xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision 4fa1452bcaa452f7fa0214b78a2e0e2032deed40)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2016 6WIND S.A.
3  * Copyright 2016 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdbool.h>
10 #include <sys/queue.h>
11 
12 #include <rte_common.h>
13 #include <rte_ether.h>
14 #include <ethdev_driver.h>
15 #include <rte_eal_paging.h>
16 #include <rte_flow.h>
17 #include <rte_cycles.h>
18 #include <rte_flow_driver.h>
19 #include <rte_malloc.h>
20 #include <rte_ip.h>
21 
22 #include <mlx5_glue.h>
23 #include <mlx5_devx_cmds.h>
24 #include <mlx5_prm.h>
25 #include <mlx5_malloc.h>
26 
27 #include "mlx5_defs.h"
28 #include "mlx5.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_flow_os.h"
31 #include "mlx5_rx.h"
32 #include "mlx5_tx.h"
33 #include "mlx5_common_os.h"
34 #include "rte_pmd_mlx5.h"
35 
36 struct tunnel_default_miss_ctx {
37 	uint16_t *queue;
38 	__extension__
39 	union {
40 		struct rte_flow_action_rss action_rss;
41 		struct rte_flow_action_queue miss_queue;
42 		struct rte_flow_action_jump miss_jump;
43 		uint8_t raw[0];
44 	};
45 };
46 
47 static int
48 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
49 			     struct rte_flow *flow,
50 			     const struct rte_flow_attr *attr,
51 			     const struct rte_flow_action *app_actions,
52 			     uint32_t flow_idx,
53 			     const struct mlx5_flow_tunnel *tunnel,
54 			     struct tunnel_default_miss_ctx *ctx,
55 			     struct rte_flow_error *error);
56 static struct mlx5_flow_tunnel *
57 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id);
58 static void
59 mlx5_flow_tunnel_free(struct rte_eth_dev *dev, struct mlx5_flow_tunnel *tunnel);
60 static uint32_t
61 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
62 				const struct mlx5_flow_tunnel *tunnel,
63 				uint32_t group, uint32_t *table,
64 				struct rte_flow_error *error);
65 
66 static struct mlx5_flow_workspace *mlx5_flow_push_thread_workspace(void);
67 static void mlx5_flow_pop_thread_workspace(void);
68 
69 
70 /** Device flow drivers. */
71 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
72 
73 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
74 
75 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
76 	[MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
77 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_INFINIBAND_VERBS_H)
78 	[MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
79 	[MLX5_FLOW_TYPE_HW] = &mlx5_flow_hw_drv_ops,
80 #endif
81 	[MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
82 	[MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
83 };
84 
85 /** Helper macro to build input graph for mlx5_flow_expand_rss(). */
86 #define MLX5_FLOW_EXPAND_RSS_NEXT(...) \
87 	(const int []){ \
88 		__VA_ARGS__, 0, \
89 	}
90 
91 /** Node object of input graph for mlx5_flow_expand_rss(). */
92 struct mlx5_flow_expand_node {
93 	const int *const next;
94 	/**<
95 	 * List of next node indexes. Index 0 is interpreted as a terminator.
96 	 */
97 	const enum rte_flow_item_type type;
98 	/**< Pattern item type of current node. */
99 	uint64_t rss_types;
100 	/**<
101 	 * RSS types bit-field associated with this node
102 	 * (see RTE_ETH_RSS_* definitions).
103 	 */
104 	uint64_t node_flags;
105 	/**<
106 	 *  Bit-fields that define how the node is used in the expansion.
107 	 * (see MLX5_EXPANSION_NODE_* definitions).
108 	 */
109 };
110 
111 /* Optional expand field. The expansion alg will not go deeper. */
112 #define MLX5_EXPANSION_NODE_OPTIONAL (UINT64_C(1) << 0)
113 
114 /* The node is not added implicitly as expansion to the flow pattern.
115  * If the node type does not match the flow pattern item type, the
116  * expansion alg will go deeper to its next items.
117  * In the current implementation, the list of next nodes indexes can
118  * have up to one node with this flag set and it has to be the last
119  * node index (before the list terminator).
120  */
121 #define MLX5_EXPANSION_NODE_EXPLICIT (UINT64_C(1) << 1)
122 
123 /** Object returned by mlx5_flow_expand_rss(). */
124 struct mlx5_flow_expand_rss {
125 	uint32_t entries;
126 	/**< Number of entries @p patterns and @p priorities. */
127 	struct {
128 		struct rte_flow_item *pattern; /**< Expanded pattern array. */
129 		uint32_t priority; /**< Priority offset for each expansion. */
130 	} entry[];
131 };
132 
133 static void
134 mlx5_dbg__print_pattern(const struct rte_flow_item *item);
135 
136 static const struct mlx5_flow_expand_node *
137 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
138 		unsigned int item_idx,
139 		const struct mlx5_flow_expand_node graph[],
140 		const struct mlx5_flow_expand_node *node);
141 
142 static bool
143 mlx5_flow_is_rss_expandable_item(const struct rte_flow_item *item)
144 {
145 	switch (item->type) {
146 	case RTE_FLOW_ITEM_TYPE_ETH:
147 	case RTE_FLOW_ITEM_TYPE_VLAN:
148 	case RTE_FLOW_ITEM_TYPE_IPV4:
149 	case RTE_FLOW_ITEM_TYPE_IPV6:
150 	case RTE_FLOW_ITEM_TYPE_UDP:
151 	case RTE_FLOW_ITEM_TYPE_TCP:
152 	case RTE_FLOW_ITEM_TYPE_VXLAN:
153 	case RTE_FLOW_ITEM_TYPE_NVGRE:
154 	case RTE_FLOW_ITEM_TYPE_GRE:
155 	case RTE_FLOW_ITEM_TYPE_GENEVE:
156 	case RTE_FLOW_ITEM_TYPE_MPLS:
157 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
158 	case RTE_FLOW_ITEM_TYPE_GRE_KEY:
159 	case RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT:
160 	case RTE_FLOW_ITEM_TYPE_GTP:
161 		return true;
162 	default:
163 		break;
164 	}
165 	return false;
166 }
167 
168 /**
169  * Network Service Header (NSH) and its next protocol values
170  * are described in RFC-8393.
171  */
172 static enum rte_flow_item_type
173 mlx5_nsh_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
174 {
175 	enum rte_flow_item_type type;
176 
177 	switch (proto_mask & proto_spec) {
178 	case 0:
179 		type = RTE_FLOW_ITEM_TYPE_VOID;
180 		break;
181 	case RTE_VXLAN_GPE_TYPE_IPV4:
182 		type = RTE_FLOW_ITEM_TYPE_IPV4;
183 		break;
184 	case RTE_VXLAN_GPE_TYPE_IPV6:
185 		type = RTE_VXLAN_GPE_TYPE_IPV6;
186 		break;
187 	case RTE_VXLAN_GPE_TYPE_ETH:
188 		type = RTE_FLOW_ITEM_TYPE_ETH;
189 		break;
190 	default:
191 		type = RTE_FLOW_ITEM_TYPE_END;
192 	}
193 	return type;
194 }
195 
196 static enum rte_flow_item_type
197 mlx5_inet_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
198 {
199 	enum rte_flow_item_type type;
200 
201 	switch (proto_mask & proto_spec) {
202 	case 0:
203 		type = RTE_FLOW_ITEM_TYPE_VOID;
204 		break;
205 	case IPPROTO_UDP:
206 		type = RTE_FLOW_ITEM_TYPE_UDP;
207 		break;
208 	case IPPROTO_TCP:
209 		type = RTE_FLOW_ITEM_TYPE_TCP;
210 		break;
211 	case IPPROTO_IPIP:
212 		type = RTE_FLOW_ITEM_TYPE_IPV4;
213 		break;
214 	case IPPROTO_IPV6:
215 		type = RTE_FLOW_ITEM_TYPE_IPV6;
216 		break;
217 	default:
218 		type = RTE_FLOW_ITEM_TYPE_END;
219 	}
220 	return type;
221 }
222 
223 static enum rte_flow_item_type
224 mlx5_ethertype_to_item_type(rte_be16_t type_spec,
225 			    rte_be16_t type_mask, bool is_tunnel)
226 {
227 	enum rte_flow_item_type type;
228 
229 	switch (rte_be_to_cpu_16(type_spec & type_mask)) {
230 	case 0:
231 		type = RTE_FLOW_ITEM_TYPE_VOID;
232 		break;
233 	case RTE_ETHER_TYPE_TEB:
234 		type = is_tunnel ?
235 		       RTE_FLOW_ITEM_TYPE_ETH : RTE_FLOW_ITEM_TYPE_END;
236 		break;
237 	case RTE_ETHER_TYPE_VLAN:
238 		type = !is_tunnel ?
239 		       RTE_FLOW_ITEM_TYPE_VLAN : RTE_FLOW_ITEM_TYPE_END;
240 		break;
241 	case RTE_ETHER_TYPE_IPV4:
242 		type = RTE_FLOW_ITEM_TYPE_IPV4;
243 		break;
244 	case RTE_ETHER_TYPE_IPV6:
245 		type = RTE_FLOW_ITEM_TYPE_IPV6;
246 		break;
247 	default:
248 		type = RTE_FLOW_ITEM_TYPE_END;
249 	}
250 	return type;
251 }
252 
253 static enum rte_flow_item_type
254 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
255 {
256 #define MLX5_XSET_ITEM_MASK_SPEC(type, fld)                              \
257 	do {                                                             \
258 		const void *m = item->mask;                              \
259 		const void *s = item->spec;                              \
260 		mask = m ?                                               \
261 			((const struct rte_flow_item_##type *)m)->fld :  \
262 			rte_flow_item_##type##_mask.fld;                 \
263 		spec = ((const struct rte_flow_item_##type *)s)->fld;    \
264 	} while (0)
265 
266 	enum rte_flow_item_type ret;
267 	uint16_t spec, mask;
268 
269 	if (item == NULL || item->spec == NULL)
270 		return RTE_FLOW_ITEM_TYPE_VOID;
271 	switch (item->type) {
272 	case RTE_FLOW_ITEM_TYPE_ETH:
273 		MLX5_XSET_ITEM_MASK_SPEC(eth, type);
274 		if (!mask)
275 			return RTE_FLOW_ITEM_TYPE_VOID;
276 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
277 		break;
278 	case RTE_FLOW_ITEM_TYPE_VLAN:
279 		MLX5_XSET_ITEM_MASK_SPEC(vlan, inner_type);
280 		if (!mask)
281 			return RTE_FLOW_ITEM_TYPE_VOID;
282 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
283 		break;
284 	case RTE_FLOW_ITEM_TYPE_IPV4:
285 		MLX5_XSET_ITEM_MASK_SPEC(ipv4, hdr.next_proto_id);
286 		if (!mask)
287 			return RTE_FLOW_ITEM_TYPE_VOID;
288 		ret = mlx5_inet_proto_to_item_type(spec, mask);
289 		break;
290 	case RTE_FLOW_ITEM_TYPE_IPV6:
291 		MLX5_XSET_ITEM_MASK_SPEC(ipv6, hdr.proto);
292 		if (!mask)
293 			return RTE_FLOW_ITEM_TYPE_VOID;
294 		ret = mlx5_inet_proto_to_item_type(spec, mask);
295 		break;
296 	case RTE_FLOW_ITEM_TYPE_GENEVE:
297 		MLX5_XSET_ITEM_MASK_SPEC(geneve, protocol);
298 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
299 		break;
300 	case RTE_FLOW_ITEM_TYPE_GRE:
301 		MLX5_XSET_ITEM_MASK_SPEC(gre, protocol);
302 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
303 		break;
304 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
305 		MLX5_XSET_ITEM_MASK_SPEC(vxlan_gpe, protocol);
306 		ret = mlx5_nsh_proto_to_item_type(spec, mask);
307 		break;
308 	default:
309 		ret = RTE_FLOW_ITEM_TYPE_VOID;
310 		break;
311 	}
312 	return ret;
313 #undef MLX5_XSET_ITEM_MASK_SPEC
314 }
315 
316 static const int *
317 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
318 		const int *next_node)
319 {
320 	const struct mlx5_flow_expand_node *node = NULL;
321 	const int *next = next_node;
322 
323 	while (next && *next) {
324 		/*
325 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
326 		 * flag set, because they were not found in the flow pattern.
327 		 */
328 		node = &graph[*next];
329 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
330 			break;
331 		next = node->next;
332 	}
333 	return next;
334 }
335 
336 #define MLX5_RSS_EXP_ELT_N 16
337 
338 /**
339  * Expand RSS flows into several possible flows according to the RSS hash
340  * fields requested and the driver capabilities.
341  *
342  * @param[out] buf
343  *   Buffer to store the result expansion.
344  * @param[in] size
345  *   Buffer size in bytes. If 0, @p buf can be NULL.
346  * @param[in] pattern
347  *   User flow pattern.
348  * @param[in] types
349  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
350  * @param[in] graph
351  *   Input graph to expand @p pattern according to @p types.
352  * @param[in] graph_root_index
353  *   Index of root node in @p graph, typically 0.
354  *
355  * @return
356  *   A positive value representing the size of @p buf in bytes regardless of
357  *   @p size on success, a negative errno value otherwise and rte_errno is
358  *   set, the following errors are defined:
359  *
360  *   -E2BIG: graph-depth @p graph is too deep.
361  *   -EINVAL: @p size has not enough space for expanded pattern.
362  */
363 static int
364 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
365 		     const struct rte_flow_item *pattern, uint64_t types,
366 		     const struct mlx5_flow_expand_node graph[],
367 		     int graph_root_index)
368 {
369 	const struct rte_flow_item *item;
370 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
371 	const int *next_node;
372 	const int *stack[MLX5_RSS_EXP_ELT_N];
373 	int stack_pos = 0;
374 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
375 	unsigned int i, item_idx, last_expand_item_idx = 0;
376 	size_t lsize;
377 	size_t user_pattern_size = 0;
378 	void *addr = NULL;
379 	const struct mlx5_flow_expand_node *next = NULL;
380 	struct rte_flow_item missed_item;
381 	int missed = 0;
382 	int elt = 0;
383 	const struct rte_flow_item *last_expand_item = NULL;
384 
385 	memset(&missed_item, 0, sizeof(missed_item));
386 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
387 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
388 	if (lsize > size)
389 		return -EINVAL;
390 	buf->entry[0].priority = 0;
391 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
392 	buf->entries = 0;
393 	addr = buf->entry[0].pattern;
394 	for (item = pattern, item_idx = 0;
395 			item->type != RTE_FLOW_ITEM_TYPE_END;
396 			item++, item_idx++) {
397 		if (!mlx5_flow_is_rss_expandable_item(item)) {
398 			user_pattern_size += sizeof(*item);
399 			continue;
400 		}
401 		last_expand_item = item;
402 		last_expand_item_idx = item_idx;
403 		i = 0;
404 		while (node->next && node->next[i]) {
405 			next = &graph[node->next[i]];
406 			if (next->type == item->type)
407 				break;
408 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
409 				node = next;
410 				i = 0;
411 			} else {
412 				++i;
413 			}
414 		}
415 		if (next)
416 			node = next;
417 		user_pattern_size += sizeof(*item);
418 	}
419 	user_pattern_size += sizeof(*item); /* Handle END item. */
420 	lsize += user_pattern_size;
421 	if (lsize > size)
422 		return -EINVAL;
423 	/* Copy the user pattern in the first entry of the buffer. */
424 	rte_memcpy(addr, pattern, user_pattern_size);
425 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
426 	buf->entries = 1;
427 	/* Start expanding. */
428 	memset(flow_items, 0, sizeof(flow_items));
429 	user_pattern_size -= sizeof(*item);
430 	/*
431 	 * Check if the last valid item has spec set, need complete pattern,
432 	 * and the pattern can be used for expansion.
433 	 */
434 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
435 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
436 		/* Item type END indicates expansion is not required. */
437 		return lsize;
438 	}
439 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
440 		next = NULL;
441 		missed = 1;
442 		i = 0;
443 		while (node->next && node->next[i]) {
444 			next = &graph[node->next[i]];
445 			if (next->type == missed_item.type) {
446 				flow_items[0].type = missed_item.type;
447 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
448 				break;
449 			}
450 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
451 				node = next;
452 				i = 0;
453 			} else {
454 				++i;
455 			}
456 			next = NULL;
457 		}
458 	}
459 	if (next && missed) {
460 		elt = 2; /* missed item + item end. */
461 		node = next;
462 		lsize += elt * sizeof(*item) + user_pattern_size;
463 		if (lsize > size)
464 			return -EINVAL;
465 		if (node->rss_types & types) {
466 			buf->entry[buf->entries].priority = 1;
467 			buf->entry[buf->entries].pattern = addr;
468 			buf->entries++;
469 			rte_memcpy(addr, buf->entry[0].pattern,
470 				   user_pattern_size);
471 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
472 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
473 			addr = (void *)(((uintptr_t)addr) +
474 					elt * sizeof(*item));
475 		}
476 	} else if (last_expand_item != NULL) {
477 		node = mlx5_flow_expand_rss_adjust_node(pattern,
478 				last_expand_item_idx, graph, node);
479 	}
480 	memset(flow_items, 0, sizeof(flow_items));
481 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
482 			node->next);
483 	stack[stack_pos] = next_node;
484 	node = next_node ? &graph[*next_node] : NULL;
485 	while (node) {
486 		flow_items[stack_pos].type = node->type;
487 		if (node->rss_types & types) {
488 			size_t n;
489 			/*
490 			 * compute the number of items to copy from the
491 			 * expansion and copy it.
492 			 * When the stack_pos is 0, there are 1 element in it,
493 			 * plus the addition END item.
494 			 */
495 			elt = stack_pos + 2;
496 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
497 			lsize += elt * sizeof(*item) + user_pattern_size;
498 			if (lsize > size)
499 				return -EINVAL;
500 			n = elt * sizeof(*item);
501 			buf->entry[buf->entries].priority =
502 				stack_pos + 1 + missed;
503 			buf->entry[buf->entries].pattern = addr;
504 			buf->entries++;
505 			rte_memcpy(addr, buf->entry[0].pattern,
506 				   user_pattern_size);
507 			addr = (void *)(((uintptr_t)addr) +
508 					user_pattern_size);
509 			rte_memcpy(addr, &missed_item,
510 				   missed * sizeof(*item));
511 			addr = (void *)(((uintptr_t)addr) +
512 				missed * sizeof(*item));
513 			rte_memcpy(addr, flow_items, n);
514 			addr = (void *)(((uintptr_t)addr) + n);
515 		}
516 		/* Go deeper. */
517 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
518 				node->next) {
519 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
520 					node->next);
521 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
522 				rte_errno = E2BIG;
523 				return -rte_errno;
524 			}
525 			stack[stack_pos] = next_node;
526 		} else if (*(next_node + 1)) {
527 			/* Follow up with the next possibility. */
528 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
529 					++next_node);
530 		} else if (!stack_pos) {
531 			/*
532 			 * Completing the traverse over the different paths.
533 			 * The next_node is advanced to the terminator.
534 			 */
535 			++next_node;
536 		} else {
537 			/* Move to the next path. */
538 			while (stack_pos) {
539 				next_node = stack[--stack_pos];
540 				next_node++;
541 				if (*next_node)
542 					break;
543 			}
544 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
545 					next_node);
546 			stack[stack_pos] = next_node;
547 		}
548 		node = next_node && *next_node ? &graph[*next_node] : NULL;
549 	};
550 	return lsize;
551 }
552 
553 enum mlx5_expansion {
554 	MLX5_EXPANSION_ROOT,
555 	MLX5_EXPANSION_ROOT_OUTER,
556 	MLX5_EXPANSION_OUTER_ETH,
557 	MLX5_EXPANSION_OUTER_VLAN,
558 	MLX5_EXPANSION_OUTER_IPV4,
559 	MLX5_EXPANSION_OUTER_IPV4_UDP,
560 	MLX5_EXPANSION_OUTER_IPV4_TCP,
561 	MLX5_EXPANSION_OUTER_IPV6,
562 	MLX5_EXPANSION_OUTER_IPV6_UDP,
563 	MLX5_EXPANSION_OUTER_IPV6_TCP,
564 	MLX5_EXPANSION_VXLAN,
565 	MLX5_EXPANSION_STD_VXLAN,
566 	MLX5_EXPANSION_L3_VXLAN,
567 	MLX5_EXPANSION_VXLAN_GPE,
568 	MLX5_EXPANSION_GRE,
569 	MLX5_EXPANSION_NVGRE,
570 	MLX5_EXPANSION_GRE_KEY,
571 	MLX5_EXPANSION_MPLS,
572 	MLX5_EXPANSION_ETH,
573 	MLX5_EXPANSION_VLAN,
574 	MLX5_EXPANSION_IPV4,
575 	MLX5_EXPANSION_IPV4_UDP,
576 	MLX5_EXPANSION_IPV4_TCP,
577 	MLX5_EXPANSION_IPV6,
578 	MLX5_EXPANSION_IPV6_UDP,
579 	MLX5_EXPANSION_IPV6_TCP,
580 	MLX5_EXPANSION_IPV6_FRAG_EXT,
581 	MLX5_EXPANSION_GTP,
582 	MLX5_EXPANSION_GENEVE,
583 };
584 
585 /** Supported expansion of items. */
586 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
587 	[MLX5_EXPANSION_ROOT] = {
588 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
589 						  MLX5_EXPANSION_IPV4,
590 						  MLX5_EXPANSION_IPV6),
591 		.type = RTE_FLOW_ITEM_TYPE_END,
592 	},
593 	[MLX5_EXPANSION_ROOT_OUTER] = {
594 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
595 						  MLX5_EXPANSION_OUTER_IPV4,
596 						  MLX5_EXPANSION_OUTER_IPV6),
597 		.type = RTE_FLOW_ITEM_TYPE_END,
598 	},
599 	[MLX5_EXPANSION_OUTER_ETH] = {
600 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
601 		.type = RTE_FLOW_ITEM_TYPE_ETH,
602 		.rss_types = 0,
603 	},
604 	[MLX5_EXPANSION_OUTER_VLAN] = {
605 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
606 						  MLX5_EXPANSION_OUTER_IPV6),
607 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
608 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
609 	},
610 	[MLX5_EXPANSION_OUTER_IPV4] = {
611 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
612 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
613 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
614 			 MLX5_EXPANSION_GRE,
615 			 MLX5_EXPANSION_NVGRE,
616 			 MLX5_EXPANSION_IPV4,
617 			 MLX5_EXPANSION_IPV6),
618 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
619 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
620 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
621 	},
622 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
623 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
624 						  MLX5_EXPANSION_VXLAN_GPE,
625 						  MLX5_EXPANSION_MPLS,
626 						  MLX5_EXPANSION_GENEVE,
627 						  MLX5_EXPANSION_GTP),
628 		.type = RTE_FLOW_ITEM_TYPE_UDP,
629 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
630 	},
631 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
632 		.type = RTE_FLOW_ITEM_TYPE_TCP,
633 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
634 	},
635 	[MLX5_EXPANSION_OUTER_IPV6] = {
636 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
637 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
638 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
639 			 MLX5_EXPANSION_IPV4,
640 			 MLX5_EXPANSION_IPV6,
641 			 MLX5_EXPANSION_GRE,
642 			 MLX5_EXPANSION_NVGRE),
643 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
644 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
645 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
646 	},
647 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
648 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
649 						  MLX5_EXPANSION_VXLAN_GPE,
650 						  MLX5_EXPANSION_MPLS,
651 						  MLX5_EXPANSION_GENEVE,
652 						  MLX5_EXPANSION_GTP),
653 		.type = RTE_FLOW_ITEM_TYPE_UDP,
654 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
655 	},
656 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
657 		.type = RTE_FLOW_ITEM_TYPE_TCP,
658 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
659 	},
660 	[MLX5_EXPANSION_VXLAN] = {
661 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
662 						  MLX5_EXPANSION_IPV4,
663 						  MLX5_EXPANSION_IPV6),
664 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
665 	},
666 	[MLX5_EXPANSION_STD_VXLAN] = {
667 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
668 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
669 	},
670 	[MLX5_EXPANSION_L3_VXLAN] = {
671 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
672 					MLX5_EXPANSION_IPV6),
673 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
674 	},
675 	[MLX5_EXPANSION_VXLAN_GPE] = {
676 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
677 						  MLX5_EXPANSION_IPV4,
678 						  MLX5_EXPANSION_IPV6),
679 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
680 	},
681 	[MLX5_EXPANSION_GRE] = {
682 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
683 						  MLX5_EXPANSION_IPV4,
684 						  MLX5_EXPANSION_IPV6,
685 						  MLX5_EXPANSION_GRE_KEY,
686 						  MLX5_EXPANSION_MPLS),
687 		.type = RTE_FLOW_ITEM_TYPE_GRE,
688 	},
689 	[MLX5_EXPANSION_GRE_KEY] = {
690 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
691 						  MLX5_EXPANSION_IPV6,
692 						  MLX5_EXPANSION_MPLS),
693 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
694 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
695 	},
696 	[MLX5_EXPANSION_NVGRE] = {
697 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
698 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
699 	},
700 	[MLX5_EXPANSION_MPLS] = {
701 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
702 						  MLX5_EXPANSION_IPV6,
703 						  MLX5_EXPANSION_ETH),
704 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
705 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
706 	},
707 	[MLX5_EXPANSION_ETH] = {
708 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
709 		.type = RTE_FLOW_ITEM_TYPE_ETH,
710 	},
711 	[MLX5_EXPANSION_VLAN] = {
712 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
713 						  MLX5_EXPANSION_IPV6),
714 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
715 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
716 	},
717 	[MLX5_EXPANSION_IPV4] = {
718 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
719 						  MLX5_EXPANSION_IPV4_TCP),
720 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
721 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
722 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
723 	},
724 	[MLX5_EXPANSION_IPV4_UDP] = {
725 		.type = RTE_FLOW_ITEM_TYPE_UDP,
726 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
727 	},
728 	[MLX5_EXPANSION_IPV4_TCP] = {
729 		.type = RTE_FLOW_ITEM_TYPE_TCP,
730 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
731 	},
732 	[MLX5_EXPANSION_IPV6] = {
733 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
734 						  MLX5_EXPANSION_IPV6_TCP,
735 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
736 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
737 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
738 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
739 	},
740 	[MLX5_EXPANSION_IPV6_UDP] = {
741 		.type = RTE_FLOW_ITEM_TYPE_UDP,
742 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
743 	},
744 	[MLX5_EXPANSION_IPV6_TCP] = {
745 		.type = RTE_FLOW_ITEM_TYPE_TCP,
746 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
747 	},
748 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
749 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
750 	},
751 	[MLX5_EXPANSION_GTP] = {
752 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
753 						  MLX5_EXPANSION_IPV6),
754 		.type = RTE_FLOW_ITEM_TYPE_GTP,
755 	},
756 	[MLX5_EXPANSION_GENEVE] = {
757 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
758 						  MLX5_EXPANSION_IPV4,
759 						  MLX5_EXPANSION_IPV6),
760 		.type = RTE_FLOW_ITEM_TYPE_GENEVE,
761 	},
762 };
763 
764 static struct rte_flow_action_handle *
765 mlx5_action_handle_create(struct rte_eth_dev *dev,
766 			  const struct rte_flow_indir_action_conf *conf,
767 			  const struct rte_flow_action *action,
768 			  struct rte_flow_error *error);
769 static int mlx5_action_handle_destroy
770 				(struct rte_eth_dev *dev,
771 				 struct rte_flow_action_handle *handle,
772 				 struct rte_flow_error *error);
773 static int mlx5_action_handle_update
774 				(struct rte_eth_dev *dev,
775 				 struct rte_flow_action_handle *handle,
776 				 const void *update,
777 				 struct rte_flow_error *error);
778 static int mlx5_action_handle_query
779 				(struct rte_eth_dev *dev,
780 				 const struct rte_flow_action_handle *handle,
781 				 void *data,
782 				 struct rte_flow_error *error);
783 static int
784 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
785 		    struct rte_flow_tunnel *app_tunnel,
786 		    struct rte_flow_action **actions,
787 		    uint32_t *num_of_actions,
788 		    struct rte_flow_error *error);
789 static int
790 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
791 		       struct rte_flow_tunnel *app_tunnel,
792 		       struct rte_flow_item **items,
793 		       uint32_t *num_of_items,
794 		       struct rte_flow_error *error);
795 static int
796 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
797 			      struct rte_flow_item *pmd_items,
798 			      uint32_t num_items, struct rte_flow_error *err);
799 static int
800 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
801 				struct rte_flow_action *pmd_actions,
802 				uint32_t num_actions,
803 				struct rte_flow_error *err);
804 static int
805 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
806 				  struct rte_mbuf *m,
807 				  struct rte_flow_restore_info *info,
808 				  struct rte_flow_error *err);
809 static struct rte_flow_item_flex_handle *
810 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
811 			   const struct rte_flow_item_flex_conf *conf,
812 			   struct rte_flow_error *error);
813 static int
814 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
815 			    const struct rte_flow_item_flex_handle *handle,
816 			    struct rte_flow_error *error);
817 static int
818 mlx5_flow_info_get(struct rte_eth_dev *dev,
819 		   struct rte_flow_port_info *port_info,
820 		   struct rte_flow_queue_info *queue_info,
821 		   struct rte_flow_error *error);
822 static int
823 mlx5_flow_port_configure(struct rte_eth_dev *dev,
824 			 const struct rte_flow_port_attr *port_attr,
825 			 uint16_t nb_queue,
826 			 const struct rte_flow_queue_attr *queue_attr[],
827 			 struct rte_flow_error *err);
828 
829 static struct rte_flow_pattern_template *
830 mlx5_flow_pattern_template_create(struct rte_eth_dev *dev,
831 		const struct rte_flow_pattern_template_attr *attr,
832 		const struct rte_flow_item items[],
833 		struct rte_flow_error *error);
834 
835 static int
836 mlx5_flow_pattern_template_destroy(struct rte_eth_dev *dev,
837 				   struct rte_flow_pattern_template *template,
838 				   struct rte_flow_error *error);
839 static struct rte_flow_actions_template *
840 mlx5_flow_actions_template_create(struct rte_eth_dev *dev,
841 			const struct rte_flow_actions_template_attr *attr,
842 			const struct rte_flow_action actions[],
843 			const struct rte_flow_action masks[],
844 			struct rte_flow_error *error);
845 static int
846 mlx5_flow_actions_template_destroy(struct rte_eth_dev *dev,
847 				   struct rte_flow_actions_template *template,
848 				   struct rte_flow_error *error);
849 
850 static struct rte_flow_template_table *
851 mlx5_flow_table_create(struct rte_eth_dev *dev,
852 		       const struct rte_flow_template_table_attr *attr,
853 		       struct rte_flow_pattern_template *item_templates[],
854 		       uint8_t nb_item_templates,
855 		       struct rte_flow_actions_template *action_templates[],
856 		       uint8_t nb_action_templates,
857 		       struct rte_flow_error *error);
858 static int
859 mlx5_flow_table_destroy(struct rte_eth_dev *dev,
860 			struct rte_flow_template_table *table,
861 			struct rte_flow_error *error);
862 static struct rte_flow *
863 mlx5_flow_async_flow_create(struct rte_eth_dev *dev,
864 			    uint32_t queue,
865 			    const struct rte_flow_op_attr *attr,
866 			    struct rte_flow_template_table *table,
867 			    const struct rte_flow_item items[],
868 			    uint8_t pattern_template_index,
869 			    const struct rte_flow_action actions[],
870 			    uint8_t action_template_index,
871 			    void *user_data,
872 			    struct rte_flow_error *error);
873 static int
874 mlx5_flow_async_flow_destroy(struct rte_eth_dev *dev,
875 			     uint32_t queue,
876 			     const struct rte_flow_op_attr *attr,
877 			     struct rte_flow *flow,
878 			     void *user_data,
879 			     struct rte_flow_error *error);
880 static int
881 mlx5_flow_pull(struct rte_eth_dev *dev,
882 	       uint32_t queue,
883 	       struct rte_flow_op_result res[],
884 	       uint16_t n_res,
885 	       struct rte_flow_error *error);
886 static int
887 mlx5_flow_push(struct rte_eth_dev *dev,
888 	       uint32_t queue,
889 	       struct rte_flow_error *error);
890 
891 static struct rte_flow_action_handle *
892 mlx5_flow_async_action_handle_create(struct rte_eth_dev *dev, uint32_t queue,
893 				 const struct rte_flow_op_attr *attr,
894 				 const struct rte_flow_indir_action_conf *conf,
895 				 const struct rte_flow_action *action,
896 				 void *user_data,
897 				 struct rte_flow_error *error);
898 
899 static int
900 mlx5_flow_async_action_handle_update(struct rte_eth_dev *dev, uint32_t queue,
901 				 const struct rte_flow_op_attr *attr,
902 				 struct rte_flow_action_handle *handle,
903 				 const void *update,
904 				 void *user_data,
905 				 struct rte_flow_error *error);
906 
907 static int
908 mlx5_flow_async_action_handle_destroy(struct rte_eth_dev *dev, uint32_t queue,
909 				  const struct rte_flow_op_attr *attr,
910 				  struct rte_flow_action_handle *handle,
911 				  void *user_data,
912 				  struct rte_flow_error *error);
913 
914 static const struct rte_flow_ops mlx5_flow_ops = {
915 	.validate = mlx5_flow_validate,
916 	.create = mlx5_flow_create,
917 	.destroy = mlx5_flow_destroy,
918 	.flush = mlx5_flow_flush,
919 	.isolate = mlx5_flow_isolate,
920 	.query = mlx5_flow_query,
921 	.dev_dump = mlx5_flow_dev_dump,
922 	.get_aged_flows = mlx5_flow_get_aged_flows,
923 	.action_handle_create = mlx5_action_handle_create,
924 	.action_handle_destroy = mlx5_action_handle_destroy,
925 	.action_handle_update = mlx5_action_handle_update,
926 	.action_handle_query = mlx5_action_handle_query,
927 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
928 	.tunnel_match = mlx5_flow_tunnel_match,
929 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
930 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
931 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
932 	.flex_item_create = mlx5_flow_flex_item_create,
933 	.flex_item_release = mlx5_flow_flex_item_release,
934 	.info_get = mlx5_flow_info_get,
935 	.configure = mlx5_flow_port_configure,
936 	.pattern_template_create = mlx5_flow_pattern_template_create,
937 	.pattern_template_destroy = mlx5_flow_pattern_template_destroy,
938 	.actions_template_create = mlx5_flow_actions_template_create,
939 	.actions_template_destroy = mlx5_flow_actions_template_destroy,
940 	.template_table_create = mlx5_flow_table_create,
941 	.template_table_destroy = mlx5_flow_table_destroy,
942 	.async_create = mlx5_flow_async_flow_create,
943 	.async_destroy = mlx5_flow_async_flow_destroy,
944 	.pull = mlx5_flow_pull,
945 	.push = mlx5_flow_push,
946 	.async_action_handle_create = mlx5_flow_async_action_handle_create,
947 	.async_action_handle_update = mlx5_flow_async_action_handle_update,
948 	.async_action_handle_destroy = mlx5_flow_async_action_handle_destroy,
949 };
950 
951 /* Tunnel information. */
952 struct mlx5_flow_tunnel_info {
953 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
954 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
955 };
956 
957 static struct mlx5_flow_tunnel_info tunnels_info[] = {
958 	{
959 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
960 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
961 	},
962 	{
963 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
964 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
965 	},
966 	{
967 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
968 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
969 	},
970 	{
971 		.tunnel = MLX5_FLOW_LAYER_GRE,
972 		.ptype = RTE_PTYPE_TUNNEL_GRE,
973 	},
974 	{
975 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
976 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
977 	},
978 	{
979 		.tunnel = MLX5_FLOW_LAYER_MPLS,
980 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
981 	},
982 	{
983 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
984 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
985 	},
986 	{
987 		.tunnel = MLX5_FLOW_LAYER_IPIP,
988 		.ptype = RTE_PTYPE_TUNNEL_IP,
989 	},
990 	{
991 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
992 		.ptype = RTE_PTYPE_TUNNEL_IP,
993 	},
994 	{
995 		.tunnel = MLX5_FLOW_LAYER_GTP,
996 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
997 	},
998 };
999 
1000 
1001 
1002 /**
1003  * Translate tag ID to register.
1004  *
1005  * @param[in] dev
1006  *   Pointer to the Ethernet device structure.
1007  * @param[in] feature
1008  *   The feature that request the register.
1009  * @param[in] id
1010  *   The request register ID.
1011  * @param[out] error
1012  *   Error description in case of any.
1013  *
1014  * @return
1015  *   The request register on success, a negative errno
1016  *   value otherwise and rte_errno is set.
1017  */
1018 int
1019 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
1020 		     enum mlx5_feature_name feature,
1021 		     uint32_t id,
1022 		     struct rte_flow_error *error)
1023 {
1024 	struct mlx5_priv *priv = dev->data->dev_private;
1025 	struct mlx5_sh_config *config = &priv->sh->config;
1026 	enum modify_reg start_reg;
1027 	bool skip_mtr_reg = false;
1028 
1029 	switch (feature) {
1030 	case MLX5_HAIRPIN_RX:
1031 		return REG_B;
1032 	case MLX5_HAIRPIN_TX:
1033 		return REG_A;
1034 	case MLX5_METADATA_RX:
1035 		switch (config->dv_xmeta_en) {
1036 		case MLX5_XMETA_MODE_LEGACY:
1037 			return REG_B;
1038 		case MLX5_XMETA_MODE_META16:
1039 			return REG_C_0;
1040 		case MLX5_XMETA_MODE_META32:
1041 			return REG_C_1;
1042 		}
1043 		break;
1044 	case MLX5_METADATA_TX:
1045 		return REG_A;
1046 	case MLX5_METADATA_FDB:
1047 		switch (config->dv_xmeta_en) {
1048 		case MLX5_XMETA_MODE_LEGACY:
1049 			return REG_NON;
1050 		case MLX5_XMETA_MODE_META16:
1051 			return REG_C_0;
1052 		case MLX5_XMETA_MODE_META32:
1053 			return REG_C_1;
1054 		}
1055 		break;
1056 	case MLX5_FLOW_MARK:
1057 		switch (config->dv_xmeta_en) {
1058 		case MLX5_XMETA_MODE_LEGACY:
1059 			return REG_NON;
1060 		case MLX5_XMETA_MODE_META16:
1061 			return REG_C_1;
1062 		case MLX5_XMETA_MODE_META32:
1063 			return REG_C_0;
1064 		}
1065 		break;
1066 	case MLX5_MTR_ID:
1067 		/*
1068 		 * If meter color and meter id share one register, flow match
1069 		 * should use the meter color register for match.
1070 		 */
1071 		if (priv->mtr_reg_share)
1072 			return priv->mtr_color_reg;
1073 		else
1074 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
1075 			       REG_C_3;
1076 	case MLX5_MTR_COLOR:
1077 	case MLX5_ASO_FLOW_HIT:
1078 	case MLX5_ASO_CONNTRACK:
1079 	case MLX5_SAMPLE_ID:
1080 		/* All features use the same REG_C. */
1081 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
1082 		return priv->mtr_color_reg;
1083 	case MLX5_COPY_MARK:
1084 		/*
1085 		 * Metadata COPY_MARK register using is in meter suffix sub
1086 		 * flow while with meter. It's safe to share the same register.
1087 		 */
1088 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
1089 	case MLX5_APP_TAG:
1090 		/*
1091 		 * If meter is enable, it will engage the register for color
1092 		 * match and flow match. If meter color match is not using the
1093 		 * REG_C_2, need to skip the REG_C_x be used by meter color
1094 		 * match.
1095 		 * If meter is disable, free to use all available registers.
1096 		 */
1097 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
1098 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
1099 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
1100 		if (id > (uint32_t)(REG_C_7 - start_reg))
1101 			return rte_flow_error_set(error, EINVAL,
1102 						  RTE_FLOW_ERROR_TYPE_ITEM,
1103 						  NULL, "invalid tag id");
1104 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
1105 			return rte_flow_error_set(error, ENOTSUP,
1106 						  RTE_FLOW_ERROR_TYPE_ITEM,
1107 						  NULL, "unsupported tag id");
1108 		/*
1109 		 * This case means meter is using the REG_C_x great than 2.
1110 		 * Take care not to conflict with meter color REG_C_x.
1111 		 * If the available index REG_C_y >= REG_C_x, skip the
1112 		 * color register.
1113 		 */
1114 		if (skip_mtr_reg && priv->sh->flow_mreg_c
1115 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
1116 			if (id >= (uint32_t)(REG_C_7 - start_reg))
1117 				return rte_flow_error_set(error, EINVAL,
1118 						       RTE_FLOW_ERROR_TYPE_ITEM,
1119 							NULL, "invalid tag id");
1120 			if (priv->sh->flow_mreg_c
1121 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
1122 				return priv->sh->flow_mreg_c
1123 					       [id + 1 + start_reg - REG_C_0];
1124 			return rte_flow_error_set(error, ENOTSUP,
1125 						  RTE_FLOW_ERROR_TYPE_ITEM,
1126 						  NULL, "unsupported tag id");
1127 		}
1128 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
1129 	}
1130 	MLX5_ASSERT(false);
1131 	return rte_flow_error_set(error, EINVAL,
1132 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1133 				  NULL, "invalid feature name");
1134 }
1135 
1136 /**
1137  * Check extensive flow metadata register support.
1138  *
1139  * @param dev
1140  *   Pointer to rte_eth_dev structure.
1141  *
1142  * @return
1143  *   True if device supports extensive flow metadata register, otherwise false.
1144  */
1145 bool
1146 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1147 {
1148 	struct mlx5_priv *priv = dev->data->dev_private;
1149 
1150 	/*
1151 	 * Having available reg_c can be regarded inclusively as supporting
1152 	 * extensive flow metadata register, which could mean,
1153 	 * - metadata register copy action by modify header.
1154 	 * - 16 modify header actions is supported.
1155 	 * - reg_c's are preserved across different domain (FDB and NIC) on
1156 	 *   packet loopback by flow lookup miss.
1157 	 */
1158 	return priv->sh->flow_mreg_c[2] != REG_NON;
1159 }
1160 
1161 /**
1162  * Get the lowest priority.
1163  *
1164  * @param[in] dev
1165  *   Pointer to the Ethernet device structure.
1166  * @param[in] attributes
1167  *   Pointer to device flow rule attributes.
1168  *
1169  * @return
1170  *   The value of lowest priority of flow.
1171  */
1172 uint32_t
1173 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1174 			  const struct rte_flow_attr *attr)
1175 {
1176 	struct mlx5_priv *priv = dev->data->dev_private;
1177 
1178 	if (!attr->group && !attr->transfer)
1179 		return priv->sh->flow_max_priority - 2;
1180 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1181 }
1182 
1183 /**
1184  * Calculate matcher priority of the flow.
1185  *
1186  * @param[in] dev
1187  *   Pointer to the Ethernet device structure.
1188  * @param[in] attr
1189  *   Pointer to device flow rule attributes.
1190  * @param[in] subpriority
1191  *   The priority based on the items.
1192  * @param[in] external
1193  *   Flow is user flow.
1194  * @return
1195  *   The matcher priority of the flow.
1196  */
1197 uint16_t
1198 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1199 			  const struct rte_flow_attr *attr,
1200 			  uint32_t subpriority, bool external)
1201 {
1202 	uint16_t priority = (uint16_t)attr->priority;
1203 	struct mlx5_priv *priv = dev->data->dev_private;
1204 
1205 	if (!attr->group && !attr->transfer) {
1206 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1207 			priority = priv->sh->flow_max_priority - 1;
1208 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1209 	} else if (!external && attr->transfer && attr->group == 0 &&
1210 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1211 		return (priv->sh->flow_max_priority - 1) * 3;
1212 	}
1213 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1214 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1215 	return priority * 3 + subpriority;
1216 }
1217 
1218 /**
1219  * Verify the @p item specifications (spec, last, mask) are compatible with the
1220  * NIC capabilities.
1221  *
1222  * @param[in] item
1223  *   Item specification.
1224  * @param[in] mask
1225  *   @p item->mask or flow default bit-masks.
1226  * @param[in] nic_mask
1227  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1228  * @param[in] size
1229  *   Bit-masks size in bytes.
1230  * @param[in] range_accepted
1231  *   True if range of values is accepted for specific fields, false otherwise.
1232  * @param[out] error
1233  *   Pointer to error structure.
1234  *
1235  * @return
1236  *   0 on success, a negative errno value otherwise and rte_errno is set.
