xref: /dpdk/drivers/net/mlx5/mlx5_flow.c (revision d84c3cf7662c6abca83df803aaa5136b73d3831d)
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 
809 static const struct rte_flow_ops mlx5_flow_ops = {
810 	.validate = mlx5_flow_validate,
811 	.create = mlx5_flow_create,
812 	.destroy = mlx5_flow_destroy,
813 	.flush = mlx5_flow_flush,
814 	.isolate = mlx5_flow_isolate,
815 	.query = mlx5_flow_query,
816 	.dev_dump = mlx5_flow_dev_dump,
817 	.get_aged_flows = mlx5_flow_get_aged_flows,
818 	.action_handle_create = mlx5_action_handle_create,
819 	.action_handle_destroy = mlx5_action_handle_destroy,
820 	.action_handle_update = mlx5_action_handle_update,
821 	.action_handle_query = mlx5_action_handle_query,
822 	.tunnel_decap_set = mlx5_flow_tunnel_decap_set,
823 	.tunnel_match = mlx5_flow_tunnel_match,
824 	.tunnel_action_decap_release = mlx5_flow_tunnel_action_release,
825 	.tunnel_item_release = mlx5_flow_tunnel_item_release,
826 	.get_restore_info = mlx5_flow_tunnel_get_restore_info,
827 	.flex_item_create = mlx5_flow_flex_item_create,
828 	.flex_item_release = mlx5_flow_flex_item_release,
829 };
830 
831 /* Tunnel information. */
832 struct mlx5_flow_tunnel_info {
833 	uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
834 	uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
835 };
836 
837 static struct mlx5_flow_tunnel_info tunnels_info[] = {
838 	{
839 		.tunnel = MLX5_FLOW_LAYER_VXLAN,
840 		.ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
841 	},
842 	{
843 		.tunnel = MLX5_FLOW_LAYER_GENEVE,
844 		.ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
845 	},
846 	{
847 		.tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
848 		.ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
849 	},
850 	{
851 		.tunnel = MLX5_FLOW_LAYER_GRE,
852 		.ptype = RTE_PTYPE_TUNNEL_GRE,
853 	},
854 	{
855 		.tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
856 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
857 	},
858 	{
859 		.tunnel = MLX5_FLOW_LAYER_MPLS,
860 		.ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
861 	},
862 	{
863 		.tunnel = MLX5_FLOW_LAYER_NVGRE,
864 		.ptype = RTE_PTYPE_TUNNEL_NVGRE,
865 	},
866 	{
867 		.tunnel = MLX5_FLOW_LAYER_IPIP,
868 		.ptype = RTE_PTYPE_TUNNEL_IP,
869 	},
870 	{
871 		.tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
872 		.ptype = RTE_PTYPE_TUNNEL_IP,
873 	},
874 	{
875 		.tunnel = MLX5_FLOW_LAYER_GTP,
876 		.ptype = RTE_PTYPE_TUNNEL_GTPU,
877 	},
878 };
879 
880 
881 
882 /**
883  * Translate tag ID to register.
884  *
885  * @param[in] dev
886  *   Pointer to the Ethernet device structure.
887  * @param[in] feature
888  *   The feature that request the register.
889  * @param[in] id
890  *   The request register ID.
891  * @param[out] error
892  *   Error description in case of any.
893  *
894  * @return
895  *   The request register on success, a negative errno
896  *   value otherwise and rte_errno is set.
897  */
898 int
899 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
900 		     enum mlx5_feature_name feature,
901 		     uint32_t id,
902 		     struct rte_flow_error *error)
903 {
904 	struct mlx5_priv *priv = dev->data->dev_private;
905 	struct mlx5_sh_config *config = &priv->sh->config;
906 	enum modify_reg start_reg;
907 	bool skip_mtr_reg = false;
908 
909 	switch (feature) {
910 	case MLX5_HAIRPIN_RX:
911 		return REG_B;
912 	case MLX5_HAIRPIN_TX:
913 		return REG_A;
914 	case MLX5_METADATA_RX:
915 		switch (config->dv_xmeta_en) {
916 		case MLX5_XMETA_MODE_LEGACY:
917 			return REG_B;
918 		case MLX5_XMETA_MODE_META16:
919 			return REG_C_0;
920 		case MLX5_XMETA_MODE_META32:
921 			return REG_C_1;
922 		}
923 		break;
924 	case MLX5_METADATA_TX:
925 		return REG_A;
926 	case MLX5_METADATA_FDB:
927 		switch (config->dv_xmeta_en) {
928 		case MLX5_XMETA_MODE_LEGACY:
929 			return REG_NON;
930 		case MLX5_XMETA_MODE_META16:
931 			return REG_C_0;
932 		case MLX5_XMETA_MODE_META32:
933 			return REG_C_1;
934 		}
935 		break;
936 	case MLX5_FLOW_MARK:
937 		switch (config->dv_xmeta_en) {
938 		case MLX5_XMETA_MODE_LEGACY:
939 			return REG_NON;
940 		case MLX5_XMETA_MODE_META16:
941 			return REG_C_1;
942 		case MLX5_XMETA_MODE_META32:
943 			return REG_C_0;
944 		}
945 		break;
946 	case MLX5_MTR_ID:
947 		/*
948 		 * If meter color and meter id share one register, flow match
949 		 * should use the meter color register for match.
950 		 */
951 		if (priv->mtr_reg_share)
952 			return priv->mtr_color_reg;
953 		else
954 			return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
955 			       REG_C_3;
956 	case MLX5_MTR_COLOR:
957 	case MLX5_ASO_FLOW_HIT:
958 	case MLX5_ASO_CONNTRACK:
959 	case MLX5_SAMPLE_ID:
960 		/* All features use the same REG_C. */
961 		MLX5_ASSERT(priv->mtr_color_reg != REG_NON);
962 		return priv->mtr_color_reg;
963 	case MLX5_COPY_MARK:
964 		/*
965 		 * Metadata COPY_MARK register using is in meter suffix sub
966 		 * flow while with meter. It's safe to share the same register.
967 		 */
968 		return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
969 	case MLX5_APP_TAG:
970 		/*
971 		 * If meter is enable, it will engage the register for color
972 		 * match and flow match. If meter color match is not using the
973 		 * REG_C_2, need to skip the REG_C_x be used by meter color
974 		 * match.
975 		 * If meter is disable, free to use all available registers.
976 		 */
977 		start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
978 			    (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
979 		skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
980 		if (id > (uint32_t)(REG_C_7 - start_reg))
981 			return rte_flow_error_set(error, EINVAL,
982 						  RTE_FLOW_ERROR_TYPE_ITEM,
983 						  NULL, "invalid tag id");
984 		if (priv->sh->flow_mreg_c[id + start_reg - REG_C_0] == REG_NON)
985 			return rte_flow_error_set(error, ENOTSUP,
986 						  RTE_FLOW_ERROR_TYPE_ITEM,
987 						  NULL, "unsupported tag id");
988 		/*
989 		 * This case means meter is using the REG_C_x great than 2.
990 		 * Take care not to conflict with meter color REG_C_x.
991 		 * If the available index REG_C_y >= REG_C_x, skip the
992 		 * color register.
993 		 */
994 		if (skip_mtr_reg && priv->sh->flow_mreg_c
995 		    [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
996 			if (id >= (uint32_t)(REG_C_7 - start_reg))
997 				return rte_flow_error_set(error, EINVAL,
998 						       RTE_FLOW_ERROR_TYPE_ITEM,
999 							NULL, "invalid tag id");
1000 			if (priv->sh->flow_mreg_c
1001 			    [id + 1 + start_reg - REG_C_0] != REG_NON)
1002 				return priv->sh->flow_mreg_c
1003 					       [id + 1 + start_reg - REG_C_0];
1004 			return rte_flow_error_set(error, ENOTSUP,
1005 						  RTE_FLOW_ERROR_TYPE_ITEM,
1006 						  NULL, "unsupported tag id");
1007 		}
1008 		return priv->sh->flow_mreg_c[id + start_reg - REG_C_0];
1009 	}
1010 	MLX5_ASSERT(false);
1011 	return rte_flow_error_set(error, EINVAL,
1012 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1013 				  NULL, "invalid feature name");
1014 }
1015 
1016 /**
1017  * Check extensive flow metadata register support.
1018  *
1019  * @param dev
1020  *   Pointer to rte_eth_dev structure.
1021  *
1022  * @return
1023  *   True if device supports extensive flow metadata register, otherwise false.
1024  */
1025 bool
1026 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
1027 {
1028 	struct mlx5_priv *priv = dev->data->dev_private;
1029 
1030 	/*
1031 	 * Having available reg_c can be regarded inclusively as supporting
1032 	 * extensive flow metadata register, which could mean,
1033 	 * - metadata register copy action by modify header.
1034 	 * - 16 modify header actions is supported.
1035 	 * - reg_c's are preserved across different domain (FDB and NIC) on
1036 	 *   packet loopback by flow lookup miss.
1037 	 */
1038 	return priv->sh->flow_mreg_c[2] != REG_NON;
1039 }
1040 
1041 /**
1042  * Get the lowest priority.
1043  *
1044  * @param[in] dev
1045  *   Pointer to the Ethernet device structure.
1046  * @param[in] attributes
1047  *   Pointer to device flow rule attributes.
1048  *
1049  * @return
1050  *   The value of lowest priority of flow.
1051  */
1052 uint32_t
1053 mlx5_get_lowest_priority(struct rte_eth_dev *dev,
1054 			  const struct rte_flow_attr *attr)
1055 {
1056 	struct mlx5_priv *priv = dev->data->dev_private;
1057 
1058 	if (!attr->group && !attr->transfer)
1059 		return priv->sh->flow_max_priority - 2;
1060 	return MLX5_NON_ROOT_FLOW_MAX_PRIO - 1;
1061 }
1062 
1063 /**
1064  * Calculate matcher priority of the flow.
1065  *
1066  * @param[in] dev
1067  *   Pointer to the Ethernet device structure.
1068  * @param[in] attr
1069  *   Pointer to device flow rule attributes.
1070  * @param[in] subpriority
1071  *   The priority based on the items.
1072  * @param[in] external
1073  *   Flow is user flow.
1074  * @return
1075  *   The matcher priority of the flow.
1076  */
1077 uint16_t
1078 mlx5_get_matcher_priority(struct rte_eth_dev *dev,
1079 			  const struct rte_flow_attr *attr,
1080 			  uint32_t subpriority, bool external)
1081 {
1082 	uint16_t priority = (uint16_t)attr->priority;
1083 	struct mlx5_priv *priv = dev->data->dev_private;
1084 
1085 	if (!attr->group && !attr->transfer) {
1086 		if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1087 			priority = priv->sh->flow_max_priority - 1;
1088 		return mlx5_os_flow_adjust_priority(dev, priority, subpriority);
1089 	} else if (!external && attr->transfer && attr->group == 0 &&
1090 		   attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR) {
1091 		return (priv->sh->flow_max_priority - 1) * 3;
1092 	}
1093 	if (attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1094 		priority = MLX5_NON_ROOT_FLOW_MAX_PRIO;
1095 	return priority * 3 + subpriority;
1096 }
1097 
1098 /**
1099  * Verify the @p item specifications (spec, last, mask) are compatible with the
1100  * NIC capabilities.
1101  *
1102  * @param[in] item
1103  *   Item specification.
1104  * @param[in] mask
1105  *   @p item->mask or flow default bit-masks.
1106  * @param[in] nic_mask
1107  *   Bit-masks covering supported fields by the NIC to compare with user mask.
1108  * @param[in] size
1109  *   Bit-masks size in bytes.
1110  * @param[in] range_accepted
1111  *   True if range of values is accepted for specific fields, false otherwise.
1112  * @param[out] error
1113  *   Pointer to error structure.
1114  *
1115  * @return
1116  *   0 on success, a negative errno value otherwise and rte_errno is set.
1117  */
1118 int
1119 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
1120 			  const uint8_t *mask,
1121 			  const uint8_t *nic_mask,
1122 			  unsigned int size,
1123 			  bool range_accepted,
1124 			  struct rte_flow_error *error)
1125 {
1126 	unsigned int i;
1127 
1128 	MLX5_ASSERT(nic_mask);
1129 	for (i = 0; i < size; ++i)
1130 		if ((nic_mask[i] | mask[i]) != nic_mask[i])
1131 			return rte_flow_error_set(error, ENOTSUP,
1132 						  RTE_FLOW_ERROR_TYPE_ITEM,
1133 						  item,
1134 						  "mask enables non supported"
1135 						  " bits");
1136 	if (!item->spec && (item->mask || item->last))
1137 		return rte_flow_error_set(error, EINVAL,
1138 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1139 					  "mask/last without a spec is not"
1140 					  " supported");
1141 	if (item->spec && item->last && !range_accepted) {
1142 		uint8_t spec[size];
1143 		uint8_t last[size];
1144 		unsigned int i;
1145 		int ret;
1146 
1147 		for (i = 0; i < size; ++i) {
1148 			spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
1149 			last[i] = ((const uint8_t *)item->last)[i] & mask[i];
1150 		}
1151 		ret = memcmp(spec, last, size);
1152 		if (ret != 0)
1153 			return rte_flow_error_set(error, EINVAL,
1154 						  RTE_FLOW_ERROR_TYPE_ITEM,
1155 						  item,
1156 						  "range is not valid");
1157 	}
1158 	return 0;
1159 }
1160 
1161 /**
1162  * Adjust the hash fields according to the @p flow information.
1163  *
1164  * @param[in] dev_flow.
1165  *   Pointer to the mlx5_flow.
1166  * @param[in] tunnel
1167  *   1 when the hash field is for a tunnel item.
1168  * @param[in] layer_types
1169  *   RTE_ETH_RSS_* types.
1170  * @param[in] hash_fields
1171  *   Item hash fields.
1172  *
1173  * @return
1174  *   The hash fields that should be used.
1175  */
1176 uint64_t
1177 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
1178 			    int tunnel __rte_unused, uint64_t layer_types,
1179 			    uint64_t hash_fields)
1180 {
1181 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1182 	int rss_request_inner = rss_desc->level >= 2;
1183 
1184 	/* Check RSS hash level for tunnel. */
1185 	if (tunnel && rss_request_inner)
1186 		hash_fields |= IBV_RX_HASH_INNER;
1187 	else if (tunnel || rss_request_inner)
1188 		return 0;
1189 #endif
1190 	/* Check if requested layer matches RSS hash fields. */
1191 	if (!(rss_desc->types & layer_types))
1192 		return 0;
1193 	return hash_fields;
1194 }
1195 
1196 /**
1197  * Lookup and set the ptype in the data Rx part.  A single Ptype can be used,
1198  * if several tunnel rules are used on this queue, the tunnel ptype will be
1199  * cleared.
1200  *
1201  * @param rxq_ctrl
1202  *   Rx queue to update.
1203  */
1204 static void
1205 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
1206 {
1207 	unsigned int i;
1208 	uint32_t tunnel_ptype = 0;
1209 
1210 	/* Look up for the ptype to use. */
1211 	for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
1212 		if (!rxq_ctrl->flow_tunnels_n[i])
1213 			continue;
1214 		if (!tunnel_ptype) {
1215 			tunnel_ptype = tunnels_info[i].ptype;
1216 		} else {
1217 			tunnel_ptype = 0;
1218 			break;
1219 		}
1220 	}
1221 	rxq_ctrl->rxq.tunnel = tunnel_ptype;
1222 }
1223 
1224 /**
1225  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the device
1226  * flow.
1227  *
1228  * @param[in] dev
1229  *   Pointer to the Ethernet device structure.
1230  * @param[in] dev_handle
1231  *   Pointer to device flow handle structure.
1232  */
1233 void
1234 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
1235 		       struct mlx5_flow_handle *dev_handle)
1236 {
1237 	struct mlx5_priv *priv = dev->data->dev_private;
1238 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1239 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1240 	unsigned int i;
1241 
1242 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1243 		struct mlx5_hrxq *hrxq;
1244 
1245 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1246 			      dev_handle->rix_hrxq);
1247 		if (hrxq)
1248 			ind_tbl = hrxq->ind_table;
1249 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1250 		struct mlx5_shared_action_rss *shared_rss;
1251 
1252 		shared_rss = mlx5_ipool_get
1253 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1254 			 dev_handle->rix_srss);
1255 		if (shared_rss)
1256 			ind_tbl = shared_rss->ind_tbl;
1257 	}
1258 	if (!ind_tbl)
1259 		return;
1260 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1261 		int idx = ind_tbl->queues[i];
1262 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1263 
1264 		MLX5_ASSERT(rxq_ctrl != NULL);
1265 		if (rxq_ctrl == NULL)
1266 			continue;
1267 		/*
1268 		 * To support metadata register copy on Tx loopback,
1269 		 * this must be always enabled (metadata may arive
1270 		 * from other port - not from local flows only.
1271 		 */
1272 		if (tunnel) {
1273 			unsigned int j;
1274 
1275 			/* Increase the counter matching the flow. */
1276 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1277 				if ((tunnels_info[j].tunnel &
1278 				     dev_handle->layers) ==
1279 				    tunnels_info[j].tunnel) {
1280 					rxq_ctrl->flow_tunnels_n[j]++;
1281 					break;
1282 				}
1283 			}
1284 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1285 		}
1286 	}
1287 }
1288 
1289 static void
1290 flow_rxq_mark_flag_set(struct rte_eth_dev *dev)
1291 {
1292 	struct mlx5_priv *priv = dev->data->dev_private;
1293 	struct mlx5_rxq_ctrl *rxq_ctrl;
1294 
1295 	if (priv->mark_enabled)
1296 		return;
1297 	LIST_FOREACH(rxq_ctrl, &priv->rxqsctrl, next) {
1298 		rxq_ctrl->rxq.mark = 1;
1299 	}
1300 	priv->mark_enabled = 1;
1301 }
1302 
1303 /**
1304  * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
1305  *
1306  * @param[in] dev
1307  *   Pointer to the Ethernet device structure.
1308  * @param[in] flow
1309  *   Pointer to flow structure.
1310  */
1311 static void
1312 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
1313 {
1314 	struct mlx5_priv *priv = dev->data->dev_private;
1315 	uint32_t handle_idx;
1316 	struct mlx5_flow_handle *dev_handle;
1317 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1318 
1319 	MLX5_ASSERT(wks);
1320 	if (wks->mark)
1321 		flow_rxq_mark_flag_set(dev);
1322 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1323 		       handle_idx, dev_handle, next)
1324 		flow_drv_rxq_flags_set(dev, dev_handle);
1325 }
1326 
1327 /**
1328  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1329  * device flow if no other flow uses it with the same kind of request.
1330  *
1331  * @param dev
1332  *   Pointer to Ethernet device.
1333  * @param[in] dev_handle
1334  *   Pointer to the device flow handle structure.
1335  */
1336 static void
1337 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
1338 			struct mlx5_flow_handle *dev_handle)
1339 {
1340 	struct mlx5_priv *priv = dev->data->dev_private;
1341 	const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
1342 	struct mlx5_ind_table_obj *ind_tbl = NULL;
1343 	unsigned int i;
1344 
1345 	if (dev_handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1346 		struct mlx5_hrxq *hrxq;
1347 
1348 		hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
1349 			      dev_handle->rix_hrxq);
1350 		if (hrxq)
1351 			ind_tbl = hrxq->ind_table;
1352 	} else if (dev_handle->fate_action == MLX5_FLOW_FATE_SHARED_RSS) {
1353 		struct mlx5_shared_action_rss *shared_rss;
1354 
1355 		shared_rss = mlx5_ipool_get
1356 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
1357 			 dev_handle->rix_srss);
1358 		if (shared_rss)
1359 			ind_tbl = shared_rss->ind_tbl;
1360 	}
1361 	if (!ind_tbl)
1362 		return;
1363 	MLX5_ASSERT(dev->data->dev_started);
1364 	for (i = 0; i != ind_tbl->queues_n; ++i) {
1365 		int idx = ind_tbl->queues[i];
1366 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev, idx);
1367 
1368 		MLX5_ASSERT(rxq_ctrl != NULL);
1369 		if (rxq_ctrl == NULL)
1370 			continue;
1371 		if (tunnel) {
1372 			unsigned int j;
1373 
1374 			/* Decrease the counter matching the flow. */
1375 			for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
1376 				if ((tunnels_info[j].tunnel &
1377 				     dev_handle->layers) ==
1378 				    tunnels_info[j].tunnel) {
1379 					rxq_ctrl->flow_tunnels_n[j]--;
1380 					break;
1381 				}
1382 			}
1383 			flow_rxq_tunnel_ptype_update(rxq_ctrl);
1384 		}
1385 	}
1386 }
1387 
1388 /**
1389  * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
1390  * @p flow if no other flow uses it with the same kind of request.
1391  *
1392  * @param dev
1393  *   Pointer to Ethernet device.
1394  * @param[in] flow
1395  *   Pointer to the flow.
1396  */
1397 static void
1398 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
1399 {
1400 	struct mlx5_priv *priv = dev->data->dev_private;
1401 	uint32_t handle_idx;
1402 	struct mlx5_flow_handle *dev_handle;
1403 
1404 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1405 		       handle_idx, dev_handle, next)
1406 		flow_drv_rxq_flags_trim(dev, dev_handle);
1407 }
1408 
1409 /**
1410  * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
1411  *
1412  * @param dev
1413  *   Pointer to Ethernet device.
1414  */
1415 static void
1416 flow_rxq_flags_clear(struct rte_eth_dev *dev)
1417 {
1418 	struct mlx5_priv *priv = dev->data->dev_private;
1419 	unsigned int i;
1420 
1421 	for (i = 0; i != priv->rxqs_n; ++i) {
1422 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1423 		unsigned int j;
1424 
1425 		if (rxq == NULL || rxq->ctrl == NULL)
1426 			continue;
1427 		rxq->ctrl->rxq.mark = 0;
1428 		for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
1429 			rxq->ctrl->flow_tunnels_n[j] = 0;
1430 		rxq->ctrl->rxq.tunnel = 0;
1431 	}
1432 	priv->mark_enabled = 0;
1433 }
1434 
1435 /**
1436  * Set the Rx queue dynamic metadata (mask and offset) for a flow
1437  *
1438  * @param[in] dev
1439  *   Pointer to the Ethernet device structure.
1440  */
1441 void
1442 mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev)
1443 {
1444 	struct mlx5_priv *priv = dev->data->dev_private;
1445 	unsigned int i;
1446 
1447 	for (i = 0; i != priv->rxqs_n; ++i) {
1448 		struct mlx5_rxq_priv *rxq = mlx5_rxq_get(dev, i);
1449 		struct mlx5_rxq_data *data;
1450 
1451 		if (rxq == NULL || rxq->ctrl == NULL)
1452 			continue;
1453 		data = &rxq->ctrl->rxq;
1454 		if (!rte_flow_dynf_metadata_avail()) {
1455 			data->dynf_meta = 0;
1456 			data->flow_meta_mask = 0;
1457 			data->flow_meta_offset = -1;
1458 			data->flow_meta_port_mask = 0;
1459 		} else {
1460 			data->dynf_meta = 1;
1461 			data->flow_meta_mask = rte_flow_dynf_metadata_mask;
1462 			data->flow_meta_offset = rte_flow_dynf_metadata_offs;
1463 			data->flow_meta_port_mask = priv->sh->dv_meta_mask;
1464 		}
1465 	}
1466 }
1467 
1468 /*
1469  * return a pointer to the desired action in the list of actions.
1470  *
1471  * @param[in] actions
1472  *   The list of actions to search the action in.
1473  * @param[in] action
1474  *   The action to find.
1475  *
1476  * @return
1477  *   Pointer to the action in the list, if found. NULL otherwise.
1478  */
1479 const struct rte_flow_action *
1480 mlx5_flow_find_action(const struct rte_flow_action *actions,
1481 		      enum rte_flow_action_type action)
1482 {
1483 	if (actions == NULL)
1484 		return NULL;
1485 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
1486 		if (actions->type == action)
1487 			return actions;
1488 	return NULL;
1489 }
1490 
1491 /*
1492  * Validate the flag action.
1493  *
1494  * @param[in] action_flags
1495  *   Bit-fields that holds the actions detected until now.
1496  * @param[in] attr
1497  *   Attributes of flow that includes this action.
1498  * @param[out] error
1499  *   Pointer to error structure.
1500  *
1501  * @return
1502  *   0 on success, a negative errno value otherwise and rte_errno is set.
1503  */
1504 int
1505 mlx5_flow_validate_action_flag(uint64_t action_flags,
1506 			       const struct rte_flow_attr *attr,
1507 			       struct rte_flow_error *error)
1508 {
1509 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1510 		return rte_flow_error_set(error, EINVAL,
1511 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1512 					  "can't mark and flag in same flow");
1513 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1514 		return rte_flow_error_set(error, EINVAL,
1515 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1516 					  "can't have 2 flag"
1517 					  " actions in same flow");
1518 	if (attr->egress)
1519 		return rte_flow_error_set(error, ENOTSUP,
1520 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1521 					  "flag action not supported for "
1522 					  "egress");
1523 	return 0;
1524 }
1525 
1526 /*
1527  * Validate the mark action.
1528  *
1529  * @param[in] action
1530  *   Pointer to the queue action.
1531  * @param[in] action_flags
1532  *   Bit-fields that holds the actions detected until now.
1533  * @param[in] attr
1534  *   Attributes of flow that includes this action.
1535  * @param[out] error
1536  *   Pointer to error structure.
1537  *
1538  * @return
1539  *   0 on success, a negative errno value otherwise and rte_errno is set.
1540  */
1541 int
1542 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
1543 			       uint64_t action_flags,
1544 			       const struct rte_flow_attr *attr,
1545 			       struct rte_flow_error *error)
1546 {
1547 	const struct rte_flow_action_mark *mark = action->conf;
1548 
1549 	if (!mark)
1550 		return rte_flow_error_set(error, EINVAL,
1551 					  RTE_FLOW_ERROR_TYPE_ACTION,
1552 					  action,
1553 					  "configuration cannot be null");
1554 	if (mark->id >= MLX5_FLOW_MARK_MAX)
1555 		return rte_flow_error_set(error, EINVAL,
1556 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1557 					  &mark->id,
1558 					  "mark id must in 0 <= id < "
1559 					  RTE_STR(MLX5_FLOW_MARK_MAX));
1560 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
1561 		return rte_flow_error_set(error, EINVAL,
1562 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1563 					  "can't flag and mark in same flow");
1564 	if (action_flags & MLX5_FLOW_ACTION_MARK)
1565 		return rte_flow_error_set(error, EINVAL,
1566 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1567 					  "can't have 2 mark actions in same"
1568 					  " flow");
1569 	if (attr->egress)
1570 		return rte_flow_error_set(error, ENOTSUP,
1571 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1572 					  "mark action not supported for "
1573 					  "egress");
1574 	return 0;
1575 }
1576 
1577 /*
1578  * Validate the drop action.
1579  *
1580  * @param[in] action_flags
1581  *   Bit-fields that holds the actions detected until now.
1582  * @param[in] attr
1583  *   Attributes of flow that includes this action.
1584  * @param[out] error
1585  *   Pointer to error structure.
1586  *
1587  * @return
1588  *   0 on success, a negative errno value otherwise and rte_errno is set.
1589  */
1590 int
1591 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1592 			       const struct rte_flow_attr *attr,
1593 			       struct rte_flow_error *error)
1594 {
1595 	if (attr->egress)
1596 		return rte_flow_error_set(error, ENOTSUP,
1597 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1598 					  "drop action not supported for "
1599 					  "egress");
1600 	return 0;
1601 }
1602 
1603 /*
1604  * Validate the queue action.
1605  *
1606  * @param[in] action
1607  *   Pointer to the queue action.
1608  * @param[in] action_flags
1609  *   Bit-fields that holds the actions detected until now.
1610  * @param[in] dev
1611  *   Pointer to the Ethernet device structure.
1612  * @param[in] attr
1613  *   Attributes of flow that includes this action.
1614  * @param[out] error
1615  *   Pointer to error structure.
1616  *
1617  * @return
1618  *   0 on success, a negative errno value otherwise and rte_errno is set.
1619  */
1620 int
1621 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1622 				uint64_t action_flags,
1623 				struct rte_eth_dev *dev,
1624 				const struct rte_flow_attr *attr,
1625 				struct rte_flow_error *error)
1626 {
1627 	struct mlx5_priv *priv = dev->data->dev_private;
1628 	const struct rte_flow_action_queue *queue = action->conf;
1629 
1630 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1631 		return rte_flow_error_set(error, EINVAL,
1632 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1633 					  "can't have 2 fate actions in"
1634 					  " same flow");
1635 	if (!priv->rxqs_n)
1636 		return rte_flow_error_set(error, EINVAL,
1637 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1638 					  NULL, "No Rx queues configured");
1639 	if (queue->index >= priv->rxqs_n)
1640 		return rte_flow_error_set(error, EINVAL,
1641 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1642 					  &queue->index,
1643 					  "queue index out of range");
1644 	if (mlx5_rxq_get(dev, queue->index) == NULL)
1645 		return rte_flow_error_set(error, EINVAL,
1646 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1647 					  &queue->index,
1648 					  "queue is not configured");
1649 	if (attr->egress)
1650 		return rte_flow_error_set(error, ENOTSUP,
1651 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1652 					  "queue action not supported for "
1653 					  "egress");
1654 	return 0;
1655 }
1656 
1657 /**
1658  * Validate queue numbers for device RSS.
1659  *
1660  * @param[in] dev
1661  *   Configured device.
1662  * @param[in] queues
1663  *   Array of queue numbers.
1664  * @param[in] queues_n
1665  *   Size of the @p queues array.
1666  * @param[out] error
1667  *   On error, filled with a textual error description.
1668  * @param[out] queue
1669  *   On error, filled with an offending queue index in @p queues array.
1670  *
1671  * @return
1672  *   0 on success, a negative errno code on error.
1673  */
1674 static int
1675 mlx5_validate_rss_queues(struct rte_eth_dev *dev,
1676 			 const uint16_t *queues, uint32_t queues_n,
1677 			 const char **error, uint32_t *queue_idx)
1678 {
1679 	const struct mlx5_priv *priv = dev->data->dev_private;
1680 	enum mlx5_rxq_type rxq_type = MLX5_RXQ_TYPE_UNDEFINED;
1681 	uint32_t i;
1682 
1683 	for (i = 0; i != queues_n; ++i) {
1684 		struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_ctrl_get(dev,
1685 								   queues[i]);
1686 
1687 		if (queues[i] >= priv->rxqs_n) {
1688 			*error = "queue index out of range";
1689 			*queue_idx = i;
1690 			return -EINVAL;
1691 		}
1692 		if (rxq_ctrl == NULL) {
1693 			*error =  "queue is not configured";
1694 			*queue_idx = i;
1695 			return -EINVAL;
1696 		}
1697 		if (i == 0)
1698 			rxq_type = rxq_ctrl->type;
1699 		if (rxq_type != rxq_ctrl->type) {
1700 			*error = "combining hairpin and regular RSS queues is not supported";
1701 			*queue_idx = i;
1702 			return -ENOTSUP;
1703 		}
1704 	}
1705 	return 0;
1706 }
1707 
1708 /*
1709  * Validate the rss action.
1710  *
1711  * @param[in] dev
1712  *   Pointer to the Ethernet device structure.
1713  * @param[in] action
1714  *   Pointer to the queue action.
1715  * @param[out] error
1716  *   Pointer to error structure.
1717  *
1718  * @return
1719  *   0 on success, a negative errno value otherwise and rte_errno is set.
1720  */
1721 int
1722 mlx5_validate_action_rss(struct rte_eth_dev *dev,
1723 			 const struct rte_flow_action *action,
1724 			 struct rte_flow_error *error)
1725 {
1726 	struct mlx5_priv *priv = dev->data->dev_private;
1727 	const struct rte_flow_action_rss *rss = action->conf;
1728 	int ret;
1729 	const char *message;
1730 	uint32_t queue_idx;
1731 
1732 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1733 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1734 		return rte_flow_error_set(error, ENOTSUP,
1735 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1736 					  &rss->func,
1737 					  "RSS hash function not supported");
1738 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1739 	if (rss->level > 2)
1740 #else
1741 	if (rss->level > 1)
1742 #endif
1743 		return rte_flow_error_set(error, ENOTSUP,
1744 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1745 					  &rss->level,
1746 					  "tunnel RSS is not supported");
1747 	/* allow RSS key_len 0 in case of NULL (default) RSS key. */
1748 	if (rss->key_len == 0 && rss->key != NULL)
1749 		return rte_flow_error_set(error, ENOTSUP,
1750 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1751 					  &rss->key_len,
1752 					  "RSS hash key length 0");
1753 	if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1754 		return rte_flow_error_set(error, ENOTSUP,
1755 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1756 					  &rss->key_len,
1757 					  "RSS hash key too small");
1758 	if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1759 		return rte_flow_error_set(error, ENOTSUP,
1760 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1761 					  &rss->key_len,
1762 					  "RSS hash key too large");
1763 	if (rss->queue_num > priv->sh->dev_cap.ind_table_max_size)
1764 		return rte_flow_error_set(error, ENOTSUP,
1765 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1766 					  &rss->queue_num,
1767 					  "number of queues too large");
1768 	if (rss->types & MLX5_RSS_HF_MASK)
1769 		return rte_flow_error_set(error, ENOTSUP,
1770 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1771 					  &rss->types,
1772 					  "some RSS protocols are not"
1773 					  " supported");
1774 	if ((rss->types & (RTE_ETH_RSS_L3_SRC_ONLY | RTE_ETH_RSS_L3_DST_ONLY)) &&
1775 	    !(rss->types & RTE_ETH_RSS_IP))
1776 		return rte_flow_error_set(error, EINVAL,
1777 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1778 					  "L3 partial RSS requested but L3 RSS"
1779 					  " type not specified");
1780 	if ((rss->types & (RTE_ETH_RSS_L4_SRC_ONLY | RTE_ETH_RSS_L4_DST_ONLY)) &&
1781 	    !(rss->types & (RTE_ETH_RSS_UDP | RTE_ETH_RSS_TCP)))
1782 		return rte_flow_error_set(error, EINVAL,
1783 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1784 					  "L4 partial RSS requested but L4 RSS"
1785 					  " type not specified");
1786 	if (!priv->rxqs_n)
1787 		return rte_flow_error_set(error, EINVAL,
1788 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1789 					  NULL, "No Rx queues configured");
1790 	if (!rss->queue_num)
1791 		return rte_flow_error_set(error, EINVAL,
1792 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1793 					  NULL, "No queues configured");
1794 	ret = mlx5_validate_rss_queues(dev, rss->queue, rss->queue_num,
1795 				       &message, &queue_idx);
1796 	if (ret != 0) {
1797 		return rte_flow_error_set(error, -ret,
1798 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1799 					  &rss->queue[queue_idx], message);
1800 	}
1801 	return 0;
1802 }
1803 
1804 /*
1805  * Validate the rss action.
1806  *
1807  * @param[in] action
1808  *   Pointer to the queue action.
1809  * @param[in] action_flags
1810  *   Bit-fields that holds the actions detected until now.
1811  * @param[in] dev
1812  *   Pointer to the Ethernet device structure.
1813  * @param[in] attr
1814  *   Attributes of flow that includes this action.
1815  * @param[in] item_flags
1816  *   Items that were detected.
1817  * @param[out] error
1818  *   Pointer to error structure.
1819  *
1820  * @return
1821  *   0 on success, a negative errno value otherwise and rte_errno is set.
1822  */
1823 int
1824 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1825 			      uint64_t action_flags,
1826 			      struct rte_eth_dev *dev,
1827 			      const struct rte_flow_attr *attr,
1828 			      uint64_t item_flags,
1829 			      struct rte_flow_error *error)
1830 {
1831 	const struct rte_flow_action_rss *rss = action->conf;
1832 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1833 	int ret;
1834 
1835 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1836 		return rte_flow_error_set(error, EINVAL,
1837 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1838 					  "can't have 2 fate actions"
1839 					  " in same flow");
1840 	ret = mlx5_validate_action_rss(dev, action, error);
1841 	if (ret)
1842 		return ret;
1843 	if (attr->egress)
1844 		return rte_flow_error_set(error, ENOTSUP,
1845 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1846 					  "rss action not supported for "
1847 					  "egress");
1848 	if (rss->level > 1 && !tunnel)
1849 		return rte_flow_error_set(error, EINVAL,
1850 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1851 					  "inner RSS is not supported for "
1852 					  "non-tunnel flows");
1853 	if ((item_flags & MLX5_FLOW_LAYER_ECPRI) &&
1854 	    !(item_flags & MLX5_FLOW_LAYER_INNER_L4_UDP)) {
1855 		return rte_flow_error_set(error, EINVAL,
1856 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1857 					  "RSS on eCPRI is not supported now");
1858 	}
1859 	if ((item_flags & MLX5_FLOW_LAYER_MPLS) &&
1860 	    !(item_flags &
1861 	      (MLX5_FLOW_LAYER_INNER_L2 | MLX5_FLOW_LAYER_INNER_L3)) &&
1862 	    rss->level > 1)
1863 		return rte_flow_error_set(error, EINVAL,
1864 					  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1865 					  "MPLS inner RSS needs to specify inner L2/L3 items after MPLS in pattern");
1866 	return 0;
1867 }
1868 
1869 /*
1870  * Validate the default miss action.
1871  *
1872  * @param[in] action_flags
1873  *   Bit-fields that holds the actions detected until now.
1874  * @param[out] error
1875  *   Pointer to error structure.
1876  *
1877  * @return
1878  *   0 on success, a negative errno value otherwise and rte_errno is set.
1879  */
1880 int
1881 mlx5_flow_validate_action_default_miss(uint64_t action_flags,
1882 				const struct rte_flow_attr *attr,
1883 				struct rte_flow_error *error)
1884 {
1885 	if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1886 		return rte_flow_error_set(error, EINVAL,
1887 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1888 					  "can't have 2 fate actions in"
1889 					  " same flow");
1890 	if (attr->egress)
1891 		return rte_flow_error_set(error, ENOTSUP,
1892 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1893 					  "default miss action not supported "
1894 					  "for egress");
1895 	if (attr->group)
1896 		return rte_flow_error_set(error, ENOTSUP,
1897 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP, NULL,
1898 					  "only group 0 is supported");
1899 	if (attr->transfer)
1900 		return rte_flow_error_set(error, ENOTSUP,
1901 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1902 					  NULL, "transfer is not supported");
1903 	return 0;
1904 }
1905 
1906 /*
1907  * Validate the count action.
1908  *
1909  * @param[in] dev
1910  *   Pointer to the Ethernet device structure.
1911  * @param[in] attr
1912  *   Attributes of flow that includes this action.
1913  * @param[out] error
1914  *   Pointer to error structure.
1915  *
1916  * @return
1917  *   0 on success, a negative errno value otherwise and rte_errno is set.
1918  */
1919 int
1920 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1921 				const struct rte_flow_attr *attr,
1922 				struct rte_flow_error *error)
1923 {
1924 	if (attr->egress)
1925 		return rte_flow_error_set(error, ENOTSUP,
1926 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1927 					  "count action not supported for "
1928 					  "egress");
1929 	return 0;
1930 }
1931 
1932 /*
1933  * Validate the ASO CT action.
1934  *
1935  * @param[in] dev
1936  *   Pointer to the Ethernet device structure.
1937  * @param[in] conntrack
1938  *   Pointer to the CT action profile.
1939  * @param[out] error
1940  *   Pointer to error structure.
1941  *
1942  * @return
1943  *   0 on success, a negative errno value otherwise and rte_errno is set.
1944  */
1945 int
1946 mlx5_validate_action_ct(struct rte_eth_dev *dev,
1947 			const struct rte_flow_action_conntrack *conntrack,
1948 			struct rte_flow_error *error)
1949 {
1950 	RTE_SET_USED(dev);
1951 
1952 	if (conntrack->state > RTE_FLOW_CONNTRACK_STATE_TIME_WAIT)
1953 		return rte_flow_error_set(error, EINVAL,
1954 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1955 					  "Invalid CT state");
1956 	if (conntrack->last_index > RTE_FLOW_CONNTRACK_FLAG_RST)
1957 		return rte_flow_error_set(error, EINVAL,
1958 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1959 					  "Invalid last TCP packet flag");
1960 	return 0;
1961 }
1962 
1963 /**
1964  * Verify the @p attributes will be correctly understood by the NIC and store
1965  * them in the @p flow if everything is correct.
1966  *
1967  * @param[in] dev
1968  *   Pointer to the Ethernet device structure.
1969  * @param[in] attributes
1970  *   Pointer to flow attributes
1971  * @param[out] error
1972  *   Pointer to error structure.
1973  *
1974  * @return
1975  *   0 on success, a negative errno value otherwise and rte_errno is set.
1976  */
1977 int
1978 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1979 			      const struct rte_flow_attr *attributes,
1980 			      struct rte_flow_error *error)
1981 {
1982 	struct mlx5_priv *priv = dev->data->dev_private;
1983 	uint32_t priority_max = priv->sh->flow_max_priority - 1;
1984 
1985 	if (attributes->group)
1986 		return rte_flow_error_set(error, ENOTSUP,
1987 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1988 					  NULL, "groups is not supported");
1989 	if (attributes->priority != MLX5_FLOW_LOWEST_PRIO_INDICATOR &&
1990 	    attributes->priority >= priority_max)
1991 		return rte_flow_error_set(error, ENOTSUP,
1992 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1993 					  NULL, "priority out of range");
1994 	if (attributes->egress)
1995 		return rte_flow_error_set(error, ENOTSUP,
1996 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1997 					  "egress is not supported");
1998 	if (attributes->transfer && !priv->sh->config.dv_esw_en)
1999 		return rte_flow_error_set(error, ENOTSUP,
2000 					  RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
2001 					  NULL, "transfer is not supported");
2002 	if (!attributes->ingress)
2003 		return rte_flow_error_set(error, EINVAL,
2004 					  RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
2005 					  NULL,
2006 					  "ingress attribute is mandatory");
2007 	return 0;
2008 }
2009 
2010 /**
2011  * Validate ICMP6 item.
2012  *
2013  * @param[in] item
2014  *   Item specification.
2015  * @param[in] item_flags
2016  *   Bit-fields that holds the items detected until now.
2017  * @param[in] ext_vlan_sup
2018  *   Whether extended VLAN features are supported or not.
2019  * @param[out] error
2020  *   Pointer to error structure.
2021  *
2022  * @return
2023  *   0 on success, a negative errno value otherwise and rte_errno is set.
