xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision b401400db24ee12b817ab9b0823ac2b0706585bb)
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 RTE_VXLAN_GPE_TYPE_IPV4:
179 		type = RTE_FLOW_ITEM_TYPE_IPV4;
180 		break;
181 	case RTE_VXLAN_GPE_TYPE_IPV6:
182 		type = RTE_VXLAN_GPE_TYPE_IPV6;
183 		break;
184 	case RTE_VXLAN_GPE_TYPE_ETH:
185 		type = RTE_FLOW_ITEM_TYPE_ETH;
186 		break;
187 	default:
188 		type = RTE_FLOW_ITEM_TYPE_END;
189 	}
190 	return type;
191 }
192 
193 static enum rte_flow_item_type
194 mlx5_inet_proto_to_item_type(uint8_t proto_spec, uint8_t proto_mask)
195 {
196 	enum rte_flow_item_type type;
197 
198 	switch (proto_mask & proto_spec) {
199 	case IPPROTO_UDP:
200 		type = RTE_FLOW_ITEM_TYPE_UDP;
201 		break;
202 	case IPPROTO_TCP:
203 		type = RTE_FLOW_ITEM_TYPE_TCP;
204 		break;
205 	case IPPROTO_IP:
206 		type = RTE_FLOW_ITEM_TYPE_IPV4;
207 		break;
208 	case IPPROTO_IPV6:
209 		type = RTE_FLOW_ITEM_TYPE_IPV6;
210 		break;
211 	default:
212 		type = RTE_FLOW_ITEM_TYPE_END;
213 	}
214 	return type;
215 }
216 
217 static enum rte_flow_item_type
218 mlx5_ethertype_to_item_type(rte_be16_t type_spec,
219 			    rte_be16_t type_mask, bool is_tunnel)
220 {
221 	enum rte_flow_item_type type;
222 
223 	switch (rte_be_to_cpu_16(type_spec & type_mask)) {
224 	case RTE_ETHER_TYPE_TEB:
225 		type = is_tunnel ?
226 		       RTE_FLOW_ITEM_TYPE_ETH : RTE_FLOW_ITEM_TYPE_END;
227 		break;
228 	case RTE_ETHER_TYPE_VLAN:
229 		type = !is_tunnel ?
230 		       RTE_FLOW_ITEM_TYPE_VLAN : RTE_FLOW_ITEM_TYPE_END;
231 		break;
232 	case RTE_ETHER_TYPE_IPV4:
233 		type = RTE_FLOW_ITEM_TYPE_IPV4;
234 		break;
235 	case RTE_ETHER_TYPE_IPV6:
236 		type = RTE_FLOW_ITEM_TYPE_IPV6;
237 		break;
238 	default:
239 		type = RTE_FLOW_ITEM_TYPE_END;
240 	}
241 	return type;
242 }
243 
244 static enum rte_flow_item_type
245 mlx5_flow_expand_rss_item_complete(const struct rte_flow_item *item)
246 {
247 #define MLX5_XSET_ITEM_MASK_SPEC(type, fld)                              \
248 	do {                                                             \
249 		const void *m = item->mask;                              \
250 		const void *s = item->spec;                              \
251 		mask = m ?                                               \
252 			((const struct rte_flow_item_##type *)m)->fld :  \
253 			rte_flow_item_##type##_mask.fld;                 \
254 		spec = ((const struct rte_flow_item_##type *)s)->fld;    \
255 	} while (0)
256 
257 	enum rte_flow_item_type ret;
258 	uint16_t spec, mask;
259 
260 	if (item == NULL || item->spec == NULL)
261 		return RTE_FLOW_ITEM_TYPE_VOID;
262 	switch (item->type) {
263 	case RTE_FLOW_ITEM_TYPE_ETH:
264 		MLX5_XSET_ITEM_MASK_SPEC(eth, type);
265 		if (!mask)
266 			return RTE_FLOW_ITEM_TYPE_VOID;
267 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
268 		break;
269 	case RTE_FLOW_ITEM_TYPE_VLAN:
270 		MLX5_XSET_ITEM_MASK_SPEC(vlan, inner_type);
271 		if (!mask)
272 			return RTE_FLOW_ITEM_TYPE_VOID;
273 		ret = mlx5_ethertype_to_item_type(spec, mask, false);
274 		break;
275 	case RTE_FLOW_ITEM_TYPE_IPV4:
276 		MLX5_XSET_ITEM_MASK_SPEC(ipv4, hdr.next_proto_id);
277 		if (!mask)
278 			return RTE_FLOW_ITEM_TYPE_VOID;
279 		ret = mlx5_inet_proto_to_item_type(spec, mask);
280 		break;
281 	case RTE_FLOW_ITEM_TYPE_IPV6:
282 		MLX5_XSET_ITEM_MASK_SPEC(ipv6, hdr.proto);
283 		if (!mask)
284 			return RTE_FLOW_ITEM_TYPE_VOID;
285 		ret = mlx5_inet_proto_to_item_type(spec, mask);
286 		break;
287 	case RTE_FLOW_ITEM_TYPE_GENEVE:
288 		MLX5_XSET_ITEM_MASK_SPEC(geneve, protocol);
289 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
290 		break;
291 	case RTE_FLOW_ITEM_TYPE_GRE:
292 		MLX5_XSET_ITEM_MASK_SPEC(gre, protocol);
293 		ret = mlx5_ethertype_to_item_type(spec, mask, true);
294 		break;
295 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
296 		MLX5_XSET_ITEM_MASK_SPEC(vxlan_gpe, protocol);
297 		ret = mlx5_nsh_proto_to_item_type(spec, mask);
298 		break;
299 	default:
300 		ret = RTE_FLOW_ITEM_TYPE_VOID;
301 		break;
302 	}
303 	return ret;
304 #undef MLX5_XSET_ITEM_MASK_SPEC
305 }
306 
307 static const int *
308 mlx5_flow_expand_rss_skip_explicit(const struct mlx5_flow_expand_node graph[],
309 		const int *next_node)
310 {
311 	const struct mlx5_flow_expand_node *node = NULL;
312 	const int *next = next_node;
313 
314 	while (next && *next) {
315 		/*
316 		 * Skip the nodes with the MLX5_EXPANSION_NODE_EXPLICIT
317 		 * flag set, because they were not found in the flow pattern.
318 		 */
319 		node = &graph[*next];
320 		if (!(node->node_flags & MLX5_EXPANSION_NODE_EXPLICIT))
321 			break;
322 		next = node->next;
323 	}
324 	return next;
325 }
326 
327 #define MLX5_RSS_EXP_ELT_N 16
328 
329 /**
330  * Expand RSS flows into several possible flows according to the RSS hash
331  * fields requested and the driver capabilities.
332  *
333  * @param[out] buf
334  *   Buffer to store the result expansion.
335  * @param[in] size
336  *   Buffer size in bytes. If 0, @p buf can be NULL.
337  * @param[in] pattern
338  *   User flow pattern.
339  * @param[in] types
340  *   RSS types to expand (see RTE_ETH_RSS_* definitions).
341  * @param[in] graph
342  *   Input graph to expand @p pattern according to @p types.
343  * @param[in] graph_root_index
344  *   Index of root node in @p graph, typically 0.
345  *
346  * @return
347  *   A positive value representing the size of @p buf in bytes regardless of
348  *   @p size on success, a negative errno value otherwise and rte_errno is
349  *   set, the following errors are defined:
350  *
351  *   -E2BIG: graph-depth @p graph is too deep.
352  *   -EINVAL: @p size has not enough space for expanded pattern.
353  */
354 static int
355 mlx5_flow_expand_rss(struct mlx5_flow_expand_rss *buf, size_t size,
356 		     const struct rte_flow_item *pattern, uint64_t types,
357 		     const struct mlx5_flow_expand_node graph[],
358 		     int graph_root_index)
359 {
360 	const struct rte_flow_item *item;
361 	const struct mlx5_flow_expand_node *node = &graph[graph_root_index];
362 	const int *next_node;
363 	const int *stack[MLX5_RSS_EXP_ELT_N];
364 	int stack_pos = 0;
365 	struct rte_flow_item flow_items[MLX5_RSS_EXP_ELT_N];
366 	unsigned int i, item_idx, last_expand_item_idx = 0;
367 	size_t lsize;
368 	size_t user_pattern_size = 0;
369 	void *addr = NULL;
370 	const struct mlx5_flow_expand_node *next = NULL;
371 	struct rte_flow_item missed_item;
372 	int missed = 0;
373 	int elt = 0;
374 	const struct rte_flow_item *last_expand_item = NULL;
375 
376 	memset(&missed_item, 0, sizeof(missed_item));
377 	lsize = offsetof(struct mlx5_flow_expand_rss, entry) +
378 		MLX5_RSS_EXP_ELT_N * sizeof(buf->entry[0]);
379 	if (lsize > size)
380 		return -EINVAL;
381 	buf->entry[0].priority = 0;
382 	buf->entry[0].pattern = (void *)&buf->entry[MLX5_RSS_EXP_ELT_N];
383 	buf->entries = 0;
384 	addr = buf->entry[0].pattern;
385 	for (item = pattern, item_idx = 0;
386 			item->type != RTE_FLOW_ITEM_TYPE_END;
387 			item++, item_idx++) {
388 		if (!mlx5_flow_is_rss_expandable_item(item)) {
389 			user_pattern_size += sizeof(*item);
390 			continue;
391 		}
392 		last_expand_item = item;
393 		last_expand_item_idx = item_idx;
394 		i = 0;
395 		while (node->next && node->next[i]) {
396 			next = &graph[node->next[i]];
397 			if (next->type == item->type)
398 				break;
399 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
400 				node = next;
401 				i = 0;
402 			} else {
403 				++i;
404 			}
405 		}
406 		if (next)
407 			node = next;
408 		user_pattern_size += sizeof(*item);
409 	}
410 	user_pattern_size += sizeof(*item); /* Handle END item. */
411 	lsize += user_pattern_size;
412 	if (lsize > size)
413 		return -EINVAL;
414 	/* Copy the user pattern in the first entry of the buffer. */
415 	rte_memcpy(addr, pattern, user_pattern_size);
416 	addr = (void *)(((uintptr_t)addr) + user_pattern_size);
417 	buf->entries = 1;
418 	/* Start expanding. */
419 	memset(flow_items, 0, sizeof(flow_items));
420 	user_pattern_size -= sizeof(*item);
421 	/*
422 	 * Check if the last valid item has spec set, need complete pattern,
423 	 * and the pattern can be used for expansion.
424 	 */
425 	missed_item.type = mlx5_flow_expand_rss_item_complete(last_expand_item);
426 	if (missed_item.type == RTE_FLOW_ITEM_TYPE_END) {
427 		/* Item type END indicates expansion is not required. */
428 		return lsize;
429 	}
430 	if (missed_item.type != RTE_FLOW_ITEM_TYPE_VOID) {
431 		next = NULL;
432 		missed = 1;
433 		i = 0;
434 		while (node->next && node->next[i]) {
435 			next = &graph[node->next[i]];
436 			if (next->type == missed_item.type) {
437 				flow_items[0].type = missed_item.type;
438 				flow_items[1].type = RTE_FLOW_ITEM_TYPE_END;
439 				break;
440 			}
441 			if (next->node_flags & MLX5_EXPANSION_NODE_EXPLICIT) {
442 				node = next;
443 				i = 0;
444 			} else {
445 				++i;
446 			}
447 			next = NULL;
448 		}
449 	}
450 	if (next && missed) {
451 		elt = 2; /* missed item + item end. */
452 		node = next;
453 		lsize += elt * sizeof(*item) + user_pattern_size;
454 		if (lsize > size)
455 			return -EINVAL;
456 		if (node->rss_types & types) {
457 			buf->entry[buf->entries].priority = 1;
458 			buf->entry[buf->entries].pattern = addr;
459 			buf->entries++;
460 			rte_memcpy(addr, buf->entry[0].pattern,
461 				   user_pattern_size);
462 			addr = (void *)(((uintptr_t)addr) + user_pattern_size);
463 			rte_memcpy(addr, flow_items, elt * sizeof(*item));
464 			addr = (void *)(((uintptr_t)addr) +
465 					elt * sizeof(*item));
466 		}
467 	} else if (last_expand_item != NULL) {
468 		node = mlx5_flow_expand_rss_adjust_node(pattern,
469 				last_expand_item_idx, graph, node);
470 	}
471 	memset(flow_items, 0, sizeof(flow_items));
472 	next_node = mlx5_flow_expand_rss_skip_explicit(graph,
473 			node->next);
474 	stack[stack_pos] = next_node;
475 	node = next_node ? &graph[*next_node] : NULL;
476 	while (node) {
477 		flow_items[stack_pos].type = node->type;
478 		if (node->rss_types & types) {
479 			size_t n;
480 			/*
481 			 * compute the number of items to copy from the
482 			 * expansion and copy it.
483 			 * When the stack_pos is 0, there are 1 element in it,
484 			 * plus the addition END item.
485 			 */
486 			elt = stack_pos + 2;
487 			flow_items[stack_pos + 1].type = RTE_FLOW_ITEM_TYPE_END;
488 			lsize += elt * sizeof(*item) + user_pattern_size;
489 			if (lsize > size)
490 				return -EINVAL;
491 			n = elt * sizeof(*item);
492 			buf->entry[buf->entries].priority =
493 				stack_pos + 1 + missed;
494 			buf->entry[buf->entries].pattern = addr;
495 			buf->entries++;
496 			rte_memcpy(addr, buf->entry[0].pattern,
497 				   user_pattern_size);
498 			addr = (void *)(((uintptr_t)addr) +
499 					user_pattern_size);
500 			rte_memcpy(addr, &missed_item,
501 				   missed * sizeof(*item));
502 			addr = (void *)(((uintptr_t)addr) +
503 				missed * sizeof(*item));
504 			rte_memcpy(addr, flow_items, n);
505 			addr = (void *)(((uintptr_t)addr) + n);
506 		}
507 		/* Go deeper. */
508 		if (!(node->node_flags & MLX5_EXPANSION_NODE_OPTIONAL) &&
509 				node->next) {
510 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
511 					node->next);
512 			if (stack_pos++ == MLX5_RSS_EXP_ELT_N) {
513 				rte_errno = E2BIG;
514 				return -rte_errno;
515 			}
516 			stack[stack_pos] = next_node;
517 		} else if (*(next_node + 1)) {
518 			/* Follow up with the next possibility. */
519 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
520 					++next_node);
521 		} else if (!stack_pos) {
522 			/*
523 			 * Completing the traverse over the different paths.
524 			 * The next_node is advanced to the terminator.
525 			 */
526 			++next_node;
527 		} else {
528 			/* Move to the next path. */
529 			while (stack_pos) {
530 				next_node = stack[--stack_pos];
531 				next_node++;
532 				if (*next_node)
533 					break;
534 			}
535 			next_node = mlx5_flow_expand_rss_skip_explicit(graph,
536 					next_node);
537 			stack[stack_pos] = next_node;
538 		}
539 		node = next_node && *next_node ? &graph[*next_node] : NULL;
540 	};
541 	return lsize;
542 }
543 
544 enum mlx5_expansion {
545 	MLX5_EXPANSION_ROOT,
546 	MLX5_EXPANSION_ROOT_OUTER,
547 	MLX5_EXPANSION_OUTER_ETH,
548 	MLX5_EXPANSION_OUTER_VLAN,
549 	MLX5_EXPANSION_OUTER_IPV4,
550 	MLX5_EXPANSION_OUTER_IPV4_UDP,
551 	MLX5_EXPANSION_OUTER_IPV4_TCP,
552 	MLX5_EXPANSION_OUTER_IPV6,
553 	MLX5_EXPANSION_OUTER_IPV6_UDP,
554 	MLX5_EXPANSION_OUTER_IPV6_TCP,
555 	MLX5_EXPANSION_VXLAN,
556 	MLX5_EXPANSION_STD_VXLAN,
557 	MLX5_EXPANSION_L3_VXLAN,
558 	MLX5_EXPANSION_VXLAN_GPE,
559 	MLX5_EXPANSION_GRE,
560 	MLX5_EXPANSION_NVGRE,
561 	MLX5_EXPANSION_GRE_KEY,
562 	MLX5_EXPANSION_MPLS,
563 	MLX5_EXPANSION_ETH,
564 	MLX5_EXPANSION_VLAN,
565 	MLX5_EXPANSION_IPV4,
566 	MLX5_EXPANSION_IPV4_UDP,
567 	MLX5_EXPANSION_IPV4_TCP,
568 	MLX5_EXPANSION_IPV6,
569 	MLX5_EXPANSION_IPV6_UDP,
570 	MLX5_EXPANSION_IPV6_TCP,
571 	MLX5_EXPANSION_IPV6_FRAG_EXT,
572 	MLX5_EXPANSION_GTP,
573 	MLX5_EXPANSION_GENEVE,
574 };
575 
576 /** Supported expansion of items. */
577 static const struct mlx5_flow_expand_node mlx5_support_expansion[] = {
578 	[MLX5_EXPANSION_ROOT] = {
579 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
580 						  MLX5_EXPANSION_IPV4,
581 						  MLX5_EXPANSION_IPV6),
582 		.type = RTE_FLOW_ITEM_TYPE_END,
583 	},
584 	[MLX5_EXPANSION_ROOT_OUTER] = {
585 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
586 						  MLX5_EXPANSION_OUTER_IPV4,
587 						  MLX5_EXPANSION_OUTER_IPV6),
588 		.type = RTE_FLOW_ITEM_TYPE_END,
589 	},
590 	[MLX5_EXPANSION_OUTER_ETH] = {
591 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
592 		.type = RTE_FLOW_ITEM_TYPE_ETH,
593 		.rss_types = 0,
594 	},
595 	[MLX5_EXPANSION_OUTER_VLAN] = {
596 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
597 						  MLX5_EXPANSION_OUTER_IPV6),
598 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
599 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
600 	},
601 	[MLX5_EXPANSION_OUTER_IPV4] = {
602 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
603 			(MLX5_EXPANSION_OUTER_IPV4_UDP,
604 			 MLX5_EXPANSION_OUTER_IPV4_TCP,
605 			 MLX5_EXPANSION_GRE,
606 			 MLX5_EXPANSION_NVGRE,
607 			 MLX5_EXPANSION_IPV4,
608 			 MLX5_EXPANSION_IPV6),
609 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
610 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
611 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
612 	},
613 	[MLX5_EXPANSION_OUTER_IPV4_UDP] = {
614 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
615 						  MLX5_EXPANSION_VXLAN_GPE,
616 						  MLX5_EXPANSION_MPLS,
617 						  MLX5_EXPANSION_GENEVE,
618 						  MLX5_EXPANSION_GTP),
619 		.type = RTE_FLOW_ITEM_TYPE_UDP,
620 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
621 	},
622 	[MLX5_EXPANSION_OUTER_IPV4_TCP] = {
623 		.type = RTE_FLOW_ITEM_TYPE_TCP,
624 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
625 	},
626 	[MLX5_EXPANSION_OUTER_IPV6] = {
627 		.next = MLX5_FLOW_EXPAND_RSS_NEXT
628 			(MLX5_EXPANSION_OUTER_IPV6_UDP,
629 			 MLX5_EXPANSION_OUTER_IPV6_TCP,
630 			 MLX5_EXPANSION_IPV4,
631 			 MLX5_EXPANSION_IPV6,
632 			 MLX5_EXPANSION_GRE,
633 			 MLX5_EXPANSION_NVGRE),
634 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
635 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
636 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
637 	},
638 	[MLX5_EXPANSION_OUTER_IPV6_UDP] = {
639 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
640 						  MLX5_EXPANSION_VXLAN_GPE,
641 						  MLX5_EXPANSION_MPLS,
642 						  MLX5_EXPANSION_GENEVE,
643 						  MLX5_EXPANSION_GTP),
644 		.type = RTE_FLOW_ITEM_TYPE_UDP,
645 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
646 	},
647 	[MLX5_EXPANSION_OUTER_IPV6_TCP] = {
648 		.type = RTE_FLOW_ITEM_TYPE_TCP,
649 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
650 	},
651 	[MLX5_EXPANSION_VXLAN] = {
652 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
653 						  MLX5_EXPANSION_IPV4,
654 						  MLX5_EXPANSION_IPV6),
655 		.type = RTE_FLOW_ITEM_TYPE_VXLAN,
656 	},
657 	[MLX5_EXPANSION_STD_VXLAN] = {
658 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
659 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
660 	},
661 	[MLX5_EXPANSION_L3_VXLAN] = {
662 			.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
663 					MLX5_EXPANSION_IPV6),
664 					.type = RTE_FLOW_ITEM_TYPE_VXLAN,
665 	},
666 	[MLX5_EXPANSION_VXLAN_GPE] = {
667 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
668 						  MLX5_EXPANSION_IPV4,
669 						  MLX5_EXPANSION_IPV6),
670 		.type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
671 	},
672 	[MLX5_EXPANSION_GRE] = {
673 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
674 						  MLX5_EXPANSION_IPV4,
675 						  MLX5_EXPANSION_IPV6,
676 						  MLX5_EXPANSION_GRE_KEY,
677 						  MLX5_EXPANSION_MPLS),
678 		.type = RTE_FLOW_ITEM_TYPE_GRE,
679 	},
680 	[MLX5_EXPANSION_GRE_KEY] = {
681 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
682 						  MLX5_EXPANSION_IPV6,
683 						  MLX5_EXPANSION_MPLS),
684 		.type = RTE_FLOW_ITEM_TYPE_GRE_KEY,
685 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
686 	},
687 	[MLX5_EXPANSION_NVGRE] = {
688 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
689 		.type = RTE_FLOW_ITEM_TYPE_NVGRE,
690 	},
691 	[MLX5_EXPANSION_MPLS] = {
692 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
693 						  MLX5_EXPANSION_IPV6,
694 						  MLX5_EXPANSION_ETH),
695 		.type = RTE_FLOW_ITEM_TYPE_MPLS,
696 		.node_flags = MLX5_EXPANSION_NODE_OPTIONAL,
697 	},
698 	[MLX5_EXPANSION_ETH] = {
699 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
700 		.type = RTE_FLOW_ITEM_TYPE_ETH,
701 	},
702 	[MLX5_EXPANSION_VLAN] = {
703 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
704 						  MLX5_EXPANSION_IPV6),
705 		.type = RTE_FLOW_ITEM_TYPE_VLAN,
706 		.node_flags = MLX5_EXPANSION_NODE_EXPLICIT,
707 	},
708 	[MLX5_EXPANSION_IPV4] = {
709 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
710 						  MLX5_EXPANSION_IPV4_TCP),
711 		.type = RTE_FLOW_ITEM_TYPE_IPV4,
712 		.rss_types = RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_FRAG_IPV4 |
713 			RTE_ETH_RSS_NONFRAG_IPV4_OTHER,
714 	},
715 	[MLX5_EXPANSION_IPV4_UDP] = {
716 		.type = RTE_FLOW_ITEM_TYPE_UDP,
717 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_UDP,
718 	},
719 	[MLX5_EXPANSION_IPV4_TCP] = {
720 		.type = RTE_FLOW_ITEM_TYPE_TCP,
721 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV4_TCP,
722 	},
723 	[MLX5_EXPANSION_IPV6] = {
724 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
725 						  MLX5_EXPANSION_IPV6_TCP,
726 						  MLX5_EXPANSION_IPV6_FRAG_EXT),
727 		.type = RTE_FLOW_ITEM_TYPE_IPV6,
728 		.rss_types = RTE_ETH_RSS_IPV6 | RTE_ETH_RSS_FRAG_IPV6 |
729 			RTE_ETH_RSS_NONFRAG_IPV6_OTHER,
730 	},
731 	[MLX5_EXPANSION_IPV6_UDP] = {
732 		.type = RTE_FLOW_ITEM_TYPE_UDP,
733 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_UDP,
734 	},
735 	[MLX5_EXPANSION_IPV6_TCP] = {
736 		.type = RTE_FLOW_ITEM_TYPE_TCP,
737 		.rss_types = RTE_ETH_RSS_NONFRAG_IPV6_TCP,
738 	},
739 	[MLX5_EXPANSION_IPV6_FRAG_EXT] = {
740 		.type = RTE_FLOW_ITEM_TYPE_IPV6_FRAG_EXT,
741 	},
742 	[MLX5_EXPANSION_GTP] = {
743 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
744 						  MLX5_EXPANSION_IPV6),
745 		.type = RTE_FLOW_ITEM_TYPE_GTP,
746 	},
747 	[MLX5_EXPANSION_GENEVE] = {
748 		.next = MLX5_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
749 						  MLX5_EXPANSION_IPV4,
750 						  MLX5_EXPANSION_IPV6),
751 		.type = RTE_FLOW_ITEM_TYPE_GENEVE,
752 	},
753 };
754 
755 static struct rte_flow_action_handle *
756 mlx5_action_handle_create(struct rte_eth_dev *dev,
757 			  const struct rte_flow_indir_action_conf *conf,
758 			  const struct rte_flow_action *action,
759 			  struct rte_flow_error *error);
760 static int mlx5_action_handle_destroy
761 				(struct rte_eth_dev *dev,
762 				 struct rte_flow_action_handle *handle,
763 				 struct rte_flow_error *error);
764 static int mlx5_action_handle_update
765 				(struct rte_eth_dev *dev,
766 				 struct rte_flow_action_handle *handle,
767 				 const void *update,
768 				 struct rte_flow_error *error);
769 static int mlx5_action_handle_query
770 				(struct rte_eth_dev *dev,
771 				 const struct rte_flow_action_handle *handle,
772 				 void *data,
773 				 struct rte_flow_error *error);
774 static int
775 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
776 		    struct rte_flow_tunnel *app_tunnel,
777 		    struct rte_flow_action **actions,
778 		    uint32_t *num_of_actions,
779 		    struct rte_flow_error *error);
780 static int
781 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
782 		       struct rte_flow_tunnel *app_tunnel,
783 		       struct rte_flow_item **items,
784 		       uint32_t *num_of_items,
785 		       struct rte_flow_error *error);
786 static int
787 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
788 			      struct rte_flow_item *pmd_items,
789 			      uint32_t num_items, struct rte_flow_error *err);
790 static int
791 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
792 				struct rte_flow_action *pmd_actions,
793 				uint32_t num_actions,
794 				struct rte_flow_error *err);
795 static int
796 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
797 				  struct rte_mbuf *m,
798 				  struct rte_flow_restore_info *info,
799 				  struct rte_flow_error *err);
800 static struct rte_flow_item_flex_handle *
801 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
802 			   const struct rte_flow_item_flex_conf *conf,
803 			   struct rte_flow_error *error);
804 static int
805 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
806 			    const struct rte_flow_item_flex_handle *handle,
807 			    struct rte_flow_error *error);
808 static int
809 mlx5_flow_info_get(struct rte_eth_dev *dev,
810 		   struct rte_flow_port_info *port_info,
811 		   struct rte_flow_queue_info *queue_info,
812 		   struct rte_flow_error *error);
813 static int
814 mlx5_flow_port_configure(struct rte_eth_dev *dev,
815 			 const struct rte_flow_port_attr *port_attr,
816 			 uint16_t nb_queue,
817 			 const struct rte_flow_queue_attr *queue_attr[],
818 			 struct rte_flow_error *err);
819 
820 static const struct rte_flow_ops mlx5_flow_ops = {
821 	.validate = mlx5_flow_validate,
822 	.create = mlx5_flow_create,
823 	.destroy = mlx5_flow_destroy,
824 	.flush = mlx5_flow_flush,
825 	.isolate = mlx5_flow_isolate,
826 	.query = mlx5_flow_query,
827 	.dev_dump = mlx5_flow_dev_dump,
828 	.get_aged_flows = mlx5_flow_get_aged_flows,
829 	.action_handle_create = mlx5_action_handle_create,
830 	.action_handle_destroy = mlx5_action_handle_destroy,
831 	.action_handle_update = mlx5_action_handle_update,
832 	.action_handle_query = mlx5_action_handle_query,
833 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
834 	.tunnel_match = mlx5_flow_tunnel_match,
835 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
836 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
837 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
838 	.flex_item_create = mlx5_flow_flex_item_create,
839 	.flex_item_release = mlx5_flow_flex_item_release,
840 	.info_get = mlx5_flow_info_get,
841 	.configure = mlx5_flow_port_configure,
842 };
843 
844 /* Tunnel information. */
845 struct mlx5_flow_tunnel_info {
846 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
847 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
848 };
849 
850 static struct mlx5_flow_tunnel_info tunnels_info[] = {
851 	{
852 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
853 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
854 	},
855 	{
856 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
857 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
858 	},
859 	{
860 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
861 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
862 	},
863 	{
864 		.tunnel = MLX5_FLOW_LAYER_GRE,
865 		.ptype = RTE_PTYPE_TUNNEL_GRE,
866 	},
867 	{
868 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
869 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
870 	},
871 	{
872 		.tunnel = MLX5_FLOW_LAYER_MPLS,
873 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
874 	},
875 	{
876 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
877 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
878 	},
879 	{
880 		.tunnel = MLX5_FLOW_LAYER_IPIP,
881 		.ptype = RTE_PTYPE_TUNNEL_IP,
882 	},
883 	{
884 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
885 		.ptype = RTE_PTYPE_TUNNEL_IP,
886 	},
887 	{
888 		.tunnel = MLX5_FLOW_LAYER_GTP,
889 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
890 	},
891 };
892 
893 
894 
895 /**
896  * Translate tag ID to register.
897  *
898  * @param[in] dev
899  *   Pointer to the Ethernet device structure.
900  * @param[in] feature
901  *   The feature that request the register.
902  * @param[in] id
903  *   The request register ID.
904  * @param[out] error
905  *   Error description in case of any.
906  *
907  * @return
908  *   The request register on success, a negative errno
909  *   value otherwise and rte_errno is set.
910  */
911 int
912 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
913 		     enum mlx5_feature_name feature,
914 		     uint32_t id,
915 		     struct rte_flow_error *error)
916 {
917 	struct mlx5_priv *priv = dev->data->dev_private;
918 	struct mlx5_sh_config *config = &priv->sh->config;
919 	enum modify_reg start_reg;
920 	bool skip_mtr_reg = false;
921 
922 	switch (feature) {
923 	case MLX5_HAIRPIN_RX:
924 		return REG_B;
925 	case MLX5_HAIRPIN_TX:
926 		return REG_A;
927 	case MLX5_METADATA_RX:
928 		switch (config->dv_xmeta_en) {
929 		case MLX5_XMETA_MODE_LEGACY:
930 			return REG_B;
931 		case MLX5_XMETA_MODE_META16:
932 			return REG_C_0;
933 		case MLX5_XMETA_MODE_META32:
934 			return REG_C_1;
935 		}
936 		break;
937 	case MLX5_METADATA_TX:
938 		return REG_A;
939 	case MLX5_METADATA_FDB:
940 		switch (config->dv_xmeta_en) {
941 		case MLX5_XMETA_MODE_LEGACY:
942 			return REG_NON;
943 		case MLX5_XMETA_MODE_META16:
944 			return REG_C_0;
945 		case MLX5_XMETA_MODE_META32:
946 			return REG_C_1;
947 		}
948 		break;
949 	case MLX5_FLOW_MARK:
950 		switch (config->dv_xmeta_en) {
951 		case MLX5_XMETA_MODE_LEGACY:
952 			return REG_NON;
953 		case MLX5_XMETA_MODE_META16:
954 			return REG_C_1;
955 		case MLX5_XMETA_MODE_META32:
956 			return REG_C_0;
957 		}
958 		break;
959 	case MLX5_MTR_ID:
960 		/*
961 		 * If meter color and meter id share one register, flow match
962 		 * should use the meter color register for match.
963 		 */
964 		if (priv->mtr_reg_share)
965 			return priv->mtr_color_reg;
966 		else
967 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
968 			       REG_C_3;
969 	case MLX5_MTR_COLOR:
970 	case MLX5_ASO_FLOW_HIT:
971 	case MLX5_ASO_CONNTRACK:
972 	case MLX5_SAMPLE_ID:
973 		/* All features use the same REG_C. */
974 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
975 		return priv->mtr_color_reg;
976 	case MLX5_COPY_MARK:
977 		/*
978 		 * Metadata COPY_MARK register using is in meter suffix sub
979 		 * flow while with meter. It's safe to share the same register.
980 		 */
981 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
982 	case MLX5_APP_TAG:
983 		/*
984 		 * If meter is enable, it will engage the register for color
985 		 * match and flow match. If meter color match is not using the
986 		 * REG_C_2, need to skip the REG_C_x be used by meter color
987 		 * match.
988 		 * If meter is disable, free to use all available registers.
989 		 */
990 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
991 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
992 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
993 		if (id > (uint32_t)(REG_C_7 - start_reg))
994 			return rte_flow_error_set(error, EINVAL,
995 						  RTE_FLOW_ERROR_TYPE_ITEM,
996 						  NULL, "invalid tag id");
997 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
998 			return rte_flow_error_set(error, ENOTSUP,
999 						  RTE_FLOW_ERROR_TYPE_ITEM,
1000 						  NULL, "unsupported tag id");
1001 		/*
1002 		 * This case means meter is using the REG_C_x great than 2.
1003 		 * Take care not to conflict with meter color REG_C_x.
1004 		 * If the available index REG_C_y >= REG_C_x, skip the
1005 		 * color register.
1006 		 */
1007 		if (skip_mtr_reg && priv->sh->flow_mreg_c
1008 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
1009 			if (id >= (uint32_t)(REG_C_7 - start_reg))
1010 				return rte_flow_error_set(error, EINVAL,
1011 						       RTE_FLOW_ERROR_TYPE_ITEM,
1012 							NULL, "invalid tag id");
1013 			if (priv->sh->flow_mreg_c
1014 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
1015 				return priv->sh->flow_mreg_c
1016 					       [id + 1 + start_reg - REG_C_0];
1017 			return rte_flow_error_set(error, ENOTSUP,
1018 						  RTE_FLOW_ERROR_TYPE_ITEM,
1019 						  NULL, "unsupported tag id");
1020 		}
1021 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
1022 	}
1023 	MLX5_ASSERT(false);
1024 	return rte_flow_error_set(error, EINVAL,
1025 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1026 				  NULL, "invalid feature name");
1027 }
1028 
1029 /**
1030  * Check extensive flow metadata register support.
1031  *
1032  * @param dev
1033  *   Pointer to rte_eth_dev structure.
1034  *
1035  * @return
1036  *   True if device supports extensive flow metadata register, otherwise false.
1037  */
1038 bool
1039 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1040 {
1041 	struct mlx5_priv *priv = dev->data->dev_private;
1042 
1043 	/*
1044 	 * Having available reg_c can be regarded inclusively as supporting
1045 	 * extensive flow metadata register, which could mean,
1046 	 * - metadata register copy action by modify header.
1047 	 * - 16 modify header actions is supported.
1048 	 * - reg_c's are preserved across different domain (FDB and NIC) on
1049 	 *   packet loopback by flow lookup miss.
1050 	 */
1051 	return priv->sh->flow_mreg_c[2] != REG_NON;
1052 }
1053 
1054 /**
1055  * Get the lowest priority.
1056  *
1057  * @param[in] dev
1058  *   Pointer to the Ethernet device structure.
1059  * @param[in] attributes
1060  *   Pointer to device flow rule attributes.
1061  *
1062  * @return
1063  *   The value of lowest priority of flow.
1064  */
1065 uint32_t
1066 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1067 			  const struct rte_flow_attr *attr)
1068 {
1069 	struct mlx5_priv *priv = dev->data->dev_private;
1070 
1071 	if (!attr->group && !attr->transfer)
1072 		return priv->sh->flow_max_priority - 2;
1073 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1074 }
1075 
1076 /**
1077  * Calculate matcher priority of the flow.
1078  *
1079  * @param[in] dev
1080  *   Pointer to the Ethernet device structure.
1081  * @param[in] attr
1082  *   Pointer to device flow rule attributes.
1083  * @param[in] subpriority
1084  *   The priority based on the items.
1085  * @param[in] external
1086  *   Flow is user flow.
1087  * @return
1088  *   The matcher priority of the flow.
1089  */
1090 uint16_t
1091 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1092 			  const struct rte_flow_attr *attr,
1093 			  uint32_t subpriority, bool external)
1094 {
1095 	uint16_t priority = (uint16_t)attr->priority;
1096 	struct mlx5_priv *priv = dev->data->dev_private;
1097 
1098 	if (!attr->group && !attr->transfer) {
1099 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1100 			priority = priv->sh->flow_max_priority - 1;
1101 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1102 	} else if (!external && attr->transfer && attr->group == 0 &&
1103 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1104 		return (priv->sh->flow_max_priority - 1) * 3;
1105 	}
1106 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1107 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1108 	return priority * 3 + subpriority;
1109 }
1110 
1111 /**
1112  * Verify the @p item specifications (spec, last, mask) are compatible with the
1113  * NIC capabilities.
1114  *
1115  * @param[in] item
1116  *   Item specification.
1117  * @param[in] mask
1118  *   @p item->mask or flow default bit-masks.
1119  * @param[in] nic_mask
1120  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1121  * @param[in] size
1122  *   Bit-masks size in bytes.
1123  * @param[in] range_accepted
1124  *   True if range of values is accepted for specific fields, false otherwise.
1125  * @param[out] error
1126  *   Pointer to error structure.
1127  *
1128  * @return
1129  *   0 on success, a negative errno value otherwise and rte_errno is set.
1130  */
1131 int
1132 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1133 			  const uint8_t *mask,
1134 			  const uint8_t *nic_mask,
1135 			  unsigned int size,
1136 			  bool range_accepted,
1137 			  struct rte_flow_error *error)
1138 {
1139 	unsigned int i;
1140 
1141 	MLX5_ASSERT(nic_mask);
1142 	for (i = 0; i < size; ++i)
1143 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1144 			return rte_flow_error_set(error, ENOTSUP,
1145 						  RTE_FLOW_ERROR_TYPE_ITEM,
1146 						  item,
1147 						  "mask enables non supported"
1148 						  " bits");
1149 	if (!item->spec && (item->mask || item->last))
1150 		return rte_flow_error_set(error, EINVAL,
1151 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1152 					  "mask/last without a spec is not"
1153 					  " supported");
1154 	if (item->spec && item->last && !range_accepted) {
1155 		uint8_t spec[size];
1156 		uint8_t last[size];
1157 		unsigned int i;
1158 		int ret;
1159 
1160 		for (i = 0; i < size; ++i) {
1161 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1162 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1163 		}
1164 		ret = memcmp(spec, last, size);
1165 		if (ret != 0)
1166 			return rte_flow_error_set(error, EINVAL,
1167 						  RTE_FLOW_ERROR_TYPE_ITEM,
1168 						  item,
1169 						  "range is not valid");
1170 	}
1171 	return 0;
1172 }
1173 
1174 /**
1175  * Adjust the hash fields according to the @p flow information.
1176  *
1177  * @param[in] dev_flow.
1178  *   Pointer to the mlx5_flow.
1179  * @param[in] tunnel
1180  *   1 when the hash field is for a tunnel item.
1181  * @param[in] layer_types
1182  *   RTE_ETH_RSS_* types.
1183  * @param[in] hash_fields
1184  *   Item hash fields.
1185  *
1186  * @return
1187  *   The hash fields that should be used.
1188  */
1189 uint64_t
1190 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1191 			    int tunnel __rte_unused, uint64_t layer_types,
1192 			    uint64_t hash_fields)
1193 {
1194 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1195 	int rss_request_inner = rss_desc->level >= 2;
1196 
1197 	/* Check RSS hash level for tunnel. */
1198 	if (tunnel && rss_request_inner)
1199 		hash_fields |= IBV_RX_HASH_INNER;
1200 	else if (tunnel || rss_request_inner)
1201 		return 0;
1202 #endif
1203 	/* Check if requested layer matches RSS hash fields. */
1204 	if (!(rss_desc->types & layer_types))
1205 		return 0;
1206 	return hash_fields;
1207 }
1208 
1209 /**
1210  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1211  * if several tunnel rules are used on this queue, the tunnel ptype will be
1212  * cleared.
1213  *
1214  * @param rxq_ctrl
1215  *   Rx queue to update.
1216  */
1217 static void
1218 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1219 {
1220 	unsigned int i;
1221 	uint32_t tunnel_ptype = 0;
1222 
1223 	/* Look up for the ptype to use. */
1224 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1225 		if (!rxq_ctrl->flow_tunnels_n[i])
1226 			continue;
1227 		if (!tunnel_ptype) {
1228 			tunnel_ptype = tunnels_info[i].ptype;
1229 		} else {
1230 			tunnel_ptype = 0;
1231 			break;
1232 		}
1233 	}
1234 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1235 }
1236 
1237 /**
1238  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
1239  * flow.