1237  */
1238 int
1239 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1240 			  const uint8_t *mask,
1241 			  const uint8_t *nic_mask,
1242 			  unsigned int size,
1243 			  bool range_accepted,
1244 			  struct rte_flow_error *error)
1245 {
1246 	unsigned int i;
1247 
1248 	MLX5_ASSERT(nic_mask);
1249 	for (i = 0; i < size; ++i)
1250 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1251 			return rte_flow_error_set(error, ENOTSUP,
1252 						  RTE_FLOW_ERROR_TYPE_ITEM,
1253 						  item,
1254 						  "mask enables non supported"
1255 						  " bits");
1256 	if (!item->spec && (item->mask || item->last))
1257 		return rte_flow_error_set(error, EINVAL,
1258 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1259 					  "mask/last without a spec is not"
1260 					  " supported");
1261 	if (item->spec && item->last && !range_accepted) {
1262 		uint8_t spec[size];
1263 		uint8_t last[size];
1264 		unsigned int i;
1265 		int ret;
1266 
1267 		for (i = 0; i < size; ++i) {
1268 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1269 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1270 		}
1271 		ret = memcmp(spec, last, size);
1272 		if (ret != 0)
1273 			return rte_flow_error_set(error, EINVAL,
1274 						  RTE_FLOW_ERROR_TYPE_ITEM,
1275 						  item,
1276 						  "range is not valid");
1277 	}
1278 	return 0;
1279 }
1280 
1281 /**
1282  * Adjust the hash fields according to the @p flow information.
1283  *
1284  * @param[in] dev_flow.
1285  *   Pointer to the mlx5_flow.
1286  * @param[in] tunnel
1287  *   1 when the hash field is for a tunnel item.
1288  * @param[in] layer_types
1289  *   RTE_ETH_RSS_* types.
1290  * @param[in] hash_fields
1291  *   Item hash fields.
1292  *
1293  * @return
1294  *   The hash fields that should be used.
1295  */
1296 uint64_t
1297 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1298 			    int tunnel __rte_unused, uint64_t layer_types,
1299 			    uint64_t hash_fields)
1300 {
1301 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1302 	int rss_request_inner = rss_desc->level >= 2;
1303 
1304 	/* Check RSS hash level for tunnel. */
1305 	if (tunnel && rss_request_inner)
1306 		hash_fields |= IBV_RX_HASH_INNER;
1307 	else if (tunnel || rss_request_inner)
1308 		return 0;
1309 #endif
1310 	/* Check if requested layer matches RSS hash fields. */
1311 	if (!(rss_desc->types & layer_types))
1312 		return 0;
1313 	return hash_fields;
1314 }
1315 
1316 /**
1317  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1318  * if several tunnel rules are used on this queue, the tunnel ptype will be
1319  * cleared.
1320  *
1321  * @param rxq_ctrl
1322  *   Rx queue to update.
1323  */
1324 static void
1325 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1326 {
1327 	unsigned int i;
1328 	uint32_t tunnel_ptype = 0;
1329 
1330 	/* Look up for the ptype to use. */
1331 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1332 		if (!rxq_ctrl->flow_tunnels_n[i])
1333 			continue;
1334 		if (!tunnel_ptype) {
1335 			tunnel_ptype = tunnels_info[i].ptype;
1336 		} else {
1337 			tunnel_ptype = 0;
1338 			break;
1339 		}
1340 	}
1341 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1342 }
1343 
1344 /**
1345  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
1346  * flow.
1347  *
1348  * @param[in] dev
1349  *   Pointer to the Ethernet device structure.
1350  * @param[in] dev_handle
1351  *   Pointer to device flow handle structure.
1352  */
1353 void
1354 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1355 		       struct mlx5_flow_handle *dev_handle)
1356 {
1357 	struct mlx5_priv *priv = dev->data->dev_private;
1358 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1359 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1360 	unsigned int i;
1361 
1362 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1363 		struct mlx5_hrxq *hrxq;
1364 
1365 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1366 			      dev_handle->rix_hrxq);
1367 		if (hrxq)
1368 			ind_tbl = hrxq->ind_table;
1369 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1370 		struct mlx5_shared_action_rss *shared_rss;
1371 
1372 		shared_rss = mlx5_ipool_get
1373 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1374 			 dev_handle->rix_srss);
1375 		if (shared_rss)
1376 			ind_tbl = shared_rss->ind_tbl;
1377 	}
1378 	if (!ind_tbl)
1379 		return;
1380 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1381 		int idx = ind_tbl->queues[i];
1382 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1383 
1384 		MLX5_ASSERT(rxq_ctrl != NULL);
1385 		if (rxq_ctrl == NULL)
1386 			continue;
1387 		/*
1388 		 * To support metadata register copy on Tx loopback,
1389 		 * this must be always enabled (metadata may arive
1390 		 * from other port - not from local flows only.
1391 		 */
1392 		if (tunnel) {
1393 			unsigned int j;
1394 
1395 			/* Increase the counter matching the flow. */
1396 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1397 				if ((tunnels_info[j].tunnel &
1398 				     dev_handle->layers) ==
1399 				    tunnels_info[j].tunnel) {
1400 					rxq_ctrl->flow_tunnels_n[j]++;
1401 					break;
1402 				}
1403 			}
1404 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1405 		}
1406 	}
1407 }
1408 
1409 static void
1410 flow_rxq_mark_flag_set(struct rte_eth_dev *dev)
1411 {
1412 	struct mlx5_priv *priv = dev->data->dev_private;
1413 	struct mlx5_rxq_ctrl *rxq_ctrl;
1414 
1415 	if (priv->mark_enabled)
1416 		return;
1417 	LIST_FOREACH(rxq_ctrl, &priv->rxqsctrl, next) {
1418 		rxq_ctrl->rxq.mark = 1;
1419 	}
1420 	priv->mark_enabled = 1;
1421 }
1422 
1423 /**
1424  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1425  *
1426  * @param[in] dev
1427  *   Pointer to the Ethernet device structure.
1428  * @param[in] flow
1429  *   Pointer to flow structure.
1430  */
1431 static void
1432 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1433 {
1434 	struct mlx5_priv *priv = dev->data->dev_private;
1435 	uint32_t handle_idx;
1436 	struct mlx5_flow_handle *dev_handle;
1437 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1438 
1439 	MLX5_ASSERT(wks);
1440 	if (wks->mark)
1441 		flow_rxq_mark_flag_set(dev);
1442 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1443 		       handle_idx, dev_handle, next)
1444 		flow_drv_rxq_flags_set(dev, dev_handle);
1445 }
1446 
1447 /**
1448  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1449  * device flow if no other flow uses it with the same kind of request.
1450  *
1451  * @param dev
1452  *   Pointer to Ethernet device.
1453  * @param[in] dev_handle
1454  *   Pointer to the device flow handle structure.
1455  */
1456 static void
1457 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1458 			struct mlx5_flow_handle *dev_handle)
1459 {
1460 	struct mlx5_priv *priv = dev->data->dev_private;
1461 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1462 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1463 	unsigned int i;
1464 
1465 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1466 		struct mlx5_hrxq *hrxq;
1467 
1468 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1469 			      dev_handle->rix_hrxq);
1470 		if (hrxq)
1471 			ind_tbl = hrxq->ind_table;
1472 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1473 		struct mlx5_shared_action_rss *shared_rss;
1474 
1475 		shared_rss = mlx5_ipool_get
1476 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1477 			 dev_handle->rix_srss);
1478 		if (shared_rss)
1479 			ind_tbl = shared_rss->ind_tbl;
1480 	}
1481 	if (!ind_tbl)
1482 		return;
1483 	MLX5_ASSERT(dev->data->dev_started);
1484 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1485 		int idx = ind_tbl->queues[i];
1486 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1487 
1488 		MLX5_ASSERT(rxq_ctrl != NULL);
1489 		if (rxq_ctrl == NULL)
1490 			continue;
1491 		if (tunnel) {
1492 			unsigned int j;
1493 
1494 			/* Decrease the counter matching the flow. */
1495 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1496 				if ((tunnels_info[j].tunnel &
1497 				     dev_handle->layers) ==
1498 				    tunnels_info[j].tunnel) {
1499 					rxq_ctrl->flow_tunnels_n[j]--;
1500 					break;
1501 				}
1502 			}
1503 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1504 		}
1505 	}
1506 }
1507 
1508 /**
1509  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1510  * @p flow if no other flow uses it with the same kind of request.
1511  *
1512  * @param dev
1513  *   Pointer to Ethernet device.
1514  * @param[in] flow
1515  *   Pointer to the flow.
1516  */
1517 static void
1518 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1519 {
1520 	struct mlx5_priv *priv = dev->data->dev_private;
1521 	uint32_t handle_idx;
1522 	struct mlx5_flow_handle *dev_handle;
1523 
1524 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1525 		       handle_idx, dev_handle, next)
1526 		flow_drv_rxq_flags_trim(dev, dev_handle);
1527 }
1528 
1529 /**
1530  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1531  *
1532  * @param dev
1533  *   Pointer to Ethernet device.
1534  */
1535 static void
1536 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1537 {
1538 	struct mlx5_priv *priv = dev->data->dev_private;
1539 	unsigned int i;
1540 
1541 	for (i = 0; i != priv->rxqs_n; ++i) {
1542 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1543 		unsigned int j;
1544 
1545 		if (rxq == NULL || rxq->ctrl == NULL)
1546 			continue;
1547 		rxq->ctrl->rxq.mark = 0;
1548 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1549 			rxq->ctrl->flow_tunnels_n[j] = 0;
1550 		rxq->ctrl->rxq.tunnel = 0;
1551 	}
1552 	priv->mark_enabled = 0;
1553 }
1554 
1555 /**
1556  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1557  *
1558  * @param[in] dev
1559  *   Pointer to the Ethernet device structure.
1560  */
1561 void
1562 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1563 {
1564 	struct mlx5_priv *priv = dev->data->dev_private;
1565 	unsigned int i;
1566 
1567 	for (i = 0; i != priv->rxqs_n; ++i) {
1568 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1569 		struct mlx5_rxq_data *data;
1570 
1571 		if (rxq == NULL || rxq->ctrl == NULL)
1572 			continue;
1573 		data = &rxq->ctrl->rxq;
1574 		if (!rte_flow_dynf_metadata_avail()) {
1575 			data->dynf_meta = 0;
1576 			data->flow_meta_mask = 0;
1577 			data->flow_meta_offset = -1;
1578 			data->flow_meta_port_mask = 0;
1579 		} else {
1580 			data->dynf_meta = 1;
1581 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1582 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1583 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1584 		}
1585 	}
1586 }
1587 
1588 /*
1589  * return a pointer to the desired action in the list of actions.
1590  *
1591  * @param[in] actions
1592  *   The list of actions to search the action in.
1593  * @param[in] action
1594  *   The action to find.
1595  *
1596  * @return
1597  *   Pointer to the action in the list, if found. NULL otherwise.
1598  */
1599 const struct rte_flow_action *
1600 mlx5_flow_find_action(const struct rte_flow_action *actions,
1601 		      enum rte_flow_action_type action)
1602 {
1603 	if (actions == NULL)
1604 		return NULL;
1605 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1606 		if (actions->type == action)
1607 			return actions;
1608 	return NULL;
1609 }
1610 
1611 /*
1612  * Validate the flag action.
1613  *
1614  * @param[in] action_flags
1615  *   Bit-fields that holds the actions detected until now.
1616  * @param[in] attr
1617  *   Attributes of flow that includes this action.
1618  * @param[out] error
1619  *   Pointer to error structure.
1620  *
1621  * @return
1622  *   0 on success, a negative errno value otherwise and rte_errno is set.
1623  */
1624 int
1625 mlx5_flow_validate_action_flag(uint64_t action_flags,
1626 			       const struct rte_flow_attr *attr,
1627 			       struct rte_flow_error *error)
1628 {
1629 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1630 		return rte_flow_error_set(error, EINVAL,
1631 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1632 					  "can't mark and flag in same flow");
1633 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1634 		return rte_flow_error_set(error, EINVAL,
1635 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1636 					  "can't have 2 flag"
1637 					  " actions in same flow");
1638 	if (attr->egress)
1639 		return rte_flow_error_set(error, ENOTSUP,
1640 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1641 					  "flag action not supported for "
1642 					  "egress");
1643 	return 0;
1644 }
1645 
1646 /*
1647  * Validate the mark action.
1648  *
1649  * @param[in] action
1650  *   Pointer to the queue action.
1651  * @param[in] action_flags
1652  *   Bit-fields that holds the actions detected until now.
1653  * @param[in] attr
1654  *   Attributes of flow that includes this action.
1655  * @param[out] error
1656  *   Pointer to error structure.
1657  *
1658  * @return
1659  *   0 on success, a negative errno value otherwise and rte_errno is set.
1660  */
1661 int
1662 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1663 			       uint64_t action_flags,
1664 			       const struct rte_flow_attr *attr,
1665 			       struct rte_flow_error *error)
1666 {
1667 	const struct rte_flow_action_mark *mark = action->conf;
1668 
1669 	if (!mark)
1670 		return rte_flow_error_set(error, EINVAL,
1671 					  RTE_FLOW_ERROR_TYPE_ACTION,
1672 					  action,
1673 					  "configuration cannot be null");
1674 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1675 		return rte_flow_error_set(error, EINVAL,
1676 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1677 					  &mark->id,
1678 					  "mark id must in 0 <= id < "
1679 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1680 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1681 		return rte_flow_error_set(error, EINVAL,
1682 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1683 					  "can't flag and mark in same flow");
1684 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1685 		return rte_flow_error_set(error, EINVAL,
1686 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1687 					  "can't have 2 mark actions in same"
1688 					  " flow");
1689 	if (attr->egress)
1690 		return rte_flow_error_set(error, ENOTSUP,
1691 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1692 					  "mark action not supported for "
1693 					  "egress");
1694 	return 0;
1695 }
1696 
1697 /*
1698  * Validate the drop action.
1699  *
1700  * @param[in] action_flags
1701  *   Bit-fields that holds the actions detected until now.
1702  * @param[in] attr
1703  *   Attributes of flow that includes this action.
1704  * @param[out] error
1705  *   Pointer to error structure.
1706  *
1707  * @return
1708  *   0 on success, a negative errno value otherwise and rte_errno is set.
1709  */
1710 int
1711 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1712 			       const struct rte_flow_attr *attr,
1713 			       struct rte_flow_error *error)
1714 {
1715 	if (attr->egress)
1716 		return rte_flow_error_set(error, ENOTSUP,
1717 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1718 					  "drop action not supported for "
1719 					  "egress");
1720 	return 0;
1721 }
1722 
1723 /*
1724  * Validate the queue action.
1725  *
1726  * @param[in] action
1727  *   Pointer to the queue action.
1728  * @param[in] action_flags
1729  *   Bit-fields that holds the actions detected until now.
1730  * @param[in] dev
1731  *   Pointer to the Ethernet device structure.
1732  * @param[in] attr
1733  *   Attributes of flow that includes this action.
1734  * @param[out] error
1735  *   Pointer to error structure.
1736  *
1737  * @return
1738  *   0 on success, a negative errno value otherwise and rte_errno is set.
1739  */
1740 int
1741 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1742 				uint64_t action_flags,
1743 				struct rte_eth_dev *dev,
1744 				const struct rte_flow_attr *attr,
1745 				struct rte_flow_error *error)
1746 {
1747 	struct mlx5_priv *priv = dev->data->dev_private;
1748 	const struct rte_flow_action_queue *queue = action->conf;
1749 
1750 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1751 		return rte_flow_error_set(error, EINVAL,
1752 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1753 					  "can't have 2 fate actions in"
1754 					  " same flow");
1755 	if (attr->egress)
1756 		return rte_flow_error_set(error, ENOTSUP,
1757 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1758 					  "queue action not supported for egress.");
1759 	if (mlx5_is_external_rxq(dev, queue->index))
1760 		return 0;
1761 	if (!priv->rxqs_n)
1762 		return rte_flow_error_set(error, EINVAL,
1763 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1764 					  NULL, "No Rx queues configured");
1765 	if (queue->index >= priv->rxqs_n)
1766 		return rte_flow_error_set(error, EINVAL,
1767 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1768 					  &queue->index,
1769 					  "queue index out of range");
1770 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1771 		return rte_flow_error_set(error, EINVAL,
1772 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1773 					  &queue->index,
1774 					  "queue is not configured");
1775 	return 0;
1776 }
1777 
1778 /**
1779  * Validate queue numbers for device RSS.
1780  *
1781  * @param[in] dev
1782  *   Configured device.
1783  * @param[in] queues
1784  *   Array of queue numbers.
1785  * @param[in] queues_n
1786  *   Size of the @p queues array.
1787  * @param[out] error
1788  *   On error, filled with a textual error description.
1789  * @param[out] queue_idx
1790  *   On error, filled with an offending queue index in @p queues array.
1791  *
1792  * @return
1793  *   0 on success, a negative errno code on error.
1794  */
1795 static int
1796 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1797 			 const uint16_t *queues, uint32_t queues_n,
1798 			 const char **error, uint32_t *queue_idx)
1799 {
1800 	const struct mlx5_priv *priv = dev->data->dev_private;
1801 	bool is_hairpin = false;
1802 	bool is_ext_rss = false;
1803 	uint32_t i;
1804 
1805 	for (i = 0; i != queues_n; ++i) {
1806 		struct mlx5_rxq_ctrl *rxq_ctrl;
1807 
1808 		if (mlx5_is_external_rxq(dev, queues[0])) {
1809 			is_ext_rss = true;
1810 			continue;
1811 		}
1812 		if (is_ext_rss) {
1813 			*error = "Combining external and regular RSS queues is not supported";
1814 			*queue_idx = i;
1815 			return -ENOTSUP;
1816 		}
1817 		if (queues[i] >= priv->rxqs_n) {
1818 			*error = "queue index out of range";
1819 			*queue_idx = i;
1820 			return -EINVAL;
1821 		}
1822 		rxq_ctrl = mlx5_rxq_ctrl_get(dev, queues[i]);
1823 		if (rxq_ctrl == NULL) {
1824 			*error =  "queue is not configured";
1825 			*queue_idx = i;
1826 			return -EINVAL;
1827 		}
1828 		if (i == 0 && rxq_ctrl->is_hairpin)
1829 			is_hairpin = true;
1830 		if (is_hairpin != rxq_ctrl->is_hairpin) {
1831 			*error = "combining hairpin and regular RSS queues is not supported";
1832 			*queue_idx = i;
1833 			return -ENOTSUP;
1834 		}
1835 	}
1836 	return 0;
1837 }
1838 
1839 /*
1840  * Validate the rss action.
1841  *
1842  * @param[in] dev
1843  *   Pointer to the Ethernet device structure.
1844  * @param[in] action
1845  *   Pointer to the queue action.
1846  * @param[out] error
1847  *   Pointer to error structure.
1848  *
1849  * @return
1850  *   0 on success, a negative errno value otherwise and rte_errno is set.
1851  */
1852 int
1853 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1854 			 const struct rte_flow_action *action,
1855 			 struct rte_flow_error *error)
1856 {
1857 	struct mlx5_priv *priv = dev->data->dev_private;
1858 	const struct rte_flow_action_rss *rss = action->conf;
1859 	int ret;
1860 	const char *message;
1861 	uint32_t queue_idx;
1862 
1863 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1864 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1865 		return rte_flow_error_set(error, ENOTSUP,
1866 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1867 					  &rss->func,
1868 					  "RSS hash function not supported");
1869 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1870 	if (rss->level > 2)
1871 #else
1872 	if (rss->level > 1)
1873 #endif
1874 		return rte_flow_error_set(error, ENOTSUP,
1875 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1876 					  &rss->level,
1877 					  "tunnel RSS is not supported");
1878 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1879 	if (rss->key_len == 0 && rss->key != NULL)
1880 		return rte_flow_error_set(error, ENOTSUP,
1881 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1882 					  &rss->key_len,
1883 					  "RSS hash key length 0");
1884 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1885 		return rte_flow_error_set(error, ENOTSUP,
1886 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1887 					  &rss->key_len,
1888 					  "RSS hash key too small");
1889 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1890 		return rte_flow_error_set(error, ENOTSUP,
1891 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1892 					  &rss->key_len,
1893 					  "RSS hash key too large");
1894 	if (rss->queue_num > priv->sh->dev_cap.ind_table_max_size)
1895 		return rte_flow_error_set(error, ENOTSUP,
1896 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1897 					  &rss->queue_num,
1898 					  "number of queues too large");
1899 	if (rss->types & MLX5_RSS_HF_MASK)
1900 		return rte_flow_error_set(error, ENOTSUP,
1901 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1902 					  &rss->types,
1903 					  "some RSS protocols are not"
1904 					  " supported");
1905 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1906 	    !(rss->types & RTE_ETH_RSS_IP))
1907 		return rte_flow_error_set(error, EINVAL,
1908 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1909 					  "L3 partial RSS requested but L3 RSS"
1910 					  " type not specified");
1911 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1912 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1913 		return rte_flow_error_set(error, EINVAL,
1914 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1915 					  "L4 partial RSS requested but L4 RSS"
1916 					  " type not specified");
1917 	if (!priv->rxqs_n && priv->ext_rxqs == NULL)
1918 		return rte_flow_error_set(error, EINVAL,
1919 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1920 					  NULL, "No Rx queues configured");
1921 	if (!rss->queue_num)
1922 		return rte_flow_error_set(error, EINVAL,
1923 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1924 					  NULL, "No queues configured");
1925 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1926 				       &message, &queue_idx);
1927 	if (ret != 0) {
1928 		return rte_flow_error_set(error, -ret,
1929 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1930 					  &rss->queue[queue_idx], message);
1931 	}
1932 	return 0;
1933 }
1934 
1935 /*
1936  * Validate the rss action.
1937  *
1938  * @param[in] action
1939  *   Pointer to the queue action.
1940  * @param[in] action_flags
1941  *   Bit-fields that holds the actions detected until now.
1942  * @param[in] dev
1943  *   Pointer to the Ethernet device structure.
1944  * @param[in] attr
1945  *   Attributes of flow that includes this action.
1946  * @param[in] item_flags
1947  *   Items that were detected.
1948  * @param[out] error
1949  *   Pointer to error structure.
1950  *
1951  * @return
1952  *   0 on success, a negative errno value otherwise and rte_errno is set.
1953  */
1954 int
1955 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1956 			      uint64_t action_flags,
1957 			      struct rte_eth_dev *dev,
1958 			      const struct rte_flow_attr *attr,
1959 			      uint64_t item_flags,
1960 			      struct rte_flow_error *error)
1961 {
1962 	const struct rte_flow_action_rss *rss = action->conf;
1963 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1964 	int ret;
1965 
1966 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1967 		return rte_flow_error_set(error, EINVAL,
1968 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1969 					  "can't have 2 fate actions"
1970 					  " in same flow");
1971 	ret = mlx5_validate_action_rss(dev, action, error);
1972 	if (ret)
1973 		return ret;
1974 	if (attr->egress)
1975 		return rte_flow_error_set(error, ENOTSUP,
1976 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1977 					  "rss action not supported for "
1978 					  "egress");
1979 	if (rss->level > 1 && !tunnel)
1980 		return rte_flow_error_set(error, EINVAL,
1981 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1982 					  "inner RSS is not supported for "
1983 					  "non-tunnel flows");
1984 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1985 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1986 		return rte_flow_error_set(error, EINVAL,
1987 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1988 					  "RSS on eCPRI is not supported now");
1989 	}
1990 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1991 	    !(item_flags &
1992 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1993 	    rss->level > 1)
1994 		return rte_flow_error_set(error, EINVAL,
1995 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1996 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1997 	return 0;
1998 }
1999 
2000 /*
2001  * Validate the default miss action.
2002  *
2003  * @param[in] action_flags
2004  *   Bit-fields that holds the actions detected until now.
2005  * @param[out] error
2006  *   Pointer to error structure.
2007  *
2008  * @return
2009  *   0 on success, a negative errno value otherwise and rte_errno is set.
2010  */
2011 int
2012 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
2013 				const struct rte_flow_attr *attr,
2014 				struct rte_flow_error *error)
2015 {
2016 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
2017 		return rte_flow_error_set(error, EINVAL,
2018 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2019 					  "can't have 2 fate actions in"
2020 					  " same flow");
2021 	if (attr->egress)
2022 		return rte_flow_error_set(error, ENOTSUP,
2023 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
2024 					  "default miss action not supported "
2025 					  "for egress");
2026 	if (attr->group)
2027 		return rte_flow_error_set(error, ENOTSUP,
2028 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
2029 					  "only group 0 is supported");
2030 	if (attr->transfer)
2031 		return rte_flow_error_set(error, ENOTSUP,
2032 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2033 					  NULL, "transfer is not supported");
2034 	return 0;
2035 }
2036 
2037 /*
2038  * Validate the count action.
2039  *
2040  * @param[in] dev
2041  *   Pointer to the Ethernet device structure.
2042  * @param[in] attr
2043  *   Attributes of flow that includes this action.
2044  * @param[out] error
2045  *   Pointer to error structure.
2046  *
2047  * @return
2048  *   0 on success, a negative errno value otherwise and rte_errno is set.
2049  */
2050 int
2051 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
2052 				const struct rte_flow_attr *attr,
2053 				struct rte_flow_error *error)
2054 {
2055 	if (attr->egress)
2056 		return rte_flow_error_set(error, ENOTSUP,
2057 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
2058 					  "count action not supported for "
2059 					  "egress");
2060 	return 0;
2061 }
2062 
2063 /*
2064  * Validate the ASO CT action.
2065  *
2066  * @param[in] dev
2067  *   Pointer to the Ethernet device structure.
2068  * @param[in] conntrack
2069  *   Pointer to the CT action profile.
2070  * @param[out] error
2071  *   Pointer to error structure.
2072  *
2073  * @return
2074  *   0 on success, a negative errno value otherwise and rte_errno is set.
2075  */
2076 int
2077 mlx5_validate_action_ct(struct rte_eth_dev *dev,
2078 			const struct rte_flow_action_conntrack *conntrack,
2079 			struct rte_flow_error *error)
2080 {
2081 	RTE_SET_USED(dev);
2082 
2083 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
2084 		return rte_flow_error_set(error, EINVAL,
2085 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2086 					  "Invalid CT state");
2087 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
2088 		return rte_flow_error_set(error, EINVAL,
2089 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2090 					  "Invalid last TCP packet flag");
2091 	return 0;
2092 }
2093 
2094 /**
2095  * Verify the @p attributes will be correctly understood by the NIC and store
2096  * them in the @p flow if everything is correct.
2097  *
2098  * @param[in] dev
2099  *   Pointer to the Ethernet device structure.
2100  * @param[in] attributes
2101  *   Pointer to flow attributes
2102  * @param[out] error
2103  *   Pointer to error structure.
2104  *
2105  * @return
2106  *   0 on success, a negative errno value otherwise and rte_errno is set.
2107  */
2108 int
2109 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
2110 			      const struct rte_flow_attr *attributes,
2111 			      struct rte_flow_error *error)
2112 {
2113 	struct mlx5_priv *priv = dev->data->dev_private;
2114 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
2115 
2116 	if (attributes->group)
2117 		return rte_flow_error_set(error, ENOTSUP,
2118 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
2119 					  NULL, "groups is not supported");
2120 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
2121 	    attributes->priority >= priority_max)
2122 		return rte_flow_error_set(error, ENOTSUP,
2123 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
2124 					  NULL, "priority out of range");
2125 	if (attributes->egress)
2126 		return rte_flow_error_set(error, ENOTSUP,
2127 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
2128 					  "egress is not supported");
2129 	if (attributes->transfer && !priv->sh->config.dv_esw_en)
2130 		return rte_flow_error_set(error, ENOTSUP,
2131 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2132 					  NULL, "transfer is not supported");
2133 	if (!attributes->ingress)
2134 		return rte_flow_error_set(error, EINVAL,
2135 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2136 					  NULL,
2137 					  "ingress attribute is mandatory");
2138 	return 0;
2139 }
2140 
2141 /**
2142  * Validate ICMP6 item.
2143  *
2144  * @param[in] item
2145  *   Item specification.
2146  * @param[in] item_flags
2147  *   Bit-fields that holds the items detected until now.
2148  * @param[in] ext_vlan_sup
2149  *   Whether extended VLAN features are supported or not.
2150  * @param[out] error
2151  *   Pointer to error structure.
2152  *
2153  * @return
2154  *   0 on success, a negative errno value otherwise and rte_errno is set.
2155  */
2156 int
2157 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2158 			       uint64_t item_flags,
2159 			       uint8_t target_protocol,
2160 			       struct rte_flow_error *error)
2161 {
2162 	const struct rte_flow_item_icmp6 *mask = item->mask;
2163 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2164 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2165 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2166 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2167 				      MLX5_FLOW_LAYER_OUTER_L4;
2168 	int ret;
2169 
2170 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2171 		return rte_flow_error_set(error, EINVAL,
2172 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2173 					  "protocol filtering not compatible"
2174 					  " with ICMP6 layer");
2175 	if (!(item_flags & l3m))
2176 		return rte_flow_error_set(error, EINVAL,
2177 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2178 					  "IPv6 is mandatory to filter on"
2179 					  " ICMP6");
2180 	if (item_flags & l4m)
2181 		return rte_flow_error_set(error, EINVAL,
2182 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2183 					  "multiple L4 layers not supported");
2184 	if (!mask)
2185 		mask = &rte_flow_item_icmp6_mask;
2186 	ret = mlx5_flow_item_acceptable
2187 		(item, (const uint8_t *)mask,
2188 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2189 		 sizeof(struct rte_flow_item_icmp6),
2190 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2191 	if (ret < 0)
2192 		return ret;
2193 	return 0;
2194 }
2195 
2196 /**
2197  * Validate ICMP item.
2198  *
2199  * @param[in] item
2200  *   Item specification.
2201  * @param[in] item_flags
2202  *   Bit-fields that holds the items detected until now.
2203  * @param[out] error
2204  *   Pointer to error structure.
2205  *
2206  * @return
2207  *   0 on success, a negative errno value otherwise and rte_errno is set.
2208  */
2209 int
2210 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2211 			     uint64_t item_flags,
2212 			     uint8_t target_protocol,
2213 			     struct rte_flow_error *error)
2214 {
2215 	const struct rte_flow_item_icmp *mask = item->mask;
2216 	const struct rte_flow_item_icmp nic_mask = {
2217 		.hdr.icmp_type = 0xff,
2218 		.hdr.icmp_code = 0xff,
2219 		.hdr.icmp_ident = RTE_BE16(0xffff),
2220 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2221 	};
2222 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2223 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2224 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2225 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2226 				      MLX5_FLOW_LAYER_OUTER_L4;
2227 	int ret;
2228 
2229 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2230 		return rte_flow_error_set(error, EINVAL,
2231 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2232 					  "protocol filtering not compatible"
2233 					  " with ICMP layer");
2234 	if (!(item_flags & l3m))
2235 		return rte_flow_error_set(error, EINVAL,
2236 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2237 					  "IPv4 is mandatory to filter"
2238 					  " on ICMP");
2239 	if (item_flags & l4m)
2240 		return rte_flow_error_set(error, EINVAL,
2241 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2242 					  "multiple L4 layers not supported");
2243 	if (!mask)
2244 		mask = &nic_mask;
2245 	ret = mlx5_flow_item_acceptable
2246 		(item, (const uint8_t *)mask,
2247 		 (const uint8_t *)&nic_mask,
2248 		 sizeof(struct rte_flow_item_icmp),
2249 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2250 	if (ret < 0)
2251 		return ret;
2252 	return 0;
2253 }
2254 
2255 /**
2256  * Validate Ethernet item.
2257  *
2258  * @param[in] item
2259  *   Item specification.
2260  * @param[in] item_flags
2261  *   Bit-fields that holds the items detected until now.
2262  * @param[out] error
2263  *   Pointer to error structure.
2264  *
2265  * @return
2266  *   0 on success, a negative errno value otherwise and rte_errno is set.
2267  */
2268 int
2269 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2270 			    uint64_t item_flags, bool ext_vlan_sup,
2271 			    struct rte_flow_error *error)
2272 {
2273 	const struct rte_flow_item_eth *mask = item->mask;
2274 	const struct rte_flow_item_eth nic_mask = {
2275 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2276 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2277 		.type = RTE_BE16(0xffff),
2278 		.has_vlan = ext_vlan_sup ? 1 : 0,
2279 	};
2280 	int ret;
2281 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2282 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2283 				       MLX5_FLOW_LAYER_OUTER_L2;
2284 
2285 	if (item_flags & ethm)
2286 		return rte_flow_error_set(error, ENOTSUP,
2287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2288 					  "multiple L2 layers not supported");
2289 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2290 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2291 		return rte_flow_error_set(error, EINVAL,
2292 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 					  "L2 layer should not follow "
2294 					  "L3 layers");
2295 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2296 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2297 		return rte_flow_error_set(error, EINVAL,
2298 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2299 					  "L2 layer should not follow VLAN");
2300 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2301 		return rte_flow_error_set(error, EINVAL,
2302 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 					  "L2 layer should not follow GTP");
2304 	if (!mask)
2305 		mask = &rte_flow_item_eth_mask;
2306 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2307 					(const uint8_t *)&nic_mask,
2308 					sizeof(struct rte_flow_item_eth),
2309 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2310 	return ret;
2311 }
2312 
2313 /**
2314  * Validate VLAN item.
2315  *
2316  * @param[in] item
2317  *   Item specification.
2318  * @param[in] item_flags
2319  *   Bit-fields that holds the items detected until now.
2320  * @param[in] dev
2321  *   Ethernet device flow is being created on.
2322  * @param[out] error
2323  *   Pointer to error structure.
2324  *
2325  * @return
2326  *   0 on success, a negative errno value otherwise and rte_errno is set.
2327  */
2328 int
2329 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2330 			     uint64_t item_flags,
2331 			     struct rte_eth_dev *dev,
2332 			     struct rte_flow_error *error)
2333 {
2334 	const struct rte_flow_item_vlan *spec = item->spec;
2335 	const struct rte_flow_item_vlan *mask = item->mask;
2336 	const struct rte_flow_item_vlan nic_mask = {
2337 		.tci = RTE_BE16(UINT16_MAX),
2338 		.inner_type = RTE_BE16(UINT16_MAX),
2339 	};
2340 	uint16_t vlan_tag = 0;
2341 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2342 	int ret;
2343 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2344 					MLX5_FLOW_LAYER_INNER_L4) :
2345 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2346 					MLX5_FLOW_LAYER_OUTER_L4);
2347 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2348 					MLX5_FLOW_LAYER_OUTER_VLAN;
2349 
2350 	if (item_flags & vlanm)
2351 		return rte_flow_error_set(error, EINVAL,
2352 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2353 					  "multiple VLAN layers not supported");
2354 	else if ((item_flags & l34m) != 0)
2355 		return rte_flow_error_set(error, EINVAL,
2356 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2357 					  "VLAN cannot follow L3/L4 layer");
2358 	if (!mask)
2359 		mask = &rte_flow_item_vlan_mask;
2360 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2361 					(const uint8_t *)&nic_mask,
2362 					sizeof(struct rte_flow_item_vlan),
2363 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2364 	if (ret)
2365 		return ret;
2366 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2367 		struct mlx5_priv *priv = dev->data->dev_private;
2368 
2369 		if (priv->vmwa_context) {
2370 			/*
2371 			 * Non-NULL context means we have a virtual machine
2372 			 * and SR-IOV enabled, we have to create VLAN interface
2373 			 * to make hypervisor to setup E-Switch vport
2374 			 * context correctly. We avoid creating the multiple
2375 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2376 			 */
2377 			return rte_flow_error_set(error, EINVAL,
2378 						  RTE_FLOW_ERROR_TYPE_ITEM,
2379 						  item,
2380 						  "VLAN tag mask is not"
2381 						  " supported in virtual"
2382 						  " environment");
2383 		}
2384 	}
2385 	if (spec) {
2386 		vlan_tag = spec->tci;
2387 		vlan_tag &= mask->tci;
2388 	}
2389 	/*
2390 	 * From verbs perspective an empty VLAN is equivalent
2391 	 * to a packet without VLAN layer.
2392 	 */
2393 	if (!vlan_tag)
2394 		return rte_flow_error_set(error, EINVAL,
2395 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2396 					  item->spec,
2397 					  "VLAN cannot be empty");
2398 	return 0;
2399 }
2400 
2401 /**
2402  * Validate IPV4 item.
2403  *
2404  * @param[in] item
2405  *   Item specification.
2406  * @param[in] item_flags
2407  *   Bit-fields that holds the items detected until now.
2408  * @param[in] last_item
2409  *   Previous validated item in the pattern items.
2410  * @param[in] ether_type
2411  *   Type in the ethernet layer header (including dot1q).
2412  * @param[in] acc_mask
2413  *   Acceptable mask, if NULL default internal default mask
2414  *   will be used to check whether item fields are supported.
2415  * @param[in] range_accepted
2416  *   True if range of values is accepted for specific fields, false otherwise.
2417  * @param[out] error
2418  *   Pointer to error structure.
2419  *
2420  * @return
2421  *   0 on success, a negative errno value otherwise and rte_errno is set.
2422  */
2423 int
2424 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2425 			     uint64_t item_flags,
2426 			     uint64_t last_item,
2427 			     uint16_t ether_type,
2428 			     const struct rte_flow_item_ipv4 *acc_mask,
2429 			     bool range_accepted,
2430 			     struct rte_flow_error *error)
2431 {
2432 	const struct rte_flow_item_ipv4 *mask = item->mask;
2433 	const struct rte_flow_item_ipv4 *spec = item->spec;
2434 	const struct rte_flow_item_ipv4 nic_mask = {
2435 		.hdr = {
2436 			.src_addr = RTE_BE32(0xffffffff),
2437 			.dst_addr = RTE_BE32(0xffffffff),
2438 			.type_of_service = 0xff,
2439 			.next_proto_id = 0xff,
2440 		},
2441 	};
2442 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2443 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2444 				      MLX5_FLOW_LAYER_OUTER_L3;
2445 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2446 				      MLX5_FLOW_LAYER_OUTER_L4;
2447 	int ret;
2448 	uint8_t next_proto = 0xFF;
2449 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2450 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2451 				  MLX5_FLOW_LAYER_INNER_VLAN);
2452 
2453 	if ((last_item & l2_vlan) && ether_type &&
2454 	    ether_type != RTE_ETHER_TYPE_IPV4)
2455 		return rte_flow_error_set(error, EINVAL,
2456 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2457 					  "IPv4 cannot follow L2/VLAN layer "
2458 					  "which ether type is not IPv4");
2459 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2460 		if (mask && spec)
2461 			next_proto = mask->hdr.next_proto_id &
2462 				     spec->hdr.next_proto_id;
2463 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2464 			return rte_flow_error_set(error, EINVAL,
2465 						  RTE_FLOW_ERROR_TYPE_ITEM,
2466 						  item,
2467 						  "multiple tunnel "
2468 						  "not supported");
2469 	}
2470 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2471 		return rte_flow_error_set(error, EINVAL,
2472 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2473 					  "wrong tunnel type - IPv6 specified "
2474 					  "but IPv4 item provided");
2475 	if (item_flags & l3m)
2476 		return rte_flow_error_set(error, ENOTSUP,
2477 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2478 					  "multiple L3 layers not supported");
2479 	else if (item_flags & l4m)
2480 		return rte_flow_error_set(error, EINVAL,
2481 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2482 					  "L3 cannot follow an L4 layer.");
2483 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2484 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2485 		return rte_flow_error_set(error, EINVAL,
2486 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2487 					  "L3 cannot follow an NVGRE layer.");
2488 	if (!mask)
2489 		mask = &rte_flow_item_ipv4_mask;
2490 	else if (mask->hdr.next_proto_id != 0 &&
2491 		 mask->hdr.next_proto_id != 0xff)
2492 		return rte_flow_error_set(error, EINVAL,
2493 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2494 					  "partial mask is not supported"
2495 					  " for protocol");
2496 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2497 					acc_mask ? (const uint8_t *)acc_mask
2498 						 : (const uint8_t *)&nic_mask,
2499 					sizeof(struct rte_flow_item_ipv4),
2500 					range_accepted, error);
2501 	if (ret < 0)
2502 		return ret;
2503 	return 0;
2504 }
2505 
2506 /**
2507  * Validate IPV6 item.
2508  *
2509  * @param[in] item
2510  *   Item specification.
2511  * @param[in] item_flags
2512  *   Bit-fields that holds the items detected until now.
2513  * @param[in] last_item
2514  *   Previous validated item in the pattern items.
2515  * @param[in] ether_type
2516  *   Type in the ethernet layer header (including dot1q).
2517  * @param[in] acc_mask
2518  *   Acceptable mask, if NULL default internal default mask
2519  *   will be used to check whether item fields are supported.
2520  * @param[out] error
2521  *   Pointer to error structure.
2522  *
2523  * @return
2524  *   0 on success, a negative errno value otherwise and rte_errno is set.
2525  */
2526 int
2527 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2528 			     uint64_t item_flags,
2529 			     uint64_t last_item,
2530 			     uint16_t ether_type,
2531 			     const struct rte_flow_item_ipv6 *acc_mask,
2532 			     struct rte_flow_error *error)
2533 {
2534 	const struct rte_flow_item_ipv6 *mask = item->mask;
2535 	const struct rte_flow_item_ipv6 *spec = item->spec;
2536 	const struct rte_flow_item_ipv6 nic_mask = {
2537 		.hdr = {
2538 			.src_addr =
2539 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2540 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2541 			.dst_addr =
2542 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2543 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2544 			.vtc_flow = RTE_BE32(0xffffffff),
2545 			.proto = 0xff,
2546 		},
2547 	};
2548 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2549 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2550 				      MLX5_FLOW_LAYER_OUTER_L3;
2551 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2552 				      MLX5_FLOW_LAYER_OUTER_L4;
2553 	int ret;
2554 	uint8_t next_proto = 0xFF;
2555 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2556 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2557 				  MLX5_FLOW_LAYER_INNER_VLAN);
2558 
2559 	if ((last_item & l2_vlan) && ether_type &&
2560 	    ether_type != RTE_ETHER_TYPE_IPV6)
2561 		return rte_flow_error_set(error, EINVAL,
2562 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2563 					  "IPv6 cannot follow L2/VLAN layer "
2564 					  "which ether type is not IPv6");
2565 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2566 		next_proto = spec->hdr.proto;
2567 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2568 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2569 			return rte_flow_error_set(error, EINVAL,
2570 						  RTE_FLOW_ERROR_TYPE_ITEM,
2571 						  item,
2572 						  "multiple tunnel "
2573 						  "not supported");
2574 	}
2575 	if (next_proto == IPPROTO_HOPOPTS  ||
2576 	    next_proto == IPPROTO_ROUTING  ||
2577 	    next_proto == IPPROTO_FRAGMENT ||
2578 	    next_proto == IPPROTO_ESP	   ||
2579 	    next_proto == IPPROTO_AH	   ||
2580 	    next_proto == IPPROTO_DSTOPTS)
2581 		return rte_flow_error_set(error, EINVAL,
2582 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2583 					  "IPv6 proto (next header) should "
2584 					  "not be set as extension header");
2585 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2586 		return rte_flow_error_set(error, EINVAL,
2587 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2588 					  "wrong tunnel type - IPv4 specified "
2589 					  "but IPv6 item provided");
2590 	if (item_flags & l3m)
2591 		return rte_flow_error_set(error, ENOTSUP,
2592 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2593 					  "multiple L3 layers not supported");
2594 	else if (item_flags & l4m)
2595 		return rte_flow_error_set(error, EINVAL,
2596 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2597 					  "L3 cannot follow an L4 layer.");
2598 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2599 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2600 		return rte_flow_error_set(error, EINVAL,
2601 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2602 					  "L3 cannot follow an NVGRE layer.");
2603 	if (!mask)
2604 		mask = &rte_flow_item_ipv6_mask;
2605 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2606 					acc_mask ? (const uint8_t *)acc_mask
2607 						 : (const uint8_t *)&nic_mask,
2608 					sizeof(struct rte_flow_item_ipv6),
2609 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2610 	if (ret < 0)
2611 		return ret;
2612 	return 0;
2613 }
2614 
2615 /**
2616  * Validate UDP item.
2617  *
2618  * @param[in] item
2619  *   Item specification.
2620  * @param[in] item_flags
2621  *   Bit-fields that holds the items detected until now.
2622  * @param[in] target_protocol
2623  *   The next protocol in the previous item.
2624  * @param[in] flow_mask
2625  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2626  * @param[out] error
2627  *   Pointer to error structure.
2628  *
2629  * @return
2630  *   0 on success, a negative errno value otherwise and rte_errno is set.
