xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 6859e67ef64a8d44a5f692a7aefda3a6024a9c96)
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_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rxtx.h"
32 
33 /** Device flow drivers. */
34 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
35 
36 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
37 
38 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
39 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
40 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
41 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
42 #endif
43 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
44 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
45 };
46 
47 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
48 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
49 	(const int []){ \
50 		__VA_ARGS__, 0, \
51 	}
52 
53 /** Node object of input graph for mlx5_flow_expand_rss(). */
54 struct mlx5_flow_expand_node {
55 	const int *const next;
56 	/**<
57 	 * List of next node indexes. Index 0 is interpreted as a terminator.
58 	 */
59 	const enum rte_flow_item_type type;
60 	/**< Pattern item type of current node. */
61 	uint64_t rss_types;
62 	/**<
63 	 * RSS types bit-field associated with this node
64 	 * (see ETH_RSS_* definitions).
65 	 */
66 };
67 
68 /** Object returned by mlx5_flow_expand_rss(). */
69 struct mlx5_flow_expand_rss {
70 	uint32_t entries;
71 	/**< Number of entries @p patterns and @p priorities. */
72 	struct {
73 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
74 		uint32_t priority; /**< Priority offset for each expansion. */
75 	} entry[];
76 };
77 
78 static enum rte_flow_item_type
79 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
80 {
81 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
82 	uint16_t ether_type = 0;
83 	uint16_t ether_type_m;
84 	uint8_t ip_next_proto = 0;
85 	uint8_t ip_next_proto_m;
86 
87 	if (item == NULL || item->spec == NULL)
88 		return ret;
89 	switch (item->type) {
90 	case RTE_FLOW_ITEM_TYPE_ETH:
91 		if (item->mask)
92 			ether_type_m = ((const struct rte_flow_item_eth *)
93 						(item->mask))->type;
94 		else
95 			ether_type_m = rte_flow_item_eth_mask.type;
96 		if (ether_type_m != RTE_BE16(0xFFFF))
97 			break;
98 		ether_type = ((const struct rte_flow_item_eth *)
99 				(item->spec))->type;
100 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
101 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
102 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
103 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
104 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
105 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
106 		else
107 			ret = RTE_FLOW_ITEM_TYPE_END;
108 		break;
109 	case RTE_FLOW_ITEM_TYPE_VLAN:
110 		if (item->mask)
111 			ether_type_m = ((const struct rte_flow_item_vlan *)
112 						(item->mask))->inner_type;
113 		else
114 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
115 		if (ether_type_m != RTE_BE16(0xFFFF))
116 			break;
117 		ether_type = ((const struct rte_flow_item_vlan *)
118 				(item->spec))->inner_type;
119 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
120 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
121 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
122 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
123 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
124 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
125 		else
126 			ret = RTE_FLOW_ITEM_TYPE_END;
127 		break;
128 	case RTE_FLOW_ITEM_TYPE_IPV4:
129 		if (item->mask)
130 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
131 					(item->mask))->hdr.next_proto_id;
132 		else
133 			ip_next_proto_m =
134 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
135 		if (ip_next_proto_m != 0xFF)
136 			break;
137 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
138 				(item->spec))->hdr.next_proto_id;
139 		if (ip_next_proto == IPPROTO_UDP)
140 			ret = RTE_FLOW_ITEM_TYPE_UDP;
141 		else if (ip_next_proto == IPPROTO_TCP)
142 			ret = RTE_FLOW_ITEM_TYPE_TCP;
143 		else if (ip_next_proto == IPPROTO_IP)
144 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
145 		else if (ip_next_proto == IPPROTO_IPV6)
146 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
147 		else
148 			ret = RTE_FLOW_ITEM_TYPE_END;
149 		break;
150 	case RTE_FLOW_ITEM_TYPE_IPV6:
151 		if (item->mask)
152 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
153 						(item->mask))->hdr.proto;
154 		else
155 			ip_next_proto_m =
156 				rte_flow_item_ipv6_mask.hdr.proto;
157 		if (ip_next_proto_m != 0xFF)
158 			break;
159 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
160 				(item->spec))->hdr.proto;
161 		if (ip_next_proto == IPPROTO_UDP)
162 			ret = RTE_FLOW_ITEM_TYPE_UDP;
163 		else if (ip_next_proto == IPPROTO_TCP)
164 			ret = RTE_FLOW_ITEM_TYPE_TCP;
165 		else if (ip_next_proto == IPPROTO_IP)
166 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
167 		else if (ip_next_proto == IPPROTO_IPV6)
168 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
169 		else
170 			ret = RTE_FLOW_ITEM_TYPE_END;
171 		break;
172 	default:
173 		ret = RTE_FLOW_ITEM_TYPE_VOID;
174 		break;
175 	}
176 	return ret;
177 }
178 
179 /**
180  * Expand RSS flows into several possible flows according to the RSS hash
181  * fields requested and the driver capabilities.
182  *
183  * @param[out] buf
184  *   Buffer to store the result expansion.
185  * @param[in] size
186  *   Buffer size in bytes. If 0, @p buf can be NULL.
187  * @param[in] pattern
188  *   User flow pattern.
189  * @param[in] types
190  *   RSS types to expand (see ETH_RSS_* definitions).
191  * @param[in] graph
192  *   Input graph to expand @p pattern according to @p types.
193  * @param[in] graph_root_index
194  *   Index of root node in @p graph, typically 0.
195  *
196  * @return
197  *   A positive value representing the size of @p buf in bytes regardless of
198  *   @p size on success, a negative errno value otherwise and rte_errno is
199  *   set, the following errors are defined:
200  *
201  *   -E2BIG: graph-depth @p graph is too deep.
202  */
203 static int
204 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
205 		     const struct rte_flow_item *pattern, uint64_t types,
206 		     const struct mlx5_flow_expand_node graph[],
207 		     int graph_root_index)
208 {
209 	const int elt_n = 8;
210 	const struct rte_flow_item *item;
211 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
212 	const int *next_node;
213 	const int *stack[elt_n];
214 	int stack_pos = 0;
215 	struct rte_flow_item flow_items[elt_n];
216 	unsigned int i;
217 	size_t lsize;
218 	size_t user_pattern_size = 0;
219 	void *addr = NULL;
220 	const struct mlx5_flow_expand_node *next = NULL;
221 	struct rte_flow_item missed_item;
222 	int missed = 0;
223 	int elt = 0;
224 	const struct rte_flow_item *last_item = NULL;
225 
226 	memset(&missed_item, 0, sizeof(missed_item));
227 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
228 		elt_n * sizeof(buf->entry[0]);
229 	if (lsize <= size) {
230 		buf->entry[0].priority = 0;
231 		buf->entry[0].pattern = (void *)&buf->entry[elt_n];
232 		buf->entries = 0;
233 		addr = buf->entry[0].pattern;
234 	}
235 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
236 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
237 			last_item = item;
238 		for (i = 0; node->next && node->next[i]; ++i) {
239 			next = &graph[node->next[i]];
240 			if (next->type == item->type)
241 				break;
242 		}
243 		if (next)
244 			node = next;
245 		user_pattern_size += sizeof(*item);
246 	}
247 	user_pattern_size += sizeof(*item); /* Handle END item. */
248 	lsize += user_pattern_size;
249 	/* Copy the user pattern in the first entry of the buffer. */
250 	if (lsize <= size) {
251 		rte_memcpy(addr, pattern, user_pattern_size);
252 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
253 		buf->entries = 1;
254 	}
255 	/* Start expanding. */
256 	memset(flow_items, 0, sizeof(flow_items));
257 	user_pattern_size -= sizeof(*item);
258 	/*
259 	 * Check if the last valid item has spec set, need complete pattern,
260 	 * and the pattern can be used for expansion.
261 	 */
262 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
263 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
264 		/* Item type END indicates expansion is not required. */
265 		return lsize;
266 	}
267 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
268 		next = NULL;
269 		missed = 1;
270 		for (i = 0; node->next && node->next[i]; ++i) {
271 			next = &graph[node->next[i]];
272 			if (next->type == missed_item.type) {
273 				flow_items[0].type = missed_item.type;
274 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
275 				break;
276 			}
277 			next = NULL;
278 		}
279 	}
280 	if (next && missed) {
281 		elt = 2; /* missed item + item end. */
282 		node = next;
283 		lsize += elt * sizeof(*item) + user_pattern_size;
284 		if ((node->rss_types & types) && lsize <= size) {
285 			buf->entry[buf->entries].priority = 1;
286 			buf->entry[buf->entries].pattern = addr;
287 			buf->entries++;
288 			rte_memcpy(addr, buf->entry[0].pattern,
289 				   user_pattern_size);
290 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
291 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
292 			addr = (void *)(((uintptr_t)addr) +
293 					elt * sizeof(*item));
294 		}
295 	}
296 	memset(flow_items, 0, sizeof(flow_items));
297 	next_node = node->next;
298 	stack[stack_pos] = next_node;
299 	node = next_node ? &graph[*next_node] : NULL;
300 	while (node) {
301 		flow_items[stack_pos].type = node->type;
302 		if (node->rss_types & types) {
303 			/*
304 			 * compute the number of items to copy from the
305 			 * expansion and copy it.
306 			 * When the stack_pos is 0, there are 1 element in it,
307 			 * plus the addition END item.
308 			 */
309 			elt = stack_pos + 2;
310 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
311 			lsize += elt * sizeof(*item) + user_pattern_size;
312 			if (lsize <= size) {
313 				size_t n = elt * sizeof(*item);
314 
315 				buf->entry[buf->entries].priority =
316 					stack_pos + 1 + missed;
317 				buf->entry[buf->entries].pattern = addr;
318 				buf->entries++;
319 				rte_memcpy(addr, buf->entry[0].pattern,
320 					   user_pattern_size);
321 				addr = (void *)(((uintptr_t)addr) +
322 						user_pattern_size);
323 				rte_memcpy(addr, &missed_item,
324 					   missed * sizeof(*item));
325 				addr = (void *)(((uintptr_t)addr) +
326 					missed * sizeof(*item));
327 				rte_memcpy(addr, flow_items, n);
328 				addr = (void *)(((uintptr_t)addr) + n);
329 			}
330 		}
331 		/* Go deeper. */
332 		if (node->next) {
333 			next_node = node->next;
334 			if (stack_pos++ == elt_n) {
335 				rte_errno = E2BIG;
336 				return -rte_errno;
337 			}
338 			stack[stack_pos] = next_node;
339 		} else if (*(next_node + 1)) {
340 			/* Follow up with the next possibility. */
341 			++next_node;
342 		} else {
343 			/* Move to the next path. */
344 			if (stack_pos)
345 				next_node = stack[--stack_pos];
346 			next_node++;
347 			stack[stack_pos] = next_node;
348 		}
349 		node = *next_node ? &graph[*next_node] : NULL;
350 	};
351 	/* no expanded flows but we have missed item, create one rule for it */
352 	if (buf->entries == 1 && missed != 0) {
353 		elt = 2;
354 		lsize += elt * sizeof(*item) + user_pattern_size;
355 		if (lsize <= size) {
356 			buf->entry[buf->entries].priority = 1;
357 			buf->entry[buf->entries].pattern = addr;
358 			buf->entries++;
359 			flow_items[0].type = missed_item.type;
360 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
361 			rte_memcpy(addr, buf->entry[0].pattern,
362 				   user_pattern_size);
363 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
364 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
365 			addr = (void *)(((uintptr_t)addr) +
366 					elt * sizeof(*item));
367 		}
368 	}
369 	return lsize;
370 }
371 
372 enum mlx5_expansion {
373 	MLX5_EXPANSION_ROOT,
374 	MLX5_EXPANSION_ROOT_OUTER,
375 	MLX5_EXPANSION_ROOT_ETH_VLAN,
376 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
377 	MLX5_EXPANSION_OUTER_ETH,
378 	MLX5_EXPANSION_OUTER_ETH_VLAN,
379 	MLX5_EXPANSION_OUTER_VLAN,
380 	MLX5_EXPANSION_OUTER_IPV4,
381 	MLX5_EXPANSION_OUTER_IPV4_UDP,
382 	MLX5_EXPANSION_OUTER_IPV4_TCP,
383 	MLX5_EXPANSION_OUTER_IPV6,
384 	MLX5_EXPANSION_OUTER_IPV6_UDP,
385 	MLX5_EXPANSION_OUTER_IPV6_TCP,
386 	MLX5_EXPANSION_VXLAN,
387 	MLX5_EXPANSION_VXLAN_GPE,
388 	MLX5_EXPANSION_GRE,
389 	MLX5_EXPANSION_MPLS,
390 	MLX5_EXPANSION_ETH,
391 	MLX5_EXPANSION_ETH_VLAN,
392 	MLX5_EXPANSION_VLAN,
393 	MLX5_EXPANSION_IPV4,
394 	MLX5_EXPANSION_IPV4_UDP,
395 	MLX5_EXPANSION_IPV4_TCP,
396 	MLX5_EXPANSION_IPV6,
397 	MLX5_EXPANSION_IPV6_UDP,
398 	MLX5_EXPANSION_IPV6_TCP,
399 };
400 
401 /** Supported expansion of items. */
402 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
403 	[MLX5_EXPANSION_ROOT] = {
404 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
405 						  MLX5_EXPANSION_IPV4,
406 						  MLX5_EXPANSION_IPV6),
407 		.type = RTE_FLOW_ITEM_TYPE_END,
408 	},
409 	[MLX5_EXPANSION_ROOT_OUTER] = {
410 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
411 						  MLX5_EXPANSION_OUTER_IPV4,
412 						  MLX5_EXPANSION_OUTER_IPV6),
413 		.type = RTE_FLOW_ITEM_TYPE_END,
414 	},
415 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
416 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
417 		.type = RTE_FLOW_ITEM_TYPE_END,
418 	},
419 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
420 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
421 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
422 		.type = RTE_FLOW_ITEM_TYPE_END,
423 	},
424 	[MLX5_EXPANSION_OUTER_ETH] = {
425 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
426 						  MLX5_EXPANSION_OUTER_IPV6,
427 						  MLX5_EXPANSION_MPLS),
428 		.type = RTE_FLOW_ITEM_TYPE_ETH,
429 		.rss_types = 0,
430 	},
431 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
432 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
433 		.type = RTE_FLOW_ITEM_TYPE_ETH,
434 		.rss_types = 0,
435 	},
436 	[MLX5_EXPANSION_OUTER_VLAN] = {
437 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
438 						  MLX5_EXPANSION_OUTER_IPV6),
439 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
440 	},
441 	[MLX5_EXPANSION_OUTER_IPV4] = {
442 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
443 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
444 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
445 			 MLX5_EXPANSION_GRE,
446 			 MLX5_EXPANSION_IPV4,
447 			 MLX5_EXPANSION_IPV6),
448 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
449 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
450 			ETH_RSS_NONFRAG_IPV4_OTHER,
451 	},
452 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
453 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
454 						  MLX5_EXPANSION_VXLAN_GPE),
455 		.type = RTE_FLOW_ITEM_TYPE_UDP,
456 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
457 	},
458 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
459 		.type = RTE_FLOW_ITEM_TYPE_TCP,
460 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
461 	},
462 	[MLX5_EXPANSION_OUTER_IPV6] = {
463 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
464 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
465 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
466 			 MLX5_EXPANSION_IPV4,
467 			 MLX5_EXPANSION_IPV6),
468 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
469 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
470 			ETH_RSS_NONFRAG_IPV6_OTHER,
471 	},
472 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
473 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
474 						  MLX5_EXPANSION_VXLAN_GPE),
475 		.type = RTE_FLOW_ITEM_TYPE_UDP,
476 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
477 	},
478 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
479 		.type = RTE_FLOW_ITEM_TYPE_TCP,
480 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
481 	},
482 	[MLX5_EXPANSION_VXLAN] = {
483 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
484 						  MLX5_EXPANSION_IPV4,
485 						  MLX5_EXPANSION_IPV6),
486 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
487 	},
488 	[MLX5_EXPANSION_VXLAN_GPE] = {
489 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
490 						  MLX5_EXPANSION_IPV4,
491 						  MLX5_EXPANSION_IPV6),
492 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
493 	},
494 	[MLX5_EXPANSION_GRE] = {
495 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
496 		.type = RTE_FLOW_ITEM_TYPE_GRE,
497 	},
498 	[MLX5_EXPANSION_MPLS] = {
499 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
500 						  MLX5_EXPANSION_IPV6),
501 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
502 	},
503 	[MLX5_EXPANSION_ETH] = {
504 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
505 						  MLX5_EXPANSION_IPV6),
506 		.type = RTE_FLOW_ITEM_TYPE_ETH,
507 	},
508 	[MLX5_EXPANSION_ETH_VLAN] = {
509 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
510 		.type = RTE_FLOW_ITEM_TYPE_ETH,
511 	},
512 	[MLX5_EXPANSION_VLAN] = {
513 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
514 						  MLX5_EXPANSION_IPV6),
515 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
516 	},
517 	[MLX5_EXPANSION_IPV4] = {
518 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
519 						  MLX5_EXPANSION_IPV4_TCP),
520 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
521 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
522 			ETH_RSS_NONFRAG_IPV4_OTHER,
523 	},
524 	[MLX5_EXPANSION_IPV4_UDP] = {
525 		.type = RTE_FLOW_ITEM_TYPE_UDP,
526 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
527 	},
528 	[MLX5_EXPANSION_IPV4_TCP] = {
529 		.type = RTE_FLOW_ITEM_TYPE_TCP,
530 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
531 	},
532 	[MLX5_EXPANSION_IPV6] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
534 						  MLX5_EXPANSION_IPV6_TCP),
535 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
536 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
537 			ETH_RSS_NONFRAG_IPV6_OTHER,
538 	},
539 	[MLX5_EXPANSION_IPV6_UDP] = {
540 		.type = RTE_FLOW_ITEM_TYPE_UDP,
541 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
542 	},
543 	[MLX5_EXPANSION_IPV6_TCP] = {
544 		.type = RTE_FLOW_ITEM_TYPE_TCP,
545 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
546 	},
547 };
548 
549 static const struct rte_flow_ops mlx5_flow_ops = {
550 	.validate = mlx5_flow_validate,
551 	.create = mlx5_flow_create,
552 	.destroy = mlx5_flow_destroy,
553 	.flush = mlx5_flow_flush,
554 	.isolate = mlx5_flow_isolate,
555 	.query = mlx5_flow_query,
556 	.dev_dump = mlx5_flow_dev_dump,
557 	.get_aged_flows = mlx5_flow_get_aged_flows,
558 };
559 
560 /* Convert FDIR request to Generic flow. */
561 struct mlx5_fdir {
562 	struct rte_flow_attr attr;
563 	struct rte_flow_item items[4];
564 	struct rte_flow_item_eth l2;
565 	struct rte_flow_item_eth l2_mask;
566 	union {
567 		struct rte_flow_item_ipv4 ipv4;
568 		struct rte_flow_item_ipv6 ipv6;
569 	} l3;
570 	union {
571 		struct rte_flow_item_ipv4 ipv4;
572 		struct rte_flow_item_ipv6 ipv6;
573 	} l3_mask;
574 	union {
575 		struct rte_flow_item_udp udp;
576 		struct rte_flow_item_tcp tcp;
577 	} l4;
578 	union {
579 		struct rte_flow_item_udp udp;
580 		struct rte_flow_item_tcp tcp;
581 	} l4_mask;
582 	struct rte_flow_action actions[2];
583 	struct rte_flow_action_queue queue;
584 };
585 
586 /* Tunnel information. */
587 struct mlx5_flow_tunnel_info {
588 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
589 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
590 };
591 
592 static struct mlx5_flow_tunnel_info tunnels_info[] = {
593 	{
594 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
595 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
596 	},
597 	{
598 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
599 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
600 	},
601 	{
602 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
603 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
604 	},
605 	{
606 		.tunnel = MLX5_FLOW_LAYER_GRE,
607 		.ptype = RTE_PTYPE_TUNNEL_GRE,
608 	},
609 	{
610 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
611 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
612 	},
613 	{
614 		.tunnel = MLX5_FLOW_LAYER_MPLS,
615 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
616 	},
617 	{
618 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
619 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
620 	},
621 	{
622 		.tunnel = MLX5_FLOW_LAYER_IPIP,
623 		.ptype = RTE_PTYPE_TUNNEL_IP,
624 	},
625 	{
626 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
627 		.ptype = RTE_PTYPE_TUNNEL_IP,
628 	},
629 	{
630 		.tunnel = MLX5_FLOW_LAYER_GTP,
631 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
632 	},
633 };
634 
635 /**
636  * Translate tag ID to register.
637  *
638  * @param[in] dev
639  *   Pointer to the Ethernet device structure.
640  * @param[in] feature
641  *   The feature that request the register.
642  * @param[in] id
643  *   The request register ID.
644  * @param[out] error
645  *   Error description in case of any.
646  *
647  * @return
648  *   The request register on success, a negative errno
649  *   value otherwise and rte_errno is set.
650  */
651 int
652 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
653 		     enum mlx5_feature_name feature,
654 		     uint32_t id,
655 		     struct rte_flow_error *error)
656 {
657 	struct mlx5_priv *priv = dev->data->dev_private;
658 	struct mlx5_dev_config *config = &priv->config;
659 	enum modify_reg start_reg;
660 	bool skip_mtr_reg = false;
661 
662 	switch (feature) {
663 	case MLX5_HAIRPIN_RX:
664 		return REG_B;
665 	case MLX5_HAIRPIN_TX:
666 		return REG_A;
667 	case MLX5_METADATA_RX:
668 		switch (config->dv_xmeta_en) {
669 		case MLX5_XMETA_MODE_LEGACY:
670 			return REG_B;
671 		case MLX5_XMETA_MODE_META16:
672 			return REG_C_0;
673 		case MLX5_XMETA_MODE_META32:
674 			return REG_C_1;
675 		}
676 		break;
677 	case MLX5_METADATA_TX:
678 		return REG_A;
679 	case MLX5_METADATA_FDB:
680 		switch (config->dv_xmeta_en) {
681 		case MLX5_XMETA_MODE_LEGACY:
682 			return REG_NON;
683 		case MLX5_XMETA_MODE_META16:
684 			return REG_C_0;
685 		case MLX5_XMETA_MODE_META32:
686 			return REG_C_1;
687 		}
688 		break;
689 	case MLX5_FLOW_MARK:
690 		switch (config->dv_xmeta_en) {
691 		case MLX5_XMETA_MODE_LEGACY:
692 			return REG_NON;
693 		case MLX5_XMETA_MODE_META16:
694 			return REG_C_1;
695 		case MLX5_XMETA_MODE_META32:
696 			return REG_C_0;
697 		}
698 		break;
699 	case MLX5_MTR_SFX:
700 		/*
701 		 * If meter color and flow match share one register, flow match
702 		 * should use the meter color register for match.
703 		 */
704 		if (priv->mtr_reg_share)
705 			return priv->mtr_color_reg;
706 		else
707 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
708 			       REG_C_3;
709 	case MLX5_MTR_COLOR:
710 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
711 		return priv->mtr_color_reg;
712 	case MLX5_COPY_MARK:
713 		/*
714 		 * Metadata COPY_MARK register using is in meter suffix sub
715 		 * flow while with meter. It's safe to share the same register.
716 		 */
717 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
718 	case MLX5_APP_TAG:
719 		/*
720 		 * If meter is enable, it will engage the register for color
721 		 * match and flow match. If meter color match is not using the
722 		 * REG_C_2, need to skip the REG_C_x be used by meter color
723 		 * match.
724 		 * If meter is disable, free to use all available registers.
725 		 */
726 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
727 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
728 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
729 		if (id > (REG_C_7 - start_reg))
730 			return rte_flow_error_set(error, EINVAL,
731 						  RTE_FLOW_ERROR_TYPE_ITEM,
732 						  NULL, "invalid tag id");
733 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
734 			return rte_flow_error_set(error, ENOTSUP,
735 						  RTE_FLOW_ERROR_TYPE_ITEM,
736 						  NULL, "unsupported tag id");
737 		/*
738 		 * This case means meter is using the REG_C_x great than 2.
739 		 * Take care not to conflict with meter color REG_C_x.
740 		 * If the available index REG_C_y >= REG_C_x, skip the
741 		 * color register.
742 		 */
743 		if (skip_mtr_reg && config->flow_mreg_c
744 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
745 			if (id >= (REG_C_7 - start_reg))
746 				return rte_flow_error_set(error, EINVAL,
747 						       RTE_FLOW_ERROR_TYPE_ITEM,
748 							NULL, "invalid tag id");
749 			if (config->flow_mreg_c
750 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
751 				return config->flow_mreg_c
752 					       [id + 1 + start_reg - REG_C_0];
753 			return rte_flow_error_set(error, ENOTSUP,
754 						  RTE_FLOW_ERROR_TYPE_ITEM,
755 						  NULL, "unsupported tag id");
756 		}
757 		return config->flow_mreg_c[id + start_reg - REG_C_0];
758 	}
759 	MLX5_ASSERT(false);
760 	return rte_flow_error_set(error, EINVAL,
761 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
762 				  NULL, "invalid feature name");
763 }
764 
765 /**
766  * Check extensive flow metadata register support.
767  *
768  * @param dev
769  *   Pointer to rte_eth_dev structure.
770  *
771  * @return
772  *   True if device supports extensive flow metadata register, otherwise false.
773  */
774 bool
775 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
776 {
777 	struct mlx5_priv *priv = dev->data->dev_private;
778 	struct mlx5_dev_config *config = &priv->config;
779 
780 	/*
781 	 * Having available reg_c can be regarded inclusively as supporting
782 	 * extensive flow metadata register, which could mean,
783 	 * - metadata register copy action by modify header.
784 	 * - 16 modify header actions is supported.
785 	 * - reg_c's are preserved across different domain (FDB and NIC) on
786 	 *   packet loopback by flow lookup miss.
787 	 */
788 	return config->flow_mreg_c[2] != REG_NON;
789 }
790 
791 /**
792  * Verify the @p item specifications (spec, last, mask) are compatible with the
793  * NIC capabilities.
794  *
795  * @param[in] item
796  *   Item specification.
797  * @param[in] mask
798  *   @p item->mask or flow default bit-masks.
799  * @param[in] nic_mask
800  *   Bit-masks covering supported fields by the NIC to compare with user mask.
801  * @param[in] size
802  *   Bit-masks size in bytes.
803  * @param[in] range_accepted
804  *   True if range of values is accepted for specific fields, false otherwise.
805  * @param[out] error
806  *   Pointer to error structure.
807  *
808  * @return
809  *   0 on success, a negative errno value otherwise and rte_errno is set.
810  */
811 int
812 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
813 			  const uint8_t *mask,
814 			  const uint8_t *nic_mask,
815 			  unsigned int size,
816 			  bool range_accepted,
817 			  struct rte_flow_error *error)
818 {
819 	unsigned int i;
820 
821 	MLX5_ASSERT(nic_mask);
822 	for (i = 0; i < size; ++i)
823 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
824 			return rte_flow_error_set(error, ENOTSUP,
825 						  RTE_FLOW_ERROR_TYPE_ITEM,
826 						  item,
827 						  "mask enables non supported"
828 						  " bits");
829 	if (!item->spec && (item->mask || item->last))
830 		return rte_flow_error_set(error, EINVAL,
831 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
832 					  "mask/last without a spec is not"
833 					  " supported");
834 	if (item->spec && item->last && !range_accepted) {
835 		uint8_t spec[size];
836 		uint8_t last[size];
837 		unsigned int i;
838 		int ret;
839 
840 		for (i = 0; i < size; ++i) {
841 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
842 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
843 		}
844 		ret = memcmp(spec, last, size);
845 		if (ret != 0)
846 			return rte_flow_error_set(error, EINVAL,
847 						  RTE_FLOW_ERROR_TYPE_ITEM,
848 						  item,
849 						  "range is not valid");
850 	}
851 	return 0;
852 }
853 
854 /**
855  * Adjust the hash fields according to the @p flow information.
856  *
857  * @param[in] dev_flow.
858  *   Pointer to the mlx5_flow.
859  * @param[in] tunnel
860  *   1 when the hash field is for a tunnel item.
861  * @param[in] layer_types
862  *   ETH_RSS_* types.
863  * @param[in] hash_fields
864  *   Item hash fields.
865  *
866  * @return
867  *   The hash fields that should be used.