1240  *
1241  * @param[in] dev
1242  *   Pointer to the Ethernet device structure.
1243  * @param[in] dev_handle
1244  *   Pointer to device flow handle structure.
1245  */
1246 void
1247 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1248 		       struct mlx5_flow_handle *dev_handle)
1249 {
1250 	struct mlx5_priv *priv = dev->data->dev_private;
1251 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1252 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1253 	unsigned int i;
1254 
1255 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1256 		struct mlx5_hrxq *hrxq;
1257 
1258 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1259 			      dev_handle->rix_hrxq);
1260 		if (hrxq)
1261 			ind_tbl = hrxq->ind_table;
1262 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1263 		struct mlx5_shared_action_rss *shared_rss;
1264 
1265 		shared_rss = mlx5_ipool_get
1266 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1267 			 dev_handle->rix_srss);
1268 		if (shared_rss)
1269 			ind_tbl = shared_rss->ind_tbl;
1270 	}
1271 	if (!ind_tbl)
1272 		return;
1273 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1274 		int idx = ind_tbl->queues[i];
1275 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1276 
1277 		MLX5_ASSERT(rxq_ctrl != NULL);
1278 		if (rxq_ctrl == NULL)
1279 			continue;
1280 		/*
1281 		 * To support metadata register copy on Tx loopback,
1282 		 * this must be always enabled (metadata may arive
1283 		 * from other port - not from local flows only.
1284 		 */
1285 		if (tunnel) {
1286 			unsigned int j;
1287 
1288 			/* Increase the counter matching the flow. */
1289 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1290 				if ((tunnels_info[j].tunnel &
1291 				     dev_handle->layers) ==
1292 				    tunnels_info[j].tunnel) {
1293 					rxq_ctrl->flow_tunnels_n[j]++;
1294 					break;
1295 				}
1296 			}
1297 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1298 		}
1299 	}
1300 }
1301 
1302 static void
1303 flow_rxq_mark_flag_set(struct rte_eth_dev *dev)
1304 {
1305 	struct mlx5_priv *priv = dev->data->dev_private;
1306 	struct mlx5_rxq_ctrl *rxq_ctrl;
1307 
1308 	if (priv->mark_enabled)
1309 		return;
1310 	LIST_FOREACH(rxq_ctrl, &priv->rxqsctrl, next) {
1311 		rxq_ctrl->rxq.mark = 1;
1312 	}
1313 	priv->mark_enabled = 1;
1314 }
1315 
1316 /**
1317  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1318  *
1319  * @param[in] dev
1320  *   Pointer to the Ethernet device structure.
1321  * @param[in] flow
1322  *   Pointer to flow structure.
1323  */
1324 static void
1325 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1326 {
1327 	struct mlx5_priv *priv = dev->data->dev_private;
1328 	uint32_t handle_idx;
1329 	struct mlx5_flow_handle *dev_handle;
1330 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1331 
1332 	MLX5_ASSERT(wks);
1333 	if (wks->mark)
1334 		flow_rxq_mark_flag_set(dev);
1335 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1336 		       handle_idx, dev_handle, next)
1337 		flow_drv_rxq_flags_set(dev, dev_handle);
1338 }
1339 
1340 /**
1341  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1342  * device flow if no other flow uses it with the same kind of request.
1343  *
1344  * @param dev
1345  *   Pointer to Ethernet device.
1346  * @param[in] dev_handle
1347  *   Pointer to the device flow handle structure.
1348  */
1349 static void
1350 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1351 			struct mlx5_flow_handle *dev_handle)
1352 {
1353 	struct mlx5_priv *priv = dev->data->dev_private;
1354 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1355 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1356 	unsigned int i;
1357 
1358 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1359 		struct mlx5_hrxq *hrxq;
1360 
1361 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1362 			      dev_handle->rix_hrxq);
1363 		if (hrxq)
1364 			ind_tbl = hrxq->ind_table;
1365 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1366 		struct mlx5_shared_action_rss *shared_rss;
1367 
1368 		shared_rss = mlx5_ipool_get
1369 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1370 			 dev_handle->rix_srss);
1371 		if (shared_rss)
1372 			ind_tbl = shared_rss->ind_tbl;
1373 	}
1374 	if (!ind_tbl)
1375 		return;
1376 	MLX5_ASSERT(dev->data->dev_started);
1377 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1378 		int idx = ind_tbl->queues[i];
1379 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1380 
1381 		MLX5_ASSERT(rxq_ctrl != NULL);
1382 		if (rxq_ctrl == NULL)
1383 			continue;
1384 		if (tunnel) {
1385 			unsigned int j;
1386 
1387 			/* Decrease the counter matching the flow. */
1388 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1389 				if ((tunnels_info[j].tunnel &
1390 				     dev_handle->layers) ==
1391 				    tunnels_info[j].tunnel) {
1392 					rxq_ctrl->flow_tunnels_n[j]--;
1393 					break;
1394 				}
1395 			}
1396 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1397 		}
1398 	}
1399 }
1400 
1401 /**
1402  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1403  * @p flow if no other flow uses it with the same kind of request.
1404  *
1405  * @param dev
1406  *   Pointer to Ethernet device.
1407  * @param[in] flow
1408  *   Pointer to the flow.
1409  */
1410 static void
1411 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1412 {
1413 	struct mlx5_priv *priv = dev->data->dev_private;
1414 	uint32_t handle_idx;
1415 	struct mlx5_flow_handle *dev_handle;
1416 
1417 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1418 		       handle_idx, dev_handle, next)
1419 		flow_drv_rxq_flags_trim(dev, dev_handle);
1420 }
1421 
1422 /**
1423  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1424  *
1425  * @param dev
1426  *   Pointer to Ethernet device.
1427  */
1428 static void
1429 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1430 {
1431 	struct mlx5_priv *priv = dev->data->dev_private;
1432 	unsigned int i;
1433 
1434 	for (i = 0; i != priv->rxqs_n; ++i) {
1435 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1436 		unsigned int j;
1437 
1438 		if (rxq == NULL || rxq->ctrl == NULL)
1439 			continue;
1440 		rxq->ctrl->rxq.mark = 0;
1441 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1442 			rxq->ctrl->flow_tunnels_n[j] = 0;
1443 		rxq->ctrl->rxq.tunnel = 0;
1444 	}
1445 	priv->mark_enabled = 0;
1446 }
1447 
1448 /**
1449  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1450  *
1451  * @param[in] dev
1452  *   Pointer to the Ethernet device structure.
1453  */
1454 void
1455 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1456 {
1457 	struct mlx5_priv *priv = dev->data->dev_private;
1458 	unsigned int i;
1459 
1460 	for (i = 0; i != priv->rxqs_n; ++i) {
1461 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1462 		struct mlx5_rxq_data *data;
1463 
1464 		if (rxq == NULL || rxq->ctrl == NULL)
1465 			continue;
1466 		data = &rxq->ctrl->rxq;
1467 		if (!rte_flow_dynf_metadata_avail()) {
1468 			data->dynf_meta = 0;
1469 			data->flow_meta_mask = 0;
1470 			data->flow_meta_offset = -1;
1471 			data->flow_meta_port_mask = 0;
1472 		} else {
1473 			data->dynf_meta = 1;
1474 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1475 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1476 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1477 		}
1478 	}
1479 }
1480 
1481 /*
1482  * return a pointer to the desired action in the list of actions.
1483  *
1484  * @param[in] actions
1485  *   The list of actions to search the action in.
1486  * @param[in] action
1487  *   The action to find.
1488  *
1489  * @return
1490  *   Pointer to the action in the list, if found. NULL otherwise.
1491  */
1492 const struct rte_flow_action *
1493 mlx5_flow_find_action(const struct rte_flow_action *actions,
1494 		      enum rte_flow_action_type action)
1495 {
1496 	if (actions == NULL)
1497 		return NULL;
1498 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1499 		if (actions->type == action)
1500 			return actions;
1501 	return NULL;
1502 }
1503 
1504 /*
1505  * Validate the flag action.
1506  *
1507  * @param[in] action_flags
1508  *   Bit-fields that holds the actions detected until now.
1509  * @param[in] attr
1510  *   Attributes of flow that includes this action.
1511  * @param[out] error
1512  *   Pointer to error structure.
1513  *
1514  * @return
1515  *   0 on success, a negative errno value otherwise and rte_errno is set.
1516  */
1517 int
1518 mlx5_flow_validate_action_flag(uint64_t action_flags,
1519 			       const struct rte_flow_attr *attr,
1520 			       struct rte_flow_error *error)
1521 {
1522 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1523 		return rte_flow_error_set(error, EINVAL,
1524 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1525 					  "can't mark and flag in same flow");
1526 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1527 		return rte_flow_error_set(error, EINVAL,
1528 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1529 					  "can't have 2 flag"
1530 					  " actions in same flow");
1531 	if (attr->egress)
1532 		return rte_flow_error_set(error, ENOTSUP,
1533 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1534 					  "flag action not supported for "
1535 					  "egress");
1536 	return 0;
1537 }
1538 
1539 /*
1540  * Validate the mark action.
1541  *
1542  * @param[in] action
1543  *   Pointer to the queue action.
1544  * @param[in] action_flags
1545  *   Bit-fields that holds the actions detected until now.
1546  * @param[in] attr
1547  *   Attributes of flow that includes this action.
1548  * @param[out] error
1549  *   Pointer to error structure.
1550  *
1551  * @return
1552  *   0 on success, a negative errno value otherwise and rte_errno is set.
1553  */
1554 int
1555 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1556 			       uint64_t action_flags,
1557 			       const struct rte_flow_attr *attr,
1558 			       struct rte_flow_error *error)
1559 {
1560 	const struct rte_flow_action_mark *mark = action->conf;
1561 
1562 	if (!mark)
1563 		return rte_flow_error_set(error, EINVAL,
1564 					  RTE_FLOW_ERROR_TYPE_ACTION,
1565 					  action,
1566 					  "configuration cannot be null");
1567 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1568 		return rte_flow_error_set(error, EINVAL,
1569 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1570 					  &mark->id,
1571 					  "mark id must in 0 <= id < "
1572 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1573 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1574 		return rte_flow_error_set(error, EINVAL,
1575 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1576 					  "can't flag and mark in same flow");
1577 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1578 		return rte_flow_error_set(error, EINVAL,
1579 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1580 					  "can't have 2 mark actions in same"
1581 					  " flow");
1582 	if (attr->egress)
1583 		return rte_flow_error_set(error, ENOTSUP,
1584 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1585 					  "mark action not supported for "
1586 					  "egress");
1587 	return 0;
1588 }
1589 
1590 /*
1591  * Validate the drop action.
1592  *
1593  * @param[in] action_flags
1594  *   Bit-fields that holds the actions detected until now.
1595  * @param[in] attr
1596  *   Attributes of flow that includes this action.
1597  * @param[out] error
1598  *   Pointer to error structure.
1599  *
1600  * @return
1601  *   0 on success, a negative errno value otherwise and rte_errno is set.
1602  */
1603 int
1604 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1605 			       const struct rte_flow_attr *attr,
1606 			       struct rte_flow_error *error)
1607 {
1608 	if (attr->egress)
1609 		return rte_flow_error_set(error, ENOTSUP,
1610 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1611 					  "drop action not supported for "
1612 					  "egress");
1613 	return 0;
1614 }
1615 
1616 /*
1617  * Validate the queue action.
1618  *
1619  * @param[in] action
1620  *   Pointer to the queue action.
1621  * @param[in] action_flags
1622  *   Bit-fields that holds the actions detected until now.
1623  * @param[in] dev
1624  *   Pointer to the Ethernet device structure.
1625  * @param[in] attr
1626  *   Attributes of flow that includes this action.
1627  * @param[out] error
1628  *   Pointer to error structure.
1629  *
1630  * @return
1631  *   0 on success, a negative errno value otherwise and rte_errno is set.
1632  */
1633 int
1634 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1635 				uint64_t action_flags,
1636 				struct rte_eth_dev *dev,
1637 				const struct rte_flow_attr *attr,
1638 				struct rte_flow_error *error)
1639 {
1640 	struct mlx5_priv *priv = dev->data->dev_private;
1641 	const struct rte_flow_action_queue *queue = action->conf;
1642 
1643 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1644 		return rte_flow_error_set(error, EINVAL,
1645 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1646 					  "can't have 2 fate actions in"
1647 					  " same flow");
1648 	if (!priv->rxqs_n)
1649 		return rte_flow_error_set(error, EINVAL,
1650 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1651 					  NULL, "No Rx queues configured");
1652 	if (queue->index >= priv->rxqs_n)
1653 		return rte_flow_error_set(error, EINVAL,
1654 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1655 					  &queue->index,
1656 					  "queue index out of range");
1657 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1658 		return rte_flow_error_set(error, EINVAL,
1659 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1660 					  &queue->index,
1661 					  "queue is not configured");
1662 	if (attr->egress)
1663 		return rte_flow_error_set(error, ENOTSUP,
1664 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1665 					  "queue action not supported for "
1666 					  "egress");
1667 	return 0;
1668 }
1669 
1670 /**
1671  * Validate queue numbers for device RSS.
1672  *
1673  * @param[in] dev
1674  *   Configured device.
1675  * @param[in] queues
1676  *   Array of queue numbers.
1677  * @param[in] queues_n
1678  *   Size of the @p queues array.
1679  * @param[out] error
1680  *   On error, filled with a textual error description.
1681  * @param[out] queue
1682  *   On error, filled with an offending queue index in @p queues array.
1683  *
1684  * @return
1685  *   0 on success, a negative errno code on error.
1686  */
1687 static int
1688 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1689 			 const uint16_t *queues, uint32_t queues_n,
1690 			 const char **error, uint32_t *queue_idx)
1691 {
1692 	const struct mlx5_priv *priv = dev->data->dev_private;
1693 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1694 	uint32_t i;
1695 
1696 	for (i = 0; i != queues_n; ++i) {
1697 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1698 								   queues[i]);
1699 
1700 		if (queues[i] >= priv->rxqs_n) {
1701 			*error = "queue index out of range";
1702 			*queue_idx = i;
1703 			return -EINVAL;
1704 		}
1705 		if (rxq_ctrl == NULL) {
1706 			*error =  "queue is not configured";
1707 			*queue_idx = i;
1708 			return -EINVAL;
1709 		}
1710 		if (i == 0)
1711 			rxq_type = rxq_ctrl->type;
1712 		if (rxq_type != rxq_ctrl->type) {
1713 			*error = "combining hairpin and regular RSS queues is not supported";
1714 			*queue_idx = i;
1715 			return -ENOTSUP;
1716 		}
1717 	}
1718 	return 0;
1719 }
1720 
1721 /*
1722  * Validate the rss action.
1723  *
1724  * @param[in] dev
1725  *   Pointer to the Ethernet device structure.
1726  * @param[in] action
1727  *   Pointer to the queue action.
1728  * @param[out] error
1729  *   Pointer to error structure.
1730  *
1731  * @return
1732  *   0 on success, a negative errno value otherwise and rte_errno is set.
1733  */
1734 int
1735 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1736 			 const struct rte_flow_action *action,
1737 			 struct rte_flow_error *error)
1738 {
1739 	struct mlx5_priv *priv = dev->data->dev_private;
1740 	const struct rte_flow_action_rss *rss = action->conf;
1741 	int ret;
1742 	const char *message;
1743 	uint32_t queue_idx;
1744 
1745 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1746 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1747 		return rte_flow_error_set(error, ENOTSUP,
1748 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1749 					  &rss->func,
1750 					  "RSS hash function not supported");
1751 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1752 	if (rss->level > 2)
1753 #else
1754 	if (rss->level > 1)
1755 #endif
1756 		return rte_flow_error_set(error, ENOTSUP,
1757 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1758 					  &rss->level,
1759 					  "tunnel RSS is not supported");
1760 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1761 	if (rss->key_len == 0 && rss->key != NULL)
1762 		return rte_flow_error_set(error, ENOTSUP,
1763 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1764 					  &rss->key_len,
1765 					  "RSS hash key length 0");
1766 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1767 		return rte_flow_error_set(error, ENOTSUP,
1768 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1769 					  &rss->key_len,
1770 					  "RSS hash key too small");
1771 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1772 		return rte_flow_error_set(error, ENOTSUP,
1773 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1774 					  &rss->key_len,
1775 					  "RSS hash key too large");
1776 	if (rss->queue_num > priv->sh->dev_cap.ind_table_max_size)
1777 		return rte_flow_error_set(error, ENOTSUP,
1778 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1779 					  &rss->queue_num,
1780 					  "number of queues too large");
1781 	if (rss->types & MLX5_RSS_HF_MASK)
1782 		return rte_flow_error_set(error, ENOTSUP,
1783 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1784 					  &rss->types,
1785 					  "some RSS protocols are not"
1786 					  " supported");
1787 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1788 	    !(rss->types & RTE_ETH_RSS_IP))
1789 		return rte_flow_error_set(error, EINVAL,
1790 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1791 					  "L3 partial RSS requested but L3 RSS"
1792 					  " type not specified");
1793 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1794 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1795 		return rte_flow_error_set(error, EINVAL,
1796 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1797 					  "L4 partial RSS requested but L4 RSS"
1798 					  " type not specified");
1799 	if (!priv->rxqs_n)
1800 		return rte_flow_error_set(error, EINVAL,
1801 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1802 					  NULL, "No Rx queues configured");
1803 	if (!rss->queue_num)
1804 		return rte_flow_error_set(error, EINVAL,
1805 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1806 					  NULL, "No queues configured");
1807 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1808 				       &message, &queue_idx);
1809 	if (ret != 0) {
1810 		return rte_flow_error_set(error, -ret,
1811 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1812 					  &rss->queue[queue_idx], message);
1813 	}
1814 	return 0;
1815 }
1816 
1817 /*
1818  * Validate the rss action.
1819  *
1820  * @param[in] action
1821  *   Pointer to the queue action.
1822  * @param[in] action_flags
1823  *   Bit-fields that holds the actions detected until now.
1824  * @param[in] dev
1825  *   Pointer to the Ethernet device structure.
1826  * @param[in] attr
1827  *   Attributes of flow that includes this action.
1828  * @param[in] item_flags
1829  *   Items that were detected.
1830  * @param[out] error
1831  *   Pointer to error structure.
1832  *
1833  * @return
1834  *   0 on success, a negative errno value otherwise and rte_errno is set.
1835  */
1836 int
1837 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1838 			      uint64_t action_flags,
1839 			      struct rte_eth_dev *dev,
1840 			      const struct rte_flow_attr *attr,
1841 			      uint64_t item_flags,
1842 			      struct rte_flow_error *error)
1843 {
1844 	const struct rte_flow_action_rss *rss = action->conf;
1845 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1846 	int ret;
1847 
1848 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1849 		return rte_flow_error_set(error, EINVAL,
1850 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1851 					  "can't have 2 fate actions"
1852 					  " in same flow");
1853 	ret = mlx5_validate_action_rss(dev, action, error);
1854 	if (ret)
1855 		return ret;
1856 	if (attr->egress)
1857 		return rte_flow_error_set(error, ENOTSUP,
1858 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1859 					  "rss action not supported for "
1860 					  "egress");
1861 	if (rss->level > 1 && !tunnel)
1862 		return rte_flow_error_set(error, EINVAL,
1863 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1864 					  "inner RSS is not supported for "
1865 					  "non-tunnel flows");
1866 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1867 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1868 		return rte_flow_error_set(error, EINVAL,
1869 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1870 					  "RSS on eCPRI is not supported now");
1871 	}
1872 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1873 	    !(item_flags &
1874 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1875 	    rss->level > 1)
1876 		return rte_flow_error_set(error, EINVAL,
1877 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1878 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1879 	return 0;
1880 }
1881 
1882 /*
1883  * Validate the default miss action.
1884  *
1885  * @param[in] action_flags
1886  *   Bit-fields that holds the actions detected until now.
1887  * @param[out] error
1888  *   Pointer to error structure.
1889  *
1890  * @return
1891  *   0 on success, a negative errno value otherwise and rte_errno is set.
1892  */
1893 int
1894 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1895 				const struct rte_flow_attr *attr,
1896 				struct rte_flow_error *error)
1897 {
1898 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1899 		return rte_flow_error_set(error, EINVAL,
1900 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1901 					  "can't have 2 fate actions in"
1902 					  " same flow");
1903 	if (attr->egress)
1904 		return rte_flow_error_set(error, ENOTSUP,
1905 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1906 					  "default miss action not supported "
1907 					  "for egress");
1908 	if (attr->group)
1909 		return rte_flow_error_set(error, ENOTSUP,
1910 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1911 					  "only group 0 is supported");
1912 	if (attr->transfer)
1913 		return rte_flow_error_set(error, ENOTSUP,
1914 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1915 					  NULL, "transfer is not supported");
1916 	return 0;
1917 }
1918 
1919 /*
1920  * Validate the count action.
1921  *
1922  * @param[in] dev
1923  *   Pointer to the Ethernet device structure.
1924  * @param[in] attr
1925  *   Attributes of flow that includes this action.
1926  * @param[out] error
1927  *   Pointer to error structure.
1928  *
1929  * @return
1930  *   0 on success, a negative errno value otherwise and rte_errno is set.
1931  */
1932 int
1933 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1934 				const struct rte_flow_attr *attr,
1935 				struct rte_flow_error *error)
1936 {
1937 	if (attr->egress)
1938 		return rte_flow_error_set(error, ENOTSUP,
1939 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1940 					  "count action not supported for "
1941 					  "egress");
1942 	return 0;
1943 }
1944 
1945 /*
1946  * Validate the ASO CT action.
1947  *
1948  * @param[in] dev
1949  *   Pointer to the Ethernet device structure.
1950  * @param[in] conntrack
1951  *   Pointer to the CT action profile.
1952  * @param[out] error
1953  *   Pointer to error structure.
1954  *
1955  * @return
1956  *   0 on success, a negative errno value otherwise and rte_errno is set.
1957  */
1958 int
1959 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1960 			const struct rte_flow_action_conntrack *conntrack,
1961 			struct rte_flow_error *error)
1962 {
1963 	RTE_SET_USED(dev);
1964 
1965 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1966 		return rte_flow_error_set(error, EINVAL,
1967 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1968 					  "Invalid CT state");
1969 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1970 		return rte_flow_error_set(error, EINVAL,
1971 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1972 					  "Invalid last TCP packet flag");
1973 	return 0;
1974 }
1975 
1976 /**
1977  * Verify the @p attributes will be correctly understood by the NIC and store
1978  * them in the @p flow if everything is correct.
1979  *
1980  * @param[in] dev
1981  *   Pointer to the Ethernet device structure.
1982  * @param[in] attributes
1983  *   Pointer to flow attributes
1984  * @param[out] error
1985  *   Pointer to error structure.
1986  *
1987  * @return
1988  *   0 on success, a negative errno value otherwise and rte_errno is set.
1989  */
1990 int
1991 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1992 			      const struct rte_flow_attr *attributes,
1993 			      struct rte_flow_error *error)
1994 {
1995 	struct mlx5_priv *priv = dev->data->dev_private;
1996 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1997 
1998 	if (attributes->group)
1999 		return rte_flow_error_set(error, ENOTSUP,
2000 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
2001 					  NULL, "groups is not supported");
2002 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
2003 	    attributes->priority >= priority_max)
2004 		return rte_flow_error_set(error, ENOTSUP,
2005 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
2006 					  NULL, "priority out of range");
2007 	if (attributes->egress)
2008 		return rte_flow_error_set(error, ENOTSUP,
2009 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
2010 					  "egress is not supported");
2011 	if (attributes->transfer && !priv->sh->config.dv_esw_en)
2012 		return rte_flow_error_set(error, ENOTSUP,
2013 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2014 					  NULL, "transfer is not supported");
2015 	if (!attributes->ingress)
2016 		return rte_flow_error_set(error, EINVAL,
2017 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2018 					  NULL,
2019 					  "ingress attribute is mandatory");
2020 	return 0;
2021 }
2022 
2023 /**
2024  * Validate ICMP6 item.
2025  *
2026  * @param[in] item
2027  *   Item specification.
2028  * @param[in] item_flags
2029  *   Bit-fields that holds the items detected until now.
2030  * @param[in] ext_vlan_sup
2031  *   Whether extended VLAN features are supported or not.
2032  * @param[out] error
2033  *   Pointer to error structure.
2034  *
2035  * @return
2036  *   0 on success, a negative errno value otherwise and rte_errno is set.
2037  */
2038 int
2039 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2040 			       uint64_t item_flags,
2041 			       uint8_t target_protocol,
2042 			       struct rte_flow_error *error)
2043 {
2044 	const struct rte_flow_item_icmp6 *mask = item->mask;
2045 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2046 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2047 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2048 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2049 				      MLX5_FLOW_LAYER_OUTER_L4;
2050 	int ret;
2051 
2052 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2053 		return rte_flow_error_set(error, EINVAL,
2054 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2055 					  "protocol filtering not compatible"
2056 					  " with ICMP6 layer");
2057 	if (!(item_flags & l3m))
2058 		return rte_flow_error_set(error, EINVAL,
2059 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2060 					  "IPv6 is mandatory to filter on"
2061 					  " ICMP6");
2062 	if (item_flags & l4m)
2063 		return rte_flow_error_set(error, EINVAL,
2064 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2065 					  "multiple L4 layers not supported");
2066 	if (!mask)
2067 		mask = &rte_flow_item_icmp6_mask;
2068 	ret = mlx5_flow_item_acceptable
2069 		(item, (const uint8_t *)mask,
2070 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2071 		 sizeof(struct rte_flow_item_icmp6),
2072 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2073 	if (ret < 0)
2074 		return ret;
2075 	return 0;
2076 }
2077 
2078 /**
2079  * Validate ICMP item.
2080  *
2081  * @param[in] item
2082  *   Item specification.
2083  * @param[in] item_flags
2084  *   Bit-fields that holds the items detected until now.
2085  * @param[out] error
2086  *   Pointer to error structure.
2087  *
2088  * @return
2089  *   0 on success, a negative errno value otherwise and rte_errno is set.
2090  */
2091 int
2092 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2093 			     uint64_t item_flags,
2094 			     uint8_t target_protocol,
2095 			     struct rte_flow_error *error)
2096 {
2097 	const struct rte_flow_item_icmp *mask = item->mask;
2098 	const struct rte_flow_item_icmp nic_mask = {
2099 		.hdr.icmp_type = 0xff,
2100 		.hdr.icmp_code = 0xff,
2101 		.hdr.icmp_ident = RTE_BE16(0xffff),
2102 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2103 	};
2104 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2105 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2106 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2107 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2108 				      MLX5_FLOW_LAYER_OUTER_L4;
2109 	int ret;
2110 
2111 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2112 		return rte_flow_error_set(error, EINVAL,
2113 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2114 					  "protocol filtering not compatible"
2115 					  " with ICMP layer");
2116 	if (!(item_flags & l3m))
2117 		return rte_flow_error_set(error, EINVAL,
2118 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2119 					  "IPv4 is mandatory to filter"
2120 					  " on ICMP");
2121 	if (item_flags & l4m)
2122 		return rte_flow_error_set(error, EINVAL,
2123 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2124 					  "multiple L4 layers not supported");
2125 	if (!mask)
2126 		mask = &nic_mask;
2127 	ret = mlx5_flow_item_acceptable
2128 		(item, (const uint8_t *)mask,
2129 		 (const uint8_t *)&nic_mask,
2130 		 sizeof(struct rte_flow_item_icmp),
2131 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2132 	if (ret < 0)
2133 		return ret;
2134 	return 0;
2135 }
2136 
2137 /**
2138  * Validate Ethernet item.
2139  *
2140  * @param[in] item
2141  *   Item specification.
2142  * @param[in] item_flags
2143  *   Bit-fields that holds the items detected until now.
2144  * @param[out] error
2145  *   Pointer to error structure.
2146  *
2147  * @return
2148  *   0 on success, a negative errno value otherwise and rte_errno is set.
2149  */
2150 int
2151 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2152 			    uint64_t item_flags, bool ext_vlan_sup,
2153 			    struct rte_flow_error *error)
2154 {
2155 	const struct rte_flow_item_eth *mask = item->mask;
2156 	const struct rte_flow_item_eth nic_mask = {
2157 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2158 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2159 		.type = RTE_BE16(0xffff),
2160 		.has_vlan = ext_vlan_sup ? 1 : 0,
2161 	};
2162 	int ret;
2163 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2164 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2165 				       MLX5_FLOW_LAYER_OUTER_L2;
2166 
2167 	if (item_flags & ethm)
2168 		return rte_flow_error_set(error, ENOTSUP,
2169 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2170 					  "multiple L2 layers not supported");
2171 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2172 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2173 		return rte_flow_error_set(error, EINVAL,
2174 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2175 					  "L2 layer should not follow "
2176 					  "L3 layers");
2177 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2178 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2179 		return rte_flow_error_set(error, EINVAL,
2180 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2181 					  "L2 layer should not follow VLAN");
2182 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2183 		return rte_flow_error_set(error, EINVAL,
2184 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2185 					  "L2 layer should not follow GTP");
2186 	if (!mask)
2187 		mask = &rte_flow_item_eth_mask;
2188 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2189 					(const uint8_t *)&nic_mask,
2190 					sizeof(struct rte_flow_item_eth),
2191 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2192 	return ret;
2193 }
2194 
2195 /**
2196  * Validate VLAN item.
2197  *
2198  * @param[in] item
2199  *   Item specification.
2200  * @param[in] item_flags
2201  *   Bit-fields that holds the items detected until now.
2202  * @param[in] dev
2203  *   Ethernet device flow is being created on.
2204  * @param[out] error
2205  *   Pointer to error structure.
2206  *
2207  * @return
2208  *   0 on success, a negative errno value otherwise and rte_errno is set.
2209  */
2210 int
2211 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2212 			     uint64_t item_flags,
2213 			     struct rte_eth_dev *dev,
2214 			     struct rte_flow_error *error)
2215 {
2216 	const struct rte_flow_item_vlan *spec = item->spec;
2217 	const struct rte_flow_item_vlan *mask = item->mask;
2218 	const struct rte_flow_item_vlan nic_mask = {
2219 		.tci = RTE_BE16(UINT16_MAX),
2220 		.inner_type = RTE_BE16(UINT16_MAX),
2221 	};
2222 	uint16_t vlan_tag = 0;
2223 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2224 	int ret;
2225 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2226 					MLX5_FLOW_LAYER_INNER_L4) :
2227 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2228 					MLX5_FLOW_LAYER_OUTER_L4);
2229 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2230 					MLX5_FLOW_LAYER_OUTER_VLAN;
2231 
2232 	if (item_flags & vlanm)
2233 		return rte_flow_error_set(error, EINVAL,
2234 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2235 					  "multiple VLAN layers not supported");
2236 	else if ((item_flags & l34m) != 0)
2237 		return rte_flow_error_set(error, EINVAL,
2238 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2239 					  "VLAN cannot follow L3/L4 layer");
2240 	if (!mask)
2241 		mask = &rte_flow_item_vlan_mask;
2242 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2243 					(const uint8_t *)&nic_mask,
2244 					sizeof(struct rte_flow_item_vlan),
2245 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2246 	if (ret)
2247 		return ret;
2248 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2249 		struct mlx5_priv *priv = dev->data->dev_private;
2250 
2251 		if (priv->vmwa_context) {
2252 			/*
2253 			 * Non-NULL context means we have a virtual machine
2254 			 * and SR-IOV enabled, we have to create VLAN interface
2255 			 * to make hypervisor to setup E-Switch vport
2256 			 * context correctly. We avoid creating the multiple
2257 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2258 			 */
2259 			return rte_flow_error_set(error, EINVAL,
2260 						  RTE_FLOW_ERROR_TYPE_ITEM,
2261 						  item,
2262 						  "VLAN tag mask is not"
2263 						  " supported in virtual"
2264 						  " environment");
2265 		}
2266 	}
2267 	if (spec) {
2268 		vlan_tag = spec->tci;
2269 		vlan_tag &= mask->tci;
2270 	}
2271 	/*
2272 	 * From verbs perspective an empty VLAN is equivalent
2273 	 * to a packet without VLAN layer.
2274 	 */
2275 	if (!vlan_tag)
2276 		return rte_flow_error_set(error, EINVAL,
2277 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2278 					  item->spec,
2279 					  "VLAN cannot be empty");
2280 	return 0;
2281 }
2282 
2283 /**
2284  * Validate IPV4 item.
2285  *
2286  * @param[in] item
2287  *   Item specification.
2288  * @param[in] item_flags
2289  *   Bit-fields that holds the items detected until now.
2290  * @param[in] last_item
2291  *   Previous validated item in the pattern items.
2292  * @param[in] ether_type
2293  *   Type in the ethernet layer header (including dot1q).
2294  * @param[in] acc_mask
2295  *   Acceptable mask, if NULL default internal default mask
2296  *   will be used to check whether item fields are supported.
2297  * @param[in] range_accepted
2298  *   True if range of values is accepted for specific fields, false otherwise.
2299  * @param[out] error
2300  *   Pointer to error structure.
2301  *
2302  * @return
2303  *   0 on success, a negative errno value otherwise and rte_errno is set.
2304  */
2305 int
2306 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2307 			     uint64_t item_flags,
2308 			     uint64_t last_item,
2309 			     uint16_t ether_type,
2310 			     const struct rte_flow_item_ipv4 *acc_mask,
2311 			     bool range_accepted,
2312 			     struct rte_flow_error *error)
2313 {
2314 	const struct rte_flow_item_ipv4 *mask = item->mask;
2315 	const struct rte_flow_item_ipv4 *spec = item->spec;
2316 	const struct rte_flow_item_ipv4 nic_mask = {
2317 		.hdr = {
2318 			.src_addr = RTE_BE32(0xffffffff),
2319 			.dst_addr = RTE_BE32(0xffffffff),
2320 			.type_of_service = 0xff,
2321 			.next_proto_id = 0xff,
2322 		},
2323 	};
2324 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2325 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2326 				      MLX5_FLOW_LAYER_OUTER_L3;
2327 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2328 				      MLX5_FLOW_LAYER_OUTER_L4;
2329 	int ret;
2330 	uint8_t next_proto = 0xFF;
2331 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2332 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2333 				  MLX5_FLOW_LAYER_INNER_VLAN);
2334 
2335 	if ((last_item & l2_vlan) && ether_type &&
2336 	    ether_type != RTE_ETHER_TYPE_IPV4)
2337 		return rte_flow_error_set(error, EINVAL,
2338 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2339 					  "IPv4 cannot follow L2/VLAN layer "
2340 					  "which ether type is not IPv4");
2341 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2342 		if (mask && spec)
2343 			next_proto = mask->hdr.next_proto_id &
2344 				     spec->hdr.next_proto_id;
2345 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2346 			return rte_flow_error_set(error, EINVAL,
2347 						  RTE_FLOW_ERROR_TYPE_ITEM,
2348 						  item,
2349 						  "multiple tunnel "
2350 						  "not supported");
2351 	}
2352 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2353 		return rte_flow_error_set(error, EINVAL,
2354 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2355 					  "wrong tunnel type - IPv6 specified "
2356 					  "but IPv4 item provided");
2357 	if (item_flags & l3m)
2358 		return rte_flow_error_set(error, ENOTSUP,
2359 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2360 					  "multiple L3 layers not supported");
2361 	else if (item_flags & l4m)
2362 		return rte_flow_error_set(error, EINVAL,
2363 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2364 					  "L3 cannot follow an L4 layer.");
2365 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2366 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2367 		return rte_flow_error_set(error, EINVAL,
2368 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2369 					  "L3 cannot follow an NVGRE layer.");
2370 	if (!mask)
2371 		mask = &rte_flow_item_ipv4_mask;
2372 	else if (mask->hdr.next_proto_id != 0 &&
2373 		 mask->hdr.next_proto_id != 0xff)
2374 		return rte_flow_error_set(error, EINVAL,
2375 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2376 					  "partial mask is not supported"
2377 					  " for protocol");
2378 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2379 					acc_mask ? (const uint8_t *)acc_mask
2380 						 : (const uint8_t *)&nic_mask,
2381 					sizeof(struct rte_flow_item_ipv4),
2382 					range_accepted, error);
2383 	if (ret < 0)
2384 		return ret;
2385 	return 0;
2386 }
2387 
2388 /**
2389  * Validate IPV6 item.
2390  *
2391  * @param[in] item
2392  *   Item specification.
2393  * @param[in] item_flags
2394  *   Bit-fields that holds the items detected until now.
2395  * @param[in] last_item
2396  *   Previous validated item in the pattern items.
2397  * @param[in] ether_type
2398  *   Type in the ethernet layer header (including dot1q).
2399  * @param[in] acc_mask
2400  *   Acceptable mask, if NULL default internal default mask
2401  *   will be used to check whether item fields are supported.
2402  * @param[out] error
2403  *   Pointer to error structure.
2404  *
2405  * @return
2406  *   0 on success, a negative errno value otherwise and rte_errno is set.
2407  */
2408 int
2409 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2410 			     uint64_t item_flags,
2411 			     uint64_t last_item,
2412 			     uint16_t ether_type,
2413 			     const struct rte_flow_item_ipv6 *acc_mask,
2414 			     struct rte_flow_error *error)
2415 {
2416 	const struct rte_flow_item_ipv6 *mask = item->mask;
2417 	const struct rte_flow_item_ipv6 *spec = item->spec;
2418 	const struct rte_flow_item_ipv6 nic_mask = {
2419 		.hdr = {
2420 			.src_addr =
2421 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2422 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2423 			.dst_addr =
2424 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2425 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2426 			.vtc_flow = RTE_BE32(0xffffffff),
2427 			.proto = 0xff,
2428 		},
2429 	};
2430 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2431 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2432 				      MLX5_FLOW_LAYER_OUTER_L3;
2433 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2434 				      MLX5_FLOW_LAYER_OUTER_L4;
2435 	int ret;
2436 	uint8_t next_proto = 0xFF;
2437 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2438 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2439 				  MLX5_FLOW_LAYER_INNER_VLAN);
2440 
2441 	if ((last_item & l2_vlan) && ether_type &&
2442 	    ether_type != RTE_ETHER_TYPE_IPV6)
2443 		return rte_flow_error_set(error, EINVAL,
2444 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2445 					  "IPv6 cannot follow L2/VLAN layer "
2446 					  "which ether type is not IPv6");
2447 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2448 		next_proto = spec->hdr.proto;
2449 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2450 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2451 			return rte_flow_error_set(error, EINVAL,
2452 						  RTE_FLOW_ERROR_TYPE_ITEM,
2453 						  item,
2454 						  "multiple tunnel "
2455 						  "not supported");
2456 	}
2457 	if (next_proto == IPPROTO_HOPOPTS  ||
2458 	    next_proto == IPPROTO_ROUTING  ||
2459 	    next_proto == IPPROTO_FRAGMENT ||
2460 	    next_proto == IPPROTO_ESP	   ||
2461 	    next_proto == IPPROTO_AH	   ||
2462 	    next_proto == IPPROTO_DSTOPTS)
2463 		return rte_flow_error_set(error, EINVAL,
2464 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2465 					  "IPv6 proto (next header) should "
2466 					  "not be set as extension header");
2467 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2468 		return rte_flow_error_set(error, EINVAL,
2469 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2470 					  "wrong tunnel type - IPv4 specified "
2471 					  "but IPv6 item provided");
2472 	if (item_flags & l3m)
2473 		return rte_flow_error_set(error, ENOTSUP,
2474 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2475 					  "multiple L3 layers not supported");
2476 	else if (item_flags & l4m)
2477 		return rte_flow_error_set(error, EINVAL,
2478 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2479 					  "L3 cannot follow an L4 layer.");
2480 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2481 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2482 		return rte_flow_error_set(error, EINVAL,
2483 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2484 					  "L3 cannot follow an NVGRE layer.");
2485 	if (!mask)
2486 		mask = &rte_flow_item_ipv6_mask;
2487 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2488 					acc_mask ? (const uint8_t *)acc_mask
2489 						 : (const uint8_t *)&nic_mask,
2490 					sizeof(struct rte_flow_item_ipv6),
2491 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2492 	if (ret < 0)
2493 		return ret;
2494 	return 0;
2495 }
2496 
2497 /**
2498  * Validate UDP item.
2499  *
2500  * @param[in] item
2501  *   Item specification.
