xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 8b11f9aa7bf12b097c8d7f049542e0e289eca79b)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <netinet/in.h>
7 #include <sys/queue.h>
8 #include <stdalign.h>
9 #include <stdint.h>
10 #include <string.h>
11 #include <stdbool.h>
12 
13 #include <rte_common.h>
14 #include <rte_ether.h>
15 #include <rte_ethdev_driver.h>
16 #include <rte_eal_paging.h>
17 #include <rte_flow.h>
18 #include <rte_cycles.h>
19 #include <rte_flow_driver.h>
20 #include <rte_malloc.h>
21 #include <rte_ip.h>
22 
23 #include <mlx5_glue.h>
24 #include <mlx5_devx_cmds.h>
25 #include <mlx5_prm.h>
26 #include <mlx5_malloc.h>
27 
28 #include "mlx5_defs.h"
29 #include "mlx5.h"
30 #include "mlx5_flow.h"
31 #include "mlx5_flow_os.h"
32 #include "mlx5_rxtx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 static struct mlx5_flow_tunnel *
37 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
38 static void
39 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
40 static const struct mlx5_flow_tbl_data_entry  *
41 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark);
42 static int
43 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
44 		     const struct rte_flow_tunnel *app_tunnel,
45 		     struct mlx5_flow_tunnel **tunnel);
46 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
47 static void mlx5_flow_pop_thread_workspace(void);
48 
49 
50 /** Device flow drivers. */
51 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
52 
53 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
54 
55 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
56 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
57 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
58 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
59 #endif
60 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
61 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
62 };
63 
64 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
65 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
66 	(const int []){ \
67 		__VA_ARGS__, 0, \
68 	}
69 
70 /** Node object of input graph for mlx5_flow_expand_rss(). */
71 struct mlx5_flow_expand_node {
72 	const int *const next;
73 	/**<
74 	 * List of next node indexes. Index 0 is interpreted as a terminator.
75 	 */
76 	const enum rte_flow_item_type type;
77 	/**< Pattern item type of current node. */
78 	uint64_t rss_types;
79 	/**<
80 	 * RSS types bit-field associated with this node
81 	 * (see ETH_RSS_* definitions).
82 	 */
83 };
84 
85 /** Object returned by mlx5_flow_expand_rss(). */
86 struct mlx5_flow_expand_rss {
87 	uint32_t entries;
88 	/**< Number of entries @p patterns and @p priorities. */
89 	struct {
90 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
91 		uint32_t priority; /**< Priority offset for each expansion. */
92 	} entry[];
93 };
94 
95 static enum rte_flow_item_type
96 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
97 {
98 	enum rte_flow_item_type ret = RTE_FLOW_ITEM_TYPE_VOID;
99 	uint16_t ether_type = 0;
100 	uint16_t ether_type_m;
101 	uint8_t ip_next_proto = 0;
102 	uint8_t ip_next_proto_m;
103 
104 	if (item == NULL || item->spec == NULL)
105 		return ret;
106 	switch (item->type) {
107 	case RTE_FLOW_ITEM_TYPE_ETH:
108 		if (item->mask)
109 			ether_type_m = ((const struct rte_flow_item_eth *)
110 						(item->mask))->type;
111 		else
112 			ether_type_m = rte_flow_item_eth_mask.type;
113 		if (ether_type_m != RTE_BE16(0xFFFF))
114 			break;
115 		ether_type = ((const struct rte_flow_item_eth *)
116 				(item->spec))->type;
117 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
118 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
119 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
120 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
121 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
122 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
123 		else
124 			ret = RTE_FLOW_ITEM_TYPE_END;
125 		break;
126 	case RTE_FLOW_ITEM_TYPE_VLAN:
127 		if (item->mask)
128 			ether_type_m = ((const struct rte_flow_item_vlan *)
129 						(item->mask))->inner_type;
130 		else
131 			ether_type_m = rte_flow_item_vlan_mask.inner_type;
132 		if (ether_type_m != RTE_BE16(0xFFFF))
133 			break;
134 		ether_type = ((const struct rte_flow_item_vlan *)
135 				(item->spec))->inner_type;
136 		if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV4)
137 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
138 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_IPV6)
139 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
140 		else if (rte_be_to_cpu_16(ether_type) == RTE_ETHER_TYPE_VLAN)
141 			ret = RTE_FLOW_ITEM_TYPE_VLAN;
142 		else
143 			ret = RTE_FLOW_ITEM_TYPE_END;
144 		break;
145 	case RTE_FLOW_ITEM_TYPE_IPV4:
146 		if (item->mask)
147 			ip_next_proto_m = ((const struct rte_flow_item_ipv4 *)
148 					(item->mask))->hdr.next_proto_id;
149 		else
150 			ip_next_proto_m =
151 				rte_flow_item_ipv4_mask.hdr.next_proto_id;
152 		if (ip_next_proto_m != 0xFF)
153 			break;
154 		ip_next_proto = ((const struct rte_flow_item_ipv4 *)
155 				(item->spec))->hdr.next_proto_id;
156 		if (ip_next_proto == IPPROTO_UDP)
157 			ret = RTE_FLOW_ITEM_TYPE_UDP;
158 		else if (ip_next_proto == IPPROTO_TCP)
159 			ret = RTE_FLOW_ITEM_TYPE_TCP;
160 		else if (ip_next_proto == IPPROTO_IP)
161 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
162 		else if (ip_next_proto == IPPROTO_IPV6)
163 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
164 		else
165 			ret = RTE_FLOW_ITEM_TYPE_END;
166 		break;
167 	case RTE_FLOW_ITEM_TYPE_IPV6:
168 		if (item->mask)
169 			ip_next_proto_m = ((const struct rte_flow_item_ipv6 *)
170 						(item->mask))->hdr.proto;
171 		else
172 			ip_next_proto_m =
173 				rte_flow_item_ipv6_mask.hdr.proto;
174 		if (ip_next_proto_m != 0xFF)
175 			break;
176 		ip_next_proto = ((const struct rte_flow_item_ipv6 *)
177 				(item->spec))->hdr.proto;
178 		if (ip_next_proto == IPPROTO_UDP)
179 			ret = RTE_FLOW_ITEM_TYPE_UDP;
180 		else if (ip_next_proto == IPPROTO_TCP)
181 			ret = RTE_FLOW_ITEM_TYPE_TCP;
182 		else if (ip_next_proto == IPPROTO_IP)
183 			ret = RTE_FLOW_ITEM_TYPE_IPV4;
184 		else if (ip_next_proto == IPPROTO_IPV6)
185 			ret = RTE_FLOW_ITEM_TYPE_IPV6;
186 		else
187 			ret = RTE_FLOW_ITEM_TYPE_END;
188 		break;
189 	default:
190 		ret = RTE_FLOW_ITEM_TYPE_VOID;
191 		break;
192 	}
193 	return ret;
194 }
195 
196 /**
197  * Expand RSS flows into several possible flows according to the RSS hash
198  * fields requested and the driver capabilities.
199  *
200  * @param[out] buf
201  *   Buffer to store the result expansion.
202  * @param[in] size
203  *   Buffer size in bytes. If 0, @p buf can be NULL.
204  * @param[in] pattern
205  *   User flow pattern.
206  * @param[in] types
207  *   RSS types to expand (see ETH_RSS_* definitions).
208  * @param[in] graph
209  *   Input graph to expand @p pattern according to @p types.
210  * @param[in] graph_root_index
211  *   Index of root node in @p graph, typically 0.
212  *
213  * @return
214  *   A positive value representing the size of @p buf in bytes regardless of
215  *   @p size on success, a negative errno value otherwise and rte_errno is
216  *   set, the following errors are defined:
217  *
218  *   -E2BIG: graph-depth @p graph is too deep.
219  */
220 static int
221 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
222 		     const struct rte_flow_item *pattern, uint64_t types,
223 		     const struct mlx5_flow_expand_node graph[],
224 		     int graph_root_index)
225 {
226 	const int elt_n = 8;
227 	const struct rte_flow_item *item;
228 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
229 	const int *next_node;
230 	const int *stack[elt_n];
231 	int stack_pos = 0;
232 	struct rte_flow_item flow_items[elt_n];
233 	unsigned int i;
234 	size_t lsize;
235 	size_t user_pattern_size = 0;
236 	void *addr = NULL;
237 	const struct mlx5_flow_expand_node *next = NULL;
238 	struct rte_flow_item missed_item;
239 	int missed = 0;
240 	int elt = 0;
241 	const struct rte_flow_item *last_item = NULL;
242 
243 	memset(&missed_item, 0, sizeof(missed_item));
244 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
245 		elt_n * sizeof(buf->entry[0]);
246 	if (lsize <= size) {
247 		buf->entry[0].priority = 0;
248 		buf->entry[0].pattern = (void *)&buf->entry[elt_n];
249 		buf->entries = 0;
250 		addr = buf->entry[0].pattern;
251 	}
252 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
253 		if (item->type != RTE_FLOW_ITEM_TYPE_VOID)
254 			last_item = item;
255 		for (i = 0; node->next && node->next[i]; ++i) {
256 			next = &graph[node->next[i]];
257 			if (next->type == item->type)
258 				break;
259 		}
260 		if (next)
261 			node = next;
262 		user_pattern_size += sizeof(*item);
263 	}
264 	user_pattern_size += sizeof(*item); /* Handle END item. */
265 	lsize += user_pattern_size;
266 	/* Copy the user pattern in the first entry of the buffer. */
267 	if (lsize <= size) {
268 		rte_memcpy(addr, pattern, user_pattern_size);
269 		addr = (void *)(((uintptr_t)addr) + user_pattern_size);
270 		buf->entries = 1;
271 	}
272 	/* Start expanding. */
273 	memset(flow_items, 0, sizeof(flow_items));
274 	user_pattern_size -= sizeof(*item);
275 	/*
276 	 * Check if the last valid item has spec set, need complete pattern,
277 	 * and the pattern can be used for expansion.
278 	 */
279 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_item);
280 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
281 		/* Item type END indicates expansion is not required. */
282 		return lsize;
283 	}
284 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
285 		next = NULL;
286 		missed = 1;
287 		for (i = 0; node->next && node->next[i]; ++i) {
288 			next = &graph[node->next[i]];
289 			if (next->type == missed_item.type) {
290 				flow_items[0].type = missed_item.type;
291 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
292 				break;
293 			}
294 			next = NULL;
295 		}
296 	}
297 	if (next && missed) {
298 		elt = 2; /* missed item + item end. */
299 		node = next;
300 		lsize += elt * sizeof(*item) + user_pattern_size;
301 		if ((node->rss_types & types) && lsize <= size) {
302 			buf->entry[buf->entries].priority = 1;
303 			buf->entry[buf->entries].pattern = addr;
304 			buf->entries++;
305 			rte_memcpy(addr, buf->entry[0].pattern,
306 				   user_pattern_size);
307 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
308 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
309 			addr = (void *)(((uintptr_t)addr) +
310 					elt * sizeof(*item));
311 		}
312 	}
313 	memset(flow_items, 0, sizeof(flow_items));
314 	next_node = node->next;
315 	stack[stack_pos] = next_node;
316 	node = next_node ? &graph[*next_node] : NULL;
317 	while (node) {
318 		flow_items[stack_pos].type = node->type;
319 		if (node->rss_types & types) {
320 			/*
321 			 * compute the number of items to copy from the
322 			 * expansion and copy it.
323 			 * When the stack_pos is 0, there are 1 element in it,
324 			 * plus the addition END item.
325 			 */
326 			elt = stack_pos + 2;
327 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
328 			lsize += elt * sizeof(*item) + user_pattern_size;
329 			if (lsize <= size) {
330 				size_t n = elt * sizeof(*item);
331 
332 				buf->entry[buf->entries].priority =
333 					stack_pos + 1 + missed;
334 				buf->entry[buf->entries].pattern = addr;
335 				buf->entries++;
336 				rte_memcpy(addr, buf->entry[0].pattern,
337 					   user_pattern_size);
338 				addr = (void *)(((uintptr_t)addr) +
339 						user_pattern_size);
340 				rte_memcpy(addr, &missed_item,
341 					   missed * sizeof(*item));
342 				addr = (void *)(((uintptr_t)addr) +
343 					missed * sizeof(*item));
344 				rte_memcpy(addr, flow_items, n);
345 				addr = (void *)(((uintptr_t)addr) + n);
346 			}
347 		}
348 		/* Go deeper. */
349 		if (node->next) {
350 			next_node = node->next;
351 			if (stack_pos++ == elt_n) {
352 				rte_errno = E2BIG;
353 				return -rte_errno;
354 			}
355 			stack[stack_pos] = next_node;
356 		} else if (*(next_node + 1)) {
357 			/* Follow up with the next possibility. */
358 			++next_node;
359 		} else {
360 			/* Move to the next path. */
361 			if (stack_pos)
362 				next_node = stack[--stack_pos];
363 			next_node++;
364 			stack[stack_pos] = next_node;
365 		}
366 		node = *next_node ? &graph[*next_node] : NULL;
367 	};
368 	/* no expanded flows but we have missed item, create one rule for it */
369 	if (buf->entries == 1 && missed != 0) {
370 		elt = 2;
371 		lsize += elt * sizeof(*item) + user_pattern_size;
372 		if (lsize <= size) {
373 			buf->entry[buf->entries].priority = 1;
374 			buf->entry[buf->entries].pattern = addr;
375 			buf->entries++;
376 			flow_items[0].type = missed_item.type;
377 			flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
378 			rte_memcpy(addr, buf->entry[0].pattern,
379 				   user_pattern_size);
380 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
381 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
382 		}
383 	}
384 	return lsize;
385 }
386 
387 enum mlx5_expansion {
388 	MLX5_EXPANSION_ROOT,
389 	MLX5_EXPANSION_ROOT_OUTER,
390 	MLX5_EXPANSION_ROOT_ETH_VLAN,
391 	MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
392 	MLX5_EXPANSION_OUTER_ETH,
393 	MLX5_EXPANSION_OUTER_ETH_VLAN,
394 	MLX5_EXPANSION_OUTER_VLAN,
395 	MLX5_EXPANSION_OUTER_IPV4,
396 	MLX5_EXPANSION_OUTER_IPV4_UDP,
397 	MLX5_EXPANSION_OUTER_IPV4_TCP,
398 	MLX5_EXPANSION_OUTER_IPV6,
399 	MLX5_EXPANSION_OUTER_IPV6_UDP,
400 	MLX5_EXPANSION_OUTER_IPV6_TCP,
401 	MLX5_EXPANSION_VXLAN,
402 	MLX5_EXPANSION_VXLAN_GPE,
403 	MLX5_EXPANSION_GRE,
404 	MLX5_EXPANSION_MPLS,
405 	MLX5_EXPANSION_ETH,
406 	MLX5_EXPANSION_ETH_VLAN,
407 	MLX5_EXPANSION_VLAN,
408 	MLX5_EXPANSION_IPV4,
409 	MLX5_EXPANSION_IPV4_UDP,
410 	MLX5_EXPANSION_IPV4_TCP,
411 	MLX5_EXPANSION_IPV6,
412 	MLX5_EXPANSION_IPV6_UDP,
413 	MLX5_EXPANSION_IPV6_TCP,
414 };
415 
416 /** Supported expansion of items. */
417 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
418 	[MLX5_EXPANSION_ROOT] = {
419 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
420 						  MLX5_EXPANSION_IPV4,
421 						  MLX5_EXPANSION_IPV6),
422 		.type = RTE_FLOW_ITEM_TYPE_END,
423 	},
424 	[MLX5_EXPANSION_ROOT_OUTER] = {
425 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
426 						  MLX5_EXPANSION_OUTER_IPV4,
427 						  MLX5_EXPANSION_OUTER_IPV6),
428 		.type = RTE_FLOW_ITEM_TYPE_END,
429 	},
430 	[MLX5_EXPANSION_ROOT_ETH_VLAN] = {
431 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
432 		.type = RTE_FLOW_ITEM_TYPE_END,
433 	},
434 	[MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
435 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
436 						(MLX5_EXPANSION_OUTER_ETH_VLAN),
437 		.type = RTE_FLOW_ITEM_TYPE_END,
438 	},
439 	[MLX5_EXPANSION_OUTER_ETH] = {
440 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
441 						  MLX5_EXPANSION_OUTER_IPV6,
442 						  MLX5_EXPANSION_MPLS),
443 		.type = RTE_FLOW_ITEM_TYPE_ETH,
444 		.rss_types = 0,
445 	},
446 	[MLX5_EXPANSION_OUTER_ETH_VLAN] = {
447 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
448 		.type = RTE_FLOW_ITEM_TYPE_ETH,
449 		.rss_types = 0,
450 	},
451 	[MLX5_EXPANSION_OUTER_VLAN] = {
452 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
453 						  MLX5_EXPANSION_OUTER_IPV6),
454 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
455 	},
456 	[MLX5_EXPANSION_OUTER_IPV4] = {
457 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
458 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
459 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
460 			 MLX5_EXPANSION_GRE,
461 			 MLX5_EXPANSION_IPV4,
462 			 MLX5_EXPANSION_IPV6),
463 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
464 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
465 			ETH_RSS_NONFRAG_IPV4_OTHER,
466 	},
467 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
468 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
469 						  MLX5_EXPANSION_VXLAN_GPE),
470 		.type = RTE_FLOW_ITEM_TYPE_UDP,
471 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
472 	},
473 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
474 		.type = RTE_FLOW_ITEM_TYPE_TCP,
475 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
476 	},
477 	[MLX5_EXPANSION_OUTER_IPV6] = {
478 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
479 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
480 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
481 			 MLX5_EXPANSION_IPV4,
482 			 MLX5_EXPANSION_IPV6),
483 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
484 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
485 			ETH_RSS_NONFRAG_IPV6_OTHER,
486 	},
487 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
488 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
489 						  MLX5_EXPANSION_VXLAN_GPE),
490 		.type = RTE_FLOW_ITEM_TYPE_UDP,
491 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
492 	},
493 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
494 		.type = RTE_FLOW_ITEM_TYPE_TCP,
495 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
496 	},
497 	[MLX5_EXPANSION_VXLAN] = {
498 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
499 						  MLX5_EXPANSION_IPV4,
500 						  MLX5_EXPANSION_IPV6),
501 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
502 	},
503 	[MLX5_EXPANSION_VXLAN_GPE] = {
504 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
505 						  MLX5_EXPANSION_IPV4,
506 						  MLX5_EXPANSION_IPV6),
507 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
508 	},
509 	[MLX5_EXPANSION_GRE] = {
510 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
511 		.type = RTE_FLOW_ITEM_TYPE_GRE,
512 	},
513 	[MLX5_EXPANSION_MPLS] = {
514 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
515 						  MLX5_EXPANSION_IPV6),
516 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
517 	},
518 	[MLX5_EXPANSION_ETH] = {
519 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
520 						  MLX5_EXPANSION_IPV6),
521 		.type = RTE_FLOW_ITEM_TYPE_ETH,
522 	},
523 	[MLX5_EXPANSION_ETH_VLAN] = {
524 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
525 		.type = RTE_FLOW_ITEM_TYPE_ETH,
526 	},
527 	[MLX5_EXPANSION_VLAN] = {
528 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
529 						  MLX5_EXPANSION_IPV6),
530 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
531 	},
532 	[MLX5_EXPANSION_IPV4] = {
533 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
534 						  MLX5_EXPANSION_IPV4_TCP),
535 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
536 		.rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
537 			ETH_RSS_NONFRAG_IPV4_OTHER,
538 	},
539 	[MLX5_EXPANSION_IPV4_UDP] = {
540 		.type = RTE_FLOW_ITEM_TYPE_UDP,
541 		.rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
542 	},
543 	[MLX5_EXPANSION_IPV4_TCP] = {
544 		.type = RTE_FLOW_ITEM_TYPE_TCP,
545 		.rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
546 	},
547 	[MLX5_EXPANSION_IPV6] = {
548 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
549 						  MLX5_EXPANSION_IPV6_TCP),
550 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
551 		.rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
552 			ETH_RSS_NONFRAG_IPV6_OTHER,
553 	},
554 	[MLX5_EXPANSION_IPV6_UDP] = {
555 		.type = RTE_FLOW_ITEM_TYPE_UDP,
556 		.rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
557 	},
558 	[MLX5_EXPANSION_IPV6_TCP] = {
559 		.type = RTE_FLOW_ITEM_TYPE_TCP,
560 		.rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
561 	},
562 };
563 
564 static struct rte_flow_shared_action *
565 mlx5_shared_action_create(struct rte_eth_dev *dev,
566 			  const struct rte_flow_shared_action_conf *conf,
567 			  const struct rte_flow_action *action,
568 			  struct rte_flow_error *error);
569 static int mlx5_shared_action_destroy
570 				(struct rte_eth_dev *dev,
571 				 struct rte_flow_shared_action *shared_action,
572 				 struct rte_flow_error *error);
573 static int mlx5_shared_action_update
574 				(struct rte_eth_dev *dev,
575 				 struct rte_flow_shared_action *shared_action,
576 				 const struct rte_flow_action *action,
577 				 struct rte_flow_error *error);
578 static int mlx5_shared_action_query
579 				(struct rte_eth_dev *dev,
580 				 const struct rte_flow_shared_action *action,
581 				 void *data,
582 				 struct rte_flow_error *error);
583 static inline bool
584 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
585 			  struct rte_flow_tunnel *tunnel,
586 			  const char *err_msg)
587 {
588 	err_msg = NULL;
589 	if (!is_tunnel_offload_active(dev)) {
590 		err_msg = "tunnel offload was not activated";
591 		goto out;
592 	} else if (!tunnel) {
593 		err_msg = "no application tunnel";
594 		goto out;
595 	}
596 
597 	switch (tunnel->type) {
598 	default:
599 		err_msg = "unsupported tunnel type";
600 		goto out;
601 	case RTE_FLOW_ITEM_TYPE_VXLAN:
602 		break;
603 	}
604 
605 out:
606 	return !err_msg;
607 }
608 
609 
610 static int
611 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
612 		    struct rte_flow_tunnel *app_tunnel,
613 		    struct rte_flow_action **actions,
614 		    uint32_t *num_of_actions,
615 		    struct rte_flow_error *error)
616 {
617 	int ret;
618 	struct mlx5_flow_tunnel *tunnel;
619 	const char *err_msg = NULL;
620 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
621 
622 	if (!verdict)
623 		return rte_flow_error_set(error, EINVAL,
624 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
625 					  err_msg);
626 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
627 	if (ret < 0) {
628 		return rte_flow_error_set(error, ret,
629 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
630 					  "failed to initialize pmd tunnel");
631 	}
632 	*actions = &tunnel->action;
633 	*num_of_actions = 1;
634 	return 0;
635 }
636 
637 static int
638 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
639 		       struct rte_flow_tunnel *app_tunnel,
640 		       struct rte_flow_item **items,
641 		       uint32_t *num_of_items,
642 		       struct rte_flow_error *error)
643 {
644 	int ret;
645 	struct mlx5_flow_tunnel *tunnel;
646 	const char *err_msg = NULL;
647 	bool verdict = mlx5_flow_tunnel_validate(dev, app_tunnel, err_msg);
648 
649 	if (!verdict)
650 		return rte_flow_error_set(error, EINVAL,
651 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
652 					  err_msg);
653 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
654 	if (ret < 0) {
655 		return rte_flow_error_set(error, ret,
656 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
657 					  "failed to initialize pmd tunnel");
658 	}
659 	*items = &tunnel->item;
660 	*num_of_items = 1;
661 	return 0;
662 }
663 
664 static int
665 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
666 			      struct rte_flow_item *pmd_items,
667 			      uint32_t num_items, struct rte_flow_error *err)
668 {
669 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
670 	struct mlx5_flow_tunnel *tun;
671 
672 	rte_spinlock_lock(&thub->sl);
673 	LIST_FOREACH(tun, &thub->tunnels, chain) {
674 		if (&tun->item == pmd_items) {
675 			LIST_REMOVE(tun, chain);
676 			break;
677 		}
678 	}
679 	rte_spinlock_unlock(&thub->sl);
680 	if (!tun || num_items != 1)
681 		return rte_flow_error_set(err, EINVAL,
682 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
683 					  "invalid argument");
684 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
685 		mlx5_flow_tunnel_free(dev, tun);
686 	return 0;
687 }
688 
689 static int
690 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
691 				struct rte_flow_action *pmd_actions,
692 				uint32_t num_actions,
693 				struct rte_flow_error *err)
694 {
695 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
696 	struct mlx5_flow_tunnel *tun;
697 
698 	rte_spinlock_lock(&thub->sl);
699 	LIST_FOREACH(tun, &thub->tunnels, chain) {
700 		if (&tun->action == pmd_actions) {
701 			LIST_REMOVE(tun, chain);
702 			break;
703 		}
704 	}
705 	rte_spinlock_unlock(&thub->sl);
706 	if (!tun || num_actions != 1)
707 		return rte_flow_error_set(err, EINVAL,
708 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
709 					  "invalid argument");
710 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
711 		mlx5_flow_tunnel_free(dev, tun);
712 
713 	return 0;
714 }
715 
716 static int
717 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
718 				  struct rte_mbuf *m,
719 				  struct rte_flow_restore_info *info,
720 				  struct rte_flow_error *err)
721 {
722 	uint64_t ol_flags = m->ol_flags;
723 	const struct mlx5_flow_tbl_data_entry *tble;
724 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
725 
726 	if ((ol_flags & mask) != mask)
727 		goto err;
728 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
729 	if (!tble) {
730 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
731 			dev->data->port_id, m->hash.fdir.hi);
732 		goto err;
733 	}
734 	MLX5_ASSERT(tble->tunnel);
735 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
736 	info->group_id = tble->group_id;
737 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
738 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
739 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
740 
741 	return 0;
742 
743 err:
744 	return rte_flow_error_set(err, EINVAL,
745 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
746 				  "failed to get restore info");
747 }
748 
749 static const struct rte_flow_ops mlx5_flow_ops = {
750 	.validate = mlx5_flow_validate,
751 	.create = mlx5_flow_create,
752 	.destroy = mlx5_flow_destroy,
753 	.flush = mlx5_flow_flush,
754 	.isolate = mlx5_flow_isolate,
755 	.query = mlx5_flow_query,
756 	.dev_dump = mlx5_flow_dev_dump,
757 	.get_aged_flows = mlx5_flow_get_aged_flows,
758 	.shared_action_create = mlx5_shared_action_create,
759 	.shared_action_destroy = mlx5_shared_action_destroy,
760 	.shared_action_update = mlx5_shared_action_update,
761 	.shared_action_query = mlx5_shared_action_query,
762 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
763 	.tunnel_match = mlx5_flow_tunnel_match,
764 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
765 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
766 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
767 };
768 
769 /* Tunnel information. */
770 struct mlx5_flow_tunnel_info {
771 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
772 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
773 };
774 
775 static struct mlx5_flow_tunnel_info tunnels_info[] = {
776 	{
777 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
778 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
779 	},
780 	{
781 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
782 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
783 	},
784 	{
785 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
786 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
787 	},
788 	{
789 		.tunnel = MLX5_FLOW_LAYER_GRE,
790 		.ptype = RTE_PTYPE_TUNNEL_GRE,
791 	},
792 	{
793 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
794 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
795 	},
796 	{
797 		.tunnel = MLX5_FLOW_LAYER_MPLS,
798 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
799 	},
800 	{
801 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
802 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
803 	},
804 	{
805 		.tunnel = MLX5_FLOW_LAYER_IPIP,
806 		.ptype = RTE_PTYPE_TUNNEL_IP,
807 	},
808 	{
809 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
810 		.ptype = RTE_PTYPE_TUNNEL_IP,
811 	},
812 	{
813 		.tunnel = MLX5_FLOW_LAYER_GTP,
814 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
815 	},
816 };
817 
818 /* Key of thread specific flow workspace data. */
819 static pthread_key_t key_workspace;
820 
821 /* Thread specific flow workspace data once initialization data. */
822 static pthread_once_t key_workspace_init;
823 
824 
825 /**
826  * Translate tag ID to register.
827  *
828  * @param[in] dev
829  *   Pointer to the Ethernet device structure.
830  * @param[in] feature
831  *   The feature that request the register.
832  * @param[in] id
833  *   The request register ID.
834  * @param[out] error
835  *   Error description in case of any.
836  *
837  * @return
838  *   The request register on success, a negative errno
839  *   value otherwise and rte_errno is set.
840  */
841 int
842 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
843 		     enum mlx5_feature_name feature,
844 		     uint32_t id,
845 		     struct rte_flow_error *error)
846 {
847 	struct mlx5_priv *priv = dev->data->dev_private;
848 	struct mlx5_dev_config *config = &priv->config;
849 	enum modify_reg start_reg;
850 	bool skip_mtr_reg = false;
851 
852 	switch (feature) {
853 	case MLX5_HAIRPIN_RX:
854 		return REG_B;
855 	case MLX5_HAIRPIN_TX:
856 		return REG_A;
857 	case MLX5_METADATA_RX:
858 		switch (config->dv_xmeta_en) {
859 		case MLX5_XMETA_MODE_LEGACY:
860 			return REG_B;
861 		case MLX5_XMETA_MODE_META16:
862 			return REG_C_0;
863 		case MLX5_XMETA_MODE_META32:
864 			return REG_C_1;
865 		}
866 		break;
867 	case MLX5_METADATA_TX:
868 		return REG_A;
869 	case MLX5_METADATA_FDB:
870 		switch (config->dv_xmeta_en) {
871 		case MLX5_XMETA_MODE_LEGACY:
872 			return REG_NON;
873 		case MLX5_XMETA_MODE_META16:
874 			return REG_C_0;
875 		case MLX5_XMETA_MODE_META32:
876 			return REG_C_1;
877 		}
878 		break;
879 	case MLX5_FLOW_MARK:
880 		switch (config->dv_xmeta_en) {
881 		case MLX5_XMETA_MODE_LEGACY:
882 			return REG_NON;
883 		case MLX5_XMETA_MODE_META16:
884 			return REG_C_1;
885 		case MLX5_XMETA_MODE_META32:
886 			return REG_C_0;
887 		}
888 		break;
889 	case MLX5_MTR_SFX:
890 		/*
891 		 * If meter color and flow match share one register, flow match
892 		 * should use the meter color register for match.
893 		 */
894 		if (priv->mtr_reg_share)
895 			return priv->mtr_color_reg;
896 		else
897 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
898 			       REG_C_3;
899 	case MLX5_MTR_COLOR:
900 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
901 		return priv->mtr_color_reg;
902 	case MLX5_COPY_MARK:
903 		/*
904 		 * Metadata COPY_MARK register using is in meter suffix sub
905 		 * flow while with meter. It's safe to share the same register.
906 		 */
907 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
908 	case MLX5_APP_TAG:
909 		/*
910 		 * If meter is enable, it will engage the register for color
911 		 * match and flow match. If meter color match is not using the
912 		 * REG_C_2, need to skip the REG_C_x be used by meter color
913 		 * match.
914 		 * If meter is disable, free to use all available registers.
