xref: /dpdk/drivers/net/mlx5/mlx5_flow_dv.c (revision 8809f78c7dd9f33a44a4f89c58fc91ded34296ed)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2018 Mellanox Technologies, Ltd
3  */
4 
5 #include <sys/queue.h>
6 #include <stdalign.h>
7 #include <stdint.h>
8 #include <string.h>
9 #include <unistd.h>
10 
11 #include <rte_common.h>
12 #include <rte_ether.h>
13 #include <rte_ethdev_driver.h>
14 #include <rte_flow.h>
15 #include <rte_flow_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_cycles.h>
18 #include <rte_ip.h>
19 #include <rte_gre.h>
20 #include <rte_vxlan.h>
21 #include <rte_gtp.h>
22 #include <rte_eal_paging.h>
23 #include <rte_mpls.h>
24 
25 #include <mlx5_glue.h>
26 #include <mlx5_devx_cmds.h>
27 #include <mlx5_prm.h>
28 #include <mlx5_malloc.h>
29 
30 #include "mlx5_defs.h"
31 #include "mlx5.h"
32 #include "mlx5_common_os.h"
33 #include "mlx5_flow.h"
34 #include "mlx5_flow_os.h"
35 #include "mlx5_rxtx.h"
36 
37 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
38 
39 #ifndef HAVE_IBV_FLOW_DEVX_COUNTERS
40 #define MLX5DV_FLOW_ACTION_COUNTERS_DEVX 0
41 #endif
42 
43 #ifndef HAVE_MLX5DV_DR_ESWITCH
44 #ifndef MLX5DV_FLOW_TABLE_TYPE_FDB
45 #define MLX5DV_FLOW_TABLE_TYPE_FDB 0
46 #endif
47 #endif
48 
49 #ifndef HAVE_MLX5DV_DR
50 #define MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL 1
51 #endif
52 
53 /* VLAN header definitions */
54 #define MLX5DV_FLOW_VLAN_PCP_SHIFT 13
55 #define MLX5DV_FLOW_VLAN_PCP_MASK (0x7 << MLX5DV_FLOW_VLAN_PCP_SHIFT)
56 #define MLX5DV_FLOW_VLAN_VID_MASK 0x0fff
57 #define MLX5DV_FLOW_VLAN_PCP_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK)
58 #define MLX5DV_FLOW_VLAN_VID_MASK_BE RTE_BE16(MLX5DV_FLOW_VLAN_VID_MASK)
59 
60 union flow_dv_attr {
61 	struct {
62 		uint32_t valid:1;
63 		uint32_t ipv4:1;
64 		uint32_t ipv6:1;
65 		uint32_t tcp:1;
66 		uint32_t udp:1;
67 		uint32_t reserved:27;
68 	};
69 	uint32_t attr;
70 };
71 
72 static int
73 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
74 			     struct mlx5_flow_tbl_resource *tbl);
75 
76 static int
77 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev);
78 
79 static int
80 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
81 				      uint32_t encap_decap_idx);
82 
83 static int
84 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
85 					uint32_t port_id);
86 
87 /**
88  * Initialize flow attributes structure according to flow items' types.
89  *
90  * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
91  * mode. For tunnel mode, the items to be modified are the outermost ones.
92  *
93  * @param[in] item
94  *   Pointer to item specification.
95  * @param[out] attr
96  *   Pointer to flow attributes structure.
97  * @param[in] dev_flow
98  *   Pointer to the sub flow.
99  * @param[in] tunnel_decap
100  *   Whether action is after tunnel decapsulation.
101  */
102 static void
103 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
104 		  struct mlx5_flow *dev_flow, bool tunnel_decap)
105 {
106 	uint64_t layers = dev_flow->handle->layers;
107 
108 	/*
109 	 * If layers is already initialized, it means this dev_flow is the
110 	 * suffix flow, the layers flags is set by the prefix flow. Need to
111 	 * use the layer flags from prefix flow as the suffix flow may not
112 	 * have the user defined items as the flow is split.
113 	 */
114 	if (layers) {
115 		if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
116 			attr->ipv4 = 1;
117 		else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
118 			attr->ipv6 = 1;
119 		if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
120 			attr->tcp = 1;
121 		else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
122 			attr->udp = 1;
123 		attr->valid = 1;
124 		return;
125 	}
126 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
127 		uint8_t next_protocol = 0xff;
128 		switch (item->type) {
129 		case RTE_FLOW_ITEM_TYPE_GRE:
130 		case RTE_FLOW_ITEM_TYPE_NVGRE:
131 		case RTE_FLOW_ITEM_TYPE_VXLAN:
132 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
133 		case RTE_FLOW_ITEM_TYPE_GENEVE:
134 		case RTE_FLOW_ITEM_TYPE_MPLS:
135 			if (tunnel_decap)
136 				attr->attr = 0;
137 			break;
138 		case RTE_FLOW_ITEM_TYPE_IPV4:
139 			if (!attr->ipv6)
140 				attr->ipv4 = 1;
141 			if (item->mask != NULL &&
142 			    ((const struct rte_flow_item_ipv4 *)
143 			    item->mask)->hdr.next_proto_id)
144 				next_protocol =
145 				    ((const struct rte_flow_item_ipv4 *)
146 				      (item->spec))->hdr.next_proto_id &
147 				    ((const struct rte_flow_item_ipv4 *)
148 				      (item->mask))->hdr.next_proto_id;
149 			if ((next_protocol == IPPROTO_IPIP ||
150 			    next_protocol == IPPROTO_IPV6) && tunnel_decap)
151 				attr->attr = 0;
152 			break;
153 		case RTE_FLOW_ITEM_TYPE_IPV6:
154 			if (!attr->ipv4)
155 				attr->ipv6 = 1;
156 			if (item->mask != NULL &&
157 			    ((const struct rte_flow_item_ipv6 *)
158 			    item->mask)->hdr.proto)
159 				next_protocol =
160 				    ((const struct rte_flow_item_ipv6 *)
161 				      (item->spec))->hdr.proto &
162 				    ((const struct rte_flow_item_ipv6 *)
163 				      (item->mask))->hdr.proto;
164 			if ((next_protocol == IPPROTO_IPIP ||
165 			    next_protocol == IPPROTO_IPV6) && tunnel_decap)
166 				attr->attr = 0;
167 			break;
168 		case RTE_FLOW_ITEM_TYPE_UDP:
169 			if (!attr->tcp)
170 				attr->udp = 1;
171 			break;
172 		case RTE_FLOW_ITEM_TYPE_TCP:
173 			if (!attr->udp)
174 				attr->tcp = 1;
175 			break;
176 		default:
177 			break;
178 		}
179 	}
180 	attr->valid = 1;
181 }
182 
183 /**
184  * Convert rte_mtr_color to mlx5 color.
185  *
186  * @param[in] rcol
187  *   rte_mtr_color.
188  *
189  * @return
190  *   mlx5 color.
191  */
192 static int
193 rte_col_2_mlx5_col(enum rte_color rcol)
194 {
195 	switch (rcol) {
196 	case RTE_COLOR_GREEN:
197 		return MLX5_FLOW_COLOR_GREEN;
198 	case RTE_COLOR_YELLOW:
199 		return MLX5_FLOW_COLOR_YELLOW;
200 	case RTE_COLOR_RED:
201 		return MLX5_FLOW_COLOR_RED;
202 	default:
203 		break;
204 	}
205 	return MLX5_FLOW_COLOR_UNDEFINED;
206 }
207 
208 struct field_modify_info {
209 	uint32_t size; /* Size of field in protocol header, in bytes. */
210 	uint32_t offset; /* Offset of field in protocol header, in bytes. */
211 	enum mlx5_modification_field id;
212 };
213 
214 struct field_modify_info modify_eth[] = {
215 	{4,  0, MLX5_MODI_OUT_DMAC_47_16},
216 	{2,  4, MLX5_MODI_OUT_DMAC_15_0},
217 	{4,  6, MLX5_MODI_OUT_SMAC_47_16},
218 	{2, 10, MLX5_MODI_OUT_SMAC_15_0},
219 	{0, 0, 0},
220 };
221 
222 struct field_modify_info modify_vlan_out_first_vid[] = {
223 	/* Size in bits !!! */
224 	{12, 0, MLX5_MODI_OUT_FIRST_VID},
225 	{0, 0, 0},
226 };
227 
228 struct field_modify_info modify_ipv4[] = {
229 	{1,  1, MLX5_MODI_OUT_IP_DSCP},
230 	{1,  8, MLX5_MODI_OUT_IPV4_TTL},
231 	{4, 12, MLX5_MODI_OUT_SIPV4},
232 	{4, 16, MLX5_MODI_OUT_DIPV4},
233 	{0, 0, 0},
234 };
235 
236 struct field_modify_info modify_ipv6[] = {
237 	{1,  0, MLX5_MODI_OUT_IP_DSCP},
238 	{1,  7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
239 	{4,  8, MLX5_MODI_OUT_SIPV6_127_96},
240 	{4, 12, MLX5_MODI_OUT_SIPV6_95_64},
241 	{4, 16, MLX5_MODI_OUT_SIPV6_63_32},
242 	{4, 20, MLX5_MODI_OUT_SIPV6_31_0},
243 	{4, 24, MLX5_MODI_OUT_DIPV6_127_96},
244 	{4, 28, MLX5_MODI_OUT_DIPV6_95_64},
245 	{4, 32, MLX5_MODI_OUT_DIPV6_63_32},
246 	{4, 36, MLX5_MODI_OUT_DIPV6_31_0},
247 	{0, 0, 0},
248 };
249 
250 struct field_modify_info modify_udp[] = {
251 	{2, 0, MLX5_MODI_OUT_UDP_SPORT},
252 	{2, 2, MLX5_MODI_OUT_UDP_DPORT},
253 	{0, 0, 0},
254 };
255 
256 struct field_modify_info modify_tcp[] = {
257 	{2, 0, MLX5_MODI_OUT_TCP_SPORT},
258 	{2, 2, MLX5_MODI_OUT_TCP_DPORT},
259 	{4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
260 	{4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
261 	{0, 0, 0},
262 };
263 
264 static void
265 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
266 			  uint8_t next_protocol, uint64_t *item_flags,
267 			  int *tunnel)
268 {
269 	MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
270 		    item->type == RTE_FLOW_ITEM_TYPE_IPV6);
271 	if (next_protocol == IPPROTO_IPIP) {
272 		*item_flags |= MLX5_FLOW_LAYER_IPIP;
273 		*tunnel = 1;
274 	}
275 	if (next_protocol == IPPROTO_IPV6) {
276 		*item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
277 		*tunnel = 1;
278 	}
279 }
280 
281 /**
282  * Acquire the synchronizing object to protect multithreaded access
283  * to shared dv context. Lock occurs only if context is actually
284  * shared, i.e. we have multiport IB device and representors are
285  * created.
286  *
287  * @param[in] dev
288  *   Pointer to the rte_eth_dev structure.
289  */
290 static void
291 flow_dv_shared_lock(struct rte_eth_dev *dev)
292 {
293 	struct mlx5_priv *priv = dev->data->dev_private;
294 	struct mlx5_dev_ctx_shared *sh = priv->sh;
295 
296 	if (sh->dv_refcnt > 1) {
297 		int ret;
298 
299 		ret = pthread_mutex_lock(&sh->dv_mutex);
300 		MLX5_ASSERT(!ret);
301 		(void)ret;
302 	}
303 }
304 
305 static void
306 flow_dv_shared_unlock(struct rte_eth_dev *dev)
307 {
308 	struct mlx5_priv *priv = dev->data->dev_private;
309 	struct mlx5_dev_ctx_shared *sh = priv->sh;
310 
311 	if (sh->dv_refcnt > 1) {
312 		int ret;
313 
314 		ret = pthread_mutex_unlock(&sh->dv_mutex);
315 		MLX5_ASSERT(!ret);
316 		(void)ret;
317 	}
318 }
319 
320 /* Update VLAN's VID/PCP based on input rte_flow_action.
321  *
322  * @param[in] action
323  *   Pointer to struct rte_flow_action.
324  * @param[out] vlan
325  *   Pointer to struct rte_vlan_hdr.
326  */
327 static void
328 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
329 			 struct rte_vlan_hdr *vlan)
330 {
331 	uint16_t vlan_tci;
332 	if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
333 		vlan_tci =
334 		    ((const struct rte_flow_action_of_set_vlan_pcp *)
335 					       action->conf)->vlan_pcp;
336 		vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
337 		vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
338 		vlan->vlan_tci |= vlan_tci;
339 	} else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
340 		vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
341 		vlan->vlan_tci |= rte_be_to_cpu_16
342 		    (((const struct rte_flow_action_of_set_vlan_vid *)
343 					     action->conf)->vlan_vid);
344 	}
345 }
346 
347 /**
348  * Fetch 1, 2, 3 or 4 byte field from the byte array
349  * and return as unsigned integer in host-endian format.
350  *
351  * @param[in] data
352  *   Pointer to data array.
353  * @param[in] size
354  *   Size of field to extract.
355  *
356  * @return
357  *   converted field in host endian format.
358  */
359 static inline uint32_t
360 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
361 {
362 	uint32_t ret;
363 
364 	switch (size) {
365 	case 1:
366 		ret = *data;
367 		break;
368 	case 2:
369 		ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
370 		break;
371 	case 3:
372 		ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
373 		ret = (ret << 8) | *(data + sizeof(uint16_t));
374 		break;
375 	case 4:
376 		ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
377 		break;
378 	default:
379 		MLX5_ASSERT(false);
380 		ret = 0;
381 		break;
382 	}
383 	return ret;
384 }
385 
386 /**
387  * Convert modify-header action to DV specification.
388  *
389  * Data length of each action is determined by provided field description
390  * and the item mask. Data bit offset and width of each action is determined
391  * by provided item mask.
392  *
393  * @param[in] item
394  *   Pointer to item specification.
395  * @param[in] field
396  *   Pointer to field modification information.
397  *     For MLX5_MODIFICATION_TYPE_SET specifies destination field.
398  *     For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
399  *     For MLX5_MODIFICATION_TYPE_COPY specifies source field.
400  * @param[in] dcopy
401  *   Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
402  *   Negative offset value sets the same offset as source offset.
403  *   size field is ignored, value is taken from source field.
404  * @param[in,out] resource
405  *   Pointer to the modify-header resource.
406  * @param[in] type
407  *   Type of modification.
408  * @param[out] error
409  *   Pointer to the error structure.
410  *
411  * @return
412  *   0 on success, a negative errno value otherwise and rte_errno is set.
413  */
414 static int
415 flow_dv_convert_modify_action(struct rte_flow_item *item,
416 			      struct field_modify_info *field,
417 			      struct field_modify_info *dcopy,
418 			      struct mlx5_flow_dv_modify_hdr_resource *resource,
419 			      uint32_t type, struct rte_flow_error *error)
420 {
421 	uint32_t i = resource->actions_num;
422 	struct mlx5_modification_cmd *actions = resource->actions;
423 
424 	/*
425 	 * The item and mask are provided in big-endian format.
426 	 * The fields should be presented as in big-endian format either.
427 	 * Mask must be always present, it defines the actual field width.
428 	 */
429 	MLX5_ASSERT(item->mask);
430 	MLX5_ASSERT(field->size);
431 	do {
432 		unsigned int size_b;
433 		unsigned int off_b;
434 		uint32_t mask;
435 		uint32_t data;
436 
437 		if (i >= MLX5_MAX_MODIFY_NUM)
438 			return rte_flow_error_set(error, EINVAL,
439 				 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
440 				 "too many items to modify");
441 		/* Fetch variable byte size mask from the array. */
442 		mask = flow_dv_fetch_field((const uint8_t *)item->mask +
443 					   field->offset, field->size);
444 		if (!mask) {
445 			++field;
446 			continue;
447 		}
448 		/* Deduce actual data width in bits from mask value. */
449 		off_b = rte_bsf32(mask);
450 		size_b = sizeof(uint32_t) * CHAR_BIT -
451 			 off_b - __builtin_clz(mask);
452 		MLX5_ASSERT(size_b);
453 		size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
454 		actions[i] = (struct mlx5_modification_cmd) {
455 			.action_type = type,
456 			.field = field->id,
457 			.offset = off_b,
458 			.length = size_b,
459 		};
460 		/* Convert entire record to expected big-endian format. */
461 		actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
462 		if (type == MLX5_MODIFICATION_TYPE_COPY) {
463 			MLX5_ASSERT(dcopy);
464 			actions[i].dst_field = dcopy->id;
465 			actions[i].dst_offset =
466 				(int)dcopy->offset < 0 ? off_b : dcopy->offset;
467 			/* Convert entire record to big-endian format. */
468 			actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
469 		} else {
470 			MLX5_ASSERT(item->spec);
471 			data = flow_dv_fetch_field((const uint8_t *)item->spec +
472 						   field->offset, field->size);
473 			/* Shift out the trailing masked bits from data. */
474 			data = (data & mask) >> off_b;
475 			actions[i].data1 = rte_cpu_to_be_32(data);
476 		}
477 		++i;
478 		++field;
479 	} while (field->size);
480 	if (resource->actions_num == i)
481 		return rte_flow_error_set(error, EINVAL,
482 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
483 					  "invalid modification flow item");
484 	resource->actions_num = i;
485 	return 0;
486 }
487 
488 /**
489  * Convert modify-header set IPv4 address action to DV specification.
490  *
491  * @param[in,out] resource
492  *   Pointer to the modify-header resource.
493  * @param[in] action
494  *   Pointer to action specification.
495  * @param[out] error
496  *   Pointer to the error structure.
497  *
498  * @return
499  *   0 on success, a negative errno value otherwise and rte_errno is set.
500  */
501 static int
502 flow_dv_convert_action_modify_ipv4
503 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
504 			 const struct rte_flow_action *action,
505 			 struct rte_flow_error *error)
506 {
507 	const struct rte_flow_action_set_ipv4 *conf =
508 		(const struct rte_flow_action_set_ipv4 *)(action->conf);
509 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
510 	struct rte_flow_item_ipv4 ipv4;
511 	struct rte_flow_item_ipv4 ipv4_mask;
512 
513 	memset(&ipv4, 0, sizeof(ipv4));
514 	memset(&ipv4_mask, 0, sizeof(ipv4_mask));
515 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
516 		ipv4.hdr.src_addr = conf->ipv4_addr;
517 		ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
518 	} else {
519 		ipv4.hdr.dst_addr = conf->ipv4_addr;
520 		ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
521 	}
522 	item.spec = &ipv4;
523 	item.mask = &ipv4_mask;
524 	return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
525 					     MLX5_MODIFICATION_TYPE_SET, error);
526 }
527 
528 /**
529  * Convert modify-header set IPv6 address action to DV specification.
530  *
531  * @param[in,out] resource
532  *   Pointer to the modify-header resource.
533  * @param[in] action
534  *   Pointer to action specification.
535  * @param[out] error
536  *   Pointer to the error structure.
537  *
538  * @return
539  *   0 on success, a negative errno value otherwise and rte_errno is set.
540  */
541 static int
542 flow_dv_convert_action_modify_ipv6
543 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
544 			 const struct rte_flow_action *action,
545 			 struct rte_flow_error *error)
546 {
547 	const struct rte_flow_action_set_ipv6 *conf =
548 		(const struct rte_flow_action_set_ipv6 *)(action->conf);
549 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
550 	struct rte_flow_item_ipv6 ipv6;
551 	struct rte_flow_item_ipv6 ipv6_mask;
552 
553 	memset(&ipv6, 0, sizeof(ipv6));
554 	memset(&ipv6_mask, 0, sizeof(ipv6_mask));
555 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
556 		memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
557 		       sizeof(ipv6.hdr.src_addr));
558 		memcpy(&ipv6_mask.hdr.src_addr,
559 		       &rte_flow_item_ipv6_mask.hdr.src_addr,
560 		       sizeof(ipv6.hdr.src_addr));
561 	} else {
562 		memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
563 		       sizeof(ipv6.hdr.dst_addr));
564 		memcpy(&ipv6_mask.hdr.dst_addr,
565 		       &rte_flow_item_ipv6_mask.hdr.dst_addr,
566 		       sizeof(ipv6.hdr.dst_addr));
567 	}
568 	item.spec = &ipv6;
569 	item.mask = &ipv6_mask;
570 	return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
571 					     MLX5_MODIFICATION_TYPE_SET, error);
572 }
573 
574 /**
575  * Convert modify-header set MAC address action to DV specification.
576  *
577  * @param[in,out] resource
578  *   Pointer to the modify-header resource.
579  * @param[in] action
580  *   Pointer to action specification.
581  * @param[out] error
582  *   Pointer to the error structure.
583  *
584  * @return
585  *   0 on success, a negative errno value otherwise and rte_errno is set.
586  */
587 static int
588 flow_dv_convert_action_modify_mac
589 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
590 			 const struct rte_flow_action *action,
591 			 struct rte_flow_error *error)
592 {
593 	const struct rte_flow_action_set_mac *conf =
594 		(const struct rte_flow_action_set_mac *)(action->conf);
595 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
596 	struct rte_flow_item_eth eth;
597 	struct rte_flow_item_eth eth_mask;
598 
599 	memset(&eth, 0, sizeof(eth));
600 	memset(&eth_mask, 0, sizeof(eth_mask));
601 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
602 		memcpy(&eth.src.addr_bytes, &conf->mac_addr,
603 		       sizeof(eth.src.addr_bytes));
604 		memcpy(&eth_mask.src.addr_bytes,
605 		       &rte_flow_item_eth_mask.src.addr_bytes,
606 		       sizeof(eth_mask.src.addr_bytes));
607 	} else {
608 		memcpy(&eth.dst.addr_bytes, &conf->mac_addr,
609 		       sizeof(eth.dst.addr_bytes));
610 		memcpy(&eth_mask.dst.addr_bytes,
611 		       &rte_flow_item_eth_mask.dst.addr_bytes,
612 		       sizeof(eth_mask.dst.addr_bytes));
613 	}
614 	item.spec = &eth;
615 	item.mask = &eth_mask;
616 	return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
617 					     MLX5_MODIFICATION_TYPE_SET, error);
618 }
619 
620 /**
621  * Convert modify-header set VLAN VID action to DV specification.
622  *
623  * @param[in,out] resource
624  *   Pointer to the modify-header resource.
625  * @param[in] action
626  *   Pointer to action specification.
627  * @param[out] error
628  *   Pointer to the error structure.
629  *
630  * @return
631  *   0 on success, a negative errno value otherwise and rte_errno is set.
632  */
633 static int
634 flow_dv_convert_action_modify_vlan_vid
635 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
636 			 const struct rte_flow_action *action,
637 			 struct rte_flow_error *error)
638 {
639 	const struct rte_flow_action_of_set_vlan_vid *conf =
640 		(const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
641 	int i = resource->actions_num;
642 	struct mlx5_modification_cmd *actions = resource->actions;
643 	struct field_modify_info *field = modify_vlan_out_first_vid;
644 
645 	if (i >= MLX5_MAX_MODIFY_NUM)
646 		return rte_flow_error_set(error, EINVAL,
647 			 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
648 			 "too many items to modify");
649 	actions[i] = (struct mlx5_modification_cmd) {
650 		.action_type = MLX5_MODIFICATION_TYPE_SET,
651 		.field = field->id,
652 		.length = field->size,
653 		.offset = field->offset,
654 	};
655 	actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
656 	actions[i].data1 = conf->vlan_vid;
657 	actions[i].data1 = actions[i].data1 << 16;
658 	resource->actions_num = ++i;
659 	return 0;
660 }
661 
662 /**
663  * Convert modify-header set TP action to DV specification.
664  *
665  * @param[in,out] resource
666  *   Pointer to the modify-header resource.
667  * @param[in] action
668  *   Pointer to action specification.
669  * @param[in] items
670  *   Pointer to rte_flow_item objects list.
671  * @param[in] attr
672  *   Pointer to flow attributes structure.
673  * @param[in] dev_flow
674  *   Pointer to the sub flow.
675  * @param[in] tunnel_decap
676  *   Whether action is after tunnel decapsulation.
677  * @param[out] error
678  *   Pointer to the error structure.
679  *
680  * @return
681  *   0 on success, a negative errno value otherwise and rte_errno is set.
682  */
683 static int
684 flow_dv_convert_action_modify_tp
685 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
686 			 const struct rte_flow_action *action,
687 			 const struct rte_flow_item *items,
688 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
689 			 bool tunnel_decap, struct rte_flow_error *error)
690 {
691 	const struct rte_flow_action_set_tp *conf =
692 		(const struct rte_flow_action_set_tp *)(action->conf);
693 	struct rte_flow_item item;
694 	struct rte_flow_item_udp udp;
695 	struct rte_flow_item_udp udp_mask;
696 	struct rte_flow_item_tcp tcp;
697 	struct rte_flow_item_tcp tcp_mask;
698 	struct field_modify_info *field;
699 
700 	if (!attr->valid)
701 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
702 	if (attr->udp) {
703 		memset(&udp, 0, sizeof(udp));
704 		memset(&udp_mask, 0, sizeof(udp_mask));
705 		if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
706 			udp.hdr.src_port = conf->port;
707 			udp_mask.hdr.src_port =
708 					rte_flow_item_udp_mask.hdr.src_port;
709 		} else {
710 			udp.hdr.dst_port = conf->port;
711 			udp_mask.hdr.dst_port =
712 					rte_flow_item_udp_mask.hdr.dst_port;
713 		}
714 		item.type = RTE_FLOW_ITEM_TYPE_UDP;
715 		item.spec = &udp;
716 		item.mask = &udp_mask;
717 		field = modify_udp;
718 	} else {
719 		MLX5_ASSERT(attr->tcp);
720 		memset(&tcp, 0, sizeof(tcp));
721 		memset(&tcp_mask, 0, sizeof(tcp_mask));
722 		if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
723 			tcp.hdr.src_port = conf->port;
724 			tcp_mask.hdr.src_port =
725 					rte_flow_item_tcp_mask.hdr.src_port;
726 		} else {
727 			tcp.hdr.dst_port = conf->port;
728 			tcp_mask.hdr.dst_port =
729 					rte_flow_item_tcp_mask.hdr.dst_port;
730 		}
731 		item.type = RTE_FLOW_ITEM_TYPE_TCP;
732 		item.spec = &tcp;
733 		item.mask = &tcp_mask;
734 		field = modify_tcp;
735 	}
736 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
737 					     MLX5_MODIFICATION_TYPE_SET, error);
738 }
739 
740 /**
741  * Convert modify-header set TTL action to DV specification.
742  *
743  * @param[in,out] resource
744  *   Pointer to the modify-header resource.
745  * @param[in] action
746  *   Pointer to action specification.
747  * @param[in] items
748  *   Pointer to rte_flow_item objects list.
749  * @param[in] attr
750  *   Pointer to flow attributes structure.
751  * @param[in] dev_flow
752  *   Pointer to the sub flow.
753  * @param[in] tunnel_decap
754  *   Whether action is after tunnel decapsulation.
755  * @param[out] error
756  *   Pointer to the error structure.
757  *
758  * @return
759  *   0 on success, a negative errno value otherwise and rte_errno is set.
760  */
761 static int
762 flow_dv_convert_action_modify_ttl
763 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
764 			 const struct rte_flow_action *action,
765 			 const struct rte_flow_item *items,
766 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
767 			 bool tunnel_decap, struct rte_flow_error *error)
768 {
769 	const struct rte_flow_action_set_ttl *conf =
770 		(const struct rte_flow_action_set_ttl *)(action->conf);
771 	struct rte_flow_item item;
772 	struct rte_flow_item_ipv4 ipv4;
773 	struct rte_flow_item_ipv4 ipv4_mask;
774 	struct rte_flow_item_ipv6 ipv6;
775 	struct rte_flow_item_ipv6 ipv6_mask;
776 	struct field_modify_info *field;
777 
778 	if (!attr->valid)
779 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
780 	if (attr->ipv4) {
781 		memset(&ipv4, 0, sizeof(ipv4));
782 		memset(&ipv4_mask, 0, sizeof(ipv4_mask));
783 		ipv4.hdr.time_to_live = conf->ttl_value;
784 		ipv4_mask.hdr.time_to_live = 0xFF;
785 		item.type = RTE_FLOW_ITEM_TYPE_IPV4;
786 		item.spec = &ipv4;
787 		item.mask = &ipv4_mask;
788 		field = modify_ipv4;
789 	} else {
790 		MLX5_ASSERT(attr->ipv6);
791 		memset(&ipv6, 0, sizeof(ipv6));
792 		memset(&ipv6_mask, 0, sizeof(ipv6_mask));
793 		ipv6.hdr.hop_limits = conf->ttl_value;
794 		ipv6_mask.hdr.hop_limits = 0xFF;
795 		item.type = RTE_FLOW_ITEM_TYPE_IPV6;
796 		item.spec = &ipv6;
797 		item.mask = &ipv6_mask;
798 		field = modify_ipv6;
799 	}
800 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
801 					     MLX5_MODIFICATION_TYPE_SET, error);
802 }
803 
804 /**
805  * Convert modify-header decrement TTL action to DV specification.
806  *
807  * @param[in,out] resource
808  *   Pointer to the modify-header resource.
809  * @param[in] action
810  *   Pointer to action specification.
811  * @param[in] items
812  *   Pointer to rte_flow_item objects list.
813  * @param[in] attr
814  *   Pointer to flow attributes structure.
815  * @param[in] dev_flow
816  *   Pointer to the sub flow.
817  * @param[in] tunnel_decap
818  *   Whether action is after tunnel decapsulation.
819  * @param[out] error
820  *   Pointer to the error structure.
821  *
822  * @return
823  *   0 on success, a negative errno value otherwise and rte_errno is set.
824  */
825 static int
826 flow_dv_convert_action_modify_dec_ttl
827 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
828 			 const struct rte_flow_item *items,
829 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
830 			 bool tunnel_decap, struct rte_flow_error *error)
831 {
832 	struct rte_flow_item item;
833 	struct rte_flow_item_ipv4 ipv4;
834 	struct rte_flow_item_ipv4 ipv4_mask;
835 	struct rte_flow_item_ipv6 ipv6;
836 	struct rte_flow_item_ipv6 ipv6_mask;
837 	struct field_modify_info *field;
838 
839 	if (!attr->valid)
840 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
841 	if (attr->ipv4) {
842 		memset(&ipv4, 0, sizeof(ipv4));
843 		memset(&ipv4_mask, 0, sizeof(ipv4_mask));
844 		ipv4.hdr.time_to_live = 0xFF;
845 		ipv4_mask.hdr.time_to_live = 0xFF;
846 		item.type = RTE_FLOW_ITEM_TYPE_IPV4;
847 		item.spec = &ipv4;
848 		item.mask = &ipv4_mask;
849 		field = modify_ipv4;
850 	} else {
851 		MLX5_ASSERT(attr->ipv6);
852 		memset(&ipv6, 0, sizeof(ipv6));
853 		memset(&ipv6_mask, 0, sizeof(ipv6_mask));
854 		ipv6.hdr.hop_limits = 0xFF;
855 		ipv6_mask.hdr.hop_limits = 0xFF;
856 		item.type = RTE_FLOW_ITEM_TYPE_IPV6;
857 		item.spec = &ipv6;
858 		item.mask = &ipv6_mask;
859 		field = modify_ipv6;
860 	}
861 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
862 					     MLX5_MODIFICATION_TYPE_ADD, error);
863 }
864 
865 /**
866  * Convert modify-header increment/decrement TCP Sequence number
867  * to DV specification.
868  *
869  * @param[in,out] resource
870  *   Pointer to the modify-header resource.
871  * @param[in] action
872  *   Pointer to action specification.
873  * @param[out] error
874  *   Pointer to the error structure.
875  *
876  * @return
877  *   0 on success, a negative errno value otherwise and rte_errno is set.
878  */
879 static int
880 flow_dv_convert_action_modify_tcp_seq
881 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
882 			 const struct rte_flow_action *action,
883 			 struct rte_flow_error *error)
884 {
885 	const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
886 	uint64_t value = rte_be_to_cpu_32(*conf);
887 	struct rte_flow_item item;
888 	struct rte_flow_item_tcp tcp;
889 	struct rte_flow_item_tcp tcp_mask;
890 
891 	memset(&tcp, 0, sizeof(tcp));
892 	memset(&tcp_mask, 0, sizeof(tcp_mask));
893 	if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
894 		/*
895 		 * The HW has no decrement operation, only increment operation.
896 		 * To simulate decrement X from Y using increment operation
897 		 * we need to add UINT32_MAX X times to Y.
898 		 * Each adding of UINT32_MAX decrements Y by 1.
899 		 */
900 		value *= UINT32_MAX;
901 	tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
902 	tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
903 	item.type = RTE_FLOW_ITEM_TYPE_TCP;
904 	item.spec = &tcp;
905 	item.mask = &tcp_mask;
906 	return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
907 					     MLX5_MODIFICATION_TYPE_ADD, error);
908 }
909 
910 /**
911  * Convert modify-header increment/decrement TCP Acknowledgment number
912  * to DV specification.
913  *
914  * @param[in,out] resource
915  *   Pointer to the modify-header resource.
916  * @param[in] action
917  *   Pointer to action specification.
918  * @param[out] error
919  *   Pointer to the error structure.
920  *
921  * @return
922  *   0 on success, a negative errno value otherwise and rte_errno is set.
923  */
924 static int
925 flow_dv_convert_action_modify_tcp_ack
926 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
927 			 const struct rte_flow_action *action,
928 			 struct rte_flow_error *error)
929 {
930 	const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
931 	uint64_t value = rte_be_to_cpu_32(*conf);
932 	struct rte_flow_item item;
933 	struct rte_flow_item_tcp tcp;
934 	struct rte_flow_item_tcp tcp_mask;
935 
936 	memset(&tcp, 0, sizeof(tcp));
937 	memset(&tcp_mask, 0, sizeof(tcp_mask));
938 	if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
939 		/*
940 		 * The HW has no decrement operation, only increment operation.
941 		 * To simulate decrement X from Y using increment operation
942 		 * we need to add UINT32_MAX X times to Y.
943 		 * Each adding of UINT32_MAX decrements Y by 1.
944 		 */
945 		value *= UINT32_MAX;
946 	tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
947 	tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
948 	item.type = RTE_FLOW_ITEM_TYPE_TCP;
949 	item.spec = &tcp;
950 	item.mask = &tcp_mask;
951 	return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
952 					     MLX5_MODIFICATION_TYPE_ADD, error);
953 }
954 
955 static enum mlx5_modification_field reg_to_field[] = {
956 	[REG_NON] = MLX5_MODI_OUT_NONE,
957 	[REG_A] = MLX5_MODI_META_DATA_REG_A,
958 	[REG_B] = MLX5_MODI_META_DATA_REG_B,
959 	[REG_C_0] = MLX5_MODI_META_REG_C_0,
960 	[REG_C_1] = MLX5_MODI_META_REG_C_1,
961 	[REG_C_2] = MLX5_MODI_META_REG_C_2,
962 	[REG_C_3] = MLX5_MODI_META_REG_C_3,
963 	[REG_C_4] = MLX5_MODI_META_REG_C_4,
964 	[REG_C_5] = MLX5_MODI_META_REG_C_5,
965 	[REG_C_6] = MLX5_MODI_META_REG_C_6,
966 	[REG_C_7] = MLX5_MODI_META_REG_C_7,
967 };
968 
969 /**
970  * Convert register set to DV specification.
971  *
972  * @param[in,out] resource
973  *   Pointer to the modify-header resource.
974  * @param[in] action
975  *   Pointer to action specification.
976  * @param[out] error
977  *   Pointer to the error structure.
978  *
979  * @return
980  *   0 on success, a negative errno value otherwise and rte_errno is set.
981  */
982 static int
983 flow_dv_convert_action_set_reg
984 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
985 			 const struct rte_flow_action *action,
986 			 struct rte_flow_error *error)
987 {
988 	const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
989 	struct mlx5_modification_cmd *actions = resource->actions;
990 	uint32_t i = resource->actions_num;
991 
992 	if (i >= MLX5_MAX_MODIFY_NUM)
993 		return rte_flow_error_set(error, EINVAL,
994 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
995 					  "too many items to modify");
996 	MLX5_ASSERT(conf->id != REG_NON);
997 	MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
998 	actions[i] = (struct mlx5_modification_cmd) {
999 		.action_type = MLX5_MODIFICATION_TYPE_SET,
1000 		.field = reg_to_field[conf->id],
1001 	};
1002 	actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
1003 	actions[i].data1 = rte_cpu_to_be_32(conf->data);
1004 	++i;
1005 	resource->actions_num = i;
1006 	return 0;
1007 }
1008 
1009 /**
1010  * Convert SET_TAG action to DV specification.
1011  *
1012  * @param[in] dev
1013  *   Pointer to the rte_eth_dev structure.
1014  * @param[in,out] resource
1015  *   Pointer to the modify-header resource.
1016  * @param[in] conf
1017  *   Pointer to action specification.
1018  * @param[out] error
1019  *   Pointer to the error structure.
1020  *
1021  * @return
1022  *   0 on success, a negative errno value otherwise and rte_errno is set.
1023  */
1024 static int
1025 flow_dv_convert_action_set_tag
1026 			(struct rte_eth_dev *dev,
1027 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
1028 			 const struct rte_flow_action_set_tag *conf,
1029 			 struct rte_flow_error *error)
1030 {
1031 	rte_be32_t data = rte_cpu_to_be_32(conf->data);
1032 	rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1033 	struct rte_flow_item item = {
1034 		.spec = &data,
1035 		.mask = &mask,
1036 	};
1037 	struct field_modify_info reg_c_x[] = {
1038 		[1] = {0, 0, 0},
1039 	};
1040 	enum mlx5_modification_field reg_type;
1041 	int ret;
1042 
1043 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1044 	if (ret < 0)
1045 		return ret;
1046 	MLX5_ASSERT(ret != REG_NON);
1047 	MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1048 	reg_type = reg_to_field[ret];
1049 	MLX5_ASSERT(reg_type > 0);
1050 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1051 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1052 					     MLX5_MODIFICATION_TYPE_SET, error);
1053 }
1054 
1055 /**
1056  * Convert internal COPY_REG action to DV specification.
1057  *
1058  * @param[in] dev
1059  *   Pointer to the rte_eth_dev structure.
1060  * @param[in,out] res
1061  *   Pointer to the modify-header resource.
1062  * @param[in] action
1063  *   Pointer to action specification.
1064  * @param[out] error
1065  *   Pointer to the error structure.
1066  *
1067  * @return
1068  *   0 on success, a negative errno value otherwise and rte_errno is set.
1069  */
1070 static int
1071 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1072 				 struct mlx5_flow_dv_modify_hdr_resource *res,
1073 				 const struct rte_flow_action *action,
1074 				 struct rte_flow_error *error)
1075 {
1076 	const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1077 	rte_be32_t mask = RTE_BE32(UINT32_MAX);
1078 	struct rte_flow_item item = {
1079 		.spec = NULL,
1080 		.mask = &mask,
1081 	};
1082 	struct field_modify_info reg_src[] = {
1083 		{4, 0, reg_to_field[conf->src]},
1084 		{0, 0, 0},
1085 	};
1086 	struct field_modify_info reg_dst = {
1087 		.offset = 0,
1088 		.id = reg_to_field[conf->dst],
1089 	};
1090 	/* Adjust reg_c[0] usage according to reported mask. */
1091 	if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1092 		struct mlx5_priv *priv = dev->data->dev_private;
1093 		uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1094 
1095 		MLX5_ASSERT(reg_c0);
1096 		MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1097 		if (conf->dst == REG_C_0) {
1098 			/* Copy to reg_c[0], within mask only. */
1099 			reg_dst.offset = rte_bsf32(reg_c0);
1100 			/*
1101 			 * Mask is ignoring the enianness, because
1102 			 * there is no conversion in datapath.
1103 			 */
1104 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1105 			/* Copy from destination lower bits to reg_c[0]. */
1106 			mask = reg_c0 >> reg_dst.offset;
1107 #else
1108 			/* Copy from destination upper bits to reg_c[0]. */
1109 			mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1110 					  rte_fls_u32(reg_c0));
1111 #endif
1112 		} else {
1113 			mask = rte_cpu_to_be_32(reg_c0);
1114 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1115 			/* Copy from reg_c[0] to destination lower bits. */
1116 			reg_dst.offset = 0;
1117 #else
1118 			/* Copy from reg_c[0] to destination upper bits. */
1119 			reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1120 					 (rte_fls_u32(reg_c0) -
1121 					  rte_bsf32(reg_c0));
1122 #endif
1123 		}
1124 	}
1125 	return flow_dv_convert_modify_action(&item,
1126 					     reg_src, &reg_dst, res,
1127 					     MLX5_MODIFICATION_TYPE_COPY,
1128 					     error);
1129 }
1130 
1131 /**
1132  * Convert MARK action to DV specification. This routine is used
1133  * in extensive metadata only and requires metadata register to be
1134  * handled. In legacy mode hardware tag resource is engaged.
1135  *
1136  * @param[in] dev
1137  *   Pointer to the rte_eth_dev structure.
1138  * @param[in] conf
1139  *   Pointer to MARK action specification.
1140  * @param[in,out] resource
1141  *   Pointer to the modify-header resource.
1142  * @param[out] error
1143  *   Pointer to the error structure.
1144  *
1145  * @return
1146  *   0 on success, a negative errno value otherwise and rte_errno is set.
1147  */
1148 static int
1149 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1150 			    const struct rte_flow_action_mark *conf,
1151 			    struct mlx5_flow_dv_modify_hdr_resource *resource,
1152 			    struct rte_flow_error *error)
1153 {
1154 	struct mlx5_priv *priv = dev->data->dev_private;
1155 	rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1156 					   priv->sh->dv_mark_mask);
1157 	rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1158 	struct rte_flow_item item = {
1159 		.spec = &data,
1160 		.mask = &mask,
1161 	};
1162 	struct field_modify_info reg_c_x[] = {
1163 		[1] = {0, 0, 0},
1164 	};
1165 	int reg;
1166 
1167 	if (!mask)
1168 		return rte_flow_error_set(error, EINVAL,
1169 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 					  NULL, "zero mark action mask");
1171 	reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1172 	if (reg < 0)
1173 		return reg;
1174 	MLX5_ASSERT(reg > 0);
1175 	if (reg == REG_C_0) {
1176 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1177 		uint32_t shl_c0 = rte_bsf32(msk_c0);
1178 
1179 		data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1180 		mask = rte_cpu_to_be_32(mask) & msk_c0;
1181 		mask = rte_cpu_to_be_32(mask << shl_c0);
1182 	}
1183 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1184 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1185 					     MLX5_MODIFICATION_TYPE_SET, error);
1186 }
1187 
1188 /**
1189  * Get metadata register index for specified steering domain.
1190  *
1191  * @param[in] dev
1192  *   Pointer to the rte_eth_dev structure.
1193  * @param[in] attr
1194  *   Attributes of flow to determine steering domain.
1195  * @param[out] error
1196  *   Pointer to the error structure.
1197  *
1198  * @return
1199  *   positive index on success, a negative errno value otherwise
1200  *   and rte_errno is set.
1201  */
1202 static enum modify_reg
1203 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1204 			 const struct rte_flow_attr *attr,
1205 			 struct rte_flow_error *error)
1206 {
1207 	int reg =
1208 		mlx5_flow_get_reg_id(dev, attr->transfer ?
1209 					  MLX5_METADATA_FDB :
1210 					    attr->egress ?
1211 					    MLX5_METADATA_TX :
1212 					    MLX5_METADATA_RX, 0, error);
1213 	if (reg < 0)
1214 		return rte_flow_error_set(error,
1215 					  ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1216 					  NULL, "unavailable "
1217 					  "metadata register");
1218 	return reg;
1219 }
1220 
1221 /**
1222  * Convert SET_META action to DV specification.
1223  *
1224  * @param[in] dev
1225  *   Pointer to the rte_eth_dev structure.
1226  * @param[in,out] resource
1227  *   Pointer to the modify-header resource.
1228  * @param[in] attr
1229  *   Attributes of flow that includes this item.
1230  * @param[in] conf
1231  *   Pointer to action specification.
1232  * @param[out] error
1233  *   Pointer to the error structure.
1234  *
1235  * @return
1236  *   0 on success, a negative errno value otherwise and rte_errno is set.
1237  */
1238 static int
1239 flow_dv_convert_action_set_meta
1240 			(struct rte_eth_dev *dev,
1241 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
1242 			 const struct rte_flow_attr *attr,
1243 			 const struct rte_flow_action_set_meta *conf,
1244 			 struct rte_flow_error *error)
1245 {
1246 	uint32_t data = conf->data;
1247 	uint32_t mask = conf->mask;
1248 	struct rte_flow_item item = {
1249 		.spec = &data,
1250 		.mask = &mask,
1251 	};
1252 	struct field_modify_info reg_c_x[] = {
1253 		[1] = {0, 0, 0},
1254 	};
1255 	int reg = flow_dv_get_metadata_reg(dev, attr, error);
1256 
1257 	if (reg < 0)
1258 		return reg;
1259 	/*
1260 	 * In datapath code there is no endianness
1261 	 * coversions for perfromance reasons, all
1262 	 * pattern conversions are done in rte_flow.
1263 	 */
1264 	if (reg == REG_C_0) {
1265 		struct mlx5_priv *priv = dev->data->dev_private;
1266 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1267 		uint32_t shl_c0;
1268 
1269 		MLX5_ASSERT(msk_c0);
1270 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1271 		shl_c0 = rte_bsf32(msk_c0);
1272 #else
1273 		shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1274 #endif
1275 		mask <<= shl_c0;
1276 		data <<= shl_c0;
1277 		MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1278 	}
1279 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1280 	/* The routine expects parameters in memory as big-endian ones. */
1281 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1282 					     MLX5_MODIFICATION_TYPE_SET, error);
1283 }
1284 
1285 /**
1286  * Convert modify-header set IPv4 DSCP action to DV specification.
1287  *
1288  * @param[in,out] resource
1289  *   Pointer to the modify-header resource.
1290  * @param[in] action
1291  *   Pointer to action specification.
1292  * @param[out] error
1293  *   Pointer to the error structure.
1294  *
1295  * @return
1296  *   0 on success, a negative errno value otherwise and rte_errno is set.
1297  */
1298 static int
1299 flow_dv_convert_action_modify_ipv4_dscp
1300 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
1301 			 const struct rte_flow_action *action,
1302 			 struct rte_flow_error *error)
1303 {
1304 	const struct rte_flow_action_set_dscp *conf =
1305 		(const struct rte_flow_action_set_dscp *)(action->conf);
1306 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1307 	struct rte_flow_item_ipv4 ipv4;
1308 	struct rte_flow_item_ipv4 ipv4_mask;
1309 
1310 	memset(&ipv4, 0, sizeof(ipv4));
1311 	memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1312 	ipv4.hdr.type_of_service = conf->dscp;
1313 	ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1314 	item.spec = &ipv4;
1315 	item.mask = &ipv4_mask;
1316 	return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1317 					     MLX5_MODIFICATION_TYPE_SET, error);
1318 }
1319 
1320 /**
1321  * Convert modify-header set IPv6 DSCP action to DV specification.
1322  *
1323  * @param[in,out] resource
1324  *   Pointer to the modify-header resource.
1325  * @param[in] action
1326  *   Pointer to action specification.
1327  * @param[out] error
1328  *   Pointer to the error structure.
1329  *
1330  * @return
1331  *   0 on success, a negative errno value otherwise and rte_errno is set.
1332  */
1333 static int
1334 flow_dv_convert_action_modify_ipv6_dscp
1335 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
1336 			 const struct rte_flow_action *action,
1337 			 struct rte_flow_error *error)
1338 {
1339 	const struct rte_flow_action_set_dscp *conf =
1340 		(const struct rte_flow_action_set_dscp *)(action->conf);
1341 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1342 	struct rte_flow_item_ipv6 ipv6;
1343 	struct rte_flow_item_ipv6 ipv6_mask;
1344 
1345 	memset(&ipv6, 0, sizeof(ipv6));
1346 	memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1347 	/*
1348 	 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1349 	 * rdma-core only accept the DSCP bits byte aligned start from
1350 	 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1351 	 * bits in IPv6 case as rdma-core requires byte aligned value.
1352 	 */
1353 	ipv6.hdr.vtc_flow = conf->dscp;
1354 	ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1355 	item.spec = &ipv6;
1356 	item.mask = &ipv6_mask;
1357 	return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1358 					     MLX5_MODIFICATION_TYPE_SET, error);
1359 }
1360 
1361 /**
1362  * Validate MARK item.
1363  *
1364  * @param[in] dev
1365  *   Pointer to the rte_eth_dev structure.
1366  * @param[in] item
1367  *   Item specification.
1368  * @param[in] attr
1369  *   Attributes of flow that includes this item.
1370  * @param[out] error
1371  *   Pointer to error structure.
1372  *
1373  * @return
1374  *   0 on success, a negative errno value otherwise and rte_errno is set.
1375  */
1376 static int
1377 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1378 			   const struct rte_flow_item *item,
1379 			   const struct rte_flow_attr *attr __rte_unused,
1380 			   struct rte_flow_error *error)
1381 {
1382 	struct mlx5_priv *priv = dev->data->dev_private;
1383 	struct mlx5_dev_config *config = &priv->config;
1384 	const struct rte_flow_item_mark *spec = item->spec;
1385 	const struct rte_flow_item_mark *mask = item->mask;
1386 	const struct rte_flow_item_mark nic_mask = {
1387 		.id = priv->sh->dv_mark_mask,
1388 	};
1389 	int ret;
1390 
1391 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1392 		return rte_flow_error_set(error, ENOTSUP,
1393 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1394 					  "extended metadata feature"
1395 					  " isn't enabled");
1396 	if (!mlx5_flow_ext_mreg_supported(dev))
1397 		return rte_flow_error_set(error, ENOTSUP,
1398 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1399 					  "extended metadata register"
1400 					  " isn't supported");
1401 	if (!nic_mask.id)
1402 		return rte_flow_error_set(error, ENOTSUP,
1403 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1404 					  "extended metadata register"
1405 					  " isn't available");
1406 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1407 	if (ret < 0)
1408 		return ret;
1409 	if (!spec)
1410 		return rte_flow_error_set(error, EINVAL,
1411 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1412 					  item->spec,
1413 					  "data cannot be empty");
1414 	if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1415 		return rte_flow_error_set(error, EINVAL,
1416 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1417 					  &spec->id,
1418 					  "mark id exceeds the limit");
1419 	if (!mask)
1420 		mask = &nic_mask;
1421 	if (!mask->id)
1422 		return rte_flow_error_set(error, EINVAL,
1423 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1424 					"mask cannot be zero");
1425 
1426 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1427 					(const uint8_t *)&nic_mask,
1428 					sizeof(struct rte_flow_item_mark),
1429 					error);
1430 	if (ret < 0)
1431 		return ret;
1432 	return 0;
1433 }
1434 
1435 /**
1436  * Validate META item.