2502  * @param[in] item_flags
2503  *   Bit-fields that holds the items detected until now.
2504  * @param[in] target_protocol
2505  *   The next protocol in the previous item.
2506  * @param[in] flow_mask
2507  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2508  * @param[out] error
2509  *   Pointer to error structure.
2510  *
2511  * @return
2512  *   0 on success, a negative errno value otherwise and rte_errno is set.
2513  */
2514 int
2515 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2516 			    uint64_t item_flags,
2517 			    uint8_t target_protocol,
2518 			    struct rte_flow_error *error)
2519 {
2520 	const struct rte_flow_item_udp *mask = item->mask;
2521 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2522 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2523 				      MLX5_FLOW_LAYER_OUTER_L3;
2524 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2525 				      MLX5_FLOW_LAYER_OUTER_L4;
2526 	int ret;
2527 
2528 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2529 		return rte_flow_error_set(error, EINVAL,
2530 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2531 					  "protocol filtering not compatible"
2532 					  " with UDP layer");
2533 	if (!(item_flags & l3m))
2534 		return rte_flow_error_set(error, EINVAL,
2535 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2536 					  "L3 is mandatory to filter on L4");
2537 	if (item_flags & l4m)
2538 		return rte_flow_error_set(error, EINVAL,
2539 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2540 					  "multiple L4 layers not supported");
2541 	if (!mask)
2542 		mask = &rte_flow_item_udp_mask;
2543 	ret = mlx5_flow_item_acceptable
2544 		(item, (const uint8_t *)mask,
2545 		 (const uint8_t *)&rte_flow_item_udp_mask,
2546 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2547 		 error);
2548 	if (ret < 0)
2549 		return ret;
2550 	return 0;
2551 }
2552 
2553 /**
2554  * Validate TCP item.
2555  *
2556  * @param[in] item
2557  *   Item specification.
2558  * @param[in] item_flags
2559  *   Bit-fields that holds the items detected until now.
2560  * @param[in] target_protocol
2561  *   The next protocol in the previous item.
2562  * @param[out] error
2563  *   Pointer to error structure.
2564  *
2565  * @return
2566  *   0 on success, a negative errno value otherwise and rte_errno is set.
2567  */
2568 int
2569 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2570 			    uint64_t item_flags,
2571 			    uint8_t target_protocol,
2572 			    const struct rte_flow_item_tcp *flow_mask,
2573 			    struct rte_flow_error *error)
2574 {
2575 	const struct rte_flow_item_tcp *mask = item->mask;
2576 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2577 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2578 				      MLX5_FLOW_LAYER_OUTER_L3;
2579 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2580 				      MLX5_FLOW_LAYER_OUTER_L4;
2581 	int ret;
2582 
2583 	MLX5_ASSERT(flow_mask);
2584 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2585 		return rte_flow_error_set(error, EINVAL,
2586 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2587 					  "protocol filtering not compatible"
2588 					  " with TCP layer");
2589 	if (!(item_flags & l3m))
2590 		return rte_flow_error_set(error, EINVAL,
2591 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2592 					  "L3 is mandatory to filter on L4");
2593 	if (item_flags & l4m)
2594 		return rte_flow_error_set(error, EINVAL,
2595 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2596 					  "multiple L4 layers not supported");
2597 	if (!mask)
2598 		mask = &rte_flow_item_tcp_mask;
2599 	ret = mlx5_flow_item_acceptable
2600 		(item, (const uint8_t *)mask,
2601 		 (const uint8_t *)flow_mask,
2602 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2603 		 error);
2604 	if (ret < 0)
2605 		return ret;
2606 	return 0;
2607 }
2608 
2609 /**
2610  * Validate VXLAN item.
2611  *
2612  * @param[in] dev
2613  *   Pointer to the Ethernet device structure.
2614  * @param[in] udp_dport
2615  *   UDP destination port
2616  * @param[in] item
2617  *   Item specification.
2618  * @param[in] item_flags
2619  *   Bit-fields that holds the items detected until now.
2620  * @param[in] attr
2621  *   Flow rule attributes.
2622  * @param[out] error
2623  *   Pointer to error structure.
2624  *
2625  * @return
2626  *   0 on success, a negative errno value otherwise and rte_errno is set.
2627  */
2628 int
2629 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2630 			      uint16_t udp_dport,
2631 			      const struct rte_flow_item *item,
2632 			      uint64_t item_flags,
2633 			      const struct rte_flow_attr *attr,
2634 			      struct rte_flow_error *error)
2635 {
2636 	const struct rte_flow_item_vxlan *spec = item->spec;
2637 	const struct rte_flow_item_vxlan *mask = item->mask;
2638 	int ret;
2639 	struct mlx5_priv *priv = dev->data->dev_private;
2640 	union vni {
2641 		uint32_t vlan_id;
2642 		uint8_t vni[4];
2643 	} id = { .vlan_id = 0, };
2644 	const struct rte_flow_item_vxlan nic_mask = {
2645 		.vni = "\xff\xff\xff",
2646 		.rsvd1 = 0xff,
2647 	};
2648 	const struct rte_flow_item_vxlan *valid_mask;
2649 
2650 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2651 		return rte_flow_error_set(error, ENOTSUP,
2652 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2653 					  "multiple tunnel layers not"
2654 					  " supported");
2655 	valid_mask = &rte_flow_item_vxlan_mask;
2656 	/*
2657 	 * Verify only UDPv4 is present as defined in
2658 	 * https://tools.ietf.org/html/rfc7348
2659 	 */
2660 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2661 		return rte_flow_error_set(error, EINVAL,
2662 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2663 					  "no outer UDP layer found");
2664 	if (!mask)
2665 		mask = &rte_flow_item_vxlan_mask;
2666 
2667 	if (priv->sh->steering_format_version !=
2668 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2669 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2670 		/* FDB domain & NIC domain non-zero group */
2671 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2672 			valid_mask = &nic_mask;
2673 		/* Group zero in NIC domain */
2674 		if (!attr->group && !attr->transfer &&
2675 		    priv->sh->tunnel_header_0_1)
2676 			valid_mask = &nic_mask;
2677 	}
2678 	ret = mlx5_flow_item_acceptable
2679 		(item, (const uint8_t *)mask,
2680 		 (const uint8_t *)valid_mask,
2681 		 sizeof(struct rte_flow_item_vxlan),
2682 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2683 	if (ret < 0)
2684 		return ret;
2685 	if (spec) {
2686 		memcpy(&id.vni[1], spec->vni, 3);
2687 		memcpy(&id.vni[1], mask->vni, 3);
2688 	}
2689 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2690 		return rte_flow_error_set(error, ENOTSUP,
2691 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2692 					  "VXLAN tunnel must be fully defined");
2693 	return 0;
2694 }
2695 
2696 /**
2697  * Validate VXLAN_GPE item.
2698  *
2699  * @param[in] item
2700  *   Item specification.
2701  * @param[in] item_flags
2702  *   Bit-fields that holds the items detected until now.
2703  * @param[in] priv
2704  *   Pointer to the private data structure.
2705  * @param[in] target_protocol
2706  *   The next protocol in the previous item.
2707  * @param[out] error
2708  *   Pointer to error structure.
2709  *
2710  * @return
2711  *   0 on success, a negative errno value otherwise and rte_errno is set.
2712  */
2713 int
2714 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2715 				  uint64_t item_flags,
2716 				  struct rte_eth_dev *dev,
2717 				  struct rte_flow_error *error)
2718 {
2719 	struct mlx5_priv *priv = dev->data->dev_private;
2720 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2721 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2722 	int ret;
2723 	union vni {
2724 		uint32_t vlan_id;
2725 		uint8_t vni[4];
2726 	} id = { .vlan_id = 0, };
2727 
2728 	if (!priv->sh->config.l3_vxlan_en)
2729 		return rte_flow_error_set(error, ENOTSUP,
2730 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2731 					  "L3 VXLAN is not enabled by device"
2732 					  " parameter and/or not configured in"
2733 					  " firmware");
2734 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2735 		return rte_flow_error_set(error, ENOTSUP,
2736 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2737 					  "multiple tunnel layers not"
2738 					  " supported");
2739 	/*
2740 	 * Verify only UDPv4 is present as defined in
2741 	 * https://tools.ietf.org/html/rfc7348
2742 	 */
2743 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2744 		return rte_flow_error_set(error, EINVAL,
2745 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2746 					  "no outer UDP layer found");
2747 	if (!mask)
2748 		mask = &rte_flow_item_vxlan_gpe_mask;
2749 	ret = mlx5_flow_item_acceptable
2750 		(item, (const uint8_t *)mask,
2751 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2752 		 sizeof(struct rte_flow_item_vxlan_gpe),
2753 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2754 	if (ret < 0)
2755 		return ret;
2756 	if (spec) {
2757 		if (spec->protocol)
2758 			return rte_flow_error_set(error, ENOTSUP,
2759 						  RTE_FLOW_ERROR_TYPE_ITEM,
2760 						  item,
2761 						  "VxLAN-GPE protocol"
2762 						  " not supported");
2763 		memcpy(&id.vni[1], spec->vni, 3);
2764 		memcpy(&id.vni[1], mask->vni, 3);
2765 	}
2766 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2767 		return rte_flow_error_set(error, ENOTSUP,
2768 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2769 					  "VXLAN-GPE tunnel must be fully"
2770 					  " defined");
2771 	return 0;
2772 }
2773 /**
2774  * Validate GRE Key item.
2775  *
2776  * @param[in] item
2777  *   Item specification.
2778  * @param[in] item_flags
2779  *   Bit flags to mark detected items.
2780  * @param[in] gre_item
2781  *   Pointer to gre_item
2782  * @param[out] error
2783  *   Pointer to error structure.
2784  *
2785  * @return
2786  *   0 on success, a negative errno value otherwise and rte_errno is set.
2787  */
2788 int
2789 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2790 				uint64_t item_flags,
2791 				const struct rte_flow_item *gre_item,
2792 				struct rte_flow_error *error)
2793 {
2794 	const rte_be32_t *mask = item->mask;
2795 	int ret = 0;
2796 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2797 	const struct rte_flow_item_gre *gre_spec;
2798 	const struct rte_flow_item_gre *gre_mask;
2799 
2800 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2801 		return rte_flow_error_set(error, ENOTSUP,
2802 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2803 					  "Multiple GRE key not support");
2804 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2805 		return rte_flow_error_set(error, ENOTSUP,
2806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 					  "No preceding GRE header");
2808 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2809 		return rte_flow_error_set(error, ENOTSUP,
2810 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2811 					  "GRE key following a wrong item");
2812 	gre_mask = gre_item->mask;
2813 	if (!gre_mask)
2814 		gre_mask = &rte_flow_item_gre_mask;
2815 	gre_spec = gre_item->spec;
2816 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2817 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2818 		return rte_flow_error_set(error, EINVAL,
2819 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2820 					  "Key bit must be on");
2821 
2822 	if (!mask)
2823 		mask = &gre_key_default_mask;
2824 	ret = mlx5_flow_item_acceptable
2825 		(item, (const uint8_t *)mask,
2826 		 (const uint8_t *)&gre_key_default_mask,
2827 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2828 	return ret;
2829 }
2830 
2831 /**
2832  * Validate GRE item.
2833  *
2834  * @param[in] item
2835  *   Item specification.
2836  * @param[in] item_flags
2837  *   Bit flags to mark detected items.
2838  * @param[in] target_protocol
2839  *   The next protocol in the previous item.
2840  * @param[out] error
2841  *   Pointer to error structure.
2842  *
2843  * @return
2844  *   0 on success, a negative errno value otherwise and rte_errno is set.
2845  */
2846 int
2847 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2848 			    uint64_t item_flags,
2849 			    uint8_t target_protocol,
2850 			    struct rte_flow_error *error)
2851 {
2852 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2853 	const struct rte_flow_item_gre *mask = item->mask;
2854 	int ret;
2855 	const struct rte_flow_item_gre nic_mask = {
2856 		.c_rsvd0_ver = RTE_BE16(0xB000),
2857 		.protocol = RTE_BE16(UINT16_MAX),
2858 	};
2859 
2860 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2861 		return rte_flow_error_set(error, EINVAL,
2862 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2863 					  "protocol filtering not compatible"
2864 					  " with this GRE layer");
2865 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2866 		return rte_flow_error_set(error, ENOTSUP,
2867 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2868 					  "multiple tunnel layers not"
2869 					  " supported");
2870 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2871 		return rte_flow_error_set(error, ENOTSUP,
2872 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2873 					  "L3 Layer is missing");
2874 	if (!mask)
2875 		mask = &rte_flow_item_gre_mask;
2876 	ret = mlx5_flow_item_acceptable
2877 		(item, (const uint8_t *)mask,
2878 		 (const uint8_t *)&nic_mask,
2879 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2880 		 error);
2881 	if (ret < 0)
2882 		return ret;
2883 #ifndef HAVE_MLX5DV_DR
2884 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2885 	if (spec && (spec->protocol & mask->protocol))
2886 		return rte_flow_error_set(error, ENOTSUP,
2887 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2888 					  "without MPLS support the"
2889 					  " specification cannot be used for"
2890 					  " filtering");
2891 #endif
2892 #endif
2893 	return 0;
2894 }
2895 
2896 /**
2897  * Validate Geneve item.
2898  *
2899  * @param[in] item
2900  *   Item specification.
2901  * @param[in] itemFlags
2902  *   Bit-fields that holds the items detected until now.
2903  * @param[in] enPriv
2904  *   Pointer to the private data structure.
2905  * @param[out] error
2906  *   Pointer to error structure.
2907  *
2908  * @return
2909  *   0 on success, a negative errno value otherwise and rte_errno is set.
2910  */
2911 
2912 int
2913 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2914 			       uint64_t item_flags,
2915 			       struct rte_eth_dev *dev,
2916 			       struct rte_flow_error *error)
2917 {
2918 	struct mlx5_priv *priv = dev->data->dev_private;
2919 	const struct rte_flow_item_geneve *spec = item->spec;
2920 	const struct rte_flow_item_geneve *mask = item->mask;
2921 	int ret;
2922 	uint16_t gbhdr;
2923 	uint8_t opt_len = priv->sh->cdev->config.hca_attr.geneve_max_opt_len ?
2924 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2925 	const struct rte_flow_item_geneve nic_mask = {
2926 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2927 		.vni = "\xff\xff\xff",
2928 		.protocol = RTE_BE16(UINT16_MAX),
2929 	};
2930 
2931 	if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_geneve_rx)
2932 		return rte_flow_error_set(error, ENOTSUP,
2933 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2934 					  "L3 Geneve is not enabled by device"
2935 					  " parameter and/or not configured in"
2936 					  " firmware");
2937 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2938 		return rte_flow_error_set(error, ENOTSUP,
2939 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2940 					  "multiple tunnel layers not"
2941 					  " supported");
2942 	/*
2943 	 * Verify only UDPv4 is present as defined in
2944 	 * https://tools.ietf.org/html/rfc7348
2945 	 */
2946 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2947 		return rte_flow_error_set(error, EINVAL,
2948 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2949 					  "no outer UDP layer found");
2950 	if (!mask)
2951 		mask = &rte_flow_item_geneve_mask;
2952 	ret = mlx5_flow_item_acceptable
2953 				  (item, (const uint8_t *)mask,
2954 				   (const uint8_t *)&nic_mask,
2955 				   sizeof(struct rte_flow_item_geneve),
2956 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2957 	if (ret)
2958 		return ret;
2959 	if (spec) {
2960 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2961 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2962 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2963 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2964 			return rte_flow_error_set(error, ENOTSUP,
2965 						  RTE_FLOW_ERROR_TYPE_ITEM,
2966 						  item,
2967 						  "Geneve protocol unsupported"
2968 						  " fields are being used");
2969 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2970 			return rte_flow_error_set
2971 					(error, ENOTSUP,
2972 					 RTE_FLOW_ERROR_TYPE_ITEM,
2973 					 item,
2974 					 "Unsupported Geneve options length");
2975 	}
2976 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2977 		return rte_flow_error_set
2978 				    (error, ENOTSUP,
2979 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2980 				     "Geneve tunnel must be fully defined");
2981 	return 0;
2982 }
2983 
2984 /**
2985  * Validate Geneve TLV option item.
2986  *
2987  * @param[in] item
2988  *   Item specification.
2989  * @param[in] last_item
2990  *   Previous validated item in the pattern items.
2991  * @param[in] geneve_item
2992  *   Previous GENEVE item specification.
2993  * @param[in] dev
2994  *   Pointer to the rte_eth_dev structure.
2995  * @param[out] error
2996  *   Pointer to error structure.
2997  *
2998  * @return
2999  *   0 on success, a negative errno value otherwise and rte_errno is set.
3000  */
3001 int
3002 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
3003 				   uint64_t last_item,
3004 				   const struct rte_flow_item *geneve_item,
3005 				   struct rte_eth_dev *dev,
3006 				   struct rte_flow_error *error)
3007 {
3008 	struct mlx5_priv *priv = dev->data->dev_private;
3009 	struct mlx5_dev_ctx_shared *sh = priv->sh;
3010 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
3011 	struct mlx5_hca_attr *hca_attr = &sh->cdev->config.hca_attr;
3012 	uint8_t data_max_supported =
3013 			hca_attr->max_geneve_tlv_option_data_len * 4;
3014 	const struct rte_flow_item_geneve *geneve_spec;
3015 	const struct rte_flow_item_geneve *geneve_mask;
3016 	const struct rte_flow_item_geneve_opt *spec = item->spec;
3017 	const struct rte_flow_item_geneve_opt *mask = item->mask;
3018 	unsigned int i;
3019 	unsigned int data_len;
3020 	uint8_t tlv_option_len;
3021 	uint16_t optlen_m, optlen_v;
3022 	const struct rte_flow_item_geneve_opt full_mask = {
3023 		.option_class = RTE_BE16(0xffff),
3024 		.option_type = 0xff,
3025 		.option_len = 0x1f,
3026 	};
3027 
3028 	if (!mask)
3029 		mask = &rte_flow_item_geneve_opt_mask;
3030 	if (!spec)
3031 		return rte_flow_error_set
3032 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3033 			"Geneve TLV opt class/type/length must be specified");
3034 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3035 		return rte_flow_error_set
3036 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 			"Geneve TLV opt length exceeds the limit (31)");
3038 	/* Check if class type and length masks are full. */
3039 	if (full_mask.option_class != mask->option_class ||
3040 	    full_mask.option_type != mask->option_type ||
3041 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
3042 		return rte_flow_error_set
3043 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3044 			"Geneve TLV opt class/type/length masks must be full");
3045 	/* Check if length is supported */
3046 	if ((uint32_t)spec->option_len >
3047 			hca_attr->max_geneve_tlv_option_data_len)
3048 		return rte_flow_error_set
3049 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3050 			"Geneve TLV opt length not supported");
3051 	if (hca_attr->max_geneve_tlv_options > 1)
3052 		DRV_LOG(DEBUG,
3053 			"max_geneve_tlv_options supports more than 1 option");
3054 	/* Check GENEVE item preceding. */
3055 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3056 		return rte_flow_error_set
3057 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3058 			"Geneve opt item must be preceded with Geneve item");
3059 	geneve_spec = geneve_item->spec;
3060 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3061 					  &rte_flow_item_geneve_mask;
3062 	/* Check if GENEVE TLV option size doesn't exceed option length */
3063 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3064 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3065 		tlv_option_len = spec->option_len & mask->option_len;
3066 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3067 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3068 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3069 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3070 		if ((optlen_v & optlen_m) <= tlv_option_len)
3071 			return rte_flow_error_set
3072 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3073 				 "GENEVE TLV option length exceeds optlen");
3074 	}
3075 	/* Check if length is 0 or data is 0. */
3076 	if (spec->data == NULL || spec->option_len == 0)
3077 		return rte_flow_error_set
3078 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3079 			"Geneve TLV opt with zero data/length not supported");
3080 	/* Check not all data & mask are 0. */
3081 	data_len = spec->option_len * 4;
3082 	if (mask->data == NULL) {
3083 		for (i = 0; i < data_len; i++)
3084 			if (spec->data[i])
3085 				break;
3086 		if (i == data_len)
3087 			return rte_flow_error_set(error, ENOTSUP,
3088 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3089 				"Can't match on Geneve option data 0");
3090 	} else {
3091 		for (i = 0; i < data_len; i++)
3092 			if (spec->data[i] & mask->data[i])
3093 				break;
3094 		if (i == data_len)
3095 			return rte_flow_error_set(error, ENOTSUP,
3096 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3097 				"Can't match on Geneve option data and mask 0");
3098 		/* Check data mask supported. */
3099 		for (i = data_max_supported; i < data_len ; i++)
3100 			if (mask->data[i])
3101 				return rte_flow_error_set(error, ENOTSUP,
3102 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3103 					"Data mask is of unsupported size");
3104 	}
3105 	/* Check GENEVE option is supported in NIC. */
3106 	if (!hca_attr->geneve_tlv_opt)
3107 		return rte_flow_error_set
3108 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3109 			"Geneve TLV opt not supported");
3110 	/* Check if we already have geneve option with different type/class. */
3111 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3112 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3113 	if (geneve_opt_resource != NULL)
3114 		if (geneve_opt_resource->option_class != spec->option_class ||
3115 		    geneve_opt_resource->option_type != spec->option_type ||
3116 		    geneve_opt_resource->length != spec->option_len) {
3117 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3118 			return rte_flow_error_set(error, ENOTSUP,
3119 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3120 				"Only one Geneve TLV option supported");
3121 		}
3122 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3123 	return 0;
3124 }
3125 
3126 /**
3127  * Validate MPLS item.
3128  *
3129  * @param[in] dev
3130  *   Pointer to the rte_eth_dev structure.
3131  * @param[in] item
3132  *   Item specification.
3133  * @param[in] item_flags
3134  *   Bit-fields that holds the items detected until now.
3135  * @param[in] prev_layer
3136  *   The protocol layer indicated in previous item.
3137  * @param[out] error
3138  *   Pointer to error structure.
3139  *
3140  * @return
3141  *   0 on success, a negative errno value otherwise and rte_errno is set.
3142  */
3143 int
3144 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3145 			     const struct rte_flow_item *item __rte_unused,
3146 			     uint64_t item_flags __rte_unused,
3147 			     uint64_t prev_layer __rte_unused,
3148 			     struct rte_flow_error *error)
3149 {
3150 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3151 	const struct rte_flow_item_mpls *mask = item->mask;
3152 	struct mlx5_priv *priv = dev->data->dev_private;
3153 	int ret;
3154 
3155 	if (!priv->sh->dev_cap.mpls_en)
3156 		return rte_flow_error_set(error, ENOTSUP,
3157 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3158 					  "MPLS not supported or"
3159 					  " disabled in firmware"
3160 					  " configuration.");
3161 	/* MPLS over UDP, GRE is allowed */
3162 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3163 			    MLX5_FLOW_LAYER_GRE |
3164 			    MLX5_FLOW_LAYER_GRE_KEY)))
3165 		return rte_flow_error_set(error, EINVAL,
3166 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3167 					  "protocol filtering not compatible"
3168 					  " with MPLS layer");
3169 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3170 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3171 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3172 		return rte_flow_error_set(error, ENOTSUP,
3173 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3174 					  "multiple tunnel layers not"
3175 					  " supported");
3176 	if (!mask)
3177 		mask = &rte_flow_item_mpls_mask;
3178 	ret = mlx5_flow_item_acceptable
3179 		(item, (const uint8_t *)mask,
3180 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3181 		 sizeof(struct rte_flow_item_mpls),
3182 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3183 	if (ret < 0)
3184 		return ret;
3185 	return 0;
3186 #else
3187 	return rte_flow_error_set(error, ENOTSUP,
3188 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3189 				  "MPLS is not supported by Verbs, please"
3190 				  " update.");
3191 #endif
3192 }
3193 
3194 /**
3195  * Validate NVGRE item.
3196  *
3197  * @param[in] item
3198  *   Item specification.
3199  * @param[in] item_flags
3200  *   Bit flags to mark detected items.
3201  * @param[in] target_protocol
3202  *   The next protocol in the previous item.
3203  * @param[out] error
3204  *   Pointer to error structure.
3205  *
3206  * @return
3207  *   0 on success, a negative errno value otherwise and rte_errno is set.
3208  */
3209 int
3210 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3211 			      uint64_t item_flags,
3212 			      uint8_t target_protocol,
3213 			      struct rte_flow_error *error)
3214 {
3215 	const struct rte_flow_item_nvgre *mask = item->mask;
3216 	int ret;
3217 
3218 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3219 		return rte_flow_error_set(error, EINVAL,
3220 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3221 					  "protocol filtering not compatible"
3222 					  " with this GRE layer");
3223 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3224 		return rte_flow_error_set(error, ENOTSUP,
3225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3226 					  "multiple tunnel layers not"
3227 					  " supported");
3228 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3229 		return rte_flow_error_set(error, ENOTSUP,
3230 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3231 					  "L3 Layer is missing");
3232 	if (!mask)
3233 		mask = &rte_flow_item_nvgre_mask;
3234 	ret = mlx5_flow_item_acceptable
3235 		(item, (const uint8_t *)mask,
3236 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3237 		 sizeof(struct rte_flow_item_nvgre),
3238 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3239 	if (ret < 0)
3240 		return ret;
3241 	return 0;
3242 }
3243 
3244 /**
3245  * Validate eCPRI item.
3246  *
3247  * @param[in] item
3248  *   Item specification.
3249  * @param[in] item_flags
3250  *   Bit-fields that holds the items detected until now.
3251  * @param[in] last_item
3252  *   Previous validated item in the pattern items.
3253  * @param[in] ether_type
3254  *   Type in the ethernet layer header (including dot1q).
3255  * @param[in] acc_mask
3256  *   Acceptable mask, if NULL default internal default mask
3257  *   will be used to check whether item fields are supported.
3258  * @param[out] error
3259  *   Pointer to error structure.
3260  *
3261  * @return
3262  *   0 on success, a negative errno value otherwise and rte_errno is set.
3263  */
3264 int
3265 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3266 			      uint64_t item_flags,
3267 			      uint64_t last_item,
3268 			      uint16_t ether_type,
3269 			      const struct rte_flow_item_ecpri *acc_mask,
3270 			      struct rte_flow_error *error)
3271 {
3272 	const struct rte_flow_item_ecpri *mask = item->mask;
3273 	const struct rte_flow_item_ecpri nic_mask = {
3274 		.hdr = {
3275 			.common = {
3276 				.u32 =
3277 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3278 					.type = 0xFF,
3279 					}).u32),
3280 			},
3281 			.dummy[0] = 0xFFFFFFFF,
3282 		},
3283 	};
3284 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3285 					MLX5_FLOW_LAYER_OUTER_VLAN);
3286 	struct rte_flow_item_ecpri mask_lo;
3287 
3288 	if (!(last_item & outer_l2_vlan) &&
3289 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3290 		return rte_flow_error_set(error, EINVAL,
3291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3292 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3293 	if ((last_item & outer_l2_vlan) && ether_type &&
3294 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3295 		return rte_flow_error_set(error, EINVAL,
3296 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3297 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3298 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3299 		return rte_flow_error_set(error, EINVAL,
3300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3301 					  "eCPRI with tunnel is not supported right now");
3302 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3303 		return rte_flow_error_set(error, ENOTSUP,
3304 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3305 					  "multiple L3 layers not supported");
3306 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3307 		return rte_flow_error_set(error, EINVAL,
3308 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3309 					  "eCPRI cannot coexist with a TCP layer");
3310 	/* In specification, eCPRI could be over UDP layer. */
3311 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3312 		return rte_flow_error_set(error, EINVAL,
3313 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3314 					  "eCPRI over UDP layer is not yet supported right now");
3315 	/* Mask for type field in common header could be zero. */
3316 	if (!mask)
3317 		mask = &rte_flow_item_ecpri_mask;
3318 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3319 	/* Input mask is in big-endian format. */
3320 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3321 		return rte_flow_error_set(error, EINVAL,
3322 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3323 					  "partial mask is not supported for protocol");
3324 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3325 		return rte_flow_error_set(error, EINVAL,
3326 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3327 					  "message header mask must be after a type mask");
3328 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3329 					 acc_mask ? (const uint8_t *)acc_mask
3330 						  : (const uint8_t *)&nic_mask,
3331 					 sizeof(struct rte_flow_item_ecpri),
3332 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3333 }
3334 
3335 static int
3336 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3337 		   const struct rte_flow_attr *attr __rte_unused,
3338 		   const struct rte_flow_item items[] __rte_unused,
3339 		   const struct rte_flow_action actions[] __rte_unused,
3340 		   bool external __rte_unused,
3341 		   int hairpin __rte_unused,
3342 		   struct rte_flow_error *error)
3343 {
3344 	return rte_flow_error_set(error, ENOTSUP,
3345 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3346 }
3347 
3348 static struct mlx5_flow *
3349 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
3350 		  const struct rte_flow_attr *attr __rte_unused,
3351 		  const struct rte_flow_item items[] __rte_unused,
3352 		  const struct rte_flow_action actions[] __rte_unused,
3353 		  struct rte_flow_error *error)
3354 {
3355 	rte_flow_error_set(error, ENOTSUP,
3356 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3357 	return NULL;
3358 }
3359 
3360 static int
3361 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3362 		    struct mlx5_flow *dev_flow __rte_unused,
3363 		    const struct rte_flow_attr *attr __rte_unused,
3364 		    const struct rte_flow_item items[] __rte_unused,
3365 		    const struct rte_flow_action actions[] __rte_unused,
3366 		    struct rte_flow_error *error)
3367 {
3368 	return rte_flow_error_set(error, ENOTSUP,
3369 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3370 }
3371 
3372 static int
3373 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3374 		struct rte_flow *flow __rte_unused,
3375 		struct rte_flow_error *error)
3376 {
3377 	return rte_flow_error_set(error, ENOTSUP,
3378 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3379 }
3380 
3381 static void
3382 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3383 		 struct rte_flow *flow __rte_unused)
3384 {
3385 }
3386 
3387 static void
3388 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3389 		  struct rte_flow *flow __rte_unused)
3390 {
3391 }
3392 
3393 static int
3394 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3395 		struct rte_flow *flow __rte_unused,
3396 		const struct rte_flow_action *actions __rte_unused,
3397 		void *data __rte_unused,
3398 		struct rte_flow_error *error)
3399 {
3400 	return rte_flow_error_set(error, ENOTSUP,
3401 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3402 }
3403 
3404 static int
3405 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3406 		      uint32_t domains __rte_unused,
3407 		      uint32_t flags __rte_unused)
3408 {
3409 	return 0;
3410 }
3411 
3412 /* Void driver to protect from null pointer reference. */
3413 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3414 	.validate = flow_null_validate,
3415 	.prepare = flow_null_prepare,
3416 	.translate = flow_null_translate,
3417 	.apply = flow_null_apply,
3418 	.remove = flow_null_remove,
3419 	.destroy = flow_null_destroy,
3420 	.query = flow_null_query,
3421 	.sync_domain = flow_null_sync_domain,
3422 };
3423 
3424 /**
3425  * Select flow driver type according to flow attributes and device
3426  * configuration.
3427  *
3428  * @param[in] dev
3429  *   Pointer to the dev structure.
3430  * @param[in] attr
3431  *   Pointer to the flow attributes.
3432  *
3433  * @return
3434  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3435  */
3436 static enum mlx5_flow_drv_type
3437 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3438 {
3439 	struct mlx5_priv *priv = dev->data->dev_private;
3440 	/* The OS can determine first a specific flow type (DV, VERBS) */
3441 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3442 
3443 	if (type != MLX5_FLOW_TYPE_MAX)
3444 		return type;
3445 	/*
3446 	 * Currently when dv_flow_en == 2, only HW steering engine is
3447 	 * supported. New engines can also be chosen here if ready.
3448 	 */
3449 	if (priv->sh->config.dv_flow_en == 2)
3450 		return MLX5_FLOW_TYPE_HW;
3451 	/* If no OS specific type - continue with DV/VERBS selection */
3452 	if (attr->transfer && priv->sh->config.dv_esw_en)
3453 		type = MLX5_FLOW_TYPE_DV;
3454 	if (!attr->transfer)
3455 		type = priv->sh->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3456 						     MLX5_FLOW_TYPE_VERBS;
3457 	return type;
3458 }
3459 
3460 #define flow_get_drv_ops(type) flow_drv_ops[type]
3461 
3462 /**
3463  * Flow driver validation API. This abstracts calling driver specific functions.
3464  * The type of flow driver is determined according to flow attributes.
3465  *
3466  * @param[in] dev
3467  *   Pointer to the dev structure.
3468  * @param[in] attr
3469  *   Pointer to the flow attributes.
3470  * @param[in] items
3471  *   Pointer to the list of items.
3472  * @param[in] actions
3473  *   Pointer to the list of actions.
3474  * @param[in] external
3475  *   This flow rule is created by request external to PMD.
3476  * @param[in] hairpin
3477  *   Number of hairpin TX actions, 0 means classic flow.
3478  * @param[out] error
3479  *   Pointer to the error structure.
3480  *
3481  * @return
3482  *   0 on success, a negative errno value otherwise and rte_errno is set.
3483  */
3484 static inline int
3485 flow_drv_validate(struct rte_eth_dev *dev,
3486 		  const struct rte_flow_attr *attr,
3487 		  const struct rte_flow_item items[],
3488 		  const struct rte_flow_action actions[],
3489 		  bool external, int hairpin, struct rte_flow_error *error)
3490 {
3491 	const struct mlx5_flow_driver_ops *fops;
3492 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3493 
3494 	fops = flow_get_drv_ops(type);
3495 	return fops->validate(dev, attr, items, actions, external,
3496 			      hairpin, error);
3497 }
3498 
3499 /**
3500  * Flow driver preparation API. This abstracts calling driver specific
3501  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3502  * calculates the size of memory required for device flow, allocates the memory,
3503  * initializes the device flow and returns the pointer.
3504  *
3505  * @note
3506  *   This function initializes device flow structure such as dv or verbs in
3507  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3508  *   rest. For example, adding returning device flow to flow->dev_flow list and
3509  *   setting backward reference to the flow should be done out of this function.
3510  *   layers field is not filled either.
3511  *
3512  * @param[in] dev
3513  *   Pointer to the dev structure.
3514  * @param[in] attr
3515  *   Pointer to the flow attributes.
3516  * @param[in] items
3517  *   Pointer to the list of items.
3518  * @param[in] actions
3519  *   Pointer to the list of actions.
3520  * @param[in] flow_idx
3521  *   This memory pool index to the flow.
3522  * @param[out] error
3523  *   Pointer to the error structure.
3524  *
3525  * @return
3526  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3527  */
3528 static inline struct mlx5_flow *
3529 flow_drv_prepare(struct rte_eth_dev *dev,
3530 		 const struct rte_flow *flow,
3531 		 const struct rte_flow_attr *attr,
3532 		 const struct rte_flow_item items[],
3533 		 const struct rte_flow_action actions[],
3534 		 uint32_t flow_idx,
3535 		 struct rte_flow_error *error)
3536 {
3537 	const struct mlx5_flow_driver_ops *fops;
3538 	enum mlx5_flow_drv_type type = flow->drv_type;
3539 	struct mlx5_flow *mlx5_flow = NULL;
3540 
3541 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3542 	fops = flow_get_drv_ops(type);
3543 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3544 	if (mlx5_flow)
3545 		mlx5_flow->flow_idx = flow_idx;
3546 	return mlx5_flow;
3547 }
3548 
3549 /**
3550  * Flow driver translation API. This abstracts calling driver specific
3551  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3552  * translates a generic flow into a driver flow. flow_drv_prepare() must
3553  * precede.
3554  *
3555  * @note
3556  *   dev_flow->layers could be filled as a result of parsing during translation
3557  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3558  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3559  *   flow->actions could be overwritten even though all the expanded dev_flows
3560  *   have the same actions.
3561  *
3562  * @param[in] dev
3563  *   Pointer to the rte dev structure.
3564  * @param[in, out] dev_flow
3565  *   Pointer to the mlx5 flow.
3566  * @param[in] attr
3567  *   Pointer to the flow attributes.
3568  * @param[in] items
3569  *   Pointer to the list of items.
3570  * @param[in] actions
3571  *   Pointer to the list of actions.
3572  * @param[out] error
3573  *   Pointer to the error structure.
3574  *
3575  * @return
3576  *   0 on success, a negative errno value otherwise and rte_errno is set.
3577  */
3578 static inline int
3579 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3580 		   const struct rte_flow_attr *attr,
3581 		   const struct rte_flow_item items[],
3582 		   const struct rte_flow_action actions[],
3583 		   struct rte_flow_error *error)
3584 {
3585 	const struct mlx5_flow_driver_ops *fops;
3586 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3587 
3588 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3589 	fops = flow_get_drv_ops(type);
3590 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3591 }
3592 
3593 /**
3594  * Flow driver apply API. This abstracts calling driver specific functions.
3595  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3596  * translated driver flows on to device. flow_drv_translate() must precede.
3597  *
3598  * @param[in] dev
3599  *   Pointer to Ethernet device structure.
3600  * @param[in, out] flow
3601  *   Pointer to flow structure.
3602  * @param[out] error
3603  *   Pointer to error structure.
3604  *
3605  * @return
3606  *   0 on success, a negative errno value otherwise and rte_errno is set.
3607  */
3608 static inline int
3609 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3610 	       struct rte_flow_error *error)
3611 {
3612 	const struct mlx5_flow_driver_ops *fops;
3613 	enum mlx5_flow_drv_type type = flow->drv_type;
3614 
3615 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3616 	fops = flow_get_drv_ops(type);
3617 	return fops->apply(dev, flow, error);
3618 }
3619 
3620 /**
3621  * Flow driver destroy API. This abstracts calling driver specific functions.
3622  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3623  * on device and releases resources of the flow.
3624  *
3625  * @param[in] dev
3626  *   Pointer to Ethernet device.
3627  * @param[in, out] flow
3628  *   Pointer to flow structure.
3629  */
3630 static inline void
3631 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3632 {
3633 	const struct mlx5_flow_driver_ops *fops;
3634 	enum mlx5_flow_drv_type type = flow->drv_type;
3635 
3636 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3637 	fops = flow_get_drv_ops(type);
3638 	fops->destroy(dev, flow);
3639 }
3640 
3641 /**
3642  * Flow driver find RSS policy tbl API. This abstracts calling driver
3643  * specific functions. Parent flow (rte_flow) should have driver
3644  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3645  *
3646  * @param[in] dev
3647  *   Pointer to Ethernet device.
3648  * @param[in, out] flow
3649  *   Pointer to flow structure.
3650  * @param[in] policy
3651  *   Pointer to meter policy table.
3652  * @param[in] rss_desc
3653  *   Pointer to rss_desc
3654  */
3655 static struct mlx5_flow_meter_sub_policy *
3656 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3657 		struct rte_flow *flow,
3658 		struct mlx5_flow_meter_policy *policy,
3659 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3660 {
3661 	const struct mlx5_flow_driver_ops *fops;
3662 	enum mlx5_flow_drv_type type = flow->drv_type;
3663 
3664 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3665 	fops = flow_get_drv_ops(type);
3666 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3667 }
3668 
3669 /**
3670  * Flow driver color tag rule API. This abstracts calling driver
3671  * specific functions. Parent flow (rte_flow) should have driver
3672  * type (drv_type). It will create the color tag rules in hierarchy meter.
3673  *
3674  * @param[in] dev
3675  *   Pointer to Ethernet device.
3676  * @param[in, out] flow
3677  *   Pointer to flow structure.
3678  * @param[in] fm
3679  *   Pointer to flow meter structure.
3680  * @param[in] src_port
3681  *   The src port this extra rule should use.
3682  * @param[in] item
3683  *   The src port id match item.
3684  * @param[out] error
3685  *   Pointer to error structure.
3686  */
3687 static int
3688 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3689 		struct rte_flow *flow,
3690 		struct mlx5_flow_meter_info *fm,
3691 		int32_t src_port,
3692 		const struct rte_flow_item *item,
3693 		struct rte_flow_error *error)
3694 {
3695 	const struct mlx5_flow_driver_ops *fops;
3696 	enum mlx5_flow_drv_type type = flow->drv_type;
3697 
3698 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3699 	fops = flow_get_drv_ops(type);
3700 	return fops->meter_hierarchy_rule_create(dev, fm,
3701 						src_port, item, error);
3702 }
3703 
3704 /**
3705  * Get RSS action from the action list.
3706  *
3707  * @param[in] dev
3708  *   Pointer to Ethernet device.