868  */
869 uint64_t
870 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
871 			    int tunnel __rte_unused, uint64_t layer_types,
872 			    uint64_t hash_fields)
873 {
874 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
875 	int rss_request_inner = rss_desc->level >= 2;
876 
877 	/* Check RSS hash level for tunnel. */
878 	if (tunnel && rss_request_inner)
879 		hash_fields |= IBV_RX_HASH_INNER;
880 	else if (tunnel || rss_request_inner)
881 		return 0;
882 #endif
883 	/* Check if requested layer matches RSS hash fields. */
884 	if (!(rss_desc->types & layer_types))
885 		return 0;
886 	return hash_fields;
887 }
888 
889 /**
890  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
891  * if several tunnel rules are used on this queue, the tunnel ptype will be
892  * cleared.
893  *
894  * @param rxq_ctrl
895  *   Rx queue to update.
896  */
897 static void
898 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
899 {
900 	unsigned int i;
901 	uint32_t tunnel_ptype = 0;
902 
903 	/* Look up for the ptype to use. */
904 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
905 		if (!rxq_ctrl->flow_tunnels_n[i])
906 			continue;
907 		if (!tunnel_ptype) {
908 			tunnel_ptype = tunnels_info[i].ptype;
909 		} else {
910 			tunnel_ptype = 0;
911 			break;
912 		}
913 	}
914 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
915 }
916 
917 /**
918  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
919  * flow.
920  *
921  * @param[in] dev
922  *   Pointer to the Ethernet device structure.
923  * @param[in] dev_handle
924  *   Pointer to device flow handle structure.
925  */
926 static void
927 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
928 		       struct mlx5_flow_handle *dev_handle)
929 {
930 	struct mlx5_priv *priv = dev->data->dev_private;
931 	const int mark = dev_handle->mark;
932 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
933 	struct mlx5_hrxq *hrxq;
934 	unsigned int i;
935 
936 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
937 		return;
938 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
939 			      dev_handle->rix_hrxq);
940 	if (!hrxq)
941 		return;
942 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
943 		int idx = hrxq->ind_table->queues[i];
944 		struct mlx5_rxq_ctrl *rxq_ctrl =
945 			container_of((*priv->rxqs)[idx],
946 				     struct mlx5_rxq_ctrl, rxq);
947 
948 		/*
949 		 * To support metadata register copy on Tx loopback,
950 		 * this must be always enabled (metadata may arive
951 		 * from other port - not from local flows only.
952 		 */
953 		if (priv->config.dv_flow_en &&
954 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
955 		    mlx5_flow_ext_mreg_supported(dev)) {
956 			rxq_ctrl->rxq.mark = 1;
957 			rxq_ctrl->flow_mark_n = 1;
958 		} else if (mark) {
959 			rxq_ctrl->rxq.mark = 1;
960 			rxq_ctrl->flow_mark_n++;
961 		}
962 		if (tunnel) {
963 			unsigned int j;
964 
965 			/* Increase the counter matching the flow. */
966 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
967 				if ((tunnels_info[j].tunnel &
968 				     dev_handle->layers) ==
969 				    tunnels_info[j].tunnel) {
970 					rxq_ctrl->flow_tunnels_n[j]++;
971 					break;
972 				}
973 			}
974 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
975 		}
976 	}
977 }
978 
979 /**
980  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
981  *
982  * @param[in] dev
983  *   Pointer to the Ethernet device structure.
984  * @param[in] flow
985  *   Pointer to flow structure.
986  */
987 static void
988 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
989 {
990 	struct mlx5_priv *priv = dev->data->dev_private;
991 	uint32_t handle_idx;
992 	struct mlx5_flow_handle *dev_handle;
993 
994 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
995 		       handle_idx, dev_handle, next)
996 		flow_drv_rxq_flags_set(dev, dev_handle);
997 }
998 
999 /**
1000  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1001  * device flow if no other flow uses it with the same kind of request.
1002  *
1003  * @param dev
1004  *   Pointer to Ethernet device.
1005  * @param[in] dev_handle
1006  *   Pointer to the device flow handle structure.
1007  */
1008 static void
1009 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1010 			struct mlx5_flow_handle *dev_handle)
1011 {
1012 	struct mlx5_priv *priv = dev->data->dev_private;
1013 	const int mark = dev_handle->mark;
1014 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1015 	struct mlx5_hrxq *hrxq;
1016 	unsigned int i;
1017 
1018 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1019 		return;
1020 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1021 			      dev_handle->rix_hrxq);
1022 	if (!hrxq)
1023 		return;
1024 	MLX5_ASSERT(dev->data->dev_started);
1025 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1026 		int idx = hrxq->ind_table->queues[i];
1027 		struct mlx5_rxq_ctrl *rxq_ctrl =
1028 			container_of((*priv->rxqs)[idx],
1029 				     struct mlx5_rxq_ctrl, rxq);
1030 
1031 		if (priv->config.dv_flow_en &&
1032 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1033 		    mlx5_flow_ext_mreg_supported(dev)) {
1034 			rxq_ctrl->rxq.mark = 1;
1035 			rxq_ctrl->flow_mark_n = 1;
1036 		} else if (mark) {
1037 			rxq_ctrl->flow_mark_n--;
1038 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1039 		}
1040 		if (tunnel) {
1041 			unsigned int j;
1042 
1043 			/* Decrease the counter matching the flow. */
1044 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1045 				if ((tunnels_info[j].tunnel &
1046 				     dev_handle->layers) ==
1047 				    tunnels_info[j].tunnel) {
1048 					rxq_ctrl->flow_tunnels_n[j]--;
1049 					break;
1050 				}
1051 			}
1052 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1053 		}
1054 	}
1055 }
1056 
1057 /**
1058  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1059  * @p flow if no other flow uses it with the same kind of request.
1060  *
1061  * @param dev
1062  *   Pointer to Ethernet device.
1063  * @param[in] flow
1064  *   Pointer to the flow.
1065  */
1066 static void
1067 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1068 {
1069 	struct mlx5_priv *priv = dev->data->dev_private;
1070 	uint32_t handle_idx;
1071 	struct mlx5_flow_handle *dev_handle;
1072 
1073 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1074 		       handle_idx, dev_handle, next)
1075 		flow_drv_rxq_flags_trim(dev, dev_handle);
1076 }
1077 
1078 /**
1079  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1080  *
1081  * @param dev
1082  *   Pointer to Ethernet device.
1083  */
1084 static void
1085 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1086 {
1087 	struct mlx5_priv *priv = dev->data->dev_private;
1088 	unsigned int i;
1089 
1090 	for (i = 0; i != priv->rxqs_n; ++i) {
1091 		struct mlx5_rxq_ctrl *rxq_ctrl;
1092 		unsigned int j;
1093 
1094 		if (!(*priv->rxqs)[i])
1095 			continue;
1096 		rxq_ctrl = container_of((*priv->rxqs)[i],
1097 					struct mlx5_rxq_ctrl, rxq);
1098 		rxq_ctrl->flow_mark_n = 0;
1099 		rxq_ctrl->rxq.mark = 0;
1100 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1101 			rxq_ctrl->flow_tunnels_n[j] = 0;
1102 		rxq_ctrl->rxq.tunnel = 0;
1103 	}
1104 }
1105 
1106 /**
1107  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1108  *
1109  * @param[in] dev
1110  *   Pointer to the Ethernet device structure.
1111  */
1112 void
1113 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1114 {
1115 	struct mlx5_priv *priv = dev->data->dev_private;
1116 	struct mlx5_rxq_data *data;
1117 	unsigned int i;
1118 
1119 	for (i = 0; i != priv->rxqs_n; ++i) {
1120 		if (!(*priv->rxqs)[i])
1121 			continue;
1122 		data = (*priv->rxqs)[i];
1123 		if (!rte_flow_dynf_metadata_avail()) {
1124 			data->dynf_meta = 0;
1125 			data->flow_meta_mask = 0;
1126 			data->flow_meta_offset = -1;
1127 		} else {
1128 			data->dynf_meta = 1;
1129 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1130 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1131 		}
1132 	}
1133 }
1134 
1135 /*
1136  * return a pointer to the desired action in the list of actions.
1137  *
1138  * @param[in] actions
1139  *   The list of actions to search the action in.
1140  * @param[in] action
1141  *   The action to find.
1142  *
1143  * @return
1144  *   Pointer to the action in the list, if found. NULL otherwise.
1145  */
1146 const struct rte_flow_action *
1147 mlx5_flow_find_action(const struct rte_flow_action *actions,
1148 		      enum rte_flow_action_type action)
1149 {
1150 	if (actions == NULL)
1151 		return NULL;
1152 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1153 		if (actions->type == action)
1154 			return actions;
1155 	return NULL;
1156 }
1157 
1158 /*
1159  * Validate the flag action.
1160  *
1161  * @param[in] action_flags
1162  *   Bit-fields that holds the actions detected until now.
1163  * @param[in] attr
1164  *   Attributes of flow that includes this action.
1165  * @param[out] error
1166  *   Pointer to error structure.
1167  *
1168  * @return
1169  *   0 on success, a negative errno value otherwise and rte_errno is set.
1170  */
1171 int
1172 mlx5_flow_validate_action_flag(uint64_t action_flags,
1173 			       const struct rte_flow_attr *attr,
1174 			       struct rte_flow_error *error)
1175 {
1176 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1177 		return rte_flow_error_set(error, EINVAL,
1178 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1179 					  "can't mark and flag in same flow");
1180 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1181 		return rte_flow_error_set(error, EINVAL,
1182 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1183 					  "can't have 2 flag"
1184 					  " actions in same flow");
1185 	if (attr->egress)
1186 		return rte_flow_error_set(error, ENOTSUP,
1187 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1188 					  "flag action not supported for "
1189 					  "egress");
1190 	return 0;
1191 }
1192 
1193 /*
1194  * Validate the mark action.
1195  *
1196  * @param[in] action
1197  *   Pointer to the queue action.
1198  * @param[in] action_flags
1199  *   Bit-fields that holds the actions detected until now.
1200  * @param[in] attr
1201  *   Attributes of flow that includes this action.
1202  * @param[out] error
1203  *   Pointer to error structure.
1204  *
1205  * @return
1206  *   0 on success, a negative errno value otherwise and rte_errno is set.
1207  */
1208 int
1209 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1210 			       uint64_t action_flags,
1211 			       const struct rte_flow_attr *attr,
1212 			       struct rte_flow_error *error)
1213 {
1214 	const struct rte_flow_action_mark *mark = action->conf;
1215 
1216 	if (!mark)
1217 		return rte_flow_error_set(error, EINVAL,
1218 					  RTE_FLOW_ERROR_TYPE_ACTION,
1219 					  action,
1220 					  "configuration cannot be null");
1221 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1222 		return rte_flow_error_set(error, EINVAL,
1223 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1224 					  &mark->id,
1225 					  "mark id must in 0 <= id < "
1226 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1227 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1228 		return rte_flow_error_set(error, EINVAL,
1229 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1230 					  "can't flag and mark in same flow");
1231 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1232 		return rte_flow_error_set(error, EINVAL,
1233 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1234 					  "can't have 2 mark actions in same"
1235 					  " flow");
1236 	if (attr->egress)
1237 		return rte_flow_error_set(error, ENOTSUP,
1238 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1239 					  "mark action not supported for "
1240 					  "egress");
1241 	return 0;
1242 }
1243 
1244 /*
1245  * Validate the drop action.
1246  *
1247  * @param[in] action_flags
1248  *   Bit-fields that holds the actions detected until now.
1249  * @param[in] attr
1250  *   Attributes of flow that includes this action.
1251  * @param[out] error
1252  *   Pointer to error structure.
1253  *
1254  * @return
1255  *   0 on success, a negative errno value otherwise and rte_errno is set.
1256  */
1257 int
1258 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1259 			       const struct rte_flow_attr *attr,
1260 			       struct rte_flow_error *error)
1261 {
1262 	if (attr->egress)
1263 		return rte_flow_error_set(error, ENOTSUP,
1264 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1265 					  "drop action not supported for "
1266 					  "egress");
1267 	return 0;
1268 }
1269 
1270 /*
1271  * Validate the queue action.
1272  *
1273  * @param[in] action
1274  *   Pointer to the queue action.
1275  * @param[in] action_flags
1276  *   Bit-fields that holds the actions detected until now.
1277  * @param[in] dev
1278  *   Pointer to the Ethernet device structure.
1279  * @param[in] attr
1280  *   Attributes of flow that includes this action.
1281  * @param[out] error
1282  *   Pointer to error structure.
1283  *
1284  * @return
1285  *   0 on success, a negative errno value otherwise and rte_errno is set.
1286  */
1287 int
1288 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1289 				uint64_t action_flags,
1290 				struct rte_eth_dev *dev,
1291 				const struct rte_flow_attr *attr,
1292 				struct rte_flow_error *error)
1293 {
1294 	struct mlx5_priv *priv = dev->data->dev_private;
1295 	const struct rte_flow_action_queue *queue = action->conf;
1296 
1297 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1298 		return rte_flow_error_set(error, EINVAL,
1299 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1300 					  "can't have 2 fate actions in"
1301 					  " same flow");
1302 	if (!priv->rxqs_n)
1303 		return rte_flow_error_set(error, EINVAL,
1304 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1305 					  NULL, "No Rx queues configured");
1306 	if (queue->index >= priv->rxqs_n)
1307 		return rte_flow_error_set(error, EINVAL,
1308 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1309 					  &queue->index,
1310 					  "queue index out of range");
1311 	if (!(*priv->rxqs)[queue->index])
1312 		return rte_flow_error_set(error, EINVAL,
1313 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1314 					  &queue->index,
1315 					  "queue is not configured");
1316 	if (attr->egress)
1317 		return rte_flow_error_set(error, ENOTSUP,
1318 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1319 					  "queue action not supported for "
1320 					  "egress");
1321 	return 0;
1322 }
1323 
1324 /*
1325  * Validate the rss action.
1326  *
1327  * @param[in] action
1328  *   Pointer to the queue action.
1329  * @param[in] action_flags
1330  *   Bit-fields that holds the actions detected until now.
1331  * @param[in] dev
1332  *   Pointer to the Ethernet device structure.
1333  * @param[in] attr
1334  *   Attributes of flow that includes this action.
1335  * @param[in] item_flags
1336  *   Items that were detected.
1337  * @param[out] error
1338  *   Pointer to error structure.
1339  *
1340  * @return
1341  *   0 on success, a negative errno value otherwise and rte_errno is set.
1342  */
1343 int
1344 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1345 			      uint64_t action_flags,
1346 			      struct rte_eth_dev *dev,
1347 			      const struct rte_flow_attr *attr,
1348 			      uint64_t item_flags,
1349 			      struct rte_flow_error *error)
1350 {
1351 	struct mlx5_priv *priv = dev->data->dev_private;
1352 	const struct rte_flow_action_rss *rss = action->conf;
1353 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1354 	unsigned int i;
1355 
1356 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1357 		return rte_flow_error_set(error, EINVAL,
1358 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1359 					  "can't have 2 fate actions"
1360 					  " in same flow");
1361 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1362 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1363 		return rte_flow_error_set(error, ENOTSUP,
1364 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1365 					  &rss->func,
1366 					  "RSS hash function not supported");
1367 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1368 	if (rss->level > 2)
1369 #else
1370 	if (rss->level > 1)
1371 #endif
1372 		return rte_flow_error_set(error, ENOTSUP,
1373 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1374 					  &rss->level,
1375 					  "tunnel RSS is not supported");
1376 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1377 	if (rss->key_len == 0 && rss->key != NULL)
1378 		return rte_flow_error_set(error, ENOTSUP,
1379 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1380 					  &rss->key_len,
1381 					  "RSS hash key length 0");
1382 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1383 		return rte_flow_error_set(error, ENOTSUP,
1384 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1385 					  &rss->key_len,
1386 					  "RSS hash key too small");
1387 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1388 		return rte_flow_error_set(error, ENOTSUP,
1389 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1390 					  &rss->key_len,
1391 					  "RSS hash key too large");
1392 	if (rss->queue_num > priv->config.ind_table_max_size)
1393 		return rte_flow_error_set(error, ENOTSUP,
1394 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1395 					  &rss->queue_num,
1396 					  "number of queues too large");
1397 	if (rss->types & MLX5_RSS_HF_MASK)
1398 		return rte_flow_error_set(error, ENOTSUP,
1399 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1400 					  &rss->types,
1401 					  "some RSS protocols are not"
1402 					  " supported");
1403 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1404 	    !(rss->types & ETH_RSS_IP))
1405 		return rte_flow_error_set(error, EINVAL,
1406 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1407 					  "L3 partial RSS requested but L3 RSS"
1408 					  " type not specified");
1409 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1410 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1411 		return rte_flow_error_set(error, EINVAL,
1412 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1413 					  "L4 partial RSS requested but L4 RSS"
1414 					  " type not specified");
1415 	if (!priv->rxqs_n)
1416 		return rte_flow_error_set(error, EINVAL,
1417 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1418 					  NULL, "No Rx queues configured");
1419 	if (!rss->queue_num)
1420 		return rte_flow_error_set(error, EINVAL,
1421 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1422 					  NULL, "No queues configured");
1423 	for (i = 0; i != rss->queue_num; ++i) {
1424 		if (rss->queue[i] >= priv->rxqs_n)
1425 			return rte_flow_error_set
1426 				(error, EINVAL,
1427 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1428 				 &rss->queue[i], "queue index out of range");
1429 		if (!(*priv->rxqs)[rss->queue[i]])
1430 			return rte_flow_error_set
1431 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1432 				 &rss->queue[i], "queue is not configured");
1433 	}
1434 	if (attr->egress)
1435 		return rte_flow_error_set(error, ENOTSUP,
1436 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1437 					  "rss action not supported for "
1438 					  "egress");
1439 	if (rss->level > 1 && !tunnel)
1440 		return rte_flow_error_set(error, EINVAL,
1441 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1442 					  "inner RSS is not supported for "
1443 					  "non-tunnel flows");
1444 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1445 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1446 		return rte_flow_error_set(error, EINVAL,
1447 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1448 					  "RSS on eCPRI is not supported now");
1449 	}
1450 	return 0;
1451 }
1452 
1453 /*
1454  * Validate the default miss action.
1455  *
1456  * @param[in] action_flags
1457  *   Bit-fields that holds the actions detected until now.
1458  * @param[out] error
1459  *   Pointer to error structure.
1460  *
1461  * @return
1462  *   0 on success, a negative errno value otherwise and rte_errno is set.
1463  */
1464 int
1465 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1466 				const struct rte_flow_attr *attr,
1467 				struct rte_flow_error *error)
1468 {
1469 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1470 		return rte_flow_error_set(error, EINVAL,
1471 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1472 					  "can't have 2 fate actions in"
1473 					  " same flow");
1474 	if (attr->egress)
1475 		return rte_flow_error_set(error, ENOTSUP,
1476 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1477 					  "default miss action not supported "
1478 					  "for egress");
1479 	if (attr->group)
1480 		return rte_flow_error_set(error, ENOTSUP,
1481 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1482 					  "only group 0 is supported");
1483 	if (attr->transfer)
1484 		return rte_flow_error_set(error, ENOTSUP,
1485 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1486 					  NULL, "transfer is not supported");
1487 	return 0;
1488 }
1489 
1490 /*
1491  * Validate the count action.
1492  *
1493  * @param[in] dev
1494  *   Pointer to the Ethernet device structure.
1495  * @param[in] attr
1496  *   Attributes of flow that includes this action.
1497  * @param[out] error
1498  *   Pointer to error structure.
1499  *
1500  * @return
1501  *   0 on success, a negative errno value otherwise and rte_errno is set.
1502  */
1503 int
1504 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1505 				const struct rte_flow_attr *attr,
1506 				struct rte_flow_error *error)
1507 {
1508 	if (attr->egress)
1509 		return rte_flow_error_set(error, ENOTSUP,
1510 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1511 					  "count action not supported for "
1512 					  "egress");
1513 	return 0;
1514 }
1515 
1516 /**
1517  * Verify the @p attributes will be correctly understood by the NIC and store
1518  * them in the @p flow if everything is correct.
1519  *
1520  * @param[in] dev
1521  *   Pointer to the Ethernet device structure.
1522  * @param[in] attributes
1523  *   Pointer to flow attributes
1524  * @param[out] error
1525  *   Pointer to error structure.
1526  *
1527  * @return
1528  *   0 on success, a negative errno value otherwise and rte_errno is set.
1529  */
1530 int
1531 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1532 			      const struct rte_flow_attr *attributes,
1533 			      struct rte_flow_error *error)
1534 {
1535 	struct mlx5_priv *priv = dev->data->dev_private;
1536 	uint32_t priority_max = priv->config.flow_prio - 1;
1537 
1538 	if (attributes->group)
1539 		return rte_flow_error_set(error, ENOTSUP,
1540 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1541 					  NULL, "groups is not supported");
1542 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1543 	    attributes->priority >= priority_max)
1544 		return rte_flow_error_set(error, ENOTSUP,
1545 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1546 					  NULL, "priority out of range");
1547 	if (attributes->egress)
1548 		return rte_flow_error_set(error, ENOTSUP,
1549 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1550 					  "egress is not supported");
1551 	if (attributes->transfer && !priv->config.dv_esw_en)
1552 		return rte_flow_error_set(error, ENOTSUP,
1553 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1554 					  NULL, "transfer is not supported");
1555 	if (!attributes->ingress)
1556 		return rte_flow_error_set(error, EINVAL,
1557 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1558 					  NULL,
1559 					  "ingress attribute is mandatory");
1560 	return 0;
1561 }
1562 
1563 /**
1564  * Validate ICMP6 item.
1565  *
1566  * @param[in] item
1567  *   Item specification.
1568  * @param[in] item_flags
1569  *   Bit-fields that holds the items detected until now.
1570  * @param[out] error
1571  *   Pointer to error structure.
1572  *
1573  * @return
1574  *   0 on success, a negative errno value otherwise and rte_errno is set.
1575  */
1576 int
1577 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1578 			       uint64_t item_flags,
1579 			       uint8_t target_protocol,
1580 			       struct rte_flow_error *error)
1581 {
1582 	const struct rte_flow_item_icmp6 *mask = item->mask;
1583 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1584 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1585 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1586 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1587 				      MLX5_FLOW_LAYER_OUTER_L4;
1588 	int ret;
1589 
1590 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1591 		return rte_flow_error_set(error, EINVAL,
1592 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1593 					  "protocol filtering not compatible"
1594 					  " with ICMP6 layer");
1595 	if (!(item_flags & l3m))
1596 		return rte_flow_error_set(error, EINVAL,
1597 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1598 					  "IPv6 is mandatory to filter on"
1599 					  " ICMP6");
1600 	if (item_flags & l4m)
1601 		return rte_flow_error_set(error, EINVAL,
1602 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1603 					  "multiple L4 layers not supported");
1604 	if (!mask)
1605 		mask = &rte_flow_item_icmp6_mask;
1606 	ret = mlx5_flow_item_acceptable
1607 		(item, (const uint8_t *)mask,
1608 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1609 		 sizeof(struct rte_flow_item_icmp6),
1610 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1611 	if (ret < 0)
1612 		return ret;
1613 	return 0;
1614 }
1615 
1616 /**
1617  * Validate ICMP item.
1618  *
1619  * @param[in] item
1620  *   Item specification.
1621  * @param[in] item_flags
1622  *   Bit-fields that holds the items detected until now.
1623  * @param[out] error
1624  *   Pointer to error structure.
1625  *
1626  * @return
1627  *   0 on success, a negative errno value otherwise and rte_errno is set.
1628  */
1629 int
1630 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1631 			     uint64_t item_flags,
1632 			     uint8_t target_protocol,
1633 			     struct rte_flow_error *error)
1634 {
1635 	const struct rte_flow_item_icmp *mask = item->mask;
1636 	const struct rte_flow_item_icmp nic_mask = {
1637 		.hdr.icmp_type = 0xff,
1638 		.hdr.icmp_code = 0xff,
1639 		.hdr.icmp_ident = RTE_BE16(0xffff),
1640 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1641 	};
1642 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1643 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1644 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1645 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1646 				      MLX5_FLOW_LAYER_OUTER_L4;
1647 	int ret;
1648 
1649 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1650 		return rte_flow_error_set(error, EINVAL,
1651 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1652 					  "protocol filtering not compatible"
1653 					  " with ICMP layer");
1654 	if (!(item_flags & l3m))
1655 		return rte_flow_error_set(error, EINVAL,
1656 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1657 					  "IPv4 is mandatory to filter"
1658 					  " on ICMP");
1659 	if (item_flags & l4m)
1660 		return rte_flow_error_set(error, EINVAL,
1661 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1662 					  "multiple L4 layers not supported");
1663 	if (!mask)
1664 		mask = &nic_mask;
1665 	ret = mlx5_flow_item_acceptable
1666 		(item, (const uint8_t *)mask,
1667 		 (const uint8_t *)&nic_mask,
1668 		 sizeof(struct rte_flow_item_icmp),
1669 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1670 	if (ret < 0)
1671 		return ret;
1672 	return 0;
1673 }
1674 
1675 /**
1676  * Validate Ethernet item.
1677  *
1678  * @param[in] item
1679  *   Item specification.
1680  * @param[in] item_flags
1681  *   Bit-fields that holds the items detected until now.
1682  * @param[out] error
1683  *   Pointer to error structure.
1684  *
1685  * @return
1686  *   0 on success, a negative errno value otherwise and rte_errno is set.
1687  */
1688 int
1689 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1690 			    uint64_t item_flags,
1691 			    struct rte_flow_error *error)
1692 {
1693 	const struct rte_flow_item_eth *mask = item->mask;
1694 	const struct rte_flow_item_eth nic_mask = {
1695 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1696 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1697 		.type = RTE_BE16(0xffff),
1698 	};
1699 	int ret;
1700 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1701 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1702 				       MLX5_FLOW_LAYER_OUTER_L2;
1703 
1704 	if (item_flags & ethm)
1705 		return rte_flow_error_set(error, ENOTSUP,
1706 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 					  "multiple L2 layers not supported");
1708 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1709 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1710 		return rte_flow_error_set(error, EINVAL,
1711 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1712 					  "L2 layer should not follow "
1713 					  "L3 layers");
1714 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1715 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1716 		return rte_flow_error_set(error, EINVAL,
1717 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1718 					  "L2 layer should not follow VLAN");
1719 	if (!mask)
1720 		mask = &rte_flow_item_eth_mask;
1721 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1722 					(const uint8_t *)&nic_mask,
1723 					sizeof(struct rte_flow_item_eth),
1724 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1725 	return ret;
1726 }
1727 
1728 /**
1729  * Validate VLAN item.
1730  *
1731  * @param[in] item
1732  *   Item specification.
1733  * @param[in] item_flags
1734  *   Bit-fields that holds the items detected until now.
1735  * @param[in] dev
1736  *   Ethernet device flow is being created on.
1737  * @param[out] error
1738  *   Pointer to error structure.
1739  *
1740  * @return
1741  *   0 on success, a negative errno value otherwise and rte_errno is set.
1742  */
1743 int
1744 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1745 			     uint64_t item_flags,
1746 			     struct rte_eth_dev *dev,
1747 			     struct rte_flow_error *error)
1748 {
1749 	const struct rte_flow_item_vlan *spec = item->spec;
1750 	const struct rte_flow_item_vlan *mask = item->mask;
1751 	const struct rte_flow_item_vlan nic_mask = {
1752 		.tci = RTE_BE16(UINT16_MAX),
1753 		.inner_type = RTE_BE16(UINT16_MAX),
1754 	};
1755 	uint16_t vlan_tag = 0;
1756 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1757 	int ret;
1758 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1759 					MLX5_FLOW_LAYER_INNER_L4) :
1760 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1761 					MLX5_FLOW_LAYER_OUTER_L4);
1762 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1763 					MLX5_FLOW_LAYER_OUTER_VLAN;
1764 
1765 	if (item_flags & vlanm)
1766 		return rte_flow_error_set(error, EINVAL,
1767 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1768 					  "multiple VLAN layers not supported");
1769 	else if ((item_flags & l34m) != 0)
1770 		return rte_flow_error_set(error, EINVAL,
1771 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1772 					  "VLAN cannot follow L3/L4 layer");
1773 	if (!mask)
1774 		mask = &rte_flow_item_vlan_mask;
1775 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1776 					(const uint8_t *)&nic_mask,
1777 					sizeof(struct rte_flow_item_vlan),
1778 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1779 	if (ret)
1780 		return ret;
1781 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1782 		struct mlx5_priv *priv = dev->data->dev_private;
1783 
1784 		if (priv->vmwa_context) {
1785 			/*
1786 			 * Non-NULL context means we have a virtual machine
1787 			 * and SR-IOV enabled, we have to create VLAN interface
1788 			 * to make hypervisor to setup E-Switch vport
1789 			 * context correctly. We avoid creating the multiple
1790 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1791 			 */
1792 			return rte_flow_error_set(error, EINVAL,
1793 						  RTE_FLOW_ERROR_TYPE_ITEM,
1794 						  item,
1795 						  "VLAN tag mask is not"
1796 						  " supported in virtual"
1797 						  " environment");
1798 		}
1799 	}
1800 	if (spec) {
1801 		vlan_tag = spec->tci;
1802 		vlan_tag &= mask->tci;
1803 	}
1804 	/*
1805 	 * From verbs perspective an empty VLAN is equivalent
1806 	 * to a packet without VLAN layer.
