xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 657df3ca0aa632b028aedeb45a2a5bf66c973125)
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[out] error
804  *   Pointer to error structure.
805  *
806  * @return
807  *   0 on success, a negative errno value otherwise and rte_errno is set.
808  */
809 int
810 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
811 			  const uint8_t *mask,
812 			  const uint8_t *nic_mask,
813 			  unsigned int size,
814 			  struct rte_flow_error *error)
815 {
816 	unsigned int i;
817 
818 	MLX5_ASSERT(nic_mask);
819 	for (i = 0; i < size; ++i)
820 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
821 			return rte_flow_error_set(error, ENOTSUP,
822 						  RTE_FLOW_ERROR_TYPE_ITEM,
823 						  item,
824 						  "mask enables non supported"
825 						  " bits");
826 	if (!item->spec && (item->mask || item->last))
827 		return rte_flow_error_set(error, EINVAL,
828 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
829 					  "mask/last without a spec is not"
830 					  " supported");
831 	if (item->spec && item->last) {
832 		uint8_t spec[size];
833 		uint8_t last[size];
834 		unsigned int i;
835 		int ret;
836 
837 		for (i = 0; i < size; ++i) {
838 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
839 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
840 		}
841 		ret = memcmp(spec, last, size);
842 		if (ret != 0)
843 			return rte_flow_error_set(error, EINVAL,
844 						  RTE_FLOW_ERROR_TYPE_ITEM,
845 						  item,
846 						  "range is not valid");
847 	}
848 	return 0;
849 }
850 
851 /**
852  * Adjust the hash fields according to the @p flow information.
853  *
854  * @param[in] dev_flow.
855  *   Pointer to the mlx5_flow.
856  * @param[in] tunnel
857  *   1 when the hash field is for a tunnel item.
858  * @param[in] layer_types
859  *   ETH_RSS_* types.
860  * @param[in] hash_fields
861  *   Item hash fields.
862  *
863  * @return
864  *   The hash fields that should be used.
865  */
866 uint64_t
867 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
868 			    int tunnel __rte_unused, uint64_t layer_types,
869 			    uint64_t hash_fields)
870 {
871 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
872 	int rss_request_inner = rss_desc->level >= 2;
873 
874 	/* Check RSS hash level for tunnel. */
875 	if (tunnel && rss_request_inner)
876 		hash_fields |= IBV_RX_HASH_INNER;
877 	else if (tunnel || rss_request_inner)
878 		return 0;
879 #endif
880 	/* Check if requested layer matches RSS hash fields. */
881 	if (!(rss_desc->types & layer_types))
882 		return 0;
883 	return hash_fields;
884 }
885 
886 /**
887  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
888  * if several tunnel rules are used on this queue, the tunnel ptype will be
889  * cleared.
890  *
891  * @param rxq_ctrl
892  *   Rx queue to update.
893  */
894 static void
895 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
896 {
897 	unsigned int i;
898 	uint32_t tunnel_ptype = 0;
899 
900 	/* Look up for the ptype to use. */
901 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
902 		if (!rxq_ctrl->flow_tunnels_n[i])
903 			continue;
904 		if (!tunnel_ptype) {
905 			tunnel_ptype = tunnels_info[i].ptype;
906 		} else {
907 			tunnel_ptype = 0;
908 			break;
909 		}
910 	}
911 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
912 }
913 
914 /**
915  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
916  * flow.
917  *
918  * @param[in] dev
919  *   Pointer to the Ethernet device structure.
920  * @param[in] dev_handle
921  *   Pointer to device flow handle structure.
922  */
923 static void
924 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
925 		       struct mlx5_flow_handle *dev_handle)
926 {
927 	struct mlx5_priv *priv = dev->data->dev_private;
928 	const int mark = dev_handle->mark;
929 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
930 	struct mlx5_hrxq *hrxq;
931 	unsigned int i;
932 
933 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
934 		return;
935 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
936 			      dev_handle->rix_hrxq);
937 	if (!hrxq)
938 		return;
939 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
940 		int idx = hrxq->ind_table->queues[i];
941 		struct mlx5_rxq_ctrl *rxq_ctrl =
942 			container_of((*priv->rxqs)[idx],
943 				     struct mlx5_rxq_ctrl, rxq);
944 
945 		/*
946 		 * To support metadata register copy on Tx loopback,
947 		 * this must be always enabled (metadata may arive
948 		 * from other port - not from local flows only.
949 		 */
950 		if (priv->config.dv_flow_en &&
951 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
952 		    mlx5_flow_ext_mreg_supported(dev)) {
953 			rxq_ctrl->rxq.mark = 1;
954 			rxq_ctrl->flow_mark_n = 1;
955 		} else if (mark) {
956 			rxq_ctrl->rxq.mark = 1;
957 			rxq_ctrl->flow_mark_n++;
958 		}
959 		if (tunnel) {
960 			unsigned int j;
961 
962 			/* Increase the counter matching the flow. */
963 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
964 				if ((tunnels_info[j].tunnel &
965 				     dev_handle->layers) ==
966 				    tunnels_info[j].tunnel) {
967 					rxq_ctrl->flow_tunnels_n[j]++;
968 					break;
969 				}
970 			}
971 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
972 		}
973 	}
974 }
975 
976 /**
977  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
978  *
979  * @param[in] dev
980  *   Pointer to the Ethernet device structure.
981  * @param[in] flow
982  *   Pointer to flow structure.
983  */
984 static void
985 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
986 {
987 	struct mlx5_priv *priv = dev->data->dev_private;
988 	uint32_t handle_idx;
989 	struct mlx5_flow_handle *dev_handle;
990 
991 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
992 		       handle_idx, dev_handle, next)
993 		flow_drv_rxq_flags_set(dev, dev_handle);
994 }
995 
996 /**
997  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
998  * device flow if no other flow uses it with the same kind of request.
999  *
1000  * @param dev
1001  *   Pointer to Ethernet device.
1002  * @param[in] dev_handle
1003  *   Pointer to the device flow handle structure.
1004  */
1005 static void
1006 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1007 			struct mlx5_flow_handle *dev_handle)
1008 {
1009 	struct mlx5_priv *priv = dev->data->dev_private;
1010 	const int mark = dev_handle->mark;
1011 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1012 	struct mlx5_hrxq *hrxq;
1013 	unsigned int i;
1014 
1015 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1016 		return;
1017 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1018 			      dev_handle->rix_hrxq);
1019 	if (!hrxq)
1020 		return;
1021 	MLX5_ASSERT(dev->data->dev_started);
1022 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1023 		int idx = hrxq->ind_table->queues[i];
1024 		struct mlx5_rxq_ctrl *rxq_ctrl =
1025 			container_of((*priv->rxqs)[idx],
1026 				     struct mlx5_rxq_ctrl, rxq);
1027 
1028 		if (priv->config.dv_flow_en &&
1029 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1030 		    mlx5_flow_ext_mreg_supported(dev)) {
1031 			rxq_ctrl->rxq.mark = 1;
1032 			rxq_ctrl->flow_mark_n = 1;
1033 		} else if (mark) {
1034 			rxq_ctrl->flow_mark_n--;
1035 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1036 		}
1037 		if (tunnel) {
1038 			unsigned int j;
1039 
1040 			/* Decrease the counter matching the flow. */
1041 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1042 				if ((tunnels_info[j].tunnel &
1043 				     dev_handle->layers) ==
1044 				    tunnels_info[j].tunnel) {
1045 					rxq_ctrl->flow_tunnels_n[j]--;
1046 					break;
1047 				}
1048 			}
1049 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1050 		}
1051 	}
1052 }
1053 
1054 /**
1055  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1056  * @p flow if no other flow uses it with the same kind of request.
1057  *
1058  * @param dev
1059  *   Pointer to Ethernet device.
1060  * @param[in] flow
1061  *   Pointer to the flow.
1062  */
1063 static void
1064 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1065 {
1066 	struct mlx5_priv *priv = dev->data->dev_private;
1067 	uint32_t handle_idx;
1068 	struct mlx5_flow_handle *dev_handle;
1069 
1070 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1071 		       handle_idx, dev_handle, next)
1072 		flow_drv_rxq_flags_trim(dev, dev_handle);
1073 }
1074 
1075 /**
1076  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1077  *
1078  * @param dev
1079  *   Pointer to Ethernet device.
1080  */
1081 static void
1082 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1083 {
1084 	struct mlx5_priv *priv = dev->data->dev_private;
1085 	unsigned int i;
1086 
1087 	for (i = 0; i != priv->rxqs_n; ++i) {
1088 		struct mlx5_rxq_ctrl *rxq_ctrl;
1089 		unsigned int j;
1090 
1091 		if (!(*priv->rxqs)[i])
1092 			continue;
1093 		rxq_ctrl = container_of((*priv->rxqs)[i],
1094 					struct mlx5_rxq_ctrl, rxq);
1095 		rxq_ctrl->flow_mark_n = 0;
1096 		rxq_ctrl->rxq.mark = 0;
1097 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1098 			rxq_ctrl->flow_tunnels_n[j] = 0;
1099 		rxq_ctrl->rxq.tunnel = 0;
1100 	}
1101 }
1102 
1103 /**
1104  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1105  *
1106  * @param[in] dev
1107  *   Pointer to the Ethernet device structure.
1108  */
1109 void
1110 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1111 {
1112 	struct mlx5_priv *priv = dev->data->dev_private;
1113 	struct mlx5_rxq_data *data;
1114 	unsigned int i;
1115 
1116 	for (i = 0; i != priv->rxqs_n; ++i) {
1117 		if (!(*priv->rxqs)[i])
1118 			continue;
1119 		data = (*priv->rxqs)[i];
1120 		if (!rte_flow_dynf_metadata_avail()) {
1121 			data->dynf_meta = 0;
1122 			data->flow_meta_mask = 0;
1123 			data->flow_meta_offset = -1;
1124 		} else {
1125 			data->dynf_meta = 1;
1126 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1127 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1128 		}
1129 	}
1130 }
1131 
1132 /*
1133  * return a pointer to the desired action in the list of actions.
1134  *
1135  * @param[in] actions
1136  *   The list of actions to search the action in.
1137  * @param[in] action
1138  *   The action to find.
1139  *
1140  * @return
1141  *   Pointer to the action in the list, if found. NULL otherwise.
1142  */
1143 const struct rte_flow_action *
1144 mlx5_flow_find_action(const struct rte_flow_action *actions,
1145 		      enum rte_flow_action_type action)
1146 {
1147 	if (actions == NULL)
1148 		return NULL;
1149 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1150 		if (actions->type == action)
1151 			return actions;
1152 	return NULL;
1153 }
1154 
1155 /*
1156  * Validate the flag action.
1157  *
1158  * @param[in] action_flags
1159  *   Bit-fields that holds the actions detected until now.
1160  * @param[in] attr
1161  *   Attributes of flow that includes this action.
1162  * @param[out] error
1163  *   Pointer to error structure.
1164  *
1165  * @return
1166  *   0 on success, a negative errno value otherwise and rte_errno is set.
1167  */
1168 int
1169 mlx5_flow_validate_action_flag(uint64_t action_flags,
1170 			       const struct rte_flow_attr *attr,
1171 			       struct rte_flow_error *error)
1172 {
1173 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1174 		return rte_flow_error_set(error, EINVAL,
1175 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1176 					  "can't mark and flag in same flow");
1177 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1178 		return rte_flow_error_set(error, EINVAL,
1179 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1180 					  "can't have 2 flag"
1181 					  " actions in same flow");
1182 	if (attr->egress)
1183 		return rte_flow_error_set(error, ENOTSUP,
1184 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1185 					  "flag action not supported for "
1186 					  "egress");
1187 	return 0;
1188 }
1189 
1190 /*
1191  * Validate the mark action.
1192  *
1193  * @param[in] action
1194  *   Pointer to the queue action.
1195  * @param[in] action_flags
1196  *   Bit-fields that holds the actions detected until now.
1197  * @param[in] attr
1198  *   Attributes of flow that includes this action.
1199  * @param[out] error
1200  *   Pointer to error structure.
1201  *
1202  * @return
1203  *   0 on success, a negative errno value otherwise and rte_errno is set.
1204  */
1205 int
1206 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1207 			       uint64_t action_flags,
1208 			       const struct rte_flow_attr *attr,
1209 			       struct rte_flow_error *error)
1210 {
1211 	const struct rte_flow_action_mark *mark = action->conf;
1212 
1213 	if (!mark)
1214 		return rte_flow_error_set(error, EINVAL,
1215 					  RTE_FLOW_ERROR_TYPE_ACTION,
1216 					  action,
1217 					  "configuration cannot be null");
1218 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1219 		return rte_flow_error_set(error, EINVAL,
1220 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1221 					  &mark->id,
1222 					  "mark id must in 0 <= id < "
1223 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1224 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1225 		return rte_flow_error_set(error, EINVAL,
1226 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1227 					  "can't flag and mark in same flow");
1228 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1229 		return rte_flow_error_set(error, EINVAL,
1230 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1231 					  "can't have 2 mark actions in same"
1232 					  " flow");
1233 	if (attr->egress)
1234 		return rte_flow_error_set(error, ENOTSUP,
1235 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1236 					  "mark action not supported for "
1237 					  "egress");
1238 	return 0;
1239 }
1240 
1241 /*
1242  * Validate the drop action.
1243  *
1244  * @param[in] action_flags
1245  *   Bit-fields that holds the actions detected until now.
1246  * @param[in] attr
1247  *   Attributes of flow that includes this action.
1248  * @param[out] error
1249  *   Pointer to error structure.
1250  *
1251  * @return
1252  *   0 on success, a negative errno value otherwise and rte_errno is set.
1253  */
1254 int
1255 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1256 			       const struct rte_flow_attr *attr,
1257 			       struct rte_flow_error *error)
1258 {
1259 	if (attr->egress)
1260 		return rte_flow_error_set(error, ENOTSUP,
1261 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1262 					  "drop action not supported for "
1263 					  "egress");
1264 	return 0;
1265 }
1266 
1267 /*
1268  * Validate the queue action.
1269  *
1270  * @param[in] action
1271  *   Pointer to the queue action.
1272  * @param[in] action_flags
1273  *   Bit-fields that holds the actions detected until now.
1274  * @param[in] dev
1275  *   Pointer to the Ethernet device structure.
1276  * @param[in] attr
1277  *   Attributes of flow that includes this action.
1278  * @param[out] error
1279  *   Pointer to error structure.
1280  *
1281  * @return
1282  *   0 on success, a negative errno value otherwise and rte_errno is set.
1283  */
1284 int
1285 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1286 				uint64_t action_flags,
1287 				struct rte_eth_dev *dev,
1288 				const struct rte_flow_attr *attr,
1289 				struct rte_flow_error *error)
1290 {
1291 	struct mlx5_priv *priv = dev->data->dev_private;
1292 	const struct rte_flow_action_queue *queue = action->conf;
1293 
1294 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1295 		return rte_flow_error_set(error, EINVAL,
1296 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1297 					  "can't have 2 fate actions in"
1298 					  " same flow");
1299 	if (!priv->rxqs_n)
1300 		return rte_flow_error_set(error, EINVAL,
1301 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1302 					  NULL, "No Rx queues configured");
1303 	if (queue->index >= priv->rxqs_n)
1304 		return rte_flow_error_set(error, EINVAL,
1305 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1306 					  &queue->index,
1307 					  "queue index out of range");
1308 	if (!(*priv->rxqs)[queue->index])
1309 		return rte_flow_error_set(error, EINVAL,
1310 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1311 					  &queue->index,
1312 					  "queue is not configured");
1313 	if (attr->egress)
1314 		return rte_flow_error_set(error, ENOTSUP,
1315 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1316 					  "queue action not supported for "
1317 					  "egress");
1318 	return 0;
1319 }
1320 
1321 /*
1322  * Validate the rss action.
1323  *
1324  * @param[in] action
1325  *   Pointer to the queue action.
1326  * @param[in] action_flags
1327  *   Bit-fields that holds the actions detected until now.
1328  * @param[in] dev
1329  *   Pointer to the Ethernet device structure.
1330  * @param[in] attr
1331  *   Attributes of flow that includes this action.
1332  * @param[in] item_flags
1333  *   Items that were detected.
1334  * @param[out] error
1335  *   Pointer to error structure.
1336  *
1337  * @return
1338  *   0 on success, a negative errno value otherwise and rte_errno is set.
1339  */
1340 int
1341 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1342 			      uint64_t action_flags,
1343 			      struct rte_eth_dev *dev,
1344 			      const struct rte_flow_attr *attr,
1345 			      uint64_t item_flags,
1346 			      struct rte_flow_error *error)
1347 {
1348 	struct mlx5_priv *priv = dev->data->dev_private;
1349 	const struct rte_flow_action_rss *rss = action->conf;
1350 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1351 	unsigned int i;
1352 
1353 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1354 		return rte_flow_error_set(error, EINVAL,
1355 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1356 					  "can't have 2 fate actions"
1357 					  " in same flow");
1358 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1359 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1360 		return rte_flow_error_set(error, ENOTSUP,
1361 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1362 					  &rss->func,
1363 					  "RSS hash function not supported");
1364 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1365 	if (rss->level > 2)
1366 #else
1367 	if (rss->level > 1)
1368 #endif
1369 		return rte_flow_error_set(error, ENOTSUP,
1370 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1371 					  &rss->level,
1372 					  "tunnel RSS is not supported");
1373 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1374 	if (rss->key_len == 0 && rss->key != NULL)
1375 		return rte_flow_error_set(error, ENOTSUP,
1376 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1377 					  &rss->key_len,
1378 					  "RSS hash key length 0");
1379 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1380 		return rte_flow_error_set(error, ENOTSUP,
1381 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1382 					  &rss->key_len,
1383 					  "RSS hash key too small");
1384 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1385 		return rte_flow_error_set(error, ENOTSUP,
1386 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1387 					  &rss->key_len,
1388 					  "RSS hash key too large");
1389 	if (rss->queue_num > priv->config.ind_table_max_size)
1390 		return rte_flow_error_set(error, ENOTSUP,
1391 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1392 					  &rss->queue_num,
1393 					  "number of queues too large");
1394 	if (rss->types & MLX5_RSS_HF_MASK)
1395 		return rte_flow_error_set(error, ENOTSUP,
1396 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1397 					  &rss->types,
1398 					  "some RSS protocols are not"
1399 					  " supported");
1400 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1401 	    !(rss->types & ETH_RSS_IP))
1402 		return rte_flow_error_set(error, EINVAL,
1403 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1404 					  "L3 partial RSS requested but L3 RSS"
1405 					  " type not specified");
1406 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1407 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1408 		return rte_flow_error_set(error, EINVAL,
1409 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1410 					  "L4 partial RSS requested but L4 RSS"
1411 					  " type not specified");
1412 	if (!priv->rxqs_n)
1413 		return rte_flow_error_set(error, EINVAL,
1414 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1415 					  NULL, "No Rx queues configured");
1416 	if (!rss->queue_num)
1417 		return rte_flow_error_set(error, EINVAL,
1418 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1419 					  NULL, "No queues configured");
1420 	for (i = 0; i != rss->queue_num; ++i) {
1421 		if (rss->queue[i] >= priv->rxqs_n)
1422 			return rte_flow_error_set
1423 				(error, EINVAL,
1424 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1425 				 &rss->queue[i], "queue index out of range");
1426 		if (!(*priv->rxqs)[rss->queue[i]])
1427 			return rte_flow_error_set
1428 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1429 				 &rss->queue[i], "queue is not configured");
1430 	}
1431 	if (attr->egress)
1432 		return rte_flow_error_set(error, ENOTSUP,
1433 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1434 					  "rss action not supported for "
1435 					  "egress");
1436 	if (rss->level > 1 && !tunnel)
1437 		return rte_flow_error_set(error, EINVAL,
1438 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1439 					  "inner RSS is not supported for "
1440 					  "non-tunnel flows");
1441 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1442 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1443 		return rte_flow_error_set(error, EINVAL,
1444 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1445 					  "RSS on eCPRI is not supported now");
1446 	}
1447 	return 0;
1448 }
1449 
1450 /*
1451  * Validate the default miss action.
1452  *
1453  * @param[in] action_flags
1454  *   Bit-fields that holds the actions detected until now.
1455  * @param[out] error
1456  *   Pointer to error structure.
1457  *
1458  * @return
1459  *   0 on success, a negative errno value otherwise and rte_errno is set.
1460  */
1461 int
1462 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1463 				const struct rte_flow_attr *attr,
1464 				struct rte_flow_error *error)
1465 {
1466 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1467 		return rte_flow_error_set(error, EINVAL,
1468 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1469 					  "can't have 2 fate actions in"
1470 					  " same flow");
1471 	if (attr->egress)
1472 		return rte_flow_error_set(error, ENOTSUP,
1473 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1474 					  "default miss action not supported "
1475 					  "for egress");
1476 	if (attr->group)
1477 		return rte_flow_error_set(error, ENOTSUP,
1478 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1479 					  "only group 0 is supported");
1480 	if (attr->transfer)
1481 		return rte_flow_error_set(error, ENOTSUP,
1482 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1483 					  NULL, "transfer is not supported");
1484 	return 0;
1485 }
1486 
1487 /*
1488  * Validate the count action.
1489  *
1490  * @param[in] dev
1491  *   Pointer to the Ethernet device structure.
1492  * @param[in] attr
1493  *   Attributes of flow that includes this action.
1494  * @param[out] error
1495  *   Pointer to error structure.
1496  *
1497  * @return
1498  *   0 on success, a negative errno value otherwise and rte_errno is set.
1499  */
1500 int
1501 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1502 				const struct rte_flow_attr *attr,
1503 				struct rte_flow_error *error)
1504 {
1505 	if (attr->egress)
1506 		return rte_flow_error_set(error, ENOTSUP,
1507 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1508 					  "count action not supported for "
1509 					  "egress");
1510 	return 0;
1511 }
1512 
1513 /**
1514  * Verify the @p attributes will be correctly understood by the NIC and store
1515  * them in the @p flow if everything is correct.
1516  *
1517  * @param[in] dev
1518  *   Pointer to the Ethernet device structure.
1519  * @param[in] attributes
1520  *   Pointer to flow attributes
1521  * @param[out] error
1522  *   Pointer to error structure.
1523  *
1524  * @return
1525  *   0 on success, a negative errno value otherwise and rte_errno is set.
1526  */
1527 int
1528 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1529 			      const struct rte_flow_attr *attributes,
1530 			      struct rte_flow_error *error)
1531 {
1532 	struct mlx5_priv *priv = dev->data->dev_private;
1533 	uint32_t priority_max = priv->config.flow_prio - 1;
1534 
1535 	if (attributes->group)
1536 		return rte_flow_error_set(error, ENOTSUP,
1537 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1538 					  NULL, "groups is not supported");
1539 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1540 	    attributes->priority >= priority_max)
1541 		return rte_flow_error_set(error, ENOTSUP,
1542 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1543 					  NULL, "priority out of range");
1544 	if (attributes->egress)
1545 		return rte_flow_error_set(error, ENOTSUP,
1546 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1547 					  "egress is not supported");
1548 	if (attributes->transfer && !priv->config.dv_esw_en)
1549 		return rte_flow_error_set(error, ENOTSUP,
1550 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1551 					  NULL, "transfer is not supported");
1552 	if (!attributes->ingress)
1553 		return rte_flow_error_set(error, EINVAL,
1554 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1555 					  NULL,
1556 					  "ingress attribute is mandatory");
1557 	return 0;
1558 }
1559 
1560 /**
1561  * Validate ICMP6 item.
1562  *
1563  * @param[in] item
1564  *   Item specification.
1565  * @param[in] item_flags
1566  *   Bit-fields that holds the items detected until now.
1567  * @param[out] error
1568  *   Pointer to error structure.
1569  *
1570  * @return
1571  *   0 on success, a negative errno value otherwise and rte_errno is set.