3709  * @param[in] actions
3710  *   Pointer to the list of actions.
3711  * @param[in] flow
3712  *   Parent flow structure pointer.
3713  *
3714  * @return
3715  *   Pointer to the RSS action if exist, else return NULL.
3716  */
3717 static const struct rte_flow_action_rss*
3718 flow_get_rss_action(struct rte_eth_dev *dev,
3719 		    const struct rte_flow_action actions[])
3720 {
3721 	struct mlx5_priv *priv = dev->data->dev_private;
3722 	const struct rte_flow_action_rss *rss = NULL;
3723 	struct mlx5_meter_policy_action_container *acg;
3724 	struct mlx5_meter_policy_action_container *acy;
3725 
3726 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3727 		switch (actions->type) {
3728 		case RTE_FLOW_ACTION_TYPE_RSS:
3729 			rss = actions->conf;
3730 			break;
3731 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3732 		{
3733 			const struct rte_flow_action_sample *sample =
3734 								actions->conf;
3735 			const struct rte_flow_action *act = sample->actions;
3736 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3737 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3738 					rss = act->conf;
3739 			break;
3740 		}
3741 		case RTE_FLOW_ACTION_TYPE_METER:
3742 		{
3743 			uint32_t mtr_idx;
3744 			struct mlx5_flow_meter_info *fm;
3745 			struct mlx5_flow_meter_policy *policy;
3746 			const struct rte_flow_action_meter *mtr = actions->conf;
3747 
3748 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3749 			if (fm && !fm->def_policy) {
3750 				policy = mlx5_flow_meter_policy_find(dev,
3751 						fm->policy_id, NULL);
3752 				MLX5_ASSERT(policy);
3753 				if (policy->is_hierarchy) {
3754 					policy =
3755 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3756 									policy);
3757 					if (!policy)
3758 						return NULL;
3759 				}
3760 				if (policy->is_rss) {
3761 					acg =
3762 					&policy->act_cnt[RTE_COLOR_GREEN];
3763 					acy =
3764 					&policy->act_cnt[RTE_COLOR_YELLOW];
3765 					if (acg->fate_action ==
3766 					    MLX5_FLOW_FATE_SHARED_RSS)
3767 						rss = acg->rss->conf;
3768 					else if (acy->fate_action ==
3769 						 MLX5_FLOW_FATE_SHARED_RSS)
3770 						rss = acy->rss->conf;
3771 				}
3772 			}
3773 			break;
3774 		}
3775 		default:
3776 			break;
3777 		}
3778 	}
3779 	return rss;
3780 }
3781 
3782 /**
3783  * Get ASO age action by index.
3784  *
3785  * @param[in] dev
3786  *   Pointer to the Ethernet device structure.
3787  * @param[in] age_idx
3788  *   Index to the ASO age action.
3789  *
3790  * @return
3791  *   The specified ASO age action.
3792  */
3793 struct mlx5_aso_age_action*
3794 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3795 {
3796 	uint16_t pool_idx = age_idx & UINT16_MAX;
3797 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3798 	struct mlx5_priv *priv = dev->data->dev_private;
3799 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3800 	struct mlx5_aso_age_pool *pool;
3801 
3802 	rte_rwlock_read_lock(&mng->resize_rwl);
3803 	pool = mng->pools[pool_idx];
3804 	rte_rwlock_read_unlock(&mng->resize_rwl);
3805 	return &pool->actions[offset - 1];
3806 }
3807 
3808 /* maps indirect action to translated direct in some actions array */
3809 struct mlx5_translated_action_handle {
3810 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3811 	int index; /**< Index in related array of rte_flow_action. */
3812 };
3813 
3814 /**
3815  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3816  * direct action if translation possible.
3817  * This functionality used to run same execution path for both direct and
3818  * indirect actions on flow create. All necessary preparations for indirect
3819  * action handling should be performed on *handle* actions list returned
3820  * from this call.
3821  *
3822  * @param[in] dev
3823  *   Pointer to Ethernet device.
3824  * @param[in] actions
3825  *   List of actions to translate.
3826  * @param[out] handle
3827  *   List to store translated indirect action object handles.
3828  * @param[in, out] indir_n
3829  *   Size of *handle* array. On return should be updated with number of
3830  *   indirect actions retrieved from the *actions* list.
3831  * @param[out] translated_actions
3832  *   List of actions where all indirect actions were translated to direct
3833  *   if possible. NULL if no translation took place.
3834  * @param[out] error
3835  *   Pointer to the error structure.
3836  *
3837  * @return
3838  *   0 on success, a negative errno value otherwise and rte_errno is set.
3839  */
3840 static int
3841 flow_action_handles_translate(struct rte_eth_dev *dev,
3842 			      const struct rte_flow_action actions[],
3843 			      struct mlx5_translated_action_handle *handle,
3844 			      int *indir_n,
3845 			      struct rte_flow_action **translated_actions,
3846 			      struct rte_flow_error *error)
3847 {
3848 	struct mlx5_priv *priv = dev->data->dev_private;
3849 	struct rte_flow_action *translated = NULL;
3850 	size_t actions_size;
3851 	int n;
3852 	int copied_n = 0;
3853 	struct mlx5_translated_action_handle *handle_end = NULL;
3854 
3855 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3856 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3857 			continue;
3858 		if (copied_n == *indir_n) {
3859 			return rte_flow_error_set
3860 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3861 				 NULL, "too many shared actions");
3862 		}
3863 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3864 			   sizeof(actions[n].conf));
3865 		handle[copied_n].index = n;
3866 		copied_n++;
3867 	}
3868 	n++;
3869 	*indir_n = copied_n;
3870 	if (!copied_n)
3871 		return 0;
3872 	actions_size = sizeof(struct rte_flow_action) * n;
3873 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3874 	if (!translated) {
3875 		rte_errno = ENOMEM;
3876 		return -ENOMEM;
3877 	}
3878 	memcpy(translated, actions, actions_size);
3879 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3880 		struct mlx5_shared_action_rss *shared_rss;
3881 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3882 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3883 		uint32_t idx = act_idx &
3884 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3885 
3886 		switch (type) {
3887 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3888 			shared_rss = mlx5_ipool_get
3889 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3890 			translated[handle->index].type =
3891 				RTE_FLOW_ACTION_TYPE_RSS;
3892 			translated[handle->index].conf =
3893 				&shared_rss->origin;
3894 			break;
3895 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3896 			translated[handle->index].type =
3897 						(enum rte_flow_action_type)
3898 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3899 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3900 			break;
3901 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3902 			if (priv->sh->flow_hit_aso_en) {
3903 				translated[handle->index].type =
3904 					(enum rte_flow_action_type)
3905 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3906 				translated[handle->index].conf =
3907 							 (void *)(uintptr_t)idx;
3908 				break;
3909 			}
3910 			/* Fall-through */
3911 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3912 			if (priv->sh->ct_aso_en) {
3913 				translated[handle->index].type =
3914 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3915 				translated[handle->index].conf =
3916 							 (void *)(uintptr_t)idx;
3917 				break;
3918 			}
3919 			/* Fall-through */
3920 		default:
3921 			mlx5_free(translated);
3922 			return rte_flow_error_set
3923 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3924 				 NULL, "invalid indirect action type");
3925 		}
3926 	}
3927 	*translated_actions = translated;
3928 	return 0;
3929 }
3930 
3931 /**
3932  * Get Shared RSS action from the action list.
3933  *
3934  * @param[in] dev
3935  *   Pointer to Ethernet device.
3936  * @param[in] shared
3937  *   Pointer to the list of actions.
3938  * @param[in] shared_n
3939  *   Actions list length.
3940  *
3941  * @return
3942  *   The MLX5 RSS action ID if exists, otherwise return 0.
3943  */
3944 static uint32_t
3945 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3946 			   struct mlx5_translated_action_handle *handle,
3947 			   int shared_n)
3948 {
3949 	struct mlx5_translated_action_handle *handle_end;
3950 	struct mlx5_priv *priv = dev->data->dev_private;
3951 	struct mlx5_shared_action_rss *shared_rss;
3952 
3953 
3954 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3955 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3956 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3957 		uint32_t idx = act_idx &
3958 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3959 		switch (type) {
3960 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3961 			shared_rss = mlx5_ipool_get
3962 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3963 									   idx);
3964 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3965 					   __ATOMIC_RELAXED);
3966 			return idx;
3967 		default:
3968 			break;
3969 		}
3970 	}
3971 	return 0;
3972 }
3973 
3974 static unsigned int
3975 find_graph_root(uint32_t rss_level)
3976 {
3977 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3978 			       MLX5_EXPANSION_ROOT_OUTER;
3979 }
3980 
3981 /**
3982  *  Get layer flags from the prefix flow.
3983  *
3984  *  Some flows may be split to several subflows, the prefix subflow gets the
3985  *  match items and the suffix sub flow gets the actions.
3986  *  Some actions need the user defined match item flags to get the detail for
3987  *  the action.
3988  *  This function helps the suffix flow to get the item layer flags from prefix
3989  *  subflow.
3990  *
3991  * @param[in] dev_flow
3992  *   Pointer the created prefix subflow.
3993  *
3994  * @return
3995  *   The layers get from prefix subflow.
3996  */
3997 static inline uint64_t
3998 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3999 {
4000 	uint64_t layers = 0;
4001 
4002 	/*
4003 	 * Layers bits could be localization, but usually the compiler will
4004 	 * help to do the optimization work for source code.
4005 	 * If no decap actions, use the layers directly.
4006 	 */
4007 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
4008 		return dev_flow->handle->layers;
4009 	/* Convert L3 layers with decap action. */
4010 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
4011 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
4012 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
4013 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
4014 	/* Convert L4 layers with decap action.  */
4015 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
4016 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
4017 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
4018 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
4019 	return layers;
4020 }
4021 
4022 /**
4023  * Get metadata split action information.
4024  *
4025  * @param[in] actions
4026  *   Pointer to the list of actions.
4027  * @param[out] qrss
4028  *   Pointer to the return pointer.
4029  * @param[out] qrss_type
4030  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
4031  *   if no QUEUE/RSS is found.
4032  * @param[out] encap_idx
4033  *   Pointer to the index of the encap action if exists, otherwise the last
4034  *   action index.
4035  *
4036  * @return
4037  *   Total number of actions.
4038  */
4039 static int
4040 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4041 				       const struct rte_flow_action **qrss,
4042 				       int *encap_idx)
4043 {
4044 	const struct rte_flow_action_raw_encap *raw_encap;
4045 	int actions_n = 0;
4046 	int raw_decap_idx = -1;
4047 
4048 	*encap_idx = -1;
4049 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4050 		switch (actions->type) {
4051 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4052 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4053 			*encap_idx = actions_n;
4054 			break;
4055 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4056 			raw_decap_idx = actions_n;
4057 			break;
4058 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4059 			raw_encap = actions->conf;
4060 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4061 				*encap_idx = raw_decap_idx != -1 ?
4062 						      raw_decap_idx : actions_n;
4063 			break;
4064 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4065 		case RTE_FLOW_ACTION_TYPE_RSS:
4066 			*qrss = actions;
4067 			break;
4068 		default:
4069 			break;
4070 		}
4071 		actions_n++;
4072 	}
4073 	if (*encap_idx == -1)
4074 		*encap_idx = actions_n;
4075 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4076 	return actions_n + 1;
4077 }
4078 
4079 /**
4080  * Check if the action will change packet.
4081  *
4082  * @param dev
4083  *   Pointer to Ethernet device.
4084  * @param[in] type
4085  *   action type.
4086  *
4087  * @return
4088  *   true if action will change packet, false otherwise.
4089  */
4090 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4091 					  enum rte_flow_action_type type)
4092 {
4093 	struct mlx5_priv *priv = dev->data->dev_private;
4094 
4095 	switch (type) {
4096 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4097 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4098 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4099 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4100 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4101 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4102 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4103 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4104 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4105 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4106 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4107 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4108 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4109 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4110 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4111 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4112 	case RTE_FLOW_ACTION_TYPE_SET_META:
4113 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4114 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4115 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4116 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4117 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4118 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4119 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4120 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4121 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4122 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4123 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4124 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4125 		return true;
4126 	case RTE_FLOW_ACTION_TYPE_FLAG:
4127 	case RTE_FLOW_ACTION_TYPE_MARK:
4128 		if (priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4129 			return true;
4130 		else
4131 			return false;
4132 	default:
4133 		return false;
4134 	}
4135 }
4136 
4137 /**
4138  * Check meter action from the action list.
4139  *
4140  * @param dev
4141  *   Pointer to Ethernet device.
4142  * @param[in] actions
4143  *   Pointer to the list of actions.
4144  * @param[out] has_mtr
4145  *   Pointer to the meter exist flag.
4146  * @param[out] has_modify
4147  *   Pointer to the flag showing there's packet change action.
4148  * @param[out] meter_id
4149  *   Pointer to the meter id.
4150  *
4151  * @return
4152  *   Total number of actions.
4153  */
4154 static int
4155 flow_check_meter_action(struct rte_eth_dev *dev,
4156 			const struct rte_flow_action actions[],
4157 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4158 {
4159 	const struct rte_flow_action_meter *mtr = NULL;
4160 	int actions_n = 0;
4161 
4162 	MLX5_ASSERT(has_mtr);
4163 	*has_mtr = false;
4164 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4165 		switch (actions->type) {
4166 		case RTE_FLOW_ACTION_TYPE_METER:
4167 			mtr = actions->conf;
4168 			*meter_id = mtr->mtr_id;
4169 			*has_mtr = true;
4170 			break;
4171 		default:
4172 			break;
4173 		}
4174 		if (!*has_mtr)
4175 			*has_modify |= flow_check_modify_action_type(dev,
4176 								actions->type);
4177 		actions_n++;
4178 	}
4179 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4180 	return actions_n + 1;
4181 }
4182 
4183 /**
4184  * Check if the flow should be split due to hairpin.
4185  * The reason for the split is that in current HW we can't
4186  * support encap and push-vlan on Rx, so if a flow contains
4187  * these actions we move it to Tx.
4188  *
4189  * @param dev
4190  *   Pointer to Ethernet device.
4191  * @param[in] attr
4192  *   Flow rule attributes.
4193  * @param[in] actions
4194  *   Associated actions (list terminated by the END action).
4195  *
4196  * @return
4197  *   > 0 the number of actions and the flow should be split,
4198  *   0 when no split required.
4199  */
4200 static int
4201 flow_check_hairpin_split(struct rte_eth_dev *dev,
4202 			 const struct rte_flow_attr *attr,
4203 			 const struct rte_flow_action actions[])
4204 {
4205 	int queue_action = 0;
4206 	int action_n = 0;
4207 	int split = 0;
4208 	const struct rte_flow_action_queue *queue;
4209 	const struct rte_flow_action_rss *rss;
4210 	const struct rte_flow_action_raw_encap *raw_encap;
4211 	const struct rte_eth_hairpin_conf *conf;
4212 
4213 	if (!attr->ingress)
4214 		return 0;
4215 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4216 		switch (actions->type) {
4217 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4218 			queue = actions->conf;
4219 			if (queue == NULL)
4220 				return 0;
4221 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4222 			if (conf == NULL || conf->tx_explicit != 0)
4223 				return 0;
4224 			queue_action = 1;
4225 			action_n++;
4226 			break;
4227 		case RTE_FLOW_ACTION_TYPE_RSS:
4228 			rss = actions->conf;
4229 			if (rss == NULL || rss->queue_num == 0)
4230 				return 0;
4231 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4232 			if (conf == NULL || conf->tx_explicit != 0)
4233 				return 0;
4234 			queue_action = 1;
4235 			action_n++;
4236 			break;
4237 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4238 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4239 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4240 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4241 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4242 			split++;
4243 			action_n++;
4244 			break;
4245 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4246 			raw_encap = actions->conf;
4247 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4248 				split++;
4249 			action_n++;
4250 			break;
4251 		default:
4252 			action_n++;
4253 			break;
4254 		}
4255 	}
4256 	if (split && queue_action)
4257 		return action_n;
4258 	return 0;
4259 }
4260 
4261 /* Declare flow create/destroy prototype in advance. */
4262 static uint32_t
4263 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4264 		 const struct rte_flow_attr *attr,
4265 		 const struct rte_flow_item items[],
4266 		 const struct rte_flow_action actions[],
4267 		 bool external, struct rte_flow_error *error);
4268 
4269 static void
4270 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4271 		  uint32_t flow_idx);
4272 
4273 int
4274 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4275 		      struct mlx5_list_entry *entry, void *cb_ctx)
4276 {
4277 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4278 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4279 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4280 
4281 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4282 }
4283 
4284 struct mlx5_list_entry *
4285 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4286 {
4287 	struct rte_eth_dev *dev = tool_ctx;
4288 	struct mlx5_priv *priv = dev->data->dev_private;
4289 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4290 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4291 	struct rte_flow_error *error = ctx->error;
4292 	uint32_t idx = 0;
4293 	int ret;
4294 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4295 	struct rte_flow_attr attr = {
4296 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4297 		.ingress = 1,
4298 	};
4299 	struct mlx5_rte_flow_item_tag tag_spec = {
4300 		.data = mark_id,
4301 	};
4302 	struct rte_flow_item items[] = {
4303 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4304 	};
4305 	struct rte_flow_action_mark ftag = {
4306 		.id = mark_id,
4307 	};
4308 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4309 		.dst = REG_B,
4310 		.src = REG_NON,
4311 	};
4312 	struct rte_flow_action_jump jump = {
4313 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4314 	};
4315 	struct rte_flow_action actions[] = {
4316 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4317 	};
4318 
4319 	/* Fill the register fields in the flow. */
4320 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4321 	if (ret < 0)
4322 		return NULL;
4323 	tag_spec.id = ret;
4324 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4325 	if (ret < 0)
4326 		return NULL;
4327 	cp_mreg.src = ret;
4328 	/* Provide the full width of FLAG specific value. */
4329 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4330 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4331 	/* Build a new flow. */
4332 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4333 		items[0] = (struct rte_flow_item){
4334 			.type = (enum rte_flow_item_type)
4335 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4336 			.spec = &tag_spec,
4337 		};
4338 		items[1] = (struct rte_flow_item){
4339 			.type = RTE_FLOW_ITEM_TYPE_END,
4340 		};
4341 		actions[0] = (struct rte_flow_action){
4342 			.type = (enum rte_flow_action_type)
4343 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4344 			.conf = &ftag,
4345 		};
4346 		actions[1] = (struct rte_flow_action){
4347 			.type = (enum rte_flow_action_type)
4348 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4349 			.conf = &cp_mreg,
4350 		};
4351 		actions[2] = (struct rte_flow_action){
4352 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4353 			.conf = &jump,
4354 		};
4355 		actions[3] = (struct rte_flow_action){
4356 			.type = RTE_FLOW_ACTION_TYPE_END,
4357 		};
4358 	} else {
4359 		/* Default rule, wildcard match. */
4360 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4361 		items[0] = (struct rte_flow_item){
4362 			.type = RTE_FLOW_ITEM_TYPE_END,
4363 		};
4364 		actions[0] = (struct rte_flow_action){
4365 			.type = (enum rte_flow_action_type)
4366 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4367 			.conf = &cp_mreg,
4368 		};
4369 		actions[1] = (struct rte_flow_action){
4370 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4371 			.conf = &jump,
4372 		};
4373 		actions[2] = (struct rte_flow_action){
4374 			.type = RTE_FLOW_ACTION_TYPE_END,
4375 		};
4376 	}
4377 	/* Build a new entry. */
4378 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4379 	if (!mcp_res) {
4380 		rte_errno = ENOMEM;
4381 		return NULL;
4382 	}
4383 	mcp_res->idx = idx;
4384 	mcp_res->mark_id = mark_id;
4385 	/*
4386 	 * The copy Flows are not included in any list. There
4387 	 * ones are referenced from other Flows and can not
4388 	 * be applied, removed, deleted in arbitrary order
4389 	 * by list traversing.
4390 	 */
4391 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4392 					&attr, items, actions, false, error);
4393 	if (!mcp_res->rix_flow) {
4394 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4395 		return NULL;
4396 	}
4397 	return &mcp_res->hlist_ent;
4398 }
4399 
4400 struct mlx5_list_entry *
4401 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4402 		      void *cb_ctx __rte_unused)
4403 {
4404 	struct rte_eth_dev *dev = tool_ctx;
4405 	struct mlx5_priv *priv = dev->data->dev_private;
4406 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4407 	uint32_t idx = 0;
4408 
4409 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4410 	if (!mcp_res) {
4411 		rte_errno = ENOMEM;
4412 		return NULL;
4413 	}
4414 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4415 	mcp_res->idx = idx;
4416 	return &mcp_res->hlist_ent;
4417 }
4418 
4419 void
4420 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4421 {
4422 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4423 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4424 	struct rte_eth_dev *dev = tool_ctx;
4425 	struct mlx5_priv *priv = dev->data->dev_private;
4426 
4427 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4428 }
4429 
4430 /**
4431  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4432  *
4433  * As mark_id is unique, if there's already a registered flow for the mark_id,
4434  * return by increasing the reference counter of the resource. Otherwise, create
4435  * the resource (mcp_res) and flow.
4436  *
4437  * Flow looks like,
4438  *   - If ingress port is ANY and reg_c[1] is mark_id,
4439  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4440  *
4441  * For default flow (zero mark_id), flow is like,
4442  *   - If ingress port is ANY,
4443  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4444  *
4445  * @param dev
4446  *   Pointer to Ethernet device.
4447  * @param mark_id
4448  *   ID of MARK action, zero means default flow for META.
4449  * @param[out] error
4450  *   Perform verbose error reporting if not NULL.
4451  *
4452  * @return
4453  *   Associated resource on success, NULL otherwise and rte_errno is set.
4454  */
4455 static struct mlx5_flow_mreg_copy_resource *
4456 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4457 			  struct rte_flow_error *error)
4458 {
4459 	struct mlx5_priv *priv = dev->data->dev_private;
4460 	struct mlx5_list_entry *entry;
4461 	struct mlx5_flow_cb_ctx ctx = {
4462 		.dev = dev,
4463 		.error = error,
4464 		.data = &mark_id,
4465 	};
4466 
4467 	/* Check if already registered. */
4468 	MLX5_ASSERT(priv->mreg_cp_tbl);
4469 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4470 	if (!entry)
4471 		return NULL;
4472 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4473 			    hlist_ent);
4474 }
4475 
4476 void
4477 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4478 {
4479 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4480 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4481 	struct rte_eth_dev *dev = tool_ctx;
4482 	struct mlx5_priv *priv = dev->data->dev_private;
4483 
4484 	MLX5_ASSERT(mcp_res->rix_flow);
4485 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4486 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4487 }
4488 
4489 /**
4490  * Release flow in RX_CP_TBL.
4491  *
4492  * @param dev
4493  *   Pointer to Ethernet device.
4494  * @flow
4495  *   Parent flow for wich copying is provided.
4496  */
4497 static void
4498 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4499 			  struct rte_flow *flow)
4500 {
4501 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4502 	struct mlx5_priv *priv = dev->data->dev_private;
4503 
4504 	if (!flow->rix_mreg_copy)
4505 		return;
4506 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4507 				 flow->rix_mreg_copy);
4508 	if (!mcp_res || !priv->mreg_cp_tbl)
4509 		return;
4510 	MLX5_ASSERT(mcp_res->rix_flow);
4511 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4512 	flow->rix_mreg_copy = 0;
4513 }
4514 
4515 /**
4516  * Remove the default copy action from RX_CP_TBL.
4517  *
4518  * This functions is called in the mlx5_dev_start(). No thread safe
4519  * is guaranteed.
4520  *
4521  * @param dev
4522  *   Pointer to Ethernet device.
4523  */
4524 static void
4525 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4526 {
4527 	struct mlx5_list_entry *entry;
4528 	struct mlx5_priv *priv = dev->data->dev_private;
4529 	struct mlx5_flow_cb_ctx ctx;
4530 	uint32_t mark_id;
4531 
4532 	/* Check if default flow is registered. */
4533 	if (!priv->mreg_cp_tbl)
4534 		return;
4535 	mark_id = MLX5_DEFAULT_COPY_ID;
4536 	ctx.data = &mark_id;
4537 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4538 	if (!entry)
4539 		return;
4540 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4541 }
4542 
4543 /**
4544  * Add the default copy action in in RX_CP_TBL.
4545  *
4546  * This functions is called in the mlx5_dev_start(). No thread safe
4547  * is guaranteed.
4548  *
4549  * @param dev
4550  *   Pointer to Ethernet device.
4551  * @param[out] error
4552  *   Perform verbose error reporting if not NULL.
4553  *
4554  * @return
4555  *   0 for success, negative value otherwise and rte_errno is set.
4556  */
4557 static int
4558 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4559 				  struct rte_flow_error *error)
4560 {
4561 	struct mlx5_priv *priv = dev->data->dev_private;
4562 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4563 	struct mlx5_flow_cb_ctx ctx;
4564 	uint32_t mark_id;
4565 
4566 	/* Check whether extensive metadata feature is engaged. */
4567 	if (!priv->sh->config.dv_flow_en ||
4568 	    priv->sh->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4569 	    !mlx5_flow_ext_mreg_supported(dev) ||
4570 	    !priv->sh->dv_regc0_mask)
4571 		return 0;
4572 	/*
4573 	 * Add default mreg copy flow may be called multiple time, but
4574 	 * only be called once in stop. Avoid register it twice.
4575 	 */
4576 	mark_id = MLX5_DEFAULT_COPY_ID;
4577 	ctx.data = &mark_id;
4578 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4579 		return 0;
4580 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4581 	if (!mcp_res)
4582 		return -rte_errno;
4583 	return 0;
4584 }
4585 
4586 /**
4587  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4588  *
4589  * All the flow having Q/RSS action should be split by
4590  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4591  * performs the following,
4592  *   - CQE->flow_tag := reg_c[1] (MARK)
4593  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4594  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4595  * but there should be a flow per each MARK ID set by MARK action.
4596  *
4597  * For the aforementioned reason, if there's a MARK action in flow's action
4598  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4599  * the MARK ID to CQE's flow_tag like,
4600  *   - If reg_c[1] is mark_id,
4601  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4602  *
4603  * For SET_META action which stores value in reg_c[0], as the destination is
4604  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4605  * MARK ID means the default flow. The default flow looks like,
4606  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4607  *
4608  * @param dev
4609  *   Pointer to Ethernet device.
4610  * @param flow
4611  *   Pointer to flow structure.
4612  * @param[in] actions
4613  *   Pointer to the list of actions.
4614  * @param[out] error
4615  *   Perform verbose error reporting if not NULL.
4616  *
4617  * @return
4618  *   0 on success, negative value otherwise and rte_errno is set.
4619  */
4620 static int
4621 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4622 			    struct rte_flow *flow,
4623 			    const struct rte_flow_action *actions,
4624 			    struct rte_flow_error *error)
4625 {
4626 	struct mlx5_priv *priv = dev->data->dev_private;
4627 	struct mlx5_sh_config *config = &priv->sh->config;
4628 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4629 	const struct rte_flow_action_mark *mark;
4630 
4631 	/* Check whether extensive metadata feature is engaged. */
4632 	if (!config->dv_flow_en ||
4633 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4634 	    !mlx5_flow_ext_mreg_supported(dev) ||
4635 	    !priv->sh->dv_regc0_mask)
4636 		return 0;
4637 	/* Find MARK action. */
4638 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4639 		switch (actions->type) {
4640 		case RTE_FLOW_ACTION_TYPE_FLAG:
4641 			mcp_res = flow_mreg_add_copy_action
4642 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4643 			if (!mcp_res)
4644 				return -rte_errno;
4645 			flow->rix_mreg_copy = mcp_res->idx;
4646 			return 0;
4647 		case RTE_FLOW_ACTION_TYPE_MARK:
4648 			mark = (const struct rte_flow_action_mark *)
4649 				actions->conf;
4650 			mcp_res =
4651 				flow_mreg_add_copy_action(dev, mark->id, error);
4652 			if (!mcp_res)
4653 				return -rte_errno;
4654 			flow->rix_mreg_copy = mcp_res->idx;
4655 			return 0;
4656 		default:
4657 			break;
4658 		}
4659 	}
4660 	return 0;
4661 }
4662 
4663 #define MLX5_MAX_SPLIT_ACTIONS 24
4664 #define MLX5_MAX_SPLIT_ITEMS 24
4665 
4666 /**
4667  * Split the hairpin flow.
4668  * Since HW can't support encap and push-vlan on Rx, we move these
4669  * actions to Tx.
4670  * If the count action is after the encap then we also
4671  * move the count action. in this case the count will also measure
4672  * the outer bytes.
4673  *
4674  * @param dev
4675  *   Pointer to Ethernet device.
4676  * @param[in] actions
4677  *   Associated actions (list terminated by the END action).
4678  * @param[out] actions_rx
4679  *   Rx flow actions.
4680  * @param[out] actions_tx
4681  *   Tx flow actions..
4682  * @param[out] pattern_tx
4683  *   The pattern items for the Tx flow.
4684  * @param[out] flow_id
4685  *   The flow ID connected to this flow.
4686  *
4687  * @return
4688  *   0 on success.
4689  */
4690 static int
4691 flow_hairpin_split(struct rte_eth_dev *dev,
4692 		   const struct rte_flow_action actions[],
4693 		   struct rte_flow_action actions_rx[],
4694 		   struct rte_flow_action actions_tx[],
4695 		   struct rte_flow_item pattern_tx[],
4696 		   uint32_t flow_id)
4697 {
4698 	const struct rte_flow_action_raw_encap *raw_encap;
4699 	const struct rte_flow_action_raw_decap *raw_decap;
4700 	struct mlx5_rte_flow_action_set_tag *set_tag;
4701 	struct rte_flow_action *tag_action;
4702 	struct mlx5_rte_flow_item_tag *tag_item;
4703 	struct rte_flow_item *item;
4704 	char *addr;
4705 	int encap = 0;
4706 
4707 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4708 		switch (actions->type) {
4709 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4710 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4711 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4712 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4713 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4714 			rte_memcpy(actions_tx, actions,
4715 			       sizeof(struct rte_flow_action));
4716 			actions_tx++;
4717 			break;
4718 		case RTE_FLOW_ACTION_TYPE_COUNT:
4719 			if (encap) {
4720 				rte_memcpy(actions_tx, actions,
4721 					   sizeof(struct rte_flow_action));
4722 				actions_tx++;
4723 			} else {
4724 				rte_memcpy(actions_rx, actions,
4725 					   sizeof(struct rte_flow_action));
4726 				actions_rx++;
4727 			}
4728 			break;
4729 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4730 			raw_encap = actions->conf;
4731 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4732 				memcpy(actions_tx, actions,
4733 				       sizeof(struct rte_flow_action));
4734 				actions_tx++;
4735 				encap = 1;
4736 			} else {
4737 				rte_memcpy(actions_rx, actions,
4738 					   sizeof(struct rte_flow_action));
4739 				actions_rx++;
4740 			}
4741 			break;
4742 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4743 			raw_decap = actions->conf;
4744 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4745 				memcpy(actions_tx, actions,
4746 				       sizeof(struct rte_flow_action));
4747 				actions_tx++;
4748 			} else {
4749 				rte_memcpy(actions_rx, actions,
4750 					   sizeof(struct rte_flow_action));
4751 				actions_rx++;
4752 			}
4753 			break;
4754 		default:
4755 			rte_memcpy(actions_rx, actions,
4756 				   sizeof(struct rte_flow_action));
4757 			actions_rx++;
4758 			break;
4759 		}
4760 	}
4761 	/* Add set meta action and end action for the Rx flow. */
4762 	tag_action = actions_rx;
4763 	tag_action->type = (enum rte_flow_action_type)
4764 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4765 	actions_rx++;
4766 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4767 	actions_rx++;
4768 	set_tag = (void *)actions_rx;
4769 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4770 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4771 		.data = flow_id,
4772 	};
4773 	MLX5_ASSERT(set_tag->id > REG_NON);
4774 	tag_action->conf = set_tag;
4775 	/* Create Tx item list. */
4776 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4777 	addr = (void *)&pattern_tx[2];
4778 	item = pattern_tx;
4779 	item->type = (enum rte_flow_item_type)
4780 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4781 	tag_item = (void *)addr;
4782 	tag_item->data = flow_id;
4783 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4784 	MLX5_ASSERT(set_tag->id > REG_NON);
4785 	item->spec = tag_item;
4786 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4787 	tag_item = (void *)addr;
4788 	tag_item->data = UINT32_MAX;
4789 	tag_item->id = UINT16_MAX;
4790 	item->mask = tag_item;
4791 	item->last = NULL;
4792 	item++;
4793 	item->type = RTE_FLOW_ITEM_TYPE_END;
4794 	return 0;
4795 }
4796 
4797 /**
4798  * The last stage of splitting chain, just creates the subflow
4799  * without any modification.
4800  *
4801  * @param[in] dev
4802  *   Pointer to Ethernet device.
4803  * @param[in] flow
4804  *   Parent flow structure pointer.
4805  * @param[in, out] sub_flow
4806  *   Pointer to return the created subflow, may be NULL.
4807  * @param[in] attr
4808  *   Flow rule attributes.
4809  * @param[in] items
4810  *   Pattern specification (list terminated by the END pattern item).
4811  * @param[in] actions
4812  *   Associated actions (list terminated by the END action).
4813  * @param[in] flow_split_info
4814  *   Pointer to flow split info structure.
4815  * @param[out] error
4816  *   Perform verbose error reporting if not NULL.
4817  * @return
4818  *   0 on success, negative value otherwise
4819  */
4820 static int
4821 flow_create_split_inner(struct rte_eth_dev *dev,
4822 			struct rte_flow *flow,
4823 			struct mlx5_flow **sub_flow,
4824 			const struct rte_flow_attr *attr,
4825 			const struct rte_flow_item items[],
4826 			const struct rte_flow_action actions[],
4827 			struct mlx5_flow_split_info *flow_split_info,
4828 			struct rte_flow_error *error)
4829 {
4830 	struct mlx5_flow *dev_flow;
4831 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
4832 
4833 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4834 				    flow_split_info->flow_idx, error);
4835 	if (!dev_flow)
4836 		return -rte_errno;
4837 	dev_flow->flow = flow;
4838 	dev_flow->external = flow_split_info->external;
4839 	dev_flow->skip_scale = flow_split_info->skip_scale;
4840 	/* Subflow object was created, we must include one in the list. */
4841 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4842 		      dev_flow->handle, next);
4843 	/*
4844 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4845 	 * flow may need some user defined item layer flags, and pass the
4846 	 * Metadata rxq mark flag to suffix flow as well.
4847 	 */
4848 	if (flow_split_info->prefix_layers)
4849 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4850 	if (flow_split_info->prefix_mark) {
4851 		MLX5_ASSERT(wks);
4852 		wks->mark = 1;
4853 	}
4854 	if (sub_flow)
4855 		*sub_flow = dev_flow;
4856 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4857 	dev_flow->dv.table_id = flow_split_info->table_id;
4858 #endif
4859 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4860 }
4861 
4862 /**
4863  * Get the sub policy of a meter.
4864  *
4865  * @param[in] dev
4866  *   Pointer to Ethernet device.
4867  * @param[in] flow
4868  *   Parent flow structure pointer.
4869  * @param wks
4870  *   Pointer to thread flow work space.
4871  * @param[in] attr
4872  *   Flow rule attributes.
4873  * @param[in] items
4874  *   Pattern specification (list terminated by the END pattern item).
4875  * @param[out] error
4876  *   Perform verbose error reporting if not NULL.
4877  *
4878  * @return
4879  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4880  */
4881 static struct mlx5_flow_meter_sub_policy *
4882 get_meter_sub_policy(struct rte_eth_dev *dev,
4883 		     struct rte_flow *flow,
4884 		     struct mlx5_flow_workspace *wks,
4885 		     const struct rte_flow_attr *attr,
4886 		     const struct rte_flow_item items[],
4887 		     struct rte_flow_error *error)
4888 {
4889 	struct mlx5_flow_meter_policy *policy;
4890 	struct mlx5_flow_meter_policy *final_policy;
4891 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4892 
4893 	policy = wks->policy;
4894 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4895 	if (final_policy->is_rss || final_policy->is_queue) {
4896 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4897 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4898 		uint32_t i;
4899 
4900 		/*
4901 		 * This is a tmp dev_flow,
4902 		 * no need to register any matcher for it in translate.
4903 		 */
4904 		wks->skip_matcher_reg = 1;
4905 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4906 			struct mlx5_flow dev_flow = {0};
4907 			struct mlx5_flow_handle dev_handle = { {0} };
4908 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4909 
4910 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4911 				const struct rte_flow_action_rss *rss_act =
4912 					final_policy->act_cnt[i].rss->conf;
4913 				struct rte_flow_action rss_actions[2] = {
4914 					[0] = {
4915 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4916 					.conf = rss_act,
4917 					},
4918 					[1] = {
4919 					.type = RTE_FLOW_ACTION_TYPE_END,
4920 					.conf = NULL,
4921 					}
4922 				};
4923 
4924 				dev_flow.handle = &dev_handle;
4925 				dev_flow.ingress = attr->ingress;
4926 				dev_flow.flow = flow;
4927 				dev_flow.external = 0;
4928 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4929 				dev_flow.dv.transfer = attr->transfer;
4930 #endif
4931 				/**
4932 				 * Translate RSS action to get rss hash fields.
4933 				 */
4934 				if (flow_drv_translate(dev, &dev_flow, attr,
4935 						items, rss_actions, error))
4936 					goto exit;
4937 				rss_desc_v[i] = wks->rss_desc;
4938 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4939 				rss_desc_v[i].hash_fields =
4940 						dev_flow.hash_fields;
4941 				rss_desc_v[i].queue_num =
4942 						rss_desc_v[i].hash_fields ?
4943 						rss_desc_v[i].queue_num : 1;
4944 				rss_desc_v[i].tunnel =
4945 						!!(dev_flow.handle->layers &
4946 						   MLX5_FLOW_LAYER_TUNNEL);
4947 				/* Use the RSS queues in the containers. */
4948 				rss_desc_v[i].queue =
4949 					(uint16_t *)(uintptr_t)rss_act->queue;
4950 				rss_desc[i] = &rss_desc_v[i];
4951 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4952 				/* This is queue action. */
4953 				rss_desc_v[i] = wks->rss_desc;
4954 				rss_desc_v[i].key_len = 0;
4955 				rss_desc_v[i].hash_fields = 0;
4956 				rss_desc_v[i].queue =
4957 					&final_policy->act_cnt[i].queue;
4958 				rss_desc_v[i].queue_num = 1;
4959 				rss_desc[i] = &rss_desc_v[i];
4960 			} else {
4961 				rss_desc[i] = NULL;
4962 			}
4963 		}
4964 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4965 						flow, policy, rss_desc);
4966 	} else {
4967 		enum mlx5_meter_domain mtr_domain =
4968 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4969 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4970 						MLX5_MTR_DOMAIN_INGRESS);
4971 		sub_policy = policy->sub_policys[mtr_domain][0];
4972 	}
4973 	if (!sub_policy)
4974 		rte_flow_error_set(error, EINVAL,
4975 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4976 				   "Failed to get meter sub-policy.");
4977 exit:
4978 	return sub_policy;
4979 }
4980 
4981 /**
4982  * Split the meter flow.
4983  *
4984  * As meter flow will split to three sub flow, other than meter
4985  * action, the other actions make sense to only meter accepts
4986  * the packet. If it need to be dropped, no other additional
4987  * actions should be take.
4988  *
4989  * One kind of special action which decapsulates the L3 tunnel
4990  * header will be in the prefix sub flow, as not to take the
4991  * L3 tunnel header into account.
4992  *
4993  * @param[in] dev
4994  *   Pointer to Ethernet device.
4995  * @param[in] flow
4996  *   Parent flow structure pointer.
4997  * @param wks
4998  *   Pointer to thread flow work space.
4999  * @param[in] attr
5000  *   Flow rule attributes.
5001  * @param[in] items
5002  *   Pattern specification (list terminated by the END pattern item).
5003  * @param[out] sfx_items
5004  *   Suffix flow match items (list terminated by the END pattern item).
5005  * @param[in] actions
5006  *   Associated actions (list terminated by the END action).
5007  * @param[out] actions_sfx
5008  *   Suffix flow actions.
5009  * @param[out] actions_pre
5010  *   Prefix flow actions.
5011  * @param[out] mtr_flow_id
5012  *   Pointer to meter flow id.
5013  * @param[out] error
5014  *   Perform verbose error reporting if not NULL.
