xref: /dpdk/drivers/net/mlx5/mlx5_flow_dv.c (revision 3cc6ecfdfe85d2577fef30e1791bb7534e3d60b3)
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 /**
80  * Initialize flow attributes structure according to flow items' types.
81  *
82  * flow_dv_validate() avoids multiple L3/L4 layers cases other than tunnel
83  * mode. For tunnel mode, the items to be modified are the outermost ones.
84  *
85  * @param[in] item
86  *   Pointer to item specification.
87  * @param[out] attr
88  *   Pointer to flow attributes structure.
89  * @param[in] dev_flow
90  *   Pointer to the sub flow.
91  * @param[in] tunnel_decap
92  *   Whether action is after tunnel decapsulation.
93  */
94 static void
95 flow_dv_attr_init(const struct rte_flow_item *item, union flow_dv_attr *attr,
96 		  struct mlx5_flow *dev_flow, bool tunnel_decap)
97 {
98 	uint64_t layers = dev_flow->handle->layers;
99 
100 	/*
101 	 * If layers is already initialized, it means this dev_flow is the
102 	 * suffix flow, the layers flags is set by the prefix flow. Need to
103 	 * use the layer flags from prefix flow as the suffix flow may not
104 	 * have the user defined items as the flow is split.
105 	 */
106 	if (layers) {
107 		if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
108 			attr->ipv4 = 1;
109 		else if (layers & MLX5_FLOW_LAYER_OUTER_L3_IPV6)
110 			attr->ipv6 = 1;
111 		if (layers & MLX5_FLOW_LAYER_OUTER_L4_TCP)
112 			attr->tcp = 1;
113 		else if (layers & MLX5_FLOW_LAYER_OUTER_L4_UDP)
114 			attr->udp = 1;
115 		attr->valid = 1;
116 		return;
117 	}
118 	for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
119 		uint8_t next_protocol = 0xff;
120 		switch (item->type) {
121 		case RTE_FLOW_ITEM_TYPE_GRE:
122 		case RTE_FLOW_ITEM_TYPE_NVGRE:
123 		case RTE_FLOW_ITEM_TYPE_VXLAN:
124 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
125 		case RTE_FLOW_ITEM_TYPE_GENEVE:
126 		case RTE_FLOW_ITEM_TYPE_MPLS:
127 			if (tunnel_decap)
128 				attr->attr = 0;
129 			break;
130 		case RTE_FLOW_ITEM_TYPE_IPV4:
131 			if (!attr->ipv6)
132 				attr->ipv4 = 1;
133 			if (item->mask != NULL &&
134 			    ((const struct rte_flow_item_ipv4 *)
135 			    item->mask)->hdr.next_proto_id)
136 				next_protocol =
137 				    ((const struct rte_flow_item_ipv4 *)
138 				      (item->spec))->hdr.next_proto_id &
139 				    ((const struct rte_flow_item_ipv4 *)
140 				      (item->mask))->hdr.next_proto_id;
141 			if ((next_protocol == IPPROTO_IPIP ||
142 			    next_protocol == IPPROTO_IPV6) && tunnel_decap)
143 				attr->attr = 0;
144 			break;
145 		case RTE_FLOW_ITEM_TYPE_IPV6:
146 			if (!attr->ipv4)
147 				attr->ipv6 = 1;
148 			if (item->mask != NULL &&
149 			    ((const struct rte_flow_item_ipv6 *)
150 			    item->mask)->hdr.proto)
151 				next_protocol =
152 				    ((const struct rte_flow_item_ipv6 *)
153 				      (item->spec))->hdr.proto &
154 				    ((const struct rte_flow_item_ipv6 *)
155 				      (item->mask))->hdr.proto;
156 			if ((next_protocol == IPPROTO_IPIP ||
157 			    next_protocol == IPPROTO_IPV6) && tunnel_decap)
158 				attr->attr = 0;
159 			break;
160 		case RTE_FLOW_ITEM_TYPE_UDP:
161 			if (!attr->tcp)
162 				attr->udp = 1;
163 			break;
164 		case RTE_FLOW_ITEM_TYPE_TCP:
165 			if (!attr->udp)
166 				attr->tcp = 1;
167 			break;
168 		default:
169 			break;
170 		}
171 	}
172 	attr->valid = 1;
173 }
174 
175 /**
176  * Convert rte_mtr_color to mlx5 color.
177  *
178  * @param[in] rcol
179  *   rte_mtr_color.
180  *
181  * @return
182  *   mlx5 color.
183  */
184 static int
185 rte_col_2_mlx5_col(enum rte_color rcol)
186 {
187 	switch (rcol) {
188 	case RTE_COLOR_GREEN:
189 		return MLX5_FLOW_COLOR_GREEN;
190 	case RTE_COLOR_YELLOW:
191 		return MLX5_FLOW_COLOR_YELLOW;
192 	case RTE_COLOR_RED:
193 		return MLX5_FLOW_COLOR_RED;
194 	default:
195 		break;
196 	}
197 	return MLX5_FLOW_COLOR_UNDEFINED;
198 }
199 
200 struct field_modify_info {
201 	uint32_t size; /* Size of field in protocol header, in bytes. */
202 	uint32_t offset; /* Offset of field in protocol header, in bytes. */
203 	enum mlx5_modification_field id;
204 };
205 
206 struct field_modify_info modify_eth[] = {
207 	{4,  0, MLX5_MODI_OUT_DMAC_47_16},
208 	{2,  4, MLX5_MODI_OUT_DMAC_15_0},
209 	{4,  6, MLX5_MODI_OUT_SMAC_47_16},
210 	{2, 10, MLX5_MODI_OUT_SMAC_15_0},
211 	{0, 0, 0},
212 };
213 
214 struct field_modify_info modify_vlan_out_first_vid[] = {
215 	/* Size in bits !!! */
216 	{12, 0, MLX5_MODI_OUT_FIRST_VID},
217 	{0, 0, 0},
218 };
219 
220 struct field_modify_info modify_ipv4[] = {
221 	{1,  1, MLX5_MODI_OUT_IP_DSCP},
222 	{1,  8, MLX5_MODI_OUT_IPV4_TTL},
223 	{4, 12, MLX5_MODI_OUT_SIPV4},
224 	{4, 16, MLX5_MODI_OUT_DIPV4},
225 	{0, 0, 0},
226 };
227 
228 struct field_modify_info modify_ipv6[] = {
229 	{1,  0, MLX5_MODI_OUT_IP_DSCP},
230 	{1,  7, MLX5_MODI_OUT_IPV6_HOPLIMIT},
231 	{4,  8, MLX5_MODI_OUT_SIPV6_127_96},
232 	{4, 12, MLX5_MODI_OUT_SIPV6_95_64},
233 	{4, 16, MLX5_MODI_OUT_SIPV6_63_32},
234 	{4, 20, MLX5_MODI_OUT_SIPV6_31_0},
235 	{4, 24, MLX5_MODI_OUT_DIPV6_127_96},
236 	{4, 28, MLX5_MODI_OUT_DIPV6_95_64},
237 	{4, 32, MLX5_MODI_OUT_DIPV6_63_32},
238 	{4, 36, MLX5_MODI_OUT_DIPV6_31_0},
239 	{0, 0, 0},
240 };
241 
242 struct field_modify_info modify_udp[] = {
243 	{2, 0, MLX5_MODI_OUT_UDP_SPORT},
244 	{2, 2, MLX5_MODI_OUT_UDP_DPORT},
245 	{0, 0, 0},
246 };
247 
248 struct field_modify_info modify_tcp[] = {
249 	{2, 0, MLX5_MODI_OUT_TCP_SPORT},
250 	{2, 2, MLX5_MODI_OUT_TCP_DPORT},
251 	{4, 4, MLX5_MODI_OUT_TCP_SEQ_NUM},
252 	{4, 8, MLX5_MODI_OUT_TCP_ACK_NUM},
253 	{0, 0, 0},
254 };
255 
256 static void
257 mlx5_flow_tunnel_ip_check(const struct rte_flow_item *item __rte_unused,
258 			  uint8_t next_protocol, uint64_t *item_flags,
259 			  int *tunnel)
260 {
261 	MLX5_ASSERT(item->type == RTE_FLOW_ITEM_TYPE_IPV4 ||
262 		    item->type == RTE_FLOW_ITEM_TYPE_IPV6);
263 	if (next_protocol == IPPROTO_IPIP) {
264 		*item_flags |= MLX5_FLOW_LAYER_IPIP;
265 		*tunnel = 1;
266 	}
267 	if (next_protocol == IPPROTO_IPV6) {
268 		*item_flags |= MLX5_FLOW_LAYER_IPV6_ENCAP;
269 		*tunnel = 1;
270 	}
271 }
272 
273 /**
274  * Acquire the synchronizing object to protect multithreaded access
275  * to shared dv context. Lock occurs only if context is actually
276  * shared, i.e. we have multiport IB device and representors are
277  * created.
278  *
279  * @param[in] dev
280  *   Pointer to the rte_eth_dev structure.
281  */
282 static void
283 flow_dv_shared_lock(struct rte_eth_dev *dev)
284 {
285 	struct mlx5_priv *priv = dev->data->dev_private;
286 	struct mlx5_dev_ctx_shared *sh = priv->sh;
287 
288 	if (sh->dv_refcnt > 1) {
289 		int ret;
290 
291 		ret = pthread_mutex_lock(&sh->dv_mutex);
292 		MLX5_ASSERT(!ret);
293 		(void)ret;
294 	}
295 }
296 
297 static void
298 flow_dv_shared_unlock(struct rte_eth_dev *dev)
299 {
300 	struct mlx5_priv *priv = dev->data->dev_private;
301 	struct mlx5_dev_ctx_shared *sh = priv->sh;
302 
303 	if (sh->dv_refcnt > 1) {
304 		int ret;
305 
306 		ret = pthread_mutex_unlock(&sh->dv_mutex);
307 		MLX5_ASSERT(!ret);
308 		(void)ret;
309 	}
310 }
311 
312 /* Update VLAN's VID/PCP based on input rte_flow_action.
313  *
314  * @param[in] action
315  *   Pointer to struct rte_flow_action.
316  * @param[out] vlan
317  *   Pointer to struct rte_vlan_hdr.