1572  */
1573 int
1574 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1575 			       uint64_t item_flags,
1576 			       uint8_t target_protocol,
1577 			       struct rte_flow_error *error)
1578 {
1579 	const struct rte_flow_item_icmp6 *mask = item->mask;
1580 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1581 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1582 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1583 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1584 				      MLX5_FLOW_LAYER_OUTER_L4;
1585 	int ret;
1586 
1587 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1588 		return rte_flow_error_set(error, EINVAL,
1589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1590 					  "protocol filtering not compatible"
1591 					  " with ICMP6 layer");
1592 	if (!(item_flags & l3m))
1593 		return rte_flow_error_set(error, EINVAL,
1594 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1595 					  "IPv6 is mandatory to filter on"
1596 					  " ICMP6");
1597 	if (item_flags & l4m)
1598 		return rte_flow_error_set(error, EINVAL,
1599 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1600 					  "multiple L4 layers not supported");
1601 	if (!mask)
1602 		mask = &rte_flow_item_icmp6_mask;
1603 	ret = mlx5_flow_item_acceptable
1604 		(item, (const uint8_t *)mask,
1605 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1606 		 sizeof(struct rte_flow_item_icmp6), error);
1607 	if (ret < 0)
1608 		return ret;
1609 	return 0;
1610 }
1611 
1612 /**
1613  * Validate ICMP item.
1614  *
1615  * @param[in] item
1616  *   Item specification.
1617  * @param[in] item_flags
1618  *   Bit-fields that holds the items detected until now.
1619  * @param[out] error
1620  *   Pointer to error structure.
1621  *
1622  * @return
1623  *   0 on success, a negative errno value otherwise and rte_errno is set.
1624  */
1625 int
1626 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1627 			     uint64_t item_flags,
1628 			     uint8_t target_protocol,
1629 			     struct rte_flow_error *error)
1630 {
1631 	const struct rte_flow_item_icmp *mask = item->mask;
1632 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1633 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1634 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1635 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1636 				      MLX5_FLOW_LAYER_OUTER_L4;
1637 	int ret;
1638 
1639 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1640 		return rte_flow_error_set(error, EINVAL,
1641 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1642 					  "protocol filtering not compatible"
1643 					  " with ICMP layer");
1644 	if (!(item_flags & l3m))
1645 		return rte_flow_error_set(error, EINVAL,
1646 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1647 					  "IPv4 is mandatory to filter"
1648 					  " on ICMP");
1649 	if (item_flags & l4m)
1650 		return rte_flow_error_set(error, EINVAL,
1651 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1652 					  "multiple L4 layers not supported");
1653 	if (!mask)
1654 		mask = &rte_flow_item_icmp_mask;
1655 	ret = mlx5_flow_item_acceptable
1656 		(item, (const uint8_t *)mask,
1657 		 (const uint8_t *)&rte_flow_item_icmp_mask,
1658 		 sizeof(struct rte_flow_item_icmp), error);
1659 	if (ret < 0)
1660 		return ret;
1661 	return 0;
1662 }
1663 
1664 /**
1665  * Validate Ethernet item.
1666  *
1667  * @param[in] item
1668  *   Item specification.
1669  * @param[in] item_flags
1670  *   Bit-fields that holds the items detected until now.
1671  * @param[out] error
1672  *   Pointer to error structure.
1673  *
1674  * @return
1675  *   0 on success, a negative errno value otherwise and rte_errno is set.
1676  */
1677 int
1678 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1679 			    uint64_t item_flags,
1680 			    struct rte_flow_error *error)
1681 {
1682 	const struct rte_flow_item_eth *mask = item->mask;
1683 	const struct rte_flow_item_eth nic_mask = {
1684 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1685 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1686 		.type = RTE_BE16(0xffff),
1687 	};
1688 	int ret;
1689 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1690 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1691 				       MLX5_FLOW_LAYER_OUTER_L2;
1692 
1693 	if (item_flags & ethm)
1694 		return rte_flow_error_set(error, ENOTSUP,
1695 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1696 					  "multiple L2 layers not supported");
1697 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1698 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1699 		return rte_flow_error_set(error, EINVAL,
1700 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1701 					  "L2 layer should not follow "
1702 					  "L3 layers");
1703 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1704 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1705 		return rte_flow_error_set(error, EINVAL,
1706 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 					  "L2 layer should not follow VLAN");
1708 	if (!mask)
1709 		mask = &rte_flow_item_eth_mask;
1710 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1711 					(const uint8_t *)&nic_mask,
1712 					sizeof(struct rte_flow_item_eth),
1713 					error);
1714 	return ret;
1715 }
1716 
1717 /**
1718  * Validate VLAN item.
1719  *
1720  * @param[in] item
1721  *   Item specification.
1722  * @param[in] item_flags
1723  *   Bit-fields that holds the items detected until now.
1724  * @param[in] dev
1725  *   Ethernet device flow is being created on.
1726  * @param[out] error
1727  *   Pointer to error structure.
1728  *
1729  * @return
1730  *   0 on success, a negative errno value otherwise and rte_errno is set.
1731  */
1732 int
1733 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1734 			     uint64_t item_flags,
1735 			     struct rte_eth_dev *dev,
1736 			     struct rte_flow_error *error)
1737 {
1738 	const struct rte_flow_item_vlan *spec = item->spec;
1739 	const struct rte_flow_item_vlan *mask = item->mask;
1740 	const struct rte_flow_item_vlan nic_mask = {
1741 		.tci = RTE_BE16(UINT16_MAX),
1742 		.inner_type = RTE_BE16(UINT16_MAX),
1743 	};
1744 	uint16_t vlan_tag = 0;
1745 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1746 	int ret;
1747 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1748 					MLX5_FLOW_LAYER_INNER_L4) :
1749 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1750 					MLX5_FLOW_LAYER_OUTER_L4);
1751 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1752 					MLX5_FLOW_LAYER_OUTER_VLAN;
1753 
1754 	if (item_flags & vlanm)
1755 		return rte_flow_error_set(error, EINVAL,
1756 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1757 					  "multiple VLAN layers not supported");
1758 	else if ((item_flags & l34m) != 0)
1759 		return rte_flow_error_set(error, EINVAL,
1760 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1761 					  "VLAN cannot follow L3/L4 layer");
1762 	if (!mask)
1763 		mask = &rte_flow_item_vlan_mask;
1764 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1765 					(const uint8_t *)&nic_mask,
1766 					sizeof(struct rte_flow_item_vlan),
1767 					error);
1768 	if (ret)
1769 		return ret;
1770 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1771 		struct mlx5_priv *priv = dev->data->dev_private;
1772 
1773 		if (priv->vmwa_context) {
1774 			/*
1775 			 * Non-NULL context means we have a virtual machine
1776 			 * and SR-IOV enabled, we have to create VLAN interface
1777 			 * to make hypervisor to setup E-Switch vport
1778 			 * context correctly. We avoid creating the multiple
1779 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1780 			 */
1781 			return rte_flow_error_set(error, EINVAL,
1782 						  RTE_FLOW_ERROR_TYPE_ITEM,
1783 						  item,
1784 						  "VLAN tag mask is not"
1785 						  " supported in virtual"
1786 						  " environment");
1787 		}
1788 	}
1789 	if (spec) {
1790 		vlan_tag = spec->tci;
1791 		vlan_tag &= mask->tci;
1792 	}
1793 	/*
1794 	 * From verbs perspective an empty VLAN is equivalent
1795 	 * to a packet without VLAN layer.
1796 	 */
1797 	if (!vlan_tag)
1798 		return rte_flow_error_set(error, EINVAL,
1799 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1800 					  item->spec,
1801 					  "VLAN cannot be empty");
1802 	return 0;
1803 }
1804 
1805 /**
1806  * Validate IPV4 item.
1807  *
1808  * @param[in] item
1809  *   Item specification.
1810  * @param[in] item_flags
1811  *   Bit-fields that holds the items detected until now.
1812  * @param[in] last_item
1813  *   Previous validated item in the pattern items.
1814  * @param[in] ether_type
1815  *   Type in the ethernet layer header (including dot1q).
1816  * @param[in] acc_mask
1817  *   Acceptable mask, if NULL default internal default mask
1818  *   will be used to check whether item fields are supported.
1819  * @param[out] error
1820  *   Pointer to error structure.
1821  *
1822  * @return
1823  *   0 on success, a negative errno value otherwise and rte_errno is set.
1824  */
1825 int
1826 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1827 			     uint64_t item_flags,
1828 			     uint64_t last_item,
1829 			     uint16_t ether_type,
1830 			     const struct rte_flow_item_ipv4 *acc_mask,
1831 			     struct rte_flow_error *error)
1832 {
1833 	const struct rte_flow_item_ipv4 *mask = item->mask;
1834 	const struct rte_flow_item_ipv4 *spec = item->spec;
1835 	const struct rte_flow_item_ipv4 nic_mask = {
1836 		.hdr = {
1837 			.src_addr = RTE_BE32(0xffffffff),
1838 			.dst_addr = RTE_BE32(0xffffffff),
1839 			.type_of_service = 0xff,
1840 			.next_proto_id = 0xff,
1841 		},
1842 	};
1843 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1844 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1845 				      MLX5_FLOW_LAYER_OUTER_L3;
1846 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1847 				      MLX5_FLOW_LAYER_OUTER_L4;
1848 	int ret;
1849 	uint8_t next_proto = 0xFF;
1850 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1851 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1852 				  MLX5_FLOW_LAYER_INNER_VLAN);
1853 
1854 	if ((last_item & l2_vlan) && ether_type &&
1855 	    ether_type != RTE_ETHER_TYPE_IPV4)
1856 		return rte_flow_error_set(error, EINVAL,
1857 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1858 					  "IPv4 cannot follow L2/VLAN layer "
1859 					  "which ether type is not IPv4");
1860 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1861 		if (mask && spec)
1862 			next_proto = mask->hdr.next_proto_id &
1863 				     spec->hdr.next_proto_id;
1864 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1865 			return rte_flow_error_set(error, EINVAL,
1866 						  RTE_FLOW_ERROR_TYPE_ITEM,
1867 						  item,
1868 						  "multiple tunnel "
1869 						  "not supported");
1870 	}
1871 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1872 		return rte_flow_error_set(error, EINVAL,
1873 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1874 					  "wrong tunnel type - IPv6 specified "
1875 					  "but IPv4 item provided");
1876 	if (item_flags & l3m)
1877 		return rte_flow_error_set(error, ENOTSUP,
1878 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1879 					  "multiple L3 layers not supported");
1880 	else if (item_flags & l4m)
1881 		return rte_flow_error_set(error, EINVAL,
1882 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1883 					  "L3 cannot follow an L4 layer.");
1884 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1885 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1886 		return rte_flow_error_set(error, EINVAL,
1887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 					  "L3 cannot follow an NVGRE layer.");
1889 	if (!mask)
1890 		mask = &rte_flow_item_ipv4_mask;
1891 	else if (mask->hdr.next_proto_id != 0 &&
1892 		 mask->hdr.next_proto_id != 0xff)
1893 		return rte_flow_error_set(error, EINVAL,
1894 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1895 					  "partial mask is not supported"
1896 					  " for protocol");
1897 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1898 					acc_mask ? (const uint8_t *)acc_mask
1899 						 : (const uint8_t *)&nic_mask,
1900 					sizeof(struct rte_flow_item_ipv4),
1901 					error);
1902 	if (ret < 0)
1903 		return ret;
1904 	return 0;
1905 }
1906 
1907 /**
1908  * Validate IPV6 item.
1909  *
1910  * @param[in] item
1911  *   Item specification.
1912  * @param[in] item_flags
1913  *   Bit-fields that holds the items detected until now.
1914  * @param[in] last_item
1915  *   Previous validated item in the pattern items.
1916  * @param[in] ether_type
1917  *   Type in the ethernet layer header (including dot1q).
1918  * @param[in] acc_mask
1919  *   Acceptable mask, if NULL default internal default mask
1920  *   will be used to check whether item fields are supported.
1921  * @param[out] error
1922  *   Pointer to error structure.
1923  *
1924  * @return
1925  *   0 on success, a negative errno value otherwise and rte_errno is set.
1926  */
1927 int
1928 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1929 			     uint64_t item_flags,
1930 			     uint64_t last_item,
1931 			     uint16_t ether_type,
1932 			     const struct rte_flow_item_ipv6 *acc_mask,
1933 			     struct rte_flow_error *error)
1934 {
1935 	const struct rte_flow_item_ipv6 *mask = item->mask;
1936 	const struct rte_flow_item_ipv6 *spec = item->spec;
1937 	const struct rte_flow_item_ipv6 nic_mask = {
1938 		.hdr = {
1939 			.src_addr =
1940 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1941 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1942 			.dst_addr =
1943 				"\xff\xff\xff\xff\xff\xff\xff\xff"
1944 				"\xff\xff\xff\xff\xff\xff\xff\xff",
1945 			.vtc_flow = RTE_BE32(0xffffffff),
1946 			.proto = 0xff,
1947 		},
1948 	};
1949 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1950 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1951 				      MLX5_FLOW_LAYER_OUTER_L3;
1952 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1953 				      MLX5_FLOW_LAYER_OUTER_L4;
1954 	int ret;
1955 	uint8_t next_proto = 0xFF;
1956 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1957 				  MLX5_FLOW_LAYER_OUTER_VLAN |
1958 				  MLX5_FLOW_LAYER_INNER_VLAN);
1959 
1960 	if ((last_item & l2_vlan) && ether_type &&
1961 	    ether_type != RTE_ETHER_TYPE_IPV6)
1962 		return rte_flow_error_set(error, EINVAL,
1963 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1964 					  "IPv6 cannot follow L2/VLAN layer "
1965 					  "which ether type is not IPv6");
1966 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1967 		if (mask && spec)
1968 			next_proto = mask->hdr.proto & spec->hdr.proto;
1969 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1970 			return rte_flow_error_set(error, EINVAL,
1971 						  RTE_FLOW_ERROR_TYPE_ITEM,
1972 						  item,
1973 						  "multiple tunnel "
1974 						  "not supported");
1975 	}
1976 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
1977 		return rte_flow_error_set(error, EINVAL,
1978 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1979 					  "wrong tunnel type - IPv4 specified "
1980 					  "but IPv6 item provided");
1981 	if (item_flags & l3m)
1982 		return rte_flow_error_set(error, ENOTSUP,
1983 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1984 					  "multiple L3 layers not supported");
1985 	else if (item_flags & l4m)
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1988 					  "L3 cannot follow an L4 layer.");
1989 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1990 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1991 		return rte_flow_error_set(error, EINVAL,
1992 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1993 					  "L3 cannot follow an NVGRE layer.");
1994 	if (!mask)
1995 		mask = &rte_flow_item_ipv6_mask;
1996 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1997 					acc_mask ? (const uint8_t *)acc_mask
1998 						 : (const uint8_t *)&nic_mask,
1999 					sizeof(struct rte_flow_item_ipv6),
2000 					error);
2001 	if (ret < 0)
2002 		return ret;
2003 	return 0;
2004 }
2005 
2006 /**
2007  * Validate UDP item.
2008  *
2009  * @param[in] item
2010  *   Item specification.
2011  * @param[in] item_flags
2012  *   Bit-fields that holds the items detected until now.
2013  * @param[in] target_protocol
2014  *   The next protocol in the previous item.
2015  * @param[in] flow_mask
2016  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2017  * @param[out] error
2018  *   Pointer to error structure.
2019  *
2020  * @return
2021  *   0 on success, a negative errno value otherwise and rte_errno is set.
2022  */
2023 int
2024 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2025 			    uint64_t item_flags,
2026 			    uint8_t target_protocol,
2027 			    struct rte_flow_error *error)
2028 {
2029 	const struct rte_flow_item_udp *mask = item->mask;
2030 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2031 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2032 				      MLX5_FLOW_LAYER_OUTER_L3;
2033 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2034 				      MLX5_FLOW_LAYER_OUTER_L4;
2035 	int ret;
2036 
2037 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2038 		return rte_flow_error_set(error, EINVAL,
2039 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2040 					  "protocol filtering not compatible"
2041 					  " with UDP layer");
2042 	if (!(item_flags & l3m))
2043 		return rte_flow_error_set(error, EINVAL,
2044 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2045 					  "L3 is mandatory to filter on L4");
2046 	if (item_flags & l4m)
2047 		return rte_flow_error_set(error, EINVAL,
2048 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2049 					  "multiple L4 layers not supported");
2050 	if (!mask)
2051 		mask = &rte_flow_item_udp_mask;
2052 	ret = mlx5_flow_item_acceptable
2053 		(item, (const uint8_t *)mask,
2054 		 (const uint8_t *)&rte_flow_item_udp_mask,
2055 		 sizeof(struct rte_flow_item_udp), error);
2056 	if (ret < 0)
2057 		return ret;
2058 	return 0;
2059 }
2060 
2061 /**
2062  * Validate TCP item.
2063  *
2064  * @param[in] item
2065  *   Item specification.
2066  * @param[in] item_flags
2067  *   Bit-fields that holds the items detected until now.
2068  * @param[in] target_protocol
2069  *   The next protocol in the previous item.
2070  * @param[out] error
2071  *   Pointer to error structure.
2072  *
2073  * @return
2074  *   0 on success, a negative errno value otherwise and rte_errno is set.
2075  */
2076 int
2077 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2078 			    uint64_t item_flags,
2079 			    uint8_t target_protocol,
2080 			    const struct rte_flow_item_tcp *flow_mask,
2081 			    struct rte_flow_error *error)
2082 {
2083 	const struct rte_flow_item_tcp *mask = item->mask;
2084 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2085 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2086 				      MLX5_FLOW_LAYER_OUTER_L3;
2087 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2088 				      MLX5_FLOW_LAYER_OUTER_L4;
2089 	int ret;
2090 
2091 	MLX5_ASSERT(flow_mask);
2092 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2093 		return rte_flow_error_set(error, EINVAL,
2094 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 					  "protocol filtering not compatible"
2096 					  " with TCP layer");
2097 	if (!(item_flags & l3m))
2098 		return rte_flow_error_set(error, EINVAL,
2099 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2100 					  "L3 is mandatory to filter on L4");
2101 	if (item_flags & l4m)
2102 		return rte_flow_error_set(error, EINVAL,
2103 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2104 					  "multiple L4 layers not supported");
2105 	if (!mask)
2106 		mask = &rte_flow_item_tcp_mask;
2107 	ret = mlx5_flow_item_acceptable
2108 		(item, (const uint8_t *)mask,
2109 		 (const uint8_t *)flow_mask,
2110 		 sizeof(struct rte_flow_item_tcp), error);
2111 	if (ret < 0)
2112 		return ret;
2113 	return 0;
2114 }
2115 
2116 /**
2117  * Validate VXLAN item.
2118  *
2119  * @param[in] item
2120  *   Item specification.
2121  * @param[in] item_flags
2122  *   Bit-fields that holds the items detected until now.
2123  * @param[in] target_protocol
2124  *   The next protocol in the previous item.
2125  * @param[out] error
2126  *   Pointer to error structure.
2127  *
2128  * @return
2129  *   0 on success, a negative errno value otherwise and rte_errno is set.
2130  */
2131 int
2132 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2133 			      uint64_t item_flags,
2134 			      struct rte_flow_error *error)
2135 {
2136 	const struct rte_flow_item_vxlan *spec = item->spec;
2137 	const struct rte_flow_item_vxlan *mask = item->mask;
2138 	int ret;
2139 	union vni {
2140 		uint32_t vlan_id;
2141 		uint8_t vni[4];
2142 	} id = { .vlan_id = 0, };
2143 
2144 
2145 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2146 		return rte_flow_error_set(error, ENOTSUP,
2147 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2148 					  "multiple tunnel layers not"
2149 					  " supported");
2150 	/*
2151 	 * Verify only UDPv4 is present as defined in
2152 	 * https://tools.ietf.org/html/rfc7348
2153 	 */
2154 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2155 		return rte_flow_error_set(error, EINVAL,
2156 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 					  "no outer UDP layer found");
2158 	if (!mask)
2159 		mask = &rte_flow_item_vxlan_mask;
2160 	ret = mlx5_flow_item_acceptable
2161 		(item, (const uint8_t *)mask,
2162 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2163 		 sizeof(struct rte_flow_item_vxlan),
2164 		 error);
2165 	if (ret < 0)
2166 		return ret;
2167 	if (spec) {
2168 		memcpy(&id.vni[1], spec->vni, 3);
2169 		memcpy(&id.vni[1], mask->vni, 3);
2170 	}
2171 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2172 		return rte_flow_error_set(error, ENOTSUP,
2173 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2174 					  "VXLAN tunnel must be fully defined");
2175 	return 0;
2176 }
2177 
2178 /**
2179  * Validate VXLAN_GPE item.
2180  *
2181  * @param[in] item
2182  *   Item specification.
2183  * @param[in] item_flags
2184  *   Bit-fields that holds the items detected until now.
2185  * @param[in] priv
2186  *   Pointer to the private data structure.
2187  * @param[in] target_protocol
2188  *   The next protocol in the previous item.
2189  * @param[out] error
2190  *   Pointer to error structure.
2191  *
2192  * @return
2193  *   0 on success, a negative errno value otherwise and rte_errno is set.
2194  */
2195 int
2196 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2197 				  uint64_t item_flags,
2198 				  struct rte_eth_dev *dev,
2199 				  struct rte_flow_error *error)
2200 {
2201 	struct mlx5_priv *priv = dev->data->dev_private;
2202 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2203 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2204 	int ret;
2205 	union vni {
2206 		uint32_t vlan_id;
2207 		uint8_t vni[4];
2208 	} id = { .vlan_id = 0, };
2209 
2210 	if (!priv->config.l3_vxlan_en)
2211 		return rte_flow_error_set(error, ENOTSUP,
2212 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2213 					  "L3 VXLAN is not enabled by device"
2214 					  " parameter and/or not configured in"
2215 					  " firmware");
2216 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2217 		return rte_flow_error_set(error, ENOTSUP,
2218 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 					  "multiple tunnel layers not"
2220 					  " supported");
2221 	/*
2222 	 * Verify only UDPv4 is present as defined in
2223 	 * https://tools.ietf.org/html/rfc7348
2224 	 */
2225 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2226 		return rte_flow_error_set(error, EINVAL,
2227 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2228 					  "no outer UDP layer found");
2229 	if (!mask)
2230 		mask = &rte_flow_item_vxlan_gpe_mask;
2231 	ret = mlx5_flow_item_acceptable
2232 		(item, (const uint8_t *)mask,
2233 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2234 		 sizeof(struct rte_flow_item_vxlan_gpe),
2235 		 error);
2236 	if (ret < 0)
2237 		return ret;
2238 	if (spec) {
2239 		if (spec->protocol)
2240 			return rte_flow_error_set(error, ENOTSUP,
2241 						  RTE_FLOW_ERROR_TYPE_ITEM,
2242 						  item,
2243 						  "VxLAN-GPE protocol"
2244 						  " not supported");
2245 		memcpy(&id.vni[1], spec->vni, 3);
2246 		memcpy(&id.vni[1], mask->vni, 3);
2247 	}
2248 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2249 		return rte_flow_error_set(error, ENOTSUP,
2250 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2251 					  "VXLAN-GPE tunnel must be fully"
2252 					  " defined");
2253 	return 0;
2254 }
2255 /**
2256  * Validate GRE Key item.
2257  *
2258  * @param[in] item
2259  *   Item specification.
2260  * @param[in] item_flags
2261  *   Bit flags to mark detected items.
2262  * @param[in] gre_item
2263  *   Pointer to gre_item
2264  * @param[out] error
2265  *   Pointer to error structure.
2266  *
2267  * @return
2268  *   0 on success, a negative errno value otherwise and rte_errno is set.