5015  *
5016  * @return
5017  *   0 on success, a negative errno value otherwise and rte_errno is set.
5018  */
5019 static int
5020 flow_meter_split_prep(struct rte_eth_dev *dev,
5021 		      struct rte_flow *flow,
5022 		      struct mlx5_flow_workspace *wks,
5023 		      const struct rte_flow_attr *attr,
5024 		      const struct rte_flow_item items[],
5025 		      struct rte_flow_item sfx_items[],
5026 		      const struct rte_flow_action actions[],
5027 		      struct rte_flow_action actions_sfx[],
5028 		      struct rte_flow_action actions_pre[],
5029 		      uint32_t *mtr_flow_id,
5030 		      struct rte_flow_error *error)
5031 {
5032 	struct mlx5_priv *priv = dev->data->dev_private;
5033 	struct mlx5_flow_meter_info *fm = wks->fm;
5034 	struct rte_flow_action *tag_action = NULL;
5035 	struct rte_flow_item *tag_item;
5036 	struct mlx5_rte_flow_action_set_tag *set_tag;
5037 	const struct rte_flow_action_raw_encap *raw_encap;
5038 	const struct rte_flow_action_raw_decap *raw_decap;
5039 	struct mlx5_rte_flow_item_tag *tag_item_spec;
5040 	struct mlx5_rte_flow_item_tag *tag_item_mask;
5041 	uint32_t tag_id = 0;
5042 	struct rte_flow_item *vlan_item_dst = NULL;
5043 	const struct rte_flow_item *vlan_item_src = NULL;
5044 	const struct rte_flow_item *orig_items = items;
5045 	struct rte_flow_action *hw_mtr_action;
5046 	struct rte_flow_action *action_pre_head = NULL;
5047 	int32_t flow_src_port = priv->representor_id;
5048 	bool mtr_first;
5049 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5050 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5051 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5052 	uint32_t flow_id = 0;
5053 	uint32_t flow_id_reversed = 0;
5054 	uint8_t flow_id_bits = 0;
5055 	int shift;
5056 
5057 	/* Prepare the suffix subflow items. */
5058 	tag_item = sfx_items++;
5059 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5060 		struct mlx5_priv *port_priv;
5061 		const struct rte_flow_item_port_id *pid_v;
5062 		int item_type = items->type;
5063 
5064 		switch (item_type) {
5065 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5066 			pid_v = items->spec;
5067 			MLX5_ASSERT(pid_v);
5068 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5069 			if (!port_priv)
5070 				return rte_flow_error_set(error,
5071 						rte_errno,
5072 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5073 						pid_v,
5074 						"Failed to get port info.");
5075 			flow_src_port = port_priv->representor_id;
5076 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5077 			    flow_src_port != priv->representor_id) {
5078 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5079 								flow, fm,
5080 								flow_src_port,
5081 								items,
5082 								error))
5083 					return -rte_errno;
5084 			}
5085 			memcpy(sfx_items, items, sizeof(*sfx_items));
5086 			sfx_items++;
5087 			break;
5088 		case RTE_FLOW_ITEM_TYPE_VLAN:
5089 			/* Determine if copy vlan item below. */
5090 			vlan_item_src = items;
5091 			vlan_item_dst = sfx_items++;
5092 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5093 			break;
5094 		default:
5095 			break;
5096 		}
5097 	}
5098 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5099 	sfx_items++;
5100 	mtr_first = priv->sh->meter_aso_en &&
5101 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5102 	/* For ASO meter, meter must be before tag in TX direction. */
5103 	if (mtr_first) {
5104 		action_pre_head = actions_pre++;
5105 		/* Leave space for tag action. */
5106 		tag_action = actions_pre++;
5107 	}
5108 	/* Prepare the actions for prefix and suffix flow. */
5109 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5110 		struct rte_flow_action *action_cur = NULL;
5111 
5112 		switch (actions->type) {
5113 		case RTE_FLOW_ACTION_TYPE_METER:
5114 			if (mtr_first) {
5115 				action_cur = action_pre_head;
5116 			} else {
5117 				/* Leave space for tag action. */
5118 				tag_action = actions_pre++;
5119 				action_cur = actions_pre++;
5120 			}
5121 			break;
5122 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5123 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5124 			action_cur = actions_pre++;
5125 			break;
5126 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5127 			raw_encap = actions->conf;
5128 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5129 				action_cur = actions_pre++;
5130 			break;
5131 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5132 			raw_decap = actions->conf;
5133 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5134 				action_cur = actions_pre++;
5135 			break;
5136 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5137 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5138 			if (vlan_item_dst && vlan_item_src) {
5139 				memcpy(vlan_item_dst, vlan_item_src,
5140 					sizeof(*vlan_item_dst));
5141 				/*
5142 				 * Convert to internal match item, it is used
5143 				 * for vlan push and set vid.
5144 				 */
5145 				vlan_item_dst->type = (enum rte_flow_item_type)
5146 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5147 			}
5148 			break;
5149 		default:
5150 			break;
5151 		}
5152 		if (!action_cur)
5153 			action_cur = (fm->def_policy) ?
5154 					actions_sfx++ : actions_pre++;
5155 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5156 	}
5157 	/* Add end action to the actions. */
5158 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5159 	if (priv->sh->meter_aso_en) {
5160 		/**
5161 		 * For ASO meter, need to add an extra jump action explicitly,
5162 		 * to jump from meter to policer table.
5163 		 */
5164 		struct mlx5_flow_meter_sub_policy *sub_policy;
5165 		struct mlx5_flow_tbl_data_entry *tbl_data;
5166 
5167 		if (!fm->def_policy) {
5168 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5169 							  attr, orig_items,
5170 							  error);
5171 			if (!sub_policy)
5172 				return -rte_errno;
5173 		} else {
5174 			enum mlx5_meter_domain mtr_domain =
5175 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5176 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5177 						MLX5_MTR_DOMAIN_INGRESS);
5178 
5179 			sub_policy =
5180 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5181 		}
5182 		tbl_data = container_of(sub_policy->tbl_rsc,
5183 					struct mlx5_flow_tbl_data_entry, tbl);
5184 		hw_mtr_action = actions_pre++;
5185 		hw_mtr_action->type = (enum rte_flow_action_type)
5186 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5187 		hw_mtr_action->conf = tbl_data->jump.action;
5188 	}
5189 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5190 	actions_pre++;
5191 	if (!tag_action)
5192 		return rte_flow_error_set(error, ENOMEM,
5193 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5194 					  NULL, "No tag action space.");
5195 	if (!mtr_flow_id) {
5196 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5197 		goto exit;
5198 	}
5199 	/* Only default-policy Meter creates mtr flow id. */
5200 	if (fm->def_policy) {
5201 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5202 		if (!tag_id)
5203 			return rte_flow_error_set(error, ENOMEM,
5204 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5205 					"Failed to allocate meter flow id.");
5206 		flow_id = tag_id - 1;
5207 		flow_id_bits = (!flow_id) ? 1 :
5208 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5209 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5210 		    mtr_reg_bits) {
5211 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5212 			return rte_flow_error_set(error, EINVAL,
5213 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5214 					"Meter flow id exceeds max limit.");
5215 		}
5216 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5217 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5218 	}
5219 	/* Build tag actions and items for meter_id/meter flow_id. */
5220 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5221 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5222 	tag_item_mask = tag_item_spec + 1;
5223 	/* Both flow_id and meter_id share the same register. */
5224 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5225 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5226 							    0, error),
5227 		.offset = mtr_id_offset,
5228 		.length = mtr_reg_bits,
5229 		.data = flow->meter,
5230 	};
5231 	/*
5232 	 * The color Reg bits used by flow_id are growing from
5233 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5234 	 */
5235 	for (shift = 0; shift < flow_id_bits; shift++)
5236 		flow_id_reversed = (flow_id_reversed << 1) |
5237 				((flow_id >> shift) & 0x1);
5238 	set_tag->data |=
5239 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5240 	tag_item_spec->id = set_tag->id;
5241 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5242 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5243 	tag_action->type = (enum rte_flow_action_type)
5244 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5245 	tag_action->conf = set_tag;
5246 	tag_item->type = (enum rte_flow_item_type)
5247 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5248 	tag_item->spec = tag_item_spec;
5249 	tag_item->last = NULL;
5250 	tag_item->mask = tag_item_mask;
5251 exit:
5252 	if (mtr_flow_id)
5253 		*mtr_flow_id = tag_id;
5254 	return 0;
5255 }
5256 
5257 /**
5258  * Split action list having QUEUE/RSS for metadata register copy.
5259  *
5260  * Once Q/RSS action is detected in user's action list, the flow action
5261  * should be split in order to copy metadata registers, which will happen in
5262  * RX_CP_TBL like,
5263  *   - CQE->flow_tag := reg_c[1] (MARK)
5264  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5265  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5266  * This is because the last action of each flow must be a terminal action
5267  * (QUEUE, RSS or DROP).
5268  *
5269  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5270  * stored and kept in the mlx5_flow structure per each sub_flow.
5271  *
5272  * The Q/RSS action is replaced with,
5273  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5274  * And the following JUMP action is added at the end,
5275  *   - JUMP, to RX_CP_TBL.
5276  *
5277  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5278  * flow_create_split_metadata() routine. The flow will look like,
5279  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5280  *
5281  * @param dev
5282  *   Pointer to Ethernet device.
5283  * @param[out] split_actions
5284  *   Pointer to store split actions to jump to CP_TBL.
5285  * @param[in] actions
5286  *   Pointer to the list of original flow actions.
5287  * @param[in] qrss
5288  *   Pointer to the Q/RSS action.
5289  * @param[in] actions_n
5290  *   Number of original actions.
5291  * @param[in] mtr_sfx
5292  *   Check if it is in meter suffix table.
5293  * @param[out] error
5294  *   Perform verbose error reporting if not NULL.
5295  *
5296  * @return
5297  *   non-zero unique flow_id on success, otherwise 0 and
5298  *   error/rte_error are set.
5299  */
5300 static uint32_t
5301 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5302 			  struct rte_flow_action *split_actions,
5303 			  const struct rte_flow_action *actions,
5304 			  const struct rte_flow_action *qrss,
5305 			  int actions_n, int mtr_sfx,
5306 			  struct rte_flow_error *error)
5307 {
5308 	struct mlx5_priv *priv = dev->data->dev_private;
5309 	struct mlx5_rte_flow_action_set_tag *set_tag;
5310 	struct rte_flow_action_jump *jump;
5311 	const int qrss_idx = qrss - actions;
5312 	uint32_t flow_id = 0;
5313 	int ret = 0;
5314 
5315 	/*
5316 	 * Given actions will be split
5317 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5318 	 * - Add jump to mreg CP_TBL.
5319 	 * As a result, there will be one more action.
5320 	 */
5321 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5322 	/* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5323 	++actions_n;
5324 	set_tag = (void *)(split_actions + actions_n);
5325 	/*
5326 	 * If we are not the meter suffix flow, add the tag action.
5327 	 * Since meter suffix flow already has the tag added.
5328 	 */
5329 	if (!mtr_sfx) {
5330 		/*
5331 		 * Allocate the new subflow ID. This one is unique within
5332 		 * device and not shared with representors. Otherwise,
5333 		 * we would have to resolve multi-thread access synch
5334 		 * issue. Each flow on the shared device is appended
5335 		 * with source vport identifier, so the resulting
5336 		 * flows will be unique in the shared (by master and
5337 		 * representors) domain even if they have coinciding
5338 		 * IDs.
5339 		 */
5340 		mlx5_ipool_malloc(priv->sh->ipool
5341 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5342 		if (!flow_id)
5343 			return rte_flow_error_set(error, ENOMEM,
5344 						  RTE_FLOW_ERROR_TYPE_ACTION,
5345 						  NULL, "can't allocate id "
5346 						  "for split Q/RSS subflow");
5347 		/* Internal SET_TAG action to set flow ID. */
5348 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5349 			.data = flow_id,
5350 		};
5351 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5352 		if (ret < 0)
5353 			return ret;
5354 		set_tag->id = ret;
5355 		/* Construct new actions array. */
5356 		/* Replace QUEUE/RSS action. */
5357 		split_actions[qrss_idx] = (struct rte_flow_action){
5358 			.type = (enum rte_flow_action_type)
5359 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5360 			.conf = set_tag,
5361 		};
5362 	} else {
5363 		/*
5364 		 * If we are the suffix flow of meter, tag already exist.
5365 		 * Set the QUEUE/RSS action to void.
5366 		 */
5367 		split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5368 	}
5369 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5370 	jump = (void *)(set_tag + 1);
5371 	*jump = (struct rte_flow_action_jump){
5372 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5373 	};
5374 	split_actions[actions_n - 2] = (struct rte_flow_action){
5375 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5376 		.conf = jump,
5377 	};
5378 	split_actions[actions_n - 1] = (struct rte_flow_action){
5379 		.type = RTE_FLOW_ACTION_TYPE_END,
5380 	};
5381 	return flow_id;
5382 }
5383 
5384 /**
5385  * Extend the given action list for Tx metadata copy.
5386  *
5387  * Copy the given action list to the ext_actions and add flow metadata register
5388  * copy action in order to copy reg_a set by WQE to reg_c[0].
5389  *
5390  * @param[out] ext_actions
5391  *   Pointer to the extended action list.
5392  * @param[in] actions
5393  *   Pointer to the list of actions.
5394  * @param[in] actions_n
5395  *   Number of actions in the list.
5396  * @param[out] error
5397  *   Perform verbose error reporting if not NULL.
5398  * @param[in] encap_idx
5399  *   The encap action index.
5400  *
5401  * @return
5402  *   0 on success, negative value otherwise
5403  */
5404 static int
5405 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5406 		       struct rte_flow_action *ext_actions,
5407 		       const struct rte_flow_action *actions,
5408 		       int actions_n, struct rte_flow_error *error,
5409 		       int encap_idx)
5410 {
5411 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5412 		(struct mlx5_flow_action_copy_mreg *)
5413 			(ext_actions + actions_n + 1);
5414 	int ret;
5415 
5416 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5417 	if (ret < 0)
5418 		return ret;
5419 	cp_mreg->dst = ret;
5420 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5421 	if (ret < 0)
5422 		return ret;
5423 	cp_mreg->src = ret;
5424 	if (encap_idx != 0)
5425 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5426 	if (encap_idx == actions_n - 1) {
5427 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5428 			.type = (enum rte_flow_action_type)
5429 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5430 			.conf = cp_mreg,
5431 		};
5432 		ext_actions[actions_n] = (struct rte_flow_action){
5433 			.type = RTE_FLOW_ACTION_TYPE_END,
5434 		};
5435 	} else {
5436 		ext_actions[encap_idx] = (struct rte_flow_action){
5437 			.type = (enum rte_flow_action_type)
5438 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5439 			.conf = cp_mreg,
5440 		};
5441 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5442 				sizeof(*ext_actions) * (actions_n - encap_idx));
5443 	}
5444 	return 0;
5445 }
5446 
5447 /**
5448  * Check the match action from the action list.
5449  *
5450  * @param[in] actions
5451  *   Pointer to the list of actions.
5452  * @param[in] attr
5453  *   Flow rule attributes.
5454  * @param[in] action
5455  *   The action to be check if exist.
5456  * @param[out] match_action_pos
5457  *   Pointer to the position of the matched action if exists, otherwise is -1.
5458  * @param[out] qrss_action_pos
5459  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5460  * @param[out] modify_after_mirror
5461  *   Pointer to the flag of modify action after FDB mirroring.
5462  *
5463  * @return
5464  *   > 0 the total number of actions.
5465  *   0 if not found match action in action list.
5466  */
5467 static int
5468 flow_check_match_action(const struct rte_flow_action actions[],
5469 			const struct rte_flow_attr *attr,
5470 			enum rte_flow_action_type action,
5471 			int *match_action_pos, int *qrss_action_pos,
5472 			int *modify_after_mirror)
5473 {
5474 	const struct rte_flow_action_sample *sample;
5475 	const struct rte_flow_action_raw_decap *decap;
5476 	int actions_n = 0;
5477 	uint32_t ratio = 0;
5478 	int sub_type = 0;
5479 	int flag = 0;
5480 	int fdb_mirror = 0;
5481 
5482 	*match_action_pos = -1;
5483 	*qrss_action_pos = -1;
5484 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5485 		if (actions->type == action) {
5486 			flag = 1;
5487 			*match_action_pos = actions_n;
5488 		}
5489 		switch (actions->type) {
5490 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5491 		case RTE_FLOW_ACTION_TYPE_RSS:
5492 			*qrss_action_pos = actions_n;
5493 			break;
5494 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5495 			sample = actions->conf;
5496 			ratio = sample->ratio;
5497 			sub_type = ((const struct rte_flow_action *)
5498 					(sample->actions))->type;
5499 			if (ratio == 1 && attr->transfer)
5500 				fdb_mirror = 1;
5501 			break;
5502 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5503 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5504 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5505 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5506 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5507 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5508 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5509 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5510 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5511 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5512 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5513 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5514 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5515 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5516 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5517 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5518 		case RTE_FLOW_ACTION_TYPE_FLAG:
5519 		case RTE_FLOW_ACTION_TYPE_MARK:
5520 		case RTE_FLOW_ACTION_TYPE_SET_META:
5521 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5522 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5523 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5524 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5525 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5526 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5527 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5528 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5529 		case RTE_FLOW_ACTION_TYPE_METER:
5530 			if (fdb_mirror)
5531 				*modify_after_mirror = 1;
5532 			break;
5533 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5534 			decap = actions->conf;
5535 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5536 				;
5537 			actions_n++;
5538 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5539 				const struct rte_flow_action_raw_encap *encap =
5540 								actions->conf;
5541 				if (decap->size <=
5542 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5543 				    encap->size >
5544 					MLX5_ENCAPSULATION_DECISION_SIZE)
5545 					/* L3 encap. */
5546 					break;
5547 			}
5548 			if (fdb_mirror)
5549 				*modify_after_mirror = 1;
5550 			break;
5551 		default:
5552 			break;
5553 		}
5554 		actions_n++;
5555 	}
5556 	if (flag && fdb_mirror && !*modify_after_mirror) {
5557 		/* FDB mirroring uses the destination array to implement
5558 		 * instead of FLOW_SAMPLER object.
5559 		 */
5560 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5561 			flag = 0;
5562 	}
5563 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5564 	return flag ? actions_n + 1 : 0;
5565 }
5566 
5567 #define SAMPLE_SUFFIX_ITEM 2
5568 
5569 /**
5570  * Split the sample flow.
5571  *
5572  * As sample flow will split to two sub flow, sample flow with
5573  * sample action, the other actions will move to new suffix flow.
5574  *
5575  * Also add unique tag id with tag action in the sample flow,
5576  * the same tag id will be as match in the suffix flow.
5577  *
5578  * @param dev
5579  *   Pointer to Ethernet device.
5580  * @param[in] add_tag
5581  *   Add extra tag action flag.
5582  * @param[out] sfx_items
5583  *   Suffix flow match items (list terminated by the END pattern item).
5584  * @param[in] actions
5585  *   Associated actions (list terminated by the END action).
5586  * @param[out] actions_sfx
5587  *   Suffix flow actions.
5588  * @param[out] actions_pre
5589  *   Prefix flow actions.
5590  * @param[in] actions_n
5591  *  The total number of actions.
5592  * @param[in] sample_action_pos
5593  *   The sample action position.
5594  * @param[in] qrss_action_pos
5595  *   The Queue/RSS action position.
5596  * @param[in] jump_table
5597  *   Add extra jump action flag.
5598  * @param[out] error
5599  *   Perform verbose error reporting if not NULL.
5600  *
5601  * @return
5602  *   0 on success, or unique flow_id, a negative errno value
5603  *   otherwise and rte_errno is set.
5604  */
5605 static int
5606 flow_sample_split_prep(struct rte_eth_dev *dev,
5607 		       int add_tag,
5608 		       struct rte_flow_item sfx_items[],
5609 		       const struct rte_flow_action actions[],
5610 		       struct rte_flow_action actions_sfx[],
5611 		       struct rte_flow_action actions_pre[],
5612 		       int actions_n,
5613 		       int sample_action_pos,
5614 		       int qrss_action_pos,
5615 		       int jump_table,
5616 		       struct rte_flow_error *error)
5617 {
5618 	struct mlx5_priv *priv = dev->data->dev_private;
5619 	struct mlx5_rte_flow_action_set_tag *set_tag;
5620 	struct mlx5_rte_flow_item_tag *tag_spec;
5621 	struct mlx5_rte_flow_item_tag *tag_mask;
5622 	struct rte_flow_action_jump *jump_action;
5623 	uint32_t tag_id = 0;
5624 	int index;
5625 	int append_index = 0;
5626 	int ret;
5627 
5628 	if (sample_action_pos < 0)
5629 		return rte_flow_error_set(error, EINVAL,
5630 					  RTE_FLOW_ERROR_TYPE_ACTION,
5631 					  NULL, "invalid position of sample "
5632 					  "action in list");
5633 	/* Prepare the actions for prefix and suffix flow. */
5634 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5635 		index = qrss_action_pos;
5636 		/* Put the preceding the Queue/RSS action into prefix flow. */
5637 		if (index != 0)
5638 			memcpy(actions_pre, actions,
5639 			       sizeof(struct rte_flow_action) * index);
5640 		/* Put others preceding the sample action into prefix flow. */
5641 		if (sample_action_pos > index + 1)
5642 			memcpy(actions_pre + index, actions + index + 1,
5643 			       sizeof(struct rte_flow_action) *
5644 			       (sample_action_pos - index - 1));
5645 		index = sample_action_pos - 1;
5646 		/* Put Queue/RSS action into Suffix flow. */
5647 		memcpy(actions_sfx, actions + qrss_action_pos,
5648 		       sizeof(struct rte_flow_action));
5649 		actions_sfx++;
5650 	} else {
5651 		index = sample_action_pos;
5652 		if (index != 0)
5653 			memcpy(actions_pre, actions,
5654 			       sizeof(struct rte_flow_action) * index);
5655 	}
5656 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5657 	 * For CX6DX and above, metadata registers Cx preserve their value,
5658 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5659 	 */
5660 	if (add_tag) {
5661 		/* Prepare the prefix tag action. */
5662 		append_index++;
5663 		set_tag = (void *)(actions_pre + actions_n + append_index);
5664 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5665 		if (ret < 0)
5666 			return ret;
5667 		mlx5_ipool_malloc(priv->sh->ipool
5668 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5669 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5670 			.id = ret,
5671 			.data = tag_id,
5672 		};
5673 		/* Prepare the suffix subflow items. */
5674 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5675 		tag_spec->data = tag_id;
5676 		tag_spec->id = set_tag->id;
5677 		tag_mask = tag_spec + 1;
5678 		tag_mask->data = UINT32_MAX;
5679 		sfx_items[0] = (struct rte_flow_item){
5680 			.type = (enum rte_flow_item_type)
5681 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5682 			.spec = tag_spec,
5683 			.last = NULL,
5684 			.mask = tag_mask,
5685 		};
5686 		sfx_items[1] = (struct rte_flow_item){
5687 			.type = (enum rte_flow_item_type)
5688 				RTE_FLOW_ITEM_TYPE_END,
5689 		};
5690 		/* Prepare the tag action in prefix subflow. */
5691 		actions_pre[index++] =
5692 			(struct rte_flow_action){
5693 			.type = (enum rte_flow_action_type)
5694 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5695 			.conf = set_tag,
5696 		};
5697 	}
5698 	memcpy(actions_pre + index, actions + sample_action_pos,
5699 	       sizeof(struct rte_flow_action));
5700 	index += 1;
5701 	/* For the modify action after the sample action in E-Switch mirroring,
5702 	 * Add the extra jump action in prefix subflow and jump into the next
5703 	 * table, then do the modify action in the new table.
5704 	 */
5705 	if (jump_table) {
5706 		/* Prepare the prefix jump action. */
5707 		append_index++;
5708 		jump_action = (void *)(actions_pre + actions_n + append_index);
5709 		jump_action->group = jump_table;
5710 		actions_pre[index++] =
5711 			(struct rte_flow_action){
5712 			.type = (enum rte_flow_action_type)
5713 				RTE_FLOW_ACTION_TYPE_JUMP,
5714 			.conf = jump_action,
5715 		};
5716 	}
5717 	actions_pre[index] = (struct rte_flow_action){
5718 		.type = (enum rte_flow_action_type)
5719 			RTE_FLOW_ACTION_TYPE_END,
5720 	};
5721 	/* Put the actions after sample into Suffix flow. */
5722 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5723 	       sizeof(struct rte_flow_action) *
5724 	       (actions_n - sample_action_pos - 1));
5725 	return tag_id;
5726 }
5727 
5728 /**
5729  * The splitting for metadata feature.
5730  *
5731  * - Q/RSS action on NIC Rx should be split in order to pass by
5732  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5733  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5734  *
5735  * - All the actions on NIC Tx should have a mreg copy action to
5736  *   copy reg_a from WQE to reg_c[0].
5737  *
5738  * @param dev
5739  *   Pointer to Ethernet device.
5740  * @param[in] flow
5741  *   Parent flow structure pointer.
5742  * @param[in] attr
5743  *   Flow rule attributes.
5744  * @param[in] items
5745  *   Pattern specification (list terminated by the END pattern item).
5746  * @param[in] actions
5747  *   Associated actions (list terminated by the END action).
5748  * @param[in] flow_split_info
5749  *   Pointer to flow split info structure.
5750  * @param[out] error
5751  *   Perform verbose error reporting if not NULL.
5752  * @return
5753  *   0 on success, negative value otherwise
5754  */
5755 static int
5756 flow_create_split_metadata(struct rte_eth_dev *dev,
5757 			   struct rte_flow *flow,
5758 			   const struct rte_flow_attr *attr,
5759 			   const struct rte_flow_item items[],
5760 			   const struct rte_flow_action actions[],
5761 			   struct mlx5_flow_split_info *flow_split_info,
5762 			   struct rte_flow_error *error)
5763 {
5764 	struct mlx5_priv *priv = dev->data->dev_private;
5765 	struct mlx5_sh_config *config = &priv->sh->config;
5766 	const struct rte_flow_action *qrss = NULL;
5767 	struct rte_flow_action *ext_actions = NULL;
5768 	struct mlx5_flow *dev_flow = NULL;
5769 	uint32_t qrss_id = 0;
5770 	int mtr_sfx = 0;
5771 	size_t act_size;
5772 	int actions_n;
5773 	int encap_idx;
5774 	int ret;
5775 
5776 	/* Check whether extensive metadata feature is engaged. */
5777 	if (!config->dv_flow_en ||
5778 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5779 	    !mlx5_flow_ext_mreg_supported(dev))
5780 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5781 					       actions, flow_split_info, error);
5782 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5783 							   &encap_idx);
5784 	if (qrss) {
5785 		/* Exclude hairpin flows from splitting. */
5786 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5787 			const struct rte_flow_action_queue *queue;
5788 
5789 			queue = qrss->conf;
5790 			if (mlx5_rxq_get_type(dev, queue->index) ==
5791 			    MLX5_RXQ_TYPE_HAIRPIN)
5792 				qrss = NULL;
5793 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5794 			const struct rte_flow_action_rss *rss;
5795 
5796 			rss = qrss->conf;
5797 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5798 			    MLX5_RXQ_TYPE_HAIRPIN)
5799 				qrss = NULL;
5800 		}
5801 	}
5802 	if (qrss) {
5803 		/* Check if it is in meter suffix table. */
5804 		mtr_sfx = attr->group == (attr->transfer ?
5805 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5806 			  MLX5_FLOW_TABLE_LEVEL_METER);
5807 		/*
5808 		 * Q/RSS action on NIC Rx should be split in order to pass by
5809 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5810 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5811 		 */
5812 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5813 			   sizeof(struct rte_flow_action_set_tag) +
5814 			   sizeof(struct rte_flow_action_jump);
5815 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5816 					  SOCKET_ID_ANY);
5817 		if (!ext_actions)
5818 			return rte_flow_error_set(error, ENOMEM,
5819 						  RTE_FLOW_ERROR_TYPE_ACTION,
5820 						  NULL, "no memory to split "
5821 						  "metadata flow");
5822 		/*
5823 		 * Create the new actions list with removed Q/RSS action
5824 		 * and appended set tag and jump to register copy table
5825 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5826 		 * in advance, because it is needed for set tag action.
5827 		 */
5828 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5829 						    qrss, actions_n,
5830 						    mtr_sfx, error);
5831 		if (!mtr_sfx && !qrss_id) {
5832 			ret = -rte_errno;
5833 			goto exit;
5834 		}
5835 	} else if (attr->egress && !attr->transfer) {
5836 		/*
5837 		 * All the actions on NIC Tx should have a metadata register
5838 		 * copy action to copy reg_a from WQE to reg_c[meta]
5839 		 */
5840 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5841 			   sizeof(struct mlx5_flow_action_copy_mreg);
5842 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5843 					  SOCKET_ID_ANY);
5844 		if (!ext_actions)
5845 			return rte_flow_error_set(error, ENOMEM,
5846 						  RTE_FLOW_ERROR_TYPE_ACTION,
5847 						  NULL, "no memory to split "
5848 						  "metadata flow");
5849 		/* Create the action list appended with copy register. */
5850 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5851 					     actions_n, error, encap_idx);
5852 		if (ret < 0)
5853 			goto exit;
5854 	}
5855 	/* Add the unmodified original or prefix subflow. */
5856 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5857 				      items, ext_actions ? ext_actions :
5858 				      actions, flow_split_info, error);
5859 	if (ret < 0)
5860 		goto exit;
5861 	MLX5_ASSERT(dev_flow);
5862 	if (qrss) {
5863 		const struct rte_flow_attr q_attr = {
5864 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5865 			.ingress = 1,
5866 		};
5867 		/* Internal PMD action to set register. */
5868 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5869 			.data = qrss_id,
5870 			.id = REG_NON,
5871 		};
5872 		struct rte_flow_item q_items[] = {
5873 			{
5874 				.type = (enum rte_flow_item_type)
5875 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5876 				.spec = &q_tag_spec,
5877 				.last = NULL,
5878 				.mask = NULL,
5879 			},
5880 			{
5881 				.type = RTE_FLOW_ITEM_TYPE_END,
5882 			},
5883 		};
5884 		struct rte_flow_action q_actions[] = {
5885 			{
5886 				.type = qrss->type,
5887 				.conf = qrss->conf,
5888 			},
5889 			{
5890 				.type = RTE_FLOW_ACTION_TYPE_END,
5891 			},
5892 		};
5893 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5894 
5895 		/*
5896 		 * Configure the tag item only if there is no meter subflow.
5897 		 * Since tag is already marked in the meter suffix subflow
5898 		 * we can just use the meter suffix items as is.
5899 		 */
5900 		if (qrss_id) {
5901 			/* Not meter subflow. */
5902 			MLX5_ASSERT(!mtr_sfx);
5903 			/*
5904 			 * Put unique id in prefix flow due to it is destroyed
5905 			 * after suffix flow and id will be freed after there
5906 			 * is no actual flows with this id and identifier
5907 			 * reallocation becomes possible (for example, for
5908 			 * other flows in other threads).
5909 			 */
5910 			dev_flow->handle->split_flow_id = qrss_id;
5911 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5912 						   error);
5913 			if (ret < 0)
5914 				goto exit;
5915 			q_tag_spec.id = ret;
5916 		}
5917 		dev_flow = NULL;
5918 		/* Add suffix subflow to execute Q/RSS. */
5919 		flow_split_info->prefix_layers = layers;
5920 		flow_split_info->prefix_mark = 0;
5921 		flow_split_info->table_id = 0;
5922 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5923 					      &q_attr, mtr_sfx ? items :
5924 					      q_items, q_actions,
5925 					      flow_split_info, error);
5926 		if (ret < 0)
5927 			goto exit;
5928 		/* qrss ID should be freed if failed. */
5929 		qrss_id = 0;
5930 		MLX5_ASSERT(dev_flow);
5931 	}
5932 
5933 exit:
5934 	/*
5935 	 * We do not destroy the partially created sub_flows in case of error.
5936 	 * These ones are included into parent flow list and will be destroyed
5937 	 * by flow_drv_destroy.
5938 	 */
5939 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5940 			qrss_id);
5941 	mlx5_free(ext_actions);
5942 	return ret;
5943 }
5944 
5945 /**
5946  * Create meter internal drop flow with the original pattern.
5947  *
5948  * @param dev
5949  *   Pointer to Ethernet device.
5950  * @param[in] flow
5951  *   Parent flow structure pointer.
5952  * @param[in] attr
5953  *   Flow rule attributes.
5954  * @param[in] items
5955  *   Pattern specification (list terminated by the END pattern item).
5956  * @param[in] flow_split_info
5957  *   Pointer to flow split info structure.
5958  * @param[in] fm
5959  *   Pointer to flow meter structure.
5960  * @param[out] error
5961  *   Perform verbose error reporting if not NULL.
5962  * @return
5963  *   0 on success, negative value otherwise
5964  */
5965 static uint32_t
5966 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5967 			struct rte_flow *flow,
5968 			const struct rte_flow_attr *attr,
5969 			const struct rte_flow_item items[],
5970 			struct mlx5_flow_split_info *flow_split_info,
5971 			struct mlx5_flow_meter_info *fm,
5972 			struct rte_flow_error *error)
5973 {
5974 	struct mlx5_flow *dev_flow = NULL;
5975 	struct rte_flow_attr drop_attr = *attr;
5976 	struct rte_flow_action drop_actions[3];
5977 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5978 
5979 	MLX5_ASSERT(fm->drop_cnt);
5980 	drop_actions[0].type =
5981 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5982 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5983 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5984 	drop_actions[1].conf = NULL;
5985 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5986 	drop_actions[2].conf = NULL;
5987 	drop_split_info.external = false;
5988 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5989 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5990 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5991 	return flow_create_split_inner(dev, flow, &dev_flow,
5992 				&drop_attr, items, drop_actions,
5993 				&drop_split_info, error);
5994 }
5995 
5996 /**
5997  * The splitting for meter feature.
5998  *
5999  * - The meter flow will be split to two flows as prefix and
6000  *   suffix flow. The packets make sense only it pass the prefix
6001  *   meter action.
6002  *
6003  * - Reg_C_5 is used for the packet to match betweend prefix and
6004  *   suffix flow.
6005  *
6006  * @param dev
6007  *   Pointer to Ethernet device.
6008  * @param[in] flow
6009  *   Parent flow structure pointer.
6010  * @param[in] attr
6011  *   Flow rule attributes.
6012  * @param[in] items
6013  *   Pattern specification (list terminated by the END pattern item).
6014  * @param[in] actions
6015  *   Associated actions (list terminated by the END action).
6016  * @param[in] flow_split_info
6017  *   Pointer to flow split info structure.
6018  * @param[out] error
6019  *   Perform verbose error reporting if not NULL.
6020  * @return
6021  *   0 on success, negative value otherwise
6022  */
6023 static int
6024 flow_create_split_meter(struct rte_eth_dev *dev,
6025 			struct rte_flow *flow,
6026 			const struct rte_flow_attr *attr,
6027 			const struct rte_flow_item items[],
6028 			const struct rte_flow_action actions[],
6029 			struct mlx5_flow_split_info *flow_split_info,
6030 			struct rte_flow_error *error)
6031 {
6032 	struct mlx5_priv *priv = dev->data->dev_private;
6033 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6034 	struct rte_flow_action *sfx_actions = NULL;
6035 	struct rte_flow_action *pre_actions = NULL;
6036 	struct rte_flow_item *sfx_items = NULL;
6037 	struct mlx5_flow *dev_flow = NULL;
6038 	struct rte_flow_attr sfx_attr = *attr;
6039 	struct mlx5_flow_meter_info *fm = NULL;
6040 	uint8_t skip_scale_restore;
6041 	bool has_mtr = false;
6042 	bool has_modify = false;
6043 	bool set_mtr_reg = true;
6044 	bool is_mtr_hierarchy = false;
6045 	uint32_t meter_id = 0;
6046 	uint32_t mtr_idx = 0;
6047 	uint32_t mtr_flow_id = 0;
6048 	size_t act_size;
6049 	size_t item_size;
6050 	int actions_n = 0;
6051 	int ret = 0;
6052 
6053 	if (priv->mtr_en)
6054 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6055 						    &has_modify, &meter_id);
6056 	if (has_mtr) {
6057 		if (flow->meter) {
6058 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6059 			if (!fm)
6060 				return rte_flow_error_set(error, EINVAL,
6061 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6062 						NULL, "Meter not found.");
6063 		} else {
6064 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6065 			if (!fm)
6066 				return rte_flow_error_set(error, EINVAL,
6067 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6068 						NULL, "Meter not found.");
6069 			ret = mlx5_flow_meter_attach(priv, fm,
6070 						     &sfx_attr, error);
6071 			if (ret)
6072 				return -rte_errno;
6073 			flow->meter = mtr_idx;
6074 		}
6075 		MLX5_ASSERT(wks);
6076 		wks->fm = fm;
6077 		if (!fm->def_policy) {
6078 			wks->policy = mlx5_flow_meter_policy_find(dev,
6079 								  fm->policy_id,
6080 								  NULL);
6081 			MLX5_ASSERT(wks->policy);
6082 			if (wks->policy->is_hierarchy) {
6083 				wks->final_policy =
6084 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6085 								wks->policy);
6086 				if (!wks->final_policy)
6087 					return rte_flow_error_set(error,
6088 					EINVAL,
6089 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6090 				"Failed to find terminal policy of hierarchy.");
6091 				is_mtr_hierarchy = true;
6092 			}
6093 		}
6094 		/*
6095 		 * If it isn't default-policy Meter, and
6096 		 * 1. There's no action in flow to change
6097 		 *    packet (modify/encap/decap etc.), OR
6098 		 * 2. No drop count needed for this meter.
6099 		 * 3. It's not meter hierarchy.
6100 		 * Then no need to use regC to save meter id anymore.
6101 		 */
6102 		if (!fm->def_policy && !is_mtr_hierarchy &&
6103 		    (!has_modify || !fm->drop_cnt))
6104 			set_mtr_reg = false;
6105 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6106 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6107 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6108 		/* Suffix items: tag, vlan, port id, end. */
6109 #define METER_SUFFIX_ITEM 4
6110 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6111 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6112 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6113 					  0, SOCKET_ID_ANY);
6114 		if (!sfx_actions)
6115 			return rte_flow_error_set(error, ENOMEM,
6116 						  RTE_FLOW_ERROR_TYPE_ACTION,
6117 						  NULL, "no memory to split "
6118 						  "meter flow");
6119 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6120 			     act_size);
6121 		/* There's no suffix flow for meter of non-default policy. */
6122 		if (!fm->def_policy)
6123 			pre_actions = sfx_actions + 1;
6124 		else
6125 			pre_actions = sfx_actions + actions_n;
6126 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6127 					    items, sfx_items, actions,
6128 					    sfx_actions, pre_actions,
6129 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6130 					    error);
6131 		if (ret) {
6132 			ret = -rte_errno;
6133 			goto exit;
6134 		}
6135 		/* Add the prefix subflow. */
6136 		skip_scale_restore = flow_split_info->skip_scale;
6137 		flow_split_info->skip_scale |=
6138 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6139 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6140 					      attr, items, pre_actions,
6141 					      flow_split_info, error);
6142 		flow_split_info->skip_scale = skip_scale_restore;
6143 		if (ret) {
6144 			if (mtr_flow_id)
6145 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6146 			ret = -rte_errno;
6147 			goto exit;
6148 		}
6149 		if (mtr_flow_id) {
6150 			dev_flow->handle->split_flow_id = mtr_flow_id;
6151 			dev_flow->handle->is_meter_flow_id = 1;
6152 		}
6153 		if (!fm->def_policy) {
6154 			if (!set_mtr_reg && fm->drop_cnt)
6155 				ret =
6156 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6157 							&sfx_attr, items,
6158 							flow_split_info,
6159 							fm, error);
6160 			goto exit;
6161 		}
6162 		/* Setting the sfx group atrr. */
6163 		sfx_attr.group = sfx_attr.transfer ?
6164 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6165 				 MLX5_FLOW_TABLE_LEVEL_METER;
6166 		flow_split_info->prefix_layers =
6167 				flow_get_prefix_layer_flags(dev_flow);
6168 		flow_split_info->prefix_mark |= wks->mark;
6169 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6170 	}
6171 	/* Add the prefix subflow. */
6172 	ret = flow_create_split_metadata(dev, flow,
6173 					 &sfx_attr, sfx_items ?
