xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision bc8e32473cc3978d763a1387eaa8244bcf75e77d)
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_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_action_release(struct rte_eth_dev *dev,
691 			 struct rte_flow_action *pmd_actions,
692 			 uint32_t num_actions, struct rte_flow_error *err)
693 {
694 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
695 	struct mlx5_flow_tunnel *tun;
696 
697 	rte_spinlock_lock(&thub->sl);
698 	LIST_FOREACH(tun, &thub->tunnels, chain) {
699 		if (&tun->action == pmd_actions) {
700 			LIST_REMOVE(tun, chain);
701 			break;
702 		}
703 	}
704 	rte_spinlock_unlock(&thub->sl);
705 	if (!tun || num_actions != 1)
706 		return rte_flow_error_set(err, EINVAL,
707 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
708 					  "invalid argument");
709 	if (!__atomic_sub_fetch(&tun->refctn, 1, __ATOMIC_RELAXED))
710 		mlx5_flow_tunnel_free(dev, tun);
711 
712 	return 0;
713 }
714 
715 static int
716 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
717 				  struct rte_mbuf *m,
718 				  struct rte_flow_restore_info *info,
719 				  struct rte_flow_error *err)
720 {
721 	uint64_t ol_flags = m->ol_flags;
722 	const struct mlx5_flow_tbl_data_entry *tble;
723 	const uint64_t mask = PKT_RX_FDIR | PKT_RX_FDIR_ID;
724 
725 	if ((ol_flags & mask) != mask)
726 		goto err;
727 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
728 	if (!tble) {
729 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
730 			dev->data->port_id, m->hash.fdir.hi);
731 		goto err;
732 	}
733 	MLX5_ASSERT(tble->tunnel);
734 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
735 	info->group_id = tble->group_id;
736 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
737 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
738 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
739 
740 	return 0;
741 
742 err:
743 	return rte_flow_error_set(err, EINVAL,
744 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
745 				  "failed to get restore info");
746 }
747 
748 static const struct rte_flow_ops mlx5_flow_ops = {
749 	.validate = mlx5_flow_validate,
750 	.create = mlx5_flow_create,
751 	.destroy = mlx5_flow_destroy,
752 	.flush = mlx5_flow_flush,
753 	.isolate = mlx5_flow_isolate,
754 	.query = mlx5_flow_query,
755 	.dev_dump = mlx5_flow_dev_dump,
756 	.get_aged_flows = mlx5_flow_get_aged_flows,
757 	.shared_action_create = mlx5_shared_action_create,
758 	.shared_action_destroy = mlx5_shared_action_destroy,
759 	.shared_action_update = mlx5_shared_action_update,
760 	.shared_action_query = mlx5_shared_action_query,
761 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
762 	.tunnel_match = mlx5_flow_tunnel_match,
763 	.tunnel_action_decap_release = mlx5_flow_action_release,
764 	.tunnel_item_release = mlx5_flow_item_release,
765 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
766 };
767 
768 /* Tunnel information. */
769 struct mlx5_flow_tunnel_info {
770 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
771 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
772 };
773 
774 static struct mlx5_flow_tunnel_info tunnels_info[] = {
775 	{
776 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
777 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
778 	},
779 	{
780 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
781 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
782 	},
783 	{
784 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
785 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
786 	},
787 	{
788 		.tunnel = MLX5_FLOW_LAYER_GRE,
789 		.ptype = RTE_PTYPE_TUNNEL_GRE,
790 	},
791 	{
792 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
793 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
794 	},
795 	{
796 		.tunnel = MLX5_FLOW_LAYER_MPLS,
797 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
798 	},
799 	{
800 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
801 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
802 	},
803 	{
804 		.tunnel = MLX5_FLOW_LAYER_IPIP,
805 		.ptype = RTE_PTYPE_TUNNEL_IP,
806 	},
807 	{
808 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
809 		.ptype = RTE_PTYPE_TUNNEL_IP,
810 	},
811 	{
812 		.tunnel = MLX5_FLOW_LAYER_GTP,
813 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
814 	},
815 };
816 
817 /* Key of thread specific flow workspace data. */
818 static pthread_key_t key_workspace;
819 
820 /* Thread specific flow workspace data once initialization data. */
821 static pthread_once_t key_workspace_init;
822 
823 
824 /**
825  * Translate tag ID to register.
826  *
827  * @param[in] dev
828  *   Pointer to the Ethernet device structure.
829  * @param[in] feature
830  *   The feature that request the register.
831  * @param[in] id
832  *   The request register ID.
833  * @param[out] error
834  *   Error description in case of any.
835  *
836  * @return
837  *   The request register on success, a negative errno
838  *   value otherwise and rte_errno is set.
839  */
840 int
841 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
842 		     enum mlx5_feature_name feature,
843 		     uint32_t id,
844 		     struct rte_flow_error *error)
845 {
846 	struct mlx5_priv *priv = dev->data->dev_private;
847 	struct mlx5_dev_config *config = &priv->config;
848 	enum modify_reg start_reg;
849 	bool skip_mtr_reg = false;
850 
851 	switch (feature) {
852 	case MLX5_HAIRPIN_RX:
853 		return REG_B;
854 	case MLX5_HAIRPIN_TX:
855 		return REG_A;
856 	case MLX5_METADATA_RX:
857 		switch (config->dv_xmeta_en) {
858 		case MLX5_XMETA_MODE_LEGACY:
859 			return REG_B;
860 		case MLX5_XMETA_MODE_META16:
861 			return REG_C_0;
862 		case MLX5_XMETA_MODE_META32:
863 			return REG_C_1;
864 		}
865 		break;
866 	case MLX5_METADATA_TX:
867 		return REG_A;
868 	case MLX5_METADATA_FDB:
869 		switch (config->dv_xmeta_en) {
870 		case MLX5_XMETA_MODE_LEGACY:
871 			return REG_NON;
872 		case MLX5_XMETA_MODE_META16:
873 			return REG_C_0;
874 		case MLX5_XMETA_MODE_META32:
875 			return REG_C_1;
876 		}
877 		break;
878 	case MLX5_FLOW_MARK:
879 		switch (config->dv_xmeta_en) {
880 		case MLX5_XMETA_MODE_LEGACY:
881 			return REG_NON;
882 		case MLX5_XMETA_MODE_META16:
883 			return REG_C_1;
884 		case MLX5_XMETA_MODE_META32:
885 			return REG_C_0;
886 		}
887 		break;
888 	case MLX5_MTR_SFX:
889 		/*
890 		 * If meter color and flow match share one register, flow match
891 		 * should use the meter color register for match.
892 		 */
893 		if (priv->mtr_reg_share)
894 			return priv->mtr_color_reg;
895 		else
896 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
897 			       REG_C_3;
898 	case MLX5_MTR_COLOR:
899 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
900 		return priv->mtr_color_reg;
901 	case MLX5_COPY_MARK:
902 		/*
903 		 * Metadata COPY_MARK register using is in meter suffix sub
904 		 * flow while with meter. It's safe to share the same register.
905 		 */
906 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
907 	case MLX5_APP_TAG:
908 		/*
909 		 * If meter is enable, it will engage the register for color
910 		 * match and flow match. If meter color match is not using the
911 		 * REG_C_2, need to skip the REG_C_x be used by meter color
912 		 * match.
913 		 * If meter is disable, free to use all available registers.
914 		 */
915 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
916 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
917 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
918 		if (id > (REG_C_7 - start_reg))
919 			return rte_flow_error_set(error, EINVAL,
920 						  RTE_FLOW_ERROR_TYPE_ITEM,
921 						  NULL, "invalid tag id");
922 		if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
923 			return rte_flow_error_set(error, ENOTSUP,
924 						  RTE_FLOW_ERROR_TYPE_ITEM,
925 						  NULL, "unsupported tag id");
926 		/*
927 		 * This case means meter is using the REG_C_x great than 2.
928 		 * Take care not to conflict with meter color REG_C_x.
929 		 * If the available index REG_C_y >= REG_C_x, skip the
930 		 * color register.
931 		 */
932 		if (skip_mtr_reg && config->flow_mreg_c
933 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
934 			if (id >= (REG_C_7 - start_reg))
935 				return rte_flow_error_set(error, EINVAL,
936 						       RTE_FLOW_ERROR_TYPE_ITEM,
937 							NULL, "invalid tag id");
938 			if (config->flow_mreg_c
939 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
940 				return config->flow_mreg_c
941 					       [id + 1 + start_reg - REG_C_0];
942 			return rte_flow_error_set(error, ENOTSUP,
943 						  RTE_FLOW_ERROR_TYPE_ITEM,
944 						  NULL, "unsupported tag id");
945 		}
946 		return config->flow_mreg_c[id + start_reg - REG_C_0];
947 	}
948 	MLX5_ASSERT(false);
949 	return rte_flow_error_set(error, EINVAL,
950 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
951 				  NULL, "invalid feature name");
952 }
953 
954 /**
955  * Check extensive flow metadata register support.
956  *
957  * @param dev
958  *   Pointer to rte_eth_dev structure.
959  *
960  * @return
961  *   True if device supports extensive flow metadata register, otherwise false.
962  */
963 bool
964 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
965 {
966 	struct mlx5_priv *priv = dev->data->dev_private;
967 	struct mlx5_dev_config *config = &priv->config;
968 
969 	/*
970 	 * Having available reg_c can be regarded inclusively as supporting
971 	 * extensive flow metadata register, which could mean,
972 	 * - metadata register copy action by modify header.
973 	 * - 16 modify header actions is supported.
974 	 * - reg_c's are preserved across different domain (FDB and NIC) on
975 	 *   packet loopback by flow lookup miss.
976 	 */
977 	return config->flow_mreg_c[2] != REG_NON;
978 }
979 
980 /**
981  * Verify the @p item specifications (spec, last, mask) are compatible with the
982  * NIC capabilities.
983  *
984  * @param[in] item
985  *   Item specification.
986  * @param[in] mask
987  *   @p item->mask or flow default bit-masks.
988  * @param[in] nic_mask
989  *   Bit-masks covering supported fields by the NIC to compare with user mask.
990  * @param[in] size
991  *   Bit-masks size in bytes.
992  * @param[in] range_accepted
993  *   True if range of values is accepted for specific fields, false otherwise.
994  * @param[out] error
995  *   Pointer to error structure.
996  *
997  * @return
998  *   0 on success, a negative errno value otherwise and rte_errno is set.
999  */
1000 int
1001 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1002 			  const uint8_t *mask,
1003 			  const uint8_t *nic_mask,
1004 			  unsigned int size,
1005 			  bool range_accepted,
1006 			  struct rte_flow_error *error)
1007 {
1008 	unsigned int i;
1009 
1010 	MLX5_ASSERT(nic_mask);
1011 	for (i = 0; i < size; ++i)
1012 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1013 			return rte_flow_error_set(error, ENOTSUP,
1014 						  RTE_FLOW_ERROR_TYPE_ITEM,
1015 						  item,
1016 						  "mask enables non supported"
1017 						  " bits");
1018 	if (!item->spec && (item->mask || item->last))
1019 		return rte_flow_error_set(error, EINVAL,
1020 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1021 					  "mask/last without a spec is not"
1022 					  " supported");
1023 	if (item->spec && item->last && !range_accepted) {
1024 		uint8_t spec[size];
1025 		uint8_t last[size];
1026 		unsigned int i;
1027 		int ret;
1028 
1029 		for (i = 0; i < size; ++i) {
1030 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1031 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1032 		}
1033 		ret = memcmp(spec, last, size);
1034 		if (ret != 0)
1035 			return rte_flow_error_set(error, EINVAL,
1036 						  RTE_FLOW_ERROR_TYPE_ITEM,
1037 						  item,
1038 						  "range is not valid");
1039 	}
1040 	return 0;
1041 }
1042 
1043 /**
1044  * Adjust the hash fields according to the @p flow information.
1045  *
1046  * @param[in] dev_flow.
1047  *   Pointer to the mlx5_flow.
1048  * @param[in] tunnel
1049  *   1 when the hash field is for a tunnel item.
1050  * @param[in] layer_types
1051  *   ETH_RSS_* types.
1052  * @param[in] hash_fields
1053  *   Item hash fields.
1054  *
1055  * @return
1056  *   The hash fields that should be used.
1057  */
1058 uint64_t
1059 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1060 			    int tunnel __rte_unused, uint64_t layer_types,
1061 			    uint64_t hash_fields)
1062 {
1063 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1064 	int rss_request_inner = rss_desc->level >= 2;
1065 
1066 	/* Check RSS hash level for tunnel. */
1067 	if (tunnel && rss_request_inner)
1068 		hash_fields |= IBV_RX_HASH_INNER;
1069 	else if (tunnel || rss_request_inner)
1070 		return 0;
1071 #endif
1072 	/* Check if requested layer matches RSS hash fields. */
1073 	if (!(rss_desc->types & layer_types))
1074 		return 0;
1075 	return hash_fields;
1076 }
1077 
1078 /**
1079  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1080  * if several tunnel rules are used on this queue, the tunnel ptype will be
1081  * cleared.
1082  *
1083  * @param rxq_ctrl
1084  *   Rx queue to update.
1085  */
1086 static void
1087 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1088 {
1089 	unsigned int i;
1090 	uint32_t tunnel_ptype = 0;
1091 
1092 	/* Look up for the ptype to use. */
1093 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1094 		if (!rxq_ctrl->flow_tunnels_n[i])
1095 			continue;
1096 		if (!tunnel_ptype) {
1097 			tunnel_ptype = tunnels_info[i].ptype;
1098 		} else {
1099 			tunnel_ptype = 0;
1100 			break;
1101 		}
1102 	}
1103 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1104 }
1105 
1106 /**
1107  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
1108  * flow.
1109  *
1110  * @param[in] dev
1111  *   Pointer to the Ethernet device structure.
1112  * @param[in] dev_handle
1113  *   Pointer to device flow handle structure.
1114  */
1115 static void
1116 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1117 		       struct mlx5_flow_handle *dev_handle)
1118 {
1119 	struct mlx5_priv *priv = dev->data->dev_private;
1120 	const int mark = dev_handle->mark;
1121 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1122 	struct mlx5_hrxq *hrxq;
1123 	unsigned int i;
1124 
1125 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1126 		return;
1127 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1128 			      dev_handle->rix_hrxq);
1129 	if (!hrxq)
1130 		return;
1131 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1132 		int idx = hrxq->ind_table->queues[i];
1133 		struct mlx5_rxq_ctrl *rxq_ctrl =
1134 			container_of((*priv->rxqs)[idx],
1135 				     struct mlx5_rxq_ctrl, rxq);
1136 
1137 		/*
1138 		 * To support metadata register copy on Tx loopback,
1139 		 * this must be always enabled (metadata may arive
1140 		 * from other port - not from local flows only.
1141 		 */
1142 		if (priv->config.dv_flow_en &&
1143 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1144 		    mlx5_flow_ext_mreg_supported(dev)) {
1145 			rxq_ctrl->rxq.mark = 1;
1146 			rxq_ctrl->flow_mark_n = 1;
1147 		} else if (mark) {
1148 			rxq_ctrl->rxq.mark = 1;
1149 			rxq_ctrl->flow_mark_n++;
1150 		}
1151 		if (tunnel) {
1152 			unsigned int j;
1153 
1154 			/* Increase the counter matching the flow. */
1155 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1156 				if ((tunnels_info[j].tunnel &
1157 				     dev_handle->layers) ==
1158 				    tunnels_info[j].tunnel) {
1159 					rxq_ctrl->flow_tunnels_n[j]++;
1160 					break;
1161 				}
1162 			}
1163 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1164 		}
1165 	}
1166 }
1167 
1168 /**
1169  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1170  *
1171  * @param[in] dev
1172  *   Pointer to the Ethernet device structure.
1173  * @param[in] flow
1174  *   Pointer to flow structure.
1175  */
1176 static void
1177 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1178 {
1179 	struct mlx5_priv *priv = dev->data->dev_private;
1180 	uint32_t handle_idx;
1181 	struct mlx5_flow_handle *dev_handle;
1182 
1183 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1184 		       handle_idx, dev_handle, next)
1185 		flow_drv_rxq_flags_set(dev, dev_handle);
1186 }
1187 
1188 /**
1189  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1190  * device flow if no other flow uses it with the same kind of request.
1191  *
1192  * @param dev
1193  *   Pointer to Ethernet device.
1194  * @param[in] dev_handle
1195  *   Pointer to the device flow handle structure.
1196  */
1197 static void
1198 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1199 			struct mlx5_flow_handle *dev_handle)
1200 {
1201 	struct mlx5_priv *priv = dev->data->dev_private;
1202 	const int mark = dev_handle->mark;
1203 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1204 	struct mlx5_hrxq *hrxq;
1205 	unsigned int i;
1206 
1207 	if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
1208 		return;
1209 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1210 			      dev_handle->rix_hrxq);
1211 	if (!hrxq)
1212 		return;
1213 	MLX5_ASSERT(dev->data->dev_started);
1214 	for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
1215 		int idx = hrxq->ind_table->queues[i];
1216 		struct mlx5_rxq_ctrl *rxq_ctrl =
1217 			container_of((*priv->rxqs)[idx],
1218 				     struct mlx5_rxq_ctrl, rxq);
1219 
1220 		if (priv->config.dv_flow_en &&
1221 		    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1222 		    mlx5_flow_ext_mreg_supported(dev)) {
1223 			rxq_ctrl->rxq.mark = 1;
1224 			rxq_ctrl->flow_mark_n = 1;
1225 		} else if (mark) {
1226 			rxq_ctrl->flow_mark_n--;
1227 			rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
1228 		}
1229 		if (tunnel) {
1230 			unsigned int j;
1231 
1232 			/* Decrease the counter matching the flow. */
1233 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1234 				if ((tunnels_info[j].tunnel &
1235 				     dev_handle->layers) ==
1236 				    tunnels_info[j].tunnel) {
1237 					rxq_ctrl->flow_tunnels_n[j]--;
1238 					break;
1239 				}
1240 			}
1241 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1242 		}
1243 	}
1244 }
1245 
1246 /**
1247  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1248  * @p flow if no other flow uses it with the same kind of request.
1249  *
1250  * @param dev
1251  *   Pointer to Ethernet device.
1252  * @param[in] flow
1253  *   Pointer to the flow.
1254  */
1255 static void
1256 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1257 {
1258 	struct mlx5_priv *priv = dev->data->dev_private;
1259 	uint32_t handle_idx;
1260 	struct mlx5_flow_handle *dev_handle;
1261 
1262 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1263 		       handle_idx, dev_handle, next)
1264 		flow_drv_rxq_flags_trim(dev, dev_handle);
1265 }
1266 
1267 /**
1268  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1269  *
1270  * @param dev
1271  *   Pointer to Ethernet device.
1272  */
1273 static void
1274 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1275 {
1276 	struct mlx5_priv *priv = dev->data->dev_private;
1277 	unsigned int i;
1278 
1279 	for (i = 0; i != priv->rxqs_n; ++i) {
1280 		struct mlx5_rxq_ctrl *rxq_ctrl;
1281 		unsigned int j;
1282 
1283 		if (!(*priv->rxqs)[i])
1284 			continue;
1285 		rxq_ctrl = container_of((*priv->rxqs)[i],
1286 					struct mlx5_rxq_ctrl, rxq);
1287 		rxq_ctrl->flow_mark_n = 0;
1288 		rxq_ctrl->rxq.mark = 0;
1289 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1290 			rxq_ctrl->flow_tunnels_n[j] = 0;
1291 		rxq_ctrl->rxq.tunnel = 0;
1292 	}
1293 }
1294 
1295 /**
1296  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1297  *
1298  * @param[in] dev
1299  *   Pointer to the Ethernet device structure.
1300  */
1301 void
1302 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1303 {
1304 	struct mlx5_priv *priv = dev->data->dev_private;
1305 	struct mlx5_rxq_data *data;
1306 	unsigned int i;
1307 
1308 	for (i = 0; i != priv->rxqs_n; ++i) {
1309 		if (!(*priv->rxqs)[i])
1310 			continue;
1311 		data = (*priv->rxqs)[i];
1312 		if (!rte_flow_dynf_metadata_avail()) {
1313 			data->dynf_meta = 0;
1314 			data->flow_meta_mask = 0;
1315 			data->flow_meta_offset = -1;
1316 		} else {
1317 			data->dynf_meta = 1;
1318 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1319 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1320 		}
1321 	}
1322 }
1323 
1324 /*
1325  * return a pointer to the desired action in the list of actions.
1326  *
1327  * @param[in] actions
1328  *   The list of actions to search the action in.
1329  * @param[in] action
1330  *   The action to find.
1331  *
1332  * @return
1333  *   Pointer to the action in the list, if found. NULL otherwise.
1334  */
1335 const struct rte_flow_action *
1336 mlx5_flow_find_action(const struct rte_flow_action *actions,
1337 		      enum rte_flow_action_type action)
1338 {
1339 	if (actions == NULL)
1340 		return NULL;
1341 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1342 		if (actions->type == action)
1343 			return actions;
1344 	return NULL;
1345 }
1346 
1347 /*
1348  * Validate the flag action.
1349  *
1350  * @param[in] action_flags
1351  *   Bit-fields that holds the actions detected until now.
1352  * @param[in] attr
1353  *   Attributes of flow that includes this action.
1354  * @param[out] error
1355  *   Pointer to error structure.
1356  *
1357  * @return
1358  *   0 on success, a negative errno value otherwise and rte_errno is set.
1359  */
1360 int
1361 mlx5_flow_validate_action_flag(uint64_t action_flags,
1362 			       const struct rte_flow_attr *attr,
1363 			       struct rte_flow_error *error)
1364 {
1365 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1366 		return rte_flow_error_set(error, EINVAL,
1367 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1368 					  "can't mark and flag in same flow");
1369 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1370 		return rte_flow_error_set(error, EINVAL,
1371 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1372 					  "can't have 2 flag"
1373 					  " actions in same flow");
1374 	if (attr->egress)
1375 		return rte_flow_error_set(error, ENOTSUP,
1376 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1377 					  "flag action not supported for "
1378 					  "egress");
1379 	return 0;
1380 }
1381 
1382 /*
1383  * Validate the mark action.
1384  *
1385  * @param[in] action
1386  *   Pointer to the queue action.
1387  * @param[in] action_flags
1388  *   Bit-fields that holds the actions detected until now.
1389  * @param[in] attr
1390  *   Attributes of flow that includes this action.
1391  * @param[out] error
1392  *   Pointer to error structure.
1393  *
1394  * @return
1395  *   0 on success, a negative errno value otherwise and rte_errno is set.
1396  */
1397 int
1398 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1399 			       uint64_t action_flags,
1400 			       const struct rte_flow_attr *attr,
1401 			       struct rte_flow_error *error)
1402 {
1403 	const struct rte_flow_action_mark *mark = action->conf;
1404 
1405 	if (!mark)
1406 		return rte_flow_error_set(error, EINVAL,
1407 					  RTE_FLOW_ERROR_TYPE_ACTION,
1408 					  action,
1409 					  "configuration cannot be null");
1410 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1411 		return rte_flow_error_set(error, EINVAL,
1412 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1413 					  &mark->id,
1414 					  "mark id must in 0 <= id < "
1415 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1416 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1417 		return rte_flow_error_set(error, EINVAL,
1418 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1419 					  "can't flag and mark in same flow");
1420 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1421 		return rte_flow_error_set(error, EINVAL,
1422 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1423 					  "can't have 2 mark actions in same"
1424 					  " flow");
1425 	if (attr->egress)
1426 		return rte_flow_error_set(error, ENOTSUP,
1427 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1428 					  "mark action not supported for "
1429 					  "egress");
1430 	return 0;
1431 }
1432 
1433 /*
1434  * Validate the drop action.
1435  *
1436  * @param[in] action_flags
1437  *   Bit-fields that holds the actions detected until now.
1438  * @param[in] attr
1439  *   Attributes of flow that includes this action.
1440  * @param[out] error
1441  *   Pointer to error structure.
1442  *
1443  * @return
1444  *   0 on success, a negative errno value otherwise and rte_errno is set.
1445  */
1446 int
1447 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1448 			       const struct rte_flow_attr *attr,
1449 			       struct rte_flow_error *error)
1450 {
1451 	if (attr->egress)
1452 		return rte_flow_error_set(error, ENOTSUP,
1453 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1454 					  "drop action not supported for "
1455 					  "egress");
1456 	return 0;
1457 }
1458 
1459 /*
1460  * Validate the queue action.
1461  *
1462  * @param[in] action
1463  *   Pointer to the queue action.
1464  * @param[in] action_flags
1465  *   Bit-fields that holds the actions detected until now.
1466  * @param[in] dev
1467  *   Pointer to the Ethernet device structure.
1468  * @param[in] attr
1469  *   Attributes of flow that includes this action.
1470  * @param[out] error
1471  *   Pointer to error structure.
1472  *
1473  * @return
1474  *   0 on success, a negative errno value otherwise and rte_errno is set.
1475  */
1476 int
1477 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1478 				uint64_t action_flags,
1479 				struct rte_eth_dev *dev,
1480 				const struct rte_flow_attr *attr,
1481 				struct rte_flow_error *error)
1482 {
1483 	struct mlx5_priv *priv = dev->data->dev_private;
1484 	const struct rte_flow_action_queue *queue = action->conf;
1485 
1486 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1487 		return rte_flow_error_set(error, EINVAL,
1488 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1489 					  "can't have 2 fate actions in"
1490 					  " same flow");
1491 	if (!priv->rxqs_n)
1492 		return rte_flow_error_set(error, EINVAL,
1493 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1494 					  NULL, "No Rx queues configured");
1495 	if (queue->index >= priv->rxqs_n)
1496 		return rte_flow_error_set(error, EINVAL,
1497 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1498 					  &queue->index,
1499 					  "queue index out of range");
1500 	if (!(*priv->rxqs)[queue->index])
1501 		return rte_flow_error_set(error, EINVAL,
1502 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1503 					  &queue->index,
1504 					  "queue is not configured");
1505 	if (attr->egress)
1506 		return rte_flow_error_set(error, ENOTSUP,
1507 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1508 					  "queue action not supported for "
1509 					  "egress");
1510 	return 0;
1511 }
1512 
1513 /*
1514  * Validate the rss action.
1515  *
1516  * @param[in] dev
1517  *   Pointer to the Ethernet device structure.
1518  * @param[in] action
1519  *   Pointer to the queue action.
1520  * @param[out] error
1521  *   Pointer to error structure.
1522  *
1523  * @return
1524  *   0 on success, a negative errno value otherwise and rte_errno is set.
1525  */
1526 int
1527 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1528 			 const struct rte_flow_action *action,
1529 			 struct rte_flow_error *error)
1530 {
1531 	struct mlx5_priv *priv = dev->data->dev_private;
1532 	const struct rte_flow_action_rss *rss = action->conf;
1533 	unsigned int i;
1534 
1535 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1536 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1537 		return rte_flow_error_set(error, ENOTSUP,
1538 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1539 					  &rss->func,
1540 					  "RSS hash function not supported");
1541 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1542 	if (rss->level > 2)
1543 #else
1544 	if (rss->level > 1)
1545 #endif
1546 		return rte_flow_error_set(error, ENOTSUP,
1547 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1548 					  &rss->level,
1549 					  "tunnel RSS is not supported");
1550 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1551 	if (rss->key_len == 0 && rss->key != NULL)
1552 		return rte_flow_error_set(error, ENOTSUP,
1553 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1554 					  &rss->key_len,
1555 					  "RSS hash key length 0");
1556 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1557 		return rte_flow_error_set(error, ENOTSUP,
1558 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1559 					  &rss->key_len,
1560 					  "RSS hash key too small");
1561 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1562 		return rte_flow_error_set(error, ENOTSUP,
1563 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1564 					  &rss->key_len,
1565 					  "RSS hash key too large");
1566 	if (rss->queue_num > priv->config.ind_table_max_size)
1567 		return rte_flow_error_set(error, ENOTSUP,
1568 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1569 					  &rss->queue_num,
1570 					  "number of queues too large");
1571 	if (rss->types & MLX5_RSS_HF_MASK)
1572 		return rte_flow_error_set(error, ENOTSUP,
1573 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1574 					  &rss->types,
1575 					  "some RSS protocols are not"
1576 					  " supported");
1577 	if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1578 	    !(rss->types & ETH_RSS_IP))
1579 		return rte_flow_error_set(error, EINVAL,
1580 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1581 					  "L3 partial RSS requested but L3 RSS"
1582 					  " type not specified");
1583 	if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1584 	    !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1585 		return rte_flow_error_set(error, EINVAL,
1586 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1587 					  "L4 partial RSS requested but L4 RSS"
1588 					  " type not specified");
1589 	if (!priv->rxqs_n)
1590 		return rte_flow_error_set(error, EINVAL,
1591 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1592 					  NULL, "No Rx queues configured");
1593 	if (!rss->queue_num)
1594 		return rte_flow_error_set(error, EINVAL,
1595 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1596 					  NULL, "No queues configured");
1597 	for (i = 0; i != rss->queue_num; ++i) {
1598 		if (rss->queue[i] >= priv->rxqs_n)
1599 			return rte_flow_error_set
1600 				(error, EINVAL,
1601 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1602 				 &rss->queue[i], "queue index out of range");
1603 		if (!(*priv->rxqs)[rss->queue[i]])
1604 			return rte_flow_error_set
1605 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1606 				 &rss->queue[i], "queue is not configured");
1607 	}
1608 	return 0;
1609 }
1610 
1611 /*
1612  * Validate the rss action.