2269  */
2270 int
2271 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2272 				uint64_t item_flags,
2273 				const struct rte_flow_item *gre_item,
2274 				struct rte_flow_error *error)
2275 {
2276 	const rte_be32_t *mask = item->mask;
2277 	int ret = 0;
2278 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2279 	const struct rte_flow_item_gre *gre_spec;
2280 	const struct rte_flow_item_gre *gre_mask;
2281 
2282 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2283 		return rte_flow_error_set(error, ENOTSUP,
2284 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 					  "Multiple GRE key not support");
2286 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2287 		return rte_flow_error_set(error, ENOTSUP,
2288 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2289 					  "No preceding GRE header");
2290 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2291 		return rte_flow_error_set(error, ENOTSUP,
2292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 					  "GRE key following a wrong item");
2294 	gre_mask = gre_item->mask;
2295 	if (!gre_mask)
2296 		gre_mask = &rte_flow_item_gre_mask;
2297 	gre_spec = gre_item->spec;
2298 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2299 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2300 		return rte_flow_error_set(error, EINVAL,
2301 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2302 					  "Key bit must be on");
2303 
2304 	if (!mask)
2305 		mask = &gre_key_default_mask;
2306 	ret = mlx5_flow_item_acceptable
2307 		(item, (const uint8_t *)mask,
2308 		 (const uint8_t *)&gre_key_default_mask,
2309 		 sizeof(rte_be32_t), error);
2310 	return ret;
2311 }
2312 
2313 /**
2314  * Validate GRE item.
2315  *
2316  * @param[in] item
2317  *   Item specification.
2318  * @param[in] item_flags
2319  *   Bit flags to mark detected items.
2320  * @param[in] target_protocol
2321  *   The next protocol in the previous item.
2322  * @param[out] error
2323  *   Pointer to error structure.
2324  *
2325  * @return
2326  *   0 on success, a negative errno value otherwise and rte_errno is set.
2327  */
2328 int
2329 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2330 			    uint64_t item_flags,
2331 			    uint8_t target_protocol,
2332 			    struct rte_flow_error *error)
2333 {
2334 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2335 	const struct rte_flow_item_gre *mask = item->mask;
2336 	int ret;
2337 	const struct rte_flow_item_gre nic_mask = {
2338 		.c_rsvd0_ver = RTE_BE16(0xB000),
2339 		.protocol = RTE_BE16(UINT16_MAX),
2340 	};
2341 
2342 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2343 		return rte_flow_error_set(error, EINVAL,
2344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2345 					  "protocol filtering not compatible"
2346 					  " with this GRE layer");
2347 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2348 		return rte_flow_error_set(error, ENOTSUP,
2349 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2350 					  "multiple tunnel layers not"
2351 					  " supported");
2352 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2353 		return rte_flow_error_set(error, ENOTSUP,
2354 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 					  "L3 Layer is missing");
2356 	if (!mask)
2357 		mask = &rte_flow_item_gre_mask;
2358 	ret = mlx5_flow_item_acceptable
2359 		(item, (const uint8_t *)mask,
2360 		 (const uint8_t *)&nic_mask,
2361 		 sizeof(struct rte_flow_item_gre), error);
2362 	if (ret < 0)
2363 		return ret;
2364 #ifndef HAVE_MLX5DV_DR
2365 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2366 	if (spec && (spec->protocol & mask->protocol))
2367 		return rte_flow_error_set(error, ENOTSUP,
2368 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2369 					  "without MPLS support the"
2370 					  " specification cannot be used for"
2371 					  " filtering");
2372 #endif
2373 #endif
2374 	return 0;
2375 }
2376 
2377 /**
2378  * Validate Geneve item.
2379  *
2380  * @param[in] item
2381  *   Item specification.
2382  * @param[in] itemFlags
2383  *   Bit-fields that holds the items detected until now.
2384  * @param[in] enPriv
2385  *   Pointer to the private data structure.
2386  * @param[out] error
2387  *   Pointer to error structure.
2388  *
2389  * @return
2390  *   0 on success, a negative errno value otherwise and rte_errno is set.
2391  */
2392 
2393 int
2394 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2395 			       uint64_t item_flags,
2396 			       struct rte_eth_dev *dev,
2397 			       struct rte_flow_error *error)
2398 {
2399 	struct mlx5_priv *priv = dev->data->dev_private;
2400 	const struct rte_flow_item_geneve *spec = item->spec;
2401 	const struct rte_flow_item_geneve *mask = item->mask;
2402 	int ret;
2403 	uint16_t gbhdr;
2404 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2405 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2406 	const struct rte_flow_item_geneve nic_mask = {
2407 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2408 		.vni = "\xff\xff\xff",
2409 		.protocol = RTE_BE16(UINT16_MAX),
2410 	};
2411 
2412 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2413 		return rte_flow_error_set(error, ENOTSUP,
2414 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2415 					  "L3 Geneve is not enabled by device"
2416 					  " parameter and/or not configured in"
2417 					  " firmware");
2418 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2419 		return rte_flow_error_set(error, ENOTSUP,
2420 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2421 					  "multiple tunnel layers not"
2422 					  " supported");
2423 	/*
2424 	 * Verify only UDPv4 is present as defined in
2425 	 * https://tools.ietf.org/html/rfc7348
2426 	 */
2427 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2428 		return rte_flow_error_set(error, EINVAL,
2429 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2430 					  "no outer UDP layer found");
2431 	if (!mask)
2432 		mask = &rte_flow_item_geneve_mask;
2433 	ret = mlx5_flow_item_acceptable
2434 				  (item, (const uint8_t *)mask,
2435 				   (const uint8_t *)&nic_mask,
2436 				   sizeof(struct rte_flow_item_geneve), error);
2437 	if (ret)
2438 		return ret;
2439 	if (spec) {
2440 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2441 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2442 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2443 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2444 			return rte_flow_error_set(error, ENOTSUP,
2445 						  RTE_FLOW_ERROR_TYPE_ITEM,
2446 						  item,
2447 						  "Geneve protocol unsupported"
2448 						  " fields are being used");
2449 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2450 			return rte_flow_error_set
2451 					(error, ENOTSUP,
2452 					 RTE_FLOW_ERROR_TYPE_ITEM,
2453 					 item,
2454 					 "Unsupported Geneve options length");
2455 	}
2456 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2457 		return rte_flow_error_set
2458 				    (error, ENOTSUP,
2459 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2460 				     "Geneve tunnel must be fully defined");
2461 	return 0;
2462 }
2463 
2464 /**
2465  * Validate MPLS item.
2466  *
2467  * @param[in] dev
2468  *   Pointer to the rte_eth_dev structure.
2469  * @param[in] item
2470  *   Item specification.
2471  * @param[in] item_flags
2472  *   Bit-fields that holds the items detected until now.
2473  * @param[in] prev_layer
2474  *   The protocol layer indicated in previous item.
2475  * @param[out] error
2476  *   Pointer to error structure.
2477  *
2478  * @return
2479  *   0 on success, a negative errno value otherwise and rte_errno is set.
2480  */
2481 int
2482 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2483 			     const struct rte_flow_item *item __rte_unused,
2484 			     uint64_t item_flags __rte_unused,
2485 			     uint64_t prev_layer __rte_unused,
2486 			     struct rte_flow_error *error)
2487 {
2488 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2489 	const struct rte_flow_item_mpls *mask = item->mask;
2490 	struct mlx5_priv *priv = dev->data->dev_private;
2491 	int ret;
2492 
2493 	if (!priv->config.mpls_en)
2494 		return rte_flow_error_set(error, ENOTSUP,
2495 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2496 					  "MPLS not supported or"
2497 					  " disabled in firmware"
2498 					  " configuration.");
2499 	/* MPLS over IP, UDP, GRE is allowed */
2500 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2501 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2502 			    MLX5_FLOW_LAYER_GRE)))
2503 		return rte_flow_error_set(error, EINVAL,
2504 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2505 					  "protocol filtering not compatible"
2506 					  " with MPLS layer");
2507 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2508 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2509 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2510 		return rte_flow_error_set(error, ENOTSUP,
2511 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2512 					  "multiple tunnel layers not"
2513 					  " supported");
2514 	if (!mask)
2515 		mask = &rte_flow_item_mpls_mask;
2516 	ret = mlx5_flow_item_acceptable
2517 		(item, (const uint8_t *)mask,
2518 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2519 		 sizeof(struct rte_flow_item_mpls), error);
2520 	if (ret < 0)
2521 		return ret;
2522 	return 0;
2523 #else
2524 	return rte_flow_error_set(error, ENOTSUP,
2525 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2526 				  "MPLS is not supported by Verbs, please"
2527 				  " update.");
2528 #endif
2529 }
2530 
2531 /**
2532  * Validate NVGRE item.
2533  *
2534  * @param[in] item
2535  *   Item specification.
2536  * @param[in] item_flags
2537  *   Bit flags to mark detected items.
2538  * @param[in] target_protocol
2539  *   The next protocol in the previous item.
2540  * @param[out] error
2541  *   Pointer to error structure.
2542  *
2543  * @return
2544  *   0 on success, a negative errno value otherwise and rte_errno is set.
2545  */
2546 int
2547 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2548 			      uint64_t item_flags,
2549 			      uint8_t target_protocol,
2550 			      struct rte_flow_error *error)
2551 {
2552 	const struct rte_flow_item_nvgre *mask = item->mask;
2553 	int ret;
2554 
2555 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2556 		return rte_flow_error_set(error, EINVAL,
2557 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2558 					  "protocol filtering not compatible"
2559 					  " with this GRE layer");
2560 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2561 		return rte_flow_error_set(error, ENOTSUP,
2562 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2563 					  "multiple tunnel layers not"
2564 					  " supported");
2565 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2566 		return rte_flow_error_set(error, ENOTSUP,
2567 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2568 					  "L3 Layer is missing");
2569 	if (!mask)
2570 		mask = &rte_flow_item_nvgre_mask;
2571 	ret = mlx5_flow_item_acceptable
2572 		(item, (const uint8_t *)mask,
2573 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2574 		 sizeof(struct rte_flow_item_nvgre), error);
2575 	if (ret < 0)
2576 		return ret;
2577 	return 0;
2578 }
2579 
2580 /**
2581  * Validate eCPRI item.
2582  *
2583  * @param[in] item
2584  *   Item specification.
2585  * @param[in] item_flags
2586  *   Bit-fields that holds the items detected until now.
2587  * @param[in] last_item
2588  *   Previous validated item in the pattern items.
2589  * @param[in] ether_type
2590  *   Type in the ethernet layer header (including dot1q).
2591  * @param[in] acc_mask
2592  *   Acceptable mask, if NULL default internal default mask
2593  *   will be used to check whether item fields are supported.
2594  * @param[out] error
2595  *   Pointer to error structure.
2596  *
2597  * @return
2598  *   0 on success, a negative errno value otherwise and rte_errno is set.
2599  */
2600 int
2601 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2602 			      uint64_t item_flags,
2603 			      uint64_t last_item,
2604 			      uint16_t ether_type,
2605 			      const struct rte_flow_item_ecpri *acc_mask,
2606 			      struct rte_flow_error *error)
2607 {
2608 	const struct rte_flow_item_ecpri *mask = item->mask;
2609 	const struct rte_flow_item_ecpri nic_mask = {
2610 		.hdr = {
2611 			.common = {
2612 				.u32 =
2613 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2614 					.type = 0xFF,
2615 					}).u32),
2616 			},
2617 			.dummy[0] = 0xFFFFFFFF,
2618 		},
2619 	};
2620 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2621 					MLX5_FLOW_LAYER_OUTER_VLAN);
2622 	struct rte_flow_item_ecpri mask_lo;
2623 
2624 	if ((last_item & outer_l2_vlan) && ether_type &&
2625 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2626 		return rte_flow_error_set(error, EINVAL,
2627 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2628 					  "eCPRI cannot follow L2/VLAN layer "
2629 					  "which ether type is not 0xAEFE.");
2630 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2631 		return rte_flow_error_set(error, EINVAL,
2632 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2633 					  "eCPRI with tunnel is not supported "
2634 					  "right now.");
2635 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2636 		return rte_flow_error_set(error, ENOTSUP,
2637 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2638 					  "multiple L3 layers not supported");
2639 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2640 		return rte_flow_error_set(error, EINVAL,
2641 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2642 					  "eCPRI cannot follow a TCP layer.");
2643 	/* In specification, eCPRI could be over UDP layer. */
2644 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2645 		return rte_flow_error_set(error, EINVAL,
2646 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2647 					  "eCPRI over UDP layer is not yet "
2648 					  "supported right now.");
2649 	/* Mask for type field in common header could be zero. */
2650 	if (!mask)
2651 		mask = &rte_flow_item_ecpri_mask;
2652 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2653 	/* Input mask is in big-endian format. */
2654 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2655 		return rte_flow_error_set(error, EINVAL,
2656 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2657 					  "partial mask is not supported "
2658 					  "for protocol");
2659 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2660 		return rte_flow_error_set(error, EINVAL,
2661 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2662 					  "message header mask must be after "
2663 					  "a type mask");
2664 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2665 					 acc_mask ? (const uint8_t *)acc_mask
2666 						  : (const uint8_t *)&nic_mask,
2667 					 sizeof(struct rte_flow_item_ecpri),
2668 					 error);
2669 }
2670 
2671 /* Allocate unique ID for the split Q/RSS subflows. */
2672 static uint32_t
2673 flow_qrss_get_id(struct rte_eth_dev *dev)
2674 {
2675 	struct mlx5_priv *priv = dev->data->dev_private;
2676 	uint32_t qrss_id, ret;
2677 
2678 	ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2679 	if (ret)
2680 		return 0;
2681 	MLX5_ASSERT(qrss_id);
2682 	return qrss_id;
2683 }
2684 
2685 /* Free unique ID for the split Q/RSS subflows. */
2686 static void
2687 flow_qrss_free_id(struct rte_eth_dev *dev,  uint32_t qrss_id)
2688 {
2689 	struct mlx5_priv *priv = dev->data->dev_private;
2690 
2691 	if (qrss_id)
2692 		mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2693 }
2694 
2695 /**
2696  * Release resource related QUEUE/RSS action split.
2697  *
2698  * @param dev
2699  *   Pointer to Ethernet device.
2700  * @param flow
2701  *   Flow to release id's from.
2702  */
2703 static void
2704 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2705 			     struct rte_flow *flow)
2706 {
2707 	struct mlx5_priv *priv = dev->data->dev_private;
2708 	uint32_t handle_idx;
2709 	struct mlx5_flow_handle *dev_handle;
2710 
2711 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2712 		       handle_idx, dev_handle, next)
2713 		if (dev_handle->split_flow_id)
2714 			flow_qrss_free_id(dev, dev_handle->split_flow_id);
2715 }
2716 
2717 static int
2718 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2719 		   const struct rte_flow_attr *attr __rte_unused,
2720 		   const struct rte_flow_item items[] __rte_unused,
2721 		   const struct rte_flow_action actions[] __rte_unused,
2722 		   bool external __rte_unused,
2723 		   int hairpin __rte_unused,
2724 		   struct rte_flow_error *error)
2725 {
2726 	return rte_flow_error_set(error, ENOTSUP,
2727 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2728 }
2729 
2730 static struct mlx5_flow *
2731 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2732 		  const struct rte_flow_attr *attr __rte_unused,
2733 		  const struct rte_flow_item items[] __rte_unused,
2734 		  const struct rte_flow_action actions[] __rte_unused,
2735 		  struct rte_flow_error *error)
2736 {
2737 	rte_flow_error_set(error, ENOTSUP,
2738 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2739 	return NULL;
2740 }
2741 
2742 static int
2743 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2744 		    struct mlx5_flow *dev_flow __rte_unused,
2745 		    const struct rte_flow_attr *attr __rte_unused,
2746 		    const struct rte_flow_item items[] __rte_unused,
2747 		    const struct rte_flow_action actions[] __rte_unused,
2748 		    struct rte_flow_error *error)
2749 {
2750 	return rte_flow_error_set(error, ENOTSUP,
2751 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2752 }
2753 
2754 static int
2755 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2756 		struct rte_flow *flow __rte_unused,
2757 		struct rte_flow_error *error)
2758 {
2759 	return rte_flow_error_set(error, ENOTSUP,
2760 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2761 }
2762 
2763 static void
2764 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2765 		 struct rte_flow *flow __rte_unused)
2766 {
2767 }
2768 
2769 static void
2770 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2771 		  struct rte_flow *flow __rte_unused)
2772 {
2773 }
2774 
2775 static int
2776 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2777 		struct rte_flow *flow __rte_unused,
2778 		const struct rte_flow_action *actions __rte_unused,
2779 		void *data __rte_unused,
2780 		struct rte_flow_error *error)
2781 {
2782 	return rte_flow_error_set(error, ENOTSUP,
2783 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2784 }
2785 
2786 /* Void driver to protect from null pointer reference. */
2787 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2788 	.validate = flow_null_validate,
2789 	.prepare = flow_null_prepare,
2790 	.translate = flow_null_translate,
2791 	.apply = flow_null_apply,
2792 	.remove = flow_null_remove,
2793 	.destroy = flow_null_destroy,
2794 	.query = flow_null_query,
2795 };
2796 
2797 /**
2798  * Select flow driver type according to flow attributes and device
2799  * configuration.
2800  *
2801  * @param[in] dev
2802  *   Pointer to the dev structure.
2803  * @param[in] attr
2804  *   Pointer to the flow attributes.
2805  *
2806  * @return
2807  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2808  */
2809 static enum mlx5_flow_drv_type
2810 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2811 {
2812 	struct mlx5_priv *priv = dev->data->dev_private;
2813 	/* The OS can determine first a specific flow type (DV, VERBS) */
2814 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
2815 
2816 	if (type != MLX5_FLOW_TYPE_MAX)
2817 		return type;
2818 	/* If no OS specific type - continue with DV/VERBS selection */
2819 	if (attr->transfer && priv->config.dv_esw_en)
2820 		type = MLX5_FLOW_TYPE_DV;
2821 	if (!attr->transfer)
2822 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2823 						 MLX5_FLOW_TYPE_VERBS;
2824 	return type;
2825 }
2826 
2827 #define flow_get_drv_ops(type) flow_drv_ops[type]
2828 
2829 /**
2830  * Flow driver validation API. This abstracts calling driver specific functions.
2831  * The type of flow driver is determined according to flow attributes.
2832  *
2833  * @param[in] dev
2834  *   Pointer to the dev structure.
2835  * @param[in] attr
2836  *   Pointer to the flow attributes.
2837  * @param[in] items
2838  *   Pointer to the list of items.
2839  * @param[in] actions
2840  *   Pointer to the list of actions.
2841  * @param[in] external
2842  *   This flow rule is created by request external to PMD.
2843  * @param[in] hairpin
2844  *   Number of hairpin TX actions, 0 means classic flow.
2845  * @param[out] error
2846  *   Pointer to the error structure.
2847  *
2848  * @return
2849  *   0 on success, a negative errno value otherwise and rte_errno is set.
2850  */
2851 static inline int
2852 flow_drv_validate(struct rte_eth_dev *dev,
2853 		  const struct rte_flow_attr *attr,
2854 		  const struct rte_flow_item items[],
2855 		  const struct rte_flow_action actions[],
2856 		  bool external, int hairpin, struct rte_flow_error *error)
2857 {
2858 	const struct mlx5_flow_driver_ops *fops;
2859 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2860 
2861 	fops = flow_get_drv_ops(type);
2862 	return fops->validate(dev, attr, items, actions, external,
2863 			      hairpin, error);
2864 }
2865 
2866 /**
2867  * Flow driver preparation API. This abstracts calling driver specific
2868  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2869  * calculates the size of memory required for device flow, allocates the memory,
2870  * initializes the device flow and returns the pointer.
2871  *
2872  * @note
2873  *   This function initializes device flow structure such as dv or verbs in
2874  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
2875  *   rest. For example, adding returning device flow to flow->dev_flow list and
2876  *   setting backward reference to the flow should be done out of this function.
2877  *   layers field is not filled either.
2878  *
2879  * @param[in] dev
2880  *   Pointer to the dev structure.
2881  * @param[in] attr
2882  *   Pointer to the flow attributes.
2883  * @param[in] items
2884  *   Pointer to the list of items.
2885  * @param[in] actions
2886  *   Pointer to the list of actions.
2887  * @param[in] flow_idx
2888  *   This memory pool index to the flow.
2889  * @param[out] error
2890  *   Pointer to the error structure.
2891  *
2892  * @return
2893  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
2894  */
2895 static inline struct mlx5_flow *
2896 flow_drv_prepare(struct rte_eth_dev *dev,
2897 		 const struct rte_flow *flow,
2898 		 const struct rte_flow_attr *attr,
2899 		 const struct rte_flow_item items[],
2900 		 const struct rte_flow_action actions[],
2901 		 uint32_t flow_idx,
2902 		 struct rte_flow_error *error)
2903 {
2904 	const struct mlx5_flow_driver_ops *fops;
2905 	enum mlx5_flow_drv_type type = flow->drv_type;
2906 	struct mlx5_flow *mlx5_flow = NULL;
2907 
2908 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2909 	fops = flow_get_drv_ops(type);
2910 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
2911 	if (mlx5_flow)
2912 		mlx5_flow->flow_idx = flow_idx;
2913 	return mlx5_flow;
2914 }
2915 
2916 /**
2917  * Flow driver translation API. This abstracts calling driver specific
2918  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2919  * translates a generic flow into a driver flow. flow_drv_prepare() must
2920  * precede.
2921  *
2922  * @note
2923  *   dev_flow->layers could be filled as a result of parsing during translation
2924  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2925  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2926  *   flow->actions could be overwritten even though all the expanded dev_flows
2927  *   have the same actions.
2928  *
2929  * @param[in] dev
2930  *   Pointer to the rte dev structure.
2931  * @param[in, out] dev_flow
2932  *   Pointer to the mlx5 flow.
2933  * @param[in] attr
2934  *   Pointer to the flow attributes.
2935  * @param[in] items
2936  *   Pointer to the list of items.
2937  * @param[in] actions
2938  *   Pointer to the list of actions.
2939  * @param[out] error
2940  *   Pointer to the error structure.
2941  *
2942  * @return
2943  *   0 on success, a negative errno value otherwise and rte_errno is set.
2944  */
2945 static inline int
2946 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2947 		   const struct rte_flow_attr *attr,
2948 		   const struct rte_flow_item items[],
2949 		   const struct rte_flow_action actions[],
2950 		   struct rte_flow_error *error)
2951 {
2952 	const struct mlx5_flow_driver_ops *fops;
2953 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2954 
2955 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2956 	fops = flow_get_drv_ops(type);
2957 	return fops->translate(dev, dev_flow, attr, items, actions, error);
2958 }
2959 
2960 /**
2961  * Flow driver apply API. This abstracts calling driver specific functions.
2962  * Parent flow (rte_flow) should have driver type (drv_type). It applies
2963  * translated driver flows on to device. flow_drv_translate() must precede.
2964  *
2965  * @param[in] dev
2966  *   Pointer to Ethernet device structure.
2967  * @param[in, out] flow
2968  *   Pointer to flow structure.
2969  * @param[out] error
2970  *   Pointer to error structure.
2971  *
2972  * @return
2973  *   0 on success, a negative errno value otherwise and rte_errno is set.
2974  */
2975 static inline int
2976 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2977 	       struct rte_flow_error *error)
2978 {
2979 	const struct mlx5_flow_driver_ops *fops;
2980 	enum mlx5_flow_drv_type type = flow->drv_type;
2981 
2982 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2983 	fops = flow_get_drv_ops(type);
2984 	return fops->apply(dev, flow, error);
2985 }
2986 
2987 /**
2988  * Flow driver remove API. This abstracts calling driver specific functions.
2989  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2990  * on device. All the resources of the flow should be freed by calling
2991  * flow_drv_destroy().
2992  *
2993  * @param[in] dev
2994  *   Pointer to Ethernet device.
2995  * @param[in, out] flow
2996  *   Pointer to flow structure.
2997  */
2998 static inline void
2999 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
3000 {
3001 	const struct mlx5_flow_driver_ops *fops;
3002 	enum mlx5_flow_drv_type type = flow->drv_type;
3003 
3004 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3005 	fops = flow_get_drv_ops(type);
3006 	fops->remove(dev, flow);
3007 }
3008 
3009 /**
3010  * Flow driver destroy API. This abstracts calling driver specific functions.