6174 					 sfx_items : items,
6175 					 sfx_actions ? sfx_actions : actions,
6176 					 flow_split_info, error);
6177 exit:
6178 	if (sfx_actions)
6179 		mlx5_free(sfx_actions);
6180 	return ret;
6181 }
6182 
6183 /**
6184  * The splitting for sample feature.
6185  *
6186  * Once Sample action is detected in the action list, the flow actions should
6187  * be split into prefix sub flow and suffix sub flow.
6188  *
6189  * The original items remain in the prefix sub flow, all actions preceding the
6190  * sample action and the sample action itself will be copied to the prefix
6191  * sub flow, the actions following the sample action will be copied to the
6192  * suffix sub flow, Queue action always be located in the suffix sub flow.
6193  *
6194  * In order to make the packet from prefix sub flow matches with suffix sub
6195  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6196  * flow uses tag item with the unique flow id.
6197  *
6198  * @param dev
6199  *   Pointer to Ethernet device.
6200  * @param[in] flow
6201  *   Parent flow structure pointer.
6202  * @param[in] attr
6203  *   Flow rule attributes.
6204  * @param[in] items
6205  *   Pattern specification (list terminated by the END pattern item).
6206  * @param[in] actions
6207  *   Associated actions (list terminated by the END action).
6208  * @param[in] flow_split_info
6209  *   Pointer to flow split info structure.
6210  * @param[out] error
6211  *   Perform verbose error reporting if not NULL.
6212  * @return
6213  *   0 on success, negative value otherwise
6214  */
6215 static int
6216 flow_create_split_sample(struct rte_eth_dev *dev,
6217 			 struct rte_flow *flow,
6218 			 const struct rte_flow_attr *attr,
6219 			 const struct rte_flow_item items[],
6220 			 const struct rte_flow_action actions[],
6221 			 struct mlx5_flow_split_info *flow_split_info,
6222 			 struct rte_flow_error *error)
6223 {
6224 	struct mlx5_priv *priv = dev->data->dev_private;
6225 	struct rte_flow_action *sfx_actions = NULL;
6226 	struct rte_flow_action *pre_actions = NULL;
6227 	struct rte_flow_item *sfx_items = NULL;
6228 	struct mlx5_flow *dev_flow = NULL;
6229 	struct rte_flow_attr sfx_attr = *attr;
6230 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6231 	struct mlx5_flow_dv_sample_resource *sample_res;
6232 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6233 	struct mlx5_flow_tbl_resource *sfx_tbl;
6234 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6235 #endif
6236 	size_t act_size;
6237 	size_t item_size;
6238 	uint32_t fdb_tx = 0;
6239 	int32_t tag_id = 0;
6240 	int actions_n = 0;
6241 	int sample_action_pos;
6242 	int qrss_action_pos;
6243 	int add_tag = 0;
6244 	int modify_after_mirror = 0;
6245 	uint16_t jump_table = 0;
6246 	const uint32_t next_ft_step = 1;
6247 	int ret = 0;
6248 
6249 	if (priv->sampler_en)
6250 		actions_n = flow_check_match_action(actions, attr,
6251 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6252 					&sample_action_pos, &qrss_action_pos,
6253 					&modify_after_mirror);
6254 	if (actions_n) {
6255 		/* The prefix actions must includes sample, tag, end. */
6256 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6257 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6258 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6259 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6260 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6261 					  item_size), 0, SOCKET_ID_ANY);
6262 		if (!sfx_actions)
6263 			return rte_flow_error_set(error, ENOMEM,
6264 						  RTE_FLOW_ERROR_TYPE_ACTION,
6265 						  NULL, "no memory to split "
6266 						  "sample flow");
6267 		/* The representor_id is UINT16_MAX for uplink. */
6268 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6269 		/*
6270 		 * When reg_c_preserve is set, metadata registers Cx preserve
6271 		 * their value even through packet duplication.
6272 		 */
6273 		add_tag = (!fdb_tx ||
6274 			   priv->sh->cdev->config.hca_attr.reg_c_preserve);
6275 		if (add_tag)
6276 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6277 					+ act_size);
6278 		if (modify_after_mirror)
6279 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6280 				     next_ft_step;
6281 		pre_actions = sfx_actions + actions_n;
6282 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6283 						actions, sfx_actions,
6284 						pre_actions, actions_n,
6285 						sample_action_pos,
6286 						qrss_action_pos, jump_table,
6287 						error);
6288 		if (tag_id < 0 || (add_tag && !tag_id)) {
6289 			ret = -rte_errno;
6290 			goto exit;
6291 		}
6292 		if (modify_after_mirror)
6293 			flow_split_info->skip_scale =
6294 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6295 		/* Add the prefix subflow. */
6296 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6297 					      items, pre_actions,
6298 					      flow_split_info, error);
6299 		if (ret) {
6300 			ret = -rte_errno;
6301 			goto exit;
6302 		}
6303 		dev_flow->handle->split_flow_id = tag_id;
6304 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6305 		if (!modify_after_mirror) {
6306 			/* Set the sfx group attr. */
6307 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6308 						dev_flow->dv.sample_res;
6309 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6310 						sample_res->normal_path_tbl;
6311 			sfx_tbl_data = container_of(sfx_tbl,
6312 						struct mlx5_flow_tbl_data_entry,
6313 						tbl);
6314 			sfx_attr.group = sfx_attr.transfer ?
6315 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6316 		} else {
6317 			MLX5_ASSERT(attr->transfer);
6318 			sfx_attr.group = jump_table;
6319 		}
6320 		flow_split_info->prefix_layers =
6321 				flow_get_prefix_layer_flags(dev_flow);
6322 		MLX5_ASSERT(wks);
6323 		flow_split_info->prefix_mark |= wks->mark;
6324 		/* Suffix group level already be scaled with factor, set
6325 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6326 		 * again in translation.
6327 		 */
6328 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6329 #endif
6330 	}
6331 	/* Add the suffix subflow. */
6332 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6333 				      sfx_items ? sfx_items : items,
6334 				      sfx_actions ? sfx_actions : actions,
6335 				      flow_split_info, error);
6336 exit:
6337 	if (sfx_actions)
6338 		mlx5_free(sfx_actions);
6339 	return ret;
6340 }
6341 
6342 /**
6343  * Split the flow to subflow set. The splitters might be linked
6344  * in the chain, like this:
6345  * flow_create_split_outer() calls:
6346  *   flow_create_split_meter() calls:
6347  *     flow_create_split_metadata(meter_subflow_0) calls:
6348  *       flow_create_split_inner(metadata_subflow_0)
6349  *       flow_create_split_inner(metadata_subflow_1)
6350  *       flow_create_split_inner(metadata_subflow_2)
6351  *     flow_create_split_metadata(meter_subflow_1) calls:
6352  *       flow_create_split_inner(metadata_subflow_0)
6353  *       flow_create_split_inner(metadata_subflow_1)
6354  *       flow_create_split_inner(metadata_subflow_2)
6355  *
6356  * This provide flexible way to add new levels of flow splitting.
6357  * The all of successfully created subflows are included to the
6358  * parent flow dev_flow list.
6359  *
6360  * @param dev
6361  *   Pointer to Ethernet device.
6362  * @param[in] flow
6363  *   Parent flow structure pointer.
6364  * @param[in] attr
6365  *   Flow rule attributes.
6366  * @param[in] items
6367  *   Pattern specification (list terminated by the END pattern item).
6368  * @param[in] actions
6369  *   Associated actions (list terminated by the END action).
6370  * @param[in] flow_split_info
6371  *   Pointer to flow split info structure.
6372  * @param[out] error
6373  *   Perform verbose error reporting if not NULL.
6374  * @return
6375  *   0 on success, negative value otherwise
6376  */
6377 static int
6378 flow_create_split_outer(struct rte_eth_dev *dev,
6379 			struct rte_flow *flow,
6380 			const struct rte_flow_attr *attr,
6381 			const struct rte_flow_item items[],
6382 			const struct rte_flow_action actions[],
6383 			struct mlx5_flow_split_info *flow_split_info,
6384 			struct rte_flow_error *error)
6385 {
6386 	int ret;
6387 
6388 	ret = flow_create_split_sample(dev, flow, attr, items,
6389 				       actions, flow_split_info, error);
6390 	MLX5_ASSERT(ret <= 0);
6391 	return ret;
6392 }
6393 
6394 static inline struct mlx5_flow_tunnel *
6395 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6396 {
6397 	struct mlx5_flow_tunnel *tunnel;
6398 
6399 #pragma GCC diagnostic push
6400 #pragma GCC diagnostic ignored "-Wcast-qual"
6401 	tunnel = (typeof(tunnel))flow->tunnel;
6402 #pragma GCC diagnostic pop
6403 
6404 	return tunnel;
6405 }
6406 
6407 /**
6408  * Adjust flow RSS workspace if needed.
6409  *
6410  * @param wks
6411  *   Pointer to thread flow work space.
6412  * @param rss_desc
6413  *   Pointer to RSS descriptor.
6414  * @param[in] nrssq_num
6415  *   New RSS queue number.
6416  *
6417  * @return
6418  *   0 on success, -1 otherwise and rte_errno is set.
6419  */
6420 static int
6421 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6422 			  struct mlx5_flow_rss_desc *rss_desc,
6423 			  uint32_t nrssq_num)
6424 {
6425 	if (likely(nrssq_num <= wks->rssq_num))
6426 		return 0;
6427 	rss_desc->queue = realloc(rss_desc->queue,
6428 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6429 	if (!rss_desc->queue) {
6430 		rte_errno = ENOMEM;
6431 		return -1;
6432 	}
6433 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6434 	return 0;
6435 }
6436 
6437 /**
6438  * Create a flow and add it to @p list.
6439  *
6440  * @param dev
6441  *   Pointer to Ethernet device.
6442  * @param list
6443  *   Pointer to a TAILQ flow list. If this parameter NULL,
6444  *   no list insertion occurred, flow is just created,
6445  *   this is caller's responsibility to track the
6446  *   created flow.
6447  * @param[in] attr
6448  *   Flow rule attributes.
6449  * @param[in] items
6450  *   Pattern specification (list terminated by the END pattern item).
6451  * @param[in] actions
6452  *   Associated actions (list terminated by the END action).
6453  * @param[in] external
6454  *   This flow rule is created by request external to PMD.
6455  * @param[out] error
6456  *   Perform verbose error reporting if not NULL.
6457  *
6458  * @return
6459  *   A flow index on success, 0 otherwise and rte_errno is set.
6460  */
6461 static uint32_t
6462 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6463 		 const struct rte_flow_attr *attr,
6464 		 const struct rte_flow_item items[],
6465 		 const struct rte_flow_action original_actions[],
6466 		 bool external, struct rte_flow_error *error)
6467 {
6468 	struct mlx5_priv *priv = dev->data->dev_private;
6469 	struct rte_flow *flow = NULL;
6470 	struct mlx5_flow *dev_flow;
6471 	const struct rte_flow_action_rss *rss = NULL;
6472 	struct mlx5_translated_action_handle
6473 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6474 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6475 	union {
6476 		struct mlx5_flow_expand_rss buf;
6477 		uint8_t buffer[4096];
6478 	} expand_buffer;
6479 	union {
6480 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6481 		uint8_t buffer[2048];
6482 	} actions_rx;
6483 	union {
6484 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6485 		uint8_t buffer[2048];
6486 	} actions_hairpin_tx;
6487 	union {
6488 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6489 		uint8_t buffer[2048];
6490 	} items_tx;
6491 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6492 	struct mlx5_flow_rss_desc *rss_desc;
6493 	const struct rte_flow_action *p_actions_rx;
6494 	uint32_t i;
6495 	uint32_t idx = 0;
6496 	int hairpin_flow;
6497 	struct rte_flow_attr attr_tx = { .priority = 0 };
6498 	const struct rte_flow_action *actions;
6499 	struct rte_flow_action *translated_actions = NULL;
6500 	struct mlx5_flow_tunnel *tunnel;
6501 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6502 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6503 	struct mlx5_flow_split_info flow_split_info = {
6504 		.external = !!external,
6505 		.skip_scale = 0,
6506 		.flow_idx = 0,
6507 		.prefix_mark = 0,
6508 		.prefix_layers = 0,
6509 		.table_id = 0
6510 	};
6511 	int ret;
6512 
6513 	MLX5_ASSERT(wks);
6514 	rss_desc = &wks->rss_desc;
6515 	ret = flow_action_handles_translate(dev, original_actions,
6516 					    indir_actions,
6517 					    &indir_actions_n,
6518 					    &translated_actions, error);
6519 	if (ret < 0) {
6520 		MLX5_ASSERT(translated_actions == NULL);
6521 		return 0;
6522 	}
6523 	actions = translated_actions ? translated_actions : original_actions;
6524 	p_actions_rx = actions;
6525 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6526 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6527 				external, hairpin_flow, error);
6528 	if (ret < 0)
6529 		goto error_before_hairpin_split;
6530 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6531 	if (!flow) {
6532 		rte_errno = ENOMEM;
6533 		goto error_before_hairpin_split;
6534 	}
6535 	if (hairpin_flow > 0) {
6536 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6537 			rte_errno = EINVAL;
6538 			goto error_before_hairpin_split;
6539 		}
6540 		flow_hairpin_split(dev, actions, actions_rx.actions,
6541 				   actions_hairpin_tx.actions, items_tx.items,
6542 				   idx);
6543 		p_actions_rx = actions_rx.actions;
6544 	}
6545 	flow_split_info.flow_idx = idx;
6546 	flow->drv_type = flow_get_drv_type(dev, attr);
6547 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6548 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6549 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6550 	/* RSS Action only works on NIC RX domain */
6551 	if (attr->ingress && !attr->transfer)
6552 		rss = flow_get_rss_action(dev, p_actions_rx);
6553 	if (rss) {
6554 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6555 			return 0;
6556 		/*
6557 		 * The following information is required by
6558 		 * mlx5_flow_hashfields_adjust() in advance.
6559 		 */
6560 		rss_desc->level = rss->level;
6561 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6562 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6563 	}
6564 	flow->dev_handles = 0;
6565 	if (rss && rss->types) {
6566 		unsigned int graph_root;
6567 
6568 		graph_root = find_graph_root(rss->level);
6569 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6570 					   items, rss->types,
6571 					   mlx5_support_expansion, graph_root);
6572 		MLX5_ASSERT(ret > 0 &&
6573 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6574 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6575 			for (i = 0; i < buf->entries; ++i)
6576 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6577 		}
6578 	} else {
6579 		buf->entries = 1;
6580 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6581 	}
6582 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6583 						      indir_actions_n);
6584 	for (i = 0; i < buf->entries; ++i) {
6585 		/* Initialize flow split data. */
6586 		flow_split_info.prefix_layers = 0;
6587 		flow_split_info.prefix_mark = 0;
6588 		flow_split_info.skip_scale = 0;
6589 		/*
6590 		 * The splitter may create multiple dev_flows,
6591 		 * depending on configuration. In the simplest
6592 		 * case it just creates unmodified original flow.
6593 		 */
6594 		ret = flow_create_split_outer(dev, flow, attr,
6595 					      buf->entry[i].pattern,
6596 					      p_actions_rx, &flow_split_info,
6597 					      error);
6598 		if (ret < 0)
6599 			goto error;
6600 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6601 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6602 							   p_actions_rx,
6603 							   idx,
6604 							   wks->flows[0].tunnel,
6605 							   &default_miss_ctx,
6606 							   error);
6607 			if (ret < 0) {
6608 				mlx5_free(default_miss_ctx.queue);
6609 				goto error;
6610 			}
6611 		}
6612 	}
6613 	/* Create the tx flow. */
6614 	if (hairpin_flow) {
6615 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6616 		attr_tx.ingress = 0;
6617 		attr_tx.egress = 1;
6618 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6619 					 actions_hairpin_tx.actions,
6620 					 idx, error);
6621 		if (!dev_flow)
6622 			goto error;
6623 		dev_flow->flow = flow;
6624 		dev_flow->external = 0;
6625 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6626 			      dev_flow->handle, next);
6627 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6628 					 items_tx.items,
6629 					 actions_hairpin_tx.actions, error);
6630 		if (ret < 0)
6631 			goto error;
6632 	}
6633 	/*
6634 	 * Update the metadata register copy table. If extensive
6635 	 * metadata feature is enabled and registers are supported
6636 	 * we might create the extra rte_flow for each unique
6637 	 * MARK/FLAG action ID.
6638 	 *
6639 	 * The table is updated for ingress Flows only, because
6640 	 * the egress Flows belong to the different device and
6641 	 * copy table should be updated in peer NIC Rx domain.
6642 	 */
6643 	if (attr->ingress &&
6644 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6645 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6646 		if (ret)
6647 			goto error;
6648 	}
6649 	/*
6650 	 * If the flow is external (from application) OR device is started,
6651 	 * OR mreg discover, then apply immediately.
6652 	 */
6653 	if (external || dev->data->dev_started ||
6654 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6655 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6656 		ret = flow_drv_apply(dev, flow, error);
6657 		if (ret < 0)
6658 			goto error;
6659 	}
6660 	flow->type = type;
6661 	flow_rxq_flags_set(dev, flow);
6662 	rte_free(translated_actions);
6663 	tunnel = flow_tunnel_from_rule(wks->flows);
6664 	if (tunnel) {
6665 		flow->tunnel = 1;
6666 		flow->tunnel_id = tunnel->tunnel_id;
6667 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6668 		mlx5_free(default_miss_ctx.queue);
6669 	}
6670 	mlx5_flow_pop_thread_workspace();
6671 	return idx;
6672 error:
6673 	MLX5_ASSERT(flow);
6674 	ret = rte_errno; /* Save rte_errno before cleanup. */
6675 	flow_mreg_del_copy_action(dev, flow);
6676 	flow_drv_destroy(dev, flow);
6677 	if (rss_desc->shared_rss)
6678 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6679 			mlx5_ipool_get
6680 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6681 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6682 	mlx5_ipool_free(priv->flows[type], idx);
6683 	rte_errno = ret; /* Restore rte_errno. */
6684 	ret = rte_errno;
6685 	rte_errno = ret;
6686 	mlx5_flow_pop_thread_workspace();
6687 error_before_hairpin_split:
6688 	rte_free(translated_actions);
6689 	return 0;
6690 }
6691 
6692 /**
6693  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6694  * incoming packets to table 1.
6695  *
6696  * Other flow rules, requested for group n, will be created in
6697  * e-switch table n+1.
6698  * Jump action to e-switch group n will be created to group n+1.
6699  *
6700  * Used when working in switchdev mode, to utilise advantages of table 1
6701  * and above.
6702  *
6703  * @param dev
6704  *   Pointer to Ethernet device.
6705  *
6706  * @return
6707  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6708  */
6709 struct rte_flow *
6710 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6711 {
6712 	const struct rte_flow_attr attr = {
6713 		.group = 0,
6714 		.priority = 0,
6715 		.ingress = 1,
6716 		.egress = 0,
6717 		.transfer = 1,
6718 	};
6719 	const struct rte_flow_item pattern = {
6720 		.type = RTE_FLOW_ITEM_TYPE_END,
6721 	};
6722 	struct rte_flow_action_jump jump = {
6723 		.group = 1,
6724 	};
6725 	const struct rte_flow_action actions[] = {
6726 		{
6727 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6728 			.conf = &jump,
6729 		},
6730 		{
6731 			.type = RTE_FLOW_ACTION_TYPE_END,
6732 		},
6733 	};
6734 	struct rte_flow_error error;
6735 
6736 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6737 						   &attr, &pattern,
6738 						   actions, false, &error);
6739 }
6740 
6741 /**
6742  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6743  * and sq number, directs all packets to peer vport.
6744  *
6745  * @param dev
6746  *   Pointer to Ethernet device.
6747  * @param txq
6748  *   Txq index.
6749  *
6750  * @return
6751  *   Flow ID on success, 0 otherwise and rte_errno is set.
6752  */
6753 uint32_t
6754 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6755 {
6756 	struct rte_flow_attr attr = {
6757 		.group = 0,
6758 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6759 		.ingress = 1,
6760 		.egress = 0,
6761 		.transfer = 1,
6762 	};
6763 	struct rte_flow_item_port_id port_spec = {
6764 		.id = MLX5_PORT_ESW_MGR,
6765 	};
6766 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6767 		.queue = txq,
6768 	};
6769 	struct rte_flow_item pattern[] = {
6770 		{
6771 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6772 			.spec = &port_spec,
6773 		},
6774 		{
6775 			.type = (enum rte_flow_item_type)
6776 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6777 			.spec = &txq_spec,
6778 		},
6779 		{
6780 			.type = RTE_FLOW_ITEM_TYPE_END,
6781 		},
6782 	};
6783 	struct rte_flow_action_jump jump = {
6784 		.group = 1,
6785 	};
6786 	struct rte_flow_action_port_id port = {
6787 		.id = dev->data->port_id,
6788 	};
6789 	struct rte_flow_action actions[] = {
6790 		{
6791 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6792 			.conf = &jump,
6793 		},
6794 		{
6795 			.type = RTE_FLOW_ACTION_TYPE_END,
6796 		},
6797 	};
6798 	struct rte_flow_error error;
6799 
6800 	/*
6801 	 * Creates group 0, highest priority jump flow.
6802 	 * Matches txq to bypass kernel packets.
6803 	 */
6804 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6805 			     false, &error) == 0)
6806 		return 0;
6807 	/* Create group 1, lowest priority redirect flow for txq. */
6808 	attr.group = 1;
6809 	actions[0].conf = &port;
6810 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6811 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6812 				actions, false, &error);
6813 }
6814 
6815 /**
6816  * Validate a flow supported by the NIC.
6817  *
6818  * @see rte_flow_validate()
6819  * @see rte_flow_ops
6820  */
6821 int
6822 mlx5_flow_validate(struct rte_eth_dev *dev,
6823 		   const struct rte_flow_attr *attr,
6824 		   const struct rte_flow_item items[],
6825 		   const struct rte_flow_action original_actions[],
6826 		   struct rte_flow_error *error)
6827 {
6828 	int hairpin_flow;
6829 	struct mlx5_translated_action_handle
6830 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6831 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6832 	const struct rte_flow_action *actions;
6833 	struct rte_flow_action *translated_actions = NULL;
6834 	int ret = flow_action_handles_translate(dev, original_actions,
6835 						indir_actions,
6836 						&indir_actions_n,
6837 						&translated_actions, error);
6838 
6839 	if (ret)
6840 		return ret;
6841 	actions = translated_actions ? translated_actions : original_actions;
6842 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6843 	ret = flow_drv_validate(dev, attr, items, actions,
6844 				true, hairpin_flow, error);
6845 	rte_free(translated_actions);
6846 	return ret;
6847 }
6848 
6849 /**
6850  * Create a flow.
6851  *
6852  * @see rte_flow_create()
6853  * @see rte_flow_ops
6854  */
6855 struct rte_flow *
6856 mlx5_flow_create(struct rte_eth_dev *dev,
6857 		 const struct rte_flow_attr *attr,
6858 		 const struct rte_flow_item items[],
6859 		 const struct rte_flow_action actions[],
6860 		 struct rte_flow_error *error)
6861 {
6862 	struct mlx5_priv *priv = dev->data->dev_private;
6863 
6864 	if (priv->sh->config.dv_flow_en == 2) {
6865 		rte_flow_error_set(error, ENOTSUP,
6866 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6867 			  NULL,
6868 			  "Flow non-Q creation not supported");
6869 		return NULL;
6870 	}
6871 	/*
6872 	 * If the device is not started yet, it is not allowed to created a
6873 	 * flow from application. PMD default flows and traffic control flows
6874 	 * are not affected.
6875 	 */
6876 	if (unlikely(!dev->data->dev_started)) {
6877 		DRV_LOG(DEBUG, "port %u is not started when "
6878 			"inserting a flow", dev->data->port_id);
6879 		rte_flow_error_set(error, ENODEV,
6880 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6881 				   NULL,
6882 				   "port not started");
6883 		return NULL;
6884 	}
6885 
6886 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6887 						   attr, items, actions,
6888 						   true, error);
6889 }
6890 
6891 /**
6892  * Destroy a flow in a list.
6893  *
6894  * @param dev
6895  *   Pointer to Ethernet device.
6896  * @param[in] flow_idx
6897  *   Index of flow to destroy.
6898  */
6899 static void
6900 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6901 		  uint32_t flow_idx)
6902 {
6903 	struct mlx5_priv *priv = dev->data->dev_private;
6904 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6905 
6906 	if (!flow)
6907 		return;
6908 	MLX5_ASSERT(flow->type == type);
6909 	/*
6910 	 * Update RX queue flags only if port is started, otherwise it is
6911 	 * already clean.
6912 	 */
6913 	if (dev->data->dev_started)
6914 		flow_rxq_flags_trim(dev, flow);
6915 	flow_drv_destroy(dev, flow);
6916 	if (flow->tunnel) {
6917 		struct mlx5_flow_tunnel *tunnel;
6918 
6919 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6920 		RTE_VERIFY(tunnel);
6921 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6922 			mlx5_flow_tunnel_free(dev, tunnel);
6923 	}
6924 	flow_mreg_del_copy_action(dev, flow);
6925 	mlx5_ipool_free(priv->flows[type], flow_idx);
6926 }
6927 
6928 /**
6929  * Destroy all flows.
6930  *
6931  * @param dev
6932  *   Pointer to Ethernet device.
6933  * @param type
6934  *   Flow type to be flushed.
6935  * @param active
6936  *   If flushing is called actively.
6937  */
6938 void
6939 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6940 		     bool active)
6941 {
6942 	struct mlx5_priv *priv = dev->data->dev_private;
6943 	uint32_t num_flushed = 0, fidx = 1;
6944 	struct rte_flow *flow;
6945 
6946 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6947 		flow_list_destroy(dev, type, fidx);
6948 		num_flushed++;
6949 	}
6950 	if (active) {
6951 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6952 			dev->data->port_id, num_flushed);
6953 	}
6954 }
6955 
6956 /**
6957  * Stop all default actions for flows.
6958  *
6959  * @param dev
6960  *   Pointer to Ethernet device.
6961  */
6962 void
6963 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6964 {
6965 	flow_mreg_del_default_copy_action(dev);
6966 	flow_rxq_flags_clear(dev);
6967 }
6968 
6969 /**
6970  * Start all default actions for flows.
6971  *
6972  * @param dev
6973  *   Pointer to Ethernet device.
6974  * @return
6975  *   0 on success, a negative errno value otherwise and rte_errno is set.
6976  */
6977 int
6978 mlx5_flow_start_default(struct rte_eth_dev *dev)
6979 {
6980 	struct rte_flow_error error;
6981 
6982 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6983 	return flow_mreg_add_default_copy_action(dev, &error);
6984 }
6985 
6986 /**
6987  * Release key of thread specific flow workspace data.
6988  */
6989 void
6990 flow_release_workspace(void *data)
6991 {
6992 	struct mlx5_flow_workspace *wks = data;
6993 	struct mlx5_flow_workspace *next;
6994 
6995 	while (wks) {
6996 		next = wks->next;
6997 		free(wks->rss_desc.queue);
6998 		free(wks);
6999 		wks = next;
7000 	}
7001 }
7002 
7003 /**
7004  * Get thread specific current flow workspace.
7005  *
7006  * @return pointer to thread specific flow workspace data, NULL on error.
7007  */
7008 struct mlx5_flow_workspace*
7009 mlx5_flow_get_thread_workspace(void)
7010 {
7011 	struct mlx5_flow_workspace *data;
7012 
7013 	data = mlx5_flow_os_get_specific_workspace();
7014 	MLX5_ASSERT(data && data->inuse);
7015 	if (!data || !data->inuse)
7016 		DRV_LOG(ERR, "flow workspace not initialized.");
7017 	return data;
7018 }
7019 
7020 /**
7021  * Allocate and init new flow workspace.
7022  *
7023  * @return pointer to flow workspace data, NULL on error.
7024  */
7025 static struct mlx5_flow_workspace*
7026 flow_alloc_thread_workspace(void)
7027 {
7028 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
7029 
7030 	if (!data) {
7031 		DRV_LOG(ERR, "Failed to allocate flow workspace "
7032 			"memory.");
7033 		return NULL;
7034 	}
7035 	data->rss_desc.queue = calloc(1,
7036 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
7037 	if (!data->rss_desc.queue)
7038 		goto err;
7039 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
7040 	return data;
7041 err:
7042 	free(data->rss_desc.queue);
7043 	free(data);
7044 	return NULL;
7045 }
7046 
7047 /**
7048  * Get new thread specific flow workspace.
7049  *
7050  * If current workspace inuse, create new one and set as current.
7051  *
7052  * @return pointer to thread specific flow workspace data, NULL on error.
7053  */
7054 static struct mlx5_flow_workspace*
7055 mlx5_flow_push_thread_workspace(void)
7056 {
7057 	struct mlx5_flow_workspace *curr;
7058 	struct mlx5_flow_workspace *data;
7059 
7060 	curr = mlx5_flow_os_get_specific_workspace();
7061 	if (!curr) {
7062 		data = flow_alloc_thread_workspace();
7063 		if (!data)
7064 			return NULL;
7065 	} else if (!curr->inuse) {
7066 		data = curr;
7067 	} else if (curr->next) {
7068 		data = curr->next;
7069 	} else {
7070 		data = flow_alloc_thread_workspace();
7071 		if (!data)
7072 			return NULL;
7073 		curr->next = data;
7074 		data->prev = curr;
7075 	}
7076 	data->inuse = 1;
7077 	data->flow_idx = 0;
7078 	/* Set as current workspace */
7079 	if (mlx5_flow_os_set_specific_workspace(data))
7080 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7081 	return data;
7082 }
7083 
7084 /**
7085  * Close current thread specific flow workspace.
7086  *
7087  * If previous workspace available, set it as current.
7088  *
7089  * @return pointer to thread specific flow workspace data, NULL on error.
7090  */
7091 static void
7092 mlx5_flow_pop_thread_workspace(void)
7093 {
7094 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7095 
7096 	if (!data)
7097 		return;
7098 	if (!data->inuse) {
7099 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7100 		return;
7101 	}
7102 	data->inuse = 0;
7103 	if (!data->prev)
7104 		return;
7105 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7106 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7107 }
7108 
7109 /**
7110  * Verify the flow list is empty
7111  *
7112  * @param dev
7113  *  Pointer to Ethernet device.
7114  *
7115  * @return the number of flows not released.
7116  */
7117 int
7118 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7119 {
7120 	struct mlx5_priv *priv = dev->data->dev_private;
7121 	struct rte_flow *flow;
7122 	uint32_t idx = 0;
7123 	int ret = 0, i;
7124 
7125 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7126 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7127 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7128 				dev->data->port_id, (void *)flow);
7129 			ret++;
7130 		}
7131 	}
7132 	return ret;
7133 }
7134 
7135 /**
7136  * Enable default hairpin egress flow.
7137  *
7138  * @param dev
7139  *   Pointer to Ethernet device.
7140  * @param queue
7141  *   The queue index.
7142  *
7143  * @return
7144  *   0 on success, a negative errno value otherwise and rte_errno is set.
7145  */
7146 int
7147 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7148 			    uint32_t queue)
7149 {
7150 	const struct rte_flow_attr attr = {
7151 		.egress = 1,
7152 		.priority = 0,
7153 	};
7154 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7155 		.queue = queue,
7156 	};
7157 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7158 		.queue = UINT32_MAX,
7159 	};
7160 	struct rte_flow_item items[] = {
7161 		{
7162 			.type = (enum rte_flow_item_type)
7163 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7164 			.spec = &queue_spec,
7165 			.last = NULL,
7166 			.mask = &queue_mask,
7167 		},
7168 		{
7169 			.type = RTE_FLOW_ITEM_TYPE_END,
7170 		},
7171 	};
7172 	struct rte_flow_action_jump jump = {
7173 		.group = MLX5_HAIRPIN_TX_TABLE,
7174 	};
7175 	struct rte_flow_action actions[2];
7176 	uint32_t flow_idx;
7177 	struct rte_flow_error error;
7178 
7179 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7180 	actions[0].conf = &jump;
7181 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7182 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7183 				    &attr, items, actions, false, &error);
7184 	if (!flow_idx) {
7185 		DRV_LOG(DEBUG,
7186 			"Failed to create ctrl flow: rte_errno(%d),"
7187 			" type(%d), message(%s)",
7188 			rte_errno, error.type,
7189 			error.message ? error.message : " (no stated reason)");
7190 		return -rte_errno;
7191 	}
7192 	return 0;
7193 }
7194 
7195 /**
7196  * Enable a control flow configured from the control plane.
7197  *
7198  * @param dev
7199  *   Pointer to Ethernet device.
7200  * @param eth_spec
7201  *   An Ethernet flow spec to apply.
7202  * @param eth_mask
7203  *   An Ethernet flow mask to apply.
7204  * @param vlan_spec
7205  *   A VLAN flow spec to apply.
7206  * @param vlan_mask
7207  *   A VLAN flow mask to apply.
7208  *
7209  * @return
7210  *   0 on success, a negative errno value otherwise and rte_errno is set.
7211  */
7212 int
7213 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7214 		    struct rte_flow_item_eth *eth_spec,
7215 		    struct rte_flow_item_eth *eth_mask,
7216 		    struct rte_flow_item_vlan *vlan_spec,
7217 		    struct rte_flow_item_vlan *vlan_mask)
7218 {
7219 	struct mlx5_priv *priv = dev->data->dev_private;
7220 	const struct rte_flow_attr attr = {
7221 		.ingress = 1,
7222 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7223 	};
7224 	struct rte_flow_item items[] = {
7225 		{
7226 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7227 			.spec = eth_spec,
7228 			.last = NULL,
7229 			.mask = eth_mask,
7230 		},
7231 		{
7232 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7233 					      RTE_FLOW_ITEM_TYPE_END,
7234 			.spec = vlan_spec,
7235 			.last = NULL,
7236 			.mask = vlan_mask,
7237 		},
7238 		{
7239 			.type = RTE_FLOW_ITEM_TYPE_END,
7240 		},
7241 	};
7242 	uint16_t queue[priv->reta_idx_n];
7243 	struct rte_flow_action_rss action_rss = {
7244 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7245 		.level = 0,
7246 		.types = priv->rss_conf.rss_hf,
7247 		.key_len = priv->rss_conf.rss_key_len,
7248 		.queue_num = priv->reta_idx_n,
7249 		.key = priv->rss_conf.rss_key,
7250 		.queue = queue,
7251 	};
7252 	struct rte_flow_action actions[] = {
7253 		{
7254 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7255 			.conf = &action_rss,
7256 		},
7257 		{
7258 			.type = RTE_FLOW_ACTION_TYPE_END,
7259 		},
7260 	};
7261 	uint32_t flow_idx;
7262 	struct rte_flow_error error;
7263 	unsigned int i;
7264 
7265 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7266 		return 0;
7267 	}
7268 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7269 		action_rss.types = 0;
7270 	for (i = 0; i != priv->reta_idx_n; ++i)
7271 		queue[i] = (*priv->reta_idx)[i];
7272 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7273 				    &attr, items, actions, false, &error);
7274 	if (!flow_idx)
7275 		return -rte_errno;
7276 	return 0;
7277 }
7278 
7279 /**
7280  * Enable a flow control configured from the control plane.
7281  *
7282  * @param dev
7283  *   Pointer to Ethernet device.
7284  * @param eth_spec
7285  *   An Ethernet flow spec to apply.
7286  * @param eth_mask
7287  *   An Ethernet flow mask to apply.
7288  *
7289  * @return
7290  *   0 on success, a negative errno value otherwise and rte_errno is set.
7291  */
7292 int
7293 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7294 	       struct rte_flow_item_eth *eth_spec,
7295 	       struct rte_flow_item_eth *eth_mask)
7296 {
7297 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7298 }
7299 
7300 /**
7301  * Create default miss flow rule matching lacp traffic
7302  *
7303  * @param dev
7304  *   Pointer to Ethernet device.
7305  * @param eth_spec
7306  *   An Ethernet flow spec to apply.
7307  *
7308  * @return
7309  *   0 on success, a negative errno value otherwise and rte_errno is set.
7310  */
7311 int
7312 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7313 {
7314 	/*
7315 	 * The LACP matching is done by only using ether type since using
7316 	 * a multicast dst mac causes kernel to give low priority to this flow.
7317 	 */
7318 	static const struct rte_flow_item_eth lacp_spec = {
7319 		.type = RTE_BE16(0x8809),
7320 	};
7321 	static const struct rte_flow_item_eth lacp_mask = {
7322 		.type = 0xffff,
7323 	};
7324 	const struct rte_flow_attr attr = {
7325 		.ingress = 1,
7326 	};
7327 	struct rte_flow_item items[] = {
7328 		{
7329 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7330 			.spec = &lacp_spec,
7331 			.mask = &lacp_mask,
7332 		},
7333 		{
7334 			.type = RTE_FLOW_ITEM_TYPE_END,
7335 		},
7336 	};
7337 	struct rte_flow_action actions[] = {
7338 		{
7339 			.type = (enum rte_flow_action_type)
7340 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7341 		},
7342 		{
7343 			.type = RTE_FLOW_ACTION_TYPE_END,
7344 		},
7345 	};
7346 	struct rte_flow_error error;
7347 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7348 					&attr, items, actions,
7349 					false, &error);
7350 
7351 	if (!flow_idx)
7352 		return -rte_errno;
7353 	return 0;
7354 }
7355 
7356 /**
7357  * Destroy a flow.
7358  *
7359  * @see rte_flow_destroy()
7360  * @see rte_flow_ops
7361  */
7362 int
7363 mlx5_flow_destroy(struct rte_eth_dev *dev,
7364 		  struct rte_flow *flow,
7365 		  struct rte_flow_error *error __rte_unused)
7366 {
7367 	struct mlx5_priv *priv = dev->data->dev_private;
7368 
7369 	if (priv->sh->config.dv_flow_en == 2)
7370 		return rte_flow_error_set(error, ENOTSUP,
7371 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7372 			  NULL,
7373 			  "Flow non-Q destruction not supported");
7374 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7375 				(uintptr_t)(void *)flow);
7376 	return 0;
7377 }
7378 
7379 /**
7380  * Destroy all flows.
7381  *
7382  * @see rte_flow_flush()
7383  * @see rte_flow_ops
7384  */
7385 int
7386 mlx5_flow_flush(struct rte_eth_dev *dev,
7387 		struct rte_flow_error *error __rte_unused)
7388 {
7389 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7390 	return 0;
7391 }
7392 
7393 /**
7394  * Isolated mode.
7395  *
7396  * @see rte_flow_isolate()
7397  * @see rte_flow_ops
7398  */
7399 int
7400 mlx5_flow_isolate(struct rte_eth_dev *dev,
7401 		  int enable,
7402 		  struct rte_flow_error *error)
7403 {
7404 	struct mlx5_priv *priv = dev->data->dev_private;
7405 
7406 	if (dev->data->dev_started) {
7407 		rte_flow_error_set(error, EBUSY,
7408 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7409 				   NULL,
7410 				   "port must be stopped first");
7411 		return -rte_errno;
7412 	}
7413 	priv->isolated = !!enable;
7414 	if (enable)
7415 		dev->dev_ops = &mlx5_dev_ops_isolate;
7416 	else
7417 		dev->dev_ops = &mlx5_dev_ops;
7418 
7419 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7420 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7421 
7422 	return 0;
7423 }
7424 
7425 /**
7426  * Query a flow.
7427  *
7428  * @see rte_flow_query()
7429  * @see rte_flow_ops
7430  */
7431 static int
7432 flow_drv_query(struct rte_eth_dev *dev,
7433 	       uint32_t flow_idx,
7434 	       const struct rte_flow_action *actions,
7435 	       void *data,
7436 	       struct rte_flow_error *error)
7437 {
7438 	struct mlx5_priv *priv = dev->data->dev_private;
7439 	const struct mlx5_flow_driver_ops *fops;
7440 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7441 					       flow_idx);
7442 	enum mlx5_flow_drv_type ftype;
7443 
7444 	if (!flow) {
7445 		return rte_flow_error_set(error, ENOENT,
7446 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7447 			  NULL,
7448 			  "invalid flow handle");
7449 	}
7450 	ftype = flow->drv_type;
7451 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7452 	fops = flow_get_drv_ops(ftype);
7453 
7454 	return fops->query(dev, flow, actions, data, error);
7455 }
7456 
7457 /**
7458  * Query a flow.