318  */
319 static void
320 mlx5_update_vlan_vid_pcp(const struct rte_flow_action *action,
321 			 struct rte_vlan_hdr *vlan)
322 {
323 	uint16_t vlan_tci;
324 	if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
325 		vlan_tci =
326 		    ((const struct rte_flow_action_of_set_vlan_pcp *)
327 					       action->conf)->vlan_pcp;
328 		vlan_tci = vlan_tci << MLX5DV_FLOW_VLAN_PCP_SHIFT;
329 		vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
330 		vlan->vlan_tci |= vlan_tci;
331 	} else if (action->type == RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
332 		vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
333 		vlan->vlan_tci |= rte_be_to_cpu_16
334 		    (((const struct rte_flow_action_of_set_vlan_vid *)
335 					     action->conf)->vlan_vid);
336 	}
337 }
338 
339 /**
340  * Fetch 1, 2, 3 or 4 byte field from the byte array
341  * and return as unsigned integer in host-endian format.
342  *
343  * @param[in] data
344  *   Pointer to data array.
345  * @param[in] size
346  *   Size of field to extract.
347  *
348  * @return
349  *   converted field in host endian format.
350  */
351 static inline uint32_t
352 flow_dv_fetch_field(const uint8_t *data, uint32_t size)
353 {
354 	uint32_t ret;
355 
356 	switch (size) {
357 	case 1:
358 		ret = *data;
359 		break;
360 	case 2:
361 		ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
362 		break;
363 	case 3:
364 		ret = rte_be_to_cpu_16(*(const unaligned_uint16_t *)data);
365 		ret = (ret << 8) | *(data + sizeof(uint16_t));
366 		break;
367 	case 4:
368 		ret = rte_be_to_cpu_32(*(const unaligned_uint32_t *)data);
369 		break;
370 	default:
371 		MLX5_ASSERT(false);
372 		ret = 0;
373 		break;
374 	}
375 	return ret;
376 }
377 
378 /**
379  * Convert modify-header action to DV specification.
380  *
381  * Data length of each action is determined by provided field description
382  * and the item mask. Data bit offset and width of each action is determined
383  * by provided item mask.
384  *
385  * @param[in] item
386  *   Pointer to item specification.
387  * @param[in] field
388  *   Pointer to field modification information.
389  *     For MLX5_MODIFICATION_TYPE_SET specifies destination field.
390  *     For MLX5_MODIFICATION_TYPE_ADD specifies destination field.
391  *     For MLX5_MODIFICATION_TYPE_COPY specifies source field.
392  * @param[in] dcopy
393  *   Destination field info for MLX5_MODIFICATION_TYPE_COPY in @type.
394  *   Negative offset value sets the same offset as source offset.
395  *   size field is ignored, value is taken from source field.
396  * @param[in,out] resource
397  *   Pointer to the modify-header resource.
398  * @param[in] type
399  *   Type of modification.
400  * @param[out] error
401  *   Pointer to the error structure.
402  *
403  * @return
404  *   0 on success, a negative errno value otherwise and rte_errno is set.
405  */
406 static int
407 flow_dv_convert_modify_action(struct rte_flow_item *item,
408 			      struct field_modify_info *field,
409 			      struct field_modify_info *dcopy,
410 			      struct mlx5_flow_dv_modify_hdr_resource *resource,
411 			      uint32_t type, struct rte_flow_error *error)
412 {
413 	uint32_t i = resource->actions_num;
414 	struct mlx5_modification_cmd *actions = resource->actions;
415 
416 	/*
417 	 * The item and mask are provided in big-endian format.
418 	 * The fields should be presented as in big-endian format either.
419 	 * Mask must be always present, it defines the actual field width.
420 	 */
421 	MLX5_ASSERT(item->mask);
422 	MLX5_ASSERT(field->size);
423 	do {
424 		unsigned int size_b;
425 		unsigned int off_b;
426 		uint32_t mask;
427 		uint32_t data;
428 
429 		if (i >= MLX5_MAX_MODIFY_NUM)
430 			return rte_flow_error_set(error, EINVAL,
431 				 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
432 				 "too many items to modify");
433 		/* Fetch variable byte size mask from the array. */
434 		mask = flow_dv_fetch_field((const uint8_t *)item->mask +
435 					   field->offset, field->size);
436 		if (!mask) {
437 			++field;
438 			continue;
439 		}
440 		/* Deduce actual data width in bits from mask value. */
441 		off_b = rte_bsf32(mask);
442 		size_b = sizeof(uint32_t) * CHAR_BIT -
443 			 off_b - __builtin_clz(mask);
444 		MLX5_ASSERT(size_b);
445 		size_b = size_b == sizeof(uint32_t) * CHAR_BIT ? 0 : size_b;
446 		actions[i] = (struct mlx5_modification_cmd) {
447 			.action_type = type,
448 			.field = field->id,
449 			.offset = off_b,
450 			.length = size_b,
451 		};
452 		/* Convert entire record to expected big-endian format. */
453 		actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
454 		if (type == MLX5_MODIFICATION_TYPE_COPY) {
455 			MLX5_ASSERT(dcopy);
456 			actions[i].dst_field = dcopy->id;
457 			actions[i].dst_offset =
458 				(int)dcopy->offset < 0 ? off_b : dcopy->offset;
459 			/* Convert entire record to big-endian format. */
460 			actions[i].data1 = rte_cpu_to_be_32(actions[i].data1);
461 		} else {
462 			MLX5_ASSERT(item->spec);
463 			data = flow_dv_fetch_field((const uint8_t *)item->spec +
464 						   field->offset, field->size);
465 			/* Shift out the trailing masked bits from data. */
466 			data = (data & mask) >> off_b;
467 			actions[i].data1 = rte_cpu_to_be_32(data);
468 		}
469 		++i;
470 		++field;
471 	} while (field->size);
472 	if (resource->actions_num == i)
473 		return rte_flow_error_set(error, EINVAL,
474 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
475 					  "invalid modification flow item");
476 	resource->actions_num = i;
477 	return 0;
478 }
479 
480 /**
481  * Convert modify-header set IPv4 address action to DV specification.
482  *
483  * @param[in,out] resource
484  *   Pointer to the modify-header resource.
485  * @param[in] action
486  *   Pointer to action specification.
487  * @param[out] error
488  *   Pointer to the error structure.
489  *
490  * @return
491  *   0 on success, a negative errno value otherwise and rte_errno is set.
492  */
493 static int
494 flow_dv_convert_action_modify_ipv4
495 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
496 			 const struct rte_flow_action *action,
497 			 struct rte_flow_error *error)
498 {
499 	const struct rte_flow_action_set_ipv4 *conf =
500 		(const struct rte_flow_action_set_ipv4 *)(action->conf);
501 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
502 	struct rte_flow_item_ipv4 ipv4;
503 	struct rte_flow_item_ipv4 ipv4_mask;
504 
505 	memset(&ipv4, 0, sizeof(ipv4));
506 	memset(&ipv4_mask, 0, sizeof(ipv4_mask));
507 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC) {
508 		ipv4.hdr.src_addr = conf->ipv4_addr;
509 		ipv4_mask.hdr.src_addr = rte_flow_item_ipv4_mask.hdr.src_addr;
510 	} else {
511 		ipv4.hdr.dst_addr = conf->ipv4_addr;
512 		ipv4_mask.hdr.dst_addr = rte_flow_item_ipv4_mask.hdr.dst_addr;
513 	}
514 	item.spec = &ipv4;
515 	item.mask = &ipv4_mask;
516 	return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
517 					     MLX5_MODIFICATION_TYPE_SET, error);
518 }
519 
520 /**
521  * Convert modify-header set IPv6 address action to DV specification.
522  *
523  * @param[in,out] resource
524  *   Pointer to the modify-header resource.
525  * @param[in] action
526  *   Pointer to action specification.
527  * @param[out] error
528  *   Pointer to the error structure.
529  *
530  * @return
531  *   0 on success, a negative errno value otherwise and rte_errno is set.
532  */
533 static int
534 flow_dv_convert_action_modify_ipv6
535 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
536 			 const struct rte_flow_action *action,
537 			 struct rte_flow_error *error)
538 {
539 	const struct rte_flow_action_set_ipv6 *conf =
540 		(const struct rte_flow_action_set_ipv6 *)(action->conf);
541 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
542 	struct rte_flow_item_ipv6 ipv6;
543 	struct rte_flow_item_ipv6 ipv6_mask;
544 
545 	memset(&ipv6, 0, sizeof(ipv6));
546 	memset(&ipv6_mask, 0, sizeof(ipv6_mask));
547 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC) {
548 		memcpy(&ipv6.hdr.src_addr, &conf->ipv6_addr,
549 		       sizeof(ipv6.hdr.src_addr));
550 		memcpy(&ipv6_mask.hdr.src_addr,
551 		       &rte_flow_item_ipv6_mask.hdr.src_addr,
552 		       sizeof(ipv6.hdr.src_addr));
553 	} else {
554 		memcpy(&ipv6.hdr.dst_addr, &conf->ipv6_addr,
555 		       sizeof(ipv6.hdr.dst_addr));
556 		memcpy(&ipv6_mask.hdr.dst_addr,
557 		       &rte_flow_item_ipv6_mask.hdr.dst_addr,
558 		       sizeof(ipv6.hdr.dst_addr));
559 	}
560 	item.spec = &ipv6;
561 	item.mask = &ipv6_mask;
562 	return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
563 					     MLX5_MODIFICATION_TYPE_SET, error);
564 }
565 
566 /**
567  * Convert modify-header set MAC address action to DV specification.
568  *
569  * @param[in,out] resource
570  *   Pointer to the modify-header resource.
571  * @param[in] action
572  *   Pointer to action specification.
573  * @param[out] error
574  *   Pointer to the error structure.
575  *
576  * @return
577  *   0 on success, a negative errno value otherwise and rte_errno is set.