3011  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3012  * on device and releases resources of the flow.
3013  *
3014  * @param[in] dev
3015  *   Pointer to Ethernet device.
3016  * @param[in, out] flow
3017  *   Pointer to flow structure.
3018  */
3019 static inline void
3020 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3021 {
3022 	const struct mlx5_flow_driver_ops *fops;
3023 	enum mlx5_flow_drv_type type = flow->drv_type;
3024 
3025 	flow_mreg_split_qrss_release(dev, flow);
3026 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3027 	fops = flow_get_drv_ops(type);
3028 	fops->destroy(dev, flow);
3029 }
3030 
3031 /**
3032  * Get RSS action from the action list.
3033  *
3034  * @param[in] actions
3035  *   Pointer to the list of actions.
3036  *
3037  * @return
3038  *   Pointer to the RSS action if exist, else return NULL.
3039  */
3040 static const struct rte_flow_action_rss*
3041 flow_get_rss_action(const struct rte_flow_action actions[])
3042 {
3043 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3044 		switch (actions->type) {
3045 		case RTE_FLOW_ACTION_TYPE_RSS:
3046 			return (const struct rte_flow_action_rss *)
3047 			       actions->conf;
3048 		default:
3049 			break;
3050 		}
3051 	}
3052 	return NULL;
3053 }
3054 
3055 static unsigned int
3056 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3057 {
3058 	const struct rte_flow_item *item;
3059 	unsigned int has_vlan = 0;
3060 
3061 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3062 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3063 			has_vlan = 1;
3064 			break;
3065 		}
3066 	}
3067 	if (has_vlan)
3068 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3069 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3070 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3071 			       MLX5_EXPANSION_ROOT_OUTER;
3072 }
3073 
3074 /**
3075  *  Get layer flags from the prefix flow.
3076  *
3077  *  Some flows may be split to several subflows, the prefix subflow gets the
3078  *  match items and the suffix sub flow gets the actions.
3079  *  Some actions need the user defined match item flags to get the detail for
3080  *  the action.
3081  *  This function helps the suffix flow to get the item layer flags from prefix
3082  *  subflow.
3083  *
3084  * @param[in] dev_flow
3085  *   Pointer the created preifx subflow.
3086  *
3087  * @return
3088  *   The layers get from prefix subflow.
3089  */
3090 static inline uint64_t
3091 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3092 {
3093 	uint64_t layers = 0;
3094 
3095 	/*
3096 	 * Layers bits could be localization, but usually the compiler will
3097 	 * help to do the optimization work for source code.
3098 	 * If no decap actions, use the layers directly.
3099 	 */
3100 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3101 		return dev_flow->handle->layers;
3102 	/* Convert L3 layers with decap action. */
3103 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3104 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3105 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3106 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3107 	/* Convert L4 layers with decap action.  */
3108 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3109 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3110 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3111 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3112 	return layers;
3113 }
3114 
3115 /**
3116  * Get metadata split action information.
3117  *
3118  * @param[in] actions
3119  *   Pointer to the list of actions.
3120  * @param[out] qrss
3121  *   Pointer to the return pointer.
3122  * @param[out] qrss_type
3123  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3124  *   if no QUEUE/RSS is found.
3125  * @param[out] encap_idx
3126  *   Pointer to the index of the encap action if exists, otherwise the last
3127  *   action index.
3128  *
3129  * @return
3130  *   Total number of actions.
3131  */
3132 static int
3133 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3134 				       const struct rte_flow_action **qrss,
3135 				       int *encap_idx)
3136 {
3137 	const struct rte_flow_action_raw_encap *raw_encap;
3138 	int actions_n = 0;
3139 	int raw_decap_idx = -1;
3140 
3141 	*encap_idx = -1;
3142 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3143 		switch (actions->type) {
3144 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3145 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3146 			*encap_idx = actions_n;
3147 			break;
3148 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3149 			raw_decap_idx = actions_n;
3150 			break;
3151 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3152 			raw_encap = actions->conf;
3153 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3154 				*encap_idx = raw_decap_idx != -1 ?
3155 						      raw_decap_idx : actions_n;
3156 			break;
3157 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3158 		case RTE_FLOW_ACTION_TYPE_RSS:
3159 			*qrss = actions;
3160 			break;
3161 		default:
3162 			break;
3163 		}
3164 		actions_n++;
3165 	}
3166 	if (*encap_idx == -1)
3167 		*encap_idx = actions_n;
3168 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3169 	return actions_n + 1;
3170 }
3171 
3172 /**
3173  * Check meter action from the action list.
3174  *
3175  * @param[in] actions
3176  *   Pointer to the list of actions.
3177  * @param[out] mtr
3178  *   Pointer to the meter exist flag.
3179  *
3180  * @return
3181  *   Total number of actions.
3182  */
3183 static int
3184 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3185 {
3186 	int actions_n = 0;
3187 
3188 	MLX5_ASSERT(mtr);
3189 	*mtr = 0;
3190 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3191 		switch (actions->type) {
3192 		case RTE_FLOW_ACTION_TYPE_METER:
3193 			*mtr = 1;
3194 			break;
3195 		default:
3196 			break;
3197 		}
3198 		actions_n++;
3199 	}
3200 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3201 	return actions_n + 1;
3202 }
3203 
3204 /**
3205  * Check if the flow should be split due to hairpin.
3206  * The reason for the split is that in current HW we can't
3207  * support encap and push-vlan on Rx, so if a flow contains
3208  * these actions we move it to Tx.
3209  *
3210  * @param dev
3211  *   Pointer to Ethernet device.
3212  * @param[in] attr
3213  *   Flow rule attributes.
3214  * @param[in] actions
3215  *   Associated actions (list terminated by the END action).
3216  *
3217  * @return
3218  *   > 0 the number of actions and the flow should be split,
3219  *   0 when no split required.
3220  */
3221 static int
3222 flow_check_hairpin_split(struct rte_eth_dev *dev,
3223 			 const struct rte_flow_attr *attr,
3224 			 const struct rte_flow_action actions[])
3225 {
3226 	int queue_action = 0;
3227 	int action_n = 0;
3228 	int split = 0;
3229 	const struct rte_flow_action_queue *queue;
3230 	const struct rte_flow_action_rss *rss;
3231 	const struct rte_flow_action_raw_encap *raw_encap;
3232 
3233 	if (!attr->ingress)
3234 		return 0;
3235 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3236 		switch (actions->type) {
3237 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3238 			queue = actions->conf;
3239 			if (queue == NULL)
3240 				return 0;
3241 			if (mlx5_rxq_get_type(dev, queue->index) !=
3242 			    MLX5_RXQ_TYPE_HAIRPIN)
3243 				return 0;
3244 			queue_action = 1;
3245 			action_n++;
3246 			break;
3247 		case RTE_FLOW_ACTION_TYPE_RSS:
3248 			rss = actions->conf;
3249 			if (rss == NULL || rss->queue_num == 0)
3250 				return 0;
3251 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
3252 			    MLX5_RXQ_TYPE_HAIRPIN)
3253 				return 0;
3254 			queue_action = 1;
3255 			action_n++;
3256 			break;
3257 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3258 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3259 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3260 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3261 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3262 			split++;
3263 			action_n++;
3264 			break;
3265 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3266 			raw_encap = actions->conf;
3267 			if (raw_encap->size >
3268 			    (sizeof(struct rte_flow_item_eth) +
3269 			     sizeof(struct rte_flow_item_ipv4)))
3270 				split++;
3271 			action_n++;
3272 			break;
3273 		default:
3274 			action_n++;
3275 			break;
3276 		}
3277 	}
3278 	if (split && queue_action)
3279 		return action_n;
3280 	return 0;
3281 }
3282 
3283 /* Declare flow create/destroy prototype in advance. */
3284 static uint32_t
3285 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3286 		 const struct rte_flow_attr *attr,
3287 		 const struct rte_flow_item items[],
3288 		 const struct rte_flow_action actions[],
3289 		 bool external, struct rte_flow_error *error);
3290 
3291 static void
3292 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3293 		  uint32_t flow_idx);
3294 
3295 /**
3296  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3297  *
3298  * As mark_id is unique, if there's already a registered flow for the mark_id,
3299  * return by increasing the reference counter of the resource. Otherwise, create
3300  * the resource (mcp_res) and flow.
3301  *
3302  * Flow looks like,
3303  *   - If ingress port is ANY and reg_c[1] is mark_id,
3304  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3305  *
3306  * For default flow (zero mark_id), flow is like,
3307  *   - If ingress port is ANY,
3308  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3309  *
3310  * @param dev
3311  *   Pointer to Ethernet device.
3312  * @param mark_id
3313  *   ID of MARK action, zero means default flow for META.
3314  * @param[out] error
3315  *   Perform verbose error reporting if not NULL.
3316  *
3317  * @return
3318  *   Associated resource on success, NULL otherwise and rte_errno is set.
3319  */
3320 static struct mlx5_flow_mreg_copy_resource *
3321 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3322 			  struct rte_flow_error *error)
3323 {
3324 	struct mlx5_priv *priv = dev->data->dev_private;
3325 	struct rte_flow_attr attr = {
3326 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3327 		.ingress = 1,
3328 	};
3329 	struct mlx5_rte_flow_item_tag tag_spec = {
3330 		.data = mark_id,
3331 	};
3332 	struct rte_flow_item items[] = {
3333 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3334 	};
3335 	struct rte_flow_action_mark ftag = {
3336 		.id = mark_id,
3337 	};
3338 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3339 		.dst = REG_B,
3340 		.src = REG_NON,
3341 	};
3342 	struct rte_flow_action_jump jump = {
3343 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3344 	};
3345 	struct rte_flow_action actions[] = {
3346 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3347 	};
3348 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3349 	uint32_t idx = 0;
3350 	int ret;
3351 
3352 	/* Fill the register fileds in the flow. */
3353 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3354 	if (ret < 0)
3355 		return NULL;
3356 	tag_spec.id = ret;
3357 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3358 	if (ret < 0)
3359 		return NULL;
3360 	cp_mreg.src = ret;
3361 	/* Check if already registered. */
3362 	MLX5_ASSERT(priv->mreg_cp_tbl);
3363 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
3364 	if (mcp_res) {
3365 		/* For non-default rule. */
3366 		if (mark_id != MLX5_DEFAULT_COPY_ID)
3367 			mcp_res->refcnt++;
3368 		MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
3369 			    mcp_res->refcnt == 1);
3370 		return mcp_res;
3371 	}
3372 	/* Provide the full width of FLAG specific value. */
3373 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3374 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3375 	/* Build a new flow. */
3376 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3377 		items[0] = (struct rte_flow_item){
3378 			.type = (enum rte_flow_item_type)
3379 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3380 			.spec = &tag_spec,
3381 		};
3382 		items[1] = (struct rte_flow_item){
3383 			.type = RTE_FLOW_ITEM_TYPE_END,
3384 		};
3385 		actions[0] = (struct rte_flow_action){
3386 			.type = (enum rte_flow_action_type)
3387 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3388 			.conf = &ftag,
3389 		};
3390 		actions[1] = (struct rte_flow_action){
3391 			.type = (enum rte_flow_action_type)
3392 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3393 			.conf = &cp_mreg,
3394 		};
3395 		actions[2] = (struct rte_flow_action){
3396 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3397 			.conf = &jump,
3398 		};
3399 		actions[3] = (struct rte_flow_action){
3400 			.type = RTE_FLOW_ACTION_TYPE_END,
3401 		};
3402 	} else {
3403 		/* Default rule, wildcard match. */
3404 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3405 		items[0] = (struct rte_flow_item){
3406 			.type = RTE_FLOW_ITEM_TYPE_END,
3407 		};
3408 		actions[0] = (struct rte_flow_action){
3409 			.type = (enum rte_flow_action_type)
3410 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3411 			.conf = &cp_mreg,
3412 		};
3413 		actions[1] = (struct rte_flow_action){
3414 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3415 			.conf = &jump,
3416 		};
3417 		actions[2] = (struct rte_flow_action){
3418 			.type = RTE_FLOW_ACTION_TYPE_END,
3419 		};
3420 	}
3421 	/* Build a new entry. */
3422 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3423 	if (!mcp_res) {
3424 		rte_errno = ENOMEM;
3425 		return NULL;
3426 	}
3427 	mcp_res->idx = idx;
3428 	/*
3429 	 * The copy Flows are not included in any list. There
3430 	 * ones are referenced from other Flows and can not
3431 	 * be applied, removed, deleted in ardbitrary order
3432 	 * by list traversing.
3433 	 */
3434 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3435 					 actions, false, error);
3436 	if (!mcp_res->rix_flow)
3437 		goto error;
3438 	mcp_res->refcnt++;
3439 	mcp_res->hlist_ent.key = mark_id;
3440 	ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3441 				&mcp_res->hlist_ent);
3442 	MLX5_ASSERT(!ret);
3443 	if (ret)
3444 		goto error;
3445 	return mcp_res;
3446 error:
3447 	if (mcp_res->rix_flow)
3448 		flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3449 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3450 	return NULL;
3451 }
3452 
3453 /**
3454  * Release flow in RX_CP_TBL.
3455  *
3456  * @param dev
3457  *   Pointer to Ethernet device.
3458  * @flow
3459  *   Parent flow for wich copying is provided.
3460  */
3461 static void
3462 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3463 			  struct rte_flow *flow)
3464 {
3465 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3466 	struct mlx5_priv *priv = dev->data->dev_private;
3467 
3468 	if (!flow->rix_mreg_copy)
3469 		return;
3470 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3471 				 flow->rix_mreg_copy);
3472 	if (!mcp_res || !priv->mreg_cp_tbl)
3473 		return;
3474 	if (flow->copy_applied) {
3475 		MLX5_ASSERT(mcp_res->appcnt);
3476 		flow->copy_applied = 0;
3477 		--mcp_res->appcnt;
3478 		if (!mcp_res->appcnt) {
3479 			struct rte_flow *mcp_flow = mlx5_ipool_get
3480 					(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3481 					mcp_res->rix_flow);
3482 
3483 			if (mcp_flow)
3484 				flow_drv_remove(dev, mcp_flow);
3485 		}
3486 	}
3487 	/*
3488 	 * We do not check availability of metadata registers here,
3489 	 * because copy resources are not allocated in this case.
3490 	 */
3491 	if (--mcp_res->refcnt)
3492 		return;
3493 	MLX5_ASSERT(mcp_res->rix_flow);
3494 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3495 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3496 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3497 	flow->rix_mreg_copy = 0;
3498 }
3499 
3500 /**
3501  * Start flow in RX_CP_TBL.
3502  *
3503  * @param dev
3504  *   Pointer to Ethernet device.
3505  * @flow
3506  *   Parent flow for wich copying is provided.
3507  *
3508  * @return
3509  *   0 on success, a negative errno value otherwise and rte_errno is set.
3510  */
3511 static int
3512 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3513 			    struct rte_flow *flow)
3514 {
3515 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3516 	struct mlx5_priv *priv = dev->data->dev_private;
3517 	int ret;
3518 
3519 	if (!flow->rix_mreg_copy || flow->copy_applied)
3520 		return 0;
3521 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3522 				 flow->rix_mreg_copy);
3523 	if (!mcp_res)
3524 		return 0;
3525 	if (!mcp_res->appcnt) {
3526 		struct rte_flow *mcp_flow = mlx5_ipool_get
3527 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3528 				mcp_res->rix_flow);
3529 
3530 		if (mcp_flow) {
3531 			ret = flow_drv_apply(dev, mcp_flow, NULL);
3532 			if (ret)
3533 				return ret;
3534 		}
3535 	}
3536 	++mcp_res->appcnt;
3537 	flow->copy_applied = 1;
3538 	return 0;
3539 }
3540 
3541 /**
3542  * Stop flow in RX_CP_TBL.
3543  *
3544  * @param dev
3545  *   Pointer to Ethernet device.
3546  * @flow
3547  *   Parent flow for wich copying is provided.
3548  */
3549 static void
3550 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3551 			   struct rte_flow *flow)
3552 {
3553 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3554 	struct mlx5_priv *priv = dev->data->dev_private;
3555 
3556 	if (!flow->rix_mreg_copy || !flow->copy_applied)
3557 		return;
3558 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3559 				 flow->rix_mreg_copy);
3560 	if (!mcp_res)
3561 		return;
3562 	MLX5_ASSERT(mcp_res->appcnt);
3563 	--mcp_res->appcnt;
3564 	flow->copy_applied = 0;
3565 	if (!mcp_res->appcnt) {
3566 		struct rte_flow *mcp_flow = mlx5_ipool_get
3567 				(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
3568 				mcp_res->rix_flow);
3569 
3570 		if (mcp_flow)
3571 			flow_drv_remove(dev, mcp_flow);
3572 	}
3573 }
3574 
3575 /**
3576  * Remove the default copy action from RX_CP_TBL.
3577  *
3578  * @param dev
3579  *   Pointer to Ethernet device.
3580  */
3581 static void
3582 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3583 {
3584 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3585 	struct mlx5_priv *priv = dev->data->dev_private;
3586 
3587 	/* Check if default flow is registered. */
3588 	if (!priv->mreg_cp_tbl)
3589 		return;
3590 	mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3591 					    MLX5_DEFAULT_COPY_ID);
3592 	if (!mcp_res)
3593 		return;
3594 	MLX5_ASSERT(mcp_res->rix_flow);
3595 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3596 	mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3597 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3598 }
3599 
3600 /**
3601  * Add the default copy action in in RX_CP_TBL.
3602  *
3603  * @param dev
3604  *   Pointer to Ethernet device.
3605  * @param[out] error
3606  *   Perform verbose error reporting if not NULL.
3607  *
3608  * @return
3609  *   0 for success, negative value otherwise and rte_errno is set.
3610  */
3611 static int
3612 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3613 				  struct rte_flow_error *error)
3614 {
3615 	struct mlx5_priv *priv = dev->data->dev_private;
3616 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3617 
3618 	/* Check whether extensive metadata feature is engaged. */
3619 	if (!priv->config.dv_flow_en ||
3620 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3621 	    !mlx5_flow_ext_mreg_supported(dev) ||
3622 	    !priv->sh->dv_regc0_mask)
3623 		return 0;
3624 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3625 	if (!mcp_res)
3626 		return -rte_errno;
3627 	return 0;
3628 }
3629 
3630 /**
3631  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3632  *
3633  * All the flow having Q/RSS action should be split by
3634  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3635  * performs the following,
3636  *   - CQE->flow_tag := reg_c[1] (MARK)
3637  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3638  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3639  * but there should be a flow per each MARK ID set by MARK action.
3640  *
3641  * For the aforementioned reason, if there's a MARK action in flow's action
3642  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3643  * the MARK ID to CQE's flow_tag like,
3644  *   - If reg_c[1] is mark_id,
3645  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3646  *
3647  * For SET_META action which stores value in reg_c[0], as the destination is
3648  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3649  * MARK ID means the default flow. The default flow looks like,
3650  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3651  *
3652  * @param dev
3653  *   Pointer to Ethernet device.
3654  * @param flow
3655  *   Pointer to flow structure.
3656  * @param[in] actions
3657  *   Pointer to the list of actions.
3658  * @param[out] error
3659  *   Perform verbose error reporting if not NULL.
3660  *
3661  * @return
3662  *   0 on success, negative value otherwise and rte_errno is set.
3663  */
3664 static int
3665 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3666 			    struct rte_flow *flow,
3667 			    const struct rte_flow_action *actions,
3668 			    struct rte_flow_error *error)
3669 {
3670 	struct mlx5_priv *priv = dev->data->dev_private;
3671 	struct mlx5_dev_config *config = &priv->config;
3672 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3673 	const struct rte_flow_action_mark *mark;
3674 
3675 	/* Check whether extensive metadata feature is engaged. */
3676 	if (!config->dv_flow_en ||
3677 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3678 	    !mlx5_flow_ext_mreg_supported(dev) ||
3679 	    !priv->sh->dv_regc0_mask)
3680 		return 0;
3681 	/* Find MARK action. */
3682 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3683 		switch (actions->type) {
3684 		case RTE_FLOW_ACTION_TYPE_FLAG:
3685 			mcp_res = flow_mreg_add_copy_action
3686 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
3687 			if (!mcp_res)
3688 				return -rte_errno;
3689 			flow->rix_mreg_copy = mcp_res->idx;
3690 			if (dev->data->dev_started) {
3691 				mcp_res->appcnt++;
3692 				flow->copy_applied = 1;
3693 			}
3694 			return 0;
3695 		case RTE_FLOW_ACTION_TYPE_MARK:
3696 			mark = (const struct rte_flow_action_mark *)
3697 				actions->conf;
3698 			mcp_res =
3699 				flow_mreg_add_copy_action(dev, mark->id, error);
3700 			if (!mcp_res)
3701 				return -rte_errno;
3702 			flow->rix_mreg_copy = mcp_res->idx;
3703 			if (dev->data->dev_started) {
3704 				mcp_res->appcnt++;
3705 				flow->copy_applied = 1;
3706 			}
3707 			return 0;
3708 		default:
3709 			break;
3710 		}
3711 	}
3712 	return 0;
3713 }
3714 
3715 #define MLX5_MAX_SPLIT_ACTIONS 24
3716 #define MLX5_MAX_SPLIT_ITEMS 24
3717 
3718 /**
3719  * Split the hairpin flow.
3720  * Since HW can't support encap and push-vlan on Rx, we move these
3721  * actions to Tx.
3722  * If the count action is after the encap then we also
3723  * move the count action. in this case the count will also measure
3724  * the outer bytes.
3725  *
3726  * @param dev
3727  *   Pointer to Ethernet device.
3728  * @param[in] actions
3729  *   Associated actions (list terminated by the END action).
3730  * @param[out] actions_rx
3731  *   Rx flow actions.
3732  * @param[out] actions_tx
3733  *   Tx flow actions..
3734  * @param[out] pattern_tx
3735  *   The pattern items for the Tx flow.
3736  * @param[out] flow_id
3737  *   The flow ID connected to this flow.
3738  *
3739  * @return
3740  *   0 on success.