7459  *
7460  * @see rte_flow_query()
7461  * @see rte_flow_ops
7462  */
7463 int
7464 mlx5_flow_query(struct rte_eth_dev *dev,
7465 		struct rte_flow *flow,
7466 		const struct rte_flow_action *actions,
7467 		void *data,
7468 		struct rte_flow_error *error)
7469 {
7470 	int ret;
7471 	struct mlx5_priv *priv = dev->data->dev_private;
7472 
7473 	if (priv->sh->config.dv_flow_en == 2)
7474 		return rte_flow_error_set(error, ENOTSUP,
7475 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7476 			  NULL,
7477 			  "Flow non-Q query not supported");
7478 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7479 			     error);
7480 	if (ret < 0)
7481 		return ret;
7482 	return 0;
7483 }
7484 
7485 /**
7486  * Get rte_flow callbacks.
7487  *
7488  * @param dev
7489  *   Pointer to Ethernet device structure.
7490  * @param ops
7491  *   Pointer to operation-specific structure.
7492  *
7493  * @return 0
7494  */
7495 int
7496 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7497 		  const struct rte_flow_ops **ops)
7498 {
7499 	*ops = &mlx5_flow_ops;
7500 	return 0;
7501 }
7502 
7503 /**
7504  * Validate meter policy actions.
7505  * Dispatcher for action type specific validation.
7506  *
7507  * @param[in] dev
7508  *   Pointer to the Ethernet device structure.
7509  * @param[in] action
7510  *   The meter policy action object to validate.
7511  * @param[in] attr
7512  *   Attributes of flow to determine steering domain.
7513  * @param[out] is_rss
7514  *   Is RSS or not.
7515  * @param[out] domain_bitmap
7516  *   Domain bitmap.
7517  * @param[out] is_def_policy
7518  *   Is default policy or not.
7519  * @param[out] error
7520  *   Perform verbose error reporting if not NULL. Initialized in case of
7521  *   error only.
7522  *
7523  * @return
7524  *   0 on success, otherwise negative errno value.
7525  */
7526 int
7527 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7528 			const struct rte_flow_action *actions[RTE_COLORS],
7529 			struct rte_flow_attr *attr,
7530 			bool *is_rss,
7531 			uint8_t *domain_bitmap,
7532 			uint8_t *policy_mode,
7533 			struct rte_mtr_error *error)
7534 {
7535 	const struct mlx5_flow_driver_ops *fops;
7536 
7537 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7538 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7539 				       domain_bitmap, policy_mode, error);
7540 }
7541 
7542 /**
7543  * Destroy the meter table set.
7544  *
7545  * @param[in] dev
7546  *   Pointer to Ethernet device.
7547  * @param[in] mtr_policy
7548  *   Meter policy struct.
7549  */
7550 void
7551 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7552 		      struct mlx5_flow_meter_policy *mtr_policy)
7553 {
7554 	const struct mlx5_flow_driver_ops *fops;
7555 
7556 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7557 	fops->destroy_mtr_acts(dev, mtr_policy);
7558 }
7559 
7560 /**
7561  * Create policy action, lock free,
7562  * (mutex should be acquired by caller).
7563  * Dispatcher for action type specific call.
7564  *
7565  * @param[in] dev
7566  *   Pointer to the Ethernet device structure.
7567  * @param[in] mtr_policy
7568  *   Meter policy struct.
7569  * @param[in] action
7570  *   Action specification used to create meter actions.
7571  * @param[out] error
7572  *   Perform verbose error reporting if not NULL. Initialized in case of
7573  *   error only.
7574  *
7575  * @return
7576  *   0 on success, otherwise negative errno value.
7577  */
7578 int
7579 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7580 		      struct mlx5_flow_meter_policy *mtr_policy,
7581 		      const struct rte_flow_action *actions[RTE_COLORS],
7582 		      struct rte_mtr_error *error)
7583 {
7584 	const struct mlx5_flow_driver_ops *fops;
7585 
7586 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7587 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7588 }
7589 
7590 /**
7591  * Create policy rules, lock free,
7592  * (mutex should be acquired by caller).
7593  * Dispatcher for action type specific call.
7594  *
7595  * @param[in] dev
7596  *   Pointer to the Ethernet device structure.
7597  * @param[in] mtr_policy
7598  *   Meter policy struct.
7599  *
7600  * @return
7601  *   0 on success, -1 otherwise.
7602  */
7603 int
7604 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7605 			     struct mlx5_flow_meter_policy *mtr_policy)
7606 {
7607 	const struct mlx5_flow_driver_ops *fops;
7608 
7609 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7610 	return fops->create_policy_rules(dev, mtr_policy);
7611 }
7612 
7613 /**
7614  * Destroy policy rules, lock free,
7615  * (mutex should be acquired by caller).
7616  * Dispatcher for action type specific call.
7617  *
7618  * @param[in] dev
7619  *   Pointer to the Ethernet device structure.
7620  * @param[in] mtr_policy
7621  *   Meter policy struct.
7622  */
7623 void
7624 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7625 			     struct mlx5_flow_meter_policy *mtr_policy)
7626 {
7627 	const struct mlx5_flow_driver_ops *fops;
7628 
7629 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7630 	fops->destroy_policy_rules(dev, mtr_policy);
7631 }
7632 
7633 /**
7634  * Destroy the default policy table set.
7635  *
7636  * @param[in] dev
7637  *   Pointer to Ethernet device.
7638  */
7639 void
7640 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7641 {
7642 	const struct mlx5_flow_driver_ops *fops;
7643 
7644 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7645 	fops->destroy_def_policy(dev);
7646 }
7647 
7648 /**
7649  * Destroy the default policy table set.
7650  *
7651  * @param[in] dev
7652  *   Pointer to Ethernet device.
7653  *
7654  * @return
7655  *   0 on success, -1 otherwise.
7656  */
7657 int
7658 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7659 {
7660 	const struct mlx5_flow_driver_ops *fops;
7661 
7662 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7663 	return fops->create_def_policy(dev);
7664 }
7665 
7666 /**
7667  * Create the needed meter and suffix tables.
7668  *
7669  * @param[in] dev
7670  *   Pointer to Ethernet device.
7671  *
7672  * @return
7673  *   0 on success, -1 otherwise.
7674  */
7675 int
7676 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7677 			struct mlx5_flow_meter_info *fm,
7678 			uint32_t mtr_idx,
7679 			uint8_t domain_bitmap)
7680 {
7681 	const struct mlx5_flow_driver_ops *fops;
7682 
7683 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7684 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7685 }
7686 
7687 /**
7688  * Destroy the meter table set.
7689  *
7690  * @param[in] dev
7691  *   Pointer to Ethernet device.
7692  * @param[in] tbl
7693  *   Pointer to the meter table set.
7694  */
7695 void
7696 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7697 			   struct mlx5_flow_meter_info *fm)
7698 {
7699 	const struct mlx5_flow_driver_ops *fops;
7700 
7701 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7702 	fops->destroy_mtr_tbls(dev, fm);
7703 }
7704 
7705 /**
7706  * Destroy the global meter drop table.
7707  *
7708  * @param[in] dev
7709  *   Pointer to Ethernet device.
7710  */
7711 void
7712 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7713 {
7714 	const struct mlx5_flow_driver_ops *fops;
7715 
7716 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7717 	fops->destroy_mtr_drop_tbls(dev);
7718 }
7719 
7720 /**
7721  * Destroy the sub policy table with RX queue.
7722  *
7723  * @param[in] dev
7724  *   Pointer to Ethernet device.
7725  * @param[in] mtr_policy
7726  *   Pointer to meter policy table.
7727  */
7728 void
7729 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7730 		struct mlx5_flow_meter_policy *mtr_policy)
7731 {
7732 	const struct mlx5_flow_driver_ops *fops;
7733 
7734 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7735 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7736 }
7737 
7738 /**
7739  * Allocate the needed aso flow meter id.
7740  *
7741  * @param[in] dev
7742  *   Pointer to Ethernet device.
7743  *
7744  * @return
7745  *   Index to aso flow meter on success, NULL otherwise.
7746  */
7747 uint32_t
7748 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7749 {
7750 	const struct mlx5_flow_driver_ops *fops;
7751 
7752 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7753 	return fops->create_meter(dev);
7754 }
7755 
7756 /**
7757  * Free the aso flow meter id.
7758  *
7759  * @param[in] dev
7760  *   Pointer to Ethernet device.
7761  * @param[in] mtr_idx
7762  *  Index to aso flow meter to be free.
7763  *
7764  * @return
7765  *   0 on success.
7766  */
7767 void
7768 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7769 {
7770 	const struct mlx5_flow_driver_ops *fops;
7771 
7772 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7773 	fops->free_meter(dev, mtr_idx);
7774 }
7775 
7776 /**
7777  * Allocate a counter.
7778  *
7779  * @param[in] dev
7780  *   Pointer to Ethernet device structure.
7781  *
7782  * @return
7783  *   Index to allocated counter  on success, 0 otherwise.
7784  */
7785 uint32_t
7786 mlx5_counter_alloc(struct rte_eth_dev *dev)
7787 {
7788 	const struct mlx5_flow_driver_ops *fops;
7789 	struct rte_flow_attr attr = { .transfer = 0 };
7790 
7791 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7792 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7793 		return fops->counter_alloc(dev);
7794 	}
7795 	DRV_LOG(ERR,
7796 		"port %u counter allocate is not supported.",
7797 		 dev->data->port_id);
7798 	return 0;
7799 }
7800 
7801 /**
7802  * Free a counter.
7803  *
7804  * @param[in] dev
7805  *   Pointer to Ethernet device structure.
7806  * @param[in] cnt
7807  *   Index to counter to be free.
7808  */
7809 void
7810 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7811 {
7812 	const struct mlx5_flow_driver_ops *fops;
7813 	struct rte_flow_attr attr = { .transfer = 0 };
7814 
7815 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7816 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7817 		fops->counter_free(dev, cnt);
7818 		return;
7819 	}
7820 	DRV_LOG(ERR,
7821 		"port %u counter free is not supported.",
7822 		 dev->data->port_id);
7823 }
7824 
7825 /**
7826  * Query counter statistics.
7827  *
7828  * @param[in] dev
7829  *   Pointer to Ethernet device structure.
7830  * @param[in] cnt
7831  *   Index to counter to query.
7832  * @param[in] clear
7833  *   Set to clear counter statistics.
7834  * @param[out] pkts
7835  *   The counter hits packets number to save.
7836  * @param[out] bytes
7837  *   The counter hits bytes number to save.
7838  *
7839  * @return
7840  *   0 on success, a negative errno value otherwise.
7841  */
7842 int
7843 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7844 		   bool clear, uint64_t *pkts, uint64_t *bytes, void **action)
7845 {
7846 	const struct mlx5_flow_driver_ops *fops;
7847 	struct rte_flow_attr attr = { .transfer = 0 };
7848 
7849 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7850 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7851 		return fops->counter_query(dev, cnt, clear, pkts,
7852 					bytes, action);
7853 	}
7854 	DRV_LOG(ERR,
7855 		"port %u counter query is not supported.",
7856 		 dev->data->port_id);
7857 	return -ENOTSUP;
7858 }
7859 
7860 /**
7861  * Get information about HWS pre-configurable resources.
7862  *
7863  * @param[in] dev
7864  *   Pointer to the rte_eth_dev structure.
7865  * @param[out] port_info
7866  *   Pointer to port information.
7867  * @param[out] queue_info
7868  *   Pointer to queue information.
7869  * @param[out] error
7870  *   Pointer to error structure.
7871  *
7872  * @return
7873  *   0 on success, a negative errno value otherwise and rte_errno is set.
7874  */
7875 static int
7876 mlx5_flow_info_get(struct rte_eth_dev *dev,
7877 		   struct rte_flow_port_info *port_info,
7878 		   struct rte_flow_queue_info *queue_info,
7879 		   struct rte_flow_error *error)
7880 {
7881 	const struct mlx5_flow_driver_ops *fops;
7882 
7883 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
7884 		return rte_flow_error_set(error, ENOTSUP,
7885 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7886 				NULL,
7887 				"info get with incorrect steering mode");
7888 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
7889 	return fops->info_get(dev, port_info, queue_info, error);
7890 }
7891 
7892 /**
7893  * Configure port HWS resources.
7894  *
7895  * @param[in] dev
7896  *   Pointer to the rte_eth_dev structure.
7897  * @param[in] port_attr
7898  *   Port configuration attributes.
7899  * @param[in] nb_queue
7900  *   Number of queue.
7901  * @param[in] queue_attr
7902  *   Array that holds attributes for each flow queue.
7903  * @param[out] error
7904  *   Pointer to error structure.
7905  *
7906  * @return
7907  *   0 on success, a negative errno value otherwise and rte_errno is set.
7908  */
7909 static int
7910 mlx5_flow_port_configure(struct rte_eth_dev *dev,
7911 			 const struct rte_flow_port_attr *port_attr,
7912 			 uint16_t nb_queue,
7913 			 const struct rte_flow_queue_attr *queue_attr[],
7914 			 struct rte_flow_error *error)
7915 {
7916 	const struct mlx5_flow_driver_ops *fops;
7917 
7918 	if (flow_get_drv_type(dev, NULL) != MLX5_FLOW_TYPE_HW)
7919 		return rte_flow_error_set(error, ENOTSUP,
7920 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7921 				NULL,
7922 				"port configure with incorrect steering mode");
7923 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_HW);
7924 	return fops->configure(dev, port_attr, nb_queue, queue_attr, error);
7925 }
7926 
7927 /**
7928  * Allocate a new memory for the counter values wrapped by all the needed
7929  * management.
7930  *
7931  * @param[in] sh
7932  *   Pointer to mlx5_dev_ctx_shared object.
7933  *
7934  * @return
7935  *   0 on success, a negative errno value otherwise.
7936  */
7937 static int
7938 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7939 {
7940 	struct mlx5_counter_stats_mem_mng *mem_mng;
7941 	volatile struct flow_counter_stats *raw_data;
7942 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7943 	int size = (sizeof(struct flow_counter_stats) *
7944 			MLX5_COUNTERS_PER_POOL +
7945 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7946 			sizeof(struct mlx5_counter_stats_mem_mng);
7947 	size_t pgsize = rte_mem_page_size();
7948 	uint8_t *mem;
7949 	int ret;
7950 	int i;
7951 
7952 	if (pgsize == (size_t)-1) {
7953 		DRV_LOG(ERR, "Failed to get mem page size");
7954 		rte_errno = ENOMEM;
7955 		return -ENOMEM;
7956 	}
7957 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7958 	if (!mem) {
7959 		rte_errno = ENOMEM;
7960 		return -ENOMEM;
7961 	}
7962 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7963 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7964 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7965 					  sh->cdev->pdn, mem, size,
7966 					  &mem_mng->wm);
7967 	if (ret) {
7968 		rte_errno = errno;
7969 		mlx5_free(mem);
7970 		return -rte_errno;
7971 	}
7972 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7973 	raw_data = (volatile struct flow_counter_stats *)mem;
7974 	for (i = 0; i < raws_n; ++i) {
7975 		mem_mng->raws[i].mem_mng = mem_mng;
7976 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7977 	}
7978 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7979 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7980 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7981 				 next);
7982 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7983 	sh->cmng.mem_mng = mem_mng;
7984 	return 0;
7985 }
7986 
7987 /**
7988  * Set the statistic memory to the new counter pool.
7989  *
7990  * @param[in] sh
7991  *   Pointer to mlx5_dev_ctx_shared object.
7992  * @param[in] pool
7993  *   Pointer to the pool to set the statistic memory.
7994  *
7995  * @return
7996  *   0 on success, a negative errno value otherwise.
7997  */
7998 static int
7999 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
8000 			       struct mlx5_flow_counter_pool *pool)
8001 {
8002 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8003 	/* Resize statistic memory once used out. */
8004 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
8005 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
8006 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
8007 		return -1;
8008 	}
8009 	rte_spinlock_lock(&pool->sl);
8010 	pool->raw = cmng->mem_mng->raws + pool->index %
8011 		    MLX5_CNT_CONTAINER_RESIZE;
8012 	rte_spinlock_unlock(&pool->sl);
8013 	pool->raw_hw = NULL;
8014 	return 0;
8015 }
8016 
8017 #define MLX5_POOL_QUERY_FREQ_US 1000000
8018 
8019 /**
8020  * Set the periodic procedure for triggering asynchronous batch queries for all
8021  * the counter pools.
8022  *
8023  * @param[in] sh
8024  *   Pointer to mlx5_dev_ctx_shared object.
8025  */
8026 void
8027 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
8028 {
8029 	uint32_t pools_n, us;
8030 
8031 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
8032 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
8033 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
8034 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
8035 		sh->cmng.query_thread_on = 0;
8036 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
8037 	} else {
8038 		sh->cmng.query_thread_on = 1;
8039 	}
8040 }
8041 
8042 /**
8043  * The periodic procedure for triggering asynchronous batch queries for all the
8044  * counter pools. This function is probably called by the host thread.
8045  *
8046  * @param[in] arg
8047  *   The parameter for the alarm process.
8048  */
8049 void
8050 mlx5_flow_query_alarm(void *arg)
8051 {
8052 	struct mlx5_dev_ctx_shared *sh = arg;
8053 	int ret;
8054 	uint16_t pool_index = sh->cmng.pool_index;
8055 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8056 	struct mlx5_flow_counter_pool *pool;
8057 	uint16_t n_valid;
8058 
8059 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
8060 		goto set_alarm;
8061 	rte_spinlock_lock(&cmng->pool_update_sl);
8062 	pool = cmng->pools[pool_index];
8063 	n_valid = cmng->n_valid;
8064 	rte_spinlock_unlock(&cmng->pool_update_sl);
8065 	/* Set the statistic memory to the new created pool. */
8066 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
8067 		goto set_alarm;
8068 	if (pool->raw_hw)
8069 		/* There is a pool query in progress. */
8070 		goto set_alarm;
8071 	pool->raw_hw =
8072 		LIST_FIRST(&sh->cmng.free_stat_raws);
8073 	if (!pool->raw_hw)
8074 		/* No free counter statistics raw memory. */
8075 		goto set_alarm;
8076 	/*
8077 	 * Identify the counters released between query trigger and query
8078 	 * handle more efficiently. The counter released in this gap period
8079 	 * should wait for a new round of query as the new arrived packets
8080 	 * will not be taken into account.
8081 	 */
8082 	pool->query_gen++;
8083 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
8084 					       MLX5_COUNTERS_PER_POOL,
8085 					       NULL, NULL,
8086 					       pool->raw_hw->mem_mng->wm.lkey,
8087 					       (void *)(uintptr_t)
8088 					       pool->raw_hw->data,
8089 					       sh->devx_comp,
8090 					       (uint64_t)(uintptr_t)pool);
8091 	if (ret) {
8092 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
8093 			" %d", pool->min_dcs->id);
8094 		pool->raw_hw = NULL;
8095 		goto set_alarm;
8096 	}
8097 	LIST_REMOVE(pool->raw_hw, next);
8098 	sh->cmng.pending_queries++;
8099 	pool_index++;
8100 	if (pool_index >= n_valid)
8101 		pool_index = 0;
8102 set_alarm:
8103 	sh->cmng.pool_index = pool_index;
8104 	mlx5_set_query_alarm(sh);
8105 }
8106 
8107 /**
8108  * Check and callback event for new aged flow in the counter pool
8109  *
8110  * @param[in] sh
8111  *   Pointer to mlx5_dev_ctx_shared object.
8112  * @param[in] pool
8113  *   Pointer to Current counter pool.
8114  */
8115 static void
8116 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
8117 		   struct mlx5_flow_counter_pool *pool)
8118 {
8119 	struct mlx5_priv *priv;
8120 	struct mlx5_flow_counter *cnt;
8121 	struct mlx5_age_info *age_info;
8122 	struct mlx5_age_param *age_param;
8123 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
8124 	struct mlx5_counter_stats_raw *prev = pool->raw;
8125 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
8126 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
8127 	uint16_t expected = AGE_CANDIDATE;
8128 	uint32_t i;
8129 
8130 	pool->time_of_last_age_check = curr_time;
8131 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8132 		cnt = MLX5_POOL_GET_CNT(pool, i);
8133 		age_param = MLX5_CNT_TO_AGE(cnt);
8134 		if (__atomic_load_n(&age_param->state,
8135 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
8136 			continue;
8137 		if (cur->data[i].hits != prev->data[i].hits) {
8138 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
8139 					 __ATOMIC_RELAXED);
8140 			continue;
8141 		}
8142 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8143 				       time_delta,
8144 				       __ATOMIC_RELAXED) <= age_param->timeout)
8145 			continue;
8146 		/**
8147 		 * Hold the lock first, or if between the
8148 		 * state AGE_TMOUT and tailq operation the
8149 		 * release happened, the release procedure
8150 		 * may delete a non-existent tailq node.
8151 		 */
8152 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8153 		age_info = GET_PORT_AGE_INFO(priv);
8154 		rte_spinlock_lock(&age_info->aged_sl);
8155 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8156 						AGE_TMOUT, false,
8157 						__ATOMIC_RELAXED,
8158 						__ATOMIC_RELAXED)) {
8159 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8160 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8161 		}
8162 		rte_spinlock_unlock(&age_info->aged_sl);
8163 	}
8164 	mlx5_age_event_prepare(sh);
8165 }
8166 
8167 /**
8168  * Handler for the HW respond about ready values from an asynchronous batch
8169  * query. This function is probably called by the host thread.
8170  *
8171  * @param[in] sh
8172  *   The pointer to the shared device context.
8173  * @param[in] async_id
8174  *   The Devx async ID.
8175  * @param[in] status
8176  *   The status of the completion.
8177  */
8178 void
8179 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8180 				  uint64_t async_id, int status)
8181 {
8182 	struct mlx5_flow_counter_pool *pool =
8183 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8184 	struct mlx5_counter_stats_raw *raw_to_free;
8185 	uint8_t query_gen = pool->query_gen ^ 1;
8186 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8187 	enum mlx5_counter_type cnt_type =
8188 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8189 				MLX5_COUNTER_TYPE_ORIGIN;
8190 
8191 	if (unlikely(status)) {
8192 		raw_to_free = pool->raw_hw;
8193 	} else {
8194 		raw_to_free = pool->raw;
8195 		if (pool->is_aged)
8196 			mlx5_flow_aging_check(sh, pool);
8197 		rte_spinlock_lock(&pool->sl);
8198 		pool->raw = pool->raw_hw;
8199 		rte_spinlock_unlock(&pool->sl);
8200 		/* Be sure the new raw counters data is updated in memory. */
8201 		rte_io_wmb();
8202 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8203 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8204 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8205 				     &pool->counters[query_gen], next);
8206 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8207 		}
8208 	}
8209 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8210 	pool->raw_hw = NULL;
8211 	sh->cmng.pending_queries--;
8212 }
8213 
8214 static int
8215 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8216 		    const struct flow_grp_info *grp_info,
8217 		    struct rte_flow_error *error)
8218 {
8219 	if (grp_info->transfer && grp_info->external &&
8220 	    grp_info->fdb_def_rule) {
8221 		if (group == UINT32_MAX)
8222 			return rte_flow_error_set
8223 						(error, EINVAL,
8224 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8225 						 NULL,
8226 						 "group index not supported");
8227 		*table = group + 1;
8228 	} else {
8229 		*table = group;
8230 	}
8231 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8232 	return 0;
8233 }
8234 
8235 /**
8236  * Translate the rte_flow group index to HW table value.
8237  *
8238  * If tunnel offload is disabled, all group ids converted to flow table
8239  * id using the standard method.
8240  * If tunnel offload is enabled, group id can be converted using the
8241  * standard or tunnel conversion method. Group conversion method
8242  * selection depends on flags in `grp_info` parameter:
8243  * - Internal (grp_info.external == 0) groups conversion uses the
8244  *   standard method.
8245  * - Group ids in JUMP action converted with the tunnel conversion.
8246  * - Group id in rule attribute conversion depends on a rule type and
8247  *   group id value:
8248  *   ** non zero group attributes converted with the tunnel method
8249  *   ** zero group attribute in non-tunnel rule is converted using the
8250  *      standard method - there's only one root table
8251  *   ** zero group attribute in steer tunnel rule is converted with the
8252  *      standard method - single root table
8253  *   ** zero group attribute in match tunnel rule is a special OvS
8254  *      case: that value is used for portability reasons. That group
8255  *      id is converted with the tunnel conversion method.
8256  *
8257  * @param[in] dev
8258  *   Port device
8259  * @param[in] tunnel
8260  *   PMD tunnel offload object
8261  * @param[in] group
8262  *   rte_flow group index value.
8263  * @param[out] table
8264  *   HW table value.
8265  * @param[in] grp_info
8266  *   flags used for conversion
8267  * @param[out] error
8268  *   Pointer to error structure.
8269  *
8270  * @return
8271  *   0 on success, a negative errno value otherwise and rte_errno is set.
8272  */
8273 int
8274 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8275 			 const struct mlx5_flow_tunnel *tunnel,
8276 			 uint32_t group, uint32_t *table,
8277 			 const struct flow_grp_info *grp_info,
8278 			 struct rte_flow_error *error)
8279 {
8280 	int ret;
8281 	bool standard_translation;
8282 
8283 	if (!grp_info->skip_scale && grp_info->external &&
8284 	    group < MLX5_MAX_TABLES_EXTERNAL)
8285 		group *= MLX5_FLOW_TABLE_FACTOR;
8286 	if (is_tunnel_offload_active(dev)) {
8287 		standard_translation = !grp_info->external ||
8288 					grp_info->std_tbl_fix;
8289 	} else {
8290 		standard_translation = true;
8291 	}
8292 	DRV_LOG(DEBUG,
8293 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8294 		dev->data->port_id, group, grp_info->transfer,
8295 		grp_info->external, grp_info->fdb_def_rule,
8296 		standard_translation ? "STANDARD" : "TUNNEL");
8297 	if (standard_translation)
8298 		ret = flow_group_to_table(dev->data->port_id, group, table,
8299 					  grp_info, error);
8300 	else
8301 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8302 						      table, error);
8303 
8304 	return ret;
8305 }
8306 
8307 /**
8308  * Discover availability of metadata reg_c's.
8309  *
8310  * Iteratively use test flows to check availability.
8311  *
8312  * @param[in] dev
8313  *   Pointer to the Ethernet device structure.
8314  *
8315  * @return
8316  *   0 on success, a negative errno value otherwise and rte_errno is set.
8317  */
8318 int
8319 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8320 {
8321 	struct mlx5_priv *priv = dev->data->dev_private;
8322 	enum modify_reg idx;
8323 	int n = 0;
8324 
8325 	/* reg_c[0] and reg_c[1] are reserved. */
8326 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8327 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8328 	/* Discover availability of other reg_c's. */
8329 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8330 		struct rte_flow_attr attr = {
8331 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8332 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8333 			.ingress = 1,
8334 		};
8335 		struct rte_flow_item items[] = {
8336 			[0] = {
8337 				.type = RTE_FLOW_ITEM_TYPE_END,
8338 			},
8339 		};
8340 		struct rte_flow_action actions[] = {
8341 			[0] = {
8342 				.type = (enum rte_flow_action_type)
8343 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8344 				.conf = &(struct mlx5_flow_action_copy_mreg){
8345 					.src = REG_C_1,
8346 					.dst = idx,
8347 				},
8348 			},
8349 			[1] = {
8350 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8351 				.conf = &(struct rte_flow_action_jump){
8352 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8353 				},
8354 			},
8355 			[2] = {
8356 				.type = RTE_FLOW_ACTION_TYPE_END,
8357 			},
8358 		};
8359 		uint32_t flow_idx;
8360 		struct rte_flow *flow;
8361 		struct rte_flow_error error;
8362 
8363 		if (!priv->sh->config.dv_flow_en)
8364 			break;
8365 		/* Create internal flow, validation skips copy action. */
8366 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8367 					items, actions, false, &error);
8368 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8369 				      flow_idx);
8370 		if (!flow)
8371 			continue;
8372 		priv->sh->flow_mreg_c[n++] = idx;
8373 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8374 	}
8375 	for (; n < MLX5_MREG_C_NUM; ++n)
8376 		priv->sh->flow_mreg_c[n] = REG_NON;
8377 	priv->sh->metadata_regc_check_flag = 1;
8378 	return 0;
8379 }
8380 
8381 int
8382 save_dump_file(const uint8_t *data, uint32_t size,
8383 	uint32_t type, uint64_t id, void *arg, FILE *file)
8384 {
8385 	char line[BUF_SIZE];
8386 	uint32_t out = 0;
8387 	uint32_t k;
8388 	uint32_t actions_num;
8389 	struct rte_flow_query_count *count;
8390 
8391 	memset(line, 0, BUF_SIZE);
8392 	switch (type) {
8393 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8394 		actions_num = *(uint32_t *)(arg);
8395 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8396 				type, id, actions_num);
8397 		break;
8398 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8399 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8400 				type, id);
8401 		break;
8402 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8403 		count = (struct rte_flow_query_count *)arg;
8404 		fprintf(file,
8405 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8406 			type, id, count->hits, count->bytes);
8407 		return 0;
8408 	default:
8409 		return -1;
8410 	}
8411 
8412 	for (k = 0; k < size; k++) {
8413 		/* Make sure we do not overrun the line buffer length. */
8414 		if (out >= BUF_SIZE - 4) {
8415 			line[out] = '\0';
8416 			break;
8417 		}
8418 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8419 				(data[k]) & 0xff);
8420 	}
8421 	fprintf(file, "%s\n", line);
8422 	return 0;
8423 }
8424 
8425 int
8426 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8427 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8428 {
8429 	struct rte_flow_action action[2];
8430 	enum mlx5_flow_drv_type ftype;
8431 	const struct mlx5_flow_driver_ops *fops;
8432 
8433 	if (!flow) {
8434 		return rte_flow_error_set(error, ENOENT,
8435 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8436 				NULL,
8437 				"invalid flow handle");
8438 	}
8439 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8440 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8441 	if (flow->counter) {
8442 		memset(count, 0, sizeof(struct rte_flow_query_count));
8443 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8444 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8445 						ftype < MLX5_FLOW_TYPE_MAX);
8446 		fops = flow_get_drv_ops(ftype);
8447 		return fops->query(dev, flow, action, count, error);
8448 	}
8449 	return -1;
8450 }
8451 
8452 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8453 /**
8454  * Dump flow ipool data to file
8455  *
8456  * @param[in] dev
8457  *   The pointer to Ethernet device.
8458  * @param[in] file
8459  *   A pointer to a file for output.
8460  * @param[out] error
8461  *   Perform verbose error reporting if not NULL. PMDs initialize this
8462  *   structure in case of error only.
8463  * @return
8464  *   0 on success, a negative value otherwise.
8465  */
8466 int
8467 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8468 	struct rte_flow *flow, FILE *file,
8469 	struct rte_flow_error *error)
8470 {
8471 	struct mlx5_priv *priv = dev->data->dev_private;
8472 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8473 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8474 	uint32_t handle_idx;
8475 	struct mlx5_flow_handle *dh;
8476 	struct rte_flow_query_count count;
8477 	uint32_t actions_num;
8478 	const uint8_t *data;
8479 	size_t size;
8480 	uint64_t id;
8481 	uint32_t type;
8482 	void *action = NULL;
8483 
8484 	if (!flow) {
8485 		return rte_flow_error_set(error, ENOENT,
8486 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8487 				NULL,
8488 				"invalid flow handle");
8489 	}
8490 	handle_idx = flow->dev_handles;
8491 	/* query counter */
8492 	if (flow->counter &&
8493 	(!mlx5_counter_query(dev, flow->counter, false,
8494 	&count.hits, &count.bytes, &action)) && action) {
8495 		id = (uint64_t)(uintptr_t)action;
8496 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8497 		save_dump_file(NULL, 0, type,
8498 			id, (void *)&count, file);
8499 	}
8500 
8501 	while (handle_idx) {
8502 		dh = mlx5_ipool_get(priv->sh->ipool
8503 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8504 		if (!dh)
8505 			continue;
8506 		handle_idx = dh->next.next;
8507 
8508 		/* Get modify_hdr and encap_decap buf from ipools. */
8509 		encap_decap = NULL;
8510 		modify_hdr = dh->dvh.modify_hdr;
8511 
8512 		if (dh->dvh.rix_encap_decap) {
8513 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8514 						[MLX5_IPOOL_DECAP_ENCAP],
8515 						dh->dvh.rix_encap_decap);
8516 		}
8517 		if (modify_hdr) {
8518 			data = (const uint8_t *)modify_hdr->actions;
8519 			size = (size_t)(modify_hdr->actions_num) * 8;
8520 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8521 			actions_num = modify_hdr->actions_num;
8522 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8523 			save_dump_file(data, size, type, id,
8524 						(void *)(&actions_num), file);
8525 		}
8526 		if (encap_decap) {
8527 			data = encap_decap->buf;
8528 			size = encap_decap->size;
8529 			id = (uint64_t)(uintptr_t)encap_decap->action;
8530 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8531 			save_dump_file(data, size, type,
8532 						id, NULL, file);
8533 		}
8534 	}
8535 	return 0;
8536 }
8537 
8538 /**
8539  * Dump all flow's encap_decap/modify_hdr/counter data to file
8540  *
8541  * @param[in] dev
8542  *   The pointer to Ethernet device.
8543  * @param[in] file
8544  *   A pointer to a file for output.
8545  * @param[out] error
8546  *   Perform verbose error reporting if not NULL. PMDs initialize this
8547  *   structure in case of error only.
8548  * @return
8549  *   0 on success, a negative value otherwise.
8550  */
8551 static int
8552 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8553 	FILE *file, struct rte_flow_error *error __rte_unused)
8554 {
8555 	struct mlx5_priv *priv = dev->data->dev_private;
8556 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8557 	struct mlx5_hlist *h;
8558 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8559 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8560 	struct rte_flow_query_count count;
8561 	uint32_t actions_num;
8562 	const uint8_t *data;
8563 	size_t size;
8564 	uint64_t id;
8565 	uint32_t type;
8566 	uint32_t i;
8567 	uint32_t j;
8568 	struct mlx5_list_inconst *l_inconst;
8569 	struct mlx5_list_entry *e;
8570 	int lcore_index;
8571 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8572 	uint32_t max;
8573 	void *action;
8574 
8575 	/* encap_decap hlist is lcore_share, get global core cache. */
8576 	i = MLX5_LIST_GLOBAL;
8577 	h = sh->encaps_decaps;
8578 	if (h) {
8579 		for (j = 0; j <= h->mask; j++) {
8580 			l_inconst = &h->buckets[j].l;
8581 			if (!l_inconst || !l_inconst->cache[i])
8582 				continue;
8583 
8584 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8585 			while (e) {
8586 				encap_decap =
8587 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8588 				data = encap_decap->buf;
8589 				size = encap_decap->size;
8590 				id = (uint64_t)(uintptr_t)encap_decap->action;
8591 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8592 				save_dump_file(data, size, type,
8593 					id, NULL, file);
8594 				e = LIST_NEXT(e, next);
8595 			}
8596 		}
8597 	}
8598 
8599 	/* get modify_hdr */
8600 	h = sh->modify_cmds;
8601 	if (h) {
8602 		lcore_index = rte_lcore_index(rte_lcore_id());
8603 		if (unlikely(lcore_index == -1)) {
8604 			lcore_index = MLX5_LIST_NLCORE;
8605 			rte_spinlock_lock(&h->l_const.lcore_lock);
8606 		}
8607 		i = lcore_index;
8608 
8609 		for (j = 0; j <= h->mask; j++) {
8610 			l_inconst = &h->buckets[j].l;
8611 			if (!l_inconst || !l_inconst->cache[i])
8612 				continue;
8613 
8614 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8615 			while (e) {
8616 				modify_hdr =
8617 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8618 				data = (const uint8_t *)modify_hdr->actions;
8619 				size = (size_t)(modify_hdr->actions_num) * 8;
8620 				actions_num = modify_hdr->actions_num;
8621 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8622 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8623 				save_dump_file(data, size, type, id,
8624 						(void *)(&actions_num), file);
8625 				e = LIST_NEXT(e, next);
8626 			}
8627 		}
8628 
8629 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8630 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8631 	}
8632 
8633 	/* get counter */
8634 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8635 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8636 	for (j = 1; j <= max; j++) {
8637 		action = NULL;
8638 		if ((!mlx5_counter_query(dev, j, false, &count.hits,
8639 		&count.bytes, &action)) && action) {
8640 			id = (uint64_t)(uintptr_t)action;
8641 			type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8642 			save_dump_file(NULL, 0, type,
8643 					id, (void *)&count, file);
8644 		}
8645 	}
8646 	return 0;
8647 }
8648 #endif
8649 
8650 /**
8651  * Dump flow raw hw data to file
8652  *
8653  * @param[in] dev
8654  *    The pointer to Ethernet device.
8655  * @param[in] file
8656  *   A pointer to a file for output.
8657  * @param[out] error
8658  *   Perform verbose error reporting if not NULL. PMDs initialize this
8659  *   structure in case of error only.
8660  * @return
8661  *   0 on success, a negative value otherwise.
8662  */
8663 int
8664 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8665 		   FILE *file,
8666 		   struct rte_flow_error *error __rte_unused)
8667 {
8668 	struct mlx5_priv *priv = dev->data->dev_private;
8669 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8670 	uint32_t handle_idx;
8671 	int ret;
8672 	struct mlx5_flow_handle *dh;
8673 	struct rte_flow *flow;
8674 
8675 	if (!sh->config.dv_flow_en) {
8676 		if (fputs("device dv flow disabled\n", file) <= 0)
8677 			return -errno;
8678 		return -ENOTSUP;
8679 	}
8680 
8681 	/* dump all */
8682 	if (!flow_idx) {
8683 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8684 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8685 			return -EINVAL;
8686 #endif
8687 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8688 					sh->rx_domain,
8689 					sh->tx_domain, file);
8690 	}
8691 	/* dump one */
8692 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8693 			(uintptr_t)(void *)flow_idx);
8694 	if (!flow)
8695 		return -EINVAL;
8696 
8697 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8698 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8699 #endif
8700 	handle_idx = flow->dev_handles;
8701 	while (handle_idx) {
8702 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8703 				handle_idx);
8704 		if (!dh)
8705 			return -ENOENT;
8706 		if (dh->drv_flow) {
8707 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8708 					file);
8709 			if (ret)
8710 				return -ENOENT;
8711 		}
8712 		handle_idx = dh->next.next;
8713 	}
8714 	return 0;
8715 }
8716 
8717 /**
8718  * Get aged-out flows.
8719  *
8720  * @param[in] dev
8721  *   Pointer to the Ethernet device structure.
8722  * @param[in] context
8723  *   The address of an array of pointers to the aged-out flows contexts.
8724  * @param[in] nb_countexts
8725  *   The length of context array pointers.
8726  * @param[out] error
8727  *   Perform verbose error reporting if not NULL. Initialized in case of
8728  *   error only.
8729  *
8730  * @return
8731  *   how many contexts get in success, otherwise negative errno value.
8732  *   if nb_contexts is 0, return the amount of all aged contexts.
8733  *   if nb_contexts is not 0 , return the amount of aged flows reported
8734  *   in the context array.
8735  */
8736 int
8737 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8738 			uint32_t nb_contexts, struct rte_flow_error *error)
8739 {
8740 	const struct mlx5_flow_driver_ops *fops;
8741 	struct rte_flow_attr attr = { .transfer = 0 };
8742 
8743 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8744 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8745 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8746 						    error);
8747 	}
8748 	DRV_LOG(ERR,
8749 		"port %u get aged flows is not supported.",
8750 		 dev->data->port_id);
8751 	return -ENOTSUP;
8752 }
8753 
8754 /* Wrapper for driver action_validate op callback */
8755 static int
8756 flow_drv_action_validate(struct rte_eth_dev *dev,
8757 			 const struct rte_flow_indir_action_conf *conf,
8758 			 const struct rte_flow_action *action,
8759 			 const struct mlx5_flow_driver_ops *fops,
8760 			 struct rte_flow_error *error)
8761 {
8762 	static const char err_msg[] = "indirect action validation unsupported";
8763 
8764 	if (!fops->action_validate) {
8765 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8766 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8767 				   NULL, err_msg);
8768 		return -rte_errno;
8769 	}
8770 	return fops->action_validate(dev, conf, action, error);
8771 }
8772 
8773 /**
8774  * Destroys the shared action by handle.
8775  *
8776  * @param dev
8777  *   Pointer to Ethernet device structure.
8778  * @param[in] handle
8779  *   Handle for the indirect action object to be destroyed.
8780  * @param[out] error
8781  *   Perform verbose error reporting if not NULL. PMDs initialize this
8782  *   structure in case of error only.
