct rte_flow_action miss_actions[3] = {
4307 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4308 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
4309 	};
4310 	const struct rte_flow_action_jump *jump_data;
4311 	uint32_t i, flow_table = 0; /* prevent compilation warning */
4312 	struct flow_grp_info grp_info = {
4313 		.external = 1,
4314 		.transfer = attr->transfer,
4315 		.fdb_def_rule = !!priv->fdb_def_rule,
4316 		.std_tbl_fix = 0,
4317 	};
4318 	int ret;
4319 
4320 	if (!attr->transfer) {
4321 		uint32_t q_size;
4322 
4323 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4324 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4325 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4326 					 0, SOCKET_ID_ANY);
4327 		if (!ctx->queue)
4328 			return rte_flow_error_set
4329 				(error, ENOMEM,
4330 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4331 				NULL, "invalid default miss RSS");
4332 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4333 		ctx->action_rss.level = 0,
4334 		ctx->action_rss.types = priv->rss_conf.rss_hf,
4335 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4336 		ctx->action_rss.queue_num = priv->reta_idx_n,
4337 		ctx->action_rss.key = priv->rss_conf.rss_key,
4338 		ctx->action_rss.queue = ctx->queue;
4339 		if (!priv->reta_idx_n || !priv->rxqs_n)
4340 			return rte_flow_error_set
4341 				(error, EINVAL,
4342 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4343 				NULL, "invalid port configuration");
4344 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4345 			ctx->action_rss.types = 0;
4346 		for (i = 0; i != priv->reta_idx_n; ++i)
4347 			ctx->queue[i] = (*priv->reta_idx)[i];
4348 	} else {
4349 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4350 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4351 	}
4352 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4353 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4354 	jump_data = app_actions->conf;
4355 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4356 	miss_attr.group = jump_data->group;
4357 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4358 				       &flow_table, grp_info, error);
4359 	if (ret)
4360 		return rte_flow_error_set(error, EINVAL,
4361 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4362 					  NULL, "invalid tunnel id");
4363 	mark_id.app_reserve = 0;
4364 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4365 	mark_id.transfer = !!attr->transfer;
4366 	mark_id._unused_ = 0;
4367 	miss_mark.id = mark_id.val;
4368 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4369 				    miss_items, miss_actions, flow_idx, error);
4370 	if (!dev_flow)
4371 		return -rte_errno;
4372 	dev_flow->flow = flow;
4373 	dev_flow->external = true;
4374 	dev_flow->tunnel = tunnel;
4375 	/* Subflow object was created, we must include one in the list. */
4376 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4377 		      dev_flow->handle, next);
4378 	DRV_LOG(DEBUG,
4379 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4380 		dev->data->port_id, tunnel->app_tunnel.type,
4381 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4382 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4383 				  miss_actions, error);
4384 	if (!ret)
4385 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4386 						  error);
4387 
4388 	return ret;
4389 }
4390 
4391 /**
4392  * The last stage of splitting chain, just creates the subflow
4393  * without any modification.
4394  *
4395  * @param[in] dev
4396  *   Pointer to Ethernet device.
4397  * @param[in] flow
4398  *   Parent flow structure pointer.
4399  * @param[in, out] sub_flow
4400  *   Pointer to return the created subflow, may be NULL.
4401  * @param[in] prefix_layers
4402  *   Prefix subflow layers, may be 0.
4403  * @param[in] prefix_mark
4404  *   Prefix subflow mark flag, may be 0.
4405  * @param[in] attr
4406  *   Flow rule attributes.
4407  * @param[in] items
4408  *   Pattern specification (list terminated by the END pattern item).
4409  * @param[in] actions
4410  *   Associated actions (list terminated by the END action).
4411  * @param[in] external
4412  *   This flow rule is created by request external to PMD.
4413  * @param[in] flow_idx
4414  *   This memory pool index to the flow.
4415  * @param[out] error
4416  *   Perform verbose error reporting if not NULL.
4417  * @return
4418  *   0 on success, negative value otherwise
4419  */
4420 static int
4421 flow_create_split_inner(struct rte_eth_dev *dev,
4422 			struct rte_flow *flow,
4423 			struct mlx5_flow **sub_flow,
4424 			uint64_t prefix_layers,
4425 			uint32_t prefix_mark,
4426 			const struct rte_flow_attr *attr,
4427 			const struct rte_flow_item items[],
4428 			const struct rte_flow_action actions[],
4429 			bool external, uint32_t flow_idx,
4430 			struct rte_flow_error *error)
4431 {
4432 	struct mlx5_flow *dev_flow;
4433 
4434 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4435 		flow_idx, error);
4436 	if (!dev_flow)
4437 		return -rte_errno;
4438 	dev_flow->flow = flow;
4439 	dev_flow->external = external;
4440 	/* Subflow object was created, we must include one in the list. */
4441 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4442 		      dev_flow->handle, next);
4443 	/*
4444 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4445 	 * flow may need some user defined item layer flags, and pass the
4446 	 * Metadate rxq mark flag to suffix flow as well.
4447 	 */
4448 	if (prefix_layers)
4449 		dev_flow->handle->layers = prefix_layers;
4450 	if (prefix_mark)
4451 		dev_flow->handle->mark = 1;
4452 	if (sub_flow)
4453 		*sub_flow = dev_flow;
4454 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4455 }
4456 
4457 /**
4458  * Split the meter flow.
4459  *
4460  * As meter flow will split to three sub flow, other than meter
4461  * action, the other actions make sense to only meter accepts
4462  * the packet. If it need to be dropped, no other additional
4463  * actions should be take.
4464  *
4465  * One kind of special action which decapsulates the L3 tunnel
4466  * header will be in the prefix sub flow, as not to take the
4467  * L3 tunnel header into account.
4468  *
4469  * @param dev
4470  *   Pointer to Ethernet device.
4471  * @param[in] items
4472  *   Pattern specification (list terminated by the END pattern item).
4473  * @param[out] sfx_items
4474  *   Suffix flow match items (list terminated by the END pattern item).
4475  * @param[in] actions
4476  *   Associated actions (list terminated by the END action).
4477  * @param[out] actions_sfx
4478  *   Suffix flow actions.
4479  * @param[out] actions_pre
4480  *   Prefix flow actions.
4481  * @param[out] pattern_sfx
4482  *   The pattern items for the suffix flow.
4483  * @param[out] tag_sfx
4484  *   Pointer to suffix flow tag.
4485  *
4486  * @return
4487  *   0 on success.
4488  */
4489 static int
4490 flow_meter_split_prep(struct rte_eth_dev *dev,
4491 		 const struct rte_flow_item items[],
4492 		 struct rte_flow_item sfx_items[],
4493 		 const struct rte_flow_action actions[],
4494 		 struct rte_flow_action actions_sfx[],
4495 		 struct rte_flow_action actions_pre[])
4496 {
4497 	struct rte_flow_action *tag_action = NULL;
4498 	struct rte_flow_item *tag_item;
4499 	struct mlx5_rte_flow_action_set_tag *set_tag;
4500 	struct rte_flow_error error;
4501 	const struct rte_flow_action_raw_encap *raw_encap;
4502 	const struct rte_flow_action_raw_decap *raw_decap;
4503 	struct mlx5_rte_flow_item_tag *tag_spec;
4504 	struct mlx5_rte_flow_item_tag *tag_mask;
4505 	uint32_t tag_id;
4506 	bool copy_vlan = false;
4507 
4508 	/* Prepare the actions for prefix and suffix flow. */
4509 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4510 		struct rte_flow_action **action_cur = NULL;
4511 
4512 		switch (actions->type) {
4513 		case RTE_FLOW_ACTION_TYPE_METER:
4514 			/* Add the extra tag action first. */
4515 			tag_action = actions_pre;
4516 			tag_action->type = (enum rte_flow_action_type)
4517 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4518 			actions_pre++;
4519 			action_cur = &actions_pre;
4520 			break;
4521 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4522 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4523 			action_cur = &actions_pre;
4524 			break;
4525 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4526 			raw_encap = actions->conf;
4527 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4528 				action_cur = &actions_pre;
4529 			break;
4530 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4531 			raw_decap = actions->conf;
4532 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4533 				action_cur = &actions_pre;
4534 			break;
4535 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4536 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4537 			copy_vlan = true;
4538 			break;
4539 		default:
4540 			break;
4541 		}
4542 		if (!action_cur)
4543 			action_cur = &actions_sfx;
4544 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4545 		(*action_cur)++;
4546 	}
4547 	/* Add end action to the actions. */
4548 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4549 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4550 	actions_pre++;
4551 	/* Set the tag. */
4552 	set_tag = (void *)actions_pre;
4553 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4554 	/*
4555 	 * Get the id from the qrss_pool to make qrss share the id with meter.
4556 	 */
4557 	tag_id = flow_qrss_get_id(dev);
4558 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4559 	assert(tag_action);
4560 	tag_action->conf = set_tag;
4561 	/* Prepare the suffix subflow items. */
4562 	tag_item = sfx_items++;
4563 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4564 		int item_type = items->type;
4565 
4566 		switch (item_type) {
4567 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4568 			memcpy(sfx_items, items, sizeof(*sfx_items));
4569 			sfx_items++;
4570 			break;
4571 		case RTE_FLOW_ITEM_TYPE_VLAN:
4572 			if (copy_vlan) {
4573 				memcpy(sfx_items, items, sizeof(*sfx_items));
4574 				/*
4575 				 * Convert to internal match item, it is used
4576 				 * for vlan push and set vid.
4577 				 */
4578 				sfx_items->type = (enum rte_flow_item_type)
4579 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4580 				sfx_items++;
4581 			}
4582 			break;
4583 		default:
4584 			break;
4585 		}
4586 	}
4587 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4588 	sfx_items++;
4589 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4590 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4591 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4592 	tag_mask = tag_spec + 1;
4593 	tag_mask->data = 0xffffff00;
4594 	tag_item->type = (enum rte_flow_item_type)
4595 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4596 	tag_item->spec = tag_spec;
4597 	tag_item->last = NULL;
4598 	tag_item->mask = tag_mask;
4599 	return tag_id;
4600 }
4601 
4602 /**
4603  * Split action list having QUEUE/RSS for metadata register copy.
4604  *
4605  * Once Q/RSS action is detected in user's action list, the flow action
4606  * should be split in order to copy metadata registers, which will happen in
4607  * RX_CP_TBL like,
4608  *   - CQE->flow_tag := reg_c[1] (MARK)
4609  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4610  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4611  * This is because the last action of each flow must be a terminal action
4612  * (QUEUE, RSS or DROP).
4613  *
4614  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4615  * stored and kept in the mlx5_flow structure per each sub_flow.
4616  *
4617  * The Q/RSS action is replaced with,
4618  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4619  * And the following JUMP action is added at the end,
4620  *   - JUMP, to RX_CP_TBL.
4621  *
4622  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4623  * flow_create_split_metadata() routine. The flow will look like,
4624  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4625  *
4626  * @param dev
4627  *   Pointer to Ethernet device.
4628  * @param[out] split_actions
4629  *   Pointer to store split actions to jump to CP_TBL.
4630  * @param[in] actions
4631  *   Pointer to the list of original flow actions.
4632  * @param[in] qrss
4633  *   Pointer to the Q/RSS action.
4634  * @param[in] actions_n
4635  *   Number of original actions.
4636  * @param[out] error
4637  *   Perform verbose error reporting if not NULL.
4638  *
4639  * @return
4640  *   non-zero unique flow_id on success, otherwise 0 and
4641  *   error/rte_error are set.
4642  */
4643 static uint32_t
4644 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4645 			  struct rte_flow_action *split_actions,
4646 			  const struct rte_flow_action *actions,
4647 			  const struct rte_flow_action *qrss,
4648 			  int actions_n, struct rte_flow_error *error)
4649 {
4650 	struct mlx5_rte_flow_action_set_tag *set_tag;
4651 	struct rte_flow_action_jump *jump;
4652 	const int qrss_idx = qrss - actions;
4653 	uint32_t flow_id = 0;
4654 	int ret = 0;
4655 
4656 	/*
4657 	 * Given actions will be split
4658 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4659 	 * - Add jump to mreg CP_TBL.
4660 	 * As a result, there will be one more action.
4661 	 */
4662 	++actions_n;
4663 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4664 	set_tag = (void *)(split_actions + actions_n);
4665 	/*
4666 	 * If tag action is not set to void(it means we are not the meter
4667 	 * suffix flow), add the tag action. Since meter suffix flow already
4668 	 * has the tag added.
4669 	 */
4670 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4671 		/*
4672 		 * Allocate the new subflow ID. This one is unique within
4673 		 * device and not shared with representors. Otherwise,
4674 		 * we would have to resolve multi-thread access synch
4675 		 * issue. Each flow on the shared device is appended
4676 		 * with source vport identifier, so the resulting
4677 		 * flows will be unique in the shared (by master and
4678 		 * representors) domain even if they have coinciding
4679 		 * IDs.
4680 		 */
4681 		flow_id = flow_qrss_get_id(dev);
4682 		if (!flow_id)
4683 			return rte_flow_error_set(error, ENOMEM,
4684 						  RTE_FLOW_ERROR_TYPE_ACTION,
4685 						  NULL, "can't allocate id "
4686 						  "for split Q/RSS subflow");
4687 		/* Internal SET_TAG action to set flow ID. */
4688 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4689 			.data = flow_id,
4690 		};
4691 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4692 		if (ret < 0)
4693 			return ret;
4694 		set_tag->id = ret;
4695 		/* Construct new actions array. */
4696 		/* Replace QUEUE/RSS action. */
4697 		split_actions[qrss_idx] = (struct rte_flow_action){
4698 			.type = (enum rte_flow_action_type)
4699 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4700 			.conf = set_tag,
4701 		};
4702 	}
4703 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4704 	jump = (void *)(set_tag + 1);
4705 	*jump = (struct rte_flow_action_jump){
4706 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4707 	};
4708 	split_actions[actions_n - 2] = (struct rte_flow_action){
4709 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4710 		.conf = jump,
4711 	};
4712 	split_actions[actions_n - 1] = (struct rte_flow_action){
4713 		.type = RTE_FLOW_ACTION_TYPE_END,
4714 	};
4715 	return flow_id;
4716 }
4717 
4718 /**
4719  * Extend the given action list for Tx metadata copy.
4720  *
4721  * Copy the given action list to the ext_actions and add flow metadata register
4722  * copy action in order to copy reg_a set by WQE to reg_c[0].
4723  *
4724  * @param[out] ext_actions
4725  *   Pointer to the extended action list.
4726  * @param[in] actions
4727  *   Pointer to the list of actions.
4728  * @param[in] actions_n
4729  *   Number of actions in the list.
4730  * @param[out] error
4731  *   Perform verbose error reporting if not NULL.
4732  * @param[in] encap_idx
4733  *   The encap action inndex.
4734  *
4735  * @return
4736  *   0 on success, negative value otherwise
4737  */
4738 static int
4739 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4740 		       struct rte_flow_action *ext_actions,
4741 		       const struct rte_flow_action *actions,
4742 		       int actions_n, struct rte_flow_error *error,
4743 		       int encap_idx)
4744 {
4745 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4746 		(struct mlx5_flow_action_copy_mreg *)
4747 			(ext_actions + actions_n + 1);
4748 	int ret;
4749 
4750 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4751 	if (ret < 0)
4752 		return ret;
4753 	cp_mreg->dst = ret;
4754 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4755 	if (ret < 0)
4756 		return ret;
4757 	cp_mreg->src = ret;
4758 	if (encap_idx != 0)
4759 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4760 	if (encap_idx == actions_n - 1) {
4761 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4762 			.type = (enum rte_flow_action_type)
4763 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4764 			.conf = cp_mreg,
4765 		};
4766 		ext_actions[actions_n] = (struct rte_flow_action){
4767 			.type = RTE_FLOW_ACTION_TYPE_END,
4768 		};
4769 	} else {
4770 		ext_actions[encap_idx] = (struct rte_flow_action){
4771 			.type = (enum rte_flow_action_type)
4772 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4773 			.conf = cp_mreg,
4774 		};
4775 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4776 				sizeof(*ext_actions) * (actions_n - encap_idx));
4777 	}
4778 	return 0;
4779 }
4780 
4781 /**
4782  * Check the match action from the action list.