578  */
579 static int
580 flow_dv_convert_action_modify_mac
581 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
582 			 const struct rte_flow_action *action,
583 			 struct rte_flow_error *error)
584 {
585 	const struct rte_flow_action_set_mac *conf =
586 		(const struct rte_flow_action_set_mac *)(action->conf);
587 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_ETH };
588 	struct rte_flow_item_eth eth;
589 	struct rte_flow_item_eth eth_mask;
590 
591 	memset(&eth, 0, sizeof(eth));
592 	memset(&eth_mask, 0, sizeof(eth_mask));
593 	if (action->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC) {
594 		memcpy(&eth.src.addr_bytes, &conf->mac_addr,
595 		       sizeof(eth.src.addr_bytes));
596 		memcpy(&eth_mask.src.addr_bytes,
597 		       &rte_flow_item_eth_mask.src.addr_bytes,
598 		       sizeof(eth_mask.src.addr_bytes));
599 	} else {
600 		memcpy(&eth.dst.addr_bytes, &conf->mac_addr,
601 		       sizeof(eth.dst.addr_bytes));
602 		memcpy(&eth_mask.dst.addr_bytes,
603 		       &rte_flow_item_eth_mask.dst.addr_bytes,
604 		       sizeof(eth_mask.dst.addr_bytes));
605 	}
606 	item.spec = &eth;
607 	item.mask = &eth_mask;
608 	return flow_dv_convert_modify_action(&item, modify_eth, NULL, resource,
609 					     MLX5_MODIFICATION_TYPE_SET, error);
610 }
611 
612 /**
613  * Convert modify-header set VLAN VID action to DV specification.
614  *
615  * @param[in,out] resource
616  *   Pointer to the modify-header resource.
617  * @param[in] action
618  *   Pointer to action specification.
619  * @param[out] error
620  *   Pointer to the error structure.
621  *
622  * @return
623  *   0 on success, a negative errno value otherwise and rte_errno is set.
624  */
625 static int
626 flow_dv_convert_action_modify_vlan_vid
627 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
628 			 const struct rte_flow_action *action,
629 			 struct rte_flow_error *error)
630 {
631 	const struct rte_flow_action_of_set_vlan_vid *conf =
632 		(const struct rte_flow_action_of_set_vlan_vid *)(action->conf);
633 	int i = resource->actions_num;
634 	struct mlx5_modification_cmd *actions = resource->actions;
635 	struct field_modify_info *field = modify_vlan_out_first_vid;
636 
637 	if (i >= MLX5_MAX_MODIFY_NUM)
638 		return rte_flow_error_set(error, EINVAL,
639 			 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
640 			 "too many items to modify");
641 	actions[i] = (struct mlx5_modification_cmd) {
642 		.action_type = MLX5_MODIFICATION_TYPE_SET,
643 		.field = field->id,
644 		.length = field->size,
645 		.offset = field->offset,
646 	};
647 	actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
648 	actions[i].data1 = conf->vlan_vid;
649 	actions[i].data1 = actions[i].data1 << 16;
650 	resource->actions_num = ++i;
651 	return 0;
652 }
653 
654 /**
655  * Convert modify-header set TP action to DV specification.
656  *
657  * @param[in,out] resource
658  *   Pointer to the modify-header resource.
659  * @param[in] action
660  *   Pointer to action specification.
661  * @param[in] items
662  *   Pointer to rte_flow_item objects list.
663  * @param[in] attr
664  *   Pointer to flow attributes structure.
665  * @param[in] dev_flow
666  *   Pointer to the sub flow.
667  * @param[in] tunnel_decap
668  *   Whether action is after tunnel decapsulation.
669  * @param[out] error
670  *   Pointer to the error structure.
671  *
672  * @return
673  *   0 on success, a negative errno value otherwise and rte_errno is set.
674  */
675 static int
676 flow_dv_convert_action_modify_tp
677 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
678 			 const struct rte_flow_action *action,
679 			 const struct rte_flow_item *items,
680 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
681 			 bool tunnel_decap, struct rte_flow_error *error)
682 {
683 	const struct rte_flow_action_set_tp *conf =
684 		(const struct rte_flow_action_set_tp *)(action->conf);
685 	struct rte_flow_item item;
686 	struct rte_flow_item_udp udp;
687 	struct rte_flow_item_udp udp_mask;
688 	struct rte_flow_item_tcp tcp;
689 	struct rte_flow_item_tcp tcp_mask;
690 	struct field_modify_info *field;
691 
692 	if (!attr->valid)
693 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
694 	if (attr->udp) {
695 		memset(&udp, 0, sizeof(udp));
696 		memset(&udp_mask, 0, sizeof(udp_mask));
697 		if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
698 			udp.hdr.src_port = conf->port;
699 			udp_mask.hdr.src_port =
700 					rte_flow_item_udp_mask.hdr.src_port;
701 		} else {
702 			udp.hdr.dst_port = conf->port;
703 			udp_mask.hdr.dst_port =
704 					rte_flow_item_udp_mask.hdr.dst_port;
705 		}
706 		item.type = RTE_FLOW_ITEM_TYPE_UDP;
707 		item.spec = &udp;
708 		item.mask = &udp_mask;
709 		field = modify_udp;
710 	} else {
711 		MLX5_ASSERT(attr->tcp);
712 		memset(&tcp, 0, sizeof(tcp));
713 		memset(&tcp_mask, 0, sizeof(tcp_mask));
714 		if (action->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC) {
715 			tcp.hdr.src_port = conf->port;
716 			tcp_mask.hdr.src_port =
717 					rte_flow_item_tcp_mask.hdr.src_port;
718 		} else {
719 			tcp.hdr.dst_port = conf->port;
720 			tcp_mask.hdr.dst_port =
721 					rte_flow_item_tcp_mask.hdr.dst_port;
722 		}
723 		item.type = RTE_FLOW_ITEM_TYPE_TCP;
724 		item.spec = &tcp;
725 		item.mask = &tcp_mask;
726 		field = modify_tcp;
727 	}
728 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
729 					     MLX5_MODIFICATION_TYPE_SET, error);
730 }
731 
732 /**
733  * Convert modify-header set TTL action to DV specification.
734  *
735  * @param[in,out] resource
736  *   Pointer to the modify-header resource.
737  * @param[in] action
738  *   Pointer to action specification.
739  * @param[in] items
740  *   Pointer to rte_flow_item objects list.
741  * @param[in] attr
742  *   Pointer to flow attributes structure.
743  * @param[in] dev_flow
744  *   Pointer to the sub flow.
745  * @param[in] tunnel_decap
746  *   Whether action is after tunnel decapsulation.
747  * @param[out] error
748  *   Pointer to the error structure.
749  *
750  * @return
751  *   0 on success, a negative errno value otherwise and rte_errno is set.
752  */
753 static int
754 flow_dv_convert_action_modify_ttl
755 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
756 			 const struct rte_flow_action *action,
757 			 const struct rte_flow_item *items,
758 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
759 			 bool tunnel_decap, struct rte_flow_error *error)
760 {
761 	const struct rte_flow_action_set_ttl *conf =
762 		(const struct rte_flow_action_set_ttl *)(action->conf);
763 	struct rte_flow_item item;
764 	struct rte_flow_item_ipv4 ipv4;
765 	struct rte_flow_item_ipv4 ipv4_mask;
766 	struct rte_flow_item_ipv6 ipv6;
767 	struct rte_flow_item_ipv6 ipv6_mask;
768 	struct field_modify_info *field;
769 
770 	if (!attr->valid)
771 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
772 	if (attr->ipv4) {
773 		memset(&ipv4, 0, sizeof(ipv4));
774 		memset(&ipv4_mask, 0, sizeof(ipv4_mask));
775 		ipv4.hdr.time_to_live = conf->ttl_value;
776 		ipv4_mask.hdr.time_to_live = 0xFF;
777 		item.type = RTE_FLOW_ITEM_TYPE_IPV4;
778 		item.spec = &ipv4;
779 		item.mask = &ipv4_mask;
780 		field = modify_ipv4;
781 	} else {
782 		MLX5_ASSERT(attr->ipv6);
783 		memset(&ipv6, 0, sizeof(ipv6));
784 		memset(&ipv6_mask, 0, sizeof(ipv6_mask));
785 		ipv6.hdr.hop_limits = conf->ttl_value;
786 		ipv6_mask.hdr.hop_limits = 0xFF;
787 		item.type = RTE_FLOW_ITEM_TYPE_IPV6;
788 		item.spec = &ipv6;
789 		item.mask = &ipv6_mask;
790 		field = modify_ipv6;
791 	}
792 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
793 					     MLX5_MODIFICATION_TYPE_SET, error);
794 }
795 
796 /**
797  * Convert modify-header decrement TTL action to DV specification.
798  *
799  * @param[in,out] resource
800  *   Pointer to the modify-header resource.
801  * @param[in] action
802  *   Pointer to action specification.
803  * @param[in] items
804  *   Pointer to rte_flow_item objects list.
805  * @param[in] attr
806  *   Pointer to flow attributes structure.
807  * @param[in] dev_flow
808  *   Pointer to the sub flow.
809  * @param[in] tunnel_decap
810  *   Whether action is after tunnel decapsulation.
811  * @param[out] error
812  *   Pointer to the error structure.
813  *
814  * @return
815  *   0 on success, a negative errno value otherwise and rte_errno is set.
816  */
817 static int
818 flow_dv_convert_action_modify_dec_ttl
819 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
820 			 const struct rte_flow_item *items,
821 			 union flow_dv_attr *attr, struct mlx5_flow *dev_flow,
822 			 bool tunnel_decap, struct rte_flow_error *error)
823 {
824 	struct rte_flow_item item;
825 	struct rte_flow_item_ipv4 ipv4;
826 	struct rte_flow_item_ipv4 ipv4_mask;
827 	struct rte_flow_item_ipv6 ipv6;
828 	struct rte_flow_item_ipv6 ipv6_mask;
829 	struct field_modify_info *field;
830 
831 	if (!attr->valid)
832 		flow_dv_attr_init(items, attr, dev_flow, tunnel_decap);
833 	if (attr->ipv4) {
834 		memset(&ipv4, 0, sizeof(ipv4));
835 		memset(&ipv4_mask, 0, sizeof(ipv4_mask));
836 		ipv4.hdr.time_to_live = 0xFF;
837 		ipv4_mask.hdr.time_to_live = 0xFF;
838 		item.type = RTE_FLOW_ITEM_TYPE_IPV4;
839 		item.spec = &ipv4;
840 		item.mask = &ipv4_mask;
841 		field = modify_ipv4;
842 	} else {
843 		MLX5_ASSERT(attr->ipv6);
844 		memset(&ipv6, 0, sizeof(ipv6));
845 		memset(&ipv6_mask, 0, sizeof(ipv6_mask));
846 		ipv6.hdr.hop_limits = 0xFF;
847 		ipv6_mask.hdr.hop_limits = 0xFF;
848 		item.type = RTE_FLOW_ITEM_TYPE_IPV6;
849 		item.spec = &ipv6;
850 		item.mask = &ipv6_mask;
851 		field = modify_ipv6;
852 	}
853 	return flow_dv_convert_modify_action(&item, field, NULL, resource,
854 					     MLX5_MODIFICATION_TYPE_ADD, error);
855 }
856 
857 /**
858  * Convert modify-header increment/decrement TCP Sequence number
859  * to DV specification.