1613  *
1614  * @param[in] action
1615  *   Pointer to the queue action.
1616  * @param[in] action_flags
1617  *   Bit-fields that holds the actions detected until now.
1618  * @param[in] dev
1619  *   Pointer to the Ethernet device structure.
1620  * @param[in] attr
1621  *   Attributes of flow that includes this action.
1622  * @param[in] item_flags
1623  *   Items that were detected.
1624  * @param[out] error
1625  *   Pointer to error structure.
1626  *
1627  * @return
1628  *   0 on success, a negative errno value otherwise and rte_errno is set.
1629  */
1630 int
1631 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1632 			      uint64_t action_flags,
1633 			      struct rte_eth_dev *dev,
1634 			      const struct rte_flow_attr *attr,
1635 			      uint64_t item_flags,
1636 			      struct rte_flow_error *error)
1637 {
1638 	const struct rte_flow_action_rss *rss = action->conf;
1639 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1640 	int ret;
1641 
1642 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1643 		return rte_flow_error_set(error, EINVAL,
1644 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1645 					  "can't have 2 fate actions"
1646 					  " in same flow");
1647 	ret = mlx5_validate_action_rss(dev, action, error);
1648 	if (ret)
1649 		return ret;
1650 	if (attr->egress)
1651 		return rte_flow_error_set(error, ENOTSUP,
1652 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1653 					  "rss action not supported for "
1654 					  "egress");
1655 	if (rss->level > 1 && !tunnel)
1656 		return rte_flow_error_set(error, EINVAL,
1657 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1658 					  "inner RSS is not supported for "
1659 					  "non-tunnel flows");
1660 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1661 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1662 		return rte_flow_error_set(error, EINVAL,
1663 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1664 					  "RSS on eCPRI is not supported now");
1665 	}
1666 	return 0;
1667 }
1668 
1669 /*
1670  * Validate the default miss action.
1671  *
1672  * @param[in] action_flags
1673  *   Bit-fields that holds the actions detected until now.
1674  * @param[out] error
1675  *   Pointer to error structure.
1676  *
1677  * @return
1678  *   0 on success, a negative errno value otherwise and rte_errno is set.
1679  */
1680 int
1681 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1682 				const struct rte_flow_attr *attr,
1683 				struct rte_flow_error *error)
1684 {
1685 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1688 					  "can't have 2 fate actions in"
1689 					  " same flow");
1690 	if (attr->egress)
1691 		return rte_flow_error_set(error, ENOTSUP,
1692 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1693 					  "default miss action not supported "
1694 					  "for egress");
1695 	if (attr->group)
1696 		return rte_flow_error_set(error, ENOTSUP,
1697 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1698 					  "only group 0 is supported");
1699 	if (attr->transfer)
1700 		return rte_flow_error_set(error, ENOTSUP,
1701 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1702 					  NULL, "transfer is not supported");
1703 	return 0;
1704 }
1705 
1706 /*
1707  * Validate the count action.
1708  *
1709  * @param[in] dev
1710  *   Pointer to the Ethernet device structure.
1711  * @param[in] attr
1712  *   Attributes of flow that includes this action.
1713  * @param[out] error
1714  *   Pointer to error structure.
1715  *
1716  * @return
1717  *   0 on success, a negative errno value otherwise and rte_errno is set.
1718  */
1719 int
1720 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1721 				const struct rte_flow_attr *attr,
1722 				struct rte_flow_error *error)
1723 {
1724 	if (attr->egress)
1725 		return rte_flow_error_set(error, ENOTSUP,
1726 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1727 					  "count action not supported for "
1728 					  "egress");
1729 	return 0;
1730 }
1731 
1732 /**
1733  * Verify the @p attributes will be correctly understood by the NIC and store
1734  * them in the @p flow if everything is correct.
1735  *
1736  * @param[in] dev
1737  *   Pointer to the Ethernet device structure.
1738  * @param[in] attributes
1739  *   Pointer to flow attributes
1740  * @param[out] error
1741  *   Pointer to error structure.
1742  *
1743  * @return
1744  *   0 on success, a negative errno value otherwise and rte_errno is set.
1745  */
1746 int
1747 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1748 			      const struct rte_flow_attr *attributes,
1749 			      struct rte_flow_error *error)
1750 {
1751 	struct mlx5_priv *priv = dev->data->dev_private;
1752 	uint32_t priority_max = priv->config.flow_prio - 1;
1753 
1754 	if (attributes->group)
1755 		return rte_flow_error_set(error, ENOTSUP,
1756 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1757 					  NULL, "groups is not supported");
1758 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1759 	    attributes->priority >= priority_max)
1760 		return rte_flow_error_set(error, ENOTSUP,
1761 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1762 					  NULL, "priority out of range");
1763 	if (attributes->egress)
1764 		return rte_flow_error_set(error, ENOTSUP,
1765 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1766 					  "egress is not supported");
1767 	if (attributes->transfer && !priv->config.dv_esw_en)
1768 		return rte_flow_error_set(error, ENOTSUP,
1769 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1770 					  NULL, "transfer is not supported");
1771 	if (!attributes->ingress)
1772 		return rte_flow_error_set(error, EINVAL,
1773 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1774 					  NULL,
1775 					  "ingress attribute is mandatory");
1776 	return 0;
1777 }
1778 
1779 /**
1780  * Validate ICMP6 item.
1781  *
1782  * @param[in] item
1783  *   Item specification.
1784  * @param[in] item_flags
1785  *   Bit-fields that holds the items detected until now.
1786  * @param[in] ext_vlan_sup
1787  *   Whether extended VLAN features are supported or not.
1788  * @param[out] error
1789  *   Pointer to error structure.
1790  *
1791  * @return
1792  *   0 on success, a negative errno value otherwise and rte_errno is set.
1793  */
1794 int
1795 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1796 			       uint64_t item_flags,
1797 			       uint8_t target_protocol,
1798 			       struct rte_flow_error *error)
1799 {
1800 	const struct rte_flow_item_icmp6 *mask = item->mask;
1801 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1802 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1803 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1804 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1805 				      MLX5_FLOW_LAYER_OUTER_L4;
1806 	int ret;
1807 
1808 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1809 		return rte_flow_error_set(error, EINVAL,
1810 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1811 					  "protocol filtering not compatible"
1812 					  " with ICMP6 layer");
1813 	if (!(item_flags & l3m))
1814 		return rte_flow_error_set(error, EINVAL,
1815 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1816 					  "IPv6 is mandatory to filter on"
1817 					  " ICMP6");
1818 	if (item_flags & l4m)
1819 		return rte_flow_error_set(error, EINVAL,
1820 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1821 					  "multiple L4 layers not supported");
1822 	if (!mask)
1823 		mask = &rte_flow_item_icmp6_mask;
1824 	ret = mlx5_flow_item_acceptable
1825 		(item, (const uint8_t *)mask,
1826 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
1827 		 sizeof(struct rte_flow_item_icmp6),
1828 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1829 	if (ret < 0)
1830 		return ret;
1831 	return 0;
1832 }
1833 
1834 /**
1835  * Validate ICMP item.
1836  *
1837  * @param[in] item
1838  *   Item specification.
1839  * @param[in] item_flags
1840  *   Bit-fields that holds the items detected until now.
1841  * @param[out] error
1842  *   Pointer to error structure.
1843  *
1844  * @return
1845  *   0 on success, a negative errno value otherwise and rte_errno is set.
1846  */
1847 int
1848 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1849 			     uint64_t item_flags,
1850 			     uint8_t target_protocol,
1851 			     struct rte_flow_error *error)
1852 {
1853 	const struct rte_flow_item_icmp *mask = item->mask;
1854 	const struct rte_flow_item_icmp nic_mask = {
1855 		.hdr.icmp_type = 0xff,
1856 		.hdr.icmp_code = 0xff,
1857 		.hdr.icmp_ident = RTE_BE16(0xffff),
1858 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
1859 	};
1860 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1861 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1862 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1863 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1864 				      MLX5_FLOW_LAYER_OUTER_L4;
1865 	int ret;
1866 
1867 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1868 		return rte_flow_error_set(error, EINVAL,
1869 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1870 					  "protocol filtering not compatible"
1871 					  " with ICMP layer");
1872 	if (!(item_flags & l3m))
1873 		return rte_flow_error_set(error, EINVAL,
1874 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1875 					  "IPv4 is mandatory to filter"
1876 					  " on ICMP");
1877 	if (item_flags & l4m)
1878 		return rte_flow_error_set(error, EINVAL,
1879 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1880 					  "multiple L4 layers not supported");
1881 	if (!mask)
1882 		mask = &nic_mask;
1883 	ret = mlx5_flow_item_acceptable
1884 		(item, (const uint8_t *)mask,
1885 		 (const uint8_t *)&nic_mask,
1886 		 sizeof(struct rte_flow_item_icmp),
1887 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1888 	if (ret < 0)
1889 		return ret;
1890 	return 0;
1891 }
1892 
1893 /**
1894  * Validate Ethernet item.
1895  *
1896  * @param[in] item
1897  *   Item specification.
1898  * @param[in] item_flags
1899  *   Bit-fields that holds the items detected until now.
1900  * @param[out] error
1901  *   Pointer to error structure.
1902  *
1903  * @return
1904  *   0 on success, a negative errno value otherwise and rte_errno is set.
1905  */
1906 int
1907 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1908 			    uint64_t item_flags, bool ext_vlan_sup,
1909 			    struct rte_flow_error *error)
1910 {
1911 	const struct rte_flow_item_eth *mask = item->mask;
1912 	const struct rte_flow_item_eth nic_mask = {
1913 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1914 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1915 		.type = RTE_BE16(0xffff),
1916 		.has_vlan = ext_vlan_sup ? 1 : 0,
1917 	};
1918 	int ret;
1919 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1920 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
1921 				       MLX5_FLOW_LAYER_OUTER_L2;
1922 
1923 	if (item_flags & ethm)
1924 		return rte_flow_error_set(error, ENOTSUP,
1925 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1926 					  "multiple L2 layers not supported");
1927 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1928 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1929 		return rte_flow_error_set(error, EINVAL,
1930 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1931 					  "L2 layer should not follow "
1932 					  "L3 layers");
1933 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1934 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1935 		return rte_flow_error_set(error, EINVAL,
1936 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1937 					  "L2 layer should not follow VLAN");
1938 	if (!mask)
1939 		mask = &rte_flow_item_eth_mask;
1940 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1941 					(const uint8_t *)&nic_mask,
1942 					sizeof(struct rte_flow_item_eth),
1943 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1944 	return ret;
1945 }
1946 
1947 /**
1948  * Validate VLAN item.
1949  *
1950  * @param[in] item
1951  *   Item specification.
1952  * @param[in] item_flags
1953  *   Bit-fields that holds the items detected until now.
1954  * @param[in] dev
1955  *   Ethernet device flow is being created on.
1956  * @param[out] error
1957  *   Pointer to error structure.
1958  *
1959  * @return
1960  *   0 on success, a negative errno value otherwise and rte_errno is set.
1961  */
1962 int
1963 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1964 			     uint64_t item_flags,
1965 			     struct rte_eth_dev *dev,
1966 			     struct rte_flow_error *error)
1967 {
1968 	const struct rte_flow_item_vlan *spec = item->spec;
1969 	const struct rte_flow_item_vlan *mask = item->mask;
1970 	const struct rte_flow_item_vlan nic_mask = {
1971 		.tci = RTE_BE16(UINT16_MAX),
1972 		.inner_type = RTE_BE16(UINT16_MAX),
1973 	};
1974 	uint16_t vlan_tag = 0;
1975 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1976 	int ret;
1977 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1978 					MLX5_FLOW_LAYER_INNER_L4) :
1979 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1980 					MLX5_FLOW_LAYER_OUTER_L4);
1981 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1982 					MLX5_FLOW_LAYER_OUTER_VLAN;
1983 
1984 	if (item_flags & vlanm)
1985 		return rte_flow_error_set(error, EINVAL,
1986 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1987 					  "multiple VLAN layers not supported");
1988 	else if ((item_flags & l34m) != 0)
1989 		return rte_flow_error_set(error, EINVAL,
1990 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1991 					  "VLAN cannot follow L3/L4 layer");
1992 	if (!mask)
1993 		mask = &rte_flow_item_vlan_mask;
1994 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1995 					(const uint8_t *)&nic_mask,
1996 					sizeof(struct rte_flow_item_vlan),
1997 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
1998 	if (ret)
1999 		return ret;
2000 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2001 		struct mlx5_priv *priv = dev->data->dev_private;
2002 
2003 		if (priv->vmwa_context) {
2004 			/*
2005 			 * Non-NULL context means we have a virtual machine
2006 			 * and SR-IOV enabled, we have to create VLAN interface
2007 			 * to make hypervisor to setup E-Switch vport
2008 			 * context correctly. We avoid creating the multiple
2009 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2010 			 */
2011 			return rte_flow_error_set(error, EINVAL,
2012 						  RTE_FLOW_ERROR_TYPE_ITEM,
2013 						  item,
2014 						  "VLAN tag mask is not"
2015 						  " supported in virtual"
2016 						  " environment");
2017 		}
2018 	}
2019 	if (spec) {
2020 		vlan_tag = spec->tci;
2021 		vlan_tag &= mask->tci;
2022 	}
2023 	/*
2024 	 * From verbs perspective an empty VLAN is equivalent
2025 	 * to a packet without VLAN layer.
2026 	 */
2027 	if (!vlan_tag)
2028 		return rte_flow_error_set(error, EINVAL,
2029 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2030 					  item->spec,
2031 					  "VLAN cannot be empty");
2032 	return 0;
2033 }
2034 
2035 /**
2036  * Validate IPV4 item.
2037  *
2038  * @param[in] item
2039  *   Item specification.
2040  * @param[in] item_flags
2041  *   Bit-fields that holds the items detected until now.
2042  * @param[in] last_item
2043  *   Previous validated item in the pattern items.
2044  * @param[in] ether_type
2045  *   Type in the ethernet layer header (including dot1q).
2046  * @param[in] acc_mask
2047  *   Acceptable mask, if NULL default internal default mask
2048  *   will be used to check whether item fields are supported.
2049  * @param[in] range_accepted
2050  *   True if range of values is accepted for specific fields, false otherwise.
2051  * @param[out] error
2052  *   Pointer to error structure.
2053  *
2054  * @return
2055  *   0 on success, a negative errno value otherwise and rte_errno is set.
2056  */
2057 int
2058 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2059 			     uint64_t item_flags,
2060 			     uint64_t last_item,
2061 			     uint16_t ether_type,
2062 			     const struct rte_flow_item_ipv4 *acc_mask,
2063 			     bool range_accepted,
2064 			     struct rte_flow_error *error)
2065 {
2066 	const struct rte_flow_item_ipv4 *mask = item->mask;
2067 	const struct rte_flow_item_ipv4 *spec = item->spec;
2068 	const struct rte_flow_item_ipv4 nic_mask = {
2069 		.hdr = {
2070 			.src_addr = RTE_BE32(0xffffffff),
2071 			.dst_addr = RTE_BE32(0xffffffff),
2072 			.type_of_service = 0xff,
2073 			.next_proto_id = 0xff,
2074 		},
2075 	};
2076 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2077 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2078 				      MLX5_FLOW_LAYER_OUTER_L3;
2079 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2080 				      MLX5_FLOW_LAYER_OUTER_L4;
2081 	int ret;
2082 	uint8_t next_proto = 0xFF;
2083 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2084 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2085 				  MLX5_FLOW_LAYER_INNER_VLAN);
2086 
2087 	if ((last_item & l2_vlan) && ether_type &&
2088 	    ether_type != RTE_ETHER_TYPE_IPV4)
2089 		return rte_flow_error_set(error, EINVAL,
2090 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2091 					  "IPv4 cannot follow L2/VLAN layer "
2092 					  "which ether type is not IPv4");
2093 	if (item_flags & MLX5_FLOW_LAYER_IPIP) {
2094 		if (mask && spec)
2095 			next_proto = mask->hdr.next_proto_id &
2096 				     spec->hdr.next_proto_id;
2097 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2098 			return rte_flow_error_set(error, EINVAL,
2099 						  RTE_FLOW_ERROR_TYPE_ITEM,
2100 						  item,
2101 						  "multiple tunnel "
2102 						  "not supported");
2103 	}
2104 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2105 		return rte_flow_error_set(error, EINVAL,
2106 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2107 					  "wrong tunnel type - IPv6 specified "
2108 					  "but IPv4 item provided");
2109 	if (item_flags & l3m)
2110 		return rte_flow_error_set(error, ENOTSUP,
2111 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2112 					  "multiple L3 layers not supported");
2113 	else if (item_flags & l4m)
2114 		return rte_flow_error_set(error, EINVAL,
2115 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2116 					  "L3 cannot follow an L4 layer.");
2117 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2118 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2119 		return rte_flow_error_set(error, EINVAL,
2120 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2121 					  "L3 cannot follow an NVGRE layer.");
2122 	if (!mask)
2123 		mask = &rte_flow_item_ipv4_mask;
2124 	else if (mask->hdr.next_proto_id != 0 &&
2125 		 mask->hdr.next_proto_id != 0xff)
2126 		return rte_flow_error_set(error, EINVAL,
2127 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2128 					  "partial mask is not supported"
2129 					  " for protocol");
2130 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2131 					acc_mask ? (const uint8_t *)acc_mask
2132 						 : (const uint8_t *)&nic_mask,
2133 					sizeof(struct rte_flow_item_ipv4),
2134 					range_accepted, error);
2135 	if (ret < 0)
2136 		return ret;
2137 	return 0;
2138 }
2139 
2140 /**
2141  * Validate IPV6 item.
2142  *
2143  * @param[in] item
2144  *   Item specification.
2145  * @param[in] item_flags
2146  *   Bit-fields that holds the items detected until now.
2147  * @param[in] last_item
2148  *   Previous validated item in the pattern items.
2149  * @param[in] ether_type
2150  *   Type in the ethernet layer header (including dot1q).
2151  * @param[in] acc_mask
2152  *   Acceptable mask, if NULL default internal default mask
2153  *   will be used to check whether item fields are supported.
2154  * @param[out] error
2155  *   Pointer to error structure.
2156  *
2157  * @return
2158  *   0 on success, a negative errno value otherwise and rte_errno is set.
2159  */
2160 int
2161 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2162 			     uint64_t item_flags,
2163 			     uint64_t last_item,
2164 			     uint16_t ether_type,
2165 			     const struct rte_flow_item_ipv6 *acc_mask,
2166 			     struct rte_flow_error *error)
2167 {
2168 	const struct rte_flow_item_ipv6 *mask = item->mask;
2169 	const struct rte_flow_item_ipv6 *spec = item->spec;
2170 	const struct rte_flow_item_ipv6 nic_mask = {
2171 		.hdr = {
2172 			.src_addr =
2173 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2174 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2175 			.dst_addr =
2176 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2177 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2178 			.vtc_flow = RTE_BE32(0xffffffff),
2179 			.proto = 0xff,
2180 		},
2181 	};
2182 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2183 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2184 				      MLX5_FLOW_LAYER_OUTER_L3;
2185 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2186 				      MLX5_FLOW_LAYER_OUTER_L4;
2187 	int ret;
2188 	uint8_t next_proto = 0xFF;
2189 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2190 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2191 				  MLX5_FLOW_LAYER_INNER_VLAN);
2192 
2193 	if ((last_item & l2_vlan) && ether_type &&
2194 	    ether_type != RTE_ETHER_TYPE_IPV6)
2195 		return rte_flow_error_set(error, EINVAL,
2196 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2197 					  "IPv6 cannot follow L2/VLAN layer "
2198 					  "which ether type is not IPv6");
2199 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2200 		next_proto = spec->hdr.proto;
2201 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
2202 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2203 			return rte_flow_error_set(error, EINVAL,
2204 						  RTE_FLOW_ERROR_TYPE_ITEM,
2205 						  item,
2206 						  "multiple tunnel "
2207 						  "not supported");
2208 	}
2209 	if (next_proto == IPPROTO_HOPOPTS  ||
2210 	    next_proto == IPPROTO_ROUTING  ||
2211 	    next_proto == IPPROTO_FRAGMENT ||
2212 	    next_proto == IPPROTO_ESP	   ||
2213 	    next_proto == IPPROTO_AH	   ||
2214 	    next_proto == IPPROTO_DSTOPTS)
2215 		return rte_flow_error_set(error, EINVAL,
2216 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2217 					  "IPv6 proto (next header) should "
2218 					  "not be set as extension header");
2219 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2220 		return rte_flow_error_set(error, EINVAL,
2221 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 					  "wrong tunnel type - IPv4 specified "
2223 					  "but IPv6 item provided");
2224 	if (item_flags & l3m)
2225 		return rte_flow_error_set(error, ENOTSUP,
2226 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2227 					  "multiple L3 layers not supported");
2228 	else if (item_flags & l4m)
2229 		return rte_flow_error_set(error, EINVAL,
2230 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2231 					  "L3 cannot follow an L4 layer.");
2232 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2233 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2234 		return rte_flow_error_set(error, EINVAL,
2235 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2236 					  "L3 cannot follow an NVGRE layer.");
2237 	if (!mask)
2238 		mask = &rte_flow_item_ipv6_mask;
2239 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2240 					acc_mask ? (const uint8_t *)acc_mask
2241 						 : (const uint8_t *)&nic_mask,
2242 					sizeof(struct rte_flow_item_ipv6),
2243 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2244 	if (ret < 0)
2245 		return ret;
2246 	return 0;
2247 }
2248 
2249 /**
2250  * Validate UDP item.
2251  *
2252  * @param[in] item
2253  *   Item specification.
2254  * @param[in] item_flags
2255  *   Bit-fields that holds the items detected until now.
2256  * @param[in] target_protocol
2257  *   The next protocol in the previous item.
2258  * @param[in] flow_mask
2259  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2260  * @param[out] error
2261  *   Pointer to error structure.
2262  *
2263  * @return
2264  *   0 on success, a negative errno value otherwise and rte_errno is set.
2265  */
2266 int
2267 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2268 			    uint64_t item_flags,
2269 			    uint8_t target_protocol,
2270 			    struct rte_flow_error *error)
2271 {
2272 	const struct rte_flow_item_udp *mask = item->mask;
2273 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2274 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2275 				      MLX5_FLOW_LAYER_OUTER_L3;
2276 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2277 				      MLX5_FLOW_LAYER_OUTER_L4;
2278 	int ret;
2279 
2280 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2281 		return rte_flow_error_set(error, EINVAL,
2282 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2283 					  "protocol filtering not compatible"
2284 					  " with UDP layer");
2285 	if (!(item_flags & l3m))
2286 		return rte_flow_error_set(error, EINVAL,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "L3 is mandatory to filter on L4");
2289 	if (item_flags & l4m)
2290 		return rte_flow_error_set(error, EINVAL,
2291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2292 					  "multiple L4 layers not supported");
2293 	if (!mask)
2294 		mask = &rte_flow_item_udp_mask;
2295 	ret = mlx5_flow_item_acceptable
2296 		(item, (const uint8_t *)mask,
2297 		 (const uint8_t *)&rte_flow_item_udp_mask,
2298 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2299 		 error);
2300 	if (ret < 0)
2301 		return ret;
2302 	return 0;
2303 }
2304 
2305 /**
2306  * Validate TCP item.
2307  *
2308  * @param[in] item
2309  *   Item specification.
2310  * @param[in] item_flags
2311  *   Bit-fields that holds the items detected until now.
2312  * @param[in] target_protocol
2313  *   The next protocol in the previous item.
2314  * @param[out] error
2315  *   Pointer to error structure.
2316  *
2317  * @return
2318  *   0 on success, a negative errno value otherwise and rte_errno is set.
2319  */
2320 int
2321 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2322 			    uint64_t item_flags,
2323 			    uint8_t target_protocol,
2324 			    const struct rte_flow_item_tcp *flow_mask,
2325 			    struct rte_flow_error *error)
2326 {
2327 	const struct rte_flow_item_tcp *mask = item->mask;
2328 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2329 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2330 				      MLX5_FLOW_LAYER_OUTER_L3;
2331 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2332 				      MLX5_FLOW_LAYER_OUTER_L4;
2333 	int ret;
2334 
2335 	MLX5_ASSERT(flow_mask);
2336 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2337 		return rte_flow_error_set(error, EINVAL,
2338 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2339 					  "protocol filtering not compatible"
2340 					  " with TCP layer");
2341 	if (!(item_flags & l3m))
2342 		return rte_flow_error_set(error, EINVAL,
2343 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2344 					  "L3 is mandatory to filter on L4");
2345 	if (item_flags & l4m)
2346 		return rte_flow_error_set(error, EINVAL,
2347 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2348 					  "multiple L4 layers not supported");
2349 	if (!mask)
2350 		mask = &rte_flow_item_tcp_mask;
2351 	ret = mlx5_flow_item_acceptable
2352 		(item, (const uint8_t *)mask,
2353 		 (const uint8_t *)flow_mask,
2354 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2355 		 error);
2356 	if (ret < 0)
2357 		return ret;
2358 	return 0;
2359 }
2360 
2361 /**
2362  * Validate VXLAN item.
2363  *
2364  * @param[in] item
2365  *   Item specification.
2366  * @param[in] item_flags
2367  *   Bit-fields that holds the items detected until now.
2368  * @param[in] target_protocol
2369  *   The next protocol in the previous item.
2370  * @param[out] error
2371  *   Pointer to error structure.
2372  *
2373  * @return
2374  *   0 on success, a negative errno value otherwise and rte_errno is set.
2375  */
2376 int
2377 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
2378 			      uint64_t item_flags,
2379 			      struct rte_flow_error *error)
2380 {
2381 	const struct rte_flow_item_vxlan *spec = item->spec;
2382 	const struct rte_flow_item_vxlan *mask = item->mask;
2383 	int ret;
2384 	union vni {
2385 		uint32_t vlan_id;
2386 		uint8_t vni[4];
2387 	} id = { .vlan_id = 0, };
2388 
2389 
2390 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2391 		return rte_flow_error_set(error, ENOTSUP,
2392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2393 					  "multiple tunnel layers not"
2394 					  " supported");
2395 	/*
2396 	 * Verify only UDPv4 is present as defined in
2397 	 * https://tools.ietf.org/html/rfc7348
2398 	 */
2399 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2400 		return rte_flow_error_set(error, EINVAL,
2401 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2402 					  "no outer UDP layer found");
2403 	if (!mask)
2404 		mask = &rte_flow_item_vxlan_mask;
2405 	ret = mlx5_flow_item_acceptable
2406 		(item, (const uint8_t *)mask,
2407 		 (const uint8_t *)&rte_flow_item_vxlan_mask,
2408 		 sizeof(struct rte_flow_item_vxlan),
2409 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2410 	if (ret < 0)
2411 		return ret;
2412 	if (spec) {
2413 		memcpy(&id.vni[1], spec->vni, 3);
2414 		memcpy(&id.vni[1], mask->vni, 3);
2415 	}
2416 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2417 		return rte_flow_error_set(error, ENOTSUP,
2418 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2419 					  "VXLAN tunnel must be fully defined");
2420 	return 0;
2421 }
2422 
2423 /**
2424  * Validate VXLAN_GPE item.
2425  *
2426  * @param[in] item
2427  *   Item specification.
2428  * @param[in] item_flags
2429  *   Bit-fields that holds the items detected until now.
2430  * @param[in] priv
2431  *   Pointer to the private data structure.
2432  * @param[in] target_protocol
2433  *   The next protocol in the previous item.
2434  * @param[out] error
2435  *   Pointer to error structure.
2436  *
2437  * @return
2438  *   0 on success, a negative errno value otherwise and rte_errno is set.
2439  */
2440 int
2441 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2442 				  uint64_t item_flags,
2443 				  struct rte_eth_dev *dev,
2444 				  struct rte_flow_error *error)
2445 {
2446 	struct mlx5_priv *priv = dev->data->dev_private;
2447 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2448 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2449 	int ret;
2450 	union vni {
2451 		uint32_t vlan_id;
2452 		uint8_t vni[4];
2453 	} id = { .vlan_id = 0, };
2454 
2455 	if (!priv->config.l3_vxlan_en)
2456 		return rte_flow_error_set(error, ENOTSUP,
2457 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2458 					  "L3 VXLAN is not enabled by device"
2459 					  " parameter and/or not configured in"
2460 					  " firmware");
2461 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2462 		return rte_flow_error_set(error, ENOTSUP,
2463 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2464 					  "multiple tunnel layers not"
2465 					  " supported");
2466 	/*
2467 	 * Verify only UDPv4 is present as defined in
2468 	 * https://tools.ietf.org/html/rfc7348
2469 	 */
2470 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2471 		return rte_flow_error_set(error, EINVAL,
2472 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2473 					  "no outer UDP layer found");
2474 	if (!mask)
2475 		mask = &rte_flow_item_vxlan_gpe_mask;
2476 	ret = mlx5_flow_item_acceptable
2477 		(item, (const uint8_t *)mask,
2478 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2479 		 sizeof(struct rte_flow_item_vxlan_gpe),
2480 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2481 	if (ret < 0)
2482 		return ret;
2483 	if (spec) {
2484 		if (spec->protocol)
2485 			return rte_flow_error_set(error, ENOTSUP,
2486 						  RTE_FLOW_ERROR_TYPE_ITEM,
2487 						  item,
2488 						  "VxLAN-GPE protocol"
2489 						  " not supported");
2490 		memcpy(&id.vni[1], spec->vni, 3);
2491 		memcpy(&id.vni[1], mask->vni, 3);
2492 	}
2493 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2494 		return rte_flow_error_set(error, ENOTSUP,
2495 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2496 					  "VXLAN-GPE tunnel must be fully"
2497 					  " defined");
2498 	return 0;
2499 }
2500 /**
2501  * Validate GRE Key item.