4783  *
4784  * @param[in] actions
4785  *   Pointer to the list of actions.
4786  * @param[in] attr
4787  *   Flow rule attributes.
4788  * @param[in] action
4789  *   The action to be check if exist.
4790  * @param[out] match_action_pos
4791  *   Pointer to the position of the matched action if exists, otherwise is -1.
4792  * @param[out] qrss_action_pos
4793  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4794  *
4795  * @return
4796  *   > 0 the total number of actions.
4797  *   0 if not found match action in action list.
4798  */
4799 static int
4800 flow_check_match_action(const struct rte_flow_action actions[],
4801 			const struct rte_flow_attr *attr,
4802 			enum rte_flow_action_type action,
4803 			int *match_action_pos, int *qrss_action_pos)
4804 {
4805 	const struct rte_flow_action_sample *sample;
4806 	int actions_n = 0;
4807 	int jump_flag = 0;
4808 	uint32_t ratio = 0;
4809 	int sub_type = 0;
4810 	int flag = 0;
4811 
4812 	*match_action_pos = -1;
4813 	*qrss_action_pos = -1;
4814 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4815 		if (actions->type == action) {
4816 			flag = 1;
4817 			*match_action_pos = actions_n;
4818 		}
4819 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4820 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4821 			*qrss_action_pos = actions_n;
4822 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4823 			jump_flag = 1;
4824 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4825 			sample = actions->conf;
4826 			ratio = sample->ratio;
4827 			sub_type = ((const struct rte_flow_action *)
4828 					(sample->actions))->type;
4829 		}
4830 		actions_n++;
4831 	}
4832 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4833 		if (ratio == 1) {
4834 			/* JUMP Action not support for Mirroring;
4835 			 * Mirroring support multi-destination;
4836 			 */
4837 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4838 				flag = 0;
4839 		}
4840 	}
4841 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4842 	return flag ? actions_n + 1 : 0;
4843 }
4844 
4845 #define SAMPLE_SUFFIX_ITEM 2
4846 
4847 /**
4848  * Split the sample flow.
4849  *
4850  * As sample flow will split to two sub flow, sample flow with
4851  * sample action, the other actions will move to new suffix flow.
4852  *
4853  * Also add unique tag id with tag action in the sample flow,
4854  * the same tag id will be as match in the suffix flow.
4855  *
4856  * @param dev
4857  *   Pointer to Ethernet device.
4858  * @param[in] fdb_tx
4859  *   FDB egress flow flag.
4860  * @param[out] sfx_items
4861  *   Suffix flow match items (list terminated by the END pattern item).
4862  * @param[in] actions
4863  *   Associated actions (list terminated by the END action).
4864  * @param[out] actions_sfx
4865  *   Suffix flow actions.
4866  * @param[out] actions_pre
4867  *   Prefix flow actions.
4868  * @param[in] actions_n
4869  *  The total number of actions.
4870  * @param[in] sample_action_pos
4871  *   The sample action position.
4872  * @param[in] qrss_action_pos
4873  *   The Queue/RSS action position.
4874  * @param[out] error
4875  *   Perform verbose error reporting if not NULL.
4876  *
4877  * @return
4878  *   0 on success, or unique flow_id, a negative errno value
4879  *   otherwise and rte_errno is set.
4880  */
4881 static int
4882 flow_sample_split_prep(struct rte_eth_dev *dev,
4883 		       uint32_t fdb_tx,
4884 		       struct rte_flow_item sfx_items[],
4885 		       const struct rte_flow_action actions[],
4886 		       struct rte_flow_action actions_sfx[],
4887 		       struct rte_flow_action actions_pre[],
4888 		       int actions_n,
4889 		       int sample_action_pos,
4890 		       int qrss_action_pos,
4891 		       struct rte_flow_error *error)
4892 {
4893 	struct mlx5_rte_flow_action_set_tag *set_tag;
4894 	struct mlx5_rte_flow_item_tag *tag_spec;
4895 	struct mlx5_rte_flow_item_tag *tag_mask;
4896 	uint32_t tag_id = 0;
4897 	int index;
4898 	int ret;
4899 
4900 	if (sample_action_pos < 0)
4901 		return rte_flow_error_set(error, EINVAL,
4902 					  RTE_FLOW_ERROR_TYPE_ACTION,
4903 					  NULL, "invalid position of sample "
4904 					  "action in list");
4905 	if (!fdb_tx) {
4906 		/* Prepare the prefix tag action. */
4907 		set_tag = (void *)(actions_pre + actions_n + 1);
4908 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4909 		if (ret < 0)
4910 			return ret;
4911 		set_tag->id = ret;
4912 		tag_id = flow_qrss_get_id(dev);
4913 		set_tag->data = tag_id;
4914 		/* Prepare the suffix subflow items. */
4915 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4916 		tag_spec->data = tag_id;
4917 		tag_spec->id = set_tag->id;
4918 		tag_mask = tag_spec + 1;
4919 		tag_mask->data = UINT32_MAX;
4920 		sfx_items[0] = (struct rte_flow_item){
4921 			.type = (enum rte_flow_item_type)
4922 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4923 			.spec = tag_spec,
4924 			.last = NULL,
4925 			.mask = tag_mask,
4926 		};
4927 		sfx_items[1] = (struct rte_flow_item){
4928 			.type = (enum rte_flow_item_type)
4929 				RTE_FLOW_ITEM_TYPE_END,
4930 		};
4931 	}
4932 	/* Prepare the actions for prefix and suffix flow. */
4933 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4934 		index = qrss_action_pos;
4935 		/* Put the preceding the Queue/RSS action into prefix flow. */
4936 		if (index != 0)
4937 			memcpy(actions_pre, actions,
4938 			       sizeof(struct rte_flow_action) * index);
4939 		/* Put others preceding the sample action into prefix flow. */
4940 		if (sample_action_pos > index + 1)
4941 			memcpy(actions_pre + index, actions + index + 1,
4942 			       sizeof(struct rte_flow_action) *
4943 			       (sample_action_pos - index - 1));
4944 		index = sample_action_pos - 1;
4945 		/* Put Queue/RSS action into Suffix flow. */
4946 		memcpy(actions_sfx, actions + qrss_action_pos,
4947 		       sizeof(struct rte_flow_action));
4948 		actions_sfx++;
4949 	} else {
4950 		index = sample_action_pos;
4951 		if (index != 0)
4952 			memcpy(actions_pre, actions,
4953 			       sizeof(struct rte_flow_action) * index);
4954 	}
4955 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4956 	if (!fdb_tx) {
4957 		actions_pre[index++] =
4958 			(struct rte_flow_action){
4959 			.type = (enum rte_flow_action_type)
4960 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4961 			.conf = set_tag,
4962 		};
4963 	}
4964 	memcpy(actions_pre + index, actions + sample_action_pos,
4965 	       sizeof(struct rte_flow_action));
4966 	index += 1;
4967 	actions_pre[index] = (struct rte_flow_action){
4968 		.type = (enum rte_flow_action_type)
4969 			RTE_FLOW_ACTION_TYPE_END,
4970 	};
4971 	/* Put the actions after sample into Suffix flow. */
4972 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4973 	       sizeof(struct rte_flow_action) *
4974 	       (actions_n - sample_action_pos - 1));
4975 	return tag_id;
4976 }
4977 
4978 /**
4979  * The splitting for metadata feature.
4980  *
4981  * - Q/RSS action on NIC Rx should be split in order to pass by
4982  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4983  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4984  *
4985  * - All the actions on NIC Tx should have a mreg copy action to
4986  *   copy reg_a from WQE to reg_c[0].
4987  *
4988  * @param dev
4989  *   Pointer to Ethernet device.
4990  * @param[in] flow
4991  *   Parent flow structure pointer.
4992  * @param[in] prefix_layers
4993  *   Prefix flow layer flags.
4994  * @param[in] prefix_mark
4995  *   Prefix subflow mark flag, may be 0.
4996  * @param[in] attr
4997  *   Flow rule attributes.
4998  * @param[in] items
4999  *   Pattern specification (list terminated by the END pattern item).
5000  * @param[in] actions
5001  *   Associated actions (list terminated by the END action).
5002  * @param[in] external
5003  *   This flow rule is created by request external to PMD.
5004  * @param[in] flow_idx
5005  *   This memory pool index to the flow.
5006  * @param[out] error
5007  *   Perform verbose error reporting if not NULL.
5008  * @return
5009  *   0 on success, negative value otherwise
5010  */
5011 static int
5012 flow_create_split_metadata(struct rte_eth_dev *dev,
5013 			   struct rte_flow *flow,
5014 			   uint64_t prefix_layers,
5015 			   uint32_t prefix_mark,
5016 			   const struct rte_flow_attr *attr,
5017 			   const struct rte_flow_item items[],
5018 			   const struct rte_flow_action actions[],
5019 			   bool external, uint32_t flow_idx,
5020 			   struct rte_flow_error *error)
5021 {
5022 	struct mlx5_priv *priv = dev->data->dev_private;
5023 	struct mlx5_dev_config *config = &priv->config;
5024 	const struct rte_flow_action *qrss = NULL;
5025 	struct rte_flow_action *ext_actions = NULL;
5026 	struct mlx5_flow *dev_flow = NULL;
5027 	uint32_t qrss_id = 0;
5028 	int mtr_sfx = 0;
5029 	size_t act_size;
5030 	int actions_n;
5031 	int encap_idx;
5032 	int ret;
5033 
5034 	/* Check whether extensive metadata feature is engaged. */
5035 	if (!config->dv_flow_en ||
5036 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5037 	    !mlx5_flow_ext_mreg_supported(dev))
5038 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
5039 					       prefix_mark, attr, items,
5040 					       actions, external, flow_idx,
5041 					       error);
5042 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5043 							   &encap_idx);
5044 	if (qrss) {
5045 		/* Exclude hairpin flows from splitting. */
5046 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5047 			const struct rte_flow_action_queue *queue;
5048 
5049 			queue = qrss->conf;
5050 			if (mlx5_rxq_get_type(dev, queue->index) ==
5051 			    MLX5_RXQ_TYPE_HAIRPIN)
5052 				qrss = NULL;
5053 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5054 			const struct rte_flow_action_rss *rss;
5055 
5056 			rss = qrss->conf;
5057 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5058 			    MLX5_RXQ_TYPE_HAIRPIN)
5059 				qrss = NULL;
5060 		}
5061 	}
5062 	if (qrss) {
5063 		/* Check if it is in meter suffix table. */
5064 		mtr_sfx = attr->group == (attr->transfer ?
5065 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5066 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
5067 		/*
5068 		 * Q/RSS action on NIC Rx should be split in order to pass by
5069 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5070 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5071 		 */
5072 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5073 			   sizeof(struct rte_flow_action_set_tag) +
5074 			   sizeof(struct rte_flow_action_jump);
5075 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5076 					  SOCKET_ID_ANY);
5077 		if (!ext_actions)
5078 			return rte_flow_error_set(error, ENOMEM,
5079 						  RTE_FLOW_ERROR_TYPE_ACTION,
5080 						  NULL, "no memory to split "
5081 						  "metadata flow");
5082 		/*
5083 		 * If we are the suffix flow of meter, tag already exist.
5084 		 * Set the tag action to void.
5085 		 */
5086 		if (mtr_sfx)
5087 			ext_actions[qrss - actions].type =
5088 						RTE_FLOW_ACTION_TYPE_VOID;
5089 		else
5090 			ext_actions[qrss - actions].type =
5091 						(enum rte_flow_action_type)
5092 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5093 		/*
5094 		 * Create the new actions list with removed Q/RSS action
5095 		 * and appended set tag and jump to register copy table
5096 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5097 		 * in advance, because it is needed for set tag action.
5098 		 */
5099 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5100 						    qrss, actions_n, error);
5101 		if (!mtr_sfx && !qrss_id) {
5102 			ret = -rte_errno;
5103 			goto exit;
5104 		}
5105 	} else if (attr->egress && !attr->transfer) {
5106 		/*
5107 		 * All the actions on NIC Tx should have a metadata register
5108 		 * copy action to copy reg_a from WQE to reg_c[meta]
5109 		 */
5110 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5111 			   sizeof(struct mlx5_flow_action_copy_mreg);
5112 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5113 					  SOCKET_ID_ANY);
5114 		if (!ext_actions)
5115 			return rte_flow_error_set(error, ENOMEM,
5116 						  RTE_FLOW_ERROR_TYPE_ACTION,
5117 						  NULL, "no memory to split "
5118 						  "metadata flow");
5119 		/* Create the action list appended with copy register. */
5120 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5121 					     actions_n, error, encap_idx);
5122 		if (ret < 0)
5123 			goto exit;
5124 	}
5125 	/* Add the unmodified original or prefix subflow. */
5126 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers,
5127 				      prefix_mark, attr,
5128 				      items, ext_actions ? ext_actions :
5129 				      actions, external, flow_idx, error);
5130 	if (ret < 0)
5131 		goto exit;
5132 	MLX5_ASSERT(dev_flow);
5133 	if (qrss) {
5134 		const struct rte_flow_attr q_attr = {
5135 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5136 			.ingress = 1,
5137 		};
5138 		/* Internal PMD action to set register. */
5139 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5140 			.data = qrss_id,
5141 			.id = REG_NON,
5142 		};
5143 		struct rte_flow_item q_items[] = {
5144 			{
5145 				.type = (enum rte_flow_item_type)
5146 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5147 				.spec = &q_tag_spec,
5148 				.last = NULL,
5149 				.mask = NULL,
5150 			},
5151 			{
5152 				.type = RTE_FLOW_ITEM_TYPE_END,
5153 			},
5154 		};
5155 		struct rte_flow_action q_actions[] = {
5156 			{
5157 				.type = qrss->type,
5158 				.conf = qrss->conf,
5159 			},
5160 			{
5161 				.type = RTE_FLOW_ACTION_TYPE_END,
5162 			},
5163 		};
5164 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5165 
5166 		/*
5167 		 * Configure the tag item only if there is no meter subflow.
5168 		 * Since tag is already marked in the meter suffix subflow
5169 		 * we can just use the meter suffix items as is.
5170 		 */
5171 		if (qrss_id) {
5172 			/* Not meter subflow. */
5173 			MLX5_ASSERT(!mtr_sfx);
5174 			/*
5175 			 * Put unique id in prefix flow due to it is destroyed
5176 			 * after suffix flow and id will be freed after there
5177 			 * is no actual flows with this id and identifier
5178 			 * reallocation becomes possible (for example, for
5179 			 * other flows in other threads).
5180 			 */
5181 			dev_flow->handle->split_flow_id = qrss_id;
5182 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5183 						   error);
5184 			if (ret < 0)
5185 				goto exit;
5186 			q_tag_spec.id = ret;
5187 		}
5188 		dev_flow = NULL;
5189 		/* Add suffix subflow to execute Q/RSS. */
5190 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers, 0,
5191 					      &q_attr, mtr_sfx ? items :
5192 					      q_items, q_actions,
5193 					      external, flow_idx, error);
5194 		if (ret < 0)
5195 			goto exit;
5196 		/* qrss ID should be freed if failed. */
5197 		qrss_id = 0;
5198 		MLX5_ASSERT(dev_flow);
5199 	}
5200 
5201 exit:
5202 	/*
5203 	 * We do not destroy the partially created sub_flows in case of error.
5204 	 * These ones are included into parent flow list and will be destroyed
5205 	 * by flow_drv_destroy.
5206 	 */
5207 	flow_qrss_free_id(dev, qrss_id);
5208 	mlx5_free(ext_actions);
5209 	return ret;
5210 }
5211 
5212 /**
5213  * The splitting for meter feature.
5214  *
5215  * - The meter flow will be split to two flows as prefix and
5216  *   suffix flow. The packets make sense only it pass the prefix
5217  *   meter action.
5218  *
5219  * - Reg_C_5 is used for the packet to match betweend prefix and
5220  *   suffix flow.
5221  *
5222  * @param dev
5223  *   Pointer to Ethernet device.
5224  * @param[in] flow
5225  *   Parent flow structure pointer.
5226  * @param[in] prefix_layers
5227  *   Prefix subflow layers, may be 0.
5228  * @param[in] prefix_mark
5229  *   Prefix subflow mark flag, may be 0.
5230  * @param[in] attr
5231  *   Flow rule attributes.