860  *
861  * @param[in,out] resource
862  *   Pointer to the modify-header resource.
863  * @param[in] action
864  *   Pointer to action specification.
865  * @param[out] error
866  *   Pointer to the error structure.
867  *
868  * @return
869  *   0 on success, a negative errno value otherwise and rte_errno is set.
870  */
871 static int
872 flow_dv_convert_action_modify_tcp_seq
873 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
874 			 const struct rte_flow_action *action,
875 			 struct rte_flow_error *error)
876 {
877 	const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
878 	uint64_t value = rte_be_to_cpu_32(*conf);
879 	struct rte_flow_item item;
880 	struct rte_flow_item_tcp tcp;
881 	struct rte_flow_item_tcp tcp_mask;
882 
883 	memset(&tcp, 0, sizeof(tcp));
884 	memset(&tcp_mask, 0, sizeof(tcp_mask));
885 	if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ)
886 		/*
887 		 * The HW has no decrement operation, only increment operation.
888 		 * To simulate decrement X from Y using increment operation
889 		 * we need to add UINT32_MAX X times to Y.
890 		 * Each adding of UINT32_MAX decrements Y by 1.
891 		 */
892 		value *= UINT32_MAX;
893 	tcp.hdr.sent_seq = rte_cpu_to_be_32((uint32_t)value);
894 	tcp_mask.hdr.sent_seq = RTE_BE32(UINT32_MAX);
895 	item.type = RTE_FLOW_ITEM_TYPE_TCP;
896 	item.spec = &tcp;
897 	item.mask = &tcp_mask;
898 	return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
899 					     MLX5_MODIFICATION_TYPE_ADD, error);
900 }
901 
902 /**
903  * Convert modify-header increment/decrement TCP Acknowledgment number
904  * to DV specification.
905  *
906  * @param[in,out] resource
907  *   Pointer to the modify-header resource.
908  * @param[in] action
909  *   Pointer to action specification.
910  * @param[out] error
911  *   Pointer to the error structure.
912  *
913  * @return
914  *   0 on success, a negative errno value otherwise and rte_errno is set.
915  */
916 static int
917 flow_dv_convert_action_modify_tcp_ack
918 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
919 			 const struct rte_flow_action *action,
920 			 struct rte_flow_error *error)
921 {
922 	const rte_be32_t *conf = (const rte_be32_t *)(action->conf);
923 	uint64_t value = rte_be_to_cpu_32(*conf);
924 	struct rte_flow_item item;
925 	struct rte_flow_item_tcp tcp;
926 	struct rte_flow_item_tcp tcp_mask;
927 
928 	memset(&tcp, 0, sizeof(tcp));
929 	memset(&tcp_mask, 0, sizeof(tcp_mask));
930 	if (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK)
931 		/*
932 		 * The HW has no decrement operation, only increment operation.
933 		 * To simulate decrement X from Y using increment operation
934 		 * we need to add UINT32_MAX X times to Y.
935 		 * Each adding of UINT32_MAX decrements Y by 1.
936 		 */
937 		value *= UINT32_MAX;
938 	tcp.hdr.recv_ack = rte_cpu_to_be_32((uint32_t)value);
939 	tcp_mask.hdr.recv_ack = RTE_BE32(UINT32_MAX);
940 	item.type = RTE_FLOW_ITEM_TYPE_TCP;
941 	item.spec = &tcp;
942 	item.mask = &tcp_mask;
943 	return flow_dv_convert_modify_action(&item, modify_tcp, NULL, resource,
944 					     MLX5_MODIFICATION_TYPE_ADD, error);
945 }
946 
947 static enum mlx5_modification_field reg_to_field[] = {
948 	[REG_NONE] = MLX5_MODI_OUT_NONE,
949 	[REG_A] = MLX5_MODI_META_DATA_REG_A,
950 	[REG_B] = MLX5_MODI_META_DATA_REG_B,
951 	[REG_C_0] = MLX5_MODI_META_REG_C_0,
952 	[REG_C_1] = MLX5_MODI_META_REG_C_1,
953 	[REG_C_2] = MLX5_MODI_META_REG_C_2,
954 	[REG_C_3] = MLX5_MODI_META_REG_C_3,
955 	[REG_C_4] = MLX5_MODI_META_REG_C_4,
956 	[REG_C_5] = MLX5_MODI_META_REG_C_5,
957 	[REG_C_6] = MLX5_MODI_META_REG_C_6,
958 	[REG_C_7] = MLX5_MODI_META_REG_C_7,
959 };
960 
961 /**
962  * Convert register set to DV specification.
963  *
964  * @param[in,out] resource
965  *   Pointer to the modify-header resource.
966  * @param[in] action
967  *   Pointer to action specification.
968  * @param[out] error
969  *   Pointer to the error structure.
970  *
971  * @return
972  *   0 on success, a negative errno value otherwise and rte_errno is set.
973  */
974 static int
975 flow_dv_convert_action_set_reg
976 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
977 			 const struct rte_flow_action *action,
978 			 struct rte_flow_error *error)
979 {
980 	const struct mlx5_rte_flow_action_set_tag *conf = action->conf;
981 	struct mlx5_modification_cmd *actions = resource->actions;
982 	uint32_t i = resource->actions_num;
983 
984 	if (i >= MLX5_MAX_MODIFY_NUM)
985 		return rte_flow_error_set(error, EINVAL,
986 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
987 					  "too many items to modify");
988 	MLX5_ASSERT(conf->id != REG_NONE);
989 	MLX5_ASSERT(conf->id < RTE_DIM(reg_to_field));
990 	actions[i] = (struct mlx5_modification_cmd) {
991 		.action_type = MLX5_MODIFICATION_TYPE_SET,
992 		.field = reg_to_field[conf->id],
993 	};
994 	actions[i].data0 = rte_cpu_to_be_32(actions[i].data0);
995 	actions[i].data1 = rte_cpu_to_be_32(conf->data);
996 	++i;
997 	resource->actions_num = i;
998 	return 0;
999 }
1000 
1001 /**
1002  * Convert SET_TAG action to DV specification.
1003  *
1004  * @param[in] dev
1005  *   Pointer to the rte_eth_dev structure.
1006  * @param[in,out] resource
1007  *   Pointer to the modify-header resource.
1008  * @param[in] conf
1009  *   Pointer to action specification.
1010  * @param[out] error
1011  *   Pointer to the error structure.
1012  *
1013  * @return
1014  *   0 on success, a negative errno value otherwise and rte_errno is set.
1015  */
1016 static int
1017 flow_dv_convert_action_set_tag
1018 			(struct rte_eth_dev *dev,
1019 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
1020 			 const struct rte_flow_action_set_tag *conf,
1021 			 struct rte_flow_error *error)
1022 {
1023 	rte_be32_t data = rte_cpu_to_be_32(conf->data);
1024 	rte_be32_t mask = rte_cpu_to_be_32(conf->mask);
1025 	struct rte_flow_item item = {
1026 		.spec = &data,
1027 		.mask = &mask,
1028 	};
1029 	struct field_modify_info reg_c_x[] = {
1030 		[1] = {0, 0, 0},
1031 	};
1032 	enum mlx5_modification_field reg_type;
1033 	int ret;
1034 
1035 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
1036 	if (ret < 0)
1037 		return ret;
1038 	MLX5_ASSERT(ret != REG_NONE);
1039 	MLX5_ASSERT((unsigned int)ret < RTE_DIM(reg_to_field));
1040 	reg_type = reg_to_field[ret];
1041 	MLX5_ASSERT(reg_type > 0);
1042 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_type};
1043 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1044 					     MLX5_MODIFICATION_TYPE_SET, error);
1045 }
1046 
1047 /**
1048  * Convert internal COPY_REG action to DV specification.
1049  *
1050  * @param[in] dev
1051  *   Pointer to the rte_eth_dev structure.
1052  * @param[in,out] res
1053  *   Pointer to the modify-header resource.
1054  * @param[in] action
1055  *   Pointer to action specification.
1056  * @param[out] error
1057  *   Pointer to the error structure.
1058  *
1059  * @return
1060  *   0 on success, a negative errno value otherwise and rte_errno is set.
1061  */
1062 static int
1063 flow_dv_convert_action_copy_mreg(struct rte_eth_dev *dev,
1064 				 struct mlx5_flow_dv_modify_hdr_resource *res,
1065 				 const struct rte_flow_action *action,
1066 				 struct rte_flow_error *error)
1067 {
1068 	const struct mlx5_flow_action_copy_mreg *conf = action->conf;
1069 	rte_be32_t mask = RTE_BE32(UINT32_MAX);
1070 	struct rte_flow_item item = {
1071 		.spec = NULL,
1072 		.mask = &mask,
1073 	};
1074 	struct field_modify_info reg_src[] = {
1075 		{4, 0, reg_to_field[conf->src]},
1076 		{0, 0, 0},
1077 	};
1078 	struct field_modify_info reg_dst = {
1079 		.offset = 0,
1080 		.id = reg_to_field[conf->dst],
1081 	};
1082 	/* Adjust reg_c[0] usage according to reported mask. */
1083 	if (conf->dst == REG_C_0 || conf->src == REG_C_0) {
1084 		struct mlx5_priv *priv = dev->data->dev_private;
1085 		uint32_t reg_c0 = priv->sh->dv_regc0_mask;
1086 
1087 		MLX5_ASSERT(reg_c0);
1088 		MLX5_ASSERT(priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY);
1089 		if (conf->dst == REG_C_0) {
1090 			/* Copy to reg_c[0], within mask only. */
1091 			reg_dst.offset = rte_bsf32(reg_c0);
1092 			/*
1093 			 * Mask is ignoring the enianness, because
1094 			 * there is no conversion in datapath.