1437  *
1438  * @param[in] dev
1439  *   Pointer to the rte_eth_dev structure.
1440  * @param[in] item
1441  *   Item specification.
1442  * @param[in] attr
1443  *   Attributes of flow that includes this item.
1444  * @param[out] error
1445  *   Pointer to error structure.
1446  *
1447  * @return
1448  *   0 on success, a negative errno value otherwise and rte_errno is set.
1449  */
1450 static int
1451 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1452 			   const struct rte_flow_item *item,
1453 			   const struct rte_flow_attr *attr,
1454 			   struct rte_flow_error *error)
1455 {
1456 	struct mlx5_priv *priv = dev->data->dev_private;
1457 	struct mlx5_dev_config *config = &priv->config;
1458 	const struct rte_flow_item_meta *spec = item->spec;
1459 	const struct rte_flow_item_meta *mask = item->mask;
1460 	struct rte_flow_item_meta nic_mask = {
1461 		.data = UINT32_MAX
1462 	};
1463 	int reg;
1464 	int ret;
1465 
1466 	if (!spec)
1467 		return rte_flow_error_set(error, EINVAL,
1468 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1469 					  item->spec,
1470 					  "data cannot be empty");
1471 	if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1472 		if (!mlx5_flow_ext_mreg_supported(dev))
1473 			return rte_flow_error_set(error, ENOTSUP,
1474 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1475 					  "extended metadata register"
1476 					  " isn't supported");
1477 		reg = flow_dv_get_metadata_reg(dev, attr, error);
1478 		if (reg < 0)
1479 			return reg;
1480 		if (reg == REG_B)
1481 			return rte_flow_error_set(error, ENOTSUP,
1482 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 					  "match on reg_b "
1484 					  "isn't supported");
1485 		if (reg != REG_A)
1486 			nic_mask.data = priv->sh->dv_meta_mask;
1487 	} else if (attr->transfer) {
1488 		return rte_flow_error_set(error, ENOTSUP,
1489 					RTE_FLOW_ERROR_TYPE_ITEM, item,
1490 					"extended metadata feature "
1491 					"should be enabled when "
1492 					"meta item is requested "
1493 					"with e-switch mode ");
1494 	}
1495 	if (!mask)
1496 		mask = &rte_flow_item_meta_mask;
1497 	if (!mask->data)
1498 		return rte_flow_error_set(error, EINVAL,
1499 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1500 					"mask cannot be zero");
1501 
1502 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1503 					(const uint8_t *)&nic_mask,
1504 					sizeof(struct rte_flow_item_meta),
1505 					error);
1506 	return ret;
1507 }
1508 
1509 /**
1510  * Validate TAG item.
1511  *
1512  * @param[in] dev
1513  *   Pointer to the rte_eth_dev structure.
1514  * @param[in] item
1515  *   Item specification.
1516  * @param[in] attr
1517  *   Attributes of flow that includes this item.
1518  * @param[out] error
1519  *   Pointer to error structure.
1520  *
1521  * @return
1522  *   0 on success, a negative errno value otherwise and rte_errno is set.
1523  */
1524 static int
1525 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1526 			  const struct rte_flow_item *item,
1527 			  const struct rte_flow_attr *attr __rte_unused,
1528 			  struct rte_flow_error *error)
1529 {
1530 	const struct rte_flow_item_tag *spec = item->spec;
1531 	const struct rte_flow_item_tag *mask = item->mask;
1532 	const struct rte_flow_item_tag nic_mask = {
1533 		.data = RTE_BE32(UINT32_MAX),
1534 		.index = 0xff,
1535 	};
1536 	int ret;
1537 
1538 	if (!mlx5_flow_ext_mreg_supported(dev))
1539 		return rte_flow_error_set(error, ENOTSUP,
1540 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1541 					  "extensive metadata register"
1542 					  " isn't supported");
1543 	if (!spec)
1544 		return rte_flow_error_set(error, EINVAL,
1545 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1546 					  item->spec,
1547 					  "data cannot be empty");
1548 	if (!mask)
1549 		mask = &rte_flow_item_tag_mask;
1550 	if (!mask->data)
1551 		return rte_flow_error_set(error, EINVAL,
1552 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1553 					"mask cannot be zero");
1554 
1555 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1556 					(const uint8_t *)&nic_mask,
1557 					sizeof(struct rte_flow_item_tag),
1558 					error);
1559 	if (ret < 0)
1560 		return ret;
1561 	if (mask->index != 0xff)
1562 		return rte_flow_error_set(error, EINVAL,
1563 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1564 					  "partial mask for tag index"
1565 					  " is not supported");
1566 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1567 	if (ret < 0)
1568 		return ret;
1569 	MLX5_ASSERT(ret != REG_NON);
1570 	return 0;
1571 }
1572 
1573 /**
1574  * Validate vport item.
1575  *
1576  * @param[in] dev
1577  *   Pointer to the rte_eth_dev structure.
1578  * @param[in] item
1579  *   Item specification.
1580  * @param[in] attr
1581  *   Attributes of flow that includes this item.
1582  * @param[in] item_flags
1583  *   Bit-fields that holds the items detected until now.
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 static int
1591 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1592 			      const struct rte_flow_item *item,
1593 			      const struct rte_flow_attr *attr,
1594 			      uint64_t item_flags,
1595 			      struct rte_flow_error *error)
1596 {
1597 	const struct rte_flow_item_port_id *spec = item->spec;
1598 	const struct rte_flow_item_port_id *mask = item->mask;
1599 	const struct rte_flow_item_port_id switch_mask = {
1600 			.id = 0xffffffff,
1601 	};
1602 	struct mlx5_priv *esw_priv;
1603 	struct mlx5_priv *dev_priv;
1604 	int ret;
1605 
1606 	if (!attr->transfer)
1607 		return rte_flow_error_set(error, EINVAL,
1608 					  RTE_FLOW_ERROR_TYPE_ITEM,
1609 					  NULL,
1610 					  "match on port id is valid only"
1611 					  " when transfer flag is enabled");
1612 	if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1613 		return rte_flow_error_set(error, ENOTSUP,
1614 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1615 					  "multiple source ports are not"
1616 					  " supported");
1617 	if (!mask)
1618 		mask = &switch_mask;
1619 	if (mask->id != 0xffffffff)
1620 		return rte_flow_error_set(error, ENOTSUP,
1621 					   RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1622 					   mask,
1623 					   "no support for partial mask on"
1624 					   " \"id\" field");
1625 	ret = mlx5_flow_item_acceptable
1626 				(item, (const uint8_t *)mask,
1627 				 (const uint8_t *)&rte_flow_item_port_id_mask,
1628 				 sizeof(struct rte_flow_item_port_id),
1629 				 error);
1630 	if (ret)
1631 		return ret;
1632 	if (!spec)
1633 		return 0;
1634 	esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1635 	if (!esw_priv)
1636 		return rte_flow_error_set(error, rte_errno,
1637 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1638 					  "failed to obtain E-Switch info for"
1639 					  " port");
1640 	dev_priv = mlx5_dev_to_eswitch_info(dev);
1641 	if (!dev_priv)
1642 		return rte_flow_error_set(error, rte_errno,
1643 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1644 					  NULL,
1645 					  "failed to obtain E-Switch info");
1646 	if (esw_priv->domain_id != dev_priv->domain_id)
1647 		return rte_flow_error_set(error, EINVAL,
1648 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1649 					  "cannot match on a port from a"
1650 					  " different E-Switch");
1651 	return 0;
1652 }
1653 
1654 /**
1655  * Validate VLAN item.
1656  *
1657  * @param[in] item
1658  *   Item specification.
1659  * @param[in] item_flags
1660  *   Bit-fields that holds the items detected until now.
1661  * @param[in] dev
1662  *   Ethernet device flow is being created on.
1663  * @param[out] error
1664  *   Pointer to error structure.
1665  *
1666  * @return
1667  *   0 on success, a negative errno value otherwise and rte_errno is set.
1668  */
1669 static int
1670 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1671 			   uint64_t item_flags,
1672 			   struct rte_eth_dev *dev,
1673 			   struct rte_flow_error *error)
1674 {
1675 	const struct rte_flow_item_vlan *mask = item->mask;
1676 	const struct rte_flow_item_vlan nic_mask = {
1677 		.tci = RTE_BE16(UINT16_MAX),
1678 		.inner_type = RTE_BE16(UINT16_MAX),
1679 	};
1680 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1681 	int ret;
1682 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1683 					MLX5_FLOW_LAYER_INNER_L4) :
1684 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1685 					MLX5_FLOW_LAYER_OUTER_L4);
1686 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1687 					MLX5_FLOW_LAYER_OUTER_VLAN;
1688 
1689 	if (item_flags & vlanm)
1690 		return rte_flow_error_set(error, EINVAL,
1691 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1692 					  "multiple VLAN layers not supported");
1693 	else if ((item_flags & l34m) != 0)
1694 		return rte_flow_error_set(error, EINVAL,
1695 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1696 					  "VLAN cannot follow L3/L4 layer");
1697 	if (!mask)
1698 		mask = &rte_flow_item_vlan_mask;
1699 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1700 					(const uint8_t *)&nic_mask,
1701 					sizeof(struct rte_flow_item_vlan),
1702 					error);
1703 	if (ret)
1704 		return ret;
1705 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1706 		struct mlx5_priv *priv = dev->data->dev_private;
1707 
1708 		if (priv->vmwa_context) {
1709 			/*
1710 			 * Non-NULL context means we have a virtual machine
1711 			 * and SR-IOV enabled, we have to create VLAN interface
1712 			 * to make hypervisor to setup E-Switch vport
1713 			 * context correctly. We avoid creating the multiple
1714 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1715 			 */
1716 			return rte_flow_error_set(error, EINVAL,
1717 						  RTE_FLOW_ERROR_TYPE_ITEM,
1718 						  item,
1719 						  "VLAN tag mask is not"
1720 						  " supported in virtual"
1721 						  " environment");
1722 		}
1723 	}
1724 	return 0;
1725 }
1726 
1727 /*
1728  * GTP flags are contained in 1 byte of the format:
1729  * -------------------------------------------
1730  * | bit   | 0 - 2   | 3  | 4   | 5 | 6 | 7  |
1731  * |-----------------------------------------|
1732  * | value | Version | PT | Res | E | S | PN |
1733  * -------------------------------------------
1734  *
1735  * Matching is supported only for GTP flags E, S, PN.
1736  */
1737 #define MLX5_GTP_FLAGS_MASK	0x07
1738 
1739 /**
1740  * Validate GTP item.
1741  *
1742  * @param[in] dev
1743  *   Pointer to the rte_eth_dev structure.
1744  * @param[in] item
1745  *   Item specification.
1746  * @param[in] item_flags
1747  *   Bit-fields that holds the items detected until now.
1748  * @param[out] error
1749  *   Pointer to error structure.
1750  *
1751  * @return
1752  *   0 on success, a negative errno value otherwise and rte_errno is set.
1753  */
1754 static int
1755 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1756 			  const struct rte_flow_item *item,
1757 			  uint64_t item_flags,
1758 			  struct rte_flow_error *error)
1759 {
1760 	struct mlx5_priv *priv = dev->data->dev_private;
1761 	const struct rte_flow_item_gtp *spec = item->spec;
1762 	const struct rte_flow_item_gtp *mask = item->mask;
1763 	const struct rte_flow_item_gtp nic_mask = {
1764 		.v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1765 		.msg_type = 0xff,
1766 		.teid = RTE_BE32(0xffffffff),
1767 	};
1768 
1769 	if (!priv->config.hca_attr.tunnel_stateless_gtp)
1770 		return rte_flow_error_set(error, ENOTSUP,
1771 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1772 					  "GTP support is not enabled");
1773 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1774 		return rte_flow_error_set(error, ENOTSUP,
1775 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1776 					  "multiple tunnel layers not"
1777 					  " supported");
1778 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1779 		return rte_flow_error_set(error, EINVAL,
1780 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1781 					  "no outer UDP layer found");
1782 	if (!mask)
1783 		mask = &rte_flow_item_gtp_mask;
1784 	if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1785 		return rte_flow_error_set(error, ENOTSUP,
1786 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1787 					  "Match is supported for GTP"
1788 					  " flags only");
1789 	return mlx5_flow_item_acceptable
1790 		(item, (const uint8_t *)mask,
1791 		 (const uint8_t *)&nic_mask,
1792 		 sizeof(struct rte_flow_item_gtp),
1793 		 error);
1794 }
1795 
1796 /**
1797  * Validate the pop VLAN action.
1798  *
1799  * @param[in] dev
1800  *   Pointer to the rte_eth_dev structure.
1801  * @param[in] action_flags
1802  *   Holds the actions detected until now.
1803  * @param[in] action
1804  *   Pointer to the pop vlan action.
1805  * @param[in] item_flags
1806  *   The items found in this flow rule.
1807  * @param[in] attr
1808  *   Pointer to flow attributes.
1809  * @param[out] error
1810  *   Pointer to error structure.
1811  *
1812  * @return
1813  *   0 on success, a negative errno value otherwise and rte_errno is set.
1814  */
1815 static int
1816 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1817 				 uint64_t action_flags,
1818 				 const struct rte_flow_action *action,
1819 				 uint64_t item_flags,
1820 				 const struct rte_flow_attr *attr,
1821 				 struct rte_flow_error *error)
1822 {
1823 	const struct mlx5_priv *priv = dev->data->dev_private;
1824 
1825 	(void)action;
1826 	(void)attr;
1827 	if (!priv->sh->pop_vlan_action)
1828 		return rte_flow_error_set(error, ENOTSUP,
1829 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1830 					  NULL,
1831 					  "pop vlan action is not supported");
1832 	if (attr->egress)
1833 		return rte_flow_error_set(error, ENOTSUP,
1834 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1835 					  NULL,
1836 					  "pop vlan action not supported for "
1837 					  "egress");
1838 	if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1839 		return rte_flow_error_set(error, ENOTSUP,
1840 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1841 					  "no support for multiple VLAN "
1842 					  "actions");
1843 	/* Pop VLAN with preceding Decap requires inner header with VLAN. */
1844 	if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
1845 	    !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
1846 		return rte_flow_error_set(error, ENOTSUP,
1847 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1848 					  NULL,
1849 					  "cannot pop vlan after decap without "
1850 					  "match on inner vlan in the flow");
1851 	/* Pop VLAN without preceding Decap requires outer header with VLAN. */
1852 	if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
1853 	    !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1854 		return rte_flow_error_set(error, ENOTSUP,
1855 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1856 					  NULL,
1857 					  "cannot pop vlan without a "
1858 					  "match on (outer) vlan in the flow");
1859 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1860 		return rte_flow_error_set(error, EINVAL,
1861 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1862 					  "wrong action order, port_id should "
1863 					  "be after pop VLAN action");
1864 	if (!attr->transfer && priv->representor)
1865 		return rte_flow_error_set(error, ENOTSUP,
1866 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1867 					  "pop vlan action for VF representor "
1868 					  "not supported on NIC table");
1869 	return 0;
1870 }
1871 
1872 /**
1873  * Get VLAN default info from vlan match info.
1874  *
1875  * @param[in] items
1876  *   the list of item specifications.
1877  * @param[out] vlan
1878  *   pointer VLAN info to fill to.
1879  *
1880  * @return
1881  *   0 on success, a negative errno value otherwise and rte_errno is set.
1882  */
1883 static void
1884 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1885 				  struct rte_vlan_hdr *vlan)
1886 {
1887 	const struct rte_flow_item_vlan nic_mask = {
1888 		.tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1889 				MLX5DV_FLOW_VLAN_VID_MASK),
1890 		.inner_type = RTE_BE16(0xffff),
1891 	};
1892 
1893 	if (items == NULL)
1894 		return;
1895 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1896 		int type = items->type;
1897 
1898 		if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1899 		    type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1900 			break;
1901 	}
1902 	if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1903 		const struct rte_flow_item_vlan *vlan_m = items->mask;
1904 		const struct rte_flow_item_vlan *vlan_v = items->spec;
1905 
1906 		/* If VLAN item in pattern doesn't contain data, return here. */
1907 		if (!vlan_v)
1908 			return;
1909 		if (!vlan_m)
1910 			vlan_m = &nic_mask;
1911 		/* Only full match values are accepted */
1912 		if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1913 		     MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1914 			vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1915 			vlan->vlan_tci |=
1916 				rte_be_to_cpu_16(vlan_v->tci &
1917 						 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1918 		}
1919 		if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1920 		     MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1921 			vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1922 			vlan->vlan_tci |=
1923 				rte_be_to_cpu_16(vlan_v->tci &
1924 						 MLX5DV_FLOW_VLAN_VID_MASK_BE);
1925 		}
1926 		if (vlan_m->inner_type == nic_mask.inner_type)
1927 			vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1928 							   vlan_m->inner_type);
1929 	}
1930 }
1931 
1932 /**
1933  * Validate the push VLAN action.
1934  *
1935  * @param[in] dev
1936  *   Pointer to the rte_eth_dev structure.
1937  * @param[in] action_flags
1938  *   Holds the actions detected until now.
1939  * @param[in] item_flags
1940  *   The items found in this flow rule.
1941  * @param[in] action
1942  *   Pointer to the action structure.
1943  * @param[in] attr
1944  *   Pointer to flow attributes
1945  * @param[out] error
1946  *   Pointer to error structure.
1947  *
1948  * @return
1949  *   0 on success, a negative errno value otherwise and rte_errno is set.
1950  */
1951 static int
1952 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1953 				  uint64_t action_flags,
1954 				  const struct rte_flow_item_vlan *vlan_m,
1955 				  const struct rte_flow_action *action,
1956 				  const struct rte_flow_attr *attr,
1957 				  struct rte_flow_error *error)
1958 {
1959 	const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1960 	const struct mlx5_priv *priv = dev->data->dev_private;
1961 
1962 	if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1963 	    push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1964 		return rte_flow_error_set(error, EINVAL,
1965 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1966 					  "invalid vlan ethertype");
1967 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1968 		return rte_flow_error_set(error, EINVAL,
1969 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1970 					  "wrong action order, port_id should "
1971 					  "be after push VLAN");
1972 	if (!attr->transfer && priv->representor)
1973 		return rte_flow_error_set(error, ENOTSUP,
1974 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1975 					  "push vlan action for VF representor "
1976 					  "not supported on NIC table");
1977 	if (vlan_m &&
1978 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1979 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1980 		MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1981 	    !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1982 	    !(mlx5_flow_find_action
1983 		(action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1984 		return rte_flow_error_set(error, EINVAL,
1985 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1986 					  "not full match mask on VLAN PCP and "
1987 					  "there is no of_set_vlan_pcp action, "
1988 					  "push VLAN action cannot figure out "
1989 					  "PCP value");
1990 	if (vlan_m &&
1991 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1992 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1993 		MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1994 	    !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1995 	    !(mlx5_flow_find_action
1996 		(action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1997 		return rte_flow_error_set(error, EINVAL,
1998 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1999 					  "not full match mask on VLAN VID and "
2000 					  "there is no of_set_vlan_vid action, "
2001 					  "push VLAN action cannot figure out "
2002 					  "VID value");
2003 	(void)attr;
2004 	return 0;
2005 }
2006 
2007 /**
2008  * Validate the set VLAN PCP.
2009  *
2010  * @param[in] action_flags
2011  *   Holds the actions detected until now.
2012  * @param[in] actions
2013  *   Pointer to the list of actions remaining in the flow rule.
2014  * @param[out] error
2015  *   Pointer to error structure.
2016  *
2017  * @return
2018  *   0 on success, a negative errno value otherwise and rte_errno is set.
2019  */
2020 static int
2021 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2022 				     const struct rte_flow_action actions[],
2023 				     struct rte_flow_error *error)
2024 {
2025 	const struct rte_flow_action *action = actions;
2026 	const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2027 
2028 	if (conf->vlan_pcp > 7)
2029 		return rte_flow_error_set(error, EINVAL,
2030 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2031 					  "VLAN PCP value is too big");
2032 	if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2033 		return rte_flow_error_set(error, ENOTSUP,
2034 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2035 					  "set VLAN PCP action must follow "
2036 					  "the push VLAN action");
2037 	if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2038 		return rte_flow_error_set(error, ENOTSUP,
2039 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2040 					  "Multiple VLAN PCP modification are "
2041 					  "not supported");
2042 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2043 		return rte_flow_error_set(error, EINVAL,
2044 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2045 					  "wrong action order, port_id should "
2046 					  "be after set VLAN PCP");
2047 	return 0;
2048 }
2049 
2050 /**
2051  * Validate the set VLAN VID.
2052  *
2053  * @param[in] item_flags
2054  *   Holds the items detected in this rule.
2055  * @param[in] action_flags
2056  *   Holds the actions detected until now.
2057  * @param[in] actions
2058  *   Pointer to the list of actions remaining in the flow rule.
2059  * @param[out] error
2060  *   Pointer to error structure.
2061  *
2062  * @return
2063  *   0 on success, a negative errno value otherwise and rte_errno is set.
2064  */
2065 static int
2066 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2067 				     uint64_t action_flags,
2068 				     const struct rte_flow_action actions[],
2069 				     struct rte_flow_error *error)
2070 {
2071 	const struct rte_flow_action *action = actions;
2072 	const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2073 
2074 	if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2075 		return rte_flow_error_set(error, EINVAL,
2076 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2077 					  "VLAN VID value is too big");
2078 	if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2079 	    !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2080 		return rte_flow_error_set(error, ENOTSUP,
2081 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2082 					  "set VLAN VID action must follow push"
2083 					  " VLAN action or match on VLAN item");
2084 	if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2085 		return rte_flow_error_set(error, ENOTSUP,
2086 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2087 					  "Multiple VLAN VID modifications are "
2088 					  "not supported");
2089 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2090 		return rte_flow_error_set(error, EINVAL,
2091 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2092 					  "wrong action order, port_id should "
2093 					  "be after set VLAN VID");
2094 	return 0;
2095 }
2096 
2097 /*
2098  * Validate the FLAG action.
2099  *
2100  * @param[in] dev
2101  *   Pointer to the rte_eth_dev structure.
2102  * @param[in] action_flags
2103  *   Holds the actions detected until now.
2104  * @param[in] attr
2105  *   Pointer to flow attributes
2106  * @param[out] error
2107  *   Pointer to error structure.
2108  *
2109  * @return
2110  *   0 on success, a negative errno value otherwise and rte_errno is set.
2111  */
2112 static int
2113 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2114 			     uint64_t action_flags,
2115 			     const struct rte_flow_attr *attr,
2116 			     struct rte_flow_error *error)
2117 {
2118 	struct mlx5_priv *priv = dev->data->dev_private;
2119 	struct mlx5_dev_config *config = &priv->config;
2120 	int ret;
2121 
2122 	/* Fall back if no extended metadata register support. */
2123 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2124 		return mlx5_flow_validate_action_flag(action_flags, attr,
2125 						      error);
2126 	/* Extensive metadata mode requires registers. */
2127 	if (!mlx5_flow_ext_mreg_supported(dev))
2128 		return rte_flow_error_set(error, ENOTSUP,
2129 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2130 					  "no metadata registers "
2131 					  "to support flag action");
2132 	if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2133 		return rte_flow_error_set(error, ENOTSUP,
2134 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2135 					  "extended metadata register"
2136 					  " isn't available");
2137 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2138 	if (ret < 0)
2139 		return ret;
2140 	MLX5_ASSERT(ret > 0);
2141 	if (action_flags & MLX5_FLOW_ACTION_MARK)
2142 		return rte_flow_error_set(error, EINVAL,
2143 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2144 					  "can't mark and flag in same flow");
2145 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
2146 		return rte_flow_error_set(error, EINVAL,
2147 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2148 					  "can't have 2 flag"
2149 					  " actions in same flow");
2150 	return 0;
2151 }
2152 
2153 /**
2154  * Validate MARK action.
2155  *
2156  * @param[in] dev
2157  *   Pointer to the rte_eth_dev structure.
2158  * @param[in] action
2159  *   Pointer to action.
2160  * @param[in] action_flags
2161  *   Holds the actions detected until now.
2162  * @param[in] attr
2163  *   Pointer to flow attributes
2164  * @param[out] error
2165  *   Pointer to error structure.
2166  *
2167  * @return
2168  *   0 on success, a negative errno value otherwise and rte_errno is set.
2169  */
2170 static int
2171 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2172 			     const struct rte_flow_action *action,
2173 			     uint64_t action_flags,
2174 			     const struct rte_flow_attr *attr,
2175 			     struct rte_flow_error *error)
2176 {
2177 	struct mlx5_priv *priv = dev->data->dev_private;
2178 	struct mlx5_dev_config *config = &priv->config;
2179 	const struct rte_flow_action_mark *mark = action->conf;
2180 	int ret;
2181 
2182 	/* Fall back if no extended metadata register support. */
2183 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2184 		return mlx5_flow_validate_action_mark(action, action_flags,
2185 						      attr, error);
2186 	/* Extensive metadata mode requires registers. */
2187 	if (!mlx5_flow_ext_mreg_supported(dev))
2188 		return rte_flow_error_set(error, ENOTSUP,
2189 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2190 					  "no metadata registers "
2191 					  "to support mark action");
2192 	if (!priv->sh->dv_mark_mask)
2193 		return rte_flow_error_set(error, ENOTSUP,
2194 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2195 					  "extended metadata register"
2196 					  " isn't available");
2197 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2198 	if (ret < 0)
2199 		return ret;
2200 	MLX5_ASSERT(ret > 0);
2201 	if (!mark)
2202 		return rte_flow_error_set(error, EINVAL,
2203 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2204 					  "configuration cannot be null");
2205 	if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2206 		return rte_flow_error_set(error, EINVAL,
2207 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2208 					  &mark->id,
2209 					  "mark id exceeds the limit");
2210 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
2211 		return rte_flow_error_set(error, EINVAL,
2212 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2213 					  "can't flag and mark in same flow");
2214 	if (action_flags & MLX5_FLOW_ACTION_MARK)
2215 		return rte_flow_error_set(error, EINVAL,
2216 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2217 					  "can't have 2 mark actions in same"
2218 					  " flow");
2219 	return 0;
2220 }
2221 
2222 /**
2223  * Validate SET_META action.
2224  *
2225  * @param[in] dev
2226  *   Pointer to the rte_eth_dev structure.
2227  * @param[in] action
2228  *   Pointer to the action structure.
2229  * @param[in] action_flags
2230  *   Holds the actions detected until now.
2231  * @param[in] attr
2232  *   Pointer to flow attributes
2233  * @param[out] error
2234  *   Pointer to error structure.
2235  *
2236  * @return
2237  *   0 on success, a negative errno value otherwise and rte_errno is set.
2238  */
2239 static int
2240 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2241 				 const struct rte_flow_action *action,
2242 				 uint64_t action_flags __rte_unused,
2243 				 const struct rte_flow_attr *attr,
2244 				 struct rte_flow_error *error)
2245 {
2246 	const struct rte_flow_action_set_meta *conf;
2247 	uint32_t nic_mask = UINT32_MAX;
2248 	int reg;
2249 
2250 	if (!mlx5_flow_ext_mreg_supported(dev))
2251 		return rte_flow_error_set(error, ENOTSUP,
2252 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2253 					  "extended metadata register"
2254 					  " isn't supported");
2255 	reg = flow_dv_get_metadata_reg(dev, attr, error);
2256 	if (reg < 0)
2257 		return reg;
2258 	if (reg != REG_A && reg != REG_B) {
2259 		struct mlx5_priv *priv = dev->data->dev_private;
2260 
2261 		nic_mask = priv->sh->dv_meta_mask;
2262 	}
2263 	if (!(action->conf))
2264 		return rte_flow_error_set(error, EINVAL,
2265 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2266 					  "configuration cannot be null");
2267 	conf = (const struct rte_flow_action_set_meta *)action->conf;
2268 	if (!conf->mask)
2269 		return rte_flow_error_set(error, EINVAL,
2270 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2271 					  "zero mask doesn't have any effect");
2272 	if (conf->mask & ~nic_mask)
2273 		return rte_flow_error_set(error, EINVAL,
2274 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2275 					  "meta data must be within reg C0");
2276 	return 0;
2277 }
2278 
2279 /**
2280  * Validate SET_TAG action.
2281  *
2282  * @param[in] dev
2283  *   Pointer to the rte_eth_dev structure.
2284  * @param[in] action
2285  *   Pointer to the action structure.
2286  * @param[in] action_flags
2287  *   Holds the actions detected until now.
2288  * @param[in] attr
2289  *   Pointer to flow attributes
2290  * @param[out] error
2291  *   Pointer to error structure.
2292  *
2293  * @return
2294  *   0 on success, a negative errno value otherwise and rte_errno is set.
2295  */
2296 static int
2297 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2298 				const struct rte_flow_action *action,
2299 				uint64_t action_flags,
2300 				const struct rte_flow_attr *attr,
2301 				struct rte_flow_error *error)
2302 {
2303 	const struct rte_flow_action_set_tag *conf;
2304 	const uint64_t terminal_action_flags =
2305 		MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2306 		MLX5_FLOW_ACTION_RSS;
2307 	int ret;
2308 
2309 	if (!mlx5_flow_ext_mreg_supported(dev))
2310 		return rte_flow_error_set(error, ENOTSUP,
2311 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2312 					  "extensive metadata register"
2313 					  " isn't supported");
2314 	if (!(action->conf))
2315 		return rte_flow_error_set(error, EINVAL,
2316 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2317 					  "configuration cannot be null");
2318 	conf = (const struct rte_flow_action_set_tag *)action->conf;
2319 	if (!conf->mask)
2320 		return rte_flow_error_set(error, EINVAL,
2321 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2322 					  "zero mask doesn't have any effect");
2323 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2324 	if (ret < 0)
2325 		return ret;
2326 	if (!attr->transfer && attr->ingress &&
2327 	    (action_flags & terminal_action_flags))
2328 		return rte_flow_error_set(error, EINVAL,
2329 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2330 					  "set_tag has no effect"
2331 					  " with terminal actions");
2332 	return 0;
2333 }
2334 
2335 /**
2336  * Validate count action.
2337  *
2338  * @param[in] dev
2339  *   Pointer to rte_eth_dev structure.
2340  * @param[out] error
2341  *   Pointer to error structure.
2342  *
2343  * @return
2344  *   0 on success, a negative errno value otherwise and rte_errno is set.
2345  */
2346 static int
2347 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2348 			      struct rte_flow_error *error)
2349 {
2350 	struct mlx5_priv *priv = dev->data->dev_private;
2351 
2352 	if (!priv->config.devx)
2353 		goto notsup_err;
2354 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2355 	return 0;
2356 #endif
2357 notsup_err:
2358 	return rte_flow_error_set
2359 		      (error, ENOTSUP,
2360 		       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2361 		       NULL,
2362 		       "count action not supported");
2363 }
2364 
2365 /**
2366  * Validate the L2 encap action.
2367  *
2368  * @param[in] dev
2369  *   Pointer to the rte_eth_dev structure.
2370  * @param[in] action_flags
2371  *   Holds the actions detected until now.
2372  * @param[in] action
2373  *   Pointer to the action structure.
2374  * @param[in] attr
2375  *   Pointer to flow attributes.
2376  * @param[out] error
2377  *   Pointer to error structure.
2378  *
2379  * @return
2380  *   0 on success, a negative errno value otherwise and rte_errno is set.
2381  */
2382 static int
2383 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2384 				 uint64_t action_flags,
2385 				 const struct rte_flow_action *action,
2386 				 const struct rte_flow_attr *attr,
2387 				 struct rte_flow_error *error)
2388 {
2389 	const struct mlx5_priv *priv = dev->data->dev_private;
2390 
2391 	if (!(action->conf))
2392 		return rte_flow_error_set(error, EINVAL,
2393 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2394 					  "configuration cannot be null");
2395 	if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2396 		return rte_flow_error_set(error, EINVAL,
2397 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2398 					  "can only have a single encap action "
2399 					  "in a flow");
2400 	if (!attr->transfer && priv->representor)
2401 		return rte_flow_error_set(error, ENOTSUP,
2402 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2403 					  "encap action for VF representor "
2404 					  "not supported on NIC table");
2405 	return 0;
2406 }
2407 
2408 /**
2409  * Validate a decap action.
2410  *
2411  * @param[in] dev
2412  *   Pointer to the rte_eth_dev structure.
2413  * @param[in] action_flags
2414  *   Holds the actions detected until now.
2415  * @param[in] attr
2416  *   Pointer to flow attributes
2417  * @param[out] error
2418  *   Pointer to error structure.
2419  *
2420  * @return
2421  *   0 on success, a negative errno value otherwise and rte_errno is set.
2422  */
2423 static int
2424 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2425 			      uint64_t action_flags,
2426 			      const struct rte_flow_attr *attr,
2427 			      struct rte_flow_error *error)
2428 {
2429 	const struct mlx5_priv *priv = dev->data->dev_private;
2430 
2431 	if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2432 	    !priv->config.decap_en)
2433 		return rte_flow_error_set(error, ENOTSUP,
2434 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2435 					  "decap is not enabled");
2436 	if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2437 		return rte_flow_error_set(error, ENOTSUP,
2438 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2439 					  action_flags &
2440 					  MLX5_FLOW_ACTION_DECAP ? "can only "
2441 					  "have a single decap action" : "decap "
2442 					  "after encap is not supported");
2443 	if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2444 		return rte_flow_error_set(error, EINVAL,
2445 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2446 					  "can't have decap action after"
2447 					  " modify action");
2448 	if (attr->egress)
2449 		return rte_flow_error_set(error, ENOTSUP,
2450 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2451 					  NULL,
2452 					  "decap action not supported for "
2453 					  "egress");
2454 	if (!attr->transfer && priv->representor)
2455 		return rte_flow_error_set(error, ENOTSUP,
2456 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2457 					  "decap action for VF representor "
2458 					  "not supported on NIC table");
2459 	return 0;
2460 }
2461 
2462 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2463 
2464 /**
2465  * Validate the raw encap and decap actions.
2466  *
2467  * @param[in] dev
2468  *   Pointer to the rte_eth_dev structure.
2469  * @param[in] decap
2470  *   Pointer to the decap action.
2471  * @param[in] encap
2472  *   Pointer to the encap action.
2473  * @param[in] attr
2474  *   Pointer to flow attributes
2475  * @param[in/out] action_flags
2476  *   Holds the actions detected until now.
2477  * @param[out] actions_n
2478  *   pointer to the number of actions counter.
2479  * @param[out] error
2480  *   Pointer to error structure.
2481  *
2482  * @return
2483  *   0 on success, a negative errno value otherwise and rte_errno is set.
2484  */
2485 static int
2486 flow_dv_validate_action_raw_encap_decap
2487 	(struct rte_eth_dev *dev,
2488 	 const struct rte_flow_action_raw_decap *decap,
2489 	 const struct rte_flow_action_raw_encap *encap,
2490 	 const struct rte_flow_attr *attr, uint64_t *action_flags,
2491 	 int *actions_n, struct rte_flow_error *error)
2492 {
2493 	const struct mlx5_priv *priv = dev->data->dev_private;
2494 	int ret;
2495 
2496 	if (encap && (!encap->size || !encap->data))
2497 		return rte_flow_error_set(error, EINVAL,
2498 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2499 					  "raw encap data cannot be empty");
2500 	if (decap && encap) {
2501 		if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2502 		    encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2503 			/* L3 encap. */
2504 			decap = NULL;
2505 		else if (encap->size <=
2506 			   MLX5_ENCAPSULATION_DECISION_SIZE &&
2507 			   decap->size >
2508 			   MLX5_ENCAPSULATION_DECISION_SIZE)
2509 			/* L3 decap. */
2510 			encap = NULL;
2511 		else if (encap->size >
2512 			   MLX5_ENCAPSULATION_DECISION_SIZE &&
2513 			   decap->size >
2514 			   MLX5_ENCAPSULATION_DECISION_SIZE)
2515 			/* 2 L2 actions: encap and decap. */
2516 			;
2517 		else
2518 			return rte_flow_error_set(error,
2519 				ENOTSUP,
2520 				RTE_FLOW_ERROR_TYPE_ACTION,
2521 				NULL, "unsupported too small "
2522 				"raw decap and too small raw "
2523 				"encap combination");
2524 	}
2525 	if (decap) {
2526 		ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2527 						    error);
2528 		if (ret < 0)
2529 			return ret;
2530 		*action_flags |= MLX5_FLOW_ACTION_DECAP;
2531 		++(*actions_n);
2532 	}
2533 	if (encap) {
2534 		if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2535 			return rte_flow_error_set(error, ENOTSUP,
2536 						  RTE_FLOW_ERROR_TYPE_ACTION,
2537 						  NULL,
2538 						  "small raw encap size");
2539 		if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2540 			return rte_flow_error_set(error, EINVAL,
2541 						  RTE_FLOW_ERROR_TYPE_ACTION,
2542 						  NULL,
2543 						  "more than one encap action");
2544 		if (!attr->transfer && priv->representor)
2545 			return rte_flow_error_set
2546 					(error, ENOTSUP,
2547 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2548 					 "encap action for VF representor "
2549 					 "not supported on NIC table");
2550 		*action_flags |= MLX5_FLOW_ACTION_ENCAP;
2551 		++(*actions_n);
2552 	}
2553 	return 0;
2554 }
2555 
2556 /**
2557  * Match encap_decap resource.
2558  *
2559  * @param entry
2560  *   Pointer to exist resource entry object.
2561  * @param ctx
2562  *   Pointer to new encap_decap resource.
2563  *
2564  * @return
2565  *   0 on matching, -1 otherwise.
2566  */
2567 static int
2568 flow_dv_encap_decap_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
2569 {
2570 	struct mlx5_flow_dv_encap_decap_resource *resource;
2571 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2572 
2573 	resource = (struct mlx5_flow_dv_encap_decap_resource *)ctx;
2574 	cache_resource = container_of(entry,
2575 				      struct mlx5_flow_dv_encap_decap_resource,
2576 				      entry);
2577 	if (resource->entry.key == cache_resource->entry.key &&
2578 	    resource->reformat_type == cache_resource->reformat_type &&
2579 	    resource->ft_type == cache_resource->ft_type &&
2580 	    resource->flags == cache_resource->flags &&
2581 	    resource->size == cache_resource->size &&
2582 	    !memcmp((const void *)resource->buf,
2583 		    (const void *)cache_resource->buf,
2584 		    resource->size))
2585 		return 0;
2586 	return -1;
2587 }
2588 
2589 /**
2590  * Find existing encap/decap resource or create and register a new one.
2591  *
2592  * @param[in, out] dev
2593  *   Pointer to rte_eth_dev structure.
2594  * @param[in, out] resource
2595  *   Pointer to encap/decap resource.
2596  * @parm[in, out] dev_flow
2597  *   Pointer to the dev_flow.
2598  * @param[out] error
2599  *   pointer to error structure.
2600  *
2601  * @return
2602  *   0 on success otherwise -errno and errno is set.
2603  */
2604 static int
2605 flow_dv_encap_decap_resource_register
2606 			(struct rte_eth_dev *dev,
2607 			 struct mlx5_flow_dv_encap_decap_resource *resource,
2608 			 struct mlx5_flow *dev_flow,
2609 			 struct rte_flow_error *error)
2610 {
2611 	struct mlx5_priv *priv = dev->data->dev_private;
2612 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2613 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2614 	struct mlx5dv_dr_domain *domain;
2615 	struct mlx5_hlist_entry *entry;
2616 	union mlx5_flow_encap_decap_key encap_decap_key = {
2617 		{
2618 			.ft_type = resource->ft_type,
2619 			.refmt_type = resource->reformat_type,
2620 			.buf_size = resource->size,
2621 			.table_level = !!dev_flow->dv.group,
2622 			.cksum = 0,
2623 		}
2624 	};
2625 	int ret;
2626 
2627 	resource->flags = dev_flow->dv.group ? 0 : 1;
2628 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2629 		domain = sh->fdb_domain;
2630 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2631 		domain = sh->rx_domain;
2632 	else
2633 		domain = sh->tx_domain;
2634 	encap_decap_key.cksum = __rte_raw_cksum(resource->buf,
2635 						resource->size, 0);
2636 	resource->entry.key = encap_decap_key.v64;
2637 	/* Lookup a matching resource from cache. */
2638 	entry = mlx5_hlist_lookup_ex(sh->encaps_decaps, resource->entry.key,
2639 				     flow_dv_encap_decap_resource_match,
2640 				     (void *)resource);
2641 	if (entry) {
2642 		cache_resource = container_of(entry,
2643 			struct mlx5_flow_dv_encap_decap_resource, entry);
2644 		DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2645 			(void *)cache_resource,
2646 			rte_atomic32_read(&cache_resource->refcnt));
2647 		rte_atomic32_inc(&cache_resource->refcnt);
2648 		dev_flow->handle->dvh.rix_encap_decap = cache_resource->idx;
2649 		dev_flow->dv.encap_decap = cache_resource;
2650 		return 0;
2651 	}
2652 	/* Register new encap/decap resource. */
2653 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2654 				       &dev_flow->handle->dvh.rix_encap_decap);
2655 	if (!cache_resource)
2656 		return rte_flow_error_set(error, ENOMEM,
2657 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2658 					  "cannot allocate resource memory");
2659 	*cache_resource = *resource;
2660 	cache_resource->idx = dev_flow->handle->dvh.rix_encap_decap;
2661 	ret = mlx5_flow_os_create_flow_action_packet_reformat
2662 					(sh->ctx, domain, cache_resource,
2663 					 &cache_resource->action);
2664 	if (ret) {
2665 		mlx5_free(cache_resource);
2666 		return rte_flow_error_set(error, ENOMEM,
2667 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2668 					  NULL, "cannot create action");
2669 	}
2670 	rte_atomic32_init(&cache_resource->refcnt);
2671 	rte_atomic32_inc(&cache_resource->refcnt);
2672 	if (mlx5_hlist_insert_ex(sh->encaps_decaps, &cache_resource->entry,
2673 				 flow_dv_encap_decap_resource_match,
2674 				 (void *)cache_resource)) {
2675 		claim_zero(mlx5_flow_os_destroy_flow_action
2676 						(cache_resource->action));
2677 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2678 				cache_resource->idx);
2679 		return rte_flow_error_set(error, EEXIST,
2680 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2681 					  NULL, "action exist");
2682 	}
2683 	dev_flow->dv.encap_decap = cache_resource;
2684 	DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2685 		(void *)cache_resource,
2686 		rte_atomic32_read(&cache_resource->refcnt));
2687 	return 0;
2688 }
2689 
2690 /**
2691  * Find existing table jump resource or create and register a new one.
2692  *
2693  * @param[in, out] dev
2694  *   Pointer to rte_eth_dev structure.
2695  * @param[in, out] tbl
2696  *   Pointer to flow table resource.
2697  * @parm[in, out] dev_flow
2698  *   Pointer to the dev_flow.
2699  * @param[out] error
2700  *   pointer to error structure.
2701  *
2702  * @return
2703  *   0 on success otherwise -errno and errno is set.
2704  */
2705 static int
2706 flow_dv_jump_tbl_resource_register
2707 			(struct rte_eth_dev *dev __rte_unused,
2708 			 struct mlx5_flow_tbl_resource *tbl,
2709 			 struct mlx5_flow *dev_flow,
2710 			 struct rte_flow_error *error)
2711 {
2712 	struct mlx5_flow_tbl_data_entry *tbl_data =
2713 		container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2714 	int cnt, ret;
2715 
2716 	MLX5_ASSERT(tbl);
2717 	cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2718 	if (!cnt) {
2719 		ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
2720 				(tbl->obj, &tbl_data->jump.action);
2721 		if (ret)
2722 			return rte_flow_error_set(error, ENOMEM,
2723 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2724 					NULL, "cannot create jump action");
2725 		DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2726 			(void *)&tbl_data->jump, cnt);
2727 	} else {
2728 		/* old jump should not make the table ref++. */
2729 		flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2730 		MLX5_ASSERT(tbl_data->jump.action);
2731 		DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2732 			(void *)&tbl_data->jump, cnt);
2733 	}
2734 	rte_atomic32_inc(&tbl_data->jump.refcnt);
2735 	dev_flow->handle->rix_jump = tbl_data->idx;
2736 	dev_flow->dv.jump = &tbl_data->jump;
2737 	return 0;
2738 }
2739 
2740 /**
2741  * Find existing default miss resource or create and register a new one.
2742  *
2743  * @param[in, out] dev
2744  *   Pointer to rte_eth_dev structure.
2745  * @param[out] error
2746  *   pointer to error structure.
2747  *
2748  * @return
2749  *   0 on success otherwise -errno and errno is set.
2750  */
2751 static int
2752 flow_dv_default_miss_resource_register(struct rte_eth_dev *dev,
2753 		struct rte_flow_error *error)
2754 {
2755 	struct mlx5_priv *priv = dev->data->dev_private;
2756 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2757 	struct mlx5_flow_default_miss_resource *cache_resource =
2758 			&sh->default_miss;
2759 	int cnt = rte_atomic32_read(&cache_resource->refcnt);
2760 
2761 	if (!cnt) {
2762 		MLX5_ASSERT(cache_resource->action);
2763 		cache_resource->action =
2764 		mlx5_glue->dr_create_flow_action_default_miss();
2765 		if (!cache_resource->action)
2766 			return rte_flow_error_set(error, ENOMEM,
2767 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2768 					"cannot create default miss action");
2769 		DRV_LOG(DEBUG, "new default miss resource %p: refcnt %d++",
2770 				(void *)cache_resource->action, cnt);
2771 	}
2772 	rte_atomic32_inc(&cache_resource->refcnt);
2773 	return 0;
2774 }
2775 
2776 /**
2777  * Find existing table port ID resource or create and register a new one.
2778  *
2779  * @param[in, out] dev
2780  *   Pointer to rte_eth_dev structure.
2781  * @param[in, out] resource
2782  *   Pointer to port ID action resource.
2783  * @parm[in, out] dev_flow
2784  *   Pointer to the dev_flow.
2785  * @param[out] error
2786  *   pointer to error structure.
2787  *
2788  * @return
2789  *   0 on success otherwise -errno and errno is set.
2790  */
2791 static int
2792 flow_dv_port_id_action_resource_register
2793 			(struct rte_eth_dev *dev,
2794 			 struct mlx5_flow_dv_port_id_action_resource *resource,
2795 			 struct mlx5_flow *dev_flow,
2796 			 struct rte_flow_error *error)
2797 {
2798 	struct mlx5_priv *priv = dev->data->dev_private;
2799 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2800 	struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2801 	uint32_t idx = 0;
2802 	int ret;
2803 
2804 	/* Lookup a matching resource from cache. */
2805 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2806 		      idx, cache_resource, next) {
2807 		if (resource->port_id == cache_resource->port_id) {
2808 			DRV_LOG(DEBUG, "port id action resource resource %p: "
2809 				"refcnt %d++",
2810 				(void *)cache_resource,
2811 				rte_atomic32_read(&cache_resource->refcnt));
2812 			rte_atomic32_inc(&cache_resource->refcnt);
2813 			dev_flow->handle->rix_port_id_action = idx;
2814 			dev_flow->dv.port_id_action = cache_resource;
2815 			return 0;
2816 		}
2817 	}
2818 	/* Register new port id action resource. */
2819 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2820 				       &dev_flow->handle->rix_port_id_action);
2821 	if (!cache_resource)
2822 		return rte_flow_error_set(error, ENOMEM,
2823 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2824 					  "cannot allocate resource memory");
2825 	*cache_resource = *resource;
2826 	ret = mlx5_flow_os_create_flow_action_dest_port
2827 				(priv->sh->fdb_domain, resource->port_id,
2828 				 &cache_resource->action);
2829 	if (ret) {
2830 		mlx5_free(cache_resource);
2831 		return rte_flow_error_set(error, ENOMEM,
2832 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2833 					  NULL, "cannot create action");
2834 	}
2835 	rte_atomic32_init(&cache_resource->refcnt);
2836 	rte_atomic32_inc(&cache_resource->refcnt);
2837 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2838 		     dev_flow->handle->rix_port_id_action, cache_resource,
2839 		     next);
2840 	dev_flow->dv.port_id_action = cache_resource;
2841 	DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2842 		(void *)cache_resource,
2843 		rte_atomic32_read(&cache_resource->refcnt));
2844 	return 0;
2845 }
2846 
2847 /**
2848  * Find existing push vlan resource or create and register a new one.
2849  *
2850  * @param [in, out] dev
2851  *   Pointer to rte_eth_dev structure.
2852  * @param[in, out] resource
2853  *   Pointer to port ID action resource.
2854  * @parm[in, out] dev_flow
2855  *   Pointer to the dev_flow.
2856  * @param[out] error
2857  *   pointer to error structure.
2858  *
2859  * @return
2860  *   0 on success otherwise -errno and errno is set.