1807 	 */
1808 	if (!vlan_tag)
1809 		return rte_flow_error_set(error, EINVAL,
1810 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1811 					  item->spec,
1812 					  "VLAN cannot be empty");
1813 	return 0;
1814 }
1815 
1816 /**
1817  * Validate IPV4 item.
1818  *
1819  * @param[in] item
1820  *   Item specification.
1821  * @param[in] item_flags
1822  *   Bit-fields that holds the items detected until now.
1823  * @param[in] last_item
1824  *   Previous validated item in the pattern items.
1825  * @param[in] ether_type
1826  *   Type in the ethernet layer header (including dot1q).
1827  * @param[in] acc_mask
1828  *   Acceptable mask, if NULL default internal default mask
1829  *   will be used to check whether item fields are supported.
1830  * @param[in] range_accepted
1831  *   True if range of values is accepted for specific fields, false otherwise.
1832  * @param[out] error
1833  *   Pointer to error structure.
1834  *
1835  * @return
1836  *   0 on success, a negative errno value otherwise and rte_errno is set.
1837  */
1838 int
1839 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1840 			     uint64_t item_flags,
1841 			     uint64_t last_item,
1842 			     uint16_t ether_type,
1843 			     const struct rte_flow_item_ipv4 *acc_mask,
1844 			     bool range_accepted,
1845 			     struct rte_flow_error *error)
1846 {
1847 	const struct rte_flow_item_ipv4 *mask = item->mask;
1848 	const struct rte_flow_item_ipv4 *spec = item->spec;
1849 	const struct rte_flow_item_ipv4 nic_mask = {
1850 		.hdr = {
1851 			.src_addr = RTE_BE32(0xffffffff),
1852 			.dst_addr = RTE_BE32(0xffffffff),
1853 			.type_of_service = 0xff,
1854 			.next_proto_id = 0xff,
1855 		},
1856 	};
1857 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1858 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1859 				      MLX5_FLOW_LAYER_OUTER_L3;
1860 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1861 				      MLX5_FLOW_LAYER_OUTER_L4;
1862 	int ret;
1863 	uint8_t next_proto = 0xFF;
1864 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1865 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1866 				  MLX5_FLOW_LAYER_INNER_VLAN);
1867 
1868 	if ((last_item & l2_vlan) && ether_type &&
1869 	    ether_type != RTE_ETHER_TYPE_IPV4)
1870 		return rte_flow_error_set(error, EINVAL,
1871 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1872 					  "IPv4 cannot follow L2/VLAN layer "
1873 					  "which ether type is not IPv4");
1874 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1875 		if (mask && spec)
1876 			next_proto = mask->hdr.next_proto_id &
1877 				     spec->hdr.next_proto_id;
1878 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1879 			return rte_flow_error_set(error, EINVAL,
1880 						  RTE_FLOW_ERROR_TYPE_ITEM,
1881 						  item,
1882 						  "multiple tunnel "
1883 						  "not supported");
1884 	}
1885 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1886 		return rte_flow_error_set(error, EINVAL,
1887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 					  "wrong tunnel type - IPv6 specified "
1889 					  "but IPv4 item provided");
1890 	if (item_flags & l3m)
1891 		return rte_flow_error_set(error, ENOTSUP,
1892 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 					  "multiple L3 layers not supported");
1894 	else if (item_flags & l4m)
1895 		return rte_flow_error_set(error, EINVAL,
1896 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1897 					  "L3 cannot follow an L4 layer.");
1898 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1899 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1900 		return rte_flow_error_set(error, EINVAL,
1901 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1902 					  "L3 cannot follow an NVGRE layer.");
1903 	if (!mask)
1904 		mask = &rte_flow_item_ipv4_mask;
1905 	else if (mask->hdr.next_proto_id != 0 &&
1906 		 mask->hdr.next_proto_id != 0xff)
1907 		return rte_flow_error_set(error, EINVAL,
1908 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1909 					  "partial mask is not supported"
1910 					  " for protocol");
1911 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1912 					acc_mask ? (const uint8_t *)acc_mask
1913 						 : (const uint8_t *)&nic_mask,
1914 					sizeof(struct rte_flow_item_ipv4),
1915 					range_accepted, error);
1916 	if (ret < 0)
1917 		return ret;
1918 	return 0;
1919 }
1920 
1921 /**
1922  * Validate IPV6 item.
1923  *
1924  * @param[in] item
1925  *   Item specification.
1926  * @param[in] item_flags
1927  *   Bit-fields that holds the items detected until now.
1928  * @param[in] last_item
1929  *   Previous validated item in the pattern items.
1930  * @param[in] ether_type
1931  *   Type in the ethernet layer header (including dot1q).
1932  * @param[in] acc_mask
1933  *   Acceptable mask, if NULL default internal default mask
1934  *   will be used to check whether item fields are supported.
1935  * @param[out] error
1936  *   Pointer to error structure.
1937  *
1938  * @return
1939  *   0 on success, a negative errno value otherwise and rte_errno is set.
1940  */
1941 int
1942 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1943 			     uint64_t item_flags,
1944 			     uint64_t last_item,
1945 			     uint16_t ether_type,
1946 			     const struct rte_flow_item_ipv6 *acc_mask,
1947 			     struct rte_flow_error *error)
1948 {
1949 	const struct rte_flow_item_ipv6 *mask = item->mask;
1950 	const struct rte_flow_item_ipv6 *spec = item->spec;
1951 	const struct rte_flow_item_ipv6 nic_mask = {
1952 		.hdr = {
1953 			.src_addr =
1954 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1955 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1956 			.dst_addr =
1957 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1958 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1959 			.vtc_flow = RTE_BE32(0xffffffff),
1960 			.proto = 0xff,
1961 		},
1962 	};
1963 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1964 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1965 				      MLX5_FLOW_LAYER_OUTER_L3;
1966 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1967 				      MLX5_FLOW_LAYER_OUTER_L4;
1968 	int ret;
1969 	uint8_t next_proto = 0xFF;
1970 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1971 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1972 				  MLX5_FLOW_LAYER_INNER_VLAN);
1973 
1974 	if ((last_item & l2_vlan) && ether_type &&
1975 	    ether_type != RTE_ETHER_TYPE_IPV6)
1976 		return rte_flow_error_set(error, EINVAL,
1977 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1978 					  "IPv6 cannot follow L2/VLAN layer "
1979 					  "which ether type is not IPv6");
1980 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1981 		if (mask && spec)
1982 			next_proto = mask->hdr.proto & spec->hdr.proto;
1983 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1984 			return rte_flow_error_set(error, EINVAL,
1985 						  RTE_FLOW_ERROR_TYPE_ITEM,
1986 						  item,
1987 						  "multiple tunnel "
1988 						  "not supported");
1989 	}
1990 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1991 		return rte_flow_error_set(error, EINVAL,
1992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 					  "wrong tunnel type - IPv4 specified "
1994 					  "but IPv6 item provided");
1995 	if (item_flags & l3m)
1996 		return rte_flow_error_set(error, ENOTSUP,
1997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1998 					  "multiple L3 layers not supported");
1999 	else if (item_flags & l4m)
2000 		return rte_flow_error_set(error, EINVAL,
2001 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2002 					  "L3 cannot follow an L4 layer.");
2003 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2004 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2005 		return rte_flow_error_set(error, EINVAL,
2006 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 					  "L3 cannot follow an NVGRE layer.");
2008 	if (!mask)
2009 		mask = &rte_flow_item_ipv6_mask;
2010 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2011 					acc_mask ? (const uint8_t *)acc_mask
2012 						 : (const uint8_t *)&nic_mask,
2013 					sizeof(struct rte_flow_item_ipv6),
2014 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2015 	if (ret < 0)
2016 		return ret;
2017 	return 0;
2018 }
2019 
2020 /**
2021  * Validate UDP item.
2022  *
2023  * @param[in] item
2024  *   Item specification.
2025  * @param[in] item_flags
2026  *   Bit-fields that holds the items detected until now.
2027  * @param[in] target_protocol
2028  *   The next protocol in the previous item.
2029  * @param[in] flow_mask
2030  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2031  * @param[out] error
2032  *   Pointer to error structure.
2033  *
2034  * @return
2035  *   0 on success, a negative errno value otherwise and rte_errno is set.
2036  */
2037 int
2038 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2039 			    uint64_t item_flags,
2040 			    uint8_t target_protocol,
2041 			    struct rte_flow_error *error)
2042 {
2043 	const struct rte_flow_item_udp *mask = item->mask;
2044 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2045 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2046 				      MLX5_FLOW_LAYER_OUTER_L3;
2047 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2048 				      MLX5_FLOW_LAYER_OUTER_L4;
2049 	int ret;
2050 
2051 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2052 		return rte_flow_error_set(error, EINVAL,
2053 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2054 					  "protocol filtering not compatible"
2055 					  " with UDP layer");
2056 	if (!(item_flags & l3m))
2057 		return rte_flow_error_set(error, EINVAL,
2058 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2059 					  "L3 is mandatory to filter on L4");
2060 	if (item_flags & l4m)
2061 		return rte_flow_error_set(error, EINVAL,
2062 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2063 					  "multiple L4 layers not supported");
2064 	if (!mask)
2065 		mask = &rte_flow_item_udp_mask;
2066 	ret = mlx5_flow_item_acceptable
2067 		(item, (const uint8_t *)mask,
2068 		 (const uint8_t *)&rte_flow_item_udp_mask,
2069 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2070 		 error);
2071 	if (ret < 0)
2072 		return ret;
2073 	return 0;
2074 }
2075 
2076 /**
2077  * Validate TCP item.
2078  *
2079  * @param[in] item
2080  *   Item specification.
2081  * @param[in] item_flags
2082  *   Bit-fields that holds the items detected until now.
2083  * @param[in] target_protocol
2084  *   The next protocol in the previous item.
2085  * @param[out] error
2086  *   Pointer to error structure.
2087  *
2088  * @return
2089  *   0 on success, a negative errno value otherwise and rte_errno is set.
2090  */
2091 int
2092 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2093 			    uint64_t item_flags,
2094 			    uint8_t target_protocol,
2095 			    const struct rte_flow_item_tcp *flow_mask,
2096 			    struct rte_flow_error *error)
2097 {
2098 	const struct rte_flow_item_tcp *mask = item->mask;
2099 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2100 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2101 				      MLX5_FLOW_LAYER_OUTER_L3;
2102 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2103 				      MLX5_FLOW_LAYER_OUTER_L4;
2104 	int ret;
2105 
2106 	MLX5_ASSERT(flow_mask);
2107 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2108 		return rte_flow_error_set(error, EINVAL,
2109 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2110 					  "protocol filtering not compatible"
2111 					  " with TCP layer");
2112 	if (!(item_flags & l3m))
2113 		return rte_flow_error_set(error, EINVAL,
2114 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2115 					  "L3 is mandatory to filter on L4");
2116 	if (item_flags & l4m)
2117 		return rte_flow_error_set(error, EINVAL,
2118 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2119 					  "multiple L4 layers not supported");
2120 	if (!mask)
2121 		mask = &rte_flow_item_tcp_mask;
2122 	ret = mlx5_flow_item_acceptable
2123 		(item, (const uint8_t *)mask,
2124 		 (const uint8_t *)flow_mask,
2125 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2126 		 error);
2127 	if (ret < 0)
2128 		return ret;
2129 	return 0;
2130 }
2131 
2132 /**
2133  * Validate VXLAN item.
2134  *
2135  * @param[in] item
2136  *   Item specification.
2137  * @param[in] item_flags
2138  *   Bit-fields that holds the items detected until now.
2139  * @param[in] target_protocol
2140  *   The next protocol in the previous item.
2141  * @param[out] error
2142  *   Pointer to error structure.
2143  *
2144  * @return
2145  *   0 on success, a negative errno value otherwise and rte_errno is set.
2146  */
2147 int
2148 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2149 			      uint64_t item_flags,
2150 			      struct rte_flow_error *error)
2151 {
2152 	const struct rte_flow_item_vxlan *spec = item->spec;
2153 	const struct rte_flow_item_vxlan *mask = item->mask;
2154 	int ret;
2155 	union vni {
2156 		uint32_t vlan_id;
2157 		uint8_t vni[4];
2158 	} id = { .vlan_id = 0, };
2159 
2160 
2161 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2162 		return rte_flow_error_set(error, ENOTSUP,
2163 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2164 					  "multiple tunnel layers not"
2165 					  " supported");
2166 	/*
2167 	 * Verify only UDPv4 is present as defined in
2168 	 * https://tools.ietf.org/html/rfc7348
2169 	 */
2170 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "no outer UDP layer found");
2174 	if (!mask)
2175 		mask = &rte_flow_item_vxlan_mask;
2176 	ret = mlx5_flow_item_acceptable
2177 		(item, (const uint8_t *)mask,
2178 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2179 		 sizeof(struct rte_flow_item_vxlan),
2180 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2181 	if (ret < 0)
2182 		return ret;
2183 	if (spec) {
2184 		memcpy(&id.vni[1], spec->vni, 3);
2185 		memcpy(&id.vni[1], mask->vni, 3);
2186 	}
2187 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2188 		return rte_flow_error_set(error, ENOTSUP,
2189 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2190 					  "VXLAN tunnel must be fully defined");
2191 	return 0;
2192 }
2193 
2194 /**
2195  * Validate VXLAN_GPE item.
2196  *
2197  * @param[in] item
2198  *   Item specification.
2199  * @param[in] item_flags
2200  *   Bit-fields that holds the items detected until now.
2201  * @param[in] priv
2202  *   Pointer to the private data structure.
2203  * @param[in] target_protocol
2204  *   The next protocol in the previous item.
2205  * @param[out] error
2206  *   Pointer to error structure.
2207  *
2208  * @return
2209  *   0 on success, a negative errno value otherwise and rte_errno is set.
2210  */
2211 int
2212 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2213 				  uint64_t item_flags,
2214 				  struct rte_eth_dev *dev,
2215 				  struct rte_flow_error *error)
2216 {
2217 	struct mlx5_priv *priv = dev->data->dev_private;
2218 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2219 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2220 	int ret;
2221 	union vni {
2222 		uint32_t vlan_id;
2223 		uint8_t vni[4];
2224 	} id = { .vlan_id = 0, };
2225 
2226 	if (!priv->config.l3_vxlan_en)
2227 		return rte_flow_error_set(error, ENOTSUP,
2228 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2229 					  "L3 VXLAN is not enabled by device"
2230 					  " parameter and/or not configured in"
2231 					  " firmware");
2232 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2233 		return rte_flow_error_set(error, ENOTSUP,
2234 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2235 					  "multiple tunnel layers not"
2236 					  " supported");
2237 	/*
2238 	 * Verify only UDPv4 is present as defined in
2239 	 * https://tools.ietf.org/html/rfc7348
2240 	 */
2241 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2242 		return rte_flow_error_set(error, EINVAL,
2243 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2244 					  "no outer UDP layer found");
2245 	if (!mask)
2246 		mask = &rte_flow_item_vxlan_gpe_mask;
2247 	ret = mlx5_flow_item_acceptable
2248 		(item, (const uint8_t *)mask,
2249 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2250 		 sizeof(struct rte_flow_item_vxlan_gpe),
2251 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2252 	if (ret < 0)
2253 		return ret;
2254 	if (spec) {
2255 		if (spec->protocol)
2256 			return rte_flow_error_set(error, ENOTSUP,
2257 						  RTE_FLOW_ERROR_TYPE_ITEM,
2258 						  item,
2259 						  "VxLAN-GPE protocol"
2260 						  " not supported");
2261 		memcpy(&id.vni[1], spec->vni, 3);
2262 		memcpy(&id.vni[1], mask->vni, 3);
2263 	}
2264 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2265 		return rte_flow_error_set(error, ENOTSUP,
2266 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2267 					  "VXLAN-GPE tunnel must be fully"
2268 					  " defined");
2269 	return 0;
2270 }
2271 /**
2272  * Validate GRE Key item.
2273  *
2274  * @param[in] item
2275  *   Item specification.
2276  * @param[in] item_flags
2277  *   Bit flags to mark detected items.
2278  * @param[in] gre_item
2279  *   Pointer to gre_item
2280  * @param[out] error
2281  *   Pointer to error structure.
2282  *
2283  * @return
2284  *   0 on success, a negative errno value otherwise and rte_errno is set.
2285  */
2286 int
2287 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2288 				uint64_t item_flags,
2289 				const struct rte_flow_item *gre_item,
2290 				struct rte_flow_error *error)
2291 {
2292 	const rte_be32_t *mask = item->mask;
2293 	int ret = 0;
2294 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2295 	const struct rte_flow_item_gre *gre_spec;
2296 	const struct rte_flow_item_gre *gre_mask;
2297 
2298 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2299 		return rte_flow_error_set(error, ENOTSUP,
2300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2301 					  "Multiple GRE key not support");
2302 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2303 		return rte_flow_error_set(error, ENOTSUP,
2304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2305 					  "No preceding GRE header");
2306 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2307 		return rte_flow_error_set(error, ENOTSUP,
2308 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2309 					  "GRE key following a wrong item");
2310 	gre_mask = gre_item->mask;
2311 	if (!gre_mask)
2312 		gre_mask = &rte_flow_item_gre_mask;
2313 	gre_spec = gre_item->spec;
2314 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2315 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2316 		return rte_flow_error_set(error, EINVAL,
2317 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2318 					  "Key bit must be on");
2319 
2320 	if (!mask)
2321 		mask = &gre_key_default_mask;
2322 	ret = mlx5_flow_item_acceptable
2323 		(item, (const uint8_t *)mask,
2324 		 (const uint8_t *)&gre_key_default_mask,
2325 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2326 	return ret;
2327 }
2328 
2329 /**
2330  * Validate GRE item.
2331  *
2332  * @param[in] item
2333  *   Item specification.
2334  * @param[in] item_flags
2335  *   Bit flags to mark detected items.
2336  * @param[in] target_protocol
2337  *   The next protocol in the previous item.
2338  * @param[out] error
2339  *   Pointer to error structure.
2340  *
2341  * @return
2342  *   0 on success, a negative errno value otherwise and rte_errno is set.
2343  */
2344 int
2345 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2346 			    uint64_t item_flags,
2347 			    uint8_t target_protocol,
2348 			    struct rte_flow_error *error)
2349 {
2350 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2351 	const struct rte_flow_item_gre *mask = item->mask;
2352 	int ret;
2353 	const struct rte_flow_item_gre nic_mask = {
2354 		.c_rsvd0_ver = RTE_BE16(0xB000),
2355 		.protocol = RTE_BE16(UINT16_MAX),
2356 	};
2357 
2358 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2359 		return rte_flow_error_set(error, EINVAL,
2360 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2361 					  "protocol filtering not compatible"
2362 					  " with this GRE layer");
2363 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2364 		return rte_flow_error_set(error, ENOTSUP,
2365 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2366 					  "multiple tunnel layers not"
2367 					  " supported");
2368 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2369 		return rte_flow_error_set(error, ENOTSUP,
2370 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2371 					  "L3 Layer is missing");
2372 	if (!mask)
2373 		mask = &rte_flow_item_gre_mask;
2374 	ret = mlx5_flow_item_acceptable
2375 		(item, (const uint8_t *)mask,
2376 		 (const uint8_t *)&nic_mask,
2377 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2378 		 error);
2379 	if (ret < 0)
2380 		return ret;
2381 #ifndef HAVE_MLX5DV_DR
2382 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2383 	if (spec && (spec->protocol & mask->protocol))
2384 		return rte_flow_error_set(error, ENOTSUP,
2385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2386 					  "without MPLS support the"
2387 					  " specification cannot be used for"
2388 					  " filtering");
2389 #endif
2390 #endif
2391 	return 0;
2392 }
2393 
2394 /**
2395  * Validate Geneve item.
2396  *
2397  * @param[in] item
2398  *   Item specification.
2399  * @param[in] itemFlags
2400  *   Bit-fields that holds the items detected until now.
2401  * @param[in] enPriv
2402  *   Pointer to the private data structure.
2403  * @param[out] error
2404  *   Pointer to error structure.
2405  *
2406  * @return
2407  *   0 on success, a negative errno value otherwise and rte_errno is set.
2408  */
2409 
2410 int
2411 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2412 			       uint64_t item_flags,
2413 			       struct rte_eth_dev *dev,
2414 			       struct rte_flow_error *error)
2415 {
2416 	struct mlx5_priv *priv = dev->data->dev_private;
2417 	const struct rte_flow_item_geneve *spec = item->spec;
2418 	const struct rte_flow_item_geneve *mask = item->mask;
2419 	int ret;
2420 	uint16_t gbhdr;
2421 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2422 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2423 	const struct rte_flow_item_geneve nic_mask = {
2424 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2425 		.vni = "\xff\xff\xff",
2426 		.protocol = RTE_BE16(UINT16_MAX),
2427 	};
2428 
2429 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2430 		return rte_flow_error_set(error, ENOTSUP,
2431 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2432 					  "L3 Geneve is not enabled by device"
2433 					  " parameter and/or not configured in"
2434 					  " firmware");
2435 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2436 		return rte_flow_error_set(error, ENOTSUP,
2437 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2438 					  "multiple tunnel layers not"
2439 					  " supported");
2440 	/*
2441 	 * Verify only UDPv4 is present as defined in
2442 	 * https://tools.ietf.org/html/rfc7348
2443 	 */
2444 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2445 		return rte_flow_error_set(error, EINVAL,
2446 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2447 					  "no outer UDP layer found");
2448 	if (!mask)
2449 		mask = &rte_flow_item_geneve_mask;
2450 	ret = mlx5_flow_item_acceptable
2451 				  (item, (const uint8_t *)mask,
2452 				   (const uint8_t *)&nic_mask,
2453 				   sizeof(struct rte_flow_item_geneve),
2454 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2455 	if (ret)
2456 		return ret;
2457 	if (spec) {
2458 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2459 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2460 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2461 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2462 			return rte_flow_error_set(error, ENOTSUP,
2463 						  RTE_FLOW_ERROR_TYPE_ITEM,
2464 						  item,
2465 						  "Geneve protocol unsupported"
2466 						  " fields are being used");
2467 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2468 			return rte_flow_error_set
2469 					(error, ENOTSUP,
2470 					 RTE_FLOW_ERROR_TYPE_ITEM,
2471 					 item,
2472 					 "Unsupported Geneve options length");
2473 	}
2474 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2475 		return rte_flow_error_set
2476 				    (error, ENOTSUP,
2477 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2478 				     "Geneve tunnel must be fully defined");
2479 	return 0;
2480 }
2481 
2482 /**
2483  * Validate MPLS item.
2484  *
2485  * @param[in] dev
2486  *   Pointer to the rte_eth_dev structure.
2487  * @param[in] item
2488  *   Item specification.
2489  * @param[in] item_flags
2490  *   Bit-fields that holds the items detected until now.
2491  * @param[in] prev_layer
2492  *   The protocol layer indicated in previous item.
2493  * @param[out] error
2494  *   Pointer to error structure.
2495  *
2496  * @return
2497  *   0 on success, a negative errno value otherwise and rte_errno is set.
2498  */
2499 int
2500 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2501 			     const struct rte_flow_item *item __rte_unused,
2502 			     uint64_t item_flags __rte_unused,
2503 			     uint64_t prev_layer __rte_unused,
2504 			     struct rte_flow_error *error)
2505 {
2506 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2507 	const struct rte_flow_item_mpls *mask = item->mask;
2508 	struct mlx5_priv *priv = dev->data->dev_private;
2509 	int ret;
2510 
2511 	if (!priv->config.mpls_en)
2512 		return rte_flow_error_set(error, ENOTSUP,
2513 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2514 					  "MPLS not supported or"
2515 					  " disabled in firmware"
2516 					  " configuration.");
2517 	/* MPLS over IP, UDP, GRE is allowed */
2518 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2519 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2520 			    MLX5_FLOW_LAYER_GRE)))
2521 		return rte_flow_error_set(error, EINVAL,
2522 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2523 					  "protocol filtering not compatible"
2524 					  " with MPLS layer");
2525 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2526 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2527 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2528 		return rte_flow_error_set(error, ENOTSUP,
2529 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2530 					  "multiple tunnel layers not"
2531 					  " supported");
2532 	if (!mask)
2533 		mask = &rte_flow_item_mpls_mask;
2534 	ret = mlx5_flow_item_acceptable
2535 		(item, (const uint8_t *)mask,
2536 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2537 		 sizeof(struct rte_flow_item_mpls),
2538 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2539 	if (ret < 0)
2540 		return ret;
2541 	return 0;
2542 #else
2543 	return rte_flow_error_set(error, ENOTSUP,
2544 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2545 				  "MPLS is not supported by Verbs, please"
2546 				  " update.");
2547 #endif
2548 }
2549 
2550 /**
2551  * Validate NVGRE item.
2552  *
2553  * @param[in] item
2554  *   Item specification.
2555  * @param[in] item_flags
2556  *   Bit flags to mark detected items.
2557  * @param[in] target_protocol
2558  *   The next protocol in the previous item.
2559  * @param[out] error
2560  *   Pointer to error structure.
2561  *
2562  * @return
2563  *   0 on success, a negative errno value otherwise and rte_errno is set.
2564  */
2565 int
2566 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2567 			      uint64_t item_flags,
2568 			      uint8_t target_protocol,
2569 			      struct rte_flow_error *error)
2570 {
2571 	const struct rte_flow_item_nvgre *mask = item->mask;
2572 	int ret;
2573 
2574 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2575 		return rte_flow_error_set(error, EINVAL,
2576 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2577 					  "protocol filtering not compatible"
2578 					  " with this GRE layer");
2579 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2580 		return rte_flow_error_set(error, ENOTSUP,
2581 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2582 					  "multiple tunnel layers not"
2583 					  " supported");
2584 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2585 		return rte_flow_error_set(error, ENOTSUP,
2586 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2587 					  "L3 Layer is missing");
2588 	if (!mask)
2589 		mask = &rte_flow_item_nvgre_mask;
2590 	ret = mlx5_flow_item_acceptable
2591 		(item, (const uint8_t *)mask,
2592 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2593 		 sizeof(struct rte_flow_item_nvgre),
2594 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2595 	if (ret < 0)
2596 		return ret;
2597 	return 0;
2598 }
2599 
2600 /**
2601  * Validate eCPRI item.
2602  *
2603  * @param[in] item
2604  *   Item specification.
2605  * @param[in] item_flags
2606  *   Bit-fields that holds the items detected until now.
2607  * @param[in] last_item
2608  *   Previous validated item in the pattern items.
2609  * @param[in] ether_type
2610  *   Type in the ethernet layer header (including dot1q).
2611  * @param[in] acc_mask
2612  *   Acceptable mask, if NULL default internal default mask
2613  *   will be used to check whether item fields are supported.
2614  * @param[out] error
2615  *   Pointer to error structure.
2616  *
2617  * @return
2618  *   0 on success, a negative errno value otherwise and rte_errno is set.