915 		 */
916 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
917 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
918 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
919 		if (id > (REG_C_7 - start_reg))
920 			return rte_flow_error_set(error, EINVAL,
921 						  RTE_FLOW_ERROR_TYPE_ITEM,
922 						  NULL, "invalid tag id");
923 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
924 			return rte_flow_error_set(error, ENOTSUP,
925 						  RTE_FLOW_ERROR_TYPE_ITEM,
926 						  NULL, "unsupported tag id");
927 		/*
928 		 * This case means meter is using the REG_C_x great than 2.
929 		 * Take care not to conflict with meter color REG_C_x.
930 		 * If the available index REG_C_y >= REG_C_x, skip the
931 		 * color register.
932 		 */
933 		if (skip_mtr_reg && config->flow_mreg_c
934 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
935 			if (id >= (REG_C_7 - start_reg))
936 				return rte_flow_error_set(error, EINVAL,
937 						       RTE_FLOW_ERROR_TYPE_ITEM,
938 							NULL, "invalid tag id");
939 			if (config->flow_mreg_c
940 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
941 				return config->flow_mreg_c
942 					       [id + 1 + start_reg - REG_C_0];
943 			return rte_flow_error_set(error, ENOTSUP,
944 						  RTE_FLOW_ERROR_TYPE_ITEM,
945 						  NULL, "unsupported tag id");
946 		}
947 		return config->flow_mreg_c[id + start_reg - REG_C_0];
948 	}
949 	MLX5_ASSERT(false);
950 	return rte_flow_error_set(error, EINVAL,
951 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
952 				  NULL, "invalid feature name");
953 }
954 
955 /**
956  * Check extensive flow metadata register support.
957  *
958  * @param dev
959  *   Pointer to rte_eth_dev structure.
960  *
961  * @return
962  *   True if device supports extensive flow metadata register, otherwise false.
963  */
964 bool
965 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
966 {
967 	struct mlx5_priv *priv = dev->data->dev_private;
968 	struct mlx5_dev_config *config = &priv->config;
969 
970 	/*
971 	 * Having available reg_c can be regarded inclusively as supporting
972 	 * extensive flow metadata register, which could mean,
973 	 * - metadata register copy action by modify header.
974 	 * - 16 modify header actions is supported.
975 	 * - reg_c's are preserved across different domain (FDB and NIC) on
976 	 *   packet loopback by flow lookup miss.
977 	 */
978 	return config->flow_mreg_c[2] != REG_NON;
979 }
980 
981 /**
982  * Verify the @p item specifications (spec, last, mask) are compatible with the
983  * NIC capabilities.
984  *
985  * @param[in] item
986  *   Item specification.
987  * @param[in] mask
988  *   @p item->mask or flow default bit-masks.
989  * @param[in] nic_mask
990  *   Bit-masks covering supported fields by the NIC to compare with user mask.
991  * @param[in] size
992  *   Bit-masks size in bytes.
993  * @param[in] range_accepted
994  *   True if range of values is accepted for specific fields, false otherwise.
995  * @param[out] error
996  *   Pointer to error structure.
997  *
998  * @return
999  *   0 on success, a negative errno value otherwise and rte_errno is set.
1000  */
1001 int
1002 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1003 			  const uint8_t *mask,
1004 			  const uint8_t *nic_mask,
1005 			  unsigned int size,
1006 			  bool range_accepted,
1007 			  struct rte_flow_error *error)
1008 {
1009 	unsigned int i;
1010 
1011 	MLX5_ASSERT(nic_mask);
1012 	for (i = 0; i < size; ++i)
1013 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1014 			return rte_flow_error_set(error, ENOTSUP,
1015 						  RTE_FLOW_ERROR_TYPE_ITEM,
1016 						  item,
1017 						  "mask enables non supported"
1018 						  " bits");
1019 	if (!item->spec && (item->mask || item->last))
1020 		return rte_flow_error_set(error, EINVAL,
1021 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1022 					  "mask/last without a spec is not"
1023 					  " supported");
1024 	if (item->spec && item->last && !range_accepted) {
1025 		uint8_t spec[size];
1026 		uint8_t last[size];
1027 		unsigned int i;
1028 		int ret;
1029 
1030 		for (i = 0; i < size; ++i) {
1031 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1032 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1033 		}
1034 		ret = memcmp(spec, last, size);
1035 		if (ret != 0)
1036 			return rte_flow_error_set(error, EINVAL,
1037 						  RTE_FLOW_ERROR_TYPE_ITEM,
1038 						  item,
1039 						  "range is not valid");
1040 	}
1041 	return 0;
1042 }
1043 
1044 /**
1045  * Adjust the hash fields according to the @p flow information.
1046  *
1047  * @param[in] dev_flow.
1048  *   Pointer to the mlx5_flow.
1049  * @param[in] tunnel
1050  *   1 when the hash field is for a tunnel item.
1051  * @param[in] layer_types
1052  *   ETH_RSS_* types.
1053  * @param[in] hash_fields
1054  *   Item hash fields.
1055  *
1056  * @return
1057  *   The hash fields that should be used.
1058  */
1059 uint64_t
1060 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1061 			    int tunnel __rte_unused, uint64_t layer_types,
1062 			    uint64_t hash_fields)
1063 {
1064 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1065 	int rss_request_inner = rss_desc->level >= 2;
1066 
1067 	/* Check RSS hash level for tunnel. */
1068 	if (tunnel && rss_request_inner)
1069 		hash_fields |= IBV_RX_HASH_INNER;
1070 	else if (tunnel || rss_request_inner)
1071 		return 0;
1072 #endif
1073 	/* Check if requested layer matches RSS hash fields. */
1074 	if (!(rss_desc->types & layer_types))
1075 		return 0;
1076 	return hash_fields;
1077 }
1078 
1079 /**
1080  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1081  * if several tunnel rules are used on this queue, the tunnel ptype will be
1082  * cleared.
1083  *
1084  * @param rxq_ctrl
1085  *   Rx queue to update.
1086  */
1087 static void
1088 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1089 {
1090 	unsigned int i;
1091 	uint32_t tunnel_ptype = 0;
1092 
1093 	/* Look up for the ptype to use. */
1094 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1095 		if (!rxq_ctrl->flow_tunnels_n[i])
1096 			continue;
1097 		if (!tunnel_ptype) {
1098 			tunnel_ptype = tunnels_info[i].ptype;
1099 		} else {
1100 			tunnel_ptype = 0;
1101 			break;
1102 		}
1103 	}
1104 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1105 }
1106 
1107 /**
1108  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1109  * flow.
1110  *
1111  * @param[in] dev
1112  *   Pointer to the Ethernet device structure.
1113  * @param[in] dev_handle
1114  *   Pointer to device flow handle structure.
1115  */
1116 static void
1117 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1118 		       struct mlx5_flow_handle *dev_handle)
1119 {
1120 	struct mlx5_priv *priv = dev->data->dev_private;
1121 	const int mark = dev_handle->mark;
1122 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1123 	struct mlx5_hrxq *hrxq;
1124 	unsigned int i;
1125 
1126 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1127 		return;
1128 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1129 			      dev_handle->rix_hrxq);
1130 	if (!hrxq)
1131 		return;
1132 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1133 		int idx = hrxq->ind_table->queues[i];
1134 		struct mlx5_rxq_ctrl *rxq_ctrl =
1135 			container_of((*priv->rxqs)[idx],
1136 				     struct mlx5_rxq_ctrl, rxq);
1137 
1138 		/*
1139 		 * To support metadata register copy on Tx loopback,
1140 		 * this must be always enabled (metadata may arive
1141 		 * from other port - not from local flows only.
1142 		 */
1143 		if (priv->config.dv_flow_en &&
1144 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1145 		    mlx5_flow_ext_mreg_supported(dev)) {
1146 			rxq_ctrl->rxq.mark = 1;
1147 			rxq_ctrl->flow_mark_n = 1;
1148 		} else if (mark) {
1149 			rxq_ctrl->rxq.mark = 1;
1150 			rxq_ctrl->flow_mark_n++;
1151 		}
1152 		if (tunnel) {
1153 			unsigned int j;
1154 
1155 			/* Increase the counter matching the flow. */
1156 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1157 				if ((tunnels_info[j].tunnel &
1158 				     dev_handle->layers) ==
1159 				    tunnels_info[j].tunnel) {
1160 					rxq_ctrl->flow_tunnels_n[j]++;
1161 					break;
1162 				}
1163 			}
1164 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1165 		}
1166 	}
1167 }
1168 
1169 /**
1170  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1171  *
1172  * @param[in] dev
1173  *   Pointer to the Ethernet device structure.
1174  * @param[in] flow
1175  *   Pointer to flow structure.
1176  */
1177 static void
1178 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1179 {
1180 	struct mlx5_priv *priv = dev->data->dev_private;
1181 	uint32_t handle_idx;
1182 	struct mlx5_flow_handle *dev_handle;
1183 
1184 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1185 		       handle_idx, dev_handle, next)
1186 		flow_drv_rxq_flags_set(dev, dev_handle);
1187 }
1188 
1189 /**
1190  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1191  * device flow if no other flow uses it with the same kind of request.
1192  *
1193  * @param dev
1194  *   Pointer to Ethernet device.
1195  * @param[in] dev_handle
1196  *   Pointer to the device flow handle structure.
1197  */
1198 static void
1199 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1200 			struct mlx5_flow_handle *dev_handle)
1201 {
1202 	struct mlx5_priv *priv = dev->data->dev_private;
1203 	const int mark = dev_handle->mark;
1204 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1205 	struct mlx5_hrxq *hrxq;
1206 	unsigned int i;
1207 
1208 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1209 		return;
1210 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1211 			      dev_handle->rix_hrxq);
1212 	if (!hrxq)
1213 		return;
1214 	MLX5_ASSERT(dev->data->dev_started);
1215 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1216 		int idx = hrxq->ind_table->queues[i];
1217 		struct mlx5_rxq_ctrl *rxq_ctrl =
1218 			container_of((*priv->rxqs)[idx],
1219 				     struct mlx5_rxq_ctrl, rxq);
1220 
1221 		if (priv->config.dv_flow_en &&
1222 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1223 		    mlx5_flow_ext_mreg_supported(dev)) {
1224 			rxq_ctrl->rxq.mark = 1;
1225 			rxq_ctrl->flow_mark_n = 1;
1226 		} else if (mark) {
1227 			rxq_ctrl->flow_mark_n--;
1228 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1229 		}
1230 		if (tunnel) {
1231 			unsigned int j;
1232 
1233 			/* Decrease the counter matching the flow. */
1234 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1235 				if ((tunnels_info[j].tunnel &
1236 				     dev_handle->layers) ==
1237 				    tunnels_info[j].tunnel) {
1238 					rxq_ctrl->flow_tunnels_n[j]--;
1239 					break;
1240 				}
1241 			}
1242 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1243 		}
1244 	}
1245 }
1246 
1247 /**
1248  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1249  * @p flow if no other flow uses it with the same kind of request.
1250  *
1251  * @param dev
1252  *   Pointer to Ethernet device.
1253  * @param[in] flow
1254  *   Pointer to the flow.
1255  */
1256 static void
1257 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1258 {
1259 	struct mlx5_priv *priv = dev->data->dev_private;
1260 	uint32_t handle_idx;
1261 	struct mlx5_flow_handle *dev_handle;
1262 
1263 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1264 		       handle_idx, dev_handle, next)
1265 		flow_drv_rxq_flags_trim(dev, dev_handle);
1266 }
1267 
1268 /**
1269  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1270  *
1271  * @param dev
1272  *   Pointer to Ethernet device.
1273  */
1274 static void
1275 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1276 {
1277 	struct mlx5_priv *priv = dev->data->dev_private;
1278 	unsigned int i;
1279 
1280 	for (i = 0; i != priv->rxqs_n; ++i) {
1281 		struct mlx5_rxq_ctrl *rxq_ctrl;
1282 		unsigned int j;
1283 
1284 		if (!(*priv->rxqs)[i])
1285 			continue;
1286 		rxq_ctrl = container_of((*priv->rxqs)[i],
1287 					struct mlx5_rxq_ctrl, rxq);
1288 		rxq_ctrl->flow_mark_n = 0;
1289 		rxq_ctrl->rxq.mark = 0;
1290 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1291 			rxq_ctrl->flow_tunnels_n[j] = 0;
1292 		rxq_ctrl->rxq.tunnel = 0;
1293 	}
1294 }
1295 
1296 /**
1297  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1298  *
1299  * @param[in] dev
1300  *   Pointer to the Ethernet device structure.
1301  */
1302 void
1303 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1304 {
1305 	struct mlx5_priv *priv = dev->data->dev_private;
1306 	struct mlx5_rxq_data *data;
1307 	unsigned int i;
1308 
1309 	for (i = 0; i != priv->rxqs_n; ++i) {
1310 		if (!(*priv->rxqs)[i])
1311 			continue;
1312 		data = (*priv->rxqs)[i];
1313 		if (!rte_flow_dynf_metadata_avail()) {
1314 			data->dynf_meta = 0;
1315 			data->flow_meta_mask = 0;
1316 			data->flow_meta_offset = -1;
1317 		} else {
1318 			data->dynf_meta = 1;
1319 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1320 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1321 		}
1322 	}
1323 }
1324 
1325 /*
1326  * return a pointer to the desired action in the list of actions.
1327  *
1328  * @param[in] actions
1329  *   The list of actions to search the action in.
1330  * @param[in] action
1331  *   The action to find.
1332  *
1333  * @return
1334  *   Pointer to the action in the list, if found. NULL otherwise.
1335  */
1336 const struct rte_flow_action *
1337 mlx5_flow_find_action(const struct rte_flow_action *actions,
1338 		      enum rte_flow_action_type action)
1339 {
1340 	if (actions == NULL)
1341 		return NULL;
1342 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1343 		if (actions->type == action)
1344 			return actions;
1345 	return NULL;
1346 }
1347 
1348 /*
1349  * Validate the flag action.
1350  *
1351  * @param[in] action_flags
1352  *   Bit-fields that holds the actions detected until now.
1353  * @param[in] attr
1354  *   Attributes of flow that includes this action.
1355  * @param[out] error
1356  *   Pointer to error structure.
1357  *
1358  * @return
1359  *   0 on success, a negative errno value otherwise and rte_errno is set.
1360  */
1361 int
1362 mlx5_flow_validate_action_flag(uint64_t action_flags,
1363 			       const struct rte_flow_attr *attr,
1364 			       struct rte_flow_error *error)
1365 {
1366 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1367 		return rte_flow_error_set(error, EINVAL,
1368 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1369 					  "can't mark and flag in same flow");
1370 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1371 		return rte_flow_error_set(error, EINVAL,
1372 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1373 					  "can't have 2 flag"
1374 					  " actions in same flow");
1375 	if (attr->egress)
1376 		return rte_flow_error_set(error, ENOTSUP,
1377 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1378 					  "flag action not supported for "
1379 					  "egress");
1380 	return 0;
1381 }
1382 
1383 /*
1384  * Validate the mark action.
1385  *
1386  * @param[in] action
1387  *   Pointer to the queue action.
1388  * @param[in] action_flags
1389  *   Bit-fields that holds the actions detected until now.
1390  * @param[in] attr
1391  *   Attributes of flow that includes this action.
1392  * @param[out] error
1393  *   Pointer to error structure.
1394  *
1395  * @return
1396  *   0 on success, a negative errno value otherwise and rte_errno is set.
1397  */
1398 int
1399 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1400 			       uint64_t action_flags,
1401 			       const struct rte_flow_attr *attr,
1402 			       struct rte_flow_error *error)
1403 {
1404 	const struct rte_flow_action_mark *mark = action->conf;
1405 
1406 	if (!mark)
1407 		return rte_flow_error_set(error, EINVAL,
1408 					  RTE_FLOW_ERROR_TYPE_ACTION,
1409 					  action,
1410 					  "configuration cannot be null");
1411 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1412 		return rte_flow_error_set(error, EINVAL,
1413 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1414 					  &mark->id,
1415 					  "mark id must in 0 <= id < "
1416 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1417 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1418 		return rte_flow_error_set(error, EINVAL,
1419 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1420 					  "can't flag and mark in same flow");
1421 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1422 		return rte_flow_error_set(error, EINVAL,
1423 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1424 					  "can't have 2 mark actions in same"
1425 					  " flow");
1426 	if (attr->egress)
1427 		return rte_flow_error_set(error, ENOTSUP,
1428 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1429 					  "mark action not supported for "
1430 					  "egress");
1431 	return 0;
1432 }
1433 
1434 /*
1435  * Validate the drop action.
1436  *
1437  * @param[in] action_flags
1438  *   Bit-fields that holds the actions detected until now.
1439  * @param[in] attr
1440  *   Attributes of flow that includes this action.
1441  * @param[out] error
1442  *   Pointer to error structure.
1443  *
1444  * @return
1445  *   0 on success, a negative errno value otherwise and rte_errno is set.
1446  */
1447 int
1448 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1449 			       const struct rte_flow_attr *attr,
1450 			       struct rte_flow_error *error)
1451 {
1452 	if (attr->egress)
1453 		return rte_flow_error_set(error, ENOTSUP,
1454 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1455 					  "drop action not supported for "
1456 					  "egress");
1457 	return 0;
1458 }
1459 
1460 /*
1461  * Validate the queue action.
1462  *
1463  * @param[in] action
1464  *   Pointer to the queue action.
1465  * @param[in] action_flags
1466  *   Bit-fields that holds the actions detected until now.
1467  * @param[in] dev
1468  *   Pointer to the Ethernet device structure.
1469  * @param[in] attr
1470  *   Attributes of flow that includes this action.
1471  * @param[out] error
1472  *   Pointer to error structure.
1473  *
1474  * @return
1475  *   0 on success, a negative errno value otherwise and rte_errno is set.
1476  */
1477 int
1478 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1479 				uint64_t action_flags,
1480 				struct rte_eth_dev *dev,
1481 				const struct rte_flow_attr *attr,
1482 				struct rte_flow_error *error)
1483 {
1484 	struct mlx5_priv *priv = dev->data->dev_private;
1485 	const struct rte_flow_action_queue *queue = action->conf;
1486 
1487 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1488 		return rte_flow_error_set(error, EINVAL,
1489 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1490 					  "can't have 2 fate actions in"
1491 					  " same flow");
1492 	if (!priv->rxqs_n)
1493 		return rte_flow_error_set(error, EINVAL,
1494 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1495 					  NULL, "No Rx queues configured");
1496 	if (queue->index >= priv->rxqs_n)
1497 		return rte_flow_error_set(error, EINVAL,
1498 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1499 					  &queue->index,
1500 					  "queue index out of range");
1501 	if (!(*priv->rxqs)[queue->index])
1502 		return rte_flow_error_set(error, EINVAL,
1503 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1504 					  &queue->index,
1505 					  "queue is not configured");
1506 	if (attr->egress)
1507 		return rte_flow_error_set(error, ENOTSUP,
1508 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1509 					  "queue action not supported for "
1510 					  "egress");
1511 	return 0;
1512 }
1513 
1514 /*
1515  * Validate the rss action.
1516  *
1517  * @param[in] dev
1518  *   Pointer to the Ethernet device structure.
1519  * @param[in] action
1520  *   Pointer to the queue action.
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_validate_action_rss(struct rte_eth_dev *dev,
1529 			 const struct rte_flow_action *action,
1530 			 struct rte_flow_error *error)
1531 {
1532 	struct mlx5_priv *priv = dev->data->dev_private;
1533 	const struct rte_flow_action_rss *rss = action->conf;
1534 	unsigned int i;
1535 
1536 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1537 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1538 		return rte_flow_error_set(error, ENOTSUP,
1539 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1540 					  &rss->func,
1541 					  "RSS hash function not supported");
1542 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1543 	if (rss->level > 2)
1544 #else
1545 	if (rss->level > 1)
1546 #endif
1547 		return rte_flow_error_set(error, ENOTSUP,
1548 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1549 					  &rss->level,
1550 					  "tunnel RSS is not supported");
1551 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1552 	if (rss->key_len == 0 && rss->key != NULL)
1553 		return rte_flow_error_set(error, ENOTSUP,
1554 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1555 					  &rss->key_len,
1556 					  "RSS hash key length 0");
1557 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1558 		return rte_flow_error_set(error, ENOTSUP,
1559 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1560 					  &rss->key_len,
1561 					  "RSS hash key too small");
1562 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1563 		return rte_flow_error_set(error, ENOTSUP,
1564 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1565 					  &rss->key_len,
1566 					  "RSS hash key too large");
1567 	if (rss->queue_num > priv->config.ind_table_max_size)
1568 		return rte_flow_error_set(error, ENOTSUP,
1569 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1570 					  &rss->queue_num,
1571 					  "number of queues too large");
1572 	if (rss->types & MLX5_RSS_HF_MASK)
1573 		return rte_flow_error_set(error, ENOTSUP,
1574 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1575 					  &rss->types,
1576 					  "some RSS protocols are not"
1577 					  " supported");
1578 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1579 	    !(rss->types & ETH_RSS_IP))
1580 		return rte_flow_error_set(error, EINVAL,
1581 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1582 					  "L3 partial RSS requested but L3 RSS"
1583 					  " type not specified");
1584 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1585 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1586 		return rte_flow_error_set(error, EINVAL,
1587 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1588 					  "L4 partial RSS requested but L4 RSS"
1589 					  " type not specified");
1590 	if (!priv->rxqs_n)
1591 		return rte_flow_error_set(error, EINVAL,
1592 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1593 					  NULL, "No Rx queues configured");
1594 	if (!rss->queue_num)
1595 		return rte_flow_error_set(error, EINVAL,
1596 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1597 					  NULL, "No queues configured");
1598 	for (i = 0; i != rss->queue_num; ++i) {
1599 		if (rss->queue[i] >= priv->rxqs_n)
1600 			return rte_flow_error_set
1601 				(error, EINVAL,
1602 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1603 				 &rss->queue[i], "queue index out of range");
1604 		if (!(*priv->rxqs)[rss->queue[i]])
1605 			return rte_flow_error_set
1606 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1607 				 &rss->queue[i], "queue is not configured");
1608 	}
1609 	return 0;
1610 }
1611 
1612 /*
1613  * Validate the rss action.
1614  *
1615  * @param[in] action
1616  *   Pointer to the queue action.
1617  * @param[in] action_flags
1618  *   Bit-fields that holds the actions detected until now.
1619  * @param[in] dev
1620  *   Pointer to the Ethernet device structure.
1621  * @param[in] attr
1622  *   Attributes of flow that includes this action.
1623  * @param[in] item_flags
1624  *   Items that were detected.
1625  * @param[out] error
1626  *   Pointer to error structure.
1627  *
1628  * @return
1629  *   0 on success, a negative errno value otherwise and rte_errno is set.
1630  */
1631 int
1632 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1633 			      uint64_t action_flags,
1634 			      struct rte_eth_dev *dev,
1635 			      const struct rte_flow_attr *attr,
1636 			      uint64_t item_flags,
1637 			      struct rte_flow_error *error)
1638 {
1639 	const struct rte_flow_action_rss *rss = action->conf;
1640 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1641 	int ret;
1642 
1643 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1644 		return rte_flow_error_set(error, EINVAL,
1645 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1646 					  "can't have 2 fate actions"
1647 					  " in same flow");
1648 	ret = mlx5_validate_action_rss(dev, action, error);
1649 	if (ret)
1650 		return ret;
1651 	if (attr->egress)
1652 		return rte_flow_error_set(error, ENOTSUP,
1653 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1654 					  "rss action not supported for "
1655 					  "egress");
1656 	if (rss->level > 1 && !tunnel)
1657 		return rte_flow_error_set(error, EINVAL,
1658 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1659 					  "inner RSS is not supported for "
1660 					  "non-tunnel flows");
1661 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1662 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1663 		return rte_flow_error_set(error, EINVAL,
1664 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1665 					  "RSS on eCPRI is not supported now");
1666 	}
1667 	return 0;
1668 }
1669 
1670 /*
1671  * Validate the default miss action.
1672  *
1673  * @param[in] action_flags
1674  *   Bit-fields that holds the actions detected until now.
1675  * @param[out] error
1676  *   Pointer to error structure.
1677  *
1678  * @return
1679  *   0 on success, a negative errno value otherwise and rte_errno is set.
1680  */
1681 int
1682 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1683 				const struct rte_flow_attr *attr,
1684 				struct rte_flow_error *error)
1685 {
1686 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1687 		return rte_flow_error_set(error, EINVAL,
1688 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1689 					  "can't have 2 fate actions in"
1690 					  " same flow");
1691 	if (attr->egress)
1692 		return rte_flow_error_set(error, ENOTSUP,
1693 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1694 					  "default miss action not supported "
1695 					  "for egress");
1696 	if (attr->group)
1697 		return rte_flow_error_set(error, ENOTSUP,
1698 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1699 					  "only group 0 is supported");
1700 	if (attr->transfer)
1701 		return rte_flow_error_set(error, ENOTSUP,
1702 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1703 					  NULL, "transfer is not supported");
1704 	return 0;
1705 }
1706 
1707 /*
1708  * Validate the count action.
1709  *
1710  * @param[in] dev
1711  *   Pointer to the Ethernet device structure.
1712  * @param[in] attr
1713  *   Attributes of flow that includes this action.
1714  * @param[out] error
1715  *   Pointer to error structure.
1716  *
1717  * @return
1718  *   0 on success, a negative errno value otherwise and rte_errno is set.
1719  */
1720 int
1721 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1722 				const struct rte_flow_attr *attr,
1723 				struct rte_flow_error *error)
1724 {
1725 	if (attr->egress)
1726 		return rte_flow_error_set(error, ENOTSUP,
1727 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1728 					  "count action not supported for "
1729 					  "egress");
1730 	return 0;
1731 }
1732 
1733 /**
1734  * Verify the @p attributes will be correctly understood by the NIC and store
1735  * them in the @p flow if everything is correct.
1736  *
1737  * @param[in] dev
1738  *   Pointer to the Ethernet device structure.
1739  * @param[in] attributes
1740  *   Pointer to flow attributes
1741  * @param[out] error
1742  *   Pointer to error structure.
1743  *
1744  * @return
1745  *   0 on success, a negative errno value otherwise and rte_errno is set.
1746  */
1747 int
1748 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1749 			      const struct rte_flow_attr *attributes,
1750 			      struct rte_flow_error *error)
1751 {
1752 	struct mlx5_priv *priv = dev->data->dev_private;
1753 	uint32_t priority_max = priv->config.flow_prio - 1;
1754 
1755 	if (attributes->group)
1756 		return rte_flow_error_set(error, ENOTSUP,
1757 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1758 					  NULL, "groups is not supported");
1759 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1760 	    attributes->priority >= priority_max)
1761 		return rte_flow_error_set(error, ENOTSUP,
1762 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1763 					  NULL, "priority out of range");
1764 	if (attributes->egress)
1765 		return rte_flow_error_set(error, ENOTSUP,
1766 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1767 					  "egress is not supported");
1768 	if (attributes->transfer && !priv->config.dv_esw_en)
1769 		return rte_flow_error_set(error, ENOTSUP,
1770 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1771 					  NULL, "transfer is not supported");
1772 	if (!attributes->ingress)
1773 		return rte_flow_error_set(error, EINVAL,
1774 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1775 					  NULL,
1776 					  "ingress attribute is mandatory");
1777 	return 0;
1778 }
1779 
1780 /**
1781  * Validate ICMP6 item.
1782  *
1783  * @param[in] item
1784  *   Item specification.
1785  * @param[in] item_flags
1786  *   Bit-fields that holds the items detected until now.
1787  * @param[in] ext_vlan_sup
1788  *   Whether extended VLAN features are supported or not.
1789  * @param[out] error
1790  *   Pointer to error structure.
1791  *
1792  * @return
1793  *   0 on success, a negative errno value otherwise and rte_errno is set.
1794  */
1795 int
1796 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1797 			       uint64_t item_flags,
1798 			       uint8_t target_protocol,
1799 			       struct rte_flow_error *error)
1800 {
1801 	const struct rte_flow_item_icmp6 *mask = item->mask;
1802 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1803 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1804 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1805 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1806 				      MLX5_FLOW_LAYER_OUTER_L4;
1807 	int ret;
1808 
1809 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1810 		return rte_flow_error_set(error, EINVAL,
1811 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1812 					  "protocol filtering not compatible"
1813 					  " with ICMP6 layer");
1814 	if (!(item_flags & l3m))
1815 		return rte_flow_error_set(error, EINVAL,
1816 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1817 					  "IPv6 is mandatory to filter on"
1818 					  " ICMP6");
1819 	if (item_flags & l4m)
1820 		return rte_flow_error_set(error, EINVAL,
1821 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1822 					  "multiple L4 layers not supported");
1823 	if (!mask)
1824 		mask = &rte_flow_item_icmp6_mask;
1825 	ret = mlx5_flow_item_acceptable
1826 		(item, (const uint8_t *)mask,
1827 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1828 		 sizeof(struct rte_flow_item_icmp6),
1829 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1830 	if (ret < 0)
1831 		return ret;
1832 	return 0;
1833 }
1834 
1835 /**
1836  * Validate ICMP item.
1837  *
1838  * @param[in] item
1839  *   Item specification.
1840  * @param[in] item_flags
1841  *   Bit-fields that holds the items detected until now.
1842  * @param[out] error
1843  *   Pointer to error structure.
1844  *
1845  * @return
1846  *   0 on success, a negative errno value otherwise and rte_errno is set.
1847  */
1848 int
1849 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1850 			     uint64_t item_flags,
1851 			     uint8_t target_protocol,
1852 			     struct rte_flow_error *error)
1853 {
1854 	const struct rte_flow_item_icmp *mask = item->mask;
1855 	const struct rte_flow_item_icmp nic_mask = {
1856 		.hdr.icmp_type = 0xff,
1857 		.hdr.icmp_code = 0xff,
1858 		.hdr.icmp_ident = RTE_BE16(0xffff),
1859 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1860 	};
1861 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1862 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1863 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1864 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1865 				      MLX5_FLOW_LAYER_OUTER_L4;
1866 	int ret;
1867 
1868 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1869 		return rte_flow_error_set(error, EINVAL,
1870 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1871 					  "protocol filtering not compatible"
1872 					  " with ICMP layer");
1873 	if (!(item_flags & l3m))
1874 		return rte_flow_error_set(error, EINVAL,
1875 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1876 					  "IPv4 is mandatory to filter"
1877 					  " on ICMP");
1878 	if (item_flags & l4m)
1879 		return rte_flow_error_set(error, EINVAL,
1880 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 					  "multiple L4 layers not supported");
1882 	if (!mask)
1883 		mask = &nic_mask;
1884 	ret = mlx5_flow_item_acceptable
1885 		(item, (const uint8_t *)mask,
1886 		 (const uint8_t *)&nic_mask,
1887 		 sizeof(struct rte_flow_item_icmp),
1888 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1889 	if (ret < 0)
1890 		return ret;
1891 	return 0;
1892 }
1893 
1894 /**
1895  * Validate Ethernet item.
1896  *
1897  * @param[in] item
1898  *   Item specification.
1899  * @param[in] item_flags
1900  *   Bit-fields that holds the items detected until now.
1901  * @param[out] error
1902  *   Pointer to error structure.