2861  */
2862 static int
2863 flow_dv_push_vlan_action_resource_register
2864 		       (struct rte_eth_dev *dev,
2865 			struct mlx5_flow_dv_push_vlan_action_resource *resource,
2866 			struct mlx5_flow *dev_flow,
2867 			struct rte_flow_error *error)
2868 {
2869 	struct mlx5_priv *priv = dev->data->dev_private;
2870 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2871 	struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2872 	struct mlx5dv_dr_domain *domain;
2873 	uint32_t idx = 0;
2874 	int ret;
2875 
2876 	/* Lookup a matching resource from cache. */
2877 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2878 		      sh->push_vlan_action_list, idx, cache_resource, next) {
2879 		if (resource->vlan_tag == cache_resource->vlan_tag &&
2880 		    resource->ft_type == cache_resource->ft_type) {
2881 			DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2882 				"refcnt %d++",
2883 				(void *)cache_resource,
2884 				rte_atomic32_read(&cache_resource->refcnt));
2885 			rte_atomic32_inc(&cache_resource->refcnt);
2886 			dev_flow->handle->dvh.rix_push_vlan = idx;
2887 			dev_flow->dv.push_vlan_res = cache_resource;
2888 			return 0;
2889 		}
2890 	}
2891 	/* Register new push_vlan action resource. */
2892 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2893 				       &dev_flow->handle->dvh.rix_push_vlan);
2894 	if (!cache_resource)
2895 		return rte_flow_error_set(error, ENOMEM,
2896 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2897 					  "cannot allocate resource memory");
2898 	*cache_resource = *resource;
2899 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2900 		domain = sh->fdb_domain;
2901 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2902 		domain = sh->rx_domain;
2903 	else
2904 		domain = sh->tx_domain;
2905 	ret = mlx5_flow_os_create_flow_action_push_vlan
2906 					(domain, resource->vlan_tag,
2907 					 &cache_resource->action);
2908 	if (ret) {
2909 		mlx5_free(cache_resource);
2910 		return rte_flow_error_set(error, ENOMEM,
2911 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2912 					  NULL, "cannot create action");
2913 	}
2914 	rte_atomic32_init(&cache_resource->refcnt);
2915 	rte_atomic32_inc(&cache_resource->refcnt);
2916 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2917 		     &sh->push_vlan_action_list,
2918 		     dev_flow->handle->dvh.rix_push_vlan,
2919 		     cache_resource, next);
2920 	dev_flow->dv.push_vlan_res = cache_resource;
2921 	DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2922 		(void *)cache_resource,
2923 		rte_atomic32_read(&cache_resource->refcnt));
2924 	return 0;
2925 }
2926 /**
2927  * Get the size of specific rte_flow_item_type hdr size
2928  *
2929  * @param[in] item_type
2930  *   Tested rte_flow_item_type.
2931  *
2932  * @return
2933  *   sizeof struct item_type, 0 if void or irrelevant.
2934  */
2935 static size_t
2936 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
2937 {
2938 	size_t retval;
2939 
2940 	switch (item_type) {
2941 	case RTE_FLOW_ITEM_TYPE_ETH:
2942 		retval = sizeof(struct rte_ether_hdr);
2943 		break;
2944 	case RTE_FLOW_ITEM_TYPE_VLAN:
2945 		retval = sizeof(struct rte_vlan_hdr);
2946 		break;
2947 	case RTE_FLOW_ITEM_TYPE_IPV4:
2948 		retval = sizeof(struct rte_ipv4_hdr);
2949 		break;
2950 	case RTE_FLOW_ITEM_TYPE_IPV6:
2951 		retval = sizeof(struct rte_ipv6_hdr);
2952 		break;
2953 	case RTE_FLOW_ITEM_TYPE_UDP:
2954 		retval = sizeof(struct rte_udp_hdr);
2955 		break;
2956 	case RTE_FLOW_ITEM_TYPE_TCP:
2957 		retval = sizeof(struct rte_tcp_hdr);
2958 		break;
2959 	case RTE_FLOW_ITEM_TYPE_VXLAN:
2960 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2961 		retval = sizeof(struct rte_vxlan_hdr);
2962 		break;
2963 	case RTE_FLOW_ITEM_TYPE_GRE:
2964 	case RTE_FLOW_ITEM_TYPE_NVGRE:
2965 		retval = sizeof(struct rte_gre_hdr);
2966 		break;
2967 	case RTE_FLOW_ITEM_TYPE_MPLS:
2968 		retval = sizeof(struct rte_mpls_hdr);
2969 		break;
2970 	case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2971 	default:
2972 		retval = 0;
2973 		break;
2974 	}
2975 	return retval;
2976 }
2977 
2978 #define MLX5_ENCAP_IPV4_VERSION		0x40
2979 #define MLX5_ENCAP_IPV4_IHL_MIN		0x05
2980 #define MLX5_ENCAP_IPV4_TTL_DEF		0x40
2981 #define MLX5_ENCAP_IPV6_VTC_FLOW	0x60000000
2982 #define MLX5_ENCAP_IPV6_HOP_LIMIT	0xff
2983 #define MLX5_ENCAP_VXLAN_FLAGS		0x08000000
2984 #define MLX5_ENCAP_VXLAN_GPE_FLAGS	0x04
2985 
2986 /**
2987  * Convert the encap action data from list of rte_flow_item to raw buffer
2988  *
2989  * @param[in] items
2990  *   Pointer to rte_flow_item objects list.
2991  * @param[out] buf
2992  *   Pointer to the output buffer.
2993  * @param[out] size
2994  *   Pointer to the output buffer size.
2995  * @param[out] error
2996  *   Pointer to the error structure.
2997  *
2998  * @return
2999  *   0 on success, a negative errno value otherwise and rte_errno is set.
3000  */
3001 static int
3002 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
3003 			   size_t *size, struct rte_flow_error *error)
3004 {
3005 	struct rte_ether_hdr *eth = NULL;
3006 	struct rte_vlan_hdr *vlan = NULL;
3007 	struct rte_ipv4_hdr *ipv4 = NULL;
3008 	struct rte_ipv6_hdr *ipv6 = NULL;
3009 	struct rte_udp_hdr *udp = NULL;
3010 	struct rte_vxlan_hdr *vxlan = NULL;
3011 	struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
3012 	struct rte_gre_hdr *gre = NULL;
3013 	size_t len;
3014 	size_t temp_size = 0;
3015 
3016 	if (!items)
3017 		return rte_flow_error_set(error, EINVAL,
3018 					  RTE_FLOW_ERROR_TYPE_ACTION,
3019 					  NULL, "invalid empty data");
3020 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3021 		len = flow_dv_get_item_hdr_len(items->type);
3022 		if (len + temp_size > MLX5_ENCAP_MAX_LEN)
3023 			return rte_flow_error_set(error, EINVAL,
3024 						  RTE_FLOW_ERROR_TYPE_ACTION,
3025 						  (void *)items->type,
3026 						  "items total size is too big"
3027 						  " for encap action");
3028 		rte_memcpy((void *)&buf[temp_size], items->spec, len);
3029 		switch (items->type) {
3030 		case RTE_FLOW_ITEM_TYPE_ETH:
3031 			eth = (struct rte_ether_hdr *)&buf[temp_size];
3032 			break;
3033 		case RTE_FLOW_ITEM_TYPE_VLAN:
3034 			vlan = (struct rte_vlan_hdr *)&buf[temp_size];
3035 			if (!eth)
3036 				return rte_flow_error_set(error, EINVAL,
3037 						RTE_FLOW_ERROR_TYPE_ACTION,
3038 						(void *)items->type,
3039 						"eth header not found");
3040 			if (!eth->ether_type)
3041 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
3042 			break;
3043 		case RTE_FLOW_ITEM_TYPE_IPV4:
3044 			ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
3045 			if (!vlan && !eth)
3046 				return rte_flow_error_set(error, EINVAL,
3047 						RTE_FLOW_ERROR_TYPE_ACTION,
3048 						(void *)items->type,
3049 						"neither eth nor vlan"
3050 						" header found");
3051 			if (vlan && !vlan->eth_proto)
3052 				vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3053 			else if (eth && !eth->ether_type)
3054 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
3055 			if (!ipv4->version_ihl)
3056 				ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
3057 						    MLX5_ENCAP_IPV4_IHL_MIN;
3058 			if (!ipv4->time_to_live)
3059 				ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3060 			break;
3061 		case RTE_FLOW_ITEM_TYPE_IPV6:
3062 			ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3063 			if (!vlan && !eth)
3064 				return rte_flow_error_set(error, EINVAL,
3065 						RTE_FLOW_ERROR_TYPE_ACTION,
3066 						(void *)items->type,
3067 						"neither eth nor vlan"
3068 						" header found");
3069 			if (vlan && !vlan->eth_proto)
3070 				vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3071 			else if (eth && !eth->ether_type)
3072 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3073 			if (!ipv6->vtc_flow)
3074 				ipv6->vtc_flow =
3075 					RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3076 			if (!ipv6->hop_limits)
3077 				ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3078 			break;
3079 		case RTE_FLOW_ITEM_TYPE_UDP:
3080 			udp = (struct rte_udp_hdr *)&buf[temp_size];
3081 			if (!ipv4 && !ipv6)
3082 				return rte_flow_error_set(error, EINVAL,
3083 						RTE_FLOW_ERROR_TYPE_ACTION,
3084 						(void *)items->type,
3085 						"ip header not found");
3086 			if (ipv4 && !ipv4->next_proto_id)
3087 				ipv4->next_proto_id = IPPROTO_UDP;
3088 			else if (ipv6 && !ipv6->proto)
3089 				ipv6->proto = IPPROTO_UDP;
3090 			break;
3091 		case RTE_FLOW_ITEM_TYPE_VXLAN:
3092 			vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3093 			if (!udp)
3094 				return rte_flow_error_set(error, EINVAL,
3095 						RTE_FLOW_ERROR_TYPE_ACTION,
3096 						(void *)items->type,
3097 						"udp header not found");
3098 			if (!udp->dst_port)
3099 				udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3100 			if (!vxlan->vx_flags)
3101 				vxlan->vx_flags =
3102 					RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3103 			break;
3104 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3105 			vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3106 			if (!udp)
3107 				return rte_flow_error_set(error, EINVAL,
3108 						RTE_FLOW_ERROR_TYPE_ACTION,
3109 						(void *)items->type,
3110 						"udp header not found");
3111 			if (!vxlan_gpe->proto)
3112 				return rte_flow_error_set(error, EINVAL,
3113 						RTE_FLOW_ERROR_TYPE_ACTION,
3114 						(void *)items->type,
3115 						"next protocol not found");
3116 			if (!udp->dst_port)
3117 				udp->dst_port =
3118 					RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3119 			if (!vxlan_gpe->vx_flags)
3120 				vxlan_gpe->vx_flags =
3121 						MLX5_ENCAP_VXLAN_GPE_FLAGS;
3122 			break;
3123 		case RTE_FLOW_ITEM_TYPE_GRE:
3124 		case RTE_FLOW_ITEM_TYPE_NVGRE:
3125 			gre = (struct rte_gre_hdr *)&buf[temp_size];
3126 			if (!gre->proto)
3127 				return rte_flow_error_set(error, EINVAL,
3128 						RTE_FLOW_ERROR_TYPE_ACTION,
3129 						(void *)items->type,
3130 						"next protocol not found");
3131 			if (!ipv4 && !ipv6)
3132 				return rte_flow_error_set(error, EINVAL,
3133 						RTE_FLOW_ERROR_TYPE_ACTION,
3134 						(void *)items->type,
3135 						"ip header not found");
3136 			if (ipv4 && !ipv4->next_proto_id)
3137 				ipv4->next_proto_id = IPPROTO_GRE;
3138 			else if (ipv6 && !ipv6->proto)
3139 				ipv6->proto = IPPROTO_GRE;
3140 			break;
3141 		case RTE_FLOW_ITEM_TYPE_VOID:
3142 			break;
3143 		default:
3144 			return rte_flow_error_set(error, EINVAL,
3145 						  RTE_FLOW_ERROR_TYPE_ACTION,
3146 						  (void *)items->type,
3147 						  "unsupported item type");
3148 			break;
3149 		}
3150 		temp_size += len;
3151 	}
3152 	*size = temp_size;
3153 	return 0;
3154 }
3155 
3156 static int
3157 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3158 {
3159 	struct rte_ether_hdr *eth = NULL;
3160 	struct rte_vlan_hdr *vlan = NULL;
3161 	struct rte_ipv6_hdr *ipv6 = NULL;
3162 	struct rte_udp_hdr *udp = NULL;
3163 	char *next_hdr;
3164 	uint16_t proto;
3165 
3166 	eth = (struct rte_ether_hdr *)data;
3167 	next_hdr = (char *)(eth + 1);
3168 	proto = RTE_BE16(eth->ether_type);
3169 
3170 	/* VLAN skipping */
3171 	while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3172 		vlan = (struct rte_vlan_hdr *)next_hdr;
3173 		proto = RTE_BE16(vlan->eth_proto);
3174 		next_hdr += sizeof(struct rte_vlan_hdr);
3175 	}
3176 
3177 	/* HW calculates IPv4 csum. no need to proceed */
3178 	if (proto == RTE_ETHER_TYPE_IPV4)
3179 		return 0;
3180 
3181 	/* non IPv4/IPv6 header. not supported */
3182 	if (proto != RTE_ETHER_TYPE_IPV6) {
3183 		return rte_flow_error_set(error, ENOTSUP,
3184 					  RTE_FLOW_ERROR_TYPE_ACTION,
3185 					  NULL, "Cannot offload non IPv4/IPv6");
3186 	}
3187 
3188 	ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3189 
3190 	/* ignore non UDP */
3191 	if (ipv6->proto != IPPROTO_UDP)
3192 		return 0;
3193 
3194 	udp = (struct rte_udp_hdr *)(ipv6 + 1);
3195 	udp->dgram_cksum = 0;
3196 
3197 	return 0;
3198 }
3199 
3200 /**
3201  * Convert L2 encap action to DV specification.
3202  *
3203  * @param[in] dev
3204  *   Pointer to rte_eth_dev structure.
3205  * @param[in] action
3206  *   Pointer to action structure.
3207  * @param[in, out] dev_flow
3208  *   Pointer to the mlx5_flow.
3209  * @param[in] transfer
3210  *   Mark if the flow is E-Switch flow.
3211  * @param[out] error
3212  *   Pointer to the error structure.
3213  *
3214  * @return
3215  *   0 on success, a negative errno value otherwise and rte_errno is set.
3216  */
3217 static int
3218 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3219 			       const struct rte_flow_action *action,
3220 			       struct mlx5_flow *dev_flow,
3221 			       uint8_t transfer,
3222 			       struct rte_flow_error *error)
3223 {
3224 	const struct rte_flow_item *encap_data;
3225 	const struct rte_flow_action_raw_encap *raw_encap_data;
3226 	struct mlx5_flow_dv_encap_decap_resource res = {
3227 		.reformat_type =
3228 			MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3229 		.ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3230 				      MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3231 	};
3232 
3233 	if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3234 		raw_encap_data =
3235 			(const struct rte_flow_action_raw_encap *)action->conf;
3236 		res.size = raw_encap_data->size;
3237 		memcpy(res.buf, raw_encap_data->data, res.size);
3238 	} else {
3239 		if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3240 			encap_data =
3241 				((const struct rte_flow_action_vxlan_encap *)
3242 						action->conf)->definition;
3243 		else
3244 			encap_data =
3245 				((const struct rte_flow_action_nvgre_encap *)
3246 						action->conf)->definition;
3247 		if (flow_dv_convert_encap_data(encap_data, res.buf,
3248 					       &res.size, error))
3249 			return -rte_errno;
3250 	}
3251 	if (flow_dv_zero_encap_udp_csum(res.buf, error))
3252 		return -rte_errno;
3253 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3254 		return rte_flow_error_set(error, EINVAL,
3255 					  RTE_FLOW_ERROR_TYPE_ACTION,
3256 					  NULL, "can't create L2 encap action");
3257 	return 0;
3258 }
3259 
3260 /**
3261  * Convert L2 decap action to DV specification.
3262  *
3263  * @param[in] dev
3264  *   Pointer to rte_eth_dev structure.
3265  * @param[in, out] dev_flow
3266  *   Pointer to the mlx5_flow.
3267  * @param[in] transfer
3268  *   Mark if the flow is E-Switch flow.
3269  * @param[out] error
3270  *   Pointer to the error structure.
3271  *
3272  * @return
3273  *   0 on success, a negative errno value otherwise and rte_errno is set.
3274  */
3275 static int
3276 flow_dv_create_action_l2_decap(struct rte_eth_dev *dev,
3277 			       struct mlx5_flow *dev_flow,
3278 			       uint8_t transfer,
3279 			       struct rte_flow_error *error)
3280 {
3281 	struct mlx5_flow_dv_encap_decap_resource res = {
3282 		.size = 0,
3283 		.reformat_type =
3284 			MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3285 		.ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3286 				      MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3287 	};
3288 
3289 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3290 		return rte_flow_error_set(error, EINVAL,
3291 					  RTE_FLOW_ERROR_TYPE_ACTION,
3292 					  NULL, "can't create L2 decap action");
3293 	return 0;
3294 }
3295 
3296 /**
3297  * Convert raw decap/encap (L3 tunnel) action to DV specification.
3298  *
3299  * @param[in] dev
3300  *   Pointer to rte_eth_dev structure.
3301  * @param[in] action
3302  *   Pointer to action structure.
3303  * @param[in, out] dev_flow
3304  *   Pointer to the mlx5_flow.
3305  * @param[in] attr
3306  *   Pointer to the flow attributes.
3307  * @param[out] error
3308  *   Pointer to the error structure.
3309  *
3310  * @return
3311  *   0 on success, a negative errno value otherwise and rte_errno is set.
3312  */
3313 static int
3314 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3315 				const struct rte_flow_action *action,
3316 				struct mlx5_flow *dev_flow,
3317 				const struct rte_flow_attr *attr,
3318 				struct rte_flow_error *error)
3319 {
3320 	const struct rte_flow_action_raw_encap *encap_data;
3321 	struct mlx5_flow_dv_encap_decap_resource res;
3322 
3323 	memset(&res, 0, sizeof(res));
3324 	encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3325 	res.size = encap_data->size;
3326 	memcpy(res.buf, encap_data->data, res.size);
3327 	res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3328 		MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3329 		MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3330 	if (attr->transfer)
3331 		res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3332 	else
3333 		res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3334 					     MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3335 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3336 		return rte_flow_error_set(error, EINVAL,
3337 					  RTE_FLOW_ERROR_TYPE_ACTION,
3338 					  NULL, "can't create encap action");
3339 	return 0;
3340 }
3341 
3342 /**
3343  * Create action push VLAN.
3344  *
3345  * @param[in] dev
3346  *   Pointer to rte_eth_dev structure.
3347  * @param[in] attr
3348  *   Pointer to the flow attributes.
3349  * @param[in] vlan
3350  *   Pointer to the vlan to push to the Ethernet header.
3351  * @param[in, out] dev_flow
3352  *   Pointer to the mlx5_flow.
3353  * @param[out] error
3354  *   Pointer to the error structure.
3355  *
3356  * @return
3357  *   0 on success, a negative errno value otherwise and rte_errno is set.
3358  */
3359 static int
3360 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3361 				const struct rte_flow_attr *attr,
3362 				const struct rte_vlan_hdr *vlan,
3363 				struct mlx5_flow *dev_flow,
3364 				struct rte_flow_error *error)
3365 {
3366 	struct mlx5_flow_dv_push_vlan_action_resource res;
3367 
3368 	memset(&res, 0, sizeof(res));
3369 	res.vlan_tag =
3370 		rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3371 				 vlan->vlan_tci);
3372 	if (attr->transfer)
3373 		res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3374 	else
3375 		res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3376 					     MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3377 	return flow_dv_push_vlan_action_resource_register
3378 					    (dev, &res, dev_flow, error);
3379 }
3380 
3381 static int fdb_mirror;
3382 
3383 /**
3384  * Validate the modify-header actions.
3385  *
3386  * @param[in] action_flags
3387  *   Holds the actions detected until now.
3388  * @param[in] action
3389  *   Pointer to the modify action.
3390  * @param[out] error
3391  *   Pointer to error structure.
3392  *
3393  * @return
3394  *   0 on success, a negative errno value otherwise and rte_errno is set.
3395  */
3396 static int
3397 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3398 				   const struct rte_flow_action *action,
3399 				   struct rte_flow_error *error)
3400 {
3401 	if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3402 		return rte_flow_error_set(error, EINVAL,
3403 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3404 					  NULL, "action configuration not set");
3405 	if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3406 		return rte_flow_error_set(error, EINVAL,
3407 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3408 					  "can't have encap action before"
3409 					  " modify action");
3410 	if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3411 		return rte_flow_error_set(error, EINVAL,
3412 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3413 					  "can't support sample action before"
3414 					  " modify action for E-Switch"
3415 					  " mirroring");
3416 	return 0;
3417 }
3418 
3419 /**
3420  * Validate the modify-header MAC address actions.
3421  *
3422  * @param[in] action_flags
3423  *   Holds the actions detected until now.
3424  * @param[in] action
3425  *   Pointer to the modify action.
3426  * @param[in] item_flags
3427  *   Holds the items detected.
3428  * @param[out] error
3429  *   Pointer to error structure.
3430  *
3431  * @return
3432  *   0 on success, a negative errno value otherwise and rte_errno is set.
3433  */
3434 static int
3435 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3436 				   const struct rte_flow_action *action,
3437 				   const uint64_t item_flags,
3438 				   struct rte_flow_error *error)
3439 {
3440 	int ret = 0;
3441 
3442 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3443 	if (!ret) {
3444 		if (!(item_flags & MLX5_FLOW_LAYER_L2))
3445 			return rte_flow_error_set(error, EINVAL,
3446 						  RTE_FLOW_ERROR_TYPE_ACTION,
3447 						  NULL,
3448 						  "no L2 item in pattern");
3449 	}
3450 	return ret;
3451 }
3452 
3453 /**
3454  * Validate the modify-header IPv4 address actions.
3455  *
3456  * @param[in] action_flags
3457  *   Holds the actions detected until now.
3458  * @param[in] action
3459  *   Pointer to the modify action.
3460  * @param[in] item_flags
3461  *   Holds the items detected.
3462  * @param[out] error
3463  *   Pointer to error structure.
3464  *
3465  * @return
3466  *   0 on success, a negative errno value otherwise and rte_errno is set.
3467  */
3468 static int
3469 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3470 				    const struct rte_flow_action *action,
3471 				    const uint64_t item_flags,
3472 				    struct rte_flow_error *error)
3473 {
3474 	int ret = 0;
3475 	uint64_t layer;
3476 
3477 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3478 	if (!ret) {
3479 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3480 				 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3481 				 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3482 		if (!(item_flags & layer))
3483 			return rte_flow_error_set(error, EINVAL,
3484 						  RTE_FLOW_ERROR_TYPE_ACTION,
3485 						  NULL,
3486 						  "no ipv4 item in pattern");
3487 	}
3488 	return ret;
3489 }
3490 
3491 /**
3492  * Validate the modify-header IPv6 address actions.
3493  *
3494  * @param[in] action_flags
3495  *   Holds the actions detected until now.
3496  * @param[in] action
3497  *   Pointer to the modify action.
3498  * @param[in] item_flags
3499  *   Holds the items detected.
3500  * @param[out] error
3501  *   Pointer to error structure.
3502  *
3503  * @return
3504  *   0 on success, a negative errno value otherwise and rte_errno is set.
3505  */
3506 static int
3507 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3508 				    const struct rte_flow_action *action,
3509 				    const uint64_t item_flags,
3510 				    struct rte_flow_error *error)
3511 {
3512 	int ret = 0;
3513 	uint64_t layer;
3514 
3515 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3516 	if (!ret) {
3517 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3518 				 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3519 				 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3520 		if (!(item_flags & layer))
3521 			return rte_flow_error_set(error, EINVAL,
3522 						  RTE_FLOW_ERROR_TYPE_ACTION,
3523 						  NULL,
3524 						  "no ipv6 item in pattern");
3525 	}
3526 	return ret;
3527 }
3528 
3529 /**
3530  * Validate the modify-header TP actions.
3531  *
3532  * @param[in] action_flags
3533  *   Holds the actions detected until now.
3534  * @param[in] action
3535  *   Pointer to the modify action.
3536  * @param[in] item_flags
3537  *   Holds the items detected.
3538  * @param[out] error
3539  *   Pointer to error structure.
3540  *
3541  * @return
3542  *   0 on success, a negative errno value otherwise and rte_errno is set.
3543  */
3544 static int
3545 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3546 				  const struct rte_flow_action *action,
3547 				  const uint64_t item_flags,
3548 				  struct rte_flow_error *error)
3549 {
3550 	int ret = 0;
3551 	uint64_t layer;
3552 
3553 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3554 	if (!ret) {
3555 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3556 				 MLX5_FLOW_LAYER_INNER_L4 :
3557 				 MLX5_FLOW_LAYER_OUTER_L4;
3558 		if (!(item_flags & layer))
3559 			return rte_flow_error_set(error, EINVAL,
3560 						  RTE_FLOW_ERROR_TYPE_ACTION,
3561 						  NULL, "no transport layer "
3562 						  "in pattern");
3563 	}
3564 	return ret;
3565 }
3566 
3567 /**
3568  * Validate the modify-header actions of increment/decrement
3569  * TCP Sequence-number.
3570  *
3571  * @param[in] action_flags
3572  *   Holds the actions detected until now.
3573  * @param[in] action
3574  *   Pointer to the modify action.
3575  * @param[in] item_flags
3576  *   Holds the items detected.
3577  * @param[out] error
3578  *   Pointer to error structure.
3579  *
3580  * @return
3581  *   0 on success, a negative errno value otherwise and rte_errno is set.
3582  */
3583 static int
3584 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3585 				       const struct rte_flow_action *action,
3586 				       const uint64_t item_flags,
3587 				       struct rte_flow_error *error)
3588 {
3589 	int ret = 0;
3590 	uint64_t layer;
3591 
3592 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3593 	if (!ret) {
3594 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3595 				 MLX5_FLOW_LAYER_INNER_L4_TCP :
3596 				 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3597 		if (!(item_flags & layer))
3598 			return rte_flow_error_set(error, EINVAL,
3599 						  RTE_FLOW_ERROR_TYPE_ACTION,
3600 						  NULL, "no TCP item in"
3601 						  " pattern");
3602 		if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3603 			(action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3604 		    (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3605 			(action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3606 			return rte_flow_error_set(error, EINVAL,
3607 						  RTE_FLOW_ERROR_TYPE_ACTION,
3608 						  NULL,
3609 						  "cannot decrease and increase"
3610 						  " TCP sequence number"
3611 						  " at the same time");
3612 	}
3613 	return ret;
3614 }
3615 
3616 /**
3617  * Validate the modify-header actions of increment/decrement
3618  * TCP Acknowledgment number.
3619  *
3620  * @param[in] action_flags
3621  *   Holds the actions detected until now.
3622  * @param[in] action
3623  *   Pointer to the modify action.
3624  * @param[in] item_flags
3625  *   Holds the items detected.
3626  * @param[out] error
3627  *   Pointer to error structure.
3628  *
3629  * @return
3630  *   0 on success, a negative errno value otherwise and rte_errno is set.
3631  */
3632 static int
3633 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3634 				       const struct rte_flow_action *action,
3635 				       const uint64_t item_flags,
3636 				       struct rte_flow_error *error)
3637 {
3638 	int ret = 0;
3639 	uint64_t layer;
3640 
3641 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3642 	if (!ret) {
3643 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3644 				 MLX5_FLOW_LAYER_INNER_L4_TCP :
3645 				 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3646 		if (!(item_flags & layer))
3647 			return rte_flow_error_set(error, EINVAL,
3648 						  RTE_FLOW_ERROR_TYPE_ACTION,
3649 						  NULL, "no TCP item in"
3650 						  " pattern");
3651 		if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3652 			(action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3653 		    (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3654 			(action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3655 			return rte_flow_error_set(error, EINVAL,
3656 						  RTE_FLOW_ERROR_TYPE_ACTION,
3657 						  NULL,
3658 						  "cannot decrease and increase"
3659 						  " TCP acknowledgment number"
3660 						  " at the same time");
3661 	}
3662 	return ret;
3663 }
3664 
3665 /**
3666  * Validate the modify-header TTL actions.
3667  *
3668  * @param[in] action_flags
3669  *   Holds the actions detected until now.
3670  * @param[in] action
3671  *   Pointer to the modify action.
3672  * @param[in] item_flags
3673  *   Holds the items detected.
3674  * @param[out] error
3675  *   Pointer to error structure.
3676  *
3677  * @return
3678  *   0 on success, a negative errno value otherwise and rte_errno is set.
3679  */
3680 static int
3681 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3682 				   const struct rte_flow_action *action,
3683 				   const uint64_t item_flags,
3684 				   struct rte_flow_error *error)
3685 {
3686 	int ret = 0;
3687 	uint64_t layer;
3688 
3689 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3690 	if (!ret) {
3691 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3692 				 MLX5_FLOW_LAYER_INNER_L3 :
3693 				 MLX5_FLOW_LAYER_OUTER_L3;
3694 		if (!(item_flags & layer))
3695 			return rte_flow_error_set(error, EINVAL,
3696 						  RTE_FLOW_ERROR_TYPE_ACTION,
3697 						  NULL,
3698 						  "no IP protocol in pattern");
3699 	}
3700 	return ret;
3701 }
3702 
3703 /**
3704  * Validate jump action.
3705  *
3706  * @param[in] action
3707  *   Pointer to the jump action.
3708  * @param[in] action_flags
3709  *   Holds the actions detected until now.
3710  * @param[in] attributes
3711  *   Pointer to flow attributes
3712  * @param[in] external
3713  *   Action belongs to flow rule created by request external to PMD.
3714  * @param[out] error
3715  *   Pointer to error structure.
3716  *
3717  * @return
3718  *   0 on success, a negative errno value otherwise and rte_errno is set.
3719  */
3720 static int
3721 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3722 			     uint64_t action_flags,
3723 			     const struct rte_flow_attr *attributes,
3724 			     bool external, struct rte_flow_error *error)
3725 {
3726 	uint32_t target_group, table;
3727 	int ret = 0;
3728 
3729 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3730 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3731 		return rte_flow_error_set(error, EINVAL,
3732 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3733 					  "can't have 2 fate actions in"
3734 					  " same flow");
3735 	if (action_flags & MLX5_FLOW_ACTION_METER)
3736 		return rte_flow_error_set(error, ENOTSUP,
3737 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3738 					  "jump with meter not support");
3739 	if ((action_flags & MLX5_FLOW_ACTION_SAMPLE) && fdb_mirror)
3740 		return rte_flow_error_set(error, EINVAL,
3741 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3742 					  "E-Switch mirroring can't support"
3743 					  " Sample action and jump action in"
3744 					  " same flow now");
3745 	if (!action->conf)
3746 		return rte_flow_error_set(error, EINVAL,
3747 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3748 					  NULL, "action configuration not set");
3749 	target_group =
3750 		((const struct rte_flow_action_jump *)action->conf)->group;
3751 	ret = mlx5_flow_group_to_table(attributes, external, target_group,
3752 				       true, &table, error);
3753 	if (ret)
3754 		return ret;
3755 	if (attributes->group == target_group)
3756 		return rte_flow_error_set(error, EINVAL,
3757 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3758 					  "target group must be other than"
3759 					  " the current flow group");
3760 	return 0;
3761 }
3762 
3763 /*
3764  * Validate the port_id action.
3765  *
3766  * @param[in] dev
3767  *   Pointer to rte_eth_dev structure.
3768  * @param[in] action_flags
3769  *   Bit-fields that holds the actions detected until now.
3770  * @param[in] action
3771  *   Port_id RTE action structure.
3772  * @param[in] attr
3773  *   Attributes of flow that includes this action.
3774  * @param[out] error
3775  *   Pointer to error structure.
3776  *
3777  * @return
3778  *   0 on success, a negative errno value otherwise and rte_errno is set.
3779  */
3780 static int
3781 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3782 				uint64_t action_flags,
3783 				const struct rte_flow_action *action,
3784 				const struct rte_flow_attr *attr,
3785 				struct rte_flow_error *error)
3786 {
3787 	const struct rte_flow_action_port_id *port_id;
3788 	struct mlx5_priv *act_priv;
3789 	struct mlx5_priv *dev_priv;
3790 	uint16_t port;
3791 
3792 	if (!attr->transfer)
3793 		return rte_flow_error_set(error, ENOTSUP,
3794 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3795 					  NULL,
3796 					  "port id action is valid in transfer"
3797 					  " mode only");
3798 	if (!action || !action->conf)
3799 		return rte_flow_error_set(error, ENOTSUP,
3800 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3801 					  NULL,
3802 					  "port id action parameters must be"
3803 					  " specified");
3804 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3805 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3806 		return rte_flow_error_set(error, EINVAL,
3807 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3808 					  "can have only one fate actions in"
3809 					  " a flow");
3810 	dev_priv = mlx5_dev_to_eswitch_info(dev);
3811 	if (!dev_priv)
3812 		return rte_flow_error_set(error, rte_errno,
3813 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3814 					  NULL,
3815 					  "failed to obtain E-Switch info");
3816 	port_id = action->conf;
3817 	port = port_id->original ? dev->data->port_id : port_id->id;
3818 	act_priv = mlx5_port_to_eswitch_info(port, false);
3819 	if (!act_priv)
3820 		return rte_flow_error_set
3821 				(error, rte_errno,
3822 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3823 				 "failed to obtain E-Switch port id for port");
3824 	if (act_priv->domain_id != dev_priv->domain_id)
3825 		return rte_flow_error_set
3826 				(error, EINVAL,
3827 				 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3828 				 "port does not belong to"
3829 				 " E-Switch being configured");
3830 	return 0;
3831 }
3832 
3833 /**
3834  * Get the maximum number of modify header actions.
3835  *
3836  * @param dev
3837  *   Pointer to rte_eth_dev structure.
3838  * @param flags
3839  *   Flags bits to check if root level.
3840  *
3841  * @return
3842  *   Max number of modify header actions device can support.
3843  */
3844 static inline unsigned int
3845 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
3846 			      uint64_t flags)
3847 {
3848 	/*
3849 	 * There's no way to directly query the max capacity from FW.
3850 	 * The maximal value on root table should be assumed to be supported.
3851 	 */
3852 	if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3853 		return MLX5_MAX_MODIFY_NUM;
3854 	else
3855 		return MLX5_ROOT_TBL_MODIFY_NUM;
3856 }
3857 
3858 /**
3859  * Validate the meter action.
3860  *
3861  * @param[in] dev
3862  *   Pointer to rte_eth_dev structure.
3863  * @param[in] action_flags
3864  *   Bit-fields that holds the actions detected until now.
3865  * @param[in] action
3866  *   Pointer to the meter action.
3867  * @param[in] attr
3868  *   Attributes of flow that includes this action.
3869  * @param[out] error
3870  *   Pointer to error structure.
3871  *
3872  * @return
3873  *   0 on success, a negative errno value otherwise and rte_ernno is set.
3874  */
3875 static int
3876 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3877 				uint64_t action_flags,
3878 				const struct rte_flow_action *action,
3879 				const struct rte_flow_attr *attr,
3880 				struct rte_flow_error *error)
3881 {
3882 	struct mlx5_priv *priv = dev->data->dev_private;
3883 	const struct rte_flow_action_meter *am = action->conf;
3884 	struct mlx5_flow_meter *fm;
3885 
3886 	if (!am)
3887 		return rte_flow_error_set(error, EINVAL,
3888 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3889 					  "meter action conf is NULL");
3890 
3891 	if (action_flags & MLX5_FLOW_ACTION_METER)
3892 		return rte_flow_error_set(error, ENOTSUP,
3893 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3894 					  "meter chaining not support");
3895 	if (action_flags & MLX5_FLOW_ACTION_JUMP)
3896 		return rte_flow_error_set(error, ENOTSUP,
3897 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3898 					  "meter with jump not support");
3899 	if (!priv->mtr_en)
3900 		return rte_flow_error_set(error, ENOTSUP,
3901 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3902 					  NULL,
3903 					  "meter action not supported");
3904 	fm = mlx5_flow_meter_find(priv, am->mtr_id);
3905 	if (!fm)
3906 		return rte_flow_error_set(error, EINVAL,
3907 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3908 					  "Meter not found");
3909 	if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3910 	      (!fm->ingress && !attr->ingress && attr->egress) ||
3911 	      (!fm->egress && !attr->egress && attr->ingress))))
3912 		return rte_flow_error_set(error, EINVAL,
3913 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3914 					  "Flow attributes are either invalid "
3915 					  "or have a conflict with current "
3916 					  "meter attributes");
3917 	return 0;
3918 }
3919 
3920 /**
3921  * Validate the age action.
3922  *
3923  * @param[in] action_flags
3924  *   Holds the actions detected until now.
3925  * @param[in] action
3926  *   Pointer to the age action.
3927  * @param[in] dev
3928  *   Pointer to the Ethernet device structure.
3929  * @param[out] error
3930  *   Pointer to error structure.
3931  *
3932  * @return
3933  *   0 on success, a negative errno value otherwise and rte_errno is set.
3934  */
3935 static int
3936 flow_dv_validate_action_age(uint64_t action_flags,
3937 			    const struct rte_flow_action *action,
3938 			    struct rte_eth_dev *dev,
3939 			    struct rte_flow_error *error)
3940 {
3941 	struct mlx5_priv *priv = dev->data->dev_private;
3942 	const struct rte_flow_action_age *age = action->conf;
3943 
3944 	if (!priv->config.devx || priv->counter_fallback)
3945 		return rte_flow_error_set(error, ENOTSUP,
3946 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3947 					  NULL,
3948 					  "age action not supported");
3949 	if (!(action->conf))
3950 		return rte_flow_error_set(error, EINVAL,
3951 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
3952 					  "configuration cannot be null");
3953 	if (age->timeout >= UINT16_MAX / 2 / 10)
3954 		return rte_flow_error_set(error, ENOTSUP,
3955 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
3956 					  "Max age time: 3275 seconds");
3957 	if (action_flags & MLX5_FLOW_ACTION_AGE)
3958 		return rte_flow_error_set(error, EINVAL,
3959 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3960 					  "Duplicate age ctions set");
3961 	return 0;
3962 }
3963 
3964 /**
3965  * Validate the modify-header IPv4 DSCP actions.
3966  *
3967  * @param[in] action_flags
3968  *   Holds the actions detected until now.
3969  * @param[in] action
3970  *   Pointer to the modify action.
3971  * @param[in] item_flags
3972  *   Holds the items detected.
3973  * @param[out] error
3974  *   Pointer to error structure.
3975  *
3976  * @return
3977  *   0 on success, a negative errno value otherwise and rte_errno is set.
3978  */
3979 static int
3980 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3981 					 const struct rte_flow_action *action,
3982 					 const uint64_t item_flags,
3983 					 struct rte_flow_error *error)
3984 {
3985 	int ret = 0;
3986 
3987 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3988 	if (!ret) {
3989 		if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3990 			return rte_flow_error_set(error, EINVAL,
3991 						  RTE_FLOW_ERROR_TYPE_ACTION,
3992 						  NULL,
3993 						  "no ipv4 item in pattern");
3994 	}
3995 	return ret;
3996 }
3997 
3998 /**
3999  * Validate the modify-header IPv6 DSCP actions.
4000  *
4001  * @param[in] action_flags
4002  *   Holds the actions detected until now.
4003  * @param[in] action
4004  *   Pointer to the modify action.
4005  * @param[in] item_flags
4006  *   Holds the items detected.
4007  * @param[out] error
4008  *   Pointer to error structure.
4009  *
4010  * @return
4011  *   0 on success, a negative errno value otherwise and rte_errno is set.
4012  */
4013 static int
4014 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
4015 					 const struct rte_flow_action *action,
4016 					 const uint64_t item_flags,
4017 					 struct rte_flow_error *error)
4018 {
4019 	int ret = 0;
4020 
4021 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
4022 	if (!ret) {
4023 		if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
4024 			return rte_flow_error_set(error, EINVAL,
4025 						  RTE_FLOW_ERROR_TYPE_ACTION,
4026 						  NULL,
4027 						  "no ipv6 item in pattern");
4028 	}
4029 	return ret;
4030 }
4031 
4032 /**
4033  * Match modify-header resource.
4034  *
4035  * @param entry
4036  *   Pointer to exist resource entry object.
4037  * @param ctx
4038  *   Pointer to new modify-header resource.
4039  *
4040  * @return
4041  *   0 on matching, -1 otherwise.
4042  */
4043 static int
4044 flow_dv_modify_hdr_resource_match(struct mlx5_hlist_entry *entry, void *ctx)
4045 {
4046 	struct mlx5_flow_dv_modify_hdr_resource *resource;
4047 	struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
4048 	uint32_t actions_len;
4049 
4050 	resource = (struct mlx5_flow_dv_modify_hdr_resource *)ctx;
4051 	cache_resource = container_of(entry,
4052 				      struct mlx5_flow_dv_modify_hdr_resource,
4053 				      entry);
4054 	actions_len = resource->actions_num * sizeof(resource->actions[0]);
4055 	if (resource->entry.key == cache_resource->entry.key &&
4056 	    resource->ft_type == cache_resource->ft_type &&
4057 	    resource->actions_num == cache_resource->actions_num &&
4058 	    resource->flags == cache_resource->flags &&
4059 	    !memcmp((const void *)resource->actions,
4060 		    (const void *)cache_resource->actions,
4061 		    actions_len))
4062 		return 0;
4063 	return -1;
4064 }
4065 
4066 /**
4067  * Validate the sample action.
4068  *
4069  * @param[in] action_flags
4070  *   Holds the actions detected until now.
4071  * @param[in] action
4072  *   Pointer to the sample action.
4073  * @param[in] dev
4074  *   Pointer to the Ethernet device structure.
4075  * @param[in] attr
4076  *   Attributes of flow that includes this action.
4077  * @param[out] error
4078  *   Pointer to error structure.
4079  *
4080  * @return
4081  *   0 on success, a negative errno value otherwise and rte_errno is set.
4082  */
4083 static int
4084 flow_dv_validate_action_sample(uint64_t action_flags,
4085 			       const struct rte_flow_action *action,
4086 			       struct rte_eth_dev *dev,
4087 			       const struct rte_flow_attr *attr,
4088 			       struct rte_flow_error *error)
4089 {
4090 	struct mlx5_priv *priv = dev->data->dev_private;
4091 	struct mlx5_dev_config *dev_conf = &priv->config;
4092 	const struct rte_flow_action_sample *sample = action->conf;
4093 	const struct rte_flow_action *act;
4094 	uint64_t sub_action_flags = 0;
4095 	uint16_t queue_index = 0xFFFF;
4096 	int actions_n = 0;
4097 	int ret;
4098 	fdb_mirror = 0;
4099 
4100 	if (!sample)
4101 		return rte_flow_error_set(error, EINVAL,
4102 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
4103 					  "configuration cannot be NULL");
4104 	if (sample->ratio == 0)
4105 		return rte_flow_error_set(error, EINVAL,
4106 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
4107 					  "ratio value starts from 1");
4108 	if (!priv->config.devx || (sample->ratio > 0 && !priv->sampler_en))
4109 		return rte_flow_error_set(error, ENOTSUP,
4110 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4111 					  NULL,
4112 					  "sample action not supported");
4113 	if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
4114 		return rte_flow_error_set(error, EINVAL,
4115 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4116 					  "Multiple sample actions not "
4117 					  "supported");
4118 	if (action_flags & MLX5_FLOW_ACTION_METER)
4119 		return rte_flow_error_set(error, EINVAL,
4120 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
4121 					  "wrong action order, meter should "
4122 					  "be after sample action");
4123 	if (action_flags & MLX5_FLOW_ACTION_JUMP)
4124 		return rte_flow_error_set(error, EINVAL,
4125 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
4126 					  "wrong action order, jump should "
4127 					  "be after sample action");
4128 	act = sample->actions;
4129 	for (; act->type != RTE_FLOW_ACTION_TYPE_END; act++) {
4130 		if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
4131 			return rte_flow_error_set(error, ENOTSUP,
4132 						  RTE_FLOW_ERROR_TYPE_ACTION,
4133 						  act, "too many actions");
4134 		switch (act->type) {
4135 		case RTE_FLOW_ACTION_TYPE_QUEUE:
4136 			ret = mlx5_flow_validate_action_queue(act,
4137 							      sub_action_flags,
4138 							      dev,
4139 							      attr, error);
4140 			if (ret < 0)
4141 				return ret;
4142 			queue_index = ((const struct rte_flow_action_queue *)
4143 							(act->conf))->index;
4144 			sub_action_flags |= MLX5_FLOW_ACTION_QUEUE;
4145 			++actions_n;
4146 			break;
4147 		case RTE_FLOW_ACTION_TYPE_MARK:
4148 			ret = flow_dv_validate_action_mark(dev, act,
4149 							   sub_action_flags,
4150 							   attr, error);
4151 			if (ret < 0)
4152 				return ret;
4153 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY)
4154 				sub_action_flags |= MLX5_FLOW_ACTION_MARK |
4155 						MLX5_FLOW_ACTION_MARK_EXT;
4156 			else
4157 				sub_action_flags |= MLX5_FLOW_ACTION_MARK;
4158 			++actions_n;
4159 			break;
4160 		case RTE_FLOW_ACTION_TYPE_COUNT:
4161 			ret = flow_dv_validate_action_count(dev, error);
4162 			if (ret < 0)
4163 				return ret;
4164 			sub_action_flags |= MLX5_FLOW_ACTION_COUNT;
4165 			++actions_n;
4166 			break;
4167 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
4168 			ret = flow_dv_validate_action_port_id(dev,
4169 							      sub_action_flags,
4170 							      act,
4171 							      attr,
4172 							      error);
4173 			if (ret)
4174 				return ret;
4175 			sub_action_flags |= MLX5_FLOW_ACTION_PORT_ID;
4176 			++actions_n;
4177 			break;
4178 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
4179 			ret = flow_dv_validate_action_raw_encap_decap
4180 				(dev, NULL, act->conf, attr, &sub_action_flags,
4181 				 &actions_n, error);
4182 			if (ret < 0)
4183 				return ret;
4184 			++actions_n;
4185 			break;
4186 		default:
4187 			return rte_flow_error_set(error, ENOTSUP,
4188 						  RTE_FLOW_ERROR_TYPE_ACTION,
4189 						  NULL,
4190 						  "Doesn't support optional "
4191 						  "action");
4192 		}
4193 	}
4194 	if (attr->ingress && !attr->transfer) {
4195 		if (!(sub_action_flags & MLX5_FLOW_ACTION_QUEUE))
4196 			return rte_flow_error_set(error, EINVAL,
4197 						  RTE_FLOW_ERROR_TYPE_ACTION,
4198 						  NULL,
4199 						  "Ingress must has a dest "
4200 						  "QUEUE for Sample");
4201 	} else if (attr->egress && !attr->transfer) {
4202 		return rte_flow_error_set(error, ENOTSUP,
4203 					  RTE_FLOW_ERROR_TYPE_ACTION,
4204 					  NULL,
4205 					  "Sample Only support Ingress "
4206 					  "or E-Switch");
4207 	} else if (sample->actions->type != RTE_FLOW_ACTION_TYPE_END) {
4208 		MLX5_ASSERT(attr->transfer);
4209 		if (sample->ratio > 1)
4210 			return rte_flow_error_set(error, ENOTSUP,
4211 						  RTE_FLOW_ERROR_TYPE_ACTION,
4212 						  NULL,
4213 						  "E-Switch doesn't support "
4214 						  "any optional action "
4215 						  "for sampling");
4216 		fdb_mirror = 1;
4217 		if (sub_action_flags & MLX5_FLOW_ACTION_QUEUE)
4218 			return rte_flow_error_set(error, ENOTSUP,
4219 						  RTE_FLOW_ERROR_TYPE_ACTION,
4220 						  NULL,
4221 						  "unsupported action QUEUE");
4222 		if (!(sub_action_flags & MLX5_FLOW_ACTION_PORT_ID))
4223 			return rte_flow_error_set(error, EINVAL,
4224 						  RTE_FLOW_ERROR_TYPE_ACTION,
4225 						  NULL,
4226 						  "E-Switch must has a dest "
4227 						  "port for mirroring");
4228 	}
4229 	/* Continue validation for Xcap actions.*/
4230 	if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) &&
4231 	    (queue_index == 0xFFFF ||
4232 	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
4233 		if ((sub_action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
4234 		     MLX5_FLOW_XCAP_ACTIONS)
4235 			return rte_flow_error_set(error, ENOTSUP,
4236 						  RTE_FLOW_ERROR_TYPE_ACTION,
4237 						  NULL, "encap and decap "
4238 						  "combination aren't "
4239 						  "supported");
4240 		if (!attr->transfer && attr->ingress && (sub_action_flags &
4241 							MLX5_FLOW_ACTION_ENCAP))
4242 			return rte_flow_error_set(error, ENOTSUP,
4243 						  RTE_FLOW_ERROR_TYPE_ACTION,
4244 						  NULL, "encap is not supported"
4245 						  " for ingress traffic");
4246 	}
4247 	return 0;
4248 }
4249 
4250 /**
4251  * Find existing modify-header resource or create and register a new one.
4252  *
4253  * @param dev[in, out]
4254  *   Pointer to rte_eth_dev structure.
4255  * @param[in, out] resource
4256  *   Pointer to modify-header resource.
4257  * @parm[in, out] dev_flow
4258  *   Pointer to the dev_flow.
4259  * @param[out] error
4260  *   pointer to error structure.
4261  *
4262  * @return
4263  *   0 on success otherwise -errno and errno is set.