3741  */
3742 static int
3743 flow_hairpin_split(struct rte_eth_dev *dev,
3744 		   const struct rte_flow_action actions[],
3745 		   struct rte_flow_action actions_rx[],
3746 		   struct rte_flow_action actions_tx[],
3747 		   struct rte_flow_item pattern_tx[],
3748 		   uint32_t *flow_id)
3749 {
3750 	struct mlx5_priv *priv = dev->data->dev_private;
3751 	const struct rte_flow_action_raw_encap *raw_encap;
3752 	const struct rte_flow_action_raw_decap *raw_decap;
3753 	struct mlx5_rte_flow_action_set_tag *set_tag;
3754 	struct rte_flow_action *tag_action;
3755 	struct mlx5_rte_flow_item_tag *tag_item;
3756 	struct rte_flow_item *item;
3757 	char *addr;
3758 	int encap = 0;
3759 
3760 	mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3761 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3762 		switch (actions->type) {
3763 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3764 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3765 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3766 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3767 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3768 			rte_memcpy(actions_tx, actions,
3769 			       sizeof(struct rte_flow_action));
3770 			actions_tx++;
3771 			break;
3772 		case RTE_FLOW_ACTION_TYPE_COUNT:
3773 			if (encap) {
3774 				rte_memcpy(actions_tx, actions,
3775 					   sizeof(struct rte_flow_action));
3776 				actions_tx++;
3777 			} else {
3778 				rte_memcpy(actions_rx, actions,
3779 					   sizeof(struct rte_flow_action));
3780 				actions_rx++;
3781 			}
3782 			break;
3783 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3784 			raw_encap = actions->conf;
3785 			if (raw_encap->size >
3786 			    (sizeof(struct rte_flow_item_eth) +
3787 			     sizeof(struct rte_flow_item_ipv4))) {
3788 				memcpy(actions_tx, actions,
3789 				       sizeof(struct rte_flow_action));
3790 				actions_tx++;
3791 				encap = 1;
3792 			} else {
3793 				rte_memcpy(actions_rx, actions,
3794 					   sizeof(struct rte_flow_action));
3795 				actions_rx++;
3796 			}
3797 			break;
3798 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3799 			raw_decap = actions->conf;
3800 			if (raw_decap->size <
3801 			    (sizeof(struct rte_flow_item_eth) +
3802 			     sizeof(struct rte_flow_item_ipv4))) {
3803 				memcpy(actions_tx, actions,
3804 				       sizeof(struct rte_flow_action));
3805 				actions_tx++;
3806 			} else {
3807 				rte_memcpy(actions_rx, actions,
3808 					   sizeof(struct rte_flow_action));
3809 				actions_rx++;
3810 			}
3811 			break;
3812 		default:
3813 			rte_memcpy(actions_rx, actions,
3814 				   sizeof(struct rte_flow_action));
3815 			actions_rx++;
3816 			break;
3817 		}
3818 	}
3819 	/* Add set meta action and end action for the Rx flow. */
3820 	tag_action = actions_rx;
3821 	tag_action->type = (enum rte_flow_action_type)
3822 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3823 	actions_rx++;
3824 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3825 	actions_rx++;
3826 	set_tag = (void *)actions_rx;
3827 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3828 	MLX5_ASSERT(set_tag->id > REG_NON);
3829 	set_tag->data = *flow_id;
3830 	tag_action->conf = set_tag;
3831 	/* Create Tx item list. */
3832 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3833 	addr = (void *)&pattern_tx[2];
3834 	item = pattern_tx;
3835 	item->type = (enum rte_flow_item_type)
3836 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3837 	tag_item = (void *)addr;
3838 	tag_item->data = *flow_id;
3839 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3840 	MLX5_ASSERT(set_tag->id > REG_NON);
3841 	item->spec = tag_item;
3842 	addr += sizeof(struct mlx5_rte_flow_item_tag);
3843 	tag_item = (void *)addr;
3844 	tag_item->data = UINT32_MAX;
3845 	tag_item->id = UINT16_MAX;
3846 	item->mask = tag_item;
3847 	item->last = NULL;
3848 	item++;
3849 	item->type = RTE_FLOW_ITEM_TYPE_END;
3850 	return 0;
3851 }
3852 
3853 /**
3854  * The last stage of splitting chain, just creates the subflow
3855  * without any modification.
3856  *
3857  * @param[in] dev
3858  *   Pointer to Ethernet device.
3859  * @param[in] flow
3860  *   Parent flow structure pointer.
3861  * @param[in, out] sub_flow
3862  *   Pointer to return the created subflow, may be NULL.
3863  * @param[in] prefix_layers
3864  *   Prefix subflow layers, may be 0.
3865  * @param[in] attr
3866  *   Flow rule attributes.
3867  * @param[in] items
3868  *   Pattern specification (list terminated by the END pattern item).
3869  * @param[in] actions
3870  *   Associated actions (list terminated by the END action).
3871  * @param[in] external
3872  *   This flow rule is created by request external to PMD.
3873  * @param[in] flow_idx
3874  *   This memory pool index to the flow.
3875  * @param[out] error
3876  *   Perform verbose error reporting if not NULL.
3877  * @return
3878  *   0 on success, negative value otherwise
3879  */
3880 static int
3881 flow_create_split_inner(struct rte_eth_dev *dev,
3882 			struct rte_flow *flow,
3883 			struct mlx5_flow **sub_flow,
3884 			uint64_t prefix_layers,
3885 			const struct rte_flow_attr *attr,
3886 			const struct rte_flow_item items[],
3887 			const struct rte_flow_action actions[],
3888 			bool external, uint32_t flow_idx,
3889 			struct rte_flow_error *error)
3890 {
3891 	struct mlx5_flow *dev_flow;
3892 
3893 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
3894 		flow_idx, error);
3895 	if (!dev_flow)
3896 		return -rte_errno;
3897 	dev_flow->flow = flow;
3898 	dev_flow->external = external;
3899 	/* Subflow object was created, we must include one in the list. */
3900 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3901 		      dev_flow->handle, next);
3902 	/*
3903 	 * If dev_flow is as one of the suffix flow, some actions in suffix
3904 	 * flow may need some user defined item layer flags.
3905 	 */
3906 	if (prefix_layers)
3907 		dev_flow->handle->layers = prefix_layers;
3908 	if (sub_flow)
3909 		*sub_flow = dev_flow;
3910 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3911 }
3912 
3913 /**
3914  * Split the meter flow.
3915  *
3916  * As meter flow will split to three sub flow, other than meter
3917  * action, the other actions make sense to only meter accepts
3918  * the packet. If it need to be dropped, no other additional
3919  * actions should be take.
3920  *
3921  * One kind of special action which decapsulates the L3 tunnel
3922  * header will be in the prefix sub flow, as not to take the
3923  * L3 tunnel header into account.
3924  *
3925  * @param dev
3926  *   Pointer to Ethernet device.
3927  * @param[in] items
3928  *   Pattern specification (list terminated by the END pattern item).
3929  * @param[out] sfx_items
3930  *   Suffix flow match items (list terminated by the END pattern item).
3931  * @param[in] actions
3932  *   Associated actions (list terminated by the END action).
3933  * @param[out] actions_sfx
3934  *   Suffix flow actions.
3935  * @param[out] actions_pre
3936  *   Prefix flow actions.
3937  * @param[out] pattern_sfx
3938  *   The pattern items for the suffix flow.
3939  * @param[out] tag_sfx
3940  *   Pointer to suffix flow tag.
3941  *
3942  * @return
3943  *   0 on success.
3944  */
3945 static int
3946 flow_meter_split_prep(struct rte_eth_dev *dev,
3947 		 const struct rte_flow_item items[],
3948 		 struct rte_flow_item sfx_items[],
3949 		 const struct rte_flow_action actions[],
3950 		 struct rte_flow_action actions_sfx[],
3951 		 struct rte_flow_action actions_pre[])
3952 {
3953 	struct rte_flow_action *tag_action = NULL;
3954 	struct rte_flow_item *tag_item;
3955 	struct mlx5_rte_flow_action_set_tag *set_tag;
3956 	struct rte_flow_error error;
3957 	const struct rte_flow_action_raw_encap *raw_encap;
3958 	const struct rte_flow_action_raw_decap *raw_decap;
3959 	struct mlx5_rte_flow_item_tag *tag_spec;
3960 	struct mlx5_rte_flow_item_tag *tag_mask;
3961 	uint32_t tag_id;
3962 	bool copy_vlan = false;
3963 
3964 	/* Prepare the actions for prefix and suffix flow. */
3965 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3966 		struct rte_flow_action **action_cur = NULL;
3967 
3968 		switch (actions->type) {
3969 		case RTE_FLOW_ACTION_TYPE_METER:
3970 			/* Add the extra tag action first. */
3971 			tag_action = actions_pre;
3972 			tag_action->type = (enum rte_flow_action_type)
3973 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3974 			actions_pre++;
3975 			action_cur = &actions_pre;
3976 			break;
3977 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3978 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3979 			action_cur = &actions_pre;
3980 			break;
3981 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3982 			raw_encap = actions->conf;
3983 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3984 				action_cur = &actions_pre;
3985 			break;
3986 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3987 			raw_decap = actions->conf;
3988 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3989 				action_cur = &actions_pre;
3990 			break;
3991 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3992 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3993 			copy_vlan = true;
3994 			break;
3995 		default:
3996 			break;
3997 		}
3998 		if (!action_cur)
3999 			action_cur = &actions_sfx;
4000 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4001 		(*action_cur)++;
4002 	}
4003 	/* Add end action to the actions. */
4004 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4005 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4006 	actions_pre++;
4007 	/* Set the tag. */
4008 	set_tag = (void *)actions_pre;
4009 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4010 	/*
4011 	 * Get the id from the qrss_pool to make qrss share the id with meter.
4012 	 */
4013 	tag_id = flow_qrss_get_id(dev);
4014 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4015 	assert(tag_action);
4016 	tag_action->conf = set_tag;
4017 	/* Prepare the suffix subflow items. */
4018 	tag_item = sfx_items++;
4019 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4020 		int item_type = items->type;
4021 
4022 		switch (item_type) {
4023 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4024 			memcpy(sfx_items, items, sizeof(*sfx_items));
4025 			sfx_items++;
4026 			break;
4027 		case RTE_FLOW_ITEM_TYPE_VLAN:
4028 			if (copy_vlan) {
4029 				memcpy(sfx_items, items, sizeof(*sfx_items));
4030 				/*
4031 				 * Convert to internal match item, it is used
4032 				 * for vlan push and set vid.
4033 				 */
4034 				sfx_items->type = (enum rte_flow_item_type)
4035 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4036 				sfx_items++;
4037 			}
4038 			break;
4039 		default:
4040 			break;
4041 		}
4042 	}
4043 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4044 	sfx_items++;
4045 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4046 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4047 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4048 	tag_mask = tag_spec + 1;
4049 	tag_mask->data = 0xffffff00;
4050 	tag_item->type = (enum rte_flow_item_type)
4051 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4052 	tag_item->spec = tag_spec;
4053 	tag_item->last = NULL;
4054 	tag_item->mask = tag_mask;
4055 	return tag_id;
4056 }
4057 
4058 /**
4059  * Split action list having QUEUE/RSS for metadata register copy.
4060  *
4061  * Once Q/RSS action is detected in user's action list, the flow action
4062  * should be split in order to copy metadata registers, which will happen in
4063  * RX_CP_TBL like,
4064  *   - CQE->flow_tag := reg_c[1] (MARK)
4065  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4066  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4067  * This is because the last action of each flow must be a terminal action
4068  * (QUEUE, RSS or DROP).
4069  *
4070  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4071  * stored and kept in the mlx5_flow structure per each sub_flow.
4072  *
4073  * The Q/RSS action is replaced with,
4074  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4075  * And the following JUMP action is added at the end,
4076  *   - JUMP, to RX_CP_TBL.
4077  *
4078  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4079  * flow_create_split_metadata() routine. The flow will look like,
4080  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4081  *
4082  * @param dev
4083  *   Pointer to Ethernet device.
4084  * @param[out] split_actions
4085  *   Pointer to store split actions to jump to CP_TBL.
4086  * @param[in] actions
4087  *   Pointer to the list of original flow actions.
4088  * @param[in] qrss
4089  *   Pointer to the Q/RSS action.
4090  * @param[in] actions_n
4091  *   Number of original actions.
4092  * @param[out] error
4093  *   Perform verbose error reporting if not NULL.
4094  *
4095  * @return
4096  *   non-zero unique flow_id on success, otherwise 0 and
4097  *   error/rte_error are set.
4098  */
4099 static uint32_t
4100 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4101 			  struct rte_flow_action *split_actions,
4102 			  const struct rte_flow_action *actions,
4103 			  const struct rte_flow_action *qrss,
4104 			  int actions_n, struct rte_flow_error *error)
4105 {
4106 	struct mlx5_rte_flow_action_set_tag *set_tag;
4107 	struct rte_flow_action_jump *jump;
4108 	const int qrss_idx = qrss - actions;
4109 	uint32_t flow_id = 0;
4110 	int ret = 0;
4111 
4112 	/*
4113 	 * Given actions will be split
4114 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4115 	 * - Add jump to mreg CP_TBL.
4116 	 * As a result, there will be one more action.
4117 	 */
4118 	++actions_n;
4119 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4120 	set_tag = (void *)(split_actions + actions_n);
4121 	/*
4122 	 * If tag action is not set to void(it means we are not the meter
4123 	 * suffix flow), add the tag action. Since meter suffix flow already
4124 	 * has the tag added.
4125 	 */
4126 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4127 		/*
4128 		 * Allocate the new subflow ID. This one is unique within
4129 		 * device and not shared with representors. Otherwise,
4130 		 * we would have to resolve multi-thread access synch
4131 		 * issue. Each flow on the shared device is appended
4132 		 * with source vport identifier, so the resulting
4133 		 * flows will be unique in the shared (by master and
4134 		 * representors) domain even if they have coinciding
4135 		 * IDs.
4136 		 */
4137 		flow_id = flow_qrss_get_id(dev);
4138 		if (!flow_id)
4139 			return rte_flow_error_set(error, ENOMEM,
4140 						  RTE_FLOW_ERROR_TYPE_ACTION,
4141 						  NULL, "can't allocate id "
4142 						  "for split Q/RSS subflow");
4143 		/* Internal SET_TAG action to set flow ID. */
4144 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4145 			.data = flow_id,
4146 		};
4147 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4148 		if (ret < 0)
4149 			return ret;
4150 		set_tag->id = ret;
4151 		/* Construct new actions array. */
4152 		/* Replace QUEUE/RSS action. */
4153 		split_actions[qrss_idx] = (struct rte_flow_action){
4154 			.type = (enum rte_flow_action_type)
4155 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4156 			.conf = set_tag,
4157 		};
4158 	}
4159 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4160 	jump = (void *)(set_tag + 1);
4161 	*jump = (struct rte_flow_action_jump){
4162 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4163 	};
4164 	split_actions[actions_n - 2] = (struct rte_flow_action){
4165 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4166 		.conf = jump,
4167 	};
4168 	split_actions[actions_n - 1] = (struct rte_flow_action){
4169 		.type = RTE_FLOW_ACTION_TYPE_END,
4170 	};
4171 	return flow_id;
4172 }
4173 
4174 /**
4175  * Extend the given action list for Tx metadata copy.
4176  *
4177  * Copy the given action list to the ext_actions and add flow metadata register
4178  * copy action in order to copy reg_a set by WQE to reg_c[0].
4179  *
4180  * @param[out] ext_actions
4181  *   Pointer to the extended action list.
4182  * @param[in] actions
4183  *   Pointer to the list of actions.
4184  * @param[in] actions_n
4185  *   Number of actions in the list.
4186  * @param[out] error
4187  *   Perform verbose error reporting if not NULL.
4188  * @param[in] encap_idx
4189  *   The encap action inndex.
4190  *
4191  * @return
4192  *   0 on success, negative value otherwise
4193  */
4194 static int
4195 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4196 		       struct rte_flow_action *ext_actions,
4197 		       const struct rte_flow_action *actions,
4198 		       int actions_n, struct rte_flow_error *error,
4199 		       int encap_idx)
4200 {
4201 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4202 		(struct mlx5_flow_action_copy_mreg *)
4203 			(ext_actions + actions_n + 1);
4204 	int ret;
4205 
4206 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4207 	if (ret < 0)
4208 		return ret;
4209 	cp_mreg->dst = ret;
4210 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4211 	if (ret < 0)
4212 		return ret;
4213 	cp_mreg->src = ret;
4214 	if (encap_idx != 0)
4215 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4216 	if (encap_idx == actions_n - 1) {
4217 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4218 			.type = (enum rte_flow_action_type)
4219 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4220 			.conf = cp_mreg,
4221 		};
4222 		ext_actions[actions_n] = (struct rte_flow_action){
4223 			.type = RTE_FLOW_ACTION_TYPE_END,
4224 		};
4225 	} else {
4226 		ext_actions[encap_idx] = (struct rte_flow_action){
4227 			.type = (enum rte_flow_action_type)
4228 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4229 			.conf = cp_mreg,
4230 		};
4231 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4232 				sizeof(*ext_actions) * (actions_n - encap_idx));
4233 	}
4234 	return 0;
4235 }
4236 
4237 /**
4238  * The splitting for metadata feature.
4239  *
4240  * - Q/RSS action on NIC Rx should be split in order to pass by
4241  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4242  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4243  *
4244  * - All the actions on NIC Tx should have a mreg copy action to
4245  *   copy reg_a from WQE to reg_c[0].
4246  *
4247  * @param dev
4248  *   Pointer to Ethernet device.
4249  * @param[in] flow
4250  *   Parent flow structure pointer.
4251  * @param[in] prefix_layers
4252  *   Prefix flow layer flags.
4253  * @param[in] attr
4254  *   Flow rule attributes.
4255  * @param[in] items
4256  *   Pattern specification (list terminated by the END pattern item).
4257  * @param[in] actions
4258  *   Associated actions (list terminated by the END action).
4259  * @param[in] external
4260  *   This flow rule is created by request external to PMD.
4261  * @param[in] flow_idx
4262  *   This memory pool index to the flow.
4263  * @param[out] error
4264  *   Perform verbose error reporting if not NULL.
4265  * @return
4266  *   0 on success, negative value otherwise
4267  */
4268 static int
4269 flow_create_split_metadata(struct rte_eth_dev *dev,
4270 			   struct rte_flow *flow,
4271 			   uint64_t prefix_layers,
4272 			   const struct rte_flow_attr *attr,
4273 			   const struct rte_flow_item items[],
4274 			   const struct rte_flow_action actions[],
4275 			   bool external, uint32_t flow_idx,
4276 			   struct rte_flow_error *error)
4277 {
4278 	struct mlx5_priv *priv = dev->data->dev_private;
4279 	struct mlx5_dev_config *config = &priv->config;
4280 	const struct rte_flow_action *qrss = NULL;
4281 	struct rte_flow_action *ext_actions = NULL;
4282 	struct mlx5_flow *dev_flow = NULL;
4283 	uint32_t qrss_id = 0;
4284 	int mtr_sfx = 0;
4285 	size_t act_size;
4286 	int actions_n;
4287 	int encap_idx;
4288 	int ret;
4289 
4290 	/* Check whether extensive metadata feature is engaged. */
4291 	if (!config->dv_flow_en ||
4292 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4293 	    !mlx5_flow_ext_mreg_supported(dev))
4294 		return flow_create_split_inner(dev, flow, NULL, prefix_layers,
4295 					       attr, items, actions, external,
4296 					       flow_idx, error);
4297 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4298 							   &encap_idx);
4299 	if (qrss) {
4300 		/* Exclude hairpin flows from splitting. */
4301 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4302 			const struct rte_flow_action_queue *queue;
4303 
4304 			queue = qrss->conf;
4305 			if (mlx5_rxq_get_type(dev, queue->index) ==
4306 			    MLX5_RXQ_TYPE_HAIRPIN)
4307 				qrss = NULL;
4308 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4309 			const struct rte_flow_action_rss *rss;
4310 
4311 			rss = qrss->conf;
4312 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4313 			    MLX5_RXQ_TYPE_HAIRPIN)
4314 				qrss = NULL;
4315 		}
4316 	}
4317 	if (qrss) {
4318 		/* Check if it is in meter suffix table. */
4319 		mtr_sfx = attr->group == (attr->transfer ?
4320 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4321 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4322 		/*
4323 		 * Q/RSS action on NIC Rx should be split in order to pass by
4324 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4325 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4326 		 */
4327 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4328 			   sizeof(struct rte_flow_action_set_tag) +
4329 			   sizeof(struct rte_flow_action_jump);
4330 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4331 					  SOCKET_ID_ANY);
4332 		if (!ext_actions)
4333 			return rte_flow_error_set(error, ENOMEM,
4334 						  RTE_FLOW_ERROR_TYPE_ACTION,
4335 						  NULL, "no memory to split "
4336 						  "metadata flow");
4337 		/*
4338 		 * If we are the suffix flow of meter, tag already exist.
4339 		 * Set the tag action to void.
4340 		 */
4341 		if (mtr_sfx)
4342 			ext_actions[qrss - actions].type =
4343 						RTE_FLOW_ACTION_TYPE_VOID;
4344 		else
4345 			ext_actions[qrss - actions].type =
4346 						(enum rte_flow_action_type)
4347 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4348 		/*
4349 		 * Create the new actions list with removed Q/RSS action
4350 		 * and appended set tag and jump to register copy table
4351 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4352 		 * in advance, because it is needed for set tag action.
4353 		 */
4354 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
4355 						    qrss, actions_n, error);
4356 		if (!mtr_sfx && !qrss_id) {
4357 			ret = -rte_errno;
4358 			goto exit;
4359 		}
4360 	} else if (attr->egress && !attr->transfer) {
4361 		/*
4362 		 * All the actions on NIC Tx should have a metadata register
4363 		 * copy action to copy reg_a from WQE to reg_c[meta]
4364 		 */
4365 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4366 			   sizeof(struct mlx5_flow_action_copy_mreg);
4367 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4368 					  SOCKET_ID_ANY);
4369 		if (!ext_actions)
4370 			return rte_flow_error_set(error, ENOMEM,
4371 						  RTE_FLOW_ERROR_TYPE_ACTION,
4372 						  NULL, "no memory to split "
4373 						  "metadata flow");
4374 		/* Create the action list appended with copy register. */
4375 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
4376 					     actions_n, error, encap_idx);
4377 		if (ret < 0)
4378 			goto exit;
4379 	}
4380 	/* Add the unmodified original or prefix subflow. */
4381 	ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
4382 				      items, ext_actions ? ext_actions :
4383 				      actions, external, flow_idx, error);
4384 	if (ret < 0)
4385 		goto exit;
4386 	MLX5_ASSERT(dev_flow);
4387 	if (qrss) {
4388 		const struct rte_flow_attr q_attr = {
4389 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4390 			.ingress = 1,
4391 		};
4392 		/* Internal PMD action to set register. */
4393 		struct mlx5_rte_flow_item_tag q_tag_spec = {
4394 			.data = qrss_id,
4395 			.id = REG_NON,
4396 		};
4397 		struct rte_flow_item q_items[] = {
4398 			{
4399 				.type = (enum rte_flow_item_type)
4400 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4401 				.spec = &q_tag_spec,
4402 				.last = NULL,
4403 				.mask = NULL,
4404 			},
4405 			{
4406 				.type = RTE_FLOW_ITEM_TYPE_END,
4407 			},
4408 		};
4409 		struct rte_flow_action q_actions[] = {
4410 			{
4411 				.type = qrss->type,
4412 				.conf = qrss->conf,
4413 			},
4414 			{
4415 				.type = RTE_FLOW_ACTION_TYPE_END,
4416 			},
4417 		};
4418 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4419 
4420 		/*
4421 		 * Configure the tag item only if there is no meter subflow.
4422 		 * Since tag is already marked in the meter suffix subflow
4423 		 * we can just use the meter suffix items as is.
4424 		 */
4425 		if (qrss_id) {
4426 			/* Not meter subflow. */
4427 			MLX5_ASSERT(!mtr_sfx);
4428 			/*
4429 			 * Put unique id in prefix flow due to it is destroyed
4430 			 * after suffix flow and id will be freed after there
4431 			 * is no actual flows with this id and identifier
4432 			 * reallocation becomes possible (for example, for
4433 			 * other flows in other threads).
4434 			 */
4435 			dev_flow->handle->split_flow_id = qrss_id;
4436 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4437 						   error);
4438 			if (ret < 0)
4439 				goto exit;
4440 			q_tag_spec.id = ret;
4441 		}
4442 		dev_flow = NULL;
4443 		/* Add suffix subflow to execute Q/RSS. */
4444 		ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4445 					      &q_attr, mtr_sfx ? items :
4446 					      q_items, q_actions,
4447 					      external, flow_idx, error);
4448 		if (ret < 0)
4449 			goto exit;
4450 		/* qrss ID should be freed if failed. */
4451 		qrss_id = 0;
4452 		MLX5_ASSERT(dev_flow);
4453 	}
4454 
4455 exit:
4456 	/*
4457 	 * We do not destroy the partially created sub_flows in case of error.
4458 	 * These ones are included into parent flow list and will be destroyed
4459 	 * by flow_drv_destroy.
4460 	 */
4461 	flow_qrss_free_id(dev, qrss_id);
4462 	mlx5_free(ext_actions);
4463 	return ret;
4464 }
4465 
4466 /**
4467  * The splitting for meter feature.
4468  *
4469  * - The meter flow will be split to two flows as prefix and
4470  *   suffix flow. The packets make sense only it pass the prefix
4471  *   meter action.
4472  *
4473  * - Reg_C_5 is used for the packet to match betweend prefix and
4474  *   suffix flow.
4475  *
4476  * @param dev
4477  *   Pointer to Ethernet device.