1903  *
1904  * @return
1905  *   0 on success, a negative errno value otherwise and rte_errno is set.
1906  */
1907 int
1908 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1909 			    uint64_t item_flags, bool ext_vlan_sup,
1910 			    struct rte_flow_error *error)
1911 {
1912 	const struct rte_flow_item_eth *mask = item->mask;
1913 	const struct rte_flow_item_eth nic_mask = {
1914 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1915 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1916 		.type = RTE_BE16(0xffff),
1917 		.has_vlan = ext_vlan_sup ? 1 : 0,
1918 	};
1919 	int ret;
1920 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1921 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1922 				       MLX5_FLOW_LAYER_OUTER_L2;
1923 
1924 	if (item_flags & ethm)
1925 		return rte_flow_error_set(error, ENOTSUP,
1926 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 					  "multiple L2 layers not supported");
1928 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1929 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1930 		return rte_flow_error_set(error, EINVAL,
1931 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1932 					  "L2 layer should not follow "
1933 					  "L3 layers");
1934 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1935 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1936 		return rte_flow_error_set(error, EINVAL,
1937 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1938 					  "L2 layer should not follow VLAN");
1939 	if (!mask)
1940 		mask = &rte_flow_item_eth_mask;
1941 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1942 					(const uint8_t *)&nic_mask,
1943 					sizeof(struct rte_flow_item_eth),
1944 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1945 	return ret;
1946 }
1947 
1948 /**
1949  * Validate VLAN item.
1950  *
1951  * @param[in] item
1952  *   Item specification.
1953  * @param[in] item_flags
1954  *   Bit-fields that holds the items detected until now.
1955  * @param[in] dev
1956  *   Ethernet device flow is being created on.
1957  * @param[out] error
1958  *   Pointer to error structure.
1959  *
1960  * @return
1961  *   0 on success, a negative errno value otherwise and rte_errno is set.
1962  */
1963 int
1964 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1965 			     uint64_t item_flags,
1966 			     struct rte_eth_dev *dev,
1967 			     struct rte_flow_error *error)
1968 {
1969 	const struct rte_flow_item_vlan *spec = item->spec;
1970 	const struct rte_flow_item_vlan *mask = item->mask;
1971 	const struct rte_flow_item_vlan nic_mask = {
1972 		.tci = RTE_BE16(UINT16_MAX),
1973 		.inner_type = RTE_BE16(UINT16_MAX),
1974 	};
1975 	uint16_t vlan_tag = 0;
1976 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1977 	int ret;
1978 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1979 					MLX5_FLOW_LAYER_INNER_L4) :
1980 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1981 					MLX5_FLOW_LAYER_OUTER_L4);
1982 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1983 					MLX5_FLOW_LAYER_OUTER_VLAN;
1984 
1985 	if (item_flags & vlanm)
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1988 					  "multiple VLAN layers not supported");
1989 	else if ((item_flags & l34m) != 0)
1990 		return rte_flow_error_set(error, EINVAL,
1991 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1992 					  "VLAN cannot follow L3/L4 layer");
1993 	if (!mask)
1994 		mask = &rte_flow_item_vlan_mask;
1995 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1996 					(const uint8_t *)&nic_mask,
1997 					sizeof(struct rte_flow_item_vlan),
1998 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1999 	if (ret)
2000 		return ret;
2001 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2002 		struct mlx5_priv *priv = dev->data->dev_private;
2003 
2004 		if (priv->vmwa_context) {
2005 			/*
2006 			 * Non-NULL context means we have a virtual machine
2007 			 * and SR-IOV enabled, we have to create VLAN interface
2008 			 * to make hypervisor to setup E-Switch vport
2009 			 * context correctly. We avoid creating the multiple
2010 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2011 			 */
2012 			return rte_flow_error_set(error, EINVAL,
2013 						  RTE_FLOW_ERROR_TYPE_ITEM,
2014 						  item,
2015 						  "VLAN tag mask is not"
2016 						  " supported in virtual"
2017 						  " environment");
2018 		}
2019 	}
2020 	if (spec) {
2021 		vlan_tag = spec->tci;
2022 		vlan_tag &= mask->tci;
2023 	}
2024 	/*
2025 	 * From verbs perspective an empty VLAN is equivalent
2026 	 * to a packet without VLAN layer.
2027 	 */
2028 	if (!vlan_tag)
2029 		return rte_flow_error_set(error, EINVAL,
2030 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2031 					  item->spec,
2032 					  "VLAN cannot be empty");
2033 	return 0;
2034 }
2035 
2036 /**
2037  * Validate IPV4 item.
2038  *
2039  * @param[in] item
2040  *   Item specification.
2041  * @param[in] item_flags
2042  *   Bit-fields that holds the items detected until now.
2043  * @param[in] last_item
2044  *   Previous validated item in the pattern items.
2045  * @param[in] ether_type
2046  *   Type in the ethernet layer header (including dot1q).
2047  * @param[in] acc_mask
2048  *   Acceptable mask, if NULL default internal default mask
2049  *   will be used to check whether item fields are supported.
2050  * @param[in] range_accepted
2051  *   True if range of values is accepted for specific fields, false otherwise.
2052  * @param[out] error
2053  *   Pointer to error structure.
2054  *
2055  * @return
2056  *   0 on success, a negative errno value otherwise and rte_errno is set.
2057  */
2058 int
2059 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2060 			     uint64_t item_flags,
2061 			     uint64_t last_item,
2062 			     uint16_t ether_type,
2063 			     const struct rte_flow_item_ipv4 *acc_mask,
2064 			     bool range_accepted,
2065 			     struct rte_flow_error *error)
2066 {
2067 	const struct rte_flow_item_ipv4 *mask = item->mask;
2068 	const struct rte_flow_item_ipv4 *spec = item->spec;
2069 	const struct rte_flow_item_ipv4 nic_mask = {
2070 		.hdr = {
2071 			.src_addr = RTE_BE32(0xffffffff),
2072 			.dst_addr = RTE_BE32(0xffffffff),
2073 			.type_of_service = 0xff,
2074 			.next_proto_id = 0xff,
2075 		},
2076 	};
2077 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2078 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2079 				      MLX5_FLOW_LAYER_OUTER_L3;
2080 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2081 				      MLX5_FLOW_LAYER_OUTER_L4;
2082 	int ret;
2083 	uint8_t next_proto = 0xFF;
2084 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2085 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2086 				  MLX5_FLOW_LAYER_INNER_VLAN);
2087 
2088 	if ((last_item & l2_vlan) && ether_type &&
2089 	    ether_type != RTE_ETHER_TYPE_IPV4)
2090 		return rte_flow_error_set(error, EINVAL,
2091 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2092 					  "IPv4 cannot follow L2/VLAN layer "
2093 					  "which ether type is not IPv4");
2094 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2095 		if (mask && spec)
2096 			next_proto = mask->hdr.next_proto_id &
2097 				     spec->hdr.next_proto_id;
2098 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2099 			return rte_flow_error_set(error, EINVAL,
2100 						  RTE_FLOW_ERROR_TYPE_ITEM,
2101 						  item,
2102 						  "multiple tunnel "
2103 						  "not supported");
2104 	}
2105 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2106 		return rte_flow_error_set(error, EINVAL,
2107 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2108 					  "wrong tunnel type - IPv6 specified "
2109 					  "but IPv4 item provided");
2110 	if (item_flags & l3m)
2111 		return rte_flow_error_set(error, ENOTSUP,
2112 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2113 					  "multiple L3 layers not supported");
2114 	else if (item_flags & l4m)
2115 		return rte_flow_error_set(error, EINVAL,
2116 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2117 					  "L3 cannot follow an L4 layer.");
2118 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2119 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2120 		return rte_flow_error_set(error, EINVAL,
2121 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2122 					  "L3 cannot follow an NVGRE layer.");
2123 	if (!mask)
2124 		mask = &rte_flow_item_ipv4_mask;
2125 	else if (mask->hdr.next_proto_id != 0 &&
2126 		 mask->hdr.next_proto_id != 0xff)
2127 		return rte_flow_error_set(error, EINVAL,
2128 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2129 					  "partial mask is not supported"
2130 					  " for protocol");
2131 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2132 					acc_mask ? (const uint8_t *)acc_mask
2133 						 : (const uint8_t *)&nic_mask,
2134 					sizeof(struct rte_flow_item_ipv4),
2135 					range_accepted, error);
2136 	if (ret < 0)
2137 		return ret;
2138 	return 0;
2139 }
2140 
2141 /**
2142  * Validate IPV6 item.
2143  *
2144  * @param[in] item
2145  *   Item specification.
2146  * @param[in] item_flags
2147  *   Bit-fields that holds the items detected until now.
2148  * @param[in] last_item
2149  *   Previous validated item in the pattern items.
2150  * @param[in] ether_type
2151  *   Type in the ethernet layer header (including dot1q).
2152  * @param[in] acc_mask
2153  *   Acceptable mask, if NULL default internal default mask
2154  *   will be used to check whether item fields are supported.
2155  * @param[out] error
2156  *   Pointer to error structure.
2157  *
2158  * @return
2159  *   0 on success, a negative errno value otherwise and rte_errno is set.
2160  */
2161 int
2162 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2163 			     uint64_t item_flags,
2164 			     uint64_t last_item,
2165 			     uint16_t ether_type,
2166 			     const struct rte_flow_item_ipv6 *acc_mask,
2167 			     struct rte_flow_error *error)
2168 {
2169 	const struct rte_flow_item_ipv6 *mask = item->mask;
2170 	const struct rte_flow_item_ipv6 *spec = item->spec;
2171 	const struct rte_flow_item_ipv6 nic_mask = {
2172 		.hdr = {
2173 			.src_addr =
2174 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2175 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2176 			.dst_addr =
2177 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2178 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2179 			.vtc_flow = RTE_BE32(0xffffffff),
2180 			.proto = 0xff,
2181 		},
2182 	};
2183 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2184 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2185 				      MLX5_FLOW_LAYER_OUTER_L3;
2186 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2187 				      MLX5_FLOW_LAYER_OUTER_L4;
2188 	int ret;
2189 	uint8_t next_proto = 0xFF;
2190 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2191 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2192 				  MLX5_FLOW_LAYER_INNER_VLAN);
2193 
2194 	if ((last_item & l2_vlan) && ether_type &&
2195 	    ether_type != RTE_ETHER_TYPE_IPV6)
2196 		return rte_flow_error_set(error, EINVAL,
2197 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2198 					  "IPv6 cannot follow L2/VLAN layer "
2199 					  "which ether type is not IPv6");
2200 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2201 		next_proto = spec->hdr.proto;
2202 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2203 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2204 			return rte_flow_error_set(error, EINVAL,
2205 						  RTE_FLOW_ERROR_TYPE_ITEM,
2206 						  item,
2207 						  "multiple tunnel "
2208 						  "not supported");
2209 	}
2210 	if (next_proto == IPPROTO_HOPOPTS  ||
2211 	    next_proto == IPPROTO_ROUTING  ||
2212 	    next_proto == IPPROTO_FRAGMENT ||
2213 	    next_proto == IPPROTO_ESP	   ||
2214 	    next_proto == IPPROTO_AH	   ||
2215 	    next_proto == IPPROTO_DSTOPTS)
2216 		return rte_flow_error_set(error, EINVAL,
2217 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2218 					  "IPv6 proto (next header) should "
2219 					  "not be set as extension header");
2220 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2221 		return rte_flow_error_set(error, EINVAL,
2222 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2223 					  "wrong tunnel type - IPv4 specified "
2224 					  "but IPv6 item provided");
2225 	if (item_flags & l3m)
2226 		return rte_flow_error_set(error, ENOTSUP,
2227 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2228 					  "multiple L3 layers not supported");
2229 	else if (item_flags & l4m)
2230 		return rte_flow_error_set(error, EINVAL,
2231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2232 					  "L3 cannot follow an L4 layer.");
2233 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2234 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2235 		return rte_flow_error_set(error, EINVAL,
2236 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2237 					  "L3 cannot follow an NVGRE layer.");
2238 	if (!mask)
2239 		mask = &rte_flow_item_ipv6_mask;
2240 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2241 					acc_mask ? (const uint8_t *)acc_mask
2242 						 : (const uint8_t *)&nic_mask,
2243 					sizeof(struct rte_flow_item_ipv6),
2244 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2245 	if (ret < 0)
2246 		return ret;
2247 	return 0;
2248 }
2249 
2250 /**
2251  * Validate UDP item.
2252  *
2253  * @param[in] item
2254  *   Item specification.
2255  * @param[in] item_flags
2256  *   Bit-fields that holds the items detected until now.
2257  * @param[in] target_protocol
2258  *   The next protocol in the previous item.
2259  * @param[in] flow_mask
2260  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2261  * @param[out] error
2262  *   Pointer to error structure.
2263  *
2264  * @return
2265  *   0 on success, a negative errno value otherwise and rte_errno is set.
2266  */
2267 int
2268 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2269 			    uint64_t item_flags,
2270 			    uint8_t target_protocol,
2271 			    struct rte_flow_error *error)
2272 {
2273 	const struct rte_flow_item_udp *mask = item->mask;
2274 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2275 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2276 				      MLX5_FLOW_LAYER_OUTER_L3;
2277 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2278 				      MLX5_FLOW_LAYER_OUTER_L4;
2279 	int ret;
2280 
2281 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2282 		return rte_flow_error_set(error, EINVAL,
2283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2284 					  "protocol filtering not compatible"
2285 					  " with UDP layer");
2286 	if (!(item_flags & l3m))
2287 		return rte_flow_error_set(error, EINVAL,
2288 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2289 					  "L3 is mandatory to filter on L4");
2290 	if (item_flags & l4m)
2291 		return rte_flow_error_set(error, EINVAL,
2292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 					  "multiple L4 layers not supported");
2294 	if (!mask)
2295 		mask = &rte_flow_item_udp_mask;
2296 	ret = mlx5_flow_item_acceptable
2297 		(item, (const uint8_t *)mask,
2298 		 (const uint8_t *)&rte_flow_item_udp_mask,
2299 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2300 		 error);
2301 	if (ret < 0)
2302 		return ret;
2303 	return 0;
2304 }
2305 
2306 /**
2307  * Validate TCP item.
2308  *
2309  * @param[in] item
2310  *   Item specification.
2311  * @param[in] item_flags
2312  *   Bit-fields that holds the items detected until now.
2313  * @param[in] target_protocol
2314  *   The next protocol in the previous item.
2315  * @param[out] error
2316  *   Pointer to error structure.
2317  *
2318  * @return
2319  *   0 on success, a negative errno value otherwise and rte_errno is set.
2320  */
2321 int
2322 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2323 			    uint64_t item_flags,
2324 			    uint8_t target_protocol,
2325 			    const struct rte_flow_item_tcp *flow_mask,
2326 			    struct rte_flow_error *error)
2327 {
2328 	const struct rte_flow_item_tcp *mask = item->mask;
2329 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2330 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2331 				      MLX5_FLOW_LAYER_OUTER_L3;
2332 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2333 				      MLX5_FLOW_LAYER_OUTER_L4;
2334 	int ret;
2335 
2336 	MLX5_ASSERT(flow_mask);
2337 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2338 		return rte_flow_error_set(error, EINVAL,
2339 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2340 					  "protocol filtering not compatible"
2341 					  " with TCP layer");
2342 	if (!(item_flags & l3m))
2343 		return rte_flow_error_set(error, EINVAL,
2344 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2345 					  "L3 is mandatory to filter on L4");
2346 	if (item_flags & l4m)
2347 		return rte_flow_error_set(error, EINVAL,
2348 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2349 					  "multiple L4 layers not supported");
2350 	if (!mask)
2351 		mask = &rte_flow_item_tcp_mask;
2352 	ret = mlx5_flow_item_acceptable
2353 		(item, (const uint8_t *)mask,
2354 		 (const uint8_t *)flow_mask,
2355 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2356 		 error);
2357 	if (ret < 0)
2358 		return ret;
2359 	return 0;
2360 }
2361 
2362 /**
2363  * Validate VXLAN item.
2364  *
2365  * @param[in] item
2366  *   Item specification.
2367  * @param[in] item_flags
2368  *   Bit-fields that holds the items detected until now.
2369  * @param[in] target_protocol
2370  *   The next protocol in the previous item.
2371  * @param[out] error
2372  *   Pointer to error structure.
2373  *
2374  * @return
2375  *   0 on success, a negative errno value otherwise and rte_errno is set.
2376  */
2377 int
2378 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2379 			      uint64_t item_flags,
2380 			      struct rte_flow_error *error)
2381 {
2382 	const struct rte_flow_item_vxlan *spec = item->spec;
2383 	const struct rte_flow_item_vxlan *mask = item->mask;
2384 	int ret;
2385 	union vni {
2386 		uint32_t vlan_id;
2387 		uint8_t vni[4];
2388 	} id = { .vlan_id = 0, };
2389 
2390 
2391 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2392 		return rte_flow_error_set(error, ENOTSUP,
2393 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2394 					  "multiple tunnel layers not"
2395 					  " supported");
2396 	/*
2397 	 * Verify only UDPv4 is present as defined in
2398 	 * https://tools.ietf.org/html/rfc7348
2399 	 */
2400 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2401 		return rte_flow_error_set(error, EINVAL,
2402 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2403 					  "no outer UDP layer found");
2404 	if (!mask)
2405 		mask = &rte_flow_item_vxlan_mask;
2406 	ret = mlx5_flow_item_acceptable
2407 		(item, (const uint8_t *)mask,
2408 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2409 		 sizeof(struct rte_flow_item_vxlan),
2410 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2411 	if (ret < 0)
2412 		return ret;
2413 	if (spec) {
2414 		memcpy(&id.vni[1], spec->vni, 3);
2415 		memcpy(&id.vni[1], mask->vni, 3);
2416 	}
2417 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2418 		return rte_flow_error_set(error, ENOTSUP,
2419 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2420 					  "VXLAN tunnel must be fully defined");
2421 	return 0;
2422 }
2423 
2424 /**
2425  * Validate VXLAN_GPE item.
2426  *
2427  * @param[in] item
2428  *   Item specification.
2429  * @param[in] item_flags
2430  *   Bit-fields that holds the items detected until now.
2431  * @param[in] priv
2432  *   Pointer to the private data structure.
2433  * @param[in] target_protocol
2434  *   The next protocol in the previous item.
2435  * @param[out] error
2436  *   Pointer to error structure.
2437  *
2438  * @return
2439  *   0 on success, a negative errno value otherwise and rte_errno is set.
2440  */
2441 int
2442 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2443 				  uint64_t item_flags,
2444 				  struct rte_eth_dev *dev,
2445 				  struct rte_flow_error *error)
2446 {
2447 	struct mlx5_priv *priv = dev->data->dev_private;
2448 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2449 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2450 	int ret;
2451 	union vni {
2452 		uint32_t vlan_id;
2453 		uint8_t vni[4];
2454 	} id = { .vlan_id = 0, };
2455 
2456 	if (!priv->config.l3_vxlan_en)
2457 		return rte_flow_error_set(error, ENOTSUP,
2458 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2459 					  "L3 VXLAN is not enabled by device"
2460 					  " parameter and/or not configured in"
2461 					  " firmware");
2462 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2463 		return rte_flow_error_set(error, ENOTSUP,
2464 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2465 					  "multiple tunnel layers not"
2466 					  " supported");
2467 	/*
2468 	 * Verify only UDPv4 is present as defined in
2469 	 * https://tools.ietf.org/html/rfc7348
2470 	 */
2471 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2472 		return rte_flow_error_set(error, EINVAL,
2473 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2474 					  "no outer UDP layer found");
2475 	if (!mask)
2476 		mask = &rte_flow_item_vxlan_gpe_mask;
2477 	ret = mlx5_flow_item_acceptable
2478 		(item, (const uint8_t *)mask,
2479 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2480 		 sizeof(struct rte_flow_item_vxlan_gpe),
2481 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2482 	if (ret < 0)
2483 		return ret;
2484 	if (spec) {
2485 		if (spec->protocol)
2486 			return rte_flow_error_set(error, ENOTSUP,
2487 						  RTE_FLOW_ERROR_TYPE_ITEM,
2488 						  item,
2489 						  "VxLAN-GPE protocol"
2490 						  " not supported");
2491 		memcpy(&id.vni[1], spec->vni, 3);
2492 		memcpy(&id.vni[1], mask->vni, 3);
2493 	}
2494 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2495 		return rte_flow_error_set(error, ENOTSUP,
2496 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2497 					  "VXLAN-GPE tunnel must be fully"
2498 					  " defined");
2499 	return 0;
2500 }
2501 /**
2502  * Validate GRE Key item.
2503  *
2504  * @param[in] item
2505  *   Item specification.
2506  * @param[in] item_flags
2507  *   Bit flags to mark detected items.
2508  * @param[in] gre_item
2509  *   Pointer to gre_item
2510  * @param[out] error
2511  *   Pointer to error structure.
2512  *
2513  * @return
2514  *   0 on success, a negative errno value otherwise and rte_errno is set.
2515  */
2516 int
2517 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2518 				uint64_t item_flags,
2519 				const struct rte_flow_item *gre_item,
2520 				struct rte_flow_error *error)
2521 {
2522 	const rte_be32_t *mask = item->mask;
2523 	int ret = 0;
2524 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2525 	const struct rte_flow_item_gre *gre_spec;
2526 	const struct rte_flow_item_gre *gre_mask;
2527 
2528 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2529 		return rte_flow_error_set(error, ENOTSUP,
2530 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2531 					  "Multiple GRE key not support");
2532 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2533 		return rte_flow_error_set(error, ENOTSUP,
2534 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2535 					  "No preceding GRE header");
2536 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2537 		return rte_flow_error_set(error, ENOTSUP,
2538 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2539 					  "GRE key following a wrong item");
2540 	gre_mask = gre_item->mask;
2541 	if (!gre_mask)
2542 		gre_mask = &rte_flow_item_gre_mask;
2543 	gre_spec = gre_item->spec;
2544 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2545 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2546 		return rte_flow_error_set(error, EINVAL,
2547 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2548 					  "Key bit must be on");
2549 
2550 	if (!mask)
2551 		mask = &gre_key_default_mask;
2552 	ret = mlx5_flow_item_acceptable
2553 		(item, (const uint8_t *)mask,
2554 		 (const uint8_t *)&gre_key_default_mask,
2555 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2556 	return ret;
2557 }
2558 
2559 /**
2560  * Validate GRE item.
2561  *
2562  * @param[in] item
2563  *   Item specification.
2564  * @param[in] item_flags
2565  *   Bit flags to mark detected items.
2566  * @param[in] target_protocol
2567  *   The next protocol in the previous item.
2568  * @param[out] error
2569  *   Pointer to error structure.
2570  *
2571  * @return
2572  *   0 on success, a negative errno value otherwise and rte_errno is set.
2573  */
2574 int
2575 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2576 			    uint64_t item_flags,
2577 			    uint8_t target_protocol,
2578 			    struct rte_flow_error *error)
2579 {
2580 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2581 	const struct rte_flow_item_gre *mask = item->mask;
2582 	int ret;
2583 	const struct rte_flow_item_gre nic_mask = {
2584 		.c_rsvd0_ver = RTE_BE16(0xB000),
2585 		.protocol = RTE_BE16(UINT16_MAX),
2586 	};
2587 
2588 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2589 		return rte_flow_error_set(error, EINVAL,
2590 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2591 					  "protocol filtering not compatible"
2592 					  " with this GRE layer");
2593 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2594 		return rte_flow_error_set(error, ENOTSUP,
2595 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2596 					  "multiple tunnel layers not"
2597 					  " supported");
2598 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2599 		return rte_flow_error_set(error, ENOTSUP,
2600 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2601 					  "L3 Layer is missing");
2602 	if (!mask)
2603 		mask = &rte_flow_item_gre_mask;
2604 	ret = mlx5_flow_item_acceptable
2605 		(item, (const uint8_t *)mask,
2606 		 (const uint8_t *)&nic_mask,
2607 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2608 		 error);
2609 	if (ret < 0)
2610 		return ret;
2611 #ifndef HAVE_MLX5DV_DR
2612 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2613 	if (spec && (spec->protocol & mask->protocol))
2614 		return rte_flow_error_set(error, ENOTSUP,
2615 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2616 					  "without MPLS support the"
2617 					  " specification cannot be used for"
2618 					  " filtering");
2619 #endif
2620 #endif
2621 	return 0;
2622 }
2623 
2624 /**
2625  * Validate Geneve item.
2626  *
2627  * @param[in] item
2628  *   Item specification.
2629  * @param[in] itemFlags
2630  *   Bit-fields that holds the items detected until now.
2631  * @param[in] enPriv
2632  *   Pointer to the private data structure.
2633  * @param[out] error
2634  *   Pointer to error structure.
2635  *
2636  * @return
2637  *   0 on success, a negative errno value otherwise and rte_errno is set.
2638  */
2639 
2640 int
2641 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2642 			       uint64_t item_flags,
2643 			       struct rte_eth_dev *dev,
2644 			       struct rte_flow_error *error)
2645 {
2646 	struct mlx5_priv *priv = dev->data->dev_private;
2647 	const struct rte_flow_item_geneve *spec = item->spec;
2648 	const struct rte_flow_item_geneve *mask = item->mask;
2649 	int ret;
2650 	uint16_t gbhdr;
2651 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2652 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2653 	const struct rte_flow_item_geneve nic_mask = {
2654 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2655 		.vni = "\xff\xff\xff",
2656 		.protocol = RTE_BE16(UINT16_MAX),
2657 	};
2658 
2659 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2660 		return rte_flow_error_set(error, ENOTSUP,
2661 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2662 					  "L3 Geneve is not enabled by device"
2663 					  " parameter and/or not configured in"
2664 					  " firmware");
2665 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2666 		return rte_flow_error_set(error, ENOTSUP,
2667 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2668 					  "multiple tunnel layers not"
2669 					  " supported");
2670 	/*
2671 	 * Verify only UDPv4 is present as defined in
2672 	 * https://tools.ietf.org/html/rfc7348
2673 	 */
2674 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2675 		return rte_flow_error_set(error, EINVAL,
2676 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2677 					  "no outer UDP layer found");
2678 	if (!mask)
2679 		mask = &rte_flow_item_geneve_mask;
2680 	ret = mlx5_flow_item_acceptable
2681 				  (item, (const uint8_t *)mask,
2682 				   (const uint8_t *)&nic_mask,
2683 				   sizeof(struct rte_flow_item_geneve),
2684 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2685 	if (ret)
2686 		return ret;
2687 	if (spec) {
2688 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2689 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2690 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2691 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2692 			return rte_flow_error_set(error, ENOTSUP,
2693 						  RTE_FLOW_ERROR_TYPE_ITEM,
2694 						  item,
2695 						  "Geneve protocol unsupported"
2696 						  " fields are being used");
2697 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2698 			return rte_flow_error_set
2699 					(error, ENOTSUP,
2700 					 RTE_FLOW_ERROR_TYPE_ITEM,
2701 					 item,
2702 					 "Unsupported Geneve options length");
2703 	}
2704 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2705 		return rte_flow_error_set
2706 				    (error, ENOTSUP,
2707 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2708 				     "Geneve tunnel must be fully defined");
2709 	return 0;
2710 }
2711 
2712 /**
2713  * Validate MPLS item.
2714  *
2715  * @param[in] dev
2716  *   Pointer to the rte_eth_dev structure.
2717  * @param[in] item
2718  *   Item specification.
2719  * @param[in] item_flags
2720  *   Bit-fields that holds the items detected until now.
2721  * @param[in] prev_layer
2722  *   The protocol layer indicated in previous item.
2723  * @param[out] error
2724  *   Pointer to error structure.
2725  *
2726  * @return
2727  *   0 on success, a negative errno value otherwise and rte_errno is set.
2728  */
2729 int
2730 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2731 			     const struct rte_flow_item *item __rte_unused,
2732 			     uint64_t item_flags __rte_unused,
2733 			     uint64_t prev_layer __rte_unused,
2734 			     struct rte_flow_error *error)
2735 {
2736 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2737 	const struct rte_flow_item_mpls *mask = item->mask;
2738 	struct mlx5_priv *priv = dev->data->dev_private;
2739 	int ret;
2740 
2741 	if (!priv->config.mpls_en)
2742 		return rte_flow_error_set(error, ENOTSUP,
2743 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2744 					  "MPLS not supported or"
2745 					  " disabled in firmware"
2746 					  " configuration.");
2747 	/* MPLS over IP, UDP, GRE is allowed */
2748 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2749 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2750 			    MLX5_FLOW_LAYER_GRE |
2751 			    MLX5_FLOW_LAYER_GRE_KEY)))
2752 		return rte_flow_error_set(error, EINVAL,
2753 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2754 					  "protocol filtering not compatible"
2755 					  " with MPLS layer");
2756 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2757 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2758 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2759 		return rte_flow_error_set(error, ENOTSUP,
2760 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2761 					  "multiple tunnel layers not"
2762 					  " supported");
2763 	if (!mask)
2764 		mask = &rte_flow_item_mpls_mask;
2765 	ret = mlx5_flow_item_acceptable
2766 		(item, (const uint8_t *)mask,
2767 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2768 		 sizeof(struct rte_flow_item_mpls),
2769 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2770 	if (ret < 0)
2771 		return ret;
2772 	return 0;
2773 #else
2774 	return rte_flow_error_set(error, ENOTSUP,
2775 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2776 				  "MPLS is not supported by Verbs, please"
2777 				  " update.");
2778 #endif
2779 }
2780 
2781 /**
2782  * Validate NVGRE item.
2783  *
2784  * @param[in] item
2785  *   Item specification.
2786  * @param[in] item_flags
2787  *   Bit flags to mark detected items.
2788  * @param[in] target_protocol
2789  *   The next protocol in the previous item.
2790  * @param[out] error
2791  *   Pointer to error structure.
2792  *
2793  * @return
2794  *   0 on success, a negative errno value otherwise and rte_errno is set.
2795  */
2796 int
2797 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2798 			      uint64_t item_flags,
2799 			      uint8_t target_protocol,
2800 			      struct rte_flow_error *error)
2801 {
2802 	const struct rte_flow_item_nvgre *mask = item->mask;
2803 	int ret;
2804 
2805 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2806 		return rte_flow_error_set(error, EINVAL,
2807 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2808 					  "protocol filtering not compatible"
2809 					  " with this GRE layer");
2810 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2811 		return rte_flow_error_set(error, ENOTSUP,
2812 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2813 					  "multiple tunnel layers not"
2814 					  " supported");
2815 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2816 		return rte_flow_error_set(error, ENOTSUP,
2817 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2818 					  "L3 Layer is missing");
2819 	if (!mask)
2820 		mask = &rte_flow_item_nvgre_mask;
2821 	ret = mlx5_flow_item_acceptable
2822 		(item, (const uint8_t *)mask,
2823 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2824 		 sizeof(struct rte_flow_item_nvgre),
2825 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2826 	if (ret < 0)
2827 		return ret;
2828 	return 0;
2829 }
2830 
2831 /**
2832  * Validate eCPRI item.
2833  *
2834  * @param[in] item
2835  *   Item specification.