2024  */
2025 int
2026 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
2027 			       uint64_t item_flags,
2028 			       uint8_t target_protocol,
2029 			       struct rte_flow_error *error)
2030 {
2031 	const struct rte_flow_item_icmp6 *mask = item->mask;
2032 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2033 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
2034 				      MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2035 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2036 				      MLX5_FLOW_LAYER_OUTER_L4;
2037 	int ret;
2038 
2039 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
2040 		return rte_flow_error_set(error, EINVAL,
2041 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2042 					  "protocol filtering not compatible"
2043 					  " with ICMP6 layer");
2044 	if (!(item_flags & l3m))
2045 		return rte_flow_error_set(error, EINVAL,
2046 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2047 					  "IPv6 is mandatory to filter on"
2048 					  " ICMP6");
2049 	if (item_flags & l4m)
2050 		return rte_flow_error_set(error, EINVAL,
2051 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 					  "multiple L4 layers not supported");
2053 	if (!mask)
2054 		mask = &rte_flow_item_icmp6_mask;
2055 	ret = mlx5_flow_item_acceptable
2056 		(item, (const uint8_t *)mask,
2057 		 (const uint8_t *)&rte_flow_item_icmp6_mask,
2058 		 sizeof(struct rte_flow_item_icmp6),
2059 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2060 	if (ret < 0)
2061 		return ret;
2062 	return 0;
2063 }
2064 
2065 /**
2066  * Validate ICMP item.
2067  *
2068  * @param[in] item
2069  *   Item specification.
2070  * @param[in] item_flags
2071  *   Bit-fields that holds the items detected until now.
2072  * @param[out] error
2073  *   Pointer to error structure.
2074  *
2075  * @return
2076  *   0 on success, a negative errno value otherwise and rte_errno is set.
2077  */
2078 int
2079 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
2080 			     uint64_t item_flags,
2081 			     uint8_t target_protocol,
2082 			     struct rte_flow_error *error)
2083 {
2084 	const struct rte_flow_item_icmp *mask = item->mask;
2085 	const struct rte_flow_item_icmp nic_mask = {
2086 		.hdr.icmp_type = 0xff,
2087 		.hdr.icmp_code = 0xff,
2088 		.hdr.icmp_ident = RTE_BE16(0xffff),
2089 		.hdr.icmp_seq_nb = RTE_BE16(0xffff),
2090 	};
2091 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2092 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
2093 				      MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2094 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2095 				      MLX5_FLOW_LAYER_OUTER_L4;
2096 	int ret;
2097 
2098 	if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
2099 		return rte_flow_error_set(error, EINVAL,
2100 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2101 					  "protocol filtering not compatible"
2102 					  " with ICMP layer");
2103 	if (!(item_flags & l3m))
2104 		return rte_flow_error_set(error, EINVAL,
2105 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2106 					  "IPv4 is mandatory to filter"
2107 					  " on ICMP");
2108 	if (item_flags & l4m)
2109 		return rte_flow_error_set(error, EINVAL,
2110 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2111 					  "multiple L4 layers not supported");
2112 	if (!mask)
2113 		mask = &nic_mask;
2114 	ret = mlx5_flow_item_acceptable
2115 		(item, (const uint8_t *)mask,
2116 		 (const uint8_t *)&nic_mask,
2117 		 sizeof(struct rte_flow_item_icmp),
2118 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2119 	if (ret < 0)
2120 		return ret;
2121 	return 0;
2122 }
2123 
2124 /**
2125  * Validate Ethernet item.
2126  *
2127  * @param[in] item
2128  *   Item specification.
2129  * @param[in] item_flags
2130  *   Bit-fields that holds the items detected until now.
2131  * @param[out] error
2132  *   Pointer to error structure.
2133  *
2134  * @return
2135  *   0 on success, a negative errno value otherwise and rte_errno is set.
2136  */
2137 int
2138 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
2139 			    uint64_t item_flags, bool ext_vlan_sup,
2140 			    struct rte_flow_error *error)
2141 {
2142 	const struct rte_flow_item_eth *mask = item->mask;
2143 	const struct rte_flow_item_eth nic_mask = {
2144 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2145 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
2146 		.type = RTE_BE16(0xffff),
2147 		.has_vlan = ext_vlan_sup ? 1 : 0,
2148 	};
2149 	int ret;
2150 	int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2151 	const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2	:
2152 				       MLX5_FLOW_LAYER_OUTER_L2;
2153 
2154 	if (item_flags & ethm)
2155 		return rte_flow_error_set(error, ENOTSUP,
2156 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 					  "multiple L2 layers not supported");
2158 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
2159 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
2160 		return rte_flow_error_set(error, EINVAL,
2161 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2162 					  "L2 layer should not follow "
2163 					  "L3 layers");
2164 	if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
2165 	    (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
2166 		return rte_flow_error_set(error, EINVAL,
2167 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2168 					  "L2 layer should not follow VLAN");
2169 	if (item_flags & MLX5_FLOW_LAYER_GTP)
2170 		return rte_flow_error_set(error, EINVAL,
2171 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2172 					  "L2 layer should not follow GTP");
2173 	if (!mask)
2174 		mask = &rte_flow_item_eth_mask;
2175 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2176 					(const uint8_t *)&nic_mask,
2177 					sizeof(struct rte_flow_item_eth),
2178 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2179 	return ret;
2180 }
2181 
2182 /**
2183  * Validate VLAN item.
2184  *
2185  * @param[in] item
2186  *   Item specification.
2187  * @param[in] item_flags
2188  *   Bit-fields that holds the items detected until now.
2189  * @param[in] dev
2190  *   Ethernet device flow is being created on.
2191  * @param[out] error
2192  *   Pointer to error structure.
2193  *
2194  * @return
2195  *   0 on success, a negative errno value otherwise and rte_errno is set.
2196  */
2197 int
2198 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
2199 			     uint64_t item_flags,
2200 			     struct rte_eth_dev *dev,
2201 			     struct rte_flow_error *error)
2202 {
2203 	const struct rte_flow_item_vlan *spec = item->spec;
2204 	const struct rte_flow_item_vlan *mask = item->mask;
2205 	const struct rte_flow_item_vlan nic_mask = {
2206 		.tci = RTE_BE16(UINT16_MAX),
2207 		.inner_type = RTE_BE16(UINT16_MAX),
2208 	};
2209 	uint16_t vlan_tag = 0;
2210 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2211 	int ret;
2212 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
2213 					MLX5_FLOW_LAYER_INNER_L4) :
2214 				       (MLX5_FLOW_LAYER_OUTER_L3 |
2215 					MLX5_FLOW_LAYER_OUTER_L4);
2216 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
2217 					MLX5_FLOW_LAYER_OUTER_VLAN;
2218 
2219 	if (item_flags & vlanm)
2220 		return rte_flow_error_set(error, EINVAL,
2221 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2222 					  "multiple VLAN layers not supported");
2223 	else if ((item_flags & l34m) != 0)
2224 		return rte_flow_error_set(error, EINVAL,
2225 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 					  "VLAN cannot follow L3/L4 layer");
2227 	if (!mask)
2228 		mask = &rte_flow_item_vlan_mask;
2229 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2230 					(const uint8_t *)&nic_mask,
2231 					sizeof(struct rte_flow_item_vlan),
2232 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2233 	if (ret)
2234 		return ret;
2235 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
2236 		struct mlx5_priv *priv = dev->data->dev_private;
2237 
2238 		if (priv->vmwa_context) {
2239 			/*
2240 			 * Non-NULL context means we have a virtual machine
2241 			 * and SR-IOV enabled, we have to create VLAN interface
2242 			 * to make hypervisor to setup E-Switch vport
2243 			 * context correctly. We avoid creating the multiple
2244 			 * VLAN interfaces, so we cannot support VLAN tag mask.
2245 			 */
2246 			return rte_flow_error_set(error, EINVAL,
2247 						  RTE_FLOW_ERROR_TYPE_ITEM,
2248 						  item,
2249 						  "VLAN tag mask is not"
2250 						  " supported in virtual"
2251 						  " environment");
2252 		}
2253 	}
2254 	if (spec) {
2255 		vlan_tag = spec->tci;
2256 		vlan_tag &= mask->tci;
2257 	}
2258 	/*
2259 	 * From verbs perspective an empty VLAN is equivalent
2260 	 * to a packet without VLAN layer.
2261 	 */
2262 	if (!vlan_tag)
2263 		return rte_flow_error_set(error, EINVAL,
2264 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2265 					  item->spec,
2266 					  "VLAN cannot be empty");
2267 	return 0;
2268 }
2269 
2270 /**
2271  * Validate IPV4 item.
2272  *
2273  * @param[in] item
2274  *   Item specification.
2275  * @param[in] item_flags
2276  *   Bit-fields that holds the items detected until now.
2277  * @param[in] last_item
2278  *   Previous validated item in the pattern items.
2279  * @param[in] ether_type
2280  *   Type in the ethernet layer header (including dot1q).
2281  * @param[in] acc_mask
2282  *   Acceptable mask, if NULL default internal default mask
2283  *   will be used to check whether item fields are supported.
2284  * @param[in] range_accepted
2285  *   True if range of values is accepted for specific fields, false otherwise.
2286  * @param[out] error
2287  *   Pointer to error structure.
2288  *
2289  * @return
2290  *   0 on success, a negative errno value otherwise and rte_errno is set.
2291  */
2292 int
2293 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
2294 			     uint64_t item_flags,
2295 			     uint64_t last_item,
2296 			     uint16_t ether_type,
2297 			     const struct rte_flow_item_ipv4 *acc_mask,
2298 			     bool range_accepted,
2299 			     struct rte_flow_error *error)
2300 {
2301 	const struct rte_flow_item_ipv4 *mask = item->mask;
2302 	const struct rte_flow_item_ipv4 *spec = item->spec;
2303 	const struct rte_flow_item_ipv4 nic_mask = {
2304 		.hdr = {
2305 			.src_addr = RTE_BE32(0xffffffff),
2306 			.dst_addr = RTE_BE32(0xffffffff),
2307 			.type_of_service = 0xff,
2308 			.next_proto_id = 0xff,
2309 		},
2310 	};
2311 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2312 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2313 				      MLX5_FLOW_LAYER_OUTER_L3;
2314 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2315 				      MLX5_FLOW_LAYER_OUTER_L4;
2316 	int ret;
2317 	uint8_t next_proto = 0xFF;
2318 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2319 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2320 				  MLX5_FLOW_LAYER_INNER_VLAN);
2321 
2322 	if ((last_item & l2_vlan) && ether_type &&
2323 	    ether_type != RTE_ETHER_TYPE_IPV4)
2324 		return rte_flow_error_set(error, EINVAL,
2325 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2326 					  "IPv4 cannot follow L2/VLAN layer "
2327 					  "which ether type is not IPv4");
2328 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2329 		if (mask && spec)
2330 			next_proto = mask->hdr.next_proto_id &
2331 				     spec->hdr.next_proto_id;
2332 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2333 			return rte_flow_error_set(error, EINVAL,
2334 						  RTE_FLOW_ERROR_TYPE_ITEM,
2335 						  item,
2336 						  "multiple tunnel "
2337 						  "not supported");
2338 	}
2339 	if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
2340 		return rte_flow_error_set(error, EINVAL,
2341 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2342 					  "wrong tunnel type - IPv6 specified "
2343 					  "but IPv4 item provided");
2344 	if (item_flags & l3m)
2345 		return rte_flow_error_set(error, ENOTSUP,
2346 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2347 					  "multiple L3 layers not supported");
2348 	else if (item_flags & l4m)
2349 		return rte_flow_error_set(error, EINVAL,
2350 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2351 					  "L3 cannot follow an L4 layer.");
2352 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2353 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2354 		return rte_flow_error_set(error, EINVAL,
2355 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2356 					  "L3 cannot follow an NVGRE layer.");
2357 	if (!mask)
2358 		mask = &rte_flow_item_ipv4_mask;
2359 	else if (mask->hdr.next_proto_id != 0 &&
2360 		 mask->hdr.next_proto_id != 0xff)
2361 		return rte_flow_error_set(error, EINVAL,
2362 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
2363 					  "partial mask is not supported"
2364 					  " for protocol");
2365 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2366 					acc_mask ? (const uint8_t *)acc_mask
2367 						 : (const uint8_t *)&nic_mask,
2368 					sizeof(struct rte_flow_item_ipv4),
2369 					range_accepted, error);
2370 	if (ret < 0)
2371 		return ret;
2372 	return 0;
2373 }
2374 
2375 /**
2376  * Validate IPV6 item.
2377  *
2378  * @param[in] item
2379  *   Item specification.
2380  * @param[in] item_flags
2381  *   Bit-fields that holds the items detected until now.
2382  * @param[in] last_item
2383  *   Previous validated item in the pattern items.
2384  * @param[in] ether_type
2385  *   Type in the ethernet layer header (including dot1q).
2386  * @param[in] acc_mask
2387  *   Acceptable mask, if NULL default internal default mask
2388  *   will be used to check whether item fields are supported.
2389  * @param[out] error
2390  *   Pointer to error structure.
2391  *
2392  * @return
2393  *   0 on success, a negative errno value otherwise and rte_errno is set.
2394  */
2395 int
2396 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
2397 			     uint64_t item_flags,
2398 			     uint64_t last_item,
2399 			     uint16_t ether_type,
2400 			     const struct rte_flow_item_ipv6 *acc_mask,
2401 			     struct rte_flow_error *error)
2402 {
2403 	const struct rte_flow_item_ipv6 *mask = item->mask;
2404 	const struct rte_flow_item_ipv6 *spec = item->spec;
2405 	const struct rte_flow_item_ipv6 nic_mask = {
2406 		.hdr = {
2407 			.src_addr =
2408 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2409 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2410 			.dst_addr =
2411 				"\xff\xff\xff\xff\xff\xff\xff\xff"
2412 				"\xff\xff\xff\xff\xff\xff\xff\xff",
2413 			.vtc_flow = RTE_BE32(0xffffffff),
2414 			.proto = 0xff,
2415 		},
2416 	};
2417 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2418 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2419 				      MLX5_FLOW_LAYER_OUTER_L3;
2420 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2421 				      MLX5_FLOW_LAYER_OUTER_L4;
2422 	int ret;
2423 	uint8_t next_proto = 0xFF;
2424 	const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
2425 				  MLX5_FLOW_LAYER_OUTER_VLAN |
2426 				  MLX5_FLOW_LAYER_INNER_VLAN);
2427 
2428 	if ((last_item & l2_vlan) && ether_type &&
2429 	    ether_type != RTE_ETHER_TYPE_IPV6)
2430 		return rte_flow_error_set(error, EINVAL,
2431 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2432 					  "IPv6 cannot follow L2/VLAN layer "
2433 					  "which ether type is not IPv6");
2434 	if (mask && mask->hdr.proto == UINT8_MAX && spec)
2435 		next_proto = spec->hdr.proto;
2436 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL) {
2437 		if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
2438 			return rte_flow_error_set(error, EINVAL,
2439 						  RTE_FLOW_ERROR_TYPE_ITEM,
2440 						  item,
2441 						  "multiple tunnel "
2442 						  "not supported");
2443 	}
2444 	if (next_proto == IPPROTO_HOPOPTS  ||
2445 	    next_proto == IPPROTO_ROUTING  ||
2446 	    next_proto == IPPROTO_FRAGMENT ||
2447 	    next_proto == IPPROTO_ESP	   ||
2448 	    next_proto == IPPROTO_AH	   ||
2449 	    next_proto == IPPROTO_DSTOPTS)
2450 		return rte_flow_error_set(error, EINVAL,
2451 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2452 					  "IPv6 proto (next header) should "
2453 					  "not be set as extension header");
2454 	if (item_flags & MLX5_FLOW_LAYER_IPIP)
2455 		return rte_flow_error_set(error, EINVAL,
2456 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2457 					  "wrong tunnel type - IPv4 specified "
2458 					  "but IPv6 item provided");
2459 	if (item_flags & l3m)
2460 		return rte_flow_error_set(error, ENOTSUP,
2461 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2462 					  "multiple L3 layers not supported");
2463 	else if (item_flags & l4m)
2464 		return rte_flow_error_set(error, EINVAL,
2465 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2466 					  "L3 cannot follow an L4 layer.");
2467 	else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
2468 		  !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
2469 		return rte_flow_error_set(error, EINVAL,
2470 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2471 					  "L3 cannot follow an NVGRE layer.");
2472 	if (!mask)
2473 		mask = &rte_flow_item_ipv6_mask;
2474 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
2475 					acc_mask ? (const uint8_t *)acc_mask
2476 						 : (const uint8_t *)&nic_mask,
2477 					sizeof(struct rte_flow_item_ipv6),
2478 					MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2479 	if (ret < 0)
2480 		return ret;
2481 	return 0;
2482 }
2483 
2484 /**
2485  * Validate UDP item.
2486  *
2487  * @param[in] item
2488  *   Item specification.
2489  * @param[in] item_flags
2490  *   Bit-fields that holds the items detected until now.
2491  * @param[in] target_protocol
2492  *   The next protocol in the previous item.
2493  * @param[in] flow_mask
2494  *   mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
2495  * @param[out] error
2496  *   Pointer to error structure.
2497  *
2498  * @return
2499  *   0 on success, a negative errno value otherwise and rte_errno is set.
2500  */
2501 int
2502 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
2503 			    uint64_t item_flags,
2504 			    uint8_t target_protocol,
2505 			    struct rte_flow_error *error)
2506 {
2507 	const struct rte_flow_item_udp *mask = item->mask;
2508 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2509 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2510 				      MLX5_FLOW_LAYER_OUTER_L3;
2511 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2512 				      MLX5_FLOW_LAYER_OUTER_L4;
2513 	int ret;
2514 
2515 	if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
2516 		return rte_flow_error_set(error, EINVAL,
2517 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2518 					  "protocol filtering not compatible"
2519 					  " with UDP layer");
2520 	if (!(item_flags & l3m))
2521 		return rte_flow_error_set(error, EINVAL,
2522 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2523 					  "L3 is mandatory to filter on L4");
2524 	if (item_flags & l4m)
2525 		return rte_flow_error_set(error, EINVAL,
2526 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2527 					  "multiple L4 layers not supported");
2528 	if (!mask)
2529 		mask = &rte_flow_item_udp_mask;
2530 	ret = mlx5_flow_item_acceptable
2531 		(item, (const uint8_t *)mask,
2532 		 (const uint8_t *)&rte_flow_item_udp_mask,
2533 		 sizeof(struct rte_flow_item_udp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2534 		 error);
2535 	if (ret < 0)
2536 		return ret;
2537 	return 0;
2538 }
2539 
2540 /**
2541  * Validate TCP item.
2542  *
2543  * @param[in] item
2544  *   Item specification.
2545  * @param[in] item_flags
2546  *   Bit-fields that holds the items detected until now.
2547  * @param[in] target_protocol
2548  *   The next protocol in the previous item.
2549  * @param[out] error
2550  *   Pointer to error structure.
2551  *
2552  * @return
2553  *   0 on success, a negative errno value otherwise and rte_errno is set.
2554  */
2555 int
2556 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
2557 			    uint64_t item_flags,
2558 			    uint8_t target_protocol,
2559 			    const struct rte_flow_item_tcp *flow_mask,
2560 			    struct rte_flow_error *error)
2561 {
2562 	const struct rte_flow_item_tcp *mask = item->mask;
2563 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
2564 	const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
2565 				      MLX5_FLOW_LAYER_OUTER_L3;
2566 	const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
2567 				      MLX5_FLOW_LAYER_OUTER_L4;
2568 	int ret;
2569 
2570 	MLX5_ASSERT(flow_mask);
2571 	if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
2572 		return rte_flow_error_set(error, EINVAL,
2573 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2574 					  "protocol filtering not compatible"
2575 					  " with TCP layer");
2576 	if (!(item_flags & l3m))
2577 		return rte_flow_error_set(error, EINVAL,
2578 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2579 					  "L3 is mandatory to filter on L4");
2580 	if (item_flags & l4m)
2581 		return rte_flow_error_set(error, EINVAL,
2582 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2583 					  "multiple L4 layers not supported");
2584 	if (!mask)
2585 		mask = &rte_flow_item_tcp_mask;
2586 	ret = mlx5_flow_item_acceptable
2587 		(item, (const uint8_t *)mask,
2588 		 (const uint8_t *)flow_mask,
2589 		 sizeof(struct rte_flow_item_tcp), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2590 		 error);
2591 	if (ret < 0)
2592 		return ret;
2593 	return 0;
2594 }
2595 
2596 /**
2597  * Validate VXLAN item.
2598  *
2599  * @param[in] dev
2600  *   Pointer to the Ethernet device structure.
2601  * @param[in] udp_dport
2602  *   UDP destination port
2603  * @param[in] item
2604  *   Item specification.
2605  * @param[in] item_flags
2606  *   Bit-fields that holds the items detected until now.
2607  * @param[in] attr
2608  *   Flow rule attributes.
2609  * @param[out] error
2610  *   Pointer to error structure.
2611  *
2612  * @return
2613  *   0 on success, a negative errno value otherwise and rte_errno is set.
2614  */
2615 int
2616 mlx5_flow_validate_item_vxlan(struct rte_eth_dev *dev,
2617 			      uint16_t udp_dport,
2618 			      const struct rte_flow_item *item,
2619 			      uint64_t item_flags,
2620 			      const struct rte_flow_attr *attr,
2621 			      struct rte_flow_error *error)
2622 {
2623 	const struct rte_flow_item_vxlan *spec = item->spec;
2624 	const struct rte_flow_item_vxlan *mask = item->mask;
2625 	int ret;
2626 	struct mlx5_priv *priv = dev->data->dev_private;
2627 	union vni {
2628 		uint32_t vlan_id;
2629 		uint8_t vni[4];
2630 	} id = { .vlan_id = 0, };
2631 	const struct rte_flow_item_vxlan nic_mask = {
2632 		.vni = "\xff\xff\xff",
2633 		.rsvd1 = 0xff,
2634 	};
2635 	const struct rte_flow_item_vxlan *valid_mask;
2636 
2637 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2638 		return rte_flow_error_set(error, ENOTSUP,
2639 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2640 					  "multiple tunnel layers not"
2641 					  " supported");
2642 	valid_mask = &rte_flow_item_vxlan_mask;
2643 	/*
2644 	 * Verify only UDPv4 is present as defined in
2645 	 * https://tools.ietf.org/html/rfc7348
2646 	 */
2647 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2648 		return rte_flow_error_set(error, EINVAL,
2649 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2650 					  "no outer UDP layer found");
2651 	if (!mask)
2652 		mask = &rte_flow_item_vxlan_mask;
2653 
2654 	if (priv->sh->steering_format_version !=
2655 	    MLX5_STEERING_LOGIC_FORMAT_CONNECTX_5 ||
2656 	    !udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN) {
2657 		/* FDB domain & NIC domain non-zero group */
2658 		if ((attr->transfer || attr->group) && priv->sh->misc5_cap)
2659 			valid_mask = &nic_mask;
2660 		/* Group zero in NIC domain */
2661 		if (!attr->group && !attr->transfer &&
2662 		    priv->sh->tunnel_header_0_1)
2663 			valid_mask = &nic_mask;
2664 	}
2665 	ret = mlx5_flow_item_acceptable
2666 		(item, (const uint8_t *)mask,
2667 		 (const uint8_t *)valid_mask,
2668 		 sizeof(struct rte_flow_item_vxlan),
2669 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2670 	if (ret < 0)
2671 		return ret;
2672 	if (spec) {
2673 		memcpy(&id.vni[1], spec->vni, 3);
2674 		memcpy(&id.vni[1], mask->vni, 3);
2675 	}
2676 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2677 		return rte_flow_error_set(error, ENOTSUP,
2678 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2679 					  "VXLAN tunnel must be fully defined");
2680 	return 0;
2681 }
2682 
2683 /**
2684  * Validate VXLAN_GPE item.
2685  *
2686  * @param[in] item
2687  *   Item specification.
2688  * @param[in] item_flags
2689  *   Bit-fields that holds the items detected until now.
2690  * @param[in] priv
2691  *   Pointer to the private data structure.
2692  * @param[in] target_protocol
2693  *   The next protocol in the previous item.
2694  * @param[out] error
2695  *   Pointer to error structure.
2696  *
2697  * @return
2698  *   0 on success, a negative errno value otherwise and rte_errno is set.
2699  */
2700 int
2701 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
2702 				  uint64_t item_flags,
2703 				  struct rte_eth_dev *dev,
2704 				  struct rte_flow_error *error)
2705 {
2706 	struct mlx5_priv *priv = dev->data->dev_private;
2707 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
2708 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
2709 	int ret;
2710 	union vni {
2711 		uint32_t vlan_id;
2712 		uint8_t vni[4];
2713 	} id = { .vlan_id = 0, };
2714 
2715 	if (!priv->sh->config.l3_vxlan_en)
2716 		return rte_flow_error_set(error, ENOTSUP,
2717 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2718 					  "L3 VXLAN is not enabled by device"
2719 					  " parameter and/or not configured in"
2720 					  " firmware");
2721 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2722 		return rte_flow_error_set(error, ENOTSUP,
2723 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2724 					  "multiple tunnel layers not"
2725 					  " supported");
2726 	/*
2727 	 * Verify only UDPv4 is present as defined in
2728 	 * https://tools.ietf.org/html/rfc7348
2729 	 */
2730 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2731 		return rte_flow_error_set(error, EINVAL,
2732 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2733 					  "no outer UDP layer found");
2734 	if (!mask)
2735 		mask = &rte_flow_item_vxlan_gpe_mask;
2736 	ret = mlx5_flow_item_acceptable
2737 		(item, (const uint8_t *)mask,
2738 		 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
2739 		 sizeof(struct rte_flow_item_vxlan_gpe),
2740 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2741 	if (ret < 0)
2742 		return ret;
2743 	if (spec) {
2744 		if (spec->protocol)
2745 			return rte_flow_error_set(error, ENOTSUP,
2746 						  RTE_FLOW_ERROR_TYPE_ITEM,
2747 						  item,
2748 						  "VxLAN-GPE protocol"
2749 						  " not supported");
2750 		memcpy(&id.vni[1], spec->vni, 3);
2751 		memcpy(&id.vni[1], mask->vni, 3);
2752 	}
2753 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2754 		return rte_flow_error_set(error, ENOTSUP,
2755 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2756 					  "VXLAN-GPE tunnel must be fully"
2757 					  " defined");
2758 	return 0;
2759 }
2760 /**
2761  * Validate GRE Key item.
2762  *
2763  * @param[in] item
2764  *   Item specification.
2765  * @param[in] item_flags
2766  *   Bit flags to mark detected items.
2767  * @param[in] gre_item
2768  *   Pointer to gre_item
2769  * @param[out] error
2770  *   Pointer to error structure.
2771  *
2772  * @return
2773  *   0 on success, a negative errno value otherwise and rte_errno is set.
2774  */
2775 int
2776 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2777 				uint64_t item_flags,
2778 				const struct rte_flow_item *gre_item,
2779 				struct rte_flow_error *error)
2780 {
2781 	const rte_be32_t *mask = item->mask;
2782 	int ret = 0;
2783 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2784 	const struct rte_flow_item_gre *gre_spec;
2785 	const struct rte_flow_item_gre *gre_mask;
2786 
2787 	if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2788 		return rte_flow_error_set(error, ENOTSUP,
2789 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2790 					  "Multiple GRE key not support");
2791 	if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2792 		return rte_flow_error_set(error, ENOTSUP,
2793 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2794 					  "No preceding GRE header");
2795 	if (item_flags & MLX5_FLOW_LAYER_INNER)
2796 		return rte_flow_error_set(error, ENOTSUP,
2797 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2798 					  "GRE key following a wrong item");
2799 	gre_mask = gre_item->mask;
2800 	if (!gre_mask)
2801 		gre_mask = &rte_flow_item_gre_mask;
2802 	gre_spec = gre_item->spec;
2803 	if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2804 			 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2805 		return rte_flow_error_set(error, EINVAL,
2806 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2807 					  "Key bit must be on");
2808 
2809 	if (!mask)
2810 		mask = &gre_key_default_mask;
2811 	ret = mlx5_flow_item_acceptable
2812 		(item, (const uint8_t *)mask,
2813 		 (const uint8_t *)&gre_key_default_mask,
2814 		 sizeof(rte_be32_t), MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2815 	return ret;
2816 }
2817 
2818 /**
2819  * Validate GRE item.
2820  *
2821  * @param[in] item
2822  *   Item specification.
2823  * @param[in] item_flags
2824  *   Bit flags to mark detected items.
2825  * @param[in] target_protocol
2826  *   The next protocol in the previous item.
2827  * @param[out] error
2828  *   Pointer to error structure.
2829  *
2830  * @return
2831  *   0 on success, a negative errno value otherwise and rte_errno is set.
2832  */
2833 int
2834 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2835 			    uint64_t item_flags,
2836 			    uint8_t target_protocol,
2837 			    struct rte_flow_error *error)
2838 {
2839 	const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2840 	const struct rte_flow_item_gre *mask = item->mask;
2841 	int ret;
2842 	const struct rte_flow_item_gre nic_mask = {
2843 		.c_rsvd0_ver = RTE_BE16(0xB000),
2844 		.protocol = RTE_BE16(UINT16_MAX),
2845 	};
2846 
2847 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2848 		return rte_flow_error_set(error, EINVAL,
2849 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2850 					  "protocol filtering not compatible"
2851 					  " with this GRE layer");
2852 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2853 		return rte_flow_error_set(error, ENOTSUP,
2854 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2855 					  "multiple tunnel layers not"
2856 					  " supported");
2857 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2858 		return rte_flow_error_set(error, ENOTSUP,
2859 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2860 					  "L3 Layer is missing");
2861 	if (!mask)
2862 		mask = &rte_flow_item_gre_mask;
2863 	ret = mlx5_flow_item_acceptable
2864 		(item, (const uint8_t *)mask,
2865 		 (const uint8_t *)&nic_mask,
2866 		 sizeof(struct rte_flow_item_gre), MLX5_ITEM_RANGE_NOT_ACCEPTED,
2867 		 error);
2868 	if (ret < 0)
2869 		return ret;
2870 #ifndef HAVE_MLX5DV_DR
2871 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2872 	if (spec && (spec->protocol & mask->protocol))
2873 		return rte_flow_error_set(error, ENOTSUP,
2874 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2875 					  "without MPLS support the"
2876 					  " specification cannot be used for"
2877 					  " filtering");
2878 #endif
2879 #endif
2880 	return 0;
2881 }
2882 
2883 /**
2884  * Validate Geneve item.
2885  *
2886  * @param[in] item
2887  *   Item specification.
2888  * @param[in] itemFlags
2889  *   Bit-fields that holds the items detected until now.
2890  * @param[in] enPriv
2891  *   Pointer to the private data structure.
2892  * @param[out] error
2893  *   Pointer to error structure.
2894  *
2895  * @return
2896  *   0 on success, a negative errno value otherwise and rte_errno is set.
2897  */
2898 
2899 int
2900 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2901 			       uint64_t item_flags,
2902 			       struct rte_eth_dev *dev,
2903 			       struct rte_flow_error *error)
2904 {
2905 	struct mlx5_priv *priv = dev->data->dev_private;
2906 	const struct rte_flow_item_geneve *spec = item->spec;
2907 	const struct rte_flow_item_geneve *mask = item->mask;
2908 	int ret;
2909 	uint16_t gbhdr;
2910 	uint8_t opt_len = priv->sh->cdev->config.hca_attr.geneve_max_opt_len ?
2911 			  MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2912 	const struct rte_flow_item_geneve nic_mask = {
2913 		.ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2914 		.vni = "\xff\xff\xff",
2915 		.protocol = RTE_BE16(UINT16_MAX),
2916 	};
2917 
2918 	if (!priv->sh->cdev->config.hca_attr.tunnel_stateless_geneve_rx)
2919 		return rte_flow_error_set(error, ENOTSUP,
2920 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2921 					  "L3 Geneve is not enabled by device"
2922 					  " parameter and/or not configured in"
2923 					  " firmware");
2924 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2925 		return rte_flow_error_set(error, ENOTSUP,
2926 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2927 					  "multiple tunnel layers not"
2928 					  " supported");
2929 	/*
2930 	 * Verify only UDPv4 is present as defined in
2931 	 * https://tools.ietf.org/html/rfc7348
2932 	 */
2933 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2934 		return rte_flow_error_set(error, EINVAL,
2935 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
2936 					  "no outer UDP layer found");
2937 	if (!mask)
2938 		mask = &rte_flow_item_geneve_mask;
2939 	ret = mlx5_flow_item_acceptable
2940 				  (item, (const uint8_t *)mask,
2941 				   (const uint8_t *)&nic_mask,
2942 				   sizeof(struct rte_flow_item_geneve),
2943 				   MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
2944 	if (ret)
2945 		return ret;
2946 	if (spec) {
2947 		gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2948 		if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2949 		     MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2950 		     MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2951 			return rte_flow_error_set(error, ENOTSUP,
2952 						  RTE_FLOW_ERROR_TYPE_ITEM,
2953 						  item,
2954 						  "Geneve protocol unsupported"
2955 						  " fields are being used");
2956 		if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2957 			return rte_flow_error_set
2958 					(error, ENOTSUP,
2959 					 RTE_FLOW_ERROR_TYPE_ITEM,
2960 					 item,
2961 					 "Unsupported Geneve options length");
2962 	}
2963 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2964 		return rte_flow_error_set
2965 				    (error, ENOTSUP,
2966 				     RTE_FLOW_ERROR_TYPE_ITEM, item,
2967 				     "Geneve tunnel must be fully defined");
2968 	return 0;
2969 }
2970 
2971 /**
2972  * Validate Geneve TLV option item.
2973  *
2974  * @param[in] item
2975  *   Item specification.
2976  * @param[in] last_item
2977  *   Previous validated item in the pattern items.
2978  * @param[in] geneve_item
2979  *   Previous GENEVE item specification.
2980  * @param[in] dev
2981  *   Pointer to the rte_eth_dev structure.
2982  * @param[out] error
2983  *   Pointer to error structure.
2984  *
2985  * @return
2986  *   0 on success, a negative errno value otherwise and rte_errno is set.
2987  */
2988 int
2989 mlx5_flow_validate_item_geneve_opt(const struct rte_flow_item *item,
2990 				   uint64_t last_item,
2991 				   const struct rte_flow_item *geneve_item,
2992 				   struct rte_eth_dev *dev,
2993 				   struct rte_flow_error *error)
2994 {
2995 	struct mlx5_priv *priv = dev->data->dev_private;
2996 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2997 	struct mlx5_geneve_tlv_option_resource *geneve_opt_resource;
2998 	struct mlx5_hca_attr *hca_attr = &sh->cdev->config.hca_attr;
2999 	uint8_t data_max_supported =
3000 			hca_attr->max_geneve_tlv_option_data_len * 4;
3001 	const struct rte_flow_item_geneve *geneve_spec;
3002 	const struct rte_flow_item_geneve *geneve_mask;
3003 	const struct rte_flow_item_geneve_opt *spec = item->spec;
3004 	const struct rte_flow_item_geneve_opt *mask = item->mask;
3005 	unsigned int i;
3006 	unsigned int data_len;
3007 	uint8_t tlv_option_len;
3008 	uint16_t optlen_m, optlen_v;
3009 	const struct rte_flow_item_geneve_opt full_mask = {
3010 		.option_class = RTE_BE16(0xffff),
3011 		.option_type = 0xff,
3012 		.option_len = 0x1f,
3013 	};
3014 
3015 	if (!mask)
3016 		mask = &rte_flow_item_geneve_opt_mask;
3017 	if (!spec)
3018 		return rte_flow_error_set
3019 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3020 			"Geneve TLV opt class/type/length must be specified");
3021 	if ((uint32_t)spec->option_len > MLX5_GENEVE_OPTLEN_MASK)
3022 		return rte_flow_error_set
3023 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3024 			"Geneve TLV opt length exceeds the limit (31)");
3025 	/* Check if class type and length masks are full. */
3026 	if (full_mask.option_class != mask->option_class ||
3027 	    full_mask.option_type != mask->option_type ||
3028 	    full_mask.option_len != (mask->option_len & full_mask.option_len))
3029 		return rte_flow_error_set
3030 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3031 			"Geneve TLV opt class/type/length masks must be full");
3032 	/* Check if length is supported */
3033 	if ((uint32_t)spec->option_len >
3034 			hca_attr->max_geneve_tlv_option_data_len)
3035 		return rte_flow_error_set
3036 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3037 			"Geneve TLV opt length not supported");
3038 	if (hca_attr->max_geneve_tlv_options > 1)
3039 		DRV_LOG(DEBUG,
3040 			"max_geneve_tlv_options supports more than 1 option");
3041 	/* Check GENEVE item preceding. */
3042 	if (!geneve_item || !(last_item & MLX5_FLOW_LAYER_GENEVE))
3043 		return rte_flow_error_set
3044 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3045 			"Geneve opt item must be preceded with Geneve item");
3046 	geneve_spec = geneve_item->spec;
3047 	geneve_mask = geneve_item->mask ? geneve_item->mask :
3048 					  &rte_flow_item_geneve_mask;
3049 	/* Check if GENEVE TLV option size doesn't exceed option length */
3050 	if (geneve_spec && (geneve_mask->ver_opt_len_o_c_rsvd0 ||
3051 			    geneve_spec->ver_opt_len_o_c_rsvd0)) {
3052 		tlv_option_len = spec->option_len & mask->option_len;
3053 		optlen_v = rte_be_to_cpu_16(geneve_spec->ver_opt_len_o_c_rsvd0);
3054 		optlen_v = MLX5_GENEVE_OPTLEN_VAL(optlen_v);
3055 		optlen_m = rte_be_to_cpu_16(geneve_mask->ver_opt_len_o_c_rsvd0);
3056 		optlen_m = MLX5_GENEVE_OPTLEN_VAL(optlen_m);
3057 		if ((optlen_v & optlen_m) <= tlv_option_len)
3058 			return rte_flow_error_set
3059 				(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3060 				 "GENEVE TLV option length exceeds optlen");
3061 	}
3062 	/* Check if length is 0 or data is 0. */
3063 	if (spec->data == NULL || spec->option_len == 0)
3064 		return rte_flow_error_set
3065 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3066 			"Geneve TLV opt with zero data/length not supported");
3067 	/* Check not all data & mask are 0. */
3068 	data_len = spec->option_len * 4;
3069 	if (mask->data == NULL) {
3070 		for (i = 0; i < data_len; i++)
3071 			if (spec->data[i])
3072 				break;
3073 		if (i == data_len)
3074 			return rte_flow_error_set(error, ENOTSUP,
3075 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3076 				"Can't match on Geneve option data 0");
3077 	} else {
3078 		for (i = 0; i < data_len; i++)
3079 			if (spec->data[i] & mask->data[i])
3080 				break;
3081 		if (i == data_len)
3082 			return rte_flow_error_set(error, ENOTSUP,
3083 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3084 				"Can't match on Geneve option data and mask 0");
3085 		/* Check data mask supported. */
3086 		for (i = data_max_supported; i < data_len ; i++)
3087 			if (mask->data[i])
3088 				return rte_flow_error_set(error, ENOTSUP,
3089 					RTE_FLOW_ERROR_TYPE_ITEM, item,
3090 					"Data mask is of unsupported size");
3091 	}
3092 	/* Check GENEVE option is supported in NIC. */
3093 	if (!hca_attr->geneve_tlv_opt)
3094 		return rte_flow_error_set
3095 			(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item,
3096 			"Geneve TLV opt not supported");
3097 	/* Check if we already have geneve option with different type/class. */
3098 	rte_spinlock_lock(&sh->geneve_tlv_opt_sl);
3099 	geneve_opt_resource = sh->geneve_tlv_option_resource;
3100 	if (geneve_opt_resource != NULL)
3101 		if (geneve_opt_resource->option_class != spec->option_class ||
3102 		    geneve_opt_resource->option_type != spec->option_type ||
3103 		    geneve_opt_resource->length != spec->option_len) {
3104 			rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3105 			return rte_flow_error_set(error, ENOTSUP,
3106 				RTE_FLOW_ERROR_TYPE_ITEM, item,
3107 				"Only one Geneve TLV option supported");
3108 		}
3109 	rte_spinlock_unlock(&sh->geneve_tlv_opt_sl);
3110 	return 0;
3111 }
3112 
3113 /**
3114  * Validate MPLS item.
3115  *
3116  * @param[in] dev
3117  *   Pointer to the rte_eth_dev structure.
3118  * @param[in] item
3119  *   Item specification.
3120  * @param[in] item_flags
3121  *   Bit-fields that holds the items detected until now.
3122  * @param[in] prev_layer
3123  *   The protocol layer indicated in previous item.
3124  * @param[out] error
3125  *   Pointer to error structure.
3126  *
3127  * @return
3128  *   0 on success, a negative errno value otherwise and rte_errno is set.
3129  */
3130 int
3131 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
3132 			     const struct rte_flow_item *item __rte_unused,
3133 			     uint64_t item_flags __rte_unused,
3134 			     uint64_t prev_layer __rte_unused,
3135 			     struct rte_flow_error *error)
3136 {
3137 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
3138 	const struct rte_flow_item_mpls *mask = item->mask;
3139 	struct mlx5_priv *priv = dev->data->dev_private;
3140 	int ret;
3141 
3142 	if (!priv->sh->dev_cap.mpls_en)
3143 		return rte_flow_error_set(error, ENOTSUP,
3144 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3145 					  "MPLS not supported or"
3146 					  " disabled in firmware"
3147 					  " configuration.");
3148 	/* MPLS over UDP, GRE is allowed */
3149 	if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L4_UDP |
3150 			    MLX5_FLOW_LAYER_GRE |
3151 			    MLX5_FLOW_LAYER_GRE_KEY)))
3152 		return rte_flow_error_set(error, EINVAL,
3153 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3154 					  "protocol filtering not compatible"
3155 					  " with MPLS layer");
3156 	/* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
3157 	if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
3158 	    !(item_flags & MLX5_FLOW_LAYER_GRE))
3159 		return rte_flow_error_set(error, ENOTSUP,
3160 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3161 					  "multiple tunnel layers not"
3162 					  " supported");
3163 	if (!mask)
3164 		mask = &rte_flow_item_mpls_mask;
3165 	ret = mlx5_flow_item_acceptable
3166 		(item, (const uint8_t *)mask,
3167 		 (const uint8_t *)&rte_flow_item_mpls_mask,
3168 		 sizeof(struct rte_flow_item_mpls),
3169 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3170 	if (ret < 0)
3171 		return ret;
3172 	return 0;
3173 #else
3174 	return rte_flow_error_set(error, ENOTSUP,
3175 				  RTE_FLOW_ERROR_TYPE_ITEM, item,
3176 				  "MPLS is not supported by Verbs, please"
3177 				  " update.");
3178 #endif
3179 }
3180 
3181 /**
3182  * Validate NVGRE item.
3183  *
3184  * @param[in] item
3185  *   Item specification.
3186  * @param[in] item_flags
3187  *   Bit flags to mark detected items.
3188  * @param[in] target_protocol
3189  *   The next protocol in the previous item.
3190  * @param[out] error
3191  *   Pointer to error structure.