2619  */
2620 int
2621 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2622 			      uint64_t item_flags,
2623 			      uint64_t last_item,
2624 			      uint16_t ether_type,
2625 			      const struct rte_flow_item_ecpri *acc_mask,
2626 			      struct rte_flow_error *error)
2627 {
2628 	const struct rte_flow_item_ecpri *mask = item->mask;
2629 	const struct rte_flow_item_ecpri nic_mask = {
2630 		.hdr = {
2631 			.common = {
2632 				.u32 =
2633 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2634 					.type = 0xFF,
2635 					}).u32),
2636 			},
2637 			.dummy[0] = 0xFFFFFFFF,
2638 		},
2639 	};
2640 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2641 					MLX5_FLOW_LAYER_OUTER_VLAN);
2642 	struct rte_flow_item_ecpri mask_lo;
2643 
2644 	if ((last_item & outer_l2_vlan) && ether_type &&
2645 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2646 		return rte_flow_error_set(error, EINVAL,
2647 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2648 					  "eCPRI cannot follow L2/VLAN layer "
2649 					  "which ether type is not 0xAEFE.");
2650 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2651 		return rte_flow_error_set(error, EINVAL,
2652 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2653 					  "eCPRI with tunnel is not supported "
2654 					  "right now.");
2655 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2656 		return rte_flow_error_set(error, ENOTSUP,
2657 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2658 					  "multiple L3 layers not supported");
2659 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2660 		return rte_flow_error_set(error, EINVAL,
2661 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2662 					  "eCPRI cannot follow a TCP layer.");
2663 	/* In specification, eCPRI could be over UDP layer. */
2664 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2665 		return rte_flow_error_set(error, EINVAL,
2666 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2667 					  "eCPRI over UDP layer is not yet "
2668 					  "supported right now.");
2669 	/* Mask for type field in common header could be zero. */
2670 	if (!mask)
2671 		mask = &rte_flow_item_ecpri_mask;
2672 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2673 	/* Input mask is in big-endian format. */
2674 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2675 		return rte_flow_error_set(error, EINVAL,
2676 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2677 					  "partial mask is not supported "
2678 					  "for protocol");
2679 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2680 		return rte_flow_error_set(error, EINVAL,
2681 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2682 					  "message header mask must be after "
2683 					  "a type mask");
2684 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2685 					 acc_mask ? (const uint8_t *)acc_mask
2686 						  : (const uint8_t *)&nic_mask,
2687 					 sizeof(struct rte_flow_item_ecpri),
2688 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2689 }
2690 
2691 /* Allocate unique ID for the split Q/RSS subflows. */
2692 static uint32_t
2693 flow_qrss_get_id(struct rte_eth_dev *dev)
2694 {
2695 	struct mlx5_priv *priv = dev->data->dev_private;
2696 	uint32_t qrss_id, ret;
2697 
2698 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2699 	if (ret)
2700 		return 0;
2701 	MLX5_ASSERT(qrss_id);
2702 	return qrss_id;
2703 }
2704 
2705 /* Free unique ID for the split Q/RSS subflows. */
2706 static void
2707 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2708 {
2709 	struct mlx5_priv *priv = dev->data->dev_private;
2710 
2711 	if (qrss_id)
2712 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2713 }
2714 
2715 /**
2716  * Release resource related QUEUE/RSS action split.
2717  *
2718  * @param dev
2719  *   Pointer to Ethernet device.
2720  * @param flow
2721  *   Flow to release id's from.
2722  */
2723 static void
2724 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2725 			     struct rte_flow *flow)
2726 {
2727 	struct mlx5_priv *priv = dev->data->dev_private;
2728 	uint32_t handle_idx;
2729 	struct mlx5_flow_handle *dev_handle;
2730 
2731 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2732 		       handle_idx, dev_handle, next)
2733 		if (dev_handle->split_flow_id)
2734 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2735 }
2736 
2737 static int
2738 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2739 		   const struct rte_flow_attr *attr __rte_unused,
2740 		   const struct rte_flow_item items[] __rte_unused,
2741 		   const struct rte_flow_action actions[] __rte_unused,
2742 		   bool external __rte_unused,
2743 		   int hairpin __rte_unused,
2744 		   struct rte_flow_error *error)
2745 {
2746 	return rte_flow_error_set(error, ENOTSUP,
2747 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2748 }
2749 
2750 static struct mlx5_flow *
2751 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2752 		  const struct rte_flow_attr *attr __rte_unused,
2753 		  const struct rte_flow_item items[] __rte_unused,
2754 		  const struct rte_flow_action actions[] __rte_unused,
2755 		  struct rte_flow_error *error)
2756 {
2757 	rte_flow_error_set(error, ENOTSUP,
2758 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2759 	return NULL;
2760 }
2761 
2762 static int
2763 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2764 		    struct mlx5_flow *dev_flow __rte_unused,
2765 		    const struct rte_flow_attr *attr __rte_unused,
2766 		    const struct rte_flow_item items[] __rte_unused,
2767 		    const struct rte_flow_action actions[] __rte_unused,
2768 		    struct rte_flow_error *error)
2769 {
2770 	return rte_flow_error_set(error, ENOTSUP,
2771 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2772 }
2773 
2774 static int
2775 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2776 		struct rte_flow *flow __rte_unused,
2777 		struct rte_flow_error *error)
2778 {
2779 	return rte_flow_error_set(error, ENOTSUP,
2780 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2781 }
2782 
2783 static void
2784 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2785 		 struct rte_flow *flow __rte_unused)
2786 {
2787 }
2788 
2789 static void
2790 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2791 		  struct rte_flow *flow __rte_unused)
2792 {
2793 }
2794 
2795 static int
2796 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2797 		struct rte_flow *flow __rte_unused,
2798 		const struct rte_flow_action *actions __rte_unused,
2799 		void *data __rte_unused,
2800 		struct rte_flow_error *error)
2801 {
2802 	return rte_flow_error_set(error, ENOTSUP,
2803 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2804 }
2805 
2806 /* Void driver to protect from null pointer reference. */
2807 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2808 	.validate = flow_null_validate,
2809 	.prepare = flow_null_prepare,
2810 	.translate = flow_null_translate,
2811 	.apply = flow_null_apply,
2812 	.remove = flow_null_remove,
2813 	.destroy = flow_null_destroy,
2814 	.query = flow_null_query,
2815 };
2816 
2817 /**
2818  * Select flow driver type according to flow attributes and device
2819  * configuration.
2820  *
2821  * @param[in] dev
2822  *   Pointer to the dev structure.
2823  * @param[in] attr
2824  *   Pointer to the flow attributes.
2825  *
2826  * @return
2827  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2828  */
2829 static enum mlx5_flow_drv_type
2830 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2831 {
2832 	struct mlx5_priv *priv = dev->data->dev_private;
2833 	/* The OS can determine first a specific flow type (DV, VERBS) */
2834 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2835 
2836 	if (type != MLX5_FLOW_TYPE_MAX)
2837 		return type;
2838 	/* If no OS specific type - continue with DV/VERBS selection */
2839 	if (attr->transfer && priv->config.dv_esw_en)
2840 		type = MLX5_FLOW_TYPE_DV;
2841 	if (!attr->transfer)
2842 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2843 						 MLX5_FLOW_TYPE_VERBS;
2844 	return type;
2845 }
2846 
2847 #define flow_get_drv_ops(type) flow_drv_ops[type]
2848 
2849 /**
2850  * Flow driver validation API. This abstracts calling driver specific functions.
2851  * The type of flow driver is determined according to flow attributes.
2852  *
2853  * @param[in] dev
2854  *   Pointer to the dev structure.
2855  * @param[in] attr
2856  *   Pointer to the flow attributes.
2857  * @param[in] items
2858  *   Pointer to the list of items.
2859  * @param[in] actions
2860  *   Pointer to the list of actions.
2861  * @param[in] external
2862  *   This flow rule is created by request external to PMD.
2863  * @param[in] hairpin
2864  *   Number of hairpin TX actions, 0 means classic flow.
2865  * @param[out] error
2866  *   Pointer to the error structure.
2867  *
2868  * @return
2869  *   0 on success, a negative errno value otherwise and rte_errno is set.
2870  */
2871 static inline int
2872 flow_drv_validate(struct rte_eth_dev *dev,
2873 		  const struct rte_flow_attr *attr,
2874 		  const struct rte_flow_item items[],
2875 		  const struct rte_flow_action actions[],
2876 		  bool external, int hairpin, struct rte_flow_error *error)
2877 {
2878 	const struct mlx5_flow_driver_ops *fops;
2879 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2880 
2881 	fops = flow_get_drv_ops(type);
2882 	return fops->validate(dev, attr, items, actions, external,
2883 			      hairpin, error);
2884 }
2885 
2886 /**
2887  * Flow driver preparation API. This abstracts calling driver specific
2888  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2889  * calculates the size of memory required for device flow, allocates the memory,
2890  * initializes the device flow and returns the pointer.
2891  *
2892  * @note
2893  *   This function initializes device flow structure such as dv or verbs in
2894  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2895  *   rest. For example, adding returning device flow to flow->dev_flow list and
2896  *   setting backward reference to the flow should be done out of this function.
2897  *   layers field is not filled either.
2898  *
2899  * @param[in] dev
2900  *   Pointer to the dev structure.
2901  * @param[in] attr
2902  *   Pointer to the flow attributes.
2903  * @param[in] items
2904  *   Pointer to the list of items.
2905  * @param[in] actions
2906  *   Pointer to the list of actions.
2907  * @param[in] flow_idx
2908  *   This memory pool index to the flow.
2909  * @param[out] error
2910  *   Pointer to the error structure.
2911  *
2912  * @return
2913  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2914  */
2915 static inline struct mlx5_flow *
2916 flow_drv_prepare(struct rte_eth_dev *dev,
2917 		 const struct rte_flow *flow,
2918 		 const struct rte_flow_attr *attr,
2919 		 const struct rte_flow_item items[],
2920 		 const struct rte_flow_action actions[],
2921 		 uint32_t flow_idx,
2922 		 struct rte_flow_error *error)
2923 {
2924 	const struct mlx5_flow_driver_ops *fops;
2925 	enum mlx5_flow_drv_type type = flow->drv_type;
2926 	struct mlx5_flow *mlx5_flow = NULL;
2927 
2928 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2929 	fops = flow_get_drv_ops(type);
2930 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2931 	if (mlx5_flow)
2932 		mlx5_flow->flow_idx = flow_idx;
2933 	return mlx5_flow;
2934 }
2935 
2936 /**
2937  * Flow driver translation API. This abstracts calling driver specific
2938  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2939  * translates a generic flow into a driver flow. flow_drv_prepare() must
2940  * precede.
2941  *
2942  * @note
2943  *   dev_flow->layers could be filled as a result of parsing during translation
2944  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2945  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2946  *   flow->actions could be overwritten even though all the expanded dev_flows
2947  *   have the same actions.
2948  *
2949  * @param[in] dev
2950  *   Pointer to the rte dev structure.
2951  * @param[in, out] dev_flow
2952  *   Pointer to the mlx5 flow.
2953  * @param[in] attr
2954  *   Pointer to the flow attributes.
2955  * @param[in] items
2956  *   Pointer to the list of items.
2957  * @param[in] actions
2958  *   Pointer to the list of actions.
2959  * @param[out] error
2960  *   Pointer to the error structure.
2961  *
2962  * @return
2963  *   0 on success, a negative errno value otherwise and rte_errno is set.
2964  */
2965 static inline int
2966 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2967 		   const struct rte_flow_attr *attr,
2968 		   const struct rte_flow_item items[],
2969 		   const struct rte_flow_action actions[],
2970 		   struct rte_flow_error *error)
2971 {
2972 	const struct mlx5_flow_driver_ops *fops;
2973 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2974 
2975 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2976 	fops = flow_get_drv_ops(type);
2977 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2978 }
2979 
2980 /**
2981  * Flow driver apply API. This abstracts calling driver specific functions.
2982  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2983  * translated driver flows on to device. flow_drv_translate() must precede.
2984  *
2985  * @param[in] dev
2986  *   Pointer to Ethernet device structure.
2987  * @param[in, out] flow
2988  *   Pointer to flow structure.
2989  * @param[out] error
2990  *   Pointer to error structure.
2991  *
2992  * @return
2993  *   0 on success, a negative errno value otherwise and rte_errno is set.
2994  */
2995 static inline int
2996 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2997 	       struct rte_flow_error *error)
2998 {
2999 	const struct mlx5_flow_driver_ops *fops;
3000 	enum mlx5_flow_drv_type type = flow->drv_type;
3001 
3002 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3003 	fops = flow_get_drv_ops(type);
3004 	return fops->apply(dev, flow, error);
3005 }
3006 
3007 /**
3008  * Flow driver remove API. This abstracts calling driver specific functions.
3009  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3010  * on device. All the resources of the flow should be freed by calling
3011  * flow_drv_destroy().
3012  *
3013  * @param[in] dev
3014  *   Pointer to Ethernet device.
3015  * @param[in, out] flow
3016  *   Pointer to flow structure.
3017  */
3018 static inline void
3019 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
3020 {
3021 	const struct mlx5_flow_driver_ops *fops;
3022 	enum mlx5_flow_drv_type type = flow->drv_type;
3023 
3024 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3025 	fops = flow_get_drv_ops(type);
3026 	fops->remove(dev, flow);
3027 }
3028 
3029 /**
3030  * Flow driver destroy API. This abstracts calling driver specific functions.
3031  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3032  * on device and releases resources of the flow.
3033  *
3034  * @param[in] dev
3035  *   Pointer to Ethernet device.
3036  * @param[in, out] flow
3037  *   Pointer to flow structure.
3038  */
3039 static inline void
3040 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3041 {
3042 	const struct mlx5_flow_driver_ops *fops;
3043 	enum mlx5_flow_drv_type type = flow->drv_type;
3044 
3045 	flow_mreg_split_qrss_release(dev, flow);
3046 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3047 	fops = flow_get_drv_ops(type);
3048 	fops->destroy(dev, flow);
3049 }
3050 
3051 /**
3052  * Get RSS action from the action list.
3053  *
3054  * @param[in] actions
3055  *   Pointer to the list of actions.
3056  *
3057  * @return
3058  *   Pointer to the RSS action if exist, else return NULL.
3059  */
3060 static const struct rte_flow_action_rss*
3061 flow_get_rss_action(const struct rte_flow_action actions[])
3062 {
3063 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3064 		switch (actions->type) {
3065 		case RTE_FLOW_ACTION_TYPE_RSS:
3066 			return (const struct rte_flow_action_rss *)
3067 			       actions->conf;
3068 		default:
3069 			break;
3070 		}
3071 	}
3072 	return NULL;
3073 }
3074 
3075 static unsigned int
3076 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3077 {
3078 	const struct rte_flow_item *item;
3079 	unsigned int has_vlan = 0;
3080 
3081 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3082 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3083 			has_vlan = 1;
3084 			break;
3085 		}
3086 	}
3087 	if (has_vlan)
3088 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3089 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3090 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3091 			       MLX5_EXPANSION_ROOT_OUTER;
3092 }
3093 
3094 /**
3095  *  Get layer flags from the prefix flow.
3096  *
3097  *  Some flows may be split to several subflows, the prefix subflow gets the
3098  *  match items and the suffix sub flow gets the actions.
3099  *  Some actions need the user defined match item flags to get the detail for
3100  *  the action.
3101  *  This function helps the suffix flow to get the item layer flags from prefix
3102  *  subflow.
3103  *
3104  * @param[in] dev_flow
3105  *   Pointer the created preifx subflow.
3106  *
3107  * @return
3108  *   The layers get from prefix subflow.
3109  */
3110 static inline uint64_t
3111 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3112 {
3113 	uint64_t layers = 0;
3114 
3115 	/*
3116 	 * Layers bits could be localization, but usually the compiler will
3117 	 * help to do the optimization work for source code.
3118 	 * If no decap actions, use the layers directly.
3119 	 */
3120 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3121 		return dev_flow->handle->layers;
3122 	/* Convert L3 layers with decap action. */
3123 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3124 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3125 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3126 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3127 	/* Convert L4 layers with decap action.  */
3128 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3129 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3130 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3131 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3132 	return layers;
3133 }
3134 
3135 /**
3136  * Get metadata split action information.
3137  *
3138  * @param[in] actions
3139  *   Pointer to the list of actions.
3140  * @param[out] qrss
3141  *   Pointer to the return pointer.
3142  * @param[out] qrss_type
3143  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3144  *   if no QUEUE/RSS is found.
3145  * @param[out] encap_idx
3146  *   Pointer to the index of the encap action if exists, otherwise the last
3147  *   action index.
3148  *
3149  * @return
3150  *   Total number of actions.
3151  */
3152 static int
3153 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3154 				       const struct rte_flow_action **qrss,
3155 				       int *encap_idx)
3156 {
3157 	const struct rte_flow_action_raw_encap *raw_encap;
3158 	int actions_n = 0;
3159 	int raw_decap_idx = -1;
3160 
3161 	*encap_idx = -1;
3162 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3163 		switch (actions->type) {
3164 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3165 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3166 			*encap_idx = actions_n;
3167 			break;
3168 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3169 			raw_decap_idx = actions_n;
3170 			break;
3171 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3172 			raw_encap = actions->conf;
3173 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3174 				*encap_idx = raw_decap_idx != -1 ?
3175 						      raw_decap_idx : actions_n;
3176 			break;
3177 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3178 		case RTE_FLOW_ACTION_TYPE_RSS:
3179 			*qrss = actions;
3180 			break;
3181 		default:
3182 			break;
3183 		}
3184 		actions_n++;
3185 	}
3186 	if (*encap_idx == -1)
3187 		*encap_idx = actions_n;
3188 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3189 	return actions_n + 1;
3190 }
3191 
3192 /**
3193  * Check meter action from the action list.
3194  *
3195  * @param[in] actions
3196  *   Pointer to the list of actions.
3197  * @param[out] mtr
3198  *   Pointer to the meter exist flag.
3199  *
3200  * @return
3201  *   Total number of actions.
3202  */
3203 static int
3204 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3205 {
3206 	int actions_n = 0;
3207 
3208 	MLX5_ASSERT(mtr);
3209 	*mtr = 0;
3210 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3211 		switch (actions->type) {
3212 		case RTE_FLOW_ACTION_TYPE_METER:
3213 			*mtr = 1;
3214 			break;
3215 		default:
3216 			break;
3217 		}
3218 		actions_n++;
3219 	}
3220 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3221 	return actions_n + 1;
3222 }
3223 
3224 /**
3225  * Check if the flow should be split due to hairpin.
3226  * The reason for the split is that in current HW we can't
3227  * support encap and push-vlan on Rx, so if a flow contains
3228  * these actions we move it to Tx.
3229  *
3230  * @param dev
3231  *   Pointer to Ethernet device.
3232  * @param[in] attr
3233  *   Flow rule attributes.
3234  * @param[in] actions
3235  *   Associated actions (list terminated by the END action).
3236  *
3237  * @return
3238  *   > 0 the number of actions and the flow should be split,
3239  *   0 when no split required.
3240  */
3241 static int
3242 flow_check_hairpin_split(struct rte_eth_dev *dev,
3243 			 const struct rte_flow_attr *attr,
3244 			 const struct rte_flow_action actions[])
3245 {
3246 	int queue_action = 0;
3247 	int action_n = 0;
3248 	int split = 0;
3249 	const struct rte_flow_action_queue *queue;
3250 	const struct rte_flow_action_rss *rss;
3251 	const struct rte_flow_action_raw_encap *raw_encap;
3252 
3253 	if (!attr->ingress)
3254 		return 0;
3255 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3256 		switch (actions->type) {
3257 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3258 			queue = actions->conf;
3259 			if (queue == NULL)
3260 				return 0;
3261 			if (mlx5_rxq_get_type(dev, queue->index) !=
3262 			    MLX5_RXQ_TYPE_HAIRPIN)
3263 				return 0;
3264 			queue_action = 1;
3265 			action_n++;
3266 			break;
3267 		case RTE_FLOW_ACTION_TYPE_RSS:
3268 			rss = actions->conf;
3269 			if (rss == NULL || rss->queue_num == 0)
3270 				return 0;
3271 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
3272 			    MLX5_RXQ_TYPE_HAIRPIN)
3273 				return 0;
3274 			queue_action = 1;
3275 			action_n++;
3276 			break;
3277 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3278 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3279 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3280 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3281 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3282 			split++;
3283 			action_n++;
3284 			break;
3285 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3286 			raw_encap = actions->conf;
3287 			if (raw_encap->size >
3288 			    (sizeof(struct rte_flow_item_eth) +
3289 			     sizeof(struct rte_flow_item_ipv4)))
3290 				split++;
3291 			action_n++;
3292 			break;
3293 		default:
3294 			action_n++;
3295 			break;
3296 		}
3297 	}
3298 	if (split && queue_action)
3299 		return action_n;
3300 	return 0;
3301 }
3302 
3303 /* Declare flow create/destroy prototype in advance. */
3304 static uint32_t
3305 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3306 		 const struct rte_flow_attr *attr,
3307 		 const struct rte_flow_item items[],
3308 		 const struct rte_flow_action actions[],
3309 		 bool external, struct rte_flow_error *error);
3310 
3311 static void
3312 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3313 		  uint32_t flow_idx);
3314 
3315 /**
3316  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3317  *
3318  * As mark_id is unique, if there's already a registered flow for the mark_id,
3319  * return by increasing the reference counter of the resource. Otherwise, create
3320  * the resource (mcp_res) and flow.
3321  *
3322  * Flow looks like,
3323  *   - If ingress port is ANY and reg_c[1] is mark_id,
3324  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3325  *
3326  * For default flow (zero mark_id), flow is like,
3327  *   - If ingress port is ANY,
3328  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3329  *
3330  * @param dev
3331  *   Pointer to Ethernet device.
3332  * @param mark_id
3333  *   ID of MARK action, zero means default flow for META.
3334  * @param[out] error
3335  *   Perform verbose error reporting if not NULL.
3336  *
3337  * @return
3338  *   Associated resource on success, NULL otherwise and rte_errno is set.
3339  */
3340 static struct mlx5_flow_mreg_copy_resource *
3341 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3342 			  struct rte_flow_error *error)
3343 {
3344 	struct mlx5_priv *priv = dev->data->dev_private;
3345 	struct rte_flow_attr attr = {
3346 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3347 		.ingress = 1,
3348 	};
3349 	struct mlx5_rte_flow_item_tag tag_spec = {
3350 		.data = mark_id,
3351 	};
3352 	struct rte_flow_item items[] = {
3353 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3354 	};
3355 	struct rte_flow_action_mark ftag = {
3356 		.id = mark_id,
3357 	};
3358 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3359 		.dst = REG_B,
3360 		.src = REG_NON,
3361 	};
3362 	struct rte_flow_action_jump jump = {
3363 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3364 	};
3365 	struct rte_flow_action actions[] = {
3366 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3367 	};
3368 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3369 	uint32_t idx = 0;
3370 	int ret;
3371 
3372 	/* Fill the register fileds in the flow. */
3373 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3374 	if (ret < 0)
3375 		return NULL;
3376 	tag_spec.id = ret;
3377 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3378 	if (ret < 0)
3379 		return NULL;
3380 	cp_mreg.src = ret;
3381 	/* Check if already registered. */
3382 	MLX5_ASSERT(priv->mreg_cp_tbl);
3383 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3384 	if (mcp_res) {
3385 		/* For non-default rule. */
3386 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3387 			mcp_res->refcnt++;
3388 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3389 			    mcp_res->refcnt == 1);
3390 		return mcp_res;
3391 	}
3392 	/* Provide the full width of FLAG specific value. */
3393 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3394 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3395 	/* Build a new flow. */
3396 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3397 		items[0] = (struct rte_flow_item){
3398 			.type = (enum rte_flow_item_type)
3399 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3400 			.spec = &tag_spec,
3401 		};
3402 		items[1] = (struct rte_flow_item){
3403 			.type = RTE_FLOW_ITEM_TYPE_END,
3404 		};
3405 		actions[0] = (struct rte_flow_action){
3406 			.type = (enum rte_flow_action_type)
3407 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3408 			.conf = &ftag,
3409 		};
3410 		actions[1] = (struct rte_flow_action){
3411 			.type = (enum rte_flow_action_type)
3412 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3413 			.conf = &cp_mreg,
3414 		};
3415 		actions[2] = (struct rte_flow_action){
3416 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3417 			.conf = &jump,
3418 		};
3419 		actions[3] = (struct rte_flow_action){
3420 			.type = RTE_FLOW_ACTION_TYPE_END,
3421 		};
3422 	} else {
3423 		/* Default rule, wildcard match. */
3424 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3425 		items[0] = (struct rte_flow_item){
3426 			.type = RTE_FLOW_ITEM_TYPE_END,
3427 		};
3428 		actions[0] = (struct rte_flow_action){
3429 			.type = (enum rte_flow_action_type)
3430 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3431 			.conf = &cp_mreg,
3432 		};
3433 		actions[1] = (struct rte_flow_action){
3434 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3435 			.conf = &jump,
3436 		};
3437 		actions[2] = (struct rte_flow_action){
3438 			.type = RTE_FLOW_ACTION_TYPE_END,
3439 		};
3440 	}
3441 	/* Build a new entry. */
3442 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3443 	if (!mcp_res) {
3444 		rte_errno = ENOMEM;
3445 		return NULL;
3446 	}
3447 	mcp_res->idx = idx;
3448 	/*
3449 	 * The copy Flows are not included in any list. There
3450 	 * ones are referenced from other Flows and can not
3451 	 * be applied, removed, deleted in ardbitrary order
3452 	 * by list traversing.
3453 	 */
3454 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3455 					 actions, false, error);
3456 	if (!mcp_res->rix_flow)
3457 		goto error;
3458 	mcp_res->refcnt++;
3459 	mcp_res->hlist_ent.key = mark_id;
3460 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3461 				&mcp_res->hlist_ent);
3462 	MLX5_ASSERT(!ret);
3463 	if (ret)
3464 		goto error;
3465 	return mcp_res;
3466 error:
3467 	if (mcp_res->rix_flow)
3468 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3469 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3470 	return NULL;
3471 }
3472 
3473 /**
3474  * Release flow in RX_CP_TBL.
3475  *
3476  * @param dev
3477  *   Pointer to Ethernet device.
3478  * @flow
3479  *   Parent flow for wich copying is provided.
3480  */
3481 static void
3482 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3483 			  struct rte_flow *flow)
3484 {
3485 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3486 	struct mlx5_priv *priv = dev->data->dev_private;
3487 
3488 	if (!flow->rix_mreg_copy)
3489 		return;
3490 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3491 				 flow->rix_mreg_copy);
3492 	if (!mcp_res || !priv->mreg_cp_tbl)
3493 		return;
3494 	if (flow->copy_applied) {
3495 		MLX5_ASSERT(mcp_res->appcnt);
3496 		flow->copy_applied = 0;
3497 		--mcp_res->appcnt;
3498 		if (!mcp_res->appcnt) {
3499 			struct rte_flow *mcp_flow = mlx5_ipool_get
3500 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3501 					mcp_res->rix_flow);
3502 
3503 			if (mcp_flow)
3504 				flow_drv_remove(dev, mcp_flow);
3505 		}
3506 	}
3507 	/*
3508 	 * We do not check availability of metadata registers here,
3509 	 * because copy resources are not allocated in this case.
3510 	 */
3511 	if (--mcp_res->refcnt)
3512 		return;
3513 	MLX5_ASSERT(mcp_res->rix_flow);
3514 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3515 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3516 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3517 	flow->rix_mreg_copy = 0;
3518 }
3519 
3520 /**
3521  * Start flow in RX_CP_TBL.
3522  *
3523  * @param dev
3524  *   Pointer to Ethernet device.
3525  * @flow
3526  *   Parent flow for wich copying is provided.
3527  *
3528  * @return
3529  *   0 on success, a negative errno value otherwise and rte_errno is set.
3530  */
3531 static int
3532 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3533 			    struct rte_flow *flow)
3534 {
3535 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3536 	struct mlx5_priv *priv = dev->data->dev_private;
3537 	int ret;
3538 
3539 	if (!flow->rix_mreg_copy || flow->copy_applied)
3540 		return 0;
3541 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3542 				 flow->rix_mreg_copy);
3543 	if (!mcp_res)
3544 		return 0;
3545 	if (!mcp_res->appcnt) {
3546 		struct rte_flow *mcp_flow = mlx5_ipool_get
3547 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3548 				mcp_res->rix_flow);
3549 
3550 		if (mcp_flow) {
3551 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3552 			if (ret)
3553 				return ret;
3554 		}
3555 	}
3556 	++mcp_res->appcnt;
3557 	flow->copy_applied = 1;
3558 	return 0;
3559 }
3560 
3561 /**
3562  * Stop flow in RX_CP_TBL.
3563  *
3564  * @param dev
3565  *   Pointer to Ethernet device.
3566  * @flow
3567  *   Parent flow for wich copying is provided.
3568  */
3569 static void
3570 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3571 			   struct rte_flow *flow)
3572 {
3573 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3574 	struct mlx5_priv *priv = dev->data->dev_private;
3575 
3576 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3577 		return;
3578 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3579 				 flow->rix_mreg_copy);
3580 	if (!mcp_res)
3581 		return;
3582 	MLX5_ASSERT(mcp_res->appcnt);
3583 	--mcp_res->appcnt;
3584 	flow->copy_applied = 0;
3585 	if (!mcp_res->appcnt) {
3586 		struct rte_flow *mcp_flow = mlx5_ipool_get
3587 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3588 				mcp_res->rix_flow);
3589 
3590 		if (mcp_flow)
3591 			flow_drv_remove(dev, mcp_flow);
3592 	}
3593 }
3594 
3595 /**
3596  * Remove the default copy action from RX_CP_TBL.