2502  *
2503  * @param[in] item
2504  *   Item specification.
2505  * @param[in] item_flags
2506  *   Bit flags to mark detected items.
2507  * @param[in] gre_item
2508  *   Pointer to gre_item
2509  * @param[out] error
2510  *   Pointer to error structure.
2511  *
2512  * @return
2513  *   0 on success, a negative errno value otherwise and rte_errno is set.
2514  */
2515 int
2516 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2517 				uint64_t item_flags,
2518 				const struct rte_flow_item *gre_item,
2519 				struct rte_flow_error *error)
2520 {
2521 	const rte_be32_t *mask = item->mask;
2522 	int ret = 0;
2523 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2524 	const struct rte_flow_item_gre *gre_spec;
2525 	const struct rte_flow_item_gre *gre_mask;
2526 
2527 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2528 		return rte_flow_error_set(error, ENOTSUP,
2529 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2530 					  "Multiple GRE key not support");
2531 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2532 		return rte_flow_error_set(error, ENOTSUP,
2533 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2534 					  "No preceding GRE header");
2535 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2536 		return rte_flow_error_set(error, ENOTSUP,
2537 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2538 					  "GRE key following a wrong item");
2539 	gre_mask = gre_item->mask;
2540 	if (!gre_mask)
2541 		gre_mask = &rte_flow_item_gre_mask;
2542 	gre_spec = gre_item->spec;
2543 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2544 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2545 		return rte_flow_error_set(error, EINVAL,
2546 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2547 					  "Key bit must be on");
2548 
2549 	if (!mask)
2550 		mask = &gre_key_default_mask;
2551 	ret = mlx5_flow_item_acceptable
2552 		(item, (const uint8_t *)mask,
2553 		 (const uint8_t *)&gre_key_default_mask,
2554 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2555 	return ret;
2556 }
2557 
2558 /**
2559  * Validate GRE item.
2560  *
2561  * @param[in] item
2562  *   Item specification.
2563  * @param[in] item_flags
2564  *   Bit flags to mark detected items.
2565  * @param[in] target_protocol
2566  *   The next protocol in the previous item.
2567  * @param[out] error
2568  *   Pointer to error structure.
2569  *
2570  * @return
2571  *   0 on success, a negative errno value otherwise and rte_errno is set.
2572  */
2573 int
2574 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2575 			    uint64_t item_flags,
2576 			    uint8_t target_protocol,
2577 			    struct rte_flow_error *error)
2578 {
2579 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2580 	const struct rte_flow_item_gre *mask = item->mask;
2581 	int ret;
2582 	const struct rte_flow_item_gre nic_mask = {
2583 		.c_rsvd0_ver = RTE_BE16(0xB000),
2584 		.protocol = RTE_BE16(UINT16_MAX),
2585 	};
2586 
2587 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2588 		return rte_flow_error_set(error, EINVAL,
2589 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2590 					  "protocol filtering not compatible"
2591 					  " with this GRE layer");
2592 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2593 		return rte_flow_error_set(error, ENOTSUP,
2594 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2595 					  "multiple tunnel layers not"
2596 					  " supported");
2597 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2598 		return rte_flow_error_set(error, ENOTSUP,
2599 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2600 					  "L3 Layer is missing");
2601 	if (!mask)
2602 		mask = &rte_flow_item_gre_mask;
2603 	ret = mlx5_flow_item_acceptable
2604 		(item, (const uint8_t *)mask,
2605 		 (const uint8_t *)&nic_mask,
2606 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2607 		 error);
2608 	if (ret < 0)
2609 		return ret;
2610 #ifndef HAVE_MLX5DV_DR
2611 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2612 	if (spec && (spec->protocol & mask->protocol))
2613 		return rte_flow_error_set(error, ENOTSUP,
2614 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2615 					  "without MPLS support the"
2616 					  " specification cannot be used for"
2617 					  " filtering");
2618 #endif
2619 #endif
2620 	return 0;
2621 }
2622 
2623 /**
2624  * Validate Geneve item.
2625  *
2626  * @param[in] item
2627  *   Item specification.
2628  * @param[in] itemFlags
2629  *   Bit-fields that holds the items detected until now.
2630  * @param[in] enPriv
2631  *   Pointer to the private data structure.
2632  * @param[out] error
2633  *   Pointer to error structure.
2634  *
2635  * @return
2636  *   0 on success, a negative errno value otherwise and rte_errno is set.
2637  */
2638 
2639 int
2640 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2641 			       uint64_t item_flags,
2642 			       struct rte_eth_dev *dev,
2643 			       struct rte_flow_error *error)
2644 {
2645 	struct mlx5_priv *priv = dev->data->dev_private;
2646 	const struct rte_flow_item_geneve *spec = item->spec;
2647 	const struct rte_flow_item_geneve *mask = item->mask;
2648 	int ret;
2649 	uint16_t gbhdr;
2650 	uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2651 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2652 	const struct rte_flow_item_geneve nic_mask = {
2653 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2654 		.vni = "\xff\xff\xff",
2655 		.protocol = RTE_BE16(UINT16_MAX),
2656 	};
2657 
2658 	if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2659 		return rte_flow_error_set(error, ENOTSUP,
2660 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2661 					  "L3 Geneve is not enabled by device"
2662 					  " parameter and/or not configured in"
2663 					  " firmware");
2664 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2665 		return rte_flow_error_set(error, ENOTSUP,
2666 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2667 					  "multiple tunnel layers not"
2668 					  " supported");
2669 	/*
2670 	 * Verify only UDPv4 is present as defined in
2671 	 * https://tools.ietf.org/html/rfc7348
2672 	 */
2673 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2674 		return rte_flow_error_set(error, EINVAL,
2675 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2676 					  "no outer UDP layer found");
2677 	if (!mask)
2678 		mask = &rte_flow_item_geneve_mask;
2679 	ret = mlx5_flow_item_acceptable
2680 				  (item, (const uint8_t *)mask,
2681 				   (const uint8_t *)&nic_mask,
2682 				   sizeof(struct rte_flow_item_geneve),
2683 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2684 	if (ret)
2685 		return ret;
2686 	if (spec) {
2687 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2688 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2689 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2690 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2691 			return rte_flow_error_set(error, ENOTSUP,
2692 						  RTE_FLOW_ERROR_TYPE_ITEM,
2693 						  item,
2694 						  "Geneve protocol unsupported"
2695 						  " fields are being used");
2696 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2697 			return rte_flow_error_set
2698 					(error, ENOTSUP,
2699 					 RTE_FLOW_ERROR_TYPE_ITEM,
2700 					 item,
2701 					 "Unsupported Geneve options length");
2702 	}
2703 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2704 		return rte_flow_error_set
2705 				    (error, ENOTSUP,
2706 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2707 				     "Geneve tunnel must be fully defined");
2708 	return 0;
2709 }
2710 
2711 /**
2712  * Validate MPLS item.
2713  *
2714  * @param[in] dev
2715  *   Pointer to the rte_eth_dev structure.
2716  * @param[in] item
2717  *   Item specification.
2718  * @param[in] item_flags
2719  *   Bit-fields that holds the items detected until now.
2720  * @param[in] prev_layer
2721  *   The protocol layer indicated in previous item.
2722  * @param[out] error
2723  *   Pointer to error structure.
2724  *
2725  * @return
2726  *   0 on success, a negative errno value otherwise and rte_errno is set.
2727  */
2728 int
2729 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2730 			     const struct rte_flow_item *item __rte_unused,
2731 			     uint64_t item_flags __rte_unused,
2732 			     uint64_t prev_layer __rte_unused,
2733 			     struct rte_flow_error *error)
2734 {
2735 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2736 	const struct rte_flow_item_mpls *mask = item->mask;
2737 	struct mlx5_priv *priv = dev->data->dev_private;
2738 	int ret;
2739 
2740 	if (!priv->config.mpls_en)
2741 		return rte_flow_error_set(error, ENOTSUP,
2742 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2743 					  "MPLS not supported or"
2744 					  " disabled in firmware"
2745 					  " configuration.");
2746 	/* MPLS over IP, UDP, GRE is allowed */
2747 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2748 			    MLX5_FLOW_LAYER_OUTER_L4_UDP |
2749 			    MLX5_FLOW_LAYER_GRE |
2750 			    MLX5_FLOW_LAYER_GRE_KEY)))
2751 		return rte_flow_error_set(error, EINVAL,
2752 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2753 					  "protocol filtering not compatible"
2754 					  " with MPLS layer");
2755 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2756 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2757 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
2758 		return rte_flow_error_set(error, ENOTSUP,
2759 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2760 					  "multiple tunnel layers not"
2761 					  " supported");
2762 	if (!mask)
2763 		mask = &rte_flow_item_mpls_mask;
2764 	ret = mlx5_flow_item_acceptable
2765 		(item, (const uint8_t *)mask,
2766 		 (const uint8_t *)&rte_flow_item_mpls_mask,
2767 		 sizeof(struct rte_flow_item_mpls),
2768 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2769 	if (ret < 0)
2770 		return ret;
2771 	return 0;
2772 #else
2773 	return rte_flow_error_set(error, ENOTSUP,
2774 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
2775 				  "MPLS is not supported by Verbs, please"
2776 				  " update.");
2777 #endif
2778 }
2779 
2780 /**
2781  * Validate NVGRE item.
2782  *
2783  * @param[in] item
2784  *   Item specification.
2785  * @param[in] item_flags
2786  *   Bit flags to mark detected items.
2787  * @param[in] target_protocol
2788  *   The next protocol in the previous item.
2789  * @param[out] error
2790  *   Pointer to error structure.
2791  *
2792  * @return
2793  *   0 on success, a negative errno value otherwise and rte_errno is set.
2794  */
2795 int
2796 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2797 			      uint64_t item_flags,
2798 			      uint8_t target_protocol,
2799 			      struct rte_flow_error *error)
2800 {
2801 	const struct rte_flow_item_nvgre *mask = item->mask;
2802 	int ret;
2803 
2804 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2805 		return rte_flow_error_set(error, EINVAL,
2806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 					  "protocol filtering not compatible"
2808 					  " with this GRE layer");
2809 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2810 		return rte_flow_error_set(error, ENOTSUP,
2811 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2812 					  "multiple tunnel layers not"
2813 					  " supported");
2814 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2815 		return rte_flow_error_set(error, ENOTSUP,
2816 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2817 					  "L3 Layer is missing");
2818 	if (!mask)
2819 		mask = &rte_flow_item_nvgre_mask;
2820 	ret = mlx5_flow_item_acceptable
2821 		(item, (const uint8_t *)mask,
2822 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
2823 		 sizeof(struct rte_flow_item_nvgre),
2824 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2825 	if (ret < 0)
2826 		return ret;
2827 	return 0;
2828 }
2829 
2830 /**
2831  * Validate eCPRI item.
2832  *
2833  * @param[in] item
2834  *   Item specification.
2835  * @param[in] item_flags
2836  *   Bit-fields that holds the items detected until now.
2837  * @param[in] last_item
2838  *   Previous validated item in the pattern items.
2839  * @param[in] ether_type
2840  *   Type in the ethernet layer header (including dot1q).
2841  * @param[in] acc_mask
2842  *   Acceptable mask, if NULL default internal default mask
2843  *   will be used to check whether item fields are supported.
2844  * @param[out] error
2845  *   Pointer to error structure.
2846  *
2847  * @return
2848  *   0 on success, a negative errno value otherwise and rte_errno is set.
2849  */
2850 int
2851 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
2852 			      uint64_t item_flags,
2853 			      uint64_t last_item,
2854 			      uint16_t ether_type,
2855 			      const struct rte_flow_item_ecpri *acc_mask,
2856 			      struct rte_flow_error *error)
2857 {
2858 	const struct rte_flow_item_ecpri *mask = item->mask;
2859 	const struct rte_flow_item_ecpri nic_mask = {
2860 		.hdr = {
2861 			.common = {
2862 				.u32 =
2863 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
2864 					.type = 0xFF,
2865 					}).u32),
2866 			},
2867 			.dummy[0] = 0xFFFFFFFF,
2868 		},
2869 	};
2870 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
2871 					MLX5_FLOW_LAYER_OUTER_VLAN);
2872 	struct rte_flow_item_ecpri mask_lo;
2873 
2874 	if (!(last_item & outer_l2_vlan) &&
2875 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
2876 		return rte_flow_error_set(error, EINVAL,
2877 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2878 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
2879 	if ((last_item & outer_l2_vlan) && ether_type &&
2880 	    ether_type != RTE_ETHER_TYPE_ECPRI)
2881 		return rte_flow_error_set(error, EINVAL,
2882 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2883 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
2884 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2885 		return rte_flow_error_set(error, EINVAL,
2886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2887 					  "eCPRI with tunnel is not supported right now");
2888 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
2889 		return rte_flow_error_set(error, ENOTSUP,
2890 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2891 					  "multiple L3 layers not supported");
2892 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
2893 		return rte_flow_error_set(error, EINVAL,
2894 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2895 					  "eCPRI cannot coexist with a TCP layer");
2896 	/* In specification, eCPRI could be over UDP layer. */
2897 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
2898 		return rte_flow_error_set(error, EINVAL,
2899 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2900 					  "eCPRI over UDP layer is not yet supported right now");
2901 	/* Mask for type field in common header could be zero. */
2902 	if (!mask)
2903 		mask = &rte_flow_item_ecpri_mask;
2904 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
2905 	/* Input mask is in big-endian format. */
2906 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
2907 		return rte_flow_error_set(error, EINVAL,
2908 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2909 					  "partial mask is not supported for protocol");
2910 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
2911 		return rte_flow_error_set(error, EINVAL,
2912 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2913 					  "message header mask must be after a type mask");
2914 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2915 					 acc_mask ? (const uint8_t *)acc_mask
2916 						  : (const uint8_t *)&nic_mask,
2917 					 sizeof(struct rte_flow_item_ecpri),
2918 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2919 }
2920 
2921 /**
2922  * Release resource related QUEUE/RSS action split.
2923  *
2924  * @param dev
2925  *   Pointer to Ethernet device.
2926  * @param flow
2927  *   Flow to release id's from.
2928  */
2929 static void
2930 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2931 			     struct rte_flow *flow)
2932 {
2933 	struct mlx5_priv *priv = dev->data->dev_private;
2934 	uint32_t handle_idx;
2935 	struct mlx5_flow_handle *dev_handle;
2936 
2937 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2938 		       handle_idx, dev_handle, next)
2939 		if (dev_handle->split_flow_id)
2940 			mlx5_ipool_free(priv->sh->ipool
2941 					[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
2942 					dev_handle->split_flow_id);
2943 }
2944 
2945 static int
2946 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2947 		   const struct rte_flow_attr *attr __rte_unused,
2948 		   const struct rte_flow_item items[] __rte_unused,
2949 		   const struct rte_flow_action actions[] __rte_unused,
2950 		   bool external __rte_unused,
2951 		   int hairpin __rte_unused,
2952 		   struct rte_flow_error *error)
2953 {
2954 	return rte_flow_error_set(error, ENOTSUP,
2955 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2956 }
2957 
2958 static struct mlx5_flow *
2959 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2960 		  const struct rte_flow_attr *attr __rte_unused,
2961 		  const struct rte_flow_item items[] __rte_unused,
2962 		  const struct rte_flow_action actions[] __rte_unused,
2963 		  struct rte_flow_error *error)
2964 {
2965 	rte_flow_error_set(error, ENOTSUP,
2966 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2967 	return NULL;
2968 }
2969 
2970 static int
2971 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2972 		    struct mlx5_flow *dev_flow __rte_unused,
2973 		    const struct rte_flow_attr *attr __rte_unused,
2974 		    const struct rte_flow_item items[] __rte_unused,
2975 		    const struct rte_flow_action actions[] __rte_unused,
2976 		    struct rte_flow_error *error)
2977 {
2978 	return rte_flow_error_set(error, ENOTSUP,
2979 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2980 }
2981 
2982 static int
2983 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2984 		struct rte_flow *flow __rte_unused,
2985 		struct rte_flow_error *error)
2986 {
2987 	return rte_flow_error_set(error, ENOTSUP,
2988 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2989 }
2990 
2991 static void
2992 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2993 		 struct rte_flow *flow __rte_unused)
2994 {
2995 }
2996 
2997 static void
2998 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2999 		  struct rte_flow *flow __rte_unused)
3000 {
3001 }
3002 
3003 static int
3004 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3005 		struct rte_flow *flow __rte_unused,
3006 		const struct rte_flow_action *actions __rte_unused,
3007 		void *data __rte_unused,
3008 		struct rte_flow_error *error)
3009 {
3010 	return rte_flow_error_set(error, ENOTSUP,
3011 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3012 }
3013 
3014 static int
3015 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3016 		      uint32_t domains __rte_unused,
3017 		      uint32_t flags __rte_unused)
3018 {
3019 	return 0;
3020 }
3021 
3022 /* Void driver to protect from null pointer reference. */
3023 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3024 	.validate = flow_null_validate,
3025 	.prepare = flow_null_prepare,
3026 	.translate = flow_null_translate,
3027 	.apply = flow_null_apply,
3028 	.remove = flow_null_remove,
3029 	.destroy = flow_null_destroy,
3030 	.query = flow_null_query,
3031 	.sync_domain = flow_null_sync_domain,
3032 };
3033 
3034 /**
3035  * Select flow driver type according to flow attributes and device
3036  * configuration.
3037  *
3038  * @param[in] dev
3039  *   Pointer to the dev structure.
3040  * @param[in] attr
3041  *   Pointer to the flow attributes.
3042  *
3043  * @return
3044  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3045  */
3046 static enum mlx5_flow_drv_type
3047 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3048 {
3049 	struct mlx5_priv *priv = dev->data->dev_private;
3050 	/* The OS can determine first a specific flow type (DV, VERBS) */
3051 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3052 
3053 	if (type != MLX5_FLOW_TYPE_MAX)
3054 		return type;
3055 	/* If no OS specific type - continue with DV/VERBS selection */
3056 	if (attr->transfer && priv->config.dv_esw_en)
3057 		type = MLX5_FLOW_TYPE_DV;
3058 	if (!attr->transfer)
3059 		type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3060 						 MLX5_FLOW_TYPE_VERBS;
3061 	return type;
3062 }
3063 
3064 #define flow_get_drv_ops(type) flow_drv_ops[type]
3065 
3066 /**
3067  * Flow driver validation API. This abstracts calling driver specific functions.
3068  * The type of flow driver is determined according to flow attributes.
3069  *
3070  * @param[in] dev
3071  *   Pointer to the dev structure.
3072  * @param[in] attr
3073  *   Pointer to the flow attributes.
3074  * @param[in] items
3075  *   Pointer to the list of items.
3076  * @param[in] actions
3077  *   Pointer to the list of actions.
3078  * @param[in] external
3079  *   This flow rule is created by request external to PMD.
3080  * @param[in] hairpin
3081  *   Number of hairpin TX actions, 0 means classic flow.
3082  * @param[out] error
3083  *   Pointer to the error structure.
3084  *
3085  * @return
3086  *   0 on success, a negative errno value otherwise and rte_errno is set.
3087  */
3088 static inline int
3089 flow_drv_validate(struct rte_eth_dev *dev,
3090 		  const struct rte_flow_attr *attr,
3091 		  const struct rte_flow_item items[],
3092 		  const struct rte_flow_action actions[],
3093 		  bool external, int hairpin, struct rte_flow_error *error)
3094 {
3095 	const struct mlx5_flow_driver_ops *fops;
3096 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3097 
3098 	fops = flow_get_drv_ops(type);
3099 	return fops->validate(dev, attr, items, actions, external,
3100 			      hairpin, error);
3101 }
3102 
3103 /**
3104  * Flow driver preparation API. This abstracts calling driver specific
3105  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3106  * calculates the size of memory required for device flow, allocates the memory,
3107  * initializes the device flow and returns the pointer.
3108  *
3109  * @note
3110  *   This function initializes device flow structure such as dv or verbs in
3111  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3112  *   rest. For example, adding returning device flow to flow->dev_flow list and
3113  *   setting backward reference to the flow should be done out of this function.
3114  *   layers field is not filled either.
3115  *
3116  * @param[in] dev
3117  *   Pointer to the dev structure.
3118  * @param[in] attr
3119  *   Pointer to the flow attributes.
3120  * @param[in] items
3121  *   Pointer to the list of items.
3122  * @param[in] actions
3123  *   Pointer to the list of actions.
3124  * @param[in] flow_idx
3125  *   This memory pool index to the flow.
3126  * @param[out] error
3127  *   Pointer to the error structure.
3128  *
3129  * @return
3130  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3131  */
3132 static inline struct mlx5_flow *
3133 flow_drv_prepare(struct rte_eth_dev *dev,
3134 		 const struct rte_flow *flow,
3135 		 const struct rte_flow_attr *attr,
3136 		 const struct rte_flow_item items[],
3137 		 const struct rte_flow_action actions[],
3138 		 uint32_t flow_idx,
3139 		 struct rte_flow_error *error)
3140 {
3141 	const struct mlx5_flow_driver_ops *fops;
3142 	enum mlx5_flow_drv_type type = flow->drv_type;
3143 	struct mlx5_flow *mlx5_flow = NULL;
3144 
3145 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3146 	fops = flow_get_drv_ops(type);
3147 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3148 	if (mlx5_flow)
3149 		mlx5_flow->flow_idx = flow_idx;
3150 	return mlx5_flow;
3151 }
3152 
3153 /**
3154  * Flow driver translation API. This abstracts calling driver specific
3155  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3156  * translates a generic flow into a driver flow. flow_drv_prepare() must
3157  * precede.
3158  *
3159  * @note
3160  *   dev_flow->layers could be filled as a result of parsing during translation
3161  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3162  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3163  *   flow->actions could be overwritten even though all the expanded dev_flows
3164  *   have the same actions.
3165  *
3166  * @param[in] dev
3167  *   Pointer to the rte dev structure.
3168  * @param[in, out] dev_flow
3169  *   Pointer to the mlx5 flow.
3170  * @param[in] attr
3171  *   Pointer to the flow attributes.
3172  * @param[in] items
3173  *   Pointer to the list of items.
3174  * @param[in] actions
3175  *   Pointer to the list of actions.
3176  * @param[out] error
3177  *   Pointer to the error structure.
3178  *
3179  * @return
3180  *   0 on success, a negative errno value otherwise and rte_errno is set.
3181  */
3182 static inline int
3183 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3184 		   const struct rte_flow_attr *attr,
3185 		   const struct rte_flow_item items[],
3186 		   const struct rte_flow_action actions[],
3187 		   struct rte_flow_error *error)
3188 {
3189 	const struct mlx5_flow_driver_ops *fops;
3190 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3191 
3192 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3193 	fops = flow_get_drv_ops(type);
3194 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3195 }
3196 
3197 /**
3198  * Flow driver apply API. This abstracts calling driver specific functions.
3199  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3200  * translated driver flows on to device. flow_drv_translate() must precede.
3201  *
3202  * @param[in] dev
3203  *   Pointer to Ethernet device structure.
3204  * @param[in, out] flow
3205  *   Pointer to flow structure.
3206  * @param[out] error
3207  *   Pointer to error structure.
3208  *
3209  * @return
3210  *   0 on success, a negative errno value otherwise and rte_errno is set.
3211  */
3212 static inline int
3213 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3214 	       struct rte_flow_error *error)
3215 {
3216 	const struct mlx5_flow_driver_ops *fops;
3217 	enum mlx5_flow_drv_type type = flow->drv_type;
3218 
3219 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3220 	fops = flow_get_drv_ops(type);
3221 	return fops->apply(dev, flow, error);
3222 }
3223 
3224 /**
3225  * Flow driver destroy API. This abstracts calling driver specific functions.
3226  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3227  * on device and releases resources of the flow.
3228  *
3229  * @param[in] dev
3230  *   Pointer to Ethernet device.
3231  * @param[in, out] flow
3232  *   Pointer to flow structure.
3233  */
3234 static inline void
3235 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3236 {
3237 	const struct mlx5_flow_driver_ops *fops;
3238 	enum mlx5_flow_drv_type type = flow->drv_type;
3239 
3240 	flow_mreg_split_qrss_release(dev, flow);
3241 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3242 	fops = flow_get_drv_ops(type);
3243 	fops->destroy(dev, flow);
3244 }
3245 
3246 /**
3247  * Get RSS action from the action list.
3248  *
3249  * @param[in] actions
3250  *   Pointer to the list of actions.
3251  *
3252  * @return
3253  *   Pointer to the RSS action if exist, else return NULL.
3254  */
3255 static const struct rte_flow_action_rss*
3256 flow_get_rss_action(const struct rte_flow_action actions[])
3257 {
3258 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3259 		switch (actions->type) {
3260 		case RTE_FLOW_ACTION_TYPE_RSS:
3261 			return (const struct rte_flow_action_rss *)
3262 			       actions->conf;
3263 		default:
3264 			break;
3265 		}
3266 	}
3267 	return NULL;
3268 }
3269 
3270 /**
3271  * Get ASO age action by index.
3272  *
3273  * @param[in] dev
3274  *   Pointer to the Ethernet device structure.
3275  * @param[in] age_idx
3276  *   Index to the ASO age action.
3277  *
3278  * @return
3279  *   The specified ASO age action.
3280  */
3281 struct mlx5_aso_age_action*
3282 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3283 {
3284 	uint16_t pool_idx = age_idx & UINT16_MAX;
3285 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3286 	struct mlx5_priv *priv = dev->data->dev_private;
3287 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3288 	struct mlx5_aso_age_pool *pool = mng->pools[pool_idx];
3289 
3290 	return &pool->actions[offset - 1];
3291 }
3292 
3293 /* maps shared action to translated non shared in some actions array */
3294 struct mlx5_translated_shared_action {
3295 	struct rte_flow_shared_action *action; /**< Shared action */
3296 	int index; /**< Index in related array of rte_flow_action */
3297 };
3298 
3299 /**
3300  * Translates actions of type RTE_FLOW_ACTION_TYPE_SHARED to related
3301  * non shared action if translation possible.
3302  * This functionality used to run same execution path for both shared & non
3303  * shared actions on flow create. All necessary preparations for shared
3304  * action handling should be preformed on *shared* actions list returned
3305  * from this call.
3306  *
3307  * @param[in] dev
3308  *   Pointer to Ethernet device.
3309  * @param[in] actions
3310  *   List of actions to translate.
3311  * @param[out] shared
3312  *   List to store translated shared actions.
3313  * @param[in, out] shared_n
3314  *   Size of *shared* array. On return should be updated with number of shared
3315  *   actions retrieved from the *actions* list.
3316  * @param[out] translated_actions
3317  *   List of actions where all shared actions were translated to non shared
3318  *   if possible. NULL if no translation took place.
3319  * @param[out] error
3320  *   Pointer to the error structure.
3321  *
3322  * @return
3323  *   0 on success, a negative errno value otherwise and rte_errno is set.
3324  */
3325 static int
3326 flow_shared_actions_translate(struct rte_eth_dev *dev,
3327 			      const struct rte_flow_action actions[],
3328 			      struct mlx5_translated_shared_action *shared,
3329 			      int *shared_n,
3330 			      struct rte_flow_action **translated_actions,
3331 			      struct rte_flow_error *error)
3332 {
3333 	struct mlx5_priv *priv = dev->data->dev_private;
3334 	struct rte_flow_action *translated = NULL;
3335 	size_t actions_size;
3336 	int n;
3337 	int copied_n = 0;
3338 	struct mlx5_translated_shared_action *shared_end = NULL;
3339 
3340 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3341 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_SHARED)
3342 			continue;
3343 		if (copied_n == *shared_n) {
3344 			return rte_flow_error_set
3345 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3346 				 NULL, "too many shared actions");
3347 		}
3348 		rte_memcpy(&shared[copied_n].action, &actions[n].conf,
3349 			   sizeof(actions[n].conf));
3350 		shared[copied_n].index = n;
3351 		copied_n++;
3352 	}
3353 	n++;
3354 	*shared_n = copied_n;
3355 	if (!copied_n)
3356 		return 0;
3357 	actions_size = sizeof(struct rte_flow_action) * n;
3358 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3359 	if (!translated) {
3360 		rte_errno = ENOMEM;
3361 		return -ENOMEM;
3362 	}
3363 	memcpy(translated, actions, actions_size);
3364 	for (shared_end = shared + copied_n; shared < shared_end; shared++) {
3365 		struct mlx5_shared_action_rss *shared_rss;
3366 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3367 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3368 		uint32_t idx = act_idx & ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET)
3369 									   - 1);
3370 
3371 		switch (type) {
3372 		case MLX5_SHARED_ACTION_TYPE_RSS:
3373 			shared_rss = mlx5_ipool_get
3374 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3375 			translated[shared->index].type =
3376 				RTE_FLOW_ACTION_TYPE_RSS;
3377 			translated[shared->index].conf =
3378 				&shared_rss->origin;
3379 			break;
3380 		case MLX5_SHARED_ACTION_TYPE_AGE:
3381 			if (priv->sh->flow_hit_aso_en) {
3382 				translated[shared->index].type =
3383 					(enum rte_flow_action_type)
3384 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3385 				translated[shared->index].conf =
3386 							 (void *)(uintptr_t)idx;
3387 				break;
3388 			}
3389 			/* Fall-through */
3390 		default:
3391 			mlx5_free(translated);
3392 			return rte_flow_error_set
3393 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3394 				 NULL, "invalid shared action type");
3395 		}
3396 	}
3397 	*translated_actions = translated;
3398 	return 0;
3399 }
3400 
3401 /**
3402  * Get Shared RSS action from the action list.