3192  *
3193  * @return
3194  *   0 on success, a negative errno value otherwise and rte_errno is set.
3195  */
3196 int
3197 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
3198 			      uint64_t item_flags,
3199 			      uint8_t target_protocol,
3200 			      struct rte_flow_error *error)
3201 {
3202 	const struct rte_flow_item_nvgre *mask = item->mask;
3203 	int ret;
3204 
3205 	if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
3206 		return rte_flow_error_set(error, EINVAL,
3207 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3208 					  "protocol filtering not compatible"
3209 					  " with this GRE layer");
3210 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3211 		return rte_flow_error_set(error, ENOTSUP,
3212 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3213 					  "multiple tunnel layers not"
3214 					  " supported");
3215 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
3216 		return rte_flow_error_set(error, ENOTSUP,
3217 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3218 					  "L3 Layer is missing");
3219 	if (!mask)
3220 		mask = &rte_flow_item_nvgre_mask;
3221 	ret = mlx5_flow_item_acceptable
3222 		(item, (const uint8_t *)mask,
3223 		 (const uint8_t *)&rte_flow_item_nvgre_mask,
3224 		 sizeof(struct rte_flow_item_nvgre),
3225 		 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3226 	if (ret < 0)
3227 		return ret;
3228 	return 0;
3229 }
3230 
3231 /**
3232  * Validate eCPRI item.
3233  *
3234  * @param[in] item
3235  *   Item specification.
3236  * @param[in] item_flags
3237  *   Bit-fields that holds the items detected until now.
3238  * @param[in] last_item
3239  *   Previous validated item in the pattern items.
3240  * @param[in] ether_type
3241  *   Type in the ethernet layer header (including dot1q).
3242  * @param[in] acc_mask
3243  *   Acceptable mask, if NULL default internal default mask
3244  *   will be used to check whether item fields are supported.
3245  * @param[out] error
3246  *   Pointer to error structure.
3247  *
3248  * @return
3249  *   0 on success, a negative errno value otherwise and rte_errno is set.
3250  */
3251 int
3252 mlx5_flow_validate_item_ecpri(const struct rte_flow_item *item,
3253 			      uint64_t item_flags,
3254 			      uint64_t last_item,
3255 			      uint16_t ether_type,
3256 			      const struct rte_flow_item_ecpri *acc_mask,
3257 			      struct rte_flow_error *error)
3258 {
3259 	const struct rte_flow_item_ecpri *mask = item->mask;
3260 	const struct rte_flow_item_ecpri nic_mask = {
3261 		.hdr = {
3262 			.common = {
3263 				.u32 =
3264 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
3265 					.type = 0xFF,
3266 					}).u32),
3267 			},
3268 			.dummy[0] = 0xFFFFFFFF,
3269 		},
3270 	};
3271 	const uint64_t outer_l2_vlan = (MLX5_FLOW_LAYER_OUTER_L2 |
3272 					MLX5_FLOW_LAYER_OUTER_VLAN);
3273 	struct rte_flow_item_ecpri mask_lo;
3274 
3275 	if (!(last_item & outer_l2_vlan) &&
3276 	    last_item != MLX5_FLOW_LAYER_OUTER_L4_UDP)
3277 		return rte_flow_error_set(error, EINVAL,
3278 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3279 					  "eCPRI can only follow L2/VLAN layer or UDP layer");
3280 	if ((last_item & outer_l2_vlan) && ether_type &&
3281 	    ether_type != RTE_ETHER_TYPE_ECPRI)
3282 		return rte_flow_error_set(error, EINVAL,
3283 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3284 					  "eCPRI cannot follow L2/VLAN layer which ether type is not 0xAEFE");
3285 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
3286 		return rte_flow_error_set(error, EINVAL,
3287 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3288 					  "eCPRI with tunnel is not supported right now");
3289 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3)
3290 		return rte_flow_error_set(error, ENOTSUP,
3291 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3292 					  "multiple L3 layers not supported");
3293 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
3294 		return rte_flow_error_set(error, EINVAL,
3295 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3296 					  "eCPRI cannot coexist with a TCP layer");
3297 	/* In specification, eCPRI could be over UDP layer. */
3298 	else if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
3299 		return rte_flow_error_set(error, EINVAL,
3300 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
3301 					  "eCPRI over UDP layer is not yet supported right now");
3302 	/* Mask for type field in common header could be zero. */
3303 	if (!mask)
3304 		mask = &rte_flow_item_ecpri_mask;
3305 	mask_lo.hdr.common.u32 = rte_be_to_cpu_32(mask->hdr.common.u32);
3306 	/* Input mask is in big-endian format. */
3307 	if (mask_lo.hdr.common.type != 0 && mask_lo.hdr.common.type != 0xff)
3308 		return rte_flow_error_set(error, EINVAL,
3309 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3310 					  "partial mask is not supported for protocol");
3311 	else if (mask_lo.hdr.common.type == 0 && mask->hdr.dummy[0] != 0)
3312 		return rte_flow_error_set(error, EINVAL,
3313 					  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
3314 					  "message header mask must be after a type mask");
3315 	return mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
3316 					 acc_mask ? (const uint8_t *)acc_mask
3317 						  : (const uint8_t *)&nic_mask,
3318 					 sizeof(struct rte_flow_item_ecpri),
3319 					 MLX5_ITEM_RANGE_NOT_ACCEPTED, error);
3320 }
3321 
3322 static int
3323 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
3324 		   const struct rte_flow_attr *attr __rte_unused,
3325 		   const struct rte_flow_item items[] __rte_unused,
3326 		   const struct rte_flow_action actions[] __rte_unused,
3327 		   bool external __rte_unused,
3328 		   int hairpin __rte_unused,
3329 		   struct rte_flow_error *error)
3330 {
3331 	return rte_flow_error_set(error, ENOTSUP,
3332 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3333 }
3334 
3335 static struct mlx5_flow *
3336 flow_null_prepare(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 		  struct rte_flow_error *error)
3341 {
3342 	rte_flow_error_set(error, ENOTSUP,
3343 			   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3344 	return NULL;
3345 }
3346 
3347 static int
3348 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
3349 		    struct mlx5_flow *dev_flow __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 	return rte_flow_error_set(error, ENOTSUP,
3356 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3357 }
3358 
3359 static int
3360 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
3361 		struct rte_flow *flow __rte_unused,
3362 		struct rte_flow_error *error)
3363 {
3364 	return rte_flow_error_set(error, ENOTSUP,
3365 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3366 }
3367 
3368 static void
3369 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
3370 		 struct rte_flow *flow __rte_unused)
3371 {
3372 }
3373 
3374 static void
3375 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
3376 		  struct rte_flow *flow __rte_unused)
3377 {
3378 }
3379 
3380 static int
3381 flow_null_query(struct rte_eth_dev *dev __rte_unused,
3382 		struct rte_flow *flow __rte_unused,
3383 		const struct rte_flow_action *actions __rte_unused,
3384 		void *data __rte_unused,
3385 		struct rte_flow_error *error)
3386 {
3387 	return rte_flow_error_set(error, ENOTSUP,
3388 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
3389 }
3390 
3391 static int
3392 flow_null_sync_domain(struct rte_eth_dev *dev __rte_unused,
3393 		      uint32_t domains __rte_unused,
3394 		      uint32_t flags __rte_unused)
3395 {
3396 	return 0;
3397 }
3398 
3399 /* Void driver to protect from null pointer reference. */
3400 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
3401 	.validate = flow_null_validate,
3402 	.prepare = flow_null_prepare,
3403 	.translate = flow_null_translate,
3404 	.apply = flow_null_apply,
3405 	.remove = flow_null_remove,
3406 	.destroy = flow_null_destroy,
3407 	.query = flow_null_query,
3408 	.sync_domain = flow_null_sync_domain,
3409 };
3410 
3411 /**
3412  * Select flow driver type according to flow attributes and device
3413  * configuration.
3414  *
3415  * @param[in] dev
3416  *   Pointer to the dev structure.
3417  * @param[in] attr
3418  *   Pointer to the flow attributes.
3419  *
3420  * @return
3421  *   flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
3422  */
3423 static enum mlx5_flow_drv_type
3424 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
3425 {
3426 	struct mlx5_priv *priv = dev->data->dev_private;
3427 	/* The OS can determine first a specific flow type (DV, VERBS) */
3428 	enum mlx5_flow_drv_type type = mlx5_flow_os_get_type();
3429 
3430 	if (type != MLX5_FLOW_TYPE_MAX)
3431 		return type;
3432 	/* If no OS specific type - continue with DV/VERBS selection */
3433 	if (attr->transfer && priv->sh->config.dv_esw_en)
3434 		type = MLX5_FLOW_TYPE_DV;
3435 	if (!attr->transfer)
3436 		type = priv->sh->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
3437 						     MLX5_FLOW_TYPE_VERBS;
3438 	return type;
3439 }
3440 
3441 #define flow_get_drv_ops(type) flow_drv_ops[type]
3442 
3443 /**
3444  * Flow driver validation API. This abstracts calling driver specific functions.
3445  * The type of flow driver is determined according to flow attributes.
3446  *
3447  * @param[in] dev
3448  *   Pointer to the dev structure.
3449  * @param[in] attr
3450  *   Pointer to the flow attributes.
3451  * @param[in] items
3452  *   Pointer to the list of items.
3453  * @param[in] actions
3454  *   Pointer to the list of actions.
3455  * @param[in] external
3456  *   This flow rule is created by request external to PMD.
3457  * @param[in] hairpin
3458  *   Number of hairpin TX actions, 0 means classic flow.
3459  * @param[out] error
3460  *   Pointer to the error structure.
3461  *
3462  * @return
3463  *   0 on success, a negative errno value otherwise and rte_errno is set.
3464  */
3465 static inline int
3466 flow_drv_validate(struct rte_eth_dev *dev,
3467 		  const struct rte_flow_attr *attr,
3468 		  const struct rte_flow_item items[],
3469 		  const struct rte_flow_action actions[],
3470 		  bool external, int hairpin, struct rte_flow_error *error)
3471 {
3472 	const struct mlx5_flow_driver_ops *fops;
3473 	enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
3474 
3475 	fops = flow_get_drv_ops(type);
3476 	return fops->validate(dev, attr, items, actions, external,
3477 			      hairpin, error);
3478 }
3479 
3480 /**
3481  * Flow driver preparation API. This abstracts calling driver specific
3482  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3483  * calculates the size of memory required for device flow, allocates the memory,
3484  * initializes the device flow and returns the pointer.
3485  *
3486  * @note
3487  *   This function initializes device flow structure such as dv or verbs in
3488  *   struct mlx5_flow. However, it is caller's responsibility to initialize the
3489  *   rest. For example, adding returning device flow to flow->dev_flow list and
3490  *   setting backward reference to the flow should be done out of this function.
3491  *   layers field is not filled either.
3492  *
3493  * @param[in] dev
3494  *   Pointer to the dev structure.
3495  * @param[in] attr
3496  *   Pointer to the flow attributes.
3497  * @param[in] items
3498  *   Pointer to the list of items.
3499  * @param[in] actions
3500  *   Pointer to the list of actions.
3501  * @param[in] flow_idx
3502  *   This memory pool index to the flow.
3503  * @param[out] error
3504  *   Pointer to the error structure.
3505  *
3506  * @return
3507  *   Pointer to device flow on success, otherwise NULL and rte_errno is set.
3508  */
3509 static inline struct mlx5_flow *
3510 flow_drv_prepare(struct rte_eth_dev *dev,
3511 		 const struct rte_flow *flow,
3512 		 const struct rte_flow_attr *attr,
3513 		 const struct rte_flow_item items[],
3514 		 const struct rte_flow_action actions[],
3515 		 uint32_t flow_idx,
3516 		 struct rte_flow_error *error)
3517 {
3518 	const struct mlx5_flow_driver_ops *fops;
3519 	enum mlx5_flow_drv_type type = flow->drv_type;
3520 	struct mlx5_flow *mlx5_flow = NULL;
3521 
3522 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3523 	fops = flow_get_drv_ops(type);
3524 	mlx5_flow = fops->prepare(dev, attr, items, actions, error);
3525 	if (mlx5_flow)
3526 		mlx5_flow->flow_idx = flow_idx;
3527 	return mlx5_flow;
3528 }
3529 
3530 /**
3531  * Flow driver translation API. This abstracts calling driver specific
3532  * functions. Parent flow (rte_flow) should have driver type (drv_type). It
3533  * translates a generic flow into a driver flow. flow_drv_prepare() must
3534  * precede.
3535  *
3536  * @note
3537  *   dev_flow->layers could be filled as a result of parsing during translation
3538  *   if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
3539  *   if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
3540  *   flow->actions could be overwritten even though all the expanded dev_flows
3541  *   have the same actions.
3542  *
3543  * @param[in] dev
3544  *   Pointer to the rte dev structure.
3545  * @param[in, out] dev_flow
3546  *   Pointer to the mlx5 flow.
3547  * @param[in] attr
3548  *   Pointer to the flow attributes.
3549  * @param[in] items
3550  *   Pointer to the list of items.
3551  * @param[in] actions
3552  *   Pointer to the list of actions.
3553  * @param[out] error
3554  *   Pointer to the error structure.
3555  *
3556  * @return
3557  *   0 on success, a negative errno value otherwise and rte_errno is set.
3558  */
3559 static inline int
3560 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3561 		   const struct rte_flow_attr *attr,
3562 		   const struct rte_flow_item items[],
3563 		   const struct rte_flow_action actions[],
3564 		   struct rte_flow_error *error)
3565 {
3566 	const struct mlx5_flow_driver_ops *fops;
3567 	enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
3568 
3569 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3570 	fops = flow_get_drv_ops(type);
3571 	return fops->translate(dev, dev_flow, attr, items, actions, error);
3572 }
3573 
3574 /**
3575  * Flow driver apply API. This abstracts calling driver specific functions.
3576  * Parent flow (rte_flow) should have driver type (drv_type). It applies
3577  * translated driver flows on to device. flow_drv_translate() must precede.
3578  *
3579  * @param[in] dev
3580  *   Pointer to Ethernet device structure.
3581  * @param[in, out] flow
3582  *   Pointer to flow structure.
3583  * @param[out] error
3584  *   Pointer to error structure.
3585  *
3586  * @return
3587  *   0 on success, a negative errno value otherwise and rte_errno is set.
3588  */
3589 static inline int
3590 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
3591 	       struct rte_flow_error *error)
3592 {
3593 	const struct mlx5_flow_driver_ops *fops;
3594 	enum mlx5_flow_drv_type type = flow->drv_type;
3595 
3596 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3597 	fops = flow_get_drv_ops(type);
3598 	return fops->apply(dev, flow, error);
3599 }
3600 
3601 /**
3602  * Flow driver destroy API. This abstracts calling driver specific functions.
3603  * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
3604  * on device and releases resources of the flow.
3605  *
3606  * @param[in] dev
3607  *   Pointer to Ethernet device.
3608  * @param[in, out] flow
3609  *   Pointer to flow structure.
3610  */
3611 static inline void
3612 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
3613 {
3614 	const struct mlx5_flow_driver_ops *fops;
3615 	enum mlx5_flow_drv_type type = flow->drv_type;
3616 
3617 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3618 	fops = flow_get_drv_ops(type);
3619 	fops->destroy(dev, flow);
3620 }
3621 
3622 /**
3623  * Flow driver find RSS policy tbl API. This abstracts calling driver
3624  * specific functions. Parent flow (rte_flow) should have driver
3625  * type (drv_type). It will find the RSS policy table that has the rss_desc.
3626  *
3627  * @param[in] dev
3628  *   Pointer to Ethernet device.
3629  * @param[in, out] flow
3630  *   Pointer to flow structure.
3631  * @param[in] policy
3632  *   Pointer to meter policy table.
3633  * @param[in] rss_desc
3634  *   Pointer to rss_desc
3635  */
3636 static struct mlx5_flow_meter_sub_policy *
3637 flow_drv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev,
3638 		struct rte_flow *flow,
3639 		struct mlx5_flow_meter_policy *policy,
3640 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS])
3641 {
3642 	const struct mlx5_flow_driver_ops *fops;
3643 	enum mlx5_flow_drv_type type = flow->drv_type;
3644 
3645 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3646 	fops = flow_get_drv_ops(type);
3647 	return fops->meter_sub_policy_rss_prepare(dev, policy, rss_desc);
3648 }
3649 
3650 /**
3651  * Flow driver color tag rule API. This abstracts calling driver
3652  * specific functions. Parent flow (rte_flow) should have driver
3653  * type (drv_type). It will create the color tag rules in hierarchy meter.
3654  *
3655  * @param[in] dev
3656  *   Pointer to Ethernet device.
3657  * @param[in, out] flow
3658  *   Pointer to flow structure.
3659  * @param[in] fm
3660  *   Pointer to flow meter structure.
3661  * @param[in] src_port
3662  *   The src port this extra rule should use.
3663  * @param[in] item
3664  *   The src port id match item.
3665  * @param[out] error
3666  *   Pointer to error structure.
3667  */
3668 static int
3669 flow_drv_mtr_hierarchy_rule_create(struct rte_eth_dev *dev,
3670 		struct rte_flow *flow,
3671 		struct mlx5_flow_meter_info *fm,
3672 		int32_t src_port,
3673 		const struct rte_flow_item *item,
3674 		struct rte_flow_error *error)
3675 {
3676 	const struct mlx5_flow_driver_ops *fops;
3677 	enum mlx5_flow_drv_type type = flow->drv_type;
3678 
3679 	MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
3680 	fops = flow_get_drv_ops(type);
3681 	return fops->meter_hierarchy_rule_create(dev, fm,
3682 						src_port, item, error);
3683 }
3684 
3685 /**
3686  * Get RSS action from the action list.
3687  *
3688  * @param[in] dev
3689  *   Pointer to Ethernet device.
3690  * @param[in] actions
3691  *   Pointer to the list of actions.
3692  * @param[in] flow
3693  *   Parent flow structure pointer.
3694  *
3695  * @return
3696  *   Pointer to the RSS action if exist, else return NULL.
3697  */
3698 static const struct rte_flow_action_rss*
3699 flow_get_rss_action(struct rte_eth_dev *dev,
3700 		    const struct rte_flow_action actions[])
3701 {
3702 	struct mlx5_priv *priv = dev->data->dev_private;
3703 	const struct rte_flow_action_rss *rss = NULL;
3704 	struct mlx5_meter_policy_action_container *acg;
3705 	struct mlx5_meter_policy_action_container *acy;
3706 
3707 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3708 		switch (actions->type) {
3709 		case RTE_FLOW_ACTION_TYPE_RSS:
3710 			rss = actions->conf;
3711 			break;
3712 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
3713 		{
3714 			const struct rte_flow_action_sample *sample =
3715 								actions->conf;
3716 			const struct rte_flow_action *act = sample->actions;
3717 			for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++)
3718 				if (act->type == RTE_FLOW_ACTION_TYPE_RSS)
3719 					rss = act->conf;
3720 			break;
3721 		}
3722 		case RTE_FLOW_ACTION_TYPE_METER:
3723 		{
3724 			uint32_t mtr_idx;
3725 			struct mlx5_flow_meter_info *fm;
3726 			struct mlx5_flow_meter_policy *policy;
3727 			const struct rte_flow_action_meter *mtr = actions->conf;
3728 
3729 			fm = mlx5_flow_meter_find(priv, mtr->mtr_id, &mtr_idx);
3730 			if (fm && !fm->def_policy) {
3731 				policy = mlx5_flow_meter_policy_find(dev,
3732 						fm->policy_id, NULL);
3733 				MLX5_ASSERT(policy);
3734 				if (policy->is_hierarchy) {
3735 					policy =
3736 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
3737 									policy);
3738 					if (!policy)
3739 						return NULL;
3740 				}
3741 				if (policy->is_rss) {
3742 					acg =
3743 					&policy->act_cnt[RTE_COLOR_GREEN];
3744 					acy =
3745 					&policy->act_cnt[RTE_COLOR_YELLOW];
3746 					if (acg->fate_action ==
3747 					    MLX5_FLOW_FATE_SHARED_RSS)
3748 						rss = acg->rss->conf;
3749 					else if (acy->fate_action ==
3750 						 MLX5_FLOW_FATE_SHARED_RSS)
3751 						rss = acy->rss->conf;
3752 				}
3753 			}
3754 			break;
3755 		}
3756 		default:
3757 			break;
3758 		}
3759 	}
3760 	return rss;
3761 }
3762 
3763 /**
3764  * Get ASO age action by index.
3765  *
3766  * @param[in] dev
3767  *   Pointer to the Ethernet device structure.
3768  * @param[in] age_idx
3769  *   Index to the ASO age action.
3770  *
3771  * @return
3772  *   The specified ASO age action.
3773  */
3774 struct mlx5_aso_age_action*
3775 flow_aso_age_get_by_idx(struct rte_eth_dev *dev, uint32_t age_idx)
3776 {
3777 	uint16_t pool_idx = age_idx & UINT16_MAX;
3778 	uint16_t offset = (age_idx >> 16) & UINT16_MAX;
3779 	struct mlx5_priv *priv = dev->data->dev_private;
3780 	struct mlx5_aso_age_mng *mng = priv->sh->aso_age_mng;
3781 	struct mlx5_aso_age_pool *pool;
3782 
3783 	rte_rwlock_read_lock(&mng->resize_rwl);
3784 	pool = mng->pools[pool_idx];
3785 	rte_rwlock_read_unlock(&mng->resize_rwl);
3786 	return &pool->actions[offset - 1];
3787 }
3788 
3789 /* maps indirect action to translated direct in some actions array */
3790 struct mlx5_translated_action_handle {
3791 	struct rte_flow_action_handle *action; /**< Indirect action handle. */
3792 	int index; /**< Index in related array of rte_flow_action. */
3793 };
3794 
3795 /**
3796  * Translates actions of type RTE_FLOW_ACTION_TYPE_INDIRECT to related
3797  * direct action if translation possible.
3798  * This functionality used to run same execution path for both direct and
3799  * indirect actions on flow create. All necessary preparations for indirect
3800  * action handling should be performed on *handle* actions list returned
3801  * from this call.
3802  *
3803  * @param[in] dev
3804  *   Pointer to Ethernet device.
3805  * @param[in] actions
3806  *   List of actions to translate.
3807  * @param[out] handle
3808  *   List to store translated indirect action object handles.
3809  * @param[in, out] indir_n
3810  *   Size of *handle* array. On return should be updated with number of
3811  *   indirect actions retrieved from the *actions* list.
3812  * @param[out] translated_actions
3813  *   List of actions where all indirect actions were translated to direct
3814  *   if possible. NULL if no translation took place.
3815  * @param[out] error
3816  *   Pointer to the error structure.
3817  *
3818  * @return
3819  *   0 on success, a negative errno value otherwise and rte_errno is set.
3820  */
3821 static int
3822 flow_action_handles_translate(struct rte_eth_dev *dev,
3823 			      const struct rte_flow_action actions[],
3824 			      struct mlx5_translated_action_handle *handle,
3825 			      int *indir_n,
3826 			      struct rte_flow_action **translated_actions,
3827 			      struct rte_flow_error *error)
3828 {
3829 	struct mlx5_priv *priv = dev->data->dev_private;
3830 	struct rte_flow_action *translated = NULL;
3831 	size_t actions_size;
3832 	int n;
3833 	int copied_n = 0;
3834 	struct mlx5_translated_action_handle *handle_end = NULL;
3835 
3836 	for (n = 0; actions[n].type != RTE_FLOW_ACTION_TYPE_END; n++) {
3837 		if (actions[n].type != RTE_FLOW_ACTION_TYPE_INDIRECT)
3838 			continue;
3839 		if (copied_n == *indir_n) {
3840 			return rte_flow_error_set
3841 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_NUM,
3842 				 NULL, "too many shared actions");
3843 		}
3844 		rte_memcpy(&handle[copied_n].action, &actions[n].conf,
3845 			   sizeof(actions[n].conf));
3846 		handle[copied_n].index = n;
3847 		copied_n++;
3848 	}
3849 	n++;
3850 	*indir_n = copied_n;
3851 	if (!copied_n)
3852 		return 0;
3853 	actions_size = sizeof(struct rte_flow_action) * n;
3854 	translated = mlx5_malloc(MLX5_MEM_ZERO, actions_size, 0, SOCKET_ID_ANY);
3855 	if (!translated) {
3856 		rte_errno = ENOMEM;
3857 		return -ENOMEM;
3858 	}
3859 	memcpy(translated, actions, actions_size);
3860 	for (handle_end = handle + copied_n; handle < handle_end; handle++) {
3861 		struct mlx5_shared_action_rss *shared_rss;
3862 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3863 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3864 		uint32_t idx = act_idx &
3865 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3866 
3867 		switch (type) {
3868 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3869 			shared_rss = mlx5_ipool_get
3870 			  (priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
3871 			translated[handle->index].type =
3872 				RTE_FLOW_ACTION_TYPE_RSS;
3873 			translated[handle->index].conf =
3874 				&shared_rss->origin;
3875 			break;
3876 		case MLX5_INDIRECT_ACTION_TYPE_COUNT:
3877 			translated[handle->index].type =
3878 						(enum rte_flow_action_type)
3879 						MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
3880 			translated[handle->index].conf = (void *)(uintptr_t)idx;
3881 			break;
3882 		case MLX5_INDIRECT_ACTION_TYPE_AGE:
3883 			if (priv->sh->flow_hit_aso_en) {
3884 				translated[handle->index].type =
3885 					(enum rte_flow_action_type)
3886 					MLX5_RTE_FLOW_ACTION_TYPE_AGE;
3887 				translated[handle->index].conf =
3888 							 (void *)(uintptr_t)idx;
3889 				break;
3890 			}
3891 			/* Fall-through */
3892 		case MLX5_INDIRECT_ACTION_TYPE_CT:
3893 			if (priv->sh->ct_aso_en) {
3894 				translated[handle->index].type =
3895 					RTE_FLOW_ACTION_TYPE_CONNTRACK;
3896 				translated[handle->index].conf =
3897 							 (void *)(uintptr_t)idx;
3898 				break;
3899 			}
3900 			/* Fall-through */
3901 		default:
3902 			mlx5_free(translated);
3903 			return rte_flow_error_set
3904 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
3905 				 NULL, "invalid indirect action type");
3906 		}
3907 	}
3908 	*translated_actions = translated;
3909 	return 0;
3910 }
3911 
3912 /**
3913  * Get Shared RSS action from the action list.
3914  *
3915  * @param[in] dev
3916  *   Pointer to Ethernet device.
3917  * @param[in] shared
3918  *   Pointer to the list of actions.
3919  * @param[in] shared_n
3920  *   Actions list length.
3921  *
3922  * @return
3923  *   The MLX5 RSS action ID if exists, otherwise return 0.
3924  */
3925 static uint32_t
3926 flow_get_shared_rss_action(struct rte_eth_dev *dev,
3927 			   struct mlx5_translated_action_handle *handle,
3928 			   int shared_n)
3929 {
3930 	struct mlx5_translated_action_handle *handle_end;
3931 	struct mlx5_priv *priv = dev->data->dev_private;
3932 	struct mlx5_shared_action_rss *shared_rss;
3933 
3934 
3935 	for (handle_end = handle + shared_n; handle < handle_end; handle++) {
3936 		uint32_t act_idx = (uint32_t)(uintptr_t)handle->action;
3937 		uint32_t type = act_idx >> MLX5_INDIRECT_ACTION_TYPE_OFFSET;
3938 		uint32_t idx = act_idx &
3939 			       ((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
3940 		switch (type) {
3941 		case MLX5_INDIRECT_ACTION_TYPE_RSS:
3942 			shared_rss = mlx5_ipool_get
3943 				(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
3944 									   idx);
3945 			__atomic_add_fetch(&shared_rss->refcnt, 1,
3946 					   __ATOMIC_RELAXED);
3947 			return idx;
3948 		default:
3949 			break;
3950 		}
3951 	}
3952 	return 0;
3953 }
3954 
3955 static unsigned int
3956 find_graph_root(uint32_t rss_level)
3957 {
3958 	return rss_level < 2 ? MLX5_EXPANSION_ROOT :
3959 			       MLX5_EXPANSION_ROOT_OUTER;
3960 }
3961 
3962 /**
3963  *  Get layer flags from the prefix flow.
3964  *
3965  *  Some flows may be split to several subflows, the prefix subflow gets the
3966  *  match items and the suffix sub flow gets the actions.
3967  *  Some actions need the user defined match item flags to get the detail for
3968  *  the action.
3969  *  This function helps the suffix flow to get the item layer flags from prefix
3970  *  subflow.
3971  *
3972  * @param[in] dev_flow
3973  *   Pointer the created prefix subflow.
3974  *
3975  * @return
3976  *   The layers get from prefix subflow.
3977  */
3978 static inline uint64_t
3979 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
3980 {
3981 	uint64_t layers = 0;
3982 
3983 	/*
3984 	 * Layers bits could be localization, but usually the compiler will
3985 	 * help to do the optimization work for source code.
3986 	 * If no decap actions, use the layers directly.
3987 	 */
3988 	if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
3989 		return dev_flow->handle->layers;
3990 	/* Convert L3 layers with decap action. */
3991 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
3992 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3993 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
3994 		layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3995 	/* Convert L4 layers with decap action.  */
3996 	if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
3997 		layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3998 	else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
3999 		layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
4000 	return layers;
4001 }
4002 
4003 /**
4004  * Get metadata split action information.
4005  *
4006  * @param[in] actions
4007  *   Pointer to the list of actions.
4008  * @param[out] qrss
4009  *   Pointer to the return pointer.
4010  * @param[out] qrss_type
4011  *   Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
4012  *   if no QUEUE/RSS is found.
4013  * @param[out] encap_idx
4014  *   Pointer to the index of the encap action if exists, otherwise the last
4015  *   action index.
4016  *
4017  * @return
4018  *   Total number of actions.
4019  */
4020 static int
4021 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
4022 				       const struct rte_flow_action **qrss,
4023 				       int *encap_idx)
4024 {
4025 	const struct rte_flow_action_raw_encap *raw_encap;
4026 	int actions_n = 0;
4027 	int raw_decap_idx = -1;
4028 
4029 	*encap_idx = -1;
4030 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4031 		switch (actions->type) {
4032 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4033 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4034 			*encap_idx = actions_n;
4035 			break;
4036 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4037 			raw_decap_idx = actions_n;
4038 			break;
4039 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4040 			raw_encap = actions->conf;
4041 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4042 				*encap_idx = raw_decap_idx != -1 ?
4043 						      raw_decap_idx : actions_n;
4044 			break;
4045 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4046 		case RTE_FLOW_ACTION_TYPE_RSS:
4047 			*qrss = actions;
4048 			break;
4049 		default:
4050 			break;
4051 		}
4052 		actions_n++;
4053 	}
4054 	if (*encap_idx == -1)
4055 		*encap_idx = actions_n;
4056 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4057 	return actions_n + 1;
4058 }
4059 
4060 /**
4061  * Check if the action will change packet.
4062  *
4063  * @param dev
4064  *   Pointer to Ethernet device.
4065  * @param[in] type
4066  *   action type.
4067  *
4068  * @return
4069  *   true if action will change packet, false otherwise.
4070  */
4071 static bool flow_check_modify_action_type(struct rte_eth_dev *dev,
4072 					  enum rte_flow_action_type type)
4073 {
4074 	struct mlx5_priv *priv = dev->data->dev_private;
4075 
4076 	switch (type) {
4077 	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
4078 	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
4079 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
4080 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
4081 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
4082 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
4083 	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
4084 	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
4085 	case RTE_FLOW_ACTION_TYPE_DEC_TTL:
4086 	case RTE_FLOW_ACTION_TYPE_SET_TTL:
4087 	case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
4088 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
4089 	case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
4090 	case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
4091 	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
4092 	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
4093 	case RTE_FLOW_ACTION_TYPE_SET_META:
4094 	case RTE_FLOW_ACTION_TYPE_SET_TAG:
4095 	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
4096 	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4097 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4098 	case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4099 	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4100 	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
4101 	case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4102 	case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
4103 	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4104 	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4105 	case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
4106 		return true;
4107 	case RTE_FLOW_ACTION_TYPE_FLAG:
4108 	case RTE_FLOW_ACTION_TYPE_MARK:
4109 		if (priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4110 			return true;
4111 		else
4112 			return false;
4113 	default:
4114 		return false;
4115 	}
4116 }
4117 
4118 /**
4119  * Check meter action from the action list.
4120  *
4121  * @param dev
4122  *   Pointer to Ethernet device.
4123  * @param[in] actions
4124  *   Pointer to the list of actions.
4125  * @param[out] has_mtr
4126  *   Pointer to the meter exist flag.
4127  * @param[out] has_modify
4128  *   Pointer to the flag showing there's packet change action.
4129  * @param[out] meter_id
4130  *   Pointer to the meter id.
4131  *
4132  * @return
4133  *   Total number of actions.
4134  */
4135 static int
4136 flow_check_meter_action(struct rte_eth_dev *dev,
4137 			const struct rte_flow_action actions[],
4138 			bool *has_mtr, bool *has_modify, uint32_t *meter_id)
4139 {
4140 	const struct rte_flow_action_meter *mtr = NULL;
4141 	int actions_n = 0;
4142 
4143 	MLX5_ASSERT(has_mtr);
4144 	*has_mtr = false;
4145 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4146 		switch (actions->type) {
4147 		case RTE_FLOW_ACTION_TYPE_METER:
4148 			mtr = actions->conf;
4149 			*meter_id = mtr->mtr_id;
4150 			*has_mtr = true;
4151 			break;
4152 		default:
4153 			break;
4154 		}
4155 		if (!*has_mtr)
4156 			*has_modify |= flow_check_modify_action_type(dev,
4157 								actions->type);
4158 		actions_n++;
4159 	}
4160 	/* Count RTE_FLOW_ACTION_TYPE_END. */
4161 	return actions_n + 1;
4162 }
4163 
4164 /**
4165  * Check if the flow should be split due to hairpin.
4166  * The reason for the split is that in current HW we can't
4167  * support encap and push-vlan on Rx, so if a flow contains
4168  * these actions we move it to Tx.
4169  *
4170  * @param dev
4171  *   Pointer to Ethernet device.
4172  * @param[in] attr
4173  *   Flow rule attributes.
4174  * @param[in] actions
4175  *   Associated actions (list terminated by the END action).
4176  *
4177  * @return
4178  *   > 0 the number of actions and the flow should be split,
4179  *   0 when no split required.
4180  */
4181 static int
4182 flow_check_hairpin_split(struct rte_eth_dev *dev,
4183 			 const struct rte_flow_attr *attr,
4184 			 const struct rte_flow_action actions[])
4185 {
4186 	int queue_action = 0;
4187 	int action_n = 0;
4188 	int split = 0;
4189 	const struct rte_flow_action_queue *queue;
4190 	const struct rte_flow_action_rss *rss;
4191 	const struct rte_flow_action_raw_encap *raw_encap;
4192 	const struct rte_eth_hairpin_conf *conf;
4193 
4194 	if (!attr->ingress)
4195 		return 0;
4196 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4197 		switch (actions->type) {
4198 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4199 			queue = actions->conf;
4200 			if (queue == NULL)
4201 				return 0;
4202 			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
4203 			if (conf == NULL || conf->tx_explicit != 0)
4204 				return 0;
4205 			queue_action = 1;
4206 			action_n++;
4207 			break;
4208 		case RTE_FLOW_ACTION_TYPE_RSS:
4209 			rss = actions->conf;
4210 			if (rss == NULL || rss->queue_num == 0)
4211 				return 0;
4212 			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
4213 			if (conf == NULL || conf->tx_explicit != 0)
4214 				return 0;
4215 			queue_action = 1;
4216 			action_n++;
4217 			break;
4218 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4219 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4220 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4221 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4222 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4223 			split++;
4224 			action_n++;
4225 			break;
4226 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4227 			raw_encap = actions->conf;
4228 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
4229 				split++;
4230 			action_n++;
4231 			break;
4232 		default:
4233 			action_n++;
4234 			break;
4235 		}
4236 	}
4237 	if (split && queue_action)
4238 		return action_n;
4239 	return 0;
4240 }
4241 
4242 /* Declare flow create/destroy prototype in advance. */
4243 static uint32_t
4244 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4245 		 const struct rte_flow_attr *attr,
4246 		 const struct rte_flow_item items[],
4247 		 const struct rte_flow_action actions[],
4248 		 bool external, struct rte_flow_error *error);
4249 
4250 static void
4251 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
4252 		  uint32_t flow_idx);
4253 
4254 int
4255 flow_dv_mreg_match_cb(void *tool_ctx __rte_unused,
4256 		      struct mlx5_list_entry *entry, void *cb_ctx)
4257 {
4258 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4259 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4260 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4261 
4262 	return mcp_res->mark_id != *(uint32_t *)(ctx->data);
4263 }
4264 
4265 struct mlx5_list_entry *
4266 flow_dv_mreg_create_cb(void *tool_ctx, void *cb_ctx)
4267 {
4268 	struct rte_eth_dev *dev = tool_ctx;
4269 	struct mlx5_priv *priv = dev->data->dev_private;
4270 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
4271 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4272 	struct rte_flow_error *error = ctx->error;
4273 	uint32_t idx = 0;
4274 	int ret;
4275 	uint32_t mark_id = *(uint32_t *)(ctx->data);
4276 	struct rte_flow_attr attr = {
4277 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
4278 		.ingress = 1,
4279 	};
4280 	struct mlx5_rte_flow_item_tag tag_spec = {
4281 		.data = mark_id,
4282 	};
4283 	struct rte_flow_item items[] = {
4284 		[1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
4285 	};
4286 	struct rte_flow_action_mark ftag = {
4287 		.id = mark_id,
4288 	};
4289 	struct mlx5_flow_action_copy_mreg cp_mreg = {
4290 		.dst = REG_B,
4291 		.src = REG_NON,
4292 	};
4293 	struct rte_flow_action_jump jump = {
4294 		.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
4295 	};
4296 	struct rte_flow_action actions[] = {
4297 		[3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
4298 	};
4299 
4300 	/* Fill the register fields in the flow. */
4301 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
4302 	if (ret < 0)
4303 		return NULL;
4304 	tag_spec.id = ret;
4305 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
4306 	if (ret < 0)
4307 		return NULL;
4308 	cp_mreg.src = ret;
4309 	/* Provide the full width of FLAG specific value. */
4310 	if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
4311 		tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
4312 	/* Build a new flow. */
4313 	if (mark_id != MLX5_DEFAULT_COPY_ID) {
4314 		items[0] = (struct rte_flow_item){
4315 			.type = (enum rte_flow_item_type)
4316 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
4317 			.spec = &tag_spec,
4318 		};
4319 		items[1] = (struct rte_flow_item){
4320 			.type = RTE_FLOW_ITEM_TYPE_END,
4321 		};
4322 		actions[0] = (struct rte_flow_action){
4323 			.type = (enum rte_flow_action_type)
4324 				MLX5_RTE_FLOW_ACTION_TYPE_MARK,
4325 			.conf = &ftag,
4326 		};
4327 		actions[1] = (struct rte_flow_action){
4328 			.type = (enum rte_flow_action_type)
4329 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4330 			.conf = &cp_mreg,
4331 		};
4332 		actions[2] = (struct rte_flow_action){
4333 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4334 			.conf = &jump,
4335 		};
4336 		actions[3] = (struct rte_flow_action){
4337 			.type = RTE_FLOW_ACTION_TYPE_END,
4338 		};
4339 	} else {
4340 		/* Default rule, wildcard match. */
4341 		attr.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR;
4342 		items[0] = (struct rte_flow_item){
4343 			.type = RTE_FLOW_ITEM_TYPE_END,
4344 		};
4345 		actions[0] = (struct rte_flow_action){
4346 			.type = (enum rte_flow_action_type)
4347 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
4348 			.conf = &cp_mreg,
4349 		};
4350 		actions[1] = (struct rte_flow_action){
4351 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
4352 			.conf = &jump,
4353 		};
4354 		actions[2] = (struct rte_flow_action){
4355 			.type = RTE_FLOW_ACTION_TYPE_END,
4356 		};
4357 	}
4358 	/* Build a new entry. */
4359 	mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4360 	if (!mcp_res) {
4361 		rte_errno = ENOMEM;
4362 		return NULL;
4363 	}
4364 	mcp_res->idx = idx;
4365 	mcp_res->mark_id = mark_id;
4366 	/*
4367 	 * The copy Flows are not included in any list. There
4368 	 * ones are referenced from other Flows and can not
4369 	 * be applied, removed, deleted in arbitrary order
4370 	 * by list traversing.
4371 	 */
4372 	mcp_res->rix_flow = flow_list_create(dev, MLX5_FLOW_TYPE_MCP,
4373 					&attr, items, actions, false, error);
4374 	if (!mcp_res->rix_flow) {
4375 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], idx);
4376 		return NULL;
4377 	}
4378 	return &mcp_res->hlist_ent;
4379 }
4380 
4381 struct mlx5_list_entry *
4382 flow_dv_mreg_clone_cb(void *tool_ctx, struct mlx5_list_entry *oentry,
4383 		      void *cb_ctx __rte_unused)
4384 {
4385 	struct rte_eth_dev *dev = tool_ctx;
4386 	struct mlx5_priv *priv = dev->data->dev_private;
4387 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4388 	uint32_t idx = 0;
4389 
4390 	mcp_res = mlx5_ipool_malloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
4391 	if (!mcp_res) {
4392 		rte_errno = ENOMEM;
4393 		return NULL;
4394 	}
4395 	memcpy(mcp_res, oentry, sizeof(*mcp_res));
4396 	mcp_res->idx = idx;
4397 	return &mcp_res->hlist_ent;
4398 }
4399 
4400 void
4401 flow_dv_mreg_clone_free_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4402 {
4403 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4404 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4405 	struct rte_eth_dev *dev = tool_ctx;
4406 	struct mlx5_priv *priv = dev->data->dev_private;
4407 
4408 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4409 }
4410 
4411 /**
4412  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4413  *
4414  * As mark_id is unique, if there's already a registered flow for the mark_id,
4415  * return by increasing the reference counter of the resource. Otherwise, create
4416  * the resource (mcp_res) and flow.
4417  *
4418  * Flow looks like,
4419  *   - If ingress port is ANY and reg_c[1] is mark_id,
4420  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4421  *
4422  * For default flow (zero mark_id), flow is like,
4423  *   - If ingress port is ANY,
4424  *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
4425  *
4426  * @param dev
4427  *   Pointer to Ethernet device.
4428  * @param mark_id
4429  *   ID of MARK action, zero means default flow for META.
4430  * @param[out] error
4431  *   Perform verbose error reporting if not NULL.
4432  *
4433  * @return
4434  *   Associated resource on success, NULL otherwise and rte_errno is set.