2631  */
2632 int
2633 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2634 			    uint64_t item_flags,
2635 			    uint8_t target_protocol,
2636 			    struct rte_flow_error *error)
2637 {
2638 	const struct rte_flow_item_udp *mask = item->mask;
2639 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2640 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2641 				      MLX5_FLOW_LAYER_OUTER_L3;
2642 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2643 				      MLX5_FLOW_LAYER_OUTER_L4;
2644 	int ret;
2645 
2646 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2647 		return rte_flow_error_set(error, EINVAL,
2648 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2649 					  "protocol filtering not compatible"
2650 					  " with UDP layer");
2651 	if (!(item_flags & l3m))
2652 		return rte_flow_error_set(error, EINVAL,
2653 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2654 					  "L3 is mandatory to filter on L4");
2655 	if (item_flags & l4m)
2656 		return rte_flow_error_set(error, EINVAL,
2657 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2658 					  "multiple L4 layers not supported");
2659 	if (!mask)
2660 		mask = &rte_flow_item_udp_mask;
2661 	ret = mlx5_flow_item_acceptable
2662 		(item, (const uint8_t *)mask,
2663 		 (const uint8_t *)&rte_flow_item_udp_mask,
2664 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2665 		 error);
2666 	if (ret < 0)
2667 		return ret;
2668 	return 0;
2669 }
2670 
2671 /**
2672  * Validate TCP item.
2673  *
2674  * @param[in] item
2675  *   Item specification.
2676  * @param[in] item_flags
2677  *   Bit-fields that holds the items detected until now.
2678  * @param[in] target_protocol
2679  *   The next protocol in the previous item.
2680  * @param[out] error
2681  *   Pointer to error structure.
2682  *
2683  * @return
2684  *   0 on success, a negative errno value otherwise and rte_errno is set.
2685  */
2686 int
2687 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2688 			    uint64_t item_flags,
2689 			    uint8_t target_protocol,
2690 			    const struct rte_flow_item_tcp *flow_mask,
2691 			    struct rte_flow_error *error)
2692 {
2693 	const struct rte_flow_item_tcp *mask = item->mask;
2694 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2695 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2696 				      MLX5_FLOW_LAYER_OUTER_L3;
2697 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2698 				      MLX5_FLOW_LAYER_OUTER_L4;
2699 	int ret;
2700 
2701 	MLX5_ASSERT(flow_mask);
2702 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2703 		return rte_flow_error_set(error, EINVAL,
2704 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2705 					  "protocol filtering not compatible"
2706 					  " with TCP layer");
2707 	if (!(item_flags & l3m))
2708 		return rte_flow_error_set(error, EINVAL,
2709 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2710 					  "L3 is mandatory to filter on L4");
2711 	if (item_flags & l4m)
2712 		return rte_flow_error_set(error, EINVAL,
2713 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2714 					  "multiple L4 layers not supported");
2715 	if (!mask)
2716 		mask = &rte_flow_item_tcp_mask;
2717 	ret = mlx5_flow_item_acceptable
2718 		(item, (const uint8_t *)mask,
2719 		 (const uint8_t *)flow_mask,
2720 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2721 		 error);
2722 	if (ret < 0)
2723 		return ret;
2724 	return 0;
2725 }
2726 
2727 /**
2728  * Validate VXLAN item.
2729  *
2730  * @param[in] dev
2731  *   Pointer to the Ethernet device structure.
2732  * @param[in] udp_dport
2733  *   UDP destination port
2734  * @param[in] item
2735  *   Item specification.
2736  * @param[in] item_flags
2737  *   Bit-fields that holds the items detected until now.
2738  * @param[in] attr
2739  *   Flow rule attributes.
2740  * @param[out] error
2741  *   Pointer to error structure.
2742  *
2743  * @return
2744  *   0 on success, a negative errno value otherwise and rte_errno is set.
2745  */
2746 int
2747 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2748 			      uint16_t udp_dport,
2749 			      const struct rte_flow_item *item,
2750 			      uint64_t item_flags,
2751 			      const struct rte_flow_attr *attr,
2752 			      struct rte_flow_error *error)
2753 {
2754 	const struct rte_flow_item_vxlan *spec = item->spec;
2755 	const struct rte_flow_item_vxlan *mask = item->mask;
2756 	int ret;
2757 	struct mlx5_priv *priv = dev->data->dev_private;
2758 	union vni {
2759 		uint32_t vlan_id;
2760 		uint8_t vni[4];
2761 	} id = { .vlan_id = 0, };
2762 	const struct rte_flow_item_vxlan nic_mask = {
2763 		.vni = "\xff\xff\xff",
2764 		.rsvd1 = 0xff,
2765 	};
2766 	const struct rte_flow_item_vxlan *valid_mask;
2767 
2768 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2769 		return rte_flow_error_set(error, ENOTSUP,
2770 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2771 					  "multiple tunnel layers not"
2772 					  " supported");
2773 	valid_mask = &rte_flow_item_vxlan_mask;
2774 	/*
2775 	 * Verify only UDPv4 is present as defined in
2776 	 * https://tools.ietf.org/html/rfc7348
2777 	 */
2778 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2779 		return rte_flow_error_set(error, EINVAL,
2780 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2781 					  "no outer UDP layer found");
2782 	if (!mask)
2783 		mask = &rte_flow_item_vxlan_mask;
2784 
2785 	if (priv->sh->steering_format_version !=
2786 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2787 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2788 		/* FDB domain & NIC domain non-zero group */
2789 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2790 			valid_mask = &nic_mask;
2791 		/* Group zero in NIC domain */
2792 		if (!attr->group && !attr->transfer &&
2793 		    priv->sh->tunnel_header_0_1)
2794 			valid_mask = &nic_mask;
2795 	}
2796 	ret = mlx5_flow_item_acceptable
2797 		(item, (const uint8_t *)mask,
2798 		 (const uint8_t *)valid_mask,
2799 		 sizeof(struct rte_flow_item_vxlan),
2800 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2801 	if (ret < 0)
2802 		return ret;
2803 	if (spec) {
2804 		memcpy(&id.vni[1], spec->vni, 3);
2805 		memcpy(&id.vni[1], mask->vni, 3);
2806 	}
2807 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2808 		return rte_flow_error_set(error, ENOTSUP,
2809 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2810 					  "VXLAN tunnel must be fully defined");
2811 	return 0;
2812 }
2813 
2814 /**
2815  * Validate VXLAN_GPE item.
2816  *
2817  * @param[in] item
2818  *   Item specification.
2819  * @param[in] item_flags
2820  *   Bit-fields that holds the items detected until now.
2821  * @param[in] priv
2822  *   Pointer to the private data structure.
2823  * @param[in] target_protocol
2824  *   The next protocol in the previous item.
2825  * @param[out] error
2826  *   Pointer to error structure.
2827  *
2828  * @return
2829  *   0 on success, a negative errno value otherwise and rte_errno is set.
2830  */
2831 int
2832 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2833 				  uint64_t item_flags,
2834 				  struct rte_eth_dev *dev,
2835 				  struct rte_flow_error *error)
2836 {
2837 	struct mlx5_priv *priv = dev->data->dev_private;
2838 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2839 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2840 	int ret;
2841 	union vni {
2842 		uint32_t vlan_id;
2843 		uint8_t vni[4];
2844 	} id = { .vlan_id = 0, };
2845 
2846 	if (!priv->sh->config.l3_vxlan_en)
2847 		return rte_flow_error_set(error, ENOTSUP,
2848 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2849 					  "L3 VXLAN is not enabled by device"
2850 					  " parameter and/or not configured in"
2851 					  " firmware");
2852 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2853 		return rte_flow_error_set(error, ENOTSUP,
2854 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2855 					  "multiple tunnel layers not"
2856 					  " supported");
2857 	/*
2858 	 * Verify only UDPv4 is present as defined in
2859 	 * https://tools.ietf.org/html/rfc7348
2860 	 */
2861 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2862 		return rte_flow_error_set(error, EINVAL,
2863 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2864 					  "no outer UDP layer found");
2865 	if (!mask)
2866 		mask = &rte_flow_item_vxlan_gpe_mask;
2867 	ret = mlx5_flow_item_acceptable
2868 		(item, (const uint8_t *)mask,
2869 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2870 		 sizeof(struct rte_flow_item_vxlan_gpe),
2871 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2872 	if (ret < 0)
2873 		return ret;
2874 	if (spec) {
2875 		if (spec->protocol)
2876 			return rte_flow_error_set(error, ENOTSUP,
2877 						  RTE_FLOW_ERROR_TYPE_ITEM,
2878 						  item,
2879 						  "VxLAN-GPE protocol"
2880 						  " not supported");
2881 		memcpy(&id.vni[1], spec->vni, 3);
2882 		memcpy(&id.vni[1], mask->vni, 3);
2883 	}
2884 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2885 		return rte_flow_error_set(error, ENOTSUP,
2886 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2887 					  "VXLAN-GPE tunnel must be fully"
2888 					  " defined");
2889 	return 0;
2890 }
2891 /**
2892  * Validate GRE Key item.
2893  *
2894  * @param[in] item
2895  *   Item specification.
2896  * @param[in] item_flags
2897  *   Bit flags to mark detected items.
2898  * @param[in] gre_item
2899  *   Pointer to gre_item
2900  * @param[out] error
2901  *   Pointer to error structure.
2902  *
2903  * @return
2904  *   0 on success, a negative errno value otherwise and rte_errno is set.
2905  */
2906 int
2907 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2908 				uint64_t item_flags,
2909 				const struct rte_flow_item *gre_item,
2910 				struct rte_flow_error *error)
2911 {
2912 	const rte_be32_t *mask = item->mask;
2913 	int ret = 0;
2914 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2915 	const struct rte_flow_item_gre *gre_spec;
2916 	const struct rte_flow_item_gre *gre_mask;
2917 
2918 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2919 		return rte_flow_error_set(error, ENOTSUP,
2920 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2921 					  "Multiple GRE key not support");
2922 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2923 		return rte_flow_error_set(error, ENOTSUP,
2924 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2925 					  "No preceding GRE header");
2926 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2927 		return rte_flow_error_set(error, ENOTSUP,
2928 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2929 					  "GRE key following a wrong item");
2930 	gre_mask = gre_item->mask;
2931 	if (!gre_mask)
2932 		gre_mask = &rte_flow_item_gre_mask;
2933 	gre_spec = gre_item->spec;
2934 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2935 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2936 		return rte_flow_error_set(error, EINVAL,
2937 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2938 					  "Key bit must be on");
2939 
2940 	if (!mask)
2941 		mask = &gre_key_default_mask;
2942 	ret = mlx5_flow_item_acceptable
2943 		(item, (const uint8_t *)mask,
2944 		 (const uint8_t *)&gre_key_default_mask,
2945 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2946 	return ret;
2947 }
2948 
2949 /**
2950  * Validate GRE optional item.
2951  *
2952  * @param[in] dev
2953  *   Pointer to the Ethernet device structure.
2954  * @param[in] item
2955  *   Item specification.
2956  * @param[in] item_flags
2957  *   Bit flags to mark detected items.
2958  * @param[in] attr
2959  *   Flow rule attributes.
2960  * @param[in] gre_item
2961  *   Pointer to gre_item
2962  * @param[out] error
2963  *   Pointer to error structure.
2964  *
2965  * @return
2966  *   0 on success, a negative errno value otherwise and rte_errno is set.
2967  */
2968 int
2969 mlx5_flow_validate_item_gre_option(struct rte_eth_dev *dev,
2970 				   const struct rte_flow_item *item,
2971 				   uint64_t item_flags,
2972 				   const struct rte_flow_attr *attr,
2973 				   const struct rte_flow_item *gre_item,
2974 				   struct rte_flow_error *error)
2975 {
2976 	const struct rte_flow_item_gre *gre_spec = gre_item->spec;
2977 	const struct rte_flow_item_gre *gre_mask = gre_item->mask;
2978 	const struct rte_flow_item_gre_opt *spec = item->spec;
2979 	const struct rte_flow_item_gre_opt *mask = item->mask;
2980 	struct mlx5_priv *priv = dev->data->dev_private;
2981 	int ret = 0;
2982 	struct rte_flow_item_gre_opt nic_mask = {
2983 		.checksum_rsvd = {
2984 			.checksum = RTE_BE16(UINT16_MAX),
2985 			.reserved1 = 0x0,
2986 		},
2987 		.key = {
2988 			.key = RTE_BE32(UINT32_MAX),
2989 		},
2990 		.sequence = {
2991 			.sequence = RTE_BE32(UINT32_MAX),
2992 		},
2993 	};
2994 
2995 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2996 		return rte_flow_error_set(error, ENOTSUP,
2997 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2998 					  "No preceding GRE header");
2999 	if (item_flags & MLX5_FLOW_LAYER_INNER)
3000 		return rte_flow_error_set(error, ENOTSUP,
3001 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3002 					  "GRE option following a wrong item");
3003 	if (!spec || !mask)
3004 		return rte_flow_error_set(error, EINVAL,
3005 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3006 					  "At least one field gre_option(checksum/key/sequence) must be specified");
3007 	if (!gre_mask)
3008 		gre_mask = &rte_flow_item_gre_mask;
3009 	if (mask->checksum_rsvd.checksum)
3010 		if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x8000)) &&
3011 				 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x8000)))
3012 			return rte_flow_error_set(error, EINVAL,
3013 						  RTE_FLOW_ERROR_TYPE_ITEM,
3014 						  item,
3015 						  "Checksum bit must be on");
3016 	if (mask->key.key)
3017 		if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
3018 				 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
3019 			return rte_flow_error_set(error, EINVAL,
3020 						  RTE_FLOW_ERROR_TYPE_ITEM,
3021 						  item, "Key bit must be on");
3022 	if (mask->sequence.sequence)
3023 		if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x1000)) &&
3024 				 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x1000)))
3025 			return rte_flow_error_set(error, EINVAL,
3026 						  RTE_FLOW_ERROR_TYPE_ITEM,
3027 						  item,
3028 						  "Sequence bit must be on");
3029 	if (mask->checksum_rsvd.checksum || mask->sequence.sequence) {
3030 		if (priv->sh->steering_format_version ==
3031 		    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
3032 		    ((attr->group || attr->transfer) &&
3033 		     !priv->sh->misc5_cap) ||
3034 		    (!(priv->sh->tunnel_header_0_1 &&
3035 		       priv->sh->tunnel_header_2_3) &&
3036 		    !attr->group && !attr->transfer))
3037 			return rte_flow_error_set(error, EINVAL,
3038 						  RTE_FLOW_ERROR_TYPE_ITEM,
3039 						  item,
3040 						  "Checksum/Sequence not supported");
3041 	}
3042 	ret = mlx5_flow_item_acceptable
3043 		(item, (const uint8_t *)mask,
3044 		 (const uint8_t *)&nic_mask,
3045 		 sizeof(struct rte_flow_item_gre_opt),
3046 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3047 	return ret;
3048 }
3049 
3050 /**
3051  * Validate GRE item.
3052  *
3053  * @param[in] item
3054  *   Item specification.
3055  * @param[in] item_flags
3056  *   Bit flags to mark detected items.
3057  * @param[in] target_protocol
3058  *   The next protocol in the previous item.
3059  * @param[out] error
3060  *   Pointer to error structure.
3061  *
3062  * @return
3063  *   0 on success, a negative errno value otherwise and rte_errno is set.
3064  */
3065 int
3066 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
3067 			    uint64_t item_flags,
3068 			    uint8_t target_protocol,
3069 			    struct rte_flow_error *error)
3070 {
3071 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
3072 	const struct rte_flow_item_gre *mask = item->mask;
3073 	int ret;
3074 	const struct rte_flow_item_gre nic_mask = {
3075 		.c_rsvd0_ver = RTE_BE16(0xB000),
3076 		.protocol = RTE_BE16(UINT16_MAX),
3077 	};
3078 
3079 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3080 		return rte_flow_error_set(error, EINVAL,
3081 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3082 					  "protocol filtering not compatible"
3083 					  " with this GRE layer");
3084 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3085 		return rte_flow_error_set(error, ENOTSUP,
3086 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3087 					  "multiple tunnel layers not"
3088 					  " supported");
3089 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3090 		return rte_flow_error_set(error, ENOTSUP,
3091 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3092 					  "L3 Layer is missing");
3093 	if (!mask)
3094 		mask = &rte_flow_item_gre_mask;
3095 	ret = mlx5_flow_item_acceptable
3096 		(item, (const uint8_t *)mask,
3097 		 (const uint8_t *)&nic_mask,
3098 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
3099 		 error);
3100 	if (ret < 0)
3101 		return ret;
3102 #ifndef HAVE_MLX5DV_DR
3103 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
3104 	if (spec && (spec->protocol & mask->protocol))
3105 		return rte_flow_error_set(error, ENOTSUP,
3106 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 					  "without MPLS support the"
3108 					  " specification cannot be used for"
3109 					  " filtering");
3110 #endif
3111 #endif
3112 	return 0;
3113 }
3114 
3115 /**
3116  * Validate Geneve item.
3117  *
3118  * @param[in] item
3119  *   Item specification.
3120  * @param[in] itemFlags
3121  *   Bit-fields that holds the items detected until now.
3122  * @param[in] enPriv
3123  *   Pointer to the private data structure.
3124  * @param[out] error
3125  *   Pointer to error structure.
3126  *
3127  * @return
3128  *   0 on success, a negative errno value otherwise and rte_errno is set.
3129  */
3130 
3131 int
3132 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
3133 			       uint64_t item_flags,
3134 			       struct rte_eth_dev *dev,
3135 			       struct rte_flow_error *error)
3136 {
3137 	struct mlx5_priv *priv = dev->data->dev_private;
3138 	const struct rte_flow_item_geneve *spec = item->spec;
3139 	const struct rte_flow_item_geneve *mask = item->mask;
3140 	int ret;
3141 	uint16_t gbhdr;
3142 	uint8_t opt_len = priv->sh->cdev->config.hca_attr.geneve_max_opt_len ?
3143 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
3144 	const struct rte_flow_item_geneve nic_mask = {
3145 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
3146 		.vni = "\xff\xff\xff",
3147 		.protocol = RTE_BE16(UINT16_MAX),
3148 	};
3149 
3150 	if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_geneve_rx)
3151 		return rte_flow_error_set(error, ENOTSUP,
3152 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3153 					  "L3 Geneve is not enabled by device"
3154 					  " parameter and/or not configured in"
3155 					  " firmware");
3156 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3157 		return rte_flow_error_set(error, ENOTSUP,
3158 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3159 					  "multiple tunnel layers not"
3160 					  " supported");
3161 	/*
3162 	 * Verify only UDPv4 is present as defined in
3163 	 * https://tools.ietf.org/html/rfc7348
3164 	 */
3165 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
3166 		return rte_flow_error_set(error, EINVAL,
3167 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3168 					  "no outer UDP layer found");
3169 	if (!mask)
3170 		mask = &rte_flow_item_geneve_mask;
3171 	ret = mlx5_flow_item_acceptable
3172 				  (item, (const uint8_t *)mask,
3173 				   (const uint8_t *)&nic_mask,
3174 				   sizeof(struct rte_flow_item_geneve),
3175 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3176 	if (ret)
3177 		return ret;
3178 	if (spec) {
3179 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
3180 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
3181 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
3182 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
3183 			return rte_flow_error_set(error, ENOTSUP,
3184 						  RTE_FLOW_ERROR_TYPE_ITEM,
3185 						  item,
3186 						  "Geneve protocol unsupported"
3187 						  " fields are being used");
3188 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
3189 			return rte_flow_error_set
3190 					(error, ENOTSUP,
3191 					 RTE_FLOW_ERROR_TYPE_ITEM,
3192 					 item,
3193 					 "Unsupported Geneve options length");
3194 	}
3195 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
3196 		return rte_flow_error_set
3197 				    (error, ENOTSUP,
3198 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
3199 				     "Geneve tunnel must be fully defined");
3200 	return 0;
3201 }
3202 
3203 /**
3204  * Validate Geneve TLV option item.
3205  *
3206  * @param[in] item
3207  *   Item specification.
3208  * @param[in] last_item
3209  *   Previous validated item in the pattern items.
3210  * @param[in] geneve_item
3211  *   Previous GENEVE item specification.
3212  * @param[in] dev
3213  *   Pointer to the rte_eth_dev structure.
3214  * @param[out] error
3215  *   Pointer to error structure.
3216  *
3217  * @return
3218  *   0 on success, a negative errno value otherwise and rte_errno is set.
3219  */
3220 int
3221 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
3222 				   uint64_t last_item,
3223 				   const struct rte_flow_item *geneve_item,
3224 				   struct rte_eth_dev *dev,
3225 				   struct rte_flow_error *error)
3226 {
3227 	struct mlx5_priv *priv = dev->data->dev_private;
3228 	struct mlx5_dev_ctx_shared *sh = priv->sh;
3229 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
3230 	struct mlx5_hca_attr *hca_attr = &sh->cdev->config.hca_attr;
3231 	uint8_t data_max_supported =
3232 			hca_attr->max_geneve_tlv_option_data_len * 4;
3233 	const struct rte_flow_item_geneve *geneve_spec;
3234 	const struct rte_flow_item_geneve *geneve_mask;
3235 	const struct rte_flow_item_geneve_opt *spec = item->spec;
3236 	const struct rte_flow_item_geneve_opt *mask = item->mask;
3237 	unsigned int i;
3238 	unsigned int data_len;
3239 	uint8_t tlv_option_len;
3240 	uint16_t optlen_m, optlen_v;
3241 	const struct rte_flow_item_geneve_opt full_mask = {
3242 		.option_class = RTE_BE16(0xffff),
3243 		.option_type = 0xff,
3244 		.option_len = 0x1f,
3245 	};
3246 
3247 	if (!mask)
3248 		mask = &rte_flow_item_geneve_opt_mask;
3249 	if (!spec)
3250 		return rte_flow_error_set
3251 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3252 			"Geneve TLV opt class/type/length must be specified");
3253 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3254 		return rte_flow_error_set
3255 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3256 			"Geneve TLV opt length exceeds the limit (31)");
3257 	/* Check if class type and length masks are full. */
3258 	if (full_mask.option_class != mask->option_class ||
3259 	    full_mask.option_type != mask->option_type ||
3260 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
3261 		return rte_flow_error_set
3262 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3263 			"Geneve TLV opt class/type/length masks must be full");
3264 	/* Check if length is supported */
3265 	if ((uint32_t)spec->option_len >
3266 			hca_attr->max_geneve_tlv_option_data_len)
3267 		return rte_flow_error_set
3268 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3269 			"Geneve TLV opt length not supported");
3270 	if (hca_attr->max_geneve_tlv_options > 1)
3271 		DRV_LOG(DEBUG,
3272 			"max_geneve_tlv_options supports more than 1 option");
3273 	/* Check GENEVE item preceding. */
3274 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3275 		return rte_flow_error_set
3276 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3277 			"Geneve opt item must be preceded with Geneve item");
3278 	geneve_spec = geneve_item->spec;
3279 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3280 					  &rte_flow_item_geneve_mask;
3281 	/* Check if GENEVE TLV option size doesn't exceed option length */
3282 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3283 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3284 		tlv_option_len = spec->option_len & mask->option_len;
3285 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3286 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3287 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3288 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3289 		if ((optlen_v & optlen_m) <= tlv_option_len)
3290 			return rte_flow_error_set
3291 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3292 				 "GENEVE TLV option length exceeds optlen");
3293 	}
3294 	/* Check if length is 0 or data is 0. */
3295 	if (spec->data == NULL || spec->option_len == 0)
3296 		return rte_flow_error_set
3297 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3298 			"Geneve TLV opt with zero data/length not supported");
3299 	/* Check not all data & mask are 0. */
3300 	data_len = spec->option_len * 4;
3301 	if (mask->data == NULL) {
3302 		for (i = 0; i < data_len; i++)
3303 			if (spec->data[i])
3304 				break;
3305 		if (i == data_len)
3306 			return rte_flow_error_set(error, ENOTSUP,
3307 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3308 				"Can't match on Geneve option data 0");
3309 	} else {
3310 		for (i = 0; i < data_len; i++)
3311 			if (spec->data[i] & mask->data[i])
3312 				break;
3313 		if (i == data_len)
3314 			return rte_flow_error_set(error, ENOTSUP,
3315 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3316 				"Can't match on Geneve option data and mask 0");
3317 		/* Check data mask supported. */
3318 		for (i = data_max_supported; i < data_len ; i++)
3319 			if (mask->data[i])
3320 				return rte_flow_error_set(error, ENOTSUP,
3321 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3322 					"Data mask is of unsupported size");
3323 	}
3324 	/* Check GENEVE option is supported in NIC. */
3325 	if (!hca_attr->geneve_tlv_opt)
3326 		return rte_flow_error_set
3327 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3328 			"Geneve TLV opt not supported");
3329 	/* Check if we already have geneve option with different type/class. */
3330 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3331 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3332 	if (geneve_opt_resource != NULL)
3333 		if (geneve_opt_resource->option_class != spec->option_class ||
3334 		    geneve_opt_resource->option_type != spec->option_type ||
3335 		    geneve_opt_resource->length != spec->option_len) {
3336 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3337 			return rte_flow_error_set(error, ENOTSUP,
3338 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3339 				"Only one Geneve TLV option supported");
3340 		}
3341 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3342 	return 0;
3343 }
3344 
3345 /**
3346  * Validate MPLS item.
3347  *
3348  * @param[in] dev
3349  *   Pointer to the rte_eth_dev structure.
3350  * @param[in] item
3351  *   Item specification.
3352  * @param[in] item_flags
3353  *   Bit-fields that holds the items detected until now.
3354  * @param[in] prev_layer
3355  *   The protocol layer indicated in previous item.
3356  * @param[out] error
3357  *   Pointer to error structure.
3358  *
3359  * @return
3360  *   0 on success, a negative errno value otherwise and rte_errno is set.
3361  */
3362 int
3363 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3364 			     const struct rte_flow_item *item __rte_unused,
3365 			     uint64_t item_flags __rte_unused,
3366 			     uint64_t prev_layer __rte_unused,
3367 			     struct rte_flow_error *error)
3368 {
3369 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3370 	const struct rte_flow_item_mpls *mask = item->mask;
3371 	struct mlx5_priv *priv = dev->data->dev_private;
3372 	int ret;
3373 
3374 	if (!priv->sh->dev_cap.mpls_en)
3375 		return rte_flow_error_set(error, ENOTSUP,
3376 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3377 					  "MPLS not supported or"
3378 					  " disabled in firmware"
3379 					  " configuration.");
3380 	/* MPLS over UDP, GRE is allowed */
3381 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3382 			    MLX5_FLOW_LAYER_GRE |
3383 			    MLX5_FLOW_LAYER_GRE_KEY)))
3384 		return rte_flow_error_set(error, EINVAL,
3385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3386 					  "protocol filtering not compatible"
3387 					  " with MPLS layer");
3388 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3389 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3390 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3391 		return rte_flow_error_set(error, ENOTSUP,
3392 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3393 					  "multiple tunnel layers not"
3394 					  " supported");
3395 	if (!mask)
3396 		mask = &rte_flow_item_mpls_mask;
3397 	ret = mlx5_flow_item_acceptable
3398 		(item, (const uint8_t *)mask,
3399 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3400 		 sizeof(struct rte_flow_item_mpls),
3401 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3402 	if (ret < 0)
3403 		return ret;
3404 	return 0;
3405 #else
3406 	return rte_flow_error_set(error, ENOTSUP,
3407 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3408 				  "MPLS is not supported by Verbs, please"
3409 				  " update.");
3410 #endif
3411 }
3412 
3413 /**
3414  * Validate NVGRE item.
3415  *
3416  * @param[in] item
3417  *   Item specification.
3418  * @param[in] item_flags
3419  *   Bit flags to mark detected items.
3420  * @param[in] target_protocol
3421  *   The next protocol in the previous item.
3422  * @param[out] error
3423  *   Pointer to error structure.
3424  *
3425  * @return
3426  *   0 on success, a negative errno value otherwise and rte_errno is set.
3427  */
3428 int
3429 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3430 			      uint64_t item_flags,
3431 			      uint8_t target_protocol,
3432 			      struct rte_flow_error *error)
3433 {
3434 	const struct rte_flow_item_nvgre *mask = item->mask;
3435 	int ret;
3436 
3437 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3438 		return rte_flow_error_set(error, EINVAL,
3439 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3440 					  "protocol filtering not compatible"
3441 					  " with this GRE layer");
3442 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3443 		return rte_flow_error_set(error, ENOTSUP,
3444 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3445 					  "multiple tunnel layers not"
3446 					  " supported");
3447 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3448 		return rte_flow_error_set(error, ENOTSUP,
3449 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3450 					  "L3 Layer is missing");
3451 	if (!mask)
3452 		mask = &rte_flow_item_nvgre_mask;
3453 	ret = mlx5_flow_item_acceptable
3454 		(item, (const uint8_t *)mask,
3455 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3456 		 sizeof(struct rte_flow_item_nvgre),
3457 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3458 	if (ret < 0)
3459 		return ret;
3460 	return 0;
3461 }
3462 
3463 /**
3464  * Validate eCPRI item.
3465  *
3466  * @param[in] item
3467  *   Item specification.
3468  * @param[in] item_flags
3469  *   Bit-fields that holds the items detected until now.
3470  * @param[in] last_item
3471  *   Previous validated item in the pattern items.
3472  * @param[in] ether_type
3473  *   Type in the ethernet layer header (including dot1q).
3474  * @param[in] acc_mask
3475  *   Acceptable mask, if NULL default internal default mask
3476  *   will be used to check whether item fields are supported.
3477  * @param[out] error
3478  *   Pointer to error structure.
3479  *
3480  * @return
3481  *   0 on success, a negative errno value otherwise and rte_errno is set.
3482  */
3483 int
3484 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3485 			      uint64_t item_flags,
3486 			      uint64_t last_item,
3487 			      uint16_t ether_type,
3488 			      const struct rte_flow_item_ecpri *acc_mask,
3489 			      struct rte_flow_error *error)
3490 {
3491 	const struct rte_flow_item_ecpri *mask = item->mask;
3492 	const struct rte_flow_item_ecpri nic_mask = {
3493 		.hdr = {
3494 			.common = {
3495 				.u32 =
3496 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3497 					.type = 0xFF,
3498 					}).u32),
3499 			},
3500 			.dummy[0] = 0xFFFFFFFF,
3501 		},
3502 	};
3503 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3504 					MLX5_FLOW_LAYER_OUTER_VLAN);
3505 	struct rte_flow_item_ecpri mask_lo;
3506 
3507 	if (!(last_item & outer_l2_vlan) &&
3508 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3509 		return rte_flow_error_set(error, EINVAL,
3510 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3511 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3512 	if ((last_item & outer_l2_vlan) && ether_type &&
3513 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3514 		return rte_flow_error_set(error, EINVAL,
3515 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3516 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3517 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3518 		return rte_flow_error_set(error, EINVAL,
3519 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3520 					  "eCPRI with tunnel is not supported right now");
3521 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3522 		return rte_flow_error_set(error, ENOTSUP,
3523 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3524 					  "multiple L3 layers not supported");
3525 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3526 		return rte_flow_error_set(error, EINVAL,
3527 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3528 					  "eCPRI cannot coexist with a TCP layer");
3529 	/* In specification, eCPRI could be over UDP layer. */
3530 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3531 		return rte_flow_error_set(error, EINVAL,
3532 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3533 					  "eCPRI over UDP layer is not yet supported right now");
3534 	/* Mask for type field in common header could be zero. */
3535 	if (!mask)
3536 		mask = &rte_flow_item_ecpri_mask;
3537 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3538 	/* Input mask is in big-endian format. */
3539 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3540 		return rte_flow_error_set(error, EINVAL,
3541 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3542 					  "partial mask is not supported for protocol");
3543 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3544 		return rte_flow_error_set(error, EINVAL,
3545 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3546 					  "message header mask must be after a type mask");
3547 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3548 					 acc_mask ? (const uint8_t *)acc_mask
3549 						  : (const uint8_t *)&nic_mask,
3550 					 sizeof(struct rte_flow_item_ecpri),
3551 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3552 }
3553 
3554 static int
3555 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3556 		   const struct rte_flow_attr *attr __rte_unused,
3557 		   const struct rte_flow_item items[] __rte_unused,
3558 		   const struct rte_flow_action actions[] __rte_unused,
3559 		   bool external __rte_unused,
3560 		   int hairpin __rte_unused,
3561 		   struct rte_flow_error *error)
3562 {
3563 	return rte_flow_error_set(error, ENOTSUP,
3564 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3565 }
3566 
3567 static struct mlx5_flow *
3568 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3569 		  const struct rte_flow_attr *attr __rte_unused,
3570 		  const struct rte_flow_item items[] __rte_unused,
3571 		  const struct rte_flow_action actions[] __rte_unused,
3572 		  struct rte_flow_error *error)
3573 {
3574 	rte_flow_error_set(error, ENOTSUP,
3575 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3576 	return NULL;
3577 }
3578 
3579 static int
3580 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3581 		    struct mlx5_flow *dev_flow __rte_unused,
3582 		    const struct rte_flow_attr *attr __rte_unused,
3583 		    const struct rte_flow_item items[] __rte_unused,
3584 		    const struct rte_flow_action actions[] __rte_unused,
3585 		    struct rte_flow_error *error)
3586 {
3587 	return rte_flow_error_set(error, ENOTSUP,
3588 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3589 }
3590 
3591 static int
3592 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3593 		struct rte_flow *flow __rte_unused,
3594 		struct rte_flow_error *error)
3595 {
3596 	return rte_flow_error_set(error, ENOTSUP,
3597 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3598 }
3599 
3600 static void
3601 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3602 		 struct rte_flow *flow __rte_unused)
3603 {
3604 }
3605 
3606 static void
3607 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3608 		  struct rte_flow *flow __rte_unused)
3609 {
3610 }
3611 
3612 static int
3613 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3614 		struct rte_flow *flow __rte_unused,
3615 		const struct rte_flow_action *actions __rte_unused,
3616 		void *data __rte_unused,
3617 		struct rte_flow_error *error)
3618 {
3619 	return rte_flow_error_set(error, ENOTSUP,
3620 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3621 }
3622 
3623 static int
3624 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3625 		      uint32_t domains __rte_unused,
3626 		      uint32_t flags __rte_unused)
3627 {
3628 	return 0;
3629 }
3630 
3631 /* Void driver to protect from null pointer reference. */
3632 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3633 	.validate = flow_null_validate,
3634 	.prepare = flow_null_prepare,
3635 	.translate = flow_null_translate,
3636 	.apply = flow_null_apply,
3637 	.remove = flow_null_remove,
3638 	.destroy = flow_null_destroy,
3639 	.query = flow_null_query,
3640 	.sync_domain = flow_null_sync_domain,
3641 };
3642 
3643 /**
3644  * Select flow driver type according to flow attributes and device
3645  * configuration.
3646  *
3647  * @param[in] dev
3648  *   Pointer to the dev structure.
3649  * @param[in] attr
3650  *   Pointer to the flow attributes.
3651  *
3652  * @return
3653  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3654  */
3655 static enum mlx5_flow_drv_type
3656 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3657 {
3658 	struct mlx5_priv *priv = dev->data->dev_private;
3659 	/* The OS can determine first a specific flow type (DV, VERBS) */
3660 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3661 
3662 	if (type != MLX5_FLOW_TYPE_MAX)
3663 		return type;
3664 	/*
3665 	 * Currently when dv_flow_en == 2, only HW steering engine is
3666 	 * supported. New engines can also be chosen here if ready.
3667 	 */
3668 	if (priv->sh->config.dv_flow_en == 2)
3669 		return MLX5_FLOW_TYPE_HW;
3670 	/* If no OS specific type - continue with DV/VERBS selection */
3671 	if (attr->transfer && priv->sh->config.dv_esw_en)
3672 		type = MLX5_FLOW_TYPE_DV;
3673 	if (!attr->transfer)
3674 		type = priv->sh->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3675 						     MLX5_FLOW_TYPE_VERBS;
3676 	return type;
3677 }
3678 
3679 #define flow_get_drv_ops(type) flow_drv_ops[type]
3680 
3681 /**
3682  * Flow driver validation API. This abstracts calling driver specific functions.
3683  * The type of flow driver is determined according to flow attributes.
3684  *
3685  * @param[in] dev
3686  *   Pointer to the dev structure.
3687  * @param[in] attr
3688  *   Pointer to the flow attributes.
3689  * @param[in] items
3690  *   Pointer to the list of items.
3691  * @param[in] actions
3692  *   Pointer to the list of actions.
3693  * @param[in] external
3694  *   This flow rule is created by request external to PMD.
3695  * @param[in] hairpin
3696  *   Number of hairpin TX actions, 0 means classic flow.
3697  * @param[out] error
3698  *   Pointer to the error structure.
3699  *
3700  * @return
3701  *   0 on success, a negative errno value otherwise and rte_errno is set.
3702  */
3703 static inline int
3704 flow_drv_validate(struct rte_eth_dev *dev,
3705 		  const struct rte_flow_attr *attr,
3706 		  const struct rte_flow_item items[],
3707 		  const struct rte_flow_action actions[],
3708 		  bool external, int hairpin, struct rte_flow_error *error)
3709 {
3710 	const struct mlx5_flow_driver_ops *fops;
3711 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3712 
3713 	fops = flow_get_drv_ops(type);
3714 	return fops->validate(dev, attr, items, actions, external,
3715 			      hairpin, error);
3716 }
3717 
3718 /**
3719  * Flow driver preparation API. This abstracts calling driver specific
3720  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3721  * calculates the size of memory required for device flow, allocates the memory,
3722  * initializes the device flow and returns the pointer.
3723  *
3724  * @note
3725  *   This function initializes device flow structure such as dv or verbs in
3726  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3727  *   rest. For example, adding returning device flow to flow->dev_flow list and
3728  *   setting backward reference to the flow should be done out of this function.
3729  *   layers field is not filled either.
3730  *
3731  * @param[in] dev
3732  *   Pointer to the dev structure.
3733  * @param[in] attr
3734  *   Pointer to the flow attributes.
3735  * @param[in] items
3736  *   Pointer to the list of items.
3737  * @param[in] actions
3738  *   Pointer to the list of actions.
3739  * @param[in] flow_idx
3740  *   This memory pool index to the flow.
3741  * @param[out] error
3742  *   Pointer to the error structure.
3743  *
3744  * @return
3745  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3746  */
3747 static inline struct mlx5_flow *
3748 flow_drv_prepare(struct rte_eth_dev *dev,
3749 		 const struct rte_flow *flow,
3750 		 const struct rte_flow_attr *attr,
3751 		 const struct rte_flow_item items[],
3752 		 const struct rte_flow_action actions[],
3753 		 uint32_t flow_idx,
3754 		 struct rte_flow_error *error)
3755 {
3756 	const struct mlx5_flow_driver_ops *fops;
3757 	enum mlx5_flow_drv_type type = flow->drv_type;
3758 	struct mlx5_flow *mlx5_flow = NULL;
3759 
3760 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3761 	fops = flow_get_drv_ops(type);
3762 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3763 	if (mlx5_flow)
3764 		mlx5_flow->flow_idx = flow_idx;
3765 	return mlx5_flow;
3766 }
3767 
3768 /**
3769  * Flow driver translation API. This abstracts calling driver specific
3770  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3771  * translates a generic flow into a driver flow. flow_drv_prepare() must
3772  * precede.
3773  *
3774  * @note
3775  *   dev_flow->layers could be filled as a result of parsing during translation
3776  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3777  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3778  *   flow->actions could be overwritten even though all the expanded dev_flows
3779  *   have the same actions.
3780  *
3781  * @param[in] dev
3782  *   Pointer to the rte dev structure.
3783  * @param[in, out] dev_flow
3784  *   Pointer to the mlx5 flow.
3785  * @param[in] attr
3786  *   Pointer to the flow attributes.
3787  * @param[in] items
3788  *   Pointer to the list of items.
3789  * @param[in] actions
3790  *   Pointer to the list of actions.
3791  * @param[out] error
3792  *   Pointer to the error structure.
3793  *
3794  * @return
3795  *   0 on success, a negative errno value otherwise and rte_errno is set.
3796  */
3797 static inline int
3798 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3799 		   const struct rte_flow_attr *attr,
3800 		   const struct rte_flow_item items[],
3801 		   const struct rte_flow_action actions[],
3802 		   struct rte_flow_error *error)
3803 {
3804 	const struct mlx5_flow_driver_ops *fops;
3805 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3806 
3807 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3808 	fops = flow_get_drv_ops(type);
3809 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3810 }
3811 
3812 /**
3813  * Flow driver apply API. This abstracts calling driver specific functions.
3814  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3815  * translated driver flows on to device. flow_drv_translate() must precede.
3816  *
3817  * @param[in] dev
3818  *   Pointer to Ethernet device structure.
3819  * @param[in, out] flow
3820  *   Pointer to flow structure.
3821  * @param[out] error
3822  *   Pointer to error structure.
3823  *
3824  * @return
3825  *   0 on success, a negative errno value otherwise and rte_errno is set.
3826  */
3827 static inline int
3828 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3829 	       struct rte_flow_error *error)
3830 {
3831 	const struct mlx5_flow_driver_ops *fops;
3832 	enum mlx5_flow_drv_type type = flow->drv_type;
3833 
3834 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3835 	fops = flow_get_drv_ops(type);
3836 	return fops->apply(dev, flow, error);
3837 }
3838 
3839 /**
3840  * Flow driver destroy API. This abstracts calling driver specific functions.
3841  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3842  * on device and releases resources of the flow.
3843  *
3844  * @param[in] dev
3845  *   Pointer to Ethernet device.
3846  * @param[in, out] flow
3847  *   Pointer to flow structure.
3848  */
3849 static inline void
3850 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3851 {
3852 	const struct mlx5_flow_driver_ops *fops;
3853 	enum mlx5_flow_drv_type type = flow->drv_type;
3854 
3855 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3856 	fops = flow_get_drv_ops(type);
3857 	fops->destroy(dev, flow);
3858 }
3859 
3860 /**
3861  * Flow driver find RSS policy tbl API. This abstracts calling driver
3862  * specific functions. Parent flow (rte_flow) should have driver
3863  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3864  *
3865  * @param[in] dev
3866  *   Pointer to Ethernet device.
3867  * @param[in, out] flow
3868  *   Pointer to flow structure.
3869  * @param[in] policy
3870  *   Pointer to meter policy table.
3871  * @param[in] rss_desc
3872  *   Pointer to rss_desc
3873  */
3874 static struct mlx5_flow_meter_sub_policy *
3875 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3876 		struct rte_flow *flow,
3877 		struct mlx5_flow_meter_policy *policy,
3878 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3879 {
3880 	const struct mlx5_flow_driver_ops *fops;
3881 	enum mlx5_flow_drv_type type = flow->drv_type;
3882 
3883 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3884 	fops = flow_get_drv_ops(type);
3885 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3886 }
3887 
3888 /**
3889  * Flow driver color tag rule API. This abstracts calling driver
3890  * specific functions. Parent flow (rte_flow) should have driver
3891  * type (drv_type). It will create the color tag rules in hierarchy meter.
3892  *
3893  * @param[in] dev
3894  *   Pointer to Ethernet device.
3895  * @param[in, out] flow
3896  *   Pointer to flow structure.
3897  * @param[in] fm
3898  *   Pointer to flow meter structure.
3899  * @param[in] src_port
3900  *   The src port this extra rule should use.
3901  * @param[in] item
3902  *   The src port id match item.
3903  * @param[out] error
3904  *   Pointer to error structure.
3905  */
3906 static int
3907 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3908 		struct rte_flow *flow,
3909 		struct mlx5_flow_meter_info *fm,
3910 		int32_t src_port,
3911 		const struct rte_flow_item *item,
3912 		struct rte_flow_error *error)
3913 {
3914 	const struct mlx5_flow_driver_ops *fops;
3915 	enum mlx5_flow_drv_type type = flow->drv_type;
3916 
3917 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3918 	fops = flow_get_drv_ops(type);
3919 	return fops->meter_hierarchy_rule_create(dev, fm,
3920 						src_port, item, error);
3921 }
3922 
3923 /**
3924  * Get RSS action from the action list.
3925  *
3926  * @param[in] dev
3927  *   Pointer to Ethernet device.
3928  * @param[in] actions
3929  *   Pointer to the list of actions.
3930  * @param[in] flow
3931  *   Parent flow structure pointer.
3932  *
3933  * @return
3934  *   Pointer to the RSS action if exist, else return NULL.