3403  *
3404  * @param[in] dev
3405  *   Pointer to Ethernet device.
3406  * @param[in] shared
3407  *   Pointer to the list of actions.
3408  * @param[in] shared_n
3409  *   Actions list length.
3410  *
3411  * @return
3412  *   The MLX5 RSS action ID if exists, otherwise return 0.
3413  */
3414 static uint32_t
3415 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3416 			   struct mlx5_translated_shared_action *shared,
3417 			   int shared_n)
3418 {
3419 	struct mlx5_translated_shared_action *shared_end;
3420 	struct mlx5_priv *priv = dev->data->dev_private;
3421 	struct mlx5_shared_action_rss *shared_rss;
3422 
3423 
3424 	for (shared_end = shared + shared_n; shared < shared_end; shared++) {
3425 		uint32_t act_idx = (uint32_t)(uintptr_t)shared->action;
3426 		uint32_t type = act_idx >> MLX5_SHARED_ACTION_TYPE_OFFSET;
3427 		uint32_t idx = act_idx &
3428 				   ((1u << MLX5_SHARED_ACTION_TYPE_OFFSET) - 1);
3429 		switch (type) {
3430 		case MLX5_SHARED_ACTION_TYPE_RSS:
3431 			shared_rss = mlx5_ipool_get
3432 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3433 									   idx);
3434 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3435 					   __ATOMIC_RELAXED);
3436 			return idx;
3437 		default:
3438 			break;
3439 		}
3440 	}
3441 	return 0;
3442 }
3443 
3444 static unsigned int
3445 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
3446 {
3447 	const struct rte_flow_item *item;
3448 	unsigned int has_vlan = 0;
3449 
3450 	for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
3451 		if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
3452 			has_vlan = 1;
3453 			break;
3454 		}
3455 	}
3456 	if (has_vlan)
3457 		return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
3458 				       MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
3459 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3460 			       MLX5_EXPANSION_ROOT_OUTER;
3461 }
3462 
3463 /**
3464  *  Get layer flags from the prefix flow.
3465  *
3466  *  Some flows may be split to several subflows, the prefix subflow gets the
3467  *  match items and the suffix sub flow gets the actions.
3468  *  Some actions need the user defined match item flags to get the detail for
3469  *  the action.
3470  *  This function helps the suffix flow to get the item layer flags from prefix
3471  *  subflow.
3472  *
3473  * @param[in] dev_flow
3474  *   Pointer the created preifx subflow.
3475  *
3476  * @return
3477  *   The layers get from prefix subflow.
3478  */
3479 static inline uint64_t
3480 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3481 {
3482 	uint64_t layers = 0;
3483 
3484 	/*
3485 	 * Layers bits could be localization, but usually the compiler will
3486 	 * help to do the optimization work for source code.
3487 	 * If no decap actions, use the layers directly.
3488 	 */
3489 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3490 		return dev_flow->handle->layers;
3491 	/* Convert L3 layers with decap action. */
3492 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3493 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3494 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3495 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3496 	/* Convert L4 layers with decap action.  */
3497 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3498 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3499 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3500 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3501 	return layers;
3502 }
3503 
3504 /**
3505  * Get metadata split action information.
3506  *
3507  * @param[in] actions
3508  *   Pointer to the list of actions.
3509  * @param[out] qrss
3510  *   Pointer to the return pointer.
3511  * @param[out] qrss_type
3512  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
3513  *   if no QUEUE/RSS is found.
3514  * @param[out] encap_idx
3515  *   Pointer to the index of the encap action if exists, otherwise the last
3516  *   action index.
3517  *
3518  * @return
3519  *   Total number of actions.
3520  */
3521 static int
3522 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
3523 				       const struct rte_flow_action **qrss,
3524 				       int *encap_idx)
3525 {
3526 	const struct rte_flow_action_raw_encap *raw_encap;
3527 	int actions_n = 0;
3528 	int raw_decap_idx = -1;
3529 
3530 	*encap_idx = -1;
3531 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3532 		switch (actions->type) {
3533 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3534 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3535 			*encap_idx = actions_n;
3536 			break;
3537 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3538 			raw_decap_idx = actions_n;
3539 			break;
3540 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3541 			raw_encap = actions->conf;
3542 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3543 				*encap_idx = raw_decap_idx != -1 ?
3544 						      raw_decap_idx : actions_n;
3545 			break;
3546 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3547 		case RTE_FLOW_ACTION_TYPE_RSS:
3548 			*qrss = actions;
3549 			break;
3550 		default:
3551 			break;
3552 		}
3553 		actions_n++;
3554 	}
3555 	if (*encap_idx == -1)
3556 		*encap_idx = actions_n;
3557 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3558 	return actions_n + 1;
3559 }
3560 
3561 /**
3562  * Check meter action from the action list.
3563  *
3564  * @param[in] actions
3565  *   Pointer to the list of actions.
3566  * @param[out] mtr
3567  *   Pointer to the meter exist flag.
3568  *
3569  * @return
3570  *   Total number of actions.
3571  */
3572 static int
3573 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
3574 {
3575 	int actions_n = 0;
3576 
3577 	MLX5_ASSERT(mtr);
3578 	*mtr = 0;
3579 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3580 		switch (actions->type) {
3581 		case RTE_FLOW_ACTION_TYPE_METER:
3582 			*mtr = 1;
3583 			break;
3584 		default:
3585 			break;
3586 		}
3587 		actions_n++;
3588 	}
3589 	/* Count RTE_FLOW_ACTION_TYPE_END. */
3590 	return actions_n + 1;
3591 }
3592 
3593 /**
3594  * Check if the flow should be split due to hairpin.
3595  * The reason for the split is that in current HW we can't
3596  * support encap and push-vlan on Rx, so if a flow contains
3597  * these actions we move it to Tx.
3598  *
3599  * @param dev
3600  *   Pointer to Ethernet device.
3601  * @param[in] attr
3602  *   Flow rule attributes.
3603  * @param[in] actions
3604  *   Associated actions (list terminated by the END action).
3605  *
3606  * @return
3607  *   > 0 the number of actions and the flow should be split,
3608  *   0 when no split required.
3609  */
3610 static int
3611 flow_check_hairpin_split(struct rte_eth_dev *dev,
3612 			 const struct rte_flow_attr *attr,
3613 			 const struct rte_flow_action actions[])
3614 {
3615 	int queue_action = 0;
3616 	int action_n = 0;
3617 	int split = 0;
3618 	const struct rte_flow_action_queue *queue;
3619 	const struct rte_flow_action_rss *rss;
3620 	const struct rte_flow_action_raw_encap *raw_encap;
3621 	const struct rte_eth_hairpin_conf *conf;
3622 
3623 	if (!attr->ingress)
3624 		return 0;
3625 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3626 		switch (actions->type) {
3627 		case RTE_FLOW_ACTION_TYPE_QUEUE:
3628 			queue = actions->conf;
3629 			if (queue == NULL)
3630 				return 0;
3631 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
3632 			if (conf != NULL && !!conf->tx_explicit)
3633 				return 0;
3634 			queue_action = 1;
3635 			action_n++;
3636 			break;
3637 		case RTE_FLOW_ACTION_TYPE_RSS:
3638 			rss = actions->conf;
3639 			if (rss == NULL || rss->queue_num == 0)
3640 				return 0;
3641 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
3642 			if (conf != NULL && !!conf->tx_explicit)
3643 				return 0;
3644 			queue_action = 1;
3645 			action_n++;
3646 			break;
3647 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3648 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3649 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3650 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3651 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3652 			split++;
3653 			action_n++;
3654 			break;
3655 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3656 			raw_encap = actions->conf;
3657 			if (raw_encap->size >
3658 			    (sizeof(struct rte_flow_item_eth) +
3659 			     sizeof(struct rte_flow_item_ipv4)))
3660 				split++;
3661 			action_n++;
3662 			break;
3663 		default:
3664 			action_n++;
3665 			break;
3666 		}
3667 	}
3668 	if (split && queue_action)
3669 		return action_n;
3670 	return 0;
3671 }
3672 
3673 /* Declare flow create/destroy prototype in advance. */
3674 static uint32_t
3675 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
3676 		 const struct rte_flow_attr *attr,
3677 		 const struct rte_flow_item items[],
3678 		 const struct rte_flow_action actions[],
3679 		 bool external, struct rte_flow_error *error);
3680 
3681 static void
3682 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
3683 		  uint32_t flow_idx);
3684 
3685 struct mlx5_hlist_entry *
3686 flow_dv_mreg_create_cb(struct mlx5_hlist *list, uint64_t key,
3687 		       void *cb_ctx)
3688 {
3689 	struct rte_eth_dev *dev = list->ctx;
3690 	struct mlx5_priv *priv = dev->data->dev_private;
3691 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
3692 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3693 	struct rte_flow_error *error = ctx->error;
3694 	uint32_t idx = 0;
3695 	int ret;
3696 	uint32_t mark_id = key;
3697 	struct rte_flow_attr attr = {
3698 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3699 		.ingress = 1,
3700 	};
3701 	struct mlx5_rte_flow_item_tag tag_spec = {
3702 		.data = mark_id,
3703 	};
3704 	struct rte_flow_item items[] = {
3705 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
3706 	};
3707 	struct rte_flow_action_mark ftag = {
3708 		.id = mark_id,
3709 	};
3710 	struct mlx5_flow_action_copy_mreg cp_mreg = {
3711 		.dst = REG_B,
3712 		.src = REG_NON,
3713 	};
3714 	struct rte_flow_action_jump jump = {
3715 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3716 	};
3717 	struct rte_flow_action actions[] = {
3718 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
3719 	};
3720 
3721 	/* Fill the register fileds in the flow. */
3722 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
3723 	if (ret < 0)
3724 		return NULL;
3725 	tag_spec.id = ret;
3726 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3727 	if (ret < 0)
3728 		return NULL;
3729 	cp_mreg.src = ret;
3730 	/* Provide the full width of FLAG specific value. */
3731 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3732 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3733 	/* Build a new flow. */
3734 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
3735 		items[0] = (struct rte_flow_item){
3736 			.type = (enum rte_flow_item_type)
3737 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3738 			.spec = &tag_spec,
3739 		};
3740 		items[1] = (struct rte_flow_item){
3741 			.type = RTE_FLOW_ITEM_TYPE_END,
3742 		};
3743 		actions[0] = (struct rte_flow_action){
3744 			.type = (enum rte_flow_action_type)
3745 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3746 			.conf = &ftag,
3747 		};
3748 		actions[1] = (struct rte_flow_action){
3749 			.type = (enum rte_flow_action_type)
3750 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3751 			.conf = &cp_mreg,
3752 		};
3753 		actions[2] = (struct rte_flow_action){
3754 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3755 			.conf = &jump,
3756 		};
3757 		actions[3] = (struct rte_flow_action){
3758 			.type = RTE_FLOW_ACTION_TYPE_END,
3759 		};
3760 	} else {
3761 		/* Default rule, wildcard match. */
3762 		attr.priority = MLX5_FLOW_PRIO_RSVD;
3763 		items[0] = (struct rte_flow_item){
3764 			.type = RTE_FLOW_ITEM_TYPE_END,
3765 		};
3766 		actions[0] = (struct rte_flow_action){
3767 			.type = (enum rte_flow_action_type)
3768 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3769 			.conf = &cp_mreg,
3770 		};
3771 		actions[1] = (struct rte_flow_action){
3772 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
3773 			.conf = &jump,
3774 		};
3775 		actions[2] = (struct rte_flow_action){
3776 			.type = RTE_FLOW_ACTION_TYPE_END,
3777 		};
3778 	}
3779 	/* Build a new entry. */
3780 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3781 	if (!mcp_res) {
3782 		rte_errno = ENOMEM;
3783 		return NULL;
3784 	}
3785 	mcp_res->idx = idx;
3786 	/*
3787 	 * The copy Flows are not included in any list. There
3788 	 * ones are referenced from other Flows and can not
3789 	 * be applied, removed, deleted in ardbitrary order
3790 	 * by list traversing.
3791 	 */
3792 	mcp_res->rix_flow = flow_list_create(dev, NULL, &attr, items,
3793 					 actions, false, error);
3794 	if (!mcp_res->rix_flow) {
3795 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
3796 		return NULL;
3797 	}
3798 	return &mcp_res->hlist_ent;
3799 }
3800 
3801 /**
3802  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3803  *
3804  * As mark_id is unique, if there's already a registered flow for the mark_id,
3805  * return by increasing the reference counter of the resource. Otherwise, create
3806  * the resource (mcp_res) and flow.
3807  *
3808  * Flow looks like,
3809  *   - If ingress port is ANY and reg_c[1] is mark_id,
3810  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3811  *
3812  * For default flow (zero mark_id), flow is like,
3813  *   - If ingress port is ANY,
3814  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
3815  *
3816  * @param dev
3817  *   Pointer to Ethernet device.
3818  * @param mark_id
3819  *   ID of MARK action, zero means default flow for META.
3820  * @param[out] error
3821  *   Perform verbose error reporting if not NULL.
3822  *
3823  * @return
3824  *   Associated resource on success, NULL otherwise and rte_errno is set.
3825  */
3826 static struct mlx5_flow_mreg_copy_resource *
3827 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
3828 			  struct rte_flow_error *error)
3829 {
3830 	struct mlx5_priv *priv = dev->data->dev_private;
3831 	struct mlx5_hlist_entry *entry;
3832 	struct mlx5_flow_cb_ctx ctx = {
3833 		.dev = dev,
3834 		.error = error,
3835 	};
3836 
3837 	/* Check if already registered. */
3838 	MLX5_ASSERT(priv->mreg_cp_tbl);
3839 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
3840 	if (!entry)
3841 		return NULL;
3842 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
3843 			    hlist_ent);
3844 }
3845 
3846 void
3847 flow_dv_mreg_remove_cb(struct mlx5_hlist *list, struct mlx5_hlist_entry *entry)
3848 {
3849 	struct mlx5_flow_mreg_copy_resource *mcp_res =
3850 		container_of(entry, typeof(*mcp_res), hlist_ent);
3851 	struct rte_eth_dev *dev = list->ctx;
3852 	struct mlx5_priv *priv = dev->data->dev_private;
3853 
3854 	MLX5_ASSERT(mcp_res->rix_flow);
3855 	flow_list_destroy(dev, NULL, mcp_res->rix_flow);
3856 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3857 }
3858 
3859 /**
3860  * Release flow in RX_CP_TBL.
3861  *
3862  * @param dev
3863  *   Pointer to Ethernet device.
3864  * @flow
3865  *   Parent flow for wich copying is provided.
3866  */
3867 static void
3868 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3869 			  struct rte_flow *flow)
3870 {
3871 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3872 	struct mlx5_priv *priv = dev->data->dev_private;
3873 
3874 	if (!flow->rix_mreg_copy)
3875 		return;
3876 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3877 				 flow->rix_mreg_copy);
3878 	if (!mcp_res || !priv->mreg_cp_tbl)
3879 		return;
3880 	MLX5_ASSERT(mcp_res->rix_flow);
3881 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3882 	flow->rix_mreg_copy = 0;
3883 }
3884 
3885 /**
3886  * Remove the default copy action from RX_CP_TBL.
3887  *
3888  * This functions is called in the mlx5_dev_start(). No thread safe
3889  * is guaranteed.
3890  *
3891  * @param dev
3892  *   Pointer to Ethernet device.
3893  */
3894 static void
3895 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3896 {
3897 	struct mlx5_hlist_entry *entry;
3898 	struct mlx5_priv *priv = dev->data->dev_private;
3899 
3900 	/* Check if default flow is registered. */
3901 	if (!priv->mreg_cp_tbl)
3902 		return;
3903 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl,
3904 				  MLX5_DEFAULT_COPY_ID, NULL);
3905 	if (!entry)
3906 		return;
3907 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
3908 }
3909 
3910 /**
3911  * Add the default copy action in in RX_CP_TBL.
3912  *
3913  * This functions is called in the mlx5_dev_start(). No thread safe
3914  * is guaranteed.
3915  *
3916  * @param dev
3917  *   Pointer to Ethernet device.
3918  * @param[out] error
3919  *   Perform verbose error reporting if not NULL.
3920  *
3921  * @return
3922  *   0 for success, negative value otherwise and rte_errno is set.
3923  */
3924 static int
3925 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3926 				  struct rte_flow_error *error)
3927 {
3928 	struct mlx5_priv *priv = dev->data->dev_private;
3929 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3930 
3931 	/* Check whether extensive metadata feature is engaged. */
3932 	if (!priv->config.dv_flow_en ||
3933 	    priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3934 	    !mlx5_flow_ext_mreg_supported(dev) ||
3935 	    !priv->sh->dv_regc0_mask)
3936 		return 0;
3937 	/*
3938 	 * Add default mreg copy flow may be called multiple time, but
3939 	 * only be called once in stop. Avoid register it twice.
3940 	 */
3941 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, MLX5_DEFAULT_COPY_ID, NULL))
3942 		return 0;
3943 	mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3944 	if (!mcp_res)
3945 		return -rte_errno;
3946 	return 0;
3947 }
3948 
3949 /**
3950  * Add a flow of copying flow metadata registers in RX_CP_TBL.
3951  *
3952  * All the flow having Q/RSS action should be split by
3953  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3954  * performs the following,
3955  *   - CQE->flow_tag := reg_c[1] (MARK)
3956  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3957  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3958  * but there should be a flow per each MARK ID set by MARK action.
3959  *
3960  * For the aforementioned reason, if there's a MARK action in flow's action
3961  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3962  * the MARK ID to CQE's flow_tag like,
3963  *   - If reg_c[1] is mark_id,
3964  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3965  *
3966  * For SET_META action which stores value in reg_c[0], as the destination is
3967  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3968  * MARK ID means the default flow. The default flow looks like,
3969  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3970  *
3971  * @param dev
3972  *   Pointer to Ethernet device.
3973  * @param flow
3974  *   Pointer to flow structure.
3975  * @param[in] actions
3976  *   Pointer to the list of actions.
3977  * @param[out] error
3978  *   Perform verbose error reporting if not NULL.
3979  *
3980  * @return
3981  *   0 on success, negative value otherwise and rte_errno is set.
3982  */
3983 static int
3984 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3985 			    struct rte_flow *flow,
3986 			    const struct rte_flow_action *actions,
3987 			    struct rte_flow_error *error)
3988 {
3989 	struct mlx5_priv *priv = dev->data->dev_private;
3990 	struct mlx5_dev_config *config = &priv->config;
3991 	struct mlx5_flow_mreg_copy_resource *mcp_res;
3992 	const struct rte_flow_action_mark *mark;
3993 
3994 	/* Check whether extensive metadata feature is engaged. */
3995 	if (!config->dv_flow_en ||
3996 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3997 	    !mlx5_flow_ext_mreg_supported(dev) ||
3998 	    !priv->sh->dv_regc0_mask)
3999 		return 0;
4000 	/* Find MARK action. */
4001 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4002 		switch (actions->type) {
4003 		case RTE_FLOW_ACTION_TYPE_FLAG:
4004 			mcp_res = flow_mreg_add_copy_action
4005 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4006 			if (!mcp_res)
4007 				return -rte_errno;
4008 			flow->rix_mreg_copy = mcp_res->idx;
4009 			return 0;
4010 		case RTE_FLOW_ACTION_TYPE_MARK:
4011 			mark = (const struct rte_flow_action_mark *)
4012 				actions->conf;
4013 			mcp_res =
4014 				flow_mreg_add_copy_action(dev, mark->id, error);
4015 			if (!mcp_res)
4016 				return -rte_errno;
4017 			flow->rix_mreg_copy = mcp_res->idx;
4018 			return 0;
4019 		default:
4020 			break;
4021 		}
4022 	}
4023 	return 0;
4024 }
4025 
4026 #define MLX5_MAX_SPLIT_ACTIONS 24
4027 #define MLX5_MAX_SPLIT_ITEMS 24
4028 
4029 /**
4030  * Split the hairpin flow.
4031  * Since HW can't support encap and push-vlan on Rx, we move these
4032  * actions to Tx.
4033  * If the count action is after the encap then we also
4034  * move the count action. in this case the count will also measure
4035  * the outer bytes.
4036  *
4037  * @param dev
4038  *   Pointer to Ethernet device.
4039  * @param[in] actions
4040  *   Associated actions (list terminated by the END action).
4041  * @param[out] actions_rx
4042  *   Rx flow actions.
4043  * @param[out] actions_tx
4044  *   Tx flow actions..
4045  * @param[out] pattern_tx
4046  *   The pattern items for the Tx flow.
4047  * @param[out] flow_id
4048  *   The flow ID connected to this flow.
4049  *
4050  * @return
4051  *   0 on success.
4052  */
4053 static int
4054 flow_hairpin_split(struct rte_eth_dev *dev,
4055 		   const struct rte_flow_action actions[],
4056 		   struct rte_flow_action actions_rx[],
4057 		   struct rte_flow_action actions_tx[],
4058 		   struct rte_flow_item pattern_tx[],
4059 		   uint32_t flow_id)
4060 {
4061 	const struct rte_flow_action_raw_encap *raw_encap;
4062 	const struct rte_flow_action_raw_decap *raw_decap;
4063 	struct mlx5_rte_flow_action_set_tag *set_tag;
4064 	struct rte_flow_action *tag_action;
4065 	struct mlx5_rte_flow_item_tag *tag_item;
4066 	struct rte_flow_item *item;
4067 	char *addr;
4068 	int encap = 0;
4069 
4070 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4071 		switch (actions->type) {
4072 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4073 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4074 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4075 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4076 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4077 			rte_memcpy(actions_tx, actions,
4078 			       sizeof(struct rte_flow_action));
4079 			actions_tx++;
4080 			break;
4081 		case RTE_FLOW_ACTION_TYPE_COUNT:
4082 			if (encap) {
4083 				rte_memcpy(actions_tx, actions,
4084 					   sizeof(struct rte_flow_action));
4085 				actions_tx++;
4086 			} else {
4087 				rte_memcpy(actions_rx, actions,
4088 					   sizeof(struct rte_flow_action));
4089 				actions_rx++;
4090 			}
4091 			break;
4092 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4093 			raw_encap = actions->conf;
4094 			if (raw_encap->size >
4095 			    (sizeof(struct rte_flow_item_eth) +
4096 			     sizeof(struct rte_flow_item_ipv4))) {
4097 				memcpy(actions_tx, actions,
4098 				       sizeof(struct rte_flow_action));
4099 				actions_tx++;
4100 				encap = 1;
4101 			} else {
4102 				rte_memcpy(actions_rx, actions,
4103 					   sizeof(struct rte_flow_action));
4104 				actions_rx++;
4105 			}
4106 			break;
4107 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4108 			raw_decap = actions->conf;
4109 			if (raw_decap->size <
4110 			    (sizeof(struct rte_flow_item_eth) +
4111 			     sizeof(struct rte_flow_item_ipv4))) {
4112 				memcpy(actions_tx, actions,
4113 				       sizeof(struct rte_flow_action));
4114 				actions_tx++;
4115 			} else {
4116 				rte_memcpy(actions_rx, actions,
4117 					   sizeof(struct rte_flow_action));
4118 				actions_rx++;
4119 			}
4120 			break;
4121 		default:
4122 			rte_memcpy(actions_rx, actions,
4123 				   sizeof(struct rte_flow_action));
4124 			actions_rx++;
4125 			break;
4126 		}
4127 	}
4128 	/* Add set meta action and end action for the Rx flow. */
4129 	tag_action = actions_rx;
4130 	tag_action->type = (enum rte_flow_action_type)
4131 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4132 	actions_rx++;
4133 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4134 	actions_rx++;
4135 	set_tag = (void *)actions_rx;
4136 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
4137 	MLX5_ASSERT(set_tag->id > REG_NON);
4138 	set_tag->data = flow_id;
4139 	tag_action->conf = set_tag;
4140 	/* Create Tx item list. */
4141 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4142 	addr = (void *)&pattern_tx[2];
4143 	item = pattern_tx;
4144 	item->type = (enum rte_flow_item_type)
4145 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4146 	tag_item = (void *)addr;
4147 	tag_item->data = flow_id;
4148 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4149 	MLX5_ASSERT(set_tag->id > REG_NON);
4150 	item->spec = tag_item;
4151 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4152 	tag_item = (void *)addr;
4153 	tag_item->data = UINT32_MAX;
4154 	tag_item->id = UINT16_MAX;
4155 	item->mask = tag_item;
4156 	item->last = NULL;
4157 	item++;
4158 	item->type = RTE_FLOW_ITEM_TYPE_END;
4159 	return 0;
4160 }
4161 
4162 __extension__
4163 union tunnel_offload_mark {
4164 	uint32_t val;
4165 	struct {
4166 		uint32_t app_reserve:8;
4167 		uint32_t table_id:15;
4168 		uint32_t transfer:1;
4169 		uint32_t _unused_:8;
4170 	};
4171 };
4172 
4173 struct tunnel_default_miss_ctx {
4174 	uint16_t *queue;
4175 	__extension__
4176 	union {
4177 		struct rte_flow_action_rss action_rss;
4178 		struct rte_flow_action_queue miss_queue;
4179 		struct rte_flow_action_jump miss_jump;
4180 		uint8_t raw[0];
4181 	};
4182 };
4183 
4184 static int
4185 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
4186 			     struct rte_flow *flow,
4187 			     const struct rte_flow_attr *attr,
4188 			     const struct rte_flow_action *app_actions,
4189 			     uint32_t flow_idx,
4190 			     struct tunnel_default_miss_ctx *ctx,
4191 			     struct rte_flow_error *error)
4192 {
4193 	struct mlx5_priv *priv = dev->data->dev_private;
4194 	struct mlx5_flow *dev_flow;
4195 	struct rte_flow_attr miss_attr = *attr;
4196 	const struct mlx5_flow_tunnel *tunnel = app_actions[0].conf;
4197 	const struct rte_flow_item miss_items[2] = {
4198 		{
4199 			.type = RTE_FLOW_ITEM_TYPE_ETH,
4200 			.spec = NULL,
4201 			.last = NULL,
4202 			.mask = NULL
4203 		},
4204 		{
4205 			.type = RTE_FLOW_ITEM_TYPE_END,
4206 			.spec = NULL,
4207 			.last = NULL,
4208 			.mask = NULL
4209 		}
4210 	};
4211 	union tunnel_offload_mark mark_id;
4212 	struct rte_flow_action_mark miss_mark;
4213 	struct rte_flow_action miss_actions[3] = {
4214 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
4215 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
4216 	};
4217 	const struct rte_flow_action_jump *jump_data;
4218 	uint32_t i, flow_table = 0; /* prevent compilation warning */
4219 	struct flow_grp_info grp_info = {
4220 		.external = 1,
4221 		.transfer = attr->transfer,
4222 		.fdb_def_rule = !!priv->fdb_def_rule,
4223 		.std_tbl_fix = 0,
4224 	};
4225 	int ret;
4226 
4227 	if (!attr->transfer) {
4228 		uint32_t q_size;
4229 
4230 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
4231 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
4232 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
4233 					 0, SOCKET_ID_ANY);
4234 		if (!ctx->queue)
4235 			return rte_flow_error_set
4236 				(error, ENOMEM,
4237 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4238 				NULL, "invalid default miss RSS");
4239 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4240 		ctx->action_rss.level = 0,
4241 		ctx->action_rss.types = priv->rss_conf.rss_hf,
4242 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
4243 		ctx->action_rss.queue_num = priv->reta_idx_n,
4244 		ctx->action_rss.key = priv->rss_conf.rss_key,
4245 		ctx->action_rss.queue = ctx->queue;
4246 		if (!priv->reta_idx_n || !priv->rxqs_n)
4247 			return rte_flow_error_set
4248 				(error, EINVAL,
4249 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4250 				NULL, "invalid port configuration");
4251 		if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
4252 			ctx->action_rss.types = 0;
4253 		for (i = 0; i != priv->reta_idx_n; ++i)
4254 			ctx->queue[i] = (*priv->reta_idx)[i];
4255 	} else {
4256 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
4257 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
4258 	}
4259 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
4260 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
4261 	jump_data = app_actions->conf;
4262 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
4263 	miss_attr.group = jump_data->group;
4264 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
4265 				       &flow_table, grp_info, error);
4266 	if (ret)
4267 		return rte_flow_error_set(error, EINVAL,
4268 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
4269 					  NULL, "invalid tunnel id");
4270 	mark_id.app_reserve = 0;
4271 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
4272 	mark_id.transfer = !!attr->transfer;
4273 	mark_id._unused_ = 0;
4274 	miss_mark.id = mark_id.val;
4275 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
4276 				    miss_items, miss_actions, flow_idx, error);
4277 	if (!dev_flow)
4278 		return -rte_errno;
4279 	dev_flow->flow = flow;
4280 	dev_flow->external = true;
4281 	dev_flow->tunnel = tunnel;
4282 	/* Subflow object was created, we must include one in the list. */
4283 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4284 		      dev_flow->handle, next);
4285 	DRV_LOG(DEBUG,
4286 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
4287 		dev->data->port_id, tunnel->app_tunnel.type,
4288 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
4289 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
4290 				  miss_actions, error);
4291 	if (!ret)
4292 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
4293 						  error);
4294 
4295 	return ret;
4296 }
4297 
4298 /**
4299  * The last stage of splitting chain, just creates the subflow
4300  * without any modification.