4435  */
4436 static struct mlx5_flow_mreg_copy_resource *
4437 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
4438 			  struct rte_flow_error *error)
4439 {
4440 	struct mlx5_priv *priv = dev->data->dev_private;
4441 	struct mlx5_list_entry *entry;
4442 	struct mlx5_flow_cb_ctx ctx = {
4443 		.dev = dev,
4444 		.error = error,
4445 		.data = &mark_id,
4446 	};
4447 
4448 	/* Check if already registered. */
4449 	MLX5_ASSERT(priv->mreg_cp_tbl);
4450 	entry = mlx5_hlist_register(priv->mreg_cp_tbl, mark_id, &ctx);
4451 	if (!entry)
4452 		return NULL;
4453 	return container_of(entry, struct mlx5_flow_mreg_copy_resource,
4454 			    hlist_ent);
4455 }
4456 
4457 void
4458 flow_dv_mreg_remove_cb(void *tool_ctx, struct mlx5_list_entry *entry)
4459 {
4460 	struct mlx5_flow_mreg_copy_resource *mcp_res =
4461 			       container_of(entry, typeof(*mcp_res), hlist_ent);
4462 	struct rte_eth_dev *dev = tool_ctx;
4463 	struct mlx5_priv *priv = dev->data->dev_private;
4464 
4465 	MLX5_ASSERT(mcp_res->rix_flow);
4466 	flow_list_destroy(dev, MLX5_FLOW_TYPE_MCP, mcp_res->rix_flow);
4467 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
4468 }
4469 
4470 /**
4471  * Release flow in RX_CP_TBL.
4472  *
4473  * @param dev
4474  *   Pointer to Ethernet device.
4475  * @flow
4476  *   Parent flow for wich copying is provided.
4477  */
4478 static void
4479 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
4480 			  struct rte_flow *flow)
4481 {
4482 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4483 	struct mlx5_priv *priv = dev->data->dev_private;
4484 
4485 	if (!flow->rix_mreg_copy)
4486 		return;
4487 	mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
4488 				 flow->rix_mreg_copy);
4489 	if (!mcp_res || !priv->mreg_cp_tbl)
4490 		return;
4491 	MLX5_ASSERT(mcp_res->rix_flow);
4492 	mlx5_hlist_unregister(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
4493 	flow->rix_mreg_copy = 0;
4494 }
4495 
4496 /**
4497  * Remove the default copy action from RX_CP_TBL.
4498  *
4499  * This functions is called in the mlx5_dev_start(). No thread safe
4500  * is guaranteed.
4501  *
4502  * @param dev
4503  *   Pointer to Ethernet device.
4504  */
4505 static void
4506 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
4507 {
4508 	struct mlx5_list_entry *entry;
4509 	struct mlx5_priv *priv = dev->data->dev_private;
4510 	struct mlx5_flow_cb_ctx ctx;
4511 	uint32_t mark_id;
4512 
4513 	/* Check if default flow is registered. */
4514 	if (!priv->mreg_cp_tbl)
4515 		return;
4516 	mark_id = MLX5_DEFAULT_COPY_ID;
4517 	ctx.data = &mark_id;
4518 	entry = mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx);
4519 	if (!entry)
4520 		return;
4521 	mlx5_hlist_unregister(priv->mreg_cp_tbl, entry);
4522 }
4523 
4524 /**
4525  * Add the default copy action in in RX_CP_TBL.
4526  *
4527  * This functions is called in the mlx5_dev_start(). No thread safe
4528  * is guaranteed.
4529  *
4530  * @param dev
4531  *   Pointer to Ethernet device.
4532  * @param[out] error
4533  *   Perform verbose error reporting if not NULL.
4534  *
4535  * @return
4536  *   0 for success, negative value otherwise and rte_errno is set.
4537  */
4538 static int
4539 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
4540 				  struct rte_flow_error *error)
4541 {
4542 	struct mlx5_priv *priv = dev->data->dev_private;
4543 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4544 	struct mlx5_flow_cb_ctx ctx;
4545 	uint32_t mark_id;
4546 
4547 	/* Check whether extensive metadata feature is engaged. */
4548 	if (!priv->sh->config.dv_flow_en ||
4549 	    priv->sh->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4550 	    !mlx5_flow_ext_mreg_supported(dev) ||
4551 	    !priv->sh->dv_regc0_mask)
4552 		return 0;
4553 	/*
4554 	 * Add default mreg copy flow may be called multiple time, but
4555 	 * only be called once in stop. Avoid register it twice.
4556 	 */
4557 	mark_id = MLX5_DEFAULT_COPY_ID;
4558 	ctx.data = &mark_id;
4559 	if (mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id, &ctx))
4560 		return 0;
4561 	mcp_res = flow_mreg_add_copy_action(dev, mark_id, error);
4562 	if (!mcp_res)
4563 		return -rte_errno;
4564 	return 0;
4565 }
4566 
4567 /**
4568  * Add a flow of copying flow metadata registers in RX_CP_TBL.
4569  *
4570  * All the flow having Q/RSS action should be split by
4571  * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
4572  * performs the following,
4573  *   - CQE->flow_tag := reg_c[1] (MARK)
4574  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
4575  * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
4576  * but there should be a flow per each MARK ID set by MARK action.
4577  *
4578  * For the aforementioned reason, if there's a MARK action in flow's action
4579  * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
4580  * the MARK ID to CQE's flow_tag like,
4581  *   - If reg_c[1] is mark_id,
4582  *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4583  *
4584  * For SET_META action which stores value in reg_c[0], as the destination is
4585  * also a flow metadata register (reg_b), adding a default flow is enough. Zero
4586  * MARK ID means the default flow. The default flow looks like,
4587  *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
4588  *
4589  * @param dev
4590  *   Pointer to Ethernet device.
4591  * @param flow
4592  *   Pointer to flow structure.
4593  * @param[in] actions
4594  *   Pointer to the list of actions.
4595  * @param[out] error
4596  *   Perform verbose error reporting if not NULL.
4597  *
4598  * @return
4599  *   0 on success, negative value otherwise and rte_errno is set.
4600  */
4601 static int
4602 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
4603 			    struct rte_flow *flow,
4604 			    const struct rte_flow_action *actions,
4605 			    struct rte_flow_error *error)
4606 {
4607 	struct mlx5_priv *priv = dev->data->dev_private;
4608 	struct mlx5_sh_config *config = &priv->sh->config;
4609 	struct mlx5_flow_mreg_copy_resource *mcp_res;
4610 	const struct rte_flow_action_mark *mark;
4611 
4612 	/* Check whether extensive metadata feature is engaged. */
4613 	if (!config->dv_flow_en ||
4614 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
4615 	    !mlx5_flow_ext_mreg_supported(dev) ||
4616 	    !priv->sh->dv_regc0_mask)
4617 		return 0;
4618 	/* Find MARK action. */
4619 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4620 		switch (actions->type) {
4621 		case RTE_FLOW_ACTION_TYPE_FLAG:
4622 			mcp_res = flow_mreg_add_copy_action
4623 				(dev, MLX5_FLOW_MARK_DEFAULT, error);
4624 			if (!mcp_res)
4625 				return -rte_errno;
4626 			flow->rix_mreg_copy = mcp_res->idx;
4627 			return 0;
4628 		case RTE_FLOW_ACTION_TYPE_MARK:
4629 			mark = (const struct rte_flow_action_mark *)
4630 				actions->conf;
4631 			mcp_res =
4632 				flow_mreg_add_copy_action(dev, mark->id, error);
4633 			if (!mcp_res)
4634 				return -rte_errno;
4635 			flow->rix_mreg_copy = mcp_res->idx;
4636 			return 0;
4637 		default:
4638 			break;
4639 		}
4640 	}
4641 	return 0;
4642 }
4643 
4644 #define MLX5_MAX_SPLIT_ACTIONS 24
4645 #define MLX5_MAX_SPLIT_ITEMS 24
4646 
4647 /**
4648  * Split the hairpin flow.
4649  * Since HW can't support encap and push-vlan on Rx, we move these
4650  * actions to Tx.
4651  * If the count action is after the encap then we also
4652  * move the count action. in this case the count will also measure
4653  * the outer bytes.
4654  *
4655  * @param dev
4656  *   Pointer to Ethernet device.
4657  * @param[in] actions
4658  *   Associated actions (list terminated by the END action).
4659  * @param[out] actions_rx
4660  *   Rx flow actions.
4661  * @param[out] actions_tx
4662  *   Tx flow actions..
4663  * @param[out] pattern_tx
4664  *   The pattern items for the Tx flow.
4665  * @param[out] flow_id
4666  *   The flow ID connected to this flow.
4667  *
4668  * @return
4669  *   0 on success.
4670  */
4671 static int
4672 flow_hairpin_split(struct rte_eth_dev *dev,
4673 		   const struct rte_flow_action actions[],
4674 		   struct rte_flow_action actions_rx[],
4675 		   struct rte_flow_action actions_tx[],
4676 		   struct rte_flow_item pattern_tx[],
4677 		   uint32_t flow_id)
4678 {
4679 	const struct rte_flow_action_raw_encap *raw_encap;
4680 	const struct rte_flow_action_raw_decap *raw_decap;
4681 	struct mlx5_rte_flow_action_set_tag *set_tag;
4682 	struct rte_flow_action *tag_action;
4683 	struct mlx5_rte_flow_item_tag *tag_item;
4684 	struct rte_flow_item *item;
4685 	char *addr;
4686 	int encap = 0;
4687 
4688 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
4689 		switch (actions->type) {
4690 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
4691 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
4692 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
4693 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
4694 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
4695 			rte_memcpy(actions_tx, actions,
4696 			       sizeof(struct rte_flow_action));
4697 			actions_tx++;
4698 			break;
4699 		case RTE_FLOW_ACTION_TYPE_COUNT:
4700 			if (encap) {
4701 				rte_memcpy(actions_tx, actions,
4702 					   sizeof(struct rte_flow_action));
4703 				actions_tx++;
4704 			} else {
4705 				rte_memcpy(actions_rx, actions,
4706 					   sizeof(struct rte_flow_action));
4707 				actions_rx++;
4708 			}
4709 			break;
4710 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4711 			raw_encap = actions->conf;
4712 			if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE) {
4713 				memcpy(actions_tx, actions,
4714 				       sizeof(struct rte_flow_action));
4715 				actions_tx++;
4716 				encap = 1;
4717 			} else {
4718 				rte_memcpy(actions_rx, actions,
4719 					   sizeof(struct rte_flow_action));
4720 				actions_rx++;
4721 			}
4722 			break;
4723 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
4724 			raw_decap = actions->conf;
4725 			if (raw_decap->size < MLX5_ENCAPSULATION_DECISION_SIZE) {
4726 				memcpy(actions_tx, actions,
4727 				       sizeof(struct rte_flow_action));
4728 				actions_tx++;
4729 			} else {
4730 				rte_memcpy(actions_rx, actions,
4731 					   sizeof(struct rte_flow_action));
4732 				actions_rx++;
4733 			}
4734 			break;
4735 		default:
4736 			rte_memcpy(actions_rx, actions,
4737 				   sizeof(struct rte_flow_action));
4738 			actions_rx++;
4739 			break;
4740 		}
4741 	}
4742 	/* Add set meta action and end action for the Rx flow. */
4743 	tag_action = actions_rx;
4744 	tag_action->type = (enum rte_flow_action_type)
4745 			   MLX5_RTE_FLOW_ACTION_TYPE_TAG;
4746 	actions_rx++;
4747 	rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
4748 	actions_rx++;
4749 	set_tag = (void *)actions_rx;
4750 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
4751 		.id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL),
4752 		.data = flow_id,
4753 	};
4754 	MLX5_ASSERT(set_tag->id > REG_NON);
4755 	tag_action->conf = set_tag;
4756 	/* Create Tx item list. */
4757 	rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
4758 	addr = (void *)&pattern_tx[2];
4759 	item = pattern_tx;
4760 	item->type = (enum rte_flow_item_type)
4761 		     MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4762 	tag_item = (void *)addr;
4763 	tag_item->data = flow_id;
4764 	tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
4765 	MLX5_ASSERT(set_tag->id > REG_NON);
4766 	item->spec = tag_item;
4767 	addr += sizeof(struct mlx5_rte_flow_item_tag);
4768 	tag_item = (void *)addr;
4769 	tag_item->data = UINT32_MAX;
4770 	tag_item->id = UINT16_MAX;
4771 	item->mask = tag_item;
4772 	item->last = NULL;
4773 	item++;
4774 	item->type = RTE_FLOW_ITEM_TYPE_END;
4775 	return 0;
4776 }
4777 
4778 /**
4779  * The last stage of splitting chain, just creates the subflow
4780  * without any modification.
4781  *
4782  * @param[in] dev
4783  *   Pointer to Ethernet device.
4784  * @param[in] flow
4785  *   Parent flow structure pointer.
4786  * @param[in, out] sub_flow
4787  *   Pointer to return the created subflow, may be NULL.
4788  * @param[in] attr
4789  *   Flow rule attributes.
4790  * @param[in] items
4791  *   Pattern specification (list terminated by the END pattern item).
4792  * @param[in] actions
4793  *   Associated actions (list terminated by the END action).
4794  * @param[in] flow_split_info
4795  *   Pointer to flow split info structure.
4796  * @param[out] error
4797  *   Perform verbose error reporting if not NULL.
4798  * @return
4799  *   0 on success, negative value otherwise
4800  */
4801 static int
4802 flow_create_split_inner(struct rte_eth_dev *dev,
4803 			struct rte_flow *flow,
4804 			struct mlx5_flow **sub_flow,
4805 			const struct rte_flow_attr *attr,
4806 			const struct rte_flow_item items[],
4807 			const struct rte_flow_action actions[],
4808 			struct mlx5_flow_split_info *flow_split_info,
4809 			struct rte_flow_error *error)
4810 {
4811 	struct mlx5_flow *dev_flow;
4812 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
4813 
4814 	dev_flow = flow_drv_prepare(dev, flow, attr, items, actions,
4815 				    flow_split_info->flow_idx, error);
4816 	if (!dev_flow)
4817 		return -rte_errno;
4818 	dev_flow->flow = flow;
4819 	dev_flow->external = flow_split_info->external;
4820 	dev_flow->skip_scale = flow_split_info->skip_scale;
4821 	/* Subflow object was created, we must include one in the list. */
4822 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4823 		      dev_flow->handle, next);
4824 	/*
4825 	 * If dev_flow is as one of the suffix flow, some actions in suffix
4826 	 * flow may need some user defined item layer flags, and pass the
4827 	 * Metadata rxq mark flag to suffix flow as well.
4828 	 */
4829 	if (flow_split_info->prefix_layers)
4830 		dev_flow->handle->layers = flow_split_info->prefix_layers;
4831 	if (flow_split_info->prefix_mark) {
4832 		MLX5_ASSERT(wks);
4833 		wks->mark = 1;
4834 	}
4835 	if (sub_flow)
4836 		*sub_flow = dev_flow;
4837 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4838 	dev_flow->dv.table_id = flow_split_info->table_id;
4839 #endif
4840 	return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
4841 }
4842 
4843 /**
4844  * Get the sub policy of a meter.
4845  *
4846  * @param[in] dev
4847  *   Pointer to Ethernet device.
4848  * @param[in] flow
4849  *   Parent flow structure pointer.
4850  * @param wks
4851  *   Pointer to thread flow work space.
4852  * @param[in] attr
4853  *   Flow rule attributes.
4854  * @param[in] items
4855  *   Pattern specification (list terminated by the END pattern item).
4856  * @param[out] error
4857  *   Perform verbose error reporting if not NULL.
4858  *
4859  * @return
4860  *   Pointer to the meter sub policy, NULL otherwise and rte_errno is set.
4861  */
4862 static struct mlx5_flow_meter_sub_policy *
4863 get_meter_sub_policy(struct rte_eth_dev *dev,
4864 		     struct rte_flow *flow,
4865 		     struct mlx5_flow_workspace *wks,
4866 		     const struct rte_flow_attr *attr,
4867 		     const struct rte_flow_item items[],
4868 		     struct rte_flow_error *error)
4869 {
4870 	struct mlx5_flow_meter_policy *policy;
4871 	struct mlx5_flow_meter_policy *final_policy;
4872 	struct mlx5_flow_meter_sub_policy *sub_policy = NULL;
4873 
4874 	policy = wks->policy;
4875 	final_policy = policy->is_hierarchy ? wks->final_policy : policy;
4876 	if (final_policy->is_rss || final_policy->is_queue) {
4877 		struct mlx5_flow_rss_desc rss_desc_v[MLX5_MTR_RTE_COLORS];
4878 		struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS] = {0};
4879 		uint32_t i;
4880 
4881 		/*
4882 		 * This is a tmp dev_flow,
4883 		 * no need to register any matcher for it in translate.
4884 		 */
4885 		wks->skip_matcher_reg = 1;
4886 		for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) {
4887 			struct mlx5_flow dev_flow = {0};
4888 			struct mlx5_flow_handle dev_handle = { {0} };
4889 			uint8_t fate = final_policy->act_cnt[i].fate_action;
4890 
4891 			if (fate == MLX5_FLOW_FATE_SHARED_RSS) {
4892 				const struct rte_flow_action_rss *rss_act =
4893 					final_policy->act_cnt[i].rss->conf;
4894 				struct rte_flow_action rss_actions[2] = {
4895 					[0] = {
4896 					.type = RTE_FLOW_ACTION_TYPE_RSS,
4897 					.conf = rss_act,
4898 					},
4899 					[1] = {
4900 					.type = RTE_FLOW_ACTION_TYPE_END,
4901 					.conf = NULL,
4902 					}
4903 				};
4904 
4905 				dev_flow.handle = &dev_handle;
4906 				dev_flow.ingress = attr->ingress;
4907 				dev_flow.flow = flow;
4908 				dev_flow.external = 0;
4909 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
4910 				dev_flow.dv.transfer = attr->transfer;
4911 #endif
4912 				/**
4913 				 * Translate RSS action to get rss hash fields.
4914 				 */
4915 				if (flow_drv_translate(dev, &dev_flow, attr,
4916 						items, rss_actions, error))
4917 					goto exit;
4918 				rss_desc_v[i] = wks->rss_desc;
4919 				rss_desc_v[i].key_len = MLX5_RSS_HASH_KEY_LEN;
4920 				rss_desc_v[i].hash_fields =
4921 						dev_flow.hash_fields;
4922 				rss_desc_v[i].queue_num =
4923 						rss_desc_v[i].hash_fields ?
4924 						rss_desc_v[i].queue_num : 1;
4925 				rss_desc_v[i].tunnel =
4926 						!!(dev_flow.handle->layers &
4927 						   MLX5_FLOW_LAYER_TUNNEL);
4928 				/* Use the RSS queues in the containers. */
4929 				rss_desc_v[i].queue =
4930 					(uint16_t *)(uintptr_t)rss_act->queue;
4931 				rss_desc[i] = &rss_desc_v[i];
4932 			} else if (fate == MLX5_FLOW_FATE_QUEUE) {
4933 				/* This is queue action. */
4934 				rss_desc_v[i] = wks->rss_desc;
4935 				rss_desc_v[i].key_len = 0;
4936 				rss_desc_v[i].hash_fields = 0;
4937 				rss_desc_v[i].queue =
4938 					&final_policy->act_cnt[i].queue;
4939 				rss_desc_v[i].queue_num = 1;
4940 				rss_desc[i] = &rss_desc_v[i];
4941 			} else {
4942 				rss_desc[i] = NULL;
4943 			}
4944 		}
4945 		sub_policy = flow_drv_meter_sub_policy_rss_prepare(dev,
4946 						flow, policy, rss_desc);
4947 	} else {
4948 		enum mlx5_meter_domain mtr_domain =
4949 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
4950 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
4951 						MLX5_MTR_DOMAIN_INGRESS);
4952 		sub_policy = policy->sub_policys[mtr_domain][0];
4953 	}
4954 	if (!sub_policy)
4955 		rte_flow_error_set(error, EINVAL,
4956 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4957 				   "Failed to get meter sub-policy.");
4958 exit:
4959 	return sub_policy;
4960 }
4961 
4962 /**
4963  * Split the meter flow.
4964  *
4965  * As meter flow will split to three sub flow, other than meter
4966  * action, the other actions make sense to only meter accepts
4967  * the packet. If it need to be dropped, no other additional
4968  * actions should be take.
4969  *
4970  * One kind of special action which decapsulates the L3 tunnel
4971  * header will be in the prefix sub flow, as not to take the
4972  * L3 tunnel header into account.
4973  *
4974  * @param[in] dev
4975  *   Pointer to Ethernet device.
4976  * @param[in] flow
4977  *   Parent flow structure pointer.
4978  * @param wks
4979  *   Pointer to thread flow work space.
4980  * @param[in] attr
4981  *   Flow rule attributes.
4982  * @param[in] items
4983  *   Pattern specification (list terminated by the END pattern item).
4984  * @param[out] sfx_items
4985  *   Suffix flow match items (list terminated by the END pattern item).
4986  * @param[in] actions
4987  *   Associated actions (list terminated by the END action).
4988  * @param[out] actions_sfx
4989  *   Suffix flow actions.
4990  * @param[out] actions_pre
4991  *   Prefix flow actions.
4992  * @param[out] mtr_flow_id
4993  *   Pointer to meter flow id.
4994  * @param[out] error
4995  *   Perform verbose error reporting if not NULL.
4996  *
4997  * @return
4998  *   0 on success, a negative errno value otherwise and rte_errno is set.
4999  */
5000 static int
5001 flow_meter_split_prep(struct rte_eth_dev *dev,
5002 		      struct rte_flow *flow,
5003 		      struct mlx5_flow_workspace *wks,
5004 		      const struct rte_flow_attr *attr,
5005 		      const struct rte_flow_item items[],
5006 		      struct rte_flow_item sfx_items[],
5007 		      const struct rte_flow_action actions[],
5008 		      struct rte_flow_action actions_sfx[],
5009 		      struct rte_flow_action actions_pre[],
5010 		      uint32_t *mtr_flow_id,
5011 		      struct rte_flow_error *error)
5012 {
5013 	struct mlx5_priv *priv = dev->data->dev_private;
5014 	struct mlx5_flow_meter_info *fm = wks->fm;
5015 	struct rte_flow_action *tag_action = NULL;
5016 	struct rte_flow_item *tag_item;
5017 	struct mlx5_rte_flow_action_set_tag *set_tag;
5018 	const struct rte_flow_action_raw_encap *raw_encap;
5019 	const struct rte_flow_action_raw_decap *raw_decap;
5020 	struct mlx5_rte_flow_item_tag *tag_item_spec;
5021 	struct mlx5_rte_flow_item_tag *tag_item_mask;
5022 	uint32_t tag_id = 0;
5023 	struct rte_flow_item *vlan_item_dst = NULL;
5024 	const struct rte_flow_item *vlan_item_src = NULL;
5025 	const struct rte_flow_item *orig_items = items;
5026 	struct rte_flow_action *hw_mtr_action;
5027 	struct rte_flow_action *action_pre_head = NULL;
5028 	int32_t flow_src_port = priv->representor_id;
5029 	bool mtr_first;
5030 	uint8_t mtr_id_offset = priv->mtr_reg_share ? MLX5_MTR_COLOR_BITS : 0;
5031 	uint8_t mtr_reg_bits = priv->mtr_reg_share ?
5032 				MLX5_MTR_IDLE_BITS_IN_COLOR_REG : MLX5_REG_BITS;
5033 	uint32_t flow_id = 0;
5034 	uint32_t flow_id_reversed = 0;
5035 	uint8_t flow_id_bits = 0;
5036 	int shift;
5037 
5038 	/* Prepare the suffix subflow items. */
5039 	tag_item = sfx_items++;
5040 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5041 		struct mlx5_priv *port_priv;
5042 		const struct rte_flow_item_port_id *pid_v;
5043 		int item_type = items->type;
5044 
5045 		switch (item_type) {
5046 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5047 			pid_v = items->spec;
5048 			MLX5_ASSERT(pid_v);
5049 			port_priv = mlx5_port_to_eswitch_info(pid_v->id, false);
5050 			if (!port_priv)
5051 				return rte_flow_error_set(error,
5052 						rte_errno,
5053 						RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
5054 						pid_v,
5055 						"Failed to get port info.");
5056 			flow_src_port = port_priv->representor_id;
5057 			if (!fm->def_policy && wks->policy->is_hierarchy &&
5058 			    flow_src_port != priv->representor_id) {
5059 				if (flow_drv_mtr_hierarchy_rule_create(dev,
5060 								flow, fm,
5061 								flow_src_port,
5062 								items,
5063 								error))
5064 					return -rte_errno;
5065 			}
5066 			memcpy(sfx_items, items, sizeof(*sfx_items));
5067 			sfx_items++;
5068 			break;
5069 		case RTE_FLOW_ITEM_TYPE_VLAN:
5070 			/* Determine if copy vlan item below. */
5071 			vlan_item_src = items;
5072 			vlan_item_dst = sfx_items++;
5073 			vlan_item_dst->type = RTE_FLOW_ITEM_TYPE_VOID;
5074 			break;
5075 		default:
5076 			break;
5077 		}
5078 	}
5079 	sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
5080 	sfx_items++;
5081 	mtr_first = priv->sh->meter_aso_en &&
5082 		(attr->egress || (attr->transfer && flow_src_port != UINT16_MAX));
5083 	/* For ASO meter, meter must be before tag in TX direction. */
5084 	if (mtr_first) {
5085 		action_pre_head = actions_pre++;
5086 		/* Leave space for tag action. */
5087 		tag_action = actions_pre++;
5088 	}
5089 	/* Prepare the actions for prefix and suffix flow. */
5090 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5091 		struct rte_flow_action *action_cur = NULL;
5092 
5093 		switch (actions->type) {
5094 		case RTE_FLOW_ACTION_TYPE_METER:
5095 			if (mtr_first) {
5096 				action_cur = action_pre_head;
5097 			} else {
5098 				/* Leave space for tag action. */
5099 				tag_action = actions_pre++;
5100 				action_cur = actions_pre++;
5101 			}
5102 			break;
5103 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5104 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5105 			action_cur = actions_pre++;
5106 			break;
5107 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5108 			raw_encap = actions->conf;
5109 			if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
5110 				action_cur = actions_pre++;
5111 			break;
5112 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5113 			raw_decap = actions->conf;
5114 			if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
5115 				action_cur = actions_pre++;
5116 			break;
5117 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5118 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5119 			if (vlan_item_dst && vlan_item_src) {
5120 				memcpy(vlan_item_dst, vlan_item_src,
5121 					sizeof(*vlan_item_dst));
5122 				/*
5123 				 * Convert to internal match item, it is used
5124 				 * for vlan push and set vid.
5125 				 */
5126 				vlan_item_dst->type = (enum rte_flow_item_type)
5127 						MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
5128 			}
5129 			break;
5130 		default:
5131 			break;
5132 		}
5133 		if (!action_cur)
5134 			action_cur = (fm->def_policy) ?
5135 					actions_sfx++ : actions_pre++;
5136 		memcpy(action_cur, actions, sizeof(struct rte_flow_action));
5137 	}
5138 	/* Add end action to the actions. */
5139 	actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
5140 	if (priv->sh->meter_aso_en) {
5141 		/**
5142 		 * For ASO meter, need to add an extra jump action explicitly,
5143 		 * to jump from meter to policer table.
5144 		 */
5145 		struct mlx5_flow_meter_sub_policy *sub_policy;
5146 		struct mlx5_flow_tbl_data_entry *tbl_data;
5147 
5148 		if (!fm->def_policy) {
5149 			sub_policy = get_meter_sub_policy(dev, flow, wks,
5150 							  attr, orig_items,
5151 							  error);
5152 			if (!sub_policy)
5153 				return -rte_errno;
5154 		} else {
5155 			enum mlx5_meter_domain mtr_domain =
5156 			attr->transfer ? MLX5_MTR_DOMAIN_TRANSFER :
5157 				(attr->egress ? MLX5_MTR_DOMAIN_EGRESS :
5158 						MLX5_MTR_DOMAIN_INGRESS);
5159 
5160 			sub_policy =
5161 			&priv->sh->mtrmng->def_policy[mtr_domain]->sub_policy;
5162 		}
5163 		tbl_data = container_of(sub_policy->tbl_rsc,
5164 					struct mlx5_flow_tbl_data_entry, tbl);
5165 		hw_mtr_action = actions_pre++;
5166 		hw_mtr_action->type = (enum rte_flow_action_type)
5167 				      MLX5_RTE_FLOW_ACTION_TYPE_JUMP;
5168 		hw_mtr_action->conf = tbl_data->jump.action;
5169 	}
5170 	actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
5171 	actions_pre++;
5172 	if (!tag_action)
5173 		return rte_flow_error_set(error, ENOMEM,
5174 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5175 					  NULL, "No tag action space.");
5176 	if (!mtr_flow_id) {
5177 		tag_action->type = RTE_FLOW_ACTION_TYPE_VOID;
5178 		goto exit;
5179 	}
5180 	/* Only default-policy Meter creates mtr flow id. */
5181 	if (fm->def_policy) {
5182 		mlx5_ipool_malloc(fm->flow_ipool, &tag_id);
5183 		if (!tag_id)
5184 			return rte_flow_error_set(error, ENOMEM,
5185 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5186 					"Failed to allocate meter flow id.");
5187 		flow_id = tag_id - 1;
5188 		flow_id_bits = (!flow_id) ? 1 :
5189 				(MLX5_REG_BITS - __builtin_clz(flow_id));
5190 		if ((flow_id_bits + priv->sh->mtrmng->max_mtr_bits) >
5191 		    mtr_reg_bits) {
5192 			mlx5_ipool_free(fm->flow_ipool, tag_id);
5193 			return rte_flow_error_set(error, EINVAL,
5194 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5195 					"Meter flow id exceeds max limit.");
5196 		}
5197 		if (flow_id_bits > priv->sh->mtrmng->max_mtr_flow_bits)
5198 			priv->sh->mtrmng->max_mtr_flow_bits = flow_id_bits;
5199 	}
5200 	/* Build tag actions and items for meter_id/meter flow_id. */
5201 	set_tag = (struct mlx5_rte_flow_action_set_tag *)actions_pre;
5202 	tag_item_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
5203 	tag_item_mask = tag_item_spec + 1;
5204 	/* Both flow_id and meter_id share the same register. */
5205 	*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5206 		.id = (enum modify_reg)mlx5_flow_get_reg_id(dev, MLX5_MTR_ID,
5207 							    0, error),
5208 		.offset = mtr_id_offset,
5209 		.length = mtr_reg_bits,
5210 		.data = flow->meter,
5211 	};
5212 	/*
5213 	 * The color Reg bits used by flow_id are growing from
5214 	 * msb to lsb, so must do bit reverse for flow_id val in RegC.
5215 	 */
5216 	for (shift = 0; shift < flow_id_bits; shift++)
5217 		flow_id_reversed = (flow_id_reversed << 1) |
5218 				((flow_id >> shift) & 0x1);
5219 	set_tag->data |=
5220 		flow_id_reversed << (mtr_reg_bits - flow_id_bits);
5221 	tag_item_spec->id = set_tag->id;
5222 	tag_item_spec->data = set_tag->data << mtr_id_offset;
5223 	tag_item_mask->data = UINT32_MAX << mtr_id_offset;
5224 	tag_action->type = (enum rte_flow_action_type)
5225 				MLX5_RTE_FLOW_ACTION_TYPE_TAG;
5226 	tag_action->conf = set_tag;
5227 	tag_item->type = (enum rte_flow_item_type)
5228 				MLX5_RTE_FLOW_ITEM_TYPE_TAG;
5229 	tag_item->spec = tag_item_spec;
5230 	tag_item->last = NULL;
5231 	tag_item->mask = tag_item_mask;
5232 exit:
5233 	if (mtr_flow_id)
5234 		*mtr_flow_id = tag_id;
5235 	return 0;
5236 }
5237 
5238 /**
5239  * Split action list having QUEUE/RSS for metadata register copy.
5240  *
5241  * Once Q/RSS action is detected in user's action list, the flow action
5242  * should be split in order to copy metadata registers, which will happen in
5243  * RX_CP_TBL like,
5244  *   - CQE->flow_tag := reg_c[1] (MARK)
5245  *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
5246  * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
5247  * This is because the last action of each flow must be a terminal action
5248  * (QUEUE, RSS or DROP).
5249  *
5250  * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
5251  * stored and kept in the mlx5_flow structure per each sub_flow.
5252  *
5253  * The Q/RSS action is replaced with,
5254  *   - SET_TAG, setting the allocated flow ID to reg_c[2].
5255  * And the following JUMP action is added at the end,
5256  *   - JUMP, to RX_CP_TBL.
5257  *
5258  * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
5259  * flow_create_split_metadata() routine. The flow will look like,
5260  *   - If flow ID matches (reg_c[2]), perform Q/RSS.
5261  *
5262  * @param dev
5263  *   Pointer to Ethernet device.
5264  * @param[out] split_actions
5265  *   Pointer to store split actions to jump to CP_TBL.
5266  * @param[in] actions
5267  *   Pointer to the list of original flow actions.
5268  * @param[in] qrss
5269  *   Pointer to the Q/RSS action.
5270  * @param[in] actions_n
5271  *   Number of original actions.
5272  * @param[in] mtr_sfx
5273  *   Check if it is in meter suffix table.
5274  * @param[out] error
5275  *   Perform verbose error reporting if not NULL.
5276  *
5277  * @return
5278  *   non-zero unique flow_id on success, otherwise 0 and
5279  *   error/rte_error are set.
5280  */
5281 static uint32_t
5282 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
5283 			  struct rte_flow_action *split_actions,
5284 			  const struct rte_flow_action *actions,
5285 			  const struct rte_flow_action *qrss,
5286 			  int actions_n, int mtr_sfx,
5287 			  struct rte_flow_error *error)
5288 {
5289 	struct mlx5_priv *priv = dev->data->dev_private;
5290 	struct mlx5_rte_flow_action_set_tag *set_tag;
5291 	struct rte_flow_action_jump *jump;
5292 	const int qrss_idx = qrss - actions;
5293 	uint32_t flow_id = 0;
5294 	int ret = 0;
5295 
5296 	/*
5297 	 * Given actions will be split
5298 	 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
5299 	 * - Add jump to mreg CP_TBL.
5300 	 * As a result, there will be one more action.
5301 	 */
5302 	memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
5303 	/* Count MLX5_RTE_FLOW_ACTION_TYPE_TAG. */
5304 	++actions_n;
5305 	set_tag = (void *)(split_actions + actions_n);
5306 	/*
5307 	 * If we are not the meter suffix flow, add the tag action.
5308 	 * Since meter suffix flow already has the tag added.
5309 	 */
5310 	if (!mtr_sfx) {
5311 		/*
5312 		 * Allocate the new subflow ID. This one is unique within
5313 		 * device and not shared with representors. Otherwise,
5314 		 * we would have to resolve multi-thread access synch
5315 		 * issue. Each flow on the shared device is appended
5316 		 * with source vport identifier, so the resulting
5317 		 * flows will be unique in the shared (by master and
5318 		 * representors) domain even if they have coinciding
5319 		 * IDs.
5320 		 */
5321 		mlx5_ipool_malloc(priv->sh->ipool
5322 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &flow_id);
5323 		if (!flow_id)
5324 			return rte_flow_error_set(error, ENOMEM,
5325 						  RTE_FLOW_ERROR_TYPE_ACTION,
5326 						  NULL, "can't allocate id "
5327 						  "for split Q/RSS subflow");
5328 		/* Internal SET_TAG action to set flow ID. */
5329 		*set_tag = (struct mlx5_rte_flow_action_set_tag){
5330 			.data = flow_id,
5331 		};
5332 		ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
5333 		if (ret < 0)
5334 			return ret;
5335 		set_tag->id = ret;
5336 		/* Construct new actions array. */
5337 		/* Replace QUEUE/RSS action. */
5338 		split_actions[qrss_idx] = (struct rte_flow_action){
5339 			.type = (enum rte_flow_action_type)
5340 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5341 			.conf = set_tag,
5342 		};
5343 	} else {
5344 		/*
5345 		 * If we are the suffix flow of meter, tag already exist.
5346 		 * Set the QUEUE/RSS action to void.
5347 		 */
5348 		split_actions[qrss_idx].type = RTE_FLOW_ACTION_TYPE_VOID;
5349 	}
5350 	/* JUMP action to jump to mreg copy table (CP_TBL). */
5351 	jump = (void *)(set_tag + 1);
5352 	*jump = (struct rte_flow_action_jump){
5353 		.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5354 	};
5355 	split_actions[actions_n - 2] = (struct rte_flow_action){
5356 		.type = RTE_FLOW_ACTION_TYPE_JUMP,
5357 		.conf = jump,
5358 	};
5359 	split_actions[actions_n - 1] = (struct rte_flow_action){
5360 		.type = RTE_FLOW_ACTION_TYPE_END,
5361 	};
5362 	return flow_id;
5363 }
5364 
5365 /**
5366  * Extend the given action list for Tx metadata copy.
5367  *
5368  * Copy the given action list to the ext_actions and add flow metadata register
5369  * copy action in order to copy reg_a set by WQE to reg_c[0].
5370  *
5371  * @param[out] ext_actions
5372  *   Pointer to the extended action list.
5373  * @param[in] actions
5374  *   Pointer to the list of actions.
5375  * @param[in] actions_n
5376  *   Number of actions in the list.
5377  * @param[out] error
5378  *   Perform verbose error reporting if not NULL.
5379  * @param[in] encap_idx
5380  *   The encap action index.
5381  *
5382  * @return
5383  *   0 on success, negative value otherwise
5384  */
5385 static int
5386 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
5387 		       struct rte_flow_action *ext_actions,
5388 		       const struct rte_flow_action *actions,
5389 		       int actions_n, struct rte_flow_error *error,
5390 		       int encap_idx)
5391 {
5392 	struct mlx5_flow_action_copy_mreg *cp_mreg =
5393 		(struct mlx5_flow_action_copy_mreg *)
5394 			(ext_actions + actions_n + 1);
5395 	int ret;
5396 
5397 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
5398 	if (ret < 0)
5399 		return ret;
5400 	cp_mreg->dst = ret;
5401 	ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
5402 	if (ret < 0)
5403 		return ret;
5404 	cp_mreg->src = ret;
5405 	if (encap_idx != 0)
5406 		memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
5407 	if (encap_idx == actions_n - 1) {
5408 		ext_actions[actions_n - 1] = (struct rte_flow_action){
5409 			.type = (enum rte_flow_action_type)
5410 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5411 			.conf = cp_mreg,
5412 		};
5413 		ext_actions[actions_n] = (struct rte_flow_action){
5414 			.type = RTE_FLOW_ACTION_TYPE_END,
5415 		};
5416 	} else {
5417 		ext_actions[encap_idx] = (struct rte_flow_action){
5418 			.type = (enum rte_flow_action_type)
5419 				MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5420 			.conf = cp_mreg,
5421 		};
5422 		memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
5423 				sizeof(*ext_actions) * (actions_n - encap_idx));
5424 	}
5425 	return 0;
5426 }
5427 
5428 /**
5429  * Check the match action from the action list.
5430  *
5431  * @param[in] actions
5432  *   Pointer to the list of actions.
5433  * @param[in] attr
5434  *   Flow rule attributes.
5435  * @param[in] action
5436  *   The action to be check if exist.
5437  * @param[out] match_action_pos
5438  *   Pointer to the position of the matched action if exists, otherwise is -1.
5439  * @param[out] qrss_action_pos
5440  *   Pointer to the position of the Queue/RSS action if exists, otherwise is -1.
5441  * @param[out] modify_after_mirror
5442  *   Pointer to the flag of modify action after FDB mirroring.
5443  *
5444  * @return
5445  *   > 0 the total number of actions.
5446  *   0 if not found match action in action list.
5447  */
5448 static int
5449 flow_check_match_action(const struct rte_flow_action actions[],
5450 			const struct rte_flow_attr *attr,
5451 			enum rte_flow_action_type action,
5452 			int *match_action_pos, int *qrss_action_pos,
5453 			int *modify_after_mirror)
5454 {
5455 	const struct rte_flow_action_sample *sample;
5456 	const struct rte_flow_action_raw_decap *decap;
5457 	int actions_n = 0;
5458 	uint32_t ratio = 0;
5459 	int sub_type = 0;
5460 	int flag = 0;
5461 	int fdb_mirror = 0;
5462 
5463 	*match_action_pos = -1;
5464 	*qrss_action_pos = -1;
5465 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5466 		if (actions->type == action) {
5467 			flag = 1;
5468 			*match_action_pos = actions_n;
5469 		}
5470 		switch (actions->type) {
5471 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5472 		case RTE_FLOW_ACTION_TYPE_RSS:
5473 			*qrss_action_pos = actions_n;
5474 			break;
5475 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
5476 			sample = actions->conf;
5477 			ratio = sample->ratio;
5478 			sub_type = ((const struct rte_flow_action *)
5479 					(sample->actions))->type;
5480 			if (ratio == 1 && attr->transfer)
5481 				fdb_mirror = 1;
5482 			break;
5483 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5484 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5485 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5486 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5487 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5488 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5489 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5490 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5491 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5492 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5493 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5494 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5495 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5496 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5497 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5498 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5499 		case RTE_FLOW_ACTION_TYPE_FLAG:
5500 		case RTE_FLOW_ACTION_TYPE_MARK:
5501 		case RTE_FLOW_ACTION_TYPE_SET_META:
5502 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5503 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5504 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5505 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5506 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5507 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5508 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5509 		case RTE_FLOW_ACTION_TYPE_MODIFY_FIELD:
5510 		case RTE_FLOW_ACTION_TYPE_METER:
5511 			if (fdb_mirror)
5512 				*modify_after_mirror = 1;
5513 			break;
5514 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5515 			decap = actions->conf;
5516 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5517 				;
5518 			actions_n++;
5519 			if (actions->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5520 				const struct rte_flow_action_raw_encap *encap =
5521 								actions->conf;
5522 				if (decap->size <=
5523 					MLX5_ENCAPSULATION_DECISION_SIZE &&
5524 				    encap->size >
5525 					MLX5_ENCAPSULATION_DECISION_SIZE)
5526 					/* L3 encap. */
5527 					break;
5528 			}
5529 			if (fdb_mirror)
5530 				*modify_after_mirror = 1;
5531 			break;
5532 		default:
5533 			break;
5534 		}
5535 		actions_n++;
5536 	}
5537 	if (flag && fdb_mirror && !*modify_after_mirror) {
5538 		/* FDB mirroring uses the destination array to implement
5539 		 * instead of FLOW_SAMPLER object.
5540 		 */
5541 		if (sub_type != RTE_FLOW_ACTION_TYPE_END)
5542 			flag = 0;
5543 	}
5544 	/* Count RTE_FLOW_ACTION_TYPE_END. */
5545 	return flag ? actions_n + 1 : 0;
5546 }
5547 
5548 #define SAMPLE_SUFFIX_ITEM 2
5549 
5550 /**
5551  * Split the sample flow.