3935  */
3936 static const struct rte_flow_action_rss*
3937 flow_get_rss_action(struct rte_eth_dev *dev,
3938 		    const struct rte_flow_action actions[])
3939 {
3940 	struct mlx5_priv *priv = dev->data->dev_private;
3941 	const struct rte_flow_action_rss *rss = NULL;
3942 	struct mlx5_meter_policy_action_container *acg;
3943 	struct mlx5_meter_policy_action_container *acy;
3944 
3945 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3946 		switch (actions->type) {
3947 		case RTE_FLOW_ACTION_TYPE_RSS:
3948 			rss = actions->conf;
3949 			break;
3950 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3951 		{
3952 			const struct rte_flow_action_sample *sample =
3953 								actions->conf;
3954 			const struct rte_flow_action *act = sample->actions;
3955 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3956 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3957 					rss = act->conf;
3958 			break;
3959 		}
3960 		case RTE_FLOW_ACTION_TYPE_METER:
3961 		{
3962 			uint32_t mtr_idx;
3963 			struct mlx5_flow_meter_info *fm;
3964 			struct mlx5_flow_meter_policy *policy;
3965 			const struct rte_flow_action_meter *mtr = actions->conf;
3966 
3967 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3968 			if (fm && !fm->def_policy) {
3969 				policy = mlx5_flow_meter_policy_find(dev,
3970 						fm->policy_id, NULL);
3971 				MLX5_ASSERT(policy);
3972 				if (policy->is_hierarchy) {
3973 					policy =
3974 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3975 									policy);
3976 					if (!policy)
3977 						return NULL;
3978 				}
3979 				if (policy->is_rss) {
3980 					acg =
3981 					&policy->act_cnt[RTE_COLOR_GREEN];
3982 					acy =
3983 					&policy->act_cnt[RTE_COLOR_YELLOW];
3984 					if (acg->fate_action ==
3985 					    MLX5_FLOW_FATE_SHARED_RSS)
3986 						rss = acg->rss->conf;
3987 					else if (acy->fate_action ==
3988 						 MLX5_FLOW_FATE_SHARED_RSS)
3989 						rss = acy->rss->conf;
3990 				}
3991 			}
3992 			break;
3993 		}
3994 		default:
3995 			break;
3996 		}
3997 	}
3998 	return rss;
3999 }
4000 
4001 /**
4002  * Get ASO age action by index.
4003  *
4004  * @param[in] dev
4005  *   Pointer to the Ethernet device structure.
4006  * @param[in] age_idx
4007  *   Index to the ASO age action.
4008  *
4009  * @return
4010  *   The specified ASO age action.
4011  */
4012 struct mlx5_aso_age_action*
4013 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
4014 {
4015 	uint16_t pool_idx = age_idx & UINT16_MAX;
4016 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
4017 	struct mlx5_priv *priv = dev->data->dev_private;
4018 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
4019 	struct mlx5_aso_age_pool *pool;
4020 
4021 	rte_rwlock_read_lock(&mng->resize_rwl);
4022 	pool = mng->pools[pool_idx];
4023 	rte_rwlock_read_unlock(&mng->resize_rwl);
4024 	return &pool->actions[offset - 1];
4025 }
4026 
4027 /* maps indirect action to translated direct in some actions array */
4028 struct mlx5_translated_action_handle {
4029 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
4030 	int index; /**< Index in related array of rte_flow_action. */
4031 };
4032 
4033 /**
4034  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
4035  * direct action if translation possible.
4036  * This functionality used to run same execution path for both direct and
4037  * indirect actions on flow create. All necessary preparations for indirect
4038  * action handling should be performed on *handle* actions list returned
4039  * from this call.
4040  *
4041  * @param[in] dev
4042  *   Pointer to Ethernet device.
4043  * @param[in] actions
4044  *   List of actions to translate.
4045  * @param[out] handle
4046  *   List to store translated indirect action object handles.
4047  * @param[in, out] indir_n
4048  *   Size of *handle* array. On return should be updated with number of
4049  *   indirect actions retrieved from the *actions* list.
4050  * @param[out] translated_actions
4051  *   List of actions where all indirect actions were translated to direct
4052  *   if possible. NULL if no translation took place.
4053  * @param[out] error
4054  *   Pointer to the error structure.
4055  *
4056  * @return
4057  *   0 on success, a negative errno value otherwise and rte_errno is set.
4058  */
4059 static int
4060 flow_action_handles_translate(struct rte_eth_dev *dev,
4061 			      const struct rte_flow_action actions[],
4062 			      struct mlx5_translated_action_handle *handle,
4063 			      int *indir_n,
4064 			      struct rte_flow_action **translated_actions,
4065 			      struct rte_flow_error *error)
4066 {
4067 	struct mlx5_priv *priv = dev->data->dev_private;
4068 	struct rte_flow_action *translated = NULL;
4069 	size_t actions_size;
4070 	int n;
4071 	int copied_n = 0;
4072 	struct mlx5_translated_action_handle *handle_end = NULL;
4073 
4074 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
4075 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
4076 			continue;
4077 		if (copied_n == *indir_n) {
4078 			return rte_flow_error_set
4079 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
4080 				 NULL, "too many shared actions");
4081 		}
4082 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
4083 			   sizeof(actions[n].conf));
4084 		handle[copied_n].index = n;
4085 		copied_n++;
4086 	}
4087 	n++;
4088 	*indir_n = copied_n;
4089 	if (!copied_n)
4090 		return 0;
4091 	actions_size = sizeof(struct rte_flow_action) * n;
4092 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
4093 	if (!translated) {
4094 		rte_errno = ENOMEM;
4095 		return -ENOMEM;
4096 	}
4097 	memcpy(translated, actions, actions_size);
4098 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
4099 		struct mlx5_shared_action_rss *shared_rss;
4100 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
4101 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
4102 		uint32_t idx = act_idx &
4103 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
4104 
4105 		switch (type) {
4106 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
4107 			shared_rss = mlx5_ipool_get
4108 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
4109 			translated[handle->index].type =
4110 				RTE_FLOW_ACTION_TYPE_RSS;
4111 			translated[handle->index].conf =
4112 				&shared_rss->origin;
4113 			break;
4114 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
4115 			translated[handle->index].type =
4116 						(enum rte_flow_action_type)
4117 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
4118 			translated[handle->index].conf = (void *)(uintptr_t)idx;
4119 			break;
4120 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
4121 			if (priv->sh->flow_hit_aso_en) {
4122 				translated[handle->index].type =
4123 					(enum rte_flow_action_type)
4124 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
4125 				translated[handle->index].conf =
4126 							 (void *)(uintptr_t)idx;
4127 				break;
4128 			}
4129 			/* Fall-through */
4130 		case MLX5_INDIRECT_ACTION_TYPE_CT:
4131 			if (priv->sh->ct_aso_en) {
4132 				translated[handle->index].type =
4133 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
4134 				translated[handle->index].conf =
4135 							 (void *)(uintptr_t)idx;
4136 				break;
4137 			}
4138 			/* Fall-through */
4139 		default:
4140 			mlx5_free(translated);
4141 			return rte_flow_error_set
4142 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
4143 				 NULL, "invalid indirect action type");
4144 		}
4145 	}
4146 	*translated_actions = translated;
4147 	return 0;
4148 }
4149 
4150 /**
4151  * Get Shared RSS action from the action list.
4152  *
4153  * @param[in] dev
4154  *   Pointer to Ethernet device.
4155  * @param[in] shared
4156  *   Pointer to the list of actions.
4157  * @param[in] shared_n
4158  *   Actions list length.
4159  *
4160  * @return
4161  *   The MLX5 RSS action ID if exists, otherwise return 0.
4162  */
4163 static uint32_t
4164 flow_get_shared_rss_action(struct rte_eth_dev *dev,
4165 			   struct mlx5_translated_action_handle *handle,
4166 			   int shared_n)
4167 {
4168 	struct mlx5_translated_action_handle *handle_end;
4169 	struct mlx5_priv *priv = dev->data->dev_private;
4170 	struct mlx5_shared_action_rss *shared_rss;
4171 
4172 
4173 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
4174 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
4175 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
4176 		uint32_t idx = act_idx &
4177 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
4178 		switch (type) {
4179 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
4180 			shared_rss = mlx5_ipool_get
4181 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
4182 									   idx);
4183 			__atomic_add_fetch(&shared_rss->refcnt, 1,
4184 					   __ATOMIC_RELAXED);
4185 			return idx;
4186 		default:
4187 			break;
4188 		}
4189 	}
4190 	return 0;
4191 }
4192 
4193 static unsigned int
4194 find_graph_root(uint32_t rss_level)
4195 {
4196 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
4197 			       MLX5_EXPANSION_ROOT_OUTER;
4198 }
4199 
4200 /**
4201  *  Get layer flags from the prefix flow.
4202  *
4203  *  Some flows may be split to several subflows, the prefix subflow gets the
4204  *  match items and the suffix sub flow gets the actions.
4205  *  Some actions need the user defined match item flags to get the detail for
4206  *  the action.
4207  *  This function helps the suffix flow to get the item layer flags from prefix
4208  *  subflow.
4209  *
4210  * @param[in] dev_flow
4211  *   Pointer the created prefix subflow.
4212  *
4213  * @return
4214  *   The layers get from prefix subflow.
4215  */
4216 static inline uint64_t
4217 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
4218 {
4219 	uint64_t layers = 0;
4220 
4221 	/*
4222 	 * Layers bits could be localization, but usually the compiler will
4223 	 * help to do the optimization work for source code.
4224 	 * If no decap actions, use the layers directly.
4225 	 */
4226 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
4227 		return dev_flow->handle->layers;
4228 	/* Convert L3 layers with decap action. */
4229 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
4230 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4231 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
4232 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4233 	/* Convert L4 layers with decap action.  */
4234 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
4235 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
4236 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
4237 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
4238 	return layers;
4239 }
4240 
4241 /**
4242  * Get metadata split action information.
4243  *
4244  * @param[in] actions
4245  *   Pointer to the list of actions.
4246  * @param[out] qrss
4247  *   Pointer to the return pointer.
4248  * @param[out] qrss_type
4249  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
4250  *   if no QUEUE/RSS is found.
4251  * @param[out] encap_idx
4252  *   Pointer to the index of the encap action if exists, otherwise the last
4253  *   action index.
4254  *
4255  * @return
4256  *   Total number of actions.
4257  */
4258 static int
4259 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4260 				       const struct rte_flow_action **qrss,
4261 				       int *encap_idx)
4262 {
4263 	const struct rte_flow_action_raw_encap *raw_encap;
4264 	int actions_n = 0;
4265 	int raw_decap_idx = -1;
4266 
4267 	*encap_idx = -1;
4268 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4269 		switch (actions->type) {
4270 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4271 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4272 			*encap_idx = actions_n;
4273 			break;
4274 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4275 			raw_decap_idx = actions_n;
4276 			break;
4277 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4278 			raw_encap = actions->conf;
4279 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4280 				*encap_idx = raw_decap_idx != -1 ?
4281 						      raw_decap_idx : actions_n;
4282 			break;
4283 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4284 		case RTE_FLOW_ACTION_TYPE_RSS:
4285 			*qrss = actions;
4286 			break;
4287 		default:
4288 			break;
4289 		}
4290 		actions_n++;
4291 	}
4292 	if (*encap_idx == -1)
4293 		*encap_idx = actions_n;
4294 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4295 	return actions_n + 1;
4296 }
4297 
4298 /**
4299  * Check if the action will change packet.
4300  *
4301  * @param dev
4302  *   Pointer to Ethernet device.
4303  * @param[in] type
4304  *   action type.
4305  *
4306  * @return
4307  *   true if action will change packet, false otherwise.
4308  */
4309 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4310 					  enum rte_flow_action_type type)
4311 {
4312 	struct mlx5_priv *priv = dev->data->dev_private;
4313 
4314 	switch (type) {
4315 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4316 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4317 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4318 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4319 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4320 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4321 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4322 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4323 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4324 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4325 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4326 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4327 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4328 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4329 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4330 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4331 	case RTE_FLOW_ACTION_TYPE_SET_META:
4332 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4333 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4334 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4335 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4336 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4337 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4338 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4339 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4340 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4341 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4342 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4343 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4344 		return true;
4345 	case RTE_FLOW_ACTION_TYPE_FLAG:
4346 	case RTE_FLOW_ACTION_TYPE_MARK:
4347 		if (priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4348 			return true;
4349 		else
4350 			return false;
4351 	default:
4352 		return false;
4353 	}
4354 }
4355 
4356 /**
4357  * Check meter action from the action list.
4358  *
4359  * @param dev
4360  *   Pointer to Ethernet device.
4361  * @param[in] actions
4362  *   Pointer to the list of actions.
4363  * @param[out] has_mtr
4364  *   Pointer to the meter exist flag.
4365  * @param[out] has_modify
4366  *   Pointer to the flag showing there's packet change action.
4367  * @param[out] meter_id
4368  *   Pointer to the meter id.
4369  *
4370  * @return
4371  *   Total number of actions.
4372  */
4373 static int
4374 flow_check_meter_action(struct rte_eth_dev *dev,
4375 			const struct rte_flow_action actions[],
4376 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4377 {
4378 	const struct rte_flow_action_meter *mtr = NULL;
4379 	int actions_n = 0;
4380 
4381 	MLX5_ASSERT(has_mtr);
4382 	*has_mtr = false;
4383 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4384 		switch (actions->type) {
4385 		case RTE_FLOW_ACTION_TYPE_METER:
4386 			mtr = actions->conf;
4387 			*meter_id = mtr->mtr_id;
4388 			*has_mtr = true;
4389 			break;
4390 		default:
4391 			break;
4392 		}
4393 		if (!*has_mtr)
4394 			*has_modify |= flow_check_modify_action_type(dev,
4395 								actions->type);
4396 		actions_n++;
4397 	}
4398 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4399 	return actions_n + 1;
4400 }
4401 
4402 /**
4403  * Check if the flow should be split due to hairpin.
4404  * The reason for the split is that in current HW we can't
4405  * support encap and push-vlan on Rx, so if a flow contains
4406  * these actions we move it to Tx.
4407  *
4408  * @param dev
4409  *   Pointer to Ethernet device.
4410  * @param[in] attr
4411  *   Flow rule attributes.
4412  * @param[in] actions
4413  *   Associated actions (list terminated by the END action).
4414  *
4415  * @return
4416  *   > 0 the number of actions and the flow should be split,
4417  *   0 when no split required.
4418  */
4419 static int
4420 flow_check_hairpin_split(struct rte_eth_dev *dev,
4421 			 const struct rte_flow_attr *attr,
4422 			 const struct rte_flow_action actions[])
4423 {
4424 	int queue_action = 0;
4425 	int action_n = 0;
4426 	int split = 0;
4427 	const struct rte_flow_action_queue *queue;
4428 	const struct rte_flow_action_rss *rss;
4429 	const struct rte_flow_action_raw_encap *raw_encap;
4430 	const struct rte_eth_hairpin_conf *conf;
4431 
4432 	if (!attr->ingress)
4433 		return 0;
4434 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4435 		switch (actions->type) {
4436 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4437 			queue = actions->conf;
4438 			if (queue == NULL)
4439 				return 0;
4440 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4441 			if (conf == NULL || conf->tx_explicit != 0)
4442 				return 0;
4443 			queue_action = 1;
4444 			action_n++;
4445 			break;
4446 		case RTE_FLOW_ACTION_TYPE_RSS:
4447 			rss = actions->conf;
4448 			if (rss == NULL || rss->queue_num == 0)
4449 				return 0;
4450 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4451 			if (conf == NULL || conf->tx_explicit != 0)
4452 				return 0;
4453 			queue_action = 1;
4454 			action_n++;
4455 			break;
4456 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4457 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4458 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4459 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4460 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4461 			split++;
4462 			action_n++;
4463 			break;
4464 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4465 			raw_encap = actions->conf;
4466 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4467 				split++;
4468 			action_n++;
4469 			break;
4470 		default:
4471 			action_n++;
4472 			break;
4473 		}
4474 	}
4475 	if (split && queue_action)
4476 		return action_n;
4477 	return 0;
4478 }
4479 
4480 /* Declare flow create/destroy prototype in advance. */
4481 static uint32_t
4482 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4483 		 const struct rte_flow_attr *attr,
4484 		 const struct rte_flow_item items[],
4485 		 const struct rte_flow_action actions[],
4486 		 bool external, struct rte_flow_error *error);
4487 
4488 static void
4489 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4490 		  uint32_t flow_idx);
4491 
4492 int
4493 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4494 		      struct mlx5_list_entry *entry, void *cb_ctx)
4495 {
4496 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4497 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4498 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4499 
4500 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4501 }
4502 
4503 struct mlx5_list_entry *
4504 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4505 {
4506 	struct rte_eth_dev *dev = tool_ctx;
4507 	struct mlx5_priv *priv = dev->data->dev_private;
4508 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4509 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4510 	struct rte_flow_error *error = ctx->error;
4511 	uint32_t idx = 0;
4512 	int ret;
4513 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4514 	struct rte_flow_attr attr = {
4515 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4516 		.ingress = 1,
4517 	};
4518 	struct mlx5_rte_flow_item_tag tag_spec = {
4519 		.data = mark_id,
4520 	};
4521 	struct rte_flow_item items[] = {
4522 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4523 	};
4524 	struct rte_flow_action_mark ftag = {
4525 		.id = mark_id,
4526 	};
4527 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4528 		.dst = REG_B,
4529 		.src = REG_NON,
4530 	};
4531 	struct rte_flow_action_jump jump = {
4532 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4533 	};
4534 	struct rte_flow_action actions[] = {
4535 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4536 	};
4537 
4538 	/* Fill the register fields in the flow. */
4539 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4540 	if (ret < 0)
4541 		return NULL;
4542 	tag_spec.id = ret;
4543 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4544 	if (ret < 0)
4545 		return NULL;
4546 	cp_mreg.src = ret;
4547 	/* Provide the full width of FLAG specific value. */
4548 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4549 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4550 	/* Build a new flow. */
4551 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4552 		items[0] = (struct rte_flow_item){
4553 			.type = (enum rte_flow_item_type)
4554 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4555 			.spec = &tag_spec,
4556 		};
4557 		items[1] = (struct rte_flow_item){
4558 			.type = RTE_FLOW_ITEM_TYPE_END,
4559 		};
4560 		actions[0] = (struct rte_flow_action){
4561 			.type = (enum rte_flow_action_type)
4562 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4563 			.conf = &ftag,
4564 		};
4565 		actions[1] = (struct rte_flow_action){
4566 			.type = (enum rte_flow_action_type)
4567 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4568 			.conf = &cp_mreg,
4569 		};
4570 		actions[2] = (struct rte_flow_action){
4571 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4572 			.conf = &jump,
4573 		};
4574 		actions[3] = (struct rte_flow_action){
4575 			.type = RTE_FLOW_ACTION_TYPE_END,
4576 		};
4577 	} else {
4578 		/* Default rule, wildcard match. */
4579 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4580 		items[0] = (struct rte_flow_item){
4581 			.type = RTE_FLOW_ITEM_TYPE_END,
4582 		};
4583 		actions[0] = (struct rte_flow_action){
4584 			.type = (enum rte_flow_action_type)
4585 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4586 			.conf = &cp_mreg,
4587 		};
4588 		actions[1] = (struct rte_flow_action){
4589 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4590 			.conf = &jump,
4591 		};
4592 		actions[2] = (struct rte_flow_action){
4593 			.type = RTE_FLOW_ACTION_TYPE_END,
4594 		};
4595 	}
4596 	/* Build a new entry. */
4597 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4598 	if (!mcp_res) {
4599 		rte_errno = ENOMEM;
4600 		return NULL;
4601 	}
4602 	mcp_res->idx = idx;
4603 	mcp_res->mark_id = mark_id;
4604 	/*
4605 	 * The copy Flows are not included in any list. There
4606 	 * ones are referenced from other Flows and can not
4607 	 * be applied, removed, deleted in arbitrary order
4608 	 * by list traversing.
4609 	 */
4610 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4611 					&attr, items, actions, false, error);
4612 	if (!mcp_res->rix_flow) {
4613 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4614 		return NULL;
4615 	}
4616 	return &mcp_res->hlist_ent;
4617 }
4618 
4619 struct mlx5_list_entry *
4620 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4621 		      void *cb_ctx __rte_unused)
4622 {
4623 	struct rte_eth_dev *dev = tool_ctx;
4624 	struct mlx5_priv *priv = dev->data->dev_private;
4625 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4626 	uint32_t idx = 0;
4627 
4628 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4629 	if (!mcp_res) {
4630 		rte_errno = ENOMEM;
4631 		return NULL;
4632 	}
4633 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4634 	mcp_res->idx = idx;
4635 	return &mcp_res->hlist_ent;
4636 }
4637 
4638 void
4639 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4640 {
4641 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4642 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4643 	struct rte_eth_dev *dev = tool_ctx;
4644 	struct mlx5_priv *priv = dev->data->dev_private;
4645 
4646 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4647 }
4648 
4649 /**
4650  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4651  *
4652  * As mark_id is unique, if there's already a registered flow for the mark_id,
4653  * return by increasing the reference counter of the resource. Otherwise, create
4654  * the resource (mcp_res) and flow.
4655  *
4656  * Flow looks like,
4657  *   - If ingress port is ANY and reg_c[1] is mark_id,
4658  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4659  *
4660  * For default flow (zero mark_id), flow is like,
4661  *   - If ingress port is ANY,
4662  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4663  *
4664  * @param dev
4665  *   Pointer to Ethernet device.
4666  * @param mark_id
4667  *   ID of MARK action, zero means default flow for META.
4668  * @param[out] error
4669  *   Perform verbose error reporting if not NULL.
4670  *
4671  * @return
4672  *   Associated resource on success, NULL otherwise and rte_errno is set.
4673  */
4674 static struct mlx5_flow_mreg_copy_resource *
4675 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4676 			  struct rte_flow_error *error)
4677 {
4678 	struct mlx5_priv *priv = dev->data->dev_private;
4679 	struct mlx5_list_entry *entry;
4680 	struct mlx5_flow_cb_ctx ctx = {
4681 		.dev = dev,
4682 		.error = error,
4683 		.data = &mark_id,
4684 	};
4685 
4686 	/* Check if already registered. */
4687 	MLX5_ASSERT(priv->mreg_cp_tbl);
4688 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4689 	if (!entry)
4690 		return NULL;
4691 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4692 			    hlist_ent);
4693 }
4694 
4695 void
4696 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4697 {
4698 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4699 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4700 	struct rte_eth_dev *dev = tool_ctx;
4701 	struct mlx5_priv *priv = dev->data->dev_private;
4702 
4703 	MLX5_ASSERT(mcp_res->rix_flow);
4704 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4705 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4706 }
4707 
4708 /**
4709  * Release flow in RX_CP_TBL.
4710  *
4711  * @param dev
4712  *   Pointer to Ethernet device.
4713  * @flow
4714  *   Parent flow for wich copying is provided.
4715  */
4716 static void
4717 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4718 			  struct rte_flow *flow)
4719 {
4720 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4721 	struct mlx5_priv *priv = dev->data->dev_private;
4722 
4723 	if (!flow->rix_mreg_copy)
4724 		return;
4725 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4726 				 flow->rix_mreg_copy);
4727 	if (!mcp_res || !priv->mreg_cp_tbl)
4728 		return;
4729 	MLX5_ASSERT(mcp_res->rix_flow);
4730 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4731 	flow->rix_mreg_copy = 0;
4732 }
4733 
4734 /**
4735  * Remove the default copy action from RX_CP_TBL.
4736  *
4737  * This functions is called in the mlx5_dev_start(). No thread safe
4738  * is guaranteed.
4739  *
4740  * @param dev
4741  *   Pointer to Ethernet device.
4742  */
4743 static void
4744 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4745 {
4746 	struct mlx5_list_entry *entry;
4747 	struct mlx5_priv *priv = dev->data->dev_private;
4748 	struct mlx5_flow_cb_ctx ctx;
4749 	uint32_t mark_id;
4750 
4751 	/* Check if default flow is registered. */
4752 	if (!priv->mreg_cp_tbl)
4753 		return;
4754 	mark_id = MLX5_DEFAULT_COPY_ID;
4755 	ctx.data = &mark_id;
4756 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4757 	if (!entry)
4758 		return;
4759 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4760 }
4761 
4762 /**
4763  * Add the default copy action in in RX_CP_TBL.
4764  *
4765  * This functions is called in the mlx5_dev_start(). No thread safe
4766  * is guaranteed.
4767  *
4768  * @param dev
4769  *   Pointer to Ethernet device.
4770  * @param[out] error
4771  *   Perform verbose error reporting if not NULL.
4772  *
4773  * @return
4774  *   0 for success, negative value otherwise and rte_errno is set.
4775  */
4776 static int
4777 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4778 				  struct rte_flow_error *error)
4779 {
4780 	struct mlx5_priv *priv = dev->data->dev_private;
4781 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4782 	struct mlx5_flow_cb_ctx ctx;
4783 	uint32_t mark_id;
4784 
4785 	/* Check whether extensive metadata feature is engaged. */
4786 	if (!priv->sh->config.dv_flow_en ||
4787 	    priv->sh->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4788 	    !mlx5_flow_ext_mreg_supported(dev) ||
4789 	    !priv->sh->dv_regc0_mask)
4790 		return 0;
4791 	/*
4792 	 * Add default mreg copy flow may be called multiple time, but
4793 	 * only be called once in stop. Avoid register it twice.
4794 	 */
4795 	mark_id = MLX5_DEFAULT_COPY_ID;
4796 	ctx.data = &mark_id;
4797 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4798 		return 0;
4799 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4800 	if (!mcp_res)
4801 		return -rte_errno;
4802 	return 0;
4803 }
4804 
4805 /**
4806  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4807  *
4808  * All the flow having Q/RSS action should be split by
4809  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4810  * performs the following,
4811  *   - CQE->flow_tag := reg_c[1] (MARK)
4812  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4813  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4814  * but there should be a flow per each MARK ID set by MARK action.
4815  *
4816  * For the aforementioned reason, if there's a MARK action in flow's action
4817  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4818  * the MARK ID to CQE's flow_tag like,
4819  *   - If reg_c[1] is mark_id,
4820  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4821  *
4822  * For SET_META action which stores value in reg_c[0], as the destination is
4823  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4824  * MARK ID means the default flow. The default flow looks like,
4825  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4826  *
4827  * @param dev
4828  *   Pointer to Ethernet device.
4829  * @param flow
4830  *   Pointer to flow structure.
4831  * @param[in] actions
4832  *   Pointer to the list of actions.
4833  * @param[out] error
4834  *   Perform verbose error reporting if not NULL.
4835  *
4836  * @return
4837  *   0 on success, negative value otherwise and rte_errno is set.
4838  */
4839 static int
4840 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4841 			    struct rte_flow *flow,
4842 			    const struct rte_flow_action *actions,
4843 			    struct rte_flow_error *error)
4844 {
4845 	struct mlx5_priv *priv = dev->data->dev_private;
4846 	struct mlx5_sh_config *config = &priv->sh->config;
4847 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4848 	const struct rte_flow_action_mark *mark;
4849 
4850 	/* Check whether extensive metadata feature is engaged. */
4851 	if (!config->dv_flow_en ||
4852 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4853 	    !mlx5_flow_ext_mreg_supported(dev) ||
4854 	    !priv->sh->dv_regc0_mask)
4855 		return 0;
4856 	/* Find MARK action. */
4857 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4858 		switch (actions->type) {
4859 		case RTE_FLOW_ACTION_TYPE_FLAG:
4860 			mcp_res = flow_mreg_add_copy_action
4861 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4862 			if (!mcp_res)
4863 				return -rte_errno;
4864 			flow->rix_mreg_copy = mcp_res->idx;
4865 			return 0;
4866 		case RTE_FLOW_ACTION_TYPE_MARK:
4867 			mark = (const struct rte_flow_action_mark *)
4868 				actions->conf;
4869 			mcp_res =
4870 				flow_mreg_add_copy_action(dev, mark->id, error);
4871 			if (!mcp_res)
4872 				return -rte_errno;
4873 			flow->rix_mreg_copy = mcp_res->idx;
4874 			return 0;
4875 		default:
4876 			break;
4877 		}
4878 	}
4879 	return 0;
4880 }
4881 
4882 #define MLX5_MAX_SPLIT_ACTIONS 24
4883 #define MLX5_MAX_SPLIT_ITEMS 24
4884 
4885 /**
4886  * Split the hairpin flow.
4887  * Since HW can't support encap and push-vlan on Rx, we move these
4888  * actions to Tx.
4889  * If the count action is after the encap then we also
4890  * move the count action. in this case the count will also measure
4891  * the outer bytes.
4892  *
4893  * @param dev
4894  *   Pointer to Ethernet device.
4895  * @param[in] actions
4896  *   Associated actions (list terminated by the END action).
4897  * @param[out] actions_rx
4898  *   Rx flow actions.
4899  * @param[out] actions_tx
4900  *   Tx flow actions..
4901  * @param[out] pattern_tx
4902  *   The pattern items for the Tx flow.
4903  * @param[out] flow_id
4904  *   The flow ID connected to this flow.
4905  *
4906  * @return
4907  *   0 on success.
4908  */
4909 static int
4910 flow_hairpin_split(struct rte_eth_dev *dev,
4911 		   const struct rte_flow_action actions[],
4912 		   struct rte_flow_action actions_rx[],
4913 		   struct rte_flow_action actions_tx[],
4914 		   struct rte_flow_item pattern_tx[],
4915 		   uint32_t flow_id)
4916 {
4917 	const struct rte_flow_action_raw_encap *raw_encap;
4918 	const struct rte_flow_action_raw_decap *raw_decap;
4919 	struct mlx5_rte_flow_action_set_tag *set_tag;
4920 	struct rte_flow_action *tag_action;
4921 	struct mlx5_rte_flow_item_tag *tag_item;
4922 	struct rte_flow_item *item;
4923 	char *addr;
4924 	int encap = 0;
4925 
4926 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4927 		switch (actions->type) {
4928 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4929 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4930 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4931 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4932 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4933 			rte_memcpy(actions_tx, actions,
4934 			       sizeof(struct rte_flow_action));
4935 			actions_tx++;
4936 			break;
4937 		case RTE_FLOW_ACTION_TYPE_COUNT:
4938 			if (encap) {
4939 				rte_memcpy(actions_tx, actions,
4940 					   sizeof(struct rte_flow_action));
4941 				actions_tx++;
4942 			} else {
4943 				rte_memcpy(actions_rx, actions,
4944 					   sizeof(struct rte_flow_action));
4945 				actions_rx++;
4946 			}
4947 			break;
4948 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4949 			raw_encap = actions->conf;
4950 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4951 				memcpy(actions_tx, actions,
4952 				       sizeof(struct rte_flow_action));
4953 				actions_tx++;
4954 				encap = 1;
4955 			} else {
4956 				rte_memcpy(actions_rx, actions,
4957 					   sizeof(struct rte_flow_action));
4958 				actions_rx++;
4959 			}
4960 			break;
4961 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4962 			raw_decap = actions->conf;
4963 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4964 				memcpy(actions_tx, actions,
4965 				       sizeof(struct rte_flow_action));
4966 				actions_tx++;
4967 			} else {
4968 				rte_memcpy(actions_rx, actions,
4969 					   sizeof(struct rte_flow_action));
4970 				actions_rx++;
4971 			}
4972 			break;
4973 		default:
4974 			rte_memcpy(actions_rx, actions,
4975 				   sizeof(struct rte_flow_action));
4976 			actions_rx++;
4977 			break;
4978 		}
4979 	}
4980 	/* Add set meta action and end action for the Rx flow. */
4981 	tag_action = actions_rx;
4982 	tag_action->type = (enum rte_flow_action_type)
4983 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4984 	actions_rx++;
4985 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4986 	actions_rx++;
4987 	set_tag = (void *)actions_rx;
4988 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4989 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4990 		.data = flow_id,
4991 	};
4992 	MLX5_ASSERT(set_tag->id > REG_NON);
4993 	tag_action->conf = set_tag;
4994 	/* Create Tx item list. */
4995 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4996 	addr = (void *)&pattern_tx[2];
4997 	item = pattern_tx;
4998 	item->type = (enum rte_flow_item_type)
4999 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5000 	tag_item = (void *)addr;
5001 	tag_item->data = flow_id;
5002 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
5003 	MLX5_ASSERT(set_tag->id > REG_NON);
5004 	item->spec = tag_item;
5005 	addr += sizeof(struct mlx5_rte_flow_item_tag);
5006 	tag_item = (void *)addr;
5007 	tag_item->data = UINT32_MAX;
5008 	tag_item->id = UINT16_MAX;
5009 	item->mask = tag_item;
5010 	item->last = NULL;
5011 	item++;
5012 	item->type = RTE_FLOW_ITEM_TYPE_END;
5013 	return 0;
5014 }
5015 
5016 /**
5017  * The last stage of splitting chain, just creates the subflow
5018  * without any modification.
5019  *
5020  * @param[in] dev
5021  *   Pointer to Ethernet device.
5022  * @param[in] flow
5023  *   Parent flow structure pointer.
5024  * @param[in, out] sub_flow
5025  *   Pointer to return the created subflow, may be NULL.
5026  * @param[in] attr
5027  *   Flow rule attributes.
5028  * @param[in] items
5029  *   Pattern specification (list terminated by the END pattern item).
5030  * @param[in] actions
5031  *   Associated actions (list terminated by the END action).
5032  * @param[in] flow_split_info
5033  *   Pointer to flow split info structure.
5034  * @param[out] error
5035  *   Perform verbose error reporting if not NULL.
5036  * @return
5037  *   0 on success, negative value otherwise
5038  */
5039 static int
5040 flow_create_split_inner(struct rte_eth_dev *dev,
5041 			struct rte_flow *flow,
5042 			struct mlx5_flow **sub_flow,
5043 			const struct rte_flow_attr *attr,
5044 			const struct rte_flow_item items[],
5045 			const struct rte_flow_action actions[],
5046 			struct mlx5_flow_split_info *flow_split_info,
5047 			struct rte_flow_error *error)
5048 {
5049 	struct mlx5_flow *dev_flow;
5050 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
5051 
5052 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
5053 				    flow_split_info->flow_idx, error);
5054 	if (!dev_flow)
5055 		return -rte_errno;
5056 	dev_flow->flow = flow;
5057 	dev_flow->external = flow_split_info->external;
5058 	dev_flow->skip_scale = flow_split_info->skip_scale;
5059 	/* Subflow object was created, we must include one in the list. */
5060 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
5061 		      dev_flow->handle, next);
5062 	/*
5063 	 * If dev_flow is as one of the suffix flow, some actions in suffix
5064 	 * flow may need some user defined item layer flags, and pass the
5065 	 * Metadata rxq mark flag to suffix flow as well.
5066 	 */
5067 	if (flow_split_info->prefix_layers)
5068 		dev_flow->handle->layers = flow_split_info->prefix_layers;
5069 	if (flow_split_info->prefix_mark) {
5070 		MLX5_ASSERT(wks);
5071 		wks->mark = 1;
5072 	}
5073 	if (sub_flow)
5074 		*sub_flow = dev_flow;
5075 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5076 	dev_flow->dv.table_id = flow_split_info->table_id;
5077 #endif
5078 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
5079 }
5080 
5081 /**
5082  * Get the sub policy of a meter.
5083  *
5084  * @param[in] dev
5085  *   Pointer to Ethernet device.
5086  * @param[in] flow
5087  *   Parent flow structure pointer.
5088  * @param wks
5089  *   Pointer to thread flow work space.
5090  * @param[in] attr
5091  *   Flow rule attributes.
5092  * @param[in] items
5093  *   Pattern specification (list terminated by the END pattern item).
5094  * @param[out] error
5095  *   Perform verbose error reporting if not NULL.
5096  *
5097  * @return
5098  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
5099  */
5100 static struct mlx5_flow_meter_sub_policy *
5101 get_meter_sub_policy(struct rte_eth_dev *dev,
5102 		     struct rte_flow *flow,
5103 		     struct mlx5_flow_workspace *wks,
5104 		     const struct rte_flow_attr *attr,
5105 		     const struct rte_flow_item items[],
5106 		     struct rte_flow_error *error)
5107 {
5108 	struct mlx5_flow_meter_policy *policy;
5109 	struct mlx5_flow_meter_policy *final_policy;
5110 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
5111 
5112 	policy = wks->policy;
5113 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
5114 	if (final_policy->is_rss || final_policy->is_queue) {
5115 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
5116 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
5117 		uint32_t i;
5118 
5119 		/*
5120 		 * This is a tmp dev_flow,
5121 		 * no need to register any matcher for it in translate.
5122 		 */
5123 		wks->skip_matcher_reg = 1;
5124 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
5125 			struct mlx5_flow dev_flow = {0};
5126 			struct mlx5_flow_handle dev_handle = { {0} };
5127 			uint8_t fate = final_policy->act_cnt[i].fate_action;
5128 
5129 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
5130 				const struct rte_flow_action_rss *rss_act =
5131 					final_policy->act_cnt[i].rss->conf;
5132 				struct rte_flow_action rss_actions[2] = {
5133 					[0] = {
5134 					.type = RTE_FLOW_ACTION_TYPE_RSS,
5135 					.conf = rss_act,
5136 					},
5137 					[1] = {
5138 					.type = RTE_FLOW_ACTION_TYPE_END,
5139 					.conf = NULL,
5140 					}
5141 				};
5142 
5143 				dev_flow.handle = &dev_handle;
5144 				dev_flow.ingress = attr->ingress;
5145 				dev_flow.flow = flow;
5146 				dev_flow.external = 0;
5147 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
5148 				dev_flow.dv.transfer = attr->transfer;
5149 #endif
5150 				/**
5151 				 * Translate RSS action to get rss hash fields.
5152 				 */
5153 				if (flow_drv_translate(dev, &dev_flow, attr,
5154 						items, rss_actions, error))
5155 					goto exit;
5156 				rss_desc_v[i] = wks->rss_desc;
5157 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
5158 				rss_desc_v[i].hash_fields =
5159 						dev_flow.hash_fields;
5160 				rss_desc_v[i].queue_num =
5161 						rss_desc_v[i].hash_fields ?
5162 						rss_desc_v[i].queue_num : 1;
5163 				rss_desc_v[i].tunnel =
5164 						!!(dev_flow.handle->layers &
5165 						   MLX5_FLOW_LAYER_TUNNEL);
5166 				/* Use the RSS queues in the containers. */
5167 				rss_desc_v[i].queue =
5168 					(uint16_t *)(uintptr_t)rss_act->queue;
5169 				rss_desc[i] = &rss_desc_v[i];
5170 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
5171 				/* This is queue action. */
5172 				rss_desc_v[i] = wks->rss_desc;
5173 				rss_desc_v[i].key_len = 0;
5174 				rss_desc_v[i].hash_fields = 0;
5175 				rss_desc_v[i].queue =
5176 					&final_policy->act_cnt[i].queue;
5177 				rss_desc_v[i].queue_num = 1;
5178 				rss_desc[i] = &rss_desc_v[i];
5179 			} else {
5180 				rss_desc[i] = NULL;
5181 			}
5182 		}
5183 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
5184 						flow, policy, rss_desc);
5185 	} else {
5186 		enum mlx5_meter_domain mtr_domain =
5187 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5188 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5189 						MLX5_MTR_DOMAIN_INGRESS);
5190 		sub_policy = policy->sub_policys[mtr_domain][0];
5191 	}
5192 	if (!sub_policy)
5193 		rte_flow_error_set(error, EINVAL,
5194 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5195 				   "Failed to get meter sub-policy.");
5196 exit:
5197 	return sub_policy;
5198 }
5199 
5200 /**
5201  * Split the meter flow.
5202  *
5203  * As meter flow will split to three sub flow, other than meter
5204  * action, the other actions make sense to only meter accepts
5205  * the packet. If it need to be dropped, no other additional
5206  * actions should be take.
5207  *
5208  * One kind of special action which decapsulates the L3 tunnel
5209  * header will be in the prefix sub flow, as not to take the
5210  * L3 tunnel header into account.
5211  *
5212  * @param[in] dev
5213  *   Pointer to Ethernet device.
5214  * @param[in] flow
5215  *   Parent flow structure pointer.
5216  * @param wks
5217  *   Pointer to thread flow work space.
5218  * @param[in] attr
5219  *   Flow rule attributes.
5220  * @param[in] items
5221  *   Pattern specification (list terminated by the END pattern item).
5222  * @param[out] sfx_items
5223  *   Suffix flow match items (list terminated by the END pattern item).
5224  * @param[in] actions
5225  *   Associated actions (list terminated by the END action).
5226  * @param[out] actions_sfx
5227  *   Suffix flow actions.
5228  * @param[out] actions_pre
5229  *   Prefix flow actions.
5230  * @param[out] mtr_flow_id
5231  *   Pointer to meter flow id.
5232  * @param[out] error
5233  *   Perform verbose error reporting if not NULL.
5234  *
5235  * @return
5236  *   0 on success, a negative errno value otherwise and rte_errno is set.
5237  */
5238 static int
5239 flow_meter_split_prep(struct rte_eth_dev *dev,
5240 		      struct rte_flow *flow,
5241 		      struct mlx5_flow_workspace *wks,
5242 		      const struct rte_flow_attr *attr,
5243 		      const struct rte_flow_item items[],
5244 		      struct rte_flow_item sfx_items[],
5245 		      const struct rte_flow_action actions[],
5246 		      struct rte_flow_action actions_sfx[],
5247 		      struct rte_flow_action actions_pre[],
5248 		      uint32_t *mtr_flow_id,
5249 		      struct rte_flow_error *error)
5250 {
5251 	struct mlx5_priv *priv = dev->data->dev_private;
5252 	struct mlx5_flow_meter_info *fm = wks->fm;
5253 	struct rte_flow_action *tag_action = NULL;
5254 	struct rte_flow_item *tag_item;
5255 	struct mlx5_rte_flow_action_set_tag *set_tag;
5256 	const struct rte_flow_action_raw_encap *raw_encap;
5257 	const struct rte_flow_action_raw_decap *raw_decap;
5258 	struct mlx5_rte_flow_item_tag *tag_item_spec;
5259 	struct mlx5_rte_flow_item_tag *tag_item_mask;
5260 	uint32_t tag_id = 0;
5261 	struct rte_flow_item *vlan_item_dst = NULL;
5262 	const struct rte_flow_item *vlan_item_src = NULL;
5263 	const struct rte_flow_item *orig_items = items;
5264 	struct rte_flow_action *hw_mtr_action;
5265 	struct rte_flow_action *action_pre_head = NULL;
5266 	int32_t flow_src_port = priv->representor_id;
5267 	bool mtr_first;
5268 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5269 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5270 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5271 	uint32_t flow_id = 0;
5272 	uint32_t flow_id_reversed = 0;
5273 	uint8_t flow_id_bits = 0;
5274 	bool after_meter = false;
5275 	int shift;
5276 
5277 	/* Prepare the suffix subflow items. */
5278 	tag_item = sfx_items++;
5279 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5280 		struct mlx5_priv *port_priv;
5281 		const struct rte_flow_item_port_id *pid_v;
5282 		int item_type = items->type;
5283 
5284 		switch (item_type) {
5285 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5286 			pid_v = items->spec;
5287 			MLX5_ASSERT(pid_v);
5288 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5289 			if (!port_priv)
5290 				return rte_flow_error_set(error,
5291 						rte_errno,
5292 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5293 						pid_v,
5294 						"Failed to get port info.");
5295 			flow_src_port = port_priv->representor_id;
5296 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5297 			    flow_src_port != priv->representor_id) {
5298 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5299 								flow, fm,
5300 								flow_src_port,
5301 								items,
5302 								error))
5303 					return -rte_errno;
5304 			}
5305 			memcpy(sfx_items, items, sizeof(*sfx_items));
5306 			sfx_items++;
5307 			break;
5308 		case RTE_FLOW_ITEM_TYPE_VLAN:
5309 			/* Determine if copy vlan item below. */
5310 			vlan_item_src = items;
5311 			vlan_item_dst = sfx_items++;
5312 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5313 			break;
5314 		default:
5315 			break;
5316 		}
5317 	}
5318 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5319 	sfx_items++;
5320 	mtr_first = priv->sh->meter_aso_en &&
5321 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5322 	/* For ASO meter, meter must be before tag in TX direction. */
5323 	if (mtr_first) {
5324 		action_pre_head = actions_pre++;
5325 		/* Leave space for tag action. */
5326 		tag_action = actions_pre++;
5327 	}
5328 	/* Prepare the actions for prefix and suffix flow. */
5329 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5330 		struct rte_flow_action *action_cur = NULL;
5331 
5332 		switch (actions->type) {
5333 		case RTE_FLOW_ACTION_TYPE_METER:
5334 			if (mtr_first) {
5335 				action_cur = action_pre_head;
5336 			} else {
5337 				/* Leave space for tag action. */
5338 				tag_action = actions_pre++;
5339 				action_cur = actions_pre++;
5340 			}
5341 			after_meter = true;
5342 			break;
5343 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5344 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5345 			action_cur = actions_pre++;
5346 			break;
5347 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5348 			raw_encap = actions->conf;
5349 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5350 				action_cur = actions_pre++;
5351 			break;
5352 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5353 			raw_decap = actions->conf;
5354 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5355 				action_cur = actions_pre++;
5356 			break;
5357 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5358 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5359 			if (vlan_item_dst && vlan_item_src) {
5360 				memcpy(vlan_item_dst, vlan_item_src,
5361 					sizeof(*vlan_item_dst));
5362 				/*
5363 				 * Convert to internal match item, it is used
5364 				 * for vlan push and set vid.
5365 				 */
5366 				vlan_item_dst->type = (enum rte_flow_item_type)
5367 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5368 			}
5369 			break;
5370 		case RTE_FLOW_ACTION_TYPE_COUNT:
5371 			if (fm->def_policy)
5372 				action_cur = after_meter ?