5232  * @param[in] items
5233  *   Pattern specification (list terminated by the END pattern item).
5234  * @param[in] actions
5235  *   Associated actions (list terminated by the END action).
5236  * @param[in] external
5237  *   This flow rule is created by request external to PMD.
5238  * @param[in] flow_idx
5239  *   This memory pool index to the flow.
5240  * @param[out] error
5241  *   Perform verbose error reporting if not NULL.
5242  * @return
5243  *   0 on success, negative value otherwise
5244  */
5245 static int
5246 flow_create_split_meter(struct rte_eth_dev *dev,
5247 			struct rte_flow *flow,
5248 			uint64_t prefix_layers,
5249 			uint32_t prefix_mark,
5250 			const struct rte_flow_attr *attr,
5251 			const struct rte_flow_item items[],
5252 			const struct rte_flow_action actions[],
5253 			bool external, uint32_t flow_idx,
5254 			struct rte_flow_error *error)
5255 {
5256 	struct mlx5_priv *priv = dev->data->dev_private;
5257 	struct rte_flow_action *sfx_actions = NULL;
5258 	struct rte_flow_action *pre_actions = NULL;
5259 	struct rte_flow_item *sfx_items = NULL;
5260 	struct mlx5_flow *dev_flow = NULL;
5261 	struct rte_flow_attr sfx_attr = *attr;
5262 	uint32_t mtr = 0;
5263 	uint32_t mtr_tag_id = 0;
5264 	size_t act_size;
5265 	size_t item_size;
5266 	int actions_n = 0;
5267 	int ret;
5268 
5269 	if (priv->mtr_en)
5270 		actions_n = flow_check_meter_action(actions, &mtr);
5271 	if (mtr) {
5272 		/* The five prefix actions: meter, decap, encap, tag, end. */
5273 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5274 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5275 		/* tag, vlan, port id, end. */
5276 #define METER_SUFFIX_ITEM 4
5277 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5278 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5279 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5280 					  0, SOCKET_ID_ANY);
5281 		if (!sfx_actions)
5282 			return rte_flow_error_set(error, ENOMEM,
5283 						  RTE_FLOW_ERROR_TYPE_ACTION,
5284 						  NULL, "no memory to split "
5285 						  "meter flow");
5286 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5287 			     act_size);
5288 		pre_actions = sfx_actions + actions_n;
5289 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5290 						   actions, sfx_actions,
5291 						   pre_actions);
5292 		if (!mtr_tag_id) {
5293 			ret = -rte_errno;
5294 			goto exit;
5295 		}
5296 		/* Add the prefix subflow. */
5297 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5298 					      prefix_layers, 0,
5299 					      attr, items,
5300 					      pre_actions, external,
5301 					      flow_idx, error);
5302 		if (ret) {
5303 			ret = -rte_errno;
5304 			goto exit;
5305 		}
5306 		dev_flow->handle->split_flow_id = mtr_tag_id;
5307 		/* Setting the sfx group atrr. */
5308 		sfx_attr.group = sfx_attr.transfer ?
5309 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5310 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5311 	}
5312 	/* Add the prefix subflow. */
5313 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
5314 					 flow_get_prefix_layer_flags(dev_flow) :
5315 					 prefix_layers, dev_flow ?
5316 					 dev_flow->handle->mark : prefix_mark,
5317 					 &sfx_attr, sfx_items ?
5318 					 sfx_items : items,
5319 					 sfx_actions ? sfx_actions : actions,
5320 					 external, flow_idx, error);
5321 exit:
5322 	if (sfx_actions)
5323 		mlx5_free(sfx_actions);
5324 	return ret;
5325 }
5326 
5327 /**
5328  * The splitting for sample feature.
5329  *
5330  * Once Sample action is detected in the action list, the flow actions should
5331  * be split into prefix sub flow and suffix sub flow.
5332  *
5333  * The original items remain in the prefix sub flow, all actions preceding the
5334  * sample action and the sample action itself will be copied to the prefix
5335  * sub flow, the actions following the sample action will be copied to the
5336  * suffix sub flow, Queue action always be located in the suffix sub flow.
5337  *
5338  * In order to make the packet from prefix sub flow matches with suffix sub
5339  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5340  * flow uses tag item with the unique flow id.
5341  *
5342  * @param dev
5343  *   Pointer to Ethernet device.
5344  * @param[in] flow
5345  *   Parent flow structure pointer.
5346  * @param[in] attr
5347  *   Flow rule attributes.
5348  * @param[in] items
5349  *   Pattern specification (list terminated by the END pattern item).
5350  * @param[in] actions
5351  *   Associated actions (list terminated by the END action).
5352  * @param[in] external
5353  *   This flow rule is created by request external to PMD.
5354  * @param[in] flow_idx
5355  *   This memory pool index to the flow.
5356  * @param[out] error
5357  *   Perform verbose error reporting if not NULL.
5358  * @return
5359  *   0 on success, negative value otherwise
5360  */
5361 static int
5362 flow_create_split_sample(struct rte_eth_dev *dev,
5363 			 struct rte_flow *flow,
5364 			 const struct rte_flow_attr *attr,
5365 			 const struct rte_flow_item items[],
5366 			 const struct rte_flow_action actions[],
5367 			 bool external, uint32_t flow_idx,
5368 			 struct rte_flow_error *error)
5369 {
5370 	struct mlx5_priv *priv = dev->data->dev_private;
5371 	struct rte_flow_action *sfx_actions = NULL;
5372 	struct rte_flow_action *pre_actions = NULL;
5373 	struct rte_flow_item *sfx_items = NULL;
5374 	struct mlx5_flow *dev_flow = NULL;
5375 	struct rte_flow_attr sfx_attr = *attr;
5376 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5377 	struct mlx5_flow_dv_sample_resource *sample_res;
5378 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5379 	struct mlx5_flow_tbl_resource *sfx_tbl;
5380 	union mlx5_flow_tbl_key sfx_table_key;
5381 #endif
5382 	size_t act_size;
5383 	size_t item_size;
5384 	uint32_t fdb_tx = 0;
5385 	int32_t tag_id = 0;
5386 	int actions_n = 0;
5387 	int sample_action_pos;
5388 	int qrss_action_pos;
5389 	int ret = 0;
5390 
5391 	if (priv->sampler_en)
5392 		actions_n = flow_check_match_action(actions, attr,
5393 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5394 					&sample_action_pos, &qrss_action_pos);
5395 	if (actions_n) {
5396 		/* The prefix actions must includes sample, tag, end. */
5397 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5398 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5399 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5400 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5401 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5402 					  item_size), 0, SOCKET_ID_ANY);
5403 		if (!sfx_actions)
5404 			return rte_flow_error_set(error, ENOMEM,
5405 						  RTE_FLOW_ERROR_TYPE_ACTION,
5406 						  NULL, "no memory to split "
5407 						  "sample flow");
5408 		/* The representor_id is -1 for uplink. */
5409 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5410 		if (!fdb_tx)
5411 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5412 					+ act_size);
5413 		pre_actions = sfx_actions + actions_n;
5414 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5415 						actions, sfx_actions,
5416 						pre_actions, actions_n,
5417 						sample_action_pos,
5418 						qrss_action_pos, error);
5419 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5420 			ret = -rte_errno;
5421 			goto exit;
5422 		}
5423 		/* Add the prefix subflow. */
5424 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, 0, attr,
5425 					      items, pre_actions, external,
5426 					      flow_idx, error);
5427 		if (ret) {
5428 			ret = -rte_errno;
5429 			goto exit;
5430 		}
5431 		dev_flow->handle->split_flow_id = tag_id;
5432 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5433 		/* Set the sfx group attr. */
5434 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5435 					dev_flow->dv.sample_res;
5436 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5437 					sample_res->normal_path_tbl;
5438 		sfx_tbl_data = container_of(sfx_tbl,
5439 					struct mlx5_flow_tbl_data_entry, tbl);
5440 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5441 		sfx_attr.group = sfx_attr.transfer ?
5442 					(sfx_table_key.table_id - 1) :
5443 					 sfx_table_key.table_id;
5444 #endif
5445 	}
5446 	/* Add the suffix subflow. */
5447 	ret = flow_create_split_meter(dev, flow, dev_flow ?
5448 				 flow_get_prefix_layer_flags(dev_flow) : 0,
5449 				 dev_flow ? dev_flow->handle->mark : 0,
5450 				 &sfx_attr, sfx_items ? sfx_items : items,
5451 				 sfx_actions ? sfx_actions : actions,
5452 				 external, flow_idx, error);
5453 exit:
5454 	if (sfx_actions)
5455 		mlx5_free(sfx_actions);
5456 	return ret;
5457 }
5458 
5459 /**
5460  * Split the flow to subflow set. The splitters might be linked
5461  * in the chain, like this:
5462  * flow_create_split_outer() calls:
5463  *   flow_create_split_meter() calls:
5464  *     flow_create_split_metadata(meter_subflow_0) calls:
5465  *       flow_create_split_inner(metadata_subflow_0)
5466  *       flow_create_split_inner(metadata_subflow_1)
5467  *       flow_create_split_inner(metadata_subflow_2)
5468  *     flow_create_split_metadata(meter_subflow_1) calls:
5469  *       flow_create_split_inner(metadata_subflow_0)
5470  *       flow_create_split_inner(metadata_subflow_1)
5471  *       flow_create_split_inner(metadata_subflow_2)
5472  *
5473  * This provide flexible way to add new levels of flow splitting.
5474  * The all of successfully created subflows are included to the
5475  * parent flow dev_flow list.
5476  *
5477  * @param dev
5478  *   Pointer to Ethernet device.
5479  * @param[in] flow
5480  *   Parent flow structure pointer.
5481  * @param[in] attr
5482  *   Flow rule attributes.
5483  * @param[in] items
5484  *   Pattern specification (list terminated by the END pattern item).
5485  * @param[in] actions
5486  *   Associated actions (list terminated by the END action).
5487  * @param[in] external
5488  *   This flow rule is created by request external to PMD.
5489  * @param[in] flow_idx
5490  *   This memory pool index to the flow.
5491  * @param[out] error
5492  *   Perform verbose error reporting if not NULL.
5493  * @return
5494  *   0 on success, negative value otherwise
5495  */
5496 static int
5497 flow_create_split_outer(struct rte_eth_dev *dev,
5498 			struct rte_flow *flow,
5499 			const struct rte_flow_attr *attr,
5500 			const struct rte_flow_item items[],
5501 			const struct rte_flow_action actions[],
5502 			bool external, uint32_t flow_idx,
5503 			struct rte_flow_error *error)
5504 {
5505 	int ret;
5506 
5507 	ret = flow_create_split_sample(dev, flow, attr, items,
5508 				       actions, external, flow_idx, error);
5509 	MLX5_ASSERT(ret <= 0);
5510 	return ret;
5511 }
5512 
5513 static struct mlx5_flow_tunnel *
5514 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5515 		      const struct rte_flow_attr *attr,
5516 		      const struct rte_flow_item items[],
5517 		      const struct rte_flow_action actions[])
5518 {
5519 	struct mlx5_flow_tunnel *tunnel;
5520 
5521 #pragma GCC diagnostic push
5522 #pragma GCC diagnostic ignored "-Wcast-qual"
5523 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5524 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5525 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5526 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5527 	else
5528 		tunnel = NULL;
5529 #pragma GCC diagnostic pop
5530 
5531 	return tunnel;
5532 }
5533 
5534 /**
5535  * Create a flow and add it to @p list.
5536  *
5537  * @param dev
5538  *   Pointer to Ethernet device.
5539  * @param list
5540  *   Pointer to a TAILQ flow list. If this parameter NULL,
5541  *   no list insertion occurred, flow is just created,
5542  *   this is caller's responsibility to track the
5543  *   created flow.
5544  * @param[in] attr
5545  *   Flow rule attributes.
5546  * @param[in] items
5547  *   Pattern specification (list terminated by the END pattern item).
5548  * @param[in] actions
5549  *   Associated actions (list terminated by the END action).
5550  * @param[in] external
5551  *   This flow rule is created by request external to PMD.
5552  * @param[out] error
5553  *   Perform verbose error reporting if not NULL.
5554  *
5555  * @return
5556  *   A flow index on success, 0 otherwise and rte_errno is set.
5557  */
5558 static uint32_t
5559 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5560 		 const struct rte_flow_attr *attr,
5561 		 const struct rte_flow_item items[],
5562 		 const struct rte_flow_action original_actions[],
5563 		 bool external, struct rte_flow_error *error)
5564 {
5565 	struct mlx5_priv *priv = dev->data->dev_private;
5566 	struct rte_flow *flow = NULL;
5567 	struct mlx5_flow *dev_flow;
5568 	const struct rte_flow_action_rss *rss;
5569 	struct mlx5_translated_shared_action
5570 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5571 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5572 	union {
5573 		struct mlx5_flow_expand_rss buf;
5574 		uint8_t buffer[2048];
5575 	} expand_buffer;
5576 	union {
5577 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5578 		uint8_t buffer[2048];
5579 	} actions_rx;
5580 	union {
5581 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5582 		uint8_t buffer[2048];
5583 	} actions_hairpin_tx;
5584 	union {
5585 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5586 		uint8_t buffer[2048];
5587 	} items_tx;
5588 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5589 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
5590 					      priv->rss_desc)[!!priv->flow_idx];
5591 	const struct rte_flow_action *p_actions_rx;
5592 	uint32_t i;
5593 	uint32_t idx = 0;
5594 	int hairpin_flow;
5595 	uint32_t hairpin_id = 0;
5596 	struct rte_flow_attr attr_tx = { .priority = 0 };
5597 	struct rte_flow_attr attr_factor = {0};
5598 	const struct rte_flow_action *actions;
5599 	struct rte_flow_action *translated_actions = NULL;
5600 	struct mlx5_flow_tunnel *tunnel;
5601 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5602 	int ret = flow_shared_actions_translate(original_actions,
5603 						shared_actions,
5604 						&shared_actions_n,
5605 						&translated_actions, error);
5606 
5607 	if (ret < 0) {
5608 		MLX5_ASSERT(translated_actions == NULL);
5609 		return 0;
5610 	}
5611 	actions = translated_actions ? translated_actions : original_actions;
5612 	memcpy((void *)&attr_factor, (const void *)attr, sizeof(*attr));
5613 	p_actions_rx = actions;
5614 	hairpin_flow = flow_check_hairpin_split(dev, &attr_factor, actions);
5615 	ret = flow_drv_validate(dev, &attr_factor, items, p_actions_rx,
5616 				external, hairpin_flow, error);
5617 	if (ret < 0)
5618 		goto error_before_hairpin_split;
5619 	if (hairpin_flow > 0) {
5620 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5621 			rte_errno = EINVAL;
5622 			goto error_before_hairpin_split;
5623 		}
5624 		flow_hairpin_split(dev, actions, actions_rx.actions,
5625 				   actions_hairpin_tx.actions, items_tx.items,
5626 				   &hairpin_id);
5627 		p_actions_rx = actions_rx.actions;
5628 	}
5629 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5630 	if (!flow) {
5631 		rte_errno = ENOMEM;
5632 		goto error_before_flow;
5633 	}
5634 	flow->drv_type = flow_get_drv_type(dev, &attr_factor);
5635 	if (hairpin_id != 0)
5636 		flow->hairpin_flow_id = hairpin_id;
5637 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5638 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5639 	memset(rss_desc, 0, sizeof(*rss_desc));
5640 	rss = flow_get_rss_action(p_actions_rx);
5641 	if (rss) {
5642 		/*
5643 		 * The following information is required by
5644 		 * mlx5_flow_hashfields_adjust() in advance.
5645 		 */
5646 		rss_desc->level = rss->level;
5647 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5648 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5649 	}
5650 	flow->dev_handles = 0;
5651 	if (rss && rss->types) {
5652 		unsigned int graph_root;
5653 
5654 		graph_root = find_graph_root(items, rss->level);
5655 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5656 					   items, rss->types,
5657 					   mlx5_support_expansion, graph_root);
5658 		MLX5_ASSERT(ret > 0 &&
5659 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5660 	} else {
5661 		buf->entries = 1;
5662 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5663 	}
5664 	flow->shared_rss = flow_get_shared_rss_action(shared_actions,
5665 						      shared_actions_n);
5666 	/*
5667 	 * Record the start index when there is a nested call. All sub-flows
5668 	 * need to be translated before another calling.