4264  */
4265 static int
4266 flow_dv_modify_hdr_resource_register
4267 			(struct rte_eth_dev *dev,
4268 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
4269 			 struct mlx5_flow *dev_flow,
4270 			 struct rte_flow_error *error)
4271 {
4272 	struct mlx5_priv *priv = dev->data->dev_private;
4273 	struct mlx5_dev_ctx_shared *sh = priv->sh;
4274 	struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
4275 	struct mlx5dv_dr_domain *ns;
4276 	uint32_t actions_len;
4277 	struct mlx5_hlist_entry *entry;
4278 	union mlx5_flow_modify_hdr_key hdr_mod_key = {
4279 		{
4280 			.ft_type = resource->ft_type,
4281 			.actions_num = resource->actions_num,
4282 			.group = dev_flow->dv.group,
4283 			.cksum = 0,
4284 		}
4285 	};
4286 	int ret;
4287 
4288 	resource->flags = dev_flow->dv.group ? 0 :
4289 			  MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4290 	if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
4291 				    resource->flags))
4292 		return rte_flow_error_set(error, EOVERFLOW,
4293 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
4294 					  "too many modify header items");
4295 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
4296 		ns = sh->fdb_domain;
4297 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
4298 		ns = sh->tx_domain;
4299 	else
4300 		ns = sh->rx_domain;
4301 	/* Lookup a matching resource from cache. */
4302 	actions_len = resource->actions_num * sizeof(resource->actions[0]);
4303 	hdr_mod_key.cksum = __rte_raw_cksum(resource->actions, actions_len, 0);
4304 	resource->entry.key = hdr_mod_key.v64;
4305 	entry = mlx5_hlist_lookup_ex(sh->modify_cmds, resource->entry.key,
4306 				     flow_dv_modify_hdr_resource_match,
4307 				     (void *)resource);
4308 	if (entry) {
4309 		cache_resource = container_of(entry,
4310 					struct mlx5_flow_dv_modify_hdr_resource,
4311 					entry);
4312 		DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
4313 			(void *)cache_resource,
4314 			rte_atomic32_read(&cache_resource->refcnt));
4315 		rte_atomic32_inc(&cache_resource->refcnt);
4316 		dev_flow->handle->dvh.modify_hdr = cache_resource;
4317 		return 0;
4318 
4319 	}
4320 	/* Register new modify-header resource. */
4321 	cache_resource = mlx5_malloc(MLX5_MEM_ZERO,
4322 				    sizeof(*cache_resource) + actions_len, 0,
4323 				    SOCKET_ID_ANY);
4324 	if (!cache_resource)
4325 		return rte_flow_error_set(error, ENOMEM,
4326 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4327 					  "cannot allocate resource memory");
4328 	*cache_resource = *resource;
4329 	rte_memcpy(cache_resource->actions, resource->actions, actions_len);
4330 	ret = mlx5_flow_os_create_flow_action_modify_header
4331 					(sh->ctx, ns, cache_resource,
4332 					 actions_len, &cache_resource->action);
4333 	if (ret) {
4334 		mlx5_free(cache_resource);
4335 		return rte_flow_error_set(error, ENOMEM,
4336 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4337 					  NULL, "cannot create action");
4338 	}
4339 	rte_atomic32_init(&cache_resource->refcnt);
4340 	rte_atomic32_inc(&cache_resource->refcnt);
4341 	if (mlx5_hlist_insert_ex(sh->modify_cmds, &cache_resource->entry,
4342 				 flow_dv_modify_hdr_resource_match,
4343 				 (void *)cache_resource)) {
4344 		claim_zero(mlx5_flow_os_destroy_flow_action
4345 						(cache_resource->action));
4346 		mlx5_free(cache_resource);
4347 		return rte_flow_error_set(error, EEXIST,
4348 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4349 					  NULL, "action exist");
4350 	}
4351 	dev_flow->handle->dvh.modify_hdr = cache_resource;
4352 	DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
4353 		(void *)cache_resource,
4354 		rte_atomic32_read(&cache_resource->refcnt));
4355 	return 0;
4356 }
4357 
4358 /**
4359  * Get DV flow counter by index.
4360  *
4361  * @param[in] dev
4362  *   Pointer to the Ethernet device structure.
4363  * @param[in] idx
4364  *   mlx5 flow counter index in the container.
4365  * @param[out] ppool
4366  *   mlx5 flow counter pool in the container,
4367  *
4368  * @return
4369  *   Pointer to the counter, NULL otherwise.
4370  */
4371 static struct mlx5_flow_counter *
4372 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4373 			   uint32_t idx,
4374 			   struct mlx5_flow_counter_pool **ppool)
4375 {
4376 	struct mlx5_priv *priv = dev->data->dev_private;
4377 	struct mlx5_pools_container *cont;
4378 	struct mlx5_flow_counter_pool *pool;
4379 	uint32_t batch = 0, age = 0;
4380 
4381 	idx--;
4382 	age = MLX_CNT_IS_AGE(idx);
4383 	idx = age ? idx - MLX5_CNT_AGE_OFFSET : idx;
4384 	if (idx >= MLX5_CNT_BATCH_OFFSET) {
4385 		idx -= MLX5_CNT_BATCH_OFFSET;
4386 		batch = 1;
4387 	}
4388 	cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4389 	MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
4390 	pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
4391 	MLX5_ASSERT(pool);
4392 	if (ppool)
4393 		*ppool = pool;
4394 	return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4395 }
4396 
4397 /**
4398  * Check the devx counter belongs to the pool.
4399  *
4400  * @param[in] pool
4401  *   Pointer to the counter pool.
4402  * @param[in] id
4403  *   The counter devx ID.
4404  *
4405  * @return
4406  *   True if counter belongs to the pool, false otherwise.
4407  */
4408 static bool
4409 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4410 {
4411 	int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4412 		   MLX5_COUNTERS_PER_POOL;
4413 
4414 	if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4415 		return true;
4416 	return false;
4417 }
4418 
4419 /**
4420  * Get a pool by devx counter ID.
4421  *
4422  * @param[in] cont
4423  *   Pointer to the counter container.
4424  * @param[in] id
4425  *   The counter devx ID.
4426  *
4427  * @return
4428  *   The counter pool pointer if exists, NULL otherwise,
4429  */
4430 static struct mlx5_flow_counter_pool *
4431 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
4432 {
4433 	uint32_t i;
4434 
4435 	/* Check last used pool. */
4436 	if (cont->last_pool_idx != POOL_IDX_INVALID &&
4437 	    flow_dv_is_counter_in_pool(cont->pools[cont->last_pool_idx], id))
4438 		return cont->pools[cont->last_pool_idx];
4439 	/* ID out of range means no suitable pool in the container. */
4440 	if (id > cont->max_id || id < cont->min_id)
4441 		return NULL;
4442 	/*
4443 	 * Find the pool from the end of the container, since mostly counter
4444 	 * ID is sequence increasing, and the last pool should be the needed
4445 	 * one.
4446 	 */
4447 	i = rte_atomic16_read(&cont->n_valid);
4448 	while (i--) {
4449 		struct mlx5_flow_counter_pool *pool = cont->pools[i];
4450 
4451 		if (flow_dv_is_counter_in_pool(pool, id))
4452 			return pool;
4453 	}
4454 	return NULL;
4455 }
4456 
4457 /**
4458  * Allocate a new memory for the counter values wrapped by all the needed
4459  * management.
4460  *
4461  * @param[in] dev
4462  *   Pointer to the Ethernet device structure.
4463  * @param[in] raws_n
4464  *   The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
4465  *
4466  * @return
4467  *   The new memory management pointer on success, otherwise NULL and rte_errno
4468  *   is set.
4469  */
4470 static struct mlx5_counter_stats_mem_mng *
4471 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
4472 {
4473 	struct mlx5_priv *priv = dev->data->dev_private;
4474 	struct mlx5_dev_ctx_shared *sh = priv->sh;
4475 	struct mlx5_devx_mkey_attr mkey_attr;
4476 	struct mlx5_counter_stats_mem_mng *mem_mng;
4477 	volatile struct flow_counter_stats *raw_data;
4478 	int size = (sizeof(struct flow_counter_stats) *
4479 			MLX5_COUNTERS_PER_POOL +
4480 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
4481 			sizeof(struct mlx5_counter_stats_mem_mng);
4482 	size_t pgsize = rte_mem_page_size();
4483 	if (pgsize == (size_t)-1) {
4484 		DRV_LOG(ERR, "Failed to get mem page size");
4485 		rte_errno = ENOMEM;
4486 		return NULL;
4487 	}
4488 	uint8_t *mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize,
4489 				  SOCKET_ID_ANY);
4490 	int i;
4491 
4492 	if (!mem) {
4493 		rte_errno = ENOMEM;
4494 		return NULL;
4495 	}
4496 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
4497 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
4498 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
4499 						 IBV_ACCESS_LOCAL_WRITE);
4500 	if (!mem_mng->umem) {
4501 		rte_errno = errno;
4502 		mlx5_free(mem);
4503 		return NULL;
4504 	}
4505 	mkey_attr.addr = (uintptr_t)mem;
4506 	mkey_attr.size = size;
4507 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
4508 	mkey_attr.pd = sh->pdn;
4509 	mkey_attr.log_entity_size = 0;
4510 	mkey_attr.pg_access = 0;
4511 	mkey_attr.klm_array = NULL;
4512 	mkey_attr.klm_num = 0;
4513 	if (priv->config.hca_attr.relaxed_ordering_write &&
4514 		priv->config.hca_attr.relaxed_ordering_read  &&
4515 		!haswell_broadwell_cpu)
4516 		mkey_attr.relaxed_ordering = 1;
4517 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4518 	if (!mem_mng->dm) {
4519 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
4520 		rte_errno = errno;
4521 		mlx5_free(mem);
4522 		return NULL;
4523 	}
4524 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4525 	raw_data = (volatile struct flow_counter_stats *)mem;
4526 	for (i = 0; i < raws_n; ++i) {
4527 		mem_mng->raws[i].mem_mng = mem_mng;
4528 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4529 	}
4530 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4531 	return mem_mng;
4532 }
4533 
4534 /**
4535  * Resize a counter container.
4536  *
4537  * @param[in] dev
4538  *   Pointer to the Ethernet device structure.
4539  * @param[in] batch
4540  *   Whether the pool is for counter that was allocated by batch command.
4541  * @param[in] age
4542  *   Whether the pool is for Aging counter.
4543  *
4544  * @return
4545  *   0 on success, otherwise negative errno value and rte_errno is set.
4546  */
4547 static int
4548 flow_dv_container_resize(struct rte_eth_dev *dev,
4549 				uint32_t batch, uint32_t age)
4550 {
4551 	struct mlx5_priv *priv = dev->data->dev_private;
4552 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4553 							       age);
4554 	struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4555 	void *old_pools = cont->pools;
4556 	uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4557 	uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4558 	void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4559 
4560 	if (!pools) {
4561 		rte_errno = ENOMEM;
4562 		return -ENOMEM;
4563 	}
4564 	if (old_pools)
4565 		memcpy(pools, old_pools, cont->n *
4566 				       sizeof(struct mlx5_flow_counter_pool *));
4567 	/*
4568 	 * Fallback mode query the counter directly, no background query
4569 	 * resources are needed.
4570 	 */
4571 	if (!priv->counter_fallback) {
4572 		int i;
4573 
4574 		mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4575 			  MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4576 		if (!mem_mng) {
4577 			mlx5_free(pools);
4578 			return -ENOMEM;
4579 		}
4580 		for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4581 			LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4582 					 mem_mng->raws +
4583 					 MLX5_CNT_CONTAINER_RESIZE +
4584 					 i, next);
4585 	}
4586 	rte_spinlock_lock(&cont->resize_sl);
4587 	cont->n = resize;
4588 	cont->mem_mng = mem_mng;
4589 	cont->pools = pools;
4590 	rte_spinlock_unlock(&cont->resize_sl);
4591 	if (old_pools)
4592 		mlx5_free(old_pools);
4593 	return 0;
4594 }
4595 
4596 /**
4597  * Query a devx flow counter.
4598  *
4599  * @param[in] dev
4600  *   Pointer to the Ethernet device structure.
4601  * @param[in] cnt
4602  *   Index to the flow counter.
4603  * @param[out] pkts
4604  *   The statistics value of packets.
4605  * @param[out] bytes
4606  *   The statistics value of bytes.
4607  *
4608  * @return
4609  *   0 on success, otherwise a negative errno value and rte_errno is set.
4610  */
4611 static inline int
4612 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4613 		     uint64_t *bytes)
4614 {
4615 	struct mlx5_priv *priv = dev->data->dev_private;
4616 	struct mlx5_flow_counter_pool *pool = NULL;
4617 	struct mlx5_flow_counter *cnt;
4618 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
4619 	int offset;
4620 
4621 	cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4622 	MLX5_ASSERT(pool);
4623 	if (counter < MLX5_CNT_BATCH_OFFSET) {
4624 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4625 		if (priv->counter_fallback)
4626 			return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4627 					0, pkts, bytes, 0, NULL, NULL, 0);
4628 	}
4629 
4630 	rte_spinlock_lock(&pool->sl);
4631 	/*
4632 	 * The single counters allocation may allocate smaller ID than the
4633 	 * current allocated in parallel to the host reading.
4634 	 * In this case the new counter values must be reported as 0.
4635 	 */
4636 	if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4637 		*pkts = 0;
4638 		*bytes = 0;
4639 	} else {
4640 		offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4641 		*pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4642 		*bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4643 	}
4644 	rte_spinlock_unlock(&pool->sl);
4645 	return 0;
4646 }
4647 
4648 /**
4649  * Create and initialize a new counter pool.
4650  *
4651  * @param[in] dev
4652  *   Pointer to the Ethernet device structure.
4653  * @param[out] dcs
4654  *   The devX counter handle.
4655  * @param[in] batch
4656  *   Whether the pool is for counter that was allocated by batch command.
4657  * @param[in] age
4658  *   Whether the pool is for counter that was allocated for aging.
4659  * @param[in/out] cont_cur
4660  *   Pointer to the container pointer, it will be update in pool resize.
4661  *
4662  * @return
4663  *   The pool container pointer on success, NULL otherwise and rte_errno is set.
4664  */
4665 static struct mlx5_flow_counter_pool *
4666 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4667 		    uint32_t batch, uint32_t age)
4668 {
4669 	struct mlx5_priv *priv = dev->data->dev_private;
4670 	struct mlx5_flow_counter_pool *pool;
4671 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4672 							       age);
4673 	int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4674 	uint32_t size = sizeof(*pool);
4675 
4676 	if (cont->n == n_valid && flow_dv_container_resize(dev, batch, age))
4677 		return NULL;
4678 	size += MLX5_COUNTERS_PER_POOL * CNT_SIZE;
4679 	size += (batch ? 0 : MLX5_COUNTERS_PER_POOL * CNTEXT_SIZE);
4680 	size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * AGE_SIZE);
4681 	pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4682 	if (!pool) {
4683 		rte_errno = ENOMEM;
4684 		return NULL;
4685 	}
4686 	pool->min_dcs = dcs;
4687 	if (!priv->counter_fallback)
4688 		pool->raw = cont->mem_mng->raws + n_valid %
4689 						      MLX5_CNT_CONTAINER_RESIZE;
4690 	pool->raw_hw = NULL;
4691 	pool->type = 0;
4692 	pool->type |= (batch ? 0 :  CNT_POOL_TYPE_EXT);
4693 	pool->type |= (!age ? 0 :  CNT_POOL_TYPE_AGE);
4694 	pool->query_gen = 0;
4695 	rte_spinlock_init(&pool->sl);
4696 	TAILQ_INIT(&pool->counters[0]);
4697 	TAILQ_INIT(&pool->counters[1]);
4698 	TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4699 	pool->index = n_valid;
4700 	cont->pools[n_valid] = pool;
4701 	if (!batch) {
4702 		int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4703 
4704 		if (base < cont->min_id)
4705 			cont->min_id = base;
4706 		if (base > cont->max_id)
4707 			cont->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4708 		cont->last_pool_idx = pool->index;
4709 	}
4710 	/* Pool initialization must be updated before host thread access. */
4711 	rte_io_wmb();
4712 	rte_atomic16_add(&cont->n_valid, 1);
4713 	return pool;
4714 }
4715 
4716 /**
4717  * Restore skipped counters in the pool.
4718  *
4719  * As counter pool query requires the first counter dcs
4720  * ID start with 4 alinged, if the pool counters with
4721  * min_dcs ID are not aligned with 4, the counters will
4722  * be skipped.
4723  * Once other min_dcs ID less than these skipped counter
4724  * dcs ID appears, the skipped counters will be safe to
4725  * use.
4726  * Should be called when min_dcs is updated.
4727  *
4728  * @param[in] pool
4729  *   Current counter pool.
4730  * @param[in] last_min_dcs
4731  *   Last min_dcs.
4732  */
4733 static void
4734 flow_dv_counter_restore(struct mlx5_flow_counter_pool *pool,
4735 			struct mlx5_devx_obj *last_min_dcs)
4736 {
4737 	struct mlx5_flow_counter_ext *cnt_ext;
4738 	uint32_t offset, new_offset;
4739 	uint32_t skip_cnt = 0;
4740 	uint32_t i;
4741 
4742 	if (!pool->skip_cnt)
4743 		return;
4744 	/*
4745 	 * If last min_dcs is not valid. The skipped counter may even after
4746 	 * last min_dcs, set the offset to the whole pool.
4747 	 */
4748 	if (last_min_dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4749 		offset = MLX5_COUNTERS_PER_POOL;
4750 	else
4751 		offset = last_min_dcs->id % MLX5_COUNTERS_PER_POOL;
4752 	new_offset = pool->min_dcs->id % MLX5_COUNTERS_PER_POOL;
4753 	/*
4754 	 * Check the counters from 1 to the last_min_dcs range. Counters
4755 	 * before new min_dcs indicates pool still has skipped counters.
4756 	 * Counters be skipped after new min_dcs will be ready to use.
4757 	 * Offset 0 counter must be empty or min_dcs, start from 1.
4758 	 */
4759 	for (i = 1; i < offset; i++) {
4760 		cnt_ext = MLX5_GET_POOL_CNT_EXT(pool, i);
4761 		if (cnt_ext->skipped) {
4762 			if (i > new_offset) {
4763 				cnt_ext->skipped = 0;
4764 				TAILQ_INSERT_TAIL
4765 					(&pool->counters[pool->query_gen],
4766 					 MLX5_POOL_GET_CNT(pool, i), next);
4767 			} else {
4768 				skip_cnt++;
4769 			}
4770 		}
4771 	}
4772 	if (!skip_cnt)
4773 		pool->skip_cnt = 0;
4774 }
4775 
4776 /**
4777  * Prepare a new counter and/or a new counter pool.
4778  *
4779  * @param[in] dev
4780  *   Pointer to the Ethernet device structure.
4781  * @param[out] cnt_free
4782  *   Where to put the pointer of a new counter.
4783  * @param[in] batch
4784  *   Whether the pool is for counter that was allocated by batch command.
4785  * @param[in] age
4786  *   Whether the pool is for counter that was allocated for aging.
4787  *
4788  * @return
4789  *   The counter pool pointer and @p cnt_free is set on success,
4790  *   NULL otherwise and rte_errno is set.
4791  */
4792 static struct mlx5_flow_counter_pool *
4793 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4794 			     struct mlx5_flow_counter **cnt_free,
4795 			     uint32_t batch, uint32_t age)
4796 {
4797 	struct mlx5_priv *priv = dev->data->dev_private;
4798 	struct mlx5_pools_container *cont;
4799 	struct mlx5_flow_counter_pool *pool;
4800 	struct mlx5_counters tmp_tq;
4801 	struct mlx5_devx_obj *last_min_dcs;
4802 	struct mlx5_devx_obj *dcs = NULL;
4803 	struct mlx5_flow_counter *cnt;
4804 	uint32_t add2other;
4805 	uint32_t i;
4806 
4807 	cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4808 	if (!batch) {
4809 retry:
4810 		add2other = 0;
4811 		/* bulk_bitmap must be 0 for single counter allocation. */
4812 		dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4813 		if (!dcs)
4814 			return NULL;
4815 		pool = flow_dv_find_pool_by_id(cont, dcs->id);
4816 		/* Check if counter belongs to exist pool ID range. */
4817 		if (!pool) {
4818 			pool = flow_dv_find_pool_by_id
4819 			       (MLX5_CNT_CONTAINER
4820 			       (priv->sh, batch, (age ^ 0x1)), dcs->id);
4821 			/*
4822 			 * Pool eixsts, counter will be added to the other
4823 			 * container, need to reallocate it later.
4824 			 */
4825 			if (pool) {
4826 				add2other = 1;
4827 			} else {
4828 				pool = flow_dv_pool_create(dev, dcs, batch,
4829 							   age);
4830 				if (!pool) {
4831 					mlx5_devx_cmd_destroy(dcs);
4832 					return NULL;
4833 				}
4834 			}
4835 		}
4836 		if ((dcs->id < pool->min_dcs->id ||
4837 		    pool->min_dcs->id &
4838 		    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) &&
4839 		    !(dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))) {
4840 			/*
4841 			 * Update the pool min_dcs only if current dcs is
4842 			 * valid and exist min_dcs is not valid or greater
4843 			 * than new dcs.
4844 			 */
4845 			last_min_dcs = pool->min_dcs;
4846 			rte_atomic64_set(&pool->a64_dcs,
4847 					 (int64_t)(uintptr_t)dcs);
4848 			/*
4849 			 * Restore any skipped counters if the new min_dcs
4850 			 * ID is smaller or min_dcs is not valid.
4851 			 */
4852 			if (dcs->id < last_min_dcs->id ||
4853 			    last_min_dcs->id &
4854 			    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4855 				flow_dv_counter_restore(pool, last_min_dcs);
4856 		}
4857 		i = dcs->id % MLX5_COUNTERS_PER_POOL;
4858 		cnt = MLX5_POOL_GET_CNT(pool, i);
4859 		cnt->pool = pool;
4860 		MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4861 		/*
4862 		 * If min_dcs is not valid, it means the new allocated dcs
4863 		 * also fail to become the valid min_dcs, just skip it.
4864 		 * Or if min_dcs is valid, and new dcs ID is smaller than
4865 		 * min_dcs, but not become the min_dcs, also skip it.
4866 		 */
4867 		if (pool->min_dcs->id &
4868 		    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1) ||
4869 		    dcs->id < pool->min_dcs->id) {
4870 			MLX5_GET_POOL_CNT_EXT(pool, i)->skipped = 1;
4871 			pool->skip_cnt = 1;
4872 			goto retry;
4873 		}
4874 		if (add2other) {
4875 			TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen],
4876 					  cnt, next);
4877 			goto retry;
4878 		}
4879 		*cnt_free = cnt;
4880 		return pool;
4881 	}
4882 	/* bulk_bitmap is in 128 counters units. */
4883 	if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4884 		dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4885 	if (!dcs) {
4886 		rte_errno = ENODATA;
4887 		return NULL;
4888 	}
4889 	pool = flow_dv_pool_create(dev, dcs, batch, age);
4890 	if (!pool) {
4891 		mlx5_devx_cmd_destroy(dcs);
4892 		return NULL;
4893 	}
4894 	TAILQ_INIT(&tmp_tq);
4895 	for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4896 		cnt = MLX5_POOL_GET_CNT(pool, i);
4897 		cnt->pool = pool;
4898 		TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4899 	}
4900 	rte_spinlock_lock(&cont->csl);
4901 	TAILQ_CONCAT(&cont->counters, &tmp_tq, next);
4902 	rte_spinlock_unlock(&cont->csl);
4903 	*cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4904 	(*cnt_free)->pool = pool;
4905 	return pool;
4906 }
4907 
4908 /**
4909  * Search for existed shared counter.
4910  *
4911  * @param[in] dev
4912  *   Pointer to the Ethernet device structure.
4913  * @param[in] id
4914  *   The shared counter ID to search.
4915  * @param[out] ppool
4916  *   mlx5 flow counter pool in the container,
4917  *
4918  * @return
4919  *   NULL if not existed, otherwise pointer to the shared extend counter.
4920  */
4921 static struct mlx5_flow_counter_ext *
4922 flow_dv_counter_shared_search(struct rte_eth_dev *dev, uint32_t id,
4923 			      struct mlx5_flow_counter_pool **ppool)
4924 {
4925 	struct mlx5_priv *priv = dev->data->dev_private;
4926 	union mlx5_l3t_data data;
4927 	uint32_t cnt_idx;
4928 
4929 	if (mlx5_l3t_get_entry(priv->sh->cnt_id_tbl, id, &data) || !data.dword)
4930 		return NULL;
4931 	cnt_idx = data.dword;
4932 	/*
4933 	 * Shared counters don't have age info. The counter extend is after
4934 	 * the counter datat structure.
4935 	 */
4936 	return (struct mlx5_flow_counter_ext *)
4937 	       ((flow_dv_counter_get_by_idx(dev, cnt_idx, ppool)) + 1);
4938 }
4939 
4940 /**
4941  * Allocate a flow counter.
4942  *
4943  * @param[in] dev
4944  *   Pointer to the Ethernet device structure.
4945  * @param[in] shared
4946  *   Indicate if this counter is shared with other flows.
4947  * @param[in] id
4948  *   Counter identifier.
4949  * @param[in] group
4950  *   Counter flow group.
4951  * @param[in] age
4952  *   Whether the counter was allocated for aging.
4953  *
4954  * @return
4955  *   Index to flow counter on success, 0 otherwise and rte_errno is set.
4956  */
4957 static uint32_t
4958 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4959 		      uint16_t group, uint32_t age)
4960 {
4961 	struct mlx5_priv *priv = dev->data->dev_private;
4962 	struct mlx5_flow_counter_pool *pool = NULL;
4963 	struct mlx5_flow_counter *cnt_free = NULL;
4964 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
4965 	/*
4966 	 * Currently group 0 flow counter cannot be assigned to a flow if it is
4967 	 * not the first one in the batch counter allocation, so it is better
4968 	 * to allocate counters one by one for these flows in a separate
4969 	 * container.
4970 	 * A counter can be shared between different groups so need to take
4971 	 * shared counters from the single container.
4972 	 */
4973 	uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4974 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4975 							       age);
4976 	uint32_t cnt_idx;
4977 
4978 	if (!priv->config.devx) {
4979 		rte_errno = ENOTSUP;
4980 		return 0;
4981 	}
4982 	if (shared) {
4983 		cnt_ext = flow_dv_counter_shared_search(dev, id, &pool);
4984 		if (cnt_ext) {
4985 			if (cnt_ext->ref_cnt + 1 == 0) {
4986 				rte_errno = E2BIG;
4987 				return 0;
4988 			}
4989 			cnt_ext->ref_cnt++;
4990 			cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4991 				  (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4992 				  + 1;
4993 			return cnt_idx;
4994 		}
4995 	}
4996 	/* Get free counters from container. */
4997 	rte_spinlock_lock(&cont->csl);
4998 	cnt_free = TAILQ_FIRST(&cont->counters);
4999 	if (cnt_free)
5000 		TAILQ_REMOVE(&cont->counters, cnt_free, next);
5001 	rte_spinlock_unlock(&cont->csl);
5002 	if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free,
5003 						       batch, age))
5004 		goto err;
5005 	pool = cnt_free->pool;
5006 	if (!batch)
5007 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
5008 	/* Create a DV counter action only in the first time usage. */
5009 	if (!cnt_free->action) {
5010 		uint16_t offset;
5011 		struct mlx5_devx_obj *dcs;
5012 		int ret;
5013 
5014 		if (batch) {
5015 			offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
5016 			dcs = pool->min_dcs;
5017 		} else {
5018 			offset = 0;
5019 			dcs = cnt_ext->dcs;
5020 		}
5021 		ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
5022 							    &cnt_free->action);
5023 		if (ret) {
5024 			rte_errno = errno;
5025 			goto err;
5026 		}
5027 	}
5028 	cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
5029 				MLX5_CNT_ARRAY_IDX(pool, cnt_free));
5030 	cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
5031 	cnt_idx += age * MLX5_CNT_AGE_OFFSET;
5032 	/* Update the counter reset values. */
5033 	if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
5034 				 &cnt_free->bytes))
5035 		goto err;
5036 	if (cnt_ext) {
5037 		cnt_ext->shared = shared;
5038 		cnt_ext->ref_cnt = 1;
5039 		cnt_ext->id = id;
5040 		if (shared) {
5041 			union mlx5_l3t_data data;
5042 
5043 			data.dword = cnt_idx;
5044 			if (mlx5_l3t_set_entry(priv->sh->cnt_id_tbl, id, &data))
5045 				return 0;
5046 		}
5047 	}
5048 	if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
5049 		/* Start the asynchronous batch query by the host thread. */
5050 		mlx5_set_query_alarm(priv->sh);
5051 	return cnt_idx;
5052 err:
5053 	if (cnt_free) {
5054 		cnt_free->pool = pool;
5055 		rte_spinlock_lock(&cont->csl);
5056 		TAILQ_INSERT_TAIL(&cont->counters, cnt_free, next);
5057 		rte_spinlock_unlock(&cont->csl);
5058 	}
5059 	return 0;
5060 }
5061 
5062 /**
5063  * Get age param from counter index.
5064  *
5065  * @param[in] dev
5066  *   Pointer to the Ethernet device structure.
5067  * @param[in] counter
5068  *   Index to the counter handler.
5069  *
5070  * @return
5071  *   The aging parameter specified for the counter index.
5072  */
5073 static struct mlx5_age_param*
5074 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
5075 				uint32_t counter)
5076 {
5077 	struct mlx5_flow_counter *cnt;
5078 	struct mlx5_flow_counter_pool *pool = NULL;
5079 
5080 	flow_dv_counter_get_by_idx(dev, counter, &pool);
5081 	counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
5082 	cnt = MLX5_POOL_GET_CNT(pool, counter);
5083 	return MLX5_CNT_TO_AGE(cnt);
5084 }
5085 
5086 /**
5087  * Remove a flow counter from aged counter list.
5088  *
5089  * @param[in] dev
5090  *   Pointer to the Ethernet device structure.
5091  * @param[in] counter
5092  *   Index to the counter handler.
5093  * @param[in] cnt
5094  *   Pointer to the counter handler.
5095  */
5096 static void
5097 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
5098 				uint32_t counter, struct mlx5_flow_counter *cnt)
5099 {
5100 	struct mlx5_age_info *age_info;
5101 	struct mlx5_age_param *age_param;
5102 	struct mlx5_priv *priv = dev->data->dev_private;
5103 
5104 	age_info = GET_PORT_AGE_INFO(priv);
5105 	age_param = flow_dv_counter_idx_get_age(dev, counter);
5106 	if (rte_atomic16_cmpset((volatile uint16_t *)
5107 			&age_param->state,
5108 			AGE_CANDIDATE, AGE_FREE)
5109 			!= AGE_CANDIDATE) {
5110 		/**
5111 		 * We need the lock even it is age timeout,
5112 		 * since counter may still in process.
5113 		 */
5114 		rte_spinlock_lock(&age_info->aged_sl);
5115 		TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
5116 		rte_spinlock_unlock(&age_info->aged_sl);
5117 	}
5118 	rte_atomic16_set(&age_param->state, AGE_FREE);
5119 }
5120 /**
5121  * Release a flow counter.
5122  *
5123  * @param[in] dev
5124  *   Pointer to the Ethernet device structure.
5125  * @param[in] counter
5126  *   Index to the counter handler.
5127  */
5128 static void
5129 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
5130 {
5131 	struct mlx5_priv *priv = dev->data->dev_private;
5132 	struct mlx5_flow_counter_pool *pool = NULL;
5133 	struct mlx5_flow_counter *cnt;
5134 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
5135 
5136 	if (!counter)
5137 		return;
5138 	cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
5139 	MLX5_ASSERT(pool);
5140 	if (counter < MLX5_CNT_BATCH_OFFSET) {
5141 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
5142 		if (cnt_ext) {
5143 			if (--cnt_ext->ref_cnt)
5144 				return;
5145 			if (cnt_ext->shared)
5146 				mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl,
5147 						     cnt_ext->id);
5148 		}
5149 	}
5150 	if (IS_AGE_POOL(pool))
5151 		flow_dv_counter_remove_from_age(dev, counter, cnt);
5152 	cnt->pool = pool;
5153 	/*
5154 	 * Put the counter back to list to be updated in none fallback mode.
5155 	 * Currently, we are using two list alternately, while one is in query,
5156 	 * add the freed counter to the other list based on the pool query_gen
5157 	 * value. After query finishes, add counter the list to the global
5158 	 * container counter list. The list changes while query starts. In
5159 	 * this case, lock will not be needed as query callback and release
5160 	 * function both operate with the different list.
5161 	 *
5162 	 */
5163 	if (!priv->counter_fallback)
5164 		TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
5165 	else
5166 		TAILQ_INSERT_TAIL(&((MLX5_CNT_CONTAINER
5167 				  (priv->sh, 0, 0))->counters),
5168 				  cnt, next);
5169 }
5170 
5171 /**
5172  * Verify the @p attributes will be correctly understood by the NIC and store
5173  * them in the @p flow if everything is correct.
5174  *
5175  * @param[in] dev
5176  *   Pointer to dev struct.
5177  * @param[in] attributes
5178  *   Pointer to flow attributes
5179  * @param[in] external
5180  *   This flow rule is created by request external to PMD.
5181  * @param[out] error
5182  *   Pointer to error structure.
5183  *
5184  * @return
5185  *   - 0 on success and non root table.
5186  *   - 1 on success and root table.
5187  *   - a negative errno value otherwise and rte_errno is set.
5188  */
5189 static int
5190 flow_dv_validate_attributes(struct rte_eth_dev *dev,
5191 			    const struct rte_flow_attr *attributes,
5192 			    bool external __rte_unused,
5193 			    struct rte_flow_error *error)
5194 {
5195 	struct mlx5_priv *priv = dev->data->dev_private;
5196 	uint32_t priority_max = priv->config.flow_prio - 1;
5197 	int ret = 0;
5198 
5199 #ifndef HAVE_MLX5DV_DR
5200 	if (attributes->group)
5201 		return rte_flow_error_set(error, ENOTSUP,
5202 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5203 					  NULL,
5204 					  "groups are not supported");
5205 #else
5206 	uint32_t table = 0;
5207 
5208 	ret = mlx5_flow_group_to_table(attributes, external,
5209 				       attributes->group, !!priv->fdb_def_rule,
5210 				       &table, error);
5211 	if (ret)
5212 		return ret;
5213 	if (!table)
5214 		ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
5215 #endif
5216 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
5217 	    attributes->priority >= priority_max)
5218 		return rte_flow_error_set(error, ENOTSUP,
5219 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
5220 					  NULL,
5221 					  "priority out of range");
5222 	if (attributes->transfer) {
5223 		if (!priv->config.dv_esw_en)
5224 			return rte_flow_error_set
5225 				(error, ENOTSUP,
5226 				 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5227 				 "E-Switch dr is not supported");
5228 		if (!(priv->representor || priv->master))
5229 			return rte_flow_error_set
5230 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5231 				 NULL, "E-Switch configuration can only be"
5232 				 " done by a master or a representor device");
5233 		if (attributes->egress)
5234 			return rte_flow_error_set
5235 				(error, ENOTSUP,
5236 				 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
5237 				 "egress is not supported");
5238 	}
5239 	if (!(attributes->egress ^ attributes->ingress))
5240 		return rte_flow_error_set(error, ENOTSUP,
5241 					  RTE_FLOW_ERROR_TYPE_ATTR, NULL,
5242 					  "must specify exactly one of "
5243 					  "ingress or egress");
5244 	return ret;
5245 }
5246 
5247 /**
5248  * Internal validation function. For validating both actions and items.
5249  *
5250  * @param[in] dev
5251  *   Pointer to the rte_eth_dev structure.
5252  * @param[in] attr
5253  *   Pointer to the flow attributes.
5254  * @param[in] items
5255  *   Pointer to the list of items.
5256  * @param[in] actions
5257  *   Pointer to the list of actions.
5258  * @param[in] external
5259  *   This flow rule is created by request external to PMD.
5260  * @param[in] hairpin
5261  *   Number of hairpin TX actions, 0 means classic flow.
5262  * @param[out] error
5263  *   Pointer to the error structure.
5264  *
5265  * @return
5266  *   0 on success, a negative errno value otherwise and rte_errno is set.
5267  */
5268 static int
5269 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
5270 		 const struct rte_flow_item items[],
5271 		 const struct rte_flow_action actions[],
5272 		 bool external, int hairpin, struct rte_flow_error *error)
5273 {
5274 	int ret;
5275 	uint64_t action_flags = 0;
5276 	uint64_t item_flags = 0;
5277 	uint64_t last_item = 0;
5278 	uint8_t next_protocol = 0xff;
5279 	uint16_t ether_type = 0;
5280 	int actions_n = 0;
5281 	uint8_t item_ipv6_proto = 0;
5282 	const struct rte_flow_item *gre_item = NULL;
5283 	const struct rte_flow_action_raw_decap *decap;
5284 	const struct rte_flow_action_raw_encap *encap;
5285 	const struct rte_flow_action_rss *rss;
5286 	const struct rte_flow_item_tcp nic_tcp_mask = {
5287 		.hdr = {
5288 			.tcp_flags = 0xFF,
5289 			.src_port = RTE_BE16(UINT16_MAX),
5290 			.dst_port = RTE_BE16(UINT16_MAX),
5291 		}
5292 	};
5293 	const struct rte_flow_item_ipv4 nic_ipv4_mask = {
5294 		.hdr = {
5295 			.src_addr = RTE_BE32(0xffffffff),
5296 			.dst_addr = RTE_BE32(0xffffffff),
5297 			.type_of_service = 0xff,
5298 			.next_proto_id = 0xff,
5299 			.time_to_live = 0xff,
5300 		},
5301 	};
5302 	const struct rte_flow_item_ipv6 nic_ipv6_mask = {
5303 		.hdr = {
5304 			.src_addr =
5305 			"\xff\xff\xff\xff\xff\xff\xff\xff"
5306 			"\xff\xff\xff\xff\xff\xff\xff\xff",
5307 			.dst_addr =
5308 			"\xff\xff\xff\xff\xff\xff\xff\xff"
5309 			"\xff\xff\xff\xff\xff\xff\xff\xff",
5310 			.vtc_flow = RTE_BE32(0xffffffff),
5311 			.proto = 0xff,
5312 			.hop_limits = 0xff,
5313 		},
5314 	};
5315 	const struct rte_flow_item_ecpri nic_ecpri_mask = {
5316 		.hdr = {
5317 			.common = {
5318 				.u32 =
5319 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
5320 					.type = 0xFF,
5321 					}).u32),
5322 			},
5323 			.dummy[0] = 0xffffffff,
5324 		},
5325 	};
5326 	struct mlx5_priv *priv = dev->data->dev_private;
5327 	struct mlx5_dev_config *dev_conf = &priv->config;
5328 	uint16_t queue_index = 0xFFFF;
5329 	const struct rte_flow_item_vlan *vlan_m = NULL;
5330 	int16_t rw_act_num = 0;
5331 	uint64_t is_root;
5332 
5333 	if (items == NULL)
5334 		return -1;
5335 	ret = flow_dv_validate_attributes(dev, attr, external, error);
5336 	if (ret < 0)
5337 		return ret;
5338 	is_root = (uint64_t)ret;
5339 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5340 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5341 		int type = items->type;
5342 
5343 		if (!mlx5_flow_os_item_supported(type))
5344 			return rte_flow_error_set(error, ENOTSUP,
5345 						  RTE_FLOW_ERROR_TYPE_ITEM,
5346 						  NULL, "item not supported");
5347 		switch (type) {
5348 		case RTE_FLOW_ITEM_TYPE_VOID:
5349 			break;
5350 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5351 			ret = flow_dv_validate_item_port_id
5352 					(dev, items, attr, item_flags, error);
5353 			if (ret < 0)
5354 				return ret;
5355 			last_item = MLX5_FLOW_ITEM_PORT_ID;
5356 			break;
5357 		case RTE_FLOW_ITEM_TYPE_ETH:
5358 			ret = mlx5_flow_validate_item_eth(items, item_flags,
5359 							  error);
5360 			if (ret < 0)
5361 				return ret;
5362 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5363 					     MLX5_FLOW_LAYER_OUTER_L2;
5364 			if (items->mask != NULL && items->spec != NULL) {
5365 				ether_type =
5366 					((const struct rte_flow_item_eth *)
5367 					 items->spec)->type;
5368 				ether_type &=
5369 					((const struct rte_flow_item_eth *)
5370 					 items->mask)->type;
5371 				ether_type = rte_be_to_cpu_16(ether_type);
5372 			} else {
5373 				ether_type = 0;
5374 			}
5375 			break;
5376 		case RTE_FLOW_ITEM_TYPE_VLAN:
5377 			ret = flow_dv_validate_item_vlan(items, item_flags,
5378 							 dev, error);
5379 			if (ret < 0)
5380 				return ret;
5381 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5382 					     MLX5_FLOW_LAYER_OUTER_VLAN;
5383 			if (items->mask != NULL && items->spec != NULL) {
5384 				ether_type =
5385 					((const struct rte_flow_item_vlan *)
5386 					 items->spec)->inner_type;
5387 				ether_type &=
5388 					((const struct rte_flow_item_vlan *)
5389 					 items->mask)->inner_type;
5390 				ether_type = rte_be_to_cpu_16(ether_type);
5391 			} else {
5392 				ether_type = 0;
5393 			}
5394 			/* Store outer VLAN mask for of_push_vlan action. */
5395 			if (!tunnel)
5396 				vlan_m = items->mask;
5397 			break;
5398 		case RTE_FLOW_ITEM_TYPE_IPV4:
5399 			mlx5_flow_tunnel_ip_check(items, next_protocol,
5400 						  &item_flags, &tunnel);
5401 			ret = mlx5_flow_validate_item_ipv4(items, item_flags,
5402 							   last_item,
5403 							   ether_type,
5404 							   &nic_ipv4_mask,
5405 							   error);
5406 			if (ret < 0)
5407 				return ret;
5408 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5409 					     MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5410 			if (items->mask != NULL &&
5411 			    ((const struct rte_flow_item_ipv4 *)
5412 			     items->mask)->hdr.next_proto_id) {
5413 				next_protocol =
5414 					((const struct rte_flow_item_ipv4 *)
5415 					 (items->spec))->hdr.next_proto_id;
5416 				next_protocol &=
5417 					((const struct rte_flow_item_ipv4 *)
5418 					 (items->mask))->hdr.next_proto_id;
5419 			} else {
5420 				/* Reset for inner layer. */
5421 				next_protocol = 0xff;
5422 			}
5423 			break;
5424 		case RTE_FLOW_ITEM_TYPE_IPV6:
5425 			mlx5_flow_tunnel_ip_check(items, next_protocol,
5426 						  &item_flags, &tunnel);
5427 			ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5428 							   last_item,
5429 							   ether_type,
5430 							   &nic_ipv6_mask,
5431 							   error);
5432 			if (ret < 0)
5433 				return ret;
5434 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5435 					     MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5436 			if (items->mask != NULL &&
5437 			    ((const struct rte_flow_item_ipv6 *)
5438 			     items->mask)->hdr.proto) {
5439 				item_ipv6_proto =
5440 					((const struct rte_flow_item_ipv6 *)
5441 					 items->spec)->hdr.proto;
5442 				next_protocol =
5443 					((const struct rte_flow_item_ipv6 *)
5444 					 items->spec)->hdr.proto;
5445 				next_protocol &=
5446 					((const struct rte_flow_item_ipv6 *)
5447 					 items->mask)->hdr.proto;
5448 			} else {
5449 				/* Reset for inner layer. */
5450 				next_protocol = 0xff;
5451 			}
5452 			break;
5453 		case RTE_FLOW_ITEM_TYPE_TCP:
5454 			ret = mlx5_flow_validate_item_tcp
5455 						(items, item_flags,
5456 						 next_protocol,
5457 						 &nic_tcp_mask,
5458 						 error);
5459 			if (ret < 0)
5460 				return ret;
5461 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5462 					     MLX5_FLOW_LAYER_OUTER_L4_TCP;
5463 			break;
5464 		case RTE_FLOW_ITEM_TYPE_UDP:
5465 			ret = mlx5_flow_validate_item_udp(items, item_flags,
5466 							  next_protocol,
5467 							  error);
5468 			if (ret < 0)
5469 				return ret;
5470 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5471 					     MLX5_FLOW_LAYER_OUTER_L4_UDP;
5472 			break;
5473 		case RTE_FLOW_ITEM_TYPE_GRE:
5474 			ret = mlx5_flow_validate_item_gre(items, item_flags,
5475 							  next_protocol, error);
5476 			if (ret < 0)
5477 				return ret;
5478 			gre_item = items;
5479 			last_item = MLX5_FLOW_LAYER_GRE;
5480 			break;
5481 		case RTE_FLOW_ITEM_TYPE_NVGRE:
5482 			ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5483 							    next_protocol,
5484 							    error);
5485 			if (ret < 0)
5486 				return ret;
5487 			last_item = MLX5_FLOW_LAYER_NVGRE;
5488 			break;
5489 		case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5490 			ret = mlx5_flow_validate_item_gre_key
5491 				(items, item_flags, gre_item, error);
5492 			if (ret < 0)
5493 				return ret;
5494 			last_item = MLX5_FLOW_LAYER_GRE_KEY;
5495 			break;
5496 		case RTE_FLOW_ITEM_TYPE_VXLAN:
5497 			ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5498 							    error);
5499 			if (ret < 0)
5500 				return ret;
5501 			last_item = MLX5_FLOW_LAYER_VXLAN;
5502 			break;
5503 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5504 			ret = mlx5_flow_validate_item_vxlan_gpe(items,
5505 								item_flags, dev,
5506 								error);
5507 			if (ret < 0)
5508 				return ret;
5509 			last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5510 			break;
5511 		case RTE_FLOW_ITEM_TYPE_GENEVE:
5512 			ret = mlx5_flow_validate_item_geneve(items,
5513 							     item_flags, dev,
5514 							     error);
5515 			if (ret < 0)
5516 				return ret;
5517 			last_item = MLX5_FLOW_LAYER_GENEVE;
5518 			break;
5519 		case RTE_FLOW_ITEM_TYPE_MPLS:
5520 			ret = mlx5_flow_validate_item_mpls(dev, items,
5521 							   item_flags,
5522 							   last_item, error);
5523 			if (ret < 0)
5524 				return ret;
5525 			last_item = MLX5_FLOW_LAYER_MPLS;
5526 			break;
5527 
5528 		case RTE_FLOW_ITEM_TYPE_MARK:
5529 			ret = flow_dv_validate_item_mark(dev, items, attr,
5530 							 error);
5531 			if (ret < 0)
5532 				return ret;
5533 			last_item = MLX5_FLOW_ITEM_MARK;
5534 			break;
5535 		case RTE_FLOW_ITEM_TYPE_META:
5536 			ret = flow_dv_validate_item_meta(dev, items, attr,
5537 							 error);
5538 			if (ret < 0)
5539 				return ret;
5540 			last_item = MLX5_FLOW_ITEM_METADATA;
5541 			break;
5542 		case RTE_FLOW_ITEM_TYPE_ICMP:
5543 			ret = mlx5_flow_validate_item_icmp(items, item_flags,
5544 							   next_protocol,
5545 							   error);
5546 			if (ret < 0)
5547 				return ret;
5548 			last_item = MLX5_FLOW_LAYER_ICMP;
5549 			break;
5550 		case RTE_FLOW_ITEM_TYPE_ICMP6:
5551 			ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5552 							    next_protocol,
5553 							    error);
5554 			if (ret < 0)
5555 				return ret;
5556 			item_ipv6_proto = IPPROTO_ICMPV6;
5557 			last_item = MLX5_FLOW_LAYER_ICMP6;
5558 			break;
5559 		case RTE_FLOW_ITEM_TYPE_TAG:
5560 			ret = flow_dv_validate_item_tag(dev, items,
5561 							attr, error);
5562 			if (ret < 0)
5563 				return ret;
5564 			last_item = MLX5_FLOW_ITEM_TAG;
5565 			break;
5566 		case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5567 		case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5568 			break;
5569 		case RTE_FLOW_ITEM_TYPE_GTP:
5570 			ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5571 							error);
5572 			if (ret < 0)
5573 				return ret;
5574 			last_item = MLX5_FLOW_LAYER_GTP;
5575 			break;
5576 		case RTE_FLOW_ITEM_TYPE_ECPRI:
5577 			/* Capacity will be checked in the translate stage. */
5578 			ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5579 							    last_item,
5580 							    ether_type,
5581 							    &nic_ecpri_mask,
5582 							    error);
5583 			if (ret < 0)
5584 				return ret;
5585 			last_item = MLX5_FLOW_LAYER_ECPRI;
5586 			break;
5587 		default:
5588 			return rte_flow_error_set(error, ENOTSUP,
5589 						  RTE_FLOW_ERROR_TYPE_ITEM,
5590 						  NULL, "item not supported");
5591 		}
5592 		item_flags |= last_item;
5593 	}
5594 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5595 		int type = actions->type;
5596 
5597 		if (!mlx5_flow_os_action_supported(type))
5598 			return rte_flow_error_set(error, ENOTSUP,
5599 						  RTE_FLOW_ERROR_TYPE_ACTION,
5600 						  actions,
5601 						  "action not supported");
5602 		if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5603 			return rte_flow_error_set(error, ENOTSUP,
5604 						  RTE_FLOW_ERROR_TYPE_ACTION,
5605 						  actions, "too many actions");
5606 		switch (type) {
5607 		case RTE_FLOW_ACTION_TYPE_VOID:
5608 			break;
5609 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
5610 			ret = flow_dv_validate_action_port_id(dev,
5611 							      action_flags,
5612 							      actions,
5613 							      attr,
5614 							      error);
5615 			if (ret)
5616 				return ret;
5617 			action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5618 			++actions_n;
5619 			break;
5620 		case RTE_FLOW_ACTION_TYPE_FLAG:
5621 			ret = flow_dv_validate_action_flag(dev, action_flags,
5622 							   attr, error);
5623 			if (ret < 0)
5624 				return ret;
5625 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5626 				/* Count all modify-header actions as one. */
5627 				if (!(action_flags &
5628 				      MLX5_FLOW_MODIFY_HDR_ACTIONS))
5629 					++actions_n;
5630 				action_flags |= MLX5_FLOW_ACTION_FLAG |
5631 						MLX5_FLOW_ACTION_MARK_EXT;
5632 			} else {
5633 				action_flags |= MLX5_FLOW_ACTION_FLAG;
5634 				++actions_n;
5635 			}
5636 			rw_act_num += MLX5_ACT_NUM_SET_MARK;
5637 			break;
5638 		case RTE_FLOW_ACTION_TYPE_MARK:
5639 			ret = flow_dv_validate_action_mark(dev, actions,
5640 							   action_flags,
5641 							   attr, error);
5642 			if (ret < 0)
5643 				return ret;
5644 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5645 				/* Count all modify-header actions as one. */
5646 				if (!(action_flags &
5647 				      MLX5_FLOW_MODIFY_HDR_ACTIONS))
5648 					++actions_n;
5649 				action_flags |= MLX5_FLOW_ACTION_MARK |
5650 						MLX5_FLOW_ACTION_MARK_EXT;
5651 			} else {
5652 				action_flags |= MLX5_FLOW_ACTION_MARK;
5653 				++actions_n;
5654 			}
5655 			rw_act_num += MLX5_ACT_NUM_SET_MARK;
5656 			break;
5657 		case RTE_FLOW_ACTION_TYPE_SET_META:
5658 			ret = flow_dv_validate_action_set_meta(dev, actions,
5659 							       action_flags,
5660 							       attr, error);
5661 			if (ret < 0)
5662 				return ret;
5663 			/* Count all modify-header actions as one action. */
5664 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5665 				++actions_n;
5666 			action_flags |= MLX5_FLOW_ACTION_SET_META;
5667 			rw_act_num += MLX5_ACT_NUM_SET_META;
5668 			break;
5669 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5670 			ret = flow_dv_validate_action_set_tag(dev, actions,
5671 							      action_flags,
5672 							      attr, error);
5673 			if (ret < 0)
5674 				return ret;
5675 			/* Count all modify-header actions as one action. */
5676 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5677 				++actions_n;
5678 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5679 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5680 			break;
5681 		case RTE_FLOW_ACTION_TYPE_DROP:
5682 			ret = mlx5_flow_validate_action_drop(action_flags,
5683 							     attr, error);
5684 			if (ret < 0)
5685 				return ret;
5686 			action_flags |= MLX5_FLOW_ACTION_DROP;
5687 			++actions_n;
5688 			break;
5689 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5690 			ret = mlx5_flow_validate_action_queue(actions,
5691 							      action_flags, dev,
5692 							      attr, error);
5693 			if (ret < 0)
5694 				return ret;
5695 			queue_index = ((const struct rte_flow_action_queue *)
5696 							(actions->conf))->index;
5697 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
5698 			++actions_n;
5699 			break;
5700 		case RTE_FLOW_ACTION_TYPE_RSS:
5701 			rss = actions->conf;
5702 			ret = mlx5_flow_validate_action_rss(actions,
5703 							    action_flags, dev,
5704 							    attr, item_flags,
5705 							    error);
5706 			if (ret < 0)
5707 				return ret;
5708 			if (rss != NULL && rss->queue_num)
5709 				queue_index = rss->queue[0];
5710 			action_flags |= MLX5_FLOW_ACTION_RSS;
5711 			++actions_n;
5712 			break;
5713 		case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5714 			ret =
5715 			mlx5_flow_validate_action_default_miss(action_flags,
5716 					attr, error);
5717 			if (ret < 0)
5718 				return ret;
5719 			action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5720 			++actions_n;
5721 			break;
5722 		case RTE_FLOW_ACTION_TYPE_COUNT:
5723 			ret = flow_dv_validate_action_count(dev, error);
5724 			if (ret < 0)
5725 				return ret;
5726 			action_flags |= MLX5_FLOW_ACTION_COUNT;
5727 			++actions_n;
5728 			break;
5729 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5730 			if (flow_dv_validate_action_pop_vlan(dev,
5731 							     action_flags,
5732 							     actions,
5733 							     item_flags, attr,
5734 							     error))
5735 				return -rte_errno;
5736 			action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5737 			++actions_n;
5738 			break;
5739 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5740 			ret = flow_dv_validate_action_push_vlan(dev,
5741 								action_flags,
5742 								vlan_m,
5743 								actions, attr,
5744 								error);
5745 			if (ret < 0)
5746 				return ret;
5747 			action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5748 			++actions_n;
5749 			break;
5750 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5751 			ret = flow_dv_validate_action_set_vlan_pcp
5752 						(action_flags, actions, error);
5753 			if (ret < 0)
5754 				return ret;
5755 			/* Count PCP with push_vlan command. */
5756 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5757 			break;
5758 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5759 			ret = flow_dv_validate_action_set_vlan_vid
5760 						(item_flags, action_flags,
5761 						 actions, error);
5762 			if (ret < 0)
5763 				return ret;
5764 			/* Count VID with push_vlan command. */
5765 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5766 			rw_act_num += MLX5_ACT_NUM_MDF_VID;
5767 			break;
5768 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5769 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5770 			ret = flow_dv_validate_action_l2_encap(dev,
5771 							       action_flags,
5772 							       actions, attr,
5773 							       error);
5774 			if (ret < 0)
5775 				return ret;
5776 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
5777 			++actions_n;
5778 			break;
5779 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5780 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5781 			ret = flow_dv_validate_action_decap(dev, action_flags,
5782 							    attr, error);
5783 			if (ret < 0)
5784 				return ret;
5785 			action_flags |= MLX5_FLOW_ACTION_DECAP;
5786 			++actions_n;
5787 			break;
5788 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5789 			ret = flow_dv_validate_action_raw_encap_decap
5790 				(dev, NULL, actions->conf, attr, &action_flags,
5791 				 &actions_n, error);
5792 			if (ret < 0)
5793 				return ret;
5794 			break;
5795 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5796 			decap = actions->conf;
5797 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5798 				;
5799 			if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5800 				encap = NULL;
5801 				actions--;
5802 			} else {
5803 				encap = actions->conf;
5804 			}
5805 			ret = flow_dv_validate_action_raw_encap_decap
5806 					   (dev,
5807 					    decap ? decap : &empty_decap, encap,
5808 					    attr, &action_flags, &actions_n,
5809 					    error);
5810 			if (ret < 0)
5811 				return ret;
5812 			break;
5813 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5814 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5815 			ret = flow_dv_validate_action_modify_mac(action_flags,
5816 								 actions,
5817 								 item_flags,
5818 								 error);
5819 			if (ret < 0)
5820 				return ret;
5821 			/* Count all modify-header actions as one action. */
5822 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5823 				++actions_n;
5824 			action_flags |= actions->type ==
5825 					RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5826 						MLX5_FLOW_ACTION_SET_MAC_SRC :
5827 						MLX5_FLOW_ACTION_SET_MAC_DST;
5828 			/*
5829 			 * Even if the source and destination MAC addresses have
5830 			 * overlap in the header with 4B alignment, the convert
5831 			 * function will handle them separately and 4 SW actions
5832 			 * will be created. And 2 actions will be added each
5833 			 * time no matter how many bytes of address will be set.