5373 						actions_sfx++ : actions_pre++;
5374 			break;
5375 		default:
5376 			break;
5377 		}
5378 		if (!action_cur)
5379 			action_cur = (fm->def_policy) ?
5380 					actions_sfx++ : actions_pre++;
5381 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5382 	}
5383 	/* Add end action to the actions. */
5384 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5385 	if (priv->sh->meter_aso_en) {
5386 		/**
5387 		 * For ASO meter, need to add an extra jump action explicitly,
5388 		 * to jump from meter to policer table.
5389 		 */
5390 		struct mlx5_flow_meter_sub_policy *sub_policy;
5391 		struct mlx5_flow_tbl_data_entry *tbl_data;
5392 
5393 		if (!fm->def_policy) {
5394 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5395 							  attr, orig_items,
5396 							  error);
5397 			if (!sub_policy)
5398 				return -rte_errno;
5399 		} else {
5400 			enum mlx5_meter_domain mtr_domain =
5401 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5402 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5403 						MLX5_MTR_DOMAIN_INGRESS);
5404 
5405 			sub_policy =
5406 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5407 		}
5408 		tbl_data = container_of(sub_policy->tbl_rsc,
5409 					struct mlx5_flow_tbl_data_entry, tbl);
5410 		hw_mtr_action = actions_pre++;
5411 		hw_mtr_action->type = (enum rte_flow_action_type)
5412 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5413 		hw_mtr_action->conf = tbl_data->jump.action;
5414 	}
5415 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5416 	actions_pre++;
5417 	if (!tag_action)
5418 		return rte_flow_error_set(error, ENOMEM,
5419 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5420 					  NULL, "No tag action space.");
5421 	if (!mtr_flow_id) {
5422 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5423 		goto exit;
5424 	}
5425 	/* Only default-policy Meter creates mtr flow id. */
5426 	if (fm->def_policy) {
5427 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5428 		if (!tag_id)
5429 			return rte_flow_error_set(error, ENOMEM,
5430 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5431 					"Failed to allocate meter flow id.");
5432 		flow_id = tag_id - 1;
5433 		flow_id_bits = (!flow_id) ? 1 :
5434 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5435 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5436 		    mtr_reg_bits) {
5437 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5438 			return rte_flow_error_set(error, EINVAL,
5439 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5440 					"Meter flow id exceeds max limit.");
5441 		}
5442 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5443 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5444 	}
5445 	/* Build tag actions and items for meter_id/meter flow_id. */
5446 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5447 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5448 	tag_item_mask = tag_item_spec + 1;
5449 	/* Both flow_id and meter_id share the same register. */
5450 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5451 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5452 							    0, error),
5453 		.offset = mtr_id_offset,
5454 		.length = mtr_reg_bits,
5455 		.data = flow->meter,
5456 	};
5457 	/*
5458 	 * The color Reg bits used by flow_id are growing from
5459 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5460 	 */
5461 	for (shift = 0; shift < flow_id_bits; shift++)
5462 		flow_id_reversed = (flow_id_reversed << 1) |
5463 				((flow_id >> shift) & 0x1);
5464 	set_tag->data |=
5465 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5466 	tag_item_spec->id = set_tag->id;
5467 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5468 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5469 	tag_action->type = (enum rte_flow_action_type)
5470 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5471 	tag_action->conf = set_tag;
5472 	tag_item->type = (enum rte_flow_item_type)
5473 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5474 	tag_item->spec = tag_item_spec;
5475 	tag_item->last = NULL;
5476 	tag_item->mask = tag_item_mask;
5477 exit:
5478 	if (mtr_flow_id)
5479 		*mtr_flow_id = tag_id;
5480 	return 0;
5481 }
5482 
5483 /**
5484  * Split action list having QUEUE/RSS for metadata register copy.
5485  *
5486  * Once Q/RSS action is detected in user's action list, the flow action
5487  * should be split in order to copy metadata registers, which will happen in
5488  * RX_CP_TBL like,
5489  *   - CQE->flow_tag := reg_c[1] (MARK)
5490  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5491  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5492  * This is because the last action of each flow must be a terminal action
5493  * (QUEUE, RSS or DROP).
5494  *
5495  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5496  * stored and kept in the mlx5_flow structure per each sub_flow.
5497  *
5498  * The Q/RSS action is replaced with,
5499  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5500  * And the following JUMP action is added at the end,
5501  *   - JUMP, to RX_CP_TBL.
5502  *
5503  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5504  * flow_create_split_metadata() routine. The flow will look like,
5505  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5506  *
5507  * @param dev
5508  *   Pointer to Ethernet device.
5509  * @param[out] split_actions
5510  *   Pointer to store split actions to jump to CP_TBL.
5511  * @param[in] actions
5512  *   Pointer to the list of original flow actions.
5513  * @param[in] qrss
5514  *   Pointer to the Q/RSS action.
5515  * @param[in] actions_n
5516  *   Number of original actions.
5517  * @param[in] mtr_sfx
5518  *   Check if it is in meter suffix table.
5519  * @param[out] error
5520  *   Perform verbose error reporting if not NULL.
5521  *
5522  * @return
5523  *   non-zero unique flow_id on success, otherwise 0 and
5524  *   error/rte_error are set.
5525  */
5526 static uint32_t
5527 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5528 			  struct rte_flow_action *split_actions,
5529 			  const struct rte_flow_action *actions,
5530 			  const struct rte_flow_action *qrss,
5531 			  int actions_n, int mtr_sfx,
5532 			  struct rte_flow_error *error)
5533 {
5534 	struct mlx5_priv *priv = dev->data->dev_private;
5535 	struct mlx5_rte_flow_action_set_tag *set_tag;
5536 	struct rte_flow_action_jump *jump;
5537 	const int qrss_idx = qrss - actions;
5538 	uint32_t flow_id = 0;
5539 	int ret = 0;
5540 
5541 	/*
5542 	 * Given actions will be split
5543 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5544 	 * - Add jump to mreg CP_TBL.
5545 	 * As a result, there will be one more action.
5546 	 */
5547 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5548 	/* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5549 	++actions_n;
5550 	set_tag = (void *)(split_actions + actions_n);
5551 	/*
5552 	 * If we are not the meter suffix flow, add the tag action.
5553 	 * Since meter suffix flow already has the tag added.
5554 	 */
5555 	if (!mtr_sfx) {
5556 		/*
5557 		 * Allocate the new subflow ID. This one is unique within
5558 		 * device and not shared with representors. Otherwise,
5559 		 * we would have to resolve multi-thread access synch
5560 		 * issue. Each flow on the shared device is appended
5561 		 * with source vport identifier, so the resulting
5562 		 * flows will be unique in the shared (by master and
5563 		 * representors) domain even if they have coinciding
5564 		 * IDs.
5565 		 */
5566 		mlx5_ipool_malloc(priv->sh->ipool
5567 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5568 		if (!flow_id)
5569 			return rte_flow_error_set(error, ENOMEM,
5570 						  RTE_FLOW_ERROR_TYPE_ACTION,
5571 						  NULL, "can't allocate id "
5572 						  "for split Q/RSS subflow");
5573 		/* Internal SET_TAG action to set flow ID. */
5574 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5575 			.data = flow_id,
5576 		};
5577 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5578 		if (ret < 0)
5579 			return ret;
5580 		set_tag->id = ret;
5581 		/* Construct new actions array. */
5582 		/* Replace QUEUE/RSS action. */
5583 		split_actions[qrss_idx] = (struct rte_flow_action){
5584 			.type = (enum rte_flow_action_type)
5585 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5586 			.conf = set_tag,
5587 		};
5588 	} else {
5589 		/*
5590 		 * If we are the suffix flow of meter, tag already exist.
5591 		 * Set the QUEUE/RSS action to void.
5592 		 */
5593 		split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5594 	}
5595 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5596 	jump = (void *)(set_tag + 1);
5597 	*jump = (struct rte_flow_action_jump){
5598 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5599 	};
5600 	split_actions[actions_n - 2] = (struct rte_flow_action){
5601 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5602 		.conf = jump,
5603 	};
5604 	split_actions[actions_n - 1] = (struct rte_flow_action){
5605 		.type = RTE_FLOW_ACTION_TYPE_END,
5606 	};
5607 	return flow_id;
5608 }
5609 
5610 /**
5611  * Extend the given action list for Tx metadata copy.
5612  *
5613  * Copy the given action list to the ext_actions and add flow metadata register
5614  * copy action in order to copy reg_a set by WQE to reg_c[0].
5615  *
5616  * @param[out] ext_actions
5617  *   Pointer to the extended action list.
5618  * @param[in] actions
5619  *   Pointer to the list of actions.
5620  * @param[in] actions_n
5621  *   Number of actions in the list.
5622  * @param[out] error
5623  *   Perform verbose error reporting if not NULL.
5624  * @param[in] encap_idx
5625  *   The encap action index.
5626  *
5627  * @return
5628  *   0 on success, negative value otherwise
5629  */
5630 static int
5631 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5632 		       struct rte_flow_action *ext_actions,
5633 		       const struct rte_flow_action *actions,
5634 		       int actions_n, struct rte_flow_error *error,
5635 		       int encap_idx)
5636 {
5637 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5638 		(struct mlx5_flow_action_copy_mreg *)
5639 			(ext_actions + actions_n + 1);
5640 	int ret;
5641 
5642 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5643 	if (ret < 0)
5644 		return ret;
5645 	cp_mreg->dst = ret;
5646 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5647 	if (ret < 0)
5648 		return ret;
5649 	cp_mreg->src = ret;
5650 	if (encap_idx != 0)
5651 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5652 	if (encap_idx == actions_n - 1) {
5653 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5654 			.type = (enum rte_flow_action_type)
5655 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5656 			.conf = cp_mreg,
5657 		};
5658 		ext_actions[actions_n] = (struct rte_flow_action){
5659 			.type = RTE_FLOW_ACTION_TYPE_END,
5660 		};
5661 	} else {
5662 		ext_actions[encap_idx] = (struct rte_flow_action){
5663 			.type = (enum rte_flow_action_type)
5664 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5665 			.conf = cp_mreg,
5666 		};
5667 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5668 				sizeof(*ext_actions) * (actions_n - encap_idx));
5669 	}
5670 	return 0;
5671 }
5672 
5673 /**
5674  * Check the match action from the action list.
5675  *
5676  * @param[in] actions
5677  *   Pointer to the list of actions.
5678  * @param[in] attr
5679  *   Flow rule attributes.
5680  * @param[in] action
5681  *   The action to be check if exist.
5682  * @param[out] match_action_pos
5683  *   Pointer to the position of the matched action if exists, otherwise is -1.
5684  * @param[out] qrss_action_pos
5685  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5686  * @param[out] modify_after_mirror
5687  *   Pointer to the flag of modify action after FDB mirroring.
5688  *
5689  * @return
5690  *   > 0 the total number of actions.
5691  *   0 if not found match action in action list.
5692  */
5693 static int
5694 flow_check_match_action(const struct rte_flow_action actions[],
5695 			const struct rte_flow_attr *attr,
5696 			enum rte_flow_action_type action,
5697 			int *match_action_pos, int *qrss_action_pos,
5698 			int *modify_after_mirror)
5699 {
5700 	const struct rte_flow_action_sample *sample;
5701 	const struct rte_flow_action_raw_decap *decap;
5702 	int actions_n = 0;
5703 	uint32_t ratio = 0;
5704 	int sub_type = 0;
5705 	int flag = 0;
5706 	int fdb_mirror = 0;
5707 
5708 	*match_action_pos = -1;
5709 	*qrss_action_pos = -1;
5710 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5711 		if (actions->type == action) {
5712 			flag = 1;
5713 			*match_action_pos = actions_n;
5714 		}
5715 		switch (actions->type) {
5716 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5717 		case RTE_FLOW_ACTION_TYPE_RSS:
5718 			*qrss_action_pos = actions_n;
5719 			break;
5720 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5721 			sample = actions->conf;
5722 			ratio = sample->ratio;
5723 			sub_type = ((const struct rte_flow_action *)
5724 					(sample->actions))->type;
5725 			if (ratio == 1 && attr->transfer)
5726 				fdb_mirror = 1;
5727 			break;
5728 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5729 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5730 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5731 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5732 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5733 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5734 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5735 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5736 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5737 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5738 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5739 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5740 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5741 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5742 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5743 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5744 		case RTE_FLOW_ACTION_TYPE_FLAG:
5745 		case RTE_FLOW_ACTION_TYPE_MARK:
5746 		case RTE_FLOW_ACTION_TYPE_SET_META:
5747 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5748 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5749 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5750 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5751 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5752 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5753 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5754 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5755 		case RTE_FLOW_ACTION_TYPE_METER:
5756 			if (fdb_mirror)
5757 				*modify_after_mirror = 1;
5758 			break;
5759 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5760 			decap = actions->conf;
5761 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5762 				;
5763 			actions_n++;
5764 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5765 				const struct rte_flow_action_raw_encap *encap =
5766 								actions->conf;
5767 				if (decap->size <=
5768 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5769 				    encap->size >
5770 					MLX5_ENCAPSULATION_DECISION_SIZE)
5771 					/* L3 encap. */
5772 					break;
5773 			}
5774 			if (fdb_mirror)
5775 				*modify_after_mirror = 1;
5776 			break;
5777 		default:
5778 			break;
5779 		}
5780 		actions_n++;
5781 	}
5782 	if (flag && fdb_mirror && !*modify_after_mirror) {
5783 		/* FDB mirroring uses the destination array to implement
5784 		 * instead of FLOW_SAMPLER object.
5785 		 */
5786 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5787 			flag = 0;
5788 	}
5789 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5790 	return flag ? actions_n + 1 : 0;
5791 }
5792 
5793 #define SAMPLE_SUFFIX_ITEM 3
5794 
5795 /**
5796  * Split the sample flow.
5797  *
5798  * As sample flow will split to two sub flow, sample flow with
5799  * sample action, the other actions will move to new suffix flow.
5800  *
5801  * Also add unique tag id with tag action in the sample flow,
5802  * the same tag id will be as match in the suffix flow.
5803  *
5804  * @param dev
5805  *   Pointer to Ethernet device.
5806  * @param[in] add_tag
5807  *   Add extra tag action flag.
5808  * @param[out] sfx_items
5809  *   Suffix flow match items (list terminated by the END pattern item).
5810  * @param[in] actions
5811  *   Associated actions (list terminated by the END action).
5812  * @param[out] actions_sfx
5813  *   Suffix flow actions.
5814  * @param[out] actions_pre
5815  *   Prefix flow actions.
5816  * @param[in] actions_n
5817  *  The total number of actions.
5818  * @param[in] sample_action_pos
5819  *   The sample action position.
5820  * @param[in] qrss_action_pos
5821  *   The Queue/RSS action position.
5822  * @param[in] jump_table
5823  *   Add extra jump action flag.
5824  * @param[out] error
5825  *   Perform verbose error reporting if not NULL.
5826  *
5827  * @return
5828  *   0 on success, or unique flow_id, a negative errno value
5829  *   otherwise and rte_errno is set.
5830  */
5831 static int
5832 flow_sample_split_prep(struct rte_eth_dev *dev,
5833 		       int add_tag,
5834 		       const struct rte_flow_item items[],
5835 		       struct rte_flow_item sfx_items[],
5836 		       const struct rte_flow_action actions[],
5837 		       struct rte_flow_action actions_sfx[],
5838 		       struct rte_flow_action actions_pre[],
5839 		       int actions_n,
5840 		       int sample_action_pos,
5841 		       int qrss_action_pos,
5842 		       int jump_table,
5843 		       struct rte_flow_error *error)
5844 {
5845 	struct mlx5_priv *priv = dev->data->dev_private;
5846 	struct mlx5_rte_flow_action_set_tag *set_tag;
5847 	struct mlx5_rte_flow_item_tag *tag_spec;
5848 	struct mlx5_rte_flow_item_tag *tag_mask;
5849 	struct rte_flow_action_jump *jump_action;
5850 	uint32_t tag_id = 0;
5851 	int append_index = 0;
5852 	int set_tag_idx = -1;
5853 	int index;
5854 	int ret;
5855 
5856 	if (sample_action_pos < 0)
5857 		return rte_flow_error_set(error, EINVAL,
5858 					  RTE_FLOW_ERROR_TYPE_ACTION,
5859 					  NULL, "invalid position of sample "
5860 					  "action in list");
5861 	/* Prepare the actions for prefix and suffix flow. */
5862 	if (add_tag) {
5863 		/* Update the new added tag action index preceding
5864 		 * the PUSH_VLAN or ENCAP action.
5865 		 */
5866 		const struct rte_flow_action_raw_encap *raw_encap;
5867 		const struct rte_flow_action *action = actions;
5868 		int encap_idx;
5869 		int action_idx = 0;
5870 		int raw_decap_idx = -1;
5871 		int push_vlan_idx = -1;
5872 		for (; action->type != RTE_FLOW_ACTION_TYPE_END; action++) {
5873 			switch (action->type) {
5874 			case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5875 				raw_decap_idx = action_idx;
5876 				break;
5877 			case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5878 				raw_encap = action->conf;
5879 				if (raw_encap->size >
5880 					MLX5_ENCAPSULATION_DECISION_SIZE) {
5881 					encap_idx = raw_decap_idx != -1 ?
5882 						    raw_decap_idx : action_idx;
5883 					if (encap_idx < sample_action_pos &&
5884 					    push_vlan_idx == -1)
5885 						set_tag_idx = encap_idx;
5886 				}
5887 				break;
5888 			case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5889 			case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5890 				encap_idx = action_idx;
5891 				if (encap_idx < sample_action_pos &&
5892 				    push_vlan_idx == -1)
5893 					set_tag_idx = encap_idx;
5894 				break;
5895 			case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5896 			case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5897 				push_vlan_idx = action_idx;
5898 				if (push_vlan_idx < sample_action_pos)
5899 					set_tag_idx = action_idx;
5900 				break;
5901 			default:
5902 				break;
5903 			}
5904 			action_idx++;
5905 		}
5906 	}
5907 	/* Prepare the actions for prefix and suffix flow. */
5908 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5909 		index = qrss_action_pos;
5910 		/* Put the preceding the Queue/RSS action into prefix flow. */
5911 		if (index != 0)
5912 			memcpy(actions_pre, actions,
5913 			       sizeof(struct rte_flow_action) * index);
5914 		/* Put others preceding the sample action into prefix flow. */
5915 		if (sample_action_pos > index + 1)
5916 			memcpy(actions_pre + index, actions + index + 1,
5917 			       sizeof(struct rte_flow_action) *
5918 			       (sample_action_pos - index - 1));
5919 		index = sample_action_pos - 1;
5920 		/* Put Queue/RSS action into Suffix flow. */
5921 		memcpy(actions_sfx, actions + qrss_action_pos,
5922 		       sizeof(struct rte_flow_action));
5923 		actions_sfx++;
5924 	} else if (add_tag && set_tag_idx >= 0) {
5925 		if (set_tag_idx > 0)
5926 			memcpy(actions_pre, actions,
5927 			       sizeof(struct rte_flow_action) * set_tag_idx);
5928 		memcpy(actions_pre + set_tag_idx + 1, actions + set_tag_idx,
5929 		       sizeof(struct rte_flow_action) *
5930 		       (sample_action_pos - set_tag_idx));
5931 		index = sample_action_pos;
5932 	} else {
5933 		index = sample_action_pos;
5934 		if (index != 0)
5935 			memcpy(actions_pre, actions,
5936 			       sizeof(struct rte_flow_action) * index);
5937 	}
5938 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5939 	 * For CX6DX and above, metadata registers Cx preserve their value,
5940 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5941 	 */
5942 	if (add_tag) {
5943 		/* Prepare the prefix tag action. */
5944 		append_index++;
5945 		set_tag = (void *)(actions_pre + actions_n + append_index);
5946 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5947 		/* Trust VF/SF on CX5 not supported meter so that the reserved
5948 		 * metadata regC is REG_NON, back to use application tag
5949 		 * index 0.
5950 		 */
5951 		if (unlikely(ret == REG_NON))
5952 			ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, 0, error);
5953 		if (ret < 0)
5954 			return ret;
5955 		mlx5_ipool_malloc(priv->sh->ipool
5956 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5957 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5958 			.id = ret,
5959 			.data = tag_id,
5960 		};
5961 		/* Prepare the suffix subflow items. */
5962 		for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5963 			if (items->type == RTE_FLOW_ITEM_TYPE_PORT_ID) {
5964 				memcpy(sfx_items, items, sizeof(*sfx_items));
5965 				sfx_items++;
5966 			}
5967 		}
5968 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5969 		tag_spec->data = tag_id;
5970 		tag_spec->id = set_tag->id;
5971 		tag_mask = tag_spec + 1;
5972 		tag_mask->data = UINT32_MAX;
5973 		sfx_items[0] = (struct rte_flow_item){
5974 			.type = (enum rte_flow_item_type)
5975 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5976 			.spec = tag_spec,
5977 			.last = NULL,
5978 			.mask = tag_mask,
5979 		};
5980 		sfx_items[1] = (struct rte_flow_item){
5981 			.type = (enum rte_flow_item_type)
5982 				RTE_FLOW_ITEM_TYPE_END,
5983 		};
5984 		/* Prepare the tag action in prefix subflow. */
5985 		set_tag_idx = (set_tag_idx == -1) ? index : set_tag_idx;
5986 		actions_pre[set_tag_idx] =
5987 			(struct rte_flow_action){
5988 			.type = (enum rte_flow_action_type)
5989 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5990 			.conf = set_tag,
5991 		};
5992 		/* Update next sample position due to add one tag action */
5993 		index += 1;
5994 	}
5995 	/* Copy the sample action into prefix flow. */
5996 	memcpy(actions_pre + index, actions + sample_action_pos,
5997 	       sizeof(struct rte_flow_action));
5998 	index += 1;
5999 	/* For the modify action after the sample action in E-Switch mirroring,
6000 	 * Add the extra jump action in prefix subflow and jump into the next
6001 	 * table, then do the modify action in the new table.
6002 	 */
6003 	if (jump_table) {
6004 		/* Prepare the prefix jump action. */
6005 		append_index++;
6006 		jump_action = (void *)(actions_pre + actions_n + append_index);
6007 		jump_action->group = jump_table;
6008 		actions_pre[index++] =
6009 			(struct rte_flow_action){
6010 			.type = (enum rte_flow_action_type)
6011 				RTE_FLOW_ACTION_TYPE_JUMP,
6012 			.conf = jump_action,
6013 		};
6014 	}
6015 	actions_pre[index] = (struct rte_flow_action){
6016 		.type = (enum rte_flow_action_type)
6017 			RTE_FLOW_ACTION_TYPE_END,
6018 	};
6019 	/* Put the actions after sample into Suffix flow. */
6020 	memcpy(actions_sfx, actions + sample_action_pos + 1,
6021 	       sizeof(struct rte_flow_action) *
6022 	       (actions_n - sample_action_pos - 1));
6023 	return tag_id;
6024 }
6025 
6026 /**
6027  * The splitting for metadata feature.
6028  *
6029  * - Q/RSS action on NIC Rx should be split in order to pass by
6030  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
6031  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
6032  *
6033  * - All the actions on NIC Tx should have a mreg copy action to
6034  *   copy reg_a from WQE to reg_c[0].
6035  *
6036  * @param dev
6037  *   Pointer to Ethernet device.
6038  * @param[in] flow
6039  *   Parent flow structure pointer.
6040  * @param[in] attr
6041  *   Flow rule attributes.
6042  * @param[in] items
6043  *   Pattern specification (list terminated by the END pattern item).
6044  * @param[in] actions
6045  *   Associated actions (list terminated by the END action).
6046  * @param[in] flow_split_info
6047  *   Pointer to flow split info structure.
6048  * @param[out] error
6049  *   Perform verbose error reporting if not NULL.
6050  * @return
6051  *   0 on success, negative value otherwise
6052  */
6053 static int
6054 flow_create_split_metadata(struct rte_eth_dev *dev,
6055 			   struct rte_flow *flow,
6056 			   const struct rte_flow_attr *attr,
6057 			   const struct rte_flow_item items[],
6058 			   const struct rte_flow_action actions[],
6059 			   struct mlx5_flow_split_info *flow_split_info,
6060 			   struct rte_flow_error *error)
6061 {
6062 	struct mlx5_priv *priv = dev->data->dev_private;
6063 	struct mlx5_sh_config *config = &priv->sh->config;
6064 	const struct rte_flow_action *qrss = NULL;
6065 	struct rte_flow_action *ext_actions = NULL;
6066 	struct mlx5_flow *dev_flow = NULL;
6067 	uint32_t qrss_id = 0;
6068 	int mtr_sfx = 0;
6069 	size_t act_size;
6070 	int actions_n;
6071 	int encap_idx;
6072 	int ret;
6073 
6074 	/* Check whether extensive metadata feature is engaged. */
6075 	if (!config->dv_flow_en ||
6076 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
6077 	    !mlx5_flow_ext_mreg_supported(dev))
6078 		return flow_create_split_inner(dev, flow, NULL, attr, items,
6079 					       actions, flow_split_info, error);
6080 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
6081 							   &encap_idx);
6082 	if (qrss) {
6083 		/* Exclude hairpin flows from splitting. */
6084 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
6085 			const struct rte_flow_action_queue *queue;
6086 
6087 			queue = qrss->conf;
6088 			if (mlx5_rxq_is_hairpin(dev, queue->index))
6089 				qrss = NULL;
6090 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
6091 			const struct rte_flow_action_rss *rss;
6092 
6093 			rss = qrss->conf;
6094 			if (mlx5_rxq_is_hairpin(dev, rss->queue[0]))
6095 				qrss = NULL;
6096 		}
6097 	}
6098 	if (qrss) {
6099 		/* Check if it is in meter suffix table. */
6100 		mtr_sfx = attr->group == (attr->transfer ?
6101 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6102 			  MLX5_FLOW_TABLE_LEVEL_METER);
6103 		/*
6104 		 * Q/RSS action on NIC Rx should be split in order to pass by
6105 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
6106 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
6107 		 */
6108 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
6109 			   sizeof(struct rte_flow_action_set_tag) +
6110 			   sizeof(struct rte_flow_action_jump);
6111 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
6112 					  SOCKET_ID_ANY);
6113 		if (!ext_actions)
6114 			return rte_flow_error_set(error, ENOMEM,
6115 						  RTE_FLOW_ERROR_TYPE_ACTION,
6116 						  NULL, "no memory to split "
6117 						  "metadata flow");
6118 		/*
6119 		 * Create the new actions list with removed Q/RSS action
6120 		 * and appended set tag and jump to register copy table
6121 		 * (RX_CP_TBL). We should preallocate unique tag ID here
6122 		 * in advance, because it is needed for set tag action.
6123 		 */
6124 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
6125 						    qrss, actions_n,
6126 						    mtr_sfx, error);
6127 		if (!mtr_sfx && !qrss_id) {
6128 			ret = -rte_errno;
6129 			goto exit;
6130 		}
6131 	} else if (attr->egress && !attr->transfer) {
6132 		/*
6133 		 * All the actions on NIC Tx should have a metadata register
6134 		 * copy action to copy reg_a from WQE to reg_c[meta]
6135 		 */
6136 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
6137 			   sizeof(struct mlx5_flow_action_copy_mreg);
6138 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
6139 					  SOCKET_ID_ANY);
6140 		if (!ext_actions)
6141 			return rte_flow_error_set(error, ENOMEM,
6142 						  RTE_FLOW_ERROR_TYPE_ACTION,
6143 						  NULL, "no memory to split "
6144 						  "metadata flow");
6145 		/* Create the action list appended with copy register. */
6146 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
6147 					     actions_n, error, encap_idx);
6148 		if (ret < 0)
6149 			goto exit;
6150 	}
6151 	/* Add the unmodified original or prefix subflow. */
6152 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6153 				      items, ext_actions ? ext_actions :
6154 				      actions, flow_split_info, error);
6155 	if (ret < 0)
6156 		goto exit;
6157 	MLX5_ASSERT(dev_flow);
6158 	if (qrss) {
6159 		const struct rte_flow_attr q_attr = {
6160 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
6161 			.ingress = 1,
6162 		};
6163 		/* Internal PMD action to set register. */
6164 		struct mlx5_rte_flow_item_tag q_tag_spec = {
6165 			.data = qrss_id,
6166 			.id = REG_NON,
6167 		};
6168 		struct rte_flow_item q_items[] = {
6169 			{
6170 				.type = (enum rte_flow_item_type)
6171 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
6172 				.spec = &q_tag_spec,
6173 				.last = NULL,
6174 				.mask = NULL,
6175 			},
6176 			{
6177 				.type = RTE_FLOW_ITEM_TYPE_END,
6178 			},
6179 		};
6180 		struct rte_flow_action q_actions[] = {
6181 			{
6182 				.type = qrss->type,
6183 				.conf = qrss->conf,
6184 			},
6185 			{
6186 				.type = RTE_FLOW_ACTION_TYPE_END,
6187 			},
6188 		};
6189 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
6190 
6191 		/*
6192 		 * Configure the tag item only if there is no meter subflow.
6193 		 * Since tag is already marked in the meter suffix subflow
6194 		 * we can just use the meter suffix items as is.
6195 		 */
6196 		if (qrss_id) {
6197 			/* Not meter subflow. */
6198 			MLX5_ASSERT(!mtr_sfx);
6199 			/*
6200 			 * Put unique id in prefix flow due to it is destroyed
6201 			 * after suffix flow and id will be freed after there
6202 			 * is no actual flows with this id and identifier
6203 			 * reallocation becomes possible (for example, for
6204 			 * other flows in other threads).
6205 			 */
6206 			dev_flow->handle->split_flow_id = qrss_id;
6207 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
6208 						   error);
6209 			if (ret < 0)
6210 				goto exit;
6211 			q_tag_spec.id = ret;
6212 		}
6213 		dev_flow = NULL;
6214 		/* Add suffix subflow to execute Q/RSS. */
6215 		flow_split_info->prefix_layers = layers;
6216 		flow_split_info->prefix_mark = 0;
6217 		flow_split_info->table_id = 0;
6218 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6219 					      &q_attr, mtr_sfx ? items :
6220 					      q_items, q_actions,
6221 					      flow_split_info, error);
6222 		if (ret < 0)
6223 			goto exit;
6224 		/* qrss ID should be freed if failed. */
6225 		qrss_id = 0;
6226 		MLX5_ASSERT(dev_flow);
6227 	}
6228 
6229 exit:
6230 	/*
6231 	 * We do not destroy the partially created sub_flows in case of error.
6232 	 * These ones are included into parent flow list and will be destroyed
6233 	 * by flow_drv_destroy.
6234 	 */
6235 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
6236 			qrss_id);
6237 	mlx5_free(ext_actions);
6238 	return ret;
6239 }
6240 
6241 /**
6242  * Create meter internal drop flow with the original pattern.
6243  *
6244  * @param dev
6245  *   Pointer to Ethernet device.
6246  * @param[in] flow
6247  *   Parent flow structure pointer.
6248  * @param[in] attr
6249  *   Flow rule attributes.
6250  * @param[in] items
6251  *   Pattern specification (list terminated by the END pattern item).
6252  * @param[in] flow_split_info
6253  *   Pointer to flow split info structure.
6254  * @param[in] fm
6255  *   Pointer to flow meter structure.
6256  * @param[out] error
6257  *   Perform verbose error reporting if not NULL.
6258  * @return
6259  *   0 on success, negative value otherwise
6260  */
6261 static uint32_t
6262 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
6263 			struct rte_flow *flow,
6264 			const struct rte_flow_attr *attr,
6265 			const struct rte_flow_item items[],
6266 			struct mlx5_flow_split_info *flow_split_info,
6267 			struct mlx5_flow_meter_info *fm,
6268 			struct rte_flow_error *error)
6269 {
6270 	struct mlx5_flow *dev_flow = NULL;
6271 	struct rte_flow_attr drop_attr = *attr;
6272 	struct rte_flow_action drop_actions[3];
6273 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
6274 
6275 	MLX5_ASSERT(fm->drop_cnt);
6276 	drop_actions[0].type =
6277 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
6278 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
6279 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
6280 	drop_actions[1].conf = NULL;
6281 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
6282 	drop_actions[2].conf = NULL;
6283 	drop_split_info.external = false;
6284 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6285 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
6286 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
6287 	return flow_create_split_inner(dev, flow, &dev_flow,
6288 				&drop_attr, items, drop_actions,
6289 				&drop_split_info, error);
6290 }
6291 
6292 /**
6293  * The splitting for meter feature.
6294  *
6295  * - The meter flow will be split to two flows as prefix and
6296  *   suffix flow. The packets make sense only it pass the prefix
6297  *   meter action.
6298  *
6299  * - Reg_C_5 is used for the packet to match betweend prefix and
6300  *   suffix flow.
6301  *
6302  * @param dev
6303  *   Pointer to Ethernet device.
6304  * @param[in] flow
6305  *   Parent flow structure pointer.
6306  * @param[in] attr
6307  *   Flow rule attributes.
6308  * @param[in] items
6309  *   Pattern specification (list terminated by the END pattern item).
6310  * @param[in] actions
6311  *   Associated actions (list terminated by the END action).
6312  * @param[in] flow_split_info
6313  *   Pointer to flow split info structure.
6314  * @param[out] error
6315  *   Perform verbose error reporting if not NULL.
6316  * @return
6317  *   0 on success, negative value otherwise
6318  */
6319 static int
6320 flow_create_split_meter(struct rte_eth_dev *dev,
6321 			struct rte_flow *flow,
6322 			const struct rte_flow_attr *attr,
6323 			const struct rte_flow_item items[],
6324 			const struct rte_flow_action actions[],
6325 			struct mlx5_flow_split_info *flow_split_info,
6326 			struct rte_flow_error *error)
6327 {
6328 	struct mlx5_priv *priv = dev->data->dev_private;
6329 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6330 	struct rte_flow_action *sfx_actions = NULL;
6331 	struct rte_flow_action *pre_actions = NULL;
6332 	struct rte_flow_item *sfx_items = NULL;
6333 	struct mlx5_flow *dev_flow = NULL;
6334 	struct rte_flow_attr sfx_attr = *attr;
6335 	struct mlx5_flow_meter_info *fm = NULL;
6336 	uint8_t skip_scale_restore;
6337 	bool has_mtr = false;
6338 	bool has_modify = false;
6339 	bool set_mtr_reg = true;
6340 	bool is_mtr_hierarchy = false;
6341 	uint32_t meter_id = 0;
6342 	uint32_t mtr_idx = 0;
6343 	uint32_t mtr_flow_id = 0;
6344 	size_t act_size;
6345 	size_t item_size;
6346 	int actions_n = 0;
6347 	int ret = 0;
6348 
6349 	if (priv->mtr_en)
6350 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6351 						    &has_modify, &meter_id);
6352 	if (has_mtr) {
6353 		if (flow->meter) {
6354 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6355 			if (!fm)
6356 				return rte_flow_error_set(error, EINVAL,
6357 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6358 						NULL, "Meter not found.");
6359 		} else {
6360 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6361 			if (!fm)
6362 				return rte_flow_error_set(error, EINVAL,
6363 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6364 						NULL, "Meter not found.");
6365 			ret = mlx5_flow_meter_attach(priv, fm,
6366 						     &sfx_attr, error);
6367 			if (ret)
6368 				return -rte_errno;
6369 			flow->meter = mtr_idx;
6370 		}
6371 		MLX5_ASSERT(wks);
6372 		wks->fm = fm;
6373 		if (!fm->def_policy) {
6374 			wks->policy = mlx5_flow_meter_policy_find(dev,
6375 								  fm->policy_id,
6376 								  NULL);
6377 			MLX5_ASSERT(wks->policy);
6378 			if (wks->policy->mark)
6379 				wks->mark = 1;
6380 			if (wks->policy->is_hierarchy) {
6381 				wks->final_policy =
6382 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6383 								wks->policy);
6384 				if (!wks->final_policy)
6385 					return rte_flow_error_set(error,
6386 					EINVAL,
6387 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6388 				"Failed to find terminal policy of hierarchy.");
6389 				is_mtr_hierarchy = true;
6390 			}
6391 		}
6392 		/*
6393 		 * If it isn't default-policy Meter, and
6394 		 * 1. There's no action in flow to change
6395 		 *    packet (modify/encap/decap etc.), OR
6396 		 * 2. No drop count needed for this meter.
6397 		 * 3. It's not meter hierarchy.
6398 		 * Then no need to use regC to save meter id anymore.
6399 		 */
6400 		if (!fm->def_policy && !is_mtr_hierarchy &&
6401 		    (!has_modify || !fm->drop_cnt))
6402 			set_mtr_reg = false;
6403 		/* Prefix actions: meter, decap, encap, tag, jump, end, cnt. */
6404 #define METER_PREFIX_ACTION 7
6405 		act_size = (sizeof(struct rte_flow_action) *
6406 			    (actions_n + METER_PREFIX_ACTION)) +
6407 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6408 		/* Suffix items: tag, vlan, port id, end. */
6409 #define METER_SUFFIX_ITEM 4
6410 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6411 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6412 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6413 					  0, SOCKET_ID_ANY);
6414 		if (!sfx_actions)
6415 			return rte_flow_error_set(error, ENOMEM,
6416 						  RTE_FLOW_ERROR_TYPE_ACTION,
6417 						  NULL, "no memory to split "
6418 						  "meter flow");
6419 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6420 			     act_size);
6421 		/* There's no suffix flow for meter of non-default policy. */
6422 		if (!fm->def_policy)
6423 			pre_actions = sfx_actions + 1;
6424 		else
6425 			pre_actions = sfx_actions + actions_n;
6426 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6427 					    items, sfx_items, actions,
6428 					    sfx_actions, pre_actions,
6429 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6430 					    error);
6431 		if (ret) {
6432 			ret = -rte_errno;
6433 			goto exit;
6434 		}
6435 		/* Add the prefix subflow. */
6436 		skip_scale_restore = flow_split_info->skip_scale;
6437 		flow_split_info->skip_scale |=
6438 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6439 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6440 					      attr, items, pre_actions,
6441 					      flow_split_info, error);
6442 		flow_split_info->skip_scale = skip_scale_restore;
6443 		if (ret) {
6444 			if (mtr_flow_id)
6445 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6446 			ret = -rte_errno;
6447 			goto exit;
6448 		}
6449 		if (mtr_flow_id) {
6450 			dev_flow->handle->split_flow_id = mtr_flow_id;
6451 			dev_flow->handle->is_meter_flow_id = 1;
6452 		}
6453 		if (!fm->def_policy) {
6454 			if (!set_mtr_reg && fm->drop_cnt)
6455 				ret =
6456 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6457 							&sfx_attr, items,
6458 							flow_split_info,
6459 							fm, error);
6460 			goto exit;
6461 		}
6462 		/* Setting the sfx group atrr. */
6463 		sfx_attr.group = sfx_attr.transfer ?
6464 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6465 				 MLX5_FLOW_TABLE_LEVEL_METER;
6466 		flow_split_info->prefix_layers =
6467 				flow_get_prefix_layer_flags(dev_flow);
6468 		flow_split_info->prefix_mark |= wks->mark;
6469 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6470 	}
6471 	/* Add the prefix subflow. */
6472 	ret = flow_create_split_metadata(dev, flow,
6473 					 &sfx_attr, sfx_items ?
6474 					 sfx_items : items,
6475 					 sfx_actions ? sfx_actions : actions,
6476 					 flow_split_info, error);
6477 exit:
6478 	if (sfx_actions)
6479 		mlx5_free(sfx_actions);
6480 	return ret;
6481 }
6482 
6483 /**
6484  * The splitting for sample feature.
6485  *
6486  * Once Sample action is detected in the action list, the flow actions should
6487  * be split into prefix sub flow and suffix sub flow.
6488  *
6489  * The original items remain in the prefix sub flow, all actions preceding the
6490  * sample action and the sample action itself will be copied to the prefix
6491  * sub flow, the actions following the sample action will be copied to the
6492  * suffix sub flow, Queue action always be located in the suffix sub flow.
6493  *
6494  * In order to make the packet from prefix sub flow matches with suffix sub
6495  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6496  * flow uses tag item with the unique flow id.
6497  *
6498  * @param dev
6499  *   Pointer to Ethernet device.
6500  * @param[in] flow
6501  *   Parent flow structure pointer.
6502  * @param[in] attr
6503  *   Flow rule attributes.
6504  * @param[in] items
6505  *   Pattern specification (list terminated by the END pattern item).
6506  * @param[in] actions
6507  *   Associated actions (list terminated by the END action).
6508  * @param[in] flow_split_info
6509  *   Pointer to flow split info structure.
6510  * @param[out] error
6511  *   Perform verbose error reporting if not NULL.
6512  * @return
6513  *   0 on success, negative value otherwise
6514  */
6515 static int
6516 flow_create_split_sample(struct rte_eth_dev *dev,
6517 			 struct rte_flow *flow,
6518 			 const struct rte_flow_attr *attr,
6519 			 const struct rte_flow_item items[],
6520 			 const struct rte_flow_action actions[],
6521 			 struct mlx5_flow_split_info *flow_split_info,
6522 			 struct rte_flow_error *error)
6523 {
6524 	struct mlx5_priv *priv = dev->data->dev_private;
6525 	struct rte_flow_action *sfx_actions = NULL;
6526 	struct rte_flow_action *pre_actions = NULL;
6527 	struct rte_flow_item *sfx_items = NULL;
6528 	struct mlx5_flow *dev_flow = NULL;
6529 	struct rte_flow_attr sfx_attr = *attr;
6530 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6531 	struct mlx5_flow_dv_sample_resource *sample_res;
6532 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6533 	struct mlx5_flow_tbl_resource *sfx_tbl;
6534 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6535 #endif
6536 	size_t act_size;
6537 	size_t item_size;
6538 	uint32_t fdb_tx = 0;
6539 	int32_t tag_id = 0;
6540 	int actions_n = 0;
6541 	int sample_action_pos;
6542 	int qrss_action_pos;
6543 	int add_tag = 0;
6544 	int modify_after_mirror = 0;
6545 	uint16_t jump_table = 0;
6546 	const uint32_t next_ft_step = 1;
6547 	int ret = 0;
6548 
6549 	if (priv->sampler_en)
6550 		actions_n = flow_check_match_action(actions, attr,
6551 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6552 					&sample_action_pos, &qrss_action_pos,
6553 					&modify_after_mirror);
6554 	if (actions_n) {
6555 		/* The prefix actions must includes sample, tag, end. */
6556 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6557 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6558 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6559 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6560 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6561 					  item_size), 0, SOCKET_ID_ANY);
6562 		if (!sfx_actions)
6563 			return rte_flow_error_set(error, ENOMEM,
6564 						  RTE_FLOW_ERROR_TYPE_ACTION,
6565 						  NULL, "no memory to split "
6566 						  "sample flow");
6567 		/* The representor_id is UINT16_MAX for uplink. */
6568 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6569 		/*
6570 		 * When reg_c_preserve is set, metadata registers Cx preserve
6571 		 * their value even through packet duplication.
6572 		 */
6573 		add_tag = (!fdb_tx ||
6574 			   priv->sh->cdev->config.hca_attr.reg_c_preserve);
6575 		if (add_tag)
6576 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6577 					+ act_size);
6578 		if (modify_after_mirror)
6579 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6580 				     next_ft_step;
6581 		pre_actions = sfx_actions + actions_n;
6582 		tag_id = flow_sample_split_prep(dev, add_tag, items, sfx_items,
6583 						actions, sfx_actions,
6584 						pre_actions, actions_n,
6585 						sample_action_pos,
6586 						qrss_action_pos, jump_table,
6587 						error);
6588 		if (tag_id < 0 || (add_tag && !tag_id)) {
6589 			ret = -rte_errno;
6590 			goto exit;
6591 		}
6592 		if (modify_after_mirror)
6593 			flow_split_info->skip_scale =
6594 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6595 		/* Add the prefix subflow. */
6596 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6597 					      items, pre_actions,
6598 					      flow_split_info, error);
6599 		if (ret) {
6600 			ret = -rte_errno;
6601 			goto exit;
6602 		}
6603 		dev_flow->handle->split_flow_id = tag_id;
6604 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6605 		if (!modify_after_mirror) {
6606 			/* Set the sfx group attr. */
6607 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6608 						dev_flow->dv.sample_res;
6609 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6610 						sample_res->normal_path_tbl;
6611 			sfx_tbl_data = container_of(sfx_tbl,
6612 						struct mlx5_flow_tbl_data_entry,
6613 						tbl);
6614 			sfx_attr.group = sfx_attr.transfer ?
6615 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6616 		} else {
6617 			MLX5_ASSERT(attr->transfer);
6618 			sfx_attr.group = jump_table;
6619 		}
6620 		flow_split_info->prefix_layers =
6621 				flow_get_prefix_layer_flags(dev_flow);
6622 		MLX5_ASSERT(wks);
6623 		flow_split_info->prefix_mark |= wks->mark;
6624 		/* Suffix group level already be scaled with factor, set
6625 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6626 		 * again in translation.