4478  * @param[in] flow
4479  *   Parent flow structure pointer.
4480  * @param[in] attr
4481  *   Flow rule attributes.
4482  * @param[in] items
4483  *   Pattern specification (list terminated by the END pattern item).
4484  * @param[in] actions
4485  *   Associated actions (list terminated by the END action).
4486  * @param[in] external
4487  *   This flow rule is created by request external to PMD.
4488  * @param[in] flow_idx
4489  *   This memory pool index to the flow.
4490  * @param[out] error
4491  *   Perform verbose error reporting if not NULL.
4492  * @return
4493  *   0 on success, negative value otherwise
4494  */
4495 static int
4496 flow_create_split_meter(struct rte_eth_dev *dev,
4497 			   struct rte_flow *flow,
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 rte_flow_action *sfx_actions = NULL;
4506 	struct rte_flow_action *pre_actions = NULL;
4507 	struct rte_flow_item *sfx_items = NULL;
4508 	struct mlx5_flow *dev_flow = NULL;
4509 	struct rte_flow_attr sfx_attr = *attr;
4510 	uint32_t mtr = 0;
4511 	uint32_t mtr_tag_id = 0;
4512 	size_t act_size;
4513 	size_t item_size;
4514 	int actions_n = 0;
4515 	int ret;
4516 
4517 	if (priv->mtr_en)
4518 		actions_n = flow_check_meter_action(actions, &mtr);
4519 	if (mtr) {
4520 		/* The five prefix actions: meter, decap, encap, tag, end. */
4521 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4522 			   sizeof(struct mlx5_rte_flow_action_set_tag);
4523 		/* tag, vlan, port id, end. */
4524 #define METER_SUFFIX_ITEM 4
4525 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4526 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
4527 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
4528 					  0, SOCKET_ID_ANY);
4529 		if (!sfx_actions)
4530 			return rte_flow_error_set(error, ENOMEM,
4531 						  RTE_FLOW_ERROR_TYPE_ACTION,
4532 						  NULL, "no memory to split "
4533 						  "meter flow");
4534 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4535 			     act_size);
4536 		pre_actions = sfx_actions + actions_n;
4537 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4538 						   actions, sfx_actions,
4539 						   pre_actions);
4540 		if (!mtr_tag_id) {
4541 			ret = -rte_errno;
4542 			goto exit;
4543 		}
4544 		/* Add the prefix subflow. */
4545 		ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4546 					      items, pre_actions, external,
4547 					      flow_idx, error);
4548 		if (ret) {
4549 			ret = -rte_errno;
4550 			goto exit;
4551 		}
4552 		dev_flow->handle->split_flow_id = mtr_tag_id;
4553 		/* Setting the sfx group atrr. */
4554 		sfx_attr.group = sfx_attr.transfer ?
4555 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4556 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4557 	}
4558 	/* Add the prefix subflow. */
4559 	ret = flow_create_split_metadata(dev, flow, dev_flow ?
4560 					 flow_get_prefix_layer_flags(dev_flow) :
4561 					 0, &sfx_attr,
4562 					 sfx_items ? sfx_items : items,
4563 					 sfx_actions ? sfx_actions : actions,
4564 					 external, flow_idx, error);
4565 exit:
4566 	if (sfx_actions)
4567 		mlx5_free(sfx_actions);
4568 	return ret;
4569 }
4570 
4571 /**
4572  * Split the flow to subflow set. The splitters might be linked
4573  * in the chain, like this:
4574  * flow_create_split_outer() calls:
4575  *   flow_create_split_meter() calls:
4576  *     flow_create_split_metadata(meter_subflow_0) calls:
4577  *       flow_create_split_inner(metadata_subflow_0)
4578  *       flow_create_split_inner(metadata_subflow_1)
4579  *       flow_create_split_inner(metadata_subflow_2)
4580  *     flow_create_split_metadata(meter_subflow_1) calls:
4581  *       flow_create_split_inner(metadata_subflow_0)
4582  *       flow_create_split_inner(metadata_subflow_1)
4583  *       flow_create_split_inner(metadata_subflow_2)
4584  *
4585  * This provide flexible way to add new levels of flow splitting.
4586  * The all of successfully created subflows are included to the
4587  * parent flow dev_flow list.
4588  *
4589  * @param dev
4590  *   Pointer to Ethernet device.
4591  * @param[in] flow
4592  *   Parent flow structure pointer.
4593  * @param[in] attr
4594  *   Flow rule attributes.
4595  * @param[in] items
4596  *   Pattern specification (list terminated by the END pattern item).
4597  * @param[in] actions
4598  *   Associated actions (list terminated by the END action).
4599  * @param[in] external
4600  *   This flow rule is created by request external to PMD.
4601  * @param[in] flow_idx
4602  *   This memory pool index to the flow.
4603  * @param[out] error
4604  *   Perform verbose error reporting if not NULL.
4605  * @return
4606  *   0 on success, negative value otherwise
4607  */
4608 static int
4609 flow_create_split_outer(struct rte_eth_dev *dev,
4610 			struct rte_flow *flow,
4611 			const struct rte_flow_attr *attr,
4612 			const struct rte_flow_item items[],
4613 			const struct rte_flow_action actions[],
4614 			bool external, uint32_t flow_idx,
4615 			struct rte_flow_error *error)
4616 {
4617 	int ret;
4618 
4619 	ret = flow_create_split_meter(dev, flow, attr, items,
4620 					 actions, external, flow_idx, error);
4621 	MLX5_ASSERT(ret <= 0);
4622 	return ret;
4623 }
4624 
4625 /**
4626  * Create a flow and add it to @p list.
4627  *
4628  * @param dev
4629  *   Pointer to Ethernet device.
4630  * @param list
4631  *   Pointer to a TAILQ flow list. If this parameter NULL,
4632  *   no list insertion occurred, flow is just created,
4633  *   this is caller's responsibility to track the
4634  *   created flow.
4635  * @param[in] attr
4636  *   Flow rule attributes.
4637  * @param[in] items
4638  *   Pattern specification (list terminated by the END pattern item).
4639  * @param[in] actions
4640  *   Associated actions (list terminated by the END action).
4641  * @param[in] external
4642  *   This flow rule is created by request external to PMD.
4643  * @param[out] error
4644  *   Perform verbose error reporting if not NULL.
4645  *
4646  * @return
4647  *   A flow index on success, 0 otherwise and rte_errno is set.
4648  */
4649 static uint32_t
4650 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
4651 		 const struct rte_flow_attr *attr,
4652 		 const struct rte_flow_item items[],
4653 		 const struct rte_flow_action actions[],
4654 		 bool external, struct rte_flow_error *error)
4655 {
4656 	struct mlx5_priv *priv = dev->data->dev_private;
4657 	struct rte_flow *flow = NULL;
4658 	struct mlx5_flow *dev_flow;
4659 	const struct rte_flow_action_rss *rss;
4660 	union {
4661 		struct mlx5_flow_expand_rss buf;
4662 		uint8_t buffer[2048];
4663 	} expand_buffer;
4664 	union {
4665 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4666 		uint8_t buffer[2048];
4667 	} actions_rx;
4668 	union {
4669 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4670 		uint8_t buffer[2048];
4671 	} actions_hairpin_tx;
4672 	union {
4673 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4674 		uint8_t buffer[2048];
4675 	} items_tx;
4676 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
4677 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4678 					      priv->rss_desc)[!!priv->flow_idx];
4679 	const struct rte_flow_action *p_actions_rx = actions;
4680 	uint32_t i;
4681 	uint32_t idx = 0;
4682 	int hairpin_flow;
4683 	uint32_t hairpin_id = 0;
4684 	struct rte_flow_attr attr_tx = { .priority = 0 };
4685 	int ret;
4686 
4687 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4688 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
4689 				external, hairpin_flow, error);
4690 	if (ret < 0)
4691 		return 0;
4692 	if (hairpin_flow > 0) {
4693 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4694 			rte_errno = EINVAL;
4695 			return 0;
4696 		}
4697 		flow_hairpin_split(dev, actions, actions_rx.actions,
4698 				   actions_hairpin_tx.actions, items_tx.items,
4699 				   &hairpin_id);
4700 		p_actions_rx = actions_rx.actions;
4701 	}
4702 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
4703 	if (!flow) {
4704 		rte_errno = ENOMEM;
4705 		goto error_before_flow;
4706 	}
4707 	flow->drv_type = flow_get_drv_type(dev, attr);
4708 	if (hairpin_id != 0)
4709 		flow->hairpin_flow_id = hairpin_id;
4710 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4711 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
4712 	memset(rss_desc, 0, sizeof(*rss_desc));
4713 	rss = flow_get_rss_action(p_actions_rx);
4714 	if (rss) {
4715 		/*
4716 		 * The following information is required by
4717 		 * mlx5_flow_hashfields_adjust() in advance.
4718 		 */
4719 		rss_desc->level = rss->level;
4720 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4721 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4722 	}
4723 	flow->dev_handles = 0;
4724 	if (rss && rss->types) {
4725 		unsigned int graph_root;
4726 
4727 		graph_root = find_graph_root(items, rss->level);
4728 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4729 					   items, rss->types,
4730 					   mlx5_support_expansion, graph_root);
4731 		MLX5_ASSERT(ret > 0 &&
4732 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
4733 	} else {
4734 		buf->entries = 1;
4735 		buf->entry[0].pattern = (void *)(uintptr_t)items;
4736 	}
4737 	/*
4738 	 * Record the start index when there is a nested call. All sub-flows
4739 	 * need to be translated before another calling.
4740 	 * No need to use ping-pong buffer to save memory here.
4741 	 */
4742 	if (priv->flow_idx) {
4743 		MLX5_ASSERT(!priv->flow_nested_idx);
4744 		priv->flow_nested_idx = priv->flow_idx;
4745 	}
4746 	for (i = 0; i < buf->entries; ++i) {
4747 		/*
4748 		 * The splitter may create multiple dev_flows,
4749 		 * depending on configuration. In the simplest
4750 		 * case it just creates unmodified original flow.
4751 		 */
4752 		ret = flow_create_split_outer(dev, flow, attr,
4753 					      buf->entry[i].pattern,
4754 					      p_actions_rx, external, idx,
4755 					      error);
4756 		if (ret < 0)
4757 			goto error;
4758 	}
4759 	/* Create the tx flow. */
4760 	if (hairpin_flow) {
4761 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4762 		attr_tx.ingress = 0;
4763 		attr_tx.egress = 1;
4764 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4765 					 actions_hairpin_tx.actions,
4766 					 idx, error);
4767 		if (!dev_flow)
4768 			goto error;
4769 		dev_flow->flow = flow;
4770 		dev_flow->external = 0;
4771 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4772 			      dev_flow->handle, next);
4773 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4774 					 items_tx.items,
4775 					 actions_hairpin_tx.actions, error);
4776 		if (ret < 0)
4777 			goto error;
4778 	}
4779 	/*
4780 	 * Update the metadata register copy table. If extensive
4781 	 * metadata feature is enabled and registers are supported
4782 	 * we might create the extra rte_flow for each unique
4783 	 * MARK/FLAG action ID.
4784 	 *
4785 	 * The table is updated for ingress Flows only, because
4786 	 * the egress Flows belong to the different device and
4787 	 * copy table should be updated in peer NIC Rx domain.
4788 	 */
4789 	if (attr->ingress &&
4790 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4791 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4792 		if (ret)
4793 			goto error;
4794 	}
4795 	/*
4796 	 * If the flow is external (from application) OR device is started, then
4797 	 * the flow will be applied immediately.
4798 	 */
4799 	if (external || dev->data->dev_started) {
4800 		ret = flow_drv_apply(dev, flow, error);
4801 		if (ret < 0)
4802 			goto error;
4803 	}
4804 	if (list)
4805 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
4806 			     flow, next);
4807 	flow_rxq_flags_set(dev, flow);
4808 	/* Nested flow creation index recovery. */
4809 	priv->flow_idx = priv->flow_nested_idx;
4810 	if (priv->flow_nested_idx)
4811 		priv->flow_nested_idx = 0;
4812 	return idx;
4813 error:
4814 	MLX5_ASSERT(flow);
4815 	ret = rte_errno; /* Save rte_errno before cleanup. */
4816 	flow_mreg_del_copy_action(dev, flow);
4817 	flow_drv_destroy(dev, flow);
4818 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
4819 	rte_errno = ret; /* Restore rte_errno. */
4820 error_before_flow:
4821 	ret = rte_errno;
4822 	if (hairpin_id)
4823 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4824 				     hairpin_id);
4825 	rte_errno = ret;
4826 	priv->flow_idx = priv->flow_nested_idx;
4827 	if (priv->flow_nested_idx)
4828 		priv->flow_nested_idx = 0;
4829 	return 0;
4830 }
4831 
4832 /**
4833  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4834  * incoming packets to table 1.
4835  *
4836  * Other flow rules, requested for group n, will be created in
4837  * e-switch table n+1.
4838  * Jump action to e-switch group n will be created to group n+1.
4839  *
4840  * Used when working in switchdev mode, to utilise advantages of table 1
4841  * and above.
4842  *
4843  * @param dev
4844  *   Pointer to Ethernet device.
4845  *
4846  * @return
4847  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
4848  */
4849 struct rte_flow *
4850 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4851 {
4852 	const struct rte_flow_attr attr = {
4853 		.group = 0,
4854 		.priority = 0,
4855 		.ingress = 1,
4856 		.egress = 0,
4857 		.transfer = 1,
4858 	};
4859 	const struct rte_flow_item pattern = {
4860 		.type = RTE_FLOW_ITEM_TYPE_END,
4861 	};
4862 	struct rte_flow_action_jump jump = {
4863 		.group = 1,
4864 	};
4865 	const struct rte_flow_action actions[] = {
4866 		{
4867 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4868 			.conf = &jump,
4869 		},
4870 		{
4871 			.type = RTE_FLOW_ACTION_TYPE_END,
4872 		},
4873 	};
4874 	struct mlx5_priv *priv = dev->data->dev_private;
4875 	struct rte_flow_error error;
4876 
4877 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
4878 						   &attr, &pattern,
4879 						   actions, false, &error);
4880 }
4881 
4882 /**
4883  * Validate a flow supported by the NIC.
4884  *
4885  * @see rte_flow_validate()
4886  * @see rte_flow_ops
4887  */
4888 int
4889 mlx5_flow_validate(struct rte_eth_dev *dev,
4890 		   const struct rte_flow_attr *attr,
4891 		   const struct rte_flow_item items[],
4892 		   const struct rte_flow_action actions[],
4893 		   struct rte_flow_error *error)
4894 {
4895 	int hairpin_flow;
4896 
4897 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4898 	return flow_drv_validate(dev, attr, items, actions,
4899 				true, hairpin_flow, error);
4900 }
4901 
4902 /**
4903  * Create a flow.
4904  *
4905  * @see rte_flow_create()
4906  * @see rte_flow_ops
4907  */
4908 struct rte_flow *
4909 mlx5_flow_create(struct rte_eth_dev *dev,
4910 		 const struct rte_flow_attr *attr,
4911 		 const struct rte_flow_item items[],
4912 		 const struct rte_flow_action actions[],
4913 		 struct rte_flow_error *error)
4914 {
4915 	struct mlx5_priv *priv = dev->data->dev_private;
4916 
4917 	/*
4918 	 * If the device is not started yet, it is not allowed to created a
4919 	 * flow from application. PMD default flows and traffic control flows
4920 	 * are not affected.
4921 	 */
4922 	if (unlikely(!dev->data->dev_started)) {
4923 		DRV_LOG(DEBUG, "port %u is not started when "
4924 			"inserting a flow", dev->data->port_id);
4925 		rte_flow_error_set(error, ENODEV,
4926 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4927 				   NULL,
4928 				   "port not started");
4929 		return NULL;
4930 	}
4931 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
4932 				  attr, items, actions, true, error);
4933 }
4934 
4935 /**
4936  * Destroy a flow in a list.
4937  *
4938  * @param dev
4939  *   Pointer to Ethernet device.
4940  * @param list
4941  *   Pointer to the Indexed flow list. If this parameter NULL,
4942  *   there is no flow removal from the list. Be noted that as
4943  *   flow is add to the indexed list, memory of the indexed
4944  *   list points to maybe changed as flow destroyed.
4945  * @param[in] flow_idx
4946  *   Index of flow to destroy.
4947  */
4948 static void
4949 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
4950 		  uint32_t flow_idx)
4951 {
4952 	struct mlx5_priv *priv = dev->data->dev_private;
4953 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4954 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
4955 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
4956 
4957 	if (!flow)
4958 		return;
4959 	/*
4960 	 * Update RX queue flags only if port is started, otherwise it is
4961 	 * already clean.
4962 	 */
4963 	if (dev->data->dev_started)
4964 		flow_rxq_flags_trim(dev, flow);
4965 	if (flow->hairpin_flow_id)
4966 		mlx5_flow_id_release(priv->sh->flow_id_pool,
4967 				     flow->hairpin_flow_id);
4968 	flow_drv_destroy(dev, flow);
4969 	if (list)
4970 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
4971 			     flow_idx, flow, next);
4972 	flow_mreg_del_copy_action(dev, flow);
4973 	if (flow->fdir) {
4974 		LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4975 			if (priv_fdir_flow->rix_flow == flow_idx)
4976 				break;
4977 		}
4978 		if (priv_fdir_flow) {
4979 			LIST_REMOVE(priv_fdir_flow, next);
4980 			mlx5_free(priv_fdir_flow->fdir);
4981 			mlx5_free(priv_fdir_flow);
4982 		}
4983 	}
4984 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
4985 }
4986 
4987 /**
4988  * Destroy all flows.
4989  *
4990  * @param dev
4991  *   Pointer to Ethernet device.
4992  * @param list
4993  *   Pointer to the Indexed flow list.
4994  * @param active
4995  *   If flushing is called avtively.
4996  */
4997 void
4998 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
4999 {
5000 	uint32_t num_flushed = 0;
5001 
5002 	while (*list) {
5003 		flow_list_destroy(dev, list, *list);
5004 		num_flushed++;
5005 	}
5006 	if (active) {
5007 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5008 			dev->data->port_id, num_flushed);
5009 	}
5010 }
5011 
5012 /**
5013  * Remove all flows.
5014  *
5015  * @param dev
5016  *   Pointer to Ethernet device.
5017  * @param list
5018  *   Pointer to the Indexed flow list.
5019  */
5020 void
5021 mlx5_flow_stop(struct rte_eth_dev *dev, uint32_t *list)
5022 {
5023 	struct mlx5_priv *priv = dev->data->dev_private;
5024 	struct rte_flow *flow = NULL;
5025 	uint32_t idx;
5026 
5027 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
5028 		      flow, next) {
5029 		flow_drv_remove(dev, flow);
5030 		flow_mreg_stop_copy_action(dev, flow);
5031 	}
5032 	flow_mreg_del_default_copy_action(dev);
5033 	flow_rxq_flags_clear(dev);
5034 }
5035 
5036 /**
5037  * Add all flows.
5038  *
5039  * @param dev
5040  *   Pointer to Ethernet device.
5041  * @param list
5042  *   Pointer to the Indexed flow list.
5043  *
5044  * @return
5045  *   0 on success, a negative errno value otherwise and rte_errno is set.
5046  */
5047 int
5048 mlx5_flow_start(struct rte_eth_dev *dev, uint32_t *list)
5049 {
5050 	struct mlx5_priv *priv = dev->data->dev_private;
5051 	struct rte_flow *flow = NULL;
5052 	struct rte_flow_error error;
5053 	uint32_t idx;
5054 	int ret = 0;
5055 
5056 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5057 	ret = flow_mreg_add_default_copy_action(dev, &error);
5058 	if (ret < 0)
5059 		return -rte_errno;
5060 	/* Apply Flows created by application. */
5061 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], *list, idx,
5062 		      flow, next) {
5063 		ret = flow_mreg_start_copy_action(dev, flow);
5064 		if (ret < 0)
5065 			goto error;
5066 		ret = flow_drv_apply(dev, flow, &error);
5067 		if (ret < 0)
5068 			goto error;
5069 		flow_rxq_flags_set(dev, flow);
5070 	}
5071 	return 0;
5072 error:
5073 	ret = rte_errno; /* Save rte_errno before cleanup. */
5074 	mlx5_flow_stop(dev, list);
5075 	rte_errno = ret; /* Restore rte_errno. */
5076 	return -rte_errno;
5077 }
5078 
5079 /**
5080  * Stop all default actions for flows.
5081  *
5082  * @param dev
5083  *   Pointer to Ethernet device.
5084  */
5085 void
5086 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5087 {
5088 	flow_mreg_del_default_copy_action(dev);
5089 	flow_rxq_flags_clear(dev);
5090 }
5091 
5092 /**
5093  * Start all default actions for flows.
5094  *
5095  * @param dev
5096  *   Pointer to Ethernet device.
5097  * @return
5098  *   0 on success, a negative errno value otherwise and rte_errno is set.
5099  */
5100 int
5101 mlx5_flow_start_default(struct rte_eth_dev *dev)
5102 {
5103 	struct rte_flow_error error;
5104 
5105 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5106 	return flow_mreg_add_default_copy_action(dev, &error);
5107 }
5108 
5109 /**
5110  * Allocate intermediate resources for flow creation.
5111  *
5112  * @param dev
5113  *   Pointer to Ethernet device.
5114  */
5115 void
5116 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
5117 {
5118 	struct mlx5_priv *priv = dev->data->dev_private;
5119 
5120 	if (!priv->inter_flows) {
5121 		priv->inter_flows = mlx5_malloc(MLX5_MEM_ZERO,
5122 				    MLX5_NUM_MAX_DEV_FLOWS *
5123 				    sizeof(struct mlx5_flow) +
5124 				    (sizeof(struct mlx5_flow_rss_desc) +
5125 				    sizeof(uint16_t) * UINT16_MAX) * 2, 0,
5126 				    SOCKET_ID_ANY);
5127 		if (!priv->inter_flows) {
5128 			DRV_LOG(ERR, "can't allocate intermediate memory.");
5129 			return;
5130 		}
5131 	}
5132 	priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
5133 			 [MLX5_NUM_MAX_DEV_FLOWS];
5134 	/* Reset the index. */
5135 	priv->flow_idx = 0;
5136 	priv->flow_nested_idx = 0;
5137 }
5138 
5139 /**
5140  * Free intermediate resources for flows.
5141  *
5142  * @param dev
5143  *   Pointer to Ethernet device.
5144  */
5145 void
5146 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
5147 {
5148 	struct mlx5_priv *priv = dev->data->dev_private;
5149 
5150 	mlx5_free(priv->inter_flows);
5151 	priv->inter_flows = NULL;
5152 }
5153 
5154 /**
5155  * Verify the flow list is empty
5156  *
5157  * @param dev
5158  *  Pointer to Ethernet device.
5159  *
5160  * @return the number of flows not released.
5161  */
5162 int
5163 mlx5_flow_verify(struct rte_eth_dev *dev)
5164 {
5165 	struct mlx5_priv *priv = dev->data->dev_private;
5166 	struct rte_flow *flow;
5167 	uint32_t idx;
5168 	int ret = 0;
5169 
5170 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
5171 		      flow, next) {
5172 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
5173 			dev->data->port_id, (void *)flow);
5174 		++ret;
5175 	}
5176 	return ret;
5177 }
5178 
5179 /**
5180  * Enable default hairpin egress flow.
5181  *
5182  * @param dev
5183  *   Pointer to Ethernet device.
5184  * @param queue
5185  *   The queue index.
5186  *
5187  * @return
5188  *   0 on success, a negative errno value otherwise and rte_errno is set.