5669 	 * No need to use ping-pong buffer to save memory here.
5670 	 */
5671 	if (priv->flow_idx) {
5672 		MLX5_ASSERT(!priv->flow_nested_idx);
5673 		priv->flow_nested_idx = priv->flow_idx;
5674 	}
5675 	for (i = 0; i < buf->entries; ++i) {
5676 		/*
5677 		 * The splitter may create multiple dev_flows,
5678 		 * depending on configuration. In the simplest
5679 		 * case it just creates unmodified original flow.
5680 		 */
5681 		ret = flow_create_split_outer(dev, flow, &attr_factor,
5682 					      buf->entry[i].pattern,
5683 					      p_actions_rx, external, idx,
5684 					      error);
5685 		if (ret < 0)
5686 			goto error;
5687 		if (is_flow_tunnel_steer_rule(dev, attr,
5688 					      buf->entry[i].pattern,
5689 					      p_actions_rx)) {
5690 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5691 							   p_actions_rx,
5692 							   idx,
5693 							   &default_miss_ctx,
5694 							   error);
5695 			if (ret < 0) {
5696 				mlx5_free(default_miss_ctx.queue);
5697 				goto error;
5698 			}
5699 		}
5700 	}
5701 	/* Create the tx flow. */
5702 	if (hairpin_flow) {
5703 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5704 		attr_tx.ingress = 0;
5705 		attr_tx.egress = 1;
5706 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5707 					 actions_hairpin_tx.actions,
5708 					 idx, error);
5709 		if (!dev_flow)
5710 			goto error;
5711 		dev_flow->flow = flow;
5712 		dev_flow->external = 0;
5713 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5714 			      dev_flow->handle, next);
5715 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5716 					 items_tx.items,
5717 					 actions_hairpin_tx.actions, error);
5718 		if (ret < 0)
5719 			goto error;
5720 	}
5721 	/*
5722 	 * Update the metadata register copy table. If extensive
5723 	 * metadata feature is enabled and registers are supported
5724 	 * we might create the extra rte_flow for each unique
5725 	 * MARK/FLAG action ID.
5726 	 *
5727 	 * The table is updated for ingress Flows only, because
5728 	 * the egress Flows belong to the different device and
5729 	 * copy table should be updated in peer NIC Rx domain.
5730 	 */
5731 	if (attr_factor.ingress &&
5732 	    (external || attr_factor.group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5733 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5734 		if (ret)
5735 			goto error;
5736 	}
5737 	/*
5738 	 * If the flow is external (from application) OR device is started, then
5739 	 * the flow will be applied immediately.
5740 	 */
5741 	if (external || dev->data->dev_started) {
5742 		ret = flow_drv_apply(dev, flow, error);
5743 		if (ret < 0)
5744 			goto error;
5745 	}
5746 	if (list)
5747 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5748 			     flow, next);
5749 	flow_rxq_flags_set(dev, flow);
5750 	rte_free(translated_actions);
5751 	/* Nested flow creation index recovery. */
5752 	priv->flow_idx = priv->flow_nested_idx;
5753 	if (priv->flow_nested_idx)
5754 		priv->flow_nested_idx = 0;
5755 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5756 	if (tunnel) {
5757 		flow->tunnel = 1;
5758 		flow->tunnel_id = tunnel->tunnel_id;
5759 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5760 		mlx5_free(default_miss_ctx.queue);
5761 	}
5762 	return idx;
5763 error:
5764 	MLX5_ASSERT(flow);
5765 	ret = rte_errno; /* Save rte_errno before cleanup. */
5766 	flow_mreg_del_copy_action(dev, flow);
5767 	flow_drv_destroy(dev, flow);
5768 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5769 	rte_errno = ret; /* Restore rte_errno. */
5770 error_before_flow:
5771 	ret = rte_errno;
5772 	if (hairpin_id)
5773 		mlx5_flow_id_release(priv->sh->flow_id_pool,
5774 				     hairpin_id);
5775 	rte_errno = ret;
5776 	priv->flow_idx = priv->flow_nested_idx;
5777 	if (priv->flow_nested_idx)
5778 		priv->flow_nested_idx = 0;
5779 error_before_hairpin_split:
5780 	rte_free(translated_actions);
5781 	return 0;
5782 }
5783 
5784 /**
5785  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5786  * incoming packets to table 1.
5787  *
5788  * Other flow rules, requested for group n, will be created in
5789  * e-switch table n+1.
5790  * Jump action to e-switch group n will be created to group n+1.
5791  *
5792  * Used when working in switchdev mode, to utilise advantages of table 1
5793  * and above.
5794  *
5795  * @param dev
5796  *   Pointer to Ethernet device.
5797  *
5798  * @return
5799  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5800  */
5801 struct rte_flow *
5802 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5803 {
5804 	const struct rte_flow_attr attr = {
5805 		.group = 0,
5806 		.priority = 0,
5807 		.ingress = 1,
5808 		.egress = 0,
5809 		.transfer = 1,
5810 	};
5811 	const struct rte_flow_item pattern = {
5812 		.type = RTE_FLOW_ITEM_TYPE_END,
5813 	};
5814 	struct rte_flow_action_jump jump = {
5815 		.group = 1,
5816 	};
5817 	const struct rte_flow_action actions[] = {
5818 		{
5819 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5820 			.conf = &jump,
5821 		},
5822 		{
5823 			.type = RTE_FLOW_ACTION_TYPE_END,
5824 		},
5825 	};
5826 	struct mlx5_priv *priv = dev->data->dev_private;
5827 	struct rte_flow_error error;
5828 
5829 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5830 						   &attr, &pattern,
5831 						   actions, false, &error);
5832 }
5833 
5834 /**
5835  * Validate a flow supported by the NIC.
5836  *
5837  * @see rte_flow_validate()
5838  * @see rte_flow_ops
5839  */
5840 int
5841 mlx5_flow_validate(struct rte_eth_dev *dev,
5842 		   const struct rte_flow_attr *attr,
5843 		   const struct rte_flow_item items[],
5844 		   const struct rte_flow_action original_actions[],
5845 		   struct rte_flow_error *error)
5846 {
5847 	int hairpin_flow;
5848 	struct mlx5_translated_shared_action
5849 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5850 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5851 	const struct rte_flow_action *actions;
5852 	struct rte_flow_action *translated_actions = NULL;
5853 	int ret = flow_shared_actions_translate(original_actions,
5854 						shared_actions,
5855 						&shared_actions_n,
5856 						&translated_actions, error);
5857 
5858 	if (ret)
5859 		return ret;
5860 	actions = translated_actions ? translated_actions : original_actions;
5861 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5862 	ret = flow_drv_validate(dev, attr, items, actions,
5863 				true, hairpin_flow, error);
5864 	rte_free(translated_actions);
5865 	return ret;
5866 }
5867 
5868 /**
5869  * Create a flow.
5870  *
5871  * @see rte_flow_create()
5872  * @see rte_flow_ops
5873  */
5874 struct rte_flow *
5875 mlx5_flow_create(struct rte_eth_dev *dev,
5876 		 const struct rte_flow_attr *attr,
5877 		 const struct rte_flow_item items[],
5878 		 const struct rte_flow_action actions[],
5879 		 struct rte_flow_error *error)
5880 {
5881 	struct mlx5_priv *priv = dev->data->dev_private;
5882 
5883 	/*
5884 	 * If the device is not started yet, it is not allowed to created a
5885 	 * flow from application. PMD default flows and traffic control flows
5886 	 * are not affected.
5887 	 */
5888 	if (unlikely(!dev->data->dev_started)) {
5889 		DRV_LOG(DEBUG, "port %u is not started when "
5890 			"inserting a flow", dev->data->port_id);
5891 		rte_flow_error_set(error, ENODEV,
5892 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5893 				   NULL,
5894 				   "port not started");
5895 		return NULL;
5896 	}
5897 
5898 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5899 				  attr, items, actions, true, error);
5900 }
5901 
5902 /**
5903  * Destroy a flow in a list.
5904  *
5905  * @param dev
5906  *   Pointer to Ethernet device.
5907  * @param list
5908  *   Pointer to the Indexed flow list. If this parameter NULL,
5909  *   there is no flow removal from the list. Be noted that as
5910  *   flow is add to the indexed list, memory of the indexed
5911  *   list points to maybe changed as flow destroyed.
5912  * @param[in] flow_idx
5913  *   Index of flow to destroy.
5914  */
5915 static void
5916 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5917 		  uint32_t flow_idx)
5918 {
5919 	struct mlx5_priv *priv = dev->data->dev_private;
5920 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5921 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5922 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5923 
5924 	if (!flow)
5925 		return;
5926 	/*
5927 	 * Update RX queue flags only if port is started, otherwise it is
5928 	 * already clean.
5929 	 */
5930 	if (dev->data->dev_started)
5931 		flow_rxq_flags_trim(dev, flow);
5932 	if (flow->hairpin_flow_id)
5933 		mlx5_flow_id_release(priv->sh->flow_id_pool,
5934 				     flow->hairpin_flow_id);
5935 	flow_drv_destroy(dev, flow);
5936 	if (list)
5937 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5938 			     flow_idx, flow, next);
5939 	flow_mreg_del_copy_action(dev, flow);
5940 	if (flow->fdir) {
5941 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5942 			if (priv_fdir_flow->rix_flow == flow_idx)
5943 				break;
5944 		}
5945 		if (priv_fdir_flow) {
5946 			LIST_REMOVE(priv_fdir_flow, next);
5947 			mlx5_free(priv_fdir_flow->fdir);
5948 			mlx5_free(priv_fdir_flow);
5949 		}
5950 	}
5951 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5952 	if (flow->tunnel) {
5953 		struct mlx5_flow_tunnel *tunnel;
5954 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5955 		RTE_VERIFY(tunnel);
5956 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5957 			mlx5_flow_tunnel_free(dev, tunnel);
5958 	}
5959 }
5960 
5961 /**
5962  * Destroy all flows.
5963  *
5964  * @param dev
5965  *   Pointer to Ethernet device.
5966  * @param list
5967  *   Pointer to the Indexed flow list.
5968  * @param active
5969  *   If flushing is called avtively.
5970  */
5971 void
5972 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5973 {
5974 	uint32_t num_flushed = 0;
5975 
5976 	while (*list) {
5977 		flow_list_destroy(dev, list, *list);
5978 		num_flushed++;
5979 	}
5980 	if (active) {
5981 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5982 			dev->data->port_id, num_flushed);
5983 	}
5984 }
5985 
5986 /**
5987  * Remove all flows.
5988  *
5989  * @param dev
5990  *   Pointer to Ethernet device.
5991  * @param list
5992  *   Pointer to the Indexed flow list.
5993  */
5994 void
5995 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
5996 {
5997 	struct mlx5_priv *priv = dev->data->dev_private;
5998 	struct rte_flow *flow = NULL;
5999 	uint32_t idx;
6000 
6001 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
6002 		      flow, next) {
6003 		flow_drv_remove(dev, flow);
6004 		flow_mreg_stop_copy_action(dev, flow);
6005 	}
6006 	flow_mreg_del_default_copy_action(dev);
6007 	flow_rxq_flags_clear(dev);
6008 }
6009 
6010 /**
6011  * Add all flows.
6012  *
6013  * @param dev
6014  *   Pointer to Ethernet device.
6015  * @param list
6016  *   Pointer to the Indexed flow list.
6017  *
6018  * @return
6019  *   0 on success, a negative errno value otherwise and rte_errno is set.
6020  */
6021 int
6022 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
6023 {
6024 	struct mlx5_priv *priv = dev->data->dev_private;
6025 	struct rte_flow *flow = NULL;
6026 	struct rte_flow_error error;
6027 	uint32_t idx;
6028 	int ret = 0;
6029 
6030 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6031 	ret = flow_mreg_add_default_copy_action(dev, &error);
6032 	if (ret < 0)
6033 		return -rte_errno;
6034 	/* Apply Flows created by application. */
6035 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
6036 		      flow, next) {
6037 		ret = flow_mreg_start_copy_action(dev, flow);
6038 		if (ret < 0)
6039 			goto error;
6040 		ret = flow_drv_apply(dev, flow, &error);
6041 		if (ret < 0)
6042 			goto error;
6043 		flow_rxq_flags_set(dev, flow);
6044 	}
6045 	return 0;
6046 error:
6047 	ret = rte_errno; /* Save rte_errno before cleanup. */
6048 	mlx5_flow_stop(dev, list);
6049 	rte_errno = ret; /* Restore rte_errno. */
6050 	return -rte_errno;
6051 }
6052 
6053 /**
6054  * Stop all default actions for flows.
6055  *
6056  * @param dev
6057  *   Pointer to Ethernet device.
6058  */
6059 void
6060 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6061 {
6062 	flow_mreg_del_default_copy_action(dev);
6063 	flow_rxq_flags_clear(dev);
6064 }
6065 
6066 /**
6067  * Start all default actions for flows.
6068  *
6069  * @param dev
6070  *   Pointer to Ethernet device.
6071  * @return
6072  *   0 on success, a negative errno value otherwise and rte_errno is set.
6073  */
6074 int
6075 mlx5_flow_start_default(struct rte_eth_dev *dev)
6076 {
6077 	struct rte_flow_error error;
6078 
6079 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6080 	return flow_mreg_add_default_copy_action(dev, &error);
6081 }
6082 
6083 /**
6084  * Allocate intermediate resources for flow creation.
6085  *
6086  * @param dev
6087  *   Pointer to Ethernet device.
6088  */
6089 void
6090 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
6091 {
6092 	struct mlx5_priv *priv = dev->data->dev_private;
6093 
6094 	if (!priv->inter_flows) {
6095 		priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
6096 				    MLX5_NUM_MAX_DEV_FLOWS *
6097 				    sizeof(struct mlx5_flow) +
6098 				    (sizeof(struct mlx5_flow_rss_desc) +
6099 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0,
6100 				    SOCKET_ID_ANY);
6101 		if (!priv->inter_flows) {
6102 			DRV_LOG(ERR, "can't allocate intermediate memory.");
6103 			return;
6104 		}
6105 	}
6106 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
6107 			 [MLX5_NUM_MAX_DEV_FLOWS];
6108 	/* Reset the index. */
6109 	priv->flow_idx = 0;
6110 	priv->flow_nested_idx = 0;
6111 }
6112 
6113 /**
6114  * Free intermediate resources for flows.
6115  *
6116  * @param dev
6117  *   Pointer to Ethernet device.
6118  */
6119 void
6120 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
6121 {
6122 	struct mlx5_priv *priv = dev->data->dev_private;
6123 
6124 	mlx5_free(priv->inter_flows);
6125 	priv->inter_flows = NULL;
6126 }
6127 
6128 /**
6129  * Verify the flow list is empty
6130  *
6131  * @param dev
6132  *  Pointer to Ethernet device.
6133  *
6134  * @return the number of flows not released.
6135  */
6136 int
6137 mlx5_flow_verify(struct rte_eth_dev *dev)
6138 {
6139 	struct mlx5_priv *priv = dev->data->dev_private;
6140 	struct rte_flow *flow;
6141 	uint32_t idx;
6142 	int ret = 0;
6143 
6144 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6145 		      flow, next) {
6146 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6147 			dev->data->port_id, (void *)flow);
6148 		++ret;
6149 	}
6150 	return ret;
6151 }
6152 
6153 /**
6154  * Enable default hairpin egress flow.
6155  *
6156  * @param dev
6157  *   Pointer to Ethernet device.
6158  * @param queue
6159  *   The queue index.
6160  *
6161  * @return
6162  *   0 on success, a negative errno value otherwise and rte_errno is set.