6627 		 */
6628 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6629 #endif
6630 	}
6631 	/* Add the suffix subflow. */
6632 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6633 				      sfx_items ? sfx_items : items,
6634 				      sfx_actions ? sfx_actions : actions,
6635 				      flow_split_info, error);
6636 exit:
6637 	if (sfx_actions)
6638 		mlx5_free(sfx_actions);
6639 	return ret;
6640 }
6641 
6642 /**
6643  * Split the flow to subflow set. The splitters might be linked
6644  * in the chain, like this:
6645  * flow_create_split_outer() calls:
6646  *   flow_create_split_meter() calls:
6647  *     flow_create_split_metadata(meter_subflow_0) calls:
6648  *       flow_create_split_inner(metadata_subflow_0)
6649  *       flow_create_split_inner(metadata_subflow_1)
6650  *       flow_create_split_inner(metadata_subflow_2)
6651  *     flow_create_split_metadata(meter_subflow_1) calls:
6652  *       flow_create_split_inner(metadata_subflow_0)
6653  *       flow_create_split_inner(metadata_subflow_1)
6654  *       flow_create_split_inner(metadata_subflow_2)
6655  *
6656  * This provide flexible way to add new levels of flow splitting.
6657  * The all of successfully created subflows are included to the
6658  * parent flow dev_flow list.
6659  *
6660  * @param dev
6661  *   Pointer to Ethernet device.
6662  * @param[in] flow
6663  *   Parent flow structure pointer.
6664  * @param[in] attr
6665  *   Flow rule attributes.
6666  * @param[in] items
6667  *   Pattern specification (list terminated by the END pattern item).
6668  * @param[in] actions
6669  *   Associated actions (list terminated by the END action).
6670  * @param[in] flow_split_info
6671  *   Pointer to flow split info structure.
6672  * @param[out] error
6673  *   Perform verbose error reporting if not NULL.
6674  * @return
6675  *   0 on success, negative value otherwise
6676  */
6677 static int
6678 flow_create_split_outer(struct rte_eth_dev *dev,
6679 			struct rte_flow *flow,
6680 			const struct rte_flow_attr *attr,
6681 			const struct rte_flow_item items[],
6682 			const struct rte_flow_action actions[],
6683 			struct mlx5_flow_split_info *flow_split_info,
6684 			struct rte_flow_error *error)
6685 {
6686 	int ret;
6687 
6688 	ret = flow_create_split_sample(dev, flow, attr, items,
6689 				       actions, flow_split_info, error);
6690 	MLX5_ASSERT(ret <= 0);
6691 	return ret;
6692 }
6693 
6694 static inline struct mlx5_flow_tunnel *
6695 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6696 {
6697 	struct mlx5_flow_tunnel *tunnel;
6698 
6699 #pragma GCC diagnostic push
6700 #pragma GCC diagnostic ignored "-Wcast-qual"
6701 	tunnel = (typeof(tunnel))flow->tunnel;
6702 #pragma GCC diagnostic pop
6703 
6704 	return tunnel;
6705 }
6706 
6707 /**
6708  * Adjust flow RSS workspace if needed.
6709  *
6710  * @param wks
6711  *   Pointer to thread flow work space.
6712  * @param rss_desc
6713  *   Pointer to RSS descriptor.
6714  * @param[in] nrssq_num
6715  *   New RSS queue number.
6716  *
6717  * @return
6718  *   0 on success, -1 otherwise and rte_errno is set.
6719  */
6720 static int
6721 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6722 			  struct mlx5_flow_rss_desc *rss_desc,
6723 			  uint32_t nrssq_num)
6724 {
6725 	if (likely(nrssq_num <= wks->rssq_num))
6726 		return 0;
6727 	rss_desc->queue = realloc(rss_desc->queue,
6728 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6729 	if (!rss_desc->queue) {
6730 		rte_errno = ENOMEM;
6731 		return -1;
6732 	}
6733 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6734 	return 0;
6735 }
6736 
6737 /**
6738  * Create a flow and add it to @p list.
6739  *
6740  * @param dev
6741  *   Pointer to Ethernet device.
6742  * @param list
6743  *   Pointer to a TAILQ flow list. If this parameter NULL,
6744  *   no list insertion occurred, flow is just created,
6745  *   this is caller's responsibility to track the
6746  *   created flow.
6747  * @param[in] attr
6748  *   Flow rule attributes.
6749  * @param[in] items
6750  *   Pattern specification (list terminated by the END pattern item).
6751  * @param[in] actions
6752  *   Associated actions (list terminated by the END action).
6753  * @param[in] external
6754  *   This flow rule is created by request external to PMD.
6755  * @param[out] error
6756  *   Perform verbose error reporting if not NULL.
6757  *
6758  * @return
6759  *   A flow index on success, 0 otherwise and rte_errno is set.
6760  */
6761 static uint32_t
6762 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6763 		 const struct rte_flow_attr *attr,
6764 		 const struct rte_flow_item items[],
6765 		 const struct rte_flow_action original_actions[],
6766 		 bool external, struct rte_flow_error *error)
6767 {
6768 	struct mlx5_priv *priv = dev->data->dev_private;
6769 	struct rte_flow *flow = NULL;
6770 	struct mlx5_flow *dev_flow;
6771 	const struct rte_flow_action_rss *rss = NULL;
6772 	struct mlx5_translated_action_handle
6773 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6774 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6775 	union {
6776 		struct mlx5_flow_expand_rss buf;
6777 		uint8_t buffer[4096];
6778 	} expand_buffer;
6779 	union {
6780 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6781 		uint8_t buffer[2048];
6782 	} actions_rx;
6783 	union {
6784 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6785 		uint8_t buffer[2048];
6786 	} actions_hairpin_tx;
6787 	union {
6788 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6789 		uint8_t buffer[2048];
6790 	} items_tx;
6791 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6792 	struct mlx5_flow_rss_desc *rss_desc;
6793 	const struct rte_flow_action *p_actions_rx;
6794 	uint32_t i;
6795 	uint32_t idx = 0;
6796 	int hairpin_flow;
6797 	struct rte_flow_attr attr_tx = { .priority = 0 };
6798 	const struct rte_flow_action *actions;
6799 	struct rte_flow_action *translated_actions = NULL;
6800 	struct mlx5_flow_tunnel *tunnel;
6801 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6802 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6803 	struct mlx5_flow_split_info flow_split_info = {
6804 		.external = !!external,
6805 		.skip_scale = 0,
6806 		.flow_idx = 0,
6807 		.prefix_mark = 0,
6808 		.prefix_layers = 0,
6809 		.table_id = 0
6810 	};
6811 	int ret;
6812 
6813 	MLX5_ASSERT(wks);
6814 	rss_desc = &wks->rss_desc;
6815 	ret = flow_action_handles_translate(dev, original_actions,
6816 					    indir_actions,
6817 					    &indir_actions_n,
6818 					    &translated_actions, error);
6819 	if (ret < 0) {
6820 		MLX5_ASSERT(translated_actions == NULL);
6821 		return 0;
6822 	}
6823 	actions = translated_actions ? translated_actions : original_actions;
6824 	p_actions_rx = actions;
6825 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6826 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6827 				external, hairpin_flow, error);
6828 	if (ret < 0)
6829 		goto error_before_hairpin_split;
6830 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6831 	if (!flow) {
6832 		rte_errno = ENOMEM;
6833 		goto error_before_hairpin_split;
6834 	}
6835 	if (hairpin_flow > 0) {
6836 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6837 			rte_errno = EINVAL;
6838 			goto error_before_hairpin_split;
6839 		}
6840 		flow_hairpin_split(dev, actions, actions_rx.actions,
6841 				   actions_hairpin_tx.actions, items_tx.items,
6842 				   idx);
6843 		p_actions_rx = actions_rx.actions;
6844 	}
6845 	flow_split_info.flow_idx = idx;
6846 	flow->drv_type = flow_get_drv_type(dev, attr);
6847 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6848 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6849 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6850 	/* RSS Action only works on NIC RX domain */
6851 	if (attr->ingress && !attr->transfer)
6852 		rss = flow_get_rss_action(dev, p_actions_rx);
6853 	if (rss) {
6854 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6855 			return 0;
6856 		/*
6857 		 * The following information is required by
6858 		 * mlx5_flow_hashfields_adjust() in advance.
6859 		 */
6860 		rss_desc->level = rss->level;
6861 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6862 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6863 	}
6864 	flow->dev_handles = 0;
6865 	if (rss && rss->types) {
6866 		unsigned int graph_root;
6867 
6868 		graph_root = find_graph_root(rss->level);
6869 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6870 					   items, rss->types,
6871 					   mlx5_support_expansion, graph_root);
6872 		MLX5_ASSERT(ret > 0 &&
6873 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6874 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6875 			for (i = 0; i < buf->entries; ++i)
6876 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6877 		}
6878 	} else {
6879 		buf->entries = 1;
6880 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6881 	}
6882 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6883 						      indir_actions_n);
6884 	for (i = 0; i < buf->entries; ++i) {
6885 		/* Initialize flow split data. */
6886 		flow_split_info.prefix_layers = 0;
6887 		flow_split_info.prefix_mark = 0;
6888 		flow_split_info.skip_scale = 0;
6889 		/*
6890 		 * The splitter may create multiple dev_flows,
6891 		 * depending on configuration. In the simplest
6892 		 * case it just creates unmodified original flow.
6893 		 */
6894 		ret = flow_create_split_outer(dev, flow, attr,
6895 					      buf->entry[i].pattern,
6896 					      p_actions_rx, &flow_split_info,
6897 					      error);
6898 		if (ret < 0)
6899 			goto error;
6900 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6901 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6902 							   p_actions_rx,
6903 							   idx,
6904 							   wks->flows[0].tunnel,
6905 							   &default_miss_ctx,
6906 							   error);
6907 			if (ret < 0) {
6908 				mlx5_free(default_miss_ctx.queue);
6909 				goto error;
6910 			}
6911 		}
6912 	}
6913 	/* Create the tx flow. */
6914 	if (hairpin_flow) {
6915 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6916 		attr_tx.ingress = 0;
6917 		attr_tx.egress = 1;
6918 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6919 					 actions_hairpin_tx.actions,
6920 					 idx, error);
6921 		if (!dev_flow)
6922 			goto error;
6923 		dev_flow->flow = flow;
6924 		dev_flow->external = 0;
6925 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6926 			      dev_flow->handle, next);
6927 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6928 					 items_tx.items,
6929 					 actions_hairpin_tx.actions, error);
6930 		if (ret < 0)
6931 			goto error;
6932 	}
6933 	/*
6934 	 * Update the metadata register copy table. If extensive
6935 	 * metadata feature is enabled and registers are supported
6936 	 * we might create the extra rte_flow for each unique
6937 	 * MARK/FLAG action ID.
6938 	 *
6939 	 * The table is updated for ingress Flows only, because
6940 	 * the egress Flows belong to the different device and
6941 	 * copy table should be updated in peer NIC Rx domain.
6942 	 */
6943 	if (attr->ingress &&
6944 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6945 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6946 		if (ret)
6947 			goto error;
6948 	}
6949 	/*
6950 	 * If the flow is external (from application) OR device is started,
6951 	 * OR mreg discover, then apply immediately.
6952 	 */
6953 	if (external || dev->data->dev_started ||
6954 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6955 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6956 		ret = flow_drv_apply(dev, flow, error);
6957 		if (ret < 0)
6958 			goto error;
6959 	}
6960 	flow->type = type;
6961 	flow_rxq_flags_set(dev, flow);
6962 	rte_free(translated_actions);
6963 	tunnel = flow_tunnel_from_rule(wks->flows);
6964 	if (tunnel) {
6965 		flow->tunnel = 1;
6966 		flow->tunnel_id = tunnel->tunnel_id;
6967 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6968 		mlx5_free(default_miss_ctx.queue);
6969 	}
6970 	mlx5_flow_pop_thread_workspace();
6971 	return idx;
6972 error:
6973 	MLX5_ASSERT(flow);
6974 	ret = rte_errno; /* Save rte_errno before cleanup. */
6975 	flow_mreg_del_copy_action(dev, flow);
6976 	flow_drv_destroy(dev, flow);
6977 	if (rss_desc->shared_rss)
6978 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6979 			mlx5_ipool_get
6980 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6981 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6982 	mlx5_ipool_free(priv->flows[type], idx);
6983 	rte_errno = ret; /* Restore rte_errno. */
6984 	ret = rte_errno;
6985 	rte_errno = ret;
6986 	mlx5_flow_pop_thread_workspace();
6987 error_before_hairpin_split:
6988 	rte_free(translated_actions);
6989 	return 0;
6990 }
6991 
6992 /**
6993  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6994  * incoming packets to table 1.
6995  *
6996  * Other flow rules, requested for group n, will be created in
6997  * e-switch table n+1.
6998  * Jump action to e-switch group n will be created to group n+1.
6999  *
7000  * Used when working in switchdev mode, to utilise advantages of table 1
7001  * and above.
7002  *
7003  * @param dev
7004  *   Pointer to Ethernet device.
7005  *
7006  * @return
7007  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
7008  */
7009 struct rte_flow *
7010 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
7011 {
7012 	const struct rte_flow_attr attr = {
7013 		.group = 0,
7014 		.priority = 0,
7015 		.ingress = 1,
7016 		.egress = 0,
7017 		.transfer = 1,
7018 	};
7019 	const struct rte_flow_item pattern = {
7020 		.type = RTE_FLOW_ITEM_TYPE_END,
7021 	};
7022 	struct rte_flow_action_jump jump = {
7023 		.group = 1,
7024 	};
7025 	const struct rte_flow_action actions[] = {
7026 		{
7027 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
7028 			.conf = &jump,
7029 		},
7030 		{
7031 			.type = RTE_FLOW_ACTION_TYPE_END,
7032 		},
7033 	};
7034 	struct rte_flow_error error;
7035 
7036 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7037 						   &attr, &pattern,
7038 						   actions, false, &error);
7039 }
7040 
7041 /**
7042  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
7043  * and sq number, directs all packets to peer vport.
7044  *
7045  * @param dev
7046  *   Pointer to Ethernet device.
7047  * @param txq
7048  *   Txq index.
7049  *
7050  * @return
7051  *   Flow ID on success, 0 otherwise and rte_errno is set.
7052  */
7053 uint32_t
7054 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
7055 {
7056 	struct rte_flow_attr attr = {
7057 		.group = 0,
7058 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7059 		.ingress = 1,
7060 		.egress = 0,
7061 		.transfer = 1,
7062 	};
7063 	struct rte_flow_item_port_id port_spec = {
7064 		.id = MLX5_PORT_ESW_MGR,
7065 	};
7066 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
7067 		.queue = txq,
7068 	};
7069 	struct rte_flow_item pattern[] = {
7070 		{
7071 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
7072 			.spec = &port_spec,
7073 		},
7074 		{
7075 			.type = (enum rte_flow_item_type)
7076 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7077 			.spec = &txq_spec,
7078 		},
7079 		{
7080 			.type = RTE_FLOW_ITEM_TYPE_END,
7081 		},
7082 	};
7083 	struct rte_flow_action_jump jump = {
7084 		.group = 1,
7085 	};
7086 	struct rte_flow_action_port_id port = {
7087 		.id = dev->data->port_id,
7088 	};
7089 	struct rte_flow_action actions[] = {
7090 		{
7091 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
7092 			.conf = &jump,
7093 		},
7094 		{
7095 			.type = RTE_FLOW_ACTION_TYPE_END,
7096 		},
7097 	};
7098 	struct rte_flow_error error;
7099 
7100 	/*
7101 	 * Creates group 0, highest priority jump flow.
7102 	 * Matches txq to bypass kernel packets.
7103 	 */
7104 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
7105 			     false, &error) == 0)
7106 		return 0;
7107 	/* Create group 1, lowest priority redirect flow for txq. */
7108 	attr.group = 1;
7109 	actions[0].conf = &port;
7110 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
7111 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
7112 				actions, false, &error);
7113 }
7114 
7115 /**
7116  * Validate a flow supported by the NIC.
7117  *
7118  * @see rte_flow_validate()
7119  * @see rte_flow_ops
7120  */
7121 int
7122 mlx5_flow_validate(struct rte_eth_dev *dev,
7123 		   const struct rte_flow_attr *attr,
7124 		   const struct rte_flow_item items[],
7125 		   const struct rte_flow_action original_actions[],
7126 		   struct rte_flow_error *error)
7127 {
7128 	int hairpin_flow;
7129 	struct mlx5_translated_action_handle
7130 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
7131 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
7132 	const struct rte_flow_action *actions;
7133 	struct rte_flow_action *translated_actions = NULL;
7134 	int ret = flow_action_handles_translate(dev, original_actions,
7135 						indir_actions,
7136 						&indir_actions_n,
7137 						&translated_actions, error);
7138 
7139 	if (ret)
7140 		return ret;
7141 	actions = translated_actions ? translated_actions : original_actions;
7142 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
7143 	ret = flow_drv_validate(dev, attr, items, actions,
7144 				true, hairpin_flow, error);
7145 	rte_free(translated_actions);
7146 	return ret;
7147 }
7148 
7149 /**
7150  * Create a flow.
7151  *
7152  * @see rte_flow_create()
7153  * @see rte_flow_ops
7154  */
7155 struct rte_flow *
7156 mlx5_flow_create(struct rte_eth_dev *dev,
7157 		 const struct rte_flow_attr *attr,
7158 		 const struct rte_flow_item items[],
7159 		 const struct rte_flow_action actions[],
7160 		 struct rte_flow_error *error)
7161 {
7162 	struct mlx5_priv *priv = dev->data->dev_private;
7163 
7164 	if (priv->sh->config.dv_flow_en == 2) {
7165 		rte_flow_error_set(error, ENOTSUP,
7166 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7167 			  NULL,
7168 			  "Flow non-Q creation not supported");
7169 		return NULL;
7170 	}
7171 	/*
7172 	 * If the device is not started yet, it is not allowed to created a
7173 	 * flow from application. PMD default flows and traffic control flows
7174 	 * are not affected.
7175 	 */
7176 	if (unlikely(!dev->data->dev_started)) {
7177 		DRV_LOG(DEBUG, "port %u is not started when "
7178 			"inserting a flow", dev->data->port_id);
7179 		rte_flow_error_set(error, ENODEV,
7180 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7181 				   NULL,
7182 				   "port not started");
7183 		return NULL;
7184 	}
7185 
7186 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
7187 						   attr, items, actions,
7188 						   true, error);
7189 }
7190 
7191 /**
7192  * Destroy a flow in a list.
7193  *
7194  * @param dev
7195  *   Pointer to Ethernet device.
7196  * @param[in] flow_idx
7197  *   Index of flow to destroy.
7198  */
7199 static void
7200 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
7201 		  uint32_t flow_idx)
7202 {
7203 	struct mlx5_priv *priv = dev->data->dev_private;
7204 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
7205 
7206 	if (!flow)
7207 		return;
7208 	MLX5_ASSERT(flow->type == type);
7209 	/*
7210 	 * Update RX queue flags only if port is started, otherwise it is
7211 	 * already clean.
7212 	 */
7213 	if (dev->data->dev_started)
7214 		flow_rxq_flags_trim(dev, flow);
7215 	flow_drv_destroy(dev, flow);
7216 	if (flow->tunnel) {
7217 		struct mlx5_flow_tunnel *tunnel;
7218 
7219 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
7220 		RTE_VERIFY(tunnel);
7221 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
7222 			mlx5_flow_tunnel_free(dev, tunnel);
7223 	}
7224 	flow_mreg_del_copy_action(dev, flow);
7225 	mlx5_ipool_free(priv->flows[type], flow_idx);
7226 }
7227 
7228 /**
7229  * Destroy all flows.
7230  *
7231  * @param dev
7232  *   Pointer to Ethernet device.
7233  * @param type
7234  *   Flow type to be flushed.
7235  * @param active
7236  *   If flushing is called actively.
7237  */
7238 void
7239 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
7240 		     bool active)
7241 {
7242 	struct mlx5_priv *priv = dev->data->dev_private;
7243 	uint32_t num_flushed = 0, fidx = 1;
7244 	struct rte_flow *flow;
7245 
7246 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
7247 	if (priv->sh->config.dv_flow_en == 2 &&
7248 	    type == MLX5_FLOW_TYPE_GEN) {
7249 		flow_hw_q_flow_flush(dev, NULL);
7250 		return;
7251 	}
7252 #endif
7253 
7254 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
7255 		flow_list_destroy(dev, type, fidx);
7256 		num_flushed++;
7257 	}
7258 	if (active) {
7259 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
7260 			dev->data->port_id, num_flushed);
7261 	}
7262 }
7263 
7264 /**
7265  * Stop all default actions for flows.
7266  *
7267  * @param dev
7268  *   Pointer to Ethernet device.
7269  */
7270 void
7271 mlx5_flow_stop_default(struct rte_eth_dev *dev)
7272 {
7273 	flow_mreg_del_default_copy_action(dev);
7274 	flow_rxq_flags_clear(dev);
7275 }
7276 
7277 /**
7278  * Start all default actions for flows.
7279  *
7280  * @param dev
7281  *   Pointer to Ethernet device.
7282  * @return
7283  *   0 on success, a negative errno value otherwise and rte_errno is set.
7284  */
7285 int
7286 mlx5_flow_start_default(struct rte_eth_dev *dev)
7287 {
7288 	struct rte_flow_error error;
7289 
7290 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
7291 	return flow_mreg_add_default_copy_action(dev, &error);
7292 }
7293 
7294 /**
7295  * Release key of thread specific flow workspace data.
7296  */
7297 void
7298 flow_release_workspace(void *data)
7299 {
7300 	struct mlx5_flow_workspace *wks = data;
7301 	struct mlx5_flow_workspace *next;
7302 
7303 	while (wks) {
7304 		next = wks->next;
7305 		free(wks->rss_desc.queue);
7306 		free(wks);
7307 		wks = next;
7308 	}
7309 }
7310 
7311 /**
7312  * Get thread specific current flow workspace.
7313  *
7314  * @return pointer to thread specific flow workspace data, NULL on error.
7315  */
7316 struct mlx5_flow_workspace*
7317 mlx5_flow_get_thread_workspace(void)
7318 {
7319 	struct mlx5_flow_workspace *data;
7320 
7321 	data = mlx5_flow_os_get_specific_workspace();
7322 	MLX5_ASSERT(data && data->inuse);
7323 	if (!data || !data->inuse)
7324 		DRV_LOG(ERR, "flow workspace not initialized.");
7325 	return data;
7326 }
7327 
7328 /**
7329  * Allocate and init new flow workspace.
7330  *
7331  * @return pointer to flow workspace data, NULL on error.
7332  */
7333 static struct mlx5_flow_workspace*
7334 flow_alloc_thread_workspace(void)
7335 {
7336 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
7337 
7338 	if (!data) {
7339 		DRV_LOG(ERR, "Failed to allocate flow workspace "
7340 			"memory.");
7341 		return NULL;
7342 	}
7343 	data->rss_desc.queue = calloc(1,
7344 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
7345 	if (!data->rss_desc.queue)
7346 		goto err;
7347 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
7348 	return data;
7349 err:
7350 	free(data->rss_desc.queue);
7351 	free(data);
7352 	return NULL;
7353 }
7354 
7355 /**
7356  * Get new thread specific flow workspace.
7357  *
7358  * If current workspace inuse, create new one and set as current.
7359  *
7360  * @return pointer to thread specific flow workspace data, NULL on error.
7361  */
7362 static struct mlx5_flow_workspace*
7363 mlx5_flow_push_thread_workspace(void)
7364 {
7365 	struct mlx5_flow_workspace *curr;
7366 	struct mlx5_flow_workspace *data;
7367 
7368 	curr = mlx5_flow_os_get_specific_workspace();
7369 	if (!curr) {
7370 		data = flow_alloc_thread_workspace();
7371 		if (!data)
7372 			return NULL;
7373 	} else if (!curr->inuse) {
7374 		data = curr;
7375 	} else if (curr->next) {
7376 		data = curr->next;
7377 	} else {
7378 		data = flow_alloc_thread_workspace();
7379 		if (!data)
7380 			return NULL;
7381 		curr->next = data;
7382 		data->prev = curr;
7383 	}
7384 	data->inuse = 1;
7385 	data->flow_idx = 0;
7386 	/* Set as current workspace */
7387 	if (mlx5_flow_os_set_specific_workspace(data))
7388 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7389 	return data;
7390 }
7391 
7392 /**
7393  * Close current thread specific flow workspace.
7394  *
7395  * If previous workspace available, set it as current.
7396  *
7397  * @return pointer to thread specific flow workspace data, NULL on error.
7398  */
7399 static void
7400 mlx5_flow_pop_thread_workspace(void)
7401 {
7402 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7403 
7404 	if (!data)
7405 		return;
7406 	if (!data->inuse) {
7407 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7408 		return;
7409 	}
7410 	data->inuse = 0;
7411 	if (!data->prev)
7412 		return;
7413 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7414 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7415 }
7416 
7417 /**
7418  * Verify the flow list is empty
7419  *
7420  * @param dev
7421  *  Pointer to Ethernet device.
7422  *
7423  * @return the number of flows not released.
7424  */
7425 int
7426 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7427 {
7428 	struct mlx5_priv *priv = dev->data->dev_private;
7429 	struct rte_flow *flow;
7430 	uint32_t idx = 0;
7431 	int ret = 0, i;
7432 
7433 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7434 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7435 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7436 				dev->data->port_id, (void *)flow);
7437 			ret++;
7438 		}
7439 	}
7440 	return ret;
7441 }
7442 
7443 /**
7444  * Enable default hairpin egress flow.
7445  *
7446  * @param dev
7447  *   Pointer to Ethernet device.
7448  * @param queue
7449  *   The queue index.
7450  *
7451  * @return
7452  *   0 on success, a negative errno value otherwise and rte_errno is set.
7453  */
7454 int
7455 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7456 			    uint32_t queue)
7457 {
7458 	const struct rte_flow_attr attr = {
7459 		.egress = 1,
7460 		.priority = 0,
7461 	};
7462 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7463 		.queue = queue,
7464 	};
7465 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7466 		.queue = UINT32_MAX,
7467 	};
7468 	struct rte_flow_item items[] = {
7469 		{
7470 			.type = (enum rte_flow_item_type)
7471 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7472 			.spec = &queue_spec,
7473 			.last = NULL,
7474 			.mask = &queue_mask,
7475 		},
7476 		{
7477 			.type = RTE_FLOW_ITEM_TYPE_END,
7478 		},
7479 	};
7480 	struct rte_flow_action_jump jump = {
7481 		.group = MLX5_HAIRPIN_TX_TABLE,
7482 	};
7483 	struct rte_flow_action actions[2];
7484 	uint32_t flow_idx;
7485 	struct rte_flow_error error;
7486 
7487 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7488 	actions[0].conf = &jump;
7489 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7490 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7491 				    &attr, items, actions, false, &error);
7492 	if (!flow_idx) {
7493 		DRV_LOG(DEBUG,
7494 			"Failed to create ctrl flow: rte_errno(%d),"
7495 			" type(%d), message(%s)",
7496 			rte_errno, error.type,
7497 			error.message ? error.message : " (no stated reason)");
7498 		return -rte_errno;
7499 	}
7500 	return 0;
7501 }
7502 
7503 /**
7504  * Enable a control flow configured from the control plane.
7505  *
7506  * @param dev
7507  *   Pointer to Ethernet device.
7508  * @param eth_spec
7509  *   An Ethernet flow spec to apply.
7510  * @param eth_mask
7511  *   An Ethernet flow mask to apply.
7512  * @param vlan_spec
7513  *   A VLAN flow spec to apply.
7514  * @param vlan_mask
7515  *   A VLAN flow mask to apply.
7516  *
7517  * @return
7518  *   0 on success, a negative errno value otherwise and rte_errno is set.
7519  */
7520 int
7521 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7522 		    struct rte_flow_item_eth *eth_spec,
7523 		    struct rte_flow_item_eth *eth_mask,
7524 		    struct rte_flow_item_vlan *vlan_spec,
7525 		    struct rte_flow_item_vlan *vlan_mask)
7526 {
7527 	struct mlx5_priv *priv = dev->data->dev_private;
7528 	const struct rte_flow_attr attr = {
7529 		.ingress = 1,
7530 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7531 	};
7532 	struct rte_flow_item items[] = {
7533 		{
7534 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7535 			.spec = eth_spec,
7536 			.last = NULL,
7537 			.mask = eth_mask,
7538 		},
7539 		{
7540 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7541 					      RTE_FLOW_ITEM_TYPE_END,
7542 			.spec = vlan_spec,
7543 			.last = NULL,
7544 			.mask = vlan_mask,
7545 		},
7546 		{
7547 			.type = RTE_FLOW_ITEM_TYPE_END,
7548 		},
7549 	};
7550 	uint16_t queue[priv->reta_idx_n];
7551 	struct rte_flow_action_rss action_rss = {
7552 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7553 		.level = 0,
7554 		.types = priv->rss_conf.rss_hf,
7555 		.key_len = priv->rss_conf.rss_key_len,
7556 		.queue_num = priv->reta_idx_n,
7557 		.key = priv->rss_conf.rss_key,
7558 		.queue = queue,
7559 	};
7560 	struct rte_flow_action actions[] = {
7561 		{
7562 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7563 			.conf = &action_rss,
7564 		},
7565 		{
7566 			.type = RTE_FLOW_ACTION_TYPE_END,
7567 		},
7568 	};
7569 	uint32_t flow_idx;
7570 	struct rte_flow_error error;
7571 	unsigned int i;
7572 
7573 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7574 		return 0;
7575 	}
7576 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7577 		action_rss.types = 0;
7578 	for (i = 0; i != priv->reta_idx_n; ++i)
7579 		queue[i] = (*priv->reta_idx)[i];
7580 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7581 				    &attr, items, actions, false, &error);
7582 	if (!flow_idx)
7583 		return -rte_errno;
7584 	return 0;
7585 }
7586 
7587 /**
7588  * Enable a flow control configured from the control plane.
7589  *
7590  * @param dev
7591  *   Pointer to Ethernet device.
7592  * @param eth_spec
7593  *   An Ethernet flow spec to apply.
7594  * @param eth_mask
7595  *   An Ethernet flow mask to apply.
7596  *
7597  * @return
7598  *   0 on success, a negative errno value otherwise and rte_errno is set.
7599  */
7600 int
7601 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7602 	       struct rte_flow_item_eth *eth_spec,
7603 	       struct rte_flow_item_eth *eth_mask)
7604 {
7605 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7606 }
7607 
7608 /**
7609  * Create default miss flow rule matching lacp traffic
7610  *
7611  * @param dev
7612  *   Pointer to Ethernet device.
7613  * @param eth_spec
7614  *   An Ethernet flow spec to apply.
7615  *
7616  * @return
7617  *   0 on success, a negative errno value otherwise and rte_errno is set.
7618  */
7619 int
7620 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7621 {
7622 	/*
7623 	 * The LACP matching is done by only using ether type since using
7624 	 * a multicast dst mac causes kernel to give low priority to this flow.
7625 	 */
7626 	static const struct rte_flow_item_eth lacp_spec = {
7627 		.type = RTE_BE16(0x8809),
7628 	};
7629 	static const struct rte_flow_item_eth lacp_mask = {
7630 		.type = 0xffff,
7631 	};
7632 	const struct rte_flow_attr attr = {
7633 		.ingress = 1,
7634 	};
7635 	struct rte_flow_item items[] = {
7636 		{
7637 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7638 			.spec = &lacp_spec,
7639 			.mask = &lacp_mask,
7640 		},
7641 		{
7642 			.type = RTE_FLOW_ITEM_TYPE_END,
7643 		},
7644 	};
7645 	struct rte_flow_action actions[] = {
7646 		{
7647 			.type = (enum rte_flow_action_type)
7648 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7649 		},
7650 		{
7651 			.type = RTE_FLOW_ACTION_TYPE_END,
7652 		},
7653 	};
7654 	struct rte_flow_error error;
7655 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7656 					&attr, items, actions,
7657 					false, &error);
7658 
7659 	if (!flow_idx)
7660 		return -rte_errno;
7661 	return 0;
7662 }
7663 
7664 /**
7665  * Destroy a flow.
7666  *
7667  * @see rte_flow_destroy()
7668  * @see rte_flow_ops
7669  */
7670 int
7671 mlx5_flow_destroy(struct rte_eth_dev *dev,
7672 		  struct rte_flow *flow,
7673 		  struct rte_flow_error *error __rte_unused)
7674 {
7675 	struct mlx5_priv *priv = dev->data->dev_private;
7676 
7677 	if (priv->sh->config.dv_flow_en == 2)
7678 		return rte_flow_error_set(error, ENOTSUP,
7679 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7680 			  NULL,
7681 			  "Flow non-Q destruction not supported");
7682 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7683 				(uintptr_t)(void *)flow);
7684 	return 0;
7685 }
7686 
7687 /**
7688  * Destroy all flows.
7689  *
7690  * @see rte_flow_flush()
7691  * @see rte_flow_ops
7692  */
7693 int
7694 mlx5_flow_flush(struct rte_eth_dev *dev,
7695 		struct rte_flow_error *error __rte_unused)
7696 {
7697 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7698 	return 0;
7699 }
7700 
7701 /**
7702  * Isolated mode.
7703  *
7704  * @see rte_flow_isolate()
7705  * @see rte_flow_ops
7706  */
7707 int
7708 mlx5_flow_isolate(struct rte_eth_dev *dev,
7709 		  int enable,
7710 		  struct rte_flow_error *error)
7711 {
7712 	struct mlx5_priv *priv = dev->data->dev_private;
7713 
7714 	if (dev->data->dev_started) {
7715 		rte_flow_error_set(error, EBUSY,
7716 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7717 				   NULL,
7718 				   "port must be stopped first");
7719 		return -rte_errno;
7720 	}
7721 	priv->isolated = !!enable;
7722 	if (enable)
7723 		dev->dev_ops = &mlx5_dev_ops_isolate;
7724 	else
7725 		dev->dev_ops = &mlx5_dev_ops;
7726 
7727 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7728 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7729 
7730 	return 0;
7731 }
7732 
7733 /**
7734  * Query a flow.
7735  *
7736  * @see rte_flow_query()
7737  * @see rte_flow_ops
7738  */
7739 static int
7740 flow_drv_query(struct rte_eth_dev *dev,
7741 	       uint32_t flow_idx,
7742 	       const struct rte_flow_action *actions,
7743 	       void *data,
7744 	       struct rte_flow_error *error)
7745 {
7746 	struct mlx5_priv *priv = dev->data->dev_private;
7747 	const struct mlx5_flow_driver_ops *fops;
7748 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7749 					       flow_idx);
7750 	enum mlx5_flow_drv_type ftype;
7751 
7752 	if (!flow) {
7753 		return rte_flow_error_set(error, ENOENT,
7754 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7755 			  NULL,
7756 			  "invalid flow handle");
7757 	}
7758 	ftype = flow->drv_type;
7759 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7760 	fops = flow_get_drv_ops(ftype);
7761 
7762 	return fops->query(dev, flow, actions, data, error);
7763 }
7764 
7765 /**
7766  * Query a flow.
7767  *
7768  * @see rte_flow_query()
7769  * @see rte_flow_ops
7770  */
7771 int
7772 mlx5_flow_query(struct rte_eth_dev *dev,
7773 		struct rte_flow *flow,
7774 		const struct rte_flow_action *actions,
7775 		void *data,
7776 		struct rte_flow_error *error)
7777 {
7778 	int ret;
7779 	struct mlx5_priv *priv = dev->data->dev_private;
7780 
7781 	if (priv->sh->config.dv_flow_en == 2)
7782 		return rte_flow_error_set(error, ENOTSUP,
7783 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7784 			  NULL,
7785 			  "Flow non-Q query not supported");
7786 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7787 			     error);
7788 	if (ret < 0)
7789 		return ret;
7790 	return 0;
7791 }
7792 
7793 /**
7794  * Get rte_flow callbacks.
7795  *
7796  * @param dev
7797  *   Pointer to Ethernet device structure.
7798  * @param ops
7799  *   Pointer to operation-specific structure.
7800  *
7801  * @return 0
7802  */
7803 int
7804 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7805 		  const struct rte_flow_ops **ops)
7806 {
7807 	*ops = &mlx5_flow_ops;
7808 	return 0;
7809 }
7810 
7811 /**
7812  * Validate meter policy actions.
7813  * Dispatcher for action type specific validation.
7814  *
7815  * @param[in] dev
7816  *   Pointer to the Ethernet device structure.
7817  * @param[in] action
7818  *   The meter policy action object to validate.
7819  * @param[in] attr
7820  *   Attributes of flow to determine steering domain.
7821  * @param[out] is_rss
7822  *   Is RSS or not.
7823  * @param[out] domain_bitmap
7824  *   Domain bitmap.
7825  * @param[out] is_def_policy
7826  *   Is default policy or not.
7827  * @param[out] error
7828  *   Perform verbose error reporting if not NULL. Initialized in case of
7829  *   error only.
7830  *
7831  * @return
7832  *   0 on success, otherwise negative errno value.
7833  */
7834 int
7835 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7836 			const struct rte_flow_action *actions[RTE_COLORS],
7837 			struct rte_flow_attr *attr,
7838 			bool *is_rss,
7839 			uint8_t *domain_bitmap,
7840 			uint8_t *policy_mode,
7841 			struct rte_mtr_error *error)
7842 {
7843 	const struct mlx5_flow_driver_ops *fops;
7844 
7845 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7846 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7847 				       domain_bitmap, policy_mode, error);
7848 }
7849 
7850 /**
7851  * Destroy the meter table set.
7852  *
7853  * @param[in] dev
7854  *   Pointer to Ethernet device.
7855  * @param[in] mtr_policy
7856  *   Meter policy struct.
7857  */
7858 void
7859 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7860 		      struct mlx5_flow_meter_policy *mtr_policy)
7861 {
7862 	const struct mlx5_flow_driver_ops *fops;
7863 
7864 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7865 	fops->destroy_mtr_acts(dev, mtr_policy);
7866 }
7867 
7868 /**
7869  * Create policy action, lock free,
7870  * (mutex should be acquired by caller).
7871  * Dispatcher for action type specific call.
7872  *
7873  * @param[in] dev
7874  *   Pointer to the Ethernet device structure.
7875  * @param[in] mtr_policy
7876  *   Meter policy struct.
7877  * @param[in] action
7878  *   Action specification used to create meter actions.
7879  * @param[out] error
7880  *   Perform verbose error reporting if not NULL. Initialized in case of
7881  *   error only.
7882  *
7883  * @return
7884  *   0 on success, otherwise negative errno value.
7885  */
7886 int
7887 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7888 		      struct mlx5_flow_meter_policy *mtr_policy,
7889 		      const struct rte_flow_action *actions[RTE_COLORS],
7890 		      struct rte_mtr_error *error)
7891 {
7892 	const struct mlx5_flow_driver_ops *fops;
7893 
7894 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7895 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7896 }
7897 
7898 /**
7899  * Create policy rules, lock free,
7900  * (mutex should be acquired by caller).
7901  * Dispatcher for action type specific call.
7902  *
7903  * @param[in] dev
7904  *   Pointer to the Ethernet device structure.
7905  * @param[in] mtr_policy
7906  *   Meter policy struct.
7907  *
7908  * @return
7909  *   0 on success, -1 otherwise.
7910  */
7911 int
7912 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7913 			     struct mlx5_flow_meter_policy *mtr_policy)
7914 {
7915 	const struct mlx5_flow_driver_ops *fops;
7916 
7917 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7918 	return fops->create_policy_rules(dev, mtr_policy);
7919 }
7920 
7921 /**
7922  * Destroy policy rules, lock free,
7923  * (mutex should be acquired by caller).
7924  * Dispatcher for action type specific call.
7925  *
7926  * @param[in] dev
7927  *   Pointer to the Ethernet device structure.
7928  * @param[in] mtr_policy
7929  *   Meter policy struct.
7930  */
7931 void
7932 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7933 			     struct mlx5_flow_meter_policy *mtr_policy)
7934 {
7935 	const struct mlx5_flow_driver_ops *fops;
7936 
7937 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7938 	fops->destroy_policy_rules(dev, mtr_policy);
7939 }
7940 
7941 /**
7942  * Destroy the default policy table set.
7943  *
7944  * @param[in] dev
7945  *   Pointer to Ethernet device.
7946  */
7947 void
7948 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7949 {
7950 	const struct mlx5_flow_driver_ops *fops;
7951 
7952 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7953 	fops->destroy_def_policy(dev);
7954 }
7955 
7956 /**
7957  * Destroy the default policy table set.
7958  *
7959  * @param[in] dev
7960  *   Pointer to Ethernet device.
7961  *
7962  * @return
7963  *   0 on success, -1 otherwise.
7964  */
7965 int
7966 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7967 {
7968 	const struct mlx5_flow_driver_ops *fops;
7969 
7970 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7971 	return fops->create_def_policy(dev);
7972 }
7973 
7974 /**
7975  * Create the needed meter and suffix tables.
7976  *
7977  * @param[in] dev
7978  *   Pointer to Ethernet device.
7979  *
7980  * @return
7981  *   0 on success, -1 otherwise.
7982  */
7983 int
7984 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7985 			struct mlx5_flow_meter_info *fm,
7986 			uint32_t mtr_idx,
7987 			uint8_t domain_bitmap)
7988 {
7989 	const struct mlx5_flow_driver_ops *fops;
7990 
7991 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7992 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7993 }
7994 
7995 /**
7996  * Destroy the meter table set.
7997  *
7998  * @param[in] dev
7999  *   Pointer to Ethernet device.
8000  * @param[in] tbl
8001  *   Pointer to the meter table set.
8002  */
8003 void
8004 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
8005 			   struct mlx5_flow_meter_info *fm)
8006 {
8007 	const struct mlx5_flow_driver_ops *fops;
8008 
8009 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8010 	fops->destroy_mtr_tbls(dev, fm);
8011 }
8012 
8013 /**
8014  * Destroy the global meter drop table.
8015  *
8016  * @param[in] dev
8017  *   Pointer to Ethernet device.
8018  */
8019 void
8020 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
8021 {
8022 	const struct mlx5_flow_driver_ops *fops;
8023 
8024 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8025 	fops->destroy_mtr_drop_tbls(dev);
8026 }
8027 
8028 /**
8029  * Destroy the sub policy table with RX queue.
8030  *
8031  * @param[in] dev
8032  *   Pointer to Ethernet device.
8033  * @param[in] mtr_policy
8034  *   Pointer to meter policy table.
8035  */
8036 void
8037 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
8038 		struct mlx5_flow_meter_policy *mtr_policy)
8039 {
8040 	const struct mlx5_flow_driver_ops *fops;
8041 
8042 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8043 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
8044 }
8045 
8046 /**
8047  * Allocate the needed aso flow meter id.
8048  *
8049  * @param[in] dev
8050  *   Pointer to Ethernet device.
8051  *
8052  * @return
8053  *   Index to aso flow meter on success, NULL otherwise.
8054  */
8055 uint32_t
8056 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
8057 {
8058 	const struct mlx5_flow_driver_ops *fops;
8059 
8060 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8061 	return fops->create_meter(dev);
8062 }
8063 
8064 /**
8065  * Free the aso flow meter id.
8066  *
8067  * @param[in] dev
8068  *   Pointer to Ethernet device.
8069  * @param[in] mtr_idx
8070  *  Index to aso flow meter to be free.
8071  *
8072  * @return
8073  *   0 on success.
8074  */
8075 void
8076 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
8077 {
8078 	const struct mlx5_flow_driver_ops *fops;
8079 
8080 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8081 	fops->free_meter(dev, mtr_idx);
8082 }
8083 
8084 /**
8085  * Allocate a counter.
8086  *
8087  * @param[in] dev
8088  *   Pointer to Ethernet device structure.
8089  *
8090  * @return
8091  *   Index to allocated counter  on success, 0 otherwise.
8092  */
8093 uint32_t
8094 mlx5_counter_alloc(struct rte_eth_dev *dev)
8095 {
8096 	const struct mlx5_flow_driver_ops *fops;
8097 	struct rte_flow_attr attr = { .transfer = 0 };
8098 
8099 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8100 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8101 		return fops->counter_alloc(dev);
8102 	}
8103 	DRV_LOG(ERR,
8104 		"port %u counter allocate is not supported.",
8105 		 dev->data->port_id);
8106 	return 0;
8107 }
8108 
8109 /**
8110  * Free a counter.
8111  *
8112  * @param[in] dev
8113  *   Pointer to Ethernet device structure.
8114  * @param[in] cnt
8115  *   Index to counter to be free.
8116  */
8117 void
8118 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
8119 {
8120 	const struct mlx5_flow_driver_ops *fops;
8121 	struct rte_flow_attr attr = { .transfer = 0 };
8122 
8123 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8124 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8125 		fops->counter_free(dev, cnt);
8126 		return;
8127 	}
8128 	DRV_LOG(ERR,
8129 		"port %u counter free is not supported.",
8130 		 dev->data->port_id);
8131 }
8132 
8133 /**
8134  * Query counter statistics.