5552  *
5553  * As sample flow will split to two sub flow, sample flow with
5554  * sample action, the other actions will move to new suffix flow.
5555  *
5556  * Also add unique tag id with tag action in the sample flow,
5557  * the same tag id will be as match in the suffix flow.
5558  *
5559  * @param dev
5560  *   Pointer to Ethernet device.
5561  * @param[in] add_tag
5562  *   Add extra tag action flag.
5563  * @param[out] sfx_items
5564  *   Suffix flow match items (list terminated by the END pattern item).
5565  * @param[in] actions
5566  *   Associated actions (list terminated by the END action).
5567  * @param[out] actions_sfx
5568  *   Suffix flow actions.
5569  * @param[out] actions_pre
5570  *   Prefix flow actions.
5571  * @param[in] actions_n
5572  *  The total number of actions.
5573  * @param[in] sample_action_pos
5574  *   The sample action position.
5575  * @param[in] qrss_action_pos
5576  *   The Queue/RSS action position.
5577  * @param[in] jump_table
5578  *   Add extra jump action flag.
5579  * @param[out] error
5580  *   Perform verbose error reporting if not NULL.
5581  *
5582  * @return
5583  *   0 on success, or unique flow_id, a negative errno value
5584  *   otherwise and rte_errno is set.
5585  */
5586 static int
5587 flow_sample_split_prep(struct rte_eth_dev *dev,
5588 		       int add_tag,
5589 		       struct rte_flow_item sfx_items[],
5590 		       const struct rte_flow_action actions[],
5591 		       struct rte_flow_action actions_sfx[],
5592 		       struct rte_flow_action actions_pre[],
5593 		       int actions_n,
5594 		       int sample_action_pos,
5595 		       int qrss_action_pos,
5596 		       int jump_table,
5597 		       struct rte_flow_error *error)
5598 {
5599 	struct mlx5_priv *priv = dev->data->dev_private;
5600 	struct mlx5_rte_flow_action_set_tag *set_tag;
5601 	struct mlx5_rte_flow_item_tag *tag_spec;
5602 	struct mlx5_rte_flow_item_tag *tag_mask;
5603 	struct rte_flow_action_jump *jump_action;
5604 	uint32_t tag_id = 0;
5605 	int index;
5606 	int append_index = 0;
5607 	int ret;
5608 
5609 	if (sample_action_pos < 0)
5610 		return rte_flow_error_set(error, EINVAL,
5611 					  RTE_FLOW_ERROR_TYPE_ACTION,
5612 					  NULL, "invalid position of sample "
5613 					  "action in list");
5614 	/* Prepare the actions for prefix and suffix flow. */
5615 	if (qrss_action_pos >= 0 && qrss_action_pos < sample_action_pos) {
5616 		index = qrss_action_pos;
5617 		/* Put the preceding the Queue/RSS action into prefix flow. */
5618 		if (index != 0)
5619 			memcpy(actions_pre, actions,
5620 			       sizeof(struct rte_flow_action) * index);
5621 		/* Put others preceding the sample action into prefix flow. */
5622 		if (sample_action_pos > index + 1)
5623 			memcpy(actions_pre + index, actions + index + 1,
5624 			       sizeof(struct rte_flow_action) *
5625 			       (sample_action_pos - index - 1));
5626 		index = sample_action_pos - 1;
5627 		/* Put Queue/RSS action into Suffix flow. */
5628 		memcpy(actions_sfx, actions + qrss_action_pos,
5629 		       sizeof(struct rte_flow_action));
5630 		actions_sfx++;
5631 	} else {
5632 		index = sample_action_pos;
5633 		if (index != 0)
5634 			memcpy(actions_pre, actions,
5635 			       sizeof(struct rte_flow_action) * index);
5636 	}
5637 	/* For CX5, add an extra tag action for NIC-RX and E-Switch ingress.
5638 	 * For CX6DX and above, metadata registers Cx preserve their value,
5639 	 * add an extra tag action for NIC-RX and E-Switch Domain.
5640 	 */
5641 	if (add_tag) {
5642 		/* Prepare the prefix tag action. */
5643 		append_index++;
5644 		set_tag = (void *)(actions_pre + actions_n + append_index);
5645 		ret = mlx5_flow_get_reg_id(dev, MLX5_SAMPLE_ID, 0, error);
5646 		if (ret < 0)
5647 			return ret;
5648 		mlx5_ipool_malloc(priv->sh->ipool
5649 				  [MLX5_IPOOL_RSS_EXPANTION_FLOW_ID], &tag_id);
5650 		*set_tag = (struct mlx5_rte_flow_action_set_tag) {
5651 			.id = ret,
5652 			.data = tag_id,
5653 		};
5654 		/* Prepare the suffix subflow items. */
5655 		tag_spec = (void *)(sfx_items + SAMPLE_SUFFIX_ITEM);
5656 		tag_spec->data = tag_id;
5657 		tag_spec->id = set_tag->id;
5658 		tag_mask = tag_spec + 1;
5659 		tag_mask->data = UINT32_MAX;
5660 		sfx_items[0] = (struct rte_flow_item){
5661 			.type = (enum rte_flow_item_type)
5662 				MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5663 			.spec = tag_spec,
5664 			.last = NULL,
5665 			.mask = tag_mask,
5666 		};
5667 		sfx_items[1] = (struct rte_flow_item){
5668 			.type = (enum rte_flow_item_type)
5669 				RTE_FLOW_ITEM_TYPE_END,
5670 		};
5671 		/* Prepare the tag action in prefix subflow. */
5672 		actions_pre[index++] =
5673 			(struct rte_flow_action){
5674 			.type = (enum rte_flow_action_type)
5675 				MLX5_RTE_FLOW_ACTION_TYPE_TAG,
5676 			.conf = set_tag,
5677 		};
5678 	}
5679 	memcpy(actions_pre + index, actions + sample_action_pos,
5680 	       sizeof(struct rte_flow_action));
5681 	index += 1;
5682 	/* For the modify action after the sample action in E-Switch mirroring,
5683 	 * Add the extra jump action in prefix subflow and jump into the next
5684 	 * table, then do the modify action in the new table.
5685 	 */
5686 	if (jump_table) {
5687 		/* Prepare the prefix jump action. */
5688 		append_index++;
5689 		jump_action = (void *)(actions_pre + actions_n + append_index);
5690 		jump_action->group = jump_table;
5691 		actions_pre[index++] =
5692 			(struct rte_flow_action){
5693 			.type = (enum rte_flow_action_type)
5694 				RTE_FLOW_ACTION_TYPE_JUMP,
5695 			.conf = jump_action,
5696 		};
5697 	}
5698 	actions_pre[index] = (struct rte_flow_action){
5699 		.type = (enum rte_flow_action_type)
5700 			RTE_FLOW_ACTION_TYPE_END,
5701 	};
5702 	/* Put the actions after sample into Suffix flow. */
5703 	memcpy(actions_sfx, actions + sample_action_pos + 1,
5704 	       sizeof(struct rte_flow_action) *
5705 	       (actions_n - sample_action_pos - 1));
5706 	return tag_id;
5707 }
5708 
5709 /**
5710  * The splitting for metadata feature.
5711  *
5712  * - Q/RSS action on NIC Rx should be split in order to pass by
5713  *   the mreg copy table (RX_CP_TBL) and then it jumps to the
5714  *   action table (RX_ACT_TBL) which has the split Q/RSS action.
5715  *
5716  * - All the actions on NIC Tx should have a mreg copy action to
5717  *   copy reg_a from WQE to reg_c[0].
5718  *
5719  * @param dev
5720  *   Pointer to Ethernet device.
5721  * @param[in] flow
5722  *   Parent flow structure pointer.
5723  * @param[in] attr
5724  *   Flow rule attributes.
5725  * @param[in] items
5726  *   Pattern specification (list terminated by the END pattern item).
5727  * @param[in] actions
5728  *   Associated actions (list terminated by the END action).
5729  * @param[in] flow_split_info
5730  *   Pointer to flow split info structure.
5731  * @param[out] error
5732  *   Perform verbose error reporting if not NULL.
5733  * @return
5734  *   0 on success, negative value otherwise
5735  */
5736 static int
5737 flow_create_split_metadata(struct rte_eth_dev *dev,
5738 			   struct rte_flow *flow,
5739 			   const struct rte_flow_attr *attr,
5740 			   const struct rte_flow_item items[],
5741 			   const struct rte_flow_action actions[],
5742 			   struct mlx5_flow_split_info *flow_split_info,
5743 			   struct rte_flow_error *error)
5744 {
5745 	struct mlx5_priv *priv = dev->data->dev_private;
5746 	struct mlx5_sh_config *config = &priv->sh->config;
5747 	const struct rte_flow_action *qrss = NULL;
5748 	struct rte_flow_action *ext_actions = NULL;
5749 	struct mlx5_flow *dev_flow = NULL;
5750 	uint32_t qrss_id = 0;
5751 	int mtr_sfx = 0;
5752 	size_t act_size;
5753 	int actions_n;
5754 	int encap_idx;
5755 	int ret;
5756 
5757 	/* Check whether extensive metadata feature is engaged. */
5758 	if (!config->dv_flow_en ||
5759 	    config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
5760 	    !mlx5_flow_ext_mreg_supported(dev))
5761 		return flow_create_split_inner(dev, flow, NULL, attr, items,
5762 					       actions, flow_split_info, error);
5763 	actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
5764 							   &encap_idx);
5765 	if (qrss) {
5766 		/* Exclude hairpin flows from splitting. */
5767 		if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
5768 			const struct rte_flow_action_queue *queue;
5769 
5770 			queue = qrss->conf;
5771 			if (mlx5_rxq_get_type(dev, queue->index) ==
5772 			    MLX5_RXQ_TYPE_HAIRPIN)
5773 				qrss = NULL;
5774 		} else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
5775 			const struct rte_flow_action_rss *rss;
5776 
5777 			rss = qrss->conf;
5778 			if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
5779 			    MLX5_RXQ_TYPE_HAIRPIN)
5780 				qrss = NULL;
5781 		}
5782 	}
5783 	if (qrss) {
5784 		/* Check if it is in meter suffix table. */
5785 		mtr_sfx = attr->group == (attr->transfer ?
5786 			  (MLX5_FLOW_TABLE_LEVEL_METER - 1) :
5787 			  MLX5_FLOW_TABLE_LEVEL_METER);
5788 		/*
5789 		 * Q/RSS action on NIC Rx should be split in order to pass by
5790 		 * the mreg copy table (RX_CP_TBL) and then it jumps to the
5791 		 * action table (RX_ACT_TBL) which has the split Q/RSS action.
5792 		 */
5793 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5794 			   sizeof(struct rte_flow_action_set_tag) +
5795 			   sizeof(struct rte_flow_action_jump);
5796 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5797 					  SOCKET_ID_ANY);
5798 		if (!ext_actions)
5799 			return rte_flow_error_set(error, ENOMEM,
5800 						  RTE_FLOW_ERROR_TYPE_ACTION,
5801 						  NULL, "no memory to split "
5802 						  "metadata flow");
5803 		/*
5804 		 * Create the new actions list with removed Q/RSS action
5805 		 * and appended set tag and jump to register copy table
5806 		 * (RX_CP_TBL). We should preallocate unique tag ID here
5807 		 * in advance, because it is needed for set tag action.
5808 		 */
5809 		qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
5810 						    qrss, actions_n,
5811 						    mtr_sfx, error);
5812 		if (!mtr_sfx && !qrss_id) {
5813 			ret = -rte_errno;
5814 			goto exit;
5815 		}
5816 	} else if (attr->egress && !attr->transfer) {
5817 		/*
5818 		 * All the actions on NIC Tx should have a metadata register
5819 		 * copy action to copy reg_a from WQE to reg_c[meta]
5820 		 */
5821 		act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
5822 			   sizeof(struct mlx5_flow_action_copy_mreg);
5823 		ext_actions = mlx5_malloc(MLX5_MEM_ZERO, act_size, 0,
5824 					  SOCKET_ID_ANY);
5825 		if (!ext_actions)
5826 			return rte_flow_error_set(error, ENOMEM,
5827 						  RTE_FLOW_ERROR_TYPE_ACTION,
5828 						  NULL, "no memory to split "
5829 						  "metadata flow");
5830 		/* Create the action list appended with copy register. */
5831 		ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
5832 					     actions_n, error, encap_idx);
5833 		if (ret < 0)
5834 			goto exit;
5835 	}
5836 	/* Add the unmodified original or prefix subflow. */
5837 	ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
5838 				      items, ext_actions ? ext_actions :
5839 				      actions, flow_split_info, error);
5840 	if (ret < 0)
5841 		goto exit;
5842 	MLX5_ASSERT(dev_flow);
5843 	if (qrss) {
5844 		const struct rte_flow_attr q_attr = {
5845 			.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5846 			.ingress = 1,
5847 		};
5848 		/* Internal PMD action to set register. */
5849 		struct mlx5_rte_flow_item_tag q_tag_spec = {
5850 			.data = qrss_id,
5851 			.id = REG_NON,
5852 		};
5853 		struct rte_flow_item q_items[] = {
5854 			{
5855 				.type = (enum rte_flow_item_type)
5856 					MLX5_RTE_FLOW_ITEM_TYPE_TAG,
5857 				.spec = &q_tag_spec,
5858 				.last = NULL,
5859 				.mask = NULL,
5860 			},
5861 			{
5862 				.type = RTE_FLOW_ITEM_TYPE_END,
5863 			},
5864 		};
5865 		struct rte_flow_action q_actions[] = {
5866 			{
5867 				.type = qrss->type,
5868 				.conf = qrss->conf,
5869 			},
5870 			{
5871 				.type = RTE_FLOW_ACTION_TYPE_END,
5872 			},
5873 		};
5874 		uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
5875 
5876 		/*
5877 		 * Configure the tag item only if there is no meter subflow.
5878 		 * Since tag is already marked in the meter suffix subflow
5879 		 * we can just use the meter suffix items as is.
5880 		 */
5881 		if (qrss_id) {
5882 			/* Not meter subflow. */
5883 			MLX5_ASSERT(!mtr_sfx);
5884 			/*
5885 			 * Put unique id in prefix flow due to it is destroyed
5886 			 * after suffix flow and id will be freed after there
5887 			 * is no actual flows with this id and identifier
5888 			 * reallocation becomes possible (for example, for
5889 			 * other flows in other threads).
5890 			 */
5891 			dev_flow->handle->split_flow_id = qrss_id;
5892 			ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
5893 						   error);
5894 			if (ret < 0)
5895 				goto exit;
5896 			q_tag_spec.id = ret;
5897 		}
5898 		dev_flow = NULL;
5899 		/* Add suffix subflow to execute Q/RSS. */
5900 		flow_split_info->prefix_layers = layers;
5901 		flow_split_info->prefix_mark = 0;
5902 		flow_split_info->table_id = 0;
5903 		ret = flow_create_split_inner(dev, flow, &dev_flow,
5904 					      &q_attr, mtr_sfx ? items :
5905 					      q_items, q_actions,
5906 					      flow_split_info, error);
5907 		if (ret < 0)
5908 			goto exit;
5909 		/* qrss ID should be freed if failed. */
5910 		qrss_id = 0;
5911 		MLX5_ASSERT(dev_flow);
5912 	}
5913 
5914 exit:
5915 	/*
5916 	 * We do not destroy the partially created sub_flows in case of error.
5917 	 * These ones are included into parent flow list and will be destroyed
5918 	 * by flow_drv_destroy.
5919 	 */
5920 	mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_RSS_EXPANTION_FLOW_ID],
5921 			qrss_id);
5922 	mlx5_free(ext_actions);
5923 	return ret;
5924 }
5925 
5926 /**
5927  * Create meter internal drop flow with the original pattern.
5928  *
5929  * @param dev
5930  *   Pointer to Ethernet device.
5931  * @param[in] flow
5932  *   Parent flow structure pointer.
5933  * @param[in] attr
5934  *   Flow rule attributes.
5935  * @param[in] items
5936  *   Pattern specification (list terminated by the END pattern item).
5937  * @param[in] flow_split_info
5938  *   Pointer to flow split info structure.
5939  * @param[in] fm
5940  *   Pointer to flow meter structure.
5941  * @param[out] error
5942  *   Perform verbose error reporting if not NULL.
5943  * @return
5944  *   0 on success, negative value otherwise
5945  */
5946 static uint32_t
5947 flow_meter_create_drop_flow_with_org_pattern(struct rte_eth_dev *dev,
5948 			struct rte_flow *flow,
5949 			const struct rte_flow_attr *attr,
5950 			const struct rte_flow_item items[],
5951 			struct mlx5_flow_split_info *flow_split_info,
5952 			struct mlx5_flow_meter_info *fm,
5953 			struct rte_flow_error *error)
5954 {
5955 	struct mlx5_flow *dev_flow = NULL;
5956 	struct rte_flow_attr drop_attr = *attr;
5957 	struct rte_flow_action drop_actions[3];
5958 	struct mlx5_flow_split_info drop_split_info = *flow_split_info;
5959 
5960 	MLX5_ASSERT(fm->drop_cnt);
5961 	drop_actions[0].type =
5962 		(enum rte_flow_action_type)MLX5_RTE_FLOW_ACTION_TYPE_COUNT;
5963 	drop_actions[0].conf = (void *)(uintptr_t)fm->drop_cnt;
5964 	drop_actions[1].type = RTE_FLOW_ACTION_TYPE_DROP;
5965 	drop_actions[1].conf = NULL;
5966 	drop_actions[2].type = RTE_FLOW_ACTION_TYPE_END;
5967 	drop_actions[2].conf = NULL;
5968 	drop_split_info.external = false;
5969 	drop_split_info.skip_scale |= 1 << MLX5_SCALE_FLOW_GROUP_BIT;
5970 	drop_split_info.table_id = MLX5_MTR_TABLE_ID_DROP;
5971 	drop_attr.group = MLX5_FLOW_TABLE_LEVEL_METER;
5972 	return flow_create_split_inner(dev, flow, &dev_flow,
5973 				&drop_attr, items, drop_actions,
5974 				&drop_split_info, error);
5975 }
5976 
5977 /**
5978  * The splitting for meter feature.
5979  *
5980  * - The meter flow will be split to two flows as prefix and
5981  *   suffix flow. The packets make sense only it pass the prefix
5982  *   meter action.
5983  *
5984  * - Reg_C_5 is used for the packet to match betweend prefix and
5985  *   suffix flow.
5986  *
5987  * @param dev
5988  *   Pointer to Ethernet device.
5989  * @param[in] flow
5990  *   Parent flow structure pointer.
5991  * @param[in] attr
5992  *   Flow rule attributes.
5993  * @param[in] items
5994  *   Pattern specification (list terminated by the END pattern item).
5995  * @param[in] actions
5996  *   Associated actions (list terminated by the END action).
5997  * @param[in] flow_split_info
5998  *   Pointer to flow split info structure.
5999  * @param[out] error
6000  *   Perform verbose error reporting if not NULL.
6001  * @return
6002  *   0 on success, negative value otherwise
6003  */
6004 static int
6005 flow_create_split_meter(struct rte_eth_dev *dev,
6006 			struct rte_flow *flow,
6007 			const struct rte_flow_attr *attr,
6008 			const struct rte_flow_item items[],
6009 			const struct rte_flow_action actions[],
6010 			struct mlx5_flow_split_info *flow_split_info,
6011 			struct rte_flow_error *error)
6012 {
6013 	struct mlx5_priv *priv = dev->data->dev_private;
6014 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6015 	struct rte_flow_action *sfx_actions = NULL;
6016 	struct rte_flow_action *pre_actions = NULL;
6017 	struct rte_flow_item *sfx_items = NULL;
6018 	struct mlx5_flow *dev_flow = NULL;
6019 	struct rte_flow_attr sfx_attr = *attr;
6020 	struct mlx5_flow_meter_info *fm = NULL;
6021 	uint8_t skip_scale_restore;
6022 	bool has_mtr = false;
6023 	bool has_modify = false;
6024 	bool set_mtr_reg = true;
6025 	bool is_mtr_hierarchy = false;
6026 	uint32_t meter_id = 0;
6027 	uint32_t mtr_idx = 0;
6028 	uint32_t mtr_flow_id = 0;
6029 	size_t act_size;
6030 	size_t item_size;
6031 	int actions_n = 0;
6032 	int ret = 0;
6033 
6034 	if (priv->mtr_en)
6035 		actions_n = flow_check_meter_action(dev, actions, &has_mtr,
6036 						    &has_modify, &meter_id);
6037 	if (has_mtr) {
6038 		if (flow->meter) {
6039 			fm = flow_dv_meter_find_by_idx(priv, flow->meter);
6040 			if (!fm)
6041 				return rte_flow_error_set(error, EINVAL,
6042 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6043 						NULL, "Meter not found.");
6044 		} else {
6045 			fm = mlx5_flow_meter_find(priv, meter_id, &mtr_idx);
6046 			if (!fm)
6047 				return rte_flow_error_set(error, EINVAL,
6048 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6049 						NULL, "Meter not found.");
6050 			ret = mlx5_flow_meter_attach(priv, fm,
6051 						     &sfx_attr, error);
6052 			if (ret)
6053 				return -rte_errno;
6054 			flow->meter = mtr_idx;
6055 		}
6056 		MLX5_ASSERT(wks);
6057 		wks->fm = fm;
6058 		if (!fm->def_policy) {
6059 			wks->policy = mlx5_flow_meter_policy_find(dev,
6060 								  fm->policy_id,
6061 								  NULL);
6062 			MLX5_ASSERT(wks->policy);
6063 			if (wks->policy->is_hierarchy) {
6064 				wks->final_policy =
6065 				mlx5_flow_meter_hierarchy_get_final_policy(dev,
6066 								wks->policy);
6067 				if (!wks->final_policy)
6068 					return rte_flow_error_set(error,
6069 					EINVAL,
6070 					RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6071 				"Failed to find terminal policy of hierarchy.");
6072 				is_mtr_hierarchy = true;
6073 			}
6074 		}
6075 		/*
6076 		 * If it isn't default-policy Meter, and
6077 		 * 1. There's no action in flow to change
6078 		 *    packet (modify/encap/decap etc.), OR
6079 		 * 2. No drop count needed for this meter.
6080 		 * 3. It's not meter hierarchy.
6081 		 * Then no need to use regC to save meter id anymore.
6082 		 */
6083 		if (!fm->def_policy && !is_mtr_hierarchy &&
6084 		    (!has_modify || !fm->drop_cnt))
6085 			set_mtr_reg = false;
6086 		/* Prefix actions: meter, decap, encap, tag, jump, end. */
6087 		act_size = sizeof(struct rte_flow_action) * (actions_n + 6) +
6088 			   sizeof(struct mlx5_rte_flow_action_set_tag);
6089 		/* Suffix items: tag, vlan, port id, end. */
6090 #define METER_SUFFIX_ITEM 4
6091 		item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
6092 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6093 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size + item_size),
6094 					  0, SOCKET_ID_ANY);
6095 		if (!sfx_actions)
6096 			return rte_flow_error_set(error, ENOMEM,
6097 						  RTE_FLOW_ERROR_TYPE_ACTION,
6098 						  NULL, "no memory to split "
6099 						  "meter flow");
6100 		sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
6101 			     act_size);
6102 		/* There's no suffix flow for meter of non-default policy. */
6103 		if (!fm->def_policy)
6104 			pre_actions = sfx_actions + 1;
6105 		else
6106 			pre_actions = sfx_actions + actions_n;
6107 		ret = flow_meter_split_prep(dev, flow, wks, &sfx_attr,
6108 					    items, sfx_items, actions,
6109 					    sfx_actions, pre_actions,
6110 					    (set_mtr_reg ? &mtr_flow_id : NULL),
6111 					    error);
6112 		if (ret) {
6113 			ret = -rte_errno;
6114 			goto exit;
6115 		}
6116 		/* Add the prefix subflow. */
6117 		skip_scale_restore = flow_split_info->skip_scale;
6118 		flow_split_info->skip_scale |=
6119 			1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6120 		ret = flow_create_split_inner(dev, flow, &dev_flow,
6121 					      attr, items, pre_actions,
6122 					      flow_split_info, error);
6123 		flow_split_info->skip_scale = skip_scale_restore;
6124 		if (ret) {
6125 			if (mtr_flow_id)
6126 				mlx5_ipool_free(fm->flow_ipool, mtr_flow_id);
6127 			ret = -rte_errno;
6128 			goto exit;
6129 		}
6130 		if (mtr_flow_id) {
6131 			dev_flow->handle->split_flow_id = mtr_flow_id;
6132 			dev_flow->handle->is_meter_flow_id = 1;
6133 		}
6134 		if (!fm->def_policy) {
6135 			if (!set_mtr_reg && fm->drop_cnt)
6136 				ret =
6137 			flow_meter_create_drop_flow_with_org_pattern(dev, flow,
6138 							&sfx_attr, items,
6139 							flow_split_info,
6140 							fm, error);
6141 			goto exit;
6142 		}
6143 		/* Setting the sfx group atrr. */
6144 		sfx_attr.group = sfx_attr.transfer ?
6145 				(MLX5_FLOW_TABLE_LEVEL_METER - 1) :
6146 				 MLX5_FLOW_TABLE_LEVEL_METER;
6147 		flow_split_info->prefix_layers =
6148 				flow_get_prefix_layer_flags(dev_flow);
6149 		flow_split_info->prefix_mark |= wks->mark;
6150 		flow_split_info->table_id = MLX5_MTR_TABLE_ID_SUFFIX;
6151 	}
6152 	/* Add the prefix subflow. */
6153 	ret = flow_create_split_metadata(dev, flow,
6154 					 &sfx_attr, sfx_items ?
6155 					 sfx_items : items,
6156 					 sfx_actions ? sfx_actions : actions,
6157 					 flow_split_info, error);
6158 exit:
6159 	if (sfx_actions)
6160 		mlx5_free(sfx_actions);
6161 	return ret;
6162 }
6163 
6164 /**
6165  * The splitting for sample feature.
6166  *
6167  * Once Sample action is detected in the action list, the flow actions should
6168  * be split into prefix sub flow and suffix sub flow.
6169  *
6170  * The original items remain in the prefix sub flow, all actions preceding the
6171  * sample action and the sample action itself will be copied to the prefix
6172  * sub flow, the actions following the sample action will be copied to the
6173  * suffix sub flow, Queue action always be located in the suffix sub flow.
6174  *
6175  * In order to make the packet from prefix sub flow matches with suffix sub
6176  * flow, an extra tag action be added into prefix sub flow, and the suffix sub
6177  * flow uses tag item with the unique flow id.
6178  *
6179  * @param dev
6180  *   Pointer to Ethernet device.
6181  * @param[in] flow
6182  *   Parent flow structure pointer.
6183  * @param[in] attr
6184  *   Flow rule attributes.
6185  * @param[in] items
6186  *   Pattern specification (list terminated by the END pattern item).
6187  * @param[in] actions
6188  *   Associated actions (list terminated by the END action).
6189  * @param[in] flow_split_info
6190  *   Pointer to flow split info structure.
6191  * @param[out] error
6192  *   Perform verbose error reporting if not NULL.
6193  * @return
6194  *   0 on success, negative value otherwise
6195  */
6196 static int
6197 flow_create_split_sample(struct rte_eth_dev *dev,
6198 			 struct rte_flow *flow,
6199 			 const struct rte_flow_attr *attr,
6200 			 const struct rte_flow_item items[],
6201 			 const struct rte_flow_action actions[],
6202 			 struct mlx5_flow_split_info *flow_split_info,
6203 			 struct rte_flow_error *error)
6204 {
6205 	struct mlx5_priv *priv = dev->data->dev_private;
6206 	struct rte_flow_action *sfx_actions = NULL;
6207 	struct rte_flow_action *pre_actions = NULL;
6208 	struct rte_flow_item *sfx_items = NULL;
6209 	struct mlx5_flow *dev_flow = NULL;
6210 	struct rte_flow_attr sfx_attr = *attr;
6211 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6212 	struct mlx5_flow_dv_sample_resource *sample_res;
6213 	struct mlx5_flow_tbl_data_entry *sfx_tbl_data;
6214 	struct mlx5_flow_tbl_resource *sfx_tbl;
6215 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
6216 #endif
6217 	size_t act_size;
6218 	size_t item_size;
6219 	uint32_t fdb_tx = 0;
6220 	int32_t tag_id = 0;
6221 	int actions_n = 0;
6222 	int sample_action_pos;
6223 	int qrss_action_pos;
6224 	int add_tag = 0;
6225 	int modify_after_mirror = 0;
6226 	uint16_t jump_table = 0;
6227 	const uint32_t next_ft_step = 1;
6228 	int ret = 0;
6229 
6230 	if (priv->sampler_en)
6231 		actions_n = flow_check_match_action(actions, attr,
6232 					RTE_FLOW_ACTION_TYPE_SAMPLE,
6233 					&sample_action_pos, &qrss_action_pos,
6234 					&modify_after_mirror);
6235 	if (actions_n) {
6236 		/* The prefix actions must includes sample, tag, end. */
6237 		act_size = sizeof(struct rte_flow_action) * (actions_n * 2 + 1)
6238 			   + sizeof(struct mlx5_rte_flow_action_set_tag);
6239 		item_size = sizeof(struct rte_flow_item) * SAMPLE_SUFFIX_ITEM +
6240 			    sizeof(struct mlx5_rte_flow_item_tag) * 2;
6241 		sfx_actions = mlx5_malloc(MLX5_MEM_ZERO, (act_size +
6242 					  item_size), 0, SOCKET_ID_ANY);
6243 		if (!sfx_actions)
6244 			return rte_flow_error_set(error, ENOMEM,
6245 						  RTE_FLOW_ERROR_TYPE_ACTION,
6246 						  NULL, "no memory to split "
6247 						  "sample flow");
6248 		/* The representor_id is UINT16_MAX for uplink. */
6249 		fdb_tx = (attr->transfer && priv->representor_id != UINT16_MAX);
6250 		/*
6251 		 * When reg_c_preserve is set, metadata registers Cx preserve
6252 		 * their value even through packet duplication.
6253 		 */
6254 		add_tag = (!fdb_tx ||
6255 			   priv->sh->cdev->config.hca_attr.reg_c_preserve);
6256 		if (add_tag)
6257 			sfx_items = (struct rte_flow_item *)((char *)sfx_actions
6258 					+ act_size);
6259 		if (modify_after_mirror)
6260 			jump_table = attr->group * MLX5_FLOW_TABLE_FACTOR +
6261 				     next_ft_step;
6262 		pre_actions = sfx_actions + actions_n;
6263 		tag_id = flow_sample_split_prep(dev, add_tag, sfx_items,
6264 						actions, sfx_actions,
6265 						pre_actions, actions_n,
6266 						sample_action_pos,
6267 						qrss_action_pos, jump_table,
6268 						error);
6269 		if (tag_id < 0 || (add_tag && !tag_id)) {
6270 			ret = -rte_errno;
6271 			goto exit;
6272 		}
6273 		if (modify_after_mirror)
6274 			flow_split_info->skip_scale =
6275 					1 << MLX5_SCALE_JUMP_FLOW_GROUP_BIT;
6276 		/* Add the prefix subflow. */
6277 		ret = flow_create_split_inner(dev, flow, &dev_flow, attr,
6278 					      items, pre_actions,
6279 					      flow_split_info, error);
6280 		if (ret) {
6281 			ret = -rte_errno;
6282 			goto exit;
6283 		}
6284 		dev_flow->handle->split_flow_id = tag_id;
6285 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
6286 		if (!modify_after_mirror) {
6287 			/* Set the sfx group attr. */
6288 			sample_res = (struct mlx5_flow_dv_sample_resource *)
6289 						dev_flow->dv.sample_res;
6290 			sfx_tbl = (struct mlx5_flow_tbl_resource *)
6291 						sample_res->normal_path_tbl;
6292 			sfx_tbl_data = container_of(sfx_tbl,
6293 						struct mlx5_flow_tbl_data_entry,
6294 						tbl);
6295 			sfx_attr.group = sfx_attr.transfer ?
6296 			(sfx_tbl_data->level - 1) : sfx_tbl_data->level;
6297 		} else {
6298 			MLX5_ASSERT(attr->transfer);
6299 			sfx_attr.group = jump_table;
6300 		}
6301 		flow_split_info->prefix_layers =
6302 				flow_get_prefix_layer_flags(dev_flow);
6303 		MLX5_ASSERT(wks);
6304 		flow_split_info->prefix_mark |= wks->mark;
6305 		/* Suffix group level already be scaled with factor, set
6306 		 * MLX5_SCALE_FLOW_GROUP_BIT of skip_scale to 1 to avoid scale
6307 		 * again in translation.
6308 		 */
6309 		flow_split_info->skip_scale = 1 << MLX5_SCALE_FLOW_GROUP_BIT;
6310 #endif
6311 	}
6312 	/* Add the suffix subflow. */
6313 	ret = flow_create_split_meter(dev, flow, &sfx_attr,
6314 				      sfx_items ? sfx_items : items,
6315 				      sfx_actions ? sfx_actions : actions,
6316 				      flow_split_info, error);
6317 exit:
6318 	if (sfx_actions)
6319 		mlx5_free(sfx_actions);
6320 	return ret;
6321 }
6322 
6323 /**
6324  * Split the flow to subflow set. The splitters might be linked
6325  * in the chain, like this:
6326  * flow_create_split_outer() calls:
6327  *   flow_create_split_meter() calls:
6328  *     flow_create_split_metadata(meter_subflow_0) calls:
6329  *       flow_create_split_inner(metadata_subflow_0)
6330  *       flow_create_split_inner(metadata_subflow_1)
6331  *       flow_create_split_inner(metadata_subflow_2)
6332  *     flow_create_split_metadata(meter_subflow_1) calls:
6333  *       flow_create_split_inner(metadata_subflow_0)
6334  *       flow_create_split_inner(metadata_subflow_1)
6335  *       flow_create_split_inner(metadata_subflow_2)
6336  *
6337  * This provide flexible way to add new levels of flow splitting.
6338  * The all of successfully created subflows are included to the
6339  * parent flow dev_flow list.
6340  *
6341  * @param dev
6342  *   Pointer to Ethernet device.
6343  * @param[in] flow
6344  *   Parent flow structure pointer.
6345  * @param[in] attr
6346  *   Flow rule attributes.
6347  * @param[in] items
6348  *   Pattern specification (list terminated by the END pattern item).
6349  * @param[in] actions
6350  *   Associated actions (list terminated by the END action).
6351  * @param[in] flow_split_info
6352  *   Pointer to flow split info structure.
6353  * @param[out] error
6354  *   Perform verbose error reporting if not NULL.
6355  * @return
6356  *   0 on success, negative value otherwise
6357  */
6358 static int
6359 flow_create_split_outer(struct rte_eth_dev *dev,
6360 			struct rte_flow *flow,
6361 			const struct rte_flow_attr *attr,
6362 			const struct rte_flow_item items[],
6363 			const struct rte_flow_action actions[],
6364 			struct mlx5_flow_split_info *flow_split_info,
6365 			struct rte_flow_error *error)
6366 {
6367 	int ret;
6368 
6369 	ret = flow_create_split_sample(dev, flow, attr, items,
6370 				       actions, flow_split_info, error);
6371 	MLX5_ASSERT(ret <= 0);
6372 	return ret;
6373 }
6374 
6375 static inline struct mlx5_flow_tunnel *
6376 flow_tunnel_from_rule(const struct mlx5_flow *flow)
6377 {
6378 	struct mlx5_flow_tunnel *tunnel;
6379 
6380 #pragma GCC diagnostic push
6381 #pragma GCC diagnostic ignored "-Wcast-qual"
6382 	tunnel = (typeof(tunnel))flow->tunnel;
6383 #pragma GCC diagnostic pop
6384 
6385 	return tunnel;
6386 }
6387 
6388 /**
6389  * Adjust flow RSS workspace if needed.
6390  *
6391  * @param wks
6392  *   Pointer to thread flow work space.
6393  * @param rss_desc
6394  *   Pointer to RSS descriptor.
6395  * @param[in] nrssq_num
6396  *   New RSS queue number.
6397  *
6398  * @return
6399  *   0 on success, -1 otherwise and rte_errno is set.
6400  */
6401 static int
6402 flow_rss_workspace_adjust(struct mlx5_flow_workspace *wks,
6403 			  struct mlx5_flow_rss_desc *rss_desc,
6404 			  uint32_t nrssq_num)
6405 {
6406 	if (likely(nrssq_num <= wks->rssq_num))
6407 		return 0;
6408 	rss_desc->queue = realloc(rss_desc->queue,
6409 			  sizeof(*rss_desc->queue) * RTE_ALIGN(nrssq_num, 2));
6410 	if (!rss_desc->queue) {
6411 		rte_errno = ENOMEM;
6412 		return -1;
6413 	}
6414 	wks->rssq_num = RTE_ALIGN(nrssq_num, 2);
6415 	return 0;
6416 }
6417 
6418 /**
6419  * Create a flow and add it to @p list.
6420  *
6421  * @param dev
6422  *   Pointer to Ethernet device.
6423  * @param list
6424  *   Pointer to a TAILQ flow list. If this parameter NULL,
6425  *   no list insertion occurred, flow is just created,
6426  *   this is caller's responsibility to track the
6427  *   created flow.
6428  * @param[in] attr
6429  *   Flow rule attributes.
6430  * @param[in] items
6431  *   Pattern specification (list terminated by the END pattern item).
6432  * @param[in] actions
6433  *   Associated actions (list terminated by the END action).
6434  * @param[in] external
6435  *   This flow rule is created by request external to PMD.
6436  * @param[out] error
6437  *   Perform verbose error reporting if not NULL.
6438  *
6439  * @return
6440  *   A flow index on success, 0 otherwise and rte_errno is set.
6441  */
6442 static uint32_t
6443 flow_list_create(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6444 		 const struct rte_flow_attr *attr,
6445 		 const struct rte_flow_item items[],
6446 		 const struct rte_flow_action original_actions[],
6447 		 bool external, struct rte_flow_error *error)
6448 {
6449 	struct mlx5_priv *priv = dev->data->dev_private;
6450 	struct rte_flow *flow = NULL;
6451 	struct mlx5_flow *dev_flow;
6452 	const struct rte_flow_action_rss *rss = NULL;
6453 	struct mlx5_translated_action_handle
6454 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6455 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6456 	union {
6457 		struct mlx5_flow_expand_rss buf;
6458 		uint8_t buffer[4096];
6459 	} expand_buffer;
6460 	union {
6461 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6462 		uint8_t buffer[2048];
6463 	} actions_rx;
6464 	union {
6465 		struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
6466 		uint8_t buffer[2048];
6467 	} actions_hairpin_tx;
6468 	union {
6469 		struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
6470 		uint8_t buffer[2048];
6471 	} items_tx;
6472 	struct mlx5_flow_expand_rss *buf = &expand_buffer.buf;
6473 	struct mlx5_flow_rss_desc *rss_desc;
6474 	const struct rte_flow_action *p_actions_rx;
6475 	uint32_t i;
6476 	uint32_t idx = 0;
6477 	int hairpin_flow;
6478 	struct rte_flow_attr attr_tx = { .priority = 0 };
6479 	const struct rte_flow_action *actions;
6480 	struct rte_flow_action *translated_actions = NULL;
6481 	struct mlx5_flow_tunnel *tunnel;
6482 	struct tunnel_default_miss_ctx default_miss_ctx = { 0, };
6483 	struct mlx5_flow_workspace *wks = mlx5_flow_push_thread_workspace();
6484 	struct mlx5_flow_split_info flow_split_info = {
6485 		.external = !!external,
6486 		.skip_scale = 0,
6487 		.flow_idx = 0,
6488 		.prefix_mark = 0,
6489 		.prefix_layers = 0,
6490 		.table_id = 0
6491 	};
6492 	int ret;
6493 
6494 	MLX5_ASSERT(wks);
6495 	rss_desc = &wks->rss_desc;
6496 	ret = flow_action_handles_translate(dev, original_actions,
6497 					    indir_actions,
6498 					    &indir_actions_n,
6499 					    &translated_actions, error);
6500 	if (ret < 0) {
6501 		MLX5_ASSERT(translated_actions == NULL);
6502 		return 0;
6503 	}
6504 	actions = translated_actions ? translated_actions : original_actions;
6505 	p_actions_rx = actions;
6506 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6507 	ret = flow_drv_validate(dev, attr, items, p_actions_rx,
6508 				external, hairpin_flow, error);
6509 	if (ret < 0)
6510 		goto error_before_hairpin_split;
6511 	flow = mlx5_ipool_zmalloc(priv->flows[type], &idx);
6512 	if (!flow) {
6513 		rte_errno = ENOMEM;
6514 		goto error_before_hairpin_split;
6515 	}
6516 	if (hairpin_flow > 0) {
6517 		if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
6518 			rte_errno = EINVAL;
6519 			goto error_before_hairpin_split;
6520 		}
6521 		flow_hairpin_split(dev, actions, actions_rx.actions,
6522 				   actions_hairpin_tx.actions, items_tx.items,
6523 				   idx);
6524 		p_actions_rx = actions_rx.actions;
6525 	}
6526 	flow_split_info.flow_idx = idx;
6527 	flow->drv_type = flow_get_drv_type(dev, attr);
6528 	MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
6529 		    flow->drv_type < MLX5_FLOW_TYPE_MAX);
6530 	memset(rss_desc, 0, offsetof(struct mlx5_flow_rss_desc, queue));
6531 	/* RSS Action only works on NIC RX domain */
6532 	if (attr->ingress && !attr->transfer)
6533 		rss = flow_get_rss_action(dev, p_actions_rx);
6534 	if (rss) {
6535 		if (flow_rss_workspace_adjust(wks, rss_desc, rss->queue_num))
6536 			return 0;
6537 		/*
6538 		 * The following information is required by
6539 		 * mlx5_flow_hashfields_adjust() in advance.
6540 		 */
6541 		rss_desc->level = rss->level;
6542 		/* RSS type 0 indicates default RSS type (RTE_ETH_RSS_IP). */
6543 		rss_desc->types = !rss->types ? RTE_ETH_RSS_IP : rss->types;
6544 	}
6545 	flow->dev_handles = 0;
6546 	if (rss && rss->types) {
6547 		unsigned int graph_root;
6548 
6549 		graph_root = find_graph_root(rss->level);
6550 		ret = mlx5_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
6551 					   items, rss->types,
6552 					   mlx5_support_expansion, graph_root);
6553 		MLX5_ASSERT(ret > 0 &&
6554 		       (unsigned int)ret < sizeof(expand_buffer.buffer));
6555 		if (rte_log_can_log(mlx5_logtype, RTE_LOG_DEBUG)) {
6556 			for (i = 0; i < buf->entries; ++i)
6557 				mlx5_dbg__print_pattern(buf->entry[i].pattern);
6558 		}
6559 	} else {
6560 		buf->entries = 1;
6561 		buf->entry[0].pattern = (void *)(uintptr_t)items;
6562 	}
6563 	rss_desc->shared_rss = flow_get_shared_rss_action(dev, indir_actions,
6564 						      indir_actions_n);
6565 	for (i = 0; i < buf->entries; ++i) {
6566 		/* Initialize flow split data. */
6567 		flow_split_info.prefix_layers = 0;
6568 		flow_split_info.prefix_mark = 0;
6569 		flow_split_info.skip_scale = 0;
6570 		/*
6571 		 * The splitter may create multiple dev_flows,
6572 		 * depending on configuration. In the simplest
6573 		 * case it just creates unmodified original flow.