6163  */
6164 int
6165 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6166 			    uint32_t queue)
6167 {
6168 	struct mlx5_priv *priv = dev->data->dev_private;
6169 	const struct rte_flow_attr attr = {
6170 		.egress = 1,
6171 		.priority = 0,
6172 	};
6173 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6174 		.queue = queue,
6175 	};
6176 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6177 		.queue = UINT32_MAX,
6178 	};
6179 	struct rte_flow_item items[] = {
6180 		{
6181 			.type = (enum rte_flow_item_type)
6182 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6183 			.spec = &queue_spec,
6184 			.last = NULL,
6185 			.mask = &queue_mask,
6186 		},
6187 		{
6188 			.type = RTE_FLOW_ITEM_TYPE_END,
6189 		},
6190 	};
6191 	struct rte_flow_action_jump jump = {
6192 		.group = MLX5_HAIRPIN_TX_TABLE,
6193 	};
6194 	struct rte_flow_action actions[2];
6195 	uint32_t flow_idx;
6196 	struct rte_flow_error error;
6197 
6198 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6199 	actions[0].conf = &jump;
6200 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6201 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6202 				&attr, items, actions, false, &error);
6203 	if (!flow_idx) {
6204 		DRV_LOG(DEBUG,
6205 			"Failed to create ctrl flow: rte_errno(%d),"
6206 			" type(%d), message(%s)",
6207 			rte_errno, error.type,
6208 			error.message ? error.message : " (no stated reason)");
6209 		return -rte_errno;
6210 	}
6211 	return 0;
6212 }
6213 
6214 /**
6215  * Enable a control flow configured from the control plane.
6216  *
6217  * @param dev
6218  *   Pointer to Ethernet device.
6219  * @param eth_spec
6220  *   An Ethernet flow spec to apply.
6221  * @param eth_mask
6222  *   An Ethernet flow mask to apply.
6223  * @param vlan_spec
6224  *   A VLAN flow spec to apply.
6225  * @param vlan_mask
6226  *   A VLAN flow mask to apply.
6227  *
6228  * @return
6229  *   0 on success, a negative errno value otherwise and rte_errno is set.
6230  */
6231 int
6232 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6233 		    struct rte_flow_item_eth *eth_spec,
6234 		    struct rte_flow_item_eth *eth_mask,
6235 		    struct rte_flow_item_vlan *vlan_spec,
6236 		    struct rte_flow_item_vlan *vlan_mask)
6237 {
6238 	struct mlx5_priv *priv = dev->data->dev_private;
6239 	const struct rte_flow_attr attr = {
6240 		.ingress = 1,
6241 		.priority = MLX5_FLOW_PRIO_RSVD,
6242 	};
6243 	struct rte_flow_item items[] = {
6244 		{
6245 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6246 			.spec = eth_spec,
6247 			.last = NULL,
6248 			.mask = eth_mask,
6249 		},
6250 		{
6251 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6252 					      RTE_FLOW_ITEM_TYPE_END,
6253 			.spec = vlan_spec,
6254 			.last = NULL,
6255 			.mask = vlan_mask,
6256 		},
6257 		{
6258 			.type = RTE_FLOW_ITEM_TYPE_END,
6259 		},
6260 	};
6261 	uint16_t queue[priv->reta_idx_n];
6262 	struct rte_flow_action_rss action_rss = {
6263 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6264 		.level = 0,
6265 		.types = priv->rss_conf.rss_hf,
6266 		.key_len = priv->rss_conf.rss_key_len,
6267 		.queue_num = priv->reta_idx_n,
6268 		.key = priv->rss_conf.rss_key,
6269 		.queue = queue,
6270 	};
6271 	struct rte_flow_action actions[] = {
6272 		{
6273 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6274 			.conf = &action_rss,
6275 		},
6276 		{
6277 			.type = RTE_FLOW_ACTION_TYPE_END,
6278 		},
6279 	};
6280 	uint32_t flow_idx;
6281 	struct rte_flow_error error;
6282 	unsigned int i;
6283 
6284 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6285 		return 0;
6286 	}
6287 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6288 		action_rss.types = 0;
6289 	for (i = 0; i != priv->reta_idx_n; ++i)
6290 		queue[i] = (*priv->reta_idx)[i];
6291 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6292 				&attr, items, actions, false, &error);
6293 	if (!flow_idx)
6294 		return -rte_errno;
6295 	return 0;
6296 }
6297 
6298 /**
6299  * Enable a flow control configured from the control plane.
6300  *
6301  * @param dev
6302  *   Pointer to Ethernet device.
6303  * @param eth_spec
6304  *   An Ethernet flow spec to apply.
6305  * @param eth_mask
6306  *   An Ethernet flow mask to apply.
6307  *
6308  * @return
6309  *   0 on success, a negative errno value otherwise and rte_errno is set.
6310  */
6311 int
6312 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6313 	       struct rte_flow_item_eth *eth_spec,
6314 	       struct rte_flow_item_eth *eth_mask)
6315 {
6316 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6317 }
6318 
6319 /**
6320  * Create default miss flow rule matching lacp traffic
6321  *
6322  * @param dev
6323  *   Pointer to Ethernet device.
6324  * @param eth_spec
6325  *   An Ethernet flow spec to apply.
6326  *
6327  * @return
6328  *   0 on success, a negative errno value otherwise and rte_errno is set.
6329  */
6330 int
6331 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6332 {
6333 	struct mlx5_priv *priv = dev->data->dev_private;
6334 	/*
6335 	 * The LACP matching is done by only using ether type since using
6336 	 * a multicast dst mac causes kernel to give low priority to this flow.
6337 	 */
6338 	static const struct rte_flow_item_eth lacp_spec = {
6339 		.type = RTE_BE16(0x8809),
6340 	};
6341 	static const struct rte_flow_item_eth lacp_mask = {
6342 		.type = 0xffff,
6343 	};
6344 	const struct rte_flow_attr attr = {
6345 		.ingress = 1,
6346 	};
6347 	struct rte_flow_item items[] = {
6348 		{
6349 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6350 			.spec = &lacp_spec,
6351 			.mask = &lacp_mask,
6352 		},
6353 		{
6354 			.type = RTE_FLOW_ITEM_TYPE_END,
6355 		},
6356 	};
6357 	struct rte_flow_action actions[] = {
6358 		{
6359 			.type = (enum rte_flow_action_type)
6360 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6361 		},
6362 		{
6363 			.type = RTE_FLOW_ACTION_TYPE_END,
6364 		},
6365 	};
6366 	struct rte_flow_error error;
6367 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6368 				&attr, items, actions, false, &error);
6369 
6370 	if (!flow_idx)
6371 		return -rte_errno;
6372 	return 0;
6373 }
6374 
6375 /**
6376  * Destroy a flow.
6377  *
6378  * @see rte_flow_destroy()
6379  * @see rte_flow_ops
6380  */
6381 int
6382 mlx5_flow_destroy(struct rte_eth_dev *dev,
6383 		  struct rte_flow *flow,
6384 		  struct rte_flow_error *error __rte_unused)
6385 {
6386 	struct mlx5_priv *priv = dev->data->dev_private;
6387 
6388 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6389 	return 0;
6390 }
6391 
6392 /**
6393  * Destroy all flows.
6394  *
6395  * @see rte_flow_flush()
6396  * @see rte_flow_ops
6397  */
6398 int
6399 mlx5_flow_flush(struct rte_eth_dev *dev,
6400 		struct rte_flow_error *error __rte_unused)
6401 {
6402 	struct mlx5_priv *priv = dev->data->dev_private;
6403 
6404 	mlx5_flow_list_flush(dev, &priv->flows, false);
6405 	return 0;
6406 }
6407 
6408 /**
6409  * Isolated mode.
6410  *
6411  * @see rte_flow_isolate()
6412  * @see rte_flow_ops
6413  */
6414 int
6415 mlx5_flow_isolate(struct rte_eth_dev *dev,
6416 		  int enable,
6417 		  struct rte_flow_error *error)
6418 {
6419 	struct mlx5_priv *priv = dev->data->dev_private;
6420 
6421 	if (dev->data->dev_started) {
6422 		rte_flow_error_set(error, EBUSY,
6423 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6424 				   NULL,
6425 				   "port must be stopped first");
6426 		return -rte_errno;
6427 	}
6428 	priv->isolated = !!enable;
6429 	if (enable)
6430 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6431 	else
6432 		dev->dev_ops = &mlx5_os_dev_ops;
6433 
6434 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6435 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6436 
6437 	return 0;
6438 }
6439 
6440 /**
6441  * Query a flow.
6442  *
6443  * @see rte_flow_query()
6444  * @see rte_flow_ops
6445  */
6446 static int
6447 flow_drv_query(struct rte_eth_dev *dev,
6448 	       uint32_t flow_idx,
6449 	       const struct rte_flow_action *actions,
6450 	       void *data,
6451 	       struct rte_flow_error *error)
6452 {
6453 	struct mlx5_priv *priv = dev->data->dev_private;
6454 	const struct mlx5_flow_driver_ops *fops;
6455 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6456 					       [MLX5_IPOOL_RTE_FLOW],
6457 					       flow_idx);
6458 	enum mlx5_flow_drv_type ftype;
6459 
6460 	if (!flow) {
6461 		return rte_flow_error_set(error, ENOENT,
6462 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6463 			  NULL,
6464 			  "invalid flow handle");
6465 	}
6466 	ftype = flow->drv_type;
6467 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6468 	fops = flow_get_drv_ops(ftype);
6469 
6470 	return fops->query(dev, flow, actions, data, error);
6471 }
6472 
6473 /**
6474  * Query a flow.
6475  *
6476  * @see rte_flow_query()
6477  * @see rte_flow_ops
6478  */
6479 int
6480 mlx5_flow_query(struct rte_eth_dev *dev,
6481 		struct rte_flow *flow,
6482 		const struct rte_flow_action *actions,
6483 		void *data,
6484 		struct rte_flow_error *error)
6485 {
6486 	int ret;
6487 
6488 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6489 			     error);
6490 	if (ret < 0)
6491 		return ret;
6492 	return 0;
6493 }
6494 
6495 /**
6496  * Convert a flow director filter to a generic flow.
6497  *
6498  * @param dev
6499  *   Pointer to Ethernet device.
6500  * @param fdir_filter
6501  *   Flow director filter to add.
6502  * @param attributes
6503  *   Generic flow parameters structure.
6504  *
6505  * @return
6506  *   0 on success, a negative errno value otherwise and rte_errno is set.
6507  */
6508 static int
6509 flow_fdir_filter_convert(struct rte_eth_dev *dev,
6510 			 const struct rte_eth_fdir_filter *fdir_filter,
6511 			 struct mlx5_fdir *attributes)
6512 {
6513 	struct mlx5_priv *priv = dev->data->dev_private;
6514 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
6515 	const struct rte_eth_fdir_masks *mask =
6516 		&dev->data->dev_conf.fdir_conf.mask;
6517 
6518 	/* Validate queue number. */
6519 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
6520 		DRV_LOG(ERR, "port %u invalid queue number %d",
6521 			dev->data->port_id, fdir_filter->action.rx_queue);
6522 		rte_errno = EINVAL;
6523 		return -rte_errno;
6524 	}
6525 	attributes->attr.ingress = 1;
6526 	attributes->items[0] = (struct rte_flow_item) {
6527 		.type = RTE_FLOW_ITEM_TYPE_ETH,
6528 		.spec = &attributes->l2,
6529 		.mask = &attributes->l2_mask,
6530 	};
6531 	switch (fdir_filter->action.behavior) {
6532 	case RTE_ETH_FDIR_ACCEPT:
6533 		attributes->actions[0] = (struct rte_flow_action){
6534 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
6535 			.conf = &attributes->queue,
6536 		};
6537 		break;
6538 	case RTE_ETH_FDIR_REJECT:
6539 		attributes->actions[0] = (struct rte_flow_action){
6540 			.type = RTE_FLOW_ACTION_TYPE_DROP,
6541 		};
6542 		break;
6543 	default:
6544 		DRV_LOG(ERR, "port %u invalid behavior %d",
6545 			dev->data->port_id,
6546 			fdir_filter->action.behavior);
6547 		rte_errno = ENOTSUP;
6548 		return -rte_errno;
6549 	}
6550 	attributes->queue.index = fdir_filter->action.rx_queue;
6551 	/* Handle L3. */
6552 	switch (fdir_filter->input.flow_type) {
6553 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6554 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6555 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6556 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
6557 			.src_addr = input->flow.ip4_flow.src_ip,
6558 			.dst_addr = input->flow.ip4_flow.dst_ip,
6559 			.time_to_live = input->flow.ip4_flow.ttl,
6560 			.type_of_service = input->flow.ip4_flow.tos,
6561 		};
6562 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
6563 			.src_addr = mask->ipv4_mask.src_ip,
6564 			.dst_addr = mask->ipv4_mask.dst_ip,
6565 			.time_to_live = mask->ipv4_mask.ttl,
6566 			.type_of_service = mask->ipv4_mask.tos,
6567 			.next_proto_id = mask->ipv4_mask.proto,
6568 		};
6569 		attributes->items[1] = (struct rte_flow_item){
6570 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
6571 			.spec = &attributes->l3,
6572 			.mask = &attributes->l3_mask,
6573 		};
6574 		break;
6575 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6576 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6577 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6578 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
6579 			.hop_limits = input->flow.ipv6_flow.hop_limits,
6580 			.proto = input->flow.ipv6_flow.proto,
6581 		};
6582 
6583 		memcpy(attributes->l3.ipv6.hdr.src_addr,
6584 		       input->flow.ipv6_flow.src_ip,
6585 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6586 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
6587 		       input->flow.ipv6_flow.dst_ip,
6588 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
6589 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
6590 		       mask->ipv6_mask.src_ip,
6591 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6592 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
6593 		       mask->ipv6_mask.dst_ip,
6594 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
6595 		attributes->items[1] = (struct rte_flow_item){
6596 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
6597 			.spec = &attributes->l3,
6598 			.mask = &attributes->l3_mask,
6599 		};
6600 		break;
6601 	default:
6602 		DRV_LOG(ERR, "port %u invalid flow type%d",
6603 			dev->data->port_id, fdir_filter->input.flow_type);
6604 		rte_errno = ENOTSUP;
6605 		return -rte_errno;
6606 	}
6607 	/* Handle L4. */
6608 	switch (fdir_filter->input.flow_type) {
6609 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6610 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6611 			.src_port = input->flow.udp4_flow.src_port,
6612 			.dst_port = input->flow.udp4_flow.dst_port,
6613 		};
6614 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6615 			.src_port = mask->src_port_mask,
6616 			.dst_port = mask->dst_port_mask,
6617 		};
6618 		attributes->items[2] = (struct rte_flow_item){
6619 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6620 			.spec = &attributes->l4,
6621 			.mask = &attributes->l4_mask,
6622 		};
6623 		break;
6624 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6625 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6626 			.src_port = input->flow.tcp4_flow.src_port,
6627 			.dst_port = input->flow.tcp4_flow.dst_port,
6628 		};
6629 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6630 			.src_port = mask->src_port_mask,
6631 			.dst_port = mask->dst_port_mask,
6632 		};
6633 		attributes->items[2] = (struct rte_flow_item){
6634 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6635 			.spec = &attributes->l4,
6636 			.mask = &attributes->l4_mask,
6637 		};
6638 		break;
6639 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6640 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6641 			.src_port = input->flow.udp6_flow.src_port,
6642 			.dst_port = input->flow.udp6_flow.dst_port,
6643 		};
6644 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6645 			.src_port = mask->src_port_mask,
6646 			.dst_port = mask->dst_port_mask,
6647 		};
6648 		attributes->items[2] = (struct rte_flow_item){
6649 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6650 			.spec = &attributes->l4,
6651 			.mask = &attributes->l4_mask,
6652 		};
6653 		break;
6654 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6655 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6656 			.src_port = input->flow.tcp6_flow.src_port,
6657 			.dst_port = input->flow.tcp6_flow.dst_port,
6658 		};
6659 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6660 			.src_port = mask->src_port_mask,
6661 			.dst_port = mask->dst_port_mask,
6662 		};
6663 		attributes->items[2] = (struct rte_flow_item){
6664 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6665 			.spec = &attributes->l4,
6666 			.mask = &attributes->l4_mask,
6667 		};
6668 		break;
6669 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6670 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6671 		break;
6672 	default:
6673 		DRV_LOG(ERR, "port %u invalid flow type%d",
6674 			dev->data->port_id, fdir_filter->input.flow_type);
6675 		rte_errno = ENOTSUP;
6676 		return -rte_errno;
6677 	}
6678 	return 0;
6679 }
6680 
6681 #define FLOW_FDIR_CMP(f1, f2, fld) \
6682 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
6683 
6684 /**
6685  * Compare two FDIR flows. If items and actions are identical, the two flows are
6686  * regarded as same.
6687  *
6688  * @param dev
6689  *   Pointer to Ethernet device.