5834 			 */
5835 			rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5836 			break;
5837 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5838 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5839 			ret = flow_dv_validate_action_modify_ipv4(action_flags,
5840 								  actions,
5841 								  item_flags,
5842 								  error);
5843 			if (ret < 0)
5844 				return ret;
5845 			/* Count all modify-header actions as one action. */
5846 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5847 				++actions_n;
5848 			action_flags |= actions->type ==
5849 					RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5850 						MLX5_FLOW_ACTION_SET_IPV4_SRC :
5851 						MLX5_FLOW_ACTION_SET_IPV4_DST;
5852 			rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5853 			break;
5854 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5855 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5856 			ret = flow_dv_validate_action_modify_ipv6(action_flags,
5857 								  actions,
5858 								  item_flags,
5859 								  error);
5860 			if (ret < 0)
5861 				return ret;
5862 			if (item_ipv6_proto == IPPROTO_ICMPV6)
5863 				return rte_flow_error_set(error, ENOTSUP,
5864 					RTE_FLOW_ERROR_TYPE_ACTION,
5865 					actions,
5866 					"Can't change header "
5867 					"with ICMPv6 proto");
5868 			/* Count all modify-header actions as one action. */
5869 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5870 				++actions_n;
5871 			action_flags |= actions->type ==
5872 					RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5873 						MLX5_FLOW_ACTION_SET_IPV6_SRC :
5874 						MLX5_FLOW_ACTION_SET_IPV6_DST;
5875 			rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5876 			break;
5877 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5878 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5879 			ret = flow_dv_validate_action_modify_tp(action_flags,
5880 								actions,
5881 								item_flags,
5882 								error);
5883 			if (ret < 0)
5884 				return ret;
5885 			/* Count all modify-header actions as one action. */
5886 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5887 				++actions_n;
5888 			action_flags |= actions->type ==
5889 					RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5890 						MLX5_FLOW_ACTION_SET_TP_SRC :
5891 						MLX5_FLOW_ACTION_SET_TP_DST;
5892 			rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5893 			break;
5894 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5895 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5896 			ret = flow_dv_validate_action_modify_ttl(action_flags,
5897 								 actions,
5898 								 item_flags,
5899 								 error);
5900 			if (ret < 0)
5901 				return ret;
5902 			/* Count all modify-header actions as one action. */
5903 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5904 				++actions_n;
5905 			action_flags |= actions->type ==
5906 					RTE_FLOW_ACTION_TYPE_SET_TTL ?
5907 						MLX5_FLOW_ACTION_SET_TTL :
5908 						MLX5_FLOW_ACTION_DEC_TTL;
5909 			rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5910 			break;
5911 		case RTE_FLOW_ACTION_TYPE_JUMP:
5912 			ret = flow_dv_validate_action_jump(actions,
5913 							   action_flags,
5914 							   attr, external,
5915 							   error);
5916 			if (ret)
5917 				return ret;
5918 			++actions_n;
5919 			action_flags |= MLX5_FLOW_ACTION_JUMP;
5920 			break;
5921 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5922 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5923 			ret = flow_dv_validate_action_modify_tcp_seq
5924 								(action_flags,
5925 								 actions,
5926 								 item_flags,
5927 								 error);
5928 			if (ret < 0)
5929 				return ret;
5930 			/* Count all modify-header actions as one action. */
5931 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5932 				++actions_n;
5933 			action_flags |= actions->type ==
5934 					RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5935 						MLX5_FLOW_ACTION_INC_TCP_SEQ :
5936 						MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5937 			rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5938 			break;
5939 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5940 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5941 			ret = flow_dv_validate_action_modify_tcp_ack
5942 								(action_flags,
5943 								 actions,
5944 								 item_flags,
5945 								 error);
5946 			if (ret < 0)
5947 				return ret;
5948 			/* Count all modify-header actions as one action. */
5949 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5950 				++actions_n;
5951 			action_flags |= actions->type ==
5952 					RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5953 						MLX5_FLOW_ACTION_INC_TCP_ACK :
5954 						MLX5_FLOW_ACTION_DEC_TCP_ACK;
5955 			rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5956 			break;
5957 		case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5958 			break;
5959 		case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5960 		case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5961 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5962 			break;
5963 		case RTE_FLOW_ACTION_TYPE_METER:
5964 			ret = mlx5_flow_validate_action_meter(dev,
5965 							      action_flags,
5966 							      actions, attr,
5967 							      error);
5968 			if (ret < 0)
5969 				return ret;
5970 			action_flags |= MLX5_FLOW_ACTION_METER;
5971 			++actions_n;
5972 			/* Meter action will add one more TAG action. */
5973 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5974 			break;
5975 		case RTE_FLOW_ACTION_TYPE_AGE:
5976 			ret = flow_dv_validate_action_age(action_flags,
5977 							  actions, dev,
5978 							  error);
5979 			if (ret < 0)
5980 				return ret;
5981 			action_flags |= MLX5_FLOW_ACTION_AGE;
5982 			++actions_n;
5983 			break;
5984 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5985 			ret = flow_dv_validate_action_modify_ipv4_dscp
5986 							 (action_flags,
5987 							  actions,
5988 							  item_flags,
5989 							  error);
5990 			if (ret < 0)
5991 				return ret;
5992 			/* Count all modify-header actions as one action. */
5993 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5994 				++actions_n;
5995 			action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5996 			rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5997 			break;
5998 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5999 			ret = flow_dv_validate_action_modify_ipv6_dscp
6000 								(action_flags,
6001 								 actions,
6002 								 item_flags,
6003 								 error);
6004 			if (ret < 0)
6005 				return ret;
6006 			/* Count all modify-header actions as one action. */
6007 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
6008 				++actions_n;
6009 			action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
6010 			rw_act_num += MLX5_ACT_NUM_SET_DSCP;
6011 			break;
6012 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
6013 			ret = flow_dv_validate_action_sample(action_flags,
6014 							     actions, dev,
6015 							     attr, error);
6016 			if (ret < 0)
6017 				return ret;
6018 			action_flags |= MLX5_FLOW_ACTION_SAMPLE;
6019 			++actions_n;
6020 			break;
6021 		default:
6022 			return rte_flow_error_set(error, ENOTSUP,
6023 						  RTE_FLOW_ERROR_TYPE_ACTION,
6024 						  actions,
6025 						  "action not supported");
6026 		}
6027 	}
6028 	/*
6029 	 * Validate the drop action mutual exclusion with other actions.
6030 	 * Drop action is mutually-exclusive with any other action, except for
6031 	 * Count action.
6032 	 */
6033 	if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
6034 	    (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
6035 		return rte_flow_error_set(error, EINVAL,
6036 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
6037 					  "Drop action is mutually-exclusive "
6038 					  "with any other action, except for "
6039 					  "Count action");
6040 	/* Eswitch has few restrictions on using items and actions */
6041 	if (attr->transfer) {
6042 		if (!mlx5_flow_ext_mreg_supported(dev) &&
6043 		    action_flags & MLX5_FLOW_ACTION_FLAG)
6044 			return rte_flow_error_set(error, ENOTSUP,
6045 						  RTE_FLOW_ERROR_TYPE_ACTION,
6046 						  NULL,
6047 						  "unsupported action FLAG");
6048 		if (!mlx5_flow_ext_mreg_supported(dev) &&
6049 		    action_flags & MLX5_FLOW_ACTION_MARK)
6050 			return rte_flow_error_set(error, ENOTSUP,
6051 						  RTE_FLOW_ERROR_TYPE_ACTION,
6052 						  NULL,
6053 						  "unsupported action MARK");
6054 		if (action_flags & MLX5_FLOW_ACTION_QUEUE)
6055 			return rte_flow_error_set(error, ENOTSUP,
6056 						  RTE_FLOW_ERROR_TYPE_ACTION,
6057 						  NULL,
6058 						  "unsupported action QUEUE");
6059 		if (action_flags & MLX5_FLOW_ACTION_RSS)
6060 			return rte_flow_error_set(error, ENOTSUP,
6061 						  RTE_FLOW_ERROR_TYPE_ACTION,
6062 						  NULL,
6063 						  "unsupported action RSS");
6064 		if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
6065 			return rte_flow_error_set(error, EINVAL,
6066 						  RTE_FLOW_ERROR_TYPE_ACTION,
6067 						  actions,
6068 						  "no fate action is found");
6069 	} else {
6070 		if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
6071 			return rte_flow_error_set(error, EINVAL,
6072 						  RTE_FLOW_ERROR_TYPE_ACTION,
6073 						  actions,
6074 						  "no fate action is found");
6075 	}
6076 	/* Continue validation for Xcap and VLAN actions.*/
6077 	if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
6078 			     MLX5_FLOW_VLAN_ACTIONS)) &&
6079 	    (queue_index == 0xFFFF ||
6080 	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
6081 		if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
6082 		    MLX5_FLOW_XCAP_ACTIONS)
6083 			return rte_flow_error_set(error, ENOTSUP,
6084 						  RTE_FLOW_ERROR_TYPE_ACTION,
6085 						  NULL, "encap and decap "
6086 						  "combination aren't supported");
6087 		if (!attr->transfer && attr->ingress) {
6088 			if (action_flags & MLX5_FLOW_ACTION_ENCAP)
6089 				return rte_flow_error_set
6090 						(error, ENOTSUP,
6091 						 RTE_FLOW_ERROR_TYPE_ACTION,
6092 						 NULL, "encap is not supported"
6093 						 " for ingress traffic");
6094 			else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
6095 				return rte_flow_error_set
6096 						(error, ENOTSUP,
6097 						 RTE_FLOW_ERROR_TYPE_ACTION,
6098 						 NULL, "push VLAN action not "
6099 						 "supported for ingress");
6100 			else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
6101 					MLX5_FLOW_VLAN_ACTIONS)
6102 				return rte_flow_error_set
6103 						(error, ENOTSUP,
6104 						 RTE_FLOW_ERROR_TYPE_ACTION,
6105 						 NULL, "no support for "
6106 						 "multiple VLAN actions");
6107 		}
6108 	}
6109 	/* Hairpin flow will add one more TAG action. */
6110 	if (hairpin > 0)
6111 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
6112 	/* extra metadata enabled: one more TAG action will be add. */
6113 	if (dev_conf->dv_flow_en &&
6114 	    dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
6115 	    mlx5_flow_ext_mreg_supported(dev))
6116 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
6117 	if ((uint32_t)rw_act_num >
6118 			flow_dv_modify_hdr_action_max(dev, is_root)) {
6119 		return rte_flow_error_set(error, ENOTSUP,
6120 					  RTE_FLOW_ERROR_TYPE_ACTION,
6121 					  NULL, "too many header modify"
6122 					  " actions to support");
6123 	}
6124 	return 0;
6125 }
6126 
6127 /**
6128  * Internal preparation function. Allocates the DV flow size,
6129  * this size is constant.
6130  *
6131  * @param[in] dev
6132  *   Pointer to the rte_eth_dev structure.
6133  * @param[in] attr
6134  *   Pointer to the flow attributes.
6135  * @param[in] items
6136  *   Pointer to the list of items.
6137  * @param[in] actions
6138  *   Pointer to the list of actions.
6139  * @param[out] error
6140  *   Pointer to the error structure.
6141  *
6142  * @return
6143  *   Pointer to mlx5_flow object on success,
6144  *   otherwise NULL and rte_errno is set.
6145  */
6146 static struct mlx5_flow *
6147 flow_dv_prepare(struct rte_eth_dev *dev,
6148 		const struct rte_flow_attr *attr __rte_unused,
6149 		const struct rte_flow_item items[] __rte_unused,
6150 		const struct rte_flow_action actions[] __rte_unused,
6151 		struct rte_flow_error *error)
6152 {
6153 	uint32_t handle_idx = 0;
6154 	struct mlx5_flow *dev_flow;
6155 	struct mlx5_flow_handle *dev_handle;
6156 	struct mlx5_priv *priv = dev->data->dev_private;
6157 
6158 	/* In case of corrupting the memory. */
6159 	if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
6160 		rte_flow_error_set(error, ENOSPC,
6161 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6162 				   "not free temporary device flow");
6163 		return NULL;
6164 	}
6165 	dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
6166 				   &handle_idx);
6167 	if (!dev_handle) {
6168 		rte_flow_error_set(error, ENOMEM,
6169 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
6170 				   "not enough memory to create flow handle");
6171 		return NULL;
6172 	}
6173 	/* No multi-thread supporting. */
6174 	dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
6175 	dev_flow->handle = dev_handle;
6176 	dev_flow->handle_idx = handle_idx;
6177 	/*
6178 	 * In some old rdma-core releases, before continuing, a check of the
6179 	 * length of matching parameter will be done at first. It needs to use
6180 	 * the length without misc4 param. If the flow has misc4 support, then
6181 	 * the length needs to be adjusted accordingly. Each param member is
6182 	 * aligned with a 64B boundary naturally.
6183 	 */
6184 	dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
6185 				  MLX5_ST_SZ_BYTES(fte_match_set_misc4);
6186 	/*
6187 	 * The matching value needs to be cleared to 0 before using. In the
6188 	 * past, it will be automatically cleared when using rte_*alloc
6189 	 * API. The time consumption will be almost the same as before.
6190 	 */
6191 	memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
6192 	dev_flow->ingress = attr->ingress;
6193 	dev_flow->dv.transfer = attr->transfer;
6194 	return dev_flow;
6195 }
6196 
6197 #ifdef RTE_LIBRTE_MLX5_DEBUG
6198 /**
6199  * Sanity check for match mask and value. Similar to check_valid_spec() in
6200  * kernel driver. If unmasked bit is present in value, it returns failure.
6201  *
6202  * @param match_mask
6203  *   pointer to match mask buffer.
6204  * @param match_value
6205  *   pointer to match value buffer.
6206  *
6207  * @return
6208  *   0 if valid, -EINVAL otherwise.
6209  */
6210 static int
6211 flow_dv_check_valid_spec(void *match_mask, void *match_value)
6212 {
6213 	uint8_t *m = match_mask;
6214 	uint8_t *v = match_value;
6215 	unsigned int i;
6216 
6217 	for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
6218 		if (v[i] & ~m[i]) {
6219 			DRV_LOG(ERR,
6220 				"match_value differs from match_criteria"
6221 				" %p[%u] != %p[%u]",
6222 				match_value, i, match_mask, i);
6223 			return -EINVAL;
6224 		}
6225 	}
6226 	return 0;
6227 }
6228 #endif
6229 
6230 /**
6231  * Add match of ip_version.
6232  *
6233  * @param[in] group
6234  *   Flow group.
6235  * @param[in] headers_v
6236  *   Values header pointer.
6237  * @param[in] headers_m
6238  *   Masks header pointer.
6239  * @param[in] ip_version
6240  *   The IP version to set.
6241  */
6242 static inline void
6243 flow_dv_set_match_ip_version(uint32_t group,
6244 			     void *headers_v,
6245 			     void *headers_m,
6246 			     uint8_t ip_version)
6247 {
6248 	if (group == 0)
6249 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
6250 	else
6251 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
6252 			 ip_version);
6253 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
6254 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
6255 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
6256 }
6257 
6258 /**
6259  * Add Ethernet item to matcher and to the value.
6260  *
6261  * @param[in, out] matcher
6262  *   Flow matcher.
6263  * @param[in, out] key
6264  *   Flow matcher value.
6265  * @param[in] item
6266  *   Flow pattern to translate.
6267  * @param[in] inner
6268  *   Item is inner pattern.
6269  */
6270 static void
6271 flow_dv_translate_item_eth(void *matcher, void *key,
6272 			   const struct rte_flow_item *item, int inner,
6273 			   uint32_t group)
6274 {
6275 	const struct rte_flow_item_eth *eth_m = item->mask;
6276 	const struct rte_flow_item_eth *eth_v = item->spec;
6277 	const struct rte_flow_item_eth nic_mask = {
6278 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6279 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
6280 		.type = RTE_BE16(0xffff),
6281 	};
6282 	void *headers_m;
6283 	void *headers_v;
6284 	char *l24_v;
6285 	unsigned int i;
6286 
6287 	if (!eth_v)
6288 		return;
6289 	if (!eth_m)
6290 		eth_m = &nic_mask;
6291 	if (inner) {
6292 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6293 					 inner_headers);
6294 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6295 	} else {
6296 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6297 					 outer_headers);
6298 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6299 	}
6300 	memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
6301 	       &eth_m->dst, sizeof(eth_m->dst));
6302 	/* The value must be in the range of the mask. */
6303 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
6304 	for (i = 0; i < sizeof(eth_m->dst); ++i)
6305 		l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
6306 	memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
6307 	       &eth_m->src, sizeof(eth_m->src));
6308 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
6309 	/* The value must be in the range of the mask. */
6310 	for (i = 0; i < sizeof(eth_m->dst); ++i)
6311 		l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
6312 	if (eth_v->type) {
6313 		/* When ethertype is present set mask for tagged VLAN. */
6314 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6315 		/* Set value for tagged VLAN if ethertype is 802.1Q. */
6316 		if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
6317 		    eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
6318 			MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
6319 				 1);
6320 			/* Return here to avoid setting match on ethertype. */
6321 			return;
6322 		}
6323 	}
6324 	/*
6325 	 * HW supports match on one Ethertype, the Ethertype following the last
6326 	 * VLAN tag of the packet (see PRM).
6327 	 * Set match on ethertype only if ETH header is not followed by VLAN.
6328 	 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6329 	 * ethertype, and use ip_version field instead.
6330 	 * eCPRI over Ether layer will use type value 0xAEFE.
6331 	 */
6332 	if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6333 	    eth_m->type == 0xFFFF) {
6334 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6335 	} else if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6336 		   eth_m->type == 0xFFFF) {
6337 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6338 	} else {
6339 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6340 			 rte_be_to_cpu_16(eth_m->type));
6341 		l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6342 				     ethertype);
6343 		*(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6344 	}
6345 }
6346 
6347 /**
6348  * Add VLAN item to matcher and to the value.
6349  *
6350  * @param[in, out] dev_flow
6351  *   Flow descriptor.
6352  * @param[in, out] matcher
6353  *   Flow matcher.
6354  * @param[in, out] key
6355  *   Flow matcher value.
6356  * @param[in] item
6357  *   Flow pattern to translate.
6358  * @param[in] inner
6359  *   Item is inner pattern.
6360  */
6361 static void
6362 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6363 			    void *matcher, void *key,
6364 			    const struct rte_flow_item *item,
6365 			    int inner, uint32_t group)
6366 {
6367 	const struct rte_flow_item_vlan *vlan_m = item->mask;
6368 	const struct rte_flow_item_vlan *vlan_v = item->spec;
6369 	void *headers_m;
6370 	void *headers_v;
6371 	uint16_t tci_m;
6372 	uint16_t tci_v;
6373 
6374 	if (inner) {
6375 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6376 					 inner_headers);
6377 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6378 	} else {
6379 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6380 					 outer_headers);
6381 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6382 		/*
6383 		 * This is workaround, masks are not supported,
6384 		 * and pre-validated.
6385 		 */
6386 		if (vlan_v)
6387 			dev_flow->handle->vf_vlan.tag =
6388 					rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6389 	}
6390 	/*
6391 	 * When VLAN item exists in flow, mark packet as tagged,
6392 	 * even if TCI is not specified.
6393 	 */
6394 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6395 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
6396 	if (!vlan_v)
6397 		return;
6398 	if (!vlan_m)
6399 		vlan_m = &rte_flow_item_vlan_mask;
6400 	tci_m = rte_be_to_cpu_16(vlan_m->tci);
6401 	tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6402 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
6403 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
6404 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
6405 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
6406 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
6407 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
6408 	/*
6409 	 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6410 	 * ethertype, and use ip_version field instead.
6411 	 */
6412 	if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6413 	    vlan_m->inner_type == 0xFFFF) {
6414 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6415 	} else if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6416 		   vlan_m->inner_type == 0xFFFF) {
6417 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6418 	} else {
6419 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6420 			 rte_be_to_cpu_16(vlan_m->inner_type));
6421 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
6422 			 rte_be_to_cpu_16(vlan_m->inner_type &
6423 					  vlan_v->inner_type));
6424 	}
6425 }
6426 
6427 /**
6428  * Add IPV4 item to matcher and to the value.
6429  *
6430  * @param[in, out] matcher
6431  *   Flow matcher.
6432  * @param[in, out] key
6433  *   Flow matcher value.
6434  * @param[in] item
6435  *   Flow pattern to translate.
6436  * @param[in] item_flags
6437  *   Bit-fields that holds the items detected until now.
6438  * @param[in] inner
6439  *   Item is inner pattern.
6440  * @param[in] group
6441  *   The group to insert the rule.
6442  */
6443 static void
6444 flow_dv_translate_item_ipv4(void *matcher, void *key,
6445 			    const struct rte_flow_item *item,
6446 			    const uint64_t item_flags,
6447 			    int inner, uint32_t group)
6448 {
6449 	const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6450 	const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6451 	const struct rte_flow_item_ipv4 nic_mask = {
6452 		.hdr = {
6453 			.src_addr = RTE_BE32(0xffffffff),
6454 			.dst_addr = RTE_BE32(0xffffffff),
6455 			.type_of_service = 0xff,
6456 			.next_proto_id = 0xff,
6457 			.time_to_live = 0xff,
6458 		},
6459 	};
6460 	void *headers_m;
6461 	void *headers_v;
6462 	char *l24_m;
6463 	char *l24_v;
6464 	uint8_t tos;
6465 
6466 	if (inner) {
6467 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6468 					 inner_headers);
6469 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6470 	} else {
6471 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6472 					 outer_headers);
6473 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6474 	}
6475 	flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6476 	/*
6477 	 * On outer header (which must contains L2), or inner header with L2,
6478 	 * set cvlan_tag mask bit to mark this packet as untagged.
6479 	 * This should be done even if item->spec is empty.
6480 	 */
6481 	if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6482 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6483 	if (!ipv4_v)
6484 		return;
6485 	if (!ipv4_m)
6486 		ipv4_m = &nic_mask;
6487 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6488 			     dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6489 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6490 			     dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6491 	*(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6492 	*(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6493 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6494 			  src_ipv4_src_ipv6.ipv4_layout.ipv4);
6495 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6496 			  src_ipv4_src_ipv6.ipv4_layout.ipv4);
6497 	*(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6498 	*(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6499 	tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6500 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6501 		 ipv4_m->hdr.type_of_service);
6502 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6503 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6504 		 ipv4_m->hdr.type_of_service >> 2);
6505 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6506 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6507 		 ipv4_m->hdr.next_proto_id);
6508 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6509 		 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6510 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6511 		 ipv4_m->hdr.time_to_live);
6512 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6513 		 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6514 }
6515 
6516 /**
6517  * Add IPV6 item to matcher and to the value.
6518  *
6519  * @param[in, out] matcher
6520  *   Flow matcher.
6521  * @param[in, out] key
6522  *   Flow matcher value.
6523  * @param[in] item
6524  *   Flow pattern to translate.
6525  * @param[in] item_flags
6526  *   Bit-fields that holds the items detected until now.
6527  * @param[in] inner
6528  *   Item is inner pattern.
6529  * @param[in] group
6530  *   The group to insert the rule.
6531  */
6532 static void
6533 flow_dv_translate_item_ipv6(void *matcher, void *key,
6534 			    const struct rte_flow_item *item,
6535 			    const uint64_t item_flags,
6536 			    int inner, uint32_t group)
6537 {
6538 	const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6539 	const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6540 	const struct rte_flow_item_ipv6 nic_mask = {
6541 		.hdr = {
6542 			.src_addr =
6543 				"\xff\xff\xff\xff\xff\xff\xff\xff"
6544 				"\xff\xff\xff\xff\xff\xff\xff\xff",
6545 			.dst_addr =
6546 				"\xff\xff\xff\xff\xff\xff\xff\xff"
6547 				"\xff\xff\xff\xff\xff\xff\xff\xff",
6548 			.vtc_flow = RTE_BE32(0xffffffff),
6549 			.proto = 0xff,
6550 			.hop_limits = 0xff,
6551 		},
6552 	};
6553 	void *headers_m;
6554 	void *headers_v;
6555 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6556 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6557 	char *l24_m;
6558 	char *l24_v;
6559 	uint32_t vtc_m;
6560 	uint32_t vtc_v;
6561 	int i;
6562 	int size;
6563 
6564 	if (inner) {
6565 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6566 					 inner_headers);
6567 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6568 	} else {
6569 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6570 					 outer_headers);
6571 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6572 	}
6573 	flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6574 	/*
6575 	 * On outer header (which must contains L2), or inner header with L2,
6576 	 * set cvlan_tag mask bit to mark this packet as untagged.
6577 	 * This should be done even if item->spec is empty.
6578 	 */
6579 	if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6580 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6581 	if (!ipv6_v)
6582 		return;
6583 	if (!ipv6_m)
6584 		ipv6_m = &nic_mask;
6585 	size = sizeof(ipv6_m->hdr.dst_addr);
6586 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6587 			     dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6588 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6589 			     dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6590 	memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6591 	for (i = 0; i < size; ++i)
6592 		l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6593 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6594 			     src_ipv4_src_ipv6.ipv6_layout.ipv6);
6595 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6596 			     src_ipv4_src_ipv6.ipv6_layout.ipv6);
6597 	memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6598 	for (i = 0; i < size; ++i)
6599 		l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6600 	/* TOS. */
6601 	vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6602 	vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6603 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6604 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6605 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6606 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6607 	/* Label. */
6608 	if (inner) {
6609 		MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6610 			 vtc_m);
6611 		MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6612 			 vtc_v);
6613 	} else {
6614 		MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6615 			 vtc_m);
6616 		MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6617 			 vtc_v);
6618 	}
6619 	/* Protocol. */
6620 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6621 		 ipv6_m->hdr.proto);
6622 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6623 		 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6624 	/* Hop limit. */
6625 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6626 		 ipv6_m->hdr.hop_limits);
6627 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6628 		 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6629 }
6630 
6631 /**
6632  * Add TCP item to matcher and to the value.
6633  *
6634  * @param[in, out] matcher
6635  *   Flow matcher.
6636  * @param[in, out] key
6637  *   Flow matcher value.
6638  * @param[in] item
6639  *   Flow pattern to translate.
6640  * @param[in] inner
6641  *   Item is inner pattern.
6642  */
6643 static void
6644 flow_dv_translate_item_tcp(void *matcher, void *key,
6645 			   const struct rte_flow_item *item,
6646 			   int inner)
6647 {
6648 	const struct rte_flow_item_tcp *tcp_m = item->mask;
6649 	const struct rte_flow_item_tcp *tcp_v = item->spec;
6650 	void *headers_m;
6651 	void *headers_v;
6652 
6653 	if (inner) {
6654 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6655 					 inner_headers);
6656 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6657 	} else {
6658 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6659 					 outer_headers);
6660 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6661 	}
6662 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6663 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6664 	if (!tcp_v)
6665 		return;
6666 	if (!tcp_m)
6667 		tcp_m = &rte_flow_item_tcp_mask;
6668 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6669 		 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6670 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6671 		 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6672 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6673 		 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6674 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6675 		 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6676 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6677 		 tcp_m->hdr.tcp_flags);
6678 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6679 		 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6680 }
6681 
6682 /**
6683  * Add UDP item to matcher and to the value.
6684  *
6685  * @param[in, out] matcher
6686  *   Flow matcher.
6687  * @param[in, out] key
6688  *   Flow matcher value.
6689  * @param[in] item
6690  *   Flow pattern to translate.
6691  * @param[in] inner
6692  *   Item is inner pattern.
6693  */
6694 static void
6695 flow_dv_translate_item_udp(void *matcher, void *key,
6696 			   const struct rte_flow_item *item,
6697 			   int inner)
6698 {
6699 	const struct rte_flow_item_udp *udp_m = item->mask;
6700 	const struct rte_flow_item_udp *udp_v = item->spec;
6701 	void *headers_m;
6702 	void *headers_v;
6703 
6704 	if (inner) {
6705 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6706 					 inner_headers);
6707 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6708 	} else {
6709 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6710 					 outer_headers);
6711 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6712 	}
6713 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6714 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6715 	if (!udp_v)
6716 		return;
6717 	if (!udp_m)
6718 		udp_m = &rte_flow_item_udp_mask;
6719 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6720 		 rte_be_to_cpu_16(udp_m->hdr.src_port));
6721 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6722 		 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6723 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6724 		 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6725 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6726 		 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6727 }
6728 
6729 /**
6730  * Add GRE optional Key item to matcher and to the value.
6731  *
6732  * @param[in, out] matcher
6733  *   Flow matcher.
6734  * @param[in, out] key
6735  *   Flow matcher value.
6736  * @param[in] item
6737  *   Flow pattern to translate.
6738  * @param[in] inner
6739  *   Item is inner pattern.
6740  */
6741 static void
6742 flow_dv_translate_item_gre_key(void *matcher, void *key,
6743 				   const struct rte_flow_item *item)
6744 {
6745 	const rte_be32_t *key_m = item->mask;
6746 	const rte_be32_t *key_v = item->spec;
6747 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6748 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6749 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6750 
6751 	/* GRE K bit must be on and should already be validated */
6752 	MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6753 	MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6754 	if (!key_v)
6755 		return;
6756 	if (!key_m)
6757 		key_m = &gre_key_default_mask;
6758 	MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6759 		 rte_be_to_cpu_32(*key_m) >> 8);
6760 	MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6761 		 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6762 	MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6763 		 rte_be_to_cpu_32(*key_m) & 0xFF);
6764 	MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6765 		 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6766 }
6767 
6768 /**
6769  * Add GRE item to matcher and to the value.
6770  *
6771  * @param[in, out] matcher
6772  *   Flow matcher.
6773  * @param[in, out] key
6774  *   Flow matcher value.
6775  * @param[in] item
6776  *   Flow pattern to translate.
6777  * @param[in] inner
6778  *   Item is inner pattern.
6779  */
6780 static void
6781 flow_dv_translate_item_gre(void *matcher, void *key,
6782 			   const struct rte_flow_item *item,
6783 			   int inner)
6784 {
6785 	const struct rte_flow_item_gre *gre_m = item->mask;
6786 	const struct rte_flow_item_gre *gre_v = item->spec;
6787 	void *headers_m;
6788 	void *headers_v;
6789 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6790 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6791 	struct {
6792 		union {
6793 			__extension__
6794 			struct {
6795 				uint16_t version:3;
6796 				uint16_t rsvd0:9;
6797 				uint16_t s_present:1;
6798 				uint16_t k_present:1;
6799 				uint16_t rsvd_bit1:1;
6800 				uint16_t c_present:1;
6801 			};
6802 			uint16_t value;
6803 		};
6804 	} gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6805 
6806 	if (inner) {
6807 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6808 					 inner_headers);
6809 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6810 	} else {
6811 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6812 					 outer_headers);
6813 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6814 	}
6815 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6816 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6817 	if (!gre_v)
6818 		return;
6819 	if (!gre_m)
6820 		gre_m = &rte_flow_item_gre_mask;
6821 	MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6822 		 rte_be_to_cpu_16(gre_m->protocol));
6823 	MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6824 		 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6825 	gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6826 	gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6827 	MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6828 		 gre_crks_rsvd0_ver_m.c_present);
6829 	MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6830 		 gre_crks_rsvd0_ver_v.c_present &
6831 		 gre_crks_rsvd0_ver_m.c_present);
6832 	MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6833 		 gre_crks_rsvd0_ver_m.k_present);
6834 	MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6835 		 gre_crks_rsvd0_ver_v.k_present &
6836 		 gre_crks_rsvd0_ver_m.k_present);
6837 	MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6838 		 gre_crks_rsvd0_ver_m.s_present);
6839 	MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6840 		 gre_crks_rsvd0_ver_v.s_present &
6841 		 gre_crks_rsvd0_ver_m.s_present);
6842 }
6843 
6844 /**
6845  * Add NVGRE item to matcher and to the value.
6846  *
6847  * @param[in, out] matcher
6848  *   Flow matcher.
6849  * @param[in, out] key
6850  *   Flow matcher value.
6851  * @param[in] item
6852  *   Flow pattern to translate.
6853  * @param[in] inner
6854  *   Item is inner pattern.
6855  */
6856 static void
6857 flow_dv_translate_item_nvgre(void *matcher, void *key,
6858 			     const struct rte_flow_item *item,
6859 			     int inner)
6860 {
6861 	const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6862 	const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6863 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6864 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6865 	const char *tni_flow_id_m;
6866 	const char *tni_flow_id_v;
6867 	char *gre_key_m;
6868 	char *gre_key_v;
6869 	int size;
6870 	int i;
6871 
6872 	/* For NVGRE, GRE header fields must be set with defined values. */
6873 	const struct rte_flow_item_gre gre_spec = {
6874 		.c_rsvd0_ver = RTE_BE16(0x2000),
6875 		.protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6876 	};
6877 	const struct rte_flow_item_gre gre_mask = {
6878 		.c_rsvd0_ver = RTE_BE16(0xB000),
6879 		.protocol = RTE_BE16(UINT16_MAX),
6880 	};
6881 	const struct rte_flow_item gre_item = {
6882 		.spec = &gre_spec,
6883 		.mask = &gre_mask,
6884 		.last = NULL,
6885 	};
6886 	flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6887 	if (!nvgre_v)
6888 		return;
6889 	if (!nvgre_m)
6890 		nvgre_m = &rte_flow_item_nvgre_mask;
6891 	tni_flow_id_m = (const char *)nvgre_m->tni;
6892 	tni_flow_id_v = (const char *)nvgre_v->tni;
6893 	size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6894 	gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6895 	gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6896 	memcpy(gre_key_m, tni_flow_id_m, size);
6897 	for (i = 0; i < size; ++i)
6898 		gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6899 }
6900 
6901 /**
6902  * Add VXLAN item to matcher and to the value.
6903  *
6904  * @param[in, out] matcher
6905  *   Flow matcher.
6906  * @param[in, out] key
6907  *   Flow matcher value.
6908  * @param[in] item
6909  *   Flow pattern to translate.
6910  * @param[in] inner
6911  *   Item is inner pattern.
6912  */
6913 static void
6914 flow_dv_translate_item_vxlan(void *matcher, void *key,
6915 			     const struct rte_flow_item *item,
6916 			     int inner)
6917 {
6918 	const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6919 	const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6920 	void *headers_m;
6921 	void *headers_v;
6922 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6923 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6924 	char *vni_m;
6925 	char *vni_v;
6926 	uint16_t dport;
6927 	int size;
6928 	int i;
6929 
6930 	if (inner) {
6931 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6932 					 inner_headers);
6933 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6934 	} else {
6935 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6936 					 outer_headers);
6937 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6938 	}
6939 	dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6940 		MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6941 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6942 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6943 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6944 	}
6945 	if (!vxlan_v)
6946 		return;
6947 	if (!vxlan_m)
6948 		vxlan_m = &rte_flow_item_vxlan_mask;
6949 	size = sizeof(vxlan_m->vni);
6950 	vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6951 	vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6952 	memcpy(vni_m, vxlan_m->vni, size);
6953 	for (i = 0; i < size; ++i)
6954 		vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6955 }
6956 
6957 /**
6958  * Add VXLAN-GPE item to matcher and to the value.
6959  *
6960  * @param[in, out] matcher
6961  *   Flow matcher.
6962  * @param[in, out] key
6963  *   Flow matcher value.
6964  * @param[in] item
6965  *   Flow pattern to translate.
6966  * @param[in] inner
6967  *   Item is inner pattern.
6968  */
6969 
6970 static void
6971 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6972 				 const struct rte_flow_item *item, int inner)
6973 {
6974 	const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6975 	const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6976 	void *headers_m;
6977 	void *headers_v;
6978 	void *misc_m =
6979 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6980 	void *misc_v =
6981 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6982 	char *vni_m;
6983 	char *vni_v;
6984 	uint16_t dport;
6985 	int size;
6986 	int i;
6987 	uint8_t flags_m = 0xff;
6988 	uint8_t flags_v = 0xc;
6989 
6990 	if (inner) {
6991 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6992 					 inner_headers);
6993 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6994 	} else {
6995 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6996 					 outer_headers);
6997 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6998 	}
6999 	dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
7000 		MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
7001 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7002 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7003 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7004 	}
7005 	if (!vxlan_v)
7006 		return;
7007 	if (!vxlan_m)
7008 		vxlan_m = &rte_flow_item_vxlan_gpe_mask;
7009 	size = sizeof(vxlan_m->vni);
7010 	vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
7011 	vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
7012 	memcpy(vni_m, vxlan_m->vni, size);
7013 	for (i = 0; i < size; ++i)
7014 		vni_v[i] = vni_m[i] & vxlan_v->vni[i];
7015 	if (vxlan_m->flags) {
7016 		flags_m = vxlan_m->flags;
7017 		flags_v = vxlan_v->flags;
7018 	}
7019 	MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
7020 	MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
7021 	MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
7022 		 vxlan_m->protocol);
7023 	MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
7024 		 vxlan_v->protocol);
7025 }
7026 
7027 /**
7028  * Add Geneve item to matcher and to the value.
7029  *
7030  * @param[in, out] matcher
7031  *   Flow matcher.
7032  * @param[in, out] key
7033  *   Flow matcher value.
7034  * @param[in] item
7035  *   Flow pattern to translate.
7036  * @param[in] inner
7037  *   Item is inner pattern.
7038  */
7039 
7040 static void
7041 flow_dv_translate_item_geneve(void *matcher, void *key,
7042 			      const struct rte_flow_item *item, int inner)
7043 {
7044 	const struct rte_flow_item_geneve *geneve_m = item->mask;
7045 	const struct rte_flow_item_geneve *geneve_v = item->spec;
7046 	void *headers_m;
7047 	void *headers_v;
7048 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7049 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7050 	uint16_t dport;
7051 	uint16_t gbhdr_m;
7052 	uint16_t gbhdr_v;
7053 	char *vni_m;
7054 	char *vni_v;
7055 	size_t size, i;
7056 
7057 	if (inner) {
7058 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7059 					 inner_headers);
7060 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7061 	} else {
7062 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7063 					 outer_headers);
7064 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7065 	}
7066 	dport = MLX5_UDP_PORT_GENEVE;
7067 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7068 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7069 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7070 	}
7071 	if (!geneve_v)
7072 		return;
7073 	if (!geneve_m)
7074 		geneve_m = &rte_flow_item_geneve_mask;
7075 	size = sizeof(geneve_m->vni);
7076 	vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
7077 	vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
7078 	memcpy(vni_m, geneve_m->vni, size);
7079 	for (i = 0; i < size; ++i)
7080 		vni_v[i] = vni_m[i] & geneve_v->vni[i];
7081 	MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
7082 		 rte_be_to_cpu_16(geneve_m->protocol));
7083 	MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
7084 		 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
7085 	gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
7086 	gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
7087 	MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
7088 		 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7089 	MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
7090 		 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
7091 	MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
7092 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7093 	MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
7094 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
7095 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
7096 }
7097 
7098 /**
7099  * Add MPLS item to matcher and to the value.
7100  *
7101  * @param[in, out] matcher
7102  *   Flow matcher.
7103  * @param[in, out] key
7104  *   Flow matcher value.
7105  * @param[in] item
7106  *   Flow pattern to translate.
7107  * @param[in] prev_layer
7108  *   The protocol layer indicated in previous item.
7109  * @param[in] inner
7110  *   Item is inner pattern.
7111  */
7112 static void
7113 flow_dv_translate_item_mpls(void *matcher, void *key,
7114 			    const struct rte_flow_item *item,
7115 			    uint64_t prev_layer,
7116 			    int inner)
7117 {
7118 	const uint32_t *in_mpls_m = item->mask;
7119 	const uint32_t *in_mpls_v = item->spec;
7120 	uint32_t *out_mpls_m = 0;
7121 	uint32_t *out_mpls_v = 0;
7122 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7123 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7124 	void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
7125 				     misc_parameters_2);
7126 	void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7127 	void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
7128 	void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7129 
7130 	switch (prev_layer) {
7131 	case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7132 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
7133 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
7134 			 MLX5_UDP_PORT_MPLS);
7135 		break;
7136 	case MLX5_FLOW_LAYER_GRE:
7137 		MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
7138 		MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
7139 			 RTE_ETHER_TYPE_MPLS);
7140 		break;
7141 	default:
7142 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
7143 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
7144 			 IPPROTO_MPLS);
7145 		break;
7146 	}
7147 	if (!in_mpls_v)
7148 		return;
7149 	if (!in_mpls_m)
7150 		in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
7151 	switch (prev_layer) {
7152 	case MLX5_FLOW_LAYER_OUTER_L4_UDP:
7153 		out_mpls_m =
7154 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7155 						 outer_first_mpls_over_udp);
7156 		out_mpls_v =
7157 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7158 						 outer_first_mpls_over_udp);
7159 		break;
7160 	case MLX5_FLOW_LAYER_GRE:
7161 		out_mpls_m =
7162 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
7163 						 outer_first_mpls_over_gre);
7164 		out_mpls_v =
7165 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
7166 						 outer_first_mpls_over_gre);
7167 		break;
7168 	default:
7169 		/* Inner MPLS not over GRE is not supported. */
7170 		if (!inner) {
7171 			out_mpls_m =
7172 				(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7173 							 misc2_m,
7174 							 outer_first_mpls);
7175 			out_mpls_v =
7176 				(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
7177 							 misc2_v,
7178 							 outer_first_mpls);
7179 		}
7180 		break;
7181 	}
7182 	if (out_mpls_m && out_mpls_v) {
7183 		*out_mpls_m = *in_mpls_m;
7184 		*out_mpls_v = *in_mpls_v & *in_mpls_m;
7185 	}
7186 }
7187 
7188 /**
7189  * Add metadata register item to matcher
7190  *
7191  * @param[in, out] matcher
7192  *   Flow matcher.
7193  * @param[in, out] key
7194  *   Flow matcher value.
7195  * @param[in] reg_type
7196  *   Type of device metadata register
7197  * @param[in] value
7198  *   Register value
7199  * @param[in] mask
7200  *   Register mask
7201  */
7202 static void
7203 flow_dv_match_meta_reg(void *matcher, void *key,
7204 		       enum modify_reg reg_type,
7205 		       uint32_t data, uint32_t mask)
7206 {
7207 	void *misc2_m =
7208 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
7209 	void *misc2_v =
7210 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
7211 	uint32_t temp;
7212 
7213 	data &= mask;
7214 	switch (reg_type) {
7215 	case REG_A:
7216 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
7217 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
7218 		break;
7219 	case REG_B:
7220 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
7221 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
7222 		break;
7223 	case REG_C_0:
7224 		/*
7225 		 * The metadata register C0 field might be divided into
7226 		 * source vport index and META item value, we should set
7227 		 * this field according to specified mask, not as whole one.
7228 		 */
7229 		temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
7230 		temp |= mask;
7231 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
7232 		temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
7233 		temp &= ~mask;
7234 		temp |= data;
7235 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
7236 		break;
7237 	case REG_C_1:
7238 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
7239 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
7240 		break;
7241 	case REG_C_2:
7242 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
7243 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
7244 		break;
7245 	case REG_C_3:
7246 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
7247 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
7248 		break;
7249 	case REG_C_4:
7250 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
7251 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
7252 		break;
7253 	case REG_C_5:
7254 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
7255 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
7256 		break;
7257 	case REG_C_6:
7258 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
7259 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
7260 		break;
7261 	case REG_C_7:
7262 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
7263 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
7264 		break;
7265 	default:
7266 		MLX5_ASSERT(false);
7267 		break;
7268 	}
7269 }
7270 
7271 /**
7272  * Add MARK item to matcher
7273  *
7274  * @param[in] dev
7275  *   The device to configure through.