4301  *
4302  * @param[in] dev
4303  *   Pointer to Ethernet device.
4304  * @param[in] flow
4305  *   Parent flow structure pointer.
4306  * @param[in, out] sub_flow
4307  *   Pointer to return the created subflow, may be NULL.
4308  * @param[in] attr
4309  *   Flow rule attributes.
4310  * @param[in] items
4311  *   Pattern specification (list terminated by the END pattern item).
4312  * @param[in] actions
4313  *   Associated actions (list terminated by the END action).
4314  * @param[in] flow_split_info
4315  *   Pointer to flow split info structure.
4316  * @param[out] error
4317  *   Perform verbose error reporting if not NULL.
4318  * @return
4319  *   0 on success, negative value otherwise
4320  */
4321 static int
4322 flow_create_split_inner(struct rte_eth_dev *dev,
4323 			struct rte_flow *flow,
4324 			struct mlx5_flow **sub_flow,
4325 			const struct rte_flow_attr *attr,
4326 			const struct rte_flow_item items[],
4327 			const struct rte_flow_action actions[],
4328 			struct mlx5_flow_split_info *flow_split_info,
4329 			struct rte_flow_error *error)
4330 {
4331 	struct mlx5_flow *dev_flow;
4332 
4333 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4334 				    flow_split_info->flow_idx, error);
4335 	if (!dev_flow)
4336 		return -rte_errno;
4337 	dev_flow->flow = flow;
4338 	dev_flow->external = flow_split_info->external;
4339 	dev_flow->skip_scale = flow_split_info->skip_scale;
4340 	/* Subflow object was created, we must include one in the list. */
4341 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4342 		      dev_flow->handle, next);
4343 	/*
4344 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4345 	 * flow may need some user defined item layer flags, and pass the
4346 	 * Metadate rxq mark flag to suffix flow as well.
4347 	 */
4348 	if (flow_split_info->prefix_layers)
4349 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4350 	if (flow_split_info->prefix_mark)
4351 		dev_flow->handle->mark = 1;
4352 	if (sub_flow)
4353 		*sub_flow = dev_flow;
4354 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4355 }
4356 
4357 /**
4358  * Split the meter flow.
4359  *
4360  * As meter flow will split to three sub flow, other than meter
4361  * action, the other actions make sense to only meter accepts
4362  * the packet. If it need to be dropped, no other additional
4363  * actions should be take.
4364  *
4365  * One kind of special action which decapsulates the L3 tunnel
4366  * header will be in the prefix sub flow, as not to take the
4367  * L3 tunnel header into account.
4368  *
4369  * @param dev
4370  *   Pointer to Ethernet device.
4371  * @param[in] items
4372  *   Pattern specification (list terminated by the END pattern item).
4373  * @param[out] sfx_items
4374  *   Suffix flow match items (list terminated by the END pattern item).
4375  * @param[in] actions
4376  *   Associated actions (list terminated by the END action).
4377  * @param[out] actions_sfx
4378  *   Suffix flow actions.
4379  * @param[out] actions_pre
4380  *   Prefix flow actions.
4381  * @param[out] pattern_sfx
4382  *   The pattern items for the suffix flow.
4383  * @param[out] tag_sfx
4384  *   Pointer to suffix flow tag.
4385  *
4386  * @return
4387  *   0 on success.
4388  */
4389 static int
4390 flow_meter_split_prep(struct rte_eth_dev *dev,
4391 		 const struct rte_flow_item items[],
4392 		 struct rte_flow_item sfx_items[],
4393 		 const struct rte_flow_action actions[],
4394 		 struct rte_flow_action actions_sfx[],
4395 		 struct rte_flow_action actions_pre[])
4396 {
4397 	struct mlx5_priv *priv = dev->data->dev_private;
4398 	struct rte_flow_action *tag_action = NULL;
4399 	struct rte_flow_item *tag_item;
4400 	struct mlx5_rte_flow_action_set_tag *set_tag;
4401 	struct rte_flow_error error;
4402 	const struct rte_flow_action_raw_encap *raw_encap;
4403 	const struct rte_flow_action_raw_decap *raw_decap;
4404 	struct mlx5_rte_flow_item_tag *tag_spec;
4405 	struct mlx5_rte_flow_item_tag *tag_mask;
4406 	uint32_t tag_id = 0;
4407 	bool copy_vlan = false;
4408 
4409 	/* Prepare the actions for prefix and suffix flow. */
4410 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4411 		struct rte_flow_action **action_cur = NULL;
4412 
4413 		switch (actions->type) {
4414 		case RTE_FLOW_ACTION_TYPE_METER:
4415 			/* Add the extra tag action first. */
4416 			tag_action = actions_pre;
4417 			tag_action->type = (enum rte_flow_action_type)
4418 					   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4419 			actions_pre++;
4420 			action_cur = &actions_pre;
4421 			break;
4422 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4423 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4424 			action_cur = &actions_pre;
4425 			break;
4426 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4427 			raw_encap = actions->conf;
4428 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
4429 				action_cur = &actions_pre;
4430 			break;
4431 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4432 			raw_decap = actions->conf;
4433 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4434 				action_cur = &actions_pre;
4435 			break;
4436 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4437 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4438 			copy_vlan = true;
4439 			break;
4440 		default:
4441 			break;
4442 		}
4443 		if (!action_cur)
4444 			action_cur = &actions_sfx;
4445 		memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
4446 		(*action_cur)++;
4447 	}
4448 	/* Add end action to the actions. */
4449 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
4450 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
4451 	actions_pre++;
4452 	/* Set the tag. */
4453 	set_tag = (void *)actions_pre;
4454 	set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4455 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
4456 			  &tag_id);
4457 	if (tag_id >= (1 << (sizeof(tag_id) * 8 - MLX5_MTR_COLOR_BITS))) {
4458 		DRV_LOG(ERR, "Port %u meter flow id exceed max limit.",
4459 			dev->data->port_id);
4460 		mlx5_ipool_free(priv->sh->ipool
4461 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], tag_id);
4462 		return 0;
4463 	} else if (!tag_id) {
4464 		return 0;
4465 	}
4466 	set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
4467 	assert(tag_action);
4468 	tag_action->conf = set_tag;
4469 	/* Prepare the suffix subflow items. */
4470 	tag_item = sfx_items++;
4471 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
4472 		int item_type = items->type;
4473 
4474 		switch (item_type) {
4475 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
4476 			memcpy(sfx_items, items, sizeof(*sfx_items));
4477 			sfx_items++;
4478 			break;
4479 		case RTE_FLOW_ITEM_TYPE_VLAN:
4480 			if (copy_vlan) {
4481 				memcpy(sfx_items, items, sizeof(*sfx_items));
4482 				/*
4483 				 * Convert to internal match item, it is used
4484 				 * for vlan push and set vid.
4485 				 */
4486 				sfx_items->type = (enum rte_flow_item_type)
4487 						  MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
4488 				sfx_items++;
4489 			}
4490 			break;
4491 		default:
4492 			break;
4493 		}
4494 	}
4495 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4496 	sfx_items++;
4497 	tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
4498 	tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
4499 	tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
4500 	tag_mask = tag_spec + 1;
4501 	tag_mask->data = 0xffffff00;
4502 	tag_item->type = (enum rte_flow_item_type)
4503 			 MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4504 	tag_item->spec = tag_spec;
4505 	tag_item->last = NULL;
4506 	tag_item->mask = tag_mask;
4507 	return tag_id;
4508 }
4509 
4510 /**
4511  * Split action list having QUEUE/RSS for metadata register copy.
4512  *
4513  * Once Q/RSS action is detected in user's action list, the flow action
4514  * should be split in order to copy metadata registers, which will happen in
4515  * RX_CP_TBL like,
4516  *   - CQE->flow_tag := reg_c[1] (MARK)
4517  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4518  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
4519  * This is because the last action of each flow must be a terminal action
4520  * (QUEUE, RSS or DROP).
4521  *
4522  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
4523  * stored and kept in the mlx5_flow structure per each sub_flow.
4524  *
4525  * The Q/RSS action is replaced with,
4526  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
4527  * And the following JUMP action is added at the end,
4528  *   - JUMP, to RX_CP_TBL.
4529  *
4530  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
4531  * flow_create_split_metadata() routine. The flow will look like,
4532  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
4533  *
4534  * @param dev
4535  *   Pointer to Ethernet device.
4536  * @param[out] split_actions
4537  *   Pointer to store split actions to jump to CP_TBL.
4538  * @param[in] actions
4539  *   Pointer to the list of original flow actions.
4540  * @param[in] qrss
4541  *   Pointer to the Q/RSS action.
4542  * @param[in] actions_n
4543  *   Number of original actions.
4544  * @param[out] error
4545  *   Perform verbose error reporting if not NULL.
4546  *
4547  * @return
4548  *   non-zero unique flow_id on success, otherwise 0 and
4549  *   error/rte_error are set.
4550  */
4551 static uint32_t
4552 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
4553 			  struct rte_flow_action *split_actions,
4554 			  const struct rte_flow_action *actions,
4555 			  const struct rte_flow_action *qrss,
4556 			  int actions_n, struct rte_flow_error *error)
4557 {
4558 	struct mlx5_priv *priv = dev->data->dev_private;
4559 	struct mlx5_rte_flow_action_set_tag *set_tag;
4560 	struct rte_flow_action_jump *jump;
4561 	const int qrss_idx = qrss - actions;
4562 	uint32_t flow_id = 0;
4563 	int ret = 0;
4564 
4565 	/*
4566 	 * Given actions will be split
4567 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
4568 	 * - Add jump to mreg CP_TBL.
4569 	 * As a result, there will be one more action.
4570 	 */
4571 	++actions_n;
4572 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
4573 	set_tag = (void *)(split_actions + actions_n);
4574 	/*
4575 	 * If tag action is not set to void(it means we are not the meter
4576 	 * suffix flow), add the tag action. Since meter suffix flow already
4577 	 * has the tag added.
4578 	 */
4579 	if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
4580 		/*
4581 		 * Allocate the new subflow ID. This one is unique within
4582 		 * device and not shared with representors. Otherwise,
4583 		 * we would have to resolve multi-thread access synch
4584 		 * issue. Each flow on the shared device is appended
4585 		 * with source vport identifier, so the resulting
4586 		 * flows will be unique in the shared (by master and
4587 		 * representors) domain even if they have coinciding
4588 		 * IDs.
4589 		 */
4590 		mlx5_ipool_malloc(priv->sh->ipool
4591 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
4592 		if (!flow_id)
4593 			return rte_flow_error_set(error, ENOMEM,
4594 						  RTE_FLOW_ERROR_TYPE_ACTION,
4595 						  NULL, "can't allocate id "
4596 						  "for split Q/RSS subflow");
4597 		/* Internal SET_TAG action to set flow ID. */
4598 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
4599 			.data = flow_id,
4600 		};
4601 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
4602 		if (ret < 0)
4603 			return ret;
4604 		set_tag->id = ret;
4605 		/* Construct new actions array. */
4606 		/* Replace QUEUE/RSS action. */
4607 		split_actions[qrss_idx] = (struct rte_flow_action){
4608 			.type = (enum rte_flow_action_type)
4609 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4610 			.conf = set_tag,
4611 		};
4612 	}
4613 	/* JUMP action to jump to mreg copy table (CP_TBL). */
4614 	jump = (void *)(set_tag + 1);
4615 	*jump = (struct rte_flow_action_jump){
4616 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4617 	};
4618 	split_actions[actions_n - 2] = (struct rte_flow_action){
4619 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
4620 		.conf = jump,
4621 	};
4622 	split_actions[actions_n - 1] = (struct rte_flow_action){
4623 		.type = RTE_FLOW_ACTION_TYPE_END,
4624 	};
4625 	return flow_id;
4626 }
4627 
4628 /**
4629  * Extend the given action list for Tx metadata copy.
4630  *
4631  * Copy the given action list to the ext_actions and add flow metadata register
4632  * copy action in order to copy reg_a set by WQE to reg_c[0].
4633  *
4634  * @param[out] ext_actions
4635  *   Pointer to the extended action list.
4636  * @param[in] actions
4637  *   Pointer to the list of actions.
4638  * @param[in] actions_n
4639  *   Number of actions in the list.
4640  * @param[out] error
4641  *   Perform verbose error reporting if not NULL.
4642  * @param[in] encap_idx
4643  *   The encap action inndex.
4644  *
4645  * @return
4646  *   0 on success, negative value otherwise
4647  */
4648 static int
4649 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
4650 		       struct rte_flow_action *ext_actions,
4651 		       const struct rte_flow_action *actions,
4652 		       int actions_n, struct rte_flow_error *error,
4653 		       int encap_idx)
4654 {
4655 	struct mlx5_flow_action_copy_mreg *cp_mreg =
4656 		(struct mlx5_flow_action_copy_mreg *)
4657 			(ext_actions + actions_n + 1);
4658 	int ret;
4659 
4660 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4661 	if (ret < 0)
4662 		return ret;
4663 	cp_mreg->dst = ret;
4664 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
4665 	if (ret < 0)
4666 		return ret;
4667 	cp_mreg->src = ret;
4668 	if (encap_idx != 0)
4669 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
4670 	if (encap_idx == actions_n - 1) {
4671 		ext_actions[actions_n - 1] = (struct rte_flow_action){
4672 			.type = (enum rte_flow_action_type)
4673 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4674 			.conf = cp_mreg,
4675 		};
4676 		ext_actions[actions_n] = (struct rte_flow_action){
4677 			.type = RTE_FLOW_ACTION_TYPE_END,
4678 		};
4679 	} else {
4680 		ext_actions[encap_idx] = (struct rte_flow_action){
4681 			.type = (enum rte_flow_action_type)
4682 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4683 			.conf = cp_mreg,
4684 		};
4685 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
4686 				sizeof(*ext_actions) * (actions_n - encap_idx));
4687 	}
4688 	return 0;
4689 }
4690 
4691 /**
4692  * Check the match action from the action list.
4693  *
4694  * @param[in] actions
4695  *   Pointer to the list of actions.
4696  * @param[in] attr
4697  *   Flow rule attributes.
4698  * @param[in] action
4699  *   The action to be check if exist.
4700  * @param[out] match_action_pos
4701  *   Pointer to the position of the matched action if exists, otherwise is -1.
4702  * @param[out] qrss_action_pos
4703  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
4704  *
4705  * @return
4706  *   > 0 the total number of actions.
4707  *   0 if not found match action in action list.
4708  */
4709 static int
4710 flow_check_match_action(const struct rte_flow_action actions[],
4711 			const struct rte_flow_attr *attr,
4712 			enum rte_flow_action_type action,
4713 			int *match_action_pos, int *qrss_action_pos)
4714 {
4715 	const struct rte_flow_action_sample *sample;
4716 	int actions_n = 0;
4717 	int jump_flag = 0;
4718 	uint32_t ratio = 0;
4719 	int sub_type = 0;
4720 	int flag = 0;
4721 
4722 	*match_action_pos = -1;
4723 	*qrss_action_pos = -1;
4724 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4725 		if (actions->type == action) {
4726 			flag = 1;
4727 			*match_action_pos = actions_n;
4728 		}
4729 		if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE ||
4730 		    actions->type == RTE_FLOW_ACTION_TYPE_RSS)
4731 			*qrss_action_pos = actions_n;
4732 		if (actions->type == RTE_FLOW_ACTION_TYPE_JUMP)
4733 			jump_flag = 1;
4734 		if (actions->type == RTE_FLOW_ACTION_TYPE_SAMPLE) {
4735 			sample = actions->conf;
4736 			ratio = sample->ratio;
4737 			sub_type = ((const struct rte_flow_action *)
4738 					(sample->actions))->type;
4739 		}
4740 		actions_n++;
4741 	}
4742 	if (flag && action == RTE_FLOW_ACTION_TYPE_SAMPLE && attr->transfer) {
4743 		if (ratio == 1) {
4744 			/* JUMP Action not support for Mirroring;
4745 			 * Mirroring support multi-destination;
4746 			 */
4747 			if (!jump_flag && sub_type != RTE_FLOW_ACTION_TYPE_END)
4748 				flag = 0;
4749 		}
4750 	}
4751 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4752 	return flag ? actions_n + 1 : 0;
4753 }
4754 
4755 #define SAMPLE_SUFFIX_ITEM 2
4756 
4757 /**
4758  * Split the sample flow.
4759  *
4760  * As sample flow will split to two sub flow, sample flow with
4761  * sample action, the other actions will move to new suffix flow.
4762  *
4763  * Also add unique tag id with tag action in the sample flow,
4764  * the same tag id will be as match in the suffix flow.
4765  *
4766  * @param dev
4767  *   Pointer to Ethernet device.
4768  * @param[in] fdb_tx
4769  *   FDB egress flow flag.
4770  * @param[out] sfx_items
4771  *   Suffix flow match items (list terminated by the END pattern item).
4772  * @param[in] actions
4773  *   Associated actions (list terminated by the END action).
4774  * @param[out] actions_sfx
4775  *   Suffix flow actions.
4776  * @param[out] actions_pre
4777  *   Prefix flow actions.
4778  * @param[in] actions_n
4779  *  The total number of actions.
4780  * @param[in] sample_action_pos
4781  *   The sample action position.
4782  * @param[in] qrss_action_pos
4783  *   The Queue/RSS action position.
4784  * @param[out] error
4785  *   Perform verbose error reporting if not NULL.
4786  *
4787  * @return
4788  *   0 on success, or unique flow_id, a negative errno value
4789  *   otherwise and rte_errno is set.
4790  */
4791 static int
4792 flow_sample_split_prep(struct rte_eth_dev *dev,
4793 		       uint32_t fdb_tx,
4794 		       struct rte_flow_item sfx_items[],
4795 		       const struct rte_flow_action actions[],
4796 		       struct rte_flow_action actions_sfx[],
4797 		       struct rte_flow_action actions_pre[],
4798 		       int actions_n,
4799 		       int sample_action_pos,
4800 		       int qrss_action_pos,
4801 		       struct rte_flow_error *error)
4802 {
4803 	struct mlx5_priv *priv = dev->data->dev_private;
4804 	struct mlx5_rte_flow_action_set_tag *set_tag;
4805 	struct mlx5_rte_flow_item_tag *tag_spec;
4806 	struct mlx5_rte_flow_item_tag *tag_mask;
4807 	uint32_t tag_id = 0;
4808 	int index;
4809 	int ret;
4810 
4811 	if (sample_action_pos < 0)
4812 		return rte_flow_error_set(error, EINVAL,
4813 					  RTE_FLOW_ERROR_TYPE_ACTION,
4814 					  NULL, "invalid position of sample "
4815 					  "action in list");
4816 	if (!fdb_tx) {
4817 		/* Prepare the prefix tag action. */
4818 		set_tag = (void *)(actions_pre + actions_n + 1);
4819 		ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
4820 		if (ret < 0)
4821 			return ret;
4822 		set_tag->id = ret;
4823 		mlx5_ipool_malloc(priv->sh->ipool
4824 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
4825 		set_tag->data = tag_id;
4826 		/* Prepare the suffix subflow items. */
4827 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
4828 		tag_spec->data = tag_id;
4829 		tag_spec->id = set_tag->id;
4830 		tag_mask = tag_spec + 1;
4831 		tag_mask->data = UINT32_MAX;
4832 		sfx_items[0] = (struct rte_flow_item){
4833 			.type = (enum rte_flow_item_type)
4834 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4835 			.spec = tag_spec,
4836 			.last = NULL,
4837 			.mask = tag_mask,
4838 		};
4839 		sfx_items[1] = (struct rte_flow_item){
4840 			.type = (enum rte_flow_item_type)
4841 				RTE_FLOW_ITEM_TYPE_END,
4842 		};
4843 	}
4844 	/* Prepare the actions for prefix and suffix flow. */
4845 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
4846 		index = qrss_action_pos;
4847 		/* Put the preceding the Queue/RSS action into prefix flow. */
4848 		if (index != 0)
4849 			memcpy(actions_pre, actions,
4850 			       sizeof(struct rte_flow_action) * index);
4851 		/* Put others preceding the sample action into prefix flow. */
4852 		if (sample_action_pos > index + 1)
4853 			memcpy(actions_pre + index, actions + index + 1,
4854 			       sizeof(struct rte_flow_action) *
4855 			       (sample_action_pos - index - 1));
4856 		index = sample_action_pos - 1;
4857 		/* Put Queue/RSS action into Suffix flow. */
4858 		memcpy(actions_sfx, actions + qrss_action_pos,
4859 		       sizeof(struct rte_flow_action));
4860 		actions_sfx++;
4861 	} else {
4862 		index = sample_action_pos;
4863 		if (index != 0)
4864 			memcpy(actions_pre, actions,
4865 			       sizeof(struct rte_flow_action) * index);
4866 	}
4867 	/* Add the extra tag action for NIC-RX and E-Switch ingress. */
4868 	if (!fdb_tx) {
4869 		actions_pre[index++] =
4870 			(struct rte_flow_action){
4871 			.type = (enum rte_flow_action_type)
4872 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
4873 			.conf = set_tag,
4874 		};
4875 	}
4876 	memcpy(actions_pre + index, actions + sample_action_pos,
4877 	       sizeof(struct rte_flow_action));
4878 	index += 1;
4879 	actions_pre[index] = (struct rte_flow_action){
4880 		.type = (enum rte_flow_action_type)
4881 			RTE_FLOW_ACTION_TYPE_END,
4882 	};
4883 	/* Put the actions after sample into Suffix flow. */
4884 	memcpy(actions_sfx, actions + sample_action_pos + 1,
4885 	       sizeof(struct rte_flow_action) *
4886 	       (actions_n - sample_action_pos - 1));
4887 	return tag_id;
4888 }
4889 
4890 /**
4891  * The splitting for metadata feature.
4892  *
4893  * - Q/RSS action on NIC Rx should be split in order to pass by
4894  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
4895  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
4896  *
4897  * - All the actions on NIC Tx should have a mreg copy action to
4898  *   copy reg_a from WQE to reg_c[0].
4899  *
4900  * @param dev
4901  *   Pointer to Ethernet device.
4902  * @param[in] flow
4903  *   Parent flow structure pointer.
4904  * @param[in] attr
4905  *   Flow rule attributes.
4906  * @param[in] items
4907  *   Pattern specification (list terminated by the END pattern item).
4908  * @param[in] actions
4909  *   Associated actions (list terminated by the END action).
4910  * @param[in] flow_split_info
4911  *   Pointer to flow split info structure.
4912  * @param[out] error
4913  *   Perform verbose error reporting if not NULL.
4914  * @return
4915  *   0 on success, negative value otherwise
4916  */
4917 static int
4918 flow_create_split_metadata(struct rte_eth_dev *dev,
4919 			   struct rte_flow *flow,
4920 			   const struct rte_flow_attr *attr,
4921 			   const struct rte_flow_item items[],
4922 			   const struct rte_flow_action actions[],
4923 			   struct mlx5_flow_split_info *flow_split_info,
4924 			   struct rte_flow_error *error)
4925 {
4926 	struct mlx5_priv *priv = dev->data->dev_private;
4927 	struct mlx5_dev_config *config = &priv->config;
4928 	const struct rte_flow_action *qrss = NULL;
4929 	struct rte_flow_action *ext_actions = NULL;
4930 	struct mlx5_flow *dev_flow = NULL;
4931 	uint32_t qrss_id = 0;
4932 	int mtr_sfx = 0;
4933 	size_t act_size;
4934 	int actions_n;
4935 	int encap_idx;
4936 	int ret;
4937 
4938 	/* Check whether extensive metadata feature is engaged. */
4939 	if (!config->dv_flow_en ||
4940 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4941 	    !mlx5_flow_ext_mreg_supported(dev))
4942 		return flow_create_split_inner(dev, flow, NULL, attr, items,
4943 					       actions, flow_split_info, error);
4944 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
4945 							   &encap_idx);
4946 	if (qrss) {
4947 		/* Exclude hairpin flows from splitting. */
4948 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
4949 			const struct rte_flow_action_queue *queue;
4950 
4951 			queue = qrss->conf;
4952 			if (mlx5_rxq_get_type(dev, queue->index) ==
4953 			    MLX5_RXQ_TYPE_HAIRPIN)
4954 				qrss = NULL;
4955 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
4956 			const struct rte_flow_action_rss *rss;
4957 
4958 			rss = qrss->conf;
4959 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
4960 			    MLX5_RXQ_TYPE_HAIRPIN)
4961 				qrss = NULL;
4962 		}
4963 	}
4964 	if (qrss) {
4965 		/* Check if it is in meter suffix table. */
4966 		mtr_sfx = attr->group == (attr->transfer ?
4967 			  (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4968 			  MLX5_FLOW_TABLE_LEVEL_SUFFIX);
4969 		/*
4970 		 * Q/RSS action on NIC Rx should be split in order to pass by
4971 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
4972 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
4973 		 */
4974 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
4975 			   sizeof(struct rte_flow_action_set_tag) +
4976 			   sizeof(struct rte_flow_action_jump);
4977 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
4978 					  SOCKET_ID_ANY);
4979 		if (!ext_actions)
4980 			return rte_flow_error_set(error, ENOMEM,
4981 						  RTE_FLOW_ERROR_TYPE_ACTION,
4982 						  NULL, "no memory to split "
4983 						  "metadata flow");
4984 		/*
4985 		 * If we are the suffix flow of meter, tag already exist.
4986 		 * Set the tag action to void.
4987 		 */
4988 		if (mtr_sfx)
4989 			ext_actions[qrss - actions].type =
4990 						RTE_FLOW_ACTION_TYPE_VOID;
4991 		else
4992 			ext_actions[qrss - actions].type =
4993 						(enum rte_flow_action_type)
4994 						MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4995 		/*
4996 		 * Create the new actions list with removed Q/RSS action
4997 		 * and appended set tag and jump to register copy table
4998 		 * (RX_CP_TBL). We should preallocate unique tag ID here
4999 		 * in advance, because it is needed for set tag action.
5000 		 */
5001 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5002 						    qrss, actions_n, error);
5003 		if (!mtr_sfx && !qrss_id) {
5004 			ret = -rte_errno;
5005 			goto exit;
5006 		}
5007 	} else if (attr->egress && !attr->transfer) {
5008 		/*
5009 		 * All the actions on NIC Tx should have a metadata register
5010 		 * copy action to copy reg_a from WQE to reg_c[meta]
5011 		 */
5012 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5013 			   sizeof(struct mlx5_flow_action_copy_mreg);
5014 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5015 					  SOCKET_ID_ANY);
5016 		if (!ext_actions)
5017 			return rte_flow_error_set(error, ENOMEM,
5018 						  RTE_FLOW_ERROR_TYPE_ACTION,
5019 						  NULL, "no memory to split "
5020 						  "metadata flow");
5021 		/* Create the action list appended with copy register. */
5022 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5023 					     actions_n, error, encap_idx);
5024 		if (ret < 0)
5025 			goto exit;
5026 	}
5027 	/* Add the unmodified original or prefix subflow. */
5028 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5029 				      items, ext_actions ? ext_actions :
5030 				      actions, flow_split_info, error);
5031 	if (ret < 0)
5032 		goto exit;
5033 	MLX5_ASSERT(dev_flow);
5034 	if (qrss) {
5035 		const struct rte_flow_attr q_attr = {
5036 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5037 			.ingress = 1,
5038 		};
5039 		/* Internal PMD action to set register. */
5040 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5041 			.data = qrss_id,
5042 			.id = REG_NON,
5043 		};
5044 		struct rte_flow_item q_items[] = {
5045 			{
5046 				.type = (enum rte_flow_item_type)
5047 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5048 				.spec = &q_tag_spec,
5049 				.last = NULL,
5050 				.mask = NULL,
5051 			},
5052 			{
5053 				.type = RTE_FLOW_ITEM_TYPE_END,
5054 			},
5055 		};
5056 		struct rte_flow_action q_actions[] = {
5057 			{
5058 				.type = qrss->type,
5059 				.conf = qrss->conf,
5060 			},
5061 			{
5062 				.type = RTE_FLOW_ACTION_TYPE_END,
5063 			},
5064 		};
5065 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5066 
5067 		/*
5068 		 * Configure the tag item only if there is no meter subflow.
5069 		 * Since tag is already marked in the meter suffix subflow
5070 		 * we can just use the meter suffix items as is.
5071 		 */
5072 		if (qrss_id) {
5073 			/* Not meter subflow. */
5074 			MLX5_ASSERT(!mtr_sfx);
5075 			/*
5076 			 * Put unique id in prefix flow due to it is destroyed
5077 			 * after suffix flow and id will be freed after there
5078 			 * is no actual flows with this id and identifier
5079 			 * reallocation becomes possible (for example, for
5080 			 * other flows in other threads).