6690  * @param f1
6691  *   FDIR flow to compare.
6692  * @param f2
6693  *   FDIR flow to compare.
6694  *
6695  * @return
6696  *   Zero on match, 1 otherwise.
6697  */
6698 static int
6699 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
6700 {
6701 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
6702 	    FLOW_FDIR_CMP(f1, f2, l2) ||
6703 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
6704 	    FLOW_FDIR_CMP(f1, f2, l3) ||
6705 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
6706 	    FLOW_FDIR_CMP(f1, f2, l4) ||
6707 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
6708 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
6709 		return 1;
6710 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
6711 	    FLOW_FDIR_CMP(f1, f2, queue))
6712 		return 1;
6713 	return 0;
6714 }
6715 
6716 /**
6717  * Search device flow list to find out a matched FDIR flow.
6718  *
6719  * @param dev
6720  *   Pointer to Ethernet device.
6721  * @param fdir_flow
6722  *   FDIR flow to lookup.
6723  *
6724  * @return
6725  *   Index of flow if found, 0 otherwise.
6726  */
6727 static uint32_t
6728 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
6729 {
6730 	struct mlx5_priv *priv = dev->data->dev_private;
6731 	uint32_t flow_idx = 0;
6732 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6733 
6734 	MLX5_ASSERT(fdir_flow);
6735 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6736 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
6737 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
6738 				dev->data->port_id, flow_idx);
6739 			flow_idx = priv_fdir_flow->rix_flow;
6740 			break;
6741 		}
6742 	}
6743 	return flow_idx;
6744 }
6745 
6746 /**
6747  * Add new flow director filter and store it in list.
6748  *
6749  * @param dev
6750  *   Pointer to Ethernet device.
6751  * @param fdir_filter
6752  *   Flow director filter to add.
6753  *
6754  * @return
6755  *   0 on success, a negative errno value otherwise and rte_errno is set.
6756  */
6757 static int
6758 flow_fdir_filter_add(struct rte_eth_dev *dev,
6759 		     const struct rte_eth_fdir_filter *fdir_filter)
6760 {
6761 	struct mlx5_priv *priv = dev->data->dev_private;
6762 	struct mlx5_fdir *fdir_flow;
6763 	struct rte_flow *flow;
6764 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6765 	uint32_t flow_idx;
6766 	int ret;
6767 
6768 	fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
6769 				SOCKET_ID_ANY);
6770 	if (!fdir_flow) {
6771 		rte_errno = ENOMEM;
6772 		return -rte_errno;
6773 	}
6774 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
6775 	if (ret)
6776 		goto error;
6777 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
6778 	if (flow_idx) {
6779 		rte_errno = EEXIST;
6780 		goto error;
6781 	}
6782 	priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
6783 				     sizeof(struct mlx5_fdir_flow),
6784 				     0, SOCKET_ID_ANY);
6785 	if (!priv_fdir_flow) {
6786 		rte_errno = ENOMEM;
6787 		goto error;
6788 	}
6789 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
6790 				    fdir_flow->items, fdir_flow->actions, true,
6791 				    NULL);
6792 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6793 	if (!flow)
6794 		goto error;
6795 	flow->fdir = 1;
6796 	priv_fdir_flow->fdir = fdir_flow;
6797 	priv_fdir_flow->rix_flow = flow_idx;
6798 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
6799 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
6800 		dev->data->port_id, (void *)flow);
6801 	return 0;
6802 error:
6803 	mlx5_free(priv_fdir_flow);
6804 	mlx5_free(fdir_flow);
6805 	return -rte_errno;
6806 }
6807 
6808 /**
6809  * Delete specific filter.
6810  *
6811  * @param dev
6812  *   Pointer to Ethernet device.
6813  * @param fdir_filter
6814  *   Filter to be deleted.
6815  *
6816  * @return
6817  *   0 on success, a negative errno value otherwise and rte_errno is set.
6818  */
6819 static int
6820 flow_fdir_filter_delete(struct rte_eth_dev *dev,
6821 			const struct rte_eth_fdir_filter *fdir_filter)
6822 {
6823 	struct mlx5_priv *priv = dev->data->dev_private;
6824 	uint32_t flow_idx;
6825 	struct mlx5_fdir fdir_flow = {
6826 		.attr.group = 0,
6827 	};
6828 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6829 	int ret;
6830 
6831 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
6832 	if (ret)
6833 		return -rte_errno;
6834 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6835 		/* Find the fdir in priv list */
6836 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
6837 			break;
6838 	}
6839 	if (!priv_fdir_flow)
6840 		return 0;
6841 	LIST_REMOVE(priv_fdir_flow, next);
6842 	flow_idx = priv_fdir_flow->rix_flow;
6843 	flow_list_destroy(dev, &priv->flows, flow_idx);
6844 	mlx5_free(priv_fdir_flow->fdir);
6845 	mlx5_free(priv_fdir_flow);
6846 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
6847 		dev->data->port_id, flow_idx);
6848 	return 0;
6849 }
6850 
6851 /**
6852  * Update queue for specific filter.
6853  *
6854  * @param dev
6855  *   Pointer to Ethernet device.
6856  * @param fdir_filter
6857  *   Filter to be updated.
6858  *
6859  * @return
6860  *   0 on success, a negative errno value otherwise and rte_errno is set.
6861  */
6862 static int
6863 flow_fdir_filter_update(struct rte_eth_dev *dev,
6864 			const struct rte_eth_fdir_filter *fdir_filter)
6865 {
6866 	int ret;
6867 
6868 	ret = flow_fdir_filter_delete(dev, fdir_filter);
6869 	if (ret)
6870 		return ret;
6871 	return flow_fdir_filter_add(dev, fdir_filter);
6872 }
6873 
6874 /**
6875  * Flush all filters.
6876  *
6877  * @param dev
6878  *   Pointer to Ethernet device.
6879  */
6880 static void
6881 flow_fdir_filter_flush(struct rte_eth_dev *dev)
6882 {
6883 	struct mlx5_priv *priv = dev->data->dev_private;
6884 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6885 
6886 	while (!LIST_EMPTY(&priv->fdir_flows)) {
6887 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
6888 		LIST_REMOVE(priv_fdir_flow, next);
6889 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
6890 		mlx5_free(priv_fdir_flow->fdir);
6891 		mlx5_free(priv_fdir_flow);
6892 	}
6893 }
6894 
6895 /**
6896  * Get flow director information.
6897  *
6898  * @param dev
6899  *   Pointer to Ethernet device.
6900  * @param[out] fdir_info
6901  *   Resulting flow director information.
6902  */
6903 static void
6904 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
6905 {
6906 	struct rte_eth_fdir_masks *mask =
6907 		&dev->data->dev_conf.fdir_conf.mask;
6908 
6909 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
6910 	fdir_info->guarant_spc = 0;
6911 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
6912 	fdir_info->max_flexpayload = 0;
6913 	fdir_info->flow_types_mask[0] = 0;
6914 	fdir_info->flex_payload_unit = 0;
6915 	fdir_info->max_flex_payload_segment_num = 0;
6916 	fdir_info->flex_payload_limit = 0;
6917 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
6918 }
6919 
6920 /**
6921  * Deal with flow director operations.
6922  *
6923  * @param dev
6924  *   Pointer to Ethernet device.
6925  * @param filter_op
6926  *   Operation to perform.
6927  * @param arg
6928  *   Pointer to operation-specific structure.
6929  *
6930  * @return
6931  *   0 on success, a negative errno value otherwise and rte_errno is set.
6932  */
6933 static int
6934 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
6935 		    void *arg)
6936 {
6937 	enum rte_fdir_mode fdir_mode =
6938 		dev->data->dev_conf.fdir_conf.mode;
6939 
6940 	if (filter_op == RTE_ETH_FILTER_NOP)
6941 		return 0;
6942 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
6943 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
6944 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
6945 			dev->data->port_id, fdir_mode);
6946 		rte_errno = EINVAL;
6947 		return -rte_errno;
6948 	}
6949 	switch (filter_op) {
6950 	case RTE_ETH_FILTER_ADD:
6951 		return flow_fdir_filter_add(dev, arg);
6952 	case RTE_ETH_FILTER_UPDATE:
6953 		return flow_fdir_filter_update(dev, arg);
6954 	case RTE_ETH_FILTER_DELETE:
6955 		return flow_fdir_filter_delete(dev, arg);
6956 	case RTE_ETH_FILTER_FLUSH:
6957 		flow_fdir_filter_flush(dev);
6958 		break;
6959 	case RTE_ETH_FILTER_INFO:
6960 		flow_fdir_info_get(dev, arg);
6961 		break;
6962 	default:
6963 		DRV_LOG(DEBUG, "port %u unknown operation %u",
6964 			dev->data->port_id, filter_op);
6965 		rte_errno = EINVAL;
6966 		return -rte_errno;
6967 	}
6968 	return 0;
6969 }
6970 
6971 /**
6972  * Manage filter operations.
6973  *
6974  * @param dev
6975  *   Pointer to Ethernet device structure.
6976  * @param filter_type
6977  *   Filter type.
6978  * @param filter_op
6979  *   Operation to perform.
6980  * @param arg
6981  *   Pointer to operation-specific structure.
6982  *
6983  * @return
6984  *   0 on success, a negative errno value otherwise and rte_errno is set.
6985  */
6986 int
6987 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6988 		     enum rte_filter_type filter_type,
6989 		     enum rte_filter_op filter_op,
6990 		     void *arg)
6991 {
6992 	switch (filter_type) {
6993 	case RTE_ETH_FILTER_GENERIC:
6994 		if (filter_op != RTE_ETH_FILTER_GET) {
6995 			rte_errno = EINVAL;
6996 			return -rte_errno;
6997 		}
6998 		*(const void **)arg = &mlx5_flow_ops;
6999 		return 0;
7000 	case RTE_ETH_FILTER_FDIR:
7001 		return flow_fdir_ctrl_func(dev, filter_op, arg);
7002 	default:
7003 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
7004 			dev->data->port_id, filter_type);
7005 		rte_errno = ENOTSUP;
7006 		return -rte_errno;
7007 	}
7008 	return 0;
7009 }
7010 
7011 /**
7012  * Create the needed meter and suffix tables.
7013  *
7014  * @param[in] dev
7015  *   Pointer to Ethernet device.
7016  * @param[in] fm
7017  *   Pointer to the flow meter.
7018  *
7019  * @return
7020  *   Pointer to table set on success, NULL otherwise.
7021  */
7022 struct mlx5_meter_domains_infos *
7023 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7024 			  const struct mlx5_flow_meter *fm)
7025 {
7026 	const struct mlx5_flow_driver_ops *fops;
7027 
7028 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7029 	return fops->create_mtr_tbls(dev, fm);
7030 }
7031 
7032 /**
7033  * Destroy the meter table set.
7034  *
7035  * @param[in] dev
7036  *   Pointer to Ethernet device.
7037  * @param[in] tbl
7038  *   Pointer to the meter table set.
7039  *
7040  * @return
7041  *   0 on success.
7042  */
7043 int
7044 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7045 			   struct mlx5_meter_domains_infos *tbls)
7046 {
7047 	const struct mlx5_flow_driver_ops *fops;
7048 
7049 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7050 	return fops->destroy_mtr_tbls(dev, tbls);
7051 }
7052 
7053 /**
7054  * Create policer rules.
7055  *
7056  * @param[in] dev
7057  *   Pointer to Ethernet device.
7058  * @param[in] fm
7059  *   Pointer to flow meter structure.
7060  * @param[in] attr
7061  *   Pointer to flow attributes.
7062  *
7063  * @return
7064  *   0 on success, -1 otherwise.
7065  */
7066 int
7067 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
7068 			       struct mlx5_flow_meter *fm,
7069 			       const struct rte_flow_attr *attr)
7070 {
7071 	const struct mlx5_flow_driver_ops *fops;
7072 
7073 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7074 	return fops->create_policer_rules(dev, fm, attr);
7075 }
7076 
7077 /**
7078  * Destroy policer rules.
7079  *
7080  * @param[in] fm
7081  *   Pointer to flow meter structure.
7082  * @param[in] attr
7083  *   Pointer to flow attributes.
7084  *
7085  * @return
7086  *   0 on success, -1 otherwise.
7087  */
7088 int
7089 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
7090 				struct mlx5_flow_meter *fm,
7091 				const struct rte_flow_attr *attr)
7092 {
7093 	const struct mlx5_flow_driver_ops *fops;
7094 
7095 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7096 	return fops->destroy_policer_rules(dev, fm, attr);
7097 }
7098 
7099 /**
7100  * Allocate a counter.
7101  *
7102  * @param[in] dev
7103  *   Pointer to Ethernet device structure.
7104  *
7105  * @return
7106  *   Index to allocated counter  on success, 0 otherwise.
7107  */
7108 uint32_t
7109 mlx5_counter_alloc(struct rte_eth_dev *dev)
7110 {
7111 	const struct mlx5_flow_driver_ops *fops;
7112 	struct rte_flow_attr attr = { .transfer = 0 };
7113 
7114 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7115 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7116 		return fops->counter_alloc(dev);
7117 	}
7118 	DRV_LOG(ERR,
7119 		"port %u counter allocate is not supported.",
7120 		 dev->data->port_id);
7121 	return 0;
7122 }
7123 
7124 /**
7125  * Free a counter.
7126  *
7127  * @param[in] dev
7128  *   Pointer to Ethernet device structure.
7129  * @param[in] cnt
7130  *   Index to counter to be free.
7131  */
7132 void
7133 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7134 {
7135 	const struct mlx5_flow_driver_ops *fops;
7136 	struct rte_flow_attr attr = { .transfer = 0 };
7137 
7138 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7139 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7140 		fops->counter_free(dev, cnt);
7141 		return;
7142 	}
7143 	DRV_LOG(ERR,
7144 		"port %u counter free is not supported.",
7145 		 dev->data->port_id);
7146 }
7147 
7148 /**
7149  * Query counter statistics.
7150  *
7151  * @param[in] dev
7152  *   Pointer to Ethernet device structure.
7153  * @param[in] cnt
7154  *   Index to counter to query.
7155  * @param[in] clear
7156  *   Set to clear counter statistics.
7157  * @param[out] pkts
7158  *   The counter hits packets number to save.
7159  * @param[out] bytes
7160  *   The counter hits bytes number to save.
7161  *
7162  * @return
7163  *   0 on success, a negative errno value otherwise.
7164  */
7165 int
7166 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7167 		   bool clear, uint64_t *pkts, uint64_t *bytes)
7168 {
7169 	const struct mlx5_flow_driver_ops *fops;
7170 	struct rte_flow_attr attr = { .transfer = 0 };
7171 
7172 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7173 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7174 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
7175 	}
7176 	DRV_LOG(ERR,
7177 		"port %u counter query is not supported.",
7178 		 dev->data->port_id);
7179 	return -ENOTSUP;
7180 }
7181 
7182 /**
7183  * Allocate a new memory for the counter values wrapped by all the needed
7184  * management.
7185  *
7186  * @param[in] sh
7187  *   Pointer to mlx5_dev_ctx_shared object.
7188  *
7189  * @return
7190  *   0 on success, a negative errno value otherwise.
7191  */
7192 static int
7193 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7194 {
7195 	struct mlx5_devx_mkey_attr mkey_attr;
7196 	struct mlx5_counter_stats_mem_mng *mem_mng;
7197 	volatile struct flow_counter_stats *raw_data;
7198 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7199 	int size = (sizeof(struct flow_counter_stats) *
7200 			MLX5_COUNTERS_PER_POOL +
7201 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7202 			sizeof(struct mlx5_counter_stats_mem_mng);
7203 	size_t pgsize = rte_mem_page_size();
7204 	uint8_t *mem;
7205 	int i;
7206 
7207 	if (pgsize == (size_t)-1) {
7208 		DRV_LOG(ERR, "Failed to get mem page size");
7209 		rte_errno = ENOMEM;
7210 		return -ENOMEM;
7211 	}
7212 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7213 	if (!mem) {
7214 		rte_errno = ENOMEM;
7215 		return -ENOMEM;
7216 	}
7217 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7218 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7219 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
7220 						 IBV_ACCESS_LOCAL_WRITE);
7221 	if (!mem_mng->umem) {
7222 		rte_errno = errno;
7223 		mlx5_free(mem);
7224 		return -rte_errno;
7225 	}
7226 	mkey_attr.addr = (uintptr_t)mem;
7227 	mkey_attr.size = size;
7228 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
7229 	mkey_attr.pd = sh->pdn;
7230 	mkey_attr.log_entity_size = 0;
7231 	mkey_attr.pg_access = 0;
7232 	mkey_attr.klm_array = NULL;
7233 	mkey_attr.klm_num = 0;
7234 	mkey_attr.relaxed_ordering = sh->cmng.relaxed_ordering;
7235 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
7236 	if (!mem_mng->dm) {
7237 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
7238 		rte_errno = errno;
7239 		mlx5_free(mem);
7240 		return -rte_errno;
7241 	}
7242 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7243 	raw_data = (volatile struct flow_counter_stats *)mem;
7244 	for (i = 0; i < raws_n; ++i) {
7245 		mem_mng->raws[i].mem_mng = mem_mng;
7246 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7247 	}
7248 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7249 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7250 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7251 				 next);
7252 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7253 	sh->cmng.mem_mng = mem_mng;
7254 	return 0;
7255 }
7256 
7257 /**
7258  * Set the statistic memory to the new counter pool.