2836  * @param[in] item_flags
2837  *   Bit-fields that holds the items detected until now.
2838  * @param[in] last_item
2839  *   Previous validated item in the pattern items.
2840  * @param[in] ether_type
2841  *   Type in the ethernet layer header (including dot1q).
2842  * @param[in] acc_mask
2843  *   Acceptable mask, if NULL default internal default mask
2844  *   will be used to check whether item fields are supported.
2845  * @param[out] error
2846  *   Pointer to error structure.
2847  *
2848  * @return
2849  *   0 on success, a negative errno value otherwise and rte_errno is set.
2850  */
2851 int
2852 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2853 			      uint64_t item_flags,
2854 			      uint64_t last_item,
2855 			      uint16_t ether_type,
2856 			      const struct rte_flow_item_ecpri *acc_mask,
2857 			      struct rte_flow_error *error)
2858 {
2859 	const struct rte_flow_item_ecpri *mask = item->mask;
2860 	const struct rte_flow_item_ecpri nic_mask = {
2861 		.hdr = {
2862 			.common = {
2863 				.u32 =
2864 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2865 					.type = 0xFF,
2866 					}).u32),
2867 			},
2868 			.dummy[0] = 0xFFFFFFFF,
2869 		},
2870 	};
2871 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2872 					MLX5_FLOW_LAYER_OUTER_VLAN);
2873 	struct rte_flow_item_ecpri mask_lo;
2874 
2875 	if (!(last_item & outer_l2_vlan) &&
2876 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2877 		return rte_flow_error_set(error, EINVAL,
2878 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2879 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2880 	if ((last_item & outer_l2_vlan) && ether_type &&
2881 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2882 		return rte_flow_error_set(error, EINVAL,
2883 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2884 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2885 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2886 		return rte_flow_error_set(error, EINVAL,
2887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2888 					  "eCPRI with tunnel is not supported right now");
2889 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2890 		return rte_flow_error_set(error, ENOTSUP,
2891 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2892 					  "multiple L3 layers not supported");
2893 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2894 		return rte_flow_error_set(error, EINVAL,
2895 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2896 					  "eCPRI cannot coexist with a TCP layer");
2897 	/* In specification, eCPRI could be over UDP layer. */
2898 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2899 		return rte_flow_error_set(error, EINVAL,
2900 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2901 					  "eCPRI over UDP layer is not yet supported right now");
2902 	/* Mask for type field in common header could be zero. */
2903 	if (!mask)
2904 		mask = &rte_flow_item_ecpri_mask;
2905 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2906 	/* Input mask is in big-endian format. */
2907 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2908 		return rte_flow_error_set(error, EINVAL,
2909 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2910 					  "partial mask is not supported for protocol");
2911 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2912 		return rte_flow_error_set(error, EINVAL,
2913 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2914 					  "message header mask must be after a type mask");
2915 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2916 					 acc_mask ? (const uint8_t *)acc_mask
2917 						  : (const uint8_t *)&nic_mask,
2918 					 sizeof(struct rte_flow_item_ecpri),
2919 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2920 }
2921 
2922 /**
2923  * Release resource related QUEUE/RSS action split.
2924  *
2925  * @param dev
2926  *   Pointer to Ethernet device.
2927  * @param flow
2928  *   Flow to release id's from.
2929  */
2930 static void
2931 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2932 			     struct rte_flow *flow)
2933 {
2934 	struct mlx5_priv *priv = dev->data->dev_private;
2935 	uint32_t handle_idx;
2936 	struct mlx5_flow_handle *dev_handle;
2937 
2938 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2939 		       handle_idx, dev_handle, next)
2940 		if (dev_handle->split_flow_id)
2941 			mlx5_ipool_free(priv->sh->ipool
2942 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2943 					dev_handle->split_flow_id);
2944 }
2945 
2946 static int
2947 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2948 		   const struct rte_flow_attr *attr __rte_unused,
2949 		   const struct rte_flow_item items[] __rte_unused,
2950 		   const struct rte_flow_action actions[] __rte_unused,
2951 		   bool external __rte_unused,
2952 		   int hairpin __rte_unused,
2953 		   struct rte_flow_error *error)
2954 {
2955 	return rte_flow_error_set(error, ENOTSUP,
2956 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2957 }
2958 
2959 static struct mlx5_flow *
2960 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2961 		  const struct rte_flow_attr *attr __rte_unused,
2962 		  const struct rte_flow_item items[] __rte_unused,
2963 		  const struct rte_flow_action actions[] __rte_unused,
2964 		  struct rte_flow_error *error)
2965 {
2966 	rte_flow_error_set(error, ENOTSUP,
2967 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2968 	return NULL;
2969 }
2970 
2971 static int
2972 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2973 		    struct mlx5_flow *dev_flow __rte_unused,
2974 		    const struct rte_flow_attr *attr __rte_unused,
2975 		    const struct rte_flow_item items[] __rte_unused,
2976 		    const struct rte_flow_action actions[] __rte_unused,
2977 		    struct rte_flow_error *error)
2978 {
2979 	return rte_flow_error_set(error, ENOTSUP,
2980 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2981 }
2982 
2983 static int
2984 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2985 		struct rte_flow *flow __rte_unused,
2986 		struct rte_flow_error *error)
2987 {
2988 	return rte_flow_error_set(error, ENOTSUP,
2989 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2990 }
2991 
2992 static void
2993 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2994 		 struct rte_flow *flow __rte_unused)
2995 {
2996 }
2997 
2998 static void
2999 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3000 		  struct rte_flow *flow __rte_unused)
3001 {
3002 }
3003 
3004 static int
3005 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3006 		struct rte_flow *flow __rte_unused,
3007 		const struct rte_flow_action *actions __rte_unused,
3008 		void *data __rte_unused,
3009 		struct rte_flow_error *error)
3010 {
3011 	return rte_flow_error_set(error, ENOTSUP,
3012 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3013 }
3014 
3015 static int
3016 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3017 		      uint32_t domains __rte_unused,
3018 		      uint32_t flags __rte_unused)
3019 {
3020 	return 0;
3021 }
3022 
3023 /* Void driver to protect from null pointer reference. */
3024 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3025 	.validate = flow_null_validate,
3026 	.prepare = flow_null_prepare,
3027 	.translate = flow_null_translate,
3028 	.apply = flow_null_apply,
3029 	.remove = flow_null_remove,
3030 	.destroy = flow_null_destroy,
3031 	.query = flow_null_query,
3032 	.sync_domain = flow_null_sync_domain,
3033 };
3034 
3035 /**
3036  * Select flow driver type according to flow attributes and device
3037  * configuration.
3038  *
3039  * @param[in] dev
3040  *   Pointer to the dev structure.
3041  * @param[in] attr
3042  *   Pointer to the flow attributes.
3043  *
3044  * @return
3045  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3046  */
3047 static enum mlx5_flow_drv_type
3048 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3049 {
3050 	struct mlx5_priv *priv = dev->data->dev_private;
3051 	/* The OS can determine first a specific flow type (DV, VERBS) */
3052 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3053 
3054 	if (type != MLX5_FLOW_TYPE_MAX)
3055 		return type;
3056 	/* If no OS specific type - continue with DV/VERBS selection */
3057 	if (attr->transfer && priv->config.dv_esw_en)
3058 		type = MLX5_FLOW_TYPE_DV;
3059 	if (!attr->transfer)
3060 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3061 						 MLX5_FLOW_TYPE_VERBS;
3062 	return type;
3063 }
3064 
3065 #define flow_get_drv_ops(type) flow_drv_ops[type]
3066 
3067 /**
3068  * Flow driver validation API. This abstracts calling driver specific functions.
3069  * The type of flow driver is determined according to flow attributes.
3070  *
3071  * @param[in] dev
3072  *   Pointer to the dev structure.
3073  * @param[in] attr
3074  *   Pointer to the flow attributes.
3075  * @param[in] items
3076  *   Pointer to the list of items.
3077  * @param[in] actions
3078  *   Pointer to the list of actions.
3079  * @param[in] external
3080  *   This flow rule is created by request external to PMD.
3081  * @param[in] hairpin
3082  *   Number of hairpin TX actions, 0 means classic flow.
3083  * @param[out] error
3084  *   Pointer to the error structure.
3085  *
3086  * @return
3087  *   0 on success, a negative errno value otherwise and rte_errno is set.
3088  */
3089 static inline int
3090 flow_drv_validate(struct rte_eth_dev *dev,
3091 		  const struct rte_flow_attr *attr,
3092 		  const struct rte_flow_item items[],
3093 		  const struct rte_flow_action actions[],
3094 		  bool external, int hairpin, struct rte_flow_error *error)
3095 {
3096 	const struct mlx5_flow_driver_ops *fops;
3097 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3098 
3099 	fops = flow_get_drv_ops(type);
3100 	return fops->validate(dev, attr, items, actions, external,
3101 			      hairpin, error);
3102 }
3103 
3104 /**
3105  * Flow driver preparation API. This abstracts calling driver specific
3106  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3107  * calculates the size of memory required for device flow, allocates the memory,
3108  * initializes the device flow and returns the pointer.
3109  *
3110  * @note
3111  *   This function initializes device flow structure such as dv or verbs in
3112  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3113  *   rest. For example, adding returning device flow to flow->dev_flow list and
3114  *   setting backward reference to the flow should be done out of this function.
3115  *   layers field is not filled either.
3116  *
3117  * @param[in] dev
3118  *   Pointer to the dev structure.
3119  * @param[in] attr
3120  *   Pointer to the flow attributes.
3121  * @param[in] items
3122  *   Pointer to the list of items.
3123  * @param[in] actions
3124  *   Pointer to the list of actions.
3125  * @param[in] flow_idx
3126  *   This memory pool index to the flow.
3127  * @param[out] error
3128  *   Pointer to the error structure.
3129  *
3130  * @return
3131  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3132  */
3133 static inline struct mlx5_flow *
3134 flow_drv_prepare(struct rte_eth_dev *dev,
3135 		 const struct rte_flow *flow,
3136 		 const struct rte_flow_attr *attr,
3137 		 const struct rte_flow_item items[],
3138 		 const struct rte_flow_action actions[],
3139 		 uint32_t flow_idx,
3140 		 struct rte_flow_error *error)
3141 {
3142 	const struct mlx5_flow_driver_ops *fops;
3143 	enum mlx5_flow_drv_type type = flow->drv_type;
3144 	struct mlx5_flow *mlx5_flow = NULL;
3145 
3146 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3147 	fops = flow_get_drv_ops(type);
3148 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3149 	if (mlx5_flow)
3150 		mlx5_flow->flow_idx = flow_idx;
3151 	return mlx5_flow;
3152 }
3153 
3154 /**
3155  * Flow driver translation API. This abstracts calling driver specific
3156  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3157  * translates a generic flow into a driver flow. flow_drv_prepare() must
3158  * precede.
3159  *
3160  * @note
3161  *   dev_flow->layers could be filled as a result of parsing during translation
3162  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3163  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3164  *   flow->actions could be overwritten even though all the expanded dev_flows
3165  *   have the same actions.
3166  *
3167  * @param[in] dev
3168  *   Pointer to the rte dev structure.
3169  * @param[in, out] dev_flow
3170  *   Pointer to the mlx5 flow.
3171  * @param[in] attr
3172  *   Pointer to the flow attributes.
3173  * @param[in] items
3174  *   Pointer to the list of items.
3175  * @param[in] actions
3176  *   Pointer to the list of actions.
3177  * @param[out] error
3178  *   Pointer to the error structure.
3179  *
3180  * @return
3181  *   0 on success, a negative errno value otherwise and rte_errno is set.
3182  */
3183 static inline int
3184 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3185 		   const struct rte_flow_attr *attr,
3186 		   const struct rte_flow_item items[],
3187 		   const struct rte_flow_action actions[],
3188 		   struct rte_flow_error *error)
3189 {
3190 	const struct mlx5_flow_driver_ops *fops;
3191 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3192 
3193 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3194 	fops = flow_get_drv_ops(type);
3195 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3196 }
3197 
3198 /**
3199  * Flow driver apply API. This abstracts calling driver specific functions.
3200  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3201  * translated driver flows on to device. flow_drv_translate() must precede.
3202  *
3203  * @param[in] dev
3204  *   Pointer to Ethernet device structure.
3205  * @param[in, out] flow
3206  *   Pointer to flow structure.
3207  * @param[out] error
3208  *   Pointer to error structure.
3209  *
3210  * @return
3211  *   0 on success, a negative errno value otherwise and rte_errno is set.
3212  */
3213 static inline int
3214 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3215 	       struct rte_flow_error *error)
3216 {
3217 	const struct mlx5_flow_driver_ops *fops;
3218 	enum mlx5_flow_drv_type type = flow->drv_type;
3219 
3220 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3221 	fops = flow_get_drv_ops(type);
3222 	return fops->apply(dev, flow, error);
3223 }
3224 
3225 /**
3226  * Flow driver destroy API. This abstracts calling driver specific functions.
3227  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3228  * on device and releases resources of the flow.
3229  *
3230  * @param[in] dev
3231  *   Pointer to Ethernet device.
3232  * @param[in, out] flow
3233  *   Pointer to flow structure.
3234  */
3235 static inline void
3236 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3237 {
3238 	const struct mlx5_flow_driver_ops *fops;
3239 	enum mlx5_flow_drv_type type = flow->drv_type;
3240 
3241 	flow_mreg_split_qrss_release(dev, flow);
3242 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3243 	fops = flow_get_drv_ops(type);
3244 	fops->destroy(dev, flow);
3245 }
3246 
3247 /**
3248  * Get RSS action from the action list.
3249  *
3250  * @param[in] actions
3251  *   Pointer to the list of actions.
3252  *
3253  * @return
3254  *   Pointer to the RSS action if exist, else return NULL.
3255  */
3256 static const struct rte_flow_action_rss*
3257 flow_get_rss_action(const struct rte_flow_action actions[])
3258 {
3259 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3260 		switch (actions->type) {
3261 		case RTE_FLOW_ACTION_TYPE_RSS:
3262 			return (const struct rte_flow_action_rss *)
3263 			       actions->conf;
3264 		default:
3265 			break;
3266 		}
3267 	}
3268 	return NULL;
3269 }
3270 
3271 /**
3272  * Get ASO age action by index.
3273  *
3274  * @param[in] dev
3275  *   Pointer to the Ethernet device structure.
3276  * @param[in] age_idx
3277  *   Index to the ASO age action.
3278  *
3279  * @return
3280  *   The specified ASO age action.
3281  */
3282 struct mlx5_aso_age_action*
3283 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3284 {
3285 	uint16_t pool_idx = age_idx & UINT16_MAX;
3286 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3287 	struct mlx5_priv *priv = dev->data->dev_private;
3288 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3289 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3290 
3291 	return &pool->actions[offset - 1];
3292 }
3293 
3294 /* maps shared action to translated non shared in some actions array */
3295 struct mlx5_translated_shared_action {
3296 	struct rte_flow_shared_action *action; /**< Shared action */
3297 	int index; /**< Index in related array of rte_flow_action */
3298 };
3299 
3300 /**
3301  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3302  * non shared action if translation possible.
3303  * This functionality used to run same execution path for both shared & non
3304  * shared actions on flow create. All necessary preparations for shared
3305  * action handling should be preformed on *shared* actions list returned
3306  * from this call.
3307  *
3308  * @param[in] dev
3309  *   Pointer to Ethernet device.
3310  * @param[in] actions
3311  *   List of actions to translate.
3312  * @param[out] shared
3313  *   List to store translated shared actions.
3314  * @param[in, out] shared_n
3315  *   Size of *shared* array. On return should be updated with number of shared
3316  *   actions retrieved from the *actions* list.
3317  * @param[out] translated_actions
3318  *   List of actions where all shared actions were translated to non shared
3319  *   if possible. NULL if no translation took place.
3320  * @param[out] error
3321  *   Pointer to the error structure.
3322  *
3323  * @return
3324  *   0 on success, a negative errno value otherwise and rte_errno is set.
3325  */
3326 static int
3327 flow_shared_actions_translate(struct rte_eth_dev *dev,
3328 			      const struct rte_flow_action actions[],
3329 			      struct mlx5_translated_shared_action *shared,
3330 			      int *shared_n,
3331 			      struct rte_flow_action **translated_actions,
3332 			      struct rte_flow_error *error)
3333 {
3334 	struct mlx5_priv *priv = dev->data->dev_private;
3335 	struct rte_flow_action *translated = NULL;
3336 	size_t actions_size;
3337 	int n;
3338 	int copied_n = 0;
3339 	struct mlx5_translated_shared_action *shared_end = NULL;
3340 
3341 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3342 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3343 			continue;
3344 		if (copied_n == *shared_n) {
3345 			return rte_flow_error_set
3346 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3347 				 NULL, "too many shared actions");
3348 		}
3349 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3350 			   sizeof(actions[n].conf));
3351 		shared[copied_n].index = n;
3352 		copied_n++;
3353 	}
3354 	n++;
3355 	*shared_n = copied_n;
3356 	if (!copied_n)
3357 		return 0;
3358 	actions_size = sizeof(struct rte_flow_action) * n;
3359 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3360 	if (!translated) {
3361 		rte_errno = ENOMEM;
3362 		return -ENOMEM;
3363 	}
3364 	memcpy(translated, actions, actions_size);
3365 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3366 		struct mlx5_shared_action_rss *shared_rss;
3367 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3368 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3369 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3370 									   - 1);
3371 
3372 		switch (type) {
3373 		case MLX5_SHARED_ACTION_TYPE_RSS:
3374 			shared_rss = mlx5_ipool_get
3375 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3376 			translated[shared->index].type =
3377 				RTE_FLOW_ACTION_TYPE_RSS;
3378 			translated[shared->index].conf =
3379 				&shared_rss->origin;
3380 			break;
3381 		case MLX5_SHARED_ACTION_TYPE_AGE:
3382 			if (priv->sh->flow_hit_aso_en) {
3383 				translated[shared->index].type =
3384 					(enum rte_flow_action_type)
3385 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3386 				translated[shared->index].conf =
3387 							 (void *)(uintptr_t)idx;
3388 				break;
3389 			}
3390 			/* Fall-through */
3391 		default:
3392 			mlx5_free(translated);
3393 			return rte_flow_error_set
3394 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3395 				 NULL, "invalid shared action type");
3396 		}
3397 	}
3398 	*translated_actions = translated;
3399 	return 0;
3400 }
3401 
3402 /**
3403  * Get Shared RSS action from the action list.
3404  *
3405  * @param[in] dev
3406  *   Pointer to Ethernet device.
3407  * @param[in] shared
3408  *   Pointer to the list of actions.
3409  * @param[in] shared_n
3410  *   Actions list length.
3411  *
3412  * @return
3413  *   The MLX5 RSS action ID if exists, otherwise return 0.
3414  */
3415 static uint32_t
3416 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3417 			   struct mlx5_translated_shared_action *shared,
3418 			   int shared_n)
3419 {
3420 	struct mlx5_translated_shared_action *shared_end;
3421 	struct mlx5_priv *priv = dev->data->dev_private;
3422 	struct mlx5_shared_action_rss *shared_rss;
3423 
3424 
3425 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3426 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3427 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3428 		uint32_t idx = act_idx &
3429 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3430 		switch (type) {
3431 		case MLX5_SHARED_ACTION_TYPE_RSS:
3432 			shared_rss = mlx5_ipool_get
3433 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3434 									   idx);
3435 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3436 					   __ATOMIC_RELAXED);
3437 			return idx;
3438 		default:
3439 			break;
3440 		}
3441 	}
3442 	return 0;
3443 }
3444 
3445 static unsigned int
3446 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3447 {
3448 	const struct rte_flow_item *item;
3449 	unsigned int has_vlan = 0;
3450 
3451 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3452 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3453 			has_vlan = 1;
3454 			break;
3455 		}
3456 	}
3457 	if (has_vlan)
3458 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3459 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3460 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3461 			       MLX5_EXPANSION_ROOT_OUTER;
3462 }
3463 
3464 /**
3465  *  Get layer flags from the prefix flow.
3466  *
3467  *  Some flows may be split to several subflows, the prefix subflow gets the
3468  *  match items and the suffix sub flow gets the actions.
3469  *  Some actions need the user defined match item flags to get the detail for
3470  *  the action.
3471  *  This function helps the suffix flow to get the item layer flags from prefix
3472  *  subflow.
3473  *
3474  * @param[in] dev_flow
3475  *   Pointer the created preifx subflow.
3476  *
3477  * @return
3478  *   The layers get from prefix subflow.
3479  */
3480 static inline uint64_t
3481 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3482 {
3483 	uint64_t layers = 0;
3484 
3485 	/*
3486 	 * Layers bits could be localization, but usually the compiler will
3487 	 * help to do the optimization work for source code.
3488 	 * If no decap actions, use the layers directly.
3489 	 */
3490 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3491 		return dev_flow->handle->layers;
3492 	/* Convert L3 layers with decap action. */
3493 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3494 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3495 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3496 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3497 	/* Convert L4 layers with decap action.  */
3498 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3499 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3500 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3501 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3502 	return layers;
3503 }
3504 
3505 /**
3506  * Get metadata split action information.
3507  *
3508  * @param[in] actions
3509  *   Pointer to the list of actions.
3510  * @param[out] qrss
3511  *   Pointer to the return pointer.
3512  * @param[out] qrss_type
3513  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3514  *   if no QUEUE/RSS is found.
3515  * @param[out] encap_idx
3516  *   Pointer to the index of the encap action if exists, otherwise the last
3517  *   action index.
3518  *
3519  * @return
3520  *   Total number of actions.
3521  */
3522 static int
3523 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3524 				       const struct rte_flow_action **qrss,
3525 				       int *encap_idx)
3526 {
3527 	const struct rte_flow_action_raw_encap *raw_encap;
3528 	int actions_n = 0;
3529 	int raw_decap_idx = -1;
3530 
3531 	*encap_idx = -1;
3532 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3533 		switch (actions->type) {
3534 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3535 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3536 			*encap_idx = actions_n;
3537 			break;
3538 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3539 			raw_decap_idx = actions_n;
3540 			break;
3541 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3542 			raw_encap = actions->conf;
3543 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3544 				*encap_idx = raw_decap_idx != -1 ?
3545 						      raw_decap_idx : actions_n;
3546 			break;
3547 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3548 		case RTE_FLOW_ACTION_TYPE_RSS:
3549 			*qrss = actions;
3550 			break;
3551 		default:
3552 			break;
3553 		}
3554 		actions_n++;
3555 	}
3556 	if (*encap_idx == -1)
3557 		*encap_idx = actions_n;
3558 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3559 	return actions_n + 1;
3560 }
3561 
3562 /**
3563  * Check meter action from the action list.
3564  *
3565  * @param[in] actions
3566  *   Pointer to the list of actions.
3567  * @param[out] mtr
3568  *   Pointer to the meter exist flag.
3569  *
3570  * @return
3571  *   Total number of actions.
3572  */
3573 static int
3574 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3575 {
3576 	int actions_n = 0;
3577 
3578 	MLX5_ASSERT(mtr);
3579 	*mtr = 0;
3580 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3581 		switch (actions->type) {
3582 		case RTE_FLOW_ACTION_TYPE_METER:
3583 			*mtr = 1;
3584 			break;
3585 		default:
3586 			break;
3587 		}
3588 		actions_n++;
3589 	}
3590 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3591 	return actions_n + 1;
3592 }
3593 
3594 /**
3595  * Check if the flow should be split due to hairpin.
3596  * The reason for the split is that in current HW we can't
3597  * support encap and push-vlan on Rx, so if a flow contains
3598  * these actions we move it to Tx.
3599  *
3600  * @param dev
3601  *   Pointer to Ethernet device.
3602  * @param[in] attr
3603  *   Flow rule attributes.
3604  * @param[in] actions
3605  *   Associated actions (list terminated by the END action).
3606  *
3607  * @return
3608  *   > 0 the number of actions and the flow should be split,
3609  *   0 when no split required.
3610  */
3611 static int
3612 flow_check_hairpin_split(struct rte_eth_dev *dev,
3613 			 const struct rte_flow_attr *attr,
3614 			 const struct rte_flow_action actions[])
3615 {
3616 	int queue_action = 0;
3617 	int action_n = 0;
3618 	int split = 0;
3619 	const struct rte_flow_action_queue *queue;
3620 	const struct rte_flow_action_rss *rss;
3621 	const struct rte_flow_action_raw_encap *raw_encap;
3622 	const struct rte_eth_hairpin_conf *conf;
3623 
3624 	if (!attr->ingress)
3625 		return 0;
3626 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3627 		switch (actions->type) {
3628 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3629 			queue = actions->conf;
3630 			if (queue == NULL)
3631 				return 0;
3632 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3633 			if (conf != NULL && !!conf->tx_explicit)
3634 				return 0;
3635 			queue_action = 1;
3636 			action_n++;
3637 			break;
3638 		case RTE_FLOW_ACTION_TYPE_RSS:
3639 			rss = actions->conf;
3640 			if (rss == NULL || rss->queue_num == 0)
3641 				return 0;
3642 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3643 			if (conf != NULL && !!conf->tx_explicit)
3644 				return 0;
3645 			queue_action = 1;
3646 			action_n++;
3647 			break;
3648 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3649 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3650 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3651 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3652 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3653 			split++;
3654 			action_n++;
3655 			break;
3656 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3657 			raw_encap = actions->conf;
3658 			if (raw_encap->size >
3659 			    (sizeof(struct rte_flow_item_eth) +
3660 			     sizeof(struct rte_flow_item_ipv4)))
3661 				split++;
3662 			action_n++;
3663 			break;
3664 		default:
3665 			action_n++;
3666 			break;
3667 		}
3668 	}
3669 	if (split && queue_action)
3670 		return action_n;
3671 	return 0;
3672 }
3673 
3674 /* Declare flow create/destroy prototype in advance. */
3675 static uint32_t
3676 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3677 		 const struct rte_flow_attr *attr,
3678 		 const struct rte_flow_item items[],
3679 		 const struct rte_flow_action actions[],
3680 		 bool external, struct rte_flow_error *error);
3681 
3682 static void
3683 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3684 		  uint32_t flow_idx);
3685 
3686 struct mlx5_hlist_entry *
3687 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3688 		       void *cb_ctx)
3689 {
3690 	struct rte_eth_dev *dev = list->ctx;
3691 	struct mlx5_priv *priv = dev->data->dev_private;
3692 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3693 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3694 	struct rte_flow_error *error = ctx->error;
3695 	uint32_t idx = 0;
3696 	int ret;
3697 	uint32_t mark_id = key;
3698 	struct rte_flow_attr attr = {
3699 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3700 		.ingress = 1,
3701 	};
3702 	struct mlx5_rte_flow_item_tag tag_spec = {
3703 		.data = mark_id,
3704 	};
3705 	struct rte_flow_item items[] = {
3706 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3707 	};
3708 	struct rte_flow_action_mark ftag = {
3709 		.id = mark_id,
3710 	};
3711 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3712 		.dst = REG_B,
3713 		.src = REG_NON,
3714 	};
3715 	struct rte_flow_action_jump jump = {
3716 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3717 	};
3718 	struct rte_flow_action actions[] = {
3719 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3720 	};
3721 
3722 	/* Fill the register fileds in the flow. */
3723 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3724 	if (ret < 0)
3725 		return NULL;
3726 	tag_spec.id = ret;
3727 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3728 	if (ret < 0)
3729 		return NULL;
3730 	cp_mreg.src = ret;
3731 	/* Provide the full width of FLAG specific value. */
3732 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3733 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3734 	/* Build a new flow. */
3735 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3736 		items[0] = (struct rte_flow_item){
3737 			.type = (enum rte_flow_item_type)
3738 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3739 			.spec = &tag_spec,
3740 		};
3741 		items[1] = (struct rte_flow_item){
3742 			.type = RTE_FLOW_ITEM_TYPE_END,
3743 		};
3744 		actions[0] = (struct rte_flow_action){
3745 			.type = (enum rte_flow_action_type)
3746 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3747 			.conf = &ftag,
3748 		};
3749 		actions[1] = (struct rte_flow_action){
3750 			.type = (enum rte_flow_action_type)
3751 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3752 			.conf = &cp_mreg,
3753 		};
3754 		actions[2] = (struct rte_flow_action){
3755 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3756 			.conf = &jump,
3757 		};
3758 		actions[3] = (struct rte_flow_action){
3759 			.type = RTE_FLOW_ACTION_TYPE_END,
3760 		};
3761 	} else {
3762 		/* Default rule, wildcard match. */
3763 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3764 		items[0] = (struct rte_flow_item){
3765 			.type = RTE_FLOW_ITEM_TYPE_END,
3766 		};
3767 		actions[0] = (struct rte_flow_action){
3768 			.type = (enum rte_flow_action_type)
3769 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3770 			.conf = &cp_mreg,
3771 		};
3772 		actions[1] = (struct rte_flow_action){
3773 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3774 			.conf = &jump,
3775 		};
3776 		actions[2] = (struct rte_flow_action){
3777 			.type = RTE_FLOW_ACTION_TYPE_END,
3778 		};
3779 	}
3780 	/* Build a new entry. */
3781 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3782 	if (!mcp_res) {
3783 		rte_errno = ENOMEM;
3784 		return NULL;
3785 	}
3786 	mcp_res->idx = idx;
3787 	/*
3788 	 * The copy Flows are not included in any list. There
3789 	 * ones are referenced from other Flows and can not
3790 	 * be applied, removed, deleted in ardbitrary order
3791 	 * by list traversing.
3792 	 */
3793 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3794 					 actions, false, error);
3795 	if (!mcp_res->rix_flow) {
3796 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3797 		return NULL;
3798 	}
3799 	return &mcp_res->hlist_ent;
3800 }
3801 
3802 /**
3803  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3804  *
3805  * As mark_id is unique, if there's already a registered flow for the mark_id,
3806  * return by increasing the reference counter of the resource. Otherwise, create
3807  * the resource (mcp_res) and flow.
3808  *
3809  * Flow looks like,
3810  *   - If ingress port is ANY and reg_c[1] is mark_id,
3811  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3812  *
3813  * For default flow (zero mark_id), flow is like,
3814  *   - If ingress port is ANY,
3815  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3816  *
3817  * @param dev
3818  *   Pointer to Ethernet device.
3819  * @param mark_id
3820  *   ID of MARK action, zero means default flow for META.
3821  * @param[out] error
3822  *   Perform verbose error reporting if not NULL.
3823  *
3824  * @return
3825  *   Associated resource on success, NULL otherwise and rte_errno is set.
3826  */
3827 static struct mlx5_flow_mreg_copy_resource *
3828 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3829 			  struct rte_flow_error *error)
3830 {
3831 	struct mlx5_priv *priv = dev->data->dev_private;
3832 	struct mlx5_hlist_entry *entry;
3833 	struct mlx5_flow_cb_ctx ctx = {
3834 		.dev = dev,
3835 		.error = error,
3836 	};
3837 
3838 	/* Check if already registered. */
3839 	MLX5_ASSERT(priv->mreg_cp_tbl);
3840 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3841 	if (!entry)
3842 		return NULL;
3843 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3844 			    hlist_ent);
3845 }
3846 
3847 void
3848 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3849 {
3850 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3851 		container_of(entry, typeof(*mcp_res), hlist_ent);
3852 	struct rte_eth_dev *dev = list->ctx;
3853 	struct mlx5_priv *priv = dev->data->dev_private;
3854 
3855 	MLX5_ASSERT(mcp_res->rix_flow);
3856 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3857 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3858 }
3859 
3860 /**
3861  * Release flow in RX_CP_TBL.