1095 			 */
1096 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1097 			/* Copy from destination lower bits to reg_c[0]. */
1098 			mask = reg_c0 >> reg_dst.offset;
1099 #else
1100 			/* Copy from destination upper bits to reg_c[0]. */
1101 			mask = reg_c0 << (sizeof(reg_c0) * CHAR_BIT -
1102 					  rte_fls_u32(reg_c0));
1103 #endif
1104 		} else {
1105 			mask = rte_cpu_to_be_32(reg_c0);
1106 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1107 			/* Copy from reg_c[0] to destination lower bits. */
1108 			reg_dst.offset = 0;
1109 #else
1110 			/* Copy from reg_c[0] to destination upper bits. */
1111 			reg_dst.offset = sizeof(reg_c0) * CHAR_BIT -
1112 					 (rte_fls_u32(reg_c0) -
1113 					  rte_bsf32(reg_c0));
1114 #endif
1115 		}
1116 	}
1117 	return flow_dv_convert_modify_action(&item,
1118 					     reg_src, &reg_dst, res,
1119 					     MLX5_MODIFICATION_TYPE_COPY,
1120 					     error);
1121 }
1122 
1123 /**
1124  * Convert MARK action to DV specification. This routine is used
1125  * in extensive metadata only and requires metadata register to be
1126  * handled. In legacy mode hardware tag resource is engaged.
1127  *
1128  * @param[in] dev
1129  *   Pointer to the rte_eth_dev structure.
1130  * @param[in] conf
1131  *   Pointer to MARK action specification.
1132  * @param[in,out] resource
1133  *   Pointer to the modify-header resource.
1134  * @param[out] error
1135  *   Pointer to the error structure.
1136  *
1137  * @return
1138  *   0 on success, a negative errno value otherwise and rte_errno is set.
1139  */
1140 static int
1141 flow_dv_convert_action_mark(struct rte_eth_dev *dev,
1142 			    const struct rte_flow_action_mark *conf,
1143 			    struct mlx5_flow_dv_modify_hdr_resource *resource,
1144 			    struct rte_flow_error *error)
1145 {
1146 	struct mlx5_priv *priv = dev->data->dev_private;
1147 	rte_be32_t mask = rte_cpu_to_be_32(MLX5_FLOW_MARK_MASK &
1148 					   priv->sh->dv_mark_mask);
1149 	rte_be32_t data = rte_cpu_to_be_32(conf->id) & mask;
1150 	struct rte_flow_item item = {
1151 		.spec = &data,
1152 		.mask = &mask,
1153 	};
1154 	struct field_modify_info reg_c_x[] = {
1155 		[1] = {0, 0, 0},
1156 	};
1157 	int reg;
1158 
1159 	if (!mask)
1160 		return rte_flow_error_set(error, EINVAL,
1161 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 					  NULL, "zero mark action mask");
1163 	reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1164 	if (reg < 0)
1165 		return reg;
1166 	MLX5_ASSERT(reg > 0);
1167 	if (reg == REG_C_0) {
1168 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1169 		uint32_t shl_c0 = rte_bsf32(msk_c0);
1170 
1171 		data = rte_cpu_to_be_32(rte_cpu_to_be_32(data) << shl_c0);
1172 		mask = rte_cpu_to_be_32(mask) & msk_c0;
1173 		mask = rte_cpu_to_be_32(mask << shl_c0);
1174 	}
1175 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1176 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1177 					     MLX5_MODIFICATION_TYPE_SET, error);
1178 }
1179 
1180 /**
1181  * Get metadata register index for specified steering domain.
1182  *
1183  * @param[in] dev
1184  *   Pointer to the rte_eth_dev structure.
1185  * @param[in] attr
1186  *   Attributes of flow to determine steering domain.
1187  * @param[out] error
1188  *   Pointer to the error structure.
1189  *
1190  * @return
1191  *   positive index on success, a negative errno value otherwise
1192  *   and rte_errno is set.
1193  */
1194 static enum modify_reg
1195 flow_dv_get_metadata_reg(struct rte_eth_dev *dev,
1196 			 const struct rte_flow_attr *attr,
1197 			 struct rte_flow_error *error)
1198 {
1199 	int reg =
1200 		mlx5_flow_get_reg_id(dev, attr->transfer ?
1201 					  MLX5_METADATA_FDB :
1202 					    attr->egress ?
1203 					    MLX5_METADATA_TX :
1204 					    MLX5_METADATA_RX, 0, error);
1205 	if (reg < 0)
1206 		return rte_flow_error_set(error,
1207 					  ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
1208 					  NULL, "unavailable "
1209 					  "metadata register");
1210 	return reg;
1211 }
1212 
1213 /**
1214  * Convert SET_META action to DV specification.
1215  *
1216  * @param[in] dev
1217  *   Pointer to the rte_eth_dev structure.
1218  * @param[in,out] resource
1219  *   Pointer to the modify-header resource.
1220  * @param[in] attr
1221  *   Attributes of flow that includes this item.
1222  * @param[in] conf
1223  *   Pointer to action specification.
1224  * @param[out] error
1225  *   Pointer to the error structure.
1226  *
1227  * @return
1228  *   0 on success, a negative errno value otherwise and rte_errno is set.
1229  */
1230 static int
1231 flow_dv_convert_action_set_meta
1232 			(struct rte_eth_dev *dev,
1233 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
1234 			 const struct rte_flow_attr *attr,
1235 			 const struct rte_flow_action_set_meta *conf,
1236 			 struct rte_flow_error *error)
1237 {
1238 	uint32_t data = conf->data;
1239 	uint32_t mask = conf->mask;
1240 	struct rte_flow_item item = {
1241 		.spec = &data,
1242 		.mask = &mask,
1243 	};
1244 	struct field_modify_info reg_c_x[] = {
1245 		[1] = {0, 0, 0},
1246 	};
1247 	int reg = flow_dv_get_metadata_reg(dev, attr, error);
1248 
1249 	if (reg < 0)
1250 		return reg;
1251 	/*
1252 	 * In datapath code there is no endianness
1253 	 * coversions for perfromance reasons, all
1254 	 * pattern conversions are done in rte_flow.
1255 	 */
1256 	if (reg == REG_C_0) {
1257 		struct mlx5_priv *priv = dev->data->dev_private;
1258 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
1259 		uint32_t shl_c0;
1260 
1261 		MLX5_ASSERT(msk_c0);
1262 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
1263 		shl_c0 = rte_bsf32(msk_c0);
1264 #else
1265 		shl_c0 = sizeof(msk_c0) * CHAR_BIT - rte_fls_u32(msk_c0);
1266 #endif
1267 		mask <<= shl_c0;
1268 		data <<= shl_c0;
1269 		MLX5_ASSERT(!(~msk_c0 & rte_cpu_to_be_32(mask)));
1270 	}
1271 	reg_c_x[0] = (struct field_modify_info){4, 0, reg_to_field[reg]};
1272 	/* The routine expects parameters in memory as big-endian ones. */
1273 	return flow_dv_convert_modify_action(&item, reg_c_x, NULL, resource,
1274 					     MLX5_MODIFICATION_TYPE_SET, error);
1275 }
1276 
1277 /**
1278  * Convert modify-header set IPv4 DSCP action to DV specification.
1279  *
1280  * @param[in,out] resource
1281  *   Pointer to the modify-header resource.
1282  * @param[in] action
1283  *   Pointer to action specification.
1284  * @param[out] error
1285  *   Pointer to the error structure.
1286  *
1287  * @return
1288  *   0 on success, a negative errno value otherwise and rte_errno is set.
1289  */
1290 static int
1291 flow_dv_convert_action_modify_ipv4_dscp
1292 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
1293 			 const struct rte_flow_action *action,
1294 			 struct rte_flow_error *error)
1295 {
1296 	const struct rte_flow_action_set_dscp *conf =
1297 		(const struct rte_flow_action_set_dscp *)(action->conf);
1298 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV4 };
1299 	struct rte_flow_item_ipv4 ipv4;
1300 	struct rte_flow_item_ipv4 ipv4_mask;
1301 
1302 	memset(&ipv4, 0, sizeof(ipv4));
1303 	memset(&ipv4_mask, 0, sizeof(ipv4_mask));
1304 	ipv4.hdr.type_of_service = conf->dscp;
1305 	ipv4_mask.hdr.type_of_service = RTE_IPV4_HDR_DSCP_MASK >> 2;
1306 	item.spec = &ipv4;
1307 	item.mask = &ipv4_mask;
1308 	return flow_dv_convert_modify_action(&item, modify_ipv4, NULL, resource,
1309 					     MLX5_MODIFICATION_TYPE_SET, error);
1310 }
1311 
1312 /**
1313  * Convert modify-header set IPv6 DSCP action to DV specification.
1314  *
1315  * @param[in,out] resource
1316  *   Pointer to the modify-header resource.
1317  * @param[in] action
1318  *   Pointer to action specification.
1319  * @param[out] error
1320  *   Pointer to the error structure.
1321  *
1322  * @return
1323  *   0 on success, a negative errno value otherwise and rte_errno is set.
1324  */
1325 static int
1326 flow_dv_convert_action_modify_ipv6_dscp
1327 			(struct mlx5_flow_dv_modify_hdr_resource *resource,
1328 			 const struct rte_flow_action *action,
1329 			 struct rte_flow_error *error)
1330 {
1331 	const struct rte_flow_action_set_dscp *conf =
1332 		(const struct rte_flow_action_set_dscp *)(action->conf);
1333 	struct rte_flow_item item = { .type = RTE_FLOW_ITEM_TYPE_IPV6 };
1334 	struct rte_flow_item_ipv6 ipv6;
1335 	struct rte_flow_item_ipv6 ipv6_mask;
1336 
1337 	memset(&ipv6, 0, sizeof(ipv6));
1338 	memset(&ipv6_mask, 0, sizeof(ipv6_mask));
1339 	/*
1340 	 * Even though the DSCP bits offset of IPv6 is not byte aligned,
1341 	 * rdma-core only accept the DSCP bits byte aligned start from
1342 	 * bit 0 to 5 as to be compatible with IPv4. No need to shift the
1343 	 * bits in IPv6 case as rdma-core requires byte aligned value.