7259  *
7260  * @param[in] sh
7261  *   Pointer to mlx5_dev_ctx_shared object.
7262  * @param[in] pool
7263  *   Pointer to the pool to set the statistic memory.
7264  *
7265  * @return
7266  *   0 on success, a negative errno value otherwise.
7267  */
7268 static int
7269 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7270 			       struct mlx5_flow_counter_pool *pool)
7271 {
7272 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7273 	/* Resize statistic memory once used out. */
7274 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7275 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7276 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7277 		return -1;
7278 	}
7279 	rte_spinlock_lock(&pool->sl);
7280 	pool->raw = cmng->mem_mng->raws + pool->index %
7281 		    MLX5_CNT_CONTAINER_RESIZE;
7282 	rte_spinlock_unlock(&pool->sl);
7283 	pool->raw_hw = NULL;
7284 	return 0;
7285 }
7286 
7287 #define MLX5_POOL_QUERY_FREQ_US 1000000
7288 
7289 /**
7290  * Set the periodic procedure for triggering asynchronous batch queries for all
7291  * the counter pools.
7292  *
7293  * @param[in] sh
7294  *   Pointer to mlx5_dev_ctx_shared object.
7295  */
7296 void
7297 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7298 {
7299 	uint32_t pools_n, us;
7300 
7301 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7302 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7303 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7304 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7305 		sh->cmng.query_thread_on = 0;
7306 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7307 	} else {
7308 		sh->cmng.query_thread_on = 1;
7309 	}
7310 }
7311 
7312 /**
7313  * The periodic procedure for triggering asynchronous batch queries for all the
7314  * counter pools. This function is probably called by the host thread.
7315  *
7316  * @param[in] arg
7317  *   The parameter for the alarm process.
7318  */
7319 void
7320 mlx5_flow_query_alarm(void *arg)
7321 {
7322 	struct mlx5_dev_ctx_shared *sh = arg;
7323 	int ret;
7324 	uint16_t pool_index = sh->cmng.pool_index;
7325 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7326 	struct mlx5_flow_counter_pool *pool;
7327 	uint16_t n_valid;
7328 
7329 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7330 		goto set_alarm;
7331 	rte_spinlock_lock(&cmng->pool_update_sl);
7332 	pool = cmng->pools[pool_index];
7333 	n_valid = cmng->n_valid;
7334 	rte_spinlock_unlock(&cmng->pool_update_sl);
7335 	/* Set the statistic memory to the new created pool. */
7336 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7337 		goto set_alarm;
7338 	if (pool->raw_hw)
7339 		/* There is a pool query in progress. */
7340 		goto set_alarm;
7341 	pool->raw_hw =
7342 		LIST_FIRST(&sh->cmng.free_stat_raws);
7343 	if (!pool->raw_hw)
7344 		/* No free counter statistics raw memory. */
7345 		goto set_alarm;
7346 	/*
7347 	 * Identify the counters released between query trigger and query
7348 	 * handle more efficiently. The counter released in this gap period
7349 	 * should wait for a new round of query as the new arrived packets
7350 	 * will not be taken into account.
7351 	 */
7352 	pool->query_gen++;
7353 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7354 					       MLX5_COUNTERS_PER_POOL,
7355 					       NULL, NULL,
7356 					       pool->raw_hw->mem_mng->dm->id,
7357 					       (void *)(uintptr_t)
7358 					       pool->raw_hw->data,
7359 					       sh->devx_comp,
7360 					       (uint64_t)(uintptr_t)pool);
7361 	if (ret) {
7362 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
7363 			" %d", pool->min_dcs->id);
7364 		pool->raw_hw = NULL;
7365 		goto set_alarm;
7366 	}
7367 	LIST_REMOVE(pool->raw_hw, next);
7368 	sh->cmng.pending_queries++;
7369 	pool_index++;
7370 	if (pool_index >= n_valid)
7371 		pool_index = 0;
7372 set_alarm:
7373 	sh->cmng.pool_index = pool_index;
7374 	mlx5_set_query_alarm(sh);
7375 }
7376 
7377 /**
7378  * Check and callback event for new aged flow in the counter pool
7379  *
7380  * @param[in] sh
7381  *   Pointer to mlx5_dev_ctx_shared object.
7382  * @param[in] pool
7383  *   Pointer to Current counter pool.
7384  */
7385 static void
7386 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
7387 		   struct mlx5_flow_counter_pool *pool)
7388 {
7389 	struct mlx5_priv *priv;
7390 	struct mlx5_flow_counter *cnt;
7391 	struct mlx5_age_info *age_info;
7392 	struct mlx5_age_param *age_param;
7393 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
7394 	struct mlx5_counter_stats_raw *prev = pool->raw;
7395 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
7396 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
7397 	uint16_t expected = AGE_CANDIDATE;
7398 	uint32_t i;
7399 
7400 	pool->time_of_last_age_check = curr_time;
7401 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
7402 		cnt = MLX5_POOL_GET_CNT(pool, i);
7403 		age_param = MLX5_CNT_TO_AGE(cnt);
7404 		if (__atomic_load_n(&age_param->state,
7405 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
7406 			continue;
7407 		if (cur->data[i].hits != prev->data[i].hits) {
7408 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
7409 					 __ATOMIC_RELAXED);
7410 			continue;
7411 		}
7412 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
7413 				       time_delta,
7414 				       __ATOMIC_RELAXED) <= age_param->timeout)
7415 			continue;
7416 		/**
7417 		 * Hold the lock first, or if between the
7418 		 * state AGE_TMOUT and tailq operation the
7419 		 * release happened, the release procedure
7420 		 * may delete a non-existent tailq node.
7421 		 */
7422 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
7423 		age_info = GET_PORT_AGE_INFO(priv);
7424 		rte_spinlock_lock(&age_info->aged_sl);
7425 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
7426 						AGE_TMOUT, false,
7427 						__ATOMIC_RELAXED,
7428 						__ATOMIC_RELAXED)) {
7429 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
7430 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
7431 		}
7432 		rte_spinlock_unlock(&age_info->aged_sl);
7433 	}
7434 	for (i = 0; i < sh->max_port; i++) {
7435 		age_info = &sh->port[i].age_info;
7436 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
7437 			continue;
7438 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
7439 			rte_eth_dev_callback_process
7440 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
7441 				RTE_ETH_EVENT_FLOW_AGED, NULL);
7442 		age_info->flags = 0;
7443 	}
7444 }
7445 
7446 /**
7447  * Handler for the HW respond about ready values from an asynchronous batch
7448  * query. This function is probably called by the host thread.
7449  *
7450  * @param[in] sh
7451  *   The pointer to the shared device context.
7452  * @param[in] async_id
7453  *   The Devx async ID.
7454  * @param[in] status
7455  *   The status of the completion.
7456  */
7457 void
7458 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
7459 				  uint64_t async_id, int status)
7460 {
7461 	struct mlx5_flow_counter_pool *pool =
7462 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
7463 	struct mlx5_counter_stats_raw *raw_to_free;
7464 	uint8_t query_gen = pool->query_gen ^ 1;
7465 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7466 	enum mlx5_counter_type cnt_type =
7467 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
7468 				MLX5_COUNTER_TYPE_ORIGIN;
7469 
7470 	if (unlikely(status)) {
7471 		raw_to_free = pool->raw_hw;
7472 	} else {
7473 		raw_to_free = pool->raw;
7474 		if (pool->is_aged)
7475 			mlx5_flow_aging_check(sh, pool);
7476 		rte_spinlock_lock(&pool->sl);
7477 		pool->raw = pool->raw_hw;
7478 		rte_spinlock_unlock(&pool->sl);
7479 		/* Be sure the new raw counters data is updated in memory. */
7480 		rte_io_wmb();
7481 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
7482 			rte_spinlock_lock(&cmng->csl[cnt_type]);
7483 			TAILQ_CONCAT(&cmng->counters[cnt_type],
7484 				     &pool->counters[query_gen], next);
7485 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
7486 		}
7487 	}
7488 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
7489 	pool->raw_hw = NULL;
7490 	sh->cmng.pending_queries--;
7491 }
7492 
7493 static const struct mlx5_flow_tbl_data_entry  *
7494 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
7495 {
7496 	struct mlx5_priv *priv = dev->data->dev_private;
7497 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7498 	struct mlx5_hlist_entry *he;
7499 	union tunnel_offload_mark mbits = { .val = mark };
7500 	union mlx5_flow_tbl_key table_key = {
7501 		{
7502 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
7503 			.reserved = 0,
7504 			.domain = !!mbits.transfer,
7505 			.direction = 0,
7506 		}
7507 	};
7508 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64);
7509 	return he ?
7510 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
7511 }
7512 
7513 static uint32_t
7514 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7515 				const struct mlx5_flow_tunnel *tunnel,
7516 				uint32_t group, uint32_t *table,
7517 				struct rte_flow_error *error)
7518 {
7519 	struct mlx5_hlist_entry *he;
7520 	struct tunnel_tbl_entry *tte;
7521 	union tunnel_tbl_key key = {
7522 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7523 		.group = group
7524 	};
7525 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7526 	struct mlx5_hlist *group_hash;
7527 
7528 	group_hash = tunnel ? tunnel->groups : thub->groups;
7529 	he = mlx5_hlist_lookup(group_hash, key.val);
7530 	if (!he) {
7531 		int ret;
7532 		tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
7533 				  sizeof(*tte), 0,
7534 				  SOCKET_ID_ANY);
7535 		if (!tte)
7536 			goto err;
7537 		tte->hash.key = key.val;
7538 		ret = mlx5_flow_id_get(thub->table_ids, &tte->flow_table);
7539 		if (ret) {
7540 			mlx5_free(tte);
7541 			goto err;
7542 		}
7543 		tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7544 		mlx5_hlist_insert(group_hash, &tte->hash);
7545 	} else {
7546 		tte = container_of(he, typeof(*tte), hash);
7547 	}
7548 	*table = tte->flow_table;
7549 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7550 		dev->data->port_id, key.tunnel_id, group, *table);
7551 	return 0;
7552 
7553 err:
7554 	return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7555 				  NULL, "tunnel group index not supported");
7556 }
7557 
7558 static int
7559 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7560 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
7561 {
7562 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7563 		if (group == UINT32_MAX)
7564 			return rte_flow_error_set
7565 						(error, EINVAL,
7566 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7567 						 NULL,
7568 						 "group index not supported");
7569 		*table = group + 1;
7570 	} else {
7571 		*table = group;
7572 	}
7573 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7574 	return 0;
7575 }
7576 
7577 /**
7578  * Translate the rte_flow group index to HW table value.
7579  *
7580  * If tunnel offload is disabled, all group ids converted to flow table
7581  * id using the standard method.
7582  * If tunnel offload is enabled, group id can be converted using the
7583  * standard or tunnel conversion method. Group conversion method
7584  * selection depends on flags in `grp_info` parameter:
7585  * - Internal (grp_info.external == 0) groups conversion uses the
7586  *   standard method.
7587  * - Group ids in JUMP action converted with the tunnel conversion.
7588  * - Group id in rule attribute conversion depends on a rule type and
7589  *   group id value:
7590  *   ** non zero group attributes converted with the tunnel method
7591  *   ** zero group attribute in non-tunnel rule is converted using the
7592  *      standard method - there's only one root table
7593  *   ** zero group attribute in steer tunnel rule is converted with the
7594  *      standard method - single root table
7595  *   ** zero group attribute in match tunnel rule is a special OvS
7596  *      case: that value is used for portability reasons. That group
7597  *      id is converted with the tunnel conversion method.
7598  *
7599  * @param[in] dev
7600  *   Port device
7601  * @param[in] tunnel
7602  *   PMD tunnel offload object
7603  * @param[in] group
7604  *   rte_flow group index value.
7605  * @param[out] table
7606  *   HW table value.
7607  * @param[in] grp_info
7608  *   flags used for conversion
7609  * @param[out] error
7610  *   Pointer to error structure.
7611  *
7612  * @return
7613  *   0 on success, a negative errno value otherwise and rte_errno is set.
7614  */
7615 int
7616 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7617 			 const struct mlx5_flow_tunnel *tunnel,
7618 			 uint32_t group, uint32_t *table,
7619 			 struct flow_grp_info grp_info,
7620 			 struct rte_flow_error *error)
7621 {
7622 	int ret;
7623 	bool standard_translation;
7624 
7625 	if (grp_info.external && group < MLX5_MAX_TABLES_EXTERNAL)
7626 		group *= MLX5_FLOW_TABLE_FACTOR;
7627 	if (is_tunnel_offload_active(dev)) {
7628 		standard_translation = !grp_info.external ||
7629 					grp_info.std_tbl_fix;
7630 	} else {
7631 		standard_translation = true;
7632 	}
7633 	DRV_LOG(DEBUG,
7634 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7635 		dev->data->port_id, group, grp_info.transfer,
7636 		grp_info.external, grp_info.fdb_def_rule,
7637 		standard_translation ? "STANDARD" : "TUNNEL");
7638 	if (standard_translation)
7639 		ret = flow_group_to_table(dev->data->port_id, group, table,
7640 					  grp_info, error);
7641 	else
7642 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7643 						      table, error);
7644 
7645 	return ret;
7646 }
7647 
7648 /**
7649  * Discover availability of metadata reg_c's.
7650  *
7651  * Iteratively use test flows to check availability.
7652  *
7653  * @param[in] dev
7654  *   Pointer to the Ethernet device structure.
7655  *
7656  * @return
7657  *   0 on success, a negative errno value otherwise and rte_errno is set.
7658  */
7659 int
7660 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7661 {
7662 	struct mlx5_priv *priv = dev->data->dev_private;
7663 	struct mlx5_dev_config *config = &priv->config;
7664 	enum modify_reg idx;
7665 	int n = 0;
7666 
7667 	/* reg_c[0] and reg_c[1] are reserved. */
7668 	config->flow_mreg_c[n++] = REG_C_0;
7669 	config->flow_mreg_c[n++] = REG_C_1;
7670 	/* Discover availability of other reg_c's. */
7671 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7672 		struct rte_flow_attr attr = {
7673 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7674 			.priority = MLX5_FLOW_PRIO_RSVD,
7675 			.ingress = 1,
7676 		};
7677 		struct rte_flow_item items[] = {
7678 			[0] = {
7679 				.type = RTE_FLOW_ITEM_TYPE_END,
7680 			},
7681 		};
7682 		struct rte_flow_action actions[] = {
7683 			[0] = {
7684 				.type = (enum rte_flow_action_type)
7685 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7686 				.conf = &(struct mlx5_flow_action_copy_mreg){
7687 					.src = REG_C_1,
7688 					.dst = idx,
7689 				},
7690 			},
7691 			[1] = {
7692 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7693 				.conf = &(struct rte_flow_action_jump){
7694 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7695 				},
7696 			},
7697 			[2] = {
7698 				.type = RTE_FLOW_ACTION_TYPE_END,
7699 			},
7700 		};
7701 		uint32_t flow_idx;
7702 		struct rte_flow *flow;
7703 		struct rte_flow_error error;
7704 
7705 		if (!config->dv_flow_en)
7706 			break;
7707 		/* Create internal flow, validation skips copy action. */
7708 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7709 					    actions, false, &error);
7710 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7711 				      flow_idx);
7712 		if (!flow)
7713 			continue;
7714 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
7715 			config->flow_mreg_c[n++] = idx;
7716 		flow_list_destroy(dev, NULL, flow_idx);
7717 	}
7718 	for (; n < MLX5_MREG_C_NUM; ++n)
7719 		config->flow_mreg_c[n] = REG_NON;
7720 	return 0;
7721 }
7722 
7723 /**
7724  * Dump flow raw hw data to file
7725  *
7726  * @param[in] dev
7727  *    The pointer to Ethernet device.