3862  *
3863  * @param dev
3864  *   Pointer to Ethernet device.
3865  * @flow
3866  *   Parent flow for wich copying is provided.
3867  */
3868 static void
3869 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3870 			  struct rte_flow *flow)
3871 {
3872 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3873 	struct mlx5_priv *priv = dev->data->dev_private;
3874 
3875 	if (!flow->rix_mreg_copy)
3876 		return;
3877 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3878 				 flow->rix_mreg_copy);
3879 	if (!mcp_res || !priv->mreg_cp_tbl)
3880 		return;
3881 	MLX5_ASSERT(mcp_res->rix_flow);
3882 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3883 	flow->rix_mreg_copy = 0;
3884 }
3885 
3886 /**
3887  * Remove the default copy action from RX_CP_TBL.
3888  *
3889  * This functions is called in the mlx5_dev_start(). No thread safe
3890  * is guaranteed.
3891  *
3892  * @param dev
3893  *   Pointer to Ethernet device.
3894  */
3895 static void
3896 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3897 {
3898 	struct mlx5_hlist_entry *entry;
3899 	struct mlx5_priv *priv = dev->data->dev_private;
3900 
3901 	/* Check if default flow is registered. */
3902 	if (!priv->mreg_cp_tbl)
3903 		return;
3904 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3905 				  MLX5_DEFAULT_COPY_ID, NULL);
3906 	if (!entry)
3907 		return;
3908 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3909 }
3910 
3911 /**
3912  * Add the default copy action in in RX_CP_TBL.
3913  *
3914  * This functions is called in the mlx5_dev_start(). No thread safe
3915  * is guaranteed.
3916  *
3917  * @param dev
3918  *   Pointer to Ethernet device.
3919  * @param[out] error
3920  *   Perform verbose error reporting if not NULL.
3921  *
3922  * @return
3923  *   0 for success, negative value otherwise and rte_errno is set.
3924  */
3925 static int
3926 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3927 				  struct rte_flow_error *error)
3928 {
3929 	struct mlx5_priv *priv = dev->data->dev_private;
3930 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3931 
3932 	/* Check whether extensive metadata feature is engaged. */
3933 	if (!priv->config.dv_flow_en ||
3934 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3935 	    !mlx5_flow_ext_mreg_supported(dev) ||
3936 	    !priv->sh->dv_regc0_mask)
3937 		return 0;
3938 	/*
3939 	 * Add default mreg copy flow may be called multiple time, but
3940 	 * only be called once in stop. Avoid register it twice.
3941 	 */
3942 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
3943 		return 0;
3944 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3945 	if (!mcp_res)
3946 		return -rte_errno;
3947 	return 0;
3948 }
3949 
3950 /**
3951  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3952  *
3953  * All the flow having Q/RSS action should be split by
3954  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3955  * performs the following,
3956  *   - CQE->flow_tag := reg_c[1] (MARK)
3957  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3958  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3959  * but there should be a flow per each MARK ID set by MARK action.
3960  *
3961  * For the aforementioned reason, if there's a MARK action in flow's action
3962  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3963  * the MARK ID to CQE's flow_tag like,
3964  *   - If reg_c[1] is mark_id,
3965  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3966  *
3967  * For SET_META action which stores value in reg_c[0], as the destination is
3968  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3969  * MARK ID means the default flow. The default flow looks like,
3970  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3971  *
3972  * @param dev
3973  *   Pointer to Ethernet device.
3974  * @param flow
3975  *   Pointer to flow structure.
3976  * @param[in] actions
3977  *   Pointer to the list of actions.
3978  * @param[out] error
3979  *   Perform verbose error reporting if not NULL.
3980  *
3981  * @return
3982  *   0 on success, negative value otherwise and rte_errno is set.
3983  */
3984 static int
3985 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3986 			    struct rte_flow *flow,
3987 			    const struct rte_flow_action *actions,
3988 			    struct rte_flow_error *error)
3989 {
3990 	struct mlx5_priv *priv = dev->data->dev_private;
3991 	struct mlx5_dev_config *config = &priv->config;
3992 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3993 	const struct rte_flow_action_mark *mark;
3994 
3995 	/* Check whether extensive metadata feature is engaged. */
3996 	if (!config->dv_flow_en ||
3997 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3998 	    !mlx5_flow_ext_mreg_supported(dev) ||
3999 	    !priv->sh->dv_regc0_mask)
4000 		return 0;
4001 	/* Find MARK action. */
4002 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4003 		switch (actions->type) {
4004 		case RTE_FLOW_ACTION_TYPE_FLAG:
4005 			mcp_res = flow_mreg_add_copy_action
4006 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4007 			if (!mcp_res)
4008 				return -rte_errno;
4009 			flow->rix_mreg_copy = mcp_res->idx;
4010 			return 0;
4011 		case RTE_FLOW_ACTION_TYPE_MARK:
4012 			mark = (const struct rte_flow_action_mark *)
4013 				actions->conf;
4014 			mcp_res =
4015 				flow_mreg_add_copy_action(dev, mark->id, error);
4016 			if (!mcp_res)
4017 				return -rte_errno;
4018 			flow->rix_mreg_copy = mcp_res->idx;
4019 			return 0;
4020 		default:
4021 			break;
4022 		}
4023 	}
4024 	return 0;
4025 }
4026 
4027 #define MLX5_MAX_SPLIT_ACTIONS 24
4028 #define MLX5_MAX_SPLIT_ITEMS 24
4029 
4030 /**
4031  * Split the hairpin flow.
4032  * Since HW can't support encap and push-vlan on Rx, we move these
4033  * actions to Tx.
4034  * If the count action is after the encap then we also
4035  * move the count action. in this case the count will also measure
4036  * the outer bytes.
4037  *
4038  * @param dev
4039  *   Pointer to Ethernet device.
4040  * @param[in] actions
4041  *   Associated actions (list terminated by the END action).
4042  * @param[out] actions_rx
4043  *   Rx flow actions.
4044  * @param[out] actions_tx
4045  *   Tx flow actions..
4046  * @param[out] pattern_tx
4047  *   The pattern items for the Tx flow.
4048  * @param[out] flow_id
4049  *   The flow ID connected to this flow.
4050  *
4051  * @return
4052  *   0 on success.
4053  */
4054 static int
4055 flow_hairpin_split(struct rte_eth_dev *dev,
4056 		   const struct rte_flow_action actions[],
4057 		   struct rte_flow_action actions_rx[],
4058 		   struct rte_flow_action actions_tx[],
4059 		   struct rte_flow_item pattern_tx[],
4060 		   uint32_t flow_id)
4061 {
4062 	const struct rte_flow_action_raw_encap *raw_encap;
4063 	const struct rte_flow_action_raw_decap *raw_decap;
4064 	struct mlx5_rte_flow_action_set_tag *set_tag;
4065 	struct rte_flow_action *tag_action;
4066 	struct mlx5_rte_flow_item_tag *tag_item;
4067 	struct rte_flow_item *item;
4068 	char *addr;
4069 	int encap = 0;
4070 
4071 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4072 		switch (actions->type) {
4073 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4074 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4075 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4076 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4077 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4078 			rte_memcpy(actions_tx, actions,
4079 			       sizeof(struct rte_flow_action));
4080 			actions_tx++;
4081 			break;
4082 		case RTE_FLOW_ACTION_TYPE_COUNT:
4083 			if (encap) {
4084 				rte_memcpy(actions_tx, actions,
4085 					   sizeof(struct rte_flow_action));
4086 				actions_tx++;
4087 			} else {
4088 				rte_memcpy(actions_rx, actions,
4089 					   sizeof(struct rte_flow_action));
4090 				actions_rx++;
4091 			}
4092 			break;
4093 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4094 			raw_encap = actions->conf;
4095 			if (raw_encap->size >
4096 			    (sizeof(struct rte_flow_item_eth) +
4097 			     sizeof(struct rte_flow_item_ipv4))) {
4098 				memcpy(actions_tx, actions,
4099 				       sizeof(struct rte_flow_action));
4100 				actions_tx++;
4101 				encap = 1;
4102 			} else {
4103 				rte_memcpy(actions_rx, actions,
4104 					   sizeof(struct rte_flow_action));
4105 				actions_rx++;
4106 			}
4107 			break;
4108 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4109 			raw_decap = actions->conf;
4110 			if (raw_decap->size <
4111 			    (sizeof(struct rte_flow_item_eth) +
4112 			     sizeof(struct rte_flow_item_ipv4))) {
4113 				memcpy(actions_tx, actions,
4114 				       sizeof(struct rte_flow_action));
4115 				actions_tx++;
4116 			} else {
4117 				rte_memcpy(actions_rx, actions,
4118 					   sizeof(struct rte_flow_action));
4119 				actions_rx++;
4120 			}
4121 			break;
4122 		default:
4123 			rte_memcpy(actions_rx, actions,
4124 				   sizeof(struct rte_flow_action));
4125 			actions_rx++;
4126 			break;
4127 		}
4128 	}
4129 	/* Add set meta action and end action for the Rx flow. */
4130 	tag_action = actions_rx;
4131 	tag_action->type = (enum rte_flow_action_type)
4132 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4133 	actions_rx++;
4134 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4135 	actions_rx++;
4136 	set_tag = (void *)actions_rx;
4137 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4138 	MLX5_ASSERT(set_tag->id > REG_NON);
4139 	set_tag->data = flow_id;
4140 	tag_action->conf = set_tag;
4141 	/* Create Tx item list. */
4142 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4143 	addr = (void *)&pattern_tx[2];
4144 	item = pattern_tx;
4145 	item->type = (enum rte_flow_item_type)
4146 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4147 	tag_item = (void *)addr;
4148 	tag_item->data = flow_id;
4149 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4150 	MLX5_ASSERT(set_tag->id > REG_NON);
4151 	item->spec = tag_item;
4152 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4153 	tag_item = (void *)addr;
4154 	tag_item->data = UINT32_MAX;
4155 	tag_item->id = UINT16_MAX;
4156 	item->mask = tag_item;
4157 	item->last = NULL;
4158 	item++;
4159 	item->type = RTE_FLOW_ITEM_TYPE_END;
4160 	return 0;
4161 }
4162 
4163 __extension__
4164 union tunnel_offload_mark {
4165 	uint32_t val;
4166 	struct {
4167 		uint32_t app_reserve:8;
4168 		uint32_t table_id:15;
4169 		uint32_t transfer:1;
4170 		uint32_t _unused_:8;
4171 	};
4172 };
4173 
4174 struct tunnel_default_miss_ctx {
4175 	uint16_t *queue;
4176 	__extension__
4177 	union {
4178 		struct rte_flow_action_rss action_rss;
4179 		struct rte_flow_action_queue miss_queue;
4180 		struct rte_flow_action_jump miss_jump;
4181 		uint8_t raw[0];
4182 	};
4183 };
4184 
4185 static int
4186 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4187 			     struct rte_flow *flow,
4188 			     const struct rte_flow_attr *attr,
4189 			     const struct rte_flow_action *app_actions,
4190 			     uint32_t flow_idx,
4191 			     struct tunnel_default_miss_ctx *ctx,
4192 			     struct rte_flow_error *error)
4193 {
4194 	struct mlx5_priv *priv = dev->data->dev_private;
4195 	struct mlx5_flow *dev_flow;
4196 	struct rte_flow_attr miss_attr = *attr;
4197 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4198 	const struct rte_flow_item miss_items[2] = {
4199 		{
4200 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4201 			.spec = NULL,
4202 			.last = NULL,
4203 			.mask = NULL
4204 		},
4205 		{
4206 			.type = RTE_FLOW_ITEM_TYPE_END,
4207 			.spec = NULL,
4208 			.last = NULL,
4209 			.mask = NULL
4210 		}
4211 	};
4212 	union tunnel_offload_mark mark_id;
4213 	struct rte_flow_action_mark miss_mark;
4214 	struct rte_flow_action miss_actions[3] = {
4215 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4216 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
4217 	};
4218 	const struct rte_flow_action_jump *jump_data;
4219 	uint32_t i, flow_table = 0; /* prevent compilation warning */
4220 	struct flow_grp_info grp_info = {
4221 		.external = 1,
4222 		.transfer = attr->transfer,
4223 		.fdb_def_rule = !!priv->fdb_def_rule,
4224 		.std_tbl_fix = 0,
4225 	};
4226 	int ret;
4227 
4228 	if (!attr->transfer) {
4229 		uint32_t q_size;
4230 
4231 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4232 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4233 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4234 					 0, SOCKET_ID_ANY);
4235 		if (!ctx->queue)
4236 			return rte_flow_error_set
4237 				(error, ENOMEM,
4238 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4239 				NULL, "invalid default miss RSS");
4240 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4241 		ctx->action_rss.level = 0,
4242 		ctx->action_rss.types = priv->rss_conf.rss_hf,
4243 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4244 		ctx->action_rss.queue_num = priv->reta_idx_n,
4245 		ctx->action_rss.key = priv->rss_conf.rss_key,
4246 		ctx->action_rss.queue = ctx->queue;
4247 		if (!priv->reta_idx_n || !priv->rxqs_n)
4248 			return rte_flow_error_set
4249 				(error, EINVAL,
4250 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4251 				NULL, "invalid port configuration");
4252 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4253 			ctx->action_rss.types = 0;
4254 		for (i = 0; i != priv->reta_idx_n; ++i)
4255 			ctx->queue[i] = (*priv->reta_idx)[i];
4256 	} else {
4257 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4258 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4259 	}
4260 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4261 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4262 	jump_data = app_actions->conf;
4263 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4264 	miss_attr.group = jump_data->group;
4265 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4266 				       &flow_table, grp_info, error);
4267 	if (ret)
4268 		return rte_flow_error_set(error, EINVAL,
4269 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4270 					  NULL, "invalid tunnel id");
4271 	mark_id.app_reserve = 0;
4272 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4273 	mark_id.transfer = !!attr->transfer;
4274 	mark_id._unused_ = 0;
4275 	miss_mark.id = mark_id.val;
4276 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4277 				    miss_items, miss_actions, flow_idx, error);
4278 	if (!dev_flow)
4279 		return -rte_errno;
4280 	dev_flow->flow = flow;
4281 	dev_flow->external = true;
4282 	dev_flow->tunnel = tunnel;
4283 	/* Subflow object was created, we must include one in the list. */
4284 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4285 		      dev_flow->handle, next);
4286 	DRV_LOG(DEBUG,
4287 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4288 		dev->data->port_id, tunnel->app_tunnel.type,
4289 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4290 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4291 				  miss_actions, error);
4292 	if (!ret)
4293 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4294 						  error);
4295 
4296 	return ret;
4297 }
4298 
4299 /**
4300  * The last stage of splitting chain, just creates the subflow
4301  * without any modification.
4302  *
4303  * @param[in] dev
4304  *   Pointer to Ethernet device.
4305  * @param[in] flow
4306  *   Parent flow structure pointer.
4307  * @param[in, out] sub_flow
4308  *   Pointer to return the created subflow, may be NULL.
4309  * @param[in] attr
4310  *   Flow rule attributes.
4311  * @param[in] items
4312  *   Pattern specification (list terminated by the END pattern item).
4313  * @param[in] actions
4314  *   Associated actions (list terminated by the END action).
4315  * @param[in] flow_split_info
4316  *   Pointer to flow split info structure.
4317  * @param[out] error
4318  *   Perform verbose error reporting if not NULL.
4319  * @return
4320  *   0 on success, negative value otherwise
4321  */
4322 static int
4323 flow_create_split_inner(struct rte_eth_dev *dev,
4324 			struct rte_flow *flow,
4325 			struct mlx5_flow **sub_flow,
4326 			const struct rte_flow_attr *attr,
4327 			const struct rte_flow_item items[],
4328 			const struct rte_flow_action actions[],
4329 			struct mlx5_flow_split_info *flow_split_info,
4330 			struct rte_flow_error *error)
4331 {
4332 	struct mlx5_flow *dev_flow;
4333 
4334 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4335 				    flow_split_info->flow_idx, error);
4336 	if (!dev_flow)
4337 		return -rte_errno;
4338 	dev_flow->flow = flow;
4339 	dev_flow->external = flow_split_info->external;
4340 	dev_flow->skip_scale = flow_split_info->skip_scale;
4341 	/* Subflow object was created, we must include one in the list. */
4342 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4343 		      dev_flow->handle, next);
4344 	/*
4345 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4346 	 * flow may need some user defined item layer flags, and pass the
4347 	 * Metadate rxq mark flag to suffix flow as well.
4348 	 */
4349 	if (flow_split_info->prefix_layers)
4350 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4351 	if (flow_split_info->prefix_mark)
4352 		dev_flow->handle->mark = 1;
4353 	if (sub_flow)
4354 		*sub_flow = dev_flow;
4355 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4356 }
4357 
4358 /**
4359  * Split the meter flow.
4360  *
4361  * As meter flow will split to three sub flow, other than meter
4362  * action, the other actions make sense to only meter accepts
4363  * the packet. If it need to be dropped, no other additional
4364  * actions should be take.
4365  *
4366  * One kind of special action which decapsulates the L3 tunnel
4367  * header will be in the prefix sub flow, as not to take the
4368  * L3 tunnel header into account.
4369  *
4370  * @param dev
4371  *   Pointer to Ethernet device.
4372  * @param[in] items
4373  *   Pattern specification (list terminated by the END pattern item).
4374  * @param[out] sfx_items
4375  *   Suffix flow match items (list terminated by the END pattern item).
4376  * @param[in] actions
4377  *   Associated actions (list terminated by the END action).
4378  * @param[out] actions_sfx
4379  *   Suffix flow actions.
4380  * @param[out] actions_pre
4381  *   Prefix flow actions.
4382  * @param[out] pattern_sfx
4383  *   The pattern items for the suffix flow.
4384  * @param[out] tag_sfx
4385  *   Pointer to suffix flow tag.
4386  *
4387  * @return
4388  *   0 on success.
4389  */
4390 static int
4391 flow_meter_split_prep(struct rte_eth_dev *dev,
4392 		 const struct rte_flow_item items[],
4393 		 struct rte_flow_item sfx_items[],
4394 		 const struct rte_flow_action actions[],
4395 		 struct rte_flow_action actions_sfx[],
4396 		 struct rte_flow_action actions_pre[])
4397 {
4398 	struct mlx5_priv *priv = dev->data->dev_private;
4399 	struct rte_flow_action *tag_action = NULL;
4400 	struct rte_flow_item *tag_item;
4401 	struct mlx5_rte_flow_action_set_tag *set_tag;
4402 	struct rte_flow_error error;
4403 	const struct rte_flow_action_raw_encap *raw_encap;
4404 	const struct rte_flow_action_raw_decap *raw_decap;
4405 	struct mlx5_rte_flow_item_tag *tag_spec;
4406 	struct mlx5_rte_flow_item_tag *tag_mask;
4407 	uint32_t tag_id = 0;
4408 	bool copy_vlan = false;
4409 
4410 	/* Prepare the actions for prefix and suffix flow. */
4411 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4412 		struct rte_flow_action **action_cur = NULL;
4413 
4414 		switch (actions->type) {
4415 		case RTE_FLOW_ACTION_TYPE_METER:
4416 			/* Add the extra tag action first. */
4417 			tag_action = actions_pre;
4418 			tag_action->type = (enum rte_flow_action_type)
4419 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4420 			actions_pre++;
4421 			action_cur = &actions_pre;
4422 			break;
4423 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4424 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4425 			action_cur = &actions_pre;
4426 			break;
4427 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4428 			raw_encap = actions->conf;
4429 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4430 				action_cur = &actions_pre;
4431 			break;
4432 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4433 			raw_decap = actions->conf;
4434 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4435 				action_cur = &actions_pre;
4436 			break;
4437 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4438 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4439 			copy_vlan = true;
4440 			break;
4441 		default:
4442 			break;
4443 		}
4444 		if (!action_cur)
4445 			action_cur = &actions_sfx;
4446 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4447 		(*action_cur)++;
4448 	}
4449 	/* Add end action to the actions. */
4450 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4451 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4452 	actions_pre++;
4453 	/* Set the tag. */
4454 	set_tag = (void *)actions_pre;
4455 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4456 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4457 			  &tag_id);
4458 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4459 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4460 			dev->data->port_id);
4461 		mlx5_ipool_free(priv->sh->ipool
4462 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4463 		return 0;
4464 	} else if (!tag_id) {
4465 		return 0;
4466 	}
4467 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4468 	assert(tag_action);
4469 	tag_action->conf = set_tag;
4470 	/* Prepare the suffix subflow items. */
4471 	tag_item = sfx_items++;
4472 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4473 		int item_type = items->type;
4474 
4475 		switch (item_type) {
4476 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4477 			memcpy(sfx_items, items, sizeof(*sfx_items));
4478 			sfx_items++;
4479 			break;
4480 		case RTE_FLOW_ITEM_TYPE_VLAN:
4481 			if (copy_vlan) {
4482 				memcpy(sfx_items, items, sizeof(*sfx_items));
4483 				/*
4484 				 * Convert to internal match item, it is used
4485 				 * for vlan push and set vid.
4486 				 */
4487 				sfx_items->type = (enum rte_flow_item_type)
4488 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4489 				sfx_items++;
4490 			}
4491 			break;
4492 		default:
4493 			break;
4494 		}
4495 	}
4496 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4497 	sfx_items++;
4498 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4499 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4500 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4501 	tag_mask = tag_spec + 1;
4502 	tag_mask->data = 0xffffff00;
4503 	tag_item->type = (enum rte_flow_item_type)
4504 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4505 	tag_item->spec = tag_spec;
4506 	tag_item->last = NULL;
4507 	tag_item->mask = tag_mask;
4508 	return tag_id;
4509 }
4510 
4511 /**
4512  * Split action list having QUEUE/RSS for metadata register copy.
4513  *
4514  * Once Q/RSS action is detected in user's action list, the flow action
4515  * should be split in order to copy metadata registers, which will happen in
4516  * RX_CP_TBL like,
4517  *   - CQE->flow_tag := reg_c[1] (MARK)
4518  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4519  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4520  * This is because the last action of each flow must be a terminal action
4521  * (QUEUE, RSS or DROP).
4522  *
4523  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4524  * stored and kept in the mlx5_flow structure per each sub_flow.
4525  *
4526  * The Q/RSS action is replaced with,
4527  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4528  * And the following JUMP action is added at the end,
4529  *   - JUMP, to RX_CP_TBL.
4530  *
4531  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4532  * flow_create_split_metadata() routine. The flow will look like,
4533  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4534  *
4535  * @param dev
4536  *   Pointer to Ethernet device.
4537  * @param[out] split_actions
4538  *   Pointer to store split actions to jump to CP_TBL.
4539  * @param[in] actions
4540  *   Pointer to the list of original flow actions.
4541  * @param[in] qrss
4542  *   Pointer to the Q/RSS action.
4543  * @param[in] actions_n
4544  *   Number of original actions.
4545  * @param[out] error
4546  *   Perform verbose error reporting if not NULL.
4547  *
4548  * @return
4549  *   non-zero unique flow_id on success, otherwise 0 and
4550  *   error/rte_error are set.
4551  */
4552 static uint32_t
4553 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4554 			  struct rte_flow_action *split_actions,
4555 			  const struct rte_flow_action *actions,
4556 			  const struct rte_flow_action *qrss,
4557 			  int actions_n, struct rte_flow_error *error)
4558 {
4559 	struct mlx5_priv *priv = dev->data->dev_private;
4560 	struct mlx5_rte_flow_action_set_tag *set_tag;
4561 	struct rte_flow_action_jump *jump;
4562 	const int qrss_idx = qrss - actions;
4563 	uint32_t flow_id = 0;
4564 	int ret = 0;
4565 
4566 	/*
4567 	 * Given actions will be split
4568 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4569 	 * - Add jump to mreg CP_TBL.
4570 	 * As a result, there will be one more action.
4571 	 */
4572 	++actions_n;
4573 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4574 	set_tag = (void *)(split_actions + actions_n);
4575 	/*
4576 	 * If tag action is not set to void(it means we are not the meter
4577 	 * suffix flow), add the tag action. Since meter suffix flow already
4578 	 * has the tag added.
4579 	 */
4580 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4581 		/*
4582 		 * Allocate the new subflow ID. This one is unique within
4583 		 * device and not shared with representors. Otherwise,
4584 		 * we would have to resolve multi-thread access synch
4585 		 * issue. Each flow on the shared device is appended
4586 		 * with source vport identifier, so the resulting
4587 		 * flows will be unique in the shared (by master and
4588 		 * representors) domain even if they have coinciding
4589 		 * IDs.
4590 		 */
4591 		mlx5_ipool_malloc(priv->sh->ipool
4592 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4593 		if (!flow_id)
4594 			return rte_flow_error_set(error, ENOMEM,
4595 						  RTE_FLOW_ERROR_TYPE_ACTION,
4596 						  NULL, "can't allocate id "
4597 						  "for split Q/RSS subflow");
4598 		/* Internal SET_TAG action to set flow ID. */
4599 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4600 			.data = flow_id,
4601 		};
4602 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4603 		if (ret < 0)
4604 			return ret;
4605 		set_tag->id = ret;
4606 		/* Construct new actions array. */
4607 		/* Replace QUEUE/RSS action. */
4608 		split_actions[qrss_idx] = (struct rte_flow_action){
4609 			.type = (enum rte_flow_action_type)
4610 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4611 			.conf = set_tag,
4612 		};
4613 	}
4614 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4615 	jump = (void *)(set_tag + 1);
4616 	*jump = (struct rte_flow_action_jump){
4617 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4618 	};
4619 	split_actions[actions_n - 2] = (struct rte_flow_action){
4620 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4621 		.conf = jump,
4622 	};
4623 	split_actions[actions_n - 1] = (struct rte_flow_action){
4624 		.type = RTE_FLOW_ACTION_TYPE_END,
4625 	};
4626 	return flow_id;
4627 }
4628 
4629 /**
4630  * Extend the given action list for Tx metadata copy.
4631  *
4632  * Copy the given action list to the ext_actions and add flow metadata register
4633  * copy action in order to copy reg_a set by WQE to reg_c[0].
4634  *
4635  * @param[out] ext_actions
4636  *   Pointer to the extended action list.
4637  * @param[in] actions
4638  *   Pointer to the list of actions.
4639  * @param[in] actions_n
4640  *   Number of actions in the list.
4641  * @param[out] error
4642  *   Perform verbose error reporting if not NULL.
4643  * @param[in] encap_idx
4644  *   The encap action inndex.
4645  *
4646  * @return
4647  *   0 on success, negative value otherwise
4648  */
4649 static int
4650 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4651 		       struct rte_flow_action *ext_actions,
4652 		       const struct rte_flow_action *actions,
4653 		       int actions_n, struct rte_flow_error *error,
4654 		       int encap_idx)
4655 {
4656 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4657 		(struct mlx5_flow_action_copy_mreg *)
4658 			(ext_actions + actions_n + 1);
4659 	int ret;
4660 
4661 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4662 	if (ret < 0)
4663 		return ret;
4664 	cp_mreg->dst = ret;
4665 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4666 	if (ret < 0)
4667 		return ret;
4668 	cp_mreg->src = ret;
4669 	if (encap_idx != 0)
4670 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4671 	if (encap_idx == actions_n - 1) {
4672 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4673 			.type = (enum rte_flow_action_type)
4674 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4675 			.conf = cp_mreg,
4676 		};
4677 		ext_actions[actions_n] = (struct rte_flow_action){
4678 			.type = RTE_FLOW_ACTION_TYPE_END,
4679 		};
4680 	} else {
4681 		ext_actions[encap_idx] = (struct rte_flow_action){
4682 			.type = (enum rte_flow_action_type)
4683 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4684 			.conf = cp_mreg,
4685 		};
4686 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4687 				sizeof(*ext_actions) * (actions_n - encap_idx));
4688 	}
4689 	return 0;
4690 }
4691 
4692 /**
4693  * Check the match action from the action list.
4694  *
4695  * @param[in] actions
4696  *   Pointer to the list of actions.
4697  * @param[in] attr
4698  *   Flow rule attributes.
4699  * @param[in] action
4700  *   The action to be check if exist.
4701  * @param[out] match_action_pos
4702  *   Pointer to the position of the matched action if exists, otherwise is -1.
4703  * @param[out] qrss_action_pos
4704  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4705  *
4706  * @return
4707  *   > 0 the total number of actions.
4708  *   0 if not found match action in action list.
4709  */
4710 static int
4711 flow_check_match_action(const struct rte_flow_action actions[],
4712 			const struct rte_flow_attr *attr,
4713 			enum rte_flow_action_type action,
4714 			int *match_action_pos, int *qrss_action_pos)
4715 {
4716 	const struct rte_flow_action_sample *sample;
4717 	int actions_n = 0;
4718 	int jump_flag = 0;
4719 	uint32_t ratio = 0;
4720 	int sub_type = 0;
4721 	int flag = 0;
4722 
4723 	*match_action_pos = -1;
4724 	*qrss_action_pos = -1;
4725 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4726 		if (actions->type == action) {
4727 			flag = 1;
4728 			*match_action_pos = actions_n;
4729 		}
4730 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4731 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4732 			*qrss_action_pos = actions_n;
4733 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4734 			jump_flag = 1;
4735 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4736 			sample = actions->conf;
4737 			ratio = sample->ratio;
4738 			sub_type = ((const struct rte_flow_action *)
4739 					(sample->actions))->type;
4740 		}
4741 		actions_n++;
4742 	}
4743 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4744 		if (ratio == 1) {
4745 			/* JUMP Action not support for Mirroring;
4746 			 * Mirroring support multi-destination;
4747 			 */
4748 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4749 				flag = 0;
4750 		}
4751 	}
4752 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4753 	return flag ? actions_n + 1 : 0;
4754 }
4755 
4756 #define SAMPLE_SUFFIX_ITEM 2
4757 
4758 /**
4759  * Split the sample flow.
4760  *
4761  * As sample flow will split to two sub flow, sample flow with
4762  * sample action, the other actions will move to new suffix flow.
4763  *
4764  * Also add unique tag id with tag action in the sample flow,
4765  * the same tag id will be as match in the suffix flow.
4766  *
4767  * @param dev
4768  *   Pointer to Ethernet device.
4769  * @param[in] fdb_tx
4770  *   FDB egress flow flag.
4771  * @param[out] sfx_items
4772  *   Suffix flow match items (list terminated by the END pattern item).
4773  * @param[in] actions
4774  *   Associated actions (list terminated by the END action).
4775  * @param[out] actions_sfx
4776  *   Suffix flow actions.
4777  * @param[out] actions_pre
4778  *   Prefix flow actions.
4779  * @param[in] actions_n
4780  *  The total number of actions.
4781  * @param[in] sample_action_pos
4782  *   The sample action position.