8135  *
8136  * @param[in] dev
8137  *   Pointer to Ethernet device structure.
8138  * @param[in] cnt
8139  *   Index to counter to query.
8140  * @param[in] clear
8141  *   Set to clear counter statistics.
8142  * @param[out] pkts
8143  *   The counter hits packets number to save.
8144  * @param[out] bytes
8145  *   The counter hits bytes number to save.
8146  *
8147  * @return
8148  *   0 on success, a negative errno value otherwise.
8149  */
8150 int
8151 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
8152 		   bool clear, uint64_t *pkts, uint64_t *bytes, void **action)
8153 {
8154 	const struct mlx5_flow_driver_ops *fops;
8155 	struct rte_flow_attr attr = { .transfer = 0 };
8156 
8157 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8158 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8159 		return fops->counter_query(dev, cnt, clear, pkts,
8160 					bytes, action);
8161 	}
8162 	DRV_LOG(ERR,
8163 		"port %u counter query is not supported.",
8164 		 dev->data->port_id);
8165 	return -ENOTSUP;
8166 }
8167 
8168 /**
8169  * Get information about HWS pre-configurable resources.
8170  *
8171  * @param[in] dev
8172  *   Pointer to the rte_eth_dev structure.
8173  * @param[out] port_info
8174  *   Pointer to port information.
8175  * @param[out] queue_info
8176  *   Pointer to queue information.
8177  * @param[out] error
8178  *   Pointer to error structure.
8179  *
8180  * @return
8181  *   0 on success, a negative errno value otherwise and rte_errno is set.
8182  */
8183 static int
8184 mlx5_flow_info_get(struct rte_eth_dev *dev,
8185 		   struct rte_flow_port_info *port_info,
8186 		   struct rte_flow_queue_info *queue_info,
8187 		   struct rte_flow_error *error)
8188 {
8189 	const struct mlx5_flow_driver_ops *fops;
8190 
8191 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8192 		return rte_flow_error_set(error, ENOTSUP,
8193 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8194 				NULL,
8195 				"info get with incorrect steering mode");
8196 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8197 	return fops->info_get(dev, port_info, queue_info, error);
8198 }
8199 
8200 /**
8201  * Configure port HWS resources.
8202  *
8203  * @param[in] dev
8204  *   Pointer to the rte_eth_dev structure.
8205  * @param[in] port_attr
8206  *   Port configuration attributes.
8207  * @param[in] nb_queue
8208  *   Number of queue.
8209  * @param[in] queue_attr
8210  *   Array that holds attributes for each flow queue.
8211  * @param[out] error
8212  *   Pointer to error structure.
8213  *
8214  * @return
8215  *   0 on success, a negative errno value otherwise and rte_errno is set.
8216  */
8217 static int
8218 mlx5_flow_port_configure(struct rte_eth_dev *dev,
8219 			 const struct rte_flow_port_attr *port_attr,
8220 			 uint16_t nb_queue,
8221 			 const struct rte_flow_queue_attr *queue_attr[],
8222 			 struct rte_flow_error *error)
8223 {
8224 	const struct mlx5_flow_driver_ops *fops;
8225 
8226 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8227 		return rte_flow_error_set(error, ENOTSUP,
8228 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8229 				NULL,
8230 				"port configure with incorrect steering mode");
8231 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8232 	return fops->configure(dev, port_attr, nb_queue, queue_attr, error);
8233 }
8234 
8235 /**
8236  * Create flow item template.
8237  *
8238  * @param[in] dev
8239  *   Pointer to the rte_eth_dev structure.
8240  * @param[in] attr
8241  *   Pointer to the item template attributes.
8242  * @param[in] items
8243  *   The template item pattern.
8244  * @param[out] error
8245  *   Pointer to error structure.
8246  *
8247  * @return
8248  *   0 on success, a negative errno value otherwise and rte_errno is set.
8249  */
8250 static struct rte_flow_pattern_template *
8251 mlx5_flow_pattern_template_create(struct rte_eth_dev *dev,
8252 		const struct rte_flow_pattern_template_attr *attr,
8253 		const struct rte_flow_item items[],
8254 		struct rte_flow_error *error)
8255 {
8256 	const struct mlx5_flow_driver_ops *fops;
8257 
8258 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW) {
8259 		rte_flow_error_set(error, ENOTSUP,
8260 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8261 				NULL,
8262 				"pattern create with incorrect steering mode");
8263 		return NULL;
8264 	}
8265 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8266 	return fops->pattern_template_create(dev, attr, items, error);
8267 }
8268 
8269 /**
8270  * Destroy flow item template.
8271  *
8272  * @param[in] dev
8273  *   Pointer to the rte_eth_dev structure.
8274  * @param[in] template
8275  *   Pointer to the item template to be destroyed.
8276  * @param[out] error
8277  *   Pointer to error structure.
8278  *
8279  * @return
8280  *   0 on success, a negative errno value otherwise and rte_errno is set.
8281  */
8282 static int
8283 mlx5_flow_pattern_template_destroy(struct rte_eth_dev *dev,
8284 				   struct rte_flow_pattern_template *template,
8285 				   struct rte_flow_error *error)
8286 {
8287 	const struct mlx5_flow_driver_ops *fops;
8288 
8289 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8290 		return rte_flow_error_set(error, ENOTSUP,
8291 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8292 				NULL,
8293 				"pattern destroy with incorrect steering mode");
8294 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8295 	return fops->pattern_template_destroy(dev, template, error);
8296 }
8297 
8298 /**
8299  * Create flow item template.
8300  *
8301  * @param[in] dev
8302  *   Pointer to the rte_eth_dev structure.
8303  * @param[in] attr
8304  *   Pointer to the action template attributes.
8305  * @param[in] actions
8306  *   Associated actions (list terminated by the END action).
8307  * @param[in] masks
8308  *   List of actions that marks which of the action's member is constant.
8309  * @param[out] error
8310  *   Pointer to error structure.
8311  *
8312  * @return
8313  *   0 on success, a negative errno value otherwise and rte_errno is set.
8314  */
8315 static struct rte_flow_actions_template *
8316 mlx5_flow_actions_template_create(struct rte_eth_dev *dev,
8317 			const struct rte_flow_actions_template_attr *attr,
8318 			const struct rte_flow_action actions[],
8319 			const struct rte_flow_action masks[],
8320 			struct rte_flow_error *error)
8321 {
8322 	const struct mlx5_flow_driver_ops *fops;
8323 
8324 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW) {
8325 		rte_flow_error_set(error, ENOTSUP,
8326 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8327 				NULL,
8328 				"action create with incorrect steering mode");
8329 		return NULL;
8330 	}
8331 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8332 	return fops->actions_template_create(dev, attr, actions, masks, error);
8333 }
8334 
8335 /**
8336  * Destroy flow action template.
8337  *
8338  * @param[in] dev
8339  *   Pointer to the rte_eth_dev structure.
8340  * @param[in] template
8341  *   Pointer to the action template to be destroyed.
8342  * @param[out] error
8343  *   Pointer to error structure.
8344  *
8345  * @return
8346  *   0 on success, a negative errno value otherwise and rte_errno is set.
8347  */
8348 static int
8349 mlx5_flow_actions_template_destroy(struct rte_eth_dev *dev,
8350 				   struct rte_flow_actions_template *template,
8351 				   struct rte_flow_error *error)
8352 {
8353 	const struct mlx5_flow_driver_ops *fops;
8354 
8355 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8356 		return rte_flow_error_set(error, ENOTSUP,
8357 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8358 				NULL,
8359 				"action destroy with incorrect steering mode");
8360 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8361 	return fops->actions_template_destroy(dev, template, error);
8362 }
8363 
8364 /**
8365  * Create flow table.
8366  *
8367  * @param[in] dev
8368  *   Pointer to the rte_eth_dev structure.
8369  * @param[in] attr
8370  *   Pointer to the table attributes.
8371  * @param[in] item_templates
8372  *   Item template array to be binded to the table.
8373  * @param[in] nb_item_templates
8374  *   Number of item template.
8375  * @param[in] action_templates
8376  *   Action template array to be binded to the table.
8377  * @param[in] nb_action_templates
8378  *   Number of action template.
8379  * @param[out] error
8380  *   Pointer to error structure.
8381  *
8382  * @return
8383  *    Table on success, NULL otherwise and rte_errno is set.
8384  */
8385 static struct rte_flow_template_table *
8386 mlx5_flow_table_create(struct rte_eth_dev *dev,
8387 		       const struct rte_flow_template_table_attr *attr,
8388 		       struct rte_flow_pattern_template *item_templates[],
8389 		       uint8_t nb_item_templates,
8390 		       struct rte_flow_actions_template *action_templates[],
8391 		       uint8_t nb_action_templates,
8392 		       struct rte_flow_error *error)
8393 {
8394 	const struct mlx5_flow_driver_ops *fops;
8395 
8396 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW) {
8397 		rte_flow_error_set(error, ENOTSUP,
8398 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8399 				NULL,
8400 				"table create with incorrect steering mode");
8401 		return NULL;
8402 	}
8403 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8404 	return fops->template_table_create(dev,
8405 					   attr,
8406 					   item_templates,
8407 					   nb_item_templates,
8408 					   action_templates,
8409 					   nb_action_templates,
8410 					   error);
8411 }
8412 
8413 /**
8414  * PMD destroy flow table.
8415  *
8416  * @param[in] dev
8417  *   Pointer to the rte_eth_dev structure.
8418  * @param[in] table
8419  *   Pointer to the table to be destroyed.
8420  * @param[out] error
8421  *   Pointer to error structure.
8422  *
8423  * @return
8424  *   0 on success, a negative errno value otherwise and rte_errno is set.
8425  */
8426 static int
8427 mlx5_flow_table_destroy(struct rte_eth_dev *dev,
8428 			struct rte_flow_template_table *table,
8429 			struct rte_flow_error *error)
8430 {
8431 	const struct mlx5_flow_driver_ops *fops;
8432 
8433 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8434 		return rte_flow_error_set(error, ENOTSUP,
8435 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8436 				NULL,
8437 				"table destroy with incorrect steering mode");
8438 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8439 	return fops->template_table_destroy(dev, table, error);
8440 }
8441 
8442 /**
8443  * Enqueue flow creation.
8444  *
8445  * @param[in] dev
8446  *   Pointer to the rte_eth_dev structure.
8447  * @param[in] queue_id
8448  *   The queue to create the flow.
8449  * @param[in] attr
8450  *   Pointer to the flow operation attributes.
8451  * @param[in] items
8452  *   Items with flow spec value.
8453  * @param[in] pattern_template_index
8454  *   The item pattern flow follows from the table.
8455  * @param[in] actions
8456  *   Action with flow spec value.
8457  * @param[in] action_template_index
8458  *   The action pattern flow follows from the table.
8459  * @param[in] user_data
8460  *   Pointer to the user_data.
8461  * @param[out] error
8462  *   Pointer to error structure.
8463  *
8464  * @return
8465  *    Flow pointer on success, NULL otherwise and rte_errno is set.
8466  */
8467 static struct rte_flow *
8468 mlx5_flow_async_flow_create(struct rte_eth_dev *dev,
8469 			    uint32_t queue_id,
8470 			    const struct rte_flow_op_attr *attr,
8471 			    struct rte_flow_template_table *table,
8472 			    const struct rte_flow_item items[],
8473 			    uint8_t pattern_template_index,
8474 			    const struct rte_flow_action actions[],
8475 			    uint8_t action_template_index,
8476 			    void *user_data,
8477 			    struct rte_flow_error *error)
8478 {
8479 	const struct mlx5_flow_driver_ops *fops;
8480 
8481 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW) {
8482 		rte_flow_error_set(error, ENOTSUP,
8483 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8484 				NULL,
8485 				"flow_q create with incorrect steering mode");
8486 		return NULL;
8487 	}
8488 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8489 	return fops->async_flow_create(dev, queue_id, attr, table,
8490 				       items, pattern_template_index,
8491 				       actions, action_template_index,
8492 				       user_data, error);
8493 }
8494 
8495 /**
8496  * Enqueue flow destruction.
8497  *
8498  * @param[in] dev
8499  *   Pointer to the rte_eth_dev structure.
8500  * @param[in] queue
8501  *   The queue to destroy the flow.
8502  * @param[in] attr
8503  *   Pointer to the flow operation attributes.
8504  * @param[in] flow
8505  *   Pointer to the flow to be destroyed.
8506  * @param[in] user_data
8507  *   Pointer to the user_data.
8508  * @param[out] error
8509  *   Pointer to error structure.
8510  *
8511  * @return
8512  *    0 on success, negative value otherwise and rte_errno is set.
8513  */
8514 static int
8515 mlx5_flow_async_flow_destroy(struct rte_eth_dev *dev,
8516 			     uint32_t queue,
8517 			     const struct rte_flow_op_attr *attr,
8518 			     struct rte_flow *flow,
8519 			     void *user_data,
8520 			     struct rte_flow_error *error)
8521 {
8522 	const struct mlx5_flow_driver_ops *fops;
8523 
8524 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8525 		return rte_flow_error_set(error, ENOTSUP,
8526 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8527 				NULL,
8528 				"flow_q destroy with incorrect steering mode");
8529 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8530 	return fops->async_flow_destroy(dev, queue, attr, flow,
8531 					user_data, error);
8532 }
8533 
8534 /**
8535  * Pull the enqueued flows.
8536  *
8537  * @param[in] dev
8538  *   Pointer to the rte_eth_dev structure.
8539  * @param[in] queue
8540  *   The queue to pull the result.
8541  * @param[in/out] res
8542  *   Array to save the results.
8543  * @param[in] n_res
8544  *   Available result with the array.
8545  * @param[out] error
8546  *   Pointer to error structure.
8547  *
8548  * @return
8549  *    Result number on success, negative value otherwise and rte_errno is set.
8550  */
8551 static int
8552 mlx5_flow_pull(struct rte_eth_dev *dev,
8553 	       uint32_t queue,
8554 	       struct rte_flow_op_result res[],
8555 	       uint16_t n_res,
8556 	       struct rte_flow_error *error)
8557 {
8558 	const struct mlx5_flow_driver_ops *fops;
8559 
8560 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8561 		return rte_flow_error_set(error, ENOTSUP,
8562 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8563 				NULL,
8564 				"flow_q pull with incorrect steering mode");
8565 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8566 	return fops->pull(dev, queue, res, n_res, error);
8567 }
8568 
8569 /**
8570  * Push the enqueued flows.
8571  *
8572  * @param[in] dev
8573  *   Pointer to the rte_eth_dev structure.
8574  * @param[in] queue
8575  *   The queue to push the flows.
8576  * @param[out] error
8577  *   Pointer to error structure.
8578  *
8579  * @return
8580  *    0 on success, negative value otherwise and rte_errno is set.
8581  */
8582 static int
8583 mlx5_flow_push(struct rte_eth_dev *dev,
8584 	       uint32_t queue,
8585 	       struct rte_flow_error *error)
8586 {
8587 	const struct mlx5_flow_driver_ops *fops;
8588 
8589 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
8590 		return rte_flow_error_set(error, ENOTSUP,
8591 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8592 				NULL,
8593 				"flow_q push with incorrect steering mode");
8594 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8595 	return fops->push(dev, queue, error);
8596 }
8597 
8598 /**
8599  * Create shared action.
8600  *
8601  * @param[in] dev
8602  *   Pointer to the rte_eth_dev structure.
8603  * @param[in] queue
8604  *   Which queue to be used..
8605  * @param[in] attr
8606  *   Operation attribute.
8607  * @param[in] conf
8608  *   Indirect action configuration.
8609  * @param[in] action
8610  *   rte_flow action detail.
8611  * @param[in] user_data
8612  *   Pointer to the user_data.
8613  * @param[out] error
8614  *   Pointer to error structure.
8615  *
8616  * @return
8617  *   Action handle on success, NULL otherwise and rte_errno is set.
8618  */
8619 static struct rte_flow_action_handle *
8620 mlx5_flow_async_action_handle_create(struct rte_eth_dev *dev, uint32_t queue,
8621 				 const struct rte_flow_op_attr *attr,
8622 				 const struct rte_flow_indir_action_conf *conf,
8623 				 const struct rte_flow_action *action,
8624 				 void *user_data,
8625 				 struct rte_flow_error *error)
8626 {
8627 	const struct mlx5_flow_driver_ops *fops =
8628 			flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8629 
8630 	return fops->async_action_create(dev, queue, attr, conf, action,
8631 					 user_data, error);
8632 }
8633 
8634 /**
8635  * Update shared action.
8636  *
8637  * @param[in] dev
8638  *   Pointer to the rte_eth_dev structure.
8639  * @param[in] queue
8640  *   Which queue to be used..
8641  * @param[in] attr
8642  *   Operation attribute.
8643  * @param[in] handle
8644  *   Action handle to be updated.
8645  * @param[in] update
8646  *   Update value.
8647  * @param[in] user_data
8648  *   Pointer to the user_data.
8649  * @param[out] error
8650  *   Pointer to error structure.
8651  *
8652  * @return
8653  *   0 on success, negative value otherwise and rte_errno is set.
8654  */
8655 static int
8656 mlx5_flow_async_action_handle_update(struct rte_eth_dev *dev, uint32_t queue,
8657 				     const struct rte_flow_op_attr *attr,
8658 				     struct rte_flow_action_handle *handle,
8659 				     const void *update,
8660 				     void *user_data,
8661 				     struct rte_flow_error *error)
8662 {
8663 	const struct mlx5_flow_driver_ops *fops =
8664 			flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8665 
8666 	return fops->async_action_update(dev, queue, attr, handle,
8667 					 update, user_data, error);
8668 }
8669 
8670 /**
8671  * Destroy shared action.
8672  *
8673  * @param[in] dev
8674  *   Pointer to the rte_eth_dev structure.
8675  * @param[in] queue
8676  *   Which queue to be used..
8677  * @param[in] attr
8678  *   Operation attribute.
8679  * @param[in] handle
8680  *   Action handle to be destroyed.
8681  * @param[in] user_data
8682  *   Pointer to the user_data.
8683  * @param[out] error
8684  *   Pointer to error structure.
8685  *
8686  * @return
8687  *   0 on success, negative value otherwise and rte_errno is set.
8688  */
8689 static int
8690 mlx5_flow_async_action_handle_destroy(struct rte_eth_dev *dev, uint32_t queue,
8691 				      const struct rte_flow_op_attr *attr,
8692 				      struct rte_flow_action_handle *handle,
8693 				      void *user_data,
8694 				      struct rte_flow_error *error)
8695 {
8696 	const struct mlx5_flow_driver_ops *fops =
8697 			flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
8698 
8699 	return fops->async_action_destroy(dev, queue, attr, handle,
8700 					  user_data, error);
8701 }
8702 
8703 /**
8704  * Allocate a new memory for the counter values wrapped by all the needed
8705  * management.
8706  *
8707  * @param[in] sh
8708  *   Pointer to mlx5_dev_ctx_shared object.
8709  *
8710  * @return
8711  *   0 on success, a negative errno value otherwise.
8712  */
8713 static int
8714 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
8715 {
8716 	struct mlx5_counter_stats_mem_mng *mem_mng;
8717 	volatile struct flow_counter_stats *raw_data;
8718 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
8719 	int size = (sizeof(struct flow_counter_stats) *
8720 			MLX5_COUNTERS_PER_POOL +
8721 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
8722 			sizeof(struct mlx5_counter_stats_mem_mng);
8723 	size_t pgsize = rte_mem_page_size();
8724 	uint8_t *mem;
8725 	int ret;
8726 	int i;
8727 
8728 	if (pgsize == (size_t)-1) {
8729 		DRV_LOG(ERR, "Failed to get mem page size");
8730 		rte_errno = ENOMEM;
8731 		return -ENOMEM;
8732 	}
8733 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
8734 	if (!mem) {
8735 		rte_errno = ENOMEM;
8736 		return -ENOMEM;
8737 	}
8738 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
8739 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
8740 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
8741 					  sh->cdev->pdn, mem, size,
8742 					  &mem_mng->wm);
8743 	if (ret) {
8744 		rte_errno = errno;
8745 		mlx5_free(mem);
8746 		return -rte_errno;
8747 	}
8748 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
8749 	raw_data = (volatile struct flow_counter_stats *)mem;
8750 	for (i = 0; i < raws_n; ++i) {
8751 		mem_mng->raws[i].mem_mng = mem_mng;
8752 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
8753 	}
8754 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
8755 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
8756 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
8757 				 next);
8758 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
8759 	sh->cmng.mem_mng = mem_mng;
8760 	return 0;
8761 }
8762 
8763 /**
8764  * Set the statistic memory to the new counter pool.
8765  *
8766  * @param[in] sh
8767  *   Pointer to mlx5_dev_ctx_shared object.
8768  * @param[in] pool
8769  *   Pointer to the pool to set the statistic memory.
8770  *
8771  * @return
8772  *   0 on success, a negative errno value otherwise.
8773  */
8774 static int
8775 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
8776 			       struct mlx5_flow_counter_pool *pool)
8777 {
8778 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8779 	/* Resize statistic memory once used out. */
8780 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
8781 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
8782 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
8783 		return -1;
8784 	}
8785 	rte_spinlock_lock(&pool->sl);
8786 	pool->raw = cmng->mem_mng->raws + pool->index %
8787 		    MLX5_CNT_CONTAINER_RESIZE;
8788 	rte_spinlock_unlock(&pool->sl);
8789 	pool->raw_hw = NULL;
8790 	return 0;
8791 }
8792 
8793 #define MLX5_POOL_QUERY_FREQ_US 1000000
8794 
8795 /**
8796  * Set the periodic procedure for triggering asynchronous batch queries for all
8797  * the counter pools.
8798  *
8799  * @param[in] sh
8800  *   Pointer to mlx5_dev_ctx_shared object.
8801  */
8802 void
8803 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
8804 {
8805 	uint32_t pools_n, us;
8806 
8807 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
8808 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
8809 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
8810 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
8811 		sh->cmng.query_thread_on = 0;
8812 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
8813 	} else {
8814 		sh->cmng.query_thread_on = 1;
8815 	}
8816 }
8817 
8818 /**
8819  * The periodic procedure for triggering asynchronous batch queries for all the
8820  * counter pools. This function is probably called by the host thread.
8821  *
8822  * @param[in] arg
8823  *   The parameter for the alarm process.
8824  */
8825 void
8826 mlx5_flow_query_alarm(void *arg)
8827 {
8828 	struct mlx5_dev_ctx_shared *sh = arg;
8829 	int ret;
8830 	uint16_t pool_index = sh->cmng.pool_index;
8831 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8832 	struct mlx5_flow_counter_pool *pool;
8833 	uint16_t n_valid;
8834 
8835 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
8836 		goto set_alarm;
8837 	rte_spinlock_lock(&cmng->pool_update_sl);
8838 	pool = cmng->pools[pool_index];
8839 	n_valid = cmng->n_valid;
8840 	rte_spinlock_unlock(&cmng->pool_update_sl);
8841 	/* Set the statistic memory to the new created pool. */
8842 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
8843 		goto set_alarm;
8844 	if (pool->raw_hw)
8845 		/* There is a pool query in progress. */
8846 		goto set_alarm;
8847 	pool->raw_hw =
8848 		LIST_FIRST(&sh->cmng.free_stat_raws);
8849 	if (!pool->raw_hw)
8850 		/* No free counter statistics raw memory. */
8851 		goto set_alarm;
8852 	/*
8853 	 * Identify the counters released between query trigger and query
8854 	 * handle more efficiently. The counter released in this gap period
8855 	 * should wait for a new round of query as the new arrived packets
8856 	 * will not be taken into account.
8857 	 */
8858 	pool->query_gen++;
8859 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
8860 					       MLX5_COUNTERS_PER_POOL,
8861 					       NULL, NULL,
8862 					       pool->raw_hw->mem_mng->wm.lkey,
8863 					       (void *)(uintptr_t)
8864 					       pool->raw_hw->data,
8865 					       sh->devx_comp,
8866 					       (uint64_t)(uintptr_t)pool);
8867 	if (ret) {
8868 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
8869 			" %d", pool->min_dcs->id);
8870 		pool->raw_hw = NULL;
8871 		goto set_alarm;
8872 	}
8873 	LIST_REMOVE(pool->raw_hw, next);
8874 	sh->cmng.pending_queries++;
8875 	pool_index++;
8876 	if (pool_index >= n_valid)
8877 		pool_index = 0;
8878 set_alarm:
8879 	sh->cmng.pool_index = pool_index;
8880 	mlx5_set_query_alarm(sh);
8881 }
8882 
8883 /**
8884  * Check and callback event for new aged flow in the counter pool
8885  *
8886  * @param[in] sh
8887  *   Pointer to mlx5_dev_ctx_shared object.
8888  * @param[in] pool
8889  *   Pointer to Current counter pool.
8890  */
8891 static void
8892 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
8893 		   struct mlx5_flow_counter_pool *pool)
8894 {
8895 	struct mlx5_priv *priv;
8896 	struct mlx5_flow_counter *cnt;
8897 	struct mlx5_age_info *age_info;
8898 	struct mlx5_age_param *age_param;
8899 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
8900 	struct mlx5_counter_stats_raw *prev = pool->raw;
8901 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
8902 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
8903 	uint16_t expected = AGE_CANDIDATE;
8904 	uint32_t i;
8905 
8906 	pool->time_of_last_age_check = curr_time;
8907 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8908 		cnt = MLX5_POOL_GET_CNT(pool, i);
8909 		age_param = MLX5_CNT_TO_AGE(cnt);
8910 		if (__atomic_load_n(&age_param->state,
8911 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
8912 			continue;
8913 		if (cur->data[i].hits != prev->data[i].hits) {
8914 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
8915 					 __ATOMIC_RELAXED);
8916 			continue;
8917 		}
8918 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8919 				       time_delta,
8920 				       __ATOMIC_RELAXED) <= age_param->timeout)
8921 			continue;
8922 		/**
8923 		 * Hold the lock first, or if between the
8924 		 * state AGE_TMOUT and tailq operation the
8925 		 * release happened, the release procedure
8926 		 * may delete a non-existent tailq node.
8927 		 */
8928 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8929 		age_info = GET_PORT_AGE_INFO(priv);
8930 		rte_spinlock_lock(&age_info->aged_sl);
8931 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8932 						AGE_TMOUT, false,
8933 						__ATOMIC_RELAXED,
8934 						__ATOMIC_RELAXED)) {
8935 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8936 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8937 		}
8938 		rte_spinlock_unlock(&age_info->aged_sl);
8939 	}
8940 	mlx5_age_event_prepare(sh);
8941 }
8942 
8943 /**
8944  * Handler for the HW respond about ready values from an asynchronous batch
8945  * query. This function is probably called by the host thread.
8946  *
8947  * @param[in] sh
8948  *   The pointer to the shared device context.
8949  * @param[in] async_id
8950  *   The Devx async ID.
8951  * @param[in] status
8952  *   The status of the completion.
8953  */
8954 void
8955 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8956 				  uint64_t async_id, int status)
8957 {
8958 	struct mlx5_flow_counter_pool *pool =
8959 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8960 	struct mlx5_counter_stats_raw *raw_to_free;
8961 	uint8_t query_gen = pool->query_gen ^ 1;
8962 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8963 	enum mlx5_counter_type cnt_type =
8964 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8965 				MLX5_COUNTER_TYPE_ORIGIN;
8966 
8967 	if (unlikely(status)) {
8968 		raw_to_free = pool->raw_hw;
8969 	} else {
8970 		raw_to_free = pool->raw;
8971 		if (pool->is_aged)
8972 			mlx5_flow_aging_check(sh, pool);
8973 		rte_spinlock_lock(&pool->sl);
8974 		pool->raw = pool->raw_hw;
8975 		rte_spinlock_unlock(&pool->sl);
8976 		/* Be sure the new raw counters data is updated in memory. */
8977 		rte_io_wmb();
8978 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8979 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8980 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8981 				     &pool->counters[query_gen], next);
8982 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8983 		}
8984 	}
8985 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8986 	pool->raw_hw = NULL;
8987 	sh->cmng.pending_queries--;
8988 }
8989 
8990 static int
8991 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8992 		    const struct flow_grp_info *grp_info,
8993 		    struct rte_flow_error *error)
8994 {
8995 	if (grp_info->transfer && grp_info->external &&
8996 	    grp_info->fdb_def_rule) {
8997 		if (group == UINT32_MAX)
8998 			return rte_flow_error_set
8999 						(error, EINVAL,
9000 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9001 						 NULL,
9002 						 "group index not supported");
9003 		*table = group + 1;
9004 	} else {
9005 		*table = group;
9006 	}
9007 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
9008 	return 0;
9009 }
9010 
9011 /**
9012  * Translate the rte_flow group index to HW table value.
9013  *
9014  * If tunnel offload is disabled, all group ids converted to flow table
9015  * id using the standard method.
9016  * If tunnel offload is enabled, group id can be converted using the
9017  * standard or tunnel conversion method. Group conversion method
9018  * selection depends on flags in `grp_info` parameter:
9019  * - Internal (grp_info.external == 0) groups conversion uses the
9020  *   standard method.
9021  * - Group ids in JUMP action converted with the tunnel conversion.
9022  * - Group id in rule attribute conversion depends on a rule type and
9023  *   group id value:
9024  *   ** non zero group attributes converted with the tunnel method
9025  *   ** zero group attribute in non-tunnel rule is converted using the
9026  *      standard method - there's only one root table
9027  *   ** zero group attribute in steer tunnel rule is converted with the
9028  *      standard method - single root table
9029  *   ** zero group attribute in match tunnel rule is a special OvS
9030  *      case: that value is used for portability reasons. That group
9031  *      id is converted with the tunnel conversion method.
9032  *
9033  * @param[in] dev
9034  *   Port device
9035  * @param[in] tunnel
9036  *   PMD tunnel offload object
9037  * @param[in] group
9038  *   rte_flow group index value.
9039  * @param[out] table
9040  *   HW table value.
9041  * @param[in] grp_info
9042  *   flags used for conversion
9043  * @param[out] error
9044  *   Pointer to error structure.
9045  *
9046  * @return
9047  *   0 on success, a negative errno value otherwise and rte_errno is set.
9048  */
9049 int
9050 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
9051 			 const struct mlx5_flow_tunnel *tunnel,
9052 			 uint32_t group, uint32_t *table,
9053 			 const struct flow_grp_info *grp_info,
9054 			 struct rte_flow_error *error)
9055 {
9056 	int ret;
9057 	bool standard_translation;
9058 
9059 	if (!grp_info->skip_scale && grp_info->external &&
9060 	    group < MLX5_MAX_TABLES_EXTERNAL)
9061 		group *= MLX5_FLOW_TABLE_FACTOR;
9062 	if (is_tunnel_offload_active(dev)) {
9063 		standard_translation = !grp_info->external ||
9064 					grp_info->std_tbl_fix;
9065 	} else {
9066 		standard_translation = true;
9067 	}
9068 	DRV_LOG(DEBUG,
9069 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
9070 		dev->data->port_id, group, grp_info->transfer,
9071 		grp_info->external, grp_info->fdb_def_rule,
9072 		standard_translation ? "STANDARD" : "TUNNEL");
9073 	if (standard_translation)
9074 		ret = flow_group_to_table(dev->data->port_id, group, table,
9075 					  grp_info, error);
9076 	else
9077 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
9078 						      table, error);
9079 
9080 	return ret;
9081 }
9082 
9083 /**
9084  * Discover availability of metadata reg_c's.
9085  *
9086  * Iteratively use test flows to check availability.
9087  *
9088  * @param[in] dev
9089  *   Pointer to the Ethernet device structure.
9090  *
9091  * @return
9092  *   0 on success, a negative errno value otherwise and rte_errno is set.
9093  */
9094 int
9095 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
9096 {
9097 	struct mlx5_priv *priv = dev->data->dev_private;
9098 	enum modify_reg idx;
9099 	int n = 0;
9100 
9101 	/* reg_c[0] and reg_c[1] are reserved. */
9102 	priv->sh->flow_mreg_c[n++] = REG_C_0;
9103 	priv->sh->flow_mreg_c[n++] = REG_C_1;
9104 	/* Discover availability of other reg_c's. */
9105 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
9106 		struct rte_flow_attr attr = {
9107 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
9108 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
9109 			.ingress = 1,
9110 		};
9111 		struct rte_flow_item items[] = {
9112 			[0] = {
9113 				.type = RTE_FLOW_ITEM_TYPE_END,
9114 			},
9115 		};
9116 		struct rte_flow_action actions[] = {
9117 			[0] = {
9118 				.type = (enum rte_flow_action_type)
9119 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
9120 				.conf = &(struct mlx5_flow_action_copy_mreg){
9121 					.src = REG_C_1,
9122 					.dst = idx,
9123 				},
9124 			},
9125 			[1] = {
9126 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
9127 				.conf = &(struct rte_flow_action_jump){
9128 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
9129 				},
9130 			},
9131 			[2] = {
9132 				.type = RTE_FLOW_ACTION_TYPE_END,
9133 			},
9134 		};
9135 		uint32_t flow_idx;
9136 		struct rte_flow *flow;
9137 		struct rte_flow_error error;
9138 
9139 		if (!priv->sh->config.dv_flow_en)
9140 			break;
9141 		/* Create internal flow, validation skips copy action. */
9142 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
9143 					items, actions, false, &error);
9144 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
9145 				      flow_idx);
9146 		if (!flow)
9147 			continue;
9148 		priv->sh->flow_mreg_c[n++] = idx;
9149 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
9150 	}
9151 	for (; n < MLX5_MREG_C_NUM; ++n)
9152 		priv->sh->flow_mreg_c[n] = REG_NON;
9153 	priv->sh->metadata_regc_check_flag = 1;
9154 	return 0;
9155 }
9156 
9157 int
9158 save_dump_file(const uint8_t *data, uint32_t size,
9159 	uint32_t type, uint64_t id, void *arg, FILE *file)
9160 {
9161 	char line[BUF_SIZE];
9162 	uint32_t out = 0;
9163 	uint32_t k;
9164 	uint32_t actions_num;
9165 	struct rte_flow_query_count *count;
9166 
9167 	memset(line, 0, BUF_SIZE);
9168 	switch (type) {
9169 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
9170 		actions_num = *(uint32_t *)(arg);
9171 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
9172 				type, id, actions_num);
9173 		break;
9174 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
9175 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
9176 				type, id);
9177 		break;
9178 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
9179 		count = (struct rte_flow_query_count *)arg;
9180 		fprintf(file,
9181 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
9182 			type, id, count->hits, count->bytes);
9183 		return 0;
9184 	default:
9185 		return -1;
9186 	}
9187 
9188 	for (k = 0; k < size; k++) {
9189 		/* Make sure we do not overrun the line buffer length. */
9190 		if (out >= BUF_SIZE - 4) {
9191 			line[out] = '\0';
9192 			break;
9193 		}
9194 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
9195 				(data[k]) & 0xff);
9196 	}
9197 	fprintf(file, "%s\n", line);
9198 	return 0;
9199 }
9200 
9201 int
9202 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
9203 	struct rte_flow_query_count *count, struct rte_flow_error *error)
9204 {
9205 	struct rte_flow_action action[2];
9206 	enum mlx5_flow_drv_type ftype;
9207 	const struct mlx5_flow_driver_ops *fops;
9208 
9209 	if (!flow) {
9210 		return rte_flow_error_set(error, ENOENT,
9211 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9212 				NULL,
9213 				"invalid flow handle");
9214 	}
9215 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
9216 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
9217 	if (flow->counter) {
9218 		memset(count, 0, sizeof(struct rte_flow_query_count));
9219 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
9220 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
9221 						ftype < MLX5_FLOW_TYPE_MAX);
9222 		fops = flow_get_drv_ops(ftype);
9223 		return fops->query(dev, flow, action, count, error);
9224 	}
9225 	return -1;
9226 }
9227 
9228 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9229 /**
9230  * Dump flow ipool data to file
9231  *
9232  * @param[in] dev
9233  *   The pointer to Ethernet device.
9234  * @param[in] file
9235  *   A pointer to a file for output.
9236  * @param[out] error
9237  *   Perform verbose error reporting if not NULL. PMDs initialize this
9238  *   structure in case of error only.
9239  * @return
9240  *   0 on success, a negative value otherwise.
9241  */
9242 int
9243 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
9244 	struct rte_flow *flow, FILE *file,
9245 	struct rte_flow_error *error)
9246 {
9247 	struct mlx5_priv *priv = dev->data->dev_private;
9248 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
9249 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
9250 	uint32_t handle_idx;
9251 	struct mlx5_flow_handle *dh;
9252 	struct rte_flow_query_count count;
9253 	uint32_t actions_num;
9254 	const uint8_t *data;
9255 	size_t size;
9256 	uint64_t id;
9257 	uint32_t type;
9258 	void *action = NULL;
9259 
9260 	if (!flow) {
9261 		return rte_flow_error_set(error, ENOENT,
9262 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9263 				NULL,
9264 				"invalid flow handle");
9265 	}
9266 	handle_idx = flow->dev_handles;
9267 	/* query counter */
9268 	if (flow->counter &&
9269 	(!mlx5_counter_query(dev, flow->counter, false,
9270 	&count.hits, &count.bytes, &action)) && action) {
9271 		id = (uint64_t)(uintptr_t)action;
9272 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
9273 		save_dump_file(NULL, 0, type,
9274 			id, (void *)&count, file);
9275 	}
9276 
9277 	while (handle_idx) {
9278 		dh = mlx5_ipool_get(priv->sh->ipool
9279 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
9280 		if (!dh)
9281 			continue;
9282 		handle_idx = dh->next.next;
9283 
9284 		/* Get modify_hdr and encap_decap buf from ipools. */
9285 		encap_decap = NULL;
9286 		modify_hdr = dh->dvh.modify_hdr;
9287 
9288 		if (dh->dvh.rix_encap_decap) {
9289 			encap_decap = mlx5_ipool_get(priv->sh->ipool
9290 						[MLX5_IPOOL_DECAP_ENCAP],
9291 						dh->dvh.rix_encap_decap);
9292 		}
9293 		if (modify_hdr) {
9294 			data = (const uint8_t *)modify_hdr->actions;
9295 			size = (size_t)(modify_hdr->actions_num) * 8;
9296 			id = (uint64_t)(uintptr_t)modify_hdr->action;
9297 			actions_num = modify_hdr->actions_num;
9298 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
9299 			save_dump_file(data, size, type, id,
9300 						(void *)(&actions_num), file);
9301 		}
9302 		if (encap_decap) {
9303 			data = encap_decap->buf;
9304 			size = encap_decap->size;
9305 			id = (uint64_t)(uintptr_t)encap_decap->action;
9306 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
9307 			save_dump_file(data, size, type,
9308 						id, NULL, file);
9309 		}
9310 	}
9311 	return 0;
9312 }
9313 
9314 /**
9315  * Dump all flow's encap_decap/modify_hdr/counter data to file
9316  *
9317  * @param[in] dev
9318  *   The pointer to Ethernet device.
9319  * @param[in] file
9320  *   A pointer to a file for output.
9321  * @param[out] error
9322  *   Perform verbose error reporting if not NULL. PMDs initialize this
9323  *   structure in case of error only.
9324  * @return
9325  *   0 on success, a negative value otherwise.
9326  */
9327 static int
9328 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
9329 	FILE *file, struct rte_flow_error *error __rte_unused)
9330 {
9331 	struct mlx5_priv *priv = dev->data->dev_private;
9332 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9333 	struct mlx5_hlist *h;
9334 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
9335 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
9336 	struct rte_flow_query_count count;
9337 	uint32_t actions_num;
9338 	const uint8_t *data;
9339 	size_t size;
9340 	uint64_t id;
9341 	uint32_t type;
9342 	uint32_t i;
9343 	uint32_t j;
9344 	struct mlx5_list_inconst *l_inconst;
9345 	struct mlx5_list_entry *e;
9346 	int lcore_index;
9347 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
9348 	uint32_t max;
9349 	void *action;
9350 
9351 	/* encap_decap hlist is lcore_share, get global core cache. */
9352 	i = MLX5_LIST_GLOBAL;
9353 	h = sh->encaps_decaps;
9354 	if (h) {
9355 		for (j = 0; j <= h->mask; j++) {
9356 			l_inconst = &h->buckets[j].l;
9357 			if (!l_inconst || !l_inconst->cache[i])
9358 				continue;
9359 
9360 			e = LIST_FIRST(&l_inconst->cache[i]->h);
9361 			while (e) {
9362 				encap_decap =
9363 				(struct mlx5_flow_dv_encap_decap_resource *)e;
9364 				data = encap_decap->buf;
9365 				size = encap_decap->size;
9366 				id = (uint64_t)(uintptr_t)encap_decap->action;
9367 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
9368 				save_dump_file(data, size, type,
9369 					id, NULL, file);
9370 				e = LIST_NEXT(e, next);
9371 			}
9372 		}
9373 	}
9374 
9375 	/* get modify_hdr */
9376 	h = sh->modify_cmds;
9377 	if (h) {
9378 		lcore_index = rte_lcore_index(rte_lcore_id());
9379 		if (unlikely(lcore_index == -1)) {
9380 			lcore_index = MLX5_LIST_NLCORE;
9381 			rte_spinlock_lock(&h->l_const.lcore_lock);
9382 		}
9383 		i = lcore_index;
9384 
9385 		for (j = 0; j <= h->mask; j++) {
9386 			l_inconst = &h->buckets[j].l;
9387 			if (!l_inconst || !l_inconst->cache[i])
9388 				continue;
9389 
9390 			e = LIST_FIRST(&l_inconst->cache[i]->h);
9391 			while (e) {
9392 				modify_hdr =
9393 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
9394 				data = (const uint8_t *)modify_hdr->actions;
9395 				size = (size_t)(modify_hdr->actions_num) * 8;
9396 				actions_num = modify_hdr->actions_num;
9397 				id = (uint64_t)(uintptr_t)modify_hdr->action;
9398 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
9399 				save_dump_file(data, size, type, id,
9400 						(void *)(&actions_num), file);
9401 				e = LIST_NEXT(e, next);
9402 			}
9403 		}
9404 
9405 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
9406 			rte_spinlock_unlock(&h->l_const.lcore_lock);
9407 	}
9408 
9409 	/* get counter */
9410 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
9411 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
9412 	for (j = 1; j <= max; j++) {
9413 		action = NULL;
9414 		if ((!mlx5_counter_query(dev, j, false, &count.hits,
9415 		&count.bytes, &action)) && action) {
9416 			id = (uint64_t)(uintptr_t)action;
9417 			type = DR_DUMP_REC_TYPE_PMD_COUNTER;
9418 			save_dump_file(NULL, 0, type,
9419 					id, (void *)&count, file);
9420 		}
9421 	}
9422 	return 0;
9423 }
9424 #endif
9425 
9426 /**
9427  * Dump flow raw hw data to file
9428  *
9429  * @param[in] dev
9430  *    The pointer to Ethernet device.
9431  * @param[in] file
9432  *   A pointer to a file for output.
9433  * @param[out] error
9434  *   Perform verbose error reporting if not NULL. PMDs initialize this
9435  *   structure in case of error only.
9436  * @return
9437  *   0 on success, a negative value otherwise.
9438  */
9439 int
9440 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
9441 		   FILE *file,
9442 		   struct rte_flow_error *error __rte_unused)
9443 {
9444 	struct mlx5_priv *priv = dev->data->dev_private;
9445 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9446 	uint32_t handle_idx;
9447 	int ret;
9448 	struct mlx5_flow_handle *dh;
9449 	struct rte_flow *flow;
9450 
9451 	if (!sh->config.dv_flow_en) {
9452 		if (fputs("device dv flow disabled\n", file) <= 0)
9453 			return -errno;
9454 		return -ENOTSUP;
9455 	}
9456 
9457 	/* dump all */
9458 	if (!flow_idx) {
9459 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9460 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
9461 			return -EINVAL;
9462 #endif
9463 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
9464 					sh->rx_domain,
9465 					sh->tx_domain, file);
9466 	}
9467 	/* dump one */
9468 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
9469 			(uintptr_t)(void *)flow_idx);
9470 	if (!flow)
9471 		return -EINVAL;
9472 
9473 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9474 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
9475 #endif
9476 	handle_idx = flow->dev_handles;
9477 	while (handle_idx) {
9478 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
9479 				handle_idx);
9480 		if (!dh)
9481 			return -ENOENT;
9482 		if (dh->drv_flow) {
9483 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
9484 					file);
9485 			if (ret)
9486 				return -ENOENT;
9487 		}
9488 		handle_idx = dh->next.next;
9489 	}
9490 	return 0;
9491 }
9492 
9493 /**
9494  * Get aged-out flows.
9495  *
9496  * @param[in] dev
9497  *   Pointer to the Ethernet device structure.
9498  * @param[in] context
9499  *   The address of an array of pointers to the aged-out flows contexts.