7276  * @param[in, out] matcher
7277  *   Flow matcher.
7278  * @param[in, out] key
7279  *   Flow matcher value.
7280  * @param[in] item
7281  *   Flow pattern to translate.
7282  */
7283 static void
7284 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
7285 			    void *matcher, void *key,
7286 			    const struct rte_flow_item *item)
7287 {
7288 	struct mlx5_priv *priv = dev->data->dev_private;
7289 	const struct rte_flow_item_mark *mark;
7290 	uint32_t value;
7291 	uint32_t mask;
7292 
7293 	mark = item->mask ? (const void *)item->mask :
7294 			    &rte_flow_item_mark_mask;
7295 	mask = mark->id & priv->sh->dv_mark_mask;
7296 	mark = (const void *)item->spec;
7297 	MLX5_ASSERT(mark);
7298 	value = mark->id & priv->sh->dv_mark_mask & mask;
7299 	if (mask) {
7300 		enum modify_reg reg;
7301 
7302 		/* Get the metadata register index for the mark. */
7303 		reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
7304 		MLX5_ASSERT(reg > 0);
7305 		if (reg == REG_C_0) {
7306 			struct mlx5_priv *priv = dev->data->dev_private;
7307 			uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7308 			uint32_t shl_c0 = rte_bsf32(msk_c0);
7309 
7310 			mask &= msk_c0;
7311 			mask <<= shl_c0;
7312 			value <<= shl_c0;
7313 		}
7314 		flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7315 	}
7316 }
7317 
7318 /**
7319  * Add META item to matcher
7320  *
7321  * @param[in] dev
7322  *   The devich to configure through.
7323  * @param[in, out] matcher
7324  *   Flow matcher.
7325  * @param[in, out] key
7326  *   Flow matcher value.
7327  * @param[in] attr
7328  *   Attributes of flow that includes this item.
7329  * @param[in] item
7330  *   Flow pattern to translate.
7331  */
7332 static void
7333 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7334 			    void *matcher, void *key,
7335 			    const struct rte_flow_attr *attr,
7336 			    const struct rte_flow_item *item)
7337 {
7338 	const struct rte_flow_item_meta *meta_m;
7339 	const struct rte_flow_item_meta *meta_v;
7340 
7341 	meta_m = (const void *)item->mask;
7342 	if (!meta_m)
7343 		meta_m = &rte_flow_item_meta_mask;
7344 	meta_v = (const void *)item->spec;
7345 	if (meta_v) {
7346 		int reg;
7347 		uint32_t value = meta_v->data;
7348 		uint32_t mask = meta_m->data;
7349 
7350 		reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7351 		if (reg < 0)
7352 			return;
7353 		/*
7354 		 * In datapath code there is no endianness
7355 		 * coversions for perfromance reasons, all
7356 		 * pattern conversions are done in rte_flow.
7357 		 */
7358 		value = rte_cpu_to_be_32(value);
7359 		mask = rte_cpu_to_be_32(mask);
7360 		if (reg == REG_C_0) {
7361 			struct mlx5_priv *priv = dev->data->dev_private;
7362 			uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7363 			uint32_t shl_c0 = rte_bsf32(msk_c0);
7364 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7365 			uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7366 
7367 			value >>= shr_c0;
7368 			mask >>= shr_c0;
7369 #endif
7370 			value <<= shl_c0;
7371 			mask <<= shl_c0;
7372 			MLX5_ASSERT(msk_c0);
7373 			MLX5_ASSERT(!(~msk_c0 & mask));
7374 		}
7375 		flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7376 	}
7377 }
7378 
7379 /**
7380  * Add vport metadata Reg C0 item to matcher
7381  *
7382  * @param[in, out] matcher
7383  *   Flow matcher.
7384  * @param[in, out] key
7385  *   Flow matcher value.
7386  * @param[in] reg
7387  *   Flow pattern to translate.
7388  */
7389 static void
7390 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7391 				  uint32_t value, uint32_t mask)
7392 {
7393 	flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7394 }
7395 
7396 /**
7397  * Add tag item to matcher
7398  *
7399  * @param[in] dev
7400  *   The devich to configure through.
7401  * @param[in, out] matcher
7402  *   Flow matcher.
7403  * @param[in, out] key
7404  *   Flow matcher value.
7405  * @param[in] item
7406  *   Flow pattern to translate.
7407  */
7408 static void
7409 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7410 				void *matcher, void *key,
7411 				const struct rte_flow_item *item)
7412 {
7413 	const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7414 	const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7415 	uint32_t mask, value;
7416 
7417 	MLX5_ASSERT(tag_v);
7418 	value = tag_v->data;
7419 	mask = tag_m ? tag_m->data : UINT32_MAX;
7420 	if (tag_v->id == REG_C_0) {
7421 		struct mlx5_priv *priv = dev->data->dev_private;
7422 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7423 		uint32_t shl_c0 = rte_bsf32(msk_c0);
7424 
7425 		mask &= msk_c0;
7426 		mask <<= shl_c0;
7427 		value <<= shl_c0;
7428 	}
7429 	flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7430 }
7431 
7432 /**
7433  * Add TAG item to matcher
7434  *
7435  * @param[in] dev
7436  *   The devich to configure through.
7437  * @param[in, out] matcher
7438  *   Flow matcher.
7439  * @param[in, out] key
7440  *   Flow matcher value.
7441  * @param[in] item
7442  *   Flow pattern to translate.
7443  */
7444 static void
7445 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7446 			   void *matcher, void *key,
7447 			   const struct rte_flow_item *item)
7448 {
7449 	const struct rte_flow_item_tag *tag_v = item->spec;
7450 	const struct rte_flow_item_tag *tag_m = item->mask;
7451 	enum modify_reg reg;
7452 
7453 	MLX5_ASSERT(tag_v);
7454 	tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7455 	/* Get the metadata register index for the tag. */
7456 	reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7457 	MLX5_ASSERT(reg > 0);
7458 	flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7459 }
7460 
7461 /**
7462  * Add source vport match to the specified matcher.
7463  *
7464  * @param[in, out] matcher
7465  *   Flow matcher.
7466  * @param[in, out] key
7467  *   Flow matcher value.
7468  * @param[in] port
7469  *   Source vport value to match
7470  * @param[in] mask
7471  *   Mask
7472  */
7473 static void
7474 flow_dv_translate_item_source_vport(void *matcher, void *key,
7475 				    int16_t port, uint16_t mask)
7476 {
7477 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7478 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7479 
7480 	MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7481 	MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7482 }
7483 
7484 /**
7485  * Translate port-id item to eswitch match on  port-id.
7486  *
7487  * @param[in] dev
7488  *   The devich to configure through.
7489  * @param[in, out] matcher
7490  *   Flow matcher.
7491  * @param[in, out] key
7492  *   Flow matcher value.
7493  * @param[in] item
7494  *   Flow pattern to translate.
7495  *
7496  * @return
7497  *   0 on success, a negative errno value otherwise.
7498  */
7499 static int
7500 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7501 			       void *key, const struct rte_flow_item *item)
7502 {
7503 	const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7504 	const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7505 	struct mlx5_priv *priv;
7506 	uint16_t mask, id;
7507 
7508 	mask = pid_m ? pid_m->id : 0xffff;
7509 	id = pid_v ? pid_v->id : dev->data->port_id;
7510 	priv = mlx5_port_to_eswitch_info(id, item == NULL);
7511 	if (!priv)
7512 		return -rte_errno;
7513 	/* Translate to vport field or to metadata, depending on mode. */
7514 	if (priv->vport_meta_mask)
7515 		flow_dv_translate_item_meta_vport(matcher, key,
7516 						  priv->vport_meta_tag,
7517 						  priv->vport_meta_mask);
7518 	else
7519 		flow_dv_translate_item_source_vport(matcher, key,
7520 						    priv->vport_id, mask);
7521 	return 0;
7522 }
7523 
7524 /**
7525  * Add ICMP6 item to matcher and to the value.
7526  *
7527  * @param[in, out] matcher
7528  *   Flow matcher.
7529  * @param[in, out] key
7530  *   Flow matcher value.
7531  * @param[in] item
7532  *   Flow pattern to translate.
7533  * @param[in] inner
7534  *   Item is inner pattern.
7535  */
7536 static void
7537 flow_dv_translate_item_icmp6(void *matcher, void *key,
7538 			      const struct rte_flow_item *item,
7539 			      int inner)
7540 {
7541 	const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7542 	const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7543 	void *headers_m;
7544 	void *headers_v;
7545 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7546 				     misc_parameters_3);
7547 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7548 	if (inner) {
7549 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7550 					 inner_headers);
7551 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7552 	} else {
7553 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7554 					 outer_headers);
7555 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7556 	}
7557 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7558 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7559 	if (!icmp6_v)
7560 		return;
7561 	if (!icmp6_m)
7562 		icmp6_m = &rte_flow_item_icmp6_mask;
7563 	/*
7564 	 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
7565 	 * If only the protocol is specified, no need to match the frag.
7566 	 */
7567 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7568 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7569 	MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7570 	MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7571 		 icmp6_v->type & icmp6_m->type);
7572 	MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7573 	MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7574 		 icmp6_v->code & icmp6_m->code);
7575 }
7576 
7577 /**
7578  * Add ICMP item to matcher and to the value.
7579  *
7580  * @param[in, out] matcher
7581  *   Flow matcher.
7582  * @param[in, out] key
7583  *   Flow matcher value.
7584  * @param[in] item
7585  *   Flow pattern to translate.
7586  * @param[in] inner
7587  *   Item is inner pattern.
7588  */
7589 static void
7590 flow_dv_translate_item_icmp(void *matcher, void *key,
7591 			    const struct rte_flow_item *item,
7592 			    int inner)
7593 {
7594 	const struct rte_flow_item_icmp *icmp_m = item->mask;
7595 	const struct rte_flow_item_icmp *icmp_v = item->spec;
7596 	uint32_t icmp_header_data_m = 0;
7597 	uint32_t icmp_header_data_v = 0;
7598 	void *headers_m;
7599 	void *headers_v;
7600 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7601 				     misc_parameters_3);
7602 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7603 	if (inner) {
7604 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7605 					 inner_headers);
7606 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7607 	} else {
7608 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7609 					 outer_headers);
7610 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7611 	}
7612 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7613 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7614 	if (!icmp_v)
7615 		return;
7616 	if (!icmp_m)
7617 		icmp_m = &rte_flow_item_icmp_mask;
7618 	/*
7619 	 * Force flow only to match the non-fragmented IPv4 ICMP packets.
7620 	 * If only the protocol is specified, no need to match the frag.
7621 	 */
7622 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7623 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7624 	MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7625 		 icmp_m->hdr.icmp_type);
7626 	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7627 		 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7628 	MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7629 		 icmp_m->hdr.icmp_code);
7630 	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7631 		 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7632 	icmp_header_data_m = rte_be_to_cpu_16(icmp_m->hdr.icmp_seq_nb);
7633 	icmp_header_data_m |= rte_be_to_cpu_16(icmp_m->hdr.icmp_ident) << 16;
7634 	if (icmp_header_data_m) {
7635 		icmp_header_data_v = rte_be_to_cpu_16(icmp_v->hdr.icmp_seq_nb);
7636 		icmp_header_data_v |=
7637 			 rte_be_to_cpu_16(icmp_v->hdr.icmp_ident) << 16;
7638 		MLX5_SET(fte_match_set_misc3, misc3_m, icmp_header_data,
7639 			 icmp_header_data_m);
7640 		MLX5_SET(fte_match_set_misc3, misc3_v, icmp_header_data,
7641 			 icmp_header_data_v & icmp_header_data_m);
7642 	}
7643 }
7644 
7645 /**
7646  * Add GTP item to matcher and to the value.
7647  *
7648  * @param[in, out] matcher
7649  *   Flow matcher.
7650  * @param[in, out] key
7651  *   Flow matcher value.
7652  * @param[in] item
7653  *   Flow pattern to translate.
7654  * @param[in] inner
7655  *   Item is inner pattern.
7656  */
7657 static void
7658 flow_dv_translate_item_gtp(void *matcher, void *key,
7659 			   const struct rte_flow_item *item, int inner)
7660 {
7661 	const struct rte_flow_item_gtp *gtp_m = item->mask;
7662 	const struct rte_flow_item_gtp *gtp_v = item->spec;
7663 	void *headers_m;
7664 	void *headers_v;
7665 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7666 				     misc_parameters_3);
7667 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7668 	uint16_t dport = RTE_GTPU_UDP_PORT;
7669 
7670 	if (inner) {
7671 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7672 					 inner_headers);
7673 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7674 	} else {
7675 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7676 					 outer_headers);
7677 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7678 	}
7679 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7680 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7681 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7682 	}
7683 	if (!gtp_v)
7684 		return;
7685 	if (!gtp_m)
7686 		gtp_m = &rte_flow_item_gtp_mask;
7687 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7688 		 gtp_m->v_pt_rsv_flags);
7689 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7690 		 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7691 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7692 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7693 		 gtp_v->msg_type & gtp_m->msg_type);
7694 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7695 		 rte_be_to_cpu_32(gtp_m->teid));
7696 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7697 		 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7698 }
7699 
7700 /**
7701  * Add eCPRI item to matcher and to the value.
7702  *
7703  * @param[in] dev
7704  *   The devich to configure through.
7705  * @param[in, out] matcher
7706  *   Flow matcher.
7707  * @param[in, out] key
7708  *   Flow matcher value.
7709  * @param[in] item
7710  *   Flow pattern to translate.
7711  * @param[in] samples
7712  *   Sample IDs to be used in the matching.
7713  */
7714 static void
7715 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7716 			     void *key, const struct rte_flow_item *item)
7717 {
7718 	struct mlx5_priv *priv = dev->data->dev_private;
7719 	const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7720 	const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7721 	void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7722 				     misc_parameters_4);
7723 	void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7724 	uint32_t *samples;
7725 	void *dw_m;
7726 	void *dw_v;
7727 
7728 	if (!ecpri_v)
7729 		return;
7730 	if (!ecpri_m)
7731 		ecpri_m = &rte_flow_item_ecpri_mask;
7732 	/*
7733 	 * Maximal four DW samples are supported in a single matching now.
7734 	 * Two are used now for a eCPRI matching:
7735 	 * 1. Type: one byte, mask should be 0x00ff0000 in network order
7736 	 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7737 	 *    if any.
7738 	 */
7739 	if (!ecpri_m->hdr.common.u32)
7740 		return;
7741 	samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7742 	/* Need to take the whole DW as the mask to fill the entry. */
7743 	dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7744 			    prog_sample_field_value_0);
7745 	dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7746 			    prog_sample_field_value_0);
7747 	/* Already big endian (network order) in the header. */
7748 	*(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7749 	*(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7750 	/* Sample#0, used for matching type, offset 0. */
7751 	MLX5_SET(fte_match_set_misc4, misc4_m,
7752 		 prog_sample_field_id_0, samples[0]);
7753 	/* It makes no sense to set the sample ID in the mask field. */
7754 	MLX5_SET(fte_match_set_misc4, misc4_v,
7755 		 prog_sample_field_id_0, samples[0]);
7756 	/*
7757 	 * Checking if message body part needs to be matched.
7758 	 * Some wildcard rules only matching type field should be supported.
7759 	 */
7760 	if (ecpri_m->hdr.dummy[0]) {
7761 		switch (ecpri_v->hdr.common.type) {
7762 		case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7763 		case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7764 		case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7765 			dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7766 					    prog_sample_field_value_1);
7767 			dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7768 					    prog_sample_field_value_1);
7769 			*(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7770 			*(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7771 			/* Sample#1, to match message body, offset 4. */
7772 			MLX5_SET(fte_match_set_misc4, misc4_m,
7773 				 prog_sample_field_id_1, samples[1]);
7774 			MLX5_SET(fte_match_set_misc4, misc4_v,
7775 				 prog_sample_field_id_1, samples[1]);
7776 			break;
7777 		default:
7778 			/* Others, do not match any sample ID. */
7779 			break;
7780 		}
7781 	}
7782 }
7783 
7784 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7785 
7786 #define HEADER_IS_ZERO(match_criteria, headers)				     \
7787 	!(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers),     \
7788 		 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7789 
7790 /**
7791  * Calculate flow matcher enable bitmap.
7792  *
7793  * @param match_criteria
7794  *   Pointer to flow matcher criteria.
7795  *
7796  * @return
7797  *   Bitmap of enabled fields.
7798  */
7799 static uint8_t
7800 flow_dv_matcher_enable(uint32_t *match_criteria)
7801 {
7802 	uint8_t match_criteria_enable;
7803 
7804 	match_criteria_enable =
7805 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7806 		MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7807 	match_criteria_enable |=
7808 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7809 		MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7810 	match_criteria_enable |=
7811 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7812 		MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7813 	match_criteria_enable |=
7814 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7815 		MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7816 	match_criteria_enable |=
7817 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7818 		MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7819 	match_criteria_enable |=
7820 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7821 		MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7822 	return match_criteria_enable;
7823 }
7824 
7825 
7826 /**
7827  * Get a flow table.
7828  *
7829  * @param[in, out] dev
7830  *   Pointer to rte_eth_dev structure.
7831  * @param[in] table_id
7832  *   Table id to use.
7833  * @param[in] egress
7834  *   Direction of the table.
7835  * @param[in] transfer
7836  *   E-Switch or NIC flow.
7837  * @param[out] error
7838  *   pointer to error structure.
7839  *
7840  * @return
7841  *   Returns tables resource based on the index, NULL in case of failed.
7842  */
7843 static struct mlx5_flow_tbl_resource *
7844 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7845 			 uint32_t table_id, uint8_t egress,
7846 			 uint8_t transfer,
7847 			 struct rte_flow_error *error)
7848 {
7849 	struct mlx5_priv *priv = dev->data->dev_private;
7850 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7851 	struct mlx5_flow_tbl_resource *tbl;
7852 	union mlx5_flow_tbl_key table_key = {
7853 		{
7854 			.table_id = table_id,
7855 			.reserved = 0,
7856 			.domain = !!transfer,
7857 			.direction = !!egress,
7858 		}
7859 	};
7860 	struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
7861 							 table_key.v64);
7862 	struct mlx5_flow_tbl_data_entry *tbl_data;
7863 	uint32_t idx = 0;
7864 	int ret;
7865 	void *domain;
7866 
7867 	if (pos) {
7868 		tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
7869 					entry);
7870 		tbl = &tbl_data->tbl;
7871 		rte_atomic32_inc(&tbl->refcnt);
7872 		return tbl;
7873 	}
7874 	tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7875 	if (!tbl_data) {
7876 		rte_flow_error_set(error, ENOMEM,
7877 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7878 				   NULL,
7879 				   "cannot allocate flow table data entry");
7880 		return NULL;
7881 	}
7882 	tbl_data->idx = idx;
7883 	tbl = &tbl_data->tbl;
7884 	pos = &tbl_data->entry;
7885 	if (transfer)
7886 		domain = sh->fdb_domain;
7887 	else if (egress)
7888 		domain = sh->tx_domain;
7889 	else
7890 		domain = sh->rx_domain;
7891 	ret = mlx5_flow_os_create_flow_tbl(domain, table_id, &tbl->obj);
7892 	if (ret) {
7893 		rte_flow_error_set(error, ENOMEM,
7894 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7895 				   NULL, "cannot create flow table object");
7896 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7897 		return NULL;
7898 	}
7899 	/*
7900 	 * No multi-threads now, but still better to initialize the reference
7901 	 * count before insert it into the hash list.
7902 	 */
7903 	rte_atomic32_init(&tbl->refcnt);
7904 	/* Jump action reference count is initialized here. */
7905 	rte_atomic32_init(&tbl_data->jump.refcnt);
7906 	pos->key = table_key.v64;
7907 	ret = mlx5_hlist_insert(sh->flow_tbls, pos);
7908 	if (ret < 0) {
7909 		rte_flow_error_set(error, -ret,
7910 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7911 				   "cannot insert flow table data entry");
7912 		mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7913 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7914 	}
7915 	rte_atomic32_inc(&tbl->refcnt);
7916 	return tbl;
7917 }
7918 
7919 /**
7920  * Release a flow table.
7921  *
7922  * @param[in] dev
7923  *   Pointer to rte_eth_dev structure.
7924  * @param[in] tbl
7925  *   Table resource to be released.
7926  *
7927  * @return
7928  *   Returns 0 if table was released, else return 1;
7929  */
7930 static int
7931 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7932 			     struct mlx5_flow_tbl_resource *tbl)
7933 {
7934 	struct mlx5_priv *priv = dev->data->dev_private;
7935 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7936 	struct mlx5_flow_tbl_data_entry *tbl_data =
7937 		container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7938 
7939 	if (!tbl)
7940 		return 0;
7941 	if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7942 		struct mlx5_hlist_entry *pos = &tbl_data->entry;
7943 
7944 		mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7945 		tbl->obj = NULL;
7946 		/* remove the entry from the hash list and free memory. */
7947 		mlx5_hlist_remove(sh->flow_tbls, pos);
7948 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7949 				tbl_data->idx);
7950 		return 0;
7951 	}
7952 	return 1;
7953 }
7954 
7955 /**
7956  * Register the flow matcher.
7957  *
7958  * @param[in, out] dev
7959  *   Pointer to rte_eth_dev structure.
7960  * @param[in, out] matcher
7961  *   Pointer to flow matcher.
7962  * @param[in, out] key
7963  *   Pointer to flow table key.
7964  * @parm[in, out] dev_flow
7965  *   Pointer to the dev_flow.
7966  * @param[out] error
7967  *   pointer to error structure.
7968  *
7969  * @return
7970  *   0 on success otherwise -errno and errno is set.
7971  */
7972 static int
7973 flow_dv_matcher_register(struct rte_eth_dev *dev,
7974 			 struct mlx5_flow_dv_matcher *matcher,
7975 			 union mlx5_flow_tbl_key *key,
7976 			 struct mlx5_flow *dev_flow,
7977 			 struct rte_flow_error *error)
7978 {
7979 	struct mlx5_priv *priv = dev->data->dev_private;
7980 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7981 	struct mlx5_flow_dv_matcher *cache_matcher;
7982 	struct mlx5dv_flow_matcher_attr dv_attr = {
7983 		.type = IBV_FLOW_ATTR_NORMAL,
7984 		.match_mask = (void *)&matcher->mask,
7985 	};
7986 	struct mlx5_flow_tbl_resource *tbl;
7987 	struct mlx5_flow_tbl_data_entry *tbl_data;
7988 	int ret;
7989 
7990 	tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7991 				       key->domain, error);
7992 	if (!tbl)
7993 		return -rte_errno;	/* No need to refill the error info */
7994 	tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7995 	/* Lookup from cache. */
7996 	LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7997 		if (matcher->crc == cache_matcher->crc &&
7998 		    matcher->priority == cache_matcher->priority &&
7999 		    !memcmp((const void *)matcher->mask.buf,
8000 			    (const void *)cache_matcher->mask.buf,
8001 			    cache_matcher->mask.size)) {
8002 			DRV_LOG(DEBUG,
8003 				"%s group %u priority %hd use %s "
8004 				"matcher %p: refcnt %d++",
8005 				key->domain ? "FDB" : "NIC", key->table_id,
8006 				cache_matcher->priority,
8007 				key->direction ? "tx" : "rx",
8008 				(void *)cache_matcher,
8009 				rte_atomic32_read(&cache_matcher->refcnt));
8010 			rte_atomic32_inc(&cache_matcher->refcnt);
8011 			dev_flow->handle->dvh.matcher = cache_matcher;
8012 			/* old matcher should not make the table ref++. */
8013 			flow_dv_tbl_resource_release(dev, tbl);
8014 			return 0;
8015 		}
8016 	}
8017 	/* Register new matcher. */
8018 	cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
8019 				    SOCKET_ID_ANY);
8020 	if (!cache_matcher) {
8021 		flow_dv_tbl_resource_release(dev, tbl);
8022 		return rte_flow_error_set(error, ENOMEM,
8023 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8024 					  "cannot allocate matcher memory");
8025 	}
8026 	*cache_matcher = *matcher;
8027 	dv_attr.match_criteria_enable =
8028 		flow_dv_matcher_enable(cache_matcher->mask.buf);
8029 	dv_attr.priority = matcher->priority;
8030 	if (key->direction)
8031 		dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
8032 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
8033 					       &cache_matcher->matcher_object);
8034 	if (ret) {
8035 		mlx5_free(cache_matcher);
8036 #ifdef HAVE_MLX5DV_DR
8037 		flow_dv_tbl_resource_release(dev, tbl);
8038 #endif
8039 		return rte_flow_error_set(error, ENOMEM,
8040 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8041 					  NULL, "cannot create matcher");
8042 	}
8043 	/* Save the table information */
8044 	cache_matcher->tbl = tbl;
8045 	rte_atomic32_init(&cache_matcher->refcnt);
8046 	/* only matcher ref++, table ref++ already done above in get API. */
8047 	rte_atomic32_inc(&cache_matcher->refcnt);
8048 	LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
8049 	dev_flow->handle->dvh.matcher = cache_matcher;
8050 	DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
8051 		key->domain ? "FDB" : "NIC", key->table_id,
8052 		cache_matcher->priority,
8053 		key->direction ? "tx" : "rx", (void *)cache_matcher,
8054 		rte_atomic32_read(&cache_matcher->refcnt));
8055 	return 0;
8056 }
8057 
8058 /**
8059  * Find existing tag resource or create and register a new one.
8060  *
8061  * @param dev[in, out]
8062  *   Pointer to rte_eth_dev structure.
8063  * @param[in, out] tag_be24
8064  *   Tag value in big endian then R-shift 8.
8065  * @parm[in, out] dev_flow
8066  *   Pointer to the dev_flow.
8067  * @param[out] error
8068  *   pointer to error structure.
8069  *
8070  * @return
8071  *   0 on success otherwise -errno and errno is set.
8072  */
8073 static int
8074 flow_dv_tag_resource_register
8075 			(struct rte_eth_dev *dev,
8076 			 uint32_t tag_be24,
8077 			 struct mlx5_flow *dev_flow,
8078 			 struct rte_flow_error *error)
8079 {
8080 	struct mlx5_priv *priv = dev->data->dev_private;
8081 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8082 	struct mlx5_flow_dv_tag_resource *cache_resource;
8083 	struct mlx5_hlist_entry *entry;
8084 	int ret;
8085 
8086 	/* Lookup a matching resource from cache. */
8087 	entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
8088 	if (entry) {
8089 		cache_resource = container_of
8090 			(entry, struct mlx5_flow_dv_tag_resource, entry);
8091 		rte_atomic32_inc(&cache_resource->refcnt);
8092 		dev_flow->handle->dvh.rix_tag = cache_resource->idx;
8093 		dev_flow->dv.tag_resource = cache_resource;
8094 		DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
8095 			(void *)cache_resource,
8096 			rte_atomic32_read(&cache_resource->refcnt));
8097 		return 0;
8098 	}
8099 	/* Register new resource. */
8100 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
8101 				       &dev_flow->handle->dvh.rix_tag);
8102 	if (!cache_resource)
8103 		return rte_flow_error_set(error, ENOMEM,
8104 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8105 					  "cannot allocate resource memory");
8106 	cache_resource->entry.key = (uint64_t)tag_be24;
8107 	ret = mlx5_flow_os_create_flow_action_tag(tag_be24,
8108 						  &cache_resource->action);
8109 	if (ret) {
8110 		mlx5_free(cache_resource);
8111 		return rte_flow_error_set(error, ENOMEM,
8112 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8113 					  NULL, "cannot create action");
8114 	}
8115 	rte_atomic32_init(&cache_resource->refcnt);
8116 	rte_atomic32_inc(&cache_resource->refcnt);
8117 	if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
8118 		mlx5_flow_os_destroy_flow_action(cache_resource->action);
8119 		mlx5_free(cache_resource);
8120 		return rte_flow_error_set(error, EEXIST,
8121 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8122 					  NULL, "cannot insert tag");
8123 	}
8124 	dev_flow->dv.tag_resource = cache_resource;
8125 	DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
8126 		(void *)cache_resource,
8127 		rte_atomic32_read(&cache_resource->refcnt));
8128 	return 0;
8129 }
8130 
8131 /**
8132  * Release the tag.
8133  *
8134  * @param dev
8135  *   Pointer to Ethernet device.
8136  * @param tag_idx
8137  *   Tag index.
8138  *
8139  * @return
8140  *   1 while a reference on it exists, 0 when freed.
8141  */
8142 static int
8143 flow_dv_tag_release(struct rte_eth_dev *dev,
8144 		    uint32_t tag_idx)
8145 {
8146 	struct mlx5_priv *priv = dev->data->dev_private;
8147 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8148 	struct mlx5_flow_dv_tag_resource *tag;
8149 
8150 	tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8151 	if (!tag)
8152 		return 0;
8153 	DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
8154 		dev->data->port_id, (void *)tag,
8155 		rte_atomic32_read(&tag->refcnt));
8156 	if (rte_atomic32_dec_and_test(&tag->refcnt)) {
8157 		claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
8158 		mlx5_hlist_remove(sh->tag_table, &tag->entry);
8159 		DRV_LOG(DEBUG, "port %u tag %p: removed",
8160 			dev->data->port_id, (void *)tag);
8161 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
8162 		return 0;
8163 	}
8164 	return 1;
8165 }
8166 
8167 /**
8168  * Translate port ID action to vport.
8169  *
8170  * @param[in] dev
8171  *   Pointer to rte_eth_dev structure.
8172  * @param[in] action
8173  *   Pointer to the port ID action.
8174  * @param[out] dst_port_id
8175  *   The target port ID.
8176  * @param[out] error
8177  *   Pointer to the error structure.
8178  *
8179  * @return
8180  *   0 on success, a negative errno value otherwise and rte_errno is set.
8181  */
8182 static int
8183 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
8184 				 const struct rte_flow_action *action,
8185 				 uint32_t *dst_port_id,
8186 				 struct rte_flow_error *error)
8187 {
8188 	uint32_t port;
8189 	struct mlx5_priv *priv;
8190 	const struct rte_flow_action_port_id *conf =
8191 			(const struct rte_flow_action_port_id *)action->conf;
8192 
8193 	port = conf->original ? dev->data->port_id : conf->id;
8194 	priv = mlx5_port_to_eswitch_info(port, false);
8195 	if (!priv)
8196 		return rte_flow_error_set(error, -rte_errno,
8197 					  RTE_FLOW_ERROR_TYPE_ACTION,
8198 					  NULL,
8199 					  "No eswitch info was found for port");
8200 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
8201 	/*
8202 	 * This parameter is transferred to
8203 	 * mlx5dv_dr_action_create_dest_ib_port().
8204 	 */
8205 	*dst_port_id = priv->dev_port;
8206 #else
8207 	/*
8208 	 * Legacy mode, no LAG configurations is supported.
8209 	 * This parameter is transferred to
8210 	 * mlx5dv_dr_action_create_dest_vport().
8211 	 */
8212 	*dst_port_id = priv->vport_id;
8213 #endif
8214 	return 0;
8215 }
8216 
8217 /**
8218  * Create a counter with aging configuration.
8219  *
8220  * @param[in] dev
8221  *   Pointer to rte_eth_dev structure.
8222  * @param[out] count
8223  *   Pointer to the counter action configuration.
8224  * @param[in] age
8225  *   Pointer to the aging action configuration.
8226  *
8227  * @return
8228  *   Index to flow counter on success, 0 otherwise.
8229  */
8230 static uint32_t
8231 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
8232 				struct mlx5_flow *dev_flow,
8233 				const struct rte_flow_action_count *count,
8234 				const struct rte_flow_action_age *age)
8235 {
8236 	uint32_t counter;
8237 	struct mlx5_age_param *age_param;
8238 
8239 	counter = flow_dv_counter_alloc(dev,
8240 				count ? count->shared : 0,
8241 				count ? count->id : 0,
8242 				dev_flow->dv.group, !!age);
8243 	if (!counter || age == NULL)
8244 		return counter;
8245 	age_param  = flow_dv_counter_idx_get_age(dev, counter);
8246 	/*
8247 	 * The counter age accuracy may have a bit delay. Have 3/4
8248 	 * second bias on the timeount in order to let it age in time.
8249 	 */
8250 	age_param->context = age->context ? age->context :
8251 		(void *)(uintptr_t)(dev_flow->flow_idx);
8252 	/*
8253 	 * The counter age accuracy may have a bit delay. Have 3/4
8254 	 * second bias on the timeount in order to let it age in time.
8255 	 */
8256 	age_param->timeout = age->timeout * 10 - MLX5_AGING_TIME_DELAY;
8257 	/* Set expire time in unit of 0.1 sec. */
8258 	age_param->port_id = dev->data->port_id;
8259 	age_param->expire = age_param->timeout +
8260 			rte_rdtsc() / (rte_get_tsc_hz() / 10);
8261 	rte_atomic16_set(&age_param->state, AGE_CANDIDATE);
8262 	return counter;
8263 }
8264 /**
8265  * Add Tx queue matcher
8266  *
8267  * @param[in] dev
8268  *   Pointer to the dev struct.
8269  * @param[in, out] matcher
8270  *   Flow matcher.
8271  * @param[in, out] key
8272  *   Flow matcher value.
8273  * @param[in] item
8274  *   Flow pattern to translate.
8275  * @param[in] inner
8276  *   Item is inner pattern.
8277  */
8278 static void
8279 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
8280 				void *matcher, void *key,
8281 				const struct rte_flow_item *item)
8282 {
8283 	const struct mlx5_rte_flow_item_tx_queue *queue_m;
8284 	const struct mlx5_rte_flow_item_tx_queue *queue_v;
8285 	void *misc_m =
8286 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
8287 	void *misc_v =
8288 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
8289 	struct mlx5_txq_ctrl *txq;
8290 	uint32_t queue;
8291 
8292 
8293 	queue_m = (const void *)item->mask;
8294 	if (!queue_m)
8295 		return;
8296 	queue_v = (const void *)item->spec;
8297 	if (!queue_v)
8298 		return;
8299 	txq = mlx5_txq_get(dev, queue_v->queue);
8300 	if (!txq)
8301 		return;
8302 	queue = txq->obj->sq->id;
8303 	MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
8304 	MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
8305 		 queue & queue_m->queue);
8306 	mlx5_txq_release(dev, queue_v->queue);
8307 }
8308 
8309 /**
8310  * Set the hash fields according to the @p flow information.
8311  *
8312  * @param[in] dev_flow
8313  *   Pointer to the mlx5_flow.
8314  * @param[in] rss_desc
8315  *   Pointer to the mlx5_flow_rss_desc.
8316  */
8317 static void
8318 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
8319 		       struct mlx5_flow_rss_desc *rss_desc)
8320 {
8321 	uint64_t items = dev_flow->handle->layers;
8322 	int rss_inner = 0;
8323 	uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
8324 
8325 	dev_flow->hash_fields = 0;
8326 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
8327 	if (rss_desc->level >= 2) {
8328 		dev_flow->hash_fields |= IBV_RX_HASH_INNER;
8329 		rss_inner = 1;
8330 	}
8331 #endif
8332 	if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8333 	    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8334 		if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8335 			if (rss_types & ETH_RSS_L3_SRC_ONLY)
8336 				dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8337 			else if (rss_types & ETH_RSS_L3_DST_ONLY)
8338 				dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8339 			else
8340 				dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8341 		}
8342 	} else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8343 		   (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8344 		if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8345 			if (rss_types & ETH_RSS_L3_SRC_ONLY)
8346 				dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8347 			else if (rss_types & ETH_RSS_L3_DST_ONLY)
8348 				dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8349 			else
8350 				dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8351 		}
8352 	}
8353 	if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8354 	    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8355 		if (rss_types & ETH_RSS_UDP) {
8356 			if (rss_types & ETH_RSS_L4_SRC_ONLY)
8357 				dev_flow->hash_fields |=
8358 						IBV_RX_HASH_SRC_PORT_UDP;
8359 			else if (rss_types & ETH_RSS_L4_DST_ONLY)
8360 				dev_flow->hash_fields |=
8361 						IBV_RX_HASH_DST_PORT_UDP;
8362 			else
8363 				dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8364 		}
8365 	} else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8366 		   (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8367 		if (rss_types & ETH_RSS_TCP) {
8368 			if (rss_types & ETH_RSS_L4_SRC_ONLY)
8369 				dev_flow->hash_fields |=
8370 						IBV_RX_HASH_SRC_PORT_TCP;
8371 			else if (rss_types & ETH_RSS_L4_DST_ONLY)
8372 				dev_flow->hash_fields |=
8373 						IBV_RX_HASH_DST_PORT_TCP;
8374 			else
8375 				dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8376 		}
8377 	}
8378 }
8379 
8380 /**
8381  * Create an Rx Hash queue.
8382  *
8383  * @param dev
8384  *   Pointer to Ethernet device.
8385  * @param[in] dev_flow
8386  *   Pointer to the mlx5_flow.
8387  * @param[in] rss_desc
8388  *   Pointer to the mlx5_flow_rss_desc.
8389  * @param[out] hrxq_idx
8390  *   Hash Rx queue index.
8391  *
8392  * @return
8393  *   The Verbs/DevX object initialised, NULL otherwise and rte_errno is set.
8394  */
8395 static struct mlx5_hrxq *
8396 flow_dv_handle_rx_queue(struct rte_eth_dev *dev,
8397 			struct mlx5_flow *dev_flow,
8398 			struct mlx5_flow_rss_desc *rss_desc,
8399 			uint32_t *hrxq_idx)
8400 {
8401 	struct mlx5_priv *priv = dev->data->dev_private;
8402 	struct mlx5_flow_handle *dh = dev_flow->handle;
8403 	struct mlx5_hrxq *hrxq;
8404 
8405 	MLX5_ASSERT(rss_desc->queue_num);
8406 	*hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8407 				  MLX5_RSS_HASH_KEY_LEN,
8408 				  dev_flow->hash_fields,
8409 				  rss_desc->queue,
8410 				  rss_desc->queue_num);
8411 	if (!*hrxq_idx) {
8412 		*hrxq_idx = mlx5_hrxq_new
8413 				(dev, rss_desc->key,
8414 				 MLX5_RSS_HASH_KEY_LEN,
8415 				 dev_flow->hash_fields,
8416 				 rss_desc->queue,
8417 				 rss_desc->queue_num,
8418 				 !!(dh->layers &
8419 				 MLX5_FLOW_LAYER_TUNNEL));
8420 		if (!*hrxq_idx)
8421 			return NULL;
8422 	}
8423 	hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8424 			      *hrxq_idx);
8425 	return hrxq;
8426 }
8427 
8428 /**
8429  * Find existing sample resource or create and register a new one.
8430  *
8431  * @param[in, out] dev
8432  *   Pointer to rte_eth_dev structure.
8433  * @param[in] attr
8434  *   Attributes of flow that includes this item.
8435  * @param[in] resource
8436  *   Pointer to sample resource.
8437  * @parm[in, out] dev_flow
8438  *   Pointer to the dev_flow.
8439  * @param[in, out] sample_dv_actions
8440  *   Pointer to sample actions list.
8441  * @param[out] error
8442  *   pointer to error structure.
8443  *
8444  * @return
8445  *   0 on success otherwise -errno and errno is set.
8446  */
8447 static int
8448 flow_dv_sample_resource_register(struct rte_eth_dev *dev,
8449 			 const struct rte_flow_attr *attr,
8450 			 struct mlx5_flow_dv_sample_resource *resource,
8451 			 struct mlx5_flow *dev_flow,
8452 			 void **sample_dv_actions,
8453 			 struct rte_flow_error *error)
8454 {
8455 	struct mlx5_flow_dv_sample_resource *cache_resource;
8456 	struct mlx5dv_dr_flow_sampler_attr sampler_attr;
8457 	struct mlx5_priv *priv = dev->data->dev_private;
8458 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8459 	struct mlx5_flow_tbl_resource *tbl;
8460 	uint32_t idx = 0;
8461 	const uint32_t next_ft_step = 1;
8462 	uint32_t next_ft_id = resource->ft_id +	next_ft_step;
8463 
8464 	/* Lookup a matching resource from cache. */
8465 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_SAMPLE], sh->sample_action_list,
8466 		      idx, cache_resource, next) {
8467 		if (resource->ratio == cache_resource->ratio &&
8468 		    resource->ft_type == cache_resource->ft_type &&
8469 		    resource->ft_id == cache_resource->ft_id &&
8470 		    resource->set_action == cache_resource->set_action &&
8471 		    !memcmp((void *)&resource->sample_act,
8472 			    (void *)&cache_resource->sample_act,
8473 			    sizeof(struct mlx5_flow_sub_actions_list))) {
8474 			DRV_LOG(DEBUG, "sample resource %p: refcnt %d++",
8475 				(void *)cache_resource,
8476 				__atomic_load_n(&cache_resource->refcnt,
8477 						__ATOMIC_RELAXED));
8478 			__atomic_fetch_add(&cache_resource->refcnt, 1,
8479 					   __ATOMIC_RELAXED);
8480 			dev_flow->handle->dvh.rix_sample = idx;
8481 			dev_flow->dv.sample_res = cache_resource;
8482 			return 0;
8483 		}
8484 	}
8485 	/* Register new sample resource. */
8486 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_SAMPLE],
8487 				       &dev_flow->handle->dvh.rix_sample);
8488 	if (!cache_resource)
8489 		return rte_flow_error_set(error, ENOMEM,
8490 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8491 					  NULL,
8492 					  "cannot allocate resource memory");
8493 	*cache_resource = *resource;
8494 	/* Create normal path table level */
8495 	tbl = flow_dv_tbl_resource_get(dev, next_ft_id,
8496 					attr->egress, attr->transfer, error);
8497 	if (!tbl) {
8498 		rte_flow_error_set(error, ENOMEM,
8499 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8500 					  NULL,
8501 					  "fail to create normal path table "
8502 					  "for sample");
8503 		goto error;
8504 	}
8505 	cache_resource->normal_path_tbl = tbl;
8506 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8507 		cache_resource->default_miss =
8508 				mlx5_glue->dr_create_flow_action_default_miss();
8509 		if (!cache_resource->default_miss) {
8510 			rte_flow_error_set(error, ENOMEM,
8511 						RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8512 						NULL,
8513 						"cannot create default miss "
8514 						"action");
8515 			goto error;
8516 		}
8517 		sample_dv_actions[resource->sample_act.actions_num++] =
8518 						cache_resource->default_miss;
8519 	}
8520 	/* Create a DR sample action */
8521 	sampler_attr.sample_ratio = cache_resource->ratio;
8522 	sampler_attr.default_next_table = tbl->obj;
8523 	sampler_attr.num_sample_actions = resource->sample_act.actions_num;
8524 	sampler_attr.sample_actions = (struct mlx5dv_dr_action **)
8525 							&sample_dv_actions[0];
8526 	sampler_attr.action = cache_resource->set_action;
8527 	cache_resource->verbs_action =
8528 		mlx5_glue->dr_create_flow_action_sampler(&sampler_attr);
8529 	if (!cache_resource->verbs_action) {
8530 		rte_flow_error_set(error, ENOMEM,
8531 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8532 					NULL, "cannot create sample action");
8533 		goto error;
8534 	}
8535 	__atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8536 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_SAMPLE], &sh->sample_action_list,
8537 		     dev_flow->handle->dvh.rix_sample, cache_resource,
8538 		     next);
8539 	dev_flow->dv.sample_res = cache_resource;
8540 	DRV_LOG(DEBUG, "new sample resource %p: refcnt %d++",
8541 		(void *)cache_resource,
8542 		__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8543 	return 0;
8544 error:
8545 	if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
8546 		if (cache_resource->default_miss)
8547 			claim_zero(mlx5_glue->destroy_flow_action
8548 				(cache_resource->default_miss));
8549 	} else {
8550 		if (cache_resource->sample_idx.rix_hrxq &&
8551 		    !mlx5_hrxq_release(dev,
8552 				cache_resource->sample_idx.rix_hrxq))
8553 			cache_resource->sample_idx.rix_hrxq = 0;
8554 		if (cache_resource->sample_idx.rix_tag &&
8555 		    !flow_dv_tag_release(dev,
8556 				cache_resource->sample_idx.rix_tag))
8557 			cache_resource->sample_idx.rix_tag = 0;
8558 		if (cache_resource->sample_idx.cnt) {
8559 			flow_dv_counter_release(dev,
8560 				cache_resource->sample_idx.cnt);
8561 			cache_resource->sample_idx.cnt = 0;
8562 		}
8563 	}
8564 	if (cache_resource->normal_path_tbl)
8565 		flow_dv_tbl_resource_release(dev,
8566 				cache_resource->normal_path_tbl);
8567 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_SAMPLE],
8568 				dev_flow->handle->dvh.rix_sample);
8569 	dev_flow->handle->dvh.rix_sample = 0;
8570 	return -rte_errno;
8571 }
8572 
8573 /**
8574  * Find existing destination array resource or create and register a new one.
8575  *
8576  * @param[in, out] dev
8577  *   Pointer to rte_eth_dev structure.
8578  * @param[in] attr
8579  *   Attributes of flow that includes this item.
8580  * @param[in] resource
8581  *   Pointer to destination array resource.
8582  * @parm[in, out] dev_flow
8583  *   Pointer to the dev_flow.
8584  * @param[out] error
8585  *   pointer to error structure.
8586  *
8587  * @return
8588  *   0 on success otherwise -errno and errno is set.
8589  */
8590 static int
8591 flow_dv_dest_array_resource_register(struct rte_eth_dev *dev,
8592 			 const struct rte_flow_attr *attr,
8593 			 struct mlx5_flow_dv_dest_array_resource *resource,
8594 			 struct mlx5_flow *dev_flow,
8595 			 struct rte_flow_error *error)
8596 {
8597 	struct mlx5_flow_dv_dest_array_resource *cache_resource;
8598 	struct mlx5dv_dr_action_dest_attr *dest_attr[MLX5_MAX_DEST_NUM] = { 0 };
8599 	struct mlx5dv_dr_action_dest_reformat dest_reformat[MLX5_MAX_DEST_NUM];
8600 	struct mlx5_priv *priv = dev->data->dev_private;
8601 	struct mlx5_dev_ctx_shared *sh = priv->sh;
8602 	struct mlx5_flow_sub_actions_list *sample_act;
8603 	struct mlx5dv_dr_domain *domain;
8604 	uint32_t idx = 0;
8605 
8606 	/* Lookup a matching resource from cache. */
8607 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8608 		      sh->dest_array_list,
8609 		      idx, cache_resource, next) {
8610 		if (resource->num_of_dest == cache_resource->num_of_dest &&
8611 		    resource->ft_type == cache_resource->ft_type &&
8612 		    !memcmp((void *)cache_resource->sample_act,
8613 			    (void *)resource->sample_act,
8614 			   (resource->num_of_dest *
8615 			   sizeof(struct mlx5_flow_sub_actions_list)))) {
8616 			DRV_LOG(DEBUG, "dest array resource %p: refcnt %d++",
8617 				(void *)cache_resource,
8618 				__atomic_load_n(&cache_resource->refcnt,
8619 						__ATOMIC_RELAXED));
8620 			__atomic_fetch_add(&cache_resource->refcnt, 1,
8621 					   __ATOMIC_RELAXED);
8622 			dev_flow->handle->dvh.rix_dest_array = idx;
8623 			dev_flow->dv.dest_array_res = cache_resource;
8624 			return 0;
8625 		}
8626 	}
8627 	/* Register new destination array resource. */
8628 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8629 				       &dev_flow->handle->dvh.rix_dest_array);
8630 	if (!cache_resource)
8631 		return rte_flow_error_set(error, ENOMEM,
8632 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8633 					  NULL,
8634 					  "cannot allocate resource memory");
8635 	*cache_resource = *resource;
8636 	if (attr->transfer)
8637 		domain = sh->fdb_domain;
8638 	else if (attr->ingress)
8639 		domain = sh->rx_domain;
8640 	else
8641 		domain = sh->tx_domain;
8642 	for (idx = 0; idx < resource->num_of_dest; idx++) {
8643 		dest_attr[idx] = (struct mlx5dv_dr_action_dest_attr *)
8644 				 mlx5_malloc(MLX5_MEM_ZERO,
8645 				 sizeof(struct mlx5dv_dr_action_dest_attr),
8646 				 0, SOCKET_ID_ANY);
8647 		if (!dest_attr[idx]) {
8648 			rte_flow_error_set(error, ENOMEM,
8649 					   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8650 					   NULL,
8651 					   "cannot allocate resource memory");
8652 			goto error;
8653 		}
8654 		dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST;
8655 		sample_act = &resource->sample_act[idx];
8656 		if (sample_act->action_flags == MLX5_FLOW_ACTION_QUEUE) {
8657 			dest_attr[idx]->dest = sample_act->dr_queue_action;
8658 		} else if (sample_act->action_flags ==
8659 			  (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_ENCAP)) {
8660 			dest_attr[idx]->type = MLX5DV_DR_ACTION_DEST_REFORMAT;
8661 			dest_attr[idx]->dest_reformat = &dest_reformat[idx];
8662 			dest_attr[idx]->dest_reformat->reformat =
8663 					sample_act->dr_encap_action;
8664 			dest_attr[idx]->dest_reformat->dest =
8665 					sample_act->dr_port_id_action;
8666 		} else if (sample_act->action_flags ==
8667 			   MLX5_FLOW_ACTION_PORT_ID) {
8668 			dest_attr[idx]->dest = sample_act->dr_port_id_action;
8669 		}
8670 	}
8671 	/* create a dest array actioin */
8672 	cache_resource->action = mlx5_glue->dr_create_flow_action_dest_array
8673 						(domain,
8674 						 cache_resource->num_of_dest,
8675 						 dest_attr);
8676 	if (!cache_resource->action) {
8677 		rte_flow_error_set(error, ENOMEM,
8678 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8679 				   NULL,
8680 				   "cannot create destination array action");
8681 		goto error;
8682 	}
8683 	__atomic_store_n(&cache_resource->refcnt, 1, __ATOMIC_RELAXED);
8684 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8685 		     &sh->dest_array_list,
8686 		     dev_flow->handle->dvh.rix_dest_array, cache_resource,
8687 		     next);
8688 	dev_flow->dv.dest_array_res = cache_resource;
8689 	DRV_LOG(DEBUG, "new destination array resource %p: refcnt %d++",
8690 		(void *)cache_resource,
8691 		__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
8692 	for (idx = 0; idx < resource->num_of_dest; idx++)
8693 		mlx5_free(dest_attr[idx]);
8694 	return 0;
8695 error:
8696 	for (idx = 0; idx < resource->num_of_dest; idx++) {
8697 		struct mlx5_flow_sub_actions_idx *act_res =
8698 					&cache_resource->sample_idx[idx];
8699 		if (act_res->rix_hrxq &&
8700 		    !mlx5_hrxq_release(dev,
8701 				act_res->rix_hrxq))
8702 			act_res->rix_hrxq = 0;
8703 		if (act_res->rix_encap_decap &&
8704 			!flow_dv_encap_decap_resource_release(dev,
8705 				act_res->rix_encap_decap))
8706 			act_res->rix_encap_decap = 0;
8707 		if (act_res->rix_port_id_action &&
8708 			!flow_dv_port_id_action_resource_release(dev,
8709 				act_res->rix_port_id_action))
8710 			act_res->rix_port_id_action = 0;
8711 		if (dest_attr[idx])
8712 			mlx5_free(dest_attr[idx]);
8713 	}
8714 
8715 	mlx5_ipool_free(sh->ipool[MLX5_IPOOL_DEST_ARRAY],
8716 				dev_flow->handle->dvh.rix_dest_array);
8717 	dev_flow->handle->dvh.rix_dest_array = 0;
8718 	return -rte_errno;
8719 }
8720 
8721 /**
8722  * Convert Sample action to DV specification.