6574 		 */
6575 		ret = flow_create_split_outer(dev, flow, attr,
6576 					      buf->entry[i].pattern,
6577 					      p_actions_rx, &flow_split_info,
6578 					      error);
6579 		if (ret < 0)
6580 			goto error;
6581 		if (is_flow_tunnel_steer_rule(wks->flows[0].tof_type)) {
6582 			ret = flow_tunnel_add_default_miss(dev, flow, attr,
6583 							   p_actions_rx,
6584 							   idx,
6585 							   wks->flows[0].tunnel,
6586 							   &default_miss_ctx,
6587 							   error);
6588 			if (ret < 0) {
6589 				mlx5_free(default_miss_ctx.queue);
6590 				goto error;
6591 			}
6592 		}
6593 	}
6594 	/* Create the tx flow. */
6595 	if (hairpin_flow) {
6596 		attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
6597 		attr_tx.ingress = 0;
6598 		attr_tx.egress = 1;
6599 		dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
6600 					 actions_hairpin_tx.actions,
6601 					 idx, error);
6602 		if (!dev_flow)
6603 			goto error;
6604 		dev_flow->flow = flow;
6605 		dev_flow->external = 0;
6606 		SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
6607 			      dev_flow->handle, next);
6608 		ret = flow_drv_translate(dev, dev_flow, &attr_tx,
6609 					 items_tx.items,
6610 					 actions_hairpin_tx.actions, error);
6611 		if (ret < 0)
6612 			goto error;
6613 	}
6614 	/*
6615 	 * Update the metadata register copy table. If extensive
6616 	 * metadata feature is enabled and registers are supported
6617 	 * we might create the extra rte_flow for each unique
6618 	 * MARK/FLAG action ID.
6619 	 *
6620 	 * The table is updated for ingress Flows only, because
6621 	 * the egress Flows belong to the different device and
6622 	 * copy table should be updated in peer NIC Rx domain.
6623 	 */
6624 	if (attr->ingress &&
6625 	    (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
6626 		ret = flow_mreg_update_copy_table(dev, flow, actions, error);
6627 		if (ret)
6628 			goto error;
6629 	}
6630 	/*
6631 	 * If the flow is external (from application) OR device is started,
6632 	 * OR mreg discover, then apply immediately.
6633 	 */
6634 	if (external || dev->data->dev_started ||
6635 	    (attr->group == MLX5_FLOW_MREG_CP_TABLE_GROUP &&
6636 	     attr->priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)) {
6637 		ret = flow_drv_apply(dev, flow, error);
6638 		if (ret < 0)
6639 			goto error;
6640 	}
6641 	flow->type = type;
6642 	flow_rxq_flags_set(dev, flow);
6643 	rte_free(translated_actions);
6644 	tunnel = flow_tunnel_from_rule(wks->flows);
6645 	if (tunnel) {
6646 		flow->tunnel = 1;
6647 		flow->tunnel_id = tunnel->tunnel_id;
6648 		__atomic_add_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED);
6649 		mlx5_free(default_miss_ctx.queue);
6650 	}
6651 	mlx5_flow_pop_thread_workspace();
6652 	return idx;
6653 error:
6654 	MLX5_ASSERT(flow);
6655 	ret = rte_errno; /* Save rte_errno before cleanup. */
6656 	flow_mreg_del_copy_action(dev, flow);
6657 	flow_drv_destroy(dev, flow);
6658 	if (rss_desc->shared_rss)
6659 		__atomic_sub_fetch(&((struct mlx5_shared_action_rss *)
6660 			mlx5_ipool_get
6661 			(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
6662 			rss_desc->shared_rss))->refcnt, 1, __ATOMIC_RELAXED);
6663 	mlx5_ipool_free(priv->flows[type], idx);
6664 	rte_errno = ret; /* Restore rte_errno. */
6665 	ret = rte_errno;
6666 	rte_errno = ret;
6667 	mlx5_flow_pop_thread_workspace();
6668 error_before_hairpin_split:
6669 	rte_free(translated_actions);
6670 	return 0;
6671 }
6672 
6673 /**
6674  * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
6675  * incoming packets to table 1.
6676  *
6677  * Other flow rules, requested for group n, will be created in
6678  * e-switch table n+1.
6679  * Jump action to e-switch group n will be created to group n+1.
6680  *
6681  * Used when working in switchdev mode, to utilise advantages of table 1
6682  * and above.
6683  *
6684  * @param dev
6685  *   Pointer to Ethernet device.
6686  *
6687  * @return
6688  *   Pointer to flow on success, NULL otherwise and rte_errno is set.
6689  */
6690 struct rte_flow *
6691 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
6692 {
6693 	const struct rte_flow_attr attr = {
6694 		.group = 0,
6695 		.priority = 0,
6696 		.ingress = 1,
6697 		.egress = 0,
6698 		.transfer = 1,
6699 	};
6700 	const struct rte_flow_item pattern = {
6701 		.type = RTE_FLOW_ITEM_TYPE_END,
6702 	};
6703 	struct rte_flow_action_jump jump = {
6704 		.group = 1,
6705 	};
6706 	const struct rte_flow_action actions[] = {
6707 		{
6708 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6709 			.conf = &jump,
6710 		},
6711 		{
6712 			.type = RTE_FLOW_ACTION_TYPE_END,
6713 		},
6714 	};
6715 	struct rte_flow_error error;
6716 
6717 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
6718 						   &attr, &pattern,
6719 						   actions, false, &error);
6720 }
6721 
6722 /**
6723  * Create a dedicated flow rule on e-switch table 1, matches ESW manager
6724  * and sq number, directs all packets to peer vport.
6725  *
6726  * @param dev
6727  *   Pointer to Ethernet device.
6728  * @param txq
6729  *   Txq index.
6730  *
6731  * @return
6732  *   Flow ID on success, 0 otherwise and rte_errno is set.
6733  */
6734 uint32_t
6735 mlx5_flow_create_devx_sq_miss_flow(struct rte_eth_dev *dev, uint32_t txq)
6736 {
6737 	struct rte_flow_attr attr = {
6738 		.group = 0,
6739 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
6740 		.ingress = 1,
6741 		.egress = 0,
6742 		.transfer = 1,
6743 	};
6744 	struct rte_flow_item_port_id port_spec = {
6745 		.id = MLX5_PORT_ESW_MGR,
6746 	};
6747 	struct mlx5_rte_flow_item_tx_queue txq_spec = {
6748 		.queue = txq,
6749 	};
6750 	struct rte_flow_item pattern[] = {
6751 		{
6752 			.type = RTE_FLOW_ITEM_TYPE_PORT_ID,
6753 			.spec = &port_spec,
6754 		},
6755 		{
6756 			.type = (enum rte_flow_item_type)
6757 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
6758 			.spec = &txq_spec,
6759 		},
6760 		{
6761 			.type = RTE_FLOW_ITEM_TYPE_END,
6762 		},
6763 	};
6764 	struct rte_flow_action_jump jump = {
6765 		.group = 1,
6766 	};
6767 	struct rte_flow_action_port_id port = {
6768 		.id = dev->data->port_id,
6769 	};
6770 	struct rte_flow_action actions[] = {
6771 		{
6772 			.type = RTE_FLOW_ACTION_TYPE_JUMP,
6773 			.conf = &jump,
6774 		},
6775 		{
6776 			.type = RTE_FLOW_ACTION_TYPE_END,
6777 		},
6778 	};
6779 	struct rte_flow_error error;
6780 
6781 	/*
6782 	 * Creates group 0, highest priority jump flow.
6783 	 * Matches txq to bypass kernel packets.
6784 	 */
6785 	if (flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern, actions,
6786 			     false, &error) == 0)
6787 		return 0;
6788 	/* Create group 1, lowest priority redirect flow for txq. */
6789 	attr.group = 1;
6790 	actions[0].conf = &port;
6791 	actions[0].type = RTE_FLOW_ACTION_TYPE_PORT_ID;
6792 	return flow_list_create(dev, MLX5_FLOW_TYPE_CTL, &attr, pattern,
6793 				actions, false, &error);
6794 }
6795 
6796 /**
6797  * Validate a flow supported by the NIC.
6798  *
6799  * @see rte_flow_validate()
6800  * @see rte_flow_ops
6801  */
6802 int
6803 mlx5_flow_validate(struct rte_eth_dev *dev,
6804 		   const struct rte_flow_attr *attr,
6805 		   const struct rte_flow_item items[],
6806 		   const struct rte_flow_action original_actions[],
6807 		   struct rte_flow_error *error)
6808 {
6809 	int hairpin_flow;
6810 	struct mlx5_translated_action_handle
6811 		indir_actions[MLX5_MAX_INDIRECT_ACTIONS];
6812 	int indir_actions_n = MLX5_MAX_INDIRECT_ACTIONS;
6813 	const struct rte_flow_action *actions;
6814 	struct rte_flow_action *translated_actions = NULL;
6815 	int ret = flow_action_handles_translate(dev, original_actions,
6816 						indir_actions,
6817 						&indir_actions_n,
6818 						&translated_actions, error);
6819 
6820 	if (ret)
6821 		return ret;
6822 	actions = translated_actions ? translated_actions : original_actions;
6823 	hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
6824 	ret = flow_drv_validate(dev, attr, items, actions,
6825 				true, hairpin_flow, error);
6826 	rte_free(translated_actions);
6827 	return ret;
6828 }
6829 
6830 /**
6831  * Create a flow.
6832  *
6833  * @see rte_flow_create()
6834  * @see rte_flow_ops
6835  */
6836 struct rte_flow *
6837 mlx5_flow_create(struct rte_eth_dev *dev,
6838 		 const struct rte_flow_attr *attr,
6839 		 const struct rte_flow_item items[],
6840 		 const struct rte_flow_action actions[],
6841 		 struct rte_flow_error *error)
6842 {
6843 	struct mlx5_priv *priv = dev->data->dev_private;
6844 
6845 	if (priv->sh->config.dv_flow_en == 2) {
6846 		rte_flow_error_set(error, ENOTSUP,
6847 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6848 			  NULL,
6849 			  "Flow non-Q creation not supported");
6850 		return NULL;
6851 	}
6852 	/*
6853 	 * If the device is not started yet, it is not allowed to created a
6854 	 * flow from application. PMD default flows and traffic control flows
6855 	 * are not affected.
6856 	 */
6857 	if (unlikely(!dev->data->dev_started)) {
6858 		DRV_LOG(DEBUG, "port %u is not started when "
6859 			"inserting a flow", dev->data->port_id);
6860 		rte_flow_error_set(error, ENODEV,
6861 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
6862 				   NULL,
6863 				   "port not started");
6864 		return NULL;
6865 	}
6866 
6867 	return (void *)(uintptr_t)flow_list_create(dev, MLX5_FLOW_TYPE_GEN,
6868 						   attr, items, actions,
6869 						   true, error);
6870 }
6871 
6872 /**
6873  * Destroy a flow in a list.
6874  *
6875  * @param dev
6876  *   Pointer to Ethernet device.
6877  * @param[in] flow_idx
6878  *   Index of flow to destroy.
6879  */
6880 static void
6881 flow_list_destroy(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6882 		  uint32_t flow_idx)
6883 {
6884 	struct mlx5_priv *priv = dev->data->dev_private;
6885 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[type], flow_idx);
6886 
6887 	if (!flow)
6888 		return;
6889 	MLX5_ASSERT(flow->type == type);
6890 	/*
6891 	 * Update RX queue flags only if port is started, otherwise it is
6892 	 * already clean.
6893 	 */
6894 	if (dev->data->dev_started)
6895 		flow_rxq_flags_trim(dev, flow);
6896 	flow_drv_destroy(dev, flow);
6897 	if (flow->tunnel) {
6898 		struct mlx5_flow_tunnel *tunnel;
6899 
6900 		tunnel = mlx5_find_tunnel_id(dev, flow->tunnel_id);
6901 		RTE_VERIFY(tunnel);
6902 		if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
6903 			mlx5_flow_tunnel_free(dev, tunnel);
6904 	}
6905 	flow_mreg_del_copy_action(dev, flow);
6906 	mlx5_ipool_free(priv->flows[type], flow_idx);
6907 }
6908 
6909 /**
6910  * Destroy all flows.
6911  *
6912  * @param dev
6913  *   Pointer to Ethernet device.
6914  * @param type
6915  *   Flow type to be flushed.
6916  * @param active
6917  *   If flushing is called actively.
6918  */
6919 void
6920 mlx5_flow_list_flush(struct rte_eth_dev *dev, enum mlx5_flow_type type,
6921 		     bool active)
6922 {
6923 	struct mlx5_priv *priv = dev->data->dev_private;
6924 	uint32_t num_flushed = 0, fidx = 1;
6925 	struct rte_flow *flow;
6926 
6927 	MLX5_IPOOL_FOREACH(priv->flows[type], fidx, flow) {
6928 		flow_list_destroy(dev, type, fidx);
6929 		num_flushed++;
6930 	}
6931 	if (active) {
6932 		DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
6933 			dev->data->port_id, num_flushed);
6934 	}
6935 }
6936 
6937 /**
6938  * Stop all default actions for flows.
6939  *
6940  * @param dev
6941  *   Pointer to Ethernet device.
6942  */
6943 void
6944 mlx5_flow_stop_default(struct rte_eth_dev *dev)
6945 {
6946 	flow_mreg_del_default_copy_action(dev);
6947 	flow_rxq_flags_clear(dev);
6948 }
6949 
6950 /**
6951  * Start all default actions for flows.
6952  *
6953  * @param dev
6954  *   Pointer to Ethernet device.
6955  * @return
6956  *   0 on success, a negative errno value otherwise and rte_errno is set.
6957  */
6958 int
6959 mlx5_flow_start_default(struct rte_eth_dev *dev)
6960 {
6961 	struct rte_flow_error error;
6962 
6963 	/* Make sure default copy action (reg_c[0] -> reg_b) is created. */
6964 	return flow_mreg_add_default_copy_action(dev, &error);
6965 }
6966 
6967 /**
6968  * Release key of thread specific flow workspace data.
6969  */
6970 void
6971 flow_release_workspace(void *data)
6972 {
6973 	struct mlx5_flow_workspace *wks = data;
6974 	struct mlx5_flow_workspace *next;
6975 
6976 	while (wks) {
6977 		next = wks->next;
6978 		free(wks->rss_desc.queue);
6979 		free(wks);
6980 		wks = next;
6981 	}
6982 }
6983 
6984 /**
6985  * Get thread specific current flow workspace.
6986  *
6987  * @return pointer to thread specific flow workspace data, NULL on error.
6988  */
6989 struct mlx5_flow_workspace*
6990 mlx5_flow_get_thread_workspace(void)
6991 {
6992 	struct mlx5_flow_workspace *data;
6993 
6994 	data = mlx5_flow_os_get_specific_workspace();
6995 	MLX5_ASSERT(data && data->inuse);
6996 	if (!data || !data->inuse)
6997 		DRV_LOG(ERR, "flow workspace not initialized.");
6998 	return data;
6999 }
7000 
7001 /**
7002  * Allocate and init new flow workspace.
7003  *
7004  * @return pointer to flow workspace data, NULL on error.
7005  */
7006 static struct mlx5_flow_workspace*
7007 flow_alloc_thread_workspace(void)
7008 {
7009 	struct mlx5_flow_workspace *data = calloc(1, sizeof(*data));
7010 
7011 	if (!data) {
7012 		DRV_LOG(ERR, "Failed to allocate flow workspace "
7013 			"memory.");
7014 		return NULL;
7015 	}
7016 	data->rss_desc.queue = calloc(1,
7017 			sizeof(uint16_t) * MLX5_RSSQ_DEFAULT_NUM);
7018 	if (!data->rss_desc.queue)
7019 		goto err;
7020 	data->rssq_num = MLX5_RSSQ_DEFAULT_NUM;
7021 	return data;
7022 err:
7023 	free(data->rss_desc.queue);
7024 	free(data);
7025 	return NULL;
7026 }
7027 
7028 /**
7029  * Get new thread specific flow workspace.
7030  *
7031  * If current workspace inuse, create new one and set as current.
7032  *
7033  * @return pointer to thread specific flow workspace data, NULL on error.
7034  */
7035 static struct mlx5_flow_workspace*
7036 mlx5_flow_push_thread_workspace(void)
7037 {
7038 	struct mlx5_flow_workspace *curr;
7039 	struct mlx5_flow_workspace *data;
7040 
7041 	curr = mlx5_flow_os_get_specific_workspace();
7042 	if (!curr) {
7043 		data = flow_alloc_thread_workspace();
7044 		if (!data)
7045 			return NULL;
7046 	} else if (!curr->inuse) {
7047 		data = curr;
7048 	} else if (curr->next) {
7049 		data = curr->next;
7050 	} else {
7051 		data = flow_alloc_thread_workspace();
7052 		if (!data)
7053 			return NULL;
7054 		curr->next = data;
7055 		data->prev = curr;
7056 	}
7057 	data->inuse = 1;
7058 	data->flow_idx = 0;
7059 	/* Set as current workspace */
7060 	if (mlx5_flow_os_set_specific_workspace(data))
7061 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7062 	return data;
7063 }
7064 
7065 /**
7066  * Close current thread specific flow workspace.
7067  *
7068  * If previous workspace available, set it as current.
7069  *
7070  * @return pointer to thread specific flow workspace data, NULL on error.
7071  */
7072 static void
7073 mlx5_flow_pop_thread_workspace(void)
7074 {
7075 	struct mlx5_flow_workspace *data = mlx5_flow_get_thread_workspace();
7076 
7077 	if (!data)
7078 		return;
7079 	if (!data->inuse) {
7080 		DRV_LOG(ERR, "Failed to close unused flow workspace.");
7081 		return;
7082 	}
7083 	data->inuse = 0;
7084 	if (!data->prev)
7085 		return;
7086 	if (mlx5_flow_os_set_specific_workspace(data->prev))
7087 		DRV_LOG(ERR, "Failed to set flow workspace to thread.");
7088 }
7089 
7090 /**
7091  * Verify the flow list is empty
7092  *
7093  * @param dev
7094  *  Pointer to Ethernet device.
7095  *
7096  * @return the number of flows not released.
7097  */
7098 int
7099 mlx5_flow_verify(struct rte_eth_dev *dev __rte_unused)
7100 {
7101 	struct mlx5_priv *priv = dev->data->dev_private;
7102 	struct rte_flow *flow;
7103 	uint32_t idx = 0;
7104 	int ret = 0, i;
7105 
7106 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
7107 		MLX5_IPOOL_FOREACH(priv->flows[i], idx, flow) {
7108 			DRV_LOG(DEBUG, "port %u flow %p still referenced",
7109 				dev->data->port_id, (void *)flow);
7110 			ret++;
7111 		}
7112 	}
7113 	return ret;
7114 }
7115 
7116 /**
7117  * Enable default hairpin egress flow.
7118  *
7119  * @param dev
7120  *   Pointer to Ethernet device.
7121  * @param queue
7122  *   The queue index.
7123  *
7124  * @return
7125  *   0 on success, a negative errno value otherwise and rte_errno is set.
7126  */
7127 int
7128 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
7129 			    uint32_t queue)
7130 {
7131 	const struct rte_flow_attr attr = {
7132 		.egress = 1,
7133 		.priority = 0,
7134 	};
7135 	struct mlx5_rte_flow_item_tx_queue queue_spec = {
7136 		.queue = queue,
7137 	};
7138 	struct mlx5_rte_flow_item_tx_queue queue_mask = {
7139 		.queue = UINT32_MAX,
7140 	};
7141 	struct rte_flow_item items[] = {
7142 		{
7143 			.type = (enum rte_flow_item_type)
7144 				MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
7145 			.spec = &queue_spec,
7146 			.last = NULL,
7147 			.mask = &queue_mask,
7148 		},
7149 		{
7150 			.type = RTE_FLOW_ITEM_TYPE_END,
7151 		},
7152 	};
7153 	struct rte_flow_action_jump jump = {
7154 		.group = MLX5_HAIRPIN_TX_TABLE,
7155 	};
7156 	struct rte_flow_action actions[2];
7157 	uint32_t flow_idx;
7158 	struct rte_flow_error error;
7159 
7160 	actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
7161 	actions[0].conf = &jump;
7162 	actions[1].type = RTE_FLOW_ACTION_TYPE_END;
7163 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7164 				    &attr, items, actions, false, &error);
7165 	if (!flow_idx) {
7166 		DRV_LOG(DEBUG,
7167 			"Failed to create ctrl flow: rte_errno(%d),"
7168 			" type(%d), message(%s)",
7169 			rte_errno, error.type,
7170 			error.message ? error.message : " (no stated reason)");
7171 		return -rte_errno;
7172 	}
7173 	return 0;
7174 }
7175 
7176 /**
7177  * Enable a control flow configured from the control plane.
7178  *
7179  * @param dev
7180  *   Pointer to Ethernet device.
7181  * @param eth_spec
7182  *   An Ethernet flow spec to apply.
7183  * @param eth_mask
7184  *   An Ethernet flow mask to apply.
7185  * @param vlan_spec
7186  *   A VLAN flow spec to apply.
7187  * @param vlan_mask
7188  *   A VLAN flow mask to apply.
7189  *
7190  * @return
7191  *   0 on success, a negative errno value otherwise and rte_errno is set.
7192  */
7193 int
7194 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
7195 		    struct rte_flow_item_eth *eth_spec,
7196 		    struct rte_flow_item_eth *eth_mask,
7197 		    struct rte_flow_item_vlan *vlan_spec,
7198 		    struct rte_flow_item_vlan *vlan_mask)
7199 {
7200 	struct mlx5_priv *priv = dev->data->dev_private;
7201 	const struct rte_flow_attr attr = {
7202 		.ingress = 1,
7203 		.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
7204 	};
7205 	struct rte_flow_item items[] = {
7206 		{
7207 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7208 			.spec = eth_spec,
7209 			.last = NULL,
7210 			.mask = eth_mask,
7211 		},
7212 		{
7213 			.type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
7214 					      RTE_FLOW_ITEM_TYPE_END,
7215 			.spec = vlan_spec,
7216 			.last = NULL,
7217 			.mask = vlan_mask,
7218 		},
7219 		{
7220 			.type = RTE_FLOW_ITEM_TYPE_END,
7221 		},
7222 	};
7223 	uint16_t queue[priv->reta_idx_n];
7224 	struct rte_flow_action_rss action_rss = {
7225 		.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
7226 		.level = 0,
7227 		.types = priv->rss_conf.rss_hf,
7228 		.key_len = priv->rss_conf.rss_key_len,
7229 		.queue_num = priv->reta_idx_n,
7230 		.key = priv->rss_conf.rss_key,
7231 		.queue = queue,
7232 	};
7233 	struct rte_flow_action actions[] = {
7234 		{
7235 			.type = RTE_FLOW_ACTION_TYPE_RSS,
7236 			.conf = &action_rss,
7237 		},
7238 		{
7239 			.type = RTE_FLOW_ACTION_TYPE_END,
7240 		},
7241 	};
7242 	uint32_t flow_idx;
7243 	struct rte_flow_error error;
7244 	unsigned int i;
7245 
7246 	if (!priv->reta_idx_n || !priv->rxqs_n) {
7247 		return 0;
7248 	}
7249 	if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
7250 		action_rss.types = 0;
7251 	for (i = 0; i != priv->reta_idx_n; ++i)
7252 		queue[i] = (*priv->reta_idx)[i];
7253 	flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7254 				    &attr, items, actions, false, &error);
7255 	if (!flow_idx)
7256 		return -rte_errno;
7257 	return 0;
7258 }
7259 
7260 /**
7261  * Enable a flow control configured from the control plane.
7262  *
7263  * @param dev
7264  *   Pointer to Ethernet device.
7265  * @param eth_spec
7266  *   An Ethernet flow spec to apply.
7267  * @param eth_mask
7268  *   An Ethernet flow mask to apply.
7269  *
7270  * @return
7271  *   0 on success, a negative errno value otherwise and rte_errno is set.
7272  */
7273 int
7274 mlx5_ctrl_flow(struct rte_eth_dev *dev,
7275 	       struct rte_flow_item_eth *eth_spec,
7276 	       struct rte_flow_item_eth *eth_mask)
7277 {
7278 	return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
7279 }
7280 
7281 /**
7282  * Create default miss flow rule matching lacp traffic
7283  *
7284  * @param dev
7285  *   Pointer to Ethernet device.
7286  * @param eth_spec
7287  *   An Ethernet flow spec to apply.
7288  *
7289  * @return
7290  *   0 on success, a negative errno value otherwise and rte_errno is set.
7291  */
7292 int
7293 mlx5_flow_lacp_miss(struct rte_eth_dev *dev)
7294 {
7295 	/*
7296 	 * The LACP matching is done by only using ether type since using
7297 	 * a multicast dst mac causes kernel to give low priority to this flow.
7298 	 */
7299 	static const struct rte_flow_item_eth lacp_spec = {
7300 		.type = RTE_BE16(0x8809),
7301 	};
7302 	static const struct rte_flow_item_eth lacp_mask = {
7303 		.type = 0xffff,
7304 	};
7305 	const struct rte_flow_attr attr = {
7306 		.ingress = 1,
7307 	};
7308 	struct rte_flow_item items[] = {
7309 		{
7310 			.type = RTE_FLOW_ITEM_TYPE_ETH,
7311 			.spec = &lacp_spec,
7312 			.mask = &lacp_mask,
7313 		},
7314 		{
7315 			.type = RTE_FLOW_ITEM_TYPE_END,
7316 		},
7317 	};
7318 	struct rte_flow_action actions[] = {
7319 		{
7320 			.type = (enum rte_flow_action_type)
7321 				MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS,
7322 		},
7323 		{
7324 			.type = RTE_FLOW_ACTION_TYPE_END,
7325 		},
7326 	};
7327 	struct rte_flow_error error;
7328 	uint32_t flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_CTL,
7329 					&attr, items, actions,
7330 					false, &error);
7331 
7332 	if (!flow_idx)
7333 		return -rte_errno;
7334 	return 0;
7335 }
7336 
7337 /**
7338  * Destroy a flow.
7339  *
7340  * @see rte_flow_destroy()
7341  * @see rte_flow_ops
7342  */
7343 int
7344 mlx5_flow_destroy(struct rte_eth_dev *dev,
7345 		  struct rte_flow *flow,
7346 		  struct rte_flow_error *error __rte_unused)
7347 {
7348 	struct mlx5_priv *priv = dev->data->dev_private;
7349 
7350 	if (priv->sh->config.dv_flow_en == 2)
7351 		return rte_flow_error_set(error, ENOTSUP,
7352 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7353 			  NULL,
7354 			  "Flow non-Q destruction not supported");
7355 	flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN,
7356 				(uintptr_t)(void *)flow);
7357 	return 0;
7358 }
7359 
7360 /**
7361  * Destroy all flows.
7362  *
7363  * @see rte_flow_flush()
7364  * @see rte_flow_ops
7365  */
7366 int
7367 mlx5_flow_flush(struct rte_eth_dev *dev,
7368 		struct rte_flow_error *error __rte_unused)
7369 {
7370 	mlx5_flow_list_flush(dev, MLX5_FLOW_TYPE_GEN, false);
7371 	return 0;
7372 }
7373 
7374 /**
7375  * Isolated mode.
7376  *
7377  * @see rte_flow_isolate()
7378  * @see rte_flow_ops
7379  */
7380 int
7381 mlx5_flow_isolate(struct rte_eth_dev *dev,
7382 		  int enable,
7383 		  struct rte_flow_error *error)
7384 {
7385 	struct mlx5_priv *priv = dev->data->dev_private;
7386 
7387 	if (dev->data->dev_started) {
7388 		rte_flow_error_set(error, EBUSY,
7389 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7390 				   NULL,
7391 				   "port must be stopped first");
7392 		return -rte_errno;
7393 	}
7394 	priv->isolated = !!enable;
7395 	if (enable)
7396 		dev->dev_ops = &mlx5_dev_ops_isolate;
7397 	else
7398 		dev->dev_ops = &mlx5_dev_ops;
7399 
7400 	dev->rx_descriptor_status = mlx5_rx_descriptor_status;
7401 	dev->tx_descriptor_status = mlx5_tx_descriptor_status;
7402 
7403 	return 0;
7404 }
7405 
7406 /**
7407  * Query a flow.
7408  *
7409  * @see rte_flow_query()
7410  * @see rte_flow_ops
7411  */
7412 static int
7413 flow_drv_query(struct rte_eth_dev *dev,
7414 	       uint32_t flow_idx,
7415 	       const struct rte_flow_action *actions,
7416 	       void *data,
7417 	       struct rte_flow_error *error)
7418 {
7419 	struct mlx5_priv *priv = dev->data->dev_private;
7420 	const struct mlx5_flow_driver_ops *fops;
7421 	struct rte_flow *flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
7422 					       flow_idx);
7423 	enum mlx5_flow_drv_type ftype;
7424 
7425 	if (!flow) {
7426 		return rte_flow_error_set(error, ENOENT,
7427 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7428 			  NULL,
7429 			  "invalid flow handle");
7430 	}
7431 	ftype = flow->drv_type;
7432 	MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
7433 	fops = flow_get_drv_ops(ftype);
7434 
7435 	return fops->query(dev, flow, actions, data, error);
7436 }
7437 
7438 /**
7439  * Query a flow.
7440  *
7441  * @see rte_flow_query()
7442  * @see rte_flow_ops
7443  */
7444 int
7445 mlx5_flow_query(struct rte_eth_dev *dev,
7446 		struct rte_flow *flow,
7447 		const struct rte_flow_action *actions,
7448 		void *data,
7449 		struct rte_flow_error *error)
7450 {
7451 	int ret;
7452 	struct mlx5_priv *priv = dev->data->dev_private;
7453 
7454 	if (priv->sh->config.dv_flow_en == 2)
7455 		return rte_flow_error_set(error, ENOTSUP,
7456 			  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7457 			  NULL,
7458 			  "Flow non-Q query not supported");
7459 	ret = flow_drv_query(dev, (uintptr_t)(void *)flow, actions, data,
7460 			     error);
7461 	if (ret < 0)
7462 		return ret;
7463 	return 0;
7464 }
7465 
7466 /**
7467  * Get rte_flow callbacks.
7468  *
7469  * @param dev
7470  *   Pointer to Ethernet device structure.
7471  * @param ops
7472  *   Pointer to operation-specific structure.
7473  *
7474  * @return 0
7475  */
7476 int
7477 mlx5_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
7478 		  const struct rte_flow_ops **ops)
7479 {
7480 	*ops = &mlx5_flow_ops;
7481 	return 0;
7482 }
7483 
7484 /**
7485  * Validate meter policy actions.
7486  * Dispatcher for action type specific validation.
7487  *
7488  * @param[in] dev
7489  *   Pointer to the Ethernet device structure.
7490  * @param[in] action
7491  *   The meter policy action object to validate.
7492  * @param[in] attr
7493  *   Attributes of flow to determine steering domain.
7494  * @param[out] is_rss
7495  *   Is RSS or not.
7496  * @param[out] domain_bitmap
7497  *   Domain bitmap.
7498  * @param[out] is_def_policy
7499  *   Is default policy or not.
7500  * @param[out] error
7501  *   Perform verbose error reporting if not NULL. Initialized in case of
7502  *   error only.
7503  *
7504  * @return
7505  *   0 on success, otherwise negative errno value.
7506  */
7507 int
7508 mlx5_flow_validate_mtr_acts(struct rte_eth_dev *dev,
7509 			const struct rte_flow_action *actions[RTE_COLORS],
7510 			struct rte_flow_attr *attr,
7511 			bool *is_rss,
7512 			uint8_t *domain_bitmap,
7513 			uint8_t *policy_mode,
7514 			struct rte_mtr_error *error)
7515 {
7516 	const struct mlx5_flow_driver_ops *fops;
7517 
7518 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7519 	return fops->validate_mtr_acts(dev, actions, attr, is_rss,
7520 				       domain_bitmap, policy_mode, error);
7521 }
7522 
7523 /**
7524  * Destroy the meter table set.
7525  *
7526  * @param[in] dev
7527  *   Pointer to Ethernet device.
7528  * @param[in] mtr_policy
7529  *   Meter policy struct.
7530  */
7531 void
7532 mlx5_flow_destroy_mtr_acts(struct rte_eth_dev *dev,
7533 		      struct mlx5_flow_meter_policy *mtr_policy)
7534 {
7535 	const struct mlx5_flow_driver_ops *fops;
7536 
7537 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7538 	fops->destroy_mtr_acts(dev, mtr_policy);
7539 }
7540 
7541 /**
7542  * Create policy action, lock free,
7543  * (mutex should be acquired by caller).
7544  * Dispatcher for action type specific call.
7545  *
7546  * @param[in] dev
7547  *   Pointer to the Ethernet device structure.
7548  * @param[in] mtr_policy
7549  *   Meter policy struct.
7550  * @param[in] action
7551  *   Action specification used to create meter actions.
7552  * @param[out] error
7553  *   Perform verbose error reporting if not NULL. Initialized in case of
7554  *   error only.
7555  *
7556  * @return
7557  *   0 on success, otherwise negative errno value.
7558  */
7559 int
7560 mlx5_flow_create_mtr_acts(struct rte_eth_dev *dev,
7561 		      struct mlx5_flow_meter_policy *mtr_policy,
7562 		      const struct rte_flow_action *actions[RTE_COLORS],
7563 		      struct rte_mtr_error *error)
7564 {
7565 	const struct mlx5_flow_driver_ops *fops;
7566 
7567 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7568 	return fops->create_mtr_acts(dev, mtr_policy, actions, error);
7569 }
7570 
7571 /**
7572  * Create policy rules, lock free,
7573  * (mutex should be acquired by caller).
7574  * Dispatcher for action type specific call.
7575  *
7576  * @param[in] dev
7577  *   Pointer to the Ethernet device structure.
7578  * @param[in] mtr_policy
7579  *   Meter policy struct.
7580  *
7581  * @return
7582  *   0 on success, -1 otherwise.
7583  */
7584 int
7585 mlx5_flow_create_policy_rules(struct rte_eth_dev *dev,
7586 			     struct mlx5_flow_meter_policy *mtr_policy)
7587 {
7588 	const struct mlx5_flow_driver_ops *fops;
7589 
7590 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7591 	return fops->create_policy_rules(dev, mtr_policy);
7592 }
7593 
7594 /**
7595  * Destroy policy rules, lock free,
7596  * (mutex should be acquired by caller).
7597  * Dispatcher for action type specific call.
7598  *
7599  * @param[in] dev
7600  *   Pointer to the Ethernet device structure.
7601  * @param[in] mtr_policy
7602  *   Meter policy struct.
7603  */
7604 void
7605 mlx5_flow_destroy_policy_rules(struct rte_eth_dev *dev,
7606 			     struct mlx5_flow_meter_policy *mtr_policy)
7607 {
7608 	const struct mlx5_flow_driver_ops *fops;
7609 
7610 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7611 	fops->destroy_policy_rules(dev, mtr_policy);
7612 }
7613 
7614 /**
7615  * Destroy the default policy table set.
7616  *
7617  * @param[in] dev
7618  *   Pointer to Ethernet device.
7619  */
7620 void
7621 mlx5_flow_destroy_def_policy(struct rte_eth_dev *dev)
7622 {
7623 	const struct mlx5_flow_driver_ops *fops;
7624 
7625 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7626 	fops->destroy_def_policy(dev);
7627 }
7628 
7629 /**
7630  * Destroy the default policy table set.
7631  *
7632  * @param[in] dev
7633  *   Pointer to Ethernet device.
7634  *
7635  * @return
7636  *   0 on success, -1 otherwise.
7637  */
7638 int
7639 mlx5_flow_create_def_policy(struct rte_eth_dev *dev)
7640 {
7641 	const struct mlx5_flow_driver_ops *fops;
7642 
7643 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7644 	return fops->create_def_policy(dev);
7645 }
7646 
7647 /**
7648  * Create the needed meter and suffix tables.
7649  *
7650  * @param[in] dev
7651  *   Pointer to Ethernet device.
7652  *
7653  * @return
7654  *   0 on success, -1 otherwise.
7655  */
7656 int
7657 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
7658 			struct mlx5_flow_meter_info *fm,
7659 			uint32_t mtr_idx,
7660 			uint8_t domain_bitmap)
7661 {
7662 	const struct mlx5_flow_driver_ops *fops;
7663 
7664 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7665 	return fops->create_mtr_tbls(dev, fm, mtr_idx, domain_bitmap);
7666 }
7667 
7668 /**
7669  * Destroy the meter table set.
7670  *
7671  * @param[in] dev
7672  *   Pointer to Ethernet device.
7673  * @param[in] tbl
7674  *   Pointer to the meter table set.
7675  */
7676 void
7677 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
7678 			   struct mlx5_flow_meter_info *fm)
7679 {
7680 	const struct mlx5_flow_driver_ops *fops;
7681 
7682 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7683 	fops->destroy_mtr_tbls(dev, fm);
7684 }
7685 
7686 /**
7687  * Destroy the global meter drop table.
7688  *
7689  * @param[in] dev
7690  *   Pointer to Ethernet device.
7691  */
7692 void
7693 mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev)
7694 {
7695 	const struct mlx5_flow_driver_ops *fops;
7696 
7697 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7698 	fops->destroy_mtr_drop_tbls(dev);
7699 }
7700 
7701 /**
7702  * Destroy the sub policy table with RX queue.
7703  *
7704  * @param[in] dev
7705  *   Pointer to Ethernet device.
7706  * @param[in] mtr_policy
7707  *   Pointer to meter policy table.
7708  */
7709 void
7710 mlx5_flow_destroy_sub_policy_with_rxq(struct rte_eth_dev *dev,
7711 		struct mlx5_flow_meter_policy *mtr_policy)
7712 {
7713 	const struct mlx5_flow_driver_ops *fops;
7714 
7715 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7716 	fops->destroy_sub_policy_with_rxq(dev, mtr_policy);
7717 }
7718 
7719 /**
7720  * Allocate the needed aso flow meter id.
7721  *
7722  * @param[in] dev
7723  *   Pointer to Ethernet device.
7724  *
7725  * @return
7726  *   Index to aso flow meter on success, NULL otherwise.
7727  */
7728 uint32_t
7729 mlx5_flow_mtr_alloc(struct rte_eth_dev *dev)
7730 {
7731 	const struct mlx5_flow_driver_ops *fops;
7732 
7733 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7734 	return fops->create_meter(dev);
7735 }
7736 
7737 /**
7738  * Free the aso flow meter id.
7739  *
7740  * @param[in] dev
7741  *   Pointer to Ethernet device.
7742  * @param[in] mtr_idx
7743  *  Index to aso flow meter to be free.
7744  *
7745  * @return
7746  *   0 on success.
7747  */
7748 void
7749 mlx5_flow_mtr_free(struct rte_eth_dev *dev, uint32_t mtr_idx)
7750 {
7751 	const struct mlx5_flow_driver_ops *fops;
7752 
7753 	fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7754 	fops->free_meter(dev, mtr_idx);
7755 }
7756 
7757 /**
7758  * Allocate a counter.
7759  *
7760  * @param[in] dev
7761  *   Pointer to Ethernet device structure.
7762  *
7763  * @return
7764  *   Index to allocated counter  on success, 0 otherwise.
7765  */
7766 uint32_t
7767 mlx5_counter_alloc(struct rte_eth_dev *dev)
7768 {
7769 	const struct mlx5_flow_driver_ops *fops;
7770 	struct rte_flow_attr attr = { .transfer = 0 };
7771 
7772 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7773 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7774 		return fops->counter_alloc(dev);
7775 	}
7776 	DRV_LOG(ERR,
7777 		"port %u counter allocate is not supported.",
7778 		 dev->data->port_id);
7779 	return 0;
7780 }
7781 
7782 /**
7783  * Free a counter.
7784  *
7785  * @param[in] dev
7786  *   Pointer to Ethernet device structure.
7787  * @param[in] cnt
7788  *   Index to counter to be free.
7789  */
7790 void
7791 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
7792 {
7793 	const struct mlx5_flow_driver_ops *fops;
7794 	struct rte_flow_attr attr = { .transfer = 0 };
7795 
7796 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7797 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7798 		fops->counter_free(dev, cnt);
7799 		return;
7800 	}
7801 	DRV_LOG(ERR,
7802 		"port %u counter free is not supported.",
7803 		 dev->data->port_id);
7804 }
7805 
7806 /**
7807  * Query counter statistics.
7808  *
7809  * @param[in] dev
7810  *   Pointer to Ethernet device structure.
7811  * @param[in] cnt
7812  *   Index to counter to query.
7813  * @param[in] clear
7814  *   Set to clear counter statistics.
7815  * @param[out] pkts
7816  *   The counter hits packets number to save.
7817  * @param[out] bytes
7818  *   The counter hits bytes number to save.
7819  *
7820  * @return
7821  *   0 on success, a negative errno value otherwise.
7822  */
7823 int
7824 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
7825 		   bool clear, uint64_t *pkts, uint64_t *bytes, void **action)
7826 {
7827 	const struct mlx5_flow_driver_ops *fops;
7828 	struct rte_flow_attr attr = { .transfer = 0 };
7829 
7830 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
7831 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
7832 		return fops->counter_query(dev, cnt, clear, pkts,
7833 					bytes, action);
7834 	}
7835 	DRV_LOG(ERR,
7836 		"port %u counter query is not supported.",
7837 		 dev->data->port_id);
7838 	return -ENOTSUP;
7839 }
7840 
7841 /**
7842  * Allocate a new memory for the counter values wrapped by all the needed
7843  * management.
7844  *
7845  * @param[in] sh
7846  *   Pointer to mlx5_dev_ctx_shared object.
7847  *
7848  * @return
7849  *   0 on success, a negative errno value otherwise.