5189  */
5190 int
5191 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
5192 			    uint32_t queue)
5193 {
5194 	struct mlx5_priv *priv = dev->data->dev_private;
5195 	const struct rte_flow_attr attr = {
5196 		.egress = 1,
5197 		.priority = 0,
5198 	};
5199 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
5200 		.queue = queue,
5201 	};
5202 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
5203 		.queue = UINT32_MAX,
5204 	};
5205 	struct rte_flow_item items[] = {
5206 		{
5207 			.type = (enum rte_flow_item_type)
5208 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
5209 			.spec = &queue_spec,
5210 			.last = NULL,
5211 			.mask = &queue_mask,
5212 		},
5213 		{
5214 			.type = RTE_FLOW_ITEM_TYPE_END,
5215 		},
5216 	};
5217 	struct rte_flow_action_jump jump = {
5218 		.group = MLX5_HAIRPIN_TX_TABLE,
5219 	};
5220 	struct rte_flow_action actions[2];
5221 	uint32_t flow_idx;
5222 	struct rte_flow_error error;
5223 
5224 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
5225 	actions[0].conf = &jump;
5226 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
5227 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5228 				&attr, items, actions, false, &error);
5229 	if (!flow_idx) {
5230 		DRV_LOG(DEBUG,
5231 			"Failed to create ctrl flow: rte_errno(%d),"
5232 			" type(%d), message(%s)",
5233 			rte_errno, error.type,
5234 			error.message ? error.message : " (no stated reason)");
5235 		return -rte_errno;
5236 	}
5237 	return 0;
5238 }
5239 
5240 /**
5241  * Enable a control flow configured from the control plane.
5242  *
5243  * @param dev
5244  *   Pointer to Ethernet device.
5245  * @param eth_spec
5246  *   An Ethernet flow spec to apply.
5247  * @param eth_mask
5248  *   An Ethernet flow mask to apply.
5249  * @param vlan_spec
5250  *   A VLAN flow spec to apply.
5251  * @param vlan_mask
5252  *   A VLAN flow mask to apply.
5253  *
5254  * @return
5255  *   0 on success, a negative errno value otherwise and rte_errno is set.
5256  */
5257 int
5258 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
5259 		    struct rte_flow_item_eth *eth_spec,
5260 		    struct rte_flow_item_eth *eth_mask,
5261 		    struct rte_flow_item_vlan *vlan_spec,
5262 		    struct rte_flow_item_vlan *vlan_mask)
5263 {
5264 	struct mlx5_priv *priv = dev->data->dev_private;
5265 	const struct rte_flow_attr attr = {
5266 		.ingress = 1,
5267 		.priority = MLX5_FLOW_PRIO_RSVD,
5268 	};
5269 	struct rte_flow_item items[] = {
5270 		{
5271 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5272 			.spec = eth_spec,
5273 			.last = NULL,
5274 			.mask = eth_mask,
5275 		},
5276 		{
5277 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
5278 					      RTE_FLOW_ITEM_TYPE_END,
5279 			.spec = vlan_spec,
5280 			.last = NULL,
5281 			.mask = vlan_mask,
5282 		},
5283 		{
5284 			.type = RTE_FLOW_ITEM_TYPE_END,
5285 		},
5286 	};
5287 	uint16_t queue[priv->reta_idx_n];
5288 	struct rte_flow_action_rss action_rss = {
5289 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
5290 		.level = 0,
5291 		.types = priv->rss_conf.rss_hf,
5292 		.key_len = priv->rss_conf.rss_key_len,
5293 		.queue_num = priv->reta_idx_n,
5294 		.key = priv->rss_conf.rss_key,
5295 		.queue = queue,
5296 	};
5297 	struct rte_flow_action actions[] = {
5298 		{
5299 			.type = RTE_FLOW_ACTION_TYPE_RSS,
5300 			.conf = &action_rss,
5301 		},
5302 		{
5303 			.type = RTE_FLOW_ACTION_TYPE_END,
5304 		},
5305 	};
5306 	uint32_t flow_idx;
5307 	struct rte_flow_error error;
5308 	unsigned int i;
5309 
5310 	if (!priv->reta_idx_n || !priv->rxqs_n) {
5311 		return 0;
5312 	}
5313 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
5314 		action_rss.types = 0;
5315 	for (i = 0; i != priv->reta_idx_n; ++i)
5316 		queue[i] = (*priv->reta_idx)[i];
5317 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5318 				&attr, items, actions, false, &error);
5319 	if (!flow_idx)
5320 		return -rte_errno;
5321 	return 0;
5322 }
5323 
5324 /**
5325  * Enable a flow control configured from the control plane.
5326  *
5327  * @param dev
5328  *   Pointer to Ethernet device.
5329  * @param eth_spec
5330  *   An Ethernet flow spec to apply.
5331  * @param eth_mask
5332  *   An Ethernet flow mask to apply.
5333  *
5334  * @return
5335  *   0 on success, a negative errno value otherwise and rte_errno is set.
5336  */
5337 int
5338 mlx5_ctrl_flow(struct rte_eth_dev *dev,
5339 	       struct rte_flow_item_eth *eth_spec,
5340 	       struct rte_flow_item_eth *eth_mask)
5341 {
5342 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
5343 }
5344 
5345 /**
5346  * Create default miss flow rule matching lacp traffic
5347  *
5348  * @param dev
5349  *   Pointer to Ethernet device.
5350  * @param eth_spec
5351  *   An Ethernet flow spec to apply.
5352  *
5353  * @return
5354  *   0 on success, a negative errno value otherwise and rte_errno is set.
5355  */
5356 int
5357 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
5358 {
5359 	struct mlx5_priv *priv = dev->data->dev_private;
5360 	/*
5361 	 * The LACP matching is done by only using ether type since using
5362 	 * a multicast dst mac causes kernel to give low priority to this flow.
5363 	 */
5364 	static const struct rte_flow_item_eth lacp_spec = {
5365 		.type = RTE_BE16(0x8809),
5366 	};
5367 	static const struct rte_flow_item_eth lacp_mask = {
5368 		.type = 0xffff,
5369 	};
5370 	const struct rte_flow_attr attr = {
5371 		.ingress = 1,
5372 	};
5373 	struct rte_flow_item items[] = {
5374 		{
5375 			.type = RTE_FLOW_ITEM_TYPE_ETH,
5376 			.spec = &lacp_spec,
5377 			.mask = &lacp_mask,
5378 		},
5379 		{
5380 			.type = RTE_FLOW_ITEM_TYPE_END,
5381 		},
5382 	};
5383 	struct rte_flow_action actions[] = {
5384 		{
5385 			.type = (enum rte_flow_action_type)
5386 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
5387 		},
5388 		{
5389 			.type = RTE_FLOW_ACTION_TYPE_END,
5390 		},
5391 	};
5392 	struct rte_flow_error error;
5393 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
5394 				&attr, items, actions, false, &error);
5395 
5396 	if (!flow_idx)
5397 		return -rte_errno;
5398 	return 0;
5399 }
5400 
5401 /**
5402  * Destroy a flow.
5403  *
5404  * @see rte_flow_destroy()
5405  * @see rte_flow_ops
5406  */
5407 int
5408 mlx5_flow_destroy(struct rte_eth_dev *dev,
5409 		  struct rte_flow *flow,
5410 		  struct rte_flow_error *error __rte_unused)
5411 {
5412 	struct mlx5_priv *priv = dev->data->dev_private;
5413 
5414 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
5415 	return 0;
5416 }
5417 
5418 /**
5419  * Destroy all flows.
5420  *
5421  * @see rte_flow_flush()
5422  * @see rte_flow_ops
5423  */
5424 int
5425 mlx5_flow_flush(struct rte_eth_dev *dev,
5426 		struct rte_flow_error *error __rte_unused)
5427 {
5428 	struct mlx5_priv *priv = dev->data->dev_private;
5429 
5430 	mlx5_flow_list_flush(dev, &priv->flows, false);
5431 	return 0;
5432 }
5433 
5434 /**
5435  * Isolated mode.
5436  *
5437  * @see rte_flow_isolate()
5438  * @see rte_flow_ops
5439  */
5440 int
5441 mlx5_flow_isolate(struct rte_eth_dev *dev,
5442 		  int enable,
5443 		  struct rte_flow_error *error)
5444 {
5445 	struct mlx5_priv *priv = dev->data->dev_private;
5446 
5447 	if (dev->data->dev_started) {
5448 		rte_flow_error_set(error, EBUSY,
5449 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5450 				   NULL,
5451 				   "port must be stopped first");
5452 		return -rte_errno;
5453 	}
5454 	priv->isolated = !!enable;
5455 	if (enable)
5456 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
5457 	else
5458 		dev->dev_ops = &mlx5_os_dev_ops;
5459 
5460 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
5461 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
5462 
5463 	return 0;
5464 }
5465 
5466 /**
5467  * Query a flow.
5468  *
5469  * @see rte_flow_query()
5470  * @see rte_flow_ops
5471  */
5472 static int
5473 flow_drv_query(struct rte_eth_dev *dev,
5474 	       uint32_t flow_idx,
5475 	       const struct rte_flow_action *actions,
5476 	       void *data,
5477 	       struct rte_flow_error *error)
5478 {
5479 	struct mlx5_priv *priv = dev->data->dev_private;
5480 	const struct mlx5_flow_driver_ops *fops;
5481 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5482 					       [MLX5_IPOOL_RTE_FLOW],
5483 					       flow_idx);
5484 	enum mlx5_flow_drv_type ftype;
5485 
5486 	if (!flow) {
5487 		return rte_flow_error_set(error, ENOENT,
5488 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5489 			  NULL,
5490 			  "invalid flow handle");
5491 	}
5492 	ftype = flow->drv_type;
5493 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
5494 	fops = flow_get_drv_ops(ftype);
5495 
5496 	return fops->query(dev, flow, actions, data, error);
5497 }
5498 
5499 /**
5500  * Query a flow.
5501  *
5502  * @see rte_flow_query()
5503  * @see rte_flow_ops
5504  */
5505 int
5506 mlx5_flow_query(struct rte_eth_dev *dev,
5507 		struct rte_flow *flow,
5508 		const struct rte_flow_action *actions,
5509 		void *data,
5510 		struct rte_flow_error *error)
5511 {
5512 	int ret;
5513 
5514 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
5515 			     error);
5516 	if (ret < 0)
5517 		return ret;
5518 	return 0;
5519 }
5520 
5521 /**
5522  * Convert a flow director filter to a generic flow.
5523  *
5524  * @param dev
5525  *   Pointer to Ethernet device.
5526  * @param fdir_filter
5527  *   Flow director filter to add.
5528  * @param attributes
5529  *   Generic flow parameters structure.
5530  *
5531  * @return
5532  *   0 on success, a negative errno value otherwise and rte_errno is set.
5533  */
5534 static int
5535 flow_fdir_filter_convert(struct rte_eth_dev *dev,
5536 			 const struct rte_eth_fdir_filter *fdir_filter,
5537 			 struct mlx5_fdir *attributes)
5538 {
5539 	struct mlx5_priv *priv = dev->data->dev_private;
5540 	const struct rte_eth_fdir_input *input = &fdir_filter->input;
5541 	const struct rte_eth_fdir_masks *mask =
5542 		&dev->data->dev_conf.fdir_conf.mask;
5543 
5544 	/* Validate queue number. */
5545 	if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5546 		DRV_LOG(ERR, "port %u invalid queue number %d",
5547 			dev->data->port_id, fdir_filter->action.rx_queue);
5548 		rte_errno = EINVAL;
5549 		return -rte_errno;
5550 	}
5551 	attributes->attr.ingress = 1;
5552 	attributes->items[0] = (struct rte_flow_item) {
5553 		.type = RTE_FLOW_ITEM_TYPE_ETH,
5554 		.spec = &attributes->l2,
5555 		.mask = &attributes->l2_mask,
5556 	};
5557 	switch (fdir_filter->action.behavior) {
5558 	case RTE_ETH_FDIR_ACCEPT:
5559 		attributes->actions[0] = (struct rte_flow_action){
5560 			.type = RTE_FLOW_ACTION_TYPE_QUEUE,
5561 			.conf = &attributes->queue,
5562 		};
5563 		break;
5564 	case RTE_ETH_FDIR_REJECT:
5565 		attributes->actions[0] = (struct rte_flow_action){
5566 			.type = RTE_FLOW_ACTION_TYPE_DROP,
5567 		};
5568 		break;
5569 	default:
5570 		DRV_LOG(ERR, "port %u invalid behavior %d",
5571 			dev->data->port_id,
5572 			fdir_filter->action.behavior);
5573 		rte_errno = ENOTSUP;
5574 		return -rte_errno;
5575 	}
5576 	attributes->queue.index = fdir_filter->action.rx_queue;
5577 	/* Handle L3. */
5578 	switch (fdir_filter->input.flow_type) {
5579 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5580 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5581 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5582 		attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5583 			.src_addr = input->flow.ip4_flow.src_ip,
5584 			.dst_addr = input->flow.ip4_flow.dst_ip,
5585 			.time_to_live = input->flow.ip4_flow.ttl,
5586 			.type_of_service = input->flow.ip4_flow.tos,
5587 		};
5588 		attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5589 			.src_addr = mask->ipv4_mask.src_ip,
5590 			.dst_addr = mask->ipv4_mask.dst_ip,
5591 			.time_to_live = mask->ipv4_mask.ttl,
5592 			.type_of_service = mask->ipv4_mask.tos,
5593 			.next_proto_id = mask->ipv4_mask.proto,
5594 		};
5595 		attributes->items[1] = (struct rte_flow_item){
5596 			.type = RTE_FLOW_ITEM_TYPE_IPV4,
5597 			.spec = &attributes->l3,
5598 			.mask = &attributes->l3_mask,
5599 		};
5600 		break;
5601 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5602 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5603 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5604 		attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5605 			.hop_limits = input->flow.ipv6_flow.hop_limits,
5606 			.proto = input->flow.ipv6_flow.proto,
5607 		};
5608 
5609 		memcpy(attributes->l3.ipv6.hdr.src_addr,
5610 		       input->flow.ipv6_flow.src_ip,
5611 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5612 		memcpy(attributes->l3.ipv6.hdr.dst_addr,
5613 		       input->flow.ipv6_flow.dst_ip,
5614 		       RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5615 		memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5616 		       mask->ipv6_mask.src_ip,
5617 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5618 		memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5619 		       mask->ipv6_mask.dst_ip,
5620 		       RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5621 		attributes->items[1] = (struct rte_flow_item){
5622 			.type = RTE_FLOW_ITEM_TYPE_IPV6,
5623 			.spec = &attributes->l3,
5624 			.mask = &attributes->l3_mask,
5625 		};
5626 		break;
5627 	default:
5628 		DRV_LOG(ERR, "port %u invalid flow type%d",
5629 			dev->data->port_id, fdir_filter->input.flow_type);
5630 		rte_errno = ENOTSUP;
5631 		return -rte_errno;
5632 	}
5633 	/* Handle L4. */
5634 	switch (fdir_filter->input.flow_type) {
5635 	case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5636 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5637 			.src_port = input->flow.udp4_flow.src_port,
5638 			.dst_port = input->flow.udp4_flow.dst_port,
5639 		};
5640 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5641 			.src_port = mask->src_port_mask,
5642 			.dst_port = mask->dst_port_mask,
5643 		};
5644 		attributes->items[2] = (struct rte_flow_item){
5645 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5646 			.spec = &attributes->l4,
5647 			.mask = &attributes->l4_mask,
5648 		};
5649 		break;
5650 	case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5651 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5652 			.src_port = input->flow.tcp4_flow.src_port,
5653 			.dst_port = input->flow.tcp4_flow.dst_port,
5654 		};
5655 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5656 			.src_port = mask->src_port_mask,
5657 			.dst_port = mask->dst_port_mask,
5658 		};
5659 		attributes->items[2] = (struct rte_flow_item){
5660 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5661 			.spec = &attributes->l4,
5662 			.mask = &attributes->l4_mask,
5663 		};
5664 		break;
5665 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5666 		attributes->l4.udp.hdr = (struct rte_udp_hdr){
5667 			.src_port = input->flow.udp6_flow.src_port,
5668 			.dst_port = input->flow.udp6_flow.dst_port,
5669 		};
5670 		attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5671 			.src_port = mask->src_port_mask,
5672 			.dst_port = mask->dst_port_mask,
5673 		};
5674 		attributes->items[2] = (struct rte_flow_item){
5675 			.type = RTE_FLOW_ITEM_TYPE_UDP,
5676 			.spec = &attributes->l4,
5677 			.mask = &attributes->l4_mask,
5678 		};
5679 		break;
5680 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5681 		attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5682 			.src_port = input->flow.tcp6_flow.src_port,
5683 			.dst_port = input->flow.tcp6_flow.dst_port,
5684 		};
5685 		attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5686 			.src_port = mask->src_port_mask,
5687 			.dst_port = mask->dst_port_mask,
5688 		};
5689 		attributes->items[2] = (struct rte_flow_item){
5690 			.type = RTE_FLOW_ITEM_TYPE_TCP,
5691 			.spec = &attributes->l4,
5692 			.mask = &attributes->l4_mask,
5693 		};
5694 		break;
5695 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5696 	case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5697 		break;
5698 	default:
5699 		DRV_LOG(ERR, "port %u invalid flow type%d",
5700 			dev->data->port_id, fdir_filter->input.flow_type);
5701 		rte_errno = ENOTSUP;
5702 		return -rte_errno;
5703 	}
5704 	return 0;
5705 }
5706 
5707 #define FLOW_FDIR_CMP(f1, f2, fld) \
5708 	memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5709 
5710 /**
5711  * Compare two FDIR flows. If items and actions are identical, the two flows are
5712  * regarded as same.
5713  *
5714  * @param dev
5715  *   Pointer to Ethernet device.
5716  * @param f1
5717  *   FDIR flow to compare.
5718  * @param f2
5719  *   FDIR flow to compare.
5720  *
5721  * @return
5722  *   Zero on match, 1 otherwise.
5723  */
5724 static int
5725 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5726 {
5727 	if (FLOW_FDIR_CMP(f1, f2, attr) ||
5728 	    FLOW_FDIR_CMP(f1, f2, l2) ||
5729 	    FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5730 	    FLOW_FDIR_CMP(f1, f2, l3) ||
5731 	    FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5732 	    FLOW_FDIR_CMP(f1, f2, l4) ||
5733 	    FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5734 	    FLOW_FDIR_CMP(f1, f2, actions[0].type))
5735 		return 1;
5736 	if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5737 	    FLOW_FDIR_CMP(f1, f2, queue))
5738 		return 1;
5739 	return 0;
5740 }
5741 
5742 /**
5743  * Search device flow list to find out a matched FDIR flow.
5744  *
5745  * @param dev
5746  *   Pointer to Ethernet device.
5747  * @param fdir_flow
5748  *   FDIR flow to lookup.
5749  *
5750  * @return
5751  *   Index of flow if found, 0 otherwise.
5752  */
5753 static uint32_t
5754 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5755 {
5756 	struct mlx5_priv *priv = dev->data->dev_private;
5757 	uint32_t flow_idx = 0;
5758 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5759 
5760 	MLX5_ASSERT(fdir_flow);
5761 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5762 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5763 			DRV_LOG(DEBUG, "port %u found FDIR flow %u",
5764 				dev->data->port_id, flow_idx);
5765 			flow_idx = priv_fdir_flow->rix_flow;
5766 			break;
5767 		}
5768 	}
5769 	return flow_idx;
5770 }
5771 
5772 /**
5773  * Add new flow director filter and store it in list.
5774  *
5775  * @param dev
5776  *   Pointer to Ethernet device.
5777  * @param fdir_filter
5778  *   Flow director filter to add.
5779  *
5780  * @return
5781  *   0 on success, a negative errno value otherwise and rte_errno is set.
5782  */
5783 static int
5784 flow_fdir_filter_add(struct rte_eth_dev *dev,
5785 		     const struct rte_eth_fdir_filter *fdir_filter)
5786 {
5787 	struct mlx5_priv *priv = dev->data->dev_private;
5788 	struct mlx5_fdir *fdir_flow;
5789 	struct rte_flow *flow;
5790 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5791 	uint32_t flow_idx;
5792 	int ret;
5793 
5794 	fdir_flow = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*fdir_flow), 0,
5795 				SOCKET_ID_ANY);
5796 	if (!fdir_flow) {
5797 		rte_errno = ENOMEM;
5798 		return -rte_errno;
5799 	}
5800 	ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5801 	if (ret)
5802 		goto error;
5803 	flow_idx = flow_fdir_filter_lookup(dev, fdir_flow);
5804 	if (flow_idx) {
5805 		rte_errno = EEXIST;
5806 		goto error;
5807 	}
5808 	priv_fdir_flow = mlx5_malloc(MLX5_MEM_ZERO,
5809 				     sizeof(struct mlx5_fdir_flow),
5810 				     0, SOCKET_ID_ANY);
5811 	if (!priv_fdir_flow) {
5812 		rte_errno = ENOMEM;
5813 		goto error;
5814 	}
5815 	flow_idx = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5816 				    fdir_flow->items, fdir_flow->actions, true,
5817 				    NULL);
5818 	flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5819 	if (!flow)
5820 		goto error;
5821 	flow->fdir = 1;
5822 	priv_fdir_flow->fdir = fdir_flow;
5823 	priv_fdir_flow->rix_flow = flow_idx;
5824 	LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5825 	DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5826 		dev->data->port_id, (void *)flow);
5827 	return 0;
5828 error:
5829 	mlx5_free(priv_fdir_flow);
5830 	mlx5_free(fdir_flow);
5831 	return -rte_errno;
5832 }
5833 
5834 /**
5835  * Delete specific filter.
5836  *
5837  * @param dev
5838  *   Pointer to Ethernet device.
5839  * @param fdir_filter
5840  *   Filter to be deleted.
5841  *
5842  * @return
5843  *   0 on success, a negative errno value otherwise and rte_errno is set.
5844  */
5845 static int
5846 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5847 			const struct rte_eth_fdir_filter *fdir_filter)
5848 {
5849 	struct mlx5_priv *priv = dev->data->dev_private;
5850 	uint32_t flow_idx;
5851 	struct mlx5_fdir fdir_flow = {
5852 		.attr.group = 0,
5853 	};
5854 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5855 	int ret;
5856 
5857 	ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5858 	if (ret)
5859 		return -rte_errno;
5860 	LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5861 		/* Find the fdir in priv list */
5862 		if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5863 			break;
5864 	}
5865 	if (!priv_fdir_flow)
5866 		return 0;
5867 	LIST_REMOVE(priv_fdir_flow, next);
5868 	flow_idx = priv_fdir_flow->rix_flow;
5869 	flow_list_destroy(dev, &priv->flows, flow_idx);
5870 	mlx5_free(priv_fdir_flow->fdir);
5871 	mlx5_free(priv_fdir_flow);
5872 	DRV_LOG(DEBUG, "port %u deleted FDIR flow %u",
5873 		dev->data->port_id, flow_idx);
5874 	return 0;
5875 }
5876 
5877 /**
5878  * Update queue for specific filter.
5879  *
5880  * @param dev
5881  *   Pointer to Ethernet device.
5882  * @param fdir_filter
5883  *   Filter to be updated.
5884  *
5885  * @return
5886  *   0 on success, a negative errno value otherwise and rte_errno is set.
5887  */
5888 static int
5889 flow_fdir_filter_update(struct rte_eth_dev *dev,
5890 			const struct rte_eth_fdir_filter *fdir_filter)
5891 {
5892 	int ret;
5893 
5894 	ret = flow_fdir_filter_delete(dev, fdir_filter);
5895 	if (ret)
5896 		return ret;
5897 	return flow_fdir_filter_add(dev, fdir_filter);
5898 }
5899 
5900 /**
5901  * Flush all filters.
5902  *
5903  * @param dev
5904  *   Pointer to Ethernet device.
5905  */
5906 static void
5907 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5908 {
5909 	struct mlx5_priv *priv = dev->data->dev_private;
5910 	struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5911 
5912 	while (!LIST_EMPTY(&priv->fdir_flows)) {
5913 		priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5914 		LIST_REMOVE(priv_fdir_flow, next);
5915 		flow_list_destroy(dev, &priv->flows, priv_fdir_flow->rix_flow);
5916 		mlx5_free(priv_fdir_flow->fdir);
5917 		mlx5_free(priv_fdir_flow);
5918 	}
5919 }
5920 
5921 /**
5922  * Get flow director information.