8723  *
8724  * @param[in] dev
8725  *   Pointer to rte_eth_dev structure.
8726  * @param[in] action
8727  *   Pointer to action structure.
8728  * @param[in, out] dev_flow
8729  *   Pointer to the mlx5_flow.
8730  * @param[in] attr
8731  *   Pointer to the flow attributes.
8732  * @param[in, out] num_of_dest
8733  *   Pointer to the num of destination.
8734  * @param[in, out] sample_actions
8735  *   Pointer to sample actions list.
8736  * @param[in, out] res
8737  *   Pointer to sample resource.
8738  * @param[out] error
8739  *   Pointer to the error structure.
8740  *
8741  * @return
8742  *   0 on success, a negative errno value otherwise and rte_errno is set.
8743  */
8744 static int
8745 flow_dv_translate_action_sample(struct rte_eth_dev *dev,
8746 				const struct rte_flow_action *action,
8747 				struct mlx5_flow *dev_flow,
8748 				const struct rte_flow_attr *attr,
8749 				uint32_t *num_of_dest,
8750 				void **sample_actions,
8751 				struct mlx5_flow_dv_sample_resource *res,
8752 				struct rte_flow_error *error)
8753 {
8754 	struct mlx5_priv *priv = dev->data->dev_private;
8755 	const struct rte_flow_action_sample *sample_action;
8756 	const struct rte_flow_action *sub_actions;
8757 	const struct rte_flow_action_queue *queue;
8758 	struct mlx5_flow_sub_actions_list *sample_act;
8759 	struct mlx5_flow_sub_actions_idx *sample_idx;
8760 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8761 					      priv->rss_desc)
8762 					      [!!priv->flow_nested_idx];
8763 	uint64_t action_flags = 0;
8764 
8765 	sample_act = &res->sample_act;
8766 	sample_idx = &res->sample_idx;
8767 	sample_action = (const struct rte_flow_action_sample *)action->conf;
8768 	res->ratio = sample_action->ratio;
8769 	sub_actions = sample_action->actions;
8770 	for (; sub_actions->type != RTE_FLOW_ACTION_TYPE_END; sub_actions++) {
8771 		int type = sub_actions->type;
8772 		uint32_t pre_rix = 0;
8773 		void *pre_r;
8774 		switch (type) {
8775 		case RTE_FLOW_ACTION_TYPE_QUEUE:
8776 		{
8777 			struct mlx5_hrxq *hrxq;
8778 			uint32_t hrxq_idx;
8779 
8780 			queue = sub_actions->conf;
8781 			rss_desc->queue_num = 1;
8782 			rss_desc->queue[0] = queue->index;
8783 			hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
8784 					rss_desc, &hrxq_idx);
8785 			if (!hrxq)
8786 				return rte_flow_error_set
8787 					(error, rte_errno,
8788 					 RTE_FLOW_ERROR_TYPE_ACTION,
8789 					 NULL,
8790 					 "cannot create fate queue");
8791 			sample_act->dr_queue_action = hrxq->action;
8792 			sample_idx->rix_hrxq = hrxq_idx;
8793 			sample_actions[sample_act->actions_num++] =
8794 						hrxq->action;
8795 			(*num_of_dest)++;
8796 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
8797 			if (action_flags & MLX5_FLOW_ACTION_MARK)
8798 				dev_flow->handle->rix_hrxq = hrxq_idx;
8799 			dev_flow->handle->fate_action =
8800 					MLX5_FLOW_FATE_QUEUE;
8801 			break;
8802 		}
8803 		case RTE_FLOW_ACTION_TYPE_MARK:
8804 		{
8805 			uint32_t tag_be = mlx5_flow_mark_set
8806 				(((const struct rte_flow_action_mark *)
8807 				(sub_actions->conf))->id);
8808 
8809 			dev_flow->handle->mark = 1;
8810 			pre_rix = dev_flow->handle->dvh.rix_tag;
8811 			/* Save the mark resource before sample */
8812 			pre_r = dev_flow->dv.tag_resource;
8813 			if (flow_dv_tag_resource_register(dev, tag_be,
8814 						  dev_flow, error))
8815 				return -rte_errno;
8816 			MLX5_ASSERT(dev_flow->dv.tag_resource);
8817 			sample_act->dr_tag_action =
8818 				dev_flow->dv.tag_resource->action;
8819 			sample_idx->rix_tag =
8820 				dev_flow->handle->dvh.rix_tag;
8821 			sample_actions[sample_act->actions_num++] =
8822 						sample_act->dr_tag_action;
8823 			/* Recover the mark resource after sample */
8824 			dev_flow->dv.tag_resource = pre_r;
8825 			dev_flow->handle->dvh.rix_tag = pre_rix;
8826 			action_flags |= MLX5_FLOW_ACTION_MARK;
8827 			break;
8828 		}
8829 		case RTE_FLOW_ACTION_TYPE_COUNT:
8830 		{
8831 			uint32_t counter;
8832 
8833 			counter = flow_dv_translate_create_counter(dev,
8834 					dev_flow, sub_actions->conf, 0);
8835 			if (!counter)
8836 				return rte_flow_error_set
8837 						(error, rte_errno,
8838 						 RTE_FLOW_ERROR_TYPE_ACTION,
8839 						 NULL,
8840 						 "cannot create counter"
8841 						 " object.");
8842 			sample_idx->cnt = counter;
8843 			sample_act->dr_cnt_action =
8844 				  (flow_dv_counter_get_by_idx(dev,
8845 				  counter, NULL))->action;
8846 			sample_actions[sample_act->actions_num++] =
8847 						sample_act->dr_cnt_action;
8848 			action_flags |= MLX5_FLOW_ACTION_COUNT;
8849 			break;
8850 		}
8851 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
8852 		{
8853 			struct mlx5_flow_dv_port_id_action_resource
8854 					port_id_resource;
8855 			uint32_t port_id = 0;
8856 
8857 			memset(&port_id_resource, 0, sizeof(port_id_resource));
8858 			/* Save the port id resource before sample */
8859 			pre_rix = dev_flow->handle->rix_port_id_action;
8860 			pre_r = dev_flow->dv.port_id_action;
8861 			if (flow_dv_translate_action_port_id(dev, sub_actions,
8862 							     &port_id, error))
8863 				return -rte_errno;
8864 			port_id_resource.port_id = port_id;
8865 			if (flow_dv_port_id_action_resource_register
8866 			    (dev, &port_id_resource, dev_flow, error))
8867 				return -rte_errno;
8868 			sample_act->dr_port_id_action =
8869 				dev_flow->dv.port_id_action->action;
8870 			sample_idx->rix_port_id_action =
8871 				dev_flow->handle->rix_port_id_action;
8872 			sample_actions[sample_act->actions_num++] =
8873 						sample_act->dr_port_id_action;
8874 			/* Recover the port id resource after sample */
8875 			dev_flow->dv.port_id_action = pre_r;
8876 			dev_flow->handle->rix_port_id_action = pre_rix;
8877 			(*num_of_dest)++;
8878 			action_flags |= MLX5_FLOW_ACTION_PORT_ID;
8879 			break;
8880 		}
8881 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
8882 			/* Save the encap resource before sample */
8883 			pre_rix = dev_flow->handle->dvh.rix_encap_decap;
8884 			pre_r = dev_flow->dv.encap_decap;
8885 			if (flow_dv_create_action_l2_encap(dev, sub_actions,
8886 							   dev_flow,
8887 							   attr->transfer,
8888 							   error))
8889 				return -rte_errno;
8890 			sample_act->dr_encap_action =
8891 				dev_flow->dv.encap_decap->action;
8892 			sample_idx->rix_encap_decap =
8893 				dev_flow->handle->dvh.rix_encap_decap;
8894 			sample_actions[sample_act->actions_num++] =
8895 						sample_act->dr_encap_action;
8896 			/* Recover the encap resource after sample */
8897 			dev_flow->dv.encap_decap = pre_r;
8898 			dev_flow->handle->dvh.rix_encap_decap = pre_rix;
8899 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
8900 			break;
8901 		default:
8902 			return rte_flow_error_set(error, EINVAL,
8903 				RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8904 				NULL,
8905 				"Not support for sampler action");
8906 		}
8907 	}
8908 	sample_act->action_flags = action_flags;
8909 	res->ft_id = dev_flow->dv.group;
8910 	if (attr->transfer) {
8911 		union {
8912 			uint32_t action_in[MLX5_ST_SZ_DW(set_action_in)];
8913 			uint64_t set_action;
8914 		} action_ctx = { .set_action = 0 };
8915 
8916 		res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
8917 		MLX5_SET(set_action_in, action_ctx.action_in, action_type,
8918 			 MLX5_MODIFICATION_TYPE_SET);
8919 		MLX5_SET(set_action_in, action_ctx.action_in, field,
8920 			 MLX5_MODI_META_REG_C_0);
8921 		MLX5_SET(set_action_in, action_ctx.action_in, data,
8922 			 priv->vport_meta_tag);
8923 		res->set_action = action_ctx.set_action;
8924 	} else if (attr->ingress) {
8925 		res->ft_type = MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
8926 	}
8927 	return 0;
8928 }
8929 
8930 /**
8931  * Convert Sample action to DV specification.
8932  *
8933  * @param[in] dev
8934  *   Pointer to rte_eth_dev structure.
8935  * @param[in, out] dev_flow
8936  *   Pointer to the mlx5_flow.
8937  * @param[in] attr
8938  *   Pointer to the flow attributes.
8939  * @param[in] num_of_dest
8940  *   The num of destination.
8941  * @param[in, out] res
8942  *   Pointer to sample resource.
8943  * @param[in, out] mdest_res
8944  *   Pointer to destination array resource.
8945  * @param[in] sample_actions
8946  *   Pointer to sample path actions list.
8947  * @param[in] action_flags
8948  *   Holds the actions detected until now.
8949  * @param[out] error
8950  *   Pointer to the error structure.
8951  *
8952  * @return
8953  *   0 on success, a negative errno value otherwise and rte_errno is set.
8954  */
8955 static int
8956 flow_dv_create_action_sample(struct rte_eth_dev *dev,
8957 			     struct mlx5_flow *dev_flow,
8958 			     const struct rte_flow_attr *attr,
8959 			     uint32_t num_of_dest,
8960 			     struct mlx5_flow_dv_sample_resource *res,
8961 			     struct mlx5_flow_dv_dest_array_resource *mdest_res,
8962 			     void **sample_actions,
8963 			     uint64_t action_flags,
8964 			     struct rte_flow_error *error)
8965 {
8966 	struct mlx5_priv *priv = dev->data->dev_private;
8967 	/* update normal path action resource into last index of array */
8968 	uint32_t dest_index = MLX5_MAX_DEST_NUM - 1;
8969 	struct mlx5_flow_sub_actions_list *sample_act =
8970 					&mdest_res->sample_act[dest_index];
8971 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8972 					      priv->rss_desc)
8973 					      [!!priv->flow_nested_idx];
8974 	uint32_t normal_idx = 0;
8975 	struct mlx5_hrxq *hrxq;
8976 	uint32_t hrxq_idx;
8977 
8978 	if (num_of_dest > 1) {
8979 		if (sample_act->action_flags & MLX5_FLOW_ACTION_QUEUE) {
8980 			/* Handle QP action for mirroring */
8981 			hrxq = flow_dv_handle_rx_queue(dev, dev_flow,
8982 						       rss_desc, &hrxq_idx);
8983 			if (!hrxq)
8984 				return rte_flow_error_set
8985 				     (error, rte_errno,
8986 				      RTE_FLOW_ERROR_TYPE_ACTION,
8987 				      NULL,
8988 				      "cannot create rx queue");
8989 			normal_idx++;
8990 			mdest_res->sample_idx[dest_index].rix_hrxq = hrxq_idx;
8991 			sample_act->dr_queue_action = hrxq->action;
8992 			if (action_flags & MLX5_FLOW_ACTION_MARK)
8993 				dev_flow->handle->rix_hrxq = hrxq_idx;
8994 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
8995 		}
8996 		if (sample_act->action_flags & MLX5_FLOW_ACTION_ENCAP) {
8997 			normal_idx++;
8998 			mdest_res->sample_idx[dest_index].rix_encap_decap =
8999 				dev_flow->handle->dvh.rix_encap_decap;
9000 			sample_act->dr_encap_action =
9001 				dev_flow->dv.encap_decap->action;
9002 		}
9003 		if (sample_act->action_flags & MLX5_FLOW_ACTION_PORT_ID) {
9004 			normal_idx++;
9005 			mdest_res->sample_idx[dest_index].rix_port_id_action =
9006 				dev_flow->handle->rix_port_id_action;
9007 			sample_act->dr_port_id_action =
9008 				dev_flow->dv.port_id_action->action;
9009 		}
9010 		sample_act->actions_num = normal_idx;
9011 		/* update sample action resource into first index of array */
9012 		mdest_res->ft_type = res->ft_type;
9013 		memcpy(&mdest_res->sample_idx[0], &res->sample_idx,
9014 				sizeof(struct mlx5_flow_sub_actions_idx));
9015 		memcpy(&mdest_res->sample_act[0], &res->sample_act,
9016 				sizeof(struct mlx5_flow_sub_actions_list));
9017 		mdest_res->num_of_dest = num_of_dest;
9018 		if (flow_dv_dest_array_resource_register(dev, attr, mdest_res,
9019 							 dev_flow, error))
9020 			return rte_flow_error_set(error, EINVAL,
9021 						  RTE_FLOW_ERROR_TYPE_ACTION,
9022 						  NULL, "can't create sample "
9023 						  "action");
9024 	} else {
9025 		if (flow_dv_sample_resource_register(dev, attr, res, dev_flow,
9026 						     sample_actions, error))
9027 			return rte_flow_error_set(error, EINVAL,
9028 						  RTE_FLOW_ERROR_TYPE_ACTION,
9029 						  NULL,
9030 						  "can't create sample action");
9031 	}
9032 	return 0;
9033 }
9034 
9035 /**
9036  * Fill the flow with DV spec, lock free
9037  * (mutex should be acquired by caller).
9038  *
9039  * @param[in] dev
9040  *   Pointer to rte_eth_dev structure.
9041  * @param[in, out] dev_flow
9042  *   Pointer to the sub flow.
9043  * @param[in] attr
9044  *   Pointer to the flow attributes.
9045  * @param[in] items
9046  *   Pointer to the list of items.
9047  * @param[in] actions
9048  *   Pointer to the list of actions.
9049  * @param[out] error
9050  *   Pointer to the error structure.
9051  *
9052  * @return
9053  *   0 on success, a negative errno value otherwise and rte_errno is set.
9054  */
9055 static int
9056 __flow_dv_translate(struct rte_eth_dev *dev,
9057 		    struct mlx5_flow *dev_flow,
9058 		    const struct rte_flow_attr *attr,
9059 		    const struct rte_flow_item items[],
9060 		    const struct rte_flow_action actions[],
9061 		    struct rte_flow_error *error)
9062 {
9063 	struct mlx5_priv *priv = dev->data->dev_private;
9064 	struct mlx5_dev_config *dev_conf = &priv->config;
9065 	struct rte_flow *flow = dev_flow->flow;
9066 	struct mlx5_flow_handle *handle = dev_flow->handle;
9067 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
9068 					      priv->rss_desc)
9069 					      [!!priv->flow_nested_idx];
9070 	uint64_t item_flags = 0;
9071 	uint64_t last_item = 0;
9072 	uint64_t action_flags = 0;
9073 	uint64_t priority = attr->priority;
9074 	struct mlx5_flow_dv_matcher matcher = {
9075 		.mask = {
9076 			.size = sizeof(matcher.mask.buf) -
9077 				MLX5_ST_SZ_BYTES(fte_match_set_misc4),
9078 		},
9079 	};
9080 	int actions_n = 0;
9081 	bool actions_end = false;
9082 	union {
9083 		struct mlx5_flow_dv_modify_hdr_resource res;
9084 		uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
9085 			    sizeof(struct mlx5_modification_cmd) *
9086 			    (MLX5_MAX_MODIFY_NUM + 1)];
9087 	} mhdr_dummy;
9088 	struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
9089 	const struct rte_flow_action_count *count = NULL;
9090 	const struct rte_flow_action_age *age = NULL;
9091 	union flow_dv_attr flow_attr = { .attr = 0 };
9092 	uint32_t tag_be;
9093 	union mlx5_flow_tbl_key tbl_key;
9094 	uint32_t modify_action_position = UINT32_MAX;
9095 	void *match_mask = matcher.mask.buf;
9096 	void *match_value = dev_flow->dv.value.buf;
9097 	uint8_t next_protocol = 0xff;
9098 	struct rte_vlan_hdr vlan = { 0 };
9099 	struct mlx5_flow_dv_dest_array_resource mdest_res;
9100 	struct mlx5_flow_dv_sample_resource sample_res;
9101 	void *sample_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9102 	struct mlx5_flow_sub_actions_list *sample_act;
9103 	uint32_t sample_act_pos = UINT32_MAX;
9104 	uint32_t num_of_dest = 0;
9105 	int tmp_actions_n = 0;
9106 	uint32_t table;
9107 	int ret = 0;
9108 
9109 	memset(&mdest_res, 0, sizeof(struct mlx5_flow_dv_dest_array_resource));
9110 	memset(&sample_res, 0, sizeof(struct mlx5_flow_dv_sample_resource));
9111 	mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
9112 					   MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
9113 	/* update normal path action resource into last index of array */
9114 	sample_act = &mdest_res.sample_act[MLX5_MAX_DEST_NUM - 1];
9115 	ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
9116 				       !!priv->fdb_def_rule, &table, error);
9117 	if (ret)
9118 		return ret;
9119 	dev_flow->dv.group = table;
9120 	if (attr->transfer)
9121 		mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
9122 	if (priority == MLX5_FLOW_PRIO_RSVD)
9123 		priority = dev_conf->flow_prio - 1;
9124 	/* number of actions must be set to 0 in case of dirty stack. */
9125 	mhdr_res->actions_num = 0;
9126 	for (; !actions_end ; actions++) {
9127 		const struct rte_flow_action_queue *queue;
9128 		const struct rte_flow_action_rss *rss;
9129 		const struct rte_flow_action *action = actions;
9130 		const uint8_t *rss_key;
9131 		const struct rte_flow_action_meter *mtr;
9132 		struct mlx5_flow_tbl_resource *tbl;
9133 		uint32_t port_id = 0;
9134 		struct mlx5_flow_dv_port_id_action_resource port_id_resource;
9135 		int action_type = actions->type;
9136 		const struct rte_flow_action *found_action = NULL;
9137 		struct mlx5_flow_meter *fm = NULL;
9138 		uint32_t jump_group = 0;
9139 
9140 		if (!mlx5_flow_os_action_supported(action_type))
9141 			return rte_flow_error_set(error, ENOTSUP,
9142 						  RTE_FLOW_ERROR_TYPE_ACTION,
9143 						  actions,
9144 						  "action not supported");
9145 		switch (action_type) {
9146 		case RTE_FLOW_ACTION_TYPE_VOID:
9147 			break;
9148 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
9149 			if (flow_dv_translate_action_port_id(dev, action,
9150 							     &port_id, error))
9151 				return -rte_errno;
9152 			port_id_resource.port_id = port_id;
9153 			MLX5_ASSERT(!handle->rix_port_id_action);
9154 			if (flow_dv_port_id_action_resource_register
9155 			    (dev, &port_id_resource, dev_flow, error))
9156 				return -rte_errno;
9157 			dev_flow->dv.actions[actions_n++] =
9158 					dev_flow->dv.port_id_action->action;
9159 			action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9160 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
9161 			sample_act->action_flags |= MLX5_FLOW_ACTION_PORT_ID;
9162 			num_of_dest++;
9163 			break;
9164 		case RTE_FLOW_ACTION_TYPE_FLAG:
9165 			action_flags |= MLX5_FLOW_ACTION_FLAG;
9166 			dev_flow->handle->mark = 1;
9167 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9168 				struct rte_flow_action_mark mark = {
9169 					.id = MLX5_FLOW_MARK_DEFAULT,
9170 				};
9171 
9172 				if (flow_dv_convert_action_mark(dev, &mark,
9173 								mhdr_res,
9174 								error))
9175 					return -rte_errno;
9176 				action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9177 				break;
9178 			}
9179 			tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
9180 			/*
9181 			 * Only one FLAG or MARK is supported per device flow
9182 			 * right now. So the pointer to the tag resource must be
9183 			 * zero before the register process.
9184 			 */
9185 			MLX5_ASSERT(!handle->dvh.rix_tag);
9186 			if (flow_dv_tag_resource_register(dev, tag_be,
9187 							  dev_flow, error))
9188 				return -rte_errno;
9189 			MLX5_ASSERT(dev_flow->dv.tag_resource);
9190 			dev_flow->dv.actions[actions_n++] =
9191 					dev_flow->dv.tag_resource->action;
9192 			break;
9193 		case RTE_FLOW_ACTION_TYPE_MARK:
9194 			action_flags |= MLX5_FLOW_ACTION_MARK;
9195 			dev_flow->handle->mark = 1;
9196 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
9197 				const struct rte_flow_action_mark *mark =
9198 					(const struct rte_flow_action_mark *)
9199 						actions->conf;
9200 
9201 				if (flow_dv_convert_action_mark(dev, mark,
9202 								mhdr_res,
9203 								error))
9204 					return -rte_errno;
9205 				action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
9206 				break;
9207 			}
9208 			/* Fall-through */
9209 		case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
9210 			/* Legacy (non-extensive) MARK action. */
9211 			tag_be = mlx5_flow_mark_set
9212 			      (((const struct rte_flow_action_mark *)
9213 			       (actions->conf))->id);
9214 			MLX5_ASSERT(!handle->dvh.rix_tag);
9215 			if (flow_dv_tag_resource_register(dev, tag_be,
9216 							  dev_flow, error))
9217 				return -rte_errno;
9218 			MLX5_ASSERT(dev_flow->dv.tag_resource);
9219 			dev_flow->dv.actions[actions_n++] =
9220 					dev_flow->dv.tag_resource->action;
9221 			break;
9222 		case RTE_FLOW_ACTION_TYPE_SET_META:
9223 			if (flow_dv_convert_action_set_meta
9224 				(dev, mhdr_res, attr,
9225 				 (const struct rte_flow_action_set_meta *)
9226 				  actions->conf, error))
9227 				return -rte_errno;
9228 			action_flags |= MLX5_FLOW_ACTION_SET_META;
9229 			break;
9230 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
9231 			if (flow_dv_convert_action_set_tag
9232 				(dev, mhdr_res,
9233 				 (const struct rte_flow_action_set_tag *)
9234 				  actions->conf, error))
9235 				return -rte_errno;
9236 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9237 			break;
9238 		case RTE_FLOW_ACTION_TYPE_DROP:
9239 			action_flags |= MLX5_FLOW_ACTION_DROP;
9240 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
9241 			break;
9242 		case RTE_FLOW_ACTION_TYPE_QUEUE:
9243 			queue = actions->conf;
9244 			rss_desc->queue_num = 1;
9245 			rss_desc->queue[0] = queue->index;
9246 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
9247 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9248 			sample_act->action_flags |= MLX5_FLOW_ACTION_QUEUE;
9249 			num_of_dest++;
9250 			break;
9251 		case RTE_FLOW_ACTION_TYPE_RSS:
9252 			rss = actions->conf;
9253 			memcpy(rss_desc->queue, rss->queue,
9254 			       rss->queue_num * sizeof(uint16_t));
9255 			rss_desc->queue_num = rss->queue_num;
9256 			/* NULL RSS key indicates default RSS key. */
9257 			rss_key = !rss->key ? rss_hash_default_key : rss->key;
9258 			memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
9259 			/*
9260 			 * rss->level and rss.types should be set in advance
9261 			 * when expanding items for RSS.
9262 			 */
9263 			action_flags |= MLX5_FLOW_ACTION_RSS;
9264 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
9265 			break;
9266 		case RTE_FLOW_ACTION_TYPE_AGE:
9267 		case RTE_FLOW_ACTION_TYPE_COUNT:
9268 			if (!dev_conf->devx) {
9269 				return rte_flow_error_set
9270 					      (error, ENOTSUP,
9271 					       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9272 					       NULL,
9273 					       "count action not supported");
9274 			}
9275 			/* Save information first, will apply later. */
9276 			if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
9277 				count = action->conf;
9278 			else
9279 				age = action->conf;
9280 			action_flags |= MLX5_FLOW_ACTION_COUNT;
9281 			break;
9282 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
9283 			dev_flow->dv.actions[actions_n++] =
9284 						priv->sh->pop_vlan_action;
9285 			action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
9286 			break;
9287 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
9288 			if (!(action_flags &
9289 			      MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
9290 				flow_dev_get_vlan_info_from_items(items, &vlan);
9291 			vlan.eth_proto = rte_be_to_cpu_16
9292 			     ((((const struct rte_flow_action_of_push_vlan *)
9293 						   actions->conf)->ethertype));
9294 			found_action = mlx5_flow_find_action
9295 					(actions + 1,
9296 					 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
9297 			if (found_action)
9298 				mlx5_update_vlan_vid_pcp(found_action, &vlan);
9299 			found_action = mlx5_flow_find_action
9300 					(actions + 1,
9301 					 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
9302 			if (found_action)
9303 				mlx5_update_vlan_vid_pcp(found_action, &vlan);
9304 			if (flow_dv_create_action_push_vlan
9305 					    (dev, attr, &vlan, dev_flow, error))
9306 				return -rte_errno;
9307 			dev_flow->dv.actions[actions_n++] =
9308 					dev_flow->dv.push_vlan_res->action;
9309 			action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
9310 			break;
9311 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
9312 			/* of_vlan_push action handled this action */
9313 			MLX5_ASSERT(action_flags &
9314 				    MLX5_FLOW_ACTION_OF_PUSH_VLAN);
9315 			break;
9316 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
9317 			if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
9318 				break;
9319 			flow_dev_get_vlan_info_from_items(items, &vlan);
9320 			mlx5_update_vlan_vid_pcp(actions, &vlan);
9321 			/* If no VLAN push - this is a modify header action */
9322 			if (flow_dv_convert_action_modify_vlan_vid
9323 						(mhdr_res, actions, error))
9324 				return -rte_errno;
9325 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
9326 			break;
9327 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
9328 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
9329 			if (flow_dv_create_action_l2_encap(dev, actions,
9330 							   dev_flow,
9331 							   attr->transfer,
9332 							   error))
9333 				return -rte_errno;
9334 			dev_flow->dv.actions[actions_n++] =
9335 					dev_flow->dv.encap_decap->action;
9336 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
9337 			if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9338 				sample_act->action_flags |=
9339 							MLX5_FLOW_ACTION_ENCAP;
9340 			break;
9341 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
9342 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
9343 			if (flow_dv_create_action_l2_decap(dev, dev_flow,
9344 							   attr->transfer,
9345 							   error))
9346 				return -rte_errno;
9347 			dev_flow->dv.actions[actions_n++] =
9348 					dev_flow->dv.encap_decap->action;
9349 			action_flags |= MLX5_FLOW_ACTION_DECAP;
9350 			break;
9351 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
9352 			/* Handle encap with preceding decap. */
9353 			if (action_flags & MLX5_FLOW_ACTION_DECAP) {
9354 				if (flow_dv_create_action_raw_encap
9355 					(dev, actions, dev_flow, attr, error))
9356 					return -rte_errno;
9357 				dev_flow->dv.actions[actions_n++] =
9358 					dev_flow->dv.encap_decap->action;
9359 			} else {
9360 				/* Handle encap without preceding decap. */
9361 				if (flow_dv_create_action_l2_encap
9362 				    (dev, actions, dev_flow, attr->transfer,
9363 				     error))
9364 					return -rte_errno;
9365 				dev_flow->dv.actions[actions_n++] =
9366 					dev_flow->dv.encap_decap->action;
9367 			}
9368 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
9369 			if (action_flags & MLX5_FLOW_ACTION_SAMPLE)
9370 				sample_act->action_flags |=
9371 							MLX5_FLOW_ACTION_ENCAP;
9372 			break;
9373 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
9374 			while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
9375 				;
9376 			if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
9377 				if (flow_dv_create_action_l2_decap
9378 				    (dev, dev_flow, attr->transfer, error))
9379 					return -rte_errno;
9380 				dev_flow->dv.actions[actions_n++] =
9381 					dev_flow->dv.encap_decap->action;
9382 			}
9383 			/* If decap is followed by encap, handle it at encap. */
9384 			action_flags |= MLX5_FLOW_ACTION_DECAP;
9385 			break;
9386 		case RTE_FLOW_ACTION_TYPE_JUMP:
9387 			jump_group = ((const struct rte_flow_action_jump *)
9388 							action->conf)->group;
9389 			if (dev_flow->external && jump_group <
9390 					MLX5_MAX_TABLES_EXTERNAL)
9391 				jump_group *= MLX5_FLOW_TABLE_FACTOR;
9392 			ret = mlx5_flow_group_to_table(attr, dev_flow->external,
9393 						       jump_group,
9394 						       !!priv->fdb_def_rule,
9395 						       &table, error);
9396 			if (ret)
9397 				return ret;
9398 			tbl = flow_dv_tbl_resource_get(dev, table,
9399 						       attr->egress,
9400 						       attr->transfer, error);
9401 			if (!tbl)
9402 				return rte_flow_error_set
9403 						(error, errno,
9404 						 RTE_FLOW_ERROR_TYPE_ACTION,
9405 						 NULL,
9406 						 "cannot create jump action.");
9407 			if (flow_dv_jump_tbl_resource_register
9408 			    (dev, tbl, dev_flow, error)) {
9409 				flow_dv_tbl_resource_release(dev, tbl);
9410 				return rte_flow_error_set
9411 						(error, errno,
9412 						 RTE_FLOW_ERROR_TYPE_ACTION,
9413 						 NULL,
9414 						 "cannot create jump action.");
9415 			}
9416 			dev_flow->dv.actions[actions_n++] =
9417 					dev_flow->dv.jump->action;
9418 			action_flags |= MLX5_FLOW_ACTION_JUMP;
9419 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
9420 			break;
9421 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
9422 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
9423 			if (flow_dv_convert_action_modify_mac
9424 					(mhdr_res, actions, error))
9425 				return -rte_errno;
9426 			action_flags |= actions->type ==
9427 					RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
9428 					MLX5_FLOW_ACTION_SET_MAC_SRC :
9429 					MLX5_FLOW_ACTION_SET_MAC_DST;
9430 			break;
9431 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
9432 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
9433 			if (flow_dv_convert_action_modify_ipv4
9434 					(mhdr_res, actions, error))
9435 				return -rte_errno;
9436 			action_flags |= actions->type ==
9437 					RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
9438 					MLX5_FLOW_ACTION_SET_IPV4_SRC :
9439 					MLX5_FLOW_ACTION_SET_IPV4_DST;
9440 			break;
9441 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
9442 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
9443 			if (flow_dv_convert_action_modify_ipv6
9444 					(mhdr_res, actions, error))
9445 				return -rte_errno;
9446 			action_flags |= actions->type ==
9447 					RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
9448 					MLX5_FLOW_ACTION_SET_IPV6_SRC :
9449 					MLX5_FLOW_ACTION_SET_IPV6_DST;
9450 			break;
9451 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
9452 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
9453 			if (flow_dv_convert_action_modify_tp
9454 					(mhdr_res, actions, items,
9455 					 &flow_attr, dev_flow, !!(action_flags &
9456 					 MLX5_FLOW_ACTION_DECAP), error))
9457 				return -rte_errno;
9458 			action_flags |= actions->type ==
9459 					RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
9460 					MLX5_FLOW_ACTION_SET_TP_SRC :
9461 					MLX5_FLOW_ACTION_SET_TP_DST;
9462 			break;
9463 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
9464 			if (flow_dv_convert_action_modify_dec_ttl
9465 					(mhdr_res, items, &flow_attr, dev_flow,
9466 					 !!(action_flags &
9467 					 MLX5_FLOW_ACTION_DECAP), error))
9468 				return -rte_errno;
9469 			action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
9470 			break;
9471 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
9472 			if (flow_dv_convert_action_modify_ttl
9473 					(mhdr_res, actions, items, &flow_attr,
9474 					 dev_flow, !!(action_flags &
9475 					 MLX5_FLOW_ACTION_DECAP), error))
9476 				return -rte_errno;
9477 			action_flags |= MLX5_FLOW_ACTION_SET_TTL;
9478 			break;
9479 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
9480 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
9481 			if (flow_dv_convert_action_modify_tcp_seq
9482 					(mhdr_res, actions, error))
9483 				return -rte_errno;
9484 			action_flags |= actions->type ==
9485 					RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
9486 					MLX5_FLOW_ACTION_INC_TCP_SEQ :
9487 					MLX5_FLOW_ACTION_DEC_TCP_SEQ;
9488 			break;
9489 
9490 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
9491 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
9492 			if (flow_dv_convert_action_modify_tcp_ack
9493 					(mhdr_res, actions, error))
9494 				return -rte_errno;
9495 			action_flags |= actions->type ==
9496 					RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
9497 					MLX5_FLOW_ACTION_INC_TCP_ACK :
9498 					MLX5_FLOW_ACTION_DEC_TCP_ACK;
9499 			break;
9500 		case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
9501 			if (flow_dv_convert_action_set_reg
9502 					(mhdr_res, actions, error))
9503 				return -rte_errno;
9504 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9505 			break;
9506 		case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
9507 			if (flow_dv_convert_action_copy_mreg
9508 					(dev, mhdr_res, actions, error))
9509 				return -rte_errno;
9510 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
9511 			break;
9512 		case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
9513 			action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
9514 			dev_flow->handle->fate_action =
9515 					MLX5_FLOW_FATE_DEFAULT_MISS;
9516 			break;
9517 		case RTE_FLOW_ACTION_TYPE_METER:
9518 			mtr = actions->conf;
9519 			if (!flow->meter) {
9520 				fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
9521 							    attr, error);
9522 				if (!fm)
9523 					return rte_flow_error_set(error,
9524 						rte_errno,
9525 						RTE_FLOW_ERROR_TYPE_ACTION,
9526 						NULL,
9527 						"meter not found "
9528 						"or invalid parameters");
9529 				flow->meter = fm->idx;
9530 			}
9531 			/* Set the meter action. */
9532 			if (!fm) {
9533 				fm = mlx5_ipool_get(priv->sh->ipool
9534 						[MLX5_IPOOL_MTR], flow->meter);
9535 				if (!fm)
9536 					return rte_flow_error_set(error,
9537 						rte_errno,
9538 						RTE_FLOW_ERROR_TYPE_ACTION,
9539 						NULL,
9540 						"meter not found "
9541 						"or invalid parameters");
9542 			}
9543 			dev_flow->dv.actions[actions_n++] =
9544 				fm->mfts->meter_action;
9545 			action_flags |= MLX5_FLOW_ACTION_METER;
9546 			break;
9547 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
9548 			if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
9549 							      actions, error))
9550 				return -rte_errno;
9551 			action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
9552 			break;
9553 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
9554 			if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
9555 							      actions, error))
9556 				return -rte_errno;
9557 			action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
9558 			break;
9559 		case RTE_FLOW_ACTION_TYPE_SAMPLE:
9560 			sample_act_pos = actions_n;
9561 			ret = flow_dv_translate_action_sample(dev,
9562 							      actions,
9563 							      dev_flow, attr,
9564 							      &num_of_dest,
9565 							      sample_actions,
9566 							      &sample_res,
9567 							      error);
9568 			if (ret < 0)
9569 				return ret;
9570 			actions_n++;
9571 			action_flags |= MLX5_FLOW_ACTION_SAMPLE;
9572 			/* put encap action into group if work with port id */
9573 			if ((action_flags & MLX5_FLOW_ACTION_ENCAP) &&
9574 			    (action_flags & MLX5_FLOW_ACTION_PORT_ID))
9575 				sample_act->action_flags |=
9576 							MLX5_FLOW_ACTION_ENCAP;
9577 			break;
9578 		case RTE_FLOW_ACTION_TYPE_END:
9579 			actions_end = true;
9580 			if (mhdr_res->actions_num) {
9581 				/* create modify action if needed. */
9582 				if (flow_dv_modify_hdr_resource_register
9583 					(dev, mhdr_res, dev_flow, error))
9584 					return -rte_errno;
9585 				dev_flow->dv.actions[modify_action_position] =
9586 					handle->dvh.modify_hdr->action;
9587 			}
9588 			if (action_flags & MLX5_FLOW_ACTION_COUNT) {
9589 				flow->counter =
9590 					flow_dv_translate_create_counter(dev,
9591 						dev_flow, count, age);
9592 
9593 				if (!flow->counter)
9594 					return rte_flow_error_set
9595 						(error, rte_errno,
9596 						RTE_FLOW_ERROR_TYPE_ACTION,
9597 						NULL,
9598 						"cannot create counter"
9599 						" object.");
9600 				dev_flow->dv.actions[actions_n] =
9601 					  (flow_dv_counter_get_by_idx(dev,
9602 					  flow->counter, NULL))->action;
9603 				actions_n++;
9604 			}
9605 			if (action_flags & MLX5_FLOW_ACTION_SAMPLE) {
9606 				ret = flow_dv_create_action_sample(dev,
9607 							  dev_flow, attr,
9608 							  num_of_dest,
9609 							  &sample_res,
9610 							  &mdest_res,
9611 							  sample_actions,
9612 							  action_flags,
9613 							  error);
9614 				if (ret < 0)
9615 					return rte_flow_error_set
9616 						(error, rte_errno,
9617 						RTE_FLOW_ERROR_TYPE_ACTION,
9618 						NULL,
9619 						"cannot create sample action");
9620 				if (num_of_dest > 1) {
9621 					dev_flow->dv.actions[sample_act_pos] =
9622 					dev_flow->dv.dest_array_res->action;
9623 				} else {
9624 					dev_flow->dv.actions[sample_act_pos] =
9625 					dev_flow->dv.sample_res->verbs_action;
9626 				}
9627 			}
9628 			break;
9629 		default:
9630 			break;
9631 		}
9632 		if (mhdr_res->actions_num &&
9633 		    modify_action_position == UINT32_MAX)
9634 			modify_action_position = actions_n++;
9635 	}
9636 	/*
9637 	 * For multiple destination (sample action with ratio=1), the encap
9638 	 * action and port id action will be combined into group action.
9639 	 * So need remove the original these actions in the flow and only
9640 	 * use the sample action instead of.
9641 	 */
9642 	if (num_of_dest > 1 && sample_act->dr_port_id_action) {
9643 		int i;
9644 		void *temp_actions[MLX5_DV_MAX_NUMBER_OF_ACTIONS] = {0};
9645 
9646 		for (i = 0; i < actions_n; i++) {
9647 			if ((sample_act->dr_encap_action &&
9648 				sample_act->dr_encap_action ==
9649 				dev_flow->dv.actions[i]) ||
9650 				(sample_act->dr_port_id_action &&
9651 				sample_act->dr_port_id_action ==
9652 				dev_flow->dv.actions[i]))
9653 				continue;
9654 			temp_actions[tmp_actions_n++] = dev_flow->dv.actions[i];
9655 		}
9656 		memcpy((void *)dev_flow->dv.actions,
9657 				(void *)temp_actions,
9658 				tmp_actions_n * sizeof(void *));
9659 		actions_n = tmp_actions_n;
9660 	}
9661 	dev_flow->dv.actions_n = actions_n;
9662 	dev_flow->act_flags = action_flags;
9663 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
9664 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
9665 		int item_type = items->type;
9666 
9667 		if (!mlx5_flow_os_item_supported(item_type))
9668 			return rte_flow_error_set(error, ENOTSUP,
9669 						  RTE_FLOW_ERROR_TYPE_ITEM,
9670 						  NULL, "item not supported");
9671 		switch (item_type) {
9672 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
9673 			flow_dv_translate_item_port_id(dev, match_mask,
9674 						       match_value, items);
9675 			last_item = MLX5_FLOW_ITEM_PORT_ID;
9676 			break;
9677 		case RTE_FLOW_ITEM_TYPE_ETH:
9678 			flow_dv_translate_item_eth(match_mask, match_value,
9679 						   items, tunnel,
9680 						   dev_flow->dv.group);
9681 			matcher.priority = action_flags &
9682 					MLX5_FLOW_ACTION_DEFAULT_MISS &&
9683 					!dev_flow->external ?
9684 					MLX5_PRIORITY_MAP_L3 :
9685 					MLX5_PRIORITY_MAP_L2;
9686 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
9687 					     MLX5_FLOW_LAYER_OUTER_L2;
9688 			break;
9689 		case RTE_FLOW_ITEM_TYPE_VLAN:
9690 			flow_dv_translate_item_vlan(dev_flow,
9691 						    match_mask, match_value,
9692 						    items, tunnel,
9693 						    dev_flow->dv.group);
9694 			matcher.priority = MLX5_PRIORITY_MAP_L2;
9695 			last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
9696 					      MLX5_FLOW_LAYER_INNER_VLAN) :
9697 					     (MLX5_FLOW_LAYER_OUTER_L2 |
9698 					      MLX5_FLOW_LAYER_OUTER_VLAN);
9699 			break;
9700 		case RTE_FLOW_ITEM_TYPE_IPV4:
9701 			mlx5_flow_tunnel_ip_check(items, next_protocol,
9702 						  &item_flags, &tunnel);
9703 			flow_dv_translate_item_ipv4(match_mask, match_value,
9704 						    items, item_flags, tunnel,
9705 						    dev_flow->dv.group);
9706 			matcher.priority = MLX5_PRIORITY_MAP_L3;
9707 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
9708 					     MLX5_FLOW_LAYER_OUTER_L3_IPV4;
9709 			if (items->mask != NULL &&
9710 			    ((const struct rte_flow_item_ipv4 *)
9711 			     items->mask)->hdr.next_proto_id) {
9712 				next_protocol =
9713 					((const struct rte_flow_item_ipv4 *)
9714 					 (items->spec))->hdr.next_proto_id;
9715 				next_protocol &=
9716 					((const struct rte_flow_item_ipv4 *)
9717 					 (items->mask))->hdr.next_proto_id;
9718 			} else {
9719 				/* Reset for inner layer. */
9720 				next_protocol = 0xff;
9721 			}
9722 			break;
9723 		case RTE_FLOW_ITEM_TYPE_IPV6:
9724 			mlx5_flow_tunnel_ip_check(items, next_protocol,
9725 						  &item_flags, &tunnel);
9726 			flow_dv_translate_item_ipv6(match_mask, match_value,
9727 						    items, item_flags, tunnel,
9728 						    dev_flow->dv.group);
9729 			matcher.priority = MLX5_PRIORITY_MAP_L3;
9730 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
9731 					     MLX5_FLOW_LAYER_OUTER_L3_IPV6;
9732 			if (items->mask != NULL &&
9733 			    ((const struct rte_flow_item_ipv6 *)
9734 			     items->mask)->hdr.proto) {
9735 				next_protocol =
9736 					((const struct rte_flow_item_ipv6 *)
9737 					 items->spec)->hdr.proto;
9738 				next_protocol &=
9739 					((const struct rte_flow_item_ipv6 *)
9740 					 items->mask)->hdr.proto;
9741 			} else {
9742 				/* Reset for inner layer. */
9743 				next_protocol = 0xff;
9744 			}
9745 			break;
9746 		case RTE_FLOW_ITEM_TYPE_TCP:
9747 			flow_dv_translate_item_tcp(match_mask, match_value,
9748 						   items, tunnel);
9749 			matcher.priority = MLX5_PRIORITY_MAP_L4;
9750 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
9751 					     MLX5_FLOW_LAYER_OUTER_L4_TCP;
9752 			break;
9753 		case RTE_FLOW_ITEM_TYPE_UDP:
9754 			flow_dv_translate_item_udp(match_mask, match_value,
9755 						   items, tunnel);
9756 			matcher.priority = MLX5_PRIORITY_MAP_L4;
9757 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
9758 					     MLX5_FLOW_LAYER_OUTER_L4_UDP;
9759 			break;
9760 		case RTE_FLOW_ITEM_TYPE_GRE:
9761 			flow_dv_translate_item_gre(match_mask, match_value,
9762 						   items, tunnel);
9763 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9764 			last_item = MLX5_FLOW_LAYER_GRE;
9765 			break;
9766 		case RTE_FLOW_ITEM_TYPE_GRE_KEY:
9767 			flow_dv_translate_item_gre_key(match_mask,
9768 						       match_value, items);
9769 			last_item = MLX5_FLOW_LAYER_GRE_KEY;
9770 			break;
9771 		case RTE_FLOW_ITEM_TYPE_NVGRE:
9772 			flow_dv_translate_item_nvgre(match_mask, match_value,
9773 						     items, tunnel);
9774 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9775 			last_item = MLX5_FLOW_LAYER_GRE;
9776 			break;
9777 		case RTE_FLOW_ITEM_TYPE_VXLAN:
9778 			flow_dv_translate_item_vxlan(match_mask, match_value,
9779 						     items, tunnel);
9780 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9781 			last_item = MLX5_FLOW_LAYER_VXLAN;
9782 			break;
9783 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
9784 			flow_dv_translate_item_vxlan_gpe(match_mask,
9785 							 match_value, items,
9786 							 tunnel);
9787 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9788 			last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
9789 			break;
9790 		case RTE_FLOW_ITEM_TYPE_GENEVE:
9791 			flow_dv_translate_item_geneve(match_mask, match_value,
9792 						      items, tunnel);
9793 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9794 			last_item = MLX5_FLOW_LAYER_GENEVE;
9795 			break;
9796 		case RTE_FLOW_ITEM_TYPE_MPLS:
9797 			flow_dv_translate_item_mpls(match_mask, match_value,
9798 						    items, last_item, tunnel);
9799 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9800 			last_item = MLX5_FLOW_LAYER_MPLS;
9801 			break;
9802 		case RTE_FLOW_ITEM_TYPE_MARK:
9803 			flow_dv_translate_item_mark(dev, match_mask,
9804 						    match_value, items);
9805 			last_item = MLX5_FLOW_ITEM_MARK;
9806 			break;
9807 		case RTE_FLOW_ITEM_TYPE_META:
9808 			flow_dv_translate_item_meta(dev, match_mask,
9809 						    match_value, attr, items);
9810 			last_item = MLX5_FLOW_ITEM_METADATA;
9811 			break;
9812 		case RTE_FLOW_ITEM_TYPE_ICMP:
9813 			flow_dv_translate_item_icmp(match_mask, match_value,
9814 						    items, tunnel);
9815 			last_item = MLX5_FLOW_LAYER_ICMP;
9816 			break;
9817 		case RTE_FLOW_ITEM_TYPE_ICMP6:
9818 			flow_dv_translate_item_icmp6(match_mask, match_value,
9819 						      items, tunnel);
9820 			last_item = MLX5_FLOW_LAYER_ICMP6;
9821 			break;
9822 		case RTE_FLOW_ITEM_TYPE_TAG:
9823 			flow_dv_translate_item_tag(dev, match_mask,
9824 						   match_value, items);
9825 			last_item = MLX5_FLOW_ITEM_TAG;
9826 			break;
9827 		case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
9828 			flow_dv_translate_mlx5_item_tag(dev, match_mask,
9829 							match_value, items);
9830 			last_item = MLX5_FLOW_ITEM_TAG;
9831 			break;
9832 		case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
9833 			flow_dv_translate_item_tx_queue(dev, match_mask,
9834 							match_value,
9835 							items);
9836 			last_item = MLX5_FLOW_ITEM_TX_QUEUE;
9837 			break;
9838 		case RTE_FLOW_ITEM_TYPE_GTP:
9839 			flow_dv_translate_item_gtp(match_mask, match_value,
9840 						   items, tunnel);
9841 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
9842 			last_item = MLX5_FLOW_LAYER_GTP;
9843 			break;
9844 		case RTE_FLOW_ITEM_TYPE_ECPRI:
9845 			if (!mlx5_flex_parser_ecpri_exist(dev)) {
9846 				/* Create it only the first time to be used. */
9847 				ret = mlx5_flex_parser_ecpri_alloc(dev);
9848 				if (ret)
9849 					return rte_flow_error_set
9850 						(error, -ret,
9851 						RTE_FLOW_ERROR_TYPE_ITEM,
9852 						NULL,
9853 						"cannot create eCPRI parser");
9854 			}
9855 			/* Adjust the length matcher and device flow value. */
9856 			matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
9857 			dev_flow->dv.value.size =
9858 					MLX5_ST_SZ_BYTES(fte_match_param);
9859 			flow_dv_translate_item_ecpri(dev, match_mask,
9860 						     match_value, items);
9861 			/* No other protocol should follow eCPRI layer. */
9862 			last_item = MLX5_FLOW_LAYER_ECPRI;
9863 			break;
9864 		default:
9865 			break;
9866 		}
9867 		item_flags |= last_item;
9868 	}
9869 	/*
9870 	 * When E-Switch mode is enabled, we have two cases where we need to
9871 	 * set the source port manually.