3597  *
3598  * @param dev
3599  *   Pointer to Ethernet device.
3600  */
3601 static void
3602 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3603 {
3604 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3605 	struct mlx5_priv *priv = dev->data->dev_private;
3606 
3607 	/* Check if default flow is registered. */
3608 	if (!priv->mreg_cp_tbl)
3609 		return;
3610 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3611 					    MLX5_DEFAULT_COPY_ID);
3612 	if (!mcp_res)
3613 		return;
3614 	MLX5_ASSERT(mcp_res->rix_flow);
3615 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3616 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3617 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3618 }
3619 
3620 /**
3621  * Add the default copy action in in RX_CP_TBL.
3622  *
3623  * @param dev
3624  *   Pointer to Ethernet device.
3625  * @param[out] error
3626  *   Perform verbose error reporting if not NULL.
3627  *
3628  * @return
3629  *   0 for success, negative value otherwise and rte_errno is set.
3630  */
3631 static int
3632 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3633 				  struct rte_flow_error *error)
3634 {
3635 	struct mlx5_priv *priv = dev->data->dev_private;
3636 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3637 
3638 	/* Check whether extensive metadata feature is engaged. */
3639 	if (!priv->config.dv_flow_en ||
3640 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3641 	    !mlx5_flow_ext_mreg_supported(dev) ||
3642 	    !priv->sh->dv_regc0_mask)
3643 		return 0;
3644 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3645 	if (!mcp_res)
3646 		return -rte_errno;
3647 	return 0;
3648 }
3649 
3650 /**
3651  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3652  *
3653  * All the flow having Q/RSS action should be split by
3654  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3655  * performs the following,
3656  *   - CQE->flow_tag := reg_c[1] (MARK)
3657  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3658  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3659  * but there should be a flow per each MARK ID set by MARK action.
3660  *
3661  * For the aforementioned reason, if there's a MARK action in flow's action
3662  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3663  * the MARK ID to CQE's flow_tag like,
3664  *   - If reg_c[1] is mark_id,
3665  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3666  *
3667  * For SET_META action which stores value in reg_c[0], as the destination is
3668  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3669  * MARK ID means the default flow. The default flow looks like,
3670  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3671  *
3672  * @param dev
3673  *   Pointer to Ethernet device.
3674  * @param flow
3675  *   Pointer to flow structure.
3676  * @param[in] actions
3677  *   Pointer to the list of actions.
3678  * @param[out] error
3679  *   Perform verbose error reporting if not NULL.
3680  *
3681  * @return
3682  *   0 on success, negative value otherwise and rte_errno is set.
3683  */
3684 static int
3685 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3686 			    struct rte_flow *flow,
3687 			    const struct rte_flow_action *actions,
3688 			    struct rte_flow_error *error)
3689 {
3690 	struct mlx5_priv *priv = dev->data->dev_private;
3691 	struct mlx5_dev_config *config = &priv->config;
3692 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3693 	const struct rte_flow_action_mark *mark;
3694 
3695 	/* Check whether extensive metadata feature is engaged. */
3696 	if (!config->dv_flow_en ||
3697 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3698 	    !mlx5_flow_ext_mreg_supported(dev) ||
3699 	    !priv->sh->dv_regc0_mask)
3700 		return 0;
3701 	/* Find MARK action. */
3702 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3703 		switch (actions->type) {
3704 		case RTE_FLOW_ACTION_TYPE_FLAG:
3705 			mcp_res = flow_mreg_add_copy_action
3706 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3707 			if (!mcp_res)
3708 				return -rte_errno;
3709 			flow->rix_mreg_copy = mcp_res->idx;
3710 			if (dev->data->dev_started) {
3711 				mcp_res->appcnt++;
3712 				flow->copy_applied = 1;
3713 			}
3714 			return 0;
3715 		case RTE_FLOW_ACTION_TYPE_MARK:
3716 			mark = (const struct rte_flow_action_mark *)
3717 				actions->conf;
3718 			mcp_res =
3719 				flow_mreg_add_copy_action(dev, mark->id, error);
3720 			if (!mcp_res)
3721 				return -rte_errno;
3722 			flow->rix_mreg_copy = mcp_res->idx;
3723 			if (dev->data->dev_started) {
3724 				mcp_res->appcnt++;
3725 				flow->copy_applied = 1;
3726 			}
3727 			return 0;
3728 		default:
3729 			break;
3730 		}
3731 	}
3732 	return 0;
3733 }
3734 
3735 #define MLX5_MAX_SPLIT_ACTIONS 24
3736 #define MLX5_MAX_SPLIT_ITEMS 24
3737 
3738 /**
3739  * Split the hairpin flow.
3740  * Since HW can't support encap and push-vlan on Rx, we move these
3741  * actions to Tx.
3742  * If the count action is after the encap then we also
3743  * move the count action. in this case the count will also measure
3744  * the outer bytes.
3745  *
3746  * @param dev
3747  *   Pointer to Ethernet device.
3748  * @param[in] actions
3749  *   Associated actions (list terminated by the END action).
3750  * @param[out] actions_rx
3751  *   Rx flow actions.
3752  * @param[out] actions_tx
3753  *   Tx flow actions..
3754  * @param[out] pattern_tx
3755  *   The pattern items for the Tx flow.
3756  * @param[out] flow_id
3757  *   The flow ID connected to this flow.
3758  *
3759  * @return
3760  *   0 on success.
3761  */
3762 static int
3763 flow_hairpin_split(struct rte_eth_dev *dev,
3764 		   const struct rte_flow_action actions[],
3765 		   struct rte_flow_action actions_rx[],
3766 		   struct rte_flow_action actions_tx[],
3767 		   struct rte_flow_item pattern_tx[],
3768 		   uint32_t *flow_id)
3769 {
3770 	struct mlx5_priv *priv = dev->data->dev_private;
3771 	const struct rte_flow_action_raw_encap *raw_encap;
3772 	const struct rte_flow_action_raw_decap *raw_decap;
3773 	struct mlx5_rte_flow_action_set_tag *set_tag;
3774 	struct rte_flow_action *tag_action;
3775 	struct mlx5_rte_flow_item_tag *tag_item;
3776 	struct rte_flow_item *item;
3777 	char *addr;
3778 	int encap = 0;
3779 
3780 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3781 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3782 		switch (actions->type) {
3783 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3784 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3785 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3786 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3787 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3788 			rte_memcpy(actions_tx, actions,
3789 			       sizeof(struct rte_flow_action));
3790 			actions_tx++;
3791 			break;
3792 		case RTE_FLOW_ACTION_TYPE_COUNT:
3793 			if (encap) {
3794 				rte_memcpy(actions_tx, actions,
3795 					   sizeof(struct rte_flow_action));
3796 				actions_tx++;
3797 			} else {
3798 				rte_memcpy(actions_rx, actions,
3799 					   sizeof(struct rte_flow_action));
3800 				actions_rx++;
3801 			}
3802 			break;
3803 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3804 			raw_encap = actions->conf;
3805 			if (raw_encap->size >
3806 			    (sizeof(struct rte_flow_item_eth) +
3807 			     sizeof(struct rte_flow_item_ipv4))) {
3808 				memcpy(actions_tx, actions,
3809 				       sizeof(struct rte_flow_action));
3810 				actions_tx++;
3811 				encap = 1;
3812 			} else {
3813 				rte_memcpy(actions_rx, actions,
3814 					   sizeof(struct rte_flow_action));
3815 				actions_rx++;
3816 			}
3817 			break;
3818 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3819 			raw_decap = actions->conf;
3820 			if (raw_decap->size <
3821 			    (sizeof(struct rte_flow_item_eth) +
3822 			     sizeof(struct rte_flow_item_ipv4))) {
3823 				memcpy(actions_tx, actions,
3824 				       sizeof(struct rte_flow_action));
3825 				actions_tx++;
3826 			} else {
3827 				rte_memcpy(actions_rx, actions,
3828 					   sizeof(struct rte_flow_action));
3829 				actions_rx++;
3830 			}
3831 			break;
3832 		default:
3833 			rte_memcpy(actions_rx, actions,
3834 				   sizeof(struct rte_flow_action));
3835 			actions_rx++;
3836 			break;
3837 		}
3838 	}
3839 	/* Add set meta action and end action for the Rx flow. */
3840 	tag_action = actions_rx;
3841 	tag_action->type = (enum rte_flow_action_type)
3842 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3843 	actions_rx++;
3844 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3845 	actions_rx++;
3846 	set_tag = (void *)actions_rx;
3847 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3848 	MLX5_ASSERT(set_tag->id > REG_NON);
3849 	set_tag->data = *flow_id;
3850 	tag_action->conf = set_tag;
3851 	/* Create Tx item list. */
3852 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3853 	addr = (void *)&pattern_tx[2];
3854 	item = pattern_tx;
3855 	item->type = (enum rte_flow_item_type)
3856 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3857 	tag_item = (void *)addr;
3858 	tag_item->data = *flow_id;
3859 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3860 	MLX5_ASSERT(set_tag->id > REG_NON);
3861 	item->spec = tag_item;
3862 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3863 	tag_item = (void *)addr;
3864 	tag_item->data = UINT32_MAX;
3865 	tag_item->id = UINT16_MAX;
3866 	item->mask = tag_item;
3867 	item->last = NULL;
3868 	item++;
3869 	item->type = RTE_FLOW_ITEM_TYPE_END;
3870 	return 0;
3871 }
3872 
3873 /**
3874  * The last stage of splitting chain, just creates the subflow
3875  * without any modification.
3876  *
3877  * @param[in] dev
3878  *   Pointer to Ethernet device.
3879  * @param[in] flow
3880  *   Parent flow structure pointer.
3881  * @param[in, out] sub_flow
3882  *   Pointer to return the created subflow, may be NULL.
3883  * @param[in] prefix_layers
3884  *   Prefix subflow layers, may be 0.
3885  * @param[in] prefix_mark
3886  *   Prefix subflow mark flag, may be 0.
3887  * @param[in] attr
3888  *   Flow rule attributes.
3889  * @param[in] items
3890  *   Pattern specification (list terminated by the END pattern item).
3891  * @param[in] actions
3892  *   Associated actions (list terminated by the END action).
3893  * @param[in] external
3894  *   This flow rule is created by request external to PMD.
3895  * @param[in] flow_idx
3896  *   This memory pool index to the flow.
3897  * @param[out] error
3898  *   Perform verbose error reporting if not NULL.
3899  * @return
3900  *   0 on success, negative value otherwise
3901  */
3902 static int
3903 flow_create_split_inner(struct rte_eth_dev *dev,
3904 			struct rte_flow *flow,
3905 			struct mlx5_flow **sub_flow,
3906 			uint64_t prefix_layers,
3907 			uint32_t prefix_mark,
3908 			const struct rte_flow_attr *attr,
3909 			const struct rte_flow_item items[],
3910 			const struct rte_flow_action actions[],
3911 			bool external, uint32_t flow_idx,
3912 			struct rte_flow_error *error)
3913 {
3914 	struct mlx5_flow *dev_flow;
3915 
3916 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3917 		flow_idx, error);
3918 	if (!dev_flow)
3919 		return -rte_errno;
3920 	dev_flow->flow = flow;
3921 	dev_flow->external = external;
3922 	/* Subflow object was created, we must include one in the list. */
3923 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3924 		      dev_flow->handle, next);
3925 	/*
3926 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3927 	 * flow may need some user defined item layer flags, and pass the
3928 	 * Metadate rxq mark flag to suffix flow as well.
3929 	 */
3930 	if (prefix_layers)
3931 		dev_flow->handle->layers = prefix_layers;
3932 	if (prefix_mark)
3933 		dev_flow->handle->mark = 1;
3934 	if (sub_flow)
3935 		*sub_flow = dev_flow;
3936 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3937 }
3938 
3939 /**
3940  * Split the meter flow.
3941  *
3942  * As meter flow will split to three sub flow, other than meter
3943  * action, the other actions make sense to only meter accepts
3944  * the packet. If it need to be dropped, no other additional
3945  * actions should be take.
3946  *
3947  * One kind of special action which decapsulates the L3 tunnel
3948  * header will be in the prefix sub flow, as not to take the
3949  * L3 tunnel header into account.
3950  *
3951  * @param dev
3952  *   Pointer to Ethernet device.
3953  * @param[in] items
3954  *   Pattern specification (list terminated by the END pattern item).
3955  * @param[out] sfx_items
3956  *   Suffix flow match items (list terminated by the END pattern item).
3957  * @param[in] actions
3958  *   Associated actions (list terminated by the END action).
3959  * @param[out] actions_sfx
3960  *   Suffix flow actions.
3961  * @param[out] actions_pre
3962  *   Prefix flow actions.
3963  * @param[out] pattern_sfx
3964  *   The pattern items for the suffix flow.
3965  * @param[out] tag_sfx
3966  *   Pointer to suffix flow tag.
3967  *
3968  * @return
3969  *   0 on success.
3970  */
3971 static int
3972 flow_meter_split_prep(struct rte_eth_dev *dev,
3973 		 const struct rte_flow_item items[],
3974 		 struct rte_flow_item sfx_items[],
3975 		 const struct rte_flow_action actions[],
3976 		 struct rte_flow_action actions_sfx[],
3977 		 struct rte_flow_action actions_pre[])
3978 {
3979 	struct rte_flow_action *tag_action = NULL;
3980 	struct rte_flow_item *tag_item;
3981 	struct mlx5_rte_flow_action_set_tag *set_tag;
3982 	struct rte_flow_error error;
3983 	const struct rte_flow_action_raw_encap *raw_encap;
3984 	const struct rte_flow_action_raw_decap *raw_decap;
3985 	struct mlx5_rte_flow_item_tag *tag_spec;
3986 	struct mlx5_rte_flow_item_tag *tag_mask;
3987 	uint32_t tag_id;
3988 	bool copy_vlan = false;
3989 
3990 	/* Prepare the actions for prefix and suffix flow. */
3991 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3992 		struct rte_flow_action **action_cur = NULL;
3993 
3994 		switch (actions->type) {
3995 		case RTE_FLOW_ACTION_TYPE_METER:
3996 			/* Add the extra tag action first. */
3997 			tag_action = actions_pre;
3998 			tag_action->type = (enum rte_flow_action_type)
3999 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4000 			actions_pre++;
4001 			action_cur = &actions_pre;
4002 			break;
4003 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4004 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4005 			action_cur = &actions_pre;
4006 			break;
4007 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4008 			raw_encap = actions->conf;
4009 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4010 				action_cur = &actions_pre;
4011 			break;
4012 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4013 			raw_decap = actions->conf;
4014 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4015 				action_cur = &actions_pre;
4016 			break;
4017 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4018 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4019 			copy_vlan = true;
4020 			break;
4021 		default:
4022 			break;
4023 		}
4024 		if (!action_cur)
4025 			action_cur = &actions_sfx;
4026 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4027 		(*action_cur)++;
4028 	}
4029 	/* Add end action to the actions. */
4030 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4031 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4032 	actions_pre++;
4033 	/* Set the tag. */
4034 	set_tag = (void *)actions_pre;
4035 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4036 	/*
4037 	 * Get the id from the qrss_pool to make qrss share the id with meter.
4038 	 */
4039 	tag_id = flow_qrss_get_id(dev);
4040 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4041 	assert(tag_action);
4042 	tag_action->conf = set_tag;
4043 	/* Prepare the suffix subflow items. */
4044 	tag_item = sfx_items++;
4045 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4046 		int item_type = items->type;
4047 
4048 		switch (item_type) {
4049 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4050 			memcpy(sfx_items, items, sizeof(*sfx_items));
4051 			sfx_items++;
4052 			break;
4053 		case RTE_FLOW_ITEM_TYPE_VLAN:
4054 			if (copy_vlan) {
4055 				memcpy(sfx_items, items, sizeof(*sfx_items));
4056 				/*
4057 				 * Convert to internal match item, it is used
4058 				 * for vlan push and set vid.
4059 				 */
4060 				sfx_items->type = (enum rte_flow_item_type)
4061 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4062 				sfx_items++;
4063 			}
4064 			break;
4065 		default:
4066 			break;
4067 		}
4068 	}
4069 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4070 	sfx_items++;
4071 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4072 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4073 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4074 	tag_mask = tag_spec + 1;
4075 	tag_mask->data = 0xffffff00;
4076 	tag_item->type = (enum rte_flow_item_type)
4077 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4078 	tag_item->spec = tag_spec;
4079 	tag_item->last = NULL;
4080 	tag_item->mask = tag_mask;
4081 	return tag_id;
4082 }
4083 
4084 /**
4085  * Split action list having QUEUE/RSS for metadata register copy.
4086  *
4087  * Once Q/RSS action is detected in user's action list, the flow action
4088  * should be split in order to copy metadata registers, which will happen in
4089  * RX_CP_TBL like,
4090  *   - CQE->flow_tag := reg_c[1] (MARK)
4091  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4092  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4093  * This is because the last action of each flow must be a terminal action
4094  * (QUEUE, RSS or DROP).
4095  *
4096  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4097  * stored and kept in the mlx5_flow structure per each sub_flow.
4098  *
4099  * The Q/RSS action is replaced with,
4100  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4101  * And the following JUMP action is added at the end,
4102  *   - JUMP, to RX_CP_TBL.
4103  *
4104  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4105  * flow_create_split_metadata() routine. The flow will look like,
4106  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4107  *
4108  * @param dev
4109  *   Pointer to Ethernet device.
4110  * @param[out] split_actions
4111  *   Pointer to store split actions to jump to CP_TBL.
4112  * @param[in] actions
4113  *   Pointer to the list of original flow actions.
4114  * @param[in] qrss
4115  *   Pointer to the Q/RSS action.
4116  * @param[in] actions_n
4117  *   Number of original actions.
4118  * @param[out] error
4119  *   Perform verbose error reporting if not NULL.
4120  *
4121  * @return
4122  *   non-zero unique flow_id on success, otherwise 0 and
4123  *   error/rte_error are set.
4124  */
4125 static uint32_t
4126 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4127 			  struct rte_flow_action *split_actions,
4128 			  const struct rte_flow_action *actions,
4129 			  const struct rte_flow_action *qrss,
4130 			  int actions_n, struct rte_flow_error *error)
4131 {
4132 	struct mlx5_rte_flow_action_set_tag *set_tag;
4133 	struct rte_flow_action_jump *jump;
4134 	const int qrss_idx = qrss - actions;
4135 	uint32_t flow_id = 0;
4136 	int ret = 0;
4137 
4138 	/*
4139 	 * Given actions will be split
4140 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4141 	 * - Add jump to mreg CP_TBL.
4142 	 * As a result, there will be one more action.
4143 	 */
4144 	++actions_n;
4145 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4146 	set_tag = (void *)(split_actions + actions_n);
4147 	/*
4148 	 * If tag action is not set to void(it means we are not the meter
4149 	 * suffix flow), add the tag action. Since meter suffix flow already
4150 	 * has the tag added.
4151 	 */
4152 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4153 		/*
4154 		 * Allocate the new subflow ID. This one is unique within
4155 		 * device and not shared with representors. Otherwise,
4156 		 * we would have to resolve multi-thread access synch
4157 		 * issue. Each flow on the shared device is appended
4158 		 * with source vport identifier, so the resulting
4159 		 * flows will be unique in the shared (by master and
4160 		 * representors) domain even if they have coinciding
4161 		 * IDs.
4162 		 */
4163 		flow_id = flow_qrss_get_id(dev);
4164 		if (!flow_id)
4165 			return rte_flow_error_set(error, ENOMEM,
4166 						  RTE_FLOW_ERROR_TYPE_ACTION,
4167 						  NULL, "can't allocate id "
4168 						  "for split Q/RSS subflow");
4169 		/* Internal SET_TAG action to set flow ID. */
4170 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4171 			.data = flow_id,
4172 		};
4173 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4174 		if (ret < 0)
4175 			return ret;
4176 		set_tag->id = ret;
4177 		/* Construct new actions array. */
4178 		/* Replace QUEUE/RSS action. */
4179 		split_actions[qrss_idx] = (struct rte_flow_action){
4180 			.type = (enum rte_flow_action_type)
4181 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4182 			.conf = set_tag,
4183 		};
4184 	}
4185 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4186 	jump = (void *)(set_tag + 1);
4187 	*jump = (struct rte_flow_action_jump){
4188 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4189 	};
4190 	split_actions[actions_n - 2] = (struct rte_flow_action){
4191 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4192 		.conf = jump,
4193 	};
4194 	split_actions[actions_n - 1] = (struct rte_flow_action){
4195 		.type = RTE_FLOW_ACTION_TYPE_END,
4196 	};
4197 	return flow_id;
4198 }
4199 
4200 /**
4201  * Extend the given action list for Tx metadata copy.
4202  *
4203  * Copy the given action list to the ext_actions and add flow metadata register
4204  * copy action in order to copy reg_a set by WQE to reg_c[0].
4205  *
4206  * @param[out] ext_actions
4207  *   Pointer to the extended action list.
4208  * @param[in] actions
4209  *   Pointer to the list of actions.
4210  * @param[in] actions_n
4211  *   Number of actions in the list.
4212  * @param[out] error
4213  *   Perform verbose error reporting if not NULL.
4214  * @param[in] encap_idx
4215  *   The encap action inndex.
4216  *
4217  * @return
4218  *   0 on success, negative value otherwise
4219  */
4220 static int
4221 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4222 		       struct rte_flow_action *ext_actions,
4223 		       const struct rte_flow_action *actions,
4224 		       int actions_n, struct rte_flow_error *error,
4225 		       int encap_idx)
4226 {
4227 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4228 		(struct mlx5_flow_action_copy_mreg *)
4229 			(ext_actions + actions_n + 1);
4230 	int ret;
4231 
4232 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4233 	if (ret < 0)
4234 		return ret;
4235 	cp_mreg->dst = ret;
4236 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4237 	if (ret < 0)
4238 		return ret;
4239 	cp_mreg->src = ret;
4240 	if (encap_idx != 0)
4241 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4242 	if (encap_idx == actions_n - 1) {
4243 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4244 			.type = (enum rte_flow_action_type)
4245 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4246 			.conf = cp_mreg,
4247 		};
4248 		ext_actions[actions_n] = (struct rte_flow_action){
4249 			.type = RTE_FLOW_ACTION_TYPE_END,
4250 		};
4251 	} else {
4252 		ext_actions[encap_idx] = (struct rte_flow_action){
4253 			.type = (enum rte_flow_action_type)
4254 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4255 			.conf = cp_mreg,
4256 		};
4257 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4258 				sizeof(*ext_actions) * (actions_n - encap_idx));
4259 	}
4260 	return 0;
4261 }
4262 
4263 /**
4264  * Check the match action from the action list.
4265  *
4266  * @param[in] actions
4267  *   Pointer to the list of actions.
4268  * @param[in] attr
4269  *   Flow rule attributes.
4270  * @param[in] action
4271  *   The action to be check if exist.
4272  * @param[out] match_action_pos
4273  *   Pointer to the position of the matched action if exists, otherwise is -1.
4274  * @param[out] qrss_action_pos
4275  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4276  *
4277  * @return
4278  *   > 0 the total number of actions.
4279  *   0 if not found match action in action list.
4280  */
4281 static int
4282 flow_check_match_action(const struct rte_flow_action actions[],
4283 			const struct rte_flow_attr *attr,
4284 			enum rte_flow_action_type action,
4285 			int *match_action_pos, int *qrss_action_pos)
4286 {
4287 	const struct rte_flow_action_sample *sample;
4288 	int actions_n = 0;
4289 	int jump_flag = 0;
4290 	uint32_t ratio = 0;
4291 	int sub_type = 0;
4292 	int flag = 0;
4293 
4294 	*match_action_pos = -1;
4295 	*qrss_action_pos = -1;
4296 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4297 		if (actions->type == action) {
4298 			flag = 1;
4299 			*match_action_pos = actions_n;
4300 		}
4301 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4302 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4303 			*qrss_action_pos = actions_n;
4304 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4305 			jump_flag = 1;
4306 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4307 			sample = actions->conf;
4308 			ratio = sample->ratio;
4309 			sub_type = ((const struct rte_flow_action *)
4310 					(sample->actions))->type;
4311 		}
4312 		actions_n++;
4313 	}
4314 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4315 		if (ratio == 1) {
4316 			/* JUMP Action not support for Mirroring;
4317 			 * Mirroring support multi-destination;
4318 			 */
4319 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4320 				flag = 0;
4321 		}
4322 	}
4323 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4324 	return flag ? actions_n + 1 : 0;
4325 }
4326 
4327 #define SAMPLE_SUFFIX_ITEM 2
4328 
4329 /**
4330  * Split the sample flow.
4331  *
4332  * As sample flow will split to two sub flow, sample flow with
4333  * sample action, the other actions will move to new suffix flow.
4334  *
4335  * Also add unique tag id with tag action in the sample flow,
4336  * the same tag id will be as match in the suffix flow.
4337  *
4338  * @param dev
4339  *   Pointer to Ethernet device.
4340  * @param[in] fdb_tx
4341  *   FDB egress flow flag.
4342  * @param[out] sfx_items
4343  *   Suffix flow match items (list terminated by the END pattern item).
4344  * @param[in] actions
4345  *   Associated actions (list terminated by the END action).
4346  * @param[out] actions_sfx
4347  *   Suffix flow actions.
4348  * @param[out] actions_pre
4349  *   Prefix flow actions.
4350  * @param[in] actions_n
4351  *  The total number of actions.
4352  * @param[in] sample_action_pos
4353  *   The sample action position.
4354  * @param[in] qrss_action_pos
4355  *   The Queue/RSS action position.
4356  * @param[out] error
4357  *   Perform verbose error reporting if not NULL.
4358  *
4359  * @return
4360  *   0 on success, or unique flow_id, a negative errno value
4361  *   otherwise and rte_errno is set.
4362  */
4363 static int
4364 flow_sample_split_prep(struct rte_eth_dev *dev,
4365 		       uint32_t fdb_tx,
4366 		       struct rte_flow_item sfx_items[],
4367 		       const struct rte_flow_action actions[],
4368 		       struct rte_flow_action actions_sfx[],
4369 		       struct rte_flow_action actions_pre[],
4370 		       int actions_n,
4371 		       int sample_action_pos,
4372 		       int qrss_action_pos,
4373 		       struct rte_flow_error *error)
4374 {
4375 	struct mlx5_rte_flow_action_set_tag *set_tag;
4376 	struct mlx5_rte_flow_item_tag *tag_spec;
4377 	struct mlx5_rte_flow_item_tag *tag_mask;
4378 	uint32_t tag_id = 0;
4379 	int index;
4380 	int ret;
4381 
4382 	if (sample_action_pos < 0)
4383 		return rte_flow_error_set(error, EINVAL,
4384 					  RTE_FLOW_ERROR_TYPE_ACTION,
4385 					  NULL, "invalid position of sample "
4386 					  "action in list");
4387 	if (!fdb_tx) {
4388 		/* Prepare the prefix tag action. */
4389 		set_tag = (void *)(actions_pre + actions_n + 1);
4390 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4391 		if (ret < 0)
4392 			return ret;
4393 		set_tag->id = ret;
4394 		tag_id = flow_qrss_get_id(dev);
4395 		set_tag->data = tag_id;
4396 		/* Prepare the suffix subflow items. */
4397 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4398 		tag_spec->data = tag_id;
4399 		tag_spec->id = set_tag->id;
4400 		tag_mask = tag_spec + 1;
4401 		tag_mask->data = UINT32_MAX;
4402 		sfx_items[0] = (struct rte_flow_item){
4403 			.type = (enum rte_flow_item_type)
4404 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4405 			.spec = tag_spec,
4406 			.last = NULL,
4407 			.mask = tag_mask,
4408 		};
4409 		sfx_items[1] = (struct rte_flow_item){
4410 			.type = (enum rte_flow_item_type)
4411 				RTE_FLOW_ITEM_TYPE_END,
4412 		};
4413 	}
4414 	/* Prepare the actions for prefix and suffix flow. */
4415 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4416 		index = qrss_action_pos;
4417 		/* Put the preceding the Queue/RSS action into prefix flow. */
4418 		if (index != 0)
4419 			memcpy(actions_pre, actions,
4420 			       sizeof(struct rte_flow_action) * index);
4421 		/* Put others preceding the sample action into prefix flow. */
4422 		if (sample_action_pos > index + 1)
4423 			memcpy(actions_pre + index, actions + index + 1,
4424 			       sizeof(struct rte_flow_action) *
4425 			       (sample_action_pos - index - 1));
4426 		index = sample_action_pos - 1;
4427 		/* Put Queue/RSS action into Suffix flow. */
4428 		memcpy(actions_sfx, actions + qrss_action_pos,
4429 		       sizeof(struct rte_flow_action));
4430 		actions_sfx++;
4431 	} else {
4432 		index = sample_action_pos;
4433 		if (index != 0)
4434 			memcpy(actions_pre, actions,
4435 			       sizeof(struct rte_flow_action) * index);
4436 	}
4437 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4438 	if (!fdb_tx) {
4439 		actions_pre[index++] =
4440 			(struct rte_flow_action){
4441 			.type = (enum rte_flow_action_type)
4442 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4443 			.conf = set_tag,
4444 		};
4445 	}
4446 	memcpy(actions_pre + index, actions + sample_action_pos,
4447 	       sizeof(struct rte_flow_action));
4448 	index += 1;
4449 	actions_pre[index] = (struct rte_flow_action){
4450 		.type = (enum rte_flow_action_type)
4451 			RTE_FLOW_ACTION_TYPE_END,
4452 	};
4453 	/* Put the actions after sample into Suffix flow. */
4454 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4455 	       sizeof(struct rte_flow_action) *
4456 	       (actions_n - sample_action_pos - 1));
4457 	return tag_id;
4458 }
4459 
4460 /**
4461  * The splitting for metadata feature.