1344 	 */
1345 	ipv6.hdr.vtc_flow = conf->dscp;
1346 	ipv6_mask.hdr.vtc_flow = RTE_IPV6_HDR_DSCP_MASK >> 22;
1347 	item.spec = &ipv6;
1348 	item.mask = &ipv6_mask;
1349 	return flow_dv_convert_modify_action(&item, modify_ipv6, NULL, resource,
1350 					     MLX5_MODIFICATION_TYPE_SET, error);
1351 }
1352 
1353 /**
1354  * Validate MARK item.
1355  *
1356  * @param[in] dev
1357  *   Pointer to the rte_eth_dev structure.
1358  * @param[in] item
1359  *   Item specification.
1360  * @param[in] attr
1361  *   Attributes of flow that includes this item.
1362  * @param[out] error
1363  *   Pointer to error structure.
1364  *
1365  * @return
1366  *   0 on success, a negative errno value otherwise and rte_errno is set.
1367  */
1368 static int
1369 flow_dv_validate_item_mark(struct rte_eth_dev *dev,
1370 			   const struct rte_flow_item *item,
1371 			   const struct rte_flow_attr *attr __rte_unused,
1372 			   struct rte_flow_error *error)
1373 {
1374 	struct mlx5_priv *priv = dev->data->dev_private;
1375 	struct mlx5_dev_config *config = &priv->config;
1376 	const struct rte_flow_item_mark *spec = item->spec;
1377 	const struct rte_flow_item_mark *mask = item->mask;
1378 	const struct rte_flow_item_mark nic_mask = {
1379 		.id = priv->sh->dv_mark_mask,
1380 	};
1381 	int ret;
1382 
1383 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
1384 		return rte_flow_error_set(error, ENOTSUP,
1385 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1386 					  "extended metadata feature"
1387 					  " isn't enabled");
1388 	if (!mlx5_flow_ext_mreg_supported(dev))
1389 		return rte_flow_error_set(error, ENOTSUP,
1390 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1391 					  "extended metadata register"
1392 					  " isn't supported");
1393 	if (!nic_mask.id)
1394 		return rte_flow_error_set(error, ENOTSUP,
1395 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1396 					  "extended metadata register"
1397 					  " isn't available");
1398 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
1399 	if (ret < 0)
1400 		return ret;
1401 	if (!spec)
1402 		return rte_flow_error_set(error, EINVAL,
1403 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1404 					  item->spec,
1405 					  "data cannot be empty");
1406 	if (spec->id >= (MLX5_FLOW_MARK_MAX & nic_mask.id))
1407 		return rte_flow_error_set(error, EINVAL,
1408 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1409 					  &spec->id,
1410 					  "mark id exceeds the limit");
1411 	if (!mask)
1412 		mask = &nic_mask;
1413 	if (!mask->id)
1414 		return rte_flow_error_set(error, EINVAL,
1415 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1416 					"mask cannot be zero");
1417 
1418 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1419 					(const uint8_t *)&nic_mask,
1420 					sizeof(struct rte_flow_item_mark),
1421 					error);
1422 	if (ret < 0)
1423 		return ret;
1424 	return 0;
1425 }
1426 
1427 /**
1428  * Validate META item.
1429  *
1430  * @param[in] dev
1431  *   Pointer to the rte_eth_dev structure.
1432  * @param[in] item
1433  *   Item specification.
1434  * @param[in] attr
1435  *   Attributes of flow that includes this item.
1436  * @param[out] error
1437  *   Pointer to error structure.
1438  *
1439  * @return
1440  *   0 on success, a negative errno value otherwise and rte_errno is set.
1441  */
1442 static int
1443 flow_dv_validate_item_meta(struct rte_eth_dev *dev __rte_unused,
1444 			   const struct rte_flow_item *item,
1445 			   const struct rte_flow_attr *attr,
1446 			   struct rte_flow_error *error)
1447 {
1448 	struct mlx5_priv *priv = dev->data->dev_private;
1449 	struct mlx5_dev_config *config = &priv->config;
1450 	const struct rte_flow_item_meta *spec = item->spec;
1451 	const struct rte_flow_item_meta *mask = item->mask;
1452 	struct rte_flow_item_meta nic_mask = {
1453 		.data = UINT32_MAX
1454 	};
1455 	int reg;
1456 	int ret;
1457 
1458 	if (!spec)
1459 		return rte_flow_error_set(error, EINVAL,
1460 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1461 					  item->spec,
1462 					  "data cannot be empty");
1463 	if (config->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1464 		if (!mlx5_flow_ext_mreg_supported(dev))
1465 			return rte_flow_error_set(error, ENOTSUP,
1466 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1467 					  "extended metadata register"
1468 					  " isn't supported");
1469 		reg = flow_dv_get_metadata_reg(dev, attr, error);
1470 		if (reg < 0)
1471 			return reg;
1472 		if (reg == REG_B)
1473 			return rte_flow_error_set(error, ENOTSUP,
1474 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1475 					  "match on reg_b "
1476 					  "isn't supported");
1477 		if (reg != REG_A)
1478 			nic_mask.data = priv->sh->dv_meta_mask;
1479 	} else if (attr->transfer) {
1480 		return rte_flow_error_set(error, ENOTSUP,
1481 					RTE_FLOW_ERROR_TYPE_ITEM, item,
1482 					"extended metadata feature "
1483 					"should be enabled when "
1484 					"meta item is requested "
1485 					"with e-switch mode ");
1486 	}
1487 	if (!mask)
1488 		mask = &rte_flow_item_meta_mask;
1489 	if (!mask->data)
1490 		return rte_flow_error_set(error, EINVAL,
1491 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1492 					"mask cannot be zero");
1493 
1494 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1495 					(const uint8_t *)&nic_mask,
1496 					sizeof(struct rte_flow_item_meta),
1497 					error);
1498 	return ret;
1499 }
1500 
1501 /**
1502  * Validate TAG item.
1503  *
1504  * @param[in] dev
1505  *   Pointer to the rte_eth_dev structure.
1506  * @param[in] item
1507  *   Item specification.
1508  * @param[in] attr
1509  *   Attributes of flow that includes this item.
1510  * @param[out] error
1511  *   Pointer to error structure.
1512  *
1513  * @return
1514  *   0 on success, a negative errno value otherwise and rte_errno is set.
1515  */
1516 static int
1517 flow_dv_validate_item_tag(struct rte_eth_dev *dev,
1518 			  const struct rte_flow_item *item,
1519 			  const struct rte_flow_attr *attr __rte_unused,
1520 			  struct rte_flow_error *error)
1521 {
1522 	const struct rte_flow_item_tag *spec = item->spec;
1523 	const struct rte_flow_item_tag *mask = item->mask;
1524 	const struct rte_flow_item_tag nic_mask = {
1525 		.data = RTE_BE32(UINT32_MAX),
1526 		.index = 0xff,
1527 	};
1528 	int ret;
1529 
1530 	if (!mlx5_flow_ext_mreg_supported(dev))
1531 		return rte_flow_error_set(error, ENOTSUP,
1532 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1533 					  "extensive metadata register"
1534 					  " isn't supported");
1535 	if (!spec)
1536 		return rte_flow_error_set(error, EINVAL,
1537 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1538 					  item->spec,
1539 					  "data cannot be empty");
1540 	if (!mask)
1541 		mask = &rte_flow_item_tag_mask;
1542 	if (!mask->data)
1543 		return rte_flow_error_set(error, EINVAL,
1544 					RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1545 					"mask cannot be zero");
1546 
1547 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1548 					(const uint8_t *)&nic_mask,
1549 					sizeof(struct rte_flow_item_tag),
1550 					error);
1551 	if (ret < 0)
1552 		return ret;
1553 	if (mask->index != 0xff)
1554 		return rte_flow_error_set(error, EINVAL,
1555 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, NULL,
1556 					  "partial mask for tag index"
1557 					  " is not supported");
1558 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, spec->index, error);
1559 	if (ret < 0)
1560 		return ret;
1561 	MLX5_ASSERT(ret != REG_NONE);
1562 	return 0;
1563 }
1564 
1565 /**
1566  * Validate vport item.
1567  *
1568  * @param[in] dev
1569  *   Pointer to the rte_eth_dev structure.
1570  * @param[in] item
1571  *   Item specification.
1572  * @param[in] attr
1573  *   Attributes of flow that includes this item.
1574  * @param[in] item_flags
1575  *   Bit-fields that holds the items detected until now.
1576  * @param[out] error
1577  *   Pointer to error structure.
1578  *
1579  * @return
1580  *   0 on success, a negative errno value otherwise and rte_errno is set.
1581  */
1582 static int
1583 flow_dv_validate_item_port_id(struct rte_eth_dev *dev,
1584 			      const struct rte_flow_item *item,
1585 			      const struct rte_flow_attr *attr,
1586 			      uint64_t item_flags,
1587 			      struct rte_flow_error *error)
1588 {
1589 	const struct rte_flow_item_port_id *spec = item->spec;
1590 	const struct rte_flow_item_port_id *mask = item->mask;
1591 	const struct rte_flow_item_port_id switch_mask = {
1592 			.id = 0xffffffff,
1593 	};
1594 	struct mlx5_priv *esw_priv;
1595 	struct mlx5_priv *dev_priv;
1596 	int ret;
1597 
1598 	if (!attr->transfer)
1599 		return rte_flow_error_set(error, EINVAL,
1600 					  RTE_FLOW_ERROR_TYPE_ITEM,
1601 					  NULL,
1602 					  "match on port id is valid only"
1603 					  " when transfer flag is enabled");
1604 	if (item_flags & MLX5_FLOW_ITEM_PORT_ID)
1605 		return rte_flow_error_set(error, ENOTSUP,
1606 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1607 					  "multiple source ports are not"
1608 					  " supported");
1609 	if (!mask)
1610 		mask = &switch_mask;
1611 	if (mask->id != 0xffffffff)
1612 		return rte_flow_error_set(error, ENOTSUP,
1613 					   RTE_FLOW_ERROR_TYPE_ITEM_MASK,
1614 					   mask,
1615 					   "no support for partial mask on"
1616 					   " \"id\" field");
1617 	ret = mlx5_flow_item_acceptable
1618 				(item, (const uint8_t *)mask,
1619 				 (const uint8_t *)&rte_flow_item_port_id_mask,
1620 				 sizeof(struct rte_flow_item_port_id),
1621 				 error);
1622 	if (ret)
1623 		return ret;
1624 	if (!spec)
1625 		return 0;
1626 	esw_priv = mlx5_port_to_eswitch_info(spec->id, false);
1627 	if (!esw_priv)
1628 		return rte_flow_error_set(error, rte_errno,
1629 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1630 					  "failed to obtain E-Switch info for"
1631 					  " port");
1632 	dev_priv = mlx5_dev_to_eswitch_info(dev);
1633 	if (!dev_priv)
1634 		return rte_flow_error_set(error, rte_errno,
1635 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1636 					  NULL,
1637 					  "failed to obtain E-Switch info");
1638 	if (esw_priv->domain_id != dev_priv->domain_id)
1639 		return rte_flow_error_set(error, EINVAL,
1640 					  RTE_FLOW_ERROR_TYPE_ITEM_SPEC, spec,
1641 					  "cannot match on a port from a"
1642 					  " different E-Switch");
1643 	return 0;
1644 }
1645 
1646 /**
1647  * Validate VLAN item.