5923  *
5924  * @param dev
5925  *   Pointer to Ethernet device.
5926  * @param[out] fdir_info
5927  *   Resulting flow director information.
5928  */
5929 static void
5930 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5931 {
5932 	struct rte_eth_fdir_masks *mask =
5933 		&dev->data->dev_conf.fdir_conf.mask;
5934 
5935 	fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5936 	fdir_info->guarant_spc = 0;
5937 	rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5938 	fdir_info->max_flexpayload = 0;
5939 	fdir_info->flow_types_mask[0] = 0;
5940 	fdir_info->flex_payload_unit = 0;
5941 	fdir_info->max_flex_payload_segment_num = 0;
5942 	fdir_info->flex_payload_limit = 0;
5943 	memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5944 }
5945 
5946 /**
5947  * Deal with flow director operations.
5948  *
5949  * @param dev
5950  *   Pointer to Ethernet device.
5951  * @param filter_op
5952  *   Operation to perform.
5953  * @param arg
5954  *   Pointer to operation-specific structure.
5955  *
5956  * @return
5957  *   0 on success, a negative errno value otherwise and rte_errno is set.
5958  */
5959 static int
5960 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5961 		    void *arg)
5962 {
5963 	enum rte_fdir_mode fdir_mode =
5964 		dev->data->dev_conf.fdir_conf.mode;
5965 
5966 	if (filter_op == RTE_ETH_FILTER_NOP)
5967 		return 0;
5968 	if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5969 	    fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5970 		DRV_LOG(ERR, "port %u flow director mode %d not supported",
5971 			dev->data->port_id, fdir_mode);
5972 		rte_errno = EINVAL;
5973 		return -rte_errno;
5974 	}
5975 	switch (filter_op) {
5976 	case RTE_ETH_FILTER_ADD:
5977 		return flow_fdir_filter_add(dev, arg);
5978 	case RTE_ETH_FILTER_UPDATE:
5979 		return flow_fdir_filter_update(dev, arg);
5980 	case RTE_ETH_FILTER_DELETE:
5981 		return flow_fdir_filter_delete(dev, arg);
5982 	case RTE_ETH_FILTER_FLUSH:
5983 		flow_fdir_filter_flush(dev);
5984 		break;
5985 	case RTE_ETH_FILTER_INFO:
5986 		flow_fdir_info_get(dev, arg);
5987 		break;
5988 	default:
5989 		DRV_LOG(DEBUG, "port %u unknown operation %u",
5990 			dev->data->port_id, filter_op);
5991 		rte_errno = EINVAL;
5992 		return -rte_errno;
5993 	}
5994 	return 0;
5995 }
5996 
5997 /**
5998  * Manage filter operations.
5999  *
6000  * @param dev
6001  *   Pointer to Ethernet device structure.
6002  * @param filter_type
6003  *   Filter type.
6004  * @param filter_op
6005  *   Operation to perform.
6006  * @param arg
6007  *   Pointer to operation-specific structure.
6008  *
6009  * @return
6010  *   0 on success, a negative errno value otherwise and rte_errno is set.
6011  */
6012 int
6013 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6014 		     enum rte_filter_type filter_type,
6015 		     enum rte_filter_op filter_op,
6016 		     void *arg)
6017 {
6018 	switch (filter_type) {
6019 	case RTE_ETH_FILTER_GENERIC:
6020 		if (filter_op != RTE_ETH_FILTER_GET) {
6021 			rte_errno = EINVAL;
6022 			return -rte_errno;
6023 		}
6024 		*(const void **)arg = &mlx5_flow_ops;
6025 		return 0;
6026 	case RTE_ETH_FILTER_FDIR:
6027 		return flow_fdir_ctrl_func(dev, filter_op, arg);
6028 	default:
6029 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6030 			dev->data->port_id, filter_type);
6031 		rte_errno = ENOTSUP;
6032 		return -rte_errno;
6033 	}
6034 	return 0;
6035 }
6036 
6037 /**
6038  * Create the needed meter and suffix tables.
6039  *
6040  * @param[in] dev
6041  *   Pointer to Ethernet device.
6042  * @param[in] fm
6043  *   Pointer to the flow meter.
6044  *
6045  * @return
6046  *   Pointer to table set on success, NULL otherwise.
6047  */
6048 struct mlx5_meter_domains_infos *
6049 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6050 			  const struct mlx5_flow_meter *fm)
6051 {
6052 	const struct mlx5_flow_driver_ops *fops;
6053 
6054 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6055 	return fops->create_mtr_tbls(dev, fm);
6056 }
6057 
6058 /**
6059  * Destroy the meter table set.
6060  *
6061  * @param[in] dev
6062  *   Pointer to Ethernet device.
6063  * @param[in] tbl
6064  *   Pointer to the meter table set.
6065  *
6066  * @return
6067  *   0 on success.
6068  */
6069 int
6070 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6071 			   struct mlx5_meter_domains_infos *tbls)
6072 {
6073 	const struct mlx5_flow_driver_ops *fops;
6074 
6075 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6076 	return fops->destroy_mtr_tbls(dev, tbls);
6077 }
6078 
6079 /**
6080  * Create policer rules.
6081  *
6082  * @param[in] dev
6083  *   Pointer to Ethernet device.
6084  * @param[in] fm
6085  *   Pointer to flow meter structure.
6086  * @param[in] attr
6087  *   Pointer to flow attributes.
6088  *
6089  * @return
6090  *   0 on success, -1 otherwise.
6091  */
6092 int
6093 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6094 			       struct mlx5_flow_meter *fm,
6095 			       const struct rte_flow_attr *attr)
6096 {
6097 	const struct mlx5_flow_driver_ops *fops;
6098 
6099 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6100 	return fops->create_policer_rules(dev, fm, attr);
6101 }
6102 
6103 /**
6104  * Destroy policer rules.
6105  *
6106  * @param[in] fm
6107  *   Pointer to flow meter structure.
6108  * @param[in] attr
6109  *   Pointer to flow attributes.
6110  *
6111  * @return
6112  *   0 on success, -1 otherwise.
6113  */
6114 int
6115 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6116 				struct mlx5_flow_meter *fm,
6117 				const struct rte_flow_attr *attr)
6118 {
6119 	const struct mlx5_flow_driver_ops *fops;
6120 
6121 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6122 	return fops->destroy_policer_rules(dev, fm, attr);
6123 }
6124 
6125 /**
6126  * Allocate a counter.
6127  *
6128  * @param[in] dev
6129  *   Pointer to Ethernet device structure.
6130  *
6131  * @return
6132  *   Index to allocated counter  on success, 0 otherwise.
6133  */
6134 uint32_t
6135 mlx5_counter_alloc(struct rte_eth_dev *dev)
6136 {
6137 	const struct mlx5_flow_driver_ops *fops;
6138 	struct rte_flow_attr attr = { .transfer = 0 };
6139 
6140 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6141 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6142 		return fops->counter_alloc(dev);
6143 	}
6144 	DRV_LOG(ERR,
6145 		"port %u counter allocate is not supported.",
6146 		 dev->data->port_id);
6147 	return 0;
6148 }
6149 
6150 /**
6151  * Free a counter.
6152  *
6153  * @param[in] dev
6154  *   Pointer to Ethernet device structure.
6155  * @param[in] cnt
6156  *   Index to counter to be free.
6157  */
6158 void
6159 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6160 {
6161 	const struct mlx5_flow_driver_ops *fops;
6162 	struct rte_flow_attr attr = { .transfer = 0 };
6163 
6164 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6165 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6166 		fops->counter_free(dev, cnt);
6167 		return;
6168 	}
6169 	DRV_LOG(ERR,
6170 		"port %u counter free is not supported.",
6171 		 dev->data->port_id);
6172 }
6173 
6174 /**
6175  * Query counter statistics.
6176  *
6177  * @param[in] dev
6178  *   Pointer to Ethernet device structure.
6179  * @param[in] cnt
6180  *   Index to counter to query.
6181  * @param[in] clear
6182  *   Set to clear counter statistics.
6183  * @param[out] pkts
6184  *   The counter hits packets number to save.
6185  * @param[out] bytes
6186  *   The counter hits bytes number to save.
6187  *
6188  * @return
6189  *   0 on success, a negative errno value otherwise.
6190  */
6191 int
6192 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6193 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6194 {
6195 	const struct mlx5_flow_driver_ops *fops;
6196 	struct rte_flow_attr attr = { .transfer = 0 };
6197 
6198 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6199 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6200 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6201 	}
6202 	DRV_LOG(ERR,
6203 		"port %u counter query is not supported.",
6204 		 dev->data->port_id);
6205 	return -ENOTSUP;
6206 }
6207 
6208 #define MLX5_POOL_QUERY_FREQ_US 1000000
6209 
6210 /**
6211  * Get number of all validate pools.
6212  *
6213  * @param[in] sh
6214  *   Pointer to mlx5_dev_ctx_shared object.
6215  *
6216  * @return
6217  *   The number of all validate pools.
6218  */
6219 static uint32_t
6220 mlx5_get_all_valid_pool_count(struct mlx5_dev_ctx_shared *sh)
6221 {
6222 	int i;
6223 	uint32_t pools_n = 0;
6224 
6225 	for (i = 0; i < MLX5_CCONT_TYPE_MAX; ++i)
6226 		pools_n += rte_atomic16_read(&sh->cmng.ccont[i].n_valid);
6227 	return pools_n;
6228 }
6229 
6230 /**
6231  * Set the periodic procedure for triggering asynchronous batch queries for all
6232  * the counter pools.
6233  *
6234  * @param[in] sh
6235  *   Pointer to mlx5_dev_ctx_shared object.
6236  */
6237 void
6238 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6239 {
6240 	uint32_t pools_n, us;
6241 
6242 	pools_n = mlx5_get_all_valid_pool_count(sh);
6243 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6244 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6245 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6246 		sh->cmng.query_thread_on = 0;
6247 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6248 	} else {
6249 		sh->cmng.query_thread_on = 1;
6250 	}
6251 }
6252 
6253 /**
6254  * The periodic procedure for triggering asynchronous batch queries for all the
6255  * counter pools. This function is probably called by the host thread.
6256  *
6257  * @param[in] arg
6258  *   The parameter for the alarm process.
6259  */
6260 void
6261 mlx5_flow_query_alarm(void *arg)
6262 {
6263 	struct mlx5_dev_ctx_shared *sh = arg;
6264 	struct mlx5_devx_obj *dcs;
6265 	uint16_t offset;
6266 	int ret;
6267 	uint8_t batch = sh->cmng.batch;
6268 	uint8_t age = sh->cmng.age;
6269 	uint16_t pool_index = sh->cmng.pool_index;
6270 	struct mlx5_pools_container *cont;
6271 	struct mlx5_flow_counter_pool *pool;
6272 	int cont_loop = MLX5_CCONT_TYPE_MAX;
6273 
6274 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6275 		goto set_alarm;
6276 next_container:
6277 	cont = MLX5_CNT_CONTAINER(sh, batch, age);
6278 	rte_spinlock_lock(&cont->resize_sl);
6279 	if (!cont->pools) {
6280 		rte_spinlock_unlock(&cont->resize_sl);
6281 		/* Check if all the containers are empty. */
6282 		if (unlikely(--cont_loop == 0))
6283 			goto set_alarm;
6284 		batch ^= 0x1;
6285 		pool_index = 0;
6286 		if (batch == 0 && pool_index == 0) {
6287 			age ^= 0x1;
6288 			sh->cmng.batch = batch;
6289 			sh->cmng.age = age;
6290 		}
6291 		goto next_container;
6292 	}
6293 	pool = cont->pools[pool_index];
6294 	rte_spinlock_unlock(&cont->resize_sl);
6295 	if (pool->raw_hw)
6296 		/* There is a pool query in progress. */
6297 		goto set_alarm;
6298 	pool->raw_hw =
6299 		LIST_FIRST(&sh->cmng.free_stat_raws);
6300 	if (!pool->raw_hw)
6301 		/* No free counter statistics raw memory. */
6302 		goto set_alarm;
6303 	dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
6304 							      (&pool->a64_dcs);
6305 	if (dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) {
6306 		/* Pool without valid counter. */
6307 		pool->raw_hw = NULL;
6308 		goto next_pool;
6309 	}
6310 	offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
6311 	/*
6312 	 * Identify the counters released between query trigger and query
6313 	 * handle more effiecntly. The counter released in this gap period
6314 	 * should wait for a new round of query as the new arrived packets
6315 	 * will not be taken into account.
6316 	 */
6317 	pool->query_gen++;
6318 	ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
6319 					       offset, NULL, NULL,
6320 					       pool->raw_hw->mem_mng->dm->id,
6321 					       (void *)(uintptr_t)
6322 					       (pool->raw_hw->data + offset),
6323 					       sh->devx_comp,
6324 					       (uint64_t)(uintptr_t)pool);
6325 	if (ret) {
6326 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6327 			" %d", pool->min_dcs->id);
6328 		pool->raw_hw = NULL;
6329 		goto set_alarm;
6330 	}
6331 	pool->raw_hw->min_dcs_id = dcs->id;
6332 	LIST_REMOVE(pool->raw_hw, next);
6333 	sh->cmng.pending_queries++;
6334 next_pool:
6335 	pool_index++;
6336 	if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
6337 		batch ^= 0x1;
6338 		pool_index = 0;
6339 		if (batch == 0 && pool_index == 0)
6340 			age ^= 0x1;
6341 	}
6342 set_alarm:
6343 	sh->cmng.batch = batch;
6344 	sh->cmng.pool_index = pool_index;
6345 	sh->cmng.age = age;
6346 	mlx5_set_query_alarm(sh);
6347 }
6348 
6349 /**
6350  * Check and callback event for new aged flow in the counter pool
6351  *
6352  * @param[in] sh
6353  *   Pointer to mlx5_dev_ctx_shared object.
6354  * @param[in] pool
6355  *   Pointer to Current counter pool.
6356  */
6357 static void
6358 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6359 		   struct mlx5_flow_counter_pool *pool)
6360 {
6361 	struct mlx5_priv *priv;
6362 	struct mlx5_flow_counter *cnt;
6363 	struct mlx5_age_info *age_info;
6364 	struct mlx5_age_param *age_param;
6365 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6366 	struct mlx5_counter_stats_raw *prev = pool->raw;
6367 	uint16_t curr = rte_rdtsc() / (rte_get_tsc_hz() / 10);
6368 	uint32_t i;
6369 
6370 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6371 		cnt = MLX5_POOL_GET_CNT(pool, i);
6372 		age_param = MLX5_CNT_TO_AGE(cnt);
6373 		if (rte_atomic16_read(&age_param->state) != AGE_CANDIDATE)
6374 			continue;
6375 		if (cur->data[i].hits != prev->data[i].hits) {
6376 			age_param->expire = curr + age_param->timeout;
6377 			continue;
6378 		}
6379 		if ((uint16_t)(curr - age_param->expire) >= (UINT16_MAX / 2))
6380 			continue;
6381 		/**
6382 		 * Hold the lock first, or if between the
6383 		 * state AGE_TMOUT and tailq operation the
6384 		 * release happened, the release procedure
6385 		 * may delete a non-existent tailq node.
6386 		 */
6387 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6388 		age_info = GET_PORT_AGE_INFO(priv);
6389 		rte_spinlock_lock(&age_info->aged_sl);
6390 		/* If the cpmset fails, release happens. */
6391 		if (rte_atomic16_cmpset((volatile uint16_t *)
6392 					&age_param->state,
6393 					AGE_CANDIDATE,
6394 					AGE_TMOUT) ==
6395 					AGE_CANDIDATE) {
6396 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6397 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6398 		}
6399 		rte_spinlock_unlock(&age_info->aged_sl);
6400 	}
6401 	for (i = 0; i < sh->max_port; i++) {
6402 		age_info = &sh->port[i].age_info;
6403 		if (!MLX5_AGE_GET(age_info, MLX5_AGE_EVENT_NEW))
6404 			continue;
6405 		if (MLX5_AGE_GET(age_info, MLX5_AGE_TRIGGER))
6406 			rte_eth_dev_callback_process
6407 				(&rte_eth_devices[sh->port[i].devx_ih_port_id],
6408 				RTE_ETH_EVENT_FLOW_AGED, NULL);
6409 		age_info->flags = 0;
6410 	}
6411 }
6412 
6413 /**
6414  * Handler for the HW respond about ready values from an asynchronous batch
6415  * query. This function is probably called by the host thread.
6416  *
6417  * @param[in] sh
6418  *   The pointer to the shared device context.
6419  * @param[in] async_id
6420  *   The Devx async ID.
6421  * @param[in] status
6422  *   The status of the completion.
6423  */
6424 void
6425 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6426 				  uint64_t async_id, int status)
6427 {
6428 	struct mlx5_flow_counter_pool *pool =
6429 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6430 	struct mlx5_counter_stats_raw *raw_to_free;
6431 	uint8_t age = !!IS_AGE_POOL(pool);
6432 	uint8_t query_gen = pool->query_gen ^ 1;
6433 	struct mlx5_pools_container *cont =
6434 		MLX5_CNT_CONTAINER(sh, !IS_EXT_POOL(pool), age);
6435 
6436 	if (unlikely(status)) {
6437 		raw_to_free = pool->raw_hw;
6438 	} else {
6439 		raw_to_free = pool->raw;
6440 		if (IS_AGE_POOL(pool))
6441 			mlx5_flow_aging_check(sh, pool);
6442 		rte_spinlock_lock(&pool->sl);
6443 		pool->raw = pool->raw_hw;
6444 		rte_spinlock_unlock(&pool->sl);
6445 		/* Be sure the new raw counters data is updated in memory. */
6446 		rte_io_wmb();
6447 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6448 			rte_spinlock_lock(&cont->csl);
6449 			TAILQ_CONCAT(&cont->counters,
6450 				     &pool->counters[query_gen], next);
6451 			rte_spinlock_unlock(&cont->csl);
6452 		}
6453 	}
6454 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6455 	pool->raw_hw = NULL;
6456 	sh->cmng.pending_queries--;
6457 }
6458 
6459 /**
6460  * Translate the rte_flow group index to HW table value.
6461  *
6462  * @param[in] attributes
6463  *   Pointer to flow attributes
6464  * @param[in] external
6465  *   Value is part of flow rule created by request external to PMD.
6466  * @param[in] group
6467  *   rte_flow group index value.
6468  * @param[out] fdb_def_rule
6469  *   Whether fdb jump to table 1 is configured.
6470  * @param[out] table
6471  *   HW table value.
6472  * @param[out] error
6473  *   Pointer to error structure.
6474  *
6475  * @return
6476  *   0 on success, a negative errno value otherwise and rte_errno is set.
6477  */
6478 int
6479 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
6480 			 uint32_t group, bool fdb_def_rule, uint32_t *table,
6481 			 struct rte_flow_error *error)
6482 {
6483 	if (attributes->transfer && external && fdb_def_rule) {
6484 		if (group == UINT32_MAX)
6485 			return rte_flow_error_set
6486 						(error, EINVAL,
6487 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
6488 						 NULL,
6489 						 "group index not supported");
6490 		*table = group + 1;
6491 	} else {
6492 		*table = group;
6493 	}
6494 	return 0;
6495 }
6496 
6497 /**
6498  * Discover availability of metadata reg_c's.
6499  *
6500  * Iteratively use test flows to check availability.
6501  *
6502  * @param[in] dev
6503  *   Pointer to the Ethernet device structure.
6504  *
6505  * @return
6506  *   0 on success, a negative errno value otherwise and rte_errno is set.
6507  */
6508 int
6509 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
6510 {
6511 	struct mlx5_priv *priv = dev->data->dev_private;
6512 	struct mlx5_dev_config *config = &priv->config;
6513 	enum modify_reg idx;
6514 	int n = 0;
6515 
6516 	/* reg_c[0] and reg_c[1] are reserved. */
6517 	config->flow_mreg_c[n++] = REG_C_0;
6518 	config->flow_mreg_c[n++] = REG_C_1;
6519 	/* Discover availability of other reg_c's. */
6520 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
6521 		struct rte_flow_attr attr = {
6522 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
6523 			.priority = MLX5_FLOW_PRIO_RSVD,
6524 			.ingress = 1,
6525 		};
6526 		struct rte_flow_item items[] = {
6527 			[0] = {
6528 				.type = RTE_FLOW_ITEM_TYPE_END,
6529 			},
6530 		};
6531 		struct rte_flow_action actions[] = {
6532 			[0] = {
6533 				.type = (enum rte_flow_action_type)
6534 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
6535 				.conf = &(struct mlx5_flow_action_copy_mreg){
6536 					.src = REG_C_1,
6537 					.dst = idx,
6538 				},
6539 			},
6540 			[1] = {
6541 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
6542 				.conf = &(struct rte_flow_action_jump){
6543 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6544 				},
6545 			},
6546 			[2] = {
6547 				.type = RTE_FLOW_ACTION_TYPE_END,
6548 			},
6549 		};
6550 		uint32_t flow_idx;
6551 		struct rte_flow *flow;
6552 		struct rte_flow_error error;
6553 
6554 		if (!config->dv_flow_en)
6555 			break;
6556 		/* Create internal flow, validation skips copy action. */
6557 		flow_idx = flow_list_create(dev, NULL, &attr, items,
6558 					    actions, false, &error);
6559 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
6560 				      flow_idx);
6561 		if (!flow)
6562 			continue;
6563 		if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
6564 			config->flow_mreg_c[n++] = idx;
6565 		flow_list_destroy(dev, NULL, flow_idx);
6566 	}
6567 	for (; n < MLX5_MREG_C_NUM; ++n)
6568 		config->flow_mreg_c[n] = REG_NON;
6569 	return 0;
6570 }
6571 
6572 /**
6573  * Dump flow raw hw data to file
6574  *
6575  * @param[in] dev
6576  *    The pointer to Ethernet device.
6577  * @param[in] file
6578  *   A pointer to a file for output.
6579  * @param[out] error
6580  *   Perform verbose error reporting if not NULL. PMDs initialize this
6581  *   structure in case of error only.
6582  * @return
6583  *   0 on success, a nagative value otherwise.
6584  */
6585 int
6586 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
6587 		   FILE *file,
6588 		   struct rte_flow_error *error __rte_unused)
6589 {
6590 	struct mlx5_priv *priv = dev->data->dev_private;
6591 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6592 
6593 	if (!priv->config.dv_flow_en) {
6594 		if (fputs("device dv flow disabled\n", file) <= 0)
6595 			return -errno;
6596 		return -ENOTSUP;
6597 	}
6598 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
6599 				       sh->tx_domain, file);
6600 }
6601 
6602 /**
6603  * Get aged-out flows.
6604  *
6605  * @param[in] dev
6606  *   Pointer to the Ethernet device structure.
6607  * @param[in] context
6608  *   The address of an array of pointers to the aged-out flows contexts.
6609  * @param[in] nb_countexts
6610  *   The length of context array pointers.
6611  * @param[out] error
6612  *   Perform verbose error reporting if not NULL. Initialized in case of
6613  *   error only.
6614  *
6615  * @return
6616  *   how many contexts get in success, otherwise negative errno value.
6617  *   if nb_contexts is 0, return the amount of all aged contexts.
6618  *   if nb_contexts is not 0 , return the amount of aged flows reported
6619  *   in the context array.
6620  */
6621 int
6622 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
6623 			uint32_t nb_contexts, struct rte_flow_error *error)
6624 {
6625 	const struct mlx5_flow_driver_ops *fops;
6626 	struct rte_flow_attr attr = { .transfer = 0 };
6627 
6628 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6629 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6630 		return fops->get_aged_flows(dev, contexts, nb_contexts,
6631 						    error);
6632 	}
6633 	DRV_LOG(ERR,
6634 		"port %u get aged flows is not supported.",
6635 		 dev->data->port_id);
6636 	return -ENOTSUP;
6637 }
6638