4783  * @param[in] qrss_action_pos
4784  *   The Queue/RSS action position.
4785  * @param[out] error
4786  *   Perform verbose error reporting if not NULL.
4787  *
4788  * @return
4789  *   0 on success, or unique flow_id, a negative errno value
4790  *   otherwise and rte_errno is set.
4791  */
4792 static int
4793 flow_sample_split_prep(struct rte_eth_dev *dev,
4794 		       uint32_t fdb_tx,
4795 		       struct rte_flow_item sfx_items[],
4796 		       const struct rte_flow_action actions[],
4797 		       struct rte_flow_action actions_sfx[],
4798 		       struct rte_flow_action actions_pre[],
4799 		       int actions_n,
4800 		       int sample_action_pos,
4801 		       int qrss_action_pos,
4802 		       struct rte_flow_error *error)
4803 {
4804 	struct mlx5_priv *priv = dev->data->dev_private;
4805 	struct mlx5_rte_flow_action_set_tag *set_tag;
4806 	struct mlx5_rte_flow_item_tag *tag_spec;
4807 	struct mlx5_rte_flow_item_tag *tag_mask;
4808 	uint32_t tag_id = 0;
4809 	int index;
4810 	int ret;
4811 
4812 	if (sample_action_pos < 0)
4813 		return rte_flow_error_set(error, EINVAL,
4814 					  RTE_FLOW_ERROR_TYPE_ACTION,
4815 					  NULL, "invalid position of sample "
4816 					  "action in list");
4817 	if (!fdb_tx) {
4818 		/* Prepare the prefix tag action. */
4819 		set_tag = (void *)(actions_pre + actions_n + 1);
4820 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4821 		if (ret < 0)
4822 			return ret;
4823 		set_tag->id = ret;
4824 		mlx5_ipool_malloc(priv->sh->ipool
4825 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4826 		set_tag->data = tag_id;
4827 		/* Prepare the suffix subflow items. */
4828 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4829 		tag_spec->data = tag_id;
4830 		tag_spec->id = set_tag->id;
4831 		tag_mask = tag_spec + 1;
4832 		tag_mask->data = UINT32_MAX;
4833 		sfx_items[0] = (struct rte_flow_item){
4834 			.type = (enum rte_flow_item_type)
4835 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4836 			.spec = tag_spec,
4837 			.last = NULL,
4838 			.mask = tag_mask,
4839 		};
4840 		sfx_items[1] = (struct rte_flow_item){
4841 			.type = (enum rte_flow_item_type)
4842 				RTE_FLOW_ITEM_TYPE_END,
4843 		};
4844 	}
4845 	/* Prepare the actions for prefix and suffix flow. */
4846 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4847 		index = qrss_action_pos;
4848 		/* Put the preceding the Queue/RSS action into prefix flow. */
4849 		if (index != 0)
4850 			memcpy(actions_pre, actions,
4851 			       sizeof(struct rte_flow_action) * index);
4852 		/* Put others preceding the sample action into prefix flow. */
4853 		if (sample_action_pos > index + 1)
4854 			memcpy(actions_pre + index, actions + index + 1,
4855 			       sizeof(struct rte_flow_action) *
4856 			       (sample_action_pos - index - 1));
4857 		index = sample_action_pos - 1;
4858 		/* Put Queue/RSS action into Suffix flow. */
4859 		memcpy(actions_sfx, actions + qrss_action_pos,
4860 		       sizeof(struct rte_flow_action));
4861 		actions_sfx++;
4862 	} else {
4863 		index = sample_action_pos;
4864 		if (index != 0)
4865 			memcpy(actions_pre, actions,
4866 			       sizeof(struct rte_flow_action) * index);
4867 	}
4868 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4869 	if (!fdb_tx) {
4870 		actions_pre[index++] =
4871 			(struct rte_flow_action){
4872 			.type = (enum rte_flow_action_type)
4873 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4874 			.conf = set_tag,
4875 		};
4876 	}
4877 	memcpy(actions_pre + index, actions + sample_action_pos,
4878 	       sizeof(struct rte_flow_action));
4879 	index += 1;
4880 	actions_pre[index] = (struct rte_flow_action){
4881 		.type = (enum rte_flow_action_type)
4882 			RTE_FLOW_ACTION_TYPE_END,
4883 	};
4884 	/* Put the actions after sample into Suffix flow. */
4885 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4886 	       sizeof(struct rte_flow_action) *
4887 	       (actions_n - sample_action_pos - 1));
4888 	return tag_id;
4889 }
4890 
4891 /**
4892  * The splitting for metadata feature.
4893  *
4894  * - Q/RSS action on NIC Rx should be split in order to pass by
4895  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4896  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4897  *
4898  * - All the actions on NIC Tx should have a mreg copy action to
4899  *   copy reg_a from WQE to reg_c[0].
4900  *
4901  * @param dev
4902  *   Pointer to Ethernet device.
4903  * @param[in] flow
4904  *   Parent flow structure pointer.
4905  * @param[in] attr
4906  *   Flow rule attributes.
4907  * @param[in] items
4908  *   Pattern specification (list terminated by the END pattern item).
4909  * @param[in] actions
4910  *   Associated actions (list terminated by the END action).
4911  * @param[in] flow_split_info
4912  *   Pointer to flow split info structure.
4913  * @param[out] error
4914  *   Perform verbose error reporting if not NULL.
4915  * @return
4916  *   0 on success, negative value otherwise
4917  */
4918 static int
4919 flow_create_split_metadata(struct rte_eth_dev *dev,
4920 			   struct rte_flow *flow,
4921 			   const struct rte_flow_attr *attr,
4922 			   const struct rte_flow_item items[],
4923 			   const struct rte_flow_action actions[],
4924 			   struct mlx5_flow_split_info *flow_split_info,
4925 			   struct rte_flow_error *error)
4926 {
4927 	struct mlx5_priv *priv = dev->data->dev_private;
4928 	struct mlx5_dev_config *config = &priv->config;
4929 	const struct rte_flow_action *qrss = NULL;
4930 	struct rte_flow_action *ext_actions = NULL;
4931 	struct mlx5_flow *dev_flow = NULL;
4932 	uint32_t qrss_id = 0;
4933 	int mtr_sfx = 0;
4934 	size_t act_size;
4935 	int actions_n;
4936 	int encap_idx;
4937 	int ret;
4938 
4939 	/* Check whether extensive metadata feature is engaged. */
4940 	if (!config->dv_flow_en ||
4941 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4942 	    !mlx5_flow_ext_mreg_supported(dev))
4943 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4944 					       actions, flow_split_info, error);
4945 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4946 							   &encap_idx);
4947 	if (qrss) {
4948 		/* Exclude hairpin flows from splitting. */
4949 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4950 			const struct rte_flow_action_queue *queue;
4951 
4952 			queue = qrss->conf;
4953 			if (mlx5_rxq_get_type(dev, queue->index) ==
4954 			    MLX5_RXQ_TYPE_HAIRPIN)
4955 				qrss = NULL;
4956 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4957 			const struct rte_flow_action_rss *rss;
4958 
4959 			rss = qrss->conf;
4960 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4961 			    MLX5_RXQ_TYPE_HAIRPIN)
4962 				qrss = NULL;
4963 		}
4964 	}
4965 	if (qrss) {
4966 		/* Check if it is in meter suffix table. */
4967 		mtr_sfx = attr->group == (attr->transfer ?
4968 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4969 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4970 		/*
4971 		 * Q/RSS action on NIC Rx should be split in order to pass by
4972 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4973 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4974 		 */
4975 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4976 			   sizeof(struct rte_flow_action_set_tag) +
4977 			   sizeof(struct rte_flow_action_jump);
4978 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4979 					  SOCKET_ID_ANY);
4980 		if (!ext_actions)
4981 			return rte_flow_error_set(error, ENOMEM,
4982 						  RTE_FLOW_ERROR_TYPE_ACTION,
4983 						  NULL, "no memory to split "
4984 						  "metadata flow");
4985 		/*
4986 		 * If we are the suffix flow of meter, tag already exist.
4987 		 * Set the tag action to void.
4988 		 */
4989 		if (mtr_sfx)
4990 			ext_actions[qrss - actions].type =
4991 						RTE_FLOW_ACTION_TYPE_VOID;
4992 		else
4993 			ext_actions[qrss - actions].type =
4994 						(enum rte_flow_action_type)
4995 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4996 		/*
4997 		 * Create the new actions list with removed Q/RSS action
4998 		 * and appended set tag and jump to register copy table
4999 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5000 		 * in advance, because it is needed for set tag action.
5001 		 */
5002 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5003 						    qrss, actions_n, error);
5004 		if (!mtr_sfx && !qrss_id) {
5005 			ret = -rte_errno;
5006 			goto exit;
5007 		}
5008 	} else if (attr->egress && !attr->transfer) {
5009 		/*
5010 		 * All the actions on NIC Tx should have a metadata register
5011 		 * copy action to copy reg_a from WQE to reg_c[meta]
5012 		 */
5013 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5014 			   sizeof(struct mlx5_flow_action_copy_mreg);
5015 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5016 					  SOCKET_ID_ANY);
5017 		if (!ext_actions)
5018 			return rte_flow_error_set(error, ENOMEM,
5019 						  RTE_FLOW_ERROR_TYPE_ACTION,
5020 						  NULL, "no memory to split "
5021 						  "metadata flow");
5022 		/* Create the action list appended with copy register. */
5023 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5024 					     actions_n, error, encap_idx);
5025 		if (ret < 0)
5026 			goto exit;
5027 	}
5028 	/* Add the unmodified original or prefix subflow. */
5029 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5030 				      items, ext_actions ? ext_actions :
5031 				      actions, flow_split_info, error);
5032 	if (ret < 0)
5033 		goto exit;
5034 	MLX5_ASSERT(dev_flow);
5035 	if (qrss) {
5036 		const struct rte_flow_attr q_attr = {
5037 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5038 			.ingress = 1,
5039 		};
5040 		/* Internal PMD action to set register. */
5041 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5042 			.data = qrss_id,
5043 			.id = REG_NON,
5044 		};
5045 		struct rte_flow_item q_items[] = {
5046 			{
5047 				.type = (enum rte_flow_item_type)
5048 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5049 				.spec = &q_tag_spec,
5050 				.last = NULL,
5051 				.mask = NULL,
5052 			},
5053 			{
5054 				.type = RTE_FLOW_ITEM_TYPE_END,
5055 			},
5056 		};
5057 		struct rte_flow_action q_actions[] = {
5058 			{
5059 				.type = qrss->type,
5060 				.conf = qrss->conf,
5061 			},
5062 			{
5063 				.type = RTE_FLOW_ACTION_TYPE_END,
5064 			},
5065 		};
5066 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5067 
5068 		/*
5069 		 * Configure the tag item only if there is no meter subflow.
5070 		 * Since tag is already marked in the meter suffix subflow
5071 		 * we can just use the meter suffix items as is.
5072 		 */
5073 		if (qrss_id) {
5074 			/* Not meter subflow. */
5075 			MLX5_ASSERT(!mtr_sfx);
5076 			/*
5077 			 * Put unique id in prefix flow due to it is destroyed
5078 			 * after suffix flow and id will be freed after there
5079 			 * is no actual flows with this id and identifier
5080 			 * reallocation becomes possible (for example, for
5081 			 * other flows in other threads).
5082 			 */
5083 			dev_flow->handle->split_flow_id = qrss_id;
5084 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5085 						   error);
5086 			if (ret < 0)
5087 				goto exit;
5088 			q_tag_spec.id = ret;
5089 		}
5090 		dev_flow = NULL;
5091 		/* Add suffix subflow to execute Q/RSS. */
5092 		flow_split_info->prefix_layers = layers;
5093 		flow_split_info->prefix_mark = 0;
5094 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5095 					      &q_attr, mtr_sfx ? items :
5096 					      q_items, q_actions,
5097 					      flow_split_info, error);
5098 		if (ret < 0)
5099 			goto exit;
5100 		/* qrss ID should be freed if failed. */
5101 		qrss_id = 0;
5102 		MLX5_ASSERT(dev_flow);
5103 	}
5104 
5105 exit:
5106 	/*
5107 	 * We do not destroy the partially created sub_flows in case of error.
5108 	 * These ones are included into parent flow list and will be destroyed
5109 	 * by flow_drv_destroy.
5110 	 */
5111 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5112 			qrss_id);
5113 	mlx5_free(ext_actions);
5114 	return ret;
5115 }
5116 
5117 /**
5118  * The splitting for meter feature.
5119  *
5120  * - The meter flow will be split to two flows as prefix and
5121  *   suffix flow. The packets make sense only it pass the prefix
5122  *   meter action.
5123  *
5124  * - Reg_C_5 is used for the packet to match betweend prefix and
5125  *   suffix flow.
5126  *
5127  * @param dev
5128  *   Pointer to Ethernet device.
5129  * @param[in] flow
5130  *   Parent flow structure pointer.
5131  * @param[in] attr
5132  *   Flow rule attributes.
5133  * @param[in] items
5134  *   Pattern specification (list terminated by the END pattern item).
5135  * @param[in] actions
5136  *   Associated actions (list terminated by the END action).
5137  * @param[in] flow_split_info
5138  *   Pointer to flow split info structure.
5139  * @param[out] error
5140  *   Perform verbose error reporting if not NULL.
5141  * @return
5142  *   0 on success, negative value otherwise
5143  */
5144 static int
5145 flow_create_split_meter(struct rte_eth_dev *dev,
5146 			struct rte_flow *flow,
5147 			const struct rte_flow_attr *attr,
5148 			const struct rte_flow_item items[],
5149 			const struct rte_flow_action actions[],
5150 			struct mlx5_flow_split_info *flow_split_info,
5151 			struct rte_flow_error *error)
5152 {
5153 	struct mlx5_priv *priv = dev->data->dev_private;
5154 	struct rte_flow_action *sfx_actions = NULL;
5155 	struct rte_flow_action *pre_actions = NULL;
5156 	struct rte_flow_item *sfx_items = NULL;
5157 	struct mlx5_flow *dev_flow = NULL;
5158 	struct rte_flow_attr sfx_attr = *attr;
5159 	uint32_t mtr = 0;
5160 	uint32_t mtr_tag_id = 0;
5161 	size_t act_size;
5162 	size_t item_size;
5163 	int actions_n = 0;
5164 	int ret;
5165 
5166 	if (priv->mtr_en)
5167 		actions_n = flow_check_meter_action(actions, &mtr);
5168 	if (mtr) {
5169 		/* The five prefix actions: meter, decap, encap, tag, end. */
5170 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5171 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5172 		/* tag, vlan, port id, end. */
5173 #define METER_SUFFIX_ITEM 4
5174 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5175 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5176 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5177 					  0, SOCKET_ID_ANY);
5178 		if (!sfx_actions)
5179 			return rte_flow_error_set(error, ENOMEM,
5180 						  RTE_FLOW_ERROR_TYPE_ACTION,
5181 						  NULL, "no memory to split "
5182 						  "meter flow");
5183 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5184 			     act_size);
5185 		pre_actions = sfx_actions + actions_n;
5186 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5187 						   actions, sfx_actions,
5188 						   pre_actions);
5189 		if (!mtr_tag_id) {
5190 			ret = -rte_errno;
5191 			goto exit;
5192 		}
5193 		/* Add the prefix subflow. */
5194 		flow_split_info->prefix_mark = 0;
5195 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5196 					      attr, items, pre_actions,
5197 					      flow_split_info, error);
5198 		if (ret) {
5199 			ret = -rte_errno;
5200 			goto exit;
5201 		}
5202 		dev_flow->handle->split_flow_id = mtr_tag_id;
5203 		/* Setting the sfx group atrr. */
5204 		sfx_attr.group = sfx_attr.transfer ?
5205 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5206 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5207 		flow_split_info->prefix_layers =
5208 				flow_get_prefix_layer_flags(dev_flow);
5209 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5210 	}
5211 	/* Add the prefix subflow. */
5212 	ret = flow_create_split_metadata(dev, flow,
5213 					 &sfx_attr, sfx_items ?
5214 					 sfx_items : items,
5215 					 sfx_actions ? sfx_actions : actions,
5216 					 flow_split_info, error);
5217 exit:
5218 	if (sfx_actions)
5219 		mlx5_free(sfx_actions);
5220 	return ret;
5221 }
5222 
5223 /**
5224  * The splitting for sample feature.
5225  *
5226  * Once Sample action is detected in the action list, the flow actions should
5227  * be split into prefix sub flow and suffix sub flow.
5228  *
5229  * The original items remain in the prefix sub flow, all actions preceding the
5230  * sample action and the sample action itself will be copied to the prefix
5231  * sub flow, the actions following the sample action will be copied to the
5232  * suffix sub flow, Queue action always be located in the suffix sub flow.
5233  *
5234  * In order to make the packet from prefix sub flow matches with suffix sub
5235  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5236  * flow uses tag item with the unique flow id.
5237  *
5238  * @param dev
5239  *   Pointer to Ethernet device.
5240  * @param[in] flow
5241  *   Parent flow structure pointer.
5242  * @param[in] attr
5243  *   Flow rule attributes.
5244  * @param[in] items
5245  *   Pattern specification (list terminated by the END pattern item).
5246  * @param[in] actions
5247  *   Associated actions (list terminated by the END action).
5248  * @param[in] flow_split_info
5249  *   Pointer to flow split info structure.
5250  * @param[out] error
5251  *   Perform verbose error reporting if not NULL.
5252  * @return
5253  *   0 on success, negative value otherwise
5254  */
5255 static int
5256 flow_create_split_sample(struct rte_eth_dev *dev,
5257 			 struct rte_flow *flow,
5258 			 const struct rte_flow_attr *attr,
5259 			 const struct rte_flow_item items[],
5260 			 const struct rte_flow_action actions[],
5261 			 struct mlx5_flow_split_info *flow_split_info,
5262 			 struct rte_flow_error *error)
5263 {
5264 	struct mlx5_priv *priv = dev->data->dev_private;
5265 	struct rte_flow_action *sfx_actions = NULL;
5266 	struct rte_flow_action *pre_actions = NULL;
5267 	struct rte_flow_item *sfx_items = NULL;
5268 	struct mlx5_flow *dev_flow = NULL;
5269 	struct rte_flow_attr sfx_attr = *attr;
5270 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5271 	struct mlx5_flow_dv_sample_resource *sample_res;
5272 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5273 	struct mlx5_flow_tbl_resource *sfx_tbl;
5274 	union mlx5_flow_tbl_key sfx_table_key;
5275 #endif
5276 	size_t act_size;
5277 	size_t item_size;
5278 	uint32_t fdb_tx = 0;
5279 	int32_t tag_id = 0;
5280 	int actions_n = 0;
5281 	int sample_action_pos;
5282 	int qrss_action_pos;
5283 	int ret = 0;
5284 
5285 	if (priv->sampler_en)
5286 		actions_n = flow_check_match_action(actions, attr,
5287 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5288 					&sample_action_pos, &qrss_action_pos);
5289 	if (actions_n) {
5290 		/* The prefix actions must includes sample, tag, end. */
5291 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5292 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5293 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5294 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5295 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5296 					  item_size), 0, SOCKET_ID_ANY);
5297 		if (!sfx_actions)
5298 			return rte_flow_error_set(error, ENOMEM,
5299 						  RTE_FLOW_ERROR_TYPE_ACTION,
5300 						  NULL, "no memory to split "
5301 						  "sample flow");
5302 		/* The representor_id is -1 for uplink. */
5303 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5304 		if (!fdb_tx)
5305 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5306 					+ act_size);
5307 		pre_actions = sfx_actions + actions_n;
5308 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5309 						actions, sfx_actions,
5310 						pre_actions, actions_n,
5311 						sample_action_pos,
5312 						qrss_action_pos, error);
5313 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5314 			ret = -rte_errno;
5315 			goto exit;
5316 		}
5317 		/* Add the prefix subflow. */
5318 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5319 					      items, pre_actions,
5320 					      flow_split_info, error);
5321 		if (ret) {
5322 			ret = -rte_errno;
5323 			goto exit;
5324 		}
5325 		dev_flow->handle->split_flow_id = tag_id;
5326 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5327 		/* Set the sfx group attr. */
5328 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5329 					dev_flow->dv.sample_res;
5330 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5331 					sample_res->normal_path_tbl;
5332 		sfx_tbl_data = container_of(sfx_tbl,
5333 					struct mlx5_flow_tbl_data_entry, tbl);
5334 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5335 		sfx_attr.group = sfx_attr.transfer ?
5336 					(sfx_table_key.table_id - 1) :
5337 					 sfx_table_key.table_id;
5338 		flow_split_info->prefix_layers =
5339 				flow_get_prefix_layer_flags(dev_flow);
5340 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5341 		/* Suffix group level already be scaled with factor, set
5342 		 * skip_scale to 1 to avoid scale again in translation.
5343 		 */
5344 		flow_split_info->skip_scale = 1;
5345 #endif
5346 	}
5347 	/* Add the suffix subflow. */
5348 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5349 				      sfx_items ? sfx_items : items,
5350 				      sfx_actions ? sfx_actions : actions,
5351 				      flow_split_info, error);
5352 exit:
5353 	if (sfx_actions)
5354 		mlx5_free(sfx_actions);
5355 	return ret;
5356 }
5357 
5358 /**
5359  * Split the flow to subflow set. The splitters might be linked
5360  * in the chain, like this:
5361  * flow_create_split_outer() calls:
5362  *   flow_create_split_meter() calls:
5363  *     flow_create_split_metadata(meter_subflow_0) calls:
5364  *       flow_create_split_inner(metadata_subflow_0)
5365  *       flow_create_split_inner(metadata_subflow_1)
5366  *       flow_create_split_inner(metadata_subflow_2)
5367  *     flow_create_split_metadata(meter_subflow_1) calls:
5368  *       flow_create_split_inner(metadata_subflow_0)
5369  *       flow_create_split_inner(metadata_subflow_1)
5370  *       flow_create_split_inner(metadata_subflow_2)
5371  *
5372  * This provide flexible way to add new levels of flow splitting.
5373  * The all of successfully created subflows are included to the
5374  * parent flow dev_flow list.
5375  *
5376  * @param dev
5377  *   Pointer to Ethernet device.
5378  * @param[in] flow
5379  *   Parent flow structure pointer.
5380  * @param[in] attr
5381  *   Flow rule attributes.
5382  * @param[in] items
5383  *   Pattern specification (list terminated by the END pattern item).
5384  * @param[in] actions
5385  *   Associated actions (list terminated by the END action).
5386  * @param[in] flow_split_info
5387  *   Pointer to flow split info structure.
5388  * @param[out] error
5389  *   Perform verbose error reporting if not NULL.
5390  * @return
5391  *   0 on success, negative value otherwise
5392  */
5393 static int
5394 flow_create_split_outer(struct rte_eth_dev *dev,
5395 			struct rte_flow *flow,
5396 			const struct rte_flow_attr *attr,
5397 			const struct rte_flow_item items[],
5398 			const struct rte_flow_action actions[],
5399 			struct mlx5_flow_split_info *flow_split_info,
5400 			struct rte_flow_error *error)
5401 {
5402 	int ret;
5403 
5404 	ret = flow_create_split_sample(dev, flow, attr, items,
5405 				       actions, flow_split_info, error);
5406 	MLX5_ASSERT(ret <= 0);
5407 	return ret;
5408 }
5409 
5410 static struct mlx5_flow_tunnel *
5411 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5412 		      const struct rte_flow_attr *attr,
5413 		      const struct rte_flow_item items[],
5414 		      const struct rte_flow_action actions[])
5415 {
5416 	struct mlx5_flow_tunnel *tunnel;
5417 
5418 #pragma GCC diagnostic push
5419 #pragma GCC diagnostic ignored "-Wcast-qual"
5420 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5421 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5422 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5423 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5424 	else
5425 		tunnel = NULL;
5426 #pragma GCC diagnostic pop
5427 
5428 	return tunnel;
5429 }
5430 
5431 /**
5432  * Adjust flow RSS workspace if needed.
5433  *
5434  * @param wks
5435  *   Pointer to thread flow work space.
5436  * @param rss_desc
5437  *   Pointer to RSS descriptor.
5438  * @param[in] nrssq_num
5439  *   New RSS queue number.
5440  *
5441  * @return
5442  *   0 on success, -1 otherwise and rte_errno is set.
5443  */
5444 static int
5445 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5446 			  struct mlx5_flow_rss_desc *rss_desc,
5447 			  uint32_t nrssq_num)
5448 {
5449 	if (likely(nrssq_num <= wks->rssq_num))
5450 		return 0;
5451 	rss_desc->queue = realloc(rss_desc->queue,
5452 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5453 	if (!rss_desc->queue) {
5454 		rte_errno = ENOMEM;
5455 		return -1;
5456 	}
5457 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5458 	return 0;
5459 }
5460 
5461 /**
5462  * Create a flow and add it to @p list.
5463  *
5464  * @param dev
5465  *   Pointer to Ethernet device.
5466  * @param list
5467  *   Pointer to a TAILQ flow list. If this parameter NULL,
5468  *   no list insertion occurred, flow is just created,
5469  *   this is caller's responsibility to track the
5470  *   created flow.
5471  * @param[in] attr
5472  *   Flow rule attributes.
5473  * @param[in] items
5474  *   Pattern specification (list terminated by the END pattern item).
5475  * @param[in] actions
5476  *   Associated actions (list terminated by the END action).
5477  * @param[in] external
5478  *   This flow rule is created by request external to PMD.
5479  * @param[out] error
5480  *   Perform verbose error reporting if not NULL.
5481  *
5482  * @return
5483  *   A flow index on success, 0 otherwise and rte_errno is set.
5484  */
5485 static uint32_t
5486 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5487 		 const struct rte_flow_attr *attr,
5488 		 const struct rte_flow_item items[],
5489 		 const struct rte_flow_action original_actions[],
5490 		 bool external, struct rte_flow_error *error)
5491 {
5492 	struct mlx5_priv *priv = dev->data->dev_private;
5493 	struct rte_flow *flow = NULL;
5494 	struct mlx5_flow *dev_flow;
5495 	const struct rte_flow_action_rss *rss;
5496 	struct mlx5_translated_shared_action
5497 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5498 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5499 	union {
5500 		struct mlx5_flow_expand_rss buf;
5501 		uint8_t buffer[2048];
5502 	} expand_buffer;
5503 	union {
5504 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5505 		uint8_t buffer[2048];
5506 	} actions_rx;
5507 	union {
5508 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5509 		uint8_t buffer[2048];
5510 	} actions_hairpin_tx;
5511 	union {
5512 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5513 		uint8_t buffer[2048];
5514 	} items_tx;
5515 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5516 	struct mlx5_flow_rss_desc *rss_desc;
5517 	const struct rte_flow_action *p_actions_rx;
5518 	uint32_t i;
5519 	uint32_t idx = 0;
5520 	int hairpin_flow;
5521 	struct rte_flow_attr attr_tx = { .priority = 0 };
5522 	const struct rte_flow_action *actions;
5523 	struct rte_flow_action *translated_actions = NULL;
5524 	struct mlx5_flow_tunnel *tunnel;
5525 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5526 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5527 	struct mlx5_flow_split_info flow_split_info = {
5528 		.external = !!external,
5529 		.skip_scale = 0,
5530 		.flow_idx = 0,
5531 		.prefix_mark = 0,
5532 		.prefix_layers = 0
5533 	};
5534 	int ret;
5535 
5536 	MLX5_ASSERT(wks);
5537 	rss_desc = &wks->rss_desc;
5538 	ret = flow_shared_actions_translate(dev, original_actions,
5539 					    shared_actions,
5540 					    &shared_actions_n,
5541 					    &translated_actions, error);
5542 	if (ret < 0) {
5543 		MLX5_ASSERT(translated_actions == NULL);
5544 		return 0;
5545 	}
5546 	actions = translated_actions ? translated_actions : original_actions;
5547 	p_actions_rx = actions;
5548 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5549 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5550 				external, hairpin_flow, error);
5551 	if (ret < 0)
5552 		goto error_before_hairpin_split;
5553 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5554 	if (!flow) {
5555 		rte_errno = ENOMEM;
5556 		goto error_before_hairpin_split;
5557 	}
5558 	if (hairpin_flow > 0) {
5559 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5560 			rte_errno = EINVAL;
5561 			goto error_before_hairpin_split;
5562 		}
5563 		flow_hairpin_split(dev, actions, actions_rx.actions,
5564 				   actions_hairpin_tx.actions, items_tx.items,
5565 				   idx);
5566 		p_actions_rx = actions_rx.actions;
5567 	}
5568 	flow_split_info.flow_idx = idx;
5569 	flow->drv_type = flow_get_drv_type(dev, attr);
5570 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5571 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5572 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5573 	rss = flow_get_rss_action(p_actions_rx);
5574 	if (rss) {
5575 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5576 			return 0;
5577 		/*
5578 		 * The following information is required by
5579 		 * mlx5_flow_hashfields_adjust() in advance.
5580 		 */
5581 		rss_desc->level = rss->level;
5582 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5583 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5584 	}
5585 	flow->dev_handles = 0;
5586 	if (rss && rss->types) {
5587 		unsigned int graph_root;
5588 
5589 		graph_root = find_graph_root(items, rss->level);
5590 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5591 					   items, rss->types,
5592 					   mlx5_support_expansion, graph_root);
5593 		MLX5_ASSERT(ret > 0 &&
5594 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5595 	} else {
5596 		buf->entries = 1;
5597 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5598 	}
5599 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5600 						      shared_actions_n);
5601 	for (i = 0; i < buf->entries; ++i) {
5602 		/* Initialize flow split data. */
5603 		flow_split_info.prefix_layers = 0;
5604 		flow_split_info.prefix_mark = 0;
5605 		flow_split_info.skip_scale = 0;
5606 		/*
5607 		 * The splitter may create multiple dev_flows,
5608 		 * depending on configuration. In the simplest
5609 		 * case it just creates unmodified original flow.
5610 		 */
5611 		ret = flow_create_split_outer(dev, flow, attr,
5612 					      buf->entry[i].pattern,
5613 					      p_actions_rx, &flow_split_info,
5614 					      error);
5615 		if (ret < 0)
5616 			goto error;
5617 		if (is_flow_tunnel_steer_rule(dev, attr,
5618 					      buf->entry[i].pattern,
5619 					      p_actions_rx)) {
5620 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5621 							   p_actions_rx,
5622 							   idx,
5623 							   &default_miss_ctx,
5624 							   error);
5625 			if (ret < 0) {
5626 				mlx5_free(default_miss_ctx.queue);
5627 				goto error;
5628 			}
5629 		}
5630 	}
5631 	/* Create the tx flow. */
5632 	if (hairpin_flow) {
5633 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5634 		attr_tx.ingress = 0;
5635 		attr_tx.egress = 1;
5636 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5637 					 actions_hairpin_tx.actions,
5638 					 idx, error);
5639 		if (!dev_flow)
5640 			goto error;
5641 		dev_flow->flow = flow;
5642 		dev_flow->external = 0;
5643 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5644 			      dev_flow->handle, next);
5645 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5646 					 items_tx.items,
5647 					 actions_hairpin_tx.actions, error);
5648 		if (ret < 0)
5649 			goto error;
5650 	}
5651 	/*
5652 	 * Update the metadata register copy table. If extensive
5653 	 * metadata feature is enabled and registers are supported
5654 	 * we might create the extra rte_flow for each unique
5655 	 * MARK/FLAG action ID.