1648  *
1649  * @param[in] item
1650  *   Item specification.
1651  * @param[in] item_flags
1652  *   Bit-fields that holds the items detected until now.
1653  * @param[in] dev
1654  *   Ethernet device flow is being created on.
1655  * @param[out] error
1656  *   Pointer to error structure.
1657  *
1658  * @return
1659  *   0 on success, a negative errno value otherwise and rte_errno is set.
1660  */
1661 static int
1662 flow_dv_validate_item_vlan(const struct rte_flow_item *item,
1663 			   uint64_t item_flags,
1664 			   struct rte_eth_dev *dev,
1665 			   struct rte_flow_error *error)
1666 {
1667 	const struct rte_flow_item_vlan *mask = item->mask;
1668 	const struct rte_flow_item_vlan nic_mask = {
1669 		.tci = RTE_BE16(UINT16_MAX),
1670 		.inner_type = RTE_BE16(UINT16_MAX),
1671 	};
1672 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1673 	int ret;
1674 	const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1675 					MLX5_FLOW_LAYER_INNER_L4) :
1676 				       (MLX5_FLOW_LAYER_OUTER_L3 |
1677 					MLX5_FLOW_LAYER_OUTER_L4);
1678 	const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1679 					MLX5_FLOW_LAYER_OUTER_VLAN;
1680 
1681 	if (item_flags & vlanm)
1682 		return rte_flow_error_set(error, EINVAL,
1683 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1684 					  "multiple VLAN layers not supported");
1685 	else if ((item_flags & l34m) != 0)
1686 		return rte_flow_error_set(error, EINVAL,
1687 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1688 					  "VLAN cannot follow L3/L4 layer");
1689 	if (!mask)
1690 		mask = &rte_flow_item_vlan_mask;
1691 	ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1692 					(const uint8_t *)&nic_mask,
1693 					sizeof(struct rte_flow_item_vlan),
1694 					error);
1695 	if (ret)
1696 		return ret;
1697 	if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1698 		struct mlx5_priv *priv = dev->data->dev_private;
1699 
1700 		if (priv->vmwa_context) {
1701 			/*
1702 			 * Non-NULL context means we have a virtual machine
1703 			 * and SR-IOV enabled, we have to create VLAN interface
1704 			 * to make hypervisor to setup E-Switch vport
1705 			 * context correctly. We avoid creating the multiple
1706 			 * VLAN interfaces, so we cannot support VLAN tag mask.
1707 			 */
1708 			return rte_flow_error_set(error, EINVAL,
1709 						  RTE_FLOW_ERROR_TYPE_ITEM,
1710 						  item,
1711 						  "VLAN tag mask is not"
1712 						  " supported in virtual"
1713 						  " environment");
1714 		}
1715 	}
1716 	return 0;
1717 }
1718 
1719 /*
1720  * GTP flags are contained in 1 byte of the format:
1721  * -------------------------------------------
1722  * | bit   | 0 - 2   | 3  | 4   | 5 | 6 | 7  |
1723  * |-----------------------------------------|
1724  * | value | Version | PT | Res | E | S | PN |
1725  * -------------------------------------------
1726  *
1727  * Matching is supported only for GTP flags E, S, PN.
1728  */
1729 #define MLX5_GTP_FLAGS_MASK	0x07
1730 
1731 /**
1732  * Validate GTP item.
1733  *
1734  * @param[in] dev
1735  *   Pointer to the rte_eth_dev structure.
1736  * @param[in] item
1737  *   Item specification.
1738  * @param[in] item_flags
1739  *   Bit-fields that holds the items detected until now.
1740  * @param[out] error
1741  *   Pointer to error structure.
1742  *
1743  * @return
1744  *   0 on success, a negative errno value otherwise and rte_errno is set.
1745  */
1746 static int
1747 flow_dv_validate_item_gtp(struct rte_eth_dev *dev,
1748 			  const struct rte_flow_item *item,
1749 			  uint64_t item_flags,
1750 			  struct rte_flow_error *error)
1751 {
1752 	struct mlx5_priv *priv = dev->data->dev_private;
1753 	const struct rte_flow_item_gtp *spec = item->spec;
1754 	const struct rte_flow_item_gtp *mask = item->mask;
1755 	const struct rte_flow_item_gtp nic_mask = {
1756 		.v_pt_rsv_flags = MLX5_GTP_FLAGS_MASK,
1757 		.msg_type = 0xff,
1758 		.teid = RTE_BE32(0xffffffff),
1759 	};
1760 
1761 	if (!priv->config.hca_attr.tunnel_stateless_gtp)
1762 		return rte_flow_error_set(error, ENOTSUP,
1763 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1764 					  "GTP support is not enabled");
1765 	if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1766 		return rte_flow_error_set(error, ENOTSUP,
1767 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1768 					  "multiple tunnel layers not"
1769 					  " supported");
1770 	if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1771 		return rte_flow_error_set(error, EINVAL,
1772 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1773 					  "no outer UDP layer found");
1774 	if (!mask)
1775 		mask = &rte_flow_item_gtp_mask;
1776 	if (spec && spec->v_pt_rsv_flags & ~MLX5_GTP_FLAGS_MASK)
1777 		return rte_flow_error_set(error, ENOTSUP,
1778 					  RTE_FLOW_ERROR_TYPE_ITEM, item,
1779 					  "Match is supported for GTP"
1780 					  " flags only");
1781 	return mlx5_flow_item_acceptable
1782 		(item, (const uint8_t *)mask,
1783 		 (const uint8_t *)&nic_mask,
1784 		 sizeof(struct rte_flow_item_gtp),
1785 		 error);
1786 }
1787 
1788 /**
1789  * Validate the pop VLAN action.
1790  *
1791  * @param[in] dev
1792  *   Pointer to the rte_eth_dev structure.
1793  * @param[in] action_flags
1794  *   Holds the actions detected until now.
1795  * @param[in] action
1796  *   Pointer to the pop vlan action.
1797  * @param[in] item_flags
1798  *   The items found in this flow rule.
1799  * @param[in] attr
1800  *   Pointer to flow attributes.
1801  * @param[out] error
1802  *   Pointer to error structure.
1803  *
1804  * @return
1805  *   0 on success, a negative errno value otherwise and rte_errno is set.
1806  */
1807 static int
1808 flow_dv_validate_action_pop_vlan(struct rte_eth_dev *dev,
1809 				 uint64_t action_flags,
1810 				 const struct rte_flow_action *action,
1811 				 uint64_t item_flags,
1812 				 const struct rte_flow_attr *attr,
1813 				 struct rte_flow_error *error)
1814 {
1815 	const struct mlx5_priv *priv = dev->data->dev_private;
1816 
1817 	(void)action;
1818 	(void)attr;
1819 	if (!priv->sh->pop_vlan_action)
1820 		return rte_flow_error_set(error, ENOTSUP,
1821 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1822 					  NULL,
1823 					  "pop vlan action is not supported");
1824 	if (attr->egress)
1825 		return rte_flow_error_set(error, ENOTSUP,
1826 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
1827 					  NULL,
1828 					  "pop vlan action not supported for "
1829 					  "egress");
1830 	if (action_flags & MLX5_FLOW_VLAN_ACTIONS)
1831 		return rte_flow_error_set(error, ENOTSUP,
1832 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1833 					  "no support for multiple VLAN "
1834 					  "actions");
1835 	/* Pop VLAN with preceding Decap requires inner header with VLAN. */
1836 	if ((action_flags & MLX5_FLOW_ACTION_DECAP) &&
1837 	    !(item_flags & MLX5_FLOW_LAYER_INNER_VLAN))
1838 		return rte_flow_error_set(error, ENOTSUP,
1839 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1840 					  NULL,
1841 					  "cannot pop vlan after decap without "
1842 					  "match on inner vlan in the flow");
1843 	/* Pop VLAN without preceding Decap requires outer header with VLAN. */
1844 	if (!(action_flags & MLX5_FLOW_ACTION_DECAP) &&
1845 	    !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
1846 		return rte_flow_error_set(error, ENOTSUP,
1847 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1848 					  NULL,
1849 					  "cannot pop vlan without a "
1850 					  "match on (outer) vlan in the flow");
1851 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1852 		return rte_flow_error_set(error, EINVAL,
1853 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1854 					  "wrong action order, port_id should "
1855 					  "be after pop VLAN action");
1856 	if (!attr->transfer && priv->representor)
1857 		return rte_flow_error_set(error, ENOTSUP,
1858 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1859 					  "pop vlan action for VF representor "
1860 					  "not supported on NIC table");
1861 	return 0;
1862 }
1863 
1864 /**
1865  * Get VLAN default info from vlan match info.
1866  *
1867  * @param[in] items
1868  *   the list of item specifications.
1869  * @param[out] vlan
1870  *   pointer VLAN info to fill to.
1871  *
1872  * @return
1873  *   0 on success, a negative errno value otherwise and rte_errno is set.