7850  */
7851 static int
7852 mlx5_flow_create_counter_stat_mem_mng(struct mlx5_dev_ctx_shared *sh)
7853 {
7854 	struct mlx5_counter_stats_mem_mng *mem_mng;
7855 	volatile struct flow_counter_stats *raw_data;
7856 	int raws_n = MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES;
7857 	int size = (sizeof(struct flow_counter_stats) *
7858 			MLX5_COUNTERS_PER_POOL +
7859 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
7860 			sizeof(struct mlx5_counter_stats_mem_mng);
7861 	size_t pgsize = rte_mem_page_size();
7862 	uint8_t *mem;
7863 	int ret;
7864 	int i;
7865 
7866 	if (pgsize == (size_t)-1) {
7867 		DRV_LOG(ERR, "Failed to get mem page size");
7868 		rte_errno = ENOMEM;
7869 		return -ENOMEM;
7870 	}
7871 	mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize, SOCKET_ID_ANY);
7872 	if (!mem) {
7873 		rte_errno = ENOMEM;
7874 		return -ENOMEM;
7875 	}
7876 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
7877 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
7878 	ret = mlx5_os_wrapped_mkey_create(sh->cdev->ctx, sh->cdev->pd,
7879 					  sh->cdev->pdn, mem, size,
7880 					  &mem_mng->wm);
7881 	if (ret) {
7882 		rte_errno = errno;
7883 		mlx5_free(mem);
7884 		return -rte_errno;
7885 	}
7886 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
7887 	raw_data = (volatile struct flow_counter_stats *)mem;
7888 	for (i = 0; i < raws_n; ++i) {
7889 		mem_mng->raws[i].mem_mng = mem_mng;
7890 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
7891 	}
7892 	for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
7893 		LIST_INSERT_HEAD(&sh->cmng.free_stat_raws,
7894 				 mem_mng->raws + MLX5_CNT_CONTAINER_RESIZE + i,
7895 				 next);
7896 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
7897 	sh->cmng.mem_mng = mem_mng;
7898 	return 0;
7899 }
7900 
7901 /**
7902  * Set the statistic memory to the new counter pool.
7903  *
7904  * @param[in] sh
7905  *   Pointer to mlx5_dev_ctx_shared object.
7906  * @param[in] pool
7907  *   Pointer to the pool to set the statistic memory.
7908  *
7909  * @return
7910  *   0 on success, a negative errno value otherwise.
7911  */
7912 static int
7913 mlx5_flow_set_counter_stat_mem(struct mlx5_dev_ctx_shared *sh,
7914 			       struct mlx5_flow_counter_pool *pool)
7915 {
7916 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7917 	/* Resize statistic memory once used out. */
7918 	if (!(pool->index % MLX5_CNT_CONTAINER_RESIZE) &&
7919 	    mlx5_flow_create_counter_stat_mem_mng(sh)) {
7920 		DRV_LOG(ERR, "Cannot resize counter stat mem.");
7921 		return -1;
7922 	}
7923 	rte_spinlock_lock(&pool->sl);
7924 	pool->raw = cmng->mem_mng->raws + pool->index %
7925 		    MLX5_CNT_CONTAINER_RESIZE;
7926 	rte_spinlock_unlock(&pool->sl);
7927 	pool->raw_hw = NULL;
7928 	return 0;
7929 }
7930 
7931 #define MLX5_POOL_QUERY_FREQ_US 1000000
7932 
7933 /**
7934  * Set the periodic procedure for triggering asynchronous batch queries for all
7935  * the counter pools.
7936  *
7937  * @param[in] sh
7938  *   Pointer to mlx5_dev_ctx_shared object.
7939  */
7940 void
7941 mlx5_set_query_alarm(struct mlx5_dev_ctx_shared *sh)
7942 {
7943 	uint32_t pools_n, us;
7944 
7945 	pools_n = __atomic_load_n(&sh->cmng.n_valid, __ATOMIC_RELAXED);
7946 	us = MLX5_POOL_QUERY_FREQ_US / pools_n;
7947 	DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
7948 	if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
7949 		sh->cmng.query_thread_on = 0;
7950 		DRV_LOG(ERR, "Cannot reinitialize query alarm");
7951 	} else {
7952 		sh->cmng.query_thread_on = 1;
7953 	}
7954 }
7955 
7956 /**
7957  * The periodic procedure for triggering asynchronous batch queries for all the
7958  * counter pools. This function is probably called by the host thread.
7959  *
7960  * @param[in] arg
7961  *   The parameter for the alarm process.
7962  */
7963 void
7964 mlx5_flow_query_alarm(void *arg)
7965 {
7966 	struct mlx5_dev_ctx_shared *sh = arg;
7967 	int ret;
7968 	uint16_t pool_index = sh->cmng.pool_index;
7969 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
7970 	struct mlx5_flow_counter_pool *pool;
7971 	uint16_t n_valid;
7972 
7973 	if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
7974 		goto set_alarm;
7975 	rte_spinlock_lock(&cmng->pool_update_sl);
7976 	pool = cmng->pools[pool_index];
7977 	n_valid = cmng->n_valid;
7978 	rte_spinlock_unlock(&cmng->pool_update_sl);
7979 	/* Set the statistic memory to the new created pool. */
7980 	if ((!pool->raw && mlx5_flow_set_counter_stat_mem(sh, pool)))
7981 		goto set_alarm;
7982 	if (pool->raw_hw)
7983 		/* There is a pool query in progress. */
7984 		goto set_alarm;
7985 	pool->raw_hw =
7986 		LIST_FIRST(&sh->cmng.free_stat_raws);
7987 	if (!pool->raw_hw)
7988 		/* No free counter statistics raw memory. */
7989 		goto set_alarm;
7990 	/*
7991 	 * Identify the counters released between query trigger and query
7992 	 * handle more efficiently. The counter released in this gap period
7993 	 * should wait for a new round of query as the new arrived packets
7994 	 * will not be taken into account.
7995 	 */
7996 	pool->query_gen++;
7997 	ret = mlx5_devx_cmd_flow_counter_query(pool->min_dcs, 0,
7998 					       MLX5_COUNTERS_PER_POOL,
7999 					       NULL, NULL,
8000 					       pool->raw_hw->mem_mng->wm.lkey,
8001 					       (void *)(uintptr_t)
8002 					       pool->raw_hw->data,
8003 					       sh->devx_comp,
8004 					       (uint64_t)(uintptr_t)pool);
8005 	if (ret) {
8006 		DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
8007 			" %d", pool->min_dcs->id);
8008 		pool->raw_hw = NULL;
8009 		goto set_alarm;
8010 	}
8011 	LIST_REMOVE(pool->raw_hw, next);
8012 	sh->cmng.pending_queries++;
8013 	pool_index++;
8014 	if (pool_index >= n_valid)
8015 		pool_index = 0;
8016 set_alarm:
8017 	sh->cmng.pool_index = pool_index;
8018 	mlx5_set_query_alarm(sh);
8019 }
8020 
8021 /**
8022  * Check and callback event for new aged flow in the counter pool
8023  *
8024  * @param[in] sh
8025  *   Pointer to mlx5_dev_ctx_shared object.
8026  * @param[in] pool
8027  *   Pointer to Current counter pool.
8028  */
8029 static void
8030 mlx5_flow_aging_check(struct mlx5_dev_ctx_shared *sh,
8031 		   struct mlx5_flow_counter_pool *pool)
8032 {
8033 	struct mlx5_priv *priv;
8034 	struct mlx5_flow_counter *cnt;
8035 	struct mlx5_age_info *age_info;
8036 	struct mlx5_age_param *age_param;
8037 	struct mlx5_counter_stats_raw *cur = pool->raw_hw;
8038 	struct mlx5_counter_stats_raw *prev = pool->raw;
8039 	const uint64_t curr_time = MLX5_CURR_TIME_SEC;
8040 	const uint32_t time_delta = curr_time - pool->time_of_last_age_check;
8041 	uint16_t expected = AGE_CANDIDATE;
8042 	uint32_t i;
8043 
8044 	pool->time_of_last_age_check = curr_time;
8045 	for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
8046 		cnt = MLX5_POOL_GET_CNT(pool, i);
8047 		age_param = MLX5_CNT_TO_AGE(cnt);
8048 		if (__atomic_load_n(&age_param->state,
8049 				    __ATOMIC_RELAXED) != AGE_CANDIDATE)
8050 			continue;
8051 		if (cur->data[i].hits != prev->data[i].hits) {
8052 			__atomic_store_n(&age_param->sec_since_last_hit, 0,
8053 					 __ATOMIC_RELAXED);
8054 			continue;
8055 		}
8056 		if (__atomic_add_fetch(&age_param->sec_since_last_hit,
8057 				       time_delta,
8058 				       __ATOMIC_RELAXED) <= age_param->timeout)
8059 			continue;
8060 		/**
8061 		 * Hold the lock first, or if between the
8062 		 * state AGE_TMOUT and tailq operation the
8063 		 * release happened, the release procedure
8064 		 * may delete a non-existent tailq node.
8065 		 */
8066 		priv = rte_eth_devices[age_param->port_id].data->dev_private;
8067 		age_info = GET_PORT_AGE_INFO(priv);
8068 		rte_spinlock_lock(&age_info->aged_sl);
8069 		if (__atomic_compare_exchange_n(&age_param->state, &expected,
8070 						AGE_TMOUT, false,
8071 						__ATOMIC_RELAXED,
8072 						__ATOMIC_RELAXED)) {
8073 			TAILQ_INSERT_TAIL(&age_info->aged_counters, cnt, next);
8074 			MLX5_AGE_SET(age_info, MLX5_AGE_EVENT_NEW);
8075 		}
8076 		rte_spinlock_unlock(&age_info->aged_sl);
8077 	}
8078 	mlx5_age_event_prepare(sh);
8079 }
8080 
8081 /**
8082  * Handler for the HW respond about ready values from an asynchronous batch
8083  * query. This function is probably called by the host thread.
8084  *
8085  * @param[in] sh
8086  *   The pointer to the shared device context.
8087  * @param[in] async_id
8088  *   The Devx async ID.
8089  * @param[in] status
8090  *   The status of the completion.
8091  */
8092 void
8093 mlx5_flow_async_pool_query_handle(struct mlx5_dev_ctx_shared *sh,
8094 				  uint64_t async_id, int status)
8095 {
8096 	struct mlx5_flow_counter_pool *pool =
8097 		(struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
8098 	struct mlx5_counter_stats_raw *raw_to_free;
8099 	uint8_t query_gen = pool->query_gen ^ 1;
8100 	struct mlx5_flow_counter_mng *cmng = &sh->cmng;
8101 	enum mlx5_counter_type cnt_type =
8102 		pool->is_aged ? MLX5_COUNTER_TYPE_AGE :
8103 				MLX5_COUNTER_TYPE_ORIGIN;
8104 
8105 	if (unlikely(status)) {
8106 		raw_to_free = pool->raw_hw;
8107 	} else {
8108 		raw_to_free = pool->raw;
8109 		if (pool->is_aged)
8110 			mlx5_flow_aging_check(sh, pool);
8111 		rte_spinlock_lock(&pool->sl);
8112 		pool->raw = pool->raw_hw;
8113 		rte_spinlock_unlock(&pool->sl);
8114 		/* Be sure the new raw counters data is updated in memory. */
8115 		rte_io_wmb();
8116 		if (!TAILQ_EMPTY(&pool->counters[query_gen])) {
8117 			rte_spinlock_lock(&cmng->csl[cnt_type]);
8118 			TAILQ_CONCAT(&cmng->counters[cnt_type],
8119 				     &pool->counters[query_gen], next);
8120 			rte_spinlock_unlock(&cmng->csl[cnt_type]);
8121 		}
8122 	}
8123 	LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
8124 	pool->raw_hw = NULL;
8125 	sh->cmng.pending_queries--;
8126 }
8127 
8128 static int
8129 flow_group_to_table(uint32_t port_id, uint32_t group, uint32_t *table,
8130 		    const struct flow_grp_info *grp_info,
8131 		    struct rte_flow_error *error)
8132 {
8133 	if (grp_info->transfer && grp_info->external &&
8134 	    grp_info->fdb_def_rule) {
8135 		if (group == UINT32_MAX)
8136 			return rte_flow_error_set
8137 						(error, EINVAL,
8138 						 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
8139 						 NULL,
8140 						 "group index not supported");
8141 		*table = group + 1;
8142 	} else {
8143 		*table = group;
8144 	}
8145 	DRV_LOG(DEBUG, "port %u group=%#x table=%#x", port_id, group, *table);
8146 	return 0;
8147 }
8148 
8149 /**
8150  * Translate the rte_flow group index to HW table value.
8151  *
8152  * If tunnel offload is disabled, all group ids converted to flow table
8153  * id using the standard method.
8154  * If tunnel offload is enabled, group id can be converted using the
8155  * standard or tunnel conversion method. Group conversion method
8156  * selection depends on flags in `grp_info` parameter:
8157  * - Internal (grp_info.external == 0) groups conversion uses the
8158  *   standard method.
8159  * - Group ids in JUMP action converted with the tunnel conversion.
8160  * - Group id in rule attribute conversion depends on a rule type and
8161  *   group id value:
8162  *   ** non zero group attributes converted with the tunnel method
8163  *   ** zero group attribute in non-tunnel rule is converted using the
8164  *      standard method - there's only one root table
8165  *   ** zero group attribute in steer tunnel rule is converted with the
8166  *      standard method - single root table
8167  *   ** zero group attribute in match tunnel rule is a special OvS
8168  *      case: that value is used for portability reasons. That group
8169  *      id is converted with the tunnel conversion method.
8170  *
8171  * @param[in] dev
8172  *   Port device
8173  * @param[in] tunnel
8174  *   PMD tunnel offload object
8175  * @param[in] group
8176  *   rte_flow group index value.
8177  * @param[out] table
8178  *   HW table value.
8179  * @param[in] grp_info
8180  *   flags used for conversion
8181  * @param[out] error
8182  *   Pointer to error structure.
8183  *
8184  * @return
8185  *   0 on success, a negative errno value otherwise and rte_errno is set.
8186  */
8187 int
8188 mlx5_flow_group_to_table(struct rte_eth_dev *dev,
8189 			 const struct mlx5_flow_tunnel *tunnel,
8190 			 uint32_t group, uint32_t *table,
8191 			 const struct flow_grp_info *grp_info,
8192 			 struct rte_flow_error *error)
8193 {
8194 	int ret;
8195 	bool standard_translation;
8196 
8197 	if (!grp_info->skip_scale && grp_info->external &&
8198 	    group < MLX5_MAX_TABLES_EXTERNAL)
8199 		group *= MLX5_FLOW_TABLE_FACTOR;
8200 	if (is_tunnel_offload_active(dev)) {
8201 		standard_translation = !grp_info->external ||
8202 					grp_info->std_tbl_fix;
8203 	} else {
8204 		standard_translation = true;
8205 	}
8206 	DRV_LOG(DEBUG,
8207 		"port %u group=%u transfer=%d external=%d fdb_def_rule=%d translate=%s",
8208 		dev->data->port_id, group, grp_info->transfer,
8209 		grp_info->external, grp_info->fdb_def_rule,
8210 		standard_translation ? "STANDARD" : "TUNNEL");
8211 	if (standard_translation)
8212 		ret = flow_group_to_table(dev->data->port_id, group, table,
8213 					  grp_info, error);
8214 	else
8215 		ret = tunnel_flow_group_to_flow_table(dev, tunnel, group,
8216 						      table, error);
8217 
8218 	return ret;
8219 }
8220 
8221 /**
8222  * Discover availability of metadata reg_c's.
8223  *
8224  * Iteratively use test flows to check availability.
8225  *
8226  * @param[in] dev
8227  *   Pointer to the Ethernet device structure.
8228  *
8229  * @return
8230  *   0 on success, a negative errno value otherwise and rte_errno is set.
8231  */
8232 int
8233 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
8234 {
8235 	struct mlx5_priv *priv = dev->data->dev_private;
8236 	enum modify_reg idx;
8237 	int n = 0;
8238 
8239 	/* reg_c[0] and reg_c[1] are reserved. */
8240 	priv->sh->flow_mreg_c[n++] = REG_C_0;
8241 	priv->sh->flow_mreg_c[n++] = REG_C_1;
8242 	/* Discover availability of other reg_c's. */
8243 	for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
8244 		struct rte_flow_attr attr = {
8245 			.group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
8246 			.priority = MLX5_FLOW_LOWEST_PRIO_INDICATOR,
8247 			.ingress = 1,
8248 		};
8249 		struct rte_flow_item items[] = {
8250 			[0] = {
8251 				.type = RTE_FLOW_ITEM_TYPE_END,
8252 			},
8253 		};
8254 		struct rte_flow_action actions[] = {
8255 			[0] = {
8256 				.type = (enum rte_flow_action_type)
8257 					MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
8258 				.conf = &(struct mlx5_flow_action_copy_mreg){
8259 					.src = REG_C_1,
8260 					.dst = idx,
8261 				},
8262 			},
8263 			[1] = {
8264 				.type = RTE_FLOW_ACTION_TYPE_JUMP,
8265 				.conf = &(struct rte_flow_action_jump){
8266 					.group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
8267 				},
8268 			},
8269 			[2] = {
8270 				.type = RTE_FLOW_ACTION_TYPE_END,
8271 			},
8272 		};
8273 		uint32_t flow_idx;
8274 		struct rte_flow *flow;
8275 		struct rte_flow_error error;
8276 
8277 		if (!priv->sh->config.dv_flow_en)
8278 			break;
8279 		/* Create internal flow, validation skips copy action. */
8280 		flow_idx = flow_list_create(dev, MLX5_FLOW_TYPE_GEN, &attr,
8281 					items, actions, false, &error);
8282 		flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8283 				      flow_idx);
8284 		if (!flow)
8285 			continue;
8286 		priv->sh->flow_mreg_c[n++] = idx;
8287 		flow_list_destroy(dev, MLX5_FLOW_TYPE_GEN, flow_idx);
8288 	}
8289 	for (; n < MLX5_MREG_C_NUM; ++n)
8290 		priv->sh->flow_mreg_c[n] = REG_NON;
8291 	priv->sh->metadata_regc_check_flag = 1;
8292 	return 0;
8293 }
8294 
8295 int
8296 save_dump_file(const uint8_t *data, uint32_t size,
8297 	uint32_t type, uint64_t id, void *arg, FILE *file)
8298 {
8299 	char line[BUF_SIZE];
8300 	uint32_t out = 0;
8301 	uint32_t k;
8302 	uint32_t actions_num;
8303 	struct rte_flow_query_count *count;
8304 
8305 	memset(line, 0, BUF_SIZE);
8306 	switch (type) {
8307 	case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
8308 		actions_num = *(uint32_t *)(arg);
8309 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",%d,",
8310 				type, id, actions_num);
8311 		break;
8312 	case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
8313 		out += snprintf(line + out, BUF_SIZE - out, "%d,0x%" PRIx64 ",",
8314 				type, id);
8315 		break;
8316 	case DR_DUMP_REC_TYPE_PMD_COUNTER:
8317 		count = (struct rte_flow_query_count *)arg;
8318 		fprintf(file,
8319 			"%d,0x%" PRIx64 ",%" PRIu64 ",%" PRIu64 "\n",
8320 			type, id, count->hits, count->bytes);
8321 		return 0;
8322 	default:
8323 		return -1;
8324 	}
8325 
8326 	for (k = 0; k < size; k++) {
8327 		/* Make sure we do not overrun the line buffer length. */
8328 		if (out >= BUF_SIZE - 4) {
8329 			line[out] = '\0';
8330 			break;
8331 		}
8332 		out += snprintf(line + out, BUF_SIZE - out, "%02x",
8333 				(data[k]) & 0xff);
8334 	}
8335 	fprintf(file, "%s\n", line);
8336 	return 0;
8337 }
8338 
8339 int
8340 mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
8341 	struct rte_flow_query_count *count, struct rte_flow_error *error)
8342 {
8343 	struct rte_flow_action action[2];
8344 	enum mlx5_flow_drv_type ftype;
8345 	const struct mlx5_flow_driver_ops *fops;
8346 
8347 	if (!flow) {
8348 		return rte_flow_error_set(error, ENOENT,
8349 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8350 				NULL,
8351 				"invalid flow handle");
8352 	}
8353 	action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
8354 	action[1].type = RTE_FLOW_ACTION_TYPE_END;
8355 	if (flow->counter) {
8356 		memset(count, 0, sizeof(struct rte_flow_query_count));
8357 		ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
8358 		MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
8359 						ftype < MLX5_FLOW_TYPE_MAX);
8360 		fops = flow_get_drv_ops(ftype);
8361 		return fops->query(dev, flow, action, count, error);
8362 	}
8363 	return -1;
8364 }
8365 
8366 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8367 /**
8368  * Dump flow ipool data to file
8369  *
8370  * @param[in] dev
8371  *   The pointer to Ethernet device.
8372  * @param[in] file
8373  *   A pointer to a file for output.
8374  * @param[out] error
8375  *   Perform verbose error reporting if not NULL. PMDs initialize this
8376  *   structure in case of error only.
8377  * @return
8378  *   0 on success, a negative value otherwise.
8379  */
8380 int
8381 mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
8382 	struct rte_flow *flow, FILE *file,
8383 	struct rte_flow_error *error)
8384 {
8385 	struct mlx5_priv *priv = dev->data->dev_private;
8386 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8387 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8388 	uint32_t handle_idx;
8389 	struct mlx5_flow_handle *dh;
8390 	struct rte_flow_query_count count;
8391 	uint32_t actions_num;
8392 	const uint8_t *data;
8393 	size_t size;
8394 	uint64_t id;
8395 	uint32_t type;
8396 	void *action = NULL;
8397 
8398 	if (!flow) {
8399 		return rte_flow_error_set(error, ENOENT,
8400 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8401 				NULL,
8402 				"invalid flow handle");
8403 	}
8404 	handle_idx = flow->dev_handles;
8405 	/* query counter */
8406 	if (flow->counter &&
8407 	(!mlx5_counter_query(dev, flow->counter, false,
8408 	&count.hits, &count.bytes, &action)) && action) {
8409 		id = (uint64_t)(uintptr_t)action;
8410 		type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8411 		save_dump_file(NULL, 0, type,
8412 			id, (void *)&count, file);
8413 	}
8414 
8415 	while (handle_idx) {
8416 		dh = mlx5_ipool_get(priv->sh->ipool
8417 				[MLX5_IPOOL_MLX5_FLOW], handle_idx);
8418 		if (!dh)
8419 			continue;
8420 		handle_idx = dh->next.next;
8421 
8422 		/* Get modify_hdr and encap_decap buf from ipools. */
8423 		encap_decap = NULL;
8424 		modify_hdr = dh->dvh.modify_hdr;
8425 
8426 		if (dh->dvh.rix_encap_decap) {
8427 			encap_decap = mlx5_ipool_get(priv->sh->ipool
8428 						[MLX5_IPOOL_DECAP_ENCAP],
8429 						dh->dvh.rix_encap_decap);
8430 		}
8431 		if (modify_hdr) {
8432 			data = (const uint8_t *)modify_hdr->actions;
8433 			size = (size_t)(modify_hdr->actions_num) * 8;
8434 			id = (uint64_t)(uintptr_t)modify_hdr->action;
8435 			actions_num = modify_hdr->actions_num;
8436 			type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8437 			save_dump_file(data, size, type, id,
8438 						(void *)(&actions_num), file);
8439 		}
8440 		if (encap_decap) {
8441 			data = encap_decap->buf;
8442 			size = encap_decap->size;
8443 			id = (uint64_t)(uintptr_t)encap_decap->action;
8444 			type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8445 			save_dump_file(data, size, type,
8446 						id, NULL, file);
8447 		}
8448 	}
8449 	return 0;
8450 }
8451 
8452 /**
8453  * Dump all flow's encap_decap/modify_hdr/counter data to file
8454  *
8455  * @param[in] dev
8456  *   The pointer to Ethernet device.
8457  * @param[in] file
8458  *   A pointer to a file for output.
8459  * @param[out] error
8460  *   Perform verbose error reporting if not NULL. PMDs initialize this
8461  *   structure in case of error only.
8462  * @return
8463  *   0 on success, a negative value otherwise.
8464  */
8465 static int
8466 mlx5_flow_dev_dump_sh_all(struct rte_eth_dev *dev,
8467 	FILE *file, struct rte_flow_error *error __rte_unused)
8468 {
8469 	struct mlx5_priv *priv = dev->data->dev_private;
8470 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8471 	struct mlx5_hlist *h;
8472 	struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
8473 	struct mlx5_flow_dv_encap_decap_resource *encap_decap;
8474 	struct rte_flow_query_count count;
8475 	uint32_t actions_num;
8476 	const uint8_t *data;
8477 	size_t size;
8478 	uint64_t id;
8479 	uint32_t type;
8480 	uint32_t i;
8481 	uint32_t j;
8482 	struct mlx5_list_inconst *l_inconst;
8483 	struct mlx5_list_entry *e;
8484 	int lcore_index;
8485 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
8486 	uint32_t max;
8487 	void *action;
8488 
8489 	/* encap_decap hlist is lcore_share, get global core cache. */
8490 	i = MLX5_LIST_GLOBAL;
8491 	h = sh->encaps_decaps;
8492 	if (h) {
8493 		for (j = 0; j <= h->mask; j++) {
8494 			l_inconst = &h->buckets[j].l;
8495 			if (!l_inconst || !l_inconst->cache[i])
8496 				continue;
8497 
8498 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8499 			while (e) {
8500 				encap_decap =
8501 				(struct mlx5_flow_dv_encap_decap_resource *)e;
8502 				data = encap_decap->buf;
8503 				size = encap_decap->size;
8504 				id = (uint64_t)(uintptr_t)encap_decap->action;
8505 				type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
8506 				save_dump_file(data, size, type,
8507 					id, NULL, file);
8508 				e = LIST_NEXT(e, next);
8509 			}
8510 		}
8511 	}
8512 
8513 	/* get modify_hdr */
8514 	h = sh->modify_cmds;
8515 	if (h) {
8516 		lcore_index = rte_lcore_index(rte_lcore_id());
8517 		if (unlikely(lcore_index == -1)) {
8518 			lcore_index = MLX5_LIST_NLCORE;
8519 			rte_spinlock_lock(&h->l_const.lcore_lock);
8520 		}
8521 		i = lcore_index;
8522 
8523 		for (j = 0; j <= h->mask; j++) {
8524 			l_inconst = &h->buckets[j].l;
8525 			if (!l_inconst || !l_inconst->cache[i])
8526 				continue;
8527 
8528 			e = LIST_FIRST(&l_inconst->cache[i]->h);
8529 			while (e) {
8530 				modify_hdr =
8531 				(struct mlx5_flow_dv_modify_hdr_resource *)e;
8532 				data = (const uint8_t *)modify_hdr->actions;
8533 				size = (size_t)(modify_hdr->actions_num) * 8;
8534 				actions_num = modify_hdr->actions_num;
8535 				id = (uint64_t)(uintptr_t)modify_hdr->action;
8536 				type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
8537 				save_dump_file(data, size, type, id,
8538 						(void *)(&actions_num), file);
8539 				e = LIST_NEXT(e, next);
8540 			}
8541 		}
8542 
8543 		if (unlikely(lcore_index == MLX5_LIST_NLCORE))
8544 			rte_spinlock_unlock(&h->l_const.lcore_lock);
8545 	}
8546 
8547 	/* get counter */
8548 	MLX5_ASSERT(cmng->n_valid <= cmng->n);
8549 	max = MLX5_COUNTERS_PER_POOL * cmng->n_valid;
8550 	for (j = 1; j <= max; j++) {
8551 		action = NULL;
8552 		if ((!mlx5_counter_query(dev, j, false, &count.hits,
8553 		&count.bytes, &action)) && action) {
8554 			id = (uint64_t)(uintptr_t)action;
8555 			type = DR_DUMP_REC_TYPE_PMD_COUNTER;
8556 			save_dump_file(NULL, 0, type,
8557 					id, (void *)&count, file);
8558 		}
8559 	}
8560 	return 0;
8561 }
8562 #endif
8563 
8564 /**
8565  * Dump flow raw hw data to file
8566  *
8567  * @param[in] dev
8568  *    The pointer to Ethernet device.
8569  * @param[in] file
8570  *   A pointer to a file for output.
8571  * @param[out] error
8572  *   Perform verbose error reporting if not NULL. PMDs initialize this
8573  *   structure in case of error only.
8574  * @return
8575  *   0 on success, a negative value otherwise.
8576  */
8577 int
8578 mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
8579 		   FILE *file,
8580 		   struct rte_flow_error *error __rte_unused)
8581 {
8582 	struct mlx5_priv *priv = dev->data->dev_private;
8583 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8584 	uint32_t handle_idx;
8585 	int ret;
8586 	struct mlx5_flow_handle *dh;
8587 	struct rte_flow *flow;
8588 
8589 	if (!sh->config.dv_flow_en) {
8590 		if (fputs("device dv flow disabled\n", file) <= 0)
8591 			return -errno;
8592 		return -ENOTSUP;
8593 	}
8594 
8595 	/* dump all */
8596 	if (!flow_idx) {
8597 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8598 		if (mlx5_flow_dev_dump_sh_all(dev, file, error))
8599 			return -EINVAL;
8600 #endif
8601 		return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
8602 					sh->rx_domain,
8603 					sh->tx_domain, file);
8604 	}
8605 	/* dump one */
8606 	flow = mlx5_ipool_get(priv->flows[MLX5_FLOW_TYPE_GEN],
8607 			(uintptr_t)(void *)flow_idx);
8608 	if (!flow)
8609 		return -EINVAL;
8610 
8611 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
8612 	mlx5_flow_dev_dump_ipool(dev, flow, file, error);
8613 #endif
8614 	handle_idx = flow->dev_handles;
8615 	while (handle_idx) {
8616 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
8617 				handle_idx);
8618 		if (!dh)
8619 			return -ENOENT;
8620 		if (dh->drv_flow) {
8621 			ret = mlx5_devx_cmd_flow_single_dump(dh->drv_flow,
8622 					file);
8623 			if (ret)
8624 				return -ENOENT;
8625 		}
8626 		handle_idx = dh->next.next;
8627 	}
8628 	return 0;
8629 }
8630 
8631 /**
8632  * Get aged-out flows.
8633  *
8634  * @param[in] dev
8635  *   Pointer to the Ethernet device structure.
8636  * @param[in] context
8637  *   The address of an array of pointers to the aged-out flows contexts.
8638  * @param[in] nb_countexts
8639  *   The length of context array pointers.
8640  * @param[out] error
8641  *   Perform verbose error reporting if not NULL. Initialized in case of
8642  *   error only.
8643  *
8644  * @return
8645  *   how many contexts get in success, otherwise negative errno value.
8646  *   if nb_contexts is 0, return the amount of all aged contexts.
8647  *   if nb_contexts is not 0 , return the amount of aged flows reported
8648  *   in the context array.
8649  */
8650 int
8651 mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
8652 			uint32_t nb_contexts, struct rte_flow_error *error)
8653 {
8654 	const struct mlx5_flow_driver_ops *fops;
8655 	struct rte_flow_attr attr = { .transfer = 0 };
8656 
8657 	if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
8658 		fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
8659 		return fops->get_aged_flows(dev, contexts, nb_contexts,
8660 						    error);
8661 	}
8662 	DRV_LOG(ERR,
8663 		"port %u get aged flows is not supported.",
8664 		 dev->data->port_id);
8665 	return -ENOTSUP;
8666 }
8667 
8668 /* Wrapper for driver action_validate op callback */
8669 static int
8670 flow_drv_action_validate(struct rte_eth_dev *dev,
8671 			 const struct rte_flow_indir_action_conf *conf,
8672 			 const struct rte_flow_action *action,
8673 			 const struct mlx5_flow_driver_ops *fops,
8674 			 struct rte_flow_error *error)
8675 {
8676 	static const char err_msg[] = "indirect action validation unsupported";
8677 
8678 	if (!fops->action_validate) {
8679 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8680 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8681 				   NULL, err_msg);
8682 		return -rte_errno;
8683 	}
8684 	return fops->action_validate(dev, conf, action, error);
8685 }
8686 
8687 /**
8688  * Destroys the shared action by handle.
8689  *
8690  * @param dev
8691  *   Pointer to Ethernet device structure.
8692  * @param[in] handle
8693  *   Handle for the indirect action object to be destroyed.
8694  * @param[out] error
8695  *   Perform verbose error reporting if not NULL. PMDs initialize this
8696  *   structure in case of error only.
8697  *
8698  * @return
8699  *   0 on success, a negative errno value otherwise and rte_errno is set.
8700  *
8701  * @note: wrapper for driver action_create op callback.
8702  */
8703 static int
8704 mlx5_action_handle_destroy(struct rte_eth_dev *dev,
8705 			   struct rte_flow_action_handle *handle,
8706 			   struct rte_flow_error *error)
8707 {
8708 	static const char err_msg[] = "indirect action destruction unsupported";
8709 	struct rte_flow_attr attr = { .transfer = 0 };
8710 	const struct mlx5_flow_driver_ops *fops =
8711 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8712 
8713 	if (!fops->action_destroy) {
8714 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8715 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8716 				   NULL, err_msg);
8717 		return -rte_errno;
8718 	}
8719 	return fops->action_destroy(dev, handle, error);
8720 }
8721 
8722 /* Wrapper for driver action_destroy op callback */
8723 static int
8724 flow_drv_action_update(struct rte_eth_dev *dev,
8725 		       struct rte_flow_action_handle *handle,
8726 		       const void *update,
8727 		       const struct mlx5_flow_driver_ops *fops,
8728 		       struct rte_flow_error *error)
8729 {
8730 	static const char err_msg[] = "indirect action update unsupported";
8731 
8732 	if (!fops->action_update) {
8733 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8734 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8735 				   NULL, err_msg);
8736 		return -rte_errno;
8737 	}
8738 	return fops->action_update(dev, handle, update, error);
8739 }
8740 
8741 /* Wrapper for driver action_destroy op callback */
8742 static int
8743 flow_drv_action_query(struct rte_eth_dev *dev,
8744 		      const struct rte_flow_action_handle *handle,
8745 		      void *data,
8746 		      const struct mlx5_flow_driver_ops *fops,
8747 		      struct rte_flow_error *error)
8748 {
8749 	static const char err_msg[] = "indirect action query unsupported";
8750 
8751 	if (!fops->action_query) {
8752 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8753 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8754 				   NULL, err_msg);
8755 		return -rte_errno;
8756 	}
8757 	return fops->action_query(dev, handle, data, error);
8758 }
8759 
8760 /**
8761  * Create indirect action for reuse in multiple flow rules.
8762  *
8763  * @param dev
8764  *   Pointer to Ethernet device structure.
8765  * @param conf
8766  *   Pointer to indirect action object configuration.
8767  * @param[in] action
8768  *   Action configuration for indirect action object creation.
8769  * @param[out] error
8770  *   Perform verbose error reporting if not NULL. PMDs initialize this
8771  *   structure in case of error only.
8772  * @return
8773  *   A valid handle in case of success, NULL otherwise and rte_errno is set.
8774  */
8775 static struct rte_flow_action_handle *
8776 mlx5_action_handle_create(struct rte_eth_dev *dev,
8777 			  const struct rte_flow_indir_action_conf *conf,
8778 			  const struct rte_flow_action *action,
8779 			  struct rte_flow_error *error)
8780 {
8781 	static const char err_msg[] = "indirect action creation unsupported";
8782 	struct rte_flow_attr attr = { .transfer = 0 };
8783 	const struct mlx5_flow_driver_ops *fops =
8784 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8785 
8786 	if (flow_drv_action_validate(dev, conf, action, fops, error))
8787 		return NULL;
8788 	if (!fops->action_create) {
8789 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
8790 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
8791 				   NULL, err_msg);
8792 		return NULL;
8793 	}
8794 	return fops->action_create(dev, conf, action, error);
8795 }
8796 
8797 /**
8798  * Updates inplace the indirect action configuration pointed by *handle*
8799  * with the configuration provided as *update* argument.
8800  * The update of the indirect action configuration effects all flow rules
8801  * reusing the action via handle.
8802  *
8803  * @param dev
8804  *   Pointer to Ethernet device structure.
8805  * @param[in] handle
8806  *   Handle for the indirect action to be updated.
8807  * @param[in] update
8808  *   Action specification used to modify the action pointed by handle.
8809  *   *update* could be of same type with the action pointed by the *handle*
8810  *   handle argument, or some other structures like a wrapper, depending on
8811  *   the indirect action type.
8812  * @param[out] error
8813  *   Perform verbose error reporting if not NULL. PMDs initialize this
8814  *   structure in case of error only.
8815  *
8816  * @return
8817  *   0 on success, a negative errno value otherwise and rte_errno is set.
8818  */
8819 static int
8820 mlx5_action_handle_update(struct rte_eth_dev *dev,
8821 		struct rte_flow_action_handle *handle,
8822 		const void *update,
8823 		struct rte_flow_error *error)
8824 {
8825 	struct rte_flow_attr attr = { .transfer = 0 };
8826 	const struct mlx5_flow_driver_ops *fops =
8827 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8828 	int ret;
8829 
8830 	ret = flow_drv_action_validate(dev, NULL,
8831 			(const struct rte_flow_action *)update, fops, error);
8832 	if (ret)
8833 		return ret;
8834 	return flow_drv_action_update(dev, handle, update, fops,
8835 				      error);
8836 }
8837 
8838 /**
8839  * Query the indirect action by handle.
8840  *
8841  * This function allows retrieving action-specific data such as counters.
8842  * Data is gathered by special action which may be present/referenced in
8843  * more than one flow rule definition.
8844  *
8845  * see @RTE_FLOW_ACTION_TYPE_COUNT
8846  *
8847  * @param dev
8848  *   Pointer to Ethernet device structure.
8849  * @param[in] handle
8850  *   Handle for the indirect action to query.
8851  * @param[in, out] data
8852  *   Pointer to storage for the associated query data type.
8853  * @param[out] error
8854  *   Perform verbose error reporting if not NULL. PMDs initialize this
8855  *   structure in case of error only.
8856  *
8857  * @return
8858  *   0 on success, a negative errno value otherwise and rte_errno is set.
8859  */
8860 static int
8861 mlx5_action_handle_query(struct rte_eth_dev *dev,
8862 			 const struct rte_flow_action_handle *handle,
8863 			 void *data,
8864 			 struct rte_flow_error *error)
8865 {
8866 	struct rte_flow_attr attr = { .transfer = 0 };
8867 	const struct mlx5_flow_driver_ops *fops =
8868 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
8869 
8870 	return flow_drv_action_query(dev, handle, data, fops, error);
8871 }
8872 
8873 /**
8874  * Destroy all indirect actions (shared RSS).
8875  *
8876  * @param dev
8877  *   Pointer to Ethernet device.
8878  *
8879  * @return
8880  *   0 on success, a negative errno value otherwise and rte_errno is set.
8881  */
8882 int
8883 mlx5_action_handle_flush(struct rte_eth_dev *dev)
8884 {
8885 	struct rte_flow_error error;
8886 	struct mlx5_priv *priv = dev->data->dev_private;
8887 	struct mlx5_shared_action_rss *shared_rss;
8888 	int ret = 0;
8889 	uint32_t idx;
8890 
8891 	ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS],
8892 		      priv->rss_shared_actions, idx, shared_rss, next) {
8893 		ret |= mlx5_action_handle_destroy(dev,
8894 		       (struct rte_flow_action_handle *)(uintptr_t)idx, &error);
8895 	}
8896 	return ret;
8897 }
8898 
8899 /**
8900  * Validate existing indirect actions against current device configuration
8901  * and attach them to device resources.
8902  *
8903  * @param dev
8904  *   Pointer to Ethernet device.
8905  *
8906  * @return
8907  *   0 on success, a negative errno value otherwise and rte_errno is set.
8908  */
8909 int
8910 mlx5_action_handle_attach(struct rte_eth_dev *dev)
8911 {
8912 	struct mlx5_priv *priv = dev->data->dev_private;
8913 	struct mlx5_indexed_pool *ipool =
8914 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8915 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8916 	int ret = 0;
8917 	uint32_t idx;
8918 
8919 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8920 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8921 		const char *message;
8922 		uint32_t queue_idx;
8923 
8924 		ret = mlx5_validate_rss_queues(dev, ind_tbl->queues,
8925 					       ind_tbl->queues_n,
8926 					       &message, &queue_idx);
8927 		if (ret != 0) {
8928 			DRV_LOG(ERR, "Port %u cannot use queue %u in RSS: %s",
8929 				dev->data->port_id, ind_tbl->queues[queue_idx],
8930 				message);
8931 			break;
8932 		}
8933 	}
8934 	if (ret != 0)
8935 		return ret;
8936 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8937 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8938 
8939 		ret = mlx5_ind_table_obj_attach(dev, ind_tbl);
8940 		if (ret != 0) {
8941 			DRV_LOG(ERR, "Port %u could not attach "
8942 				"indirection table obj %p",
8943 				dev->data->port_id, (void *)ind_tbl);
8944 			goto error;
8945 		}
8946 	}
8947 	return 0;
8948 error:
8949 	shared_rss_last = shared_rss;
8950 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8951 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8952 
8953 		if (shared_rss == shared_rss_last)
8954 			break;
8955 		if (mlx5_ind_table_obj_detach(dev, ind_tbl) != 0)
8956 			DRV_LOG(CRIT, "Port %u could not detach "
8957 				"indirection table obj %p on rollback",
8958 				dev->data->port_id, (void *)ind_tbl);
8959 	}
8960 	return ret;
8961 }
8962 
8963 /**
8964  * Detach indirect actions of the device from its resources.
8965  *
8966  * @param dev
8967  *   Pointer to Ethernet device.
8968  *
8969  * @return
8970  *   0 on success, a negative errno value otherwise and rte_errno is set.