5656 	 *
5657 	 * The table is updated for ingress Flows only, because
5658 	 * the egress Flows belong to the different device and
5659 	 * copy table should be updated in peer NIC Rx domain.
5660 	 */
5661 	if (attr->ingress &&
5662 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5663 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5664 		if (ret)
5665 			goto error;
5666 	}
5667 	/*
5668 	 * If the flow is external (from application) OR device is started,
5669 	 * OR mreg discover, then apply immediately.
5670 	 */
5671 	if (external || dev->data->dev_started ||
5672 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5673 	     attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5674 		ret = flow_drv_apply(dev, flow, error);
5675 		if (ret < 0)
5676 			goto error;
5677 	}
5678 	if (list) {
5679 		rte_spinlock_lock(&priv->flow_list_lock);
5680 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5681 			     flow, next);
5682 		rte_spinlock_unlock(&priv->flow_list_lock);
5683 	}
5684 	flow_rxq_flags_set(dev, flow);
5685 	rte_free(translated_actions);
5686 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5687 	if (tunnel) {
5688 		flow->tunnel = 1;
5689 		flow->tunnel_id = tunnel->tunnel_id;
5690 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5691 		mlx5_free(default_miss_ctx.queue);
5692 	}
5693 	mlx5_flow_pop_thread_workspace();
5694 	return idx;
5695 error:
5696 	MLX5_ASSERT(flow);
5697 	ret = rte_errno; /* Save rte_errno before cleanup. */
5698 	flow_mreg_del_copy_action(dev, flow);
5699 	flow_drv_destroy(dev, flow);
5700 	if (rss_desc->shared_rss)
5701 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5702 			mlx5_ipool_get
5703 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5704 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5705 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5706 	rte_errno = ret; /* Restore rte_errno. */
5707 	ret = rte_errno;
5708 	rte_errno = ret;
5709 	mlx5_flow_pop_thread_workspace();
5710 error_before_hairpin_split:
5711 	rte_free(translated_actions);
5712 	return 0;
5713 }
5714 
5715 /**
5716  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5717  * incoming packets to table 1.
5718  *
5719  * Other flow rules, requested for group n, will be created in
5720  * e-switch table n+1.
5721  * Jump action to e-switch group n will be created to group n+1.
5722  *
5723  * Used when working in switchdev mode, to utilise advantages of table 1
5724  * and above.
5725  *
5726  * @param dev
5727  *   Pointer to Ethernet device.
5728  *
5729  * @return
5730  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5731  */
5732 struct rte_flow *
5733 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5734 {
5735 	const struct rte_flow_attr attr = {
5736 		.group = 0,
5737 		.priority = 0,
5738 		.ingress = 1,
5739 		.egress = 0,
5740 		.transfer = 1,
5741 	};
5742 	const struct rte_flow_item pattern = {
5743 		.type = RTE_FLOW_ITEM_TYPE_END,
5744 	};
5745 	struct rte_flow_action_jump jump = {
5746 		.group = 1,
5747 	};
5748 	const struct rte_flow_action actions[] = {
5749 		{
5750 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5751 			.conf = &jump,
5752 		},
5753 		{
5754 			.type = RTE_FLOW_ACTION_TYPE_END,
5755 		},
5756 	};
5757 	struct mlx5_priv *priv = dev->data->dev_private;
5758 	struct rte_flow_error error;
5759 
5760 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5761 						   &attr, &pattern,
5762 						   actions, false, &error);
5763 }
5764 
5765 /**
5766  * Validate a flow supported by the NIC.
5767  *
5768  * @see rte_flow_validate()
5769  * @see rte_flow_ops
5770  */
5771 int
5772 mlx5_flow_validate(struct rte_eth_dev *dev,
5773 		   const struct rte_flow_attr *attr,
5774 		   const struct rte_flow_item items[],
5775 		   const struct rte_flow_action original_actions[],
5776 		   struct rte_flow_error *error)
5777 {
5778 	int hairpin_flow;
5779 	struct mlx5_translated_shared_action
5780 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5781 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5782 	const struct rte_flow_action *actions;
5783 	struct rte_flow_action *translated_actions = NULL;
5784 	int ret = flow_shared_actions_translate(dev, original_actions,
5785 						shared_actions,
5786 						&shared_actions_n,
5787 						&translated_actions, error);
5788 
5789 	if (ret)
5790 		return ret;
5791 	actions = translated_actions ? translated_actions : original_actions;
5792 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5793 	ret = flow_drv_validate(dev, attr, items, actions,
5794 				true, hairpin_flow, error);
5795 	rte_free(translated_actions);
5796 	return ret;
5797 }
5798 
5799 /**
5800  * Create a flow.
5801  *
5802  * @see rte_flow_create()
5803  * @see rte_flow_ops
5804  */
5805 struct rte_flow *
5806 mlx5_flow_create(struct rte_eth_dev *dev,
5807 		 const struct rte_flow_attr *attr,
5808 		 const struct rte_flow_item items[],
5809 		 const struct rte_flow_action actions[],
5810 		 struct rte_flow_error *error)
5811 {
5812 	struct mlx5_priv *priv = dev->data->dev_private;
5813 
5814 	/*
5815 	 * If the device is not started yet, it is not allowed to created a
5816 	 * flow from application. PMD default flows and traffic control flows
5817 	 * are not affected.
5818 	 */
5819 	if (unlikely(!dev->data->dev_started)) {
5820 		DRV_LOG(DEBUG, "port %u is not started when "
5821 			"inserting a flow", dev->data->port_id);
5822 		rte_flow_error_set(error, ENODEV,
5823 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5824 				   NULL,
5825 				   "port not started");
5826 		return NULL;
5827 	}
5828 
5829 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5830 				  attr, items, actions, true, error);
5831 }
5832 
5833 /**
5834  * Destroy a flow in a list.
5835  *
5836  * @param dev
5837  *   Pointer to Ethernet device.
5838  * @param list
5839  *   Pointer to the Indexed flow list. If this parameter NULL,
5840  *   there is no flow removal from the list. Be noted that as
5841  *   flow is add to the indexed list, memory of the indexed
5842  *   list points to maybe changed as flow destroyed.
5843  * @param[in] flow_idx
5844  *   Index of flow to destroy.
5845  */
5846 static void
5847 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5848 		  uint32_t flow_idx)
5849 {
5850 	struct mlx5_priv *priv = dev->data->dev_private;
5851 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5852 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5853 
5854 	if (!flow)
5855 		return;
5856 	/*
5857 	 * Update RX queue flags only if port is started, otherwise it is
5858 	 * already clean.
5859 	 */
5860 	if (dev->data->dev_started)
5861 		flow_rxq_flags_trim(dev, flow);
5862 	flow_drv_destroy(dev, flow);
5863 	if (list) {
5864 		rte_spinlock_lock(&priv->flow_list_lock);
5865 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5866 			     flow_idx, flow, next);
5867 		rte_spinlock_unlock(&priv->flow_list_lock);
5868 	}
5869 	if (flow->tunnel) {
5870 		struct mlx5_flow_tunnel *tunnel;
5871 
5872 		rte_spinlock_lock(&mlx5_tunnel_hub(dev)->sl);
5873 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5874 		RTE_VERIFY(tunnel);
5875 		LIST_REMOVE(tunnel, chain);
5876 		rte_spinlock_unlock(&mlx5_tunnel_hub(dev)->sl);
5877 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5878 			mlx5_flow_tunnel_free(dev, tunnel);
5879 	}
5880 	flow_mreg_del_copy_action(dev, flow);
5881 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5882 }
5883 
5884 /**
5885  * Destroy all flows.
5886  *
5887  * @param dev
5888  *   Pointer to Ethernet device.
5889  * @param list
5890  *   Pointer to the Indexed flow list.
5891  * @param active
5892  *   If flushing is called avtively.
5893  */
5894 void
5895 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5896 {
5897 	uint32_t num_flushed = 0;
5898 
5899 	while (*list) {
5900 		flow_list_destroy(dev, list, *list);
5901 		num_flushed++;
5902 	}
5903 	if (active) {
5904 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5905 			dev->data->port_id, num_flushed);
5906 	}
5907 }
5908 
5909 /**
5910  * Stop all default actions for flows.
5911  *
5912  * @param dev
5913  *   Pointer to Ethernet device.
5914  */
5915 void
5916 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5917 {
5918 	flow_mreg_del_default_copy_action(dev);
5919 	flow_rxq_flags_clear(dev);
5920 }
5921 
5922 /**
5923  * Start all default actions for flows.
5924  *
5925  * @param dev
5926  *   Pointer to Ethernet device.
5927  * @return
5928  *   0 on success, a negative errno value otherwise and rte_errno is set.
5929  */
5930 int
5931 mlx5_flow_start_default(struct rte_eth_dev *dev)
5932 {
5933 	struct rte_flow_error error;
5934 
5935 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5936 	return flow_mreg_add_default_copy_action(dev, &error);
5937 }
5938 
5939 /**
5940  * Release key of thread specific flow workspace data.
5941  */
5942 static void
5943 flow_release_workspace(void *data)
5944 {
5945 	struct mlx5_flow_workspace *wks = data;
5946 	struct mlx5_flow_workspace *next;
5947 
5948 	while (wks) {
5949 		next = wks->next;
5950 		free(wks->rss_desc.queue);
5951 		free(wks);
5952 		wks = next;
5953 	}
5954 }
5955 
5956 /**
5957  * Initialize key of thread specific flow workspace data.
5958  */
5959 static void
5960 flow_alloc_workspace(void)
5961 {
5962 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5963 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5964 }
5965 
5966 /**
5967  * Get thread specific current flow workspace.
5968  *
5969  * @return pointer to thread specific flow workspace data, NULL on error.
5970  */
5971 struct mlx5_flow_workspace*
5972 mlx5_flow_get_thread_workspace(void)
5973 {
5974 	struct mlx5_flow_workspace *data;
5975 
5976 	data = pthread_getspecific(key_workspace);
5977 	MLX5_ASSERT(data && data->inuse);
5978 	if (!data || !data->inuse)
5979 		DRV_LOG(ERR, "flow workspace not initialized.");
5980 	return data;
5981 }
5982 
5983 /**
5984  * Allocate and init new flow workspace.
5985  *
5986  * @return pointer to flow workspace data, NULL on error.
5987  */
5988 static struct mlx5_flow_workspace*
5989 flow_alloc_thread_workspace(void)
5990 {
5991 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5992 
5993 	if (!data) {
5994 		DRV_LOG(ERR, "Failed to allocate flow workspace "
5995 			"memory.");
5996 		return NULL;
5997 	}
5998 	data->rss_desc.queue = calloc(1,
5999 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
6000 	if (!data->rss_desc.queue)
6001 		goto err;
6002 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6003 	return data;
6004 err:
6005 	if (data->rss_desc.queue)
6006 		free(data->rss_desc.queue);
6007 	free(data);
6008 	return NULL;
6009 }
6010 
6011 /**
6012  * Get new thread specific flow workspace.
6013  *
6014  * If current workspace inuse, create new one and set as current.
6015  *
6016  * @return pointer to thread specific flow workspace data, NULL on error.
6017  */
6018 static struct mlx5_flow_workspace*
6019 mlx5_flow_push_thread_workspace(void)
6020 {
6021 	struct mlx5_flow_workspace *curr;
6022 	struct mlx5_flow_workspace *data;
6023 
6024 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
6025 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
6026 		return NULL;
6027 	}
6028 	curr = pthread_getspecific(key_workspace);
6029 	if (!curr) {
6030 		data = flow_alloc_thread_workspace();
6031 		if (!data)
6032 			return NULL;
6033 	} else if (!curr->inuse) {
6034 		data = curr;
6035 	} else if (curr->next) {
6036 		data = curr->next;
6037 	} else {
6038 		data = flow_alloc_thread_workspace();
6039 		if (!data)
6040 			return NULL;
6041 		curr->next = data;
6042 		data->prev = curr;
6043 	}
6044 	data->inuse = 1;
6045 	data->flow_idx = 0;
6046 	/* Set as current workspace */
6047 	if (pthread_setspecific(key_workspace, data))
6048 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6049 	return data;
6050 }
6051 
6052 /**
6053  * Close current thread specific flow workspace.
6054  *
6055  * If previous workspace available, set it as current.
6056  *
6057  * @return pointer to thread specific flow workspace data, NULL on error.
6058  */
6059 static void
6060 mlx5_flow_pop_thread_workspace(void)
6061 {
6062 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6063 
6064 	if (!data)
6065 		return;
6066 	if (!data->inuse) {
6067 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6068 		return;
6069 	}
6070 	data->inuse = 0;
6071 	if (!data->prev)
6072 		return;
6073 	if (pthread_setspecific(key_workspace, data->prev))
6074 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6075 }
6076 
6077 /**
6078  * Verify the flow list is empty
6079  *
6080  * @param dev
6081  *  Pointer to Ethernet device.
6082  *
6083  * @return the number of flows not released.
6084  */
6085 int
6086 mlx5_flow_verify(struct rte_eth_dev *dev)
6087 {
6088 	struct mlx5_priv *priv = dev->data->dev_private;
6089 	struct rte_flow *flow;
6090 	uint32_t idx;
6091 	int ret = 0;
6092 
6093 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6094 		      flow, next) {
6095 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6096 			dev->data->port_id, (void *)flow);
6097 		++ret;
6098 	}
6099 	return ret;
6100 }
6101 
6102 /**
6103  * Enable default hairpin egress flow.
6104  *
6105  * @param dev
6106  *   Pointer to Ethernet device.
6107  * @param queue
6108  *   The queue index.
6109  *
6110  * @return
6111  *   0 on success, a negative errno value otherwise and rte_errno is set.
6112  */
6113 int
6114 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6115 			    uint32_t queue)
6116 {
6117 	struct mlx5_priv *priv = dev->data->dev_private;
6118 	const struct rte_flow_attr attr = {
6119 		.egress = 1,
6120 		.priority = 0,
6121 	};
6122 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6123 		.queue = queue,
6124 	};
6125 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6126 		.queue = UINT32_MAX,
6127 	};
6128 	struct rte_flow_item items[] = {
6129 		{
6130 			.type = (enum rte_flow_item_type)
6131 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6132 			.spec = &queue_spec,
6133 			.last = NULL,
6134 			.mask = &queue_mask,
6135 		},
6136 		{
6137 			.type = RTE_FLOW_ITEM_TYPE_END,
6138 		},
6139 	};
6140 	struct rte_flow_action_jump jump = {
6141 		.group = MLX5_HAIRPIN_TX_TABLE,
6142 	};
6143 	struct rte_flow_action actions[2];
6144 	uint32_t flow_idx;
6145 	struct rte_flow_error error;
6146 
6147 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6148 	actions[0].conf = &jump;
6149 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6150 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6151 				&attr, items, actions, false, &error);
6152 	if (!flow_idx) {
6153 		DRV_LOG(DEBUG,
6154 			"Failed to create ctrl flow: rte_errno(%d),"
6155 			" type(%d), message(%s)",
6156 			rte_errno, error.type,
6157 			error.message ? error.message : " (no stated reason)");
6158 		return -rte_errno;
6159 	}
6160 	return 0;
6161 }
6162 
6163 /**
6164  * Enable a control flow configured from the control plane.
6165  *
6166  * @param dev
6167  *   Pointer to Ethernet device.
6168  * @param eth_spec
6169  *   An Ethernet flow spec to apply.
6170  * @param eth_mask
6171  *   An Ethernet flow mask to apply.
6172  * @param vlan_spec
6173  *   A VLAN flow spec to apply.
6174  * @param vlan_mask
6175  *   A VLAN flow mask to apply.
6176  *
6177  * @return
6178  *   0 on success, a negative errno value otherwise and rte_errno is set.
6179  */
6180 int
6181 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6182 		    struct rte_flow_item_eth *eth_spec,
6183 		    struct rte_flow_item_eth *eth_mask,
6184 		    struct rte_flow_item_vlan *vlan_spec,
6185 		    struct rte_flow_item_vlan *vlan_mask)
6186 {
6187 	struct mlx5_priv *priv = dev->data->dev_private;
6188 	const struct rte_flow_attr attr = {
6189 		.ingress = 1,
6190 		.priority = MLX5_FLOW_PRIO_RSVD,
6191 	};
6192 	struct rte_flow_item items[] = {
6193 		{
6194 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6195 			.spec = eth_spec,
6196 			.last = NULL,
6197 			.mask = eth_mask,
6198 		},
6199 		{
6200 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6201 					      RTE_FLOW_ITEM_TYPE_END,
6202 			.spec = vlan_spec,
6203 			.last = NULL,
6204 			.mask = vlan_mask,
6205 		},
6206 		{
6207 			.type = RTE_FLOW_ITEM_TYPE_END,
6208 		},
6209 	};
6210 	uint16_t queue[priv->reta_idx_n];
6211 	struct rte_flow_action_rss action_rss = {
6212 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6213 		.level = 0,
6214 		.types = priv->rss_conf.rss_hf,
6215 		.key_len = priv->rss_conf.rss_key_len,
6216 		.queue_num = priv->reta_idx_n,
6217 		.key = priv->rss_conf.rss_key,
6218 		.queue = queue,
6219 	};
6220 	struct rte_flow_action actions[] = {
6221 		{
6222 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6223 			.conf = &action_rss,
6224 		},
6225 		{
6226 			.type = RTE_FLOW_ACTION_TYPE_END,
6227 		},
6228 	};
6229 	uint32_t flow_idx;
6230 	struct rte_flow_error error;
6231 	unsigned int i;
6232 
6233 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6234 		return 0;
6235 	}
6236 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6237 		action_rss.types = 0;
6238 	for (i = 0; i != priv->reta_idx_n; ++i)
6239 		queue[i] = (*priv->reta_idx)[i];
6240 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6241 				&attr, items, actions, false, &error);
6242 	if (!flow_idx)
6243 		return -rte_errno;
6244 	return 0;
6245 }
6246 
6247 /**
6248  * Enable a flow control configured from the control plane.
6249  *
6250  * @param dev
6251  *   Pointer to Ethernet device.
6252  * @param eth_spec
6253  *   An Ethernet flow spec to apply.
6254  * @param eth_mask
6255  *   An Ethernet flow mask to apply.
6256  *
6257  * @return
6258  *   0 on success, a negative errno value otherwise and rte_errno is set.
6259  */
6260 int
6261 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6262 	       struct rte_flow_item_eth *eth_spec,
6263 	       struct rte_flow_item_eth *eth_mask)
6264 {
6265 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6266 }
6267 
6268 /**
6269  * Create default miss flow rule matching lacp traffic
6270  *
6271  * @param dev
6272  *   Pointer to Ethernet device.
6273  * @param eth_spec
6274  *   An Ethernet flow spec to apply.
6275  *
6276  * @return
6277  *   0 on success, a negative errno value otherwise and rte_errno is set.
6278  */
6279 int
6280 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6281 {
6282 	struct mlx5_priv *priv = dev->data->dev_private;
6283 	/*
6284 	 * The LACP matching is done by only using ether type since using
6285 	 * a multicast dst mac causes kernel to give low priority to this flow.
6286 	 */
6287 	static const struct rte_flow_item_eth lacp_spec = {
6288 		.type = RTE_BE16(0x8809),
6289 	};
6290 	static const struct rte_flow_item_eth lacp_mask = {
6291 		.type = 0xffff,
6292 	};
6293 	const struct rte_flow_attr attr = {
6294 		.ingress = 1,
6295 	};
6296 	struct rte_flow_item items[] = {
6297 		{
6298 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6299 			.spec = &lacp_spec,
6300 			.mask = &lacp_mask,
6301 		},
6302 		{
6303 			.type = RTE_FLOW_ITEM_TYPE_END,
6304 		},
6305 	};
6306 	struct rte_flow_action actions[] = {
6307 		{
6308 			.type = (enum rte_flow_action_type)
6309 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6310 		},
6311 		{
6312 			.type = RTE_FLOW_ACTION_TYPE_END,
6313 		},
6314 	};
6315 	struct rte_flow_error error;
6316 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6317 				&attr, items, actions, false, &error);
6318 
6319 	if (!flow_idx)
6320 		return -rte_errno;
6321 	return 0;
6322 }
6323 
6324 /**
6325  * Destroy a flow.
6326  *
6327  * @see rte_flow_destroy()
6328  * @see rte_flow_ops
6329  */
6330 int
6331 mlx5_flow_destroy(struct rte_eth_dev *dev,
6332 		  struct rte_flow *flow,
6333 		  struct rte_flow_error *error __rte_unused)
6334 {
6335 	struct mlx5_priv *priv = dev->data->dev_private;
6336 
6337 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6338 	return 0;
6339 }
6340 
6341 /**
6342  * Destroy all flows.
6343  *
6344  * @see rte_flow_flush()
6345  * @see rte_flow_ops
6346  */
6347 int
6348 mlx5_flow_flush(struct rte_eth_dev *dev,
6349 		struct rte_flow_error *error __rte_unused)
6350 {
6351 	struct mlx5_priv *priv = dev->data->dev_private;
6352 
6353 	mlx5_flow_list_flush(dev, &priv->flows, false);
6354 	return 0;
6355 }
6356 
6357 /**
6358  * Isolated mode.
6359  *
6360  * @see rte_flow_isolate()
6361  * @see rte_flow_ops
6362  */
6363 int
6364 mlx5_flow_isolate(struct rte_eth_dev *dev,
6365 		  int enable,
6366 		  struct rte_flow_error *error)
6367 {
6368 	struct mlx5_priv *priv = dev->data->dev_private;
6369 
6370 	if (dev->data->dev_started) {
6371 		rte_flow_error_set(error, EBUSY,
6372 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6373 				   NULL,
6374 				   "port must be stopped first");
6375 		return -rte_errno;
6376 	}
6377 	priv->isolated = !!enable;
6378 	if (enable)
6379 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6380 	else
6381 		dev->dev_ops = &mlx5_os_dev_ops;
6382 
6383 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6384 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6385 
6386 	return 0;
6387 }
6388 
6389 /**
6390  * Query a flow.
6391  *
6392  * @see rte_flow_query()
6393  * @see rte_flow_ops
6394  */
6395 static int
6396 flow_drv_query(struct rte_eth_dev *dev,
6397 	       uint32_t flow_idx,
6398 	       const struct rte_flow_action *actions,
6399 	       void *data,
6400 	       struct rte_flow_error *error)
6401 {
6402 	struct mlx5_priv *priv = dev->data->dev_private;
6403 	const struct mlx5_flow_driver_ops *fops;
6404 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6405 					       [MLX5_IPOOL_RTE_FLOW],
6406 					       flow_idx);
6407 	enum mlx5_flow_drv_type ftype;
6408 
6409 	if (!flow) {
6410 		return rte_flow_error_set(error, ENOENT,
6411 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6412 			  NULL,
6413 			  "invalid flow handle");
6414 	}
6415 	ftype = flow->drv_type;
6416 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6417 	fops = flow_get_drv_ops(ftype);
6418 
6419 	return fops->query(dev, flow, actions, data, error);
6420 }
6421 
6422 /**
6423  * Query a flow.
6424  *
6425  * @see rte_flow_query()
6426  * @see rte_flow_ops
6427  */
6428 int
6429 mlx5_flow_query(struct rte_eth_dev *dev,
6430 		struct rte_flow *flow,
6431 		const struct rte_flow_action *actions,
6432 		void *data,
6433 		struct rte_flow_error *error)
6434 {
6435 	int ret;
6436 
6437 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6438 			     error);
6439 	if (ret < 0)
6440 		return ret;
6441 	return 0;
6442 }
6443 
6444 /**
6445  * Manage filter operations.
6446  *
6447  * @param dev
6448  *   Pointer to Ethernet device structure.
6449  * @param filter_type
6450  *   Filter type.
6451  * @param filter_op
6452  *   Operation to perform.
6453  * @param arg
6454  *   Pointer to operation-specific structure.
6455  *
6456  * @return
6457  *   0 on success, a negative errno value otherwise and rte_errno is set.
6458  */
6459 int
6460 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6461 		     enum rte_filter_type filter_type,
6462 		     enum rte_filter_op filter_op,
6463 		     void *arg)
6464 {
6465 	switch (filter_type) {
6466 	case RTE_ETH_FILTER_GENERIC:
6467 		if (filter_op != RTE_ETH_FILTER_GET) {
6468 			rte_errno = EINVAL;
6469 			return -rte_errno;
6470 		}
6471 		*(const void **)arg = &mlx5_flow_ops;
6472 		return 0;
6473 	default:
6474 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6475 			dev->data->port_id, filter_type);
6476 		rte_errno = ENOTSUP;
6477 		return -rte_errno;
6478 	}
6479 	return 0;
6480 }
6481 
6482 /**
6483  * Create the needed meter and suffix tables.
6484  *
6485  * @param[in] dev
6486  *   Pointer to Ethernet device.
6487  * @param[in] fm
6488  *   Pointer to the flow meter.
6489  *
6490  * @return
6491  *   Pointer to table set on success, NULL otherwise.
6492  */
6493 struct mlx5_meter_domains_infos *
6494 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6495 			  const struct mlx5_flow_meter *fm)
6496 {
6497 	const struct mlx5_flow_driver_ops *fops;
6498 
6499 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6500 	return fops->create_mtr_tbls(dev, fm);
6501 }
6502 
6503 /**
6504  * Destroy the meter table set.
6505  *
6506  * @param[in] dev
6507  *   Pointer to Ethernet device.
6508  * @param[in] tbl
6509  *   Pointer to the meter table set.
6510  *
6511  * @return
6512  *   0 on success.
6513  */
6514 int
6515 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6516 			   struct mlx5_meter_domains_infos *tbls)
6517 {
6518 	const struct mlx5_flow_driver_ops *fops;
6519 
6520 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6521 	return fops->destroy_mtr_tbls(dev, tbls);
6522 }
6523 
6524 /**
6525  * Create policer rules.
6526  *
6527  * @param[in] dev
6528  *   Pointer to Ethernet device.
6529  * @param[in] fm
6530  *   Pointer to flow meter structure.
6531  * @param[in] attr
6532  *   Pointer to flow attributes.
6533  *
6534  * @return
6535  *   0 on success, -1 otherwise.
6536  */
6537 int
6538 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6539 			       struct mlx5_flow_meter *fm,
6540 			       const struct rte_flow_attr *attr)
6541 {
6542 	const struct mlx5_flow_driver_ops *fops;
6543 
6544 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6545 	return fops->create_policer_rules(dev, fm, attr);
6546 }
6547 
6548 /**
6549  * Destroy policer rules.
6550  *
6551  * @param[in] fm
6552  *   Pointer to flow meter structure.
6553  * @param[in] attr
6554  *   Pointer to flow attributes.
6555  *
6556  * @return
6557  *   0 on success, -1 otherwise.
6558  */
6559 int
6560 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6561 				struct mlx5_flow_meter *fm,
6562 				const struct rte_flow_attr *attr)
6563 {
6564 	const struct mlx5_flow_driver_ops *fops;
6565 
6566 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6567 	return fops->destroy_policer_rules(dev, fm, attr);
6568 }
6569 
6570 /**
6571  * Allocate a counter.
6572  *
6573  * @param[in] dev
6574  *   Pointer to Ethernet device structure.
6575  *
6576  * @return
6577  *   Index to allocated counter  on success, 0 otherwise.
6578  */
6579 uint32_t
6580 mlx5_counter_alloc(struct rte_eth_dev *dev)
6581 {
6582 	const struct mlx5_flow_driver_ops *fops;
6583 	struct rte_flow_attr attr = { .transfer = 0 };
6584 
6585 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6586 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6587 		return fops->counter_alloc(dev);
6588 	}
6589 	DRV_LOG(ERR,
6590 		"port %u counter allocate is not supported.",
6591 		 dev->data->port_id);
6592 	return 0;
6593 }
6594 
6595 /**
6596  * Free a counter.
6597  *
6598  * @param[in] dev
6599  *   Pointer to Ethernet device structure.
6600  * @param[in] cnt
6601  *   Index to counter to be free.
6602  */
6603 void
6604 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6605 {
6606 	const struct mlx5_flow_driver_ops *fops;
6607 	struct rte_flow_attr attr = { .transfer = 0 };
6608 
6609 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6610 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6611 		fops->counter_free(dev, cnt);
6612 		return;
6613 	}
6614 	DRV_LOG(ERR,
6615 		"port %u counter free is not supported.",
6616 		 dev->data->port_id);
6617 }
6618 
6619 /**
6620  * Query counter statistics.
6621  *
6622  * @param[in] dev
6623  *   Pointer to Ethernet device structure.
6624  * @param[in] cnt
6625  *   Index to counter to query.
6626  * @param[in] clear
6627  *   Set to clear counter statistics.
6628  * @param[out] pkts
6629  *   The counter hits packets number to save.
6630  * @param[out] bytes
6631  *   The counter hits bytes number to save.
6632  *
6633  * @return
6634  *   0 on success, a negative errno value otherwise.
6635  */
6636 int
6637 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6638 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6639 {
6640 	const struct mlx5_flow_driver_ops *fops;
6641 	struct rte_flow_attr attr = { .transfer = 0 };
6642 
6643 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6644 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6645 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6646 	}
6647 	DRV_LOG(ERR,
6648 		"port %u counter query is not supported.",
6649 		 dev->data->port_id);
6650 	return -ENOTSUP;
6651 }
6652 
6653 /**
6654  * Allocate a new memory for the counter values wrapped by all the needed
6655  * management.
6656  *
6657  * @param[in] sh
6658  *   Pointer to mlx5_dev_ctx_shared object.
6659  *
6660  * @return
6661  *   0 on success, a negative errno value otherwise.
6662  */
6663 static int
6664 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6665 {
6666 	struct mlx5_devx_mkey_attr mkey_attr;
6667 	struct mlx5_counter_stats_mem_mng *mem_mng;
6668 	volatile struct flow_counter_stats *raw_data;
6669 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6670 	int size = (sizeof(struct flow_counter_stats) *
6671 			MLX5_COUNTERS_PER_POOL +
6672 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6673 			sizeof(struct mlx5_counter_stats_mem_mng);
6674 	size_t pgsize = rte_mem_page_size();
6675 	uint8_t *mem;
6676 	int i;
6677 
6678 	if (pgsize == (size_t)-1) {
6679 		DRV_LOG(ERR, "Failed to get mem page size");
6680 		rte_errno = ENOMEM;
6681 		return -ENOMEM;
6682 	}
6683 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6684 	if (!mem) {
6685 		rte_errno = ENOMEM;
6686 		return -ENOMEM;
6687 	}
6688 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6689 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6690 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
6691 						 IBV_ACCESS_LOCAL_WRITE);
6692 	if (!mem_mng->umem) {
6693 		rte_errno = errno;
6694 		mlx5_free(mem);
6695 		return -rte_errno;
6696 	}
6697 	mkey_attr.addr = (uintptr_t)mem;
6698 	mkey_attr.size = size;
6699 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6700 	mkey_attr.pd = sh->pdn;
6701 	mkey_attr.log_entity_size = 0;
6702 	mkey_attr.pg_access = 0;
6703 	mkey_attr.klm_array = NULL;
6704 	mkey_attr.klm_num = 0;
6705 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6706 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6707 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6708 	if (!mem_mng->dm) {
6709 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
6710 		rte_errno = errno;
6711 		mlx5_free(mem);
6712 		return -rte_errno;
6713 	}
6714 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6715 	raw_data = (volatile struct flow_counter_stats *)mem;
6716 	for (i = 0; i < raws_n; ++i) {
6717 		mem_mng->raws[i].mem_mng = mem_mng;
6718 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6719 	}
6720 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6721 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6722 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6723 				 next);
6724 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6725 	sh->cmng.mem_mng = mem_mng;
6726 	return 0;
6727 }
6728 
6729 /**
6730  * Set the statistic memory to the new counter pool.