9872 	 * The first one, is in case of Nic steering rule, and the second is
9873 	 * E-Switch rule where no port_id item was found. In both cases
9874 	 * the source port is set according the current port in use.
9875 	 */
9876 	if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
9877 	    (priv->representor || priv->master)) {
9878 		if (flow_dv_translate_item_port_id(dev, match_mask,
9879 						   match_value, NULL))
9880 			return -rte_errno;
9881 	}
9882 #ifdef RTE_LIBRTE_MLX5_DEBUG
9883 	MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
9884 					      dev_flow->dv.value.buf));
9885 #endif
9886 	/*
9887 	 * Layers may be already initialized from prefix flow if this dev_flow
9888 	 * is the suffix flow.
9889 	 */
9890 	handle->layers |= item_flags;
9891 	if (action_flags & MLX5_FLOW_ACTION_RSS)
9892 		flow_dv_hashfields_set(dev_flow, rss_desc);
9893 	/* Register matcher. */
9894 	matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
9895 				    matcher.mask.size);
9896 	matcher.priority = mlx5_flow_adjust_priority(dev, priority,
9897 						     matcher.priority);
9898 	/* reserved field no needs to be set to 0 here. */
9899 	tbl_key.domain = attr->transfer;
9900 	tbl_key.direction = attr->egress;
9901 	tbl_key.table_id = dev_flow->dv.group;
9902 	if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
9903 		return -rte_errno;
9904 	return 0;
9905 }
9906 
9907 /**
9908  * Apply the flow to the NIC, lock free,
9909  * (mutex should be acquired by caller).
9910  *
9911  * @param[in] dev
9912  *   Pointer to the Ethernet device structure.
9913  * @param[in, out] flow
9914  *   Pointer to flow structure.
9915  * @param[out] error
9916  *   Pointer to error structure.
9917  *
9918  * @return
9919  *   0 on success, a negative errno value otherwise and rte_errno is set.
9920  */
9921 static int
9922 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
9923 		struct rte_flow_error *error)
9924 {
9925 	struct mlx5_flow_dv_workspace *dv;
9926 	struct mlx5_flow_handle *dh;
9927 	struct mlx5_flow_handle_dv *dv_h;
9928 	struct mlx5_flow *dev_flow;
9929 	struct mlx5_priv *priv = dev->data->dev_private;
9930 	uint32_t handle_idx;
9931 	int n;
9932 	int err;
9933 	int idx;
9934 
9935 	for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
9936 		dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
9937 		dv = &dev_flow->dv;
9938 		dh = dev_flow->handle;
9939 		dv_h = &dh->dvh;
9940 		n = dv->actions_n;
9941 		if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
9942 			if (dv->transfer) {
9943 				dv->actions[n++] = priv->sh->esw_drop_action;
9944 			} else {
9945 				struct mlx5_hrxq *drop_hrxq;
9946 				drop_hrxq = mlx5_drop_action_create(dev);
9947 				if (!drop_hrxq) {
9948 					rte_flow_error_set
9949 						(error, errno,
9950 						 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9951 						 NULL,
9952 						 "cannot get drop hash queue");
9953 					goto error;
9954 				}
9955 				/*
9956 				 * Drop queues will be released by the specify
9957 				 * mlx5_drop_action_destroy() function. Assign
9958 				 * the special index to hrxq to mark the queue
9959 				 * has been allocated.
9960 				 */
9961 				dh->rix_hrxq = UINT32_MAX;
9962 				dv->actions[n++] = drop_hrxq->action;
9963 			}
9964 		} else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE &&
9965 			   !dv_h->rix_sample && !dv_h->rix_dest_array) {
9966 			struct mlx5_hrxq *hrxq;
9967 			uint32_t hrxq_idx;
9968 			struct mlx5_flow_rss_desc *rss_desc =
9969 				&((struct mlx5_flow_rss_desc *)priv->rss_desc)
9970 				[!!priv->flow_nested_idx];
9971 
9972 			MLX5_ASSERT(rss_desc->queue_num);
9973 			hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
9974 						 MLX5_RSS_HASH_KEY_LEN,
9975 						 dev_flow->hash_fields,
9976 						 rss_desc->queue,
9977 						 rss_desc->queue_num);
9978 			if (!hrxq_idx) {
9979 				hrxq_idx = mlx5_hrxq_new
9980 						(dev, rss_desc->key,
9981 						 MLX5_RSS_HASH_KEY_LEN,
9982 						 dev_flow->hash_fields,
9983 						 rss_desc->queue,
9984 						 rss_desc->queue_num,
9985 						 !!(dh->layers &
9986 						 MLX5_FLOW_LAYER_TUNNEL));
9987 			}
9988 			hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
9989 					      hrxq_idx);
9990 			if (!hrxq) {
9991 				rte_flow_error_set
9992 					(error, rte_errno,
9993 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
9994 					 "cannot get hash queue");
9995 				goto error;
9996 			}
9997 			dh->rix_hrxq = hrxq_idx;
9998 			dv->actions[n++] = hrxq->action;
9999 		} else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
10000 			if (flow_dv_default_miss_resource_register
10001 					(dev, error)) {
10002 				rte_flow_error_set
10003 					(error, rte_errno,
10004 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
10005 					 "cannot create default miss resource");
10006 				goto error_default_miss;
10007 			}
10008 			dh->rix_default_fate =  MLX5_FLOW_FATE_DEFAULT_MISS;
10009 			dv->actions[n++] = priv->sh->default_miss.action;
10010 		}
10011 		err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
10012 					       (void *)&dv->value, n,
10013 					       dv->actions, &dh->drv_flow);
10014 		if (err) {
10015 			rte_flow_error_set(error, errno,
10016 					   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10017 					   NULL,
10018 					   "hardware refuses to create flow");
10019 			goto error;
10020 		}
10021 		if (priv->vmwa_context &&
10022 		    dh->vf_vlan.tag && !dh->vf_vlan.created) {
10023 			/*
10024 			 * The rule contains the VLAN pattern.
10025 			 * For VF we are going to create VLAN
10026 			 * interface to make hypervisor set correct
10027 			 * e-Switch vport context.
10028 			 */
10029 			mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
10030 		}
10031 	}
10032 	return 0;
10033 error:
10034 	if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
10035 		flow_dv_default_miss_resource_release(dev);
10036 error_default_miss:
10037 	err = rte_errno; /* Save rte_errno before cleanup. */
10038 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
10039 		       handle_idx, dh, next) {
10040 		/* hrxq is union, don't clear it if the flag is not set. */
10041 		if (dh->rix_hrxq) {
10042 			if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
10043 				mlx5_drop_action_destroy(dev);
10044 				dh->rix_hrxq = 0;
10045 			} else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
10046 				mlx5_hrxq_release(dev, dh->rix_hrxq);
10047 				dh->rix_hrxq = 0;
10048 			}
10049 		}
10050 		if (dh->vf_vlan.tag && dh->vf_vlan.created)
10051 			mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10052 	}
10053 	rte_errno = err; /* Restore rte_errno. */
10054 	return -rte_errno;
10055 }
10056 
10057 /**
10058  * Release the flow matcher.
10059  *
10060  * @param dev
10061  *   Pointer to Ethernet device.
10062  * @param handle
10063  *   Pointer to mlx5_flow_handle.
10064  *
10065  * @return
10066  *   1 while a reference on it exists, 0 when freed.
10067  */
10068 static int
10069 flow_dv_matcher_release(struct rte_eth_dev *dev,
10070 			struct mlx5_flow_handle *handle)
10071 {
10072 	struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
10073 
10074 	MLX5_ASSERT(matcher->matcher_object);
10075 	DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
10076 		dev->data->port_id, (void *)matcher,
10077 		rte_atomic32_read(&matcher->refcnt));
10078 	if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
10079 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10080 			   (matcher->matcher_object));
10081 		LIST_REMOVE(matcher, next);
10082 		/* table ref-- in release interface. */
10083 		flow_dv_tbl_resource_release(dev, matcher->tbl);
10084 		mlx5_free(matcher);
10085 		DRV_LOG(DEBUG, "port %u matcher %p: removed",
10086 			dev->data->port_id, (void *)matcher);
10087 		return 0;
10088 	}
10089 	return 1;
10090 }
10091 
10092 /**
10093  * Release an encap/decap resource.
10094  *
10095  * @param dev
10096  *   Pointer to Ethernet device.
10097  * @param encap_decap_idx
10098  *   Index of encap decap resource.
10099  *
10100  * @return
10101  *   1 while a reference on it exists, 0 when freed.
10102  */
10103 static int
10104 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
10105 				     uint32_t encap_decap_idx)
10106 {
10107 	struct mlx5_priv *priv = dev->data->dev_private;
10108 	uint32_t idx = encap_decap_idx;
10109 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
10110 
10111 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
10112 			 idx);
10113 	if (!cache_resource)
10114 		return 0;
10115 	MLX5_ASSERT(cache_resource->action);
10116 	DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
10117 		(void *)cache_resource,
10118 		rte_atomic32_read(&cache_resource->refcnt));
10119 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10120 		claim_zero(mlx5_flow_os_destroy_flow_action
10121 						(cache_resource->action));
10122 		mlx5_hlist_remove(priv->sh->encaps_decaps,
10123 				  &cache_resource->entry);
10124 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
10125 		DRV_LOG(DEBUG, "encap/decap resource %p: removed",
10126 			(void *)cache_resource);
10127 		return 0;
10128 	}
10129 	return 1;
10130 }
10131 
10132 /**
10133  * Release an jump to table action resource.
10134  *
10135  * @param dev
10136  *   Pointer to Ethernet device.
10137  * @param handle
10138  *   Pointer to mlx5_flow_handle.
10139  *
10140  * @return
10141  *   1 while a reference on it exists, 0 when freed.
10142  */
10143 static int
10144 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
10145 				  struct mlx5_flow_handle *handle)
10146 {
10147 	struct mlx5_priv *priv = dev->data->dev_private;
10148 	struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
10149 	struct mlx5_flow_tbl_data_entry *tbl_data;
10150 
10151 	tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
10152 			     handle->rix_jump);
10153 	if (!tbl_data)
10154 		return 0;
10155 	cache_resource = &tbl_data->jump;
10156 	MLX5_ASSERT(cache_resource->action);
10157 	DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
10158 		(void *)cache_resource,
10159 		rte_atomic32_read(&cache_resource->refcnt));
10160 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10161 		claim_zero(mlx5_flow_os_destroy_flow_action
10162 						(cache_resource->action));
10163 		/* jump action memory free is inside the table release. */
10164 		flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
10165 		DRV_LOG(DEBUG, "jump table resource %p: removed",
10166 			(void *)cache_resource);
10167 		return 0;
10168 	}
10169 	return 1;
10170 }
10171 
10172 /**
10173  * Release a default miss resource.
10174  *
10175  * @param dev
10176  *   Pointer to Ethernet device.
10177  * @return
10178  *   1 while a reference on it exists, 0 when freed.
10179  */
10180 static int
10181 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev)
10182 {
10183 	struct mlx5_priv *priv = dev->data->dev_private;
10184 	struct mlx5_dev_ctx_shared *sh = priv->sh;
10185 	struct mlx5_flow_default_miss_resource *cache_resource =
10186 			&sh->default_miss;
10187 
10188 	MLX5_ASSERT(cache_resource->action);
10189 	DRV_LOG(DEBUG, "default miss resource %p: refcnt %d--",
10190 			(void *)cache_resource->action,
10191 			rte_atomic32_read(&cache_resource->refcnt));
10192 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10193 		claim_zero(mlx5_glue->destroy_flow_action
10194 				(cache_resource->action));
10195 		DRV_LOG(DEBUG, "default miss resource %p: removed",
10196 				(void *)cache_resource->action);
10197 		return 0;
10198 	}
10199 	return 1;
10200 }
10201 
10202 /**
10203  * Release a modify-header resource.
10204  *
10205  * @param dev
10206  *   Pointer to Ethernet device.
10207  * @param handle
10208  *   Pointer to mlx5_flow_handle.
10209  *
10210  * @return
10211  *   1 while a reference on it exists, 0 when freed.
10212  */
10213 static int
10214 flow_dv_modify_hdr_resource_release(struct rte_eth_dev *dev,
10215 				    struct mlx5_flow_handle *handle)
10216 {
10217 	struct mlx5_priv *priv = dev->data->dev_private;
10218 	struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
10219 							handle->dvh.modify_hdr;
10220 
10221 	MLX5_ASSERT(cache_resource->action);
10222 	DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
10223 		(void *)cache_resource,
10224 		rte_atomic32_read(&cache_resource->refcnt));
10225 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10226 		claim_zero(mlx5_flow_os_destroy_flow_action
10227 						(cache_resource->action));
10228 		mlx5_hlist_remove(priv->sh->modify_cmds,
10229 				  &cache_resource->entry);
10230 		mlx5_free(cache_resource);
10231 		DRV_LOG(DEBUG, "modify-header resource %p: removed",
10232 			(void *)cache_resource);
10233 		return 0;
10234 	}
10235 	return 1;
10236 }
10237 
10238 /**
10239  * Release port ID action resource.
10240  *
10241  * @param dev
10242  *   Pointer to Ethernet device.
10243  * @param handle
10244  *   Pointer to mlx5_flow_handle.
10245  *
10246  * @return
10247  *   1 while a reference on it exists, 0 when freed.
10248  */
10249 static int
10250 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
10251 					uint32_t port_id)
10252 {
10253 	struct mlx5_priv *priv = dev->data->dev_private;
10254 	struct mlx5_flow_dv_port_id_action_resource *cache_resource;
10255 	uint32_t idx = port_id;
10256 
10257 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10258 					idx);
10259 	if (!cache_resource)
10260 		return 0;
10261 	MLX5_ASSERT(cache_resource->action);
10262 	DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
10263 		(void *)cache_resource,
10264 		rte_atomic32_read(&cache_resource->refcnt));
10265 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10266 		claim_zero(mlx5_flow_os_destroy_flow_action
10267 						(cache_resource->action));
10268 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
10269 			     &priv->sh->port_id_action_list, idx,
10270 			     cache_resource, next);
10271 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
10272 		DRV_LOG(DEBUG, "port id action resource %p: removed",
10273 			(void *)cache_resource);
10274 		return 0;
10275 	}
10276 	return 1;
10277 }
10278 
10279 /**
10280  * Release push vlan action resource.
10281  *
10282  * @param dev
10283  *   Pointer to Ethernet device.
10284  * @param handle
10285  *   Pointer to mlx5_flow_handle.
10286  *
10287  * @return
10288  *   1 while a reference on it exists, 0 when freed.
10289  */
10290 static int
10291 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
10292 					  struct mlx5_flow_handle *handle)
10293 {
10294 	struct mlx5_priv *priv = dev->data->dev_private;
10295 	uint32_t idx = handle->dvh.rix_push_vlan;
10296 	struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
10297 
10298 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10299 					idx);
10300 	if (!cache_resource)
10301 		return 0;
10302 	MLX5_ASSERT(cache_resource->action);
10303 	DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
10304 		(void *)cache_resource,
10305 		rte_atomic32_read(&cache_resource->refcnt));
10306 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
10307 		claim_zero(mlx5_flow_os_destroy_flow_action
10308 						(cache_resource->action));
10309 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
10310 			     &priv->sh->push_vlan_action_list, idx,
10311 			     cache_resource, next);
10312 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
10313 		DRV_LOG(DEBUG, "push vlan action resource %p: removed",
10314 			(void *)cache_resource);
10315 		return 0;
10316 	}
10317 	return 1;
10318 }
10319 
10320 /**
10321  * Release the fate resource.
10322  *
10323  * @param dev
10324  *   Pointer to Ethernet device.
10325  * @param handle
10326  *   Pointer to mlx5_flow_handle.
10327  */
10328 static void
10329 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
10330 			       struct mlx5_flow_handle *handle)
10331 {
10332 	if (!handle->rix_fate)
10333 		return;
10334 	switch (handle->fate_action) {
10335 	case MLX5_FLOW_FATE_DROP:
10336 		mlx5_drop_action_destroy(dev);
10337 		break;
10338 	case MLX5_FLOW_FATE_QUEUE:
10339 		mlx5_hrxq_release(dev, handle->rix_hrxq);
10340 		break;
10341 	case MLX5_FLOW_FATE_JUMP:
10342 		flow_dv_jump_tbl_resource_release(dev, handle);
10343 		break;
10344 	case MLX5_FLOW_FATE_PORT_ID:
10345 		flow_dv_port_id_action_resource_release(dev,
10346 				handle->rix_port_id_action);
10347 		break;
10348 	case MLX5_FLOW_FATE_DEFAULT_MISS:
10349 		flow_dv_default_miss_resource_release(dev);
10350 		break;
10351 	default:
10352 		DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
10353 		break;
10354 	}
10355 	handle->rix_fate = 0;
10356 }
10357 
10358 /**
10359  * Release an sample resource.
10360  *
10361  * @param dev
10362  *   Pointer to Ethernet device.
10363  * @param handle
10364  *   Pointer to mlx5_flow_handle.
10365  *
10366  * @return
10367  *   1 while a reference on it exists, 0 when freed.
10368  */
10369 static int
10370 flow_dv_sample_resource_release(struct rte_eth_dev *dev,
10371 				     struct mlx5_flow_handle *handle)
10372 {
10373 	struct mlx5_priv *priv = dev->data->dev_private;
10374 	uint32_t idx = handle->dvh.rix_sample;
10375 	struct mlx5_flow_dv_sample_resource *cache_resource;
10376 
10377 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10378 			 idx);
10379 	if (!cache_resource)
10380 		return 0;
10381 	MLX5_ASSERT(cache_resource->verbs_action);
10382 	DRV_LOG(DEBUG, "sample resource %p: refcnt %d--",
10383 		(void *)cache_resource,
10384 		__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10385 	if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10386 			       __ATOMIC_RELAXED) == 0) {
10387 		if (cache_resource->verbs_action)
10388 			claim_zero(mlx5_glue->destroy_flow_action
10389 					(cache_resource->verbs_action));
10390 		if (cache_resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB) {
10391 			if (cache_resource->default_miss)
10392 				claim_zero(mlx5_glue->destroy_flow_action
10393 				  (cache_resource->default_miss));
10394 		}
10395 		if (cache_resource->normal_path_tbl)
10396 			flow_dv_tbl_resource_release(dev,
10397 				cache_resource->normal_path_tbl);
10398 	}
10399 	if (cache_resource->sample_idx.rix_hrxq &&
10400 		!mlx5_hrxq_release(dev,
10401 			cache_resource->sample_idx.rix_hrxq))
10402 		cache_resource->sample_idx.rix_hrxq = 0;
10403 	if (cache_resource->sample_idx.rix_tag &&
10404 		!flow_dv_tag_release(dev,
10405 			cache_resource->sample_idx.rix_tag))
10406 		cache_resource->sample_idx.rix_tag = 0;
10407 	if (cache_resource->sample_idx.cnt) {
10408 		flow_dv_counter_release(dev,
10409 			cache_resource->sample_idx.cnt);
10410 		cache_resource->sample_idx.cnt = 0;
10411 	}
10412 	if (!__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED)) {
10413 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_SAMPLE],
10414 			     &priv->sh->sample_action_list, idx,
10415 			     cache_resource, next);
10416 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_SAMPLE], idx);
10417 		DRV_LOG(DEBUG, "sample resource %p: removed",
10418 			(void *)cache_resource);
10419 		return 0;
10420 	}
10421 	return 1;
10422 }
10423 
10424 /**
10425  * Release an destination array resource.
10426  *
10427  * @param dev
10428  *   Pointer to Ethernet device.
10429  * @param handle
10430  *   Pointer to mlx5_flow_handle.
10431  *
10432  * @return
10433  *   1 while a reference on it exists, 0 when freed.
10434  */
10435 static int
10436 flow_dv_dest_array_resource_release(struct rte_eth_dev *dev,
10437 				     struct mlx5_flow_handle *handle)
10438 {
10439 	struct mlx5_priv *priv = dev->data->dev_private;
10440 	struct mlx5_flow_dv_dest_array_resource *cache_resource;
10441 	struct mlx5_flow_sub_actions_idx *mdest_act_res;
10442 	uint32_t idx = handle->dvh.rix_dest_array;
10443 	uint32_t i = 0;
10444 
10445 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10446 			 idx);
10447 	if (!cache_resource)
10448 		return 0;
10449 	MLX5_ASSERT(cache_resource->action);
10450 	DRV_LOG(DEBUG, "destination array resource %p: refcnt %d--",
10451 		(void *)cache_resource,
10452 		__atomic_load_n(&cache_resource->refcnt, __ATOMIC_RELAXED));
10453 	if (__atomic_sub_fetch(&cache_resource->refcnt, 1,
10454 			       __ATOMIC_RELAXED) == 0) {
10455 		if (cache_resource->action)
10456 			claim_zero(mlx5_glue->destroy_flow_action
10457 						(cache_resource->action));
10458 		for (; i < cache_resource->num_of_dest; i++) {
10459 			mdest_act_res = &cache_resource->sample_idx[i];
10460 			if (mdest_act_res->rix_hrxq) {
10461 				mlx5_hrxq_release(dev,
10462 					mdest_act_res->rix_hrxq);
10463 				mdest_act_res->rix_hrxq = 0;
10464 			}
10465 			if (mdest_act_res->rix_encap_decap) {
10466 				flow_dv_encap_decap_resource_release(dev,
10467 					mdest_act_res->rix_encap_decap);
10468 				mdest_act_res->rix_encap_decap = 0;
10469 			}
10470 			if (mdest_act_res->rix_port_id_action) {
10471 				flow_dv_port_id_action_resource_release(dev,
10472 					mdest_act_res->rix_port_id_action);
10473 				mdest_act_res->rix_port_id_action = 0;
10474 			}
10475 			if (mdest_act_res->rix_tag) {
10476 				flow_dv_tag_release(dev,
10477 					mdest_act_res->rix_tag);
10478 				mdest_act_res->rix_tag = 0;
10479 			}
10480 		}
10481 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY],
10482 			     &priv->sh->dest_array_list, idx,
10483 			     cache_resource, next);
10484 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DEST_ARRAY], idx);
10485 		DRV_LOG(DEBUG, "destination array resource %p: removed",
10486 			(void *)cache_resource);
10487 		return 0;
10488 	}
10489 	return 1;
10490 }
10491 
10492 /**
10493  * Remove the flow from the NIC but keeps it in memory.
10494  * Lock free, (mutex should be acquired by caller).
10495  *
10496  * @param[in] dev
10497  *   Pointer to Ethernet device.
10498  * @param[in, out] flow
10499  *   Pointer to flow structure.
10500  */
10501 static void
10502 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10503 {
10504 	struct mlx5_flow_handle *dh;
10505 	uint32_t handle_idx;
10506 	struct mlx5_priv *priv = dev->data->dev_private;
10507 
10508 	if (!flow)
10509 		return;
10510 	handle_idx = flow->dev_handles;
10511 	while (handle_idx) {
10512 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10513 				    handle_idx);
10514 		if (!dh)
10515 			return;
10516 		if (dh->drv_flow) {
10517 			claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
10518 			dh->drv_flow = NULL;
10519 		}
10520 		if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
10521 		    dh->fate_action == MLX5_FLOW_FATE_QUEUE ||
10522 		    dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
10523 			flow_dv_fate_resource_release(dev, dh);
10524 		if (dh->vf_vlan.tag && dh->vf_vlan.created)
10525 			mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
10526 		handle_idx = dh->next.next;
10527 	}
10528 }
10529 
10530 /**
10531  * Remove the flow from the NIC and the memory.
10532  * Lock free, (mutex should be acquired by caller).
10533  *
10534  * @param[in] dev
10535  *   Pointer to the Ethernet device structure.
10536  * @param[in, out] flow
10537  *   Pointer to flow structure.
10538  */
10539 static void
10540 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10541 {
10542 	struct mlx5_flow_handle *dev_handle;
10543 	struct mlx5_priv *priv = dev->data->dev_private;
10544 
10545 	if (!flow)
10546 		return;
10547 	__flow_dv_remove(dev, flow);
10548 	if (flow->counter) {
10549 		flow_dv_counter_release(dev, flow->counter);
10550 		flow->counter = 0;
10551 	}
10552 	if (flow->meter) {
10553 		struct mlx5_flow_meter *fm;
10554 
10555 		fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
10556 				    flow->meter);
10557 		if (fm)
10558 			mlx5_flow_meter_detach(fm);
10559 		flow->meter = 0;
10560 	}
10561 	while (flow->dev_handles) {
10562 		uint32_t tmp_idx = flow->dev_handles;
10563 
10564 		dev_handle = mlx5_ipool_get(priv->sh->ipool
10565 					    [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
10566 		if (!dev_handle)
10567 			return;
10568 		flow->dev_handles = dev_handle->next.next;
10569 		if (dev_handle->dvh.matcher)
10570 			flow_dv_matcher_release(dev, dev_handle);
10571 		if (dev_handle->dvh.rix_sample)
10572 			flow_dv_sample_resource_release(dev, dev_handle);
10573 		if (dev_handle->dvh.rix_dest_array)
10574 			flow_dv_dest_array_resource_release(dev, dev_handle);
10575 		if (dev_handle->dvh.rix_encap_decap)
10576 			flow_dv_encap_decap_resource_release(dev,
10577 				dev_handle->dvh.rix_encap_decap);
10578 		if (dev_handle->dvh.modify_hdr)
10579 			flow_dv_modify_hdr_resource_release(dev, dev_handle);
10580 		if (dev_handle->dvh.rix_push_vlan)
10581 			flow_dv_push_vlan_action_resource_release(dev,
10582 								  dev_handle);
10583 		if (dev_handle->dvh.rix_tag)
10584 			flow_dv_tag_release(dev,
10585 					    dev_handle->dvh.rix_tag);
10586 		flow_dv_fate_resource_release(dev, dev_handle);
10587 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
10588 			   tmp_idx);
10589 	}
10590 }
10591 
10592 /**
10593  * Query a dv flow  rule for its statistics via devx.
10594  *
10595  * @param[in] dev
10596  *   Pointer to Ethernet device.
10597  * @param[in] flow
10598  *   Pointer to the sub flow.
10599  * @param[out] data
10600  *   data retrieved by the query.
10601  * @param[out] error
10602  *   Perform verbose error reporting if not NULL.
10603  *
10604  * @return
10605  *   0 on success, a negative errno value otherwise and rte_errno is set.
10606  */
10607 static int
10608 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
10609 		    void *data, struct rte_flow_error *error)
10610 {
10611 	struct mlx5_priv *priv = dev->data->dev_private;
10612 	struct rte_flow_query_count *qc = data;
10613 
10614 	if (!priv->config.devx)
10615 		return rte_flow_error_set(error, ENOTSUP,
10616 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10617 					  NULL,
10618 					  "counters are not supported");
10619 	if (flow->counter) {
10620 		uint64_t pkts, bytes;
10621 		struct mlx5_flow_counter *cnt;
10622 
10623 		cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
10624 						 NULL);
10625 		int err = _flow_dv_query_count(dev, flow->counter, &pkts,
10626 					       &bytes);
10627 
10628 		if (err)
10629 			return rte_flow_error_set(error, -err,
10630 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10631 					NULL, "cannot read counters");
10632 		qc->hits_set = 1;
10633 		qc->bytes_set = 1;
10634 		qc->hits = pkts - cnt->hits;
10635 		qc->bytes = bytes - cnt->bytes;
10636 		if (qc->reset) {
10637 			cnt->hits = pkts;
10638 			cnt->bytes = bytes;
10639 		}
10640 		return 0;
10641 	}
10642 	return rte_flow_error_set(error, EINVAL,
10643 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
10644 				  NULL,
10645 				  "counters are not available");
10646 }
10647 
10648 /**
10649  * Query a flow.
10650  *
10651  * @see rte_flow_query()
10652  * @see rte_flow_ops
10653  */
10654 static int
10655 flow_dv_query(struct rte_eth_dev *dev,
10656 	      struct rte_flow *flow __rte_unused,
10657 	      const struct rte_flow_action *actions __rte_unused,
10658 	      void *data __rte_unused,
10659 	      struct rte_flow_error *error __rte_unused)
10660 {
10661 	int ret = -EINVAL;
10662 
10663 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
10664 		switch (actions->type) {
10665 		case RTE_FLOW_ACTION_TYPE_VOID:
10666 			break;
10667 		case RTE_FLOW_ACTION_TYPE_COUNT:
10668 			ret = flow_dv_query_count(dev, flow, data, error);
10669 			break;
10670 		default:
10671 			return rte_flow_error_set(error, ENOTSUP,
10672 						  RTE_FLOW_ERROR_TYPE_ACTION,
10673 						  actions,
10674 						  "action not supported");
10675 		}
10676 	}
10677 	return ret;
10678 }
10679 
10680 /**
10681  * Destroy the meter table set.
10682  * Lock free, (mutex should be acquired by caller).
10683  *
10684  * @param[in] dev
10685  *   Pointer to Ethernet device.
10686  * @param[in] tbl
10687  *   Pointer to the meter table set.
10688  *
10689  * @return
10690  *   Always 0.
10691  */
10692 static int
10693 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
10694 			struct mlx5_meter_domains_infos *tbl)
10695 {
10696 	struct mlx5_priv *priv = dev->data->dev_private;
10697 	struct mlx5_meter_domains_infos *mtd =
10698 				(struct mlx5_meter_domains_infos *)tbl;
10699 
10700 	if (!mtd || !priv->config.dv_flow_en)
10701 		return 0;
10702 	if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
10703 		claim_zero(mlx5_flow_os_destroy_flow
10704 			   (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
10705 	if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
10706 		claim_zero(mlx5_flow_os_destroy_flow
10707 			   (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
10708 	if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
10709 		claim_zero(mlx5_flow_os_destroy_flow
10710 			   (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
10711 	if (mtd->egress.color_matcher)
10712 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10713 			   (mtd->egress.color_matcher));
10714 	if (mtd->egress.any_matcher)
10715 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10716 			   (mtd->egress.any_matcher));
10717 	if (mtd->egress.tbl)
10718 		flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
10719 	if (mtd->egress.sfx_tbl)
10720 		flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
10721 	if (mtd->ingress.color_matcher)
10722 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10723 			   (mtd->ingress.color_matcher));
10724 	if (mtd->ingress.any_matcher)
10725 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10726 			   (mtd->ingress.any_matcher));
10727 	if (mtd->ingress.tbl)
10728 		flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
10729 	if (mtd->ingress.sfx_tbl)
10730 		flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
10731 	if (mtd->transfer.color_matcher)
10732 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10733 			   (mtd->transfer.color_matcher));
10734 	if (mtd->transfer.any_matcher)
10735 		claim_zero(mlx5_flow_os_destroy_flow_matcher
10736 			   (mtd->transfer.any_matcher));
10737 	if (mtd->transfer.tbl)
10738 		flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
10739 	if (mtd->transfer.sfx_tbl)
10740 		flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
10741 	if (mtd->drop_actn)
10742 		claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
10743 	mlx5_free(mtd);
10744 	return 0;
10745 }
10746 
10747 /* Number of meter flow actions, count and jump or count and drop. */
10748 #define METER_ACTIONS 2
10749 
10750 /**
10751  * Create specify domain meter table and suffix table.
10752  *
10753  * @param[in] dev
10754  *   Pointer to Ethernet device.
10755  * @param[in,out] mtb
10756  *   Pointer to DV meter table set.
10757  * @param[in] egress
10758  *   Table attribute.
10759  * @param[in] transfer
10760  *   Table attribute.
10761  * @param[in] color_reg_c_idx
10762  *   Reg C index for color match.
10763  *
10764  * @return
10765  *   0 on success, -1 otherwise and rte_errno is set.
10766  */
10767 static int
10768 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
10769 			   struct mlx5_meter_domains_infos *mtb,
10770 			   uint8_t egress, uint8_t transfer,
10771 			   uint32_t color_reg_c_idx)
10772 {
10773 	struct mlx5_priv *priv = dev->data->dev_private;
10774 	struct mlx5_dev_ctx_shared *sh = priv->sh;
10775 	struct mlx5_flow_dv_match_params mask = {
10776 		.size = sizeof(mask.buf),
10777 	};
10778 	struct mlx5_flow_dv_match_params value = {
10779 		.size = sizeof(value.buf),
10780 	};
10781 	struct mlx5dv_flow_matcher_attr dv_attr = {
10782 		.type = IBV_FLOW_ATTR_NORMAL,
10783 		.priority = 0,
10784 		.match_criteria_enable = 0,
10785 		.match_mask = (void *)&mask,
10786 	};
10787 	void *actions[METER_ACTIONS];
10788 	struct mlx5_meter_domain_info *dtb;
10789 	struct rte_flow_error error;
10790 	int i = 0;
10791 	int ret;
10792 
10793 	if (transfer)
10794 		dtb = &mtb->transfer;
10795 	else if (egress)
10796 		dtb = &mtb->egress;
10797 	else
10798 		dtb = &mtb->ingress;
10799 	/* Create the meter table with METER level. */
10800 	dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
10801 					    egress, transfer, &error);
10802 	if (!dtb->tbl) {
10803 		DRV_LOG(ERR, "Failed to create meter policer table.");
10804 		return -1;
10805 	}
10806 	/* Create the meter suffix table with SUFFIX level. */
10807 	dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
10808 					    MLX5_FLOW_TABLE_LEVEL_SUFFIX,
10809 					    egress, transfer, &error);
10810 	if (!dtb->sfx_tbl) {
10811 		DRV_LOG(ERR, "Failed to create meter suffix table.");
10812 		return -1;
10813 	}
10814 	/* Create matchers, Any and Color. */
10815 	dv_attr.priority = 3;
10816 	dv_attr.match_criteria_enable = 0;
10817 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
10818 					       &dtb->any_matcher);
10819 	if (ret) {
10820 		DRV_LOG(ERR, "Failed to create meter"
10821 			     " policer default matcher.");
10822 		goto error_exit;
10823 	}
10824 	dv_attr.priority = 0;
10825 	dv_attr.match_criteria_enable =
10826 				1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
10827 	flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
10828 			       rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
10829 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
10830 					       &dtb->color_matcher);
10831 	if (ret) {
10832 		DRV_LOG(ERR, "Failed to create meter policer color matcher.");
10833 		goto error_exit;
10834 	}
10835 	if (mtb->count_actns[RTE_MTR_DROPPED])
10836 		actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
10837 	actions[i++] = mtb->drop_actn;
10838 	/* Default rule: lowest priority, match any, actions: drop. */
10839 	ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
10840 				       actions,
10841 				       &dtb->policer_rules[RTE_MTR_DROPPED]);
10842 	if (ret) {
10843 		DRV_LOG(ERR, "Failed to create meter policer drop rule.");
10844 		goto error_exit;
10845 	}
10846 	return 0;
10847 error_exit:
10848 	return -1;
10849 }
10850 
10851 /**
10852  * Create the needed meter and suffix tables.
10853  * Lock free, (mutex should be acquired by caller).
10854  *
10855  * @param[in] dev
10856  *   Pointer to Ethernet device.
10857  * @param[in] fm
10858  *   Pointer to the flow meter.
10859  *
10860  * @return
10861  *   Pointer to table set on success, NULL otherwise and rte_errno is set.
10862  */
10863 static struct mlx5_meter_domains_infos *
10864 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
10865 		       const struct mlx5_flow_meter *fm)
10866 {
10867 	struct mlx5_priv *priv = dev->data->dev_private;
10868 	struct mlx5_meter_domains_infos *mtb;
10869 	int ret;
10870 	int i;
10871 
10872 	if (!priv->mtr_en) {
10873 		rte_errno = ENOTSUP;
10874 		return NULL;
10875 	}
10876 	mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
10877 	if (!mtb) {
10878 		DRV_LOG(ERR, "Failed to allocate memory for meter.");
10879 		return NULL;
10880 	}
10881 	/* Create meter count actions */
10882 	for (i = 0; i <= RTE_MTR_DROPPED; i++) {
10883 		struct mlx5_flow_counter *cnt;
10884 		if (!fm->policer_stats.cnt[i])
10885 			continue;
10886 		cnt = flow_dv_counter_get_by_idx(dev,
10887 		      fm->policer_stats.cnt[i], NULL);
10888 		mtb->count_actns[i] = cnt->action;
10889 	}
10890 	/* Create drop action. */
10891 	ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
10892 	if (ret) {
10893 		DRV_LOG(ERR, "Failed to create drop action.");
10894 		goto error_exit;
10895 	}
10896 	/* Egress meter table. */
10897 	ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
10898 	if (ret) {
10899 		DRV_LOG(ERR, "Failed to prepare egress meter table.");
10900 		goto error_exit;
10901 	}
10902 	/* Ingress meter table. */
10903 	ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
10904 	if (ret) {
10905 		DRV_LOG(ERR, "Failed to prepare ingress meter table.");
10906 		goto error_exit;
10907 	}
10908 	/* FDB meter table. */
10909 	if (priv->config.dv_esw_en) {
10910 		ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
10911 						 priv->mtr_color_reg);
10912 		if (ret) {
10913 			DRV_LOG(ERR, "Failed to prepare fdb meter table.");
10914 			goto error_exit;
10915 		}
10916 	}
10917 	return mtb;
10918 error_exit:
10919 	flow_dv_destroy_mtr_tbl(dev, mtb);
10920 	return NULL;
10921 }
10922 
10923 /**
10924  * Destroy domain policer rule.
10925  *
10926  * @param[in] dt
10927  *   Pointer to domain table.
10928  */
10929 static void
10930 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
10931 {
10932 	int i;
10933 
10934 	for (i = 0; i < RTE_MTR_DROPPED; i++) {
10935 		if (dt->policer_rules[i]) {
10936 			claim_zero(mlx5_flow_os_destroy_flow
10937 				   (dt->policer_rules[i]));
10938 			dt->policer_rules[i] = NULL;
10939 		}
10940 	}
10941 	if (dt->jump_actn) {
10942 		claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
10943 		dt->jump_actn = NULL;
10944 	}
10945 }
10946 
10947 /**
10948  * Destroy policer rules.
10949  *
10950  * @param[in] dev
10951  *   Pointer to Ethernet device.
10952  * @param[in] fm
10953  *   Pointer to flow meter structure.
10954  * @param[in] attr
10955  *   Pointer to flow attributes.
10956  *
10957  * @return
10958  *   Always 0.
10959  */
10960 static int
10961 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
10962 			      const struct mlx5_flow_meter *fm,
10963 			      const struct rte_flow_attr *attr)
10964 {
10965 	struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
10966 
10967 	if (!mtb)
10968 		return 0;
10969 	if (attr->egress)
10970 		flow_dv_destroy_domain_policer_rule(&mtb->egress);
10971 	if (attr->ingress)
10972 		flow_dv_destroy_domain_policer_rule(&mtb->ingress);
10973 	if (attr->transfer)
10974 		flow_dv_destroy_domain_policer_rule(&mtb->transfer);
10975 	return 0;
10976 }
10977 
10978 /**
10979  * Create specify domain meter policer rule.
10980  *
10981  * @param[in] fm
10982  *   Pointer to flow meter structure.
10983  * @param[in] mtb
10984  *   Pointer to DV meter table set.
10985  * @param[in] mtr_reg_c
10986  *   Color match REG_C.
10987  *
10988  * @return
10989  *   0 on success, -1 otherwise.
10990  */
10991 static int
10992 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
10993 				    struct mlx5_meter_domain_info *dtb,
10994 				    uint8_t mtr_reg_c)
10995 {
10996 	struct mlx5_flow_dv_match_params matcher = {
10997 		.size = sizeof(matcher.buf),
10998 	};
10999 	struct mlx5_flow_dv_match_params value = {
11000 		.size = sizeof(value.buf),
11001 	};
11002 	struct mlx5_meter_domains_infos *mtb = fm->mfts;
11003 	void *actions[METER_ACTIONS];
11004 	int i;
11005 	int ret = 0;
11006 
11007 	/* Create jump action. */
11008 	if (!dtb->jump_actn)
11009 		ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
11010 				(dtb->sfx_tbl->obj, &dtb->jump_actn);
11011 	if (ret) {
11012 		DRV_LOG(ERR, "Failed to create policer jump action.");
11013 		goto error;
11014 	}
11015 	for (i = 0; i < RTE_MTR_DROPPED; i++) {
11016 		int j = 0;
11017 
11018 		flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
11019 				       rte_col_2_mlx5_col(i), UINT8_MAX);
11020 		if (mtb->count_actns[i])
11021 			actions[j++] = mtb->count_actns[i];
11022 		if (fm->action[i] == MTR_POLICER_ACTION_DROP)
11023 			actions[j++] = mtb->drop_actn;
11024 		else
11025 			actions[j++] = dtb->jump_actn;
11026 		ret = mlx5_flow_os_create_flow(dtb->color_matcher,
11027 					       (void *)&value, j, actions,
11028 					       &dtb->policer_rules[i]);
11029 		if (ret) {
11030 			DRV_LOG(ERR, "Failed to create policer rule.");
11031 			goto error;
11032 		}
11033 	}
11034 	return 0;
11035 error:
11036 	rte_errno = errno;
11037 	return -1;
11038 }
11039 
11040 /**
11041  * Create policer rules.
11042  *
11043  * @param[in] dev
11044  *   Pointer to Ethernet device.
11045  * @param[in] fm
11046  *   Pointer to flow meter structure.
11047  * @param[in] attr
11048  *   Pointer to flow attributes.
11049  *
11050  * @return
11051  *   0 on success, -1 otherwise.
11052  */
11053 static int
11054 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
11055 			     struct mlx5_flow_meter *fm,
11056 			     const struct rte_flow_attr *attr)
11057 {
11058 	struct mlx5_priv *priv = dev->data->dev_private;
11059 	struct mlx5_meter_domains_infos *mtb = fm->mfts;
11060 	int ret;
11061 
11062 	if (attr->egress) {
11063 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
11064 						priv->mtr_color_reg);
11065 		if (ret) {
11066 			DRV_LOG(ERR, "Failed to create egress policer.");
11067 			goto error;
11068 		}
11069 	}
11070 	if (attr->ingress) {
11071 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
11072 						priv->mtr_color_reg);
11073 		if (ret) {
11074 			DRV_LOG(ERR, "Failed to create ingress policer.");
11075 			goto error;
11076 		}
11077 	}
11078 	if (attr->transfer) {
11079 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
11080 						priv->mtr_color_reg);
11081 		if (ret) {
11082 			DRV_LOG(ERR, "Failed to create transfer policer.");
11083 			goto error;
11084 		}
11085 	}
11086 	return 0;
11087 error:
11088 	flow_dv_destroy_policer_rules(dev, fm, attr);
11089 	return -1;
11090 }
11091 
11092 /**
11093  * Query a devx counter.
11094  *
11095  * @param[in] dev
11096  *   Pointer to the Ethernet device structure.
11097  * @param[in] cnt
11098  *   Index to the flow counter.
11099  * @param[in] clear
11100  *   Set to clear the counter statistics.
11101  * @param[out] pkts
11102  *   The statistics value of packets.
11103  * @param[out] bytes
11104  *   The statistics value of bytes.
11105  *
11106  * @return
11107  *   0 on success, otherwise return -1.
11108  */
11109 static int
11110 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
11111 		      uint64_t *pkts, uint64_t *bytes)
11112 {
11113 	struct mlx5_priv *priv = dev->data->dev_private;
11114 	struct mlx5_flow_counter *cnt;
11115 	uint64_t inn_pkts, inn_bytes;
11116 	int ret;
11117 
11118 	if (!priv->config.devx)
11119 		return -1;
11120 
11121 	ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
11122 	if (ret)
11123 		return -1;
11124 	cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
11125 	*pkts = inn_pkts - cnt->hits;
11126 	*bytes = inn_bytes - cnt->bytes;
11127 	if (clear) {
11128 		cnt->hits = inn_pkts;
11129 		cnt->bytes = inn_bytes;
11130 	}
11131 	return 0;
11132 }
11133 
11134 /**
11135  * Get aged-out flows.
11136  *
11137  * @param[in] dev
11138  *   Pointer to the Ethernet device structure.
11139  * @param[in] context
11140  *   The address of an array of pointers to the aged-out flows contexts.
11141  * @param[in] nb_contexts
11142  *   The length of context array pointers.
11143  * @param[out] error
11144  *   Perform verbose error reporting if not NULL. Initialized in case of
11145  *   error only.
11146  *
11147  * @return
11148  *   how many contexts get in success, otherwise negative errno value.
11149  *   if nb_contexts is 0, return the amount of all aged contexts.
11150  *   if nb_contexts is not 0 , return the amount of aged flows reported
11151  *   in the context array.
11152  * @note: only stub for now
11153  */
11154 static int
11155 flow_get_aged_flows(struct rte_eth_dev *dev,
11156 		    void **context,
11157 		    uint32_t nb_contexts,
11158 		    struct rte_flow_error *error)
11159 {
11160 	struct mlx5_priv *priv = dev->data->dev_private;
11161 	struct mlx5_age_info *age_info;
11162 	struct mlx5_age_param *age_param;
11163 	struct mlx5_flow_counter *counter;
11164 	int nb_flows = 0;
11165 
11166 	if (nb_contexts && !context)
11167 		return rte_flow_error_set(error, EINVAL,
11168 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
11169 					  NULL,
11170 					  "Should assign at least one flow or"
11171 					  " context to get if nb_contexts != 0");
11172 	age_info = GET_PORT_AGE_INFO(priv);
11173 	rte_spinlock_lock(&age_info->aged_sl);
11174 	TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
11175 		nb_flows++;
11176 		if (nb_contexts) {
11177 			age_param = MLX5_CNT_TO_AGE(counter);
11178 			context[nb_flows - 1] = age_param->context;
11179 			if (!(--nb_contexts))
11180 				break;
11181 		}
11182 	}
11183 	rte_spinlock_unlock(&age_info->aged_sl);
11184 	MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
11185 	return nb_flows;
11186 }
11187 
11188 /*
11189  * Mutex-protected thunk to lock-free  __flow_dv_translate().
11190  */
11191 static int
11192 flow_dv_translate(struct rte_eth_dev *dev,
11193 		  struct mlx5_flow *dev_flow,
11194 		  const struct rte_flow_attr *attr,
11195 		  const struct rte_flow_item items[],
11196 		  const struct rte_flow_action actions[],
11197 		  struct rte_flow_error *error)
11198 {
11199 	int ret;
11200 
11201 	flow_dv_shared_lock(dev);
11202 	ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
11203 	flow_dv_shared_unlock(dev);
11204 	return ret;
11205 }
11206 
11207 /*
11208  * Mutex-protected thunk to lock-free  __flow_dv_apply().
11209  */
11210 static int
11211 flow_dv_apply(struct rte_eth_dev *dev,
11212 	      struct rte_flow *flow,
11213 	      struct rte_flow_error *error)
11214 {
11215 	int ret;
11216 
11217 	flow_dv_shared_lock(dev);
11218 	ret = __flow_dv_apply(dev, flow, error);
11219 	flow_dv_shared_unlock(dev);
11220 	return ret;
11221 }
11222 
11223 /*
11224  * Mutex-protected thunk to lock-free __flow_dv_remove().
11225  */
11226 static void
11227 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
11228 {
11229 	flow_dv_shared_lock(dev);
11230 	__flow_dv_remove(dev, flow);
11231 	flow_dv_shared_unlock(dev);
11232 }
11233 
11234 /*
11235  * Mutex-protected thunk to lock-free __flow_dv_destroy().
11236  */
11237 static void
11238 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
11239 {
11240 	flow_dv_shared_lock(dev);
11241 	__flow_dv_destroy(dev, flow);
11242 	flow_dv_shared_unlock(dev);
11243 }
11244 
11245 /*
11246  * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
11247  */
11248 static uint32_t
11249 flow_dv_counter_allocate(struct rte_eth_dev *dev)
11250 {
11251 	uint32_t cnt;
11252 
11253 	flow_dv_shared_lock(dev);
11254 	cnt = flow_dv_counter_alloc(dev, 0, 0, 1, 0);
11255 	flow_dv_shared_unlock(dev);
11256 	return cnt;
11257 }
11258 
11259 /*
11260  * Mutex-protected thunk to lock-free flow_dv_counter_release().
11261  */
11262 static void
11263 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
11264 {
11265 	flow_dv_shared_lock(dev);
11266 	flow_dv_counter_release(dev, cnt);
11267 	flow_dv_shared_unlock(dev);
11268 }
11269 
11270 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
11271 	.validate = flow_dv_validate,
11272 	.prepare = flow_dv_prepare,
11273 	.translate = flow_dv_translate,
11274 	.apply = flow_dv_apply,
11275 	.remove = flow_dv_remove,
11276 	.destroy = flow_dv_destroy,
11277 	.query = flow_dv_query,
11278 	.create_mtr_tbls = flow_dv_create_mtr_tbl,
11279 	.destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
11280 	.create_policer_rules = flow_dv_create_policer_rules,
11281 	.destroy_policer_rules = flow_dv_destroy_policer_rules,
11282 	.counter_alloc = flow_dv_counter_allocate,
11283 	.counter_free = flow_dv_counter_free,
11284 	.counter_query = flow_dv_counter_query,
11285 	.get_aged_flows = flow_get_aged_flows,
11286 };
11287 
11288 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
11289 
11290