4462  *
4463  * - Q/RSS action on NIC Rx should be split in order to pass by
4464  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4465  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4466  *
4467  * - All the actions on NIC Tx should have a mreg copy action to
4468  *   copy reg_a from WQE to reg_c[0].
4469  *
4470  * @param dev
4471  *   Pointer to Ethernet device.
4472  * @param[in] flow
4473  *   Parent flow structure pointer.
4474  * @param[in] prefix_layers
4475  *   Prefix flow layer flags.
4476  * @param[in] prefix_mark
4477  *   Prefix subflow mark flag, may be 0.
4478  * @param[in] attr
4479  *   Flow rule attributes.
4480  * @param[in] items
4481  *   Pattern specification (list terminated by the END pattern item).
4482  * @param[in] actions
4483  *   Associated actions (list terminated by the END action).
4484  * @param[in] external
4485  *   This flow rule is created by request external to PMD.
4486  * @param[in] flow_idx
4487  *   This memory pool index to the flow.
4488  * @param[out] error
4489  *   Perform verbose error reporting if not NULL.
4490  * @return
4491  *   0 on success, negative value otherwise
4492  */
4493 static int
4494 flow_create_split_metadata(struct rte_eth_dev *dev,
4495 			   struct rte_flow *flow,
4496 			   uint64_t prefix_layers,
4497 			   uint32_t prefix_mark,
4498 			   const struct rte_flow_attr *attr,
4499 			   const struct rte_flow_item items[],
4500 			   const struct rte_flow_action actions[],
4501 			   bool external, uint32_t flow_idx,
4502 			   struct rte_flow_error *error)
4503 {
4504 	struct mlx5_priv *priv = dev->data->dev_private;
4505 	struct mlx5_dev_config *config = &priv->config;
4506 	const struct rte_flow_action *qrss = NULL;
4507 	struct rte_flow_action *ext_actions = NULL;
4508 	struct mlx5_flow *dev_flow = NULL;
4509 	uint32_t qrss_id = 0;
4510 	int mtr_sfx = 0;
4511 	size_t act_size;
4512 	int actions_n;
4513 	int encap_idx;
4514 	int ret;
4515 
4516 	/* Check whether extensive metadata feature is engaged. */
4517 	if (!config->dv_flow_en ||
4518 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4519 	    !mlx5_flow_ext_mreg_supported(dev))
4520 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
4521 					       prefix_mark, attr, items,
4522 					       actions, external, flow_idx,
4523 					       error);
4524 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4525 							   &encap_idx);
4526 	if (qrss) {
4527 		/* Exclude hairpin flows from splitting. */
4528 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4529 			const struct rte_flow_action_queue *queue;
4530 
4531 			queue = qrss->conf;
4532 			if (mlx5_rxq_get_type(dev, queue->index) ==
4533 			    MLX5_RXQ_TYPE_HAIRPIN)
4534 				qrss = NULL;
4535 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4536 			const struct rte_flow_action_rss *rss;
4537 
4538 			rss = qrss->conf;
4539 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4540 			    MLX5_RXQ_TYPE_HAIRPIN)
4541 				qrss = NULL;
4542 		}
4543 	}
4544 	if (qrss) {
4545 		/* Check if it is in meter suffix table. */
4546 		mtr_sfx = attr->group == (attr->transfer ?
4547 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4548 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4549 		/*
4550 		 * Q/RSS action on NIC Rx should be split in order to pass by
4551 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4552 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4553 		 */
4554 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4555 			   sizeof(struct rte_flow_action_set_tag) +
4556 			   sizeof(struct rte_flow_action_jump);
4557 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4558 					  SOCKET_ID_ANY);
4559 		if (!ext_actions)
4560 			return rte_flow_error_set(error, ENOMEM,
4561 						  RTE_FLOW_ERROR_TYPE_ACTION,
4562 						  NULL, "no memory to split "
4563 						  "metadata flow");
4564 		/*
4565 		 * If we are the suffix flow of meter, tag already exist.
4566 		 * Set the tag action to void.
4567 		 */
4568 		if (mtr_sfx)
4569 			ext_actions[qrss - actions].type =
4570 						RTE_FLOW_ACTION_TYPE_VOID;
4571 		else
4572 			ext_actions[qrss - actions].type =
4573 						(enum rte_flow_action_type)
4574 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4575 		/*
4576 		 * Create the new actions list with removed Q/RSS action
4577 		 * and appended set tag and jump to register copy table
4578 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4579 		 * in advance, because it is needed for set tag action.
4580 		 */
4581 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4582 						    qrss, actions_n, error);
4583 		if (!mtr_sfx && !qrss_id) {
4584 			ret = -rte_errno;
4585 			goto exit;
4586 		}
4587 	} else if (attr->egress && !attr->transfer) {
4588 		/*
4589 		 * All the actions on NIC Tx should have a metadata register
4590 		 * copy action to copy reg_a from WQE to reg_c[meta]
4591 		 */
4592 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4593 			   sizeof(struct mlx5_flow_action_copy_mreg);
4594 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4595 					  SOCKET_ID_ANY);
4596 		if (!ext_actions)
4597 			return rte_flow_error_set(error, ENOMEM,
4598 						  RTE_FLOW_ERROR_TYPE_ACTION,
4599 						  NULL, "no memory to split "
4600 						  "metadata flow");
4601 		/* Create the action list appended with copy register. */
4602 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4603 					     actions_n, error, encap_idx);
4604 		if (ret < 0)
4605 			goto exit;
4606 	}
4607 	/* Add the unmodified original or prefix subflow. */
4608 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers,
4609 				      prefix_mark, attr,
4610 				      items, ext_actions ? ext_actions :
4611 				      actions, external, flow_idx, error);
4612 	if (ret < 0)
4613 		goto exit;
4614 	MLX5_ASSERT(dev_flow);
4615 	if (qrss) {
4616 		const struct rte_flow_attr q_attr = {
4617 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4618 			.ingress = 1,
4619 		};
4620 		/* Internal PMD action to set register. */
4621 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4622 			.data = qrss_id,
4623 			.id = REG_NON,
4624 		};
4625 		struct rte_flow_item q_items[] = {
4626 			{
4627 				.type = (enum rte_flow_item_type)
4628 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4629 				.spec = &q_tag_spec,
4630 				.last = NULL,
4631 				.mask = NULL,
4632 			},
4633 			{
4634 				.type = RTE_FLOW_ITEM_TYPE_END,
4635 			},
4636 		};
4637 		struct rte_flow_action q_actions[] = {
4638 			{
4639 				.type = qrss->type,
4640 				.conf = qrss->conf,
4641 			},
4642 			{
4643 				.type = RTE_FLOW_ACTION_TYPE_END,
4644 			},
4645 		};
4646 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4647 
4648 		/*
4649 		 * Configure the tag item only if there is no meter subflow.
4650 		 * Since tag is already marked in the meter suffix subflow
4651 		 * we can just use the meter suffix items as is.
4652 		 */
4653 		if (qrss_id) {
4654 			/* Not meter subflow. */
4655 			MLX5_ASSERT(!mtr_sfx);
4656 			/*
4657 			 * Put unique id in prefix flow due to it is destroyed
4658 			 * after suffix flow and id will be freed after there
4659 			 * is no actual flows with this id and identifier
4660 			 * reallocation becomes possible (for example, for
4661 			 * other flows in other threads).
4662 			 */
4663 			dev_flow->handle->split_flow_id = qrss_id;
4664 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4665 						   error);
4666 			if (ret < 0)
4667 				goto exit;
4668 			q_tag_spec.id = ret;
4669 		}
4670 		dev_flow = NULL;
4671 		/* Add suffix subflow to execute Q/RSS. */
4672 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers, 0,
4673 					      &q_attr, mtr_sfx ? items :
4674 					      q_items, q_actions,
4675 					      external, flow_idx, error);
4676 		if (ret < 0)
4677 			goto exit;
4678 		/* qrss ID should be freed if failed. */
4679 		qrss_id = 0;
4680 		MLX5_ASSERT(dev_flow);
4681 	}
4682 
4683 exit:
4684 	/*
4685 	 * We do not destroy the partially created sub_flows in case of error.
4686 	 * These ones are included into parent flow list and will be destroyed
4687 	 * by flow_drv_destroy.
4688 	 */
4689 	flow_qrss_free_id(dev, qrss_id);
4690 	mlx5_free(ext_actions);
4691 	return ret;
4692 }
4693 
4694 /**
4695  * The splitting for meter feature.
4696  *
4697  * - The meter flow will be split to two flows as prefix and
4698  *   suffix flow. The packets make sense only it pass the prefix
4699  *   meter action.
4700  *
4701  * - Reg_C_5 is used for the packet to match betweend prefix and
4702  *   suffix flow.
4703  *
4704  * @param dev
4705  *   Pointer to Ethernet device.
4706  * @param[in] flow
4707  *   Parent flow structure pointer.
4708  * @param[in] prefix_layers
4709  *   Prefix subflow layers, may be 0.
4710  * @param[in] prefix_mark
4711  *   Prefix subflow mark flag, may be 0.
4712  * @param[in] attr
4713  *   Flow rule attributes.
4714  * @param[in] items
4715  *   Pattern specification (list terminated by the END pattern item).
4716  * @param[in] actions
4717  *   Associated actions (list terminated by the END action).
4718  * @param[in] external
4719  *   This flow rule is created by request external to PMD.
4720  * @param[in] flow_idx
4721  *   This memory pool index to the flow.
4722  * @param[out] error
4723  *   Perform verbose error reporting if not NULL.
4724  * @return
4725  *   0 on success, negative value otherwise
4726  */
4727 static int
4728 flow_create_split_meter(struct rte_eth_dev *dev,
4729 			struct rte_flow *flow,
4730 			uint64_t prefix_layers,
4731 			uint32_t prefix_mark,
4732 			const struct rte_flow_attr *attr,
4733 			const struct rte_flow_item items[],
4734 			const struct rte_flow_action actions[],
4735 			bool external, uint32_t flow_idx,
4736 			struct rte_flow_error *error)
4737 {
4738 	struct mlx5_priv *priv = dev->data->dev_private;
4739 	struct rte_flow_action *sfx_actions = NULL;
4740 	struct rte_flow_action *pre_actions = NULL;
4741 	struct rte_flow_item *sfx_items = NULL;
4742 	struct mlx5_flow *dev_flow = NULL;
4743 	struct rte_flow_attr sfx_attr = *attr;
4744 	uint32_t mtr = 0;
4745 	uint32_t mtr_tag_id = 0;
4746 	size_t act_size;
4747 	size_t item_size;
4748 	int actions_n = 0;
4749 	int ret;
4750 
4751 	if (priv->mtr_en)
4752 		actions_n = flow_check_meter_action(actions, &mtr);
4753 	if (mtr) {
4754 		/* The five prefix actions: meter, decap, encap, tag, end. */
4755 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4756 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4757 		/* tag, vlan, port id, end. */
4758 #define METER_SUFFIX_ITEM 4
4759 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4760 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4761 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4762 					  0, SOCKET_ID_ANY);
4763 		if (!sfx_actions)
4764 			return rte_flow_error_set(error, ENOMEM,
4765 						  RTE_FLOW_ERROR_TYPE_ACTION,
4766 						  NULL, "no memory to split "
4767 						  "meter flow");
4768 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4769 			     act_size);
4770 		pre_actions = sfx_actions + actions_n;
4771 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4772 						   actions, sfx_actions,
4773 						   pre_actions);
4774 		if (!mtr_tag_id) {
4775 			ret = -rte_errno;
4776 			goto exit;
4777 		}
4778 		/* Add the prefix subflow. */
4779 		ret = flow_create_split_inner(dev, flow, &dev_flow,
4780 					      prefix_layers, 0,
4781 					      attr, items,
4782 					      pre_actions, external,
4783 					      flow_idx, error);
4784 		if (ret) {
4785 			ret = -rte_errno;
4786 			goto exit;
4787 		}
4788 		dev_flow->handle->split_flow_id = mtr_tag_id;
4789 		/* Setting the sfx group atrr. */
4790 		sfx_attr.group = sfx_attr.transfer ?
4791 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4792 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4793 	}
4794 	/* Add the prefix subflow. */
4795 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4796 					 flow_get_prefix_layer_flags(dev_flow) :
4797 					 prefix_layers, dev_flow ?
4798 					 dev_flow->handle->mark : prefix_mark,
4799 					 &sfx_attr, sfx_items ?
4800 					 sfx_items : items,
4801 					 sfx_actions ? sfx_actions : actions,
4802 					 external, flow_idx, error);
4803 exit:
4804 	if (sfx_actions)
4805 		mlx5_free(sfx_actions);
4806 	return ret;
4807 }
4808 
4809 /**
4810  * The splitting for sample feature.
4811  *
4812  * Once Sample action is detected in the action list, the flow actions should
4813  * be split into prefix sub flow and suffix sub flow.
4814  *
4815  * The original items remain in the prefix sub flow, all actions preceding the
4816  * sample action and the sample action itself will be copied to the prefix
4817  * sub flow, the actions following the sample action will be copied to the
4818  * suffix sub flow, Queue action always be located in the suffix sub flow.
4819  *
4820  * In order to make the packet from prefix sub flow matches with suffix sub
4821  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
4822  * flow uses tag item with the unique flow id.
4823  *
4824  * @param dev
4825  *   Pointer to Ethernet device.
4826  * @param[in] flow
4827  *   Parent flow structure pointer.
4828  * @param[in] attr
4829  *   Flow rule attributes.
4830  * @param[in] items
4831  *   Pattern specification (list terminated by the END pattern item).
4832  * @param[in] actions
4833  *   Associated actions (list terminated by the END action).
4834  * @param[in] external
4835  *   This flow rule is created by request external to PMD.
4836  * @param[in] flow_idx
4837  *   This memory pool index to the flow.
4838  * @param[out] error
4839  *   Perform verbose error reporting if not NULL.
4840  * @return
4841  *   0 on success, negative value otherwise
4842  */
4843 static int
4844 flow_create_split_sample(struct rte_eth_dev *dev,
4845 			 struct rte_flow *flow,
4846 			 const struct rte_flow_attr *attr,
4847 			 const struct rte_flow_item items[],
4848 			 const struct rte_flow_action actions[],
4849 			 bool external, uint32_t flow_idx,
4850 			 struct rte_flow_error *error)
4851 {
4852 	struct mlx5_priv *priv = dev->data->dev_private;
4853 	struct rte_flow_action *sfx_actions = NULL;
4854 	struct rte_flow_action *pre_actions = NULL;
4855 	struct rte_flow_item *sfx_items = NULL;
4856 	struct mlx5_flow *dev_flow = NULL;
4857 	struct rte_flow_attr sfx_attr = *attr;
4858 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4859 	struct mlx5_flow_dv_sample_resource *sample_res;
4860 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
4861 	struct mlx5_flow_tbl_resource *sfx_tbl;
4862 	union mlx5_flow_tbl_key sfx_table_key;
4863 #endif
4864 	size_t act_size;
4865 	size_t item_size;
4866 	uint32_t fdb_tx = 0;
4867 	int32_t tag_id = 0;
4868 	int actions_n = 0;
4869 	int sample_action_pos;
4870 	int qrss_action_pos;
4871 	int ret = 0;
4872 
4873 	if (priv->sampler_en)
4874 		actions_n = flow_check_match_action(actions, attr,
4875 					RTE_FLOW_ACTION_TYPE_SAMPLE,
4876 					&sample_action_pos, &qrss_action_pos);
4877 	if (actions_n) {
4878 		/* The prefix actions must includes sample, tag, end. */
4879 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
4880 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
4881 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
4882 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4883 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
4884 					  item_size), 0, SOCKET_ID_ANY);
4885 		if (!sfx_actions)
4886 			return rte_flow_error_set(error, ENOMEM,
4887 						  RTE_FLOW_ERROR_TYPE_ACTION,
4888 						  NULL, "no memory to split "
4889 						  "sample flow");
4890 		/* The representor_id is -1 for uplink. */
4891 		fdb_tx = (attr->transfer && priv->representor_id != -1);
4892 		if (!fdb_tx)
4893 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
4894 					+ act_size);
4895 		pre_actions = sfx_actions + actions_n;
4896 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
4897 						actions, sfx_actions,
4898 						pre_actions, actions_n,
4899 						sample_action_pos,
4900 						qrss_action_pos, error);
4901 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
4902 			ret = -rte_errno;
4903 			goto exit;
4904 		}
4905 		/* Add the prefix subflow. */
4906 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, 0, attr,
4907 					      items, pre_actions, external,
4908 					      flow_idx, error);
4909 		if (ret) {
4910 			ret = -rte_errno;
4911 			goto exit;
4912 		}
4913 		dev_flow->handle->split_flow_id = tag_id;
4914 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4915 		/* Set the sfx group attr. */
4916 		sample_res = (struct mlx5_flow_dv_sample_resource *)
4917 					dev_flow->dv.sample_res;
4918 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
4919 					sample_res->normal_path_tbl;
4920 		sfx_tbl_data = container_of(sfx_tbl,
4921 					struct mlx5_flow_tbl_data_entry, tbl);
4922 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
4923 		sfx_attr.group = sfx_attr.transfer ?
4924 					(sfx_table_key.table_id - 1) :
4925 					 sfx_table_key.table_id;
4926 #endif
4927 	}
4928 	/* Add the suffix subflow. */
4929 	ret = flow_create_split_meter(dev, flow, dev_flow ?
4930 				 flow_get_prefix_layer_flags(dev_flow) : 0,
4931 				 dev_flow ? dev_flow->handle->mark : 0,
4932 				 &sfx_attr, sfx_items ? sfx_items : items,
4933 				 sfx_actions ? sfx_actions : actions,
4934 				 external, flow_idx, error);
4935 exit:
4936 	if (sfx_actions)
4937 		mlx5_free(sfx_actions);
4938 	return ret;
4939 }
4940 
4941 /**
4942  * Split the flow to subflow set. The splitters might be linked
4943  * in the chain, like this:
4944  * flow_create_split_outer() calls:
4945  *   flow_create_split_meter() calls:
4946  *     flow_create_split_metadata(meter_subflow_0) calls:
4947  *       flow_create_split_inner(metadata_subflow_0)
4948  *       flow_create_split_inner(metadata_subflow_1)
4949  *       flow_create_split_inner(metadata_subflow_2)
4950  *     flow_create_split_metadata(meter_subflow_1) calls:
4951  *       flow_create_split_inner(metadata_subflow_0)
4952  *       flow_create_split_inner(metadata_subflow_1)
4953  *       flow_create_split_inner(metadata_subflow_2)
4954  *
4955  * This provide flexible way to add new levels of flow splitting.
4956  * The all of successfully created subflows are included to the
4957  * parent flow dev_flow list.
4958  *
4959  * @param dev
4960  *   Pointer to Ethernet device.
4961  * @param[in] flow
4962  *   Parent flow structure pointer.
4963  * @param[in] attr
4964  *   Flow rule attributes.
4965  * @param[in] items
4966  *   Pattern specification (list terminated by the END pattern item).
4967  * @param[in] actions
4968  *   Associated actions (list terminated by the END action).
4969  * @param[in] external
4970  *   This flow rule is created by request external to PMD.
4971  * @param[in] flow_idx
4972  *   This memory pool index to the flow.
4973  * @param[out] error
4974  *   Perform verbose error reporting if not NULL.
4975  * @return
4976  *   0 on success, negative value otherwise
4977  */
4978 static int
4979 flow_create_split_outer(struct rte_eth_dev *dev,
4980 			struct rte_flow *flow,
4981 			const struct rte_flow_attr *attr,
4982 			const struct rte_flow_item items[],
4983 			const struct rte_flow_action actions[],
4984 			bool external, uint32_t flow_idx,
4985 			struct rte_flow_error *error)
4986 {
4987 	int ret;
4988 
4989 	ret = flow_create_split_sample(dev, flow, attr, items,
4990 				       actions, external, flow_idx, error);
4991 	MLX5_ASSERT(ret <= 0);
4992 	return ret;
4993 }
4994 
4995 /**
4996  * Create a flow and add it to @p list.
4997  *
4998  * @param dev
4999  *   Pointer to Ethernet device.
5000  * @param list
5001  *   Pointer to a TAILQ flow list. If this parameter NULL,
5002  *   no list insertion occurred, flow is just created,
5003  *   this is caller's responsibility to track the
5004  *   created flow.
5005  * @param[in] attr
5006  *   Flow rule attributes.
5007  * @param[in] items
5008  *   Pattern specification (list terminated by the END pattern item).
5009  * @param[in] actions
5010  *   Associated actions (list terminated by the END action).
5011  * @param[in] external
5012  *   This flow rule is created by request external to PMD.
5013  * @param[out] error
5014  *   Perform verbose error reporting if not NULL.
5015  *
5016  * @return
5017  *   A flow index on success, 0 otherwise and rte_errno is set.
5018  */
5019 static uint32_t
5020 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5021 		 const struct rte_flow_attr *attr,
5022 		 const struct rte_flow_item items[],
5023 		 const struct rte_flow_action actions[],
5024 		 bool external, struct rte_flow_error *error)
5025 {
5026 	struct mlx5_priv *priv = dev->data->dev_private;
5027 	struct rte_flow *flow = NULL;
5028 	struct mlx5_flow *dev_flow;
5029 	const struct rte_flow_action_rss *rss;
5030 	union {
5031 		struct mlx5_flow_expand_rss buf;
5032 		uint8_t buffer[2048];
5033 	} expand_buffer;
5034 	union {
5035 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5036 		uint8_t buffer[2048];
5037 	} actions_rx;
5038 	union {
5039 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5040 		uint8_t buffer[2048];
5041 	} actions_hairpin_tx;
5042 	union {
5043 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5044 		uint8_t buffer[2048];
5045 	} items_tx;
5046 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5047 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
5048 					      priv->rss_desc)[!!priv->flow_idx];
5049 	const struct rte_flow_action *p_actions_rx = actions;
5050 	uint32_t i;
5051 	uint32_t idx = 0;
5052 	int hairpin_flow;
5053 	uint32_t hairpin_id = 0;
5054 	struct rte_flow_attr attr_tx = { .priority = 0 };
5055 	struct rte_flow_attr attr_factor = {0};
5056 	int ret;
5057 
5058 	memcpy((void *)&attr_factor, (const void *)attr, sizeof(*attr));
5059 	if (external)
5060 		attr_factor.group *= MLX5_FLOW_TABLE_FACTOR;
5061 	hairpin_flow = flow_check_hairpin_split(dev, &attr_factor, actions);
5062 	ret = flow_drv_validate(dev, &attr_factor, items, p_actions_rx,
5063 				external, hairpin_flow, error);
5064 	if (ret < 0)
5065 		return 0;
5066 	if (hairpin_flow > 0) {
5067 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5068 			rte_errno = EINVAL;
5069 			return 0;
5070 		}
5071 		flow_hairpin_split(dev, actions, actions_rx.actions,
5072 				   actions_hairpin_tx.actions, items_tx.items,
5073 				   &hairpin_id);
5074 		p_actions_rx = actions_rx.actions;
5075 	}
5076 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5077 	if (!flow) {
5078 		rte_errno = ENOMEM;
5079 		goto error_before_flow;
5080 	}
5081 	flow->drv_type = flow_get_drv_type(dev, &attr_factor);
5082 	if (hairpin_id != 0)
5083 		flow->hairpin_flow_id = hairpin_id;
5084 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5085 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5086 	memset(rss_desc, 0, sizeof(*rss_desc));
5087 	rss = flow_get_rss_action(p_actions_rx);
5088 	if (rss) {
5089 		/*
5090 		 * The following information is required by
5091 		 * mlx5_flow_hashfields_adjust() in advance.
5092 		 */
5093 		rss_desc->level = rss->level;
5094 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5095 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5096 	}
5097 	flow->dev_handles = 0;
5098 	if (rss && rss->types) {
5099 		unsigned int graph_root;
5100 
5101 		graph_root = find_graph_root(items, rss->level);
5102 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5103 					   items, rss->types,
5104 					   mlx5_support_expansion, graph_root);
5105 		MLX5_ASSERT(ret > 0 &&
5106 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5107 	} else {
5108 		buf->entries = 1;
5109 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5110 	}
5111 	/*
5112 	 * Record the start index when there is a nested call. All sub-flows
5113 	 * need to be translated before another calling.
5114 	 * No need to use ping-pong buffer to save memory here.
5115 	 */
5116 	if (priv->flow_idx) {
5117 		MLX5_ASSERT(!priv->flow_nested_idx);
5118 		priv->flow_nested_idx = priv->flow_idx;
5119 	}
5120 	for (i = 0; i < buf->entries; ++i) {
5121 		/*
5122 		 * The splitter may create multiple dev_flows,
5123 		 * depending on configuration. In the simplest
5124 		 * case it just creates unmodified original flow.
5125 		 */
5126 		ret = flow_create_split_outer(dev, flow, &attr_factor,
5127 					      buf->entry[i].pattern,
5128 					      p_actions_rx, external, idx,
5129 					      error);
5130 		if (ret < 0)
5131 			goto error;
5132 	}
5133 	/* Create the tx flow. */
5134 	if (hairpin_flow) {
5135 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5136 		attr_tx.ingress = 0;
5137 		attr_tx.egress = 1;
5138 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5139 					 actions_hairpin_tx.actions,
5140 					 idx, error);
5141 		if (!dev_flow)
5142 			goto error;
5143 		dev_flow->flow = flow;
5144 		dev_flow->external = 0;
5145 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5146 			      dev_flow->handle, next);
5147 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5148 					 items_tx.items,
5149 					 actions_hairpin_tx.actions, error);
5150 		if (ret < 0)
5151 			goto error;
5152 	}
5153 	/*
5154 	 * Update the metadata register copy table. If extensive
5155 	 * metadata feature is enabled and registers are supported
5156 	 * we might create the extra rte_flow for each unique
5157 	 * MARK/FLAG action ID.
5158 	 *
5159 	 * The table is updated for ingress Flows only, because
5160 	 * the egress Flows belong to the different device and
5161 	 * copy table should be updated in peer NIC Rx domain.
5162 	 */
5163 	if (attr_factor.ingress &&
5164 	    (external || attr_factor.group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5165 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5166 		if (ret)
5167 			goto error;
5168 	}
5169 	/*
5170 	 * If the flow is external (from application) OR device is started, then
5171 	 * the flow will be applied immediately.