5081 			 */
5082 			dev_flow->handle->split_flow_id = qrss_id;
5083 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5084 						   error);
5085 			if (ret < 0)
5086 				goto exit;
5087 			q_tag_spec.id = ret;
5088 		}
5089 		dev_flow = NULL;
5090 		/* Add suffix subflow to execute Q/RSS. */
5091 		flow_split_info->prefix_layers = layers;
5092 		flow_split_info->prefix_mark = 0;
5093 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5094 					      &q_attr, mtr_sfx ? items :
5095 					      q_items, q_actions,
5096 					      flow_split_info, error);
5097 		if (ret < 0)
5098 			goto exit;
5099 		/* qrss ID should be freed if failed. */
5100 		qrss_id = 0;
5101 		MLX5_ASSERT(dev_flow);
5102 	}
5103 
5104 exit:
5105 	/*
5106 	 * We do not destroy the partially created sub_flows in case of error.
5107 	 * These ones are included into parent flow list and will be destroyed
5108 	 * by flow_drv_destroy.
5109 	 */
5110 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5111 			qrss_id);
5112 	mlx5_free(ext_actions);
5113 	return ret;
5114 }
5115 
5116 /**
5117  * The splitting for meter feature.
5118  *
5119  * - The meter flow will be split to two flows as prefix and
5120  *   suffix flow. The packets make sense only it pass the prefix
5121  *   meter action.
5122  *
5123  * - Reg_C_5 is used for the packet to match betweend prefix and
5124  *   suffix flow.
5125  *
5126  * @param dev
5127  *   Pointer to Ethernet device.
5128  * @param[in] flow
5129  *   Parent flow structure pointer.
5130  * @param[in] attr
5131  *   Flow rule attributes.
5132  * @param[in] items
5133  *   Pattern specification (list terminated by the END pattern item).
5134  * @param[in] actions
5135  *   Associated actions (list terminated by the END action).
5136  * @param[in] flow_split_info
5137  *   Pointer to flow split info structure.
5138  * @param[out] error
5139  *   Perform verbose error reporting if not NULL.
5140  * @return
5141  *   0 on success, negative value otherwise
5142  */
5143 static int
5144 flow_create_split_meter(struct rte_eth_dev *dev,
5145 			struct rte_flow *flow,
5146 			const struct rte_flow_attr *attr,
5147 			const struct rte_flow_item items[],
5148 			const struct rte_flow_action actions[],
5149 			struct mlx5_flow_split_info *flow_split_info,
5150 			struct rte_flow_error *error)
5151 {
5152 	struct mlx5_priv *priv = dev->data->dev_private;
5153 	struct rte_flow_action *sfx_actions = NULL;
5154 	struct rte_flow_action *pre_actions = NULL;
5155 	struct rte_flow_item *sfx_items = NULL;
5156 	struct mlx5_flow *dev_flow = NULL;
5157 	struct rte_flow_attr sfx_attr = *attr;
5158 	uint32_t mtr = 0;
5159 	uint32_t mtr_tag_id = 0;
5160 	size_t act_size;
5161 	size_t item_size;
5162 	int actions_n = 0;
5163 	int ret;
5164 
5165 	if (priv->mtr_en)
5166 		actions_n = flow_check_meter_action(actions, &mtr);
5167 	if (mtr) {
5168 		/* The five prefix actions: meter, decap, encap, tag, end. */
5169 		act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
5170 			   sizeof(struct mlx5_rte_flow_action_set_tag);
5171 		/* tag, vlan, port id, end. */
5172 #define METER_SUFFIX_ITEM 4
5173 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
5174 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5175 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
5176 					  0, SOCKET_ID_ANY);
5177 		if (!sfx_actions)
5178 			return rte_flow_error_set(error, ENOMEM,
5179 						  RTE_FLOW_ERROR_TYPE_ACTION,
5180 						  NULL, "no memory to split "
5181 						  "meter flow");
5182 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
5183 			     act_size);
5184 		pre_actions = sfx_actions + actions_n;
5185 		mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
5186 						   actions, sfx_actions,
5187 						   pre_actions);
5188 		if (!mtr_tag_id) {
5189 			ret = -rte_errno;
5190 			goto exit;
5191 		}
5192 		/* Add the prefix subflow. */
5193 		flow_split_info->prefix_mark = 0;
5194 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5195 					      attr, items, pre_actions,
5196 					      flow_split_info, error);
5197 		if (ret) {
5198 			ret = -rte_errno;
5199 			goto exit;
5200 		}
5201 		dev_flow->handle->split_flow_id = mtr_tag_id;
5202 		/* Setting the sfx group atrr. */
5203 		sfx_attr.group = sfx_attr.transfer ?
5204 				(MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
5205 				 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
5206 		flow_split_info->prefix_layers =
5207 				flow_get_prefix_layer_flags(dev_flow);
5208 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5209 	}
5210 	/* Add the prefix subflow. */
5211 	ret = flow_create_split_metadata(dev, flow,
5212 					 &sfx_attr, sfx_items ?
5213 					 sfx_items : items,
5214 					 sfx_actions ? sfx_actions : actions,
5215 					 flow_split_info, error);
5216 exit:
5217 	if (sfx_actions)
5218 		mlx5_free(sfx_actions);
5219 	return ret;
5220 }
5221 
5222 /**
5223  * The splitting for sample feature.
5224  *
5225  * Once Sample action is detected in the action list, the flow actions should
5226  * be split into prefix sub flow and suffix sub flow.
5227  *
5228  * The original items remain in the prefix sub flow, all actions preceding the
5229  * sample action and the sample action itself will be copied to the prefix
5230  * sub flow, the actions following the sample action will be copied to the
5231  * suffix sub flow, Queue action always be located in the suffix sub flow.
5232  *
5233  * In order to make the packet from prefix sub flow matches with suffix sub
5234  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
5235  * flow uses tag item with the unique flow id.
5236  *
5237  * @param dev
5238  *   Pointer to Ethernet device.
5239  * @param[in] flow
5240  *   Parent flow structure pointer.
5241  * @param[in] attr
5242  *   Flow rule attributes.
5243  * @param[in] items
5244  *   Pattern specification (list terminated by the END pattern item).
5245  * @param[in] actions
5246  *   Associated actions (list terminated by the END action).
5247  * @param[in] flow_split_info
5248  *   Pointer to flow split info structure.
5249  * @param[out] error
5250  *   Perform verbose error reporting if not NULL.
5251  * @return
5252  *   0 on success, negative value otherwise
5253  */
5254 static int
5255 flow_create_split_sample(struct rte_eth_dev *dev,
5256 			 struct rte_flow *flow,
5257 			 const struct rte_flow_attr *attr,
5258 			 const struct rte_flow_item items[],
5259 			 const struct rte_flow_action actions[],
5260 			 struct mlx5_flow_split_info *flow_split_info,
5261 			 struct rte_flow_error *error)
5262 {
5263 	struct mlx5_priv *priv = dev->data->dev_private;
5264 	struct rte_flow_action *sfx_actions = NULL;
5265 	struct rte_flow_action *pre_actions = NULL;
5266 	struct rte_flow_item *sfx_items = NULL;
5267 	struct mlx5_flow *dev_flow = NULL;
5268 	struct rte_flow_attr sfx_attr = *attr;
5269 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5270 	struct mlx5_flow_dv_sample_resource *sample_res;
5271 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
5272 	struct mlx5_flow_tbl_resource *sfx_tbl;
5273 	union mlx5_flow_tbl_key sfx_table_key;
5274 #endif
5275 	size_t act_size;
5276 	size_t item_size;
5277 	uint32_t fdb_tx = 0;
5278 	int32_t tag_id = 0;
5279 	int actions_n = 0;
5280 	int sample_action_pos;
5281 	int qrss_action_pos;
5282 	int ret = 0;
5283 
5284 	if (priv->sampler_en)
5285 		actions_n = flow_check_match_action(actions, attr,
5286 					RTE_FLOW_ACTION_TYPE_SAMPLE,
5287 					&sample_action_pos, &qrss_action_pos);
5288 	if (actions_n) {
5289 		/* The prefix actions must includes sample, tag, end. */
5290 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
5291 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
5292 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
5293 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
5294 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
5295 					  item_size), 0, SOCKET_ID_ANY);
5296 		if (!sfx_actions)
5297 			return rte_flow_error_set(error, ENOMEM,
5298 						  RTE_FLOW_ERROR_TYPE_ACTION,
5299 						  NULL, "no memory to split "
5300 						  "sample flow");
5301 		/* The representor_id is -1 for uplink. */
5302 		fdb_tx = (attr->transfer && priv->representor_id != -1);
5303 		if (!fdb_tx)
5304 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
5305 					+ act_size);
5306 		pre_actions = sfx_actions + actions_n;
5307 		tag_id = flow_sample_split_prep(dev, fdb_tx, sfx_items,
5308 						actions, sfx_actions,
5309 						pre_actions, actions_n,
5310 						sample_action_pos,
5311 						qrss_action_pos, error);
5312 		if (tag_id < 0 || (!fdb_tx && !tag_id)) {
5313 			ret = -rte_errno;
5314 			goto exit;
5315 		}
5316 		/* Add the prefix subflow. */
5317 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5318 					      items, pre_actions,
5319 					      flow_split_info, error);
5320 		if (ret) {
5321 			ret = -rte_errno;
5322 			goto exit;
5323 		}
5324 		dev_flow->handle->split_flow_id = tag_id;
5325 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5326 		/* Set the sfx group attr. */
5327 		sample_res = (struct mlx5_flow_dv_sample_resource *)
5328 					dev_flow->dv.sample_res;
5329 		sfx_tbl = (struct mlx5_flow_tbl_resource *)
5330 					sample_res->normal_path_tbl;
5331 		sfx_tbl_data = container_of(sfx_tbl,
5332 					struct mlx5_flow_tbl_data_entry, tbl);
5333 		sfx_table_key.v64 = sfx_tbl_data->entry.key;
5334 		sfx_attr.group = sfx_attr.transfer ?
5335 					(sfx_table_key.table_id - 1) :
5336 					 sfx_table_key.table_id;
5337 		flow_split_info->prefix_layers =
5338 				flow_get_prefix_layer_flags(dev_flow);
5339 		flow_split_info->prefix_mark = dev_flow->handle->mark;
5340 		/* Suffix group level already be scaled with factor, set
5341 		 * skip_scale to 1 to avoid scale again in translation.
5342 		 */
5343 		flow_split_info->skip_scale = 1;
5344 #endif
5345 	}
5346 	/* Add the suffix subflow. */
5347 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
5348 				      sfx_items ? sfx_items : items,
5349 				      sfx_actions ? sfx_actions : actions,
5350 				      flow_split_info, error);
5351 exit:
5352 	if (sfx_actions)
5353 		mlx5_free(sfx_actions);
5354 	return ret;
5355 }
5356 
5357 /**
5358  * Split the flow to subflow set. The splitters might be linked
5359  * in the chain, like this:
5360  * flow_create_split_outer() calls:
5361  *   flow_create_split_meter() calls:
5362  *     flow_create_split_metadata(meter_subflow_0) calls:
5363  *       flow_create_split_inner(metadata_subflow_0)
5364  *       flow_create_split_inner(metadata_subflow_1)
5365  *       flow_create_split_inner(metadata_subflow_2)
5366  *     flow_create_split_metadata(meter_subflow_1) calls:
5367  *       flow_create_split_inner(metadata_subflow_0)
5368  *       flow_create_split_inner(metadata_subflow_1)
5369  *       flow_create_split_inner(metadata_subflow_2)
5370  *
5371  * This provide flexible way to add new levels of flow splitting.
5372  * The all of successfully created subflows are included to the
5373  * parent flow dev_flow list.
5374  *
5375  * @param dev
5376  *   Pointer to Ethernet device.
5377  * @param[in] flow
5378  *   Parent flow structure pointer.
5379  * @param[in] attr
5380  *   Flow rule attributes.
5381  * @param[in] items
5382  *   Pattern specification (list terminated by the END pattern item).
5383  * @param[in] actions
5384  *   Associated actions (list terminated by the END action).
5385  * @param[in] flow_split_info
5386  *   Pointer to flow split info structure.
5387  * @param[out] error
5388  *   Perform verbose error reporting if not NULL.
5389  * @return
5390  *   0 on success, negative value otherwise
5391  */
5392 static int
5393 flow_create_split_outer(struct rte_eth_dev *dev,
5394 			struct rte_flow *flow,
5395 			const struct rte_flow_attr *attr,
5396 			const struct rte_flow_item items[],
5397 			const struct rte_flow_action actions[],
5398 			struct mlx5_flow_split_info *flow_split_info,
5399 			struct rte_flow_error *error)
5400 {
5401 	int ret;
5402 
5403 	ret = flow_create_split_sample(dev, flow, attr, items,
5404 				       actions, flow_split_info, error);
5405 	MLX5_ASSERT(ret <= 0);
5406 	return ret;
5407 }
5408 
5409 static struct mlx5_flow_tunnel *
5410 flow_tunnel_from_rule(struct rte_eth_dev *dev,
5411 		      const struct rte_flow_attr *attr,
5412 		      const struct rte_flow_item items[],
5413 		      const struct rte_flow_action actions[])
5414 {
5415 	struct mlx5_flow_tunnel *tunnel;
5416 
5417 #pragma GCC diagnostic push
5418 #pragma GCC diagnostic ignored "-Wcast-qual"
5419 	if (is_flow_tunnel_match_rule(dev, attr, items, actions))
5420 		tunnel = (struct mlx5_flow_tunnel *)items[0].spec;
5421 	else if (is_flow_tunnel_steer_rule(dev, attr, items, actions))
5422 		tunnel = (struct mlx5_flow_tunnel *)actions[0].conf;
5423 	else
5424 		tunnel = NULL;
5425 #pragma GCC diagnostic pop
5426 
5427 	return tunnel;
5428 }
5429 
5430 /**
5431  * Adjust flow RSS workspace if needed.
5432  *
5433  * @param wks
5434  *   Pointer to thread flow work space.
5435  * @param rss_desc
5436  *   Pointer to RSS descriptor.
5437  * @param[in] nrssq_num
5438  *   New RSS queue number.
5439  *
5440  * @return
5441  *   0 on success, -1 otherwise and rte_errno is set.
5442  */
5443 static int
5444 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
5445 			  struct mlx5_flow_rss_desc *rss_desc,
5446 			  uint32_t nrssq_num)
5447 {
5448 	if (likely(nrssq_num <= wks->rssq_num))
5449 		return 0;
5450 	rss_desc->queue = realloc(rss_desc->queue,
5451 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
5452 	if (!rss_desc->queue) {
5453 		rte_errno = ENOMEM;
5454 		return -1;
5455 	}
5456 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
5457 	return 0;
5458 }
5459 
5460 /**
5461  * Create a flow and add it to @p list.
5462  *
5463  * @param dev
5464  *   Pointer to Ethernet device.
5465  * @param list
5466  *   Pointer to a TAILQ flow list. If this parameter NULL,
5467  *   no list insertion occurred, flow is just created,
5468  *   this is caller's responsibility to track the
5469  *   created flow.
5470  * @param[in] attr
5471  *   Flow rule attributes.
5472  * @param[in] items
5473  *   Pattern specification (list terminated by the END pattern item).
5474  * @param[in] actions
5475  *   Associated actions (list terminated by the END action).
5476  * @param[in] external
5477  *   This flow rule is created by request external to PMD.
5478  * @param[out] error
5479  *   Perform verbose error reporting if not NULL.
5480  *
5481  * @return
5482  *   A flow index on success, 0 otherwise and rte_errno is set.
5483  */
5484 static uint32_t
5485 flow_list_create(struct rte_eth_dev *dev, uint32_t *list,
5486 		 const struct rte_flow_attr *attr,
5487 		 const struct rte_flow_item items[],
5488 		 const struct rte_flow_action original_actions[],
5489 		 bool external, struct rte_flow_error *error)
5490 {
5491 	struct mlx5_priv *priv = dev->data->dev_private;
5492 	struct rte_flow *flow = NULL;
5493 	struct mlx5_flow *dev_flow;
5494 	const struct rte_flow_action_rss *rss;
5495 	struct mlx5_translated_shared_action
5496 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5497 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5498 	union {
5499 		struct mlx5_flow_expand_rss buf;
5500 		uint8_t buffer[2048];
5501 	} expand_buffer;
5502 	union {
5503 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5504 		uint8_t buffer[2048];
5505 	} actions_rx;
5506 	union {
5507 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
5508 		uint8_t buffer[2048];
5509 	} actions_hairpin_tx;
5510 	union {
5511 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
5512 		uint8_t buffer[2048];
5513 	} items_tx;
5514 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
5515 	struct mlx5_flow_rss_desc *rss_desc;
5516 	const struct rte_flow_action *p_actions_rx;
5517 	uint32_t i;
5518 	uint32_t idx = 0;
5519 	int hairpin_flow;
5520 	struct rte_flow_attr attr_tx = { .priority = 0 };
5521 	const struct rte_flow_action *actions;
5522 	struct rte_flow_action *translated_actions = NULL;
5523 	struct mlx5_flow_tunnel *tunnel;
5524 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
5525 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
5526 	struct mlx5_flow_split_info flow_split_info = {
5527 		.external = !!external,
5528 		.skip_scale = 0,
5529 		.flow_idx = 0,
5530 		.prefix_mark = 0,
5531 		.prefix_layers = 0
5532 	};
5533 	int ret;
5534 
5535 	MLX5_ASSERT(wks);
5536 	rss_desc = &wks->rss_desc;
5537 	ret = flow_shared_actions_translate(dev, original_actions,
5538 					    shared_actions,
5539 					    &shared_actions_n,
5540 					    &translated_actions, error);
5541 	if (ret < 0) {
5542 		MLX5_ASSERT(translated_actions == NULL);
5543 		return 0;
5544 	}
5545 	actions = translated_actions ? translated_actions : original_actions;
5546 	p_actions_rx = actions;
5547 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5548 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
5549 				external, hairpin_flow, error);
5550 	if (ret < 0)
5551 		goto error_before_hairpin_split;
5552 	flow = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], &idx);
5553 	if (!flow) {
5554 		rte_errno = ENOMEM;
5555 		goto error_before_hairpin_split;
5556 	}
5557 	if (hairpin_flow > 0) {
5558 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
5559 			rte_errno = EINVAL;
5560 			goto error_before_hairpin_split;
5561 		}
5562 		flow_hairpin_split(dev, actions, actions_rx.actions,
5563 				   actions_hairpin_tx.actions, items_tx.items,
5564 				   idx);
5565 		p_actions_rx = actions_rx.actions;
5566 	}
5567 	flow_split_info.flow_idx = idx;
5568 	flow->drv_type = flow_get_drv_type(dev, attr);
5569 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
5570 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
5571 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
5572 	rss = flow_get_rss_action(p_actions_rx);
5573 	if (rss) {
5574 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
5575 			return 0;
5576 		/*
5577 		 * The following information is required by
5578 		 * mlx5_flow_hashfields_adjust() in advance.
5579 		 */
5580 		rss_desc->level = rss->level;
5581 		/* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
5582 		rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
5583 	}
5584 	flow->dev_handles = 0;
5585 	if (rss && rss->types) {
5586 		unsigned int graph_root;
5587 
5588 		graph_root = find_graph_root(items, rss->level);
5589 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
5590 					   items, rss->types,
5591 					   mlx5_support_expansion, graph_root);
5592 		MLX5_ASSERT(ret > 0 &&
5593 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
5594 	} else {
5595 		buf->entries = 1;
5596 		buf->entry[0].pattern = (void *)(uintptr_t)items;
5597 	}
5598 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, shared_actions,
5599 						      shared_actions_n);
5600 	for (i = 0; i < buf->entries; ++i) {
5601 		/* Initialize flow split data. */
5602 		flow_split_info.prefix_layers = 0;
5603 		flow_split_info.prefix_mark = 0;
5604 		flow_split_info.skip_scale = 0;
5605 		/*
5606 		 * The splitter may create multiple dev_flows,
5607 		 * depending on configuration. In the simplest
5608 		 * case it just creates unmodified original flow.
5609 		 */
5610 		ret = flow_create_split_outer(dev, flow, attr,
5611 					      buf->entry[i].pattern,
5612 					      p_actions_rx, &flow_split_info,
5613 					      error);
5614 		if (ret < 0)
5615 			goto error;
5616 		if (is_flow_tunnel_steer_rule(dev, attr,
5617 					      buf->entry[i].pattern,
5618 					      p_actions_rx)) {
5619 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
5620 							   p_actions_rx,
5621 							   idx,
5622 							   &default_miss_ctx,
5623 							   error);
5624 			if (ret < 0) {
5625 				mlx5_free(default_miss_ctx.queue);
5626 				goto error;
5627 			}
5628 		}
5629 	}
5630 	/* Create the tx flow. */
5631 	if (hairpin_flow) {
5632 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
5633 		attr_tx.ingress = 0;
5634 		attr_tx.egress = 1;
5635 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
5636 					 actions_hairpin_tx.actions,
5637 					 idx, error);
5638 		if (!dev_flow)
5639 			goto error;
5640 		dev_flow->flow = flow;
5641 		dev_flow->external = 0;
5642 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5643 			      dev_flow->handle, next);
5644 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
5645 					 items_tx.items,
5646 					 actions_hairpin_tx.actions, error);
5647 		if (ret < 0)
5648 			goto error;
5649 	}
5650 	/*
5651 	 * Update the metadata register copy table. If extensive
5652 	 * metadata feature is enabled and registers are supported
5653 	 * we might create the extra rte_flow for each unique
5654 	 * MARK/FLAG action ID.
5655 	 *
5656 	 * The table is updated for ingress Flows only, because
5657 	 * the egress Flows belong to the different device and
5658 	 * copy table should be updated in peer NIC Rx domain.
5659 	 */
5660 	if (attr->ingress &&
5661 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
5662 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
5663 		if (ret)
5664 			goto error;
5665 	}
5666 	/*
5667 	 * If the flow is external (from application) OR device is started,
5668 	 * OR mreg discover, then apply immediately.
5669 	 */
5670 	if (external || dev->data->dev_started ||
5671 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
5672 	     attr->priority == MLX5_FLOW_PRIO_RSVD)) {
5673 		ret = flow_drv_apply(dev, flow, error);
5674 		if (ret < 0)
5675 			goto error;
5676 	}
5677 	if (list) {
5678 		rte_spinlock_lock(&priv->flow_list_lock);
5679 		ILIST_INSERT(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list, idx,
5680 			     flow, next);
5681 		rte_spinlock_unlock(&priv->flow_list_lock);
5682 	}
5683 	flow_rxq_flags_set(dev, flow);
5684 	rte_free(translated_actions);
5685 	tunnel = flow_tunnel_from_rule(dev, attr, items, actions);
5686 	if (tunnel) {
5687 		flow->tunnel = 1;
5688 		flow->tunnel_id = tunnel->tunnel_id;
5689 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
5690 		mlx5_free(default_miss_ctx.queue);
5691 	}
5692 	mlx5_flow_pop_thread_workspace();
5693 	return idx;
5694 error:
5695 	MLX5_ASSERT(flow);
5696 	ret = rte_errno; /* Save rte_errno before cleanup. */
5697 	flow_mreg_del_copy_action(dev, flow);
5698 	flow_drv_destroy(dev, flow);
5699 	if (rss_desc->shared_rss)
5700 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
5701 			mlx5_ipool_get
5702 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
5703 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
5704 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], idx);
5705 	rte_errno = ret; /* Restore rte_errno. */
5706 	ret = rte_errno;
5707 	rte_errno = ret;
5708 	mlx5_flow_pop_thread_workspace();
5709 error_before_hairpin_split:
5710 	rte_free(translated_actions);
5711 	return 0;
5712 }
5713 
5714 /**
5715  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
5716  * incoming packets to table 1.
5717  *
5718  * Other flow rules, requested for group n, will be created in
5719  * e-switch table n+1.
5720  * Jump action to e-switch group n will be created to group n+1.
5721  *
5722  * Used when working in switchdev mode, to utilise advantages of table 1
5723  * and above.
5724  *
5725  * @param dev
5726  *   Pointer to Ethernet device.
5727  *
5728  * @return
5729  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
5730  */
5731 struct rte_flow *
5732 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
5733 {
5734 	const struct rte_flow_attr attr = {
5735 		.group = 0,
5736 		.priority = 0,
5737 		.ingress = 1,
5738 		.egress = 0,
5739 		.transfer = 1,
5740 	};
5741 	const struct rte_flow_item pattern = {
5742 		.type = RTE_FLOW_ITEM_TYPE_END,
5743 	};
5744 	struct rte_flow_action_jump jump = {
5745 		.group = 1,
5746 	};
5747 	const struct rte_flow_action actions[] = {
5748 		{
5749 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
5750 			.conf = &jump,
5751 		},
5752 		{
5753 			.type = RTE_FLOW_ACTION_TYPE_END,
5754 		},
5755 	};
5756 	struct mlx5_priv *priv = dev->data->dev_private;
5757 	struct rte_flow_error error;
5758 
5759 	return (void *)(uintptr_t)flow_list_create(dev, &priv->ctrl_flows,
5760 						   &attr, &pattern,
5761 						   actions, false, &error);
5762 }
5763 
5764 /**
5765  * Validate a flow supported by the NIC.
5766  *
5767  * @see rte_flow_validate()
5768  * @see rte_flow_ops
5769  */
5770 int
5771 mlx5_flow_validate(struct rte_eth_dev *dev,
5772 		   const struct rte_flow_attr *attr,
5773 		   const struct rte_flow_item items[],
5774 		   const struct rte_flow_action original_actions[],
5775 		   struct rte_flow_error *error)
5776 {
5777 	int hairpin_flow;
5778 	struct mlx5_translated_shared_action
5779 		shared_actions[MLX5_MAX_SHARED_ACTIONS];
5780 	int shared_actions_n = MLX5_MAX_SHARED_ACTIONS;
5781 	const struct rte_flow_action *actions;
5782 	struct rte_flow_action *translated_actions = NULL;
5783 	int ret = flow_shared_actions_translate(dev, original_actions,
5784 						shared_actions,
5785 						&shared_actions_n,
5786 						&translated_actions, error);
5787 
5788 	if (ret)
5789 		return ret;
5790 	actions = translated_actions ? translated_actions : original_actions;
5791 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
5792 	ret = flow_drv_validate(dev, attr, items, actions,
5793 				true, hairpin_flow, error);
5794 	rte_free(translated_actions);
5795 	return ret;
5796 }
5797 
5798 /**
5799  * Create a flow.
5800  *
5801  * @see rte_flow_create()
5802  * @see rte_flow_ops
5803  */
5804 struct rte_flow *
5805 mlx5_flow_create(struct rte_eth_dev *dev,
5806 		 const struct rte_flow_attr *attr,
5807 		 const struct rte_flow_item items[],
5808 		 const struct rte_flow_action actions[],
5809 		 struct rte_flow_error *error)
5810 {
5811 	struct mlx5_priv *priv = dev->data->dev_private;
5812 
5813 	/*
5814 	 * If the device is not started yet, it is not allowed to created a
5815 	 * flow from application. PMD default flows and traffic control flows
5816 	 * are not affected.
5817 	 */
5818 	if (unlikely(!dev->data->dev_started)) {
5819 		DRV_LOG(DEBUG, "port %u is not started when "
5820 			"inserting a flow", dev->data->port_id);
5821 		rte_flow_error_set(error, ENODEV,
5822 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5823 				   NULL,
5824 				   "port not started");
5825 		return NULL;
5826 	}
5827 
5828 	return (void *)(uintptr_t)flow_list_create(dev, &priv->flows,
5829 				  attr, items, actions, true, error);
5830 }
5831 
5832 /**
5833  * Destroy a flow in a list.
5834  *
5835  * @param dev
5836  *   Pointer to Ethernet device.
5837  * @param list
5838  *   Pointer to the Indexed flow list. If this parameter NULL,
5839  *   there is no flow removal from the list. Be noted that as
5840  *   flow is add to the indexed list, memory of the indexed
5841  *   list points to maybe changed as flow destroyed.
5842  * @param[in] flow_idx
5843  *   Index of flow to destroy.
5844  */
5845 static void
5846 flow_list_destroy(struct rte_eth_dev *dev, uint32_t *list,
5847 		  uint32_t flow_idx)
5848 {
5849 	struct mlx5_priv *priv = dev->data->dev_private;
5850 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
5851 					       [MLX5_IPOOL_RTE_FLOW], flow_idx);
5852 
5853 	if (!flow)
5854 		return;
5855 	/*
5856 	 * Update RX queue flags only if port is started, otherwise it is
5857 	 * already clean.
5858 	 */
5859 	if (dev->data->dev_started)
5860 		flow_rxq_flags_trim(dev, flow);
5861 	flow_drv_destroy(dev, flow);
5862 	if (list) {
5863 		rte_spinlock_lock(&priv->flow_list_lock);
5864 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], list,
5865 			     flow_idx, flow, next);
5866 		rte_spinlock_unlock(&priv->flow_list_lock);
5867 	}
5868 	if (flow->tunnel) {
5869 		struct mlx5_flow_tunnel *tunnel;
5870 
5871 		rte_spinlock_lock(&mlx5_tunnel_hub(dev)->sl);
5872 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
5873 		RTE_VERIFY(tunnel);
5874 		LIST_REMOVE(tunnel, chain);
5875 		rte_spinlock_unlock(&mlx5_tunnel_hub(dev)->sl);
5876 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
5877 			mlx5_flow_tunnel_free(dev, tunnel);
5878 	}
5879 	flow_mreg_del_copy_action(dev, flow);
5880 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], flow_idx);
5881 }
5882 
5883 /**
5884  * Destroy all flows.