6731  *
6732  * @param[in] sh
6733  *   Pointer to mlx5_dev_ctx_shared object.
6734  * @param[in] pool
6735  *   Pointer to the pool to set the statistic memory.
6736  *
6737  * @return
6738  *   0 on success, a negative errno value otherwise.
6739  */
6740 static int
6741 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6742 			       struct mlx5_flow_counter_pool *pool)
6743 {
6744 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6745 	/* Resize statistic memory once used out. */
6746 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6747 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6748 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6749 		return -1;
6750 	}
6751 	rte_spinlock_lock(&pool->sl);
6752 	pool->raw = cmng->mem_mng->raws + pool->index %
6753 		    MLX5_CNT_CONTAINER_RESIZE;
6754 	rte_spinlock_unlock(&pool->sl);
6755 	pool->raw_hw = NULL;
6756 	return 0;
6757 }
6758 
6759 #define MLX5_POOL_QUERY_FREQ_US 1000000
6760 
6761 /**
6762  * Set the periodic procedure for triggering asynchronous batch queries for all
6763  * the counter pools.
6764  *
6765  * @param[in] sh
6766  *   Pointer to mlx5_dev_ctx_shared object.
6767  */
6768 void
6769 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6770 {
6771 	uint32_t pools_n, us;
6772 
6773 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6774 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6775 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6776 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6777 		sh->cmng.query_thread_on = 0;
6778 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6779 	} else {
6780 		sh->cmng.query_thread_on = 1;
6781 	}
6782 }
6783 
6784 /**
6785  * The periodic procedure for triggering asynchronous batch queries for all the
6786  * counter pools. This function is probably called by the host thread.
6787  *
6788  * @param[in] arg
6789  *   The parameter for the alarm process.
6790  */
6791 void
6792 mlx5_flow_query_alarm(void *arg)
6793 {
6794 	struct mlx5_dev_ctx_shared *sh = arg;
6795 	int ret;
6796 	uint16_t pool_index = sh->cmng.pool_index;
6797 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6798 	struct mlx5_flow_counter_pool *pool;
6799 	uint16_t n_valid;
6800 
6801 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6802 		goto set_alarm;
6803 	rte_spinlock_lock(&cmng->pool_update_sl);
6804 	pool = cmng->pools[pool_index];
6805 	n_valid = cmng->n_valid;
6806 	rte_spinlock_unlock(&cmng->pool_update_sl);
6807 	/* Set the statistic memory to the new created pool. */
6808 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6809 		goto set_alarm;
6810 	if (pool->raw_hw)
6811 		/* There is a pool query in progress. */
6812 		goto set_alarm;
6813 	pool->raw_hw =
6814 		LIST_FIRST(&sh->cmng.free_stat_raws);
6815 	if (!pool->raw_hw)
6816 		/* No free counter statistics raw memory. */
6817 		goto set_alarm;
6818 	/*
6819 	 * Identify the counters released between query trigger and query
6820 	 * handle more efficiently. The counter released in this gap period
6821 	 * should wait for a new round of query as the new arrived packets
6822 	 * will not be taken into account.
6823 	 */
6824 	pool->query_gen++;
6825 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6826 					       MLX5_COUNTERS_PER_POOL,
6827 					       NULL, NULL,
6828 					       pool->raw_hw->mem_mng->dm->id,
6829 					       (void *)(uintptr_t)
6830 					       pool->raw_hw->data,
6831 					       sh->devx_comp,
6832 					       (uint64_t)(uintptr_t)pool);
6833 	if (ret) {
6834 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6835 			" %d", pool->min_dcs->id);
6836 		pool->raw_hw = NULL;
6837 		goto set_alarm;
6838 	}
6839 	LIST_REMOVE(pool->raw_hw, next);
6840 	sh->cmng.pending_queries++;
6841 	pool_index++;
6842 	if (pool_index >= n_valid)
6843 		pool_index = 0;
6844 set_alarm:
6845 	sh->cmng.pool_index = pool_index;
6846 	mlx5_set_query_alarm(sh);
6847 }
6848 
6849 /**
6850  * Check and callback event for new aged flow in the counter pool
6851  *
6852  * @param[in] sh
6853  *   Pointer to mlx5_dev_ctx_shared object.
6854  * @param[in] pool
6855  *   Pointer to Current counter pool.
6856  */
6857 static void
6858 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6859 		   struct mlx5_flow_counter_pool *pool)
6860 {
6861 	struct mlx5_priv *priv;
6862 	struct mlx5_flow_counter *cnt;
6863 	struct mlx5_age_info *age_info;
6864 	struct mlx5_age_param *age_param;
6865 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6866 	struct mlx5_counter_stats_raw *prev = pool->raw;
6867 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6868 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6869 	uint16_t expected = AGE_CANDIDATE;
6870 	uint32_t i;
6871 
6872 	pool->time_of_last_age_check = curr_time;
6873 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6874 		cnt = MLX5_POOL_GET_CNT(pool, i);
6875 		age_param = MLX5_CNT_TO_AGE(cnt);
6876 		if (__atomic_load_n(&age_param->state,
6877 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6878 			continue;
6879 		if (cur->data[i].hits != prev->data[i].hits) {
6880 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6881 					 __ATOMIC_RELAXED);
6882 			continue;
6883 		}
6884 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6885 				       time_delta,
6886 				       __ATOMIC_RELAXED) <= age_param->timeout)
6887 			continue;
6888 		/**
6889 		 * Hold the lock first, or if between the
6890 		 * state AGE_TMOUT and tailq operation the
6891 		 * release happened, the release procedure
6892 		 * may delete a non-existent tailq node.
6893 		 */
6894 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6895 		age_info = GET_PORT_AGE_INFO(priv);
6896 		rte_spinlock_lock(&age_info->aged_sl);
6897 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6898 						AGE_TMOUT, false,
6899 						__ATOMIC_RELAXED,
6900 						__ATOMIC_RELAXED)) {
6901 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6902 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6903 		}
6904 		rte_spinlock_unlock(&age_info->aged_sl);
6905 	}
6906 	mlx5_age_event_prepare(sh);
6907 }
6908 
6909 /**
6910  * Handler for the HW respond about ready values from an asynchronous batch
6911  * query. This function is probably called by the host thread.
6912  *
6913  * @param[in] sh
6914  *   The pointer to the shared device context.
6915  * @param[in] async_id
6916  *   The Devx async ID.
6917  * @param[in] status
6918  *   The status of the completion.
6919  */
6920 void
6921 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6922 				  uint64_t async_id, int status)
6923 {
6924 	struct mlx5_flow_counter_pool *pool =
6925 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6926 	struct mlx5_counter_stats_raw *raw_to_free;
6927 	uint8_t query_gen = pool->query_gen ^ 1;
6928 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6929 	enum mlx5_counter_type cnt_type =
6930 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6931 				MLX5_COUNTER_TYPE_ORIGIN;
6932 
6933 	if (unlikely(status)) {
6934 		raw_to_free = pool->raw_hw;
6935 	} else {
6936 		raw_to_free = pool->raw;
6937 		if (pool->is_aged)
6938 			mlx5_flow_aging_check(sh, pool);
6939 		rte_spinlock_lock(&pool->sl);
6940 		pool->raw = pool->raw_hw;
6941 		rte_spinlock_unlock(&pool->sl);
6942 		/* Be sure the new raw counters data is updated in memory. */
6943 		rte_io_wmb();
6944 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6945 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6946 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6947 				     &pool->counters[query_gen], next);
6948 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6949 		}
6950 	}
6951 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6952 	pool->raw_hw = NULL;
6953 	sh->cmng.pending_queries--;
6954 }
6955 
6956 static const struct mlx5_flow_tbl_data_entry  *
6957 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
6958 {
6959 	struct mlx5_priv *priv = dev->data->dev_private;
6960 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6961 	struct mlx5_hlist_entry *he;
6962 	union tunnel_offload_mark mbits = { .val = mark };
6963 	union mlx5_flow_tbl_key table_key = {
6964 		{
6965 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
6966 			.dummy = 0,
6967 			.domain = !!mbits.transfer,
6968 			.direction = 0,
6969 		}
6970 	};
6971 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
6972 	return he ?
6973 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
6974 }
6975 
6976 static void
6977 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
6978 				   struct mlx5_hlist_entry *entry)
6979 {
6980 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6981 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
6982 
6983 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6984 			tunnel_flow_tbl_to_id(tte->flow_table));
6985 	mlx5_free(tte);
6986 }
6987 
6988 static struct mlx5_hlist_entry *
6989 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list,
6990 				   uint64_t key __rte_unused,
6991 				   void *ctx __rte_unused)
6992 {
6993 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6994 	struct tunnel_tbl_entry *tte;
6995 
6996 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
6997 			  sizeof(*tte), 0,
6998 			  SOCKET_ID_ANY);
6999 	if (!tte)
7000 		goto err;
7001 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7002 			  &tte->flow_table);
7003 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7004 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7005 			tte->flow_table);
7006 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7007 				tte->flow_table);
7008 		goto err;
7009 	} else if (!tte->flow_table) {
7010 		goto err;
7011 	}
7012 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7013 	return &tte->hash;
7014 err:
7015 	if (tte)
7016 		mlx5_free(tte);
7017 	return NULL;
7018 }
7019 
7020 static uint32_t
7021 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7022 				const struct mlx5_flow_tunnel *tunnel,
7023 				uint32_t group, uint32_t *table,
7024 				struct rte_flow_error *error)
7025 {
7026 	struct mlx5_hlist_entry *he;
7027 	struct tunnel_tbl_entry *tte;
7028 	union tunnel_tbl_key key = {
7029 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7030 		.group = group
7031 	};
7032 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7033 	struct mlx5_hlist *group_hash;
7034 
7035 	group_hash = tunnel ? tunnel->groups : thub->groups;
7036 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7037 	if (!he)
7038 		return rte_flow_error_set(error, EINVAL,
7039 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7040 					  NULL,
7041 					  "tunnel group index not supported");
7042 	tte = container_of(he, typeof(*tte), hash);
7043 	*table = tte->flow_table;
7044 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7045 		dev->data->port_id, key.tunnel_id, group, *table);
7046 	return 0;
7047 }
7048 
7049 static int
7050 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7051 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
7052 {
7053 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7054 		if (group == UINT32_MAX)
7055 			return rte_flow_error_set
7056 						(error, EINVAL,
7057 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7058 						 NULL,
7059 						 "group index not supported");
7060 		*table = group + 1;
7061 	} else {
7062 		*table = group;
7063 	}
7064 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7065 	return 0;
7066 }
7067 
7068 /**
7069  * Translate the rte_flow group index to HW table value.
7070  *
7071  * If tunnel offload is disabled, all group ids converted to flow table
7072  * id using the standard method.
7073  * If tunnel offload is enabled, group id can be converted using the
7074  * standard or tunnel conversion method. Group conversion method
7075  * selection depends on flags in `grp_info` parameter:
7076  * - Internal (grp_info.external == 0) groups conversion uses the
7077  *   standard method.
7078  * - Group ids in JUMP action converted with the tunnel conversion.
7079  * - Group id in rule attribute conversion depends on a rule type and
7080  *   group id value:
7081  *   ** non zero group attributes converted with the tunnel method
7082  *   ** zero group attribute in non-tunnel rule is converted using the
7083  *      standard method - there's only one root table
7084  *   ** zero group attribute in steer tunnel rule is converted with the
7085  *      standard method - single root table
7086  *   ** zero group attribute in match tunnel rule is a special OvS
7087  *      case: that value is used for portability reasons. That group
7088  *      id is converted with the tunnel conversion method.
7089  *
7090  * @param[in] dev
7091  *   Port device
7092  * @param[in] tunnel
7093  *   PMD tunnel offload object
7094  * @param[in] group
7095  *   rte_flow group index value.
7096  * @param[out] table
7097  *   HW table value.
7098  * @param[in] grp_info
7099  *   flags used for conversion
7100  * @param[out] error
7101  *   Pointer to error structure.
7102  *
7103  * @return
7104  *   0 on success, a negative errno value otherwise and rte_errno is set.
7105  */
7106 int
7107 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7108 			 const struct mlx5_flow_tunnel *tunnel,
7109 			 uint32_t group, uint32_t *table,
7110 			 struct flow_grp_info grp_info,
7111 			 struct rte_flow_error *error)
7112 {
7113 	int ret;
7114 	bool standard_translation;
7115 
7116 	if (!grp_info.skip_scale && grp_info.external &&
7117 	    group < MLX5_MAX_TABLES_EXTERNAL)
7118 		group *= MLX5_FLOW_TABLE_FACTOR;
7119 	if (is_tunnel_offload_active(dev)) {
7120 		standard_translation = !grp_info.external ||
7121 					grp_info.std_tbl_fix;
7122 	} else {
7123 		standard_translation = true;
7124 	}
7125 	DRV_LOG(DEBUG,
7126 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7127 		dev->data->port_id, group, grp_info.transfer,
7128 		grp_info.external, grp_info.fdb_def_rule,
7129 		standard_translation ? "STANDARD" : "TUNNEL");
7130 	if (standard_translation)
7131 		ret = flow_group_to_table(dev->data->port_id, group, table,
7132 					  grp_info, error);
7133 	else
7134 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7135 						      table, error);
7136 
7137 	return ret;
7138 }
7139 
7140 /**
7141  * Discover availability of metadata reg_c's.
7142  *
7143  * Iteratively use test flows to check availability.
7144  *
7145  * @param[in] dev
7146  *   Pointer to the Ethernet device structure.
7147  *
7148  * @return
7149  *   0 on success, a negative errno value otherwise and rte_errno is set.
7150  */
7151 int
7152 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7153 {
7154 	struct mlx5_priv *priv = dev->data->dev_private;
7155 	struct mlx5_dev_config *config = &priv->config;
7156 	enum modify_reg idx;
7157 	int n = 0;
7158 
7159 	/* reg_c[0] and reg_c[1] are reserved. */
7160 	config->flow_mreg_c[n++] = REG_C_0;
7161 	config->flow_mreg_c[n++] = REG_C_1;
7162 	/* Discover availability of other reg_c's. */
7163 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7164 		struct rte_flow_attr attr = {
7165 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7166 			.priority = MLX5_FLOW_PRIO_RSVD,
7167 			.ingress = 1,
7168 		};
7169 		struct rte_flow_item items[] = {
7170 			[0] = {
7171 				.type = RTE_FLOW_ITEM_TYPE_END,
7172 			},
7173 		};
7174 		struct rte_flow_action actions[] = {
7175 			[0] = {
7176 				.type = (enum rte_flow_action_type)
7177 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7178 				.conf = &(struct mlx5_flow_action_copy_mreg){
7179 					.src = REG_C_1,
7180 					.dst = idx,
7181 				},
7182 			},
7183 			[1] = {
7184 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7185 				.conf = &(struct rte_flow_action_jump){
7186 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7187 				},
7188 			},
7189 			[2] = {
7190 				.type = RTE_FLOW_ACTION_TYPE_END,
7191 			},
7192 		};
7193 		uint32_t flow_idx;
7194 		struct rte_flow *flow;
7195 		struct rte_flow_error error;
7196 
7197 		if (!config->dv_flow_en)
7198 			break;
7199 		/* Create internal flow, validation skips copy action. */
7200 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7201 					    actions, false, &error);
7202 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7203 				      flow_idx);
7204 		if (!flow)
7205 			continue;
7206 		config->flow_mreg_c[n++] = idx;
7207 		flow_list_destroy(dev, NULL, flow_idx);
7208 	}
7209 	for (; n < MLX5_MREG_C_NUM; ++n)
7210 		config->flow_mreg_c[n] = REG_NON;
7211 	return 0;
7212 }
7213 
7214 /**
7215  * Dump flow raw hw data to file
7216  *
7217  * @param[in] dev
7218  *    The pointer to Ethernet device.
7219  * @param[in] file
7220  *   A pointer to a file for output.
7221  * @param[out] error
7222  *   Perform verbose error reporting if not NULL. PMDs initialize this
7223  *   structure in case of error only.
7224  * @return
7225  *   0 on success, a nagative value otherwise.
7226  */
7227 int
7228 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7229 		   FILE *file,
7230 		   struct rte_flow_error *error __rte_unused)
7231 {
7232 	struct mlx5_priv *priv = dev->data->dev_private;
7233 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7234 
7235 	if (!priv->config.dv_flow_en) {
7236 		if (fputs("device dv flow disabled\n", file) <= 0)
7237 			return -errno;
7238 		return -ENOTSUP;
7239 	}
7240 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7241 				       sh->tx_domain, file);
7242 }
7243 
7244 /**
7245  * Get aged-out flows.
7246  *
7247  * @param[in] dev
7248  *   Pointer to the Ethernet device structure.
7249  * @param[in] context
7250  *   The address of an array of pointers to the aged-out flows contexts.
7251  * @param[in] nb_countexts
7252  *   The length of context array pointers.
7253  * @param[out] error
7254  *   Perform verbose error reporting if not NULL. Initialized in case of
7255  *   error only.
7256  *
7257  * @return
7258  *   how many contexts get in success, otherwise negative errno value.
7259  *   if nb_contexts is 0, return the amount of all aged contexts.
7260  *   if nb_contexts is not 0 , return the amount of aged flows reported
7261  *   in the context array.
7262  */
7263 int
7264 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7265 			uint32_t nb_contexts, struct rte_flow_error *error)
7266 {
7267 	const struct mlx5_flow_driver_ops *fops;
7268 	struct rte_flow_attr attr = { .transfer = 0 };
7269 
7270 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7271 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7272 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7273 						    error);
7274 	}
7275 	DRV_LOG(ERR,
7276 		"port %u get aged flows is not supported.",
7277 		 dev->data->port_id);
7278 	return -ENOTSUP;
7279 }
7280 
7281 /* Wrapper for driver action_validate op callback */
7282 static int
7283 flow_drv_action_validate(struct rte_eth_dev *dev,
7284 			 const struct rte_flow_shared_action_conf *conf,
7285 			 const struct rte_flow_action *action,
7286 			 const struct mlx5_flow_driver_ops *fops,
7287 			 struct rte_flow_error *error)
7288 {
7289 	static const char err_msg[] = "shared action validation unsupported";
7290 
7291 	if (!fops->action_validate) {
7292 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7293 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7294 				   NULL, err_msg);
7295 		return -rte_errno;
7296 	}
7297 	return fops->action_validate(dev, conf, action, error);
7298 }
7299 
7300 /**
7301  * Destroys the shared action by handle.
7302  *
7303  * @param dev
7304  *   Pointer to Ethernet device structure.
7305  * @param[in] action
7306  *   Handle for the shared action to be destroyed.
7307  * @param[out] error
7308  *   Perform verbose error reporting if not NULL. PMDs initialize this
7309  *   structure in case of error only.
7310  *
7311  * @return
7312  *   0 on success, a negative errno value otherwise and rte_errno is set.
7313  *
7314  * @note: wrapper for driver action_create op callback.
7315  */
7316 static int
7317 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7318 			   struct rte_flow_shared_action *action,
7319 			   struct rte_flow_error *error)
7320 {
7321 	static const char err_msg[] = "shared action destruction unsupported";
7322 	struct rte_flow_attr attr = { .transfer = 0 };
7323 	const struct mlx5_flow_driver_ops *fops =
7324 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7325 
7326 	if (!fops->action_destroy) {
7327 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7328 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7329 				   NULL, err_msg);
7330 		return -rte_errno;
7331 	}
7332 	return fops->action_destroy(dev, action, error);
7333 }
7334 
7335 /* Wrapper for driver action_destroy op callback */
7336 static int
7337 flow_drv_action_update(struct rte_eth_dev *dev,
7338 		       struct rte_flow_shared_action *action,
7339 		       const void *action_conf,
7340 		       const struct mlx5_flow_driver_ops *fops,
7341 		       struct rte_flow_error *error)
7342 {
7343 	static const char err_msg[] = "shared action update unsupported";
7344 
7345 	if (!fops->action_update) {
7346 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7347 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7348 				   NULL, err_msg);
7349 		return -rte_errno;
7350 	}
7351 	return fops->action_update(dev, action, action_conf, error);
7352 }
7353 
7354 /* Wrapper for driver action_destroy op callback */
7355 static int
7356 flow_drv_action_query(struct rte_eth_dev *dev,
7357 		      const struct rte_flow_shared_action *action,
7358 		      void *data,
7359 		      const struct mlx5_flow_driver_ops *fops,
7360 		      struct rte_flow_error *error)
7361 {
7362 	static const char err_msg[] = "shared action query unsupported";
7363 
7364 	if (!fops->action_query) {
7365 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7366 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7367 				   NULL, err_msg);
7368 		return -rte_errno;
7369 	}
7370 	return fops->action_query(dev, action, data, error);
7371 }
7372 
7373 /**
7374  * Create shared action for reuse in multiple flow rules.
7375  *
7376  * @param dev
7377  *   Pointer to Ethernet device structure.
7378  * @param[in] action
7379  *   Action configuration for shared action creation.
7380  * @param[out] error
7381  *   Perform verbose error reporting if not NULL. PMDs initialize this
7382  *   structure in case of error only.
7383  * @return
7384  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7385  */
7386 static struct rte_flow_shared_action *
7387 mlx5_shared_action_create(struct rte_eth_dev *dev,
7388 			  const struct rte_flow_shared_action_conf *conf,
7389 			  const struct rte_flow_action *action,
7390 			  struct rte_flow_error *error)
7391 {
7392 	static const char err_msg[] = "shared action creation unsupported";
7393 	struct rte_flow_attr attr = { .transfer = 0 };
7394 	const struct mlx5_flow_driver_ops *fops =
7395 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7396 
7397 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7398 		return NULL;
7399 	if (!fops->action_create) {
7400 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7401 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7402 				   NULL, err_msg);
7403 		return NULL;
7404 	}
7405 	return fops->action_create(dev, conf, action, error);
7406 }
7407 
7408 /**
7409  * Updates inplace the shared action configuration pointed by *action* handle
7410  * with the configuration provided as *action* argument.
7411  * The update of the shared action configuration effects all flow rules reusing
7412  * the action via handle.
7413  *
7414  * @param dev
7415  *   Pointer to Ethernet device structure.
7416  * @param[in] shared_action
7417  *   Handle for the shared action to be updated.
7418  * @param[in] action
7419  *   Action specification used to modify the action pointed by handle.
7420  *   *action* should be of same type with the action pointed by the *action*
7421  *   handle argument, otherwise considered as invalid.
7422  * @param[out] error
7423  *   Perform verbose error reporting if not NULL. PMDs initialize this
7424  *   structure in case of error only.
7425  *
7426  * @return
7427  *   0 on success, a negative errno value otherwise and rte_errno is set.
7428  */
7429 static int
7430 mlx5_shared_action_update(struct rte_eth_dev *dev,
7431 		struct rte_flow_shared_action *shared_action,
7432 		const struct rte_flow_action *action,
7433 		struct rte_flow_error *error)
7434 {
7435 	struct rte_flow_attr attr = { .transfer = 0 };
7436 	const struct mlx5_flow_driver_ops *fops =
7437 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7438 	int ret;
7439 
7440 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7441 	if (ret)
7442 		return ret;
7443 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7444 				      error);
7445 }
7446 
7447 /**
7448  * Query the shared action by handle.
7449  *
7450  * This function allows retrieving action-specific data such as counters.
7451  * Data is gathered by special action which may be present/referenced in
7452  * more than one flow rule definition.
7453  *
7454  * \see RTE_FLOW_ACTION_TYPE_COUNT
7455  *
7456  * @param dev
7457  *   Pointer to Ethernet device structure.
7458  * @param[in] action
7459  *   Handle for the shared action to query.
7460  * @param[in, out] data
7461  *   Pointer to storage for the associated query data type.
7462  * @param[out] error
7463  *   Perform verbose error reporting if not NULL. PMDs initialize this
7464  *   structure in case of error only.
7465  *
7466  * @return
7467  *   0 on success, a negative errno value otherwise and rte_errno is set.
7468  */
7469 static int
7470 mlx5_shared_action_query(struct rte_eth_dev *dev,
7471 			 const struct rte_flow_shared_action *action,
7472 			 void *data,
7473 			 struct rte_flow_error *error)
7474 {
7475 	struct rte_flow_attr attr = { .transfer = 0 };
7476 	const struct mlx5_flow_driver_ops *fops =
7477 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7478 
7479 	return flow_drv_action_query(dev, action, data, fops, error);
7480 }
7481 
7482 /**
7483  * Destroy all shared actions.
7484  *
7485  * @param dev
7486  *   Pointer to Ethernet device.
7487  *
7488  * @return
7489  *   0 on success, a negative errno value otherwise and rte_errno is set.
7490  */
7491 int
7492 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7493 {
7494 	struct rte_flow_error error;
7495 	struct mlx5_priv *priv = dev->data->dev_private;
7496 	struct mlx5_shared_action_rss *action;
7497 	int ret = 0;
7498 	uint32_t idx;
7499 
7500 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7501 		      priv->rss_shared_actions, idx, action, next) {
7502 		ret |= mlx5_shared_action_destroy(dev,
7503 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7504 	}
7505 	return ret;
7506 }
7507 
7508 static void
7509 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7510 		      struct mlx5_flow_tunnel *tunnel)
7511 {
7512 	struct mlx5_priv *priv = dev->data->dev_private;
7513 
7514 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7515 		dev->data->port_id, tunnel->tunnel_id);
7516 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
7517 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
7518 			tunnel->tunnel_id);
7519 	mlx5_hlist_destroy(tunnel->groups);
7520 	mlx5_free(tunnel);
7521 }
7522 
7523 static struct mlx5_flow_tunnel *
7524 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7525 {
7526 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7527 	struct mlx5_flow_tunnel *tun;
7528 
7529 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7530 		if (tun->tunnel_id == id)
7531 			break;
7532 	}
7533 
7534 	return tun;
7535 }
7536 
7537 static struct mlx5_flow_tunnel *
7538 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7539 			  const struct rte_flow_tunnel *app_tunnel)
7540 {
7541 	struct mlx5_priv *priv = dev->data->dev_private;
7542 	struct mlx5_flow_tunnel *tunnel;
7543 	uint32_t id;
7544 
7545 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
7546 			  &id);
7547 	if (id >= MLX5_MAX_TUNNELS) {
7548 		mlx5_ipool_free(priv->sh->ipool
7549 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7550 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7551 		return NULL;
7552 	} else if (!id) {
7553 		return NULL;
7554 	}
7555 	/**
7556 	 * mlx5 flow tunnel is an auxlilary data structure
7557 	 * It's not part of IO. No need to allocate it from
7558 	 * huge pages pools dedicated for IO
7559 	 */
7560 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
7561 			     0, SOCKET_ID_ANY);
7562 	if (!tunnel) {
7563 		mlx5_ipool_free(priv->sh->ipool
7564 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7565 		return NULL;
7566 	}
7567 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7568 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7569 					   NULL,
7570 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7571 	if (!tunnel->groups) {
7572 		mlx5_ipool_free(priv->sh->ipool
7573 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7574 		mlx5_free(tunnel);
7575 		return NULL;
7576 	}
7577 	tunnel->groups->ctx = priv->sh;
7578 	/* initiate new PMD tunnel */
7579 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7580 	tunnel->tunnel_id = id;
7581 	tunnel->action.type = (typeof(tunnel->action.type))
7582 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7583 	tunnel->action.conf = tunnel;
7584 	tunnel->item.type = (typeof(tunnel->item.type))
7585 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7586 	tunnel->item.spec = tunnel;
7587 	tunnel->item.last = NULL;
7588 	tunnel->item.mask = NULL;
7589 
7590 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7591 		dev->data->port_id, tunnel->tunnel_id);
7592 
7593 	return tunnel;
7594 }
7595 
7596 static int
7597 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7598 		     const struct rte_flow_tunnel *app_tunnel,
7599 		     struct mlx5_flow_tunnel **tunnel)
7600 {
7601 	int ret;
7602 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7603 	struct mlx5_flow_tunnel *tun;
7604 
7605 	rte_spinlock_lock(&thub->sl);
7606 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7607 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7608 			    sizeof(*app_tunnel))) {
7609 			*tunnel = tun;
7610 			ret = 0;
7611 			break;
7612 		}
7613 	}
7614 	if (!tun) {
7615 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
7616 		if (tun) {
7617 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
7618 			*tunnel = tun;
7619 		} else {
7620 			ret = -ENOMEM;
7621 		}
7622 	}
7623 	rte_spinlock_unlock(&thub->sl);
7624 	if (tun)
7625 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
7626 
7627 	return ret;
7628 }
7629 
7630 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7631 {
7632 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7633 
7634 	if (!thub)
7635 		return;
7636 	if (!LIST_EMPTY(&thub->tunnels))
7637 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7638 	mlx5_hlist_destroy(thub->groups);
7639 	mlx5_free(thub);
7640 }
7641 
7642 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7643 {
7644 	int err;
7645 	struct mlx5_flow_tunnel_hub *thub;
7646 
7647 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7648 			   0, SOCKET_ID_ANY);
7649 	if (!thub)
7650 		return -ENOMEM;
7651 	LIST_INIT(&thub->tunnels);
7652 	rte_spinlock_init(&thub->sl);
7653 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7654 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7655 					 NULL,
7656 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7657 	if (!thub->groups) {
7658 		err = -rte_errno;
7659 		goto err;
7660 	}
7661 	thub->groups->ctx = sh;
7662 	sh->tunnel_hub = thub;
7663 
7664 	return 0;
7665 
7666 err:
7667 	if (thub->groups)
7668 		mlx5_hlist_destroy(thub->groups);
7669 	if (thub)
7670 		mlx5_free(thub);
7671 	return err;
7672 }
7673 
7674 #ifndef HAVE_MLX5DV_DR
7675 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7676 #else
7677 #define MLX5_DOMAIN_SYNC_FLOW \
7678 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7679 #endif
7680 
7681 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7682 {
7683 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7684 	const struct mlx5_flow_driver_ops *fops;
7685 	int ret;
7686 	struct rte_flow_attr attr = { .transfer = 0 };
7687 
7688 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7689 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7690 	if (ret > 0)
7691 		ret = -ret;
7692 	return ret;
7693 }
7694