5172 	 */
5173 	if (external || dev->data->dev_started) {
5174 		ret = flow_drv_apply(dev, flow, error);
5175 		if (ret < 0)
5176 			goto error;
5177 	}
5178 	if (list)
5179 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5180 			     flow, next);
5181 	flow_rxq_flags_set(dev, flow);
5182 	/* Nested flow creation index recovery. */
5183 	priv->flow_idx = priv->flow_nested_idx;
5184 	if (priv->flow_nested_idx)
5185 		priv->flow_nested_idx = 0;
5186 	return idx;
5187 error:
5188 	MLX5_ASSERT(flow);
5189 	ret = rte_errno; /* Save rte_errno before cleanup. */
5190 	flow_mreg_del_copy_action(dev, flow);
5191 	flow_drv_destroy(dev, flow);
5192 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5193 	rte_errno = ret; /* Restore rte_errno. */
5194 error_before_flow:
5195 	ret = rte_errno;
5196 	if (hairpin_id)
5197 		mlx5_flow_id_release(priv->sh->flow_id_pool,
5198 				     hairpin_id);
5199 	rte_errno = ret;
5200 	priv->flow_idx = priv->flow_nested_idx;
5201 	if (priv->flow_nested_idx)
5202 		priv->flow_nested_idx = 0;
5203 	return 0;
5204 }
5205 
5206 /**
5207  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5208  * incoming packets to table 1.
5209  *
5210  * Other flow rules, requested for group n, will be created in
5211  * e-switch table n+1.
5212  * Jump action to e-switch group n will be created to group n+1.
5213  *
5214  * Used when working in switchdev mode, to utilise advantages of table 1
5215  * and above.
5216  *
5217  * @param dev
5218  *   Pointer to Ethernet device.
5219  *
5220  * @return
5221  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5222  */
5223 struct rte_flow *
5224 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5225 {
5226 	const struct rte_flow_attr attr = {
5227 		.group = 0,
5228 		.priority = 0,
5229 		.ingress = 1,
5230 		.egress = 0,
5231 		.transfer = 1,
5232 	};
5233 	const struct rte_flow_item pattern = {
5234 		.type = RTE_FLOW_ITEM_TYPE_END,
5235 	};
5236 	struct rte_flow_action_jump jump = {
5237 		.group = 1,
5238 	};
5239 	const struct rte_flow_action actions[] = {
5240 		{
5241 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5242 			.conf = &jump,
5243 		},
5244 		{
5245 			.type = RTE_FLOW_ACTION_TYPE_END,
5246 		},
5247 	};
5248 	struct mlx5_priv *priv = dev->data->dev_private;
5249 	struct rte_flow_error error;
5250 
5251 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5252 						   &attr, &pattern,
5253 						   actions, false, &error);
5254 }
5255 
5256 /**
5257  * Validate a flow supported by the NIC.
5258  *
5259  * @see rte_flow_validate()
5260  * @see rte_flow_ops
5261  */
5262 int
5263 mlx5_flow_validate(struct rte_eth_dev *dev,
5264 		   const struct rte_flow_attr *attr,
5265 		   const struct rte_flow_item items[],
5266 		   const struct rte_flow_action actions[],
5267 		   struct rte_flow_error *error)
5268 {
5269 	int hairpin_flow;
5270 
5271 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5272 	return flow_drv_validate(dev, attr, items, actions,
5273 				true, hairpin_flow, error);
5274 }
5275 
5276 /**
5277  * Create a flow.
5278  *
5279  * @see rte_flow_create()
5280  * @see rte_flow_ops
5281  */
5282 struct rte_flow *
5283 mlx5_flow_create(struct rte_eth_dev *dev,
5284 		 const struct rte_flow_attr *attr,
5285 		 const struct rte_flow_item items[],
5286 		 const struct rte_flow_action actions[],
5287 		 struct rte_flow_error *error)
5288 {
5289 	struct mlx5_priv *priv = dev->data->dev_private;
5290 
5291 	/*
5292 	 * If the device is not started yet, it is not allowed to created a
5293 	 * flow from application. PMD default flows and traffic control flows
5294 	 * are not affected.
5295 	 */
5296 	if (unlikely(!dev->data->dev_started)) {
5297 		DRV_LOG(DEBUG, "port %u is not started when "
5298 			"inserting a flow", dev->data->port_id);
5299 		rte_flow_error_set(error, ENODEV,
5300 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5301 				   NULL,
5302 				   "port not started");
5303 		return NULL;
5304 	}
5305 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5306 				  attr, items, actions, true, error);
5307 }
5308 
5309 /**
5310  * Destroy a flow in a list.
5311  *
5312  * @param dev
5313  *   Pointer to Ethernet device.
5314  * @param list
5315  *   Pointer to the Indexed flow list. If this parameter NULL,
5316  *   there is no flow removal from the list. Be noted that as
5317  *   flow is add to the indexed list, memory of the indexed
5318  *   list points to maybe changed as flow destroyed.
5319  * @param[in] flow_idx
5320  *   Index of flow to destroy.
5321  */
5322 static void
5323 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5324 		  uint32_t flow_idx)
5325 {
5326 	struct mlx5_priv *priv = dev->data->dev_private;
5327 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5328 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5329 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5330 
5331 	if (!flow)
5332 		return;
5333 	/*
5334 	 * Update RX queue flags only if port is started, otherwise it is
5335 	 * already clean.
5336 	 */
5337 	if (dev->data->dev_started)
5338 		flow_rxq_flags_trim(dev, flow);
5339 	if (flow->hairpin_flow_id)
5340 		mlx5_flow_id_release(priv->sh->flow_id_pool,
5341 				     flow->hairpin_flow_id);
5342 	flow_drv_destroy(dev, flow);
5343 	if (list)
5344 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5345 			     flow_idx, flow, next);
5346 	flow_mreg_del_copy_action(dev, flow);
5347 	if (flow->fdir) {
5348 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5349 			if (priv_fdir_flow->rix_flow == flow_idx)
5350 				break;
5351 		}
5352 		if (priv_fdir_flow) {
5353 			LIST_REMOVE(priv_fdir_flow, next);
5354 			mlx5_free(priv_fdir_flow->fdir);
5355 			mlx5_free(priv_fdir_flow);
5356 		}
5357 	}
5358 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5359 }
5360 
5361 /**
5362  * Destroy all flows.
5363  *
5364  * @param dev
5365  *   Pointer to Ethernet device.
5366  * @param list
5367  *   Pointer to the Indexed flow list.
5368  * @param active
5369  *   If flushing is called avtively.
5370  */
5371 void
5372 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5373 {
5374 	uint32_t num_flushed = 0;
5375 
5376 	while (*list) {
5377 		flow_list_destroy(dev, list, *list);
5378 		num_flushed++;
5379 	}
5380 	if (active) {
5381 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5382 			dev->data->port_id, num_flushed);
5383 	}
5384 }
5385 
5386 /**
5387  * Remove all flows.
5388  *
5389  * @param dev
5390  *   Pointer to Ethernet device.
5391  * @param list
5392  *   Pointer to the Indexed flow list.
5393  */
5394 void
5395 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
5396 {
5397 	struct mlx5_priv *priv = dev->data->dev_private;
5398 	struct rte_flow *flow = NULL;
5399 	uint32_t idx;
5400 
5401 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
5402 		      flow, next) {
5403 		flow_drv_remove(dev, flow);
5404 		flow_mreg_stop_copy_action(dev, flow);
5405 	}
5406 	flow_mreg_del_default_copy_action(dev);
5407 	flow_rxq_flags_clear(dev);
5408 }
5409 
5410 /**
5411  * Add all flows.
5412  *
5413  * @param dev
5414  *   Pointer to Ethernet device.
5415  * @param list
5416  *   Pointer to the Indexed flow list.
5417  *
5418  * @return
5419  *   0 on success, a negative errno value otherwise and rte_errno is set.
5420  */
5421 int
5422 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
5423 {
5424 	struct mlx5_priv *priv = dev->data->dev_private;
5425 	struct rte_flow *flow = NULL;
5426 	struct rte_flow_error error;
5427 	uint32_t idx;
5428 	int ret = 0;
5429 
5430 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5431 	ret = flow_mreg_add_default_copy_action(dev, &error);
5432 	if (ret < 0)
5433 		return -rte_errno;
5434 	/* Apply Flows created by application. */
5435 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
5436 		      flow, next) {
5437 		ret = flow_mreg_start_copy_action(dev, flow);
5438 		if (ret < 0)
5439 			goto error;
5440 		ret = flow_drv_apply(dev, flow, &error);
5441 		if (ret < 0)
5442 			goto error;
5443 		flow_rxq_flags_set(dev, flow);
5444 	}
5445 	return 0;
5446 error:
5447 	ret = rte_errno; /* Save rte_errno before cleanup. */
5448 	mlx5_flow_stop(dev, list);
5449 	rte_errno = ret; /* Restore rte_errno. */
5450 	return -rte_errno;
5451 }
5452 
5453 /**
5454  * Stop all default actions for flows.
5455  *
5456  * @param dev
5457  *   Pointer to Ethernet device.
5458  */
5459 void
5460 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5461 {
5462 	flow_mreg_del_default_copy_action(dev);
5463 	flow_rxq_flags_clear(dev);
5464 }
5465 
5466 /**
5467  * Start all default actions for flows.
5468  *
5469  * @param dev
5470  *   Pointer to Ethernet device.
5471  * @return
5472  *   0 on success, a negative errno value otherwise and rte_errno is set.
5473  */
5474 int
5475 mlx5_flow_start_default(struct rte_eth_dev *dev)
5476 {
5477 	struct rte_flow_error error;
5478 
5479 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5480 	return flow_mreg_add_default_copy_action(dev, &error);
5481 }
5482 
5483 /**
5484  * Allocate intermediate resources for flow creation.
5485  *
5486  * @param dev
5487  *   Pointer to Ethernet device.
5488  */
5489 void
5490 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
5491 {
5492 	struct mlx5_priv *priv = dev->data->dev_private;
5493 
5494 	if (!priv->inter_flows) {
5495 		priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
5496 				    MLX5_NUM_MAX_DEV_FLOWS *
5497 				    sizeof(struct mlx5_flow) +
5498 				    (sizeof(struct mlx5_flow_rss_desc) +
5499 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0,
5500 				    SOCKET_ID_ANY);
5501 		if (!priv->inter_flows) {
5502 			DRV_LOG(ERR, "can't allocate intermediate memory.");
5503 			return;
5504 		}
5505 	}
5506 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
5507 			 [MLX5_NUM_MAX_DEV_FLOWS];
5508 	/* Reset the index. */
5509 	priv->flow_idx = 0;
5510 	priv->flow_nested_idx = 0;
5511 }
5512 
5513 /**
5514  * Free intermediate resources for flows.
5515  *
5516  * @param dev
5517  *   Pointer to Ethernet device.
5518  */
5519 void
5520 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
5521 {
5522 	struct mlx5_priv *priv = dev->data->dev_private;
5523 
5524 	mlx5_free(priv->inter_flows);
5525 	priv->inter_flows = NULL;
5526 }
5527 
5528 /**
5529  * Verify the flow list is empty
5530  *
5531  * @param dev
5532  *  Pointer to Ethernet device.
5533  *
5534  * @return the number of flows not released.
5535  */
5536 int
5537 mlx5_flow_verify(struct rte_eth_dev *dev)
5538 {
5539 	struct mlx5_priv *priv = dev->data->dev_private;
5540 	struct rte_flow *flow;
5541 	uint32_t idx;
5542 	int ret = 0;
5543 
5544 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
5545 		      flow, next) {
5546 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
5547 			dev->data->port_id, (void *)flow);
5548 		++ret;
5549 	}
5550 	return ret;
5551 }
5552 
5553 /**
5554  * Enable default hairpin egress flow.
5555  *
5556  * @param dev
5557  *   Pointer to Ethernet device.
5558  * @param queue
5559  *   The queue index.
5560  *
5561  * @return
5562  *   0 on success, a negative errno value otherwise and rte_errno is set.
5563  */
5564 int
5565 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
5566 			    uint32_t queue)
5567 {
5568 	struct mlx5_priv *priv = dev->data->dev_private;
5569 	const struct rte_flow_attr attr = {
5570 		.egress = 1,
5571 		.priority = 0,
5572 	};
5573 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
5574 		.queue = queue,
5575 	};
5576 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
5577 		.queue = UINT32_MAX,
5578 	};
5579 	struct rte_flow_item items[] = {
5580 		{
5581 			.type = (enum rte_flow_item_type)
5582 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
5583 			.spec = &queue_spec,
5584 			.last = NULL,
5585 			.mask = &queue_mask,
5586 		},
5587 		{
5588 			.type = RTE_FLOW_ITEM_TYPE_END,
5589 		},
5590 	};
5591 	struct rte_flow_action_jump jump = {
5592 		.group = MLX5_HAIRPIN_TX_TABLE,
5593 	};
5594 	struct rte_flow_action actions[2];
5595 	uint32_t flow_idx;
5596 	struct rte_flow_error error;
5597 
5598 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
5599 	actions[0].conf = &jump;
5600 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
5601 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5602 				&attr, items, actions, false, &error);
5603 	if (!flow_idx) {
5604 		DRV_LOG(DEBUG,
5605 			"Failed to create ctrl flow: rte_errno(%d),"
5606 			" type(%d), message(%s)",
5607 			rte_errno, error.type,
5608 			error.message ? error.message : " (no stated reason)");
5609 		return -rte_errno;
5610 	}
5611 	return 0;
5612 }
5613 
5614 /**
5615  * Enable a control flow configured from the control plane.
5616  *
5617  * @param dev
5618  *   Pointer to Ethernet device.
5619  * @param eth_spec
5620  *   An Ethernet flow spec to apply.
5621  * @param eth_mask
5622  *   An Ethernet flow mask to apply.
5623  * @param vlan_spec
5624  *   A VLAN flow spec to apply.
5625  * @param vlan_mask
5626  *   A VLAN flow mask to apply.
5627  *
5628  * @return
5629  *   0 on success, a negative errno value otherwise and rte_errno is set.
5630  */
5631 int
5632 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
5633 		    struct rte_flow_item_eth *eth_spec,
5634 		    struct rte_flow_item_eth *eth_mask,
5635 		    struct rte_flow_item_vlan *vlan_spec,
5636 		    struct rte_flow_item_vlan *vlan_mask)
5637 {
5638 	struct mlx5_priv *priv = dev->data->dev_private;
5639 	const struct rte_flow_attr attr = {
5640 		.ingress = 1,
5641 		.priority = MLX5_FLOW_PRIO_RSVD,
5642 	};
5643 	struct rte_flow_item items[] = {
5644 		{
5645 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5646 			.spec = eth_spec,
5647 			.last = NULL,
5648 			.mask = eth_mask,
5649 		},
5650 		{
5651 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
5652 					      RTE_FLOW_ITEM_TYPE_END,
5653 			.spec = vlan_spec,
5654 			.last = NULL,
5655 			.mask = vlan_mask,
5656 		},
5657 		{
5658 			.type = RTE_FLOW_ITEM_TYPE_END,
5659 		},
5660 	};
5661 	uint16_t queue[priv->reta_idx_n];
5662 	struct rte_flow_action_rss action_rss = {
5663 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
5664 		.level = 0,
5665 		.types = priv->rss_conf.rss_hf,
5666 		.key_len = priv->rss_conf.rss_key_len,
5667 		.queue_num = priv->reta_idx_n,
5668 		.key = priv->rss_conf.rss_key,
5669 		.queue = queue,
5670 	};
5671 	struct rte_flow_action actions[] = {
5672 		{
5673 			.type = RTE_FLOW_ACTION_TYPE_RSS,
5674 			.conf = &action_rss,
5675 		},
5676 		{
5677 			.type = RTE_FLOW_ACTION_TYPE_END,
5678 		},
5679 	};
5680 	uint32_t flow_idx;
5681 	struct rte_flow_error error;
5682 	unsigned int i;
5683 
5684 	if (!priv->reta_idx_n || !priv->rxqs_n) {
5685 		return 0;
5686 	}
5687 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
5688 		action_rss.types = 0;
5689 	for (i = 0; i != priv->reta_idx_n; ++i)
5690 		queue[i] = (*priv->reta_idx)[i];
5691 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5692 				&attr, items, actions, false, &error);
5693 	if (!flow_idx)
5694 		return -rte_errno;
5695 	return 0;
5696 }
5697 
5698 /**
5699  * Enable a flow control configured from the control plane.
5700  *
5701  * @param dev
5702  *   Pointer to Ethernet device.
5703  * @param eth_spec
5704  *   An Ethernet flow spec to apply.
5705  * @param eth_mask
5706  *   An Ethernet flow mask to apply.
5707  *
5708  * @return
5709  *   0 on success, a negative errno value otherwise and rte_errno is set.
5710  */
5711 int
5712 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5713 	       struct rte_flow_item_eth *eth_spec,
5714 	       struct rte_flow_item_eth *eth_mask)
5715 {
5716 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5717 }
5718 
5719 /**
5720  * Create default miss flow rule matching lacp traffic
5721  *
5722  * @param dev
5723  *   Pointer to Ethernet device.
5724  * @param eth_spec
5725  *   An Ethernet flow spec to apply.
5726  *
5727  * @return
5728  *   0 on success, a negative errno value otherwise and rte_errno is set.
5729  */
5730 int
5731 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5732 {
5733 	struct mlx5_priv *priv = dev->data->dev_private;
5734 	/*
5735 	 * The LACP matching is done by only using ether type since using
5736 	 * a multicast dst mac causes kernel to give low priority to this flow.
5737 	 */
5738 	static const struct rte_flow_item_eth lacp_spec = {
5739 		.type = RTE_BE16(0x8809),
5740 	};
5741 	static const struct rte_flow_item_eth lacp_mask = {
5742 		.type = 0xffff,
5743 	};
5744 	const struct rte_flow_attr attr = {
5745 		.ingress = 1,
5746 	};
5747 	struct rte_flow_item items[] = {
5748 		{
5749 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5750 			.spec = &lacp_spec,
5751 			.mask = &lacp_mask,
5752 		},
5753 		{
5754 			.type = RTE_FLOW_ITEM_TYPE_END,
5755 		},
5756 	};
5757 	struct rte_flow_action actions[] = {
5758 		{
5759 			.type = (enum rte_flow_action_type)
5760 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5761 		},
5762 		{
5763 			.type = RTE_FLOW_ACTION_TYPE_END,
5764 		},
5765 	};
5766 	struct rte_flow_error error;
5767 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5768 				&attr, items, actions, false, &error);
5769 
5770 	if (!flow_idx)
5771 		return -rte_errno;
5772 	return 0;
5773 }
5774 
5775 /**
5776  * Destroy a flow.
5777  *
5778  * @see rte_flow_destroy()
5779  * @see rte_flow_ops
5780  */
5781 int
5782 mlx5_flow_destroy(struct rte_eth_dev *dev,
5783 		  struct rte_flow *flow,
5784 		  struct rte_flow_error *error __rte_unused)
5785 {
5786 	struct mlx5_priv *priv = dev->data->dev_private;
5787 
5788 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5789 	return 0;
5790 }
5791 
5792 /**
5793  * Destroy all flows.
5794  *
5795  * @see rte_flow_flush()
5796  * @see rte_flow_ops
5797  */
5798 int
5799 mlx5_flow_flush(struct rte_eth_dev *dev,
5800 		struct rte_flow_error *error __rte_unused)
5801 {
5802 	struct mlx5_priv *priv = dev->data->dev_private;
5803 
5804 	mlx5_flow_list_flush(dev, &priv->flows, false);
5805 	return 0;
5806 }
5807 
5808 /**
5809  * Isolated mode.
5810  *
5811  * @see rte_flow_isolate()
5812  * @see rte_flow_ops
5813  */
5814 int
5815 mlx5_flow_isolate(struct rte_eth_dev *dev,
5816 		  int enable,
5817 		  struct rte_flow_error *error)
5818 {
5819 	struct mlx5_priv *priv = dev->data->dev_private;
5820 
5821 	if (dev->data->dev_started) {
5822 		rte_flow_error_set(error, EBUSY,
5823 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5824 				   NULL,
5825 				   "port must be stopped first");
5826 		return -rte_errno;
5827 	}
5828 	priv->isolated = !!enable;
5829 	if (enable)
5830 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
5831 	else
5832 		dev->dev_ops = &mlx5_os_dev_ops;
5833 
5834 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
5835 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
5836 
5837 	return 0;
5838 }
5839 
5840 /**
5841  * Query a flow.
5842  *
5843  * @see rte_flow_query()
5844  * @see rte_flow_ops
5845  */
5846 static int
5847 flow_drv_query(struct rte_eth_dev *dev,
5848 	       uint32_t flow_idx,
5849 	       const struct rte_flow_action *actions,
5850 	       void *data,
5851 	       struct rte_flow_error *error)
5852 {
5853 	struct mlx5_priv *priv = dev->data->dev_private;
5854 	const struct mlx5_flow_driver_ops *fops;
5855 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5856 					       [MLX5_IPOOL_RTE_FLOW],
5857 					       flow_idx);
5858 	enum mlx5_flow_drv_type ftype;
5859 
5860 	if (!flow) {
5861 		return rte_flow_error_set(error, ENOENT,
5862 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5863 			  NULL,
5864 			  "invalid flow handle");
5865 	}
5866 	ftype = flow->drv_type;
5867 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5868 	fops = flow_get_drv_ops(ftype);
5869 
5870 	return fops->query(dev, flow, actions, data, error);
5871 }
5872 
5873 /**
5874  * Query a flow.
5875  *
5876  * @see rte_flow_query()
5877  * @see rte_flow_ops
5878  */
5879 int
5880 mlx5_flow_query(struct rte_eth_dev *dev,
5881 		struct rte_flow *flow,
5882 		const struct rte_flow_action *actions,
5883 		void *data,
5884 		struct rte_flow_error *error)
5885 {
5886 	int ret;
5887 
5888 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5889 			     error);
5890 	if (ret < 0)
5891 		return ret;
5892 	return 0;
5893 }
5894 
5895 /**
5896  * Convert a flow director filter to a generic flow.
5897  *
5898  * @param dev
5899  *   Pointer to Ethernet device.
5900  * @param fdir_filter
5901  *   Flow director filter to add.
5902  * @param attributes
5903  *   Generic flow parameters structure.
5904  *
5905  * @return
5906  *   0 on success, a negative errno value otherwise and rte_errno is set.
5907  */
5908 static int
5909 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5910 			 const struct rte_eth_fdir_filter *fdir_filter,
5911 			 struct mlx5_fdir *attributes)
5912 {
5913 	struct mlx5_priv *priv = dev->data->dev_private;
5914 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5915 	const struct rte_eth_fdir_masks *mask =
5916 		&dev->data->dev_conf.fdir_conf.mask;
5917 
5918 	/* Validate queue number. */
5919 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5920 		DRV_LOG(ERR, "port %u invalid queue number %d",
5921 			dev->data->port_id, fdir_filter->action.rx_queue);
5922 		rte_errno = EINVAL;
5923 		return -rte_errno;
5924 	}
5925 	attributes->attr.ingress = 1;
5926 	attributes->items[0] = (struct rte_flow_item) {
5927 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5928 		.spec = &attributes->l2,
5929 		.mask = &attributes->l2_mask,
5930 	};
5931 	switch (fdir_filter->action.behavior) {
5932 	case RTE_ETH_FDIR_ACCEPT:
5933 		attributes->actions[0] = (struct rte_flow_action){
5934 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5935 			.conf = &attributes->queue,
5936 		};
5937 		break;
5938 	case RTE_ETH_FDIR_REJECT:
5939 		attributes->actions[0] = (struct rte_flow_action){
5940 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5941 		};
5942 		break;
5943 	default:
5944 		DRV_LOG(ERR, "port %u invalid behavior %d",
5945 			dev->data->port_id,
5946 			fdir_filter->action.behavior);
5947 		rte_errno = ENOTSUP;
5948 		return -rte_errno;
5949 	}
5950 	attributes->queue.index = fdir_filter->action.rx_queue;
5951 	/* Handle L3. */
5952 	switch (fdir_filter->input.flow_type) {
5953 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5954 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5955 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5956 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5957 			.src_addr = input->flow.ip4_flow.src_ip,
5958 			.dst_addr = input->flow.ip4_flow.dst_ip,
5959 			.time_to_live = input->flow.ip4_flow.ttl,
5960 			.type_of_service = input->flow.ip4_flow.tos,
5961 		};
5962 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5963 			.src_addr = mask->ipv4_mask.src_ip,
5964 			.dst_addr = mask->ipv4_mask.dst_ip,
5965 			.time_to_live = mask->ipv4_mask.ttl,
5966 			.type_of_service = mask->ipv4_mask.tos,
5967 			.next_proto_id = mask->ipv4_mask.proto,
5968 		};
5969 		attributes->items[1] = (struct rte_flow_item){
5970 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5971 			.spec = &attributes->l3,
5972 			.mask = &attributes->l3_mask,
5973 		};
5974 		break;
5975 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5976 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5977 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5978 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5979 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5980 			.proto = input->flow.ipv6_flow.proto,
5981 		};
5982 
5983 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5984 		       input->flow.ipv6_flow.src_ip,
5985 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5986 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5987 		       input->flow.ipv6_flow.dst_ip,
5988 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5989 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5990 		       mask->ipv6_mask.src_ip,
5991 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5992 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5993 		       mask->ipv6_mask.dst_ip,
5994 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5995 		attributes->items[1] = (struct rte_flow_item){
5996 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5997 			.spec = &attributes->l3,
5998 			.mask = &attributes->l3_mask,
5999 		};
6000 		break;
6001 	default:
6002 		DRV_LOG(ERR, "port %u invalid flow type%d",
6003 			dev->data->port_id, fdir_filter->input.flow_type);
6004 		rte_errno = ENOTSUP;
6005 		return -rte_errno;
6006 	}
6007 	/* Handle L4. */
6008 	switch (fdir_filter->input.flow_type) {
6009 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
6010 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6011 			.src_port = input->flow.udp4_flow.src_port,
6012 			.dst_port = input->flow.udp4_flow.dst_port,
6013 		};
6014 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6015 			.src_port = mask->src_port_mask,
6016 			.dst_port = mask->dst_port_mask,
6017 		};
6018 		attributes->items[2] = (struct rte_flow_item){
6019 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6020 			.spec = &attributes->l4,
6021 			.mask = &attributes->l4_mask,
6022 		};
6023 		break;
6024 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
6025 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6026 			.src_port = input->flow.tcp4_flow.src_port,
6027 			.dst_port = input->flow.tcp4_flow.dst_port,
6028 		};
6029 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6030 			.src_port = mask->src_port_mask,
6031 			.dst_port = mask->dst_port_mask,
6032 		};
6033 		attributes->items[2] = (struct rte_flow_item){
6034 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6035 			.spec = &attributes->l4,
6036 			.mask = &attributes->l4_mask,
6037 		};
6038 		break;
6039 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
6040 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
6041 			.src_port = input->flow.udp6_flow.src_port,
6042 			.dst_port = input->flow.udp6_flow.dst_port,
6043 		};
6044 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
6045 			.src_port = mask->src_port_mask,
6046 			.dst_port = mask->dst_port_mask,
6047 		};
6048 		attributes->items[2] = (struct rte_flow_item){
6049 			.type = RTE_FLOW_ITEM_TYPE_UDP,
6050 			.spec = &attributes->l4,
6051 			.mask = &attributes->l4_mask,
6052 		};
6053 		break;
6054 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
6055 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
6056 			.src_port = input->flow.tcp6_flow.src_port,
6057 			.dst_port = input->flow.tcp6_flow.dst_port,
6058 		};
6059 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
6060 			.src_port = mask->src_port_mask,
6061 			.dst_port = mask->dst_port_mask,
6062 		};
6063 		attributes->items[2] = (struct rte_flow_item){
6064 			.type = RTE_FLOW_ITEM_TYPE_TCP,
6065 			.spec = &attributes->l4,
6066 			.mask = &attributes->l4_mask,
6067 		};
6068 		break;
6069 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
6070 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
6071 		break;
6072 	default:
6073 		DRV_LOG(ERR, "port %u invalid flow type%d",
6074 			dev->data->port_id, fdir_filter->input.flow_type);
6075 		rte_errno = ENOTSUP;
6076 		return -rte_errno;
6077 	}
6078 	return 0;
6079 }
6080 
6081 #define FLOW_FDIR_CMP(f1, f2, fld) \
6082 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
6083 
6084 /**
6085  * Compare two FDIR flows. If items and actions are identical, the two flows are
6086  * regarded as same.
6087  *
6088  * @param dev
6089  *   Pointer to Ethernet device.
6090  * @param f1
6091  *   FDIR flow to compare.
6092  * @param f2
6093  *   FDIR flow to compare.
6094  *
6095  * @return
6096  *   Zero on match, 1 otherwise.