5885  *
5886  * @param dev
5887  *   Pointer to Ethernet device.
5888  * @param list
5889  *   Pointer to the Indexed flow list.
5890  * @param active
5891  *   If flushing is called avtively.
5892  */
5893 void
5894 mlx5_flow_list_flush(struct rte_eth_dev *dev, uint32_t *list, bool active)
5895 {
5896 	uint32_t num_flushed = 0;
5897 
5898 	while (*list) {
5899 		flow_list_destroy(dev, list, *list);
5900 		num_flushed++;
5901 	}
5902 	if (active) {
5903 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
5904 			dev->data->port_id, num_flushed);
5905 	}
5906 }
5907 
5908 /**
5909  * Stop all default actions for flows.
5910  *
5911  * @param dev
5912  *   Pointer to Ethernet device.
5913  */
5914 void
5915 mlx5_flow_stop_default(struct rte_eth_dev *dev)
5916 {
5917 	flow_mreg_del_default_copy_action(dev);
5918 	flow_rxq_flags_clear(dev);
5919 }
5920 
5921 /**
5922  * Start all default actions for flows.
5923  *
5924  * @param dev
5925  *   Pointer to Ethernet device.
5926  * @return
5927  *   0 on success, a negative errno value otherwise and rte_errno is set.
5928  */
5929 int
5930 mlx5_flow_start_default(struct rte_eth_dev *dev)
5931 {
5932 	struct rte_flow_error error;
5933 
5934 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
5935 	return flow_mreg_add_default_copy_action(dev, &error);
5936 }
5937 
5938 /**
5939  * Release key of thread specific flow workspace data.
5940  */
5941 static void
5942 flow_release_workspace(void *data)
5943 {
5944 	struct mlx5_flow_workspace *wks = data;
5945 	struct mlx5_flow_workspace *next;
5946 
5947 	while (wks) {
5948 		next = wks->next;
5949 		free(wks->rss_desc.queue);
5950 		free(wks);
5951 		wks = next;
5952 	}
5953 }
5954 
5955 /**
5956  * Initialize key of thread specific flow workspace data.
5957  */
5958 static void
5959 flow_alloc_workspace(void)
5960 {
5961 	if (pthread_key_create(&key_workspace, flow_release_workspace))
5962 		DRV_LOG(ERR, "Can't create flow workspace data thread key.");
5963 }
5964 
5965 /**
5966  * Get thread specific current flow workspace.
5967  *
5968  * @return pointer to thread specific flow workspace data, NULL on error.
5969  */
5970 struct mlx5_flow_workspace*
5971 mlx5_flow_get_thread_workspace(void)
5972 {
5973 	struct mlx5_flow_workspace *data;
5974 
5975 	data = pthread_getspecific(key_workspace);
5976 	MLX5_ASSERT(data && data->inuse);
5977 	if (!data || !data->inuse)
5978 		DRV_LOG(ERR, "flow workspace not initialized.");
5979 	return data;
5980 }
5981 
5982 /**
5983  * Allocate and init new flow workspace.
5984  *
5985  * @return pointer to flow workspace data, NULL on error.
5986  */
5987 static struct mlx5_flow_workspace*
5988 flow_alloc_thread_workspace(void)
5989 {
5990 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
5991 
5992 	if (!data) {
5993 		DRV_LOG(ERR, "Failed to allocate flow workspace "
5994 			"memory.");
5995 		return NULL;
5996 	}
5997 	data->rss_desc.queue = calloc(1,
5998 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
5999 	if (!data->rss_desc.queue)
6000 		goto err;
6001 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
6002 	return data;
6003 err:
6004 	if (data->rss_desc.queue)
6005 		free(data->rss_desc.queue);
6006 	free(data);
6007 	return NULL;
6008 }
6009 
6010 /**
6011  * Get new thread specific flow workspace.
6012  *
6013  * If current workspace inuse, create new one and set as current.
6014  *
6015  * @return pointer to thread specific flow workspace data, NULL on error.
6016  */
6017 static struct mlx5_flow_workspace*
6018 mlx5_flow_push_thread_workspace(void)
6019 {
6020 	struct mlx5_flow_workspace *curr;
6021 	struct mlx5_flow_workspace *data;
6022 
6023 	if (pthread_once(&key_workspace_init, flow_alloc_workspace)) {
6024 		DRV_LOG(ERR, "Failed to init flow workspace data thread key.");
6025 		return NULL;
6026 	}
6027 	curr = pthread_getspecific(key_workspace);
6028 	if (!curr) {
6029 		data = flow_alloc_thread_workspace();
6030 		if (!data)
6031 			return NULL;
6032 	} else if (!curr->inuse) {
6033 		data = curr;
6034 	} else if (curr->next) {
6035 		data = curr->next;
6036 	} else {
6037 		data = flow_alloc_thread_workspace();
6038 		if (!data)
6039 			return NULL;
6040 		curr->next = data;
6041 		data->prev = curr;
6042 	}
6043 	data->inuse = 1;
6044 	data->flow_idx = 0;
6045 	/* Set as current workspace */
6046 	if (pthread_setspecific(key_workspace, data))
6047 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6048 	return data;
6049 }
6050 
6051 /**
6052  * Close current thread specific flow workspace.
6053  *
6054  * If previous workspace available, set it as current.
6055  *
6056  * @return pointer to thread specific flow workspace data, NULL on error.
6057  */
6058 static void
6059 mlx5_flow_pop_thread_workspace(void)
6060 {
6061 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
6062 
6063 	if (!data)
6064 		return;
6065 	if (!data->inuse) {
6066 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
6067 		return;
6068 	}
6069 	data->inuse = 0;
6070 	if (!data->prev)
6071 		return;
6072 	if (pthread_setspecific(key_workspace, data->prev))
6073 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
6074 }
6075 
6076 /**
6077  * Verify the flow list is empty
6078  *
6079  * @param dev
6080  *  Pointer to Ethernet device.
6081  *
6082  * @return the number of flows not released.
6083  */
6084 int
6085 mlx5_flow_verify(struct rte_eth_dev *dev)
6086 {
6087 	struct mlx5_priv *priv = dev->data->dev_private;
6088 	struct rte_flow *flow;
6089 	uint32_t idx;
6090 	int ret = 0;
6091 
6092 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW], priv->flows, idx,
6093 		      flow, next) {
6094 		DRV_LOG(DEBUG, "port %u flow %p still referenced",
6095 			dev->data->port_id, (void *)flow);
6096 		++ret;
6097 	}
6098 	return ret;
6099 }
6100 
6101 /**
6102  * Enable default hairpin egress flow.
6103  *
6104  * @param dev
6105  *   Pointer to Ethernet device.
6106  * @param queue
6107  *   The queue index.
6108  *
6109  * @return
6110  *   0 on success, a negative errno value otherwise and rte_errno is set.
6111  */
6112 int
6113 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
6114 			    uint32_t queue)
6115 {
6116 	struct mlx5_priv *priv = dev->data->dev_private;
6117 	const struct rte_flow_attr attr = {
6118 		.egress = 1,
6119 		.priority = 0,
6120 	};
6121 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
6122 		.queue = queue,
6123 	};
6124 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
6125 		.queue = UINT32_MAX,
6126 	};
6127 	struct rte_flow_item items[] = {
6128 		{
6129 			.type = (enum rte_flow_item_type)
6130 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6131 			.spec = &queue_spec,
6132 			.last = NULL,
6133 			.mask = &queue_mask,
6134 		},
6135 		{
6136 			.type = RTE_FLOW_ITEM_TYPE_END,
6137 		},
6138 	};
6139 	struct rte_flow_action_jump jump = {
6140 		.group = MLX5_HAIRPIN_TX_TABLE,
6141 	};
6142 	struct rte_flow_action actions[2];
6143 	uint32_t flow_idx;
6144 	struct rte_flow_error error;
6145 
6146 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
6147 	actions[0].conf = &jump;
6148 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
6149 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6150 				&attr, items, actions, false, &error);
6151 	if (!flow_idx) {
6152 		DRV_LOG(DEBUG,
6153 			"Failed to create ctrl flow: rte_errno(%d),"
6154 			" type(%d), message(%s)",
6155 			rte_errno, error.type,
6156 			error.message ? error.message : " (no stated reason)");
6157 		return -rte_errno;
6158 	}
6159 	return 0;
6160 }
6161 
6162 /**
6163  * Enable a control flow configured from the control plane.
6164  *
6165  * @param dev
6166  *   Pointer to Ethernet device.
6167  * @param eth_spec
6168  *   An Ethernet flow spec to apply.
6169  * @param eth_mask
6170  *   An Ethernet flow mask to apply.
6171  * @param vlan_spec
6172  *   A VLAN flow spec to apply.
6173  * @param vlan_mask
6174  *   A VLAN flow mask to apply.
6175  *
6176  * @return
6177  *   0 on success, a negative errno value otherwise and rte_errno is set.
6178  */
6179 int
6180 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
6181 		    struct rte_flow_item_eth *eth_spec,
6182 		    struct rte_flow_item_eth *eth_mask,
6183 		    struct rte_flow_item_vlan *vlan_spec,
6184 		    struct rte_flow_item_vlan *vlan_mask)
6185 {
6186 	struct mlx5_priv *priv = dev->data->dev_private;
6187 	const struct rte_flow_attr attr = {
6188 		.ingress = 1,
6189 		.priority = MLX5_FLOW_PRIO_RSVD,
6190 	};
6191 	struct rte_flow_item items[] = {
6192 		{
6193 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6194 			.spec = eth_spec,
6195 			.last = NULL,
6196 			.mask = eth_mask,
6197 		},
6198 		{
6199 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
6200 					      RTE_FLOW_ITEM_TYPE_END,
6201 			.spec = vlan_spec,
6202 			.last = NULL,
6203 			.mask = vlan_mask,
6204 		},
6205 		{
6206 			.type = RTE_FLOW_ITEM_TYPE_END,
6207 		},
6208 	};
6209 	uint16_t queue[priv->reta_idx_n];
6210 	struct rte_flow_action_rss action_rss = {
6211 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
6212 		.level = 0,
6213 		.types = priv->rss_conf.rss_hf,
6214 		.key_len = priv->rss_conf.rss_key_len,
6215 		.queue_num = priv->reta_idx_n,
6216 		.key = priv->rss_conf.rss_key,
6217 		.queue = queue,
6218 	};
6219 	struct rte_flow_action actions[] = {
6220 		{
6221 			.type = RTE_FLOW_ACTION_TYPE_RSS,
6222 			.conf = &action_rss,
6223 		},
6224 		{
6225 			.type = RTE_FLOW_ACTION_TYPE_END,
6226 		},
6227 	};
6228 	uint32_t flow_idx;
6229 	struct rte_flow_error error;
6230 	unsigned int i;
6231 
6232 	if (!priv->reta_idx_n || !priv->rxqs_n) {
6233 		return 0;
6234 	}
6235 	if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
6236 		action_rss.types = 0;
6237 	for (i = 0; i != priv->reta_idx_n; ++i)
6238 		queue[i] = (*priv->reta_idx)[i];
6239 	flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6240 				&attr, items, actions, false, &error);
6241 	if (!flow_idx)
6242 		return -rte_errno;
6243 	return 0;
6244 }
6245 
6246 /**
6247  * Enable a flow control configured from the control plane.
6248  *
6249  * @param dev
6250  *   Pointer to Ethernet device.
6251  * @param eth_spec
6252  *   An Ethernet flow spec to apply.
6253  * @param eth_mask
6254  *   An Ethernet flow mask to apply.
6255  *
6256  * @return
6257  *   0 on success, a negative errno value otherwise and rte_errno is set.
6258  */
6259 int
6260 mlx5_ctrl_flow(struct rte_eth_dev *dev,
6261 	       struct rte_flow_item_eth *eth_spec,
6262 	       struct rte_flow_item_eth *eth_mask)
6263 {
6264 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
6265 }
6266 
6267 /**
6268  * Create default miss flow rule matching lacp traffic
6269  *
6270  * @param dev
6271  *   Pointer to Ethernet device.
6272  * @param eth_spec
6273  *   An Ethernet flow spec to apply.
6274  *
6275  * @return
6276  *   0 on success, a negative errno value otherwise and rte_errno is set.
6277  */
6278 int
6279 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
6280 {
6281 	struct mlx5_priv *priv = dev->data->dev_private;
6282 	/*
6283 	 * The LACP matching is done by only using ether type since using
6284 	 * a multicast dst mac causes kernel to give low priority to this flow.
6285 	 */
6286 	static const struct rte_flow_item_eth lacp_spec = {
6287 		.type = RTE_BE16(0x8809),
6288 	};
6289 	static const struct rte_flow_item_eth lacp_mask = {
6290 		.type = 0xffff,
6291 	};
6292 	const struct rte_flow_attr attr = {
6293 		.ingress = 1,
6294 	};
6295 	struct rte_flow_item items[] = {
6296 		{
6297 			.type = RTE_FLOW_ITEM_TYPE_ETH,
6298 			.spec = &lacp_spec,
6299 			.mask = &lacp_mask,
6300 		},
6301 		{
6302 			.type = RTE_FLOW_ITEM_TYPE_END,
6303 		},
6304 	};
6305 	struct rte_flow_action actions[] = {
6306 		{
6307 			.type = (enum rte_flow_action_type)
6308 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
6309 		},
6310 		{
6311 			.type = RTE_FLOW_ACTION_TYPE_END,
6312 		},
6313 	};
6314 	struct rte_flow_error error;
6315 	uint32_t flow_idx = flow_list_create(dev, &priv->ctrl_flows,
6316 				&attr, items, actions, false, &error);
6317 
6318 	if (!flow_idx)
6319 		return -rte_errno;
6320 	return 0;
6321 }
6322 
6323 /**
6324  * Destroy a flow.
6325  *
6326  * @see rte_flow_destroy()
6327  * @see rte_flow_ops
6328  */
6329 int
6330 mlx5_flow_destroy(struct rte_eth_dev *dev,
6331 		  struct rte_flow *flow,
6332 		  struct rte_flow_error *error __rte_unused)
6333 {
6334 	struct mlx5_priv *priv = dev->data->dev_private;
6335 
6336 	flow_list_destroy(dev, &priv->flows, (uintptr_t)(void *)flow);
6337 	return 0;
6338 }
6339 
6340 /**
6341  * Destroy all flows.
6342  *
6343  * @see rte_flow_flush()
6344  * @see rte_flow_ops
6345  */
6346 int
6347 mlx5_flow_flush(struct rte_eth_dev *dev,
6348 		struct rte_flow_error *error __rte_unused)
6349 {
6350 	struct mlx5_priv *priv = dev->data->dev_private;
6351 
6352 	mlx5_flow_list_flush(dev, &priv->flows, false);
6353 	return 0;
6354 }
6355 
6356 /**
6357  * Isolated mode.
6358  *
6359  * @see rte_flow_isolate()
6360  * @see rte_flow_ops
6361  */
6362 int
6363 mlx5_flow_isolate(struct rte_eth_dev *dev,
6364 		  int enable,
6365 		  struct rte_flow_error *error)
6366 {
6367 	struct mlx5_priv *priv = dev->data->dev_private;
6368 
6369 	if (dev->data->dev_started) {
6370 		rte_flow_error_set(error, EBUSY,
6371 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6372 				   NULL,
6373 				   "port must be stopped first");
6374 		return -rte_errno;
6375 	}
6376 	priv->isolated = !!enable;
6377 	if (enable)
6378 		dev->dev_ops = &mlx5_os_dev_ops_isolate;
6379 	else
6380 		dev->dev_ops = &mlx5_os_dev_ops;
6381 
6382 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
6383 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
6384 
6385 	return 0;
6386 }
6387 
6388 /**
6389  * Query a flow.
6390  *
6391  * @see rte_flow_query()
6392  * @see rte_flow_ops
6393  */
6394 static int
6395 flow_drv_query(struct rte_eth_dev *dev,
6396 	       uint32_t flow_idx,
6397 	       const struct rte_flow_action *actions,
6398 	       void *data,
6399 	       struct rte_flow_error *error)
6400 {
6401 	struct mlx5_priv *priv = dev->data->dev_private;
6402 	const struct mlx5_flow_driver_ops *fops;
6403 	struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
6404 					       [MLX5_IPOOL_RTE_FLOW],
6405 					       flow_idx);
6406 	enum mlx5_flow_drv_type ftype;
6407 
6408 	if (!flow) {
6409 		return rte_flow_error_set(error, ENOENT,
6410 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6411 			  NULL,
6412 			  "invalid flow handle");
6413 	}
6414 	ftype = flow->drv_type;
6415 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
6416 	fops = flow_get_drv_ops(ftype);
6417 
6418 	return fops->query(dev, flow, actions, data, error);
6419 }
6420 
6421 /**
6422  * Query a flow.
6423  *
6424  * @see rte_flow_query()
6425  * @see rte_flow_ops
6426  */
6427 int
6428 mlx5_flow_query(struct rte_eth_dev *dev,
6429 		struct rte_flow *flow,
6430 		const struct rte_flow_action *actions,
6431 		void *data,
6432 		struct rte_flow_error *error)
6433 {
6434 	int ret;
6435 
6436 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
6437 			     error);
6438 	if (ret < 0)
6439 		return ret;
6440 	return 0;
6441 }
6442 
6443 /**
6444  * Manage filter operations.
6445  *
6446  * @param dev
6447  *   Pointer to Ethernet device structure.
6448  * @param filter_type
6449  *   Filter type.
6450  * @param filter_op
6451  *   Operation to perform.
6452  * @param arg
6453  *   Pointer to operation-specific structure.
6454  *
6455  * @return
6456  *   0 on success, a negative errno value otherwise and rte_errno is set.
6457  */
6458 int
6459 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
6460 		     enum rte_filter_type filter_type,
6461 		     enum rte_filter_op filter_op,
6462 		     void *arg)
6463 {
6464 	switch (filter_type) {
6465 	case RTE_ETH_FILTER_GENERIC:
6466 		if (filter_op != RTE_ETH_FILTER_GET) {
6467 			rte_errno = EINVAL;
6468 			return -rte_errno;
6469 		}
6470 		*(const void **)arg = &mlx5_flow_ops;
6471 		return 0;
6472 	default:
6473 		DRV_LOG(ERR, "port %u filter type (%d) not supported",
6474 			dev->data->port_id, filter_type);
6475 		rte_errno = ENOTSUP;
6476 		return -rte_errno;
6477 	}
6478 	return 0;
6479 }
6480 
6481 /**
6482  * Create the needed meter and suffix tables.
6483  *
6484  * @param[in] dev
6485  *   Pointer to Ethernet device.
6486  * @param[in] fm
6487  *   Pointer to the flow meter.
6488  *
6489  * @return
6490  *   Pointer to table set on success, NULL otherwise.
6491  */
6492 struct mlx5_meter_domains_infos *
6493 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
6494 			  const struct mlx5_flow_meter *fm)
6495 {
6496 	const struct mlx5_flow_driver_ops *fops;
6497 
6498 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6499 	return fops->create_mtr_tbls(dev, fm);
6500 }
6501 
6502 /**
6503  * Destroy the meter table set.
6504  *
6505  * @param[in] dev
6506  *   Pointer to Ethernet device.
6507  * @param[in] tbl
6508  *   Pointer to the meter table set.
6509  *
6510  * @return
6511  *   0 on success.
6512  */
6513 int
6514 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
6515 			   struct mlx5_meter_domains_infos *tbls)
6516 {
6517 	const struct mlx5_flow_driver_ops *fops;
6518 
6519 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6520 	return fops->destroy_mtr_tbls(dev, tbls);
6521 }
6522 
6523 /**
6524  * Create policer rules.
6525  *
6526  * @param[in] dev
6527  *   Pointer to Ethernet device.
6528  * @param[in] fm
6529  *   Pointer to flow meter structure.
6530  * @param[in] attr
6531  *   Pointer to flow attributes.
6532  *
6533  * @return
6534  *   0 on success, -1 otherwise.
6535  */
6536 int
6537 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
6538 			       struct mlx5_flow_meter *fm,
6539 			       const struct rte_flow_attr *attr)
6540 {
6541 	const struct mlx5_flow_driver_ops *fops;
6542 
6543 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6544 	return fops->create_policer_rules(dev, fm, attr);
6545 }
6546 
6547 /**
6548  * Destroy policer rules.
6549  *
6550  * @param[in] fm
6551  *   Pointer to flow meter structure.
6552  * @param[in] attr
6553  *   Pointer to flow attributes.
6554  *
6555  * @return
6556  *   0 on success, -1 otherwise.
6557  */
6558 int
6559 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
6560 				struct mlx5_flow_meter *fm,
6561 				const struct rte_flow_attr *attr)
6562 {
6563 	const struct mlx5_flow_driver_ops *fops;
6564 
6565 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6566 	return fops->destroy_policer_rules(dev, fm, attr);
6567 }
6568 
6569 /**
6570  * Allocate a counter.
6571  *
6572  * @param[in] dev
6573  *   Pointer to Ethernet device structure.
6574  *
6575  * @return
6576  *   Index to allocated counter  on success, 0 otherwise.
6577  */
6578 uint32_t
6579 mlx5_counter_alloc(struct rte_eth_dev *dev)
6580 {
6581 	const struct mlx5_flow_driver_ops *fops;
6582 	struct rte_flow_attr attr = { .transfer = 0 };
6583 
6584 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6585 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6586 		return fops->counter_alloc(dev);
6587 	}
6588 	DRV_LOG(ERR,
6589 		"port %u counter allocate is not supported.",
6590 		 dev->data->port_id);
6591 	return 0;
6592 }
6593 
6594 /**
6595  * Free a counter.
6596  *
6597  * @param[in] dev
6598  *   Pointer to Ethernet device structure.
6599  * @param[in] cnt
6600  *   Index to counter to be free.
6601  */
6602 void
6603 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
6604 {
6605 	const struct mlx5_flow_driver_ops *fops;
6606 	struct rte_flow_attr attr = { .transfer = 0 };
6607 
6608 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6609 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6610 		fops->counter_free(dev, cnt);
6611 		return;
6612 	}
6613 	DRV_LOG(ERR,
6614 		"port %u counter free is not supported.",
6615 		 dev->data->port_id);
6616 }
6617 
6618 /**
6619  * Query counter statistics.
6620  *
6621  * @param[in] dev
6622  *   Pointer to Ethernet device structure.
6623  * @param[in] cnt
6624  *   Index to counter to query.
6625  * @param[in] clear
6626  *   Set to clear counter statistics.
6627  * @param[out] pkts
6628  *   The counter hits packets number to save.
6629  * @param[out] bytes
6630  *   The counter hits bytes number to save.
6631  *
6632  * @return
6633  *   0 on success, a negative errno value otherwise.
6634  */
6635 int
6636 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
6637 		   bool clear, uint64_t *pkts, uint64_t *bytes)
6638 {
6639 	const struct mlx5_flow_driver_ops *fops;
6640 	struct rte_flow_attr attr = { .transfer = 0 };
6641 
6642 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
6643 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
6644 		return fops->counter_query(dev, cnt, clear, pkts, bytes);
6645 	}
6646 	DRV_LOG(ERR,
6647 		"port %u counter query is not supported.",
6648 		 dev->data->port_id);
6649 	return -ENOTSUP;
6650 }
6651 
6652 /**
6653  * Allocate a new memory for the counter values wrapped by all the needed
6654  * management.
6655  *
6656  * @param[in] sh
6657  *   Pointer to mlx5_dev_ctx_shared object.
6658  *
6659  * @return
6660  *   0 on success, a negative errno value otherwise.
6661  */
6662 static int
6663 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
6664 {
6665 	struct mlx5_devx_mkey_attr mkey_attr;
6666 	struct mlx5_counter_stats_mem_mng *mem_mng;
6667 	volatile struct flow_counter_stats *raw_data;
6668 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
6669 	int size = (sizeof(struct flow_counter_stats) *
6670 			MLX5_COUNTERS_PER_POOL +
6671 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
6672 			sizeof(struct mlx5_counter_stats_mem_mng);
6673 	size_t pgsize = rte_mem_page_size();
6674 	uint8_t *mem;
6675 	int i;
6676 
6677 	if (pgsize == (size_t)-1) {
6678 		DRV_LOG(ERR, "Failed to get mem page size");
6679 		rte_errno = ENOMEM;
6680 		return -ENOMEM;
6681 	}
6682 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
6683 	if (!mem) {
6684 		rte_errno = ENOMEM;
6685 		return -ENOMEM;
6686 	}
6687 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
6688 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
6689 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
6690 						 IBV_ACCESS_LOCAL_WRITE);
6691 	if (!mem_mng->umem) {
6692 		rte_errno = errno;
6693 		mlx5_free(mem);
6694 		return -rte_errno;
6695 	}
6696 	mkey_attr.addr = (uintptr_t)mem;
6697 	mkey_attr.size = size;
6698 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
6699 	mkey_attr.pd = sh->pdn;
6700 	mkey_attr.log_entity_size = 0;
6701 	mkey_attr.pg_access = 0;
6702 	mkey_attr.klm_array = NULL;
6703 	mkey_attr.klm_num = 0;
6704 	mkey_attr.relaxed_ordering_write = sh->cmng.relaxed_ordering_write;
6705 	mkey_attr.relaxed_ordering_read = sh->cmng.relaxed_ordering_read;
6706 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
6707 	if (!mem_mng->dm) {
6708 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
6709 		rte_errno = errno;
6710 		mlx5_free(mem);
6711 		return -rte_errno;
6712 	}
6713 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
6714 	raw_data = (volatile struct flow_counter_stats *)mem;
6715 	for (i = 0; i < raws_n; ++i) {
6716 		mem_mng->raws[i].mem_mng = mem_mng;
6717 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
6718 	}
6719 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
6720 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
6721 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
6722 				 next);
6723 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
6724 	sh->cmng.mem_mng = mem_mng;
6725 	return 0;
6726 }
6727 
6728 /**
6729  * Set the statistic memory to the new counter pool.
6730  *
6731  * @param[in] sh
6732  *   Pointer to mlx5_dev_ctx_shared object.
6733  * @param[in] pool
6734  *   Pointer to the pool to set the statistic memory.
6735  *
6736  * @return
6737  *   0 on success, a negative errno value otherwise.
6738  */
6739 static int
6740 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
6741 			       struct mlx5_flow_counter_pool *pool)
6742 {
6743 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6744 	/* Resize statistic memory once used out. */
6745 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
6746 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
6747 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
6748 		return -1;
6749 	}
6750 	rte_spinlock_lock(&pool->sl);
6751 	pool->raw = cmng->mem_mng->raws + pool->index %
6752 		    MLX5_CNT_CONTAINER_RESIZE;
6753 	rte_spinlock_unlock(&pool->sl);
6754 	pool->raw_hw = NULL;
6755 	return 0;
6756 }
6757 
6758 #define MLX5_POOL_QUERY_FREQ_US 1000000
6759 
6760 /**
6761  * Set the periodic procedure for triggering asynchronous batch queries for all
6762  * the counter pools.
6763  *
6764  * @param[in] sh
6765  *   Pointer to mlx5_dev_ctx_shared object.
6766  */
6767 void
6768 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
6769 {
6770 	uint32_t pools_n, us;
6771 
6772 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
6773 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
6774 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
6775 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
6776 		sh->cmng.query_thread_on = 0;
6777 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
6778 	} else {
6779 		sh->cmng.query_thread_on = 1;
6780 	}
6781 }
6782 
6783 /**
6784  * The periodic procedure for triggering asynchronous batch queries for all the
6785  * counter pools. This function is probably called by the host thread.
6786  *
6787  * @param[in] arg
6788  *   The parameter for the alarm process.
6789  */
6790 void
6791 mlx5_flow_query_alarm(void *arg)
6792 {
6793 	struct mlx5_dev_ctx_shared *sh = arg;
6794 	int ret;
6795 	uint16_t pool_index = sh->cmng.pool_index;
6796 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6797 	struct mlx5_flow_counter_pool *pool;
6798 	uint16_t n_valid;
6799 
6800 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
6801 		goto set_alarm;
6802 	rte_spinlock_lock(&cmng->pool_update_sl);
6803 	pool = cmng->pools[pool_index];
6804 	n_valid = cmng->n_valid;
6805 	rte_spinlock_unlock(&cmng->pool_update_sl);
6806 	/* Set the statistic memory to the new created pool. */
6807 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
6808 		goto set_alarm;
6809 	if (pool->raw_hw)
6810 		/* There is a pool query in progress. */
6811 		goto set_alarm;
6812 	pool->raw_hw =
6813 		LIST_FIRST(&sh->cmng.free_stat_raws);
6814 	if (!pool->raw_hw)
6815 		/* No free counter statistics raw memory. */
6816 		goto set_alarm;
6817 	/*
6818 	 * Identify the counters released between query trigger and query
6819 	 * handle more efficiently. The counter released in this gap period
6820 	 * should wait for a new round of query as the new arrived packets
6821 	 * will not be taken into account.