7728  * @param[in] file
7729  *   A pointer to a file for output.
7730  * @param[out] error
7731  *   Perform verbose error reporting if not NULL. PMDs initialize this
7732  *   structure in case of error only.
7733  * @return
7734  *   0 on success, a nagative value otherwise.
7735  */
7736 int
7737 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7738 		   FILE *file,
7739 		   struct rte_flow_error *error __rte_unused)
7740 {
7741 	struct mlx5_priv *priv = dev->data->dev_private;
7742 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7743 
7744 	if (!priv->config.dv_flow_en) {
7745 		if (fputs("device dv flow disabled\n", file) <= 0)
7746 			return -errno;
7747 		return -ENOTSUP;
7748 	}
7749 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7750 				       sh->tx_domain, file);
7751 }
7752 
7753 /**
7754  * Get aged-out flows.
7755  *
7756  * @param[in] dev
7757  *   Pointer to the Ethernet device structure.
7758  * @param[in] context
7759  *   The address of an array of pointers to the aged-out flows contexts.
7760  * @param[in] nb_countexts
7761  *   The length of context array pointers.
7762  * @param[out] error
7763  *   Perform verbose error reporting if not NULL. Initialized in case of
7764  *   error only.
7765  *
7766  * @return
7767  *   how many contexts get in success, otherwise negative errno value.
7768  *   if nb_contexts is 0, return the amount of all aged contexts.
7769  *   if nb_contexts is not 0 , return the amount of aged flows reported
7770  *   in the context array.
7771  */
7772 int
7773 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7774 			uint32_t nb_contexts, struct rte_flow_error *error)
7775 {
7776 	const struct mlx5_flow_driver_ops *fops;
7777 	struct rte_flow_attr attr = { .transfer = 0 };
7778 
7779 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7780 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7781 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7782 						    error);
7783 	}
7784 	DRV_LOG(ERR,
7785 		"port %u get aged flows is not supported.",
7786 		 dev->data->port_id);
7787 	return -ENOTSUP;
7788 }
7789 
7790 /* Wrapper for driver action_validate op callback */
7791 static int
7792 flow_drv_action_validate(struct rte_eth_dev *dev,
7793 			 const struct rte_flow_shared_action_conf *conf,
7794 			 const struct rte_flow_action *action,
7795 			 const struct mlx5_flow_driver_ops *fops,
7796 			 struct rte_flow_error *error)
7797 {
7798 	static const char err_msg[] = "shared action validation unsupported";
7799 
7800 	if (!fops->action_validate) {
7801 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7802 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7803 				   NULL, err_msg);
7804 		return -rte_errno;
7805 	}
7806 	return fops->action_validate(dev, conf, action, error);
7807 }
7808 
7809 /**
7810  * Destroys the shared action by handle.
7811  *
7812  * @param dev
7813  *   Pointer to Ethernet device structure.
7814  * @param[in] action
7815  *   Handle for the shared action to be destroyed.
7816  * @param[out] error
7817  *   Perform verbose error reporting if not NULL. PMDs initialize this
7818  *   structure in case of error only.
7819  *
7820  * @return
7821  *   0 on success, a negative errno value otherwise and rte_errno is set.
7822  *
7823  * @note: wrapper for driver action_create op callback.
7824  */
7825 static int
7826 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7827 			   struct rte_flow_shared_action *action,
7828 			   struct rte_flow_error *error)
7829 {
7830 	static const char err_msg[] = "shared action destruction unsupported";
7831 	struct rte_flow_attr attr = { .transfer = 0 };
7832 	const struct mlx5_flow_driver_ops *fops =
7833 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7834 
7835 	if (!fops->action_destroy) {
7836 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7837 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7838 				   NULL, err_msg);
7839 		return -rte_errno;
7840 	}
7841 	return fops->action_destroy(dev, action, error);
7842 }
7843 
7844 /* Wrapper for driver action_destroy op callback */
7845 static int
7846 flow_drv_action_update(struct rte_eth_dev *dev,
7847 		       struct rte_flow_shared_action *action,
7848 		       const void *action_conf,
7849 		       const struct mlx5_flow_driver_ops *fops,
7850 		       struct rte_flow_error *error)
7851 {
7852 	static const char err_msg[] = "shared action update unsupported";
7853 
7854 	if (!fops->action_update) {
7855 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7856 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7857 				   NULL, err_msg);
7858 		return -rte_errno;
7859 	}
7860 	return fops->action_update(dev, action, action_conf, error);
7861 }
7862 
7863 /**
7864  * Create shared action for reuse in multiple flow rules.
7865  *
7866  * @param dev
7867  *   Pointer to Ethernet device structure.
7868  * @param[in] action
7869  *   Action configuration for shared action creation.
7870  * @param[out] error
7871  *   Perform verbose error reporting if not NULL. PMDs initialize this
7872  *   structure in case of error only.
7873  * @return
7874  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7875  */
7876 static struct rte_flow_shared_action *
7877 mlx5_shared_action_create(struct rte_eth_dev *dev,
7878 			  const struct rte_flow_shared_action_conf *conf,
7879 			  const struct rte_flow_action *action,
7880 			  struct rte_flow_error *error)
7881 {
7882 	static const char err_msg[] = "shared action creation unsupported";
7883 	struct rte_flow_attr attr = { .transfer = 0 };
7884 	const struct mlx5_flow_driver_ops *fops =
7885 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7886 
7887 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7888 		return NULL;
7889 	if (!fops->action_create) {
7890 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7891 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7892 				   NULL, err_msg);
7893 		return NULL;
7894 	}
7895 	return fops->action_create(dev, conf, action, error);
7896 }
7897 
7898 /**
7899  * Updates inplace the shared action configuration pointed by *action* handle
7900  * with the configuration provided as *action* argument.
7901  * The update of the shared action configuration effects all flow rules reusing
7902  * the action via handle.
7903  *
7904  * @param dev
7905  *   Pointer to Ethernet device structure.
7906  * @param[in] shared_action
7907  *   Handle for the shared action to be updated.
7908  * @param[in] action
7909  *   Action specification used to modify the action pointed by handle.
7910  *   *action* should be of same type with the action pointed by the *action*
7911  *   handle argument, otherwise considered as invalid.
7912  * @param[out] error
7913  *   Perform verbose error reporting if not NULL. PMDs initialize this
7914  *   structure in case of error only.
7915  *
7916  * @return
7917  *   0 on success, a negative errno value otherwise and rte_errno is set.
7918  */
7919 static int
7920 mlx5_shared_action_update(struct rte_eth_dev *dev,
7921 		struct rte_flow_shared_action *shared_action,
7922 		const struct rte_flow_action *action,
7923 		struct rte_flow_error *error)
7924 {
7925 	struct rte_flow_attr attr = { .transfer = 0 };
7926 	const struct mlx5_flow_driver_ops *fops =
7927 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7928 	int ret;
7929 
7930 	switch (shared_action->type) {
7931 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7932 		if (action->type != RTE_FLOW_ACTION_TYPE_RSS) {
7933 			return rte_flow_error_set(error, EINVAL,
7934 						  RTE_FLOW_ERROR_TYPE_ACTION,
7935 						  NULL,
7936 						  "update action type invalid");
7937 		}
7938 		ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7939 		if (ret)
7940 			return ret;
7941 		return flow_drv_action_update(dev, shared_action, action->conf,
7942 					      fops, error);
7943 	default:
7944 		return rte_flow_error_set(error, ENOTSUP,
7945 					  RTE_FLOW_ERROR_TYPE_ACTION,
7946 					  NULL,
7947 					  "action type not supported");
7948 	}
7949 }
7950 
7951 /**
7952  * Query the shared action by handle.
7953  *
7954  * This function allows retrieving action-specific data such as counters.
7955  * Data is gathered by special action which may be present/referenced in
7956  * more than one flow rule definition.
7957  *
7958  * \see RTE_FLOW_ACTION_TYPE_COUNT
7959  *
7960  * @param dev
7961  *   Pointer to Ethernet device structure.
7962  * @param[in] action
7963  *   Handle for the shared action to query.
7964  * @param[in, out] data
7965  *   Pointer to storage for the associated query data type.
7966  * @param[out] error
7967  *   Perform verbose error reporting if not NULL. PMDs initialize this
7968  *   structure in case of error only.
7969  *
7970  * @return
7971  *   0 on success, a negative errno value otherwise and rte_errno is set.
7972  */
7973 static int
7974 mlx5_shared_action_query(struct rte_eth_dev *dev,
7975 			 const struct rte_flow_shared_action *action,
7976 			 void *data,
7977 			 struct rte_flow_error *error)
7978 {
7979 	(void)dev;
7980 	switch (action->type) {
7981 	case MLX5_RTE_FLOW_ACTION_TYPE_SHARED_RSS:
7982 		__atomic_load(&action->refcnt, (uint32_t *)data,
7983 			      __ATOMIC_RELAXED);
7984 		return 0;
7985 	default:
7986 		return rte_flow_error_set(error, ENOTSUP,
7987 					  RTE_FLOW_ERROR_TYPE_ACTION,
7988 					  NULL,
7989 					  "action type not supported");
7990 	}
7991 }
7992 
7993 /**
7994  * Destroy all shared actions.
7995  *
7996  * @param dev
7997  *   Pointer to Ethernet device.
7998  *
7999  * @return
8000  *   0 on success, a negative errno value otherwise and rte_errno is set.
8001  */
8002 int
8003 mlx5_shared_action_flush(struct rte_eth_dev *dev)
8004 {
8005 	struct rte_flow_error error;
8006 	struct mlx5_priv *priv = dev->data->dev_private;
8007 	struct rte_flow_shared_action *action;
8008 	int ret = 0;
8009 
8010 	while (!LIST_EMPTY(&priv->shared_actions)) {
8011 		action = LIST_FIRST(&priv->shared_actions);
8012 		ret = mlx5_shared_action_destroy(dev, action, &error);
8013 	}
8014 	return ret;
8015 }
8016 
8017 static void
8018 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
8019 		      struct mlx5_flow_tunnel *tunnel)
8020 {
8021 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8022 	struct mlx5_flow_id_pool *id_pool = thub->tunnel_ids;
8023 
8024 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
8025 		dev->data->port_id, tunnel->tunnel_id);
8026 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
8027 	LIST_REMOVE(tunnel, chain);
8028 	mlx5_flow_id_release(id_pool, tunnel->tunnel_id);
8029 	mlx5_hlist_destroy(tunnel->groups, NULL, NULL);
8030 	mlx5_free(tunnel);
8031 }
8032 
8033 static struct mlx5_flow_tunnel *
8034 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
8035 {
8036 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8037 	struct mlx5_flow_tunnel *tun;
8038 
8039 	LIST_FOREACH(tun, &thub->tunnels, chain) {
8040 		if (tun->tunnel_id == id)
8041 			break;
8042 	}
8043 
8044 	return tun;
8045 }
8046 
8047 static struct mlx5_flow_tunnel *
8048 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
8049 			  const struct rte_flow_tunnel *app_tunnel)
8050 {
8051 	int ret;
8052 	struct mlx5_flow_tunnel *tunnel;
8053 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8054 	struct mlx5_flow_id_pool *id_pool = thub->tunnel_ids;
8055 	uint32_t id;
8056 
8057 	ret = mlx5_flow_id_get(id_pool, &id);
8058 	if (ret)
8059 		return NULL;
8060 	/**
8061 	 * mlx5 flow tunnel is an auxlilary data structure
8062 	 * It's not part of IO. No need to allocate it from
8063 	 * huge pages pools dedicated for IO
8064 	 */
8065 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
8066 			     0, SOCKET_ID_ANY);
8067 	if (!tunnel) {
8068 		mlx5_flow_id_pool_release(id_pool);
8069 		return NULL;
8070 	}
8071 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024);
8072 	if (!tunnel->groups) {
8073 		mlx5_flow_id_pool_release(id_pool);
8074 		mlx5_free(tunnel);
8075 		return NULL;
8076 	}
8077 	/* initiate new PMD tunnel */
8078 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
8079 	tunnel->tunnel_id = id;
8080 	tunnel->action.type = (typeof(tunnel->action.type))
8081 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
8082 	tunnel->action.conf = tunnel;
8083 	tunnel->item.type = (typeof(tunnel->item.type))
8084 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
8085 	tunnel->item.spec = tunnel;
8086 	tunnel->item.last = NULL;
8087 	tunnel->item.mask = NULL;
8088 
8089 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
8090 		dev->data->port_id, tunnel->tunnel_id);
8091 
8092 	return tunnel;
8093 }
8094 
8095 static int
8096 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
8097 		     const struct rte_flow_tunnel *app_tunnel,
8098 		     struct mlx5_flow_tunnel **tunnel)
8099 {
8100 	int ret;
8101 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
8102 	struct mlx5_flow_tunnel *tun;
8103 
8104 	LIST_FOREACH(tun, &thub->tunnels, chain) {
8105 		if (!memcmp(app_tunnel, &tun->app_tunnel,
8106 			    sizeof(*app_tunnel))) {
8107 			*tunnel = tun;
8108 			ret = 0;
8109 			break;
8110 		}
8111 	}
8112 	if (!tun) {
8113 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
8114 		if (tun) {
8115 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
8116 			*tunnel = tun;
8117 		} else {
8118 			ret = -ENOMEM;
8119 		}
8120 	}
8121 	if (tun)
8122 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
8123 
8124 	return ret;
8125 }
8126 
8127 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
8128 {
8129 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
8130 
8131 	if (!thub)
8132 		return;
8133 	if (!LIST_EMPTY(&thub->tunnels))
8134 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
8135 	mlx5_flow_id_pool_release(thub->tunnel_ids);
8136 	mlx5_flow_id_pool_release(thub->table_ids);
8137 	mlx5_hlist_destroy(thub->groups, NULL, NULL);
8138 	mlx5_free(thub);
8139 }
8140 
8141 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
8142 {
8143 	int err;
8144 	struct mlx5_flow_tunnel_hub *thub;
8145 
8146 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
8147 			   0, SOCKET_ID_ANY);
8148 	if (!thub)
8149 		return -ENOMEM;
8150 	LIST_INIT(&thub->tunnels);
8151 	thub->tunnel_ids = mlx5_flow_id_pool_alloc(MLX5_MAX_TUNNELS);
8152 	if (!thub->tunnel_ids) {
8153 		err = -rte_errno;
8154 		goto err;
8155 	}
8156 	thub->table_ids = mlx5_flow_id_pool_alloc(MLX5_MAX_TABLES);
8157 	if (!thub->table_ids) {
8158 		err = -rte_errno;
8159 		goto err;
8160 	}
8161 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES);
8162 	if (!thub->groups) {
8163 		err = -rte_errno;
8164 		goto err;
8165 	}
8166 	sh->tunnel_hub = thub;
8167 
8168 	return 0;
8169 
8170 err:
8171 	if (thub->groups)
8172 		mlx5_hlist_destroy(thub->groups, NULL, NULL);
8173 	if (thub->table_ids)
8174 		mlx5_flow_id_pool_release(thub->table_ids);
8175 	if (thub->tunnel_ids)
8176 		mlx5_flow_id_pool_release(thub->tunnel_ids);
8177 	if (thub)
8178 		mlx5_free(thub);
8179 	return err;
8180 }
8181 
8182 #ifndef HAVE_MLX5DV_DR
8183 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
8184 #else
8185 #define MLX5_DOMAIN_SYNC_FLOW \
8186 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
8187 #endif
8188 
8189 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
8190 {
8191 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
8192 	const struct mlx5_flow_driver_ops *fops;
8193 	int ret;
8194 	struct rte_flow_attr attr = { .transfer = 0 };
8195 
8196 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8197 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
8198 	if (ret > 0)
8199 		ret = -ret;
8200 	return ret;
8201 }
8202