6097  */
6098 static int
6099 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
6100 {
6101 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
6102 	    FLOW_FDIR_CMP(f1, f2, l2) ||
6103 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
6104 	    FLOW_FDIR_CMP(f1, f2, l3) ||
6105 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
6106 	    FLOW_FDIR_CMP(f1, f2, l4) ||
6107 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
6108 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
6109 		return 1;
6110 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
6111 	    FLOW_FDIR_CMP(f1, f2, queue))
6112 		return 1;
6113 	return 0;
6114 }
6115 
6116 /**
6117  * Search device flow list to find out a matched FDIR flow.
6118  *
6119  * @param dev
6120  *   Pointer to Ethernet device.
6121  * @param fdir_flow
6122  *   FDIR flow to lookup.
6123  *
6124  * @return
6125  *   Index of flow if found, 0 otherwise.
6126  */
6127 static uint32_t
6128 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
6129 {
6130 	struct mlx5_priv *priv = dev->data->dev_private;
6131 	uint32_t flow_idx = 0;
6132 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6133 
6134 	MLX5_ASSERT(fdir_flow);
6135 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6136 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
6137 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
6138 				dev->data->port_id, flow_idx);
6139 			flow_idx = priv_fdir_flow->rix_flow;
6140 			break;
6141 		}
6142 	}
6143 	return flow_idx;
6144 }
6145 
6146 /**
6147  * Add new flow director filter and store it in list.
6148  *
6149  * @param dev
6150  *   Pointer to Ethernet device.
6151  * @param fdir_filter
6152  *   Flow director filter to add.
6153  *
6154  * @return
6155  *   0 on success, a negative errno value otherwise and rte_errno is set.
6156  */
6157 static int
6158 flow_fdir_filter_add(struct rte_eth_dev *dev,
6159 		     const struct rte_eth_fdir_filter *fdir_filter)
6160 {
6161 	struct mlx5_priv *priv = dev->data->dev_private;
6162 	struct mlx5_fdir *fdir_flow;
6163 	struct rte_flow *flow;
6164 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6165 	uint32_t flow_idx;
6166 	int ret;
6167 
6168 	fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
6169 				SOCKET_ID_ANY);
6170 	if (!fdir_flow) {
6171 		rte_errno = ENOMEM;
6172 		return -rte_errno;
6173 	}
6174 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
6175 	if (ret)
6176 		goto error;
6177 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
6178 	if (flow_idx) {
6179 		rte_errno = EEXIST;
6180 		goto error;
6181 	}
6182 	priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
6183 				     sizeof(struct mlx5_fdir_flow),
6184 				     0, SOCKET_ID_ANY);
6185 	if (!priv_fdir_flow) {
6186 		rte_errno = ENOMEM;
6187 		goto error;
6188 	}
6189 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
6190 				    fdir_flow->items, fdir_flow->actions, true,
6191 				    NULL);
6192 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
6193 	if (!flow)
6194 		goto error;
6195 	flow->fdir = 1;
6196 	priv_fdir_flow->fdir = fdir_flow;
6197 	priv_fdir_flow->rix_flow = flow_idx;
6198 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
6199 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
6200 		dev->data->port_id, (void *)flow);
6201 	return 0;
6202 error:
6203 	mlx5_free(priv_fdir_flow);
6204 	mlx5_free(fdir_flow);
6205 	return -rte_errno;
6206 }
6207 
6208 /**
6209  * Delete specific filter.
6210  *
6211  * @param dev
6212  *   Pointer to Ethernet device.
6213  * @param fdir_filter
6214  *   Filter to be deleted.
6215  *
6216  * @return
6217  *   0 on success, a negative errno value otherwise and rte_errno is set.
6218  */
6219 static int
6220 flow_fdir_filter_delete(struct rte_eth_dev *dev,
6221 			const struct rte_eth_fdir_filter *fdir_filter)
6222 {
6223 	struct mlx5_priv *priv = dev->data->dev_private;
6224 	uint32_t flow_idx;
6225 	struct mlx5_fdir fdir_flow = {
6226 		.attr.group = 0,
6227 	};
6228 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6229 	int ret;
6230 
6231 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
6232 	if (ret)
6233 		return -rte_errno;
6234 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
6235 		/* Find the fdir in priv list */
6236 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
6237 			break;
6238 	}
6239 	if (!priv_fdir_flow)
6240 		return 0;
6241 	LIST_REMOVE(priv_fdir_flow, next);
6242 	flow_idx = priv_fdir_flow->rix_flow;
6243 	flow_list_destroy(dev, &priv->flows, flow_idx);
6244 	mlx5_free(priv_fdir_flow->fdir);
6245 	mlx5_free(priv_fdir_flow);
6246 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
6247 		dev->data->port_id, flow_idx);
6248 	return 0;
6249 }
6250 
6251 /**
6252  * Update queue for specific filter.
6253  *
6254  * @param dev
6255  *   Pointer to Ethernet device.
6256  * @param fdir_filter
6257  *   Filter to be updated.
6258  *
6259  * @return
6260  *   0 on success, a negative errno value otherwise and rte_errno is set.
6261  */
6262 static int
6263 flow_fdir_filter_update(struct rte_eth_dev *dev,
6264 			const struct rte_eth_fdir_filter *fdir_filter)
6265 {
6266 	int ret;
6267 
6268 	ret = flow_fdir_filter_delete(dev, fdir_filter);
6269 	if (ret)
6270 		return ret;
6271 	return flow_fdir_filter_add(dev, fdir_filter);
6272 }
6273 
6274 /**
6275  * Flush all filters.
6276  *
6277  * @param dev
6278  *   Pointer to Ethernet device.
6279  */
6280 static void
6281 flow_fdir_filter_flush(struct rte_eth_dev *dev)
6282 {
6283 	struct mlx5_priv *priv = dev->data->dev_private;
6284 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
6285 
6286 	while (!LIST_EMPTY(&priv->fdir_flows)) {
6287 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
6288 		LIST_REMOVE(priv_fdir_flow, next);
6289 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
6290 		mlx5_free(priv_fdir_flow->fdir);
6291 		mlx5_free(priv_fdir_flow);
6292 	}
6293 }
6294 
6295 /**
6296  * Get flow director information.
6297  *
6298  * @param dev
6299  *   Pointer to Ethernet device.
6300  * @param[out] fdir_info
6301  *   Resulting flow director information.
6302  */
6303 static void
6304 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
6305 {
6306 	struct rte_eth_fdir_masks *mask =
6307 		&dev->data->dev_conf.fdir_conf.mask;
6308 
6309 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
6310 	fdir_info->guarant_spc = 0;
6311 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
6312 	fdir_info->max_flexpayload = 0;
6313 	fdir_info->flow_types_mask[0] = 0;
6314 	fdir_info->flex_payload_unit = 0;
6315 	fdir_info->max_flex_payload_segment_num = 0;
6316 	fdir_info->flex_payload_limit = 0;
6317 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
6318 }
6319 
6320 /**
6321  * Deal with flow director operations.
6322  *
6323  * @param dev
6324  *   Pointer to Ethernet device.
6325  * @param filter_op
6326  *   Operation to perform.
6327  * @param arg
6328  *   Pointer to operation-specific structure.
6329  *
6330  * @return
6331  *   0 on success, a negative errno value otherwise and rte_errno is set.
6332  */
6333 static int
6334 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
6335 		    void *arg)
6336 {
6337 	enum rte_fdir_mode fdir_mode =
6338 		dev->data->dev_conf.fdir_conf.mode;
6339 
6340 	if (filter_op == RTE_ETH_FILTER_NOP)
6341 		return 0;
6342 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
6343 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
6344 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
6345 			dev->data->port_id, fdir_mode);
6346 		rte_errno = EINVAL;
6347 		return -rte_errno;
6348 	}
6349 	switch (filter_op) {
6350 	case RTE_ETH_FILTER_ADD:
6351 		return flow_fdir_filter_add(dev, arg);
6352 	case RTE_ETH_FILTER_UPDATE:
6353 		return flow_fdir_filter_update(dev, arg);
6354 	case RTE_ETH_FILTER_DELETE:
6355 		return flow_fdir_filter_delete(dev, arg);
6356 	case RTE_ETH_FILTER_FLUSH:
6357 		flow_fdir_filter_flush(dev);
6358 		break;
6359 	case RTE_ETH_FILTER_INFO:
6360 		flow_fdir_info_get(dev, arg);
6361 		break;
6362 	default:
6363 		DRV_LOG(DEBUG, "port %u unknown operation %u",
6364 			dev->data->port_id, filter_op);
6365 		rte_errno = EINVAL;
6366 		return -rte_errno;
6367 	}
6368 	return 0;
6369 }
6370 
6371 /**
6372  * Manage filter operations.
6373  *
6374  * @param dev
6375  *   Pointer to Ethernet device structure.
6376  * @param filter_type
6377  *   Filter type.
6378  * @param filter_op
6379  *   Operation to perform.
6380  * @param arg
6381  *   Pointer to operation-specific structure.
6382  *
6383  * @return
6384  *   0 on success, a negative errno value otherwise and rte_errno is set.
6385  */
6386 int
6387 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6388 		     enum rte_filter_type filter_type,
6389 		     enum rte_filter_op filter_op,
6390 		     void *arg)
6391 {
6392 	switch (filter_type) {
6393 	case RTE_ETH_FILTER_GENERIC:
6394 		if (filter_op != RTE_ETH_FILTER_GET) {
6395 			rte_errno = EINVAL;
6396 			return -rte_errno;
6397 		}
6398 		*(const void **)arg = &mlx5_flow_ops;
6399 		return 0;
6400 	case RTE_ETH_FILTER_FDIR:
6401 		return flow_fdir_ctrl_func(dev, filter_op, arg);
6402 	default:
6403 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6404 			dev->data->port_id, filter_type);
6405 		rte_errno = ENOTSUP;
6406 		return -rte_errno;
6407 	}
6408 	return 0;
6409 }
6410 
6411 /**
6412  * Create the needed meter and suffix tables.
6413  *
6414  * @param[in] dev
6415  *   Pointer to Ethernet device.
6416  * @param[in] fm
6417  *   Pointer to the flow meter.
6418  *
6419  * @return
6420  *   Pointer to table set on success, NULL otherwise.
6421  */
6422 struct mlx5_meter_domains_infos *
6423 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6424 			  const struct mlx5_flow_meter *fm)
6425 {
6426 	const struct mlx5_flow_driver_ops *fops;
6427 
6428 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6429 	return fops->create_mtr_tbls(dev, fm);
6430 }
6431 
6432 /**
6433  * Destroy the meter table set.
6434  *
6435  * @param[in] dev
6436  *   Pointer to Ethernet device.
6437  * @param[in] tbl
6438  *   Pointer to the meter table set.
6439  *
6440  * @return
6441  *   0 on success.
6442  */
6443 int
6444 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6445 			   struct mlx5_meter_domains_infos *tbls)
6446 {
6447 	const struct mlx5_flow_driver_ops *fops;
6448 
6449 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6450 	return fops->destroy_mtr_tbls(dev, tbls);
6451 }
6452 
6453 /**
6454  * Create policer rules.
6455  *
6456  * @param[in] dev
6457  *   Pointer to Ethernet device.
6458  * @param[in] fm
6459  *   Pointer to flow meter structure.
6460  * @param[in] attr
6461  *   Pointer to flow attributes.
6462  *
6463  * @return
6464  *   0 on success, -1 otherwise.
6465  */
6466 int
6467 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6468 			       struct mlx5_flow_meter *fm,
6469 			       const struct rte_flow_attr *attr)
6470 {
6471 	const struct mlx5_flow_driver_ops *fops;
6472 
6473 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6474 	return fops->create_policer_rules(dev, fm, attr);
6475 }
6476 
6477 /**
6478  * Destroy policer rules.
6479  *
6480  * @param[in] fm
6481  *   Pointer to flow meter structure.
6482  * @param[in] attr
6483  *   Pointer to flow attributes.
6484  *
6485  * @return
6486  *   0 on success, -1 otherwise.
6487  */
6488 int
6489 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6490 				struct mlx5_flow_meter *fm,
6491 				const struct rte_flow_attr *attr)
6492 {
6493 	const struct mlx5_flow_driver_ops *fops;
6494 
6495 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6496 	return fops->destroy_policer_rules(dev, fm, attr);
6497 }
6498 
6499 /**
6500  * Allocate a counter.
6501  *
6502  * @param[in] dev
6503  *   Pointer to Ethernet device structure.
6504  *
6505  * @return
6506  *   Index to allocated counter  on success, 0 otherwise.
6507  */
6508 uint32_t
6509 mlx5_counter_alloc(struct rte_eth_dev *dev)
6510 {
6511 	const struct mlx5_flow_driver_ops *fops;
6512 	struct rte_flow_attr attr = { .transfer = 0 };
6513 
6514 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6515 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6516 		return fops->counter_alloc(dev);
6517 	}
6518 	DRV_LOG(ERR,
6519 		"port %u counter allocate is not supported.",
6520 		 dev->data->port_id);
6521 	return 0;
6522 }
6523 
6524 /**
6525  * Free a counter.
6526  *
6527  * @param[in] dev
6528  *   Pointer to Ethernet device structure.
6529  * @param[in] cnt
6530  *   Index to counter to be free.
6531  */
6532 void
6533 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6534 {
6535 	const struct mlx5_flow_driver_ops *fops;
6536 	struct rte_flow_attr attr = { .transfer = 0 };
6537 
6538 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6539 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6540 		fops->counter_free(dev, cnt);
6541 		return;
6542 	}
6543 	DRV_LOG(ERR,
6544 		"port %u counter free is not supported.",
6545 		 dev->data->port_id);
6546 }
6547 
6548 /**
6549  * Query counter statistics.
6550  *
6551  * @param[in] dev
6552  *   Pointer to Ethernet device structure.
6553  * @param[in] cnt
6554  *   Index to counter to query.
6555  * @param[in] clear
6556  *   Set to clear counter statistics.
6557  * @param[out] pkts
6558  *   The counter hits packets number to save.
6559  * @param[out] bytes
6560  *   The counter hits bytes number to save.
6561  *
6562  * @return
6563  *   0 on success, a negative errno value otherwise.
6564  */
6565 int
6566 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6567 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6568 {
6569 	const struct mlx5_flow_driver_ops *fops;
6570 	struct rte_flow_attr attr = { .transfer = 0 };
6571 
6572 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6573 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6574 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6575 	}
6576 	DRV_LOG(ERR,
6577 		"port %u counter query is not supported.",
6578 		 dev->data->port_id);
6579 	return -ENOTSUP;
6580 }
6581 
6582 #define MLX5_POOL_QUERY_FREQ_US 1000000
6583 
6584 /**
6585  * Get number of all validate pools.
6586  *
6587  * @param[in] sh
6588  *   Pointer to mlx5_dev_ctx_shared object.
6589  *
6590  * @return
6591  *   The number of all validate pools.
6592  */
6593 static uint32_t
6594 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
6595 {
6596 	int i;
6597 	uint32_t pools_n = 0;
6598 
6599 	for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
6600 		pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
6601 	return pools_n;
6602 }
6603 
6604 /**
6605  * Set the periodic procedure for triggering asynchronous batch queries for all
6606  * the counter pools.
6607  *
6608  * @param[in] sh
6609  *   Pointer to mlx5_dev_ctx_shared object.
6610  */
6611 void
6612 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6613 {
6614 	uint32_t pools_n, us;
6615 
6616 	pools_n = mlx5_get_all_valid_pool_count(sh);
6617 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6618 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6619 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6620 		sh->cmng.query_thread_on = 0;
6621 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6622 	} else {
6623 		sh->cmng.query_thread_on = 1;
6624 	}
6625 }
6626 
6627 /**
6628  * The periodic procedure for triggering asynchronous batch queries for all the
6629  * counter pools. This function is probably called by the host thread.
6630  *
6631  * @param[in] arg
6632  *   The parameter for the alarm process.
6633  */
6634 void
6635 mlx5_flow_query_alarm(void *arg)
6636 {
6637 	struct mlx5_dev_ctx_shared *sh = arg;
6638 	struct mlx5_devx_obj *dcs;
6639 	uint16_t offset;
6640 	int ret;
6641 	uint8_t batch = sh->cmng.batch;
6642 	uint8_t age = sh->cmng.age;
6643 	uint16_t pool_index = sh->cmng.pool_index;
6644 	struct mlx5_pools_container *cont;
6645 	struct mlx5_flow_counter_pool *pool;
6646 	int cont_loop = MLX5_CCONT_TYPE_MAX;
6647 
6648 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6649 		goto set_alarm;
6650 next_container:
6651 	cont = MLX5_CNT_CONTAINER(sh, batch, age);
6652 	rte_spinlock_lock(&cont->resize_sl);
6653 	if (!cont->pools) {
6654 		rte_spinlock_unlock(&cont->resize_sl);
6655 		/* Check if all the containers are empty. */
6656 		if (unlikely(--cont_loop == 0))
6657 			goto set_alarm;
6658 		batch ^= 0x1;
6659 		pool_index = 0;
6660 		if (batch == 0 && pool_index == 0) {
6661 			age ^= 0x1;
6662 			sh->cmng.batch = batch;
6663 			sh->cmng.age = age;
6664 		}
6665 		goto next_container;
6666 	}
6667 	pool = cont->pools[pool_index];
6668 	rte_spinlock_unlock(&cont->resize_sl);
6669 	if (pool->raw_hw)
6670 		/* There is a pool query in progress. */
6671 		goto set_alarm;
6672 	pool->raw_hw =
6673 		LIST_FIRST(&sh->cmng.free_stat_raws);
6674 	if (!pool->raw_hw)
6675 		/* No free counter statistics raw memory. */
6676 		goto set_alarm;
6677 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
6678 							      (&pool->a64_dcs);
6679 	if (dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) {
6680 		/* Pool without valid counter. */
6681 		pool->raw_hw = NULL;
6682 		goto next_pool;
6683 	}
6684 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
6685 	/*
6686 	 * Identify the counters released between query trigger and query
6687 	 * handle more effiecntly. The counter released in this gap period
6688 	 * should wait for a new round of query as the new arrived packets
6689 	 * will not be taken into account.
6690 	 */
6691 	pool->query_gen++;
6692 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
6693 					       offset, NULL, NULL,
6694 					       pool->raw_hw->mem_mng->dm->id,
6695 					       (void *)(uintptr_t)
6696 					       (pool->raw_hw->data + offset),
6697 					       sh->devx_comp,
6698 					       (uint64_t)(uintptr_t)pool);
6699 	if (ret) {
6700 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6701 			" %d", pool->min_dcs->id);
6702 		pool->raw_hw = NULL;
6703 		goto set_alarm;
6704 	}
6705 	pool->raw_hw->min_dcs_id = dcs->id;
6706 	LIST_REMOVE(pool->raw_hw, next);
6707 	sh->cmng.pending_queries++;
6708 next_pool:
6709 	pool_index++;
6710 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6711 		batch ^= 0x1;
6712 		pool_index = 0;
6713 		if (batch == 0 && pool_index == 0)
6714 			age ^= 0x1;
6715 	}
6716 set_alarm:
6717 	sh->cmng.batch = batch;
6718 	sh->cmng.pool_index = pool_index;
6719 	sh->cmng.age = age;
6720 	mlx5_set_query_alarm(sh);
6721 }
6722 
6723 /**
6724  * Check and callback event for new aged flow in the counter pool
6725  *
6726  * @param[in] sh
6727  *   Pointer to mlx5_dev_ctx_shared object.
6728  * @param[in] pool
6729  *   Pointer to Current counter pool.
6730  */
6731 static void
6732 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6733 		   struct mlx5_flow_counter_pool *pool)
6734 {
6735 	struct mlx5_priv *priv;
6736 	struct mlx5_flow_counter *cnt;
6737 	struct mlx5_age_info *age_info;
6738 	struct mlx5_age_param *age_param;
6739 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6740 	struct mlx5_counter_stats_raw *prev = pool->raw;
6741 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6742 	uint32_t i;
6743 
6744 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6745 		cnt = MLX5_POOL_GET_CNT(pool, i);
6746 		age_param = MLX5_CNT_TO_AGE(cnt);
6747 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6748 			continue;
6749 		if (cur->data[i].hits != prev->data[i].hits) {
6750 			age_param->expire = curr + age_param->timeout;
6751 			continue;
6752 		}
6753 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6754 			continue;
6755 		/**
6756 		 * Hold the lock first, or if between the
6757 		 * state AGE_TMOUT and tailq operation the
6758 		 * release happened, the release procedure
6759 		 * may delete a non-existent tailq node.
6760 		 */
6761 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6762 		age_info = GET_PORT_AGE_INFO(priv);
6763 		rte_spinlock_lock(&age_info->aged_sl);
6764 		/* If the cpmset fails, release happens. */
6765 		if (rte_atomic16_cmpset((volatile uint16_t *)
6766 					&age_param->state,
6767 					AGE_CANDIDATE,
6768 					AGE_TMOUT) ==
6769 					AGE_CANDIDATE) {
6770 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6771 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6772 		}
6773 		rte_spinlock_unlock(&age_info->aged_sl);
6774 	}
6775 	for (i = 0; i < sh->max_port; i++) {
6776 		age_info = &sh->port[i].age_info;
6777 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6778 			continue;
6779 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6780 			rte_eth_dev_callback_process
6781 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
6782 				RTE_ETH_EVENT_FLOW_AGED, NULL);
6783 		age_info->flags = 0;
6784 	}
6785 }
6786 
6787 /**
6788  * Handler for the HW respond about ready values from an asynchronous batch
6789  * query. This function is probably called by the host thread.
6790  *
6791  * @param[in] sh
6792  *   The pointer to the shared device context.
6793  * @param[in] async_id
6794  *   The Devx async ID.
6795  * @param[in] status
6796  *   The status of the completion.
6797  */
6798 void
6799 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6800 				  uint64_t async_id, int status)
6801 {
6802 	struct mlx5_flow_counter_pool *pool =
6803 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6804 	struct mlx5_counter_stats_raw *raw_to_free;
6805 	uint8_t age = !!IS_AGE_POOL(pool);
6806 	uint8_t query_gen = pool->query_gen ^ 1;
6807 	struct mlx5_pools_container *cont =
6808 		MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6809 
6810 	if (unlikely(status)) {
6811 		raw_to_free = pool->raw_hw;
6812 	} else {
6813 		raw_to_free = pool->raw;
6814 		if (IS_AGE_POOL(pool))
6815 			mlx5_flow_aging_check(sh, pool);
6816 		rte_spinlock_lock(&pool->sl);
6817 		pool->raw = pool->raw_hw;
6818 		rte_spinlock_unlock(&pool->sl);
6819 		/* Be sure the new raw counters data is updated in memory. */
6820 		rte_io_wmb();
6821 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6822 			rte_spinlock_lock(&cont->csl);
6823 			TAILQ_CONCAT(&cont->counters,
6824 				     &pool->counters[query_gen], next);
6825 			rte_spinlock_unlock(&cont->csl);
6826 		}
6827 	}
6828 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6829 	pool->raw_hw = NULL;
6830 	sh->cmng.pending_queries--;
6831 }
6832 
6833 /**
6834  * Translate the rte_flow group index to HW table value.
6835  *
6836  * @param[in] attributes
6837  *   Pointer to flow attributes
6838  * @param[in] external
6839  *   Value is part of flow rule created by request external to PMD.
6840  * @param[in] group
6841  *   rte_flow group index value.
6842  * @param[out] fdb_def_rule
6843  *   Whether fdb jump to table 1 is configured.
6844  * @param[out] table
6845  *   HW table value.
6846  * @param[out] error
6847  *   Pointer to error structure.
6848  *
6849  * @return
6850  *   0 on success, a negative errno value otherwise and rte_errno is set.
6851  */
6852 int
6853 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6854 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6855 			 struct rte_flow_error *error)
6856 {
6857 	if (attributes->transfer && external && fdb_def_rule) {
6858 		if (group == UINT32_MAX)
6859 			return rte_flow_error_set
6860 						(error, EINVAL,
6861 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6862 						 NULL,
6863 						 "group index not supported");
6864 		*table = group + 1;
6865 	} else {
6866 		*table = group;
6867 	}
6868 	return 0;
6869 }
6870 
6871 /**
6872  * Discover availability of metadata reg_c's.
6873  *
6874  * Iteratively use test flows to check availability.
6875  *
6876  * @param[in] dev
6877  *   Pointer to the Ethernet device structure.
6878  *
6879  * @return
6880  *   0 on success, a negative errno value otherwise and rte_errno is set.
6881  */
6882 int
6883 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6884 {
6885 	struct mlx5_priv *priv = dev->data->dev_private;
6886 	struct mlx5_dev_config *config = &priv->config;
6887 	enum modify_reg idx;
6888 	int n = 0;
6889 
6890 	/* reg_c[0] and reg_c[1] are reserved. */
6891 	config->flow_mreg_c[n++] = REG_C_0;
6892 	config->flow_mreg_c[n++] = REG_C_1;
6893 	/* Discover availability of other reg_c's. */
6894 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6895 		struct rte_flow_attr attr = {
6896 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6897 			.priority = MLX5_FLOW_PRIO_RSVD,
6898 			.ingress = 1,
6899 		};
6900 		struct rte_flow_item items[] = {
6901 			[0] = {
6902 				.type = RTE_FLOW_ITEM_TYPE_END,
6903 			},
6904 		};
6905 		struct rte_flow_action actions[] = {
6906 			[0] = {
6907 				.type = (enum rte_flow_action_type)
6908 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6909 				.conf = &(struct mlx5_flow_action_copy_mreg){
6910 					.src = REG_C_1,
6911 					.dst = idx,
6912 				},
6913 			},
6914 			[1] = {
6915 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6916 				.conf = &(struct rte_flow_action_jump){
6917 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6918 				},
6919 			},
6920 			[2] = {
6921 				.type = RTE_FLOW_ACTION_TYPE_END,
6922 			},
6923 		};
6924 		uint32_t flow_idx;
6925 		struct rte_flow *flow;
6926 		struct rte_flow_error error;
6927 
6928 		if (!config->dv_flow_en)
6929 			break;
6930 		/* Create internal flow, validation skips copy action. */
6931 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6932 					    actions, false, &error);
6933 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6934 				      flow_idx);
6935 		if (!flow)
6936 			continue;
6937 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6938 			config->flow_mreg_c[n++] = idx;
6939 		flow_list_destroy(dev, NULL, flow_idx);
6940 	}
6941 	for (; n < MLX5_MREG_C_NUM; ++n)
6942 		config->flow_mreg_c[n] = REG_NON;
6943 	return 0;
6944 }
6945 
6946 /**
6947  * Dump flow raw hw data to file
6948  *
6949  * @param[in] dev
6950  *    The pointer to Ethernet device.
6951  * @param[in] file
6952  *   A pointer to a file for output.
6953  * @param[out] error
6954  *   Perform verbose error reporting if not NULL. PMDs initialize this
6955  *   structure in case of error only.
6956  * @return
6957  *   0 on success, a nagative value otherwise.
6958  */
6959 int
6960 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6961 		   FILE *file,
6962 		   struct rte_flow_error *error __rte_unused)
6963 {
6964 	struct mlx5_priv *priv = dev->data->dev_private;
6965 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6966 
6967 	if (!priv->config.dv_flow_en) {
6968 		if (fputs("device dv flow disabled\n", file) <= 0)
6969 			return -errno;
6970 		return -ENOTSUP;
6971 	}
6972 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6973 				       sh->tx_domain, file);
6974 }
6975 
6976 /**
6977  * Get aged-out flows.
6978  *
6979  * @param[in] dev
6980  *   Pointer to the Ethernet device structure.
6981  * @param[in] context
6982  *   The address of an array of pointers to the aged-out flows contexts.
6983  * @param[in] nb_countexts
6984  *   The length of context array pointers.
6985  * @param[out] error
6986  *   Perform verbose error reporting if not NULL. Initialized in case of
6987  *   error only.
6988  *
6989  * @return
6990  *   how many contexts get in success, otherwise negative errno value.
6991  *   if nb_contexts is 0, return the amount of all aged contexts.
6992  *   if nb_contexts is not 0 , return the amount of aged flows reported
6993  *   in the context array.
6994  */
6995 int
6996 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6997 			uint32_t nb_contexts, struct rte_flow_error *error)
6998 {
6999 	const struct mlx5_flow_driver_ops *fops;
7000 	struct rte_flow_attr attr = { .transfer = 0 };
7001 
7002 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7003 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7004 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7005 						    error);
7006 	}
7007 	DRV_LOG(ERR,
7008 		"port %u get aged flows is not supported.",
7009 		 dev->data->port_id);
7010 	return -ENOTSUP;
7011 }
7012