6822 	 */
6823 	pool->query_gen++;
6824 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
6825 					       MLX5_COUNTERS_PER_POOL,
6826 					       NULL, NULL,
6827 					       pool->raw_hw->mem_mng->dm->id,
6828 					       (void *)(uintptr_t)
6829 					       pool->raw_hw->data,
6830 					       sh->devx_comp,
6831 					       (uint64_t)(uintptr_t)pool);
6832 	if (ret) {
6833 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
6834 			" %d", pool->min_dcs->id);
6835 		pool->raw_hw = NULL;
6836 		goto set_alarm;
6837 	}
6838 	LIST_REMOVE(pool->raw_hw, next);
6839 	sh->cmng.pending_queries++;
6840 	pool_index++;
6841 	if (pool_index >= n_valid)
6842 		pool_index = 0;
6843 set_alarm:
6844 	sh->cmng.pool_index = pool_index;
6845 	mlx5_set_query_alarm(sh);
6846 }
6847 
6848 /**
6849  * Check and callback event for new aged flow in the counter pool
6850  *
6851  * @param[in] sh
6852  *   Pointer to mlx5_dev_ctx_shared object.
6853  * @param[in] pool
6854  *   Pointer to Current counter pool.
6855  */
6856 static void
6857 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
6858 		   struct mlx5_flow_counter_pool *pool)
6859 {
6860 	struct mlx5_priv *priv;
6861 	struct mlx5_flow_counter *cnt;
6862 	struct mlx5_age_info *age_info;
6863 	struct mlx5_age_param *age_param;
6864 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
6865 	struct mlx5_counter_stats_raw *prev = pool->raw;
6866 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
6867 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
6868 	uint16_t expected = AGE_CANDIDATE;
6869 	uint32_t i;
6870 
6871 	pool->time_of_last_age_check = curr_time;
6872 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
6873 		cnt = MLX5_POOL_GET_CNT(pool, i);
6874 		age_param = MLX5_CNT_TO_AGE(cnt);
6875 		if (__atomic_load_n(&age_param->state,
6876 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
6877 			continue;
6878 		if (cur->data[i].hits != prev->data[i].hits) {
6879 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
6880 					 __ATOMIC_RELAXED);
6881 			continue;
6882 		}
6883 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
6884 				       time_delta,
6885 				       __ATOMIC_RELAXED) <= age_param->timeout)
6886 			continue;
6887 		/**
6888 		 * Hold the lock first, or if between the
6889 		 * state AGE_TMOUT and tailq operation the
6890 		 * release happened, the release procedure
6891 		 * may delete a non-existent tailq node.
6892 		 */
6893 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
6894 		age_info = GET_PORT_AGE_INFO(priv);
6895 		rte_spinlock_lock(&age_info->aged_sl);
6896 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
6897 						AGE_TMOUT, false,
6898 						__ATOMIC_RELAXED,
6899 						__ATOMIC_RELAXED)) {
6900 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
6901 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
6902 		}
6903 		rte_spinlock_unlock(&age_info->aged_sl);
6904 	}
6905 	mlx5_age_event_prepare(sh);
6906 }
6907 
6908 /**
6909  * Handler for the HW respond about ready values from an asynchronous batch
6910  * query. This function is probably called by the host thread.
6911  *
6912  * @param[in] sh
6913  *   The pointer to the shared device context.
6914  * @param[in] async_id
6915  *   The Devx async ID.
6916  * @param[in] status
6917  *   The status of the completion.
6918  */
6919 void
6920 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
6921 				  uint64_t async_id, int status)
6922 {
6923 	struct mlx5_flow_counter_pool *pool =
6924 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
6925 	struct mlx5_counter_stats_raw *raw_to_free;
6926 	uint8_t query_gen = pool->query_gen ^ 1;
6927 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
6928 	enum mlx5_counter_type cnt_type =
6929 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
6930 				MLX5_COUNTER_TYPE_ORIGIN;
6931 
6932 	if (unlikely(status)) {
6933 		raw_to_free = pool->raw_hw;
6934 	} else {
6935 		raw_to_free = pool->raw;
6936 		if (pool->is_aged)
6937 			mlx5_flow_aging_check(sh, pool);
6938 		rte_spinlock_lock(&pool->sl);
6939 		pool->raw = pool->raw_hw;
6940 		rte_spinlock_unlock(&pool->sl);
6941 		/* Be sure the new raw counters data is updated in memory. */
6942 		rte_io_wmb();
6943 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
6944 			rte_spinlock_lock(&cmng->csl[cnt_type]);
6945 			TAILQ_CONCAT(&cmng->counters[cnt_type],
6946 				     &pool->counters[query_gen], next);
6947 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
6948 		}
6949 	}
6950 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
6951 	pool->raw_hw = NULL;
6952 	sh->cmng.pending_queries--;
6953 }
6954 
6955 static const struct mlx5_flow_tbl_data_entry  *
6956 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
6957 {
6958 	struct mlx5_priv *priv = dev->data->dev_private;
6959 	struct mlx5_dev_ctx_shared *sh = priv->sh;
6960 	struct mlx5_hlist_entry *he;
6961 	union tunnel_offload_mark mbits = { .val = mark };
6962 	union mlx5_flow_tbl_key table_key = {
6963 		{
6964 			.table_id = tunnel_id_to_flow_tbl(mbits.table_id),
6965 			.dummy = 0,
6966 			.domain = !!mbits.transfer,
6967 			.direction = 0,
6968 		}
6969 	};
6970 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, NULL);
6971 	return he ?
6972 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
6973 }
6974 
6975 static void
6976 mlx5_flow_tunnel_grp2tbl_remove_cb(struct mlx5_hlist *list,
6977 				   struct mlx5_hlist_entry *entry)
6978 {
6979 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6980 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
6981 
6982 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
6983 			tunnel_flow_tbl_to_id(tte->flow_table));
6984 	mlx5_free(tte);
6985 }
6986 
6987 static struct mlx5_hlist_entry *
6988 mlx5_flow_tunnel_grp2tbl_create_cb(struct mlx5_hlist *list,
6989 				   uint64_t key __rte_unused,
6990 				   void *ctx __rte_unused)
6991 {
6992 	struct mlx5_dev_ctx_shared *sh = list->ctx;
6993 	struct tunnel_tbl_entry *tte;
6994 
6995 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
6996 			  sizeof(*tte), 0,
6997 			  SOCKET_ID_ANY);
6998 	if (!tte)
6999 		goto err;
7000 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7001 			  &tte->flow_table);
7002 	if (tte->flow_table >= MLX5_MAX_TABLES) {
7003 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
7004 			tte->flow_table);
7005 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
7006 				tte->flow_table);
7007 		goto err;
7008 	} else if (!tte->flow_table) {
7009 		goto err;
7010 	}
7011 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
7012 	return &tte->hash;
7013 err:
7014 	if (tte)
7015 		mlx5_free(tte);
7016 	return NULL;
7017 }
7018 
7019 static uint32_t
7020 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
7021 				const struct mlx5_flow_tunnel *tunnel,
7022 				uint32_t group, uint32_t *table,
7023 				struct rte_flow_error *error)
7024 {
7025 	struct mlx5_hlist_entry *he;
7026 	struct tunnel_tbl_entry *tte;
7027 	union tunnel_tbl_key key = {
7028 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
7029 		.group = group
7030 	};
7031 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7032 	struct mlx5_hlist *group_hash;
7033 
7034 	group_hash = tunnel ? tunnel->groups : thub->groups;
7035 	he = mlx5_hlist_register(group_hash, key.val, NULL);
7036 	if (!he)
7037 		return rte_flow_error_set(error, EINVAL,
7038 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7039 					  NULL,
7040 					  "tunnel group index not supported");
7041 	tte = container_of(he, typeof(*tte), hash);
7042 	*table = tte->flow_table;
7043 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
7044 		dev->data->port_id, key.tunnel_id, group, *table);
7045 	return 0;
7046 }
7047 
7048 static int
7049 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
7050 		    struct flow_grp_info grp_info, struct rte_flow_error *error)
7051 {
7052 	if (grp_info.transfer && grp_info.external && grp_info.fdb_def_rule) {
7053 		if (group == UINT32_MAX)
7054 			return rte_flow_error_set
7055 						(error, EINVAL,
7056 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
7057 						 NULL,
7058 						 "group index not supported");
7059 		*table = group + 1;
7060 	} else {
7061 		*table = group;
7062 	}
7063 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
7064 	return 0;
7065 }
7066 
7067 /**
7068  * Translate the rte_flow group index to HW table value.
7069  *
7070  * If tunnel offload is disabled, all group ids converted to flow table
7071  * id using the standard method.
7072  * If tunnel offload is enabled, group id can be converted using the
7073  * standard or tunnel conversion method. Group conversion method
7074  * selection depends on flags in `grp_info` parameter:
7075  * - Internal (grp_info.external == 0) groups conversion uses the
7076  *   standard method.
7077  * - Group ids in JUMP action converted with the tunnel conversion.
7078  * - Group id in rule attribute conversion depends on a rule type and
7079  *   group id value:
7080  *   ** non zero group attributes converted with the tunnel method
7081  *   ** zero group attribute in non-tunnel rule is converted using the
7082  *      standard method - there's only one root table
7083  *   ** zero group attribute in steer tunnel rule is converted with the
7084  *      standard method - single root table
7085  *   ** zero group attribute in match tunnel rule is a special OvS
7086  *      case: that value is used for portability reasons. That group
7087  *      id is converted with the tunnel conversion method.
7088  *
7089  * @param[in] dev
7090  *   Port device
7091  * @param[in] tunnel
7092  *   PMD tunnel offload object
7093  * @param[in] group
7094  *   rte_flow group index value.
7095  * @param[out] table
7096  *   HW table value.
7097  * @param[in] grp_info
7098  *   flags used for conversion
7099  * @param[out] error
7100  *   Pointer to error structure.
7101  *
7102  * @return
7103  *   0 on success, a negative errno value otherwise and rte_errno is set.
7104  */
7105 int
7106 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
7107 			 const struct mlx5_flow_tunnel *tunnel,
7108 			 uint32_t group, uint32_t *table,
7109 			 struct flow_grp_info grp_info,
7110 			 struct rte_flow_error *error)
7111 {
7112 	int ret;
7113 	bool standard_translation;
7114 
7115 	if (!grp_info.skip_scale && grp_info.external &&
7116 	    group < MLX5_MAX_TABLES_EXTERNAL)
7117 		group *= MLX5_FLOW_TABLE_FACTOR;
7118 	if (is_tunnel_offload_active(dev)) {
7119 		standard_translation = !grp_info.external ||
7120 					grp_info.std_tbl_fix;
7121 	} else {
7122 		standard_translation = true;
7123 	}
7124 	DRV_LOG(DEBUG,
7125 		"port %u group=%#x transfer=%d external=%d fdb_def_rule=%d translate=%s",
7126 		dev->data->port_id, group, grp_info.transfer,
7127 		grp_info.external, grp_info.fdb_def_rule,
7128 		standard_translation ? "STANDARD" : "TUNNEL");
7129 	if (standard_translation)
7130 		ret = flow_group_to_table(dev->data->port_id, group, table,
7131 					  grp_info, error);
7132 	else
7133 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
7134 						      table, error);
7135 
7136 	return ret;
7137 }
7138 
7139 /**
7140  * Discover availability of metadata reg_c's.
7141  *
7142  * Iteratively use test flows to check availability.
7143  *
7144  * @param[in] dev
7145  *   Pointer to the Ethernet device structure.
7146  *
7147  * @return
7148  *   0 on success, a negative errno value otherwise and rte_errno is set.
7149  */
7150 int
7151 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
7152 {
7153 	struct mlx5_priv *priv = dev->data->dev_private;
7154 	struct mlx5_dev_config *config = &priv->config;
7155 	enum modify_reg idx;
7156 	int n = 0;
7157 
7158 	/* reg_c[0] and reg_c[1] are reserved. */
7159 	config->flow_mreg_c[n++] = REG_C_0;
7160 	config->flow_mreg_c[n++] = REG_C_1;
7161 	/* Discover availability of other reg_c's. */
7162 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
7163 		struct rte_flow_attr attr = {
7164 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
7165 			.priority = MLX5_FLOW_PRIO_RSVD,
7166 			.ingress = 1,
7167 		};
7168 		struct rte_flow_item items[] = {
7169 			[0] = {
7170 				.type = RTE_FLOW_ITEM_TYPE_END,
7171 			},
7172 		};
7173 		struct rte_flow_action actions[] = {
7174 			[0] = {
7175 				.type = (enum rte_flow_action_type)
7176 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
7177 				.conf = &(struct mlx5_flow_action_copy_mreg){
7178 					.src = REG_C_1,
7179 					.dst = idx,
7180 				},
7181 			},
7182 			[1] = {
7183 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
7184 				.conf = &(struct rte_flow_action_jump){
7185 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
7186 				},
7187 			},
7188 			[2] = {
7189 				.type = RTE_FLOW_ACTION_TYPE_END,
7190 			},
7191 		};
7192 		uint32_t flow_idx;
7193 		struct rte_flow *flow;
7194 		struct rte_flow_error error;
7195 
7196 		if (!config->dv_flow_en)
7197 			break;
7198 		/* Create internal flow, validation skips copy action. */
7199 		flow_idx = flow_list_create(dev, NULL, &attr, items,
7200 					    actions, false, &error);
7201 		flow = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
7202 				      flow_idx);
7203 		if (!flow)
7204 			continue;
7205 		config->flow_mreg_c[n++] = idx;
7206 		flow_list_destroy(dev, NULL, flow_idx);
7207 	}
7208 	for (; n < MLX5_MREG_C_NUM; ++n)
7209 		config->flow_mreg_c[n] = REG_NON;
7210 	return 0;
7211 }
7212 
7213 /**
7214  * Dump flow raw hw data to file
7215  *
7216  * @param[in] dev
7217  *    The pointer to Ethernet device.
7218  * @param[in] file
7219  *   A pointer to a file for output.
7220  * @param[out] error
7221  *   Perform verbose error reporting if not NULL. PMDs initialize this
7222  *   structure in case of error only.
7223  * @return
7224  *   0 on success, a nagative value otherwise.
7225  */
7226 int
7227 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
7228 		   FILE *file,
7229 		   struct rte_flow_error *error __rte_unused)
7230 {
7231 	struct mlx5_priv *priv = dev->data->dev_private;
7232 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7233 
7234 	if (!priv->config.dv_flow_en) {
7235 		if (fputs("device dv flow disabled\n", file) <= 0)
7236 			return -errno;
7237 		return -ENOTSUP;
7238 	}
7239 	return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
7240 				       sh->tx_domain, file);
7241 }
7242 
7243 /**
7244  * Get aged-out flows.
7245  *
7246  * @param[in] dev
7247  *   Pointer to the Ethernet device structure.
7248  * @param[in] context
7249  *   The address of an array of pointers to the aged-out flows contexts.
7250  * @param[in] nb_countexts
7251  *   The length of context array pointers.
7252  * @param[out] error
7253  *   Perform verbose error reporting if not NULL. Initialized in case of
7254  *   error only.
7255  *
7256  * @return
7257  *   how many contexts get in success, otherwise negative errno value.
7258  *   if nb_contexts is 0, return the amount of all aged contexts.
7259  *   if nb_contexts is not 0 , return the amount of aged flows reported
7260  *   in the context array.
7261  */
7262 int
7263 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
7264 			uint32_t nb_contexts, struct rte_flow_error *error)
7265 {
7266 	const struct mlx5_flow_driver_ops *fops;
7267 	struct rte_flow_attr attr = { .transfer = 0 };
7268 
7269 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7270 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7271 		return fops->get_aged_flows(dev, contexts, nb_contexts,
7272 						    error);
7273 	}
7274 	DRV_LOG(ERR,
7275 		"port %u get aged flows is not supported.",
7276 		 dev->data->port_id);
7277 	return -ENOTSUP;
7278 }
7279 
7280 /* Wrapper for driver action_validate op callback */
7281 static int
7282 flow_drv_action_validate(struct rte_eth_dev *dev,
7283 			 const struct rte_flow_shared_action_conf *conf,
7284 			 const struct rte_flow_action *action,
7285 			 const struct mlx5_flow_driver_ops *fops,
7286 			 struct rte_flow_error *error)
7287 {
7288 	static const char err_msg[] = "shared action validation unsupported";
7289 
7290 	if (!fops->action_validate) {
7291 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7292 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7293 				   NULL, err_msg);
7294 		return -rte_errno;
7295 	}
7296 	return fops->action_validate(dev, conf, action, error);
7297 }
7298 
7299 /**
7300  * Destroys the shared action by handle.
7301  *
7302  * @param dev
7303  *   Pointer to Ethernet device structure.
7304  * @param[in] action
7305  *   Handle for the shared action to be destroyed.
7306  * @param[out] error
7307  *   Perform verbose error reporting if not NULL. PMDs initialize this
7308  *   structure in case of error only.
7309  *
7310  * @return
7311  *   0 on success, a negative errno value otherwise and rte_errno is set.
7312  *
7313  * @note: wrapper for driver action_create op callback.
7314  */
7315 static int
7316 mlx5_shared_action_destroy(struct rte_eth_dev *dev,
7317 			   struct rte_flow_shared_action *action,
7318 			   struct rte_flow_error *error)
7319 {
7320 	static const char err_msg[] = "shared action destruction unsupported";
7321 	struct rte_flow_attr attr = { .transfer = 0 };
7322 	const struct mlx5_flow_driver_ops *fops =
7323 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7324 
7325 	if (!fops->action_destroy) {
7326 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7327 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7328 				   NULL, err_msg);
7329 		return -rte_errno;
7330 	}
7331 	return fops->action_destroy(dev, action, error);
7332 }
7333 
7334 /* Wrapper for driver action_destroy op callback */
7335 static int
7336 flow_drv_action_update(struct rte_eth_dev *dev,
7337 		       struct rte_flow_shared_action *action,
7338 		       const void *action_conf,
7339 		       const struct mlx5_flow_driver_ops *fops,
7340 		       struct rte_flow_error *error)
7341 {
7342 	static const char err_msg[] = "shared action update unsupported";
7343 
7344 	if (!fops->action_update) {
7345 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7346 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7347 				   NULL, err_msg);
7348 		return -rte_errno;
7349 	}
7350 	return fops->action_update(dev, action, action_conf, error);
7351 }
7352 
7353 /* Wrapper for driver action_destroy op callback */
7354 static int
7355 flow_drv_action_query(struct rte_eth_dev *dev,
7356 		      const struct rte_flow_shared_action *action,
7357 		      void *data,
7358 		      const struct mlx5_flow_driver_ops *fops,
7359 		      struct rte_flow_error *error)
7360 {
7361 	static const char err_msg[] = "shared action query unsupported";
7362 
7363 	if (!fops->action_query) {
7364 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7365 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7366 				   NULL, err_msg);
7367 		return -rte_errno;
7368 	}
7369 	return fops->action_query(dev, action, data, error);
7370 }
7371 
7372 /**
7373  * Create shared action for reuse in multiple flow rules.
7374  *
7375  * @param dev
7376  *   Pointer to Ethernet device structure.
7377  * @param[in] action
7378  *   Action configuration for shared action creation.
7379  * @param[out] error
7380  *   Perform verbose error reporting if not NULL. PMDs initialize this
7381  *   structure in case of error only.
7382  * @return
7383  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
7384  */
7385 static struct rte_flow_shared_action *
7386 mlx5_shared_action_create(struct rte_eth_dev *dev,
7387 			  const struct rte_flow_shared_action_conf *conf,
7388 			  const struct rte_flow_action *action,
7389 			  struct rte_flow_error *error)
7390 {
7391 	static const char err_msg[] = "shared action creation unsupported";
7392 	struct rte_flow_attr attr = { .transfer = 0 };
7393 	const struct mlx5_flow_driver_ops *fops =
7394 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7395 
7396 	if (flow_drv_action_validate(dev, conf, action, fops, error))
7397 		return NULL;
7398 	if (!fops->action_create) {
7399 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
7400 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
7401 				   NULL, err_msg);
7402 		return NULL;
7403 	}
7404 	return fops->action_create(dev, conf, action, error);
7405 }
7406 
7407 /**
7408  * Updates inplace the shared action configuration pointed by *action* handle
7409  * with the configuration provided as *action* argument.
7410  * The update of the shared action configuration effects all flow rules reusing
7411  * the action via handle.
7412  *
7413  * @param dev
7414  *   Pointer to Ethernet device structure.
7415  * @param[in] shared_action
7416  *   Handle for the shared action to be updated.
7417  * @param[in] action
7418  *   Action specification used to modify the action pointed by handle.
7419  *   *action* should be of same type with the action pointed by the *action*
7420  *   handle argument, otherwise considered as invalid.
7421  * @param[out] error
7422  *   Perform verbose error reporting if not NULL. PMDs initialize this
7423  *   structure in case of error only.
7424  *
7425  * @return
7426  *   0 on success, a negative errno value otherwise and rte_errno is set.
7427  */
7428 static int
7429 mlx5_shared_action_update(struct rte_eth_dev *dev,
7430 		struct rte_flow_shared_action *shared_action,
7431 		const struct rte_flow_action *action,
7432 		struct rte_flow_error *error)
7433 {
7434 	struct rte_flow_attr attr = { .transfer = 0 };
7435 	const struct mlx5_flow_driver_ops *fops =
7436 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7437 	int ret;
7438 
7439 	ret = flow_drv_action_validate(dev, NULL, action, fops, error);
7440 	if (ret)
7441 		return ret;
7442 	return flow_drv_action_update(dev, shared_action, action->conf, fops,
7443 				      error);
7444 }
7445 
7446 /**
7447  * Query the shared action by handle.
7448  *
7449  * This function allows retrieving action-specific data such as counters.
7450  * Data is gathered by special action which may be present/referenced in
7451  * more than one flow rule definition.
7452  *
7453  * \see RTE_FLOW_ACTION_TYPE_COUNT
7454  *
7455  * @param dev
7456  *   Pointer to Ethernet device structure.
7457  * @param[in] action
7458  *   Handle for the shared action to query.
7459  * @param[in, out] data
7460  *   Pointer to storage for the associated query data type.
7461  * @param[out] error
7462  *   Perform verbose error reporting if not NULL. PMDs initialize this
7463  *   structure in case of error only.
7464  *
7465  * @return
7466  *   0 on success, a negative errno value otherwise and rte_errno is set.
7467  */
7468 static int
7469 mlx5_shared_action_query(struct rte_eth_dev *dev,
7470 			 const struct rte_flow_shared_action *action,
7471 			 void *data,
7472 			 struct rte_flow_error *error)
7473 {
7474 	struct rte_flow_attr attr = { .transfer = 0 };
7475 	const struct mlx5_flow_driver_ops *fops =
7476 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7477 
7478 	return flow_drv_action_query(dev, action, data, fops, error);
7479 }
7480 
7481 /**
7482  * Destroy all shared actions.
7483  *
7484  * @param dev
7485  *   Pointer to Ethernet device.
7486  *
7487  * @return
7488  *   0 on success, a negative errno value otherwise and rte_errno is set.
7489  */
7490 int
7491 mlx5_shared_action_flush(struct rte_eth_dev *dev)
7492 {
7493 	struct rte_flow_error error;
7494 	struct mlx5_priv *priv = dev->data->dev_private;
7495 	struct mlx5_shared_action_rss *action;
7496 	int ret = 0;
7497 	uint32_t idx;
7498 
7499 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
7500 		      priv->rss_shared_actions, idx, action, next) {
7501 		ret |= mlx5_shared_action_destroy(dev,
7502 		       (struct rte_flow_shared_action *)(uintptr_t)idx, &error);
7503 	}
7504 	return ret;
7505 }
7506 
7507 static void
7508 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
7509 		      struct mlx5_flow_tunnel *tunnel)
7510 {
7511 	struct mlx5_priv *priv = dev->data->dev_private;
7512 
7513 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
7514 		dev->data->port_id, tunnel->tunnel_id);
7515 	RTE_VERIFY(!__atomic_load_n(&tunnel->refctn, __ATOMIC_RELAXED));
7516 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID],
7517 			tunnel->tunnel_id);
7518 	mlx5_hlist_destroy(tunnel->groups);
7519 	mlx5_free(tunnel);
7520 }
7521 
7522 static struct mlx5_flow_tunnel *
7523 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
7524 {
7525 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7526 	struct mlx5_flow_tunnel *tun;
7527 
7528 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7529 		if (tun->tunnel_id == id)
7530 			break;
7531 	}
7532 
7533 	return tun;
7534 }
7535 
7536 static struct mlx5_flow_tunnel *
7537 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
7538 			  const struct rte_flow_tunnel *app_tunnel)
7539 {
7540 	struct mlx5_priv *priv = dev->data->dev_private;
7541 	struct mlx5_flow_tunnel *tunnel;
7542 	uint32_t id;
7543 
7544 	mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
7545 			  &id);
7546 	if (id >= MLX5_MAX_TUNNELS) {
7547 		mlx5_ipool_free(priv->sh->ipool
7548 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7549 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
7550 		return NULL;
7551 	} else if (!id) {
7552 		return NULL;
7553 	}
7554 	/**
7555 	 * mlx5 flow tunnel is an auxlilary data structure
7556 	 * It's not part of IO. No need to allocate it from
7557 	 * huge pages pools dedicated for IO
7558 	 */
7559 	tunnel = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*tunnel),
7560 			     0, SOCKET_ID_ANY);
7561 	if (!tunnel) {
7562 		mlx5_ipool_free(priv->sh->ipool
7563 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7564 		return NULL;
7565 	}
7566 	tunnel->groups = mlx5_hlist_create("tunnel groups", 1024, 0, 0,
7567 					   mlx5_flow_tunnel_grp2tbl_create_cb,
7568 					   NULL,
7569 					   mlx5_flow_tunnel_grp2tbl_remove_cb);
7570 	if (!tunnel->groups) {
7571 		mlx5_ipool_free(priv->sh->ipool
7572 				[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], id);
7573 		mlx5_free(tunnel);
7574 		return NULL;
7575 	}
7576 	tunnel->groups->ctx = priv->sh;
7577 	/* initiate new PMD tunnel */
7578 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
7579 	tunnel->tunnel_id = id;
7580 	tunnel->action.type = (typeof(tunnel->action.type))
7581 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
7582 	tunnel->action.conf = tunnel;
7583 	tunnel->item.type = (typeof(tunnel->item.type))
7584 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
7585 	tunnel->item.spec = tunnel;
7586 	tunnel->item.last = NULL;
7587 	tunnel->item.mask = NULL;
7588 
7589 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
7590 		dev->data->port_id, tunnel->tunnel_id);
7591 
7592 	return tunnel;
7593 }
7594 
7595 static int
7596 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
7597 		     const struct rte_flow_tunnel *app_tunnel,
7598 		     struct mlx5_flow_tunnel **tunnel)
7599 {
7600 	int ret;
7601 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
7602 	struct mlx5_flow_tunnel *tun;
7603 
7604 	rte_spinlock_lock(&thub->sl);
7605 	LIST_FOREACH(tun, &thub->tunnels, chain) {
7606 		if (!memcmp(app_tunnel, &tun->app_tunnel,
7607 			    sizeof(*app_tunnel))) {
7608 			*tunnel = tun;
7609 			ret = 0;
7610 			break;
7611 		}
7612 	}
7613 	if (!tun) {
7614 		tun = mlx5_flow_tunnel_allocate(dev, app_tunnel);
7615 		if (tun) {
7616 			LIST_INSERT_HEAD(&thub->tunnels, tun, chain);
7617 			*tunnel = tun;
7618 		} else {
7619 			ret = -ENOMEM;
7620 		}
7621 	}
7622 	rte_spinlock_unlock(&thub->sl);
7623 	if (tun)
7624 		__atomic_add_fetch(&tun->refctn, 1, __ATOMIC_RELAXED);
7625 
7626 	return ret;
7627 }
7628 
7629 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
7630 {
7631 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
7632 
7633 	if (!thub)
7634 		return;
7635 	if (!LIST_EMPTY(&thub->tunnels))
7636 		DRV_LOG(WARNING, "port %u tunnels present\n", port_id);
7637 	mlx5_hlist_destroy(thub->groups);
7638 	mlx5_free(thub);
7639 }
7640 
7641 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
7642 {
7643 	int err;
7644 	struct mlx5_flow_tunnel_hub *thub;
7645 
7646 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
7647 			   0, SOCKET_ID_ANY);
7648 	if (!thub)
7649 		return -ENOMEM;
7650 	LIST_INIT(&thub->tunnels);
7651 	rte_spinlock_init(&thub->sl);
7652 	thub->groups = mlx5_hlist_create("flow groups", MLX5_MAX_TABLES, 0,
7653 					 0, mlx5_flow_tunnel_grp2tbl_create_cb,
7654 					 NULL,
7655 					 mlx5_flow_tunnel_grp2tbl_remove_cb);
7656 	if (!thub->groups) {
7657 		err = -rte_errno;
7658 		goto err;
7659 	}
7660 	thub->groups->ctx = sh;
7661 	sh->tunnel_hub = thub;
7662 
7663 	return 0;
7664 
7665 err:
7666 	if (thub->groups)
7667 		mlx5_hlist_destroy(thub->groups);
7668 	if (thub)
7669 		mlx5_free(thub);
7670 	return err;
7671 }
7672 
7673 #ifndef HAVE_MLX5DV_DR
7674 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
7675 #else
7676 #define MLX5_DOMAIN_SYNC_FLOW \
7677 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
7678 #endif
7679 
7680 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
7681 {
7682 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
7683 	const struct mlx5_flow_driver_ops *fops;
7684 	int ret;
7685 	struct rte_flow_attr attr = { .transfer = 0 };
7686 
7687 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
7688 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
7689 	if (ret > 0)
7690 		ret = -ret;
7691 	return ret;
7692 }
7693