8783  *
8784  * @return
8785  *   0 on success, a negative errno value otherwise and rte_errno is set.
8786  *
8787  * @note: wrapper for driver action_create op callback.
8788  */
8789 static int
8790 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8791 			   struct rte_flow_action_handle *handle,
8792 			   struct rte_flow_error *error)
8793 {
8794 	static const char err_msg[] = "indirect action destruction unsupported";
8795 	struct rte_flow_attr attr = { .transfer = 0 };
8796 	const struct mlx5_flow_driver_ops *fops =
8797 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8798 
8799 	if (!fops->action_destroy) {
8800 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8801 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8802 				   NULL, err_msg);
8803 		return -rte_errno;
8804 	}
8805 	return fops->action_destroy(dev, handle, error);
8806 }
8807 
8808 /* Wrapper for driver action_destroy op callback */
8809 static int
8810 flow_drv_action_update(struct rte_eth_dev *dev,
8811 		       struct rte_flow_action_handle *handle,
8812 		       const void *update,
8813 		       const struct mlx5_flow_driver_ops *fops,
8814 		       struct rte_flow_error *error)
8815 {
8816 	static const char err_msg[] = "indirect action update unsupported";
8817 
8818 	if (!fops->action_update) {
8819 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8820 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8821 				   NULL, err_msg);
8822 		return -rte_errno;
8823 	}
8824 	return fops->action_update(dev, handle, update, error);
8825 }
8826 
8827 /* Wrapper for driver action_destroy op callback */
8828 static int
8829 flow_drv_action_query(struct rte_eth_dev *dev,
8830 		      const struct rte_flow_action_handle *handle,
8831 		      void *data,
8832 		      const struct mlx5_flow_driver_ops *fops,
8833 		      struct rte_flow_error *error)
8834 {
8835 	static const char err_msg[] = "indirect action query unsupported";
8836 
8837 	if (!fops->action_query) {
8838 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8839 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8840 				   NULL, err_msg);
8841 		return -rte_errno;
8842 	}
8843 	return fops->action_query(dev, handle, data, error);
8844 }
8845 
8846 /**
8847  * Create indirect action for reuse in multiple flow rules.
8848  *
8849  * @param dev
8850  *   Pointer to Ethernet device structure.
8851  * @param conf
8852  *   Pointer to indirect action object configuration.
8853  * @param[in] action
8854  *   Action configuration for indirect action object creation.
8855  * @param[out] error
8856  *   Perform verbose error reporting if not NULL. PMDs initialize this
8857  *   structure in case of error only.
8858  * @return
8859  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8860  */
8861 static struct rte_flow_action_handle *
8862 mlx5_action_handle_create(struct rte_eth_dev *dev,
8863 			  const struct rte_flow_indir_action_conf *conf,
8864 			  const struct rte_flow_action *action,
8865 			  struct rte_flow_error *error)
8866 {
8867 	static const char err_msg[] = "indirect action creation unsupported";
8868 	struct rte_flow_attr attr = { .transfer = 0 };
8869 	const struct mlx5_flow_driver_ops *fops =
8870 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8871 
8872 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8873 		return NULL;
8874 	if (!fops->action_create) {
8875 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8876 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8877 				   NULL, err_msg);
8878 		return NULL;
8879 	}
8880 	return fops->action_create(dev, conf, action, error);
8881 }
8882 
8883 /**
8884  * Updates inplace the indirect action configuration pointed by *handle*
8885  * with the configuration provided as *update* argument.
8886  * The update of the indirect action configuration effects all flow rules
8887  * reusing the action via handle.
8888  *
8889  * @param dev
8890  *   Pointer to Ethernet device structure.
8891  * @param[in] handle
8892  *   Handle for the indirect action to be updated.
8893  * @param[in] update
8894  *   Action specification used to modify the action pointed by handle.
8895  *   *update* could be of same type with the action pointed by the *handle*
8896  *   handle argument, or some other structures like a wrapper, depending on
8897  *   the indirect action type.
8898  * @param[out] error
8899  *   Perform verbose error reporting if not NULL. PMDs initialize this
8900  *   structure in case of error only.
8901  *
8902  * @return
8903  *   0 on success, a negative errno value otherwise and rte_errno is set.
8904  */
8905 static int
8906 mlx5_action_handle_update(struct rte_eth_dev *dev,
8907 		struct rte_flow_action_handle *handle,
8908 		const void *update,
8909 		struct rte_flow_error *error)
8910 {
8911 	struct rte_flow_attr attr = { .transfer = 0 };
8912 	const struct mlx5_flow_driver_ops *fops =
8913 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8914 	int ret;
8915 
8916 	ret = flow_drv_action_validate(dev, NULL,
8917 			(const struct rte_flow_action *)update, fops, error);
8918 	if (ret)
8919 		return ret;
8920 	return flow_drv_action_update(dev, handle, update, fops,
8921 				      error);
8922 }
8923 
8924 /**
8925  * Query the indirect action by handle.
8926  *
8927  * This function allows retrieving action-specific data such as counters.
8928  * Data is gathered by special action which may be present/referenced in
8929  * more than one flow rule definition.
8930  *
8931  * see @RTE_FLOW_ACTION_TYPE_COUNT
8932  *
8933  * @param dev
8934  *   Pointer to Ethernet device structure.
8935  * @param[in] handle
8936  *   Handle for the indirect action to query.
8937  * @param[in, out] data
8938  *   Pointer to storage for the associated query data type.
8939  * @param[out] error
8940  *   Perform verbose error reporting if not NULL. PMDs initialize this
8941  *   structure in case of error only.
8942  *
8943  * @return
8944  *   0 on success, a negative errno value otherwise and rte_errno is set.
8945  */
8946 static int
8947 mlx5_action_handle_query(struct rte_eth_dev *dev,
8948 			 const struct rte_flow_action_handle *handle,
8949 			 void *data,
8950 			 struct rte_flow_error *error)
8951 {
8952 	struct rte_flow_attr attr = { .transfer = 0 };
8953 	const struct mlx5_flow_driver_ops *fops =
8954 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8955 
8956 	return flow_drv_action_query(dev, handle, data, fops, error);
8957 }
8958 
8959 /**
8960  * Destroy all indirect actions (shared RSS).
8961  *
8962  * @param dev
8963  *   Pointer to Ethernet device.
8964  *
8965  * @return
8966  *   0 on success, a negative errno value otherwise and rte_errno is set.
8967  */
8968 int
8969 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8970 {
8971 	struct rte_flow_error error;
8972 	struct mlx5_priv *priv = dev->data->dev_private;
8973 	struct mlx5_shared_action_rss *shared_rss;
8974 	int ret = 0;
8975 	uint32_t idx;
8976 
8977 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8978 		      priv->rss_shared_actions, idx, shared_rss, next) {
8979 		ret |= mlx5_action_handle_destroy(dev,
8980 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8981 	}
8982 	return ret;
8983 }
8984 
8985 /**
8986  * Validate existing indirect actions against current device configuration
8987  * and attach them to device resources.
8988  *
8989  * @param dev
8990  *   Pointer to Ethernet device.
8991  *
8992  * @return
8993  *   0 on success, a negative errno value otherwise and rte_errno is set.
8994  */
8995 int
8996 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8997 {
8998 	struct mlx5_priv *priv = dev->data->dev_private;
8999 	struct mlx5_indexed_pool *ipool =
9000 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
9001 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
9002 	int ret = 0;
9003 	uint32_t idx;
9004 
9005 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
9006 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
9007 		const char *message;
9008 		uint32_t queue_idx;
9009 
9010 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
9011 					       ind_tbl->queues_n,
9012 					       &message, &queue_idx);
9013 		if (ret != 0) {
9014 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
9015 				dev->data->port_id, ind_tbl->queues[queue_idx],
9016 				message);
9017 			break;
9018 		}
9019 	}
9020 	if (ret != 0)
9021 		return ret;
9022 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
9023 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
9024 
9025 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
9026 		if (ret != 0) {
9027 			DRV_LOG(ERR, "Port %u could not attach "
9028 				"indirection table obj %p",
9029 				dev->data->port_id, (void *)ind_tbl);
9030 			goto error;
9031 		}
9032 	}
9033 	return 0;
9034 error:
9035 	shared_rss_last = shared_rss;
9036 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
9037 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
9038 
9039 		if (shared_rss == shared_rss_last)
9040 			break;
9041 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
9042 			DRV_LOG(CRIT, "Port %u could not detach "
9043 				"indirection table obj %p on rollback",
9044 				dev->data->port_id, (void *)ind_tbl);
9045 	}
9046 	return ret;
9047 }
9048 
9049 /**
9050  * Detach indirect actions of the device from its resources.
9051  *
9052  * @param dev
9053  *   Pointer to Ethernet device.
9054  *
9055  * @return
9056  *   0 on success, a negative errno value otherwise and rte_errno is set.
9057  */
9058 int
9059 mlx5_action_handle_detach(struct rte_eth_dev *dev)
9060 {
9061 	struct mlx5_priv *priv = dev->data->dev_private;
9062 	struct mlx5_indexed_pool *ipool =
9063 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
9064 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
9065 	int ret = 0;
9066 	uint32_t idx;
9067 
9068 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
9069 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
9070 
9071 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
9072 		if (ret != 0) {
9073 			DRV_LOG(ERR, "Port %u could not detach "
9074 				"indirection table obj %p",
9075 				dev->data->port_id, (void *)ind_tbl);
9076 			goto error;
9077 		}
9078 	}
9079 	return 0;
9080 error:
9081 	shared_rss_last = shared_rss;
9082 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
9083 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
9084 
9085 		if (shared_rss == shared_rss_last)
9086 			break;
9087 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
9088 			DRV_LOG(CRIT, "Port %u could not attach "
9089 				"indirection table obj %p on rollback",
9090 				dev->data->port_id, (void *)ind_tbl);
9091 	}
9092 	return ret;
9093 }
9094 
9095 #ifndef HAVE_MLX5DV_DR
9096 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
9097 #else
9098 #define MLX5_DOMAIN_SYNC_FLOW \
9099 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
9100 #endif
9101 
9102 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
9103 {
9104 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
9105 	const struct mlx5_flow_driver_ops *fops;
9106 	int ret;
9107 	struct rte_flow_attr attr = { .transfer = 0 };
9108 
9109 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9110 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
9111 	if (ret > 0)
9112 		ret = -ret;
9113 	return ret;
9114 }
9115 
9116 const struct mlx5_flow_tunnel *
9117 mlx5_get_tof(const struct rte_flow_item *item,
9118 	     const struct rte_flow_action *action,
9119 	     enum mlx5_tof_rule_type *rule_type)
9120 {
9121 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9122 		if (item->type == (typeof(item->type))
9123 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
9124 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
9125 			return flow_items_to_tunnel(item);
9126 		}
9127 	}
9128 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9129 		if (action->type == (typeof(action->type))
9130 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9131 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9132 			return flow_actions_to_tunnel(action);
9133 		}
9134 	}
9135 	return NULL;
9136 }
9137 
9138 /**
9139  * tunnel offload functionality is defined for DV environment only
9140  */
9141 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9142 __extension__
9143 union tunnel_offload_mark {
9144 	uint32_t val;
9145 	struct {
9146 		uint32_t app_reserve:8;
9147 		uint32_t table_id:15;
9148 		uint32_t transfer:1;
9149 		uint32_t _unused_:8;
9150 	};
9151 };
9152 
9153 static bool
9154 mlx5_access_tunnel_offload_db
9155 	(struct rte_eth_dev *dev,
9156 	 bool (*match)(struct rte_eth_dev *,
9157 		       struct mlx5_flow_tunnel *, const void *),
9158 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9159 	 void (*miss)(struct rte_eth_dev *, void *),
9160 	 void *ctx, bool lock_op);
9161 
9162 static int
9163 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9164 			     struct rte_flow *flow,
9165 			     const struct rte_flow_attr *attr,
9166 			     const struct rte_flow_action *app_actions,
9167 			     uint32_t flow_idx,
9168 			     const struct mlx5_flow_tunnel *tunnel,
9169 			     struct tunnel_default_miss_ctx *ctx,
9170 			     struct rte_flow_error *error)
9171 {
9172 	struct mlx5_priv *priv = dev->data->dev_private;
9173 	struct mlx5_flow *dev_flow;
9174 	struct rte_flow_attr miss_attr = *attr;
9175 	const struct rte_flow_item miss_items[2] = {
9176 		{
9177 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9178 			.spec = NULL,
9179 			.last = NULL,
9180 			.mask = NULL
9181 		},
9182 		{
9183 			.type = RTE_FLOW_ITEM_TYPE_END,
9184 			.spec = NULL,
9185 			.last = NULL,
9186 			.mask = NULL
9187 		}
9188 	};
9189 	union tunnel_offload_mark mark_id;
9190 	struct rte_flow_action_mark miss_mark;
9191 	struct rte_flow_action miss_actions[3] = {
9192 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9193 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9194 	};
9195 	const struct rte_flow_action_jump *jump_data;
9196 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9197 	struct flow_grp_info grp_info = {
9198 		.external = 1,
9199 		.transfer = attr->transfer,
9200 		.fdb_def_rule = !!priv->fdb_def_rule,
9201 		.std_tbl_fix = 0,
9202 	};
9203 	int ret;
9204 
9205 	if (!attr->transfer) {
9206 		uint32_t q_size;
9207 
9208 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9209 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9210 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9211 					 0, SOCKET_ID_ANY);
9212 		if (!ctx->queue)
9213 			return rte_flow_error_set
9214 				(error, ENOMEM,
9215 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9216 				NULL, "invalid default miss RSS");
9217 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9218 		ctx->action_rss.level = 0,
9219 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9220 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9221 		ctx->action_rss.queue_num = priv->reta_idx_n,
9222 		ctx->action_rss.key = priv->rss_conf.rss_key,
9223 		ctx->action_rss.queue = ctx->queue;
9224 		if (!priv->reta_idx_n || !priv->rxqs_n)
9225 			return rte_flow_error_set
9226 				(error, EINVAL,
9227 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9228 				NULL, "invalid port configuration");
9229 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9230 			ctx->action_rss.types = 0;
9231 		for (i = 0; i != priv->reta_idx_n; ++i)
9232 			ctx->queue[i] = (*priv->reta_idx)[i];
9233 	} else {
9234 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9235 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9236 	}
9237 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9238 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9239 	jump_data = app_actions->conf;
9240 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9241 	miss_attr.group = jump_data->group;
9242 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9243 				       &flow_table, &grp_info, error);
9244 	if (ret)
9245 		return rte_flow_error_set(error, EINVAL,
9246 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9247 					  NULL, "invalid tunnel id");
9248 	mark_id.app_reserve = 0;
9249 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9250 	mark_id.transfer = !!attr->transfer;
9251 	mark_id._unused_ = 0;
9252 	miss_mark.id = mark_id.val;
9253 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9254 				    miss_items, miss_actions, flow_idx, error);
9255 	if (!dev_flow)
9256 		return -rte_errno;
9257 	dev_flow->flow = flow;
9258 	dev_flow->external = true;
9259 	dev_flow->tunnel = tunnel;
9260 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9261 	/* Subflow object was created, we must include one in the list. */
9262 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9263 		      dev_flow->handle, next);
9264 	DRV_LOG(DEBUG,
9265 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9266 		dev->data->port_id, tunnel->app_tunnel.type,
9267 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9268 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9269 				  miss_actions, error);
9270 	if (!ret)
9271 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9272 						  error);
9273 
9274 	return ret;
9275 }
9276 
9277 static const struct mlx5_flow_tbl_data_entry  *
9278 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9279 {
9280 	struct mlx5_priv *priv = dev->data->dev_private;
9281 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9282 	struct mlx5_list_entry *he;
9283 	union tunnel_offload_mark mbits = { .val = mark };
9284 	union mlx5_flow_tbl_key table_key = {
9285 		{
9286 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9287 			.id = 0,
9288 			.reserved = 0,
9289 			.dummy = 0,
9290 			.is_fdb = !!mbits.transfer,
9291 			.is_egress = 0,
9292 		}
9293 	};
9294 	struct mlx5_flow_cb_ctx ctx = {
9295 		.data = &table_key.v64,
9296 	};
9297 
9298 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9299 	return he ?
9300 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9301 }
9302 
9303 static void
9304 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9305 				   struct mlx5_list_entry *entry)
9306 {
9307 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9308 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9309 
9310 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9311 			tunnel_flow_tbl_to_id(tte->flow_table));
9312 	mlx5_free(tte);
9313 }
9314 
9315 static int
9316 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9317 				  struct mlx5_list_entry *entry, void *cb_ctx)
9318 {
9319 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9320 	union tunnel_tbl_key tbl = {
9321 		.val = *(uint64_t *)(ctx->data),
9322 	};
9323 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9324 
9325 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9326 }
9327 
9328 static struct mlx5_list_entry *
9329 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9330 {
9331 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9332 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9333 	struct tunnel_tbl_entry *tte;
9334 	union tunnel_tbl_key tbl = {
9335 		.val = *(uint64_t *)(ctx->data),
9336 	};
9337 
9338 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9339 			  sizeof(*tte), 0,
9340 			  SOCKET_ID_ANY);
9341 	if (!tte)
9342 		goto err;
9343 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9344 			  &tte->flow_table);
9345 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9346 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9347 			tte->flow_table);
9348 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9349 				tte->flow_table);
9350 		goto err;
9351 	} else if (!tte->flow_table) {
9352 		goto err;
9353 	}
9354 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9355 	tte->tunnel_id = tbl.tunnel_id;
9356 	tte->group = tbl.group;
9357 	return &tte->hash;
9358 err:
9359 	if (tte)
9360 		mlx5_free(tte);
9361 	return NULL;
9362 }
9363 
9364 static struct mlx5_list_entry *
9365 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9366 				  struct mlx5_list_entry *oentry,
9367 				  void *cb_ctx __rte_unused)
9368 {
9369 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9370 						   0, SOCKET_ID_ANY);
9371 
9372 	if (!tte)
9373 		return NULL;
9374 	memcpy(tte, oentry, sizeof(*tte));
9375 	return &tte->hash;
9376 }
9377 
9378 static void
9379 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9380 				       struct mlx5_list_entry *entry)
9381 {
9382 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9383 
9384 	mlx5_free(tte);
9385 }
9386 
9387 static uint32_t
9388 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9389 				const struct mlx5_flow_tunnel *tunnel,
9390 				uint32_t group, uint32_t *table,
9391 				struct rte_flow_error *error)
9392 {
9393 	struct mlx5_list_entry *he;
9394 	struct tunnel_tbl_entry *tte;
9395 	union tunnel_tbl_key key = {
9396 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9397 		.group = group
9398 	};
9399 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9400 	struct mlx5_hlist *group_hash;
9401 	struct mlx5_flow_cb_ctx ctx = {
9402 		.data = &key.val,
9403 	};
9404 
9405 	group_hash = tunnel ? tunnel->groups : thub->groups;
9406 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9407 	if (!he)
9408 		return rte_flow_error_set(error, EINVAL,
9409 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9410 					  NULL,
9411 					  "tunnel group index not supported");
9412 	tte = container_of(he, typeof(*tte), hash);
9413 	*table = tte->flow_table;
9414 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9415 		dev->data->port_id, key.tunnel_id, group, *table);
9416 	return 0;
9417 }
9418 
9419 static void
9420 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9421 		      struct mlx5_flow_tunnel *tunnel)
9422 {
9423 	struct mlx5_priv *priv = dev->data->dev_private;
9424 	struct mlx5_indexed_pool *ipool;
9425 
9426 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9427 		dev->data->port_id, tunnel->tunnel_id);
9428 	LIST_REMOVE(tunnel, chain);
9429 	mlx5_hlist_destroy(tunnel->groups);
9430 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9431 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9432 }
9433 
9434 static bool
9435 mlx5_access_tunnel_offload_db
9436 	(struct rte_eth_dev *dev,
9437 	 bool (*match)(struct rte_eth_dev *,
9438 		       struct mlx5_flow_tunnel *, const void *),
9439 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9440 	 void (*miss)(struct rte_eth_dev *, void *),
9441 	 void *ctx, bool lock_op)
9442 {
9443 	bool verdict = false;
9444 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9445 	struct mlx5_flow_tunnel *tunnel;
9446 
9447 	rte_spinlock_lock(&thub->sl);
9448 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9449 		verdict = match(dev, tunnel, (const void *)ctx);
9450 		if (verdict)
9451 			break;
9452 	}
9453 	if (!lock_op)
9454 		rte_spinlock_unlock(&thub->sl);
9455 	if (verdict && hit)
9456 		hit(dev, tunnel, ctx);
9457 	if (!verdict && miss)
9458 		miss(dev, ctx);
9459 	if (lock_op)
9460 		rte_spinlock_unlock(&thub->sl);
9461 
9462 	return verdict;
9463 }
9464 
9465 struct tunnel_db_find_tunnel_id_ctx {
9466 	uint32_t tunnel_id;
9467 	struct mlx5_flow_tunnel *tunnel;
9468 };
9469 
9470 static bool
9471 find_tunnel_id_match(struct rte_eth_dev *dev,
9472 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9473 {
9474 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9475 
9476 	RTE_SET_USED(dev);
9477 	return tunnel->tunnel_id == ctx->tunnel_id;
9478 }
9479 
9480 static void
9481 find_tunnel_id_hit(struct rte_eth_dev *dev,
9482 		   struct mlx5_flow_tunnel *tunnel, void *x)
9483 {
9484 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9485 	RTE_SET_USED(dev);
9486 	ctx->tunnel = tunnel;
9487 }
9488 
9489 static struct mlx5_flow_tunnel *
9490 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9491 {
9492 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9493 		.tunnel_id = id,
9494 	};
9495 
9496 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9497 				      find_tunnel_id_hit, NULL, &ctx, true);
9498 
9499 	return ctx.tunnel;
9500 }
9501 
9502 static struct mlx5_flow_tunnel *
9503 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9504 			  const struct rte_flow_tunnel *app_tunnel)
9505 {
9506 	struct mlx5_priv *priv = dev->data->dev_private;
9507 	struct mlx5_indexed_pool *ipool;
9508 	struct mlx5_flow_tunnel *tunnel;
9509 	uint32_t id;
9510 
9511 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9512 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9513 	if (!tunnel)
9514 		return NULL;
9515 	if (id >= MLX5_MAX_TUNNELS) {
9516 		mlx5_ipool_free(ipool, id);
9517 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9518 		return NULL;
9519 	}
9520 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9521 					   priv->sh,
9522 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9523 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9524 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9525 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9526 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9527 	if (!tunnel->groups) {
9528 		mlx5_ipool_free(ipool, id);
9529 		return NULL;
9530 	}
9531 	/* initiate new PMD tunnel */
9532 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9533 	tunnel->tunnel_id = id;
9534 	tunnel->action.type = (typeof(tunnel->action.type))
9535 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9536 	tunnel->action.conf = tunnel;
9537 	tunnel->item.type = (typeof(tunnel->item.type))
9538 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9539 	tunnel->item.spec = tunnel;
9540 	tunnel->item.last = NULL;
9541 	tunnel->item.mask = NULL;
9542 
9543 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9544 		dev->data->port_id, tunnel->tunnel_id);
9545 
9546 	return tunnel;
9547 }
9548 
9549 struct tunnel_db_get_tunnel_ctx {
9550 	const struct rte_flow_tunnel *app_tunnel;
9551 	struct mlx5_flow_tunnel *tunnel;
9552 };
9553 
9554 static bool get_tunnel_match(struct rte_eth_dev *dev,
9555 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9556 {
9557 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9558 
9559 	RTE_SET_USED(dev);
9560 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9561 		       sizeof(*ctx->app_tunnel));
9562 }
9563 
9564 static void get_tunnel_hit(struct rte_eth_dev *dev,
9565 			   struct mlx5_flow_tunnel *tunnel, void *x)
9566 {
9567 	/* called under tunnel spinlock protection */
9568 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9569 
9570 	RTE_SET_USED(dev);
9571 	tunnel->refctn++;
9572 	ctx->tunnel = tunnel;
9573 }
9574 
9575 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9576 {
9577 	/* called under tunnel spinlock protection */
9578 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9579 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9580 
9581 	rte_spinlock_unlock(&thub->sl);
9582 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9583 	rte_spinlock_lock(&thub->sl);
9584 	if (ctx->tunnel) {
9585 		ctx->tunnel->refctn = 1;
9586 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9587 	}
9588 }
9589 
9590 
9591 static int
9592 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9593 		     const struct rte_flow_tunnel *app_tunnel,
9594 		     struct mlx5_flow_tunnel **tunnel)
9595 {
9596 	struct tunnel_db_get_tunnel_ctx ctx = {
9597 		.app_tunnel = app_tunnel,
9598 	};
9599 
9600 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9601 				      get_tunnel_miss, &ctx, true);
9602 	*tunnel = ctx.tunnel;
9603 	return ctx.tunnel ? 0 : -ENOMEM;
9604 }
9605 
9606 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9607 {
9608 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9609 
9610 	if (!thub)
9611 		return;
9612 	if (!LIST_EMPTY(&thub->tunnels))
9613 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9614 	mlx5_hlist_destroy(thub->groups);
9615 	mlx5_free(thub);
9616 }
9617 
9618 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9619 {
9620 	int err;
9621 	struct mlx5_flow_tunnel_hub *thub;
9622 
9623 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9624 			   0, SOCKET_ID_ANY);
9625 	if (!thub)
9626 		return -ENOMEM;
9627 	LIST_INIT(&thub->tunnels);
9628 	rte_spinlock_init(&thub->sl);
9629 	thub->groups = mlx5_hlist_create("flow groups", 64,
9630 					 false, true, sh,
9631 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9632 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9633 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9634 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9635 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9636 	if (!thub->groups) {
9637 		err = -rte_errno;
9638 		goto err;
9639 	}
9640 	sh->tunnel_hub = thub;
9641 
9642 	return 0;
9643 
9644 err:
9645 	if (thub->groups)
9646 		mlx5_hlist_destroy(thub->groups);
9647 	if (thub)
9648 		mlx5_free(thub);
9649 	return err;
9650 }
9651 
9652 static inline int
9653 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9654 			  struct rte_flow_tunnel *tunnel,
9655 			  struct rte_flow_error *error)
9656 {
9657 	struct mlx5_priv *priv = dev->data->dev_private;
9658 
9659 	if (!priv->sh->config.dv_flow_en)
9660 		return rte_flow_error_set(error, ENOTSUP,
9661 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9662 					  "flow DV interface is off");
9663 	if (!is_tunnel_offload_active(dev))
9664 		return rte_flow_error_set(error, ENOTSUP,
9665 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9666 					  "tunnel offload was not activated");
9667 	if (!tunnel)
9668 		return rte_flow_error_set(error, EINVAL,
9669 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9670 					  "no application tunnel");
9671 	switch (tunnel->type) {
9672 	default:
9673 		return rte_flow_error_set(error, EINVAL,
9674 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9675 					  "unsupported tunnel type");
9676 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9677 	case RTE_FLOW_ITEM_TYPE_GRE:
9678 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9679 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9680 		break;
9681 	}
9682 	return 0;
9683 }
9684 
9685 static int
9686 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9687 		    struct rte_flow_tunnel *app_tunnel,
9688 		    struct rte_flow_action **actions,
9689 		    uint32_t *num_of_actions,
9690 		    struct rte_flow_error *error)
9691 {
9692 	struct mlx5_flow_tunnel *tunnel;
9693 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9694 
9695 	if (ret)
9696 		return ret;
9697 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9698 	if (ret < 0) {
9699 		return rte_flow_error_set(error, ret,
9700 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9701 					  "failed to initialize pmd tunnel");
9702 	}
9703 	*actions = &tunnel->action;
9704 	*num_of_actions = 1;
9705 	return 0;
9706 }
9707 
9708 static int
9709 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9710 		       struct rte_flow_tunnel *app_tunnel,
9711 		       struct rte_flow_item **items,
9712 		       uint32_t *num_of_items,
9713 		       struct rte_flow_error *error)
9714 {
9715 	struct mlx5_flow_tunnel *tunnel;
9716 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9717 
9718 	if (ret)
9719 		return ret;
9720 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9721 	if (ret < 0) {
9722 		return rte_flow_error_set(error, ret,
9723 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9724 					  "failed to initialize pmd tunnel");
9725 	}
9726 	*items = &tunnel->item;
9727 	*num_of_items = 1;
9728 	return 0;
9729 }
9730 
9731 struct tunnel_db_element_release_ctx {
9732 	struct rte_flow_item *items;
9733 	struct rte_flow_action *actions;
9734 	uint32_t num_elements;
9735 	struct rte_flow_error *error;
9736 	int ret;
9737 };
9738 
9739 static bool
9740 tunnel_element_release_match(struct rte_eth_dev *dev,
9741 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9742 {
9743 	const struct tunnel_db_element_release_ctx *ctx = x;
9744 
9745 	RTE_SET_USED(dev);
9746 	if (ctx->num_elements != 1)
9747 		return false;
9748 	else if (ctx->items)
9749 		return ctx->items == &tunnel->item;
9750 	else if (ctx->actions)
9751 		return ctx->actions == &tunnel->action;
9752 
9753 	return false;
9754 }
9755 
9756 static void
9757 tunnel_element_release_hit(struct rte_eth_dev *dev,
9758 			   struct mlx5_flow_tunnel *tunnel, void *x)
9759 {
9760 	struct tunnel_db_element_release_ctx *ctx = x;
9761 	ctx->ret = 0;
9762 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9763 		mlx5_flow_tunnel_free(dev, tunnel);
9764 }
9765 
9766 static void
9767 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9768 {
9769 	struct tunnel_db_element_release_ctx *ctx = x;
9770 	RTE_SET_USED(dev);
9771 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9772 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9773 				      "invalid argument");
9774 }
9775 
9776 static int
9777 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9778 		       struct rte_flow_item *pmd_items,
9779 		       uint32_t num_items, struct rte_flow_error *err)
9780 {
9781 	struct tunnel_db_element_release_ctx ctx = {
9782 		.items = pmd_items,
9783 		.actions = NULL,
9784 		.num_elements = num_items,
9785 		.error = err,
9786 	};
9787 
9788 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9789 				      tunnel_element_release_hit,
9790 				      tunnel_element_release_miss, &ctx, false);
9791 
9792 	return ctx.ret;
9793 }
9794 
9795 static int
9796 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9797 			 struct rte_flow_action *pmd_actions,
9798 			 uint32_t num_actions, struct rte_flow_error *err)
9799 {
9800 	struct tunnel_db_element_release_ctx ctx = {
9801 		.items = NULL,
9802 		.actions = pmd_actions,
9803 		.num_elements = num_actions,
9804 		.error = err,
9805 	};
9806 
9807 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9808 				      tunnel_element_release_hit,
9809 				      tunnel_element_release_miss, &ctx, false);
9810 
9811 	return ctx.ret;
9812 }
9813 
9814 static int
9815 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9816 				  struct rte_mbuf *m,
9817 				  struct rte_flow_restore_info *info,
9818 				  struct rte_flow_error *err)
9819 {
9820 	uint64_t ol_flags = m->ol_flags;
9821 	const struct mlx5_flow_tbl_data_entry *tble;
9822 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9823 
9824 	if (!is_tunnel_offload_active(dev)) {
9825 		info->flags = 0;
9826 		return 0;
9827 	}
9828 
9829 	if ((ol_flags & mask) != mask)
9830 		goto err;
9831 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9832 	if (!tble) {
9833 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9834 			dev->data->port_id, m->hash.fdir.hi);
9835 		goto err;
9836 	}
9837 	MLX5_ASSERT(tble->tunnel);
9838 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9839 	info->group_id = tble->group_id;
9840 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9841 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9842 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9843 
9844 	return 0;
9845 
9846 err:
9847 	return rte_flow_error_set(err, EINVAL,
9848 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9849 				  "failed to get restore info");
9850 }
9851 
9852 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9853 static int
9854 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9855 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9856 			   __rte_unused struct rte_flow_action **actions,
9857 			   __rte_unused uint32_t *num_of_actions,
9858 			   __rte_unused struct rte_flow_error *error)
9859 {
9860 	return -ENOTSUP;
9861 }
9862 
9863 static int
9864 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9865 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9866 		       __rte_unused struct rte_flow_item **items,
9867 		       __rte_unused uint32_t *num_of_items,
9868 		       __rte_unused struct rte_flow_error *error)
9869 {
9870 	return -ENOTSUP;
9871 }
9872 
9873 static int
9874 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9875 			      __rte_unused struct rte_flow_item *pmd_items,
9876 			      __rte_unused uint32_t num_items,
9877 			      __rte_unused struct rte_flow_error *err)
9878 {
9879 	return -ENOTSUP;
9880 }
9881 
9882 static int
9883 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9884 				__rte_unused struct rte_flow_action *pmd_action,
9885 				__rte_unused uint32_t num_actions,
9886 				__rte_unused struct rte_flow_error *err)
9887 {
9888 	return -ENOTSUP;
9889 }
9890 
9891 static int
9892 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9893 				  __rte_unused struct rte_mbuf *m,
9894 				  __rte_unused struct rte_flow_restore_info *i,
9895 				  __rte_unused struct rte_flow_error *err)
9896 {
9897 	return -ENOTSUP;
9898 }
9899 
9900 static int
9901 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9902 			     __rte_unused struct rte_flow *flow,
9903 			     __rte_unused const struct rte_flow_attr *attr,
9904 			     __rte_unused const struct rte_flow_action *actions,
9905 			     __rte_unused uint32_t flow_idx,
9906 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9907 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9908 			     __rte_unused struct rte_flow_error *error)
9909 {
9910 	return -ENOTSUP;
9911 }
9912 
9913 static struct mlx5_flow_tunnel *
9914 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9915 		    __rte_unused uint32_t id)
9916 {
9917 	return NULL;
9918 }
9919 
9920 static void
9921 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9922 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9923 {
9924 }
9925 
9926 static uint32_t
9927 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9928 				__rte_unused const struct mlx5_flow_tunnel *t,
9929 				__rte_unused uint32_t group,
9930 				__rte_unused uint32_t *table,
9931 				struct rte_flow_error *error)
9932 {
9933 	return rte_flow_error_set(error, ENOTSUP,
9934 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9935 				  "tunnel offload requires DV support");
9936 }
9937 
9938 void
9939 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9940 			__rte_unused  uint16_t port_id)
9941 {
9942 }
9943 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9944 
9945 /* Flex flow item API */
9946 static struct rte_flow_item_flex_handle *
9947 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9948 			   const struct rte_flow_item_flex_conf *conf,
9949 			   struct rte_flow_error *error)
9950 {
9951 	static const char err_msg[] = "flex item creation unsupported";
9952 	struct rte_flow_attr attr = { .transfer = 0 };
9953 	const struct mlx5_flow_driver_ops *fops =
9954 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9955 
9956 	if (!fops->item_create) {
9957 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9958 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9959 				   NULL, err_msg);
9960 		return NULL;
9961 	}
9962 	return fops->item_create(dev, conf, error);
9963 }
9964 
9965 static int
9966 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9967 			    const struct rte_flow_item_flex_handle *handle,
9968 			    struct rte_flow_error *error)
9969 {
9970 	static const char err_msg[] = "flex item release unsupported";
9971 	struct rte_flow_attr attr = { .transfer = 0 };
9972 	const struct mlx5_flow_driver_ops *fops =
9973 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9974 
9975 	if (!fops->item_release) {
9976 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9977 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9978 				   NULL, err_msg);
9979 		return -rte_errno;
9980 	}
9981 	return fops->item_release(dev, handle, error);
9982 }
9983 
9984 static void
9985 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9986 {
9987 	int ret;
9988 	struct rte_flow_error error;
9989 
9990 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9991 		char *item_name;
9992 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9993 				    sizeof(item_name),
9994 				    (void *)(uintptr_t)item->type, &error);
9995 		if (ret > 0)
9996 			printf("%s ", item_name);
9997 		else
9998 			printf("%d\n", (int)item->type);
9999 	}
10000 	printf("END\n");
10001 }
10002 
10003 static int
10004 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
10005 {
10006 	const struct rte_flow_item_udp *spec = udp_item->spec;
10007 	const struct rte_flow_item_udp *mask = udp_item->mask;
10008 	uint16_t udp_dport = 0;
10009 
10010 	if (spec != NULL) {
10011 		if (!mask)
10012 			mask = &rte_flow_item_udp_mask;
10013 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
10014 				mask->hdr.dst_port);
10015 	}
10016 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
10017 }
10018 
10019 static const struct mlx5_flow_expand_node *
10020 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
10021 		unsigned int item_idx,
10022 		const struct mlx5_flow_expand_node graph[],
10023 		const struct mlx5_flow_expand_node *node)
10024 {
10025 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
10026 
10027 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
10028 			node != NULL &&
10029 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
10030 		/*
10031 		 * The expansion node is VXLAN and it is also the last
10032 		 * expandable item in the pattern, so need to continue
10033 		 * expansion of the inner tunnel.
10034 		 */
10035 		MLX5_ASSERT(item_idx > 0);
10036 		prev_item = pattern + item_idx - 1;
10037 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
10038 		if (mlx5_flow_is_std_vxlan_port(prev_item))
10039 			return &graph[MLX5_EXPANSION_STD_VXLAN];
10040 		return &graph[MLX5_EXPANSION_L3_VXLAN];
10041 	}
10042 	return node;
10043 }
10044 
10045 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
10046 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
10047 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
10048 };
10049 
10050 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
10051 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
10052 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
10053 	{ 9, 10, 11 }, { 12, 13, 14 },
10054 };
10055 
10056 /**
10057  * Discover the number of available flow priorities.
10058  *
10059  * @param dev
10060  *   Ethernet device.
10061  *
10062  * @return
10063  *   On success, number of available flow priorities.
10064  *   On failure, a negative errno-style code and rte_errno is set.
10065  */
10066 int
10067 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
10068 {
10069 	static const uint16_t vprio[] = {8, 16};
10070 	const struct mlx5_priv *priv = dev->data->dev_private;
10071 	const struct mlx5_flow_driver_ops *fops;
10072 	enum mlx5_flow_drv_type type;
10073 	int ret;
10074 
10075 	type = mlx5_flow_os_get_type();
10076 	if (type == MLX5_FLOW_TYPE_MAX) {
10077 		type = MLX5_FLOW_TYPE_VERBS;
10078 		if (priv->sh->cdev->config.devx && priv->sh->config.dv_flow_en)
10079 			type = MLX5_FLOW_TYPE_DV;
10080 	}
10081 	fops = flow_get_drv_ops(type);
10082 	if (fops->discover_priorities == NULL) {
10083 		DRV_LOG(ERR, "Priority discovery not supported");
10084 		rte_errno = ENOTSUP;
10085 		return -rte_errno;
10086 	}
10087 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
10088 	if (ret < 0)
10089 		return ret;
10090 	switch (ret) {
10091 	case 8:
10092 		ret = RTE_DIM(priority_map_3);
10093 		break;
10094 	case 16:
10095 		ret = RTE_DIM(priority_map_5);
10096 		break;
10097 	default:
10098 		rte_errno = ENOTSUP;
10099 		DRV_LOG(ERR,
10100 			"port %u maximum priority: %d expected 8/16",
10101 			dev->data->port_id, ret);
10102 		return -rte_errno;
10103 	}
10104 	DRV_LOG(INFO, "port %u supported flow priorities:"
10105 		" 0-%d for ingress or egress root table,"
10106 		" 0-%d for non-root table or transfer root table.",
10107 		dev->data->port_id, ret - 2,
10108 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
10109 	return ret;
10110 }
10111 
10112 /**
10113  * Adjust flow priority based on the highest layer and the request priority.
10114  *
10115  * @param[in] dev
10116  *   Pointer to the Ethernet device structure.
10117  * @param[in] priority
10118  *   The rule base priority.
10119  * @param[in] subpriority
10120  *   The priority based on the items.
10121  *
10122  * @return
10123  *   The new priority.
10124  */
10125 uint32_t
10126 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10127 			  uint32_t subpriority)
10128 {
10129 	uint32_t res = 0;
10130 	struct mlx5_priv *priv = dev->data->dev_private;
10131 
10132 	switch (priv->sh->flow_max_priority) {
10133 	case RTE_DIM(priority_map_3):
10134 		res = priority_map_3[priority][subpriority];
10135 		break;
10136 	case RTE_DIM(priority_map_5):
10137 		res = priority_map_5[priority][subpriority];
10138 		break;
10139 	}
10140 	return  res;
10141 }
10142