9500  * @param[in] nb_countexts
9501  *   The length of context array pointers.
9502  * @param[out] error
9503  *   Perform verbose error reporting if not NULL. Initialized in case of
9504  *   error only.
9505  *
9506  * @return
9507  *   how many contexts get in success, otherwise negative errno value.
9508  *   if nb_contexts is 0, return the amount of all aged contexts.
9509  *   if nb_contexts is not 0 , return the amount of aged flows reported
9510  *   in the context array.
9511  */
9512 int
9513 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
9514 			uint32_t nb_contexts, struct rte_flow_error *error)
9515 {
9516 	const struct mlx5_flow_driver_ops *fops;
9517 	struct rte_flow_attr attr = { .transfer = 0 };
9518 
9519 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
9520 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
9521 		return fops->get_aged_flows(dev, contexts, nb_contexts,
9522 						    error);
9523 	}
9524 	DRV_LOG(ERR,
9525 		"port %u get aged flows is not supported.",
9526 		 dev->data->port_id);
9527 	return -ENOTSUP;
9528 }
9529 
9530 /* Wrapper for driver action_validate op callback */
9531 static int
9532 flow_drv_action_validate(struct rte_eth_dev *dev,
9533 			 const struct rte_flow_indir_action_conf *conf,
9534 			 const struct rte_flow_action *action,
9535 			 const struct mlx5_flow_driver_ops *fops,
9536 			 struct rte_flow_error *error)
9537 {
9538 	static const char err_msg[] = "indirect action validation unsupported";
9539 
9540 	if (!fops->action_validate) {
9541 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9542 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9543 				   NULL, err_msg);
9544 		return -rte_errno;
9545 	}
9546 	return fops->action_validate(dev, conf, action, error);
9547 }
9548 
9549 /**
9550  * Destroys the shared action by handle.
9551  *
9552  * @param dev
9553  *   Pointer to Ethernet device structure.
9554  * @param[in] handle
9555  *   Handle for the indirect action object to be destroyed.
9556  * @param[out] error
9557  *   Perform verbose error reporting if not NULL. PMDs initialize this
9558  *   structure in case of error only.
9559  *
9560  * @return
9561  *   0 on success, a negative errno value otherwise and rte_errno is set.
9562  *
9563  * @note: wrapper for driver action_create op callback.
9564  */
9565 static int
9566 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
9567 			   struct rte_flow_action_handle *handle,
9568 			   struct rte_flow_error *error)
9569 {
9570 	static const char err_msg[] = "indirect action destruction unsupported";
9571 	struct rte_flow_attr attr = { .transfer = 0 };
9572 	const struct mlx5_flow_driver_ops *fops =
9573 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9574 
9575 	if (!fops->action_destroy) {
9576 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9577 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9578 				   NULL, err_msg);
9579 		return -rte_errno;
9580 	}
9581 	return fops->action_destroy(dev, handle, error);
9582 }
9583 
9584 /* Wrapper for driver action_destroy op callback */
9585 static int
9586 flow_drv_action_update(struct rte_eth_dev *dev,
9587 		       struct rte_flow_action_handle *handle,
9588 		       const void *update,
9589 		       const struct mlx5_flow_driver_ops *fops,
9590 		       struct rte_flow_error *error)
9591 {
9592 	static const char err_msg[] = "indirect action update unsupported";
9593 
9594 	if (!fops->action_update) {
9595 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9596 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9597 				   NULL, err_msg);
9598 		return -rte_errno;
9599 	}
9600 	return fops->action_update(dev, handle, update, error);
9601 }
9602 
9603 /* Wrapper for driver action_destroy op callback */
9604 static int
9605 flow_drv_action_query(struct rte_eth_dev *dev,
9606 		      const struct rte_flow_action_handle *handle,
9607 		      void *data,
9608 		      const struct mlx5_flow_driver_ops *fops,
9609 		      struct rte_flow_error *error)
9610 {
9611 	static const char err_msg[] = "indirect action query unsupported";
9612 
9613 	if (!fops->action_query) {
9614 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9615 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9616 				   NULL, err_msg);
9617 		return -rte_errno;
9618 	}
9619 	return fops->action_query(dev, handle, data, error);
9620 }
9621 
9622 /**
9623  * Create indirect action for reuse in multiple flow rules.
9624  *
9625  * @param dev
9626  *   Pointer to Ethernet device structure.
9627  * @param conf
9628  *   Pointer to indirect action object configuration.
9629  * @param[in] action
9630  *   Action configuration for indirect action object creation.
9631  * @param[out] error
9632  *   Perform verbose error reporting if not NULL. PMDs initialize this
9633  *   structure in case of error only.
9634  * @return
9635  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
9636  */
9637 static struct rte_flow_action_handle *
9638 mlx5_action_handle_create(struct rte_eth_dev *dev,
9639 			  const struct rte_flow_indir_action_conf *conf,
9640 			  const struct rte_flow_action *action,
9641 			  struct rte_flow_error *error)
9642 {
9643 	static const char err_msg[] = "indirect action creation unsupported";
9644 	struct rte_flow_attr attr = { .transfer = 0 };
9645 	const struct mlx5_flow_driver_ops *fops =
9646 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9647 
9648 	if (flow_drv_action_validate(dev, conf, action, fops, error))
9649 		return NULL;
9650 	if (!fops->action_create) {
9651 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9652 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9653 				   NULL, err_msg);
9654 		return NULL;
9655 	}
9656 	return fops->action_create(dev, conf, action, error);
9657 }
9658 
9659 /**
9660  * Updates inplace the indirect action configuration pointed by *handle*
9661  * with the configuration provided as *update* argument.
9662  * The update of the indirect action configuration effects all flow rules
9663  * reusing the action via handle.
9664  *
9665  * @param dev
9666  *   Pointer to Ethernet device structure.
9667  * @param[in] handle
9668  *   Handle for the indirect action to be updated.
9669  * @param[in] update
9670  *   Action specification used to modify the action pointed by handle.
9671  *   *update* could be of same type with the action pointed by the *handle*
9672  *   handle argument, or some other structures like a wrapper, depending on
9673  *   the indirect action type.
9674  * @param[out] error
9675  *   Perform verbose error reporting if not NULL. PMDs initialize this
9676  *   structure in case of error only.
9677  *
9678  * @return
9679  *   0 on success, a negative errno value otherwise and rte_errno is set.
9680  */
9681 static int
9682 mlx5_action_handle_update(struct rte_eth_dev *dev,
9683 		struct rte_flow_action_handle *handle,
9684 		const void *update,
9685 		struct rte_flow_error *error)
9686 {
9687 	struct rte_flow_attr attr = { .transfer = 0 };
9688 	const struct mlx5_flow_driver_ops *fops =
9689 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9690 	int ret;
9691 
9692 	ret = flow_drv_action_validate(dev, NULL,
9693 			(const struct rte_flow_action *)update, fops, error);
9694 	if (ret)
9695 		return ret;
9696 	return flow_drv_action_update(dev, handle, update, fops,
9697 				      error);
9698 }
9699 
9700 /**
9701  * Query the indirect action by handle.
9702  *
9703  * This function allows retrieving action-specific data such as counters.
9704  * Data is gathered by special action which may be present/referenced in
9705  * more than one flow rule definition.
9706  *
9707  * see @RTE_FLOW_ACTION_TYPE_COUNT
9708  *
9709  * @param dev
9710  *   Pointer to Ethernet device structure.
9711  * @param[in] handle
9712  *   Handle for the indirect action to query.
9713  * @param[in, out] data
9714  *   Pointer to storage for the associated query data type.
9715  * @param[out] error
9716  *   Perform verbose error reporting if not NULL. PMDs initialize this
9717  *   structure in case of error only.
9718  *
9719  * @return
9720  *   0 on success, a negative errno value otherwise and rte_errno is set.
9721  */
9722 static int
9723 mlx5_action_handle_query(struct rte_eth_dev *dev,
9724 			 const struct rte_flow_action_handle *handle,
9725 			 void *data,
9726 			 struct rte_flow_error *error)
9727 {
9728 	struct rte_flow_attr attr = { .transfer = 0 };
9729 	const struct mlx5_flow_driver_ops *fops =
9730 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9731 
9732 	return flow_drv_action_query(dev, handle, data, fops, error);
9733 }
9734 
9735 /**
9736  * Destroy all indirect actions (shared RSS).
9737  *
9738  * @param dev
9739  *   Pointer to Ethernet device.
9740  *
9741  * @return
9742  *   0 on success, a negative errno value otherwise and rte_errno is set.
9743  */
9744 int
9745 mlx5_action_handle_flush(struct rte_eth_dev *dev)
9746 {
9747 	struct rte_flow_error error;
9748 	struct mlx5_priv *priv = dev->data->dev_private;
9749 	struct mlx5_shared_action_rss *shared_rss;
9750 	int ret = 0;
9751 	uint32_t idx;
9752 
9753 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
9754 		      priv->rss_shared_actions, idx, shared_rss, next) {
9755 		ret |= mlx5_action_handle_destroy(dev,
9756 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
9757 	}
9758 	return ret;
9759 }
9760 
9761 /**
9762  * Validate existing indirect actions against current device configuration
9763  * and attach them to device resources.
9764  *
9765  * @param dev
9766  *   Pointer to Ethernet device.
9767  *
9768  * @return
9769  *   0 on success, a negative errno value otherwise and rte_errno is set.
9770  */
9771 int
9772 mlx5_action_handle_attach(struct rte_eth_dev *dev)
9773 {
9774 	struct mlx5_priv *priv = dev->data->dev_private;
9775 	int ret = 0;
9776 	struct mlx5_ind_table_obj *ind_tbl, *ind_tbl_last;
9777 
9778 	LIST_FOREACH(ind_tbl, &priv->standalone_ind_tbls, next) {
9779 		const char *message;
9780 		uint32_t queue_idx;
9781 
9782 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
9783 					       ind_tbl->queues_n,
9784 					       &message, &queue_idx);
9785 		if (ret != 0) {
9786 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
9787 				dev->data->port_id, ind_tbl->queues[queue_idx],
9788 				message);
9789 			break;
9790 		}
9791 	}
9792 	if (ret != 0)
9793 		return ret;
9794 	LIST_FOREACH(ind_tbl, &priv->standalone_ind_tbls, next) {
9795 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
9796 		if (ret != 0) {
9797 			DRV_LOG(ERR, "Port %u could not attach "
9798 				"indirection table obj %p",
9799 				dev->data->port_id, (void *)ind_tbl);
9800 			goto error;
9801 		}
9802 	}
9803 
9804 	return 0;
9805 error:
9806 	ind_tbl_last = ind_tbl;
9807 	LIST_FOREACH(ind_tbl, &priv->standalone_ind_tbls, next) {
9808 		if (ind_tbl == ind_tbl_last)
9809 			break;
9810 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
9811 			DRV_LOG(CRIT, "Port %u could not detach "
9812 				"indirection table obj %p on rollback",
9813 				dev->data->port_id, (void *)ind_tbl);
9814 	}
9815 	return ret;
9816 }
9817 
9818 /**
9819  * Detach indirect actions of the device from its resources.
9820  *
9821  * @param dev
9822  *   Pointer to Ethernet device.
9823  *
9824  * @return
9825  *   0 on success, a negative errno value otherwise and rte_errno is set.
9826  */
9827 int
9828 mlx5_action_handle_detach(struct rte_eth_dev *dev)
9829 {
9830 	struct mlx5_priv *priv = dev->data->dev_private;
9831 	int ret = 0;
9832 	struct mlx5_ind_table_obj *ind_tbl, *ind_tbl_last;
9833 
9834 	LIST_FOREACH(ind_tbl, &priv->standalone_ind_tbls, next) {
9835 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
9836 		if (ret != 0) {
9837 			DRV_LOG(ERR, "Port %u could not detach "
9838 				"indirection table obj %p",
9839 				dev->data->port_id, (void *)ind_tbl);
9840 			goto error;
9841 		}
9842 	}
9843 	return 0;
9844 error:
9845 	ind_tbl_last = ind_tbl;
9846 	LIST_FOREACH(ind_tbl, &priv->standalone_ind_tbls, next) {
9847 		if (ind_tbl == ind_tbl_last)
9848 			break;
9849 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
9850 			DRV_LOG(CRIT, "Port %u could not attach "
9851 				"indirection table obj %p on rollback",
9852 				dev->data->port_id, (void *)ind_tbl);
9853 	}
9854 	return ret;
9855 }
9856 
9857 #ifndef HAVE_MLX5DV_DR
9858 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
9859 #else
9860 #define MLX5_DOMAIN_SYNC_FLOW \
9861 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
9862 #endif
9863 
9864 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
9865 {
9866 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
9867 	const struct mlx5_flow_driver_ops *fops;
9868 	int ret;
9869 	struct rte_flow_attr attr = { .transfer = 0 };
9870 
9871 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9872 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
9873 	if (ret > 0)
9874 		ret = -ret;
9875 	return ret;
9876 }
9877 
9878 const struct mlx5_flow_tunnel *
9879 mlx5_get_tof(const struct rte_flow_item *item,
9880 	     const struct rte_flow_action *action,
9881 	     enum mlx5_tof_rule_type *rule_type)
9882 {
9883 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9884 		if (item->type == (typeof(item->type))
9885 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
9886 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
9887 			return flow_items_to_tunnel(item);
9888 		}
9889 	}
9890 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9891 		if (action->type == (typeof(action->type))
9892 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9893 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9894 			return flow_actions_to_tunnel(action);
9895 		}
9896 	}
9897 	return NULL;
9898 }
9899 
9900 /**
9901  * tunnel offload functionality is defined for DV environment only
9902  */
9903 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9904 __extension__
9905 union tunnel_offload_mark {
9906 	uint32_t val;
9907 	struct {
9908 		uint32_t app_reserve:8;
9909 		uint32_t table_id:15;
9910 		uint32_t transfer:1;
9911 		uint32_t _unused_:8;
9912 	};
9913 };
9914 
9915 static bool
9916 mlx5_access_tunnel_offload_db
9917 	(struct rte_eth_dev *dev,
9918 	 bool (*match)(struct rte_eth_dev *,
9919 		       struct mlx5_flow_tunnel *, const void *),
9920 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9921 	 void (*miss)(struct rte_eth_dev *, void *),
9922 	 void *ctx, bool lock_op);
9923 
9924 static int
9925 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9926 			     struct rte_flow *flow,
9927 			     const struct rte_flow_attr *attr,
9928 			     const struct rte_flow_action *app_actions,
9929 			     uint32_t flow_idx,
9930 			     const struct mlx5_flow_tunnel *tunnel,
9931 			     struct tunnel_default_miss_ctx *ctx,
9932 			     struct rte_flow_error *error)
9933 {
9934 	struct mlx5_priv *priv = dev->data->dev_private;
9935 	struct mlx5_flow *dev_flow;
9936 	struct rte_flow_attr miss_attr = *attr;
9937 	const struct rte_flow_item miss_items[2] = {
9938 		{
9939 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9940 			.spec = NULL,
9941 			.last = NULL,
9942 			.mask = NULL
9943 		},
9944 		{
9945 			.type = RTE_FLOW_ITEM_TYPE_END,
9946 			.spec = NULL,
9947 			.last = NULL,
9948 			.mask = NULL
9949 		}
9950 	};
9951 	union tunnel_offload_mark mark_id;
9952 	struct rte_flow_action_mark miss_mark;
9953 	struct rte_flow_action miss_actions[3] = {
9954 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9955 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9956 	};
9957 	const struct rte_flow_action_jump *jump_data;
9958 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9959 	struct flow_grp_info grp_info = {
9960 		.external = 1,
9961 		.transfer = attr->transfer,
9962 		.fdb_def_rule = !!priv->fdb_def_rule,
9963 		.std_tbl_fix = 0,
9964 	};
9965 	int ret;
9966 
9967 	if (!attr->transfer) {
9968 		uint32_t q_size;
9969 
9970 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9971 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9972 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9973 					 0, SOCKET_ID_ANY);
9974 		if (!ctx->queue)
9975 			return rte_flow_error_set
9976 				(error, ENOMEM,
9977 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9978 				NULL, "invalid default miss RSS");
9979 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9980 		ctx->action_rss.level = 0,
9981 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9982 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9983 		ctx->action_rss.queue_num = priv->reta_idx_n,
9984 		ctx->action_rss.key = priv->rss_conf.rss_key,
9985 		ctx->action_rss.queue = ctx->queue;
9986 		if (!priv->reta_idx_n || !priv->rxqs_n)
9987 			return rte_flow_error_set
9988 				(error, EINVAL,
9989 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9990 				NULL, "invalid port configuration");
9991 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9992 			ctx->action_rss.types = 0;
9993 		for (i = 0; i != priv->reta_idx_n; ++i)
9994 			ctx->queue[i] = (*priv->reta_idx)[i];
9995 	} else {
9996 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9997 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9998 	}
9999 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
10000 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
10001 	jump_data = app_actions->conf;
10002 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
10003 	miss_attr.group = jump_data->group;
10004 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
10005 				       &flow_table, &grp_info, error);
10006 	if (ret)
10007 		return rte_flow_error_set(error, EINVAL,
10008 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
10009 					  NULL, "invalid tunnel id");
10010 	mark_id.app_reserve = 0;
10011 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
10012 	mark_id.transfer = !!attr->transfer;
10013 	mark_id._unused_ = 0;
10014 	miss_mark.id = mark_id.val;
10015 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
10016 				    miss_items, miss_actions, flow_idx, error);
10017 	if (!dev_flow)
10018 		return -rte_errno;
10019 	dev_flow->flow = flow;
10020 	dev_flow->external = true;
10021 	dev_flow->tunnel = tunnel;
10022 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
10023 	/* Subflow object was created, we must include one in the list. */
10024 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
10025 		      dev_flow->handle, next);
10026 	DRV_LOG(DEBUG,
10027 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
10028 		dev->data->port_id, tunnel->app_tunnel.type,
10029 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
10030 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
10031 				  miss_actions, error);
10032 	if (!ret)
10033 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
10034 						  error);
10035 
10036 	return ret;
10037 }
10038 
10039 static const struct mlx5_flow_tbl_data_entry  *
10040 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
10041 {
10042 	struct mlx5_priv *priv = dev->data->dev_private;
10043 	struct mlx5_dev_ctx_shared *sh = priv->sh;
10044 	struct mlx5_list_entry *he;
10045 	union tunnel_offload_mark mbits = { .val = mark };
10046 	union mlx5_flow_tbl_key table_key = {
10047 		{
10048 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
10049 			.id = 0,
10050 			.reserved = 0,
10051 			.dummy = 0,
10052 			.is_fdb = !!mbits.transfer,
10053 			.is_egress = 0,
10054 		}
10055 	};
10056 	struct mlx5_flow_cb_ctx ctx = {
10057 		.data = &table_key.v64,
10058 	};
10059 
10060 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
10061 	return he ?
10062 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
10063 }
10064 
10065 static void
10066 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
10067 				   struct mlx5_list_entry *entry)
10068 {
10069 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
10070 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
10071 
10072 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
10073 			tunnel_flow_tbl_to_id(tte->flow_table));
10074 	mlx5_free(tte);
10075 }
10076 
10077 static int
10078 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
10079 				  struct mlx5_list_entry *entry, void *cb_ctx)
10080 {
10081 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10082 	union tunnel_tbl_key tbl = {
10083 		.val = *(uint64_t *)(ctx->data),
10084 	};
10085 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
10086 
10087 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
10088 }
10089 
10090 static struct mlx5_list_entry *
10091 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
10092 {
10093 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
10094 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
10095 	struct tunnel_tbl_entry *tte;
10096 	union tunnel_tbl_key tbl = {
10097 		.val = *(uint64_t *)(ctx->data),
10098 	};
10099 
10100 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
10101 			  sizeof(*tte), 0,
10102 			  SOCKET_ID_ANY);
10103 	if (!tte)
10104 		goto err;
10105 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
10106 			  &tte->flow_table);
10107 	if (tte->flow_table >= MLX5_MAX_TABLES) {
10108 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
10109 			tte->flow_table);
10110 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
10111 				tte->flow_table);
10112 		goto err;
10113 	} else if (!tte->flow_table) {
10114 		goto err;
10115 	}
10116 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
10117 	tte->tunnel_id = tbl.tunnel_id;
10118 	tte->group = tbl.group;
10119 	return &tte->hash;
10120 err:
10121 	if (tte)
10122 		mlx5_free(tte);
10123 	return NULL;
10124 }
10125 
10126 static struct mlx5_list_entry *
10127 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
10128 				  struct mlx5_list_entry *oentry,
10129 				  void *cb_ctx __rte_unused)
10130 {
10131 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
10132 						   0, SOCKET_ID_ANY);
10133 
10134 	if (!tte)
10135 		return NULL;
10136 	memcpy(tte, oentry, sizeof(*tte));
10137 	return &tte->hash;
10138 }
10139 
10140 static void
10141 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
10142 				       struct mlx5_list_entry *entry)
10143 {
10144 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
10145 
10146 	mlx5_free(tte);
10147 }
10148 
10149 static uint32_t
10150 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
10151 				const struct mlx5_flow_tunnel *tunnel,
10152 				uint32_t group, uint32_t *table,
10153 				struct rte_flow_error *error)
10154 {
10155 	struct mlx5_list_entry *he;
10156 	struct tunnel_tbl_entry *tte;
10157 	union tunnel_tbl_key key = {
10158 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
10159 		.group = group
10160 	};
10161 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
10162 	struct mlx5_hlist *group_hash;
10163 	struct mlx5_flow_cb_ctx ctx = {
10164 		.data = &key.val,
10165 	};
10166 
10167 	group_hash = tunnel ? tunnel->groups : thub->groups;
10168 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
10169 	if (!he)
10170 		return rte_flow_error_set(error, EINVAL,
10171 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
10172 					  NULL,
10173 					  "tunnel group index not supported");
10174 	tte = container_of(he, typeof(*tte), hash);
10175 	*table = tte->flow_table;
10176 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
10177 		dev->data->port_id, key.tunnel_id, group, *table);
10178 	return 0;
10179 }
10180 
10181 static void
10182 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
10183 		      struct mlx5_flow_tunnel *tunnel)
10184 {
10185 	struct mlx5_priv *priv = dev->data->dev_private;
10186 	struct mlx5_indexed_pool *ipool;
10187 
10188 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
10189 		dev->data->port_id, tunnel->tunnel_id);
10190 	LIST_REMOVE(tunnel, chain);
10191 	mlx5_hlist_destroy(tunnel->groups);
10192 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
10193 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
10194 }
10195 
10196 static bool
10197 mlx5_access_tunnel_offload_db
10198 	(struct rte_eth_dev *dev,
10199 	 bool (*match)(struct rte_eth_dev *,
10200 		       struct mlx5_flow_tunnel *, const void *),
10201 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
10202 	 void (*miss)(struct rte_eth_dev *, void *),
10203 	 void *ctx, bool lock_op)
10204 {
10205 	bool verdict = false;
10206 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
10207 	struct mlx5_flow_tunnel *tunnel;
10208 
10209 	rte_spinlock_lock(&thub->sl);
10210 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
10211 		verdict = match(dev, tunnel, (const void *)ctx);
10212 		if (verdict)
10213 			break;
10214 	}
10215 	if (!lock_op)
10216 		rte_spinlock_unlock(&thub->sl);
10217 	if (verdict && hit)
10218 		hit(dev, tunnel, ctx);
10219 	if (!verdict && miss)
10220 		miss(dev, ctx);
10221 	if (lock_op)
10222 		rte_spinlock_unlock(&thub->sl);
10223 
10224 	return verdict;
10225 }
10226 
10227 struct tunnel_db_find_tunnel_id_ctx {
10228 	uint32_t tunnel_id;
10229 	struct mlx5_flow_tunnel *tunnel;
10230 };
10231 
10232 static bool
10233 find_tunnel_id_match(struct rte_eth_dev *dev,
10234 		     struct mlx5_flow_tunnel *tunnel, const void *x)
10235 {
10236 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
10237 
10238 	RTE_SET_USED(dev);
10239 	return tunnel->tunnel_id == ctx->tunnel_id;
10240 }
10241 
10242 static void
10243 find_tunnel_id_hit(struct rte_eth_dev *dev,
10244 		   struct mlx5_flow_tunnel *tunnel, void *x)
10245 {
10246 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
10247 	RTE_SET_USED(dev);
10248 	ctx->tunnel = tunnel;
10249 }
10250 
10251 static struct mlx5_flow_tunnel *
10252 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
10253 {
10254 	struct tunnel_db_find_tunnel_id_ctx ctx = {
10255 		.tunnel_id = id,
10256 	};
10257 
10258 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
10259 				      find_tunnel_id_hit, NULL, &ctx, true);
10260 
10261 	return ctx.tunnel;
10262 }
10263 
10264 static struct mlx5_flow_tunnel *
10265 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
10266 			  const struct rte_flow_tunnel *app_tunnel)
10267 {
10268 	struct mlx5_priv *priv = dev->data->dev_private;
10269 	struct mlx5_indexed_pool *ipool;
10270 	struct mlx5_flow_tunnel *tunnel;
10271 	uint32_t id;
10272 
10273 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
10274 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
10275 	if (!tunnel)
10276 		return NULL;
10277 	if (id >= MLX5_MAX_TUNNELS) {
10278 		mlx5_ipool_free(ipool, id);
10279 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
10280 		return NULL;
10281 	}
10282 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
10283 					   priv->sh,
10284 					   mlx5_flow_tunnel_grp2tbl_create_cb,
10285 					   mlx5_flow_tunnel_grp2tbl_match_cb,
10286 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
10287 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
10288 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
10289 	if (!tunnel->groups) {
10290 		mlx5_ipool_free(ipool, id);
10291 		return NULL;
10292 	}
10293 	/* initiate new PMD tunnel */
10294 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
10295 	tunnel->tunnel_id = id;
10296 	tunnel->action.type = (typeof(tunnel->action.type))
10297 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
10298 	tunnel->action.conf = tunnel;
10299 	tunnel->item.type = (typeof(tunnel->item.type))
10300 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
10301 	tunnel->item.spec = tunnel;
10302 	tunnel->item.last = NULL;
10303 	tunnel->item.mask = NULL;
10304 
10305 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
10306 		dev->data->port_id, tunnel->tunnel_id);
10307 
10308 	return tunnel;
10309 }
10310 
10311 struct tunnel_db_get_tunnel_ctx {
10312 	const struct rte_flow_tunnel *app_tunnel;
10313 	struct mlx5_flow_tunnel *tunnel;
10314 };
10315 
10316 static bool get_tunnel_match(struct rte_eth_dev *dev,
10317 			     struct mlx5_flow_tunnel *tunnel, const void *x)
10318 {
10319 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
10320 
10321 	RTE_SET_USED(dev);
10322 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
10323 		       sizeof(*ctx->app_tunnel));
10324 }
10325 
10326 static void get_tunnel_hit(struct rte_eth_dev *dev,
10327 			   struct mlx5_flow_tunnel *tunnel, void *x)
10328 {
10329 	/* called under tunnel spinlock protection */
10330 	struct tunnel_db_get_tunnel_ctx *ctx = x;
10331 
10332 	RTE_SET_USED(dev);
10333 	tunnel->refctn++;
10334 	ctx->tunnel = tunnel;
10335 }
10336 
10337 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
10338 {
10339 	/* called under tunnel spinlock protection */
10340 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
10341 	struct tunnel_db_get_tunnel_ctx *ctx = x;
10342 
10343 	rte_spinlock_unlock(&thub->sl);
10344 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
10345 	rte_spinlock_lock(&thub->sl);
10346 	if (ctx->tunnel) {
10347 		ctx->tunnel->refctn = 1;
10348 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
10349 	}
10350 }
10351 
10352 
10353 static int
10354 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
10355 		     const struct rte_flow_tunnel *app_tunnel,
10356 		     struct mlx5_flow_tunnel **tunnel)
10357 {
10358 	struct tunnel_db_get_tunnel_ctx ctx = {
10359 		.app_tunnel = app_tunnel,
10360 	};
10361 
10362 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
10363 				      get_tunnel_miss, &ctx, true);
10364 	*tunnel = ctx.tunnel;
10365 	return ctx.tunnel ? 0 : -ENOMEM;
10366 }
10367 
10368 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
10369 {
10370 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
10371 
10372 	if (!thub)
10373 		return;
10374 	if (!LIST_EMPTY(&thub->tunnels))
10375 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
10376 	mlx5_hlist_destroy(thub->groups);
10377 	mlx5_free(thub);
10378 }
10379 
10380 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
10381 {
10382 	int err;
10383 	struct mlx5_flow_tunnel_hub *thub;
10384 
10385 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
10386 			   0, SOCKET_ID_ANY);
10387 	if (!thub)
10388 		return -ENOMEM;
10389 	LIST_INIT(&thub->tunnels);
10390 	rte_spinlock_init(&thub->sl);
10391 	thub->groups = mlx5_hlist_create("flow groups", 64,
10392 					 false, true, sh,
10393 					 mlx5_flow_tunnel_grp2tbl_create_cb,
10394 					 mlx5_flow_tunnel_grp2tbl_match_cb,
10395 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
10396 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
10397 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
10398 	if (!thub->groups) {
10399 		err = -rte_errno;
10400 		goto err;
10401 	}
10402 	sh->tunnel_hub = thub;
10403 
10404 	return 0;
10405 
10406 err:
10407 	if (thub->groups)
10408 		mlx5_hlist_destroy(thub->groups);
10409 	if (thub)
10410 		mlx5_free(thub);
10411 	return err;
10412 }
10413 
10414 static inline int
10415 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
10416 			  struct rte_flow_tunnel *tunnel,
10417 			  struct rte_flow_error *error)
10418 {
10419 	struct mlx5_priv *priv = dev->data->dev_private;
10420 
10421 	if (!priv->sh->config.dv_flow_en)
10422 		return rte_flow_error_set(error, ENOTSUP,
10423 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
10424 					  "flow DV interface is off");
10425 	if (!is_tunnel_offload_active(dev))
10426 		return rte_flow_error_set(error, ENOTSUP,
10427 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
10428 					  "tunnel offload was not activated");
10429 	if (!tunnel)
10430 		return rte_flow_error_set(error, EINVAL,
10431 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
10432 					  "no application tunnel");
10433 	switch (tunnel->type) {
10434 	default:
10435 		return rte_flow_error_set(error, EINVAL,
10436 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
10437 					  "unsupported tunnel type");
10438 	case RTE_FLOW_ITEM_TYPE_VXLAN:
10439 	case RTE_FLOW_ITEM_TYPE_GRE:
10440 	case RTE_FLOW_ITEM_TYPE_NVGRE:
10441 	case RTE_FLOW_ITEM_TYPE_GENEVE:
10442 		break;
10443 	}
10444 	return 0;
10445 }
10446 
10447 static int
10448 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
10449 		    struct rte_flow_tunnel *app_tunnel,
10450 		    struct rte_flow_action **actions,
10451 		    uint32_t *num_of_actions,
10452 		    struct rte_flow_error *error)
10453 {
10454 	struct mlx5_flow_tunnel *tunnel;
10455 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
10456 
10457 	if (ret)
10458 		return ret;
10459 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
10460 	if (ret < 0) {
10461 		return rte_flow_error_set(error, ret,
10462 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
10463 					  "failed to initialize pmd tunnel");
10464 	}
10465 	*actions = &tunnel->action;
10466 	*num_of_actions = 1;
10467 	return 0;
10468 }
10469 
10470 static int
10471 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
10472 		       struct rte_flow_tunnel *app_tunnel,
10473 		       struct rte_flow_item **items,
10474 		       uint32_t *num_of_items,
10475 		       struct rte_flow_error *error)
10476 {
10477 	struct mlx5_flow_tunnel *tunnel;
10478 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
10479 
10480 	if (ret)
10481 		return ret;
10482 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
10483 	if (ret < 0) {
10484 		return rte_flow_error_set(error, ret,
10485 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
10486 					  "failed to initialize pmd tunnel");
10487 	}
10488 	*items = &tunnel->item;
10489 	*num_of_items = 1;
10490 	return 0;
10491 }
10492 
10493 struct tunnel_db_element_release_ctx {
10494 	struct rte_flow_item *items;
10495 	struct rte_flow_action *actions;
10496 	uint32_t num_elements;
10497 	struct rte_flow_error *error;
10498 	int ret;
10499 };
10500 
10501 static bool
10502 tunnel_element_release_match(struct rte_eth_dev *dev,
10503 			     struct mlx5_flow_tunnel *tunnel, const void *x)
10504 {
10505 	const struct tunnel_db_element_release_ctx *ctx = x;
10506 
10507 	RTE_SET_USED(dev);
10508 	if (ctx->num_elements != 1)
10509 		return false;
10510 	else if (ctx->items)
10511 		return ctx->items == &tunnel->item;
10512 	else if (ctx->actions)
10513 		return ctx->actions == &tunnel->action;
10514 
10515 	return false;
10516 }
10517 
10518 static void
10519 tunnel_element_release_hit(struct rte_eth_dev *dev,
10520 			   struct mlx5_flow_tunnel *tunnel, void *x)
10521 {
10522 	struct tunnel_db_element_release_ctx *ctx = x;
10523 	ctx->ret = 0;
10524 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
10525 		mlx5_flow_tunnel_free(dev, tunnel);
10526 }
10527 
10528 static void
10529 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
10530 {
10531 	struct tunnel_db_element_release_ctx *ctx = x;
10532 	RTE_SET_USED(dev);
10533 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
10534 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
10535 				      "invalid argument");
10536 }
10537 
10538 static int
10539 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
10540 		       struct rte_flow_item *pmd_items,
10541 		       uint32_t num_items, struct rte_flow_error *err)
10542 {
10543 	struct tunnel_db_element_release_ctx ctx = {
10544 		.items = pmd_items,
10545 		.actions = NULL,
10546 		.num_elements = num_items,
10547 		.error = err,
10548 	};
10549 
10550 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
10551 				      tunnel_element_release_hit,
10552 				      tunnel_element_release_miss, &ctx, false);
10553 
10554 	return ctx.ret;
10555 }
10556 
10557 static int
10558 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
10559 			 struct rte_flow_action *pmd_actions,
10560 			 uint32_t num_actions, struct rte_flow_error *err)
10561 {
10562 	struct tunnel_db_element_release_ctx ctx = {
10563 		.items = NULL,
10564 		.actions = pmd_actions,
10565 		.num_elements = num_actions,
10566 		.error = err,
10567 	};
10568 
10569 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
10570 				      tunnel_element_release_hit,
10571 				      tunnel_element_release_miss, &ctx, false);
10572 
10573 	return ctx.ret;
10574 }
10575 
10576 static int
10577 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
10578 				  struct rte_mbuf *m,
10579 				  struct rte_flow_restore_info *info,
10580 				  struct rte_flow_error *err)
10581 {
10582 	uint64_t ol_flags = m->ol_flags;
10583 	const struct mlx5_flow_tbl_data_entry *tble;
10584 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
10585 
10586 	if (!is_tunnel_offload_active(dev)) {
10587 		info->flags = 0;
10588 		return 0;
10589 	}
10590 
10591 	if ((ol_flags & mask) != mask)
10592 		goto err;
10593 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
10594 	if (!tble) {
10595 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
10596 			dev->data->port_id, m->hash.fdir.hi);
10597 		goto err;
10598 	}
10599 	MLX5_ASSERT(tble->tunnel);
10600 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
10601 	info->group_id = tble->group_id;
10602 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
10603 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
10604 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
10605 
10606 	return 0;
10607 
10608 err:
10609 	return rte_flow_error_set(err, EINVAL,
10610 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10611 				  "failed to get restore info");
10612 }
10613 
10614 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
10615 static int
10616 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
10617 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
10618 			   __rte_unused struct rte_flow_action **actions,
10619 			   __rte_unused uint32_t *num_of_actions,
10620 			   __rte_unused struct rte_flow_error *error)
10621 {
10622 	return -ENOTSUP;
10623 }
10624 
10625 static int
10626 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
10627 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
10628 		       __rte_unused struct rte_flow_item **items,
10629 		       __rte_unused uint32_t *num_of_items,
10630 		       __rte_unused struct rte_flow_error *error)
10631 {
10632 	return -ENOTSUP;
10633 }
10634 
10635 static int
10636 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
10637 			      __rte_unused struct rte_flow_item *pmd_items,
10638 			      __rte_unused uint32_t num_items,
10639 			      __rte_unused struct rte_flow_error *err)
10640 {
10641 	return -ENOTSUP;
10642 }
10643 
10644 static int
10645 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
10646 				__rte_unused struct rte_flow_action *pmd_action,
10647 				__rte_unused uint32_t num_actions,
10648 				__rte_unused struct rte_flow_error *err)
10649 {
10650 	return -ENOTSUP;
10651 }
10652 
10653 static int
10654 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
10655 				  __rte_unused struct rte_mbuf *m,
10656 				  __rte_unused struct rte_flow_restore_info *i,
10657 				  __rte_unused struct rte_flow_error *err)
10658 {
10659 	return -ENOTSUP;
10660 }
10661 
10662 static int
10663 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
10664 			     __rte_unused struct rte_flow *flow,
10665 			     __rte_unused const struct rte_flow_attr *attr,
10666 			     __rte_unused const struct rte_flow_action *actions,
10667 			     __rte_unused uint32_t flow_idx,
10668 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
10669 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
10670 			     __rte_unused struct rte_flow_error *error)
10671 {
10672 	return -ENOTSUP;
10673 }
10674 
10675 static struct mlx5_flow_tunnel *
10676 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
10677 		    __rte_unused uint32_t id)
10678 {
10679 	return NULL;
10680 }
10681 
10682 static void
10683 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
10684 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
10685 {
10686 }
10687 
10688 static uint32_t
10689 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
10690 				__rte_unused const struct mlx5_flow_tunnel *t,
10691 				__rte_unused uint32_t group,
10692 				__rte_unused uint32_t *table,
10693 				struct rte_flow_error *error)
10694 {
10695 	return rte_flow_error_set(error, ENOTSUP,
10696 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10697 				  "tunnel offload requires DV support");
10698 }
10699 
10700 void
10701 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
10702 			__rte_unused  uint16_t port_id)
10703 {
10704 }
10705 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
10706 
10707 /* Flex flow item API */
10708 static struct rte_flow_item_flex_handle *
10709 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
10710 			   const struct rte_flow_item_flex_conf *conf,
10711 			   struct rte_flow_error *error)
10712 {
10713 	static const char err_msg[] = "flex item creation unsupported";
10714 	struct mlx5_priv *priv = dev->data->dev_private;
10715 	struct rte_flow_attr attr = { .transfer = 0 };
10716 	const struct mlx5_flow_driver_ops *fops =
10717 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
10718 
10719 	if (!priv->pci_dev) {
10720 		rte_flow_error_set(error, ENOTSUP,
10721 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10722 				   "create flex item on PF only");
10723 		return NULL;
10724 	}
10725 	switch (priv->pci_dev->id.device_id) {
10726 	case PCI_DEVICE_ID_MELLANOX_CONNECTX6DXBF:
10727 	case PCI_DEVICE_ID_MELLANOX_CONNECTX7BF:
10728 		break;
10729 	default:
10730 		rte_flow_error_set(error, ENOTSUP,
10731 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10732 				   "flex item available on BlueField ports only");
10733 		return NULL;
10734 	}
10735 	if (!fops->item_create) {
10736 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
10737 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
10738 				   NULL, err_msg);
10739 		return NULL;
10740 	}
10741 	return fops->item_create(dev, conf, error);
10742 }
10743 
10744 static int
10745 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
10746 			    const struct rte_flow_item_flex_handle *handle,
10747 			    struct rte_flow_error *error)
10748 {
10749 	static const char err_msg[] = "flex item release unsupported";
10750 	struct rte_flow_attr attr = { .transfer = 0 };
10751 	const struct mlx5_flow_driver_ops *fops =
10752 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
10753 
10754 	if (!fops->item_release) {
10755 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
10756 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
10757 				   NULL, err_msg);
10758 		return -rte_errno;
10759 	}
10760 	return fops->item_release(dev, handle, error);
10761 }
10762 
10763 static void
10764 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
10765 {
10766 	int ret;
10767 	struct rte_flow_error error;
10768 
10769 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
10770 		char *item_name;
10771 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
10772 				    sizeof(item_name),
10773 				    (void *)(uintptr_t)item->type, &error);
10774 		if (ret > 0)
10775 			printf("%s ", item_name);
10776 		else
10777 			printf("%d\n", (int)item->type);
10778 	}
10779 	printf("END\n");
10780 }
10781 
10782 static int
10783 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
10784 {
10785 	const struct rte_flow_item_udp *spec = udp_item->spec;
10786 	const struct rte_flow_item_udp *mask = udp_item->mask;
10787 	uint16_t udp_dport = 0;
10788 
10789 	if (spec != NULL) {
10790 		if (!mask)
10791 			mask = &rte_flow_item_udp_mask;
10792 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
10793 				mask->hdr.dst_port);
10794 	}
10795 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
10796 }
10797 
10798 static const struct mlx5_flow_expand_node *
10799 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
10800 		unsigned int item_idx,
10801 		const struct mlx5_flow_expand_node graph[],
10802 		const struct mlx5_flow_expand_node *node)
10803 {
10804 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
10805 
10806 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
10807 			node != NULL &&
10808 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
10809 		/*
10810 		 * The expansion node is VXLAN and it is also the last
10811 		 * expandable item in the pattern, so need to continue
10812 		 * expansion of the inner tunnel.
10813 		 */
10814 		MLX5_ASSERT(item_idx > 0);
10815 		prev_item = pattern + item_idx - 1;
10816 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
10817 		if (mlx5_flow_is_std_vxlan_port(prev_item))
10818 			return &graph[MLX5_EXPANSION_STD_VXLAN];
10819 		return &graph[MLX5_EXPANSION_L3_VXLAN];
10820 	}
10821 	return node;
10822 }
10823 
10824 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
10825 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
10826 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
10827 };
10828 
10829 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
10830 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
10831 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
10832 	{ 9, 10, 11 }, { 12, 13, 14 },
10833 };
10834 
10835 /**
10836  * Discover the number of available flow priorities.
10837  *
10838  * @param dev
10839  *   Ethernet device.
10840  *
10841  * @return
10842  *   On success, number of available flow priorities.
10843  *   On failure, a negative errno-style code and rte_errno is set.
10844  */
10845 int
10846 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
10847 {
10848 	static const uint16_t vprio[] = {8, 16};
10849 	const struct mlx5_priv *priv = dev->data->dev_private;
10850 	const struct mlx5_flow_driver_ops *fops;
10851 	enum mlx5_flow_drv_type type;
10852 	int ret;
10853 
10854 	type = mlx5_flow_os_get_type();
10855 	if (type == MLX5_FLOW_TYPE_MAX) {
10856 		type = MLX5_FLOW_TYPE_VERBS;
10857 		if (priv->sh->cdev->config.devx && priv->sh->config.dv_flow_en)
10858 			type = MLX5_FLOW_TYPE_DV;
10859 	}
10860 	fops = flow_get_drv_ops(type);
10861 	if (fops->discover_priorities == NULL) {
10862 		DRV_LOG(ERR, "Priority discovery not supported");
10863 		rte_errno = ENOTSUP;
10864 		return -rte_errno;
10865 	}
10866 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
10867 	if (ret < 0)
10868 		return ret;
10869 	switch (ret) {
10870 	case 8:
10871 		ret = RTE_DIM(priority_map_3);
10872 		break;
10873 	case 16:
10874 		ret = RTE_DIM(priority_map_5);
10875 		break;
10876 	default:
10877 		rte_errno = ENOTSUP;
10878 		DRV_LOG(ERR,
10879 			"port %u maximum priority: %d expected 8/16",
10880 			dev->data->port_id, ret);
10881 		return -rte_errno;
10882 	}
10883 	DRV_LOG(INFO, "port %u supported flow priorities:"
10884 		" 0-%d for ingress or egress root table,"
10885 		" 0-%d for non-root table or transfer root table.",
10886 		dev->data->port_id, ret - 2,
10887 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
10888 	return ret;
10889 }
10890 
10891 /**
10892  * Adjust flow priority based on the highest layer and the request priority.
10893  *
10894  * @param[in] dev
10895  *   Pointer to the Ethernet device structure.
10896  * @param[in] priority
10897  *   The rule base priority.
10898  * @param[in] subpriority
10899  *   The priority based on the items.
10900  *
10901  * @return
10902  *   The new priority.
10903  */
10904 uint32_t
10905 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10906 			  uint32_t subpriority)
10907 {
10908 	uint32_t res = 0;
10909 	struct mlx5_priv *priv = dev->data->dev_private;
10910 
10911 	switch (priv->sh->flow_max_priority) {
10912 	case RTE_DIM(priority_map_3):
10913 		res = priority_map_3[priority][subpriority];
10914 		break;
10915 	case RTE_DIM(priority_map_5):
10916 		res = priority_map_5[priority][subpriority];
10917 		break;
10918 	}
10919 	return  res;
10920 }
10921