8971  */
8972 int
8973 mlx5_action_handle_detach(struct rte_eth_dev *dev)
8974 {
8975 	struct mlx5_priv *priv = dev->data->dev_private;
8976 	struct mlx5_indexed_pool *ipool =
8977 			priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS];
8978 	struct mlx5_shared_action_rss *shared_rss, *shared_rss_last;
8979 	int ret = 0;
8980 	uint32_t idx;
8981 
8982 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8983 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8984 
8985 		ret = mlx5_ind_table_obj_detach(dev, ind_tbl);
8986 		if (ret != 0) {
8987 			DRV_LOG(ERR, "Port %u could not detach "
8988 				"indirection table obj %p",
8989 				dev->data->port_id, (void *)ind_tbl);
8990 			goto error;
8991 		}
8992 	}
8993 	return 0;
8994 error:
8995 	shared_rss_last = shared_rss;
8996 	ILIST_FOREACH(ipool, priv->rss_shared_actions, idx, shared_rss, next) {
8997 		struct mlx5_ind_table_obj *ind_tbl = shared_rss->ind_tbl;
8998 
8999 		if (shared_rss == shared_rss_last)
9000 			break;
9001 		if (mlx5_ind_table_obj_attach(dev, ind_tbl) != 0)
9002 			DRV_LOG(CRIT, "Port %u could not attach "
9003 				"indirection table obj %p on rollback",
9004 				dev->data->port_id, (void *)ind_tbl);
9005 	}
9006 	return ret;
9007 }
9008 
9009 #ifndef HAVE_MLX5DV_DR
9010 #define MLX5_DOMAIN_SYNC_FLOW ((1 << 0) | (1 << 1))
9011 #else
9012 #define MLX5_DOMAIN_SYNC_FLOW \
9013 	(MLX5DV_DR_DOMAIN_SYNC_FLAGS_SW | MLX5DV_DR_DOMAIN_SYNC_FLAGS_HW)
9014 #endif
9015 
9016 int rte_pmd_mlx5_sync_flow(uint16_t port_id, uint32_t domains)
9017 {
9018 	struct rte_eth_dev *dev = &rte_eth_devices[port_id];
9019 	const struct mlx5_flow_driver_ops *fops;
9020 	int ret;
9021 	struct rte_flow_attr attr = { .transfer = 0 };
9022 
9023 	fops = flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9024 	ret = fops->sync_domain(dev, domains, MLX5_DOMAIN_SYNC_FLOW);
9025 	if (ret > 0)
9026 		ret = -ret;
9027 	return ret;
9028 }
9029 
9030 const struct mlx5_flow_tunnel *
9031 mlx5_get_tof(const struct rte_flow_item *item,
9032 	     const struct rte_flow_action *action,
9033 	     enum mlx5_tof_rule_type *rule_type)
9034 {
9035 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9036 		if (item->type == (typeof(item->type))
9037 				  MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL) {
9038 			*rule_type = MLX5_TUNNEL_OFFLOAD_MATCH_RULE;
9039 			return flow_items_to_tunnel(item);
9040 		}
9041 	}
9042 	for (; action->conf != RTE_FLOW_ACTION_TYPE_END; action++) {
9043 		if (action->type == (typeof(action->type))
9044 				    MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET) {
9045 			*rule_type = MLX5_TUNNEL_OFFLOAD_SET_RULE;
9046 			return flow_actions_to_tunnel(action);
9047 		}
9048 	}
9049 	return NULL;
9050 }
9051 
9052 /**
9053  * tunnel offload functionality is defined for DV environment only
9054  */
9055 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
9056 __extension__
9057 union tunnel_offload_mark {
9058 	uint32_t val;
9059 	struct {
9060 		uint32_t app_reserve:8;
9061 		uint32_t table_id:15;
9062 		uint32_t transfer:1;
9063 		uint32_t _unused_:8;
9064 	};
9065 };
9066 
9067 static bool
9068 mlx5_access_tunnel_offload_db
9069 	(struct rte_eth_dev *dev,
9070 	 bool (*match)(struct rte_eth_dev *,
9071 		       struct mlx5_flow_tunnel *, const void *),
9072 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9073 	 void (*miss)(struct rte_eth_dev *, void *),
9074 	 void *ctx, bool lock_op);
9075 
9076 static int
9077 flow_tunnel_add_default_miss(struct rte_eth_dev *dev,
9078 			     struct rte_flow *flow,
9079 			     const struct rte_flow_attr *attr,
9080 			     const struct rte_flow_action *app_actions,
9081 			     uint32_t flow_idx,
9082 			     const struct mlx5_flow_tunnel *tunnel,
9083 			     struct tunnel_default_miss_ctx *ctx,
9084 			     struct rte_flow_error *error)
9085 {
9086 	struct mlx5_priv *priv = dev->data->dev_private;
9087 	struct mlx5_flow *dev_flow;
9088 	struct rte_flow_attr miss_attr = *attr;
9089 	const struct rte_flow_item miss_items[2] = {
9090 		{
9091 			.type = RTE_FLOW_ITEM_TYPE_ETH,
9092 			.spec = NULL,
9093 			.last = NULL,
9094 			.mask = NULL
9095 		},
9096 		{
9097 			.type = RTE_FLOW_ITEM_TYPE_END,
9098 			.spec = NULL,
9099 			.last = NULL,
9100 			.mask = NULL
9101 		}
9102 	};
9103 	union tunnel_offload_mark mark_id;
9104 	struct rte_flow_action_mark miss_mark;
9105 	struct rte_flow_action miss_actions[3] = {
9106 		[0] = { .type = RTE_FLOW_ACTION_TYPE_MARK, .conf = &miss_mark },
9107 		[2] = { .type = RTE_FLOW_ACTION_TYPE_END,  .conf = NULL }
9108 	};
9109 	const struct rte_flow_action_jump *jump_data;
9110 	uint32_t i, flow_table = 0; /* prevent compilation warning */
9111 	struct flow_grp_info grp_info = {
9112 		.external = 1,
9113 		.transfer = attr->transfer,
9114 		.fdb_def_rule = !!priv->fdb_def_rule,
9115 		.std_tbl_fix = 0,
9116 	};
9117 	int ret;
9118 
9119 	if (!attr->transfer) {
9120 		uint32_t q_size;
9121 
9122 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_RSS;
9123 		q_size = priv->reta_idx_n * sizeof(ctx->queue[0]);
9124 		ctx->queue = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, q_size,
9125 					 0, SOCKET_ID_ANY);
9126 		if (!ctx->queue)
9127 			return rte_flow_error_set
9128 				(error, ENOMEM,
9129 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9130 				NULL, "invalid default miss RSS");
9131 		ctx->action_rss.func = RTE_ETH_HASH_FUNCTION_DEFAULT,
9132 		ctx->action_rss.level = 0,
9133 		ctx->action_rss.types = priv->rss_conf.rss_hf,
9134 		ctx->action_rss.key_len = priv->rss_conf.rss_key_len,
9135 		ctx->action_rss.queue_num = priv->reta_idx_n,
9136 		ctx->action_rss.key = priv->rss_conf.rss_key,
9137 		ctx->action_rss.queue = ctx->queue;
9138 		if (!priv->reta_idx_n || !priv->rxqs_n)
9139 			return rte_flow_error_set
9140 				(error, EINVAL,
9141 				RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9142 				NULL, "invalid port configuration");
9143 		if (!(dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
9144 			ctx->action_rss.types = 0;
9145 		for (i = 0; i != priv->reta_idx_n; ++i)
9146 			ctx->queue[i] = (*priv->reta_idx)[i];
9147 	} else {
9148 		miss_actions[1].type = RTE_FLOW_ACTION_TYPE_JUMP;
9149 		ctx->miss_jump.group = MLX5_TNL_MISS_FDB_JUMP_GRP;
9150 	}
9151 	miss_actions[1].conf = (typeof(miss_actions[1].conf))ctx->raw;
9152 	for (; app_actions->type != RTE_FLOW_ACTION_TYPE_JUMP; app_actions++);
9153 	jump_data = app_actions->conf;
9154 	miss_attr.priority = MLX5_TNL_MISS_RULE_PRIORITY;
9155 	miss_attr.group = jump_data->group;
9156 	ret = mlx5_flow_group_to_table(dev, tunnel, jump_data->group,
9157 				       &flow_table, &grp_info, error);
9158 	if (ret)
9159 		return rte_flow_error_set(error, EINVAL,
9160 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
9161 					  NULL, "invalid tunnel id");
9162 	mark_id.app_reserve = 0;
9163 	mark_id.table_id = tunnel_flow_tbl_to_id(flow_table);
9164 	mark_id.transfer = !!attr->transfer;
9165 	mark_id._unused_ = 0;
9166 	miss_mark.id = mark_id.val;
9167 	dev_flow = flow_drv_prepare(dev, flow, &miss_attr,
9168 				    miss_items, miss_actions, flow_idx, error);
9169 	if (!dev_flow)
9170 		return -rte_errno;
9171 	dev_flow->flow = flow;
9172 	dev_flow->external = true;
9173 	dev_flow->tunnel = tunnel;
9174 	dev_flow->tof_type = MLX5_TUNNEL_OFFLOAD_MISS_RULE;
9175 	/* Subflow object was created, we must include one in the list. */
9176 	SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
9177 		      dev_flow->handle, next);
9178 	DRV_LOG(DEBUG,
9179 		"port %u tunnel type=%d id=%u miss rule priority=%u group=%u",
9180 		dev->data->port_id, tunnel->app_tunnel.type,
9181 		tunnel->tunnel_id, miss_attr.priority, miss_attr.group);
9182 	ret = flow_drv_translate(dev, dev_flow, &miss_attr, miss_items,
9183 				  miss_actions, error);
9184 	if (!ret)
9185 		ret = flow_mreg_update_copy_table(dev, flow, miss_actions,
9186 						  error);
9187 
9188 	return ret;
9189 }
9190 
9191 static const struct mlx5_flow_tbl_data_entry  *
9192 tunnel_mark_decode(struct rte_eth_dev *dev, uint32_t mark)
9193 {
9194 	struct mlx5_priv *priv = dev->data->dev_private;
9195 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9196 	struct mlx5_list_entry *he;
9197 	union tunnel_offload_mark mbits = { .val = mark };
9198 	union mlx5_flow_tbl_key table_key = {
9199 		{
9200 			.level = tunnel_id_to_flow_tbl(mbits.table_id),
9201 			.id = 0,
9202 			.reserved = 0,
9203 			.dummy = 0,
9204 			.is_fdb = !!mbits.transfer,
9205 			.is_egress = 0,
9206 		}
9207 	};
9208 	struct mlx5_flow_cb_ctx ctx = {
9209 		.data = &table_key.v64,
9210 	};
9211 
9212 	he = mlx5_hlist_lookup(sh->flow_tbls, table_key.v64, &ctx);
9213 	return he ?
9214 	       container_of(he, struct mlx5_flow_tbl_data_entry, entry) : NULL;
9215 }
9216 
9217 static void
9218 mlx5_flow_tunnel_grp2tbl_remove_cb(void *tool_ctx,
9219 				   struct mlx5_list_entry *entry)
9220 {
9221 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9222 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9223 
9224 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9225 			tunnel_flow_tbl_to_id(tte->flow_table));
9226 	mlx5_free(tte);
9227 }
9228 
9229 static int
9230 mlx5_flow_tunnel_grp2tbl_match_cb(void *tool_ctx __rte_unused,
9231 				  struct mlx5_list_entry *entry, void *cb_ctx)
9232 {
9233 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9234 	union tunnel_tbl_key tbl = {
9235 		.val = *(uint64_t *)(ctx->data),
9236 	};
9237 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9238 
9239 	return tbl.tunnel_id != tte->tunnel_id || tbl.group != tte->group;
9240 }
9241 
9242 static struct mlx5_list_entry *
9243 mlx5_flow_tunnel_grp2tbl_create_cb(void *tool_ctx, void *cb_ctx)
9244 {
9245 	struct mlx5_dev_ctx_shared *sh = tool_ctx;
9246 	struct mlx5_flow_cb_ctx *ctx = cb_ctx;
9247 	struct tunnel_tbl_entry *tte;
9248 	union tunnel_tbl_key tbl = {
9249 		.val = *(uint64_t *)(ctx->data),
9250 	};
9251 
9252 	tte = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO,
9253 			  sizeof(*tte), 0,
9254 			  SOCKET_ID_ANY);
9255 	if (!tte)
9256 		goto err;
9257 	mlx5_ipool_malloc(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9258 			  &tte->flow_table);
9259 	if (tte->flow_table >= MLX5_MAX_TABLES) {
9260 		DRV_LOG(ERR, "Tunnel TBL ID %d exceed max limit.",
9261 			tte->flow_table);
9262 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_TNL_TBL_ID],
9263 				tte->flow_table);
9264 		goto err;
9265 	} else if (!tte->flow_table) {
9266 		goto err;
9267 	}
9268 	tte->flow_table = tunnel_id_to_flow_tbl(tte->flow_table);
9269 	tte->tunnel_id = tbl.tunnel_id;
9270 	tte->group = tbl.group;
9271 	return &tte->hash;
9272 err:
9273 	if (tte)
9274 		mlx5_free(tte);
9275 	return NULL;
9276 }
9277 
9278 static struct mlx5_list_entry *
9279 mlx5_flow_tunnel_grp2tbl_clone_cb(void *tool_ctx __rte_unused,
9280 				  struct mlx5_list_entry *oentry,
9281 				  void *cb_ctx __rte_unused)
9282 {
9283 	struct tunnel_tbl_entry *tte = mlx5_malloc(MLX5_MEM_SYS, sizeof(*tte),
9284 						   0, SOCKET_ID_ANY);
9285 
9286 	if (!tte)
9287 		return NULL;
9288 	memcpy(tte, oentry, sizeof(*tte));
9289 	return &tte->hash;
9290 }
9291 
9292 static void
9293 mlx5_flow_tunnel_grp2tbl_clone_free_cb(void *tool_ctx __rte_unused,
9294 				       struct mlx5_list_entry *entry)
9295 {
9296 	struct tunnel_tbl_entry *tte = container_of(entry, typeof(*tte), hash);
9297 
9298 	mlx5_free(tte);
9299 }
9300 
9301 static uint32_t
9302 tunnel_flow_group_to_flow_table(struct rte_eth_dev *dev,
9303 				const struct mlx5_flow_tunnel *tunnel,
9304 				uint32_t group, uint32_t *table,
9305 				struct rte_flow_error *error)
9306 {
9307 	struct mlx5_list_entry *he;
9308 	struct tunnel_tbl_entry *tte;
9309 	union tunnel_tbl_key key = {
9310 		.tunnel_id = tunnel ? tunnel->tunnel_id : 0,
9311 		.group = group
9312 	};
9313 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9314 	struct mlx5_hlist *group_hash;
9315 	struct mlx5_flow_cb_ctx ctx = {
9316 		.data = &key.val,
9317 	};
9318 
9319 	group_hash = tunnel ? tunnel->groups : thub->groups;
9320 	he = mlx5_hlist_register(group_hash, key.val, &ctx);
9321 	if (!he)
9322 		return rte_flow_error_set(error, EINVAL,
9323 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
9324 					  NULL,
9325 					  "tunnel group index not supported");
9326 	tte = container_of(he, typeof(*tte), hash);
9327 	*table = tte->flow_table;
9328 	DRV_LOG(DEBUG, "port %u tunnel %u group=%#x table=%#x",
9329 		dev->data->port_id, key.tunnel_id, group, *table);
9330 	return 0;
9331 }
9332 
9333 static void
9334 mlx5_flow_tunnel_free(struct rte_eth_dev *dev,
9335 		      struct mlx5_flow_tunnel *tunnel)
9336 {
9337 	struct mlx5_priv *priv = dev->data->dev_private;
9338 	struct mlx5_indexed_pool *ipool;
9339 
9340 	DRV_LOG(DEBUG, "port %u release pmd tunnel id=0x%x",
9341 		dev->data->port_id, tunnel->tunnel_id);
9342 	LIST_REMOVE(tunnel, chain);
9343 	mlx5_hlist_destroy(tunnel->groups);
9344 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9345 	mlx5_ipool_free(ipool, tunnel->tunnel_id);
9346 }
9347 
9348 static bool
9349 mlx5_access_tunnel_offload_db
9350 	(struct rte_eth_dev *dev,
9351 	 bool (*match)(struct rte_eth_dev *,
9352 		       struct mlx5_flow_tunnel *, const void *),
9353 	 void (*hit)(struct rte_eth_dev *, struct mlx5_flow_tunnel *, void *),
9354 	 void (*miss)(struct rte_eth_dev *, void *),
9355 	 void *ctx, bool lock_op)
9356 {
9357 	bool verdict = false;
9358 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9359 	struct mlx5_flow_tunnel *tunnel;
9360 
9361 	rte_spinlock_lock(&thub->sl);
9362 	LIST_FOREACH(tunnel, &thub->tunnels, chain) {
9363 		verdict = match(dev, tunnel, (const void *)ctx);
9364 		if (verdict)
9365 			break;
9366 	}
9367 	if (!lock_op)
9368 		rte_spinlock_unlock(&thub->sl);
9369 	if (verdict && hit)
9370 		hit(dev, tunnel, ctx);
9371 	if (!verdict && miss)
9372 		miss(dev, ctx);
9373 	if (lock_op)
9374 		rte_spinlock_unlock(&thub->sl);
9375 
9376 	return verdict;
9377 }
9378 
9379 struct tunnel_db_find_tunnel_id_ctx {
9380 	uint32_t tunnel_id;
9381 	struct mlx5_flow_tunnel *tunnel;
9382 };
9383 
9384 static bool
9385 find_tunnel_id_match(struct rte_eth_dev *dev,
9386 		     struct mlx5_flow_tunnel *tunnel, const void *x)
9387 {
9388 	const struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9389 
9390 	RTE_SET_USED(dev);
9391 	return tunnel->tunnel_id == ctx->tunnel_id;
9392 }
9393 
9394 static void
9395 find_tunnel_id_hit(struct rte_eth_dev *dev,
9396 		   struct mlx5_flow_tunnel *tunnel, void *x)
9397 {
9398 	struct tunnel_db_find_tunnel_id_ctx *ctx = x;
9399 	RTE_SET_USED(dev);
9400 	ctx->tunnel = tunnel;
9401 }
9402 
9403 static struct mlx5_flow_tunnel *
9404 mlx5_find_tunnel_id(struct rte_eth_dev *dev, uint32_t id)
9405 {
9406 	struct tunnel_db_find_tunnel_id_ctx ctx = {
9407 		.tunnel_id = id,
9408 	};
9409 
9410 	mlx5_access_tunnel_offload_db(dev, find_tunnel_id_match,
9411 				      find_tunnel_id_hit, NULL, &ctx, true);
9412 
9413 	return ctx.tunnel;
9414 }
9415 
9416 static struct mlx5_flow_tunnel *
9417 mlx5_flow_tunnel_allocate(struct rte_eth_dev *dev,
9418 			  const struct rte_flow_tunnel *app_tunnel)
9419 {
9420 	struct mlx5_priv *priv = dev->data->dev_private;
9421 	struct mlx5_indexed_pool *ipool;
9422 	struct mlx5_flow_tunnel *tunnel;
9423 	uint32_t id;
9424 
9425 	ipool = priv->sh->ipool[MLX5_IPOOL_TUNNEL_ID];
9426 	tunnel = mlx5_ipool_zmalloc(ipool, &id);
9427 	if (!tunnel)
9428 		return NULL;
9429 	if (id >= MLX5_MAX_TUNNELS) {
9430 		mlx5_ipool_free(ipool, id);
9431 		DRV_LOG(ERR, "Tunnel ID %d exceed max limit.", id);
9432 		return NULL;
9433 	}
9434 	tunnel->groups = mlx5_hlist_create("tunnel groups", 64, false, true,
9435 					   priv->sh,
9436 					   mlx5_flow_tunnel_grp2tbl_create_cb,
9437 					   mlx5_flow_tunnel_grp2tbl_match_cb,
9438 					   mlx5_flow_tunnel_grp2tbl_remove_cb,
9439 					   mlx5_flow_tunnel_grp2tbl_clone_cb,
9440 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9441 	if (!tunnel->groups) {
9442 		mlx5_ipool_free(ipool, id);
9443 		return NULL;
9444 	}
9445 	/* initiate new PMD tunnel */
9446 	memcpy(&tunnel->app_tunnel, app_tunnel, sizeof(*app_tunnel));
9447 	tunnel->tunnel_id = id;
9448 	tunnel->action.type = (typeof(tunnel->action.type))
9449 			      MLX5_RTE_FLOW_ACTION_TYPE_TUNNEL_SET;
9450 	tunnel->action.conf = tunnel;
9451 	tunnel->item.type = (typeof(tunnel->item.type))
9452 			    MLX5_RTE_FLOW_ITEM_TYPE_TUNNEL;
9453 	tunnel->item.spec = tunnel;
9454 	tunnel->item.last = NULL;
9455 	tunnel->item.mask = NULL;
9456 
9457 	DRV_LOG(DEBUG, "port %u new pmd tunnel id=0x%x",
9458 		dev->data->port_id, tunnel->tunnel_id);
9459 
9460 	return tunnel;
9461 }
9462 
9463 struct tunnel_db_get_tunnel_ctx {
9464 	const struct rte_flow_tunnel *app_tunnel;
9465 	struct mlx5_flow_tunnel *tunnel;
9466 };
9467 
9468 static bool get_tunnel_match(struct rte_eth_dev *dev,
9469 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9470 {
9471 	const struct tunnel_db_get_tunnel_ctx *ctx = x;
9472 
9473 	RTE_SET_USED(dev);
9474 	return !memcmp(ctx->app_tunnel, &tunnel->app_tunnel,
9475 		       sizeof(*ctx->app_tunnel));
9476 }
9477 
9478 static void get_tunnel_hit(struct rte_eth_dev *dev,
9479 			   struct mlx5_flow_tunnel *tunnel, void *x)
9480 {
9481 	/* called under tunnel spinlock protection */
9482 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9483 
9484 	RTE_SET_USED(dev);
9485 	tunnel->refctn++;
9486 	ctx->tunnel = tunnel;
9487 }
9488 
9489 static void get_tunnel_miss(struct rte_eth_dev *dev, void *x)
9490 {
9491 	/* called under tunnel spinlock protection */
9492 	struct mlx5_flow_tunnel_hub *thub = mlx5_tunnel_hub(dev);
9493 	struct tunnel_db_get_tunnel_ctx *ctx = x;
9494 
9495 	rte_spinlock_unlock(&thub->sl);
9496 	ctx->tunnel = mlx5_flow_tunnel_allocate(dev, ctx->app_tunnel);
9497 	rte_spinlock_lock(&thub->sl);
9498 	if (ctx->tunnel) {
9499 		ctx->tunnel->refctn = 1;
9500 		LIST_INSERT_HEAD(&thub->tunnels, ctx->tunnel, chain);
9501 	}
9502 }
9503 
9504 
9505 static int
9506 mlx5_get_flow_tunnel(struct rte_eth_dev *dev,
9507 		     const struct rte_flow_tunnel *app_tunnel,
9508 		     struct mlx5_flow_tunnel **tunnel)
9509 {
9510 	struct tunnel_db_get_tunnel_ctx ctx = {
9511 		.app_tunnel = app_tunnel,
9512 	};
9513 
9514 	mlx5_access_tunnel_offload_db(dev, get_tunnel_match, get_tunnel_hit,
9515 				      get_tunnel_miss, &ctx, true);
9516 	*tunnel = ctx.tunnel;
9517 	return ctx.tunnel ? 0 : -ENOMEM;
9518 }
9519 
9520 void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id)
9521 {
9522 	struct mlx5_flow_tunnel_hub *thub = sh->tunnel_hub;
9523 
9524 	if (!thub)
9525 		return;
9526 	if (!LIST_EMPTY(&thub->tunnels))
9527 		DRV_LOG(WARNING, "port %u tunnels present", port_id);
9528 	mlx5_hlist_destroy(thub->groups);
9529 	mlx5_free(thub);
9530 }
9531 
9532 int mlx5_alloc_tunnel_hub(struct mlx5_dev_ctx_shared *sh)
9533 {
9534 	int err;
9535 	struct mlx5_flow_tunnel_hub *thub;
9536 
9537 	thub = mlx5_malloc(MLX5_MEM_SYS | MLX5_MEM_ZERO, sizeof(*thub),
9538 			   0, SOCKET_ID_ANY);
9539 	if (!thub)
9540 		return -ENOMEM;
9541 	LIST_INIT(&thub->tunnels);
9542 	rte_spinlock_init(&thub->sl);
9543 	thub->groups = mlx5_hlist_create("flow groups", 64,
9544 					 false, true, sh,
9545 					 mlx5_flow_tunnel_grp2tbl_create_cb,
9546 					 mlx5_flow_tunnel_grp2tbl_match_cb,
9547 					 mlx5_flow_tunnel_grp2tbl_remove_cb,
9548 					 mlx5_flow_tunnel_grp2tbl_clone_cb,
9549 					mlx5_flow_tunnel_grp2tbl_clone_free_cb);
9550 	if (!thub->groups) {
9551 		err = -rte_errno;
9552 		goto err;
9553 	}
9554 	sh->tunnel_hub = thub;
9555 
9556 	return 0;
9557 
9558 err:
9559 	if (thub->groups)
9560 		mlx5_hlist_destroy(thub->groups);
9561 	if (thub)
9562 		mlx5_free(thub);
9563 	return err;
9564 }
9565 
9566 static inline int
9567 mlx5_flow_tunnel_validate(struct rte_eth_dev *dev,
9568 			  struct rte_flow_tunnel *tunnel,
9569 			  struct rte_flow_error *error)
9570 {
9571 	struct mlx5_priv *priv = dev->data->dev_private;
9572 
9573 	if (!priv->sh->config.dv_flow_en)
9574 		return rte_flow_error_set(error, ENOTSUP,
9575 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9576 					  "flow DV interface is off");
9577 	if (!is_tunnel_offload_active(dev))
9578 		return rte_flow_error_set(error, ENOTSUP,
9579 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9580 					  "tunnel offload was not activated");
9581 	if (!tunnel)
9582 		return rte_flow_error_set(error, EINVAL,
9583 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9584 					  "no application tunnel");
9585 	switch (tunnel->type) {
9586 	default:
9587 		return rte_flow_error_set(error, EINVAL,
9588 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9589 					  "unsupported tunnel type");
9590 	case RTE_FLOW_ITEM_TYPE_VXLAN:
9591 	case RTE_FLOW_ITEM_TYPE_GRE:
9592 	case RTE_FLOW_ITEM_TYPE_NVGRE:
9593 	case RTE_FLOW_ITEM_TYPE_GENEVE:
9594 		break;
9595 	}
9596 	return 0;
9597 }
9598 
9599 static int
9600 mlx5_flow_tunnel_decap_set(struct rte_eth_dev *dev,
9601 		    struct rte_flow_tunnel *app_tunnel,
9602 		    struct rte_flow_action **actions,
9603 		    uint32_t *num_of_actions,
9604 		    struct rte_flow_error *error)
9605 {
9606 	struct mlx5_flow_tunnel *tunnel;
9607 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9608 
9609 	if (ret)
9610 		return ret;
9611 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9612 	if (ret < 0) {
9613 		return rte_flow_error_set(error, ret,
9614 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
9615 					  "failed to initialize pmd tunnel");
9616 	}
9617 	*actions = &tunnel->action;
9618 	*num_of_actions = 1;
9619 	return 0;
9620 }
9621 
9622 static int
9623 mlx5_flow_tunnel_match(struct rte_eth_dev *dev,
9624 		       struct rte_flow_tunnel *app_tunnel,
9625 		       struct rte_flow_item **items,
9626 		       uint32_t *num_of_items,
9627 		       struct rte_flow_error *error)
9628 {
9629 	struct mlx5_flow_tunnel *tunnel;
9630 	int ret = mlx5_flow_tunnel_validate(dev, app_tunnel, error);
9631 
9632 	if (ret)
9633 		return ret;
9634 	ret = mlx5_get_flow_tunnel(dev, app_tunnel, &tunnel);
9635 	if (ret < 0) {
9636 		return rte_flow_error_set(error, ret,
9637 					  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9638 					  "failed to initialize pmd tunnel");
9639 	}
9640 	*items = &tunnel->item;
9641 	*num_of_items = 1;
9642 	return 0;
9643 }
9644 
9645 struct tunnel_db_element_release_ctx {
9646 	struct rte_flow_item *items;
9647 	struct rte_flow_action *actions;
9648 	uint32_t num_elements;
9649 	struct rte_flow_error *error;
9650 	int ret;
9651 };
9652 
9653 static bool
9654 tunnel_element_release_match(struct rte_eth_dev *dev,
9655 			     struct mlx5_flow_tunnel *tunnel, const void *x)
9656 {
9657 	const struct tunnel_db_element_release_ctx *ctx = x;
9658 
9659 	RTE_SET_USED(dev);
9660 	if (ctx->num_elements != 1)
9661 		return false;
9662 	else if (ctx->items)
9663 		return ctx->items == &tunnel->item;
9664 	else if (ctx->actions)
9665 		return ctx->actions == &tunnel->action;
9666 
9667 	return false;
9668 }
9669 
9670 static void
9671 tunnel_element_release_hit(struct rte_eth_dev *dev,
9672 			   struct mlx5_flow_tunnel *tunnel, void *x)
9673 {
9674 	struct tunnel_db_element_release_ctx *ctx = x;
9675 	ctx->ret = 0;
9676 	if (!__atomic_sub_fetch(&tunnel->refctn, 1, __ATOMIC_RELAXED))
9677 		mlx5_flow_tunnel_free(dev, tunnel);
9678 }
9679 
9680 static void
9681 tunnel_element_release_miss(struct rte_eth_dev *dev, void *x)
9682 {
9683 	struct tunnel_db_element_release_ctx *ctx = x;
9684 	RTE_SET_USED(dev);
9685 	ctx->ret = rte_flow_error_set(ctx->error, EINVAL,
9686 				      RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
9687 				      "invalid argument");
9688 }
9689 
9690 static int
9691 mlx5_flow_tunnel_item_release(struct rte_eth_dev *dev,
9692 		       struct rte_flow_item *pmd_items,
9693 		       uint32_t num_items, struct rte_flow_error *err)
9694 {
9695 	struct tunnel_db_element_release_ctx ctx = {
9696 		.items = pmd_items,
9697 		.actions = NULL,
9698 		.num_elements = num_items,
9699 		.error = err,
9700 	};
9701 
9702 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9703 				      tunnel_element_release_hit,
9704 				      tunnel_element_release_miss, &ctx, false);
9705 
9706 	return ctx.ret;
9707 }
9708 
9709 static int
9710 mlx5_flow_tunnel_action_release(struct rte_eth_dev *dev,
9711 			 struct rte_flow_action *pmd_actions,
9712 			 uint32_t num_actions, struct rte_flow_error *err)
9713 {
9714 	struct tunnel_db_element_release_ctx ctx = {
9715 		.items = NULL,
9716 		.actions = pmd_actions,
9717 		.num_elements = num_actions,
9718 		.error = err,
9719 	};
9720 
9721 	mlx5_access_tunnel_offload_db(dev, tunnel_element_release_match,
9722 				      tunnel_element_release_hit,
9723 				      tunnel_element_release_miss, &ctx, false);
9724 
9725 	return ctx.ret;
9726 }
9727 
9728 static int
9729 mlx5_flow_tunnel_get_restore_info(struct rte_eth_dev *dev,
9730 				  struct rte_mbuf *m,
9731 				  struct rte_flow_restore_info *info,
9732 				  struct rte_flow_error *err)
9733 {
9734 	uint64_t ol_flags = m->ol_flags;
9735 	const struct mlx5_flow_tbl_data_entry *tble;
9736 	const uint64_t mask = RTE_MBUF_F_RX_FDIR | RTE_MBUF_F_RX_FDIR_ID;
9737 
9738 	if (!is_tunnel_offload_active(dev)) {
9739 		info->flags = 0;
9740 		return 0;
9741 	}
9742 
9743 	if ((ol_flags & mask) != mask)
9744 		goto err;
9745 	tble = tunnel_mark_decode(dev, m->hash.fdir.hi);
9746 	if (!tble) {
9747 		DRV_LOG(DEBUG, "port %u invalid miss tunnel mark %#x",
9748 			dev->data->port_id, m->hash.fdir.hi);
9749 		goto err;
9750 	}
9751 	MLX5_ASSERT(tble->tunnel);
9752 	memcpy(&info->tunnel, &tble->tunnel->app_tunnel, sizeof(info->tunnel));
9753 	info->group_id = tble->group_id;
9754 	info->flags = RTE_FLOW_RESTORE_INFO_TUNNEL |
9755 		      RTE_FLOW_RESTORE_INFO_GROUP_ID |
9756 		      RTE_FLOW_RESTORE_INFO_ENCAPSULATED;
9757 
9758 	return 0;
9759 
9760 err:
9761 	return rte_flow_error_set(err, EINVAL,
9762 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9763 				  "failed to get restore info");
9764 }
9765 
9766 #else /* HAVE_IBV_FLOW_DV_SUPPORT */
9767 static int
9768 mlx5_flow_tunnel_decap_set(__rte_unused struct rte_eth_dev *dev,
9769 			   __rte_unused struct rte_flow_tunnel *app_tunnel,
9770 			   __rte_unused struct rte_flow_action **actions,
9771 			   __rte_unused uint32_t *num_of_actions,
9772 			   __rte_unused struct rte_flow_error *error)
9773 {
9774 	return -ENOTSUP;
9775 }
9776 
9777 static int
9778 mlx5_flow_tunnel_match(__rte_unused struct rte_eth_dev *dev,
9779 		       __rte_unused struct rte_flow_tunnel *app_tunnel,
9780 		       __rte_unused struct rte_flow_item **items,
9781 		       __rte_unused uint32_t *num_of_items,
9782 		       __rte_unused struct rte_flow_error *error)
9783 {
9784 	return -ENOTSUP;
9785 }
9786 
9787 static int
9788 mlx5_flow_tunnel_item_release(__rte_unused struct rte_eth_dev *dev,
9789 			      __rte_unused struct rte_flow_item *pmd_items,
9790 			      __rte_unused uint32_t num_items,
9791 			      __rte_unused struct rte_flow_error *err)
9792 {
9793 	return -ENOTSUP;
9794 }
9795 
9796 static int
9797 mlx5_flow_tunnel_action_release(__rte_unused struct rte_eth_dev *dev,
9798 				__rte_unused struct rte_flow_action *pmd_action,
9799 				__rte_unused uint32_t num_actions,
9800 				__rte_unused struct rte_flow_error *err)
9801 {
9802 	return -ENOTSUP;
9803 }
9804 
9805 static int
9806 mlx5_flow_tunnel_get_restore_info(__rte_unused struct rte_eth_dev *dev,
9807 				  __rte_unused struct rte_mbuf *m,
9808 				  __rte_unused struct rte_flow_restore_info *i,
9809 				  __rte_unused struct rte_flow_error *err)
9810 {
9811 	return -ENOTSUP;
9812 }
9813 
9814 static int
9815 flow_tunnel_add_default_miss(__rte_unused struct rte_eth_dev *dev,
9816 			     __rte_unused struct rte_flow *flow,
9817 			     __rte_unused const struct rte_flow_attr *attr,
9818 			     __rte_unused const struct rte_flow_action *actions,
9819 			     __rte_unused uint32_t flow_idx,
9820 			     __rte_unused const struct mlx5_flow_tunnel *tunnel,
9821 			     __rte_unused struct tunnel_default_miss_ctx *ctx,
9822 			     __rte_unused struct rte_flow_error *error)
9823 {
9824 	return -ENOTSUP;
9825 }
9826 
9827 static struct mlx5_flow_tunnel *
9828 mlx5_find_tunnel_id(__rte_unused struct rte_eth_dev *dev,
9829 		    __rte_unused uint32_t id)
9830 {
9831 	return NULL;
9832 }
9833 
9834 static void
9835 mlx5_flow_tunnel_free(__rte_unused struct rte_eth_dev *dev,
9836 		      __rte_unused struct mlx5_flow_tunnel *tunnel)
9837 {
9838 }
9839 
9840 static uint32_t
9841 tunnel_flow_group_to_flow_table(__rte_unused struct rte_eth_dev *dev,
9842 				__rte_unused const struct mlx5_flow_tunnel *t,
9843 				__rte_unused uint32_t group,
9844 				__rte_unused uint32_t *table,
9845 				struct rte_flow_error *error)
9846 {
9847 	return rte_flow_error_set(error, ENOTSUP,
9848 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9849 				  "tunnel offload requires DV support");
9850 }
9851 
9852 void
9853 mlx5_release_tunnel_hub(__rte_unused struct mlx5_dev_ctx_shared *sh,
9854 			__rte_unused  uint16_t port_id)
9855 {
9856 }
9857 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
9858 
9859 /* Flex flow item API */
9860 static struct rte_flow_item_flex_handle *
9861 mlx5_flow_flex_item_create(struct rte_eth_dev *dev,
9862 			   const struct rte_flow_item_flex_conf *conf,
9863 			   struct rte_flow_error *error)
9864 {
9865 	static const char err_msg[] = "flex item creation unsupported";
9866 	struct rte_flow_attr attr = { .transfer = 0 };
9867 	const struct mlx5_flow_driver_ops *fops =
9868 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9869 
9870 	if (!fops->item_create) {
9871 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9872 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9873 				   NULL, err_msg);
9874 		return NULL;
9875 	}
9876 	return fops->item_create(dev, conf, error);
9877 }
9878 
9879 static int
9880 mlx5_flow_flex_item_release(struct rte_eth_dev *dev,
9881 			    const struct rte_flow_item_flex_handle *handle,
9882 			    struct rte_flow_error *error)
9883 {
9884 	static const char err_msg[] = "flex item release unsupported";
9885 	struct rte_flow_attr attr = { .transfer = 0 };
9886 	const struct mlx5_flow_driver_ops *fops =
9887 			flow_get_drv_ops(flow_get_drv_type(dev, &attr));
9888 
9889 	if (!fops->item_release) {
9890 		DRV_LOG(ERR, "port %u %s.", dev->data->port_id, err_msg);
9891 		rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
9892 				   NULL, err_msg);
9893 		return -rte_errno;
9894 	}
9895 	return fops->item_release(dev, handle, error);
9896 }
9897 
9898 static void
9899 mlx5_dbg__print_pattern(const struct rte_flow_item *item)
9900 {
9901 	int ret;
9902 	struct rte_flow_error error;
9903 
9904 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
9905 		char *item_name;
9906 		ret = rte_flow_conv(RTE_FLOW_CONV_OP_ITEM_NAME_PTR, &item_name,
9907 				    sizeof(item_name),
9908 				    (void *)(uintptr_t)item->type, &error);
9909 		if (ret > 0)
9910 			printf("%s ", item_name);
9911 		else
9912 			printf("%d\n", (int)item->type);
9913 	}
9914 	printf("END\n");
9915 }
9916 
9917 static int
9918 mlx5_flow_is_std_vxlan_port(const struct rte_flow_item *udp_item)
9919 {
9920 	const struct rte_flow_item_udp *spec = udp_item->spec;
9921 	const struct rte_flow_item_udp *mask = udp_item->mask;
9922 	uint16_t udp_dport = 0;
9923 
9924 	if (spec != NULL) {
9925 		if (!mask)
9926 			mask = &rte_flow_item_udp_mask;
9927 		udp_dport = rte_be_to_cpu_16(spec->hdr.dst_port &
9928 				mask->hdr.dst_port);
9929 	}
9930 	return (!udp_dport || udp_dport == MLX5_UDP_PORT_VXLAN);
9931 }
9932 
9933 static const struct mlx5_flow_expand_node *
9934 mlx5_flow_expand_rss_adjust_node(const struct rte_flow_item *pattern,
9935 		unsigned int item_idx,
9936 		const struct mlx5_flow_expand_node graph[],
9937 		const struct mlx5_flow_expand_node *node)
9938 {
9939 	const struct rte_flow_item *item = pattern + item_idx, *prev_item;
9940 
9941 	if (item->type == RTE_FLOW_ITEM_TYPE_VXLAN &&
9942 			node != NULL &&
9943 			node->type == RTE_FLOW_ITEM_TYPE_VXLAN) {
9944 		/*
9945 		 * The expansion node is VXLAN and it is also the last
9946 		 * expandable item in the pattern, so need to continue
9947 		 * expansion of the inner tunnel.
9948 		 */
9949 		MLX5_ASSERT(item_idx > 0);
9950 		prev_item = pattern + item_idx - 1;
9951 		MLX5_ASSERT(prev_item->type == RTE_FLOW_ITEM_TYPE_UDP);
9952 		if (mlx5_flow_is_std_vxlan_port(prev_item))
9953 			return &graph[MLX5_EXPANSION_STD_VXLAN];
9954 		return &graph[MLX5_EXPANSION_L3_VXLAN];
9955 	}
9956 	return node;
9957 }
9958 
9959 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
9960 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
9961 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
9962 };
9963 
9964 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
9965 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
9966 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
9967 	{ 9, 10, 11 }, { 12, 13, 14 },
9968 };
9969 
9970 /**
9971  * Discover the number of available flow priorities.
9972  *
9973  * @param dev
9974  *   Ethernet device.
9975  *
9976  * @return
9977  *   On success, number of available flow priorities.
9978  *   On failure, a negative errno-style code and rte_errno is set.
9979  */
9980 int
9981 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
9982 {
9983 	static const uint16_t vprio[] = {8, 16};
9984 	const struct mlx5_priv *priv = dev->data->dev_private;
9985 	const struct mlx5_flow_driver_ops *fops;
9986 	enum mlx5_flow_drv_type type;
9987 	int ret;
9988 
9989 	type = mlx5_flow_os_get_type();
9990 	if (type == MLX5_FLOW_TYPE_MAX) {
9991 		type = MLX5_FLOW_TYPE_VERBS;
9992 		if (priv->sh->cdev->config.devx && priv->sh->config.dv_flow_en)
9993 			type = MLX5_FLOW_TYPE_DV;
9994 	}
9995 	fops = flow_get_drv_ops(type);
9996 	if (fops->discover_priorities == NULL) {
9997 		DRV_LOG(ERR, "Priority discovery not supported");
9998 		rte_errno = ENOTSUP;
9999 		return -rte_errno;
10000 	}
10001 	ret = fops->discover_priorities(dev, vprio, RTE_DIM(vprio));
10002 	if (ret < 0)
10003 		return ret;
10004 	switch (ret) {
10005 	case 8:
10006 		ret = RTE_DIM(priority_map_3);
10007 		break;
10008 	case 16:
10009 		ret = RTE_DIM(priority_map_5);
10010 		break;
10011 	default:
10012 		rte_errno = ENOTSUP;
10013 		DRV_LOG(ERR,
10014 			"port %u maximum priority: %d expected 8/16",
10015 			dev->data->port_id, ret);
10016 		return -rte_errno;
10017 	}
10018 	DRV_LOG(INFO, "port %u supported flow priorities:"
10019 		" 0-%d for ingress or egress root table,"
10020 		" 0-%d for non-root table or transfer root table.",
10021 		dev->data->port_id, ret - 2,
10022 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
10023 	return ret;
10024 }
10025 
10026 /**
10027  * Adjust flow priority based on the highest layer and the request priority.
10028  *
10029  * @param[in] dev
10030  *   Pointer to the Ethernet device structure.
10031  * @param[in] priority
10032  *   The rule base priority.
10033  * @param[in] subpriority
10034  *   The priority based on the items.
10035  *
10036  * @return
10037  *   The new priority.
10038  */
10039 uint32_t
10040 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
10041 			  uint32_t subpriority)
10042 {
10043 	uint32_t res = 0;
10044 	struct mlx5_priv *priv = dev->data->dev_private;
10045 
10046 	switch (priv->sh->flow_max_priority) {
10047 	case RTE_DIM(priority_map_3):
10048 		res = priority_map_3[priority][subpriority];
10049 		break;
10050 	case RTE_DIM(priority_map_5):
10051 		res = priority_map_5[priority][subpriority];
10052 		break;
10053 	}
10054 	return  res;
10055 }
10056