1874  */
1875 static void
1876 flow_dev_get_vlan_info_from_items(const struct rte_flow_item *items,
1877 				  struct rte_vlan_hdr *vlan)
1878 {
1879 	const struct rte_flow_item_vlan nic_mask = {
1880 		.tci = RTE_BE16(MLX5DV_FLOW_VLAN_PCP_MASK |
1881 				MLX5DV_FLOW_VLAN_VID_MASK),
1882 		.inner_type = RTE_BE16(0xffff),
1883 	};
1884 
1885 	if (items == NULL)
1886 		return;
1887 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1888 		int type = items->type;
1889 
1890 		if (type == RTE_FLOW_ITEM_TYPE_VLAN ||
1891 		    type == MLX5_RTE_FLOW_ITEM_TYPE_VLAN)
1892 			break;
1893 	}
1894 	if (items->type != RTE_FLOW_ITEM_TYPE_END) {
1895 		const struct rte_flow_item_vlan *vlan_m = items->mask;
1896 		const struct rte_flow_item_vlan *vlan_v = items->spec;
1897 
1898 		/* If VLAN item in pattern doesn't contain data, return here. */
1899 		if (!vlan_v)
1900 			return;
1901 		if (!vlan_m)
1902 			vlan_m = &nic_mask;
1903 		/* Only full match values are accepted */
1904 		if ((vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) ==
1905 		     MLX5DV_FLOW_VLAN_PCP_MASK_BE) {
1906 			vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_PCP_MASK;
1907 			vlan->vlan_tci |=
1908 				rte_be_to_cpu_16(vlan_v->tci &
1909 						 MLX5DV_FLOW_VLAN_PCP_MASK_BE);
1910 		}
1911 		if ((vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) ==
1912 		     MLX5DV_FLOW_VLAN_VID_MASK_BE) {
1913 			vlan->vlan_tci &= ~MLX5DV_FLOW_VLAN_VID_MASK;
1914 			vlan->vlan_tci |=
1915 				rte_be_to_cpu_16(vlan_v->tci &
1916 						 MLX5DV_FLOW_VLAN_VID_MASK_BE);
1917 		}
1918 		if (vlan_m->inner_type == nic_mask.inner_type)
1919 			vlan->eth_proto = rte_be_to_cpu_16(vlan_v->inner_type &
1920 							   vlan_m->inner_type);
1921 	}
1922 }
1923 
1924 /**
1925  * Validate the push VLAN action.
1926  *
1927  * @param[in] dev
1928  *   Pointer to the rte_eth_dev structure.
1929  * @param[in] action_flags
1930  *   Holds the actions detected until now.
1931  * @param[in] item_flags
1932  *   The items found in this flow rule.
1933  * @param[in] action
1934  *   Pointer to the action structure.
1935  * @param[in] attr
1936  *   Pointer to flow attributes
1937  * @param[out] error
1938  *   Pointer to error structure.
1939  *
1940  * @return
1941  *   0 on success, a negative errno value otherwise and rte_errno is set.
1942  */
1943 static int
1944 flow_dv_validate_action_push_vlan(struct rte_eth_dev *dev,
1945 				  uint64_t action_flags,
1946 				  const struct rte_flow_item_vlan *vlan_m,
1947 				  const struct rte_flow_action *action,
1948 				  const struct rte_flow_attr *attr,
1949 				  struct rte_flow_error *error)
1950 {
1951 	const struct rte_flow_action_of_push_vlan *push_vlan = action->conf;
1952 	const struct mlx5_priv *priv = dev->data->dev_private;
1953 
1954 	if (push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_VLAN) &&
1955 	    push_vlan->ethertype != RTE_BE16(RTE_ETHER_TYPE_QINQ))
1956 		return rte_flow_error_set(error, EINVAL,
1957 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1958 					  "invalid vlan ethertype");
1959 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
1960 		return rte_flow_error_set(error, EINVAL,
1961 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1962 					  "wrong action order, port_id should "
1963 					  "be after push VLAN");
1964 	if (!attr->transfer && priv->representor)
1965 		return rte_flow_error_set(error, ENOTSUP,
1966 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1967 					  "push vlan action for VF representor "
1968 					  "not supported on NIC table");
1969 	if (vlan_m &&
1970 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) &&
1971 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_PCP_MASK_BE) !=
1972 		MLX5DV_FLOW_VLAN_PCP_MASK_BE &&
1973 	    !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP) &&
1974 	    !(mlx5_flow_find_action
1975 		(action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP)))
1976 		return rte_flow_error_set(error, EINVAL,
1977 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1978 					  "not full match mask on VLAN PCP and "
1979 					  "there is no of_set_vlan_pcp action, "
1980 					  "push VLAN action cannot figure out "
1981 					  "PCP value");
1982 	if (vlan_m &&
1983 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) &&
1984 	    (vlan_m->tci & MLX5DV_FLOW_VLAN_VID_MASK_BE) !=
1985 		MLX5DV_FLOW_VLAN_VID_MASK_BE &&
1986 	    !(action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID) &&
1987 	    !(mlx5_flow_find_action
1988 		(action + 1, RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID)))
1989 		return rte_flow_error_set(error, EINVAL,
1990 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
1991 					  "not full match mask on VLAN VID and "
1992 					  "there is no of_set_vlan_vid action, "
1993 					  "push VLAN action cannot figure out "
1994 					  "VID value");
1995 	(void)attr;
1996 	return 0;
1997 }
1998 
1999 /**
2000  * Validate the set VLAN PCP.
2001  *
2002  * @param[in] action_flags
2003  *   Holds the actions detected until now.
2004  * @param[in] actions
2005  *   Pointer to the list of actions remaining in the flow rule.
2006  * @param[out] error
2007  *   Pointer to error structure.
2008  *
2009  * @return
2010  *   0 on success, a negative errno value otherwise and rte_errno is set.
2011  */
2012 static int
2013 flow_dv_validate_action_set_vlan_pcp(uint64_t action_flags,
2014 				     const struct rte_flow_action actions[],
2015 				     struct rte_flow_error *error)
2016 {
2017 	const struct rte_flow_action *action = actions;
2018 	const struct rte_flow_action_of_set_vlan_pcp *conf = action->conf;
2019 
2020 	if (conf->vlan_pcp > 7)
2021 		return rte_flow_error_set(error, EINVAL,
2022 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2023 					  "VLAN PCP value is too big");
2024 	if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
2025 		return rte_flow_error_set(error, ENOTSUP,
2026 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2027 					  "set VLAN PCP action must follow "
2028 					  "the push VLAN action");
2029 	if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
2030 		return rte_flow_error_set(error, ENOTSUP,
2031 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2032 					  "Multiple VLAN PCP modification are "
2033 					  "not supported");
2034 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2035 		return rte_flow_error_set(error, EINVAL,
2036 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2037 					  "wrong action order, port_id should "
2038 					  "be after set VLAN PCP");
2039 	return 0;
2040 }
2041 
2042 /**
2043  * Validate the set VLAN VID.
2044  *
2045  * @param[in] item_flags
2046  *   Holds the items detected in this rule.
2047  * @param[in] action_flags
2048  *   Holds the actions detected until now.
2049  * @param[in] actions
2050  *   Pointer to the list of actions remaining in the flow rule.
2051  * @param[out] error
2052  *   Pointer to error structure.
2053  *
2054  * @return
2055  *   0 on success, a negative errno value otherwise and rte_errno is set.
2056  */
2057 static int
2058 flow_dv_validate_action_set_vlan_vid(uint64_t item_flags,
2059 				     uint64_t action_flags,
2060 				     const struct rte_flow_action actions[],
2061 				     struct rte_flow_error *error)
2062 {
2063 	const struct rte_flow_action *action = actions;
2064 	const struct rte_flow_action_of_set_vlan_vid *conf = action->conf;
2065 
2066 	if (rte_be_to_cpu_16(conf->vlan_vid) > 0xFFE)
2067 		return rte_flow_error_set(error, EINVAL,
2068 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2069 					  "VLAN VID value is too big");
2070 	if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2071 	    !(item_flags & MLX5_FLOW_LAYER_OUTER_VLAN))
2072 		return rte_flow_error_set(error, ENOTSUP,
2073 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2074 					  "set VLAN VID action must follow push"
2075 					  " VLAN action or match on VLAN item");
2076 	if (action_flags & MLX5_FLOW_ACTION_OF_SET_VLAN_VID)
2077 		return rte_flow_error_set(error, ENOTSUP,
2078 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2079 					  "Multiple VLAN VID modifications are "
2080 					  "not supported");
2081 	if (action_flags & MLX5_FLOW_ACTION_PORT_ID)
2082 		return rte_flow_error_set(error, EINVAL,
2083 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2084 					  "wrong action order, port_id should "
2085 					  "be after set VLAN VID");
2086 	return 0;
2087 }
2088 
2089 /*
2090  * Validate the FLAG action.
2091  *
2092  * @param[in] dev
2093  *   Pointer to the rte_eth_dev structure.
2094  * @param[in] action_flags
2095  *   Holds the actions detected until now.
2096  * @param[in] attr
2097  *   Pointer to flow attributes
2098  * @param[out] error
2099  *   Pointer to error structure.
2100  *
2101  * @return
2102  *   0 on success, a negative errno value otherwise and rte_errno is set.
2103  */
2104 static int
2105 flow_dv_validate_action_flag(struct rte_eth_dev *dev,
2106 			     uint64_t action_flags,
2107 			     const struct rte_flow_attr *attr,
2108 			     struct rte_flow_error *error)
2109 {
2110 	struct mlx5_priv *priv = dev->data->dev_private;
2111 	struct mlx5_dev_config *config = &priv->config;
2112 	int ret;
2113 
2114 	/* Fall back if no extended metadata register support. */
2115 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2116 		return mlx5_flow_validate_action_flag(action_flags, attr,
2117 						      error);
2118 	/* Extensive metadata mode requires registers. */
2119 	if (!mlx5_flow_ext_mreg_supported(dev))
2120 		return rte_flow_error_set(error, ENOTSUP,
2121 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2122 					  "no metadata registers "
2123 					  "to support flag action");
2124 	if (!(priv->sh->dv_mark_mask & MLX5_FLOW_MARK_DEFAULT))
2125 		return rte_flow_error_set(error, ENOTSUP,
2126 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2127 					  "extended metadata register"
2128 					  " isn't available");
2129 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2130 	if (ret < 0)
2131 		return ret;
2132 	MLX5_ASSERT(ret > 0);
2133 	if (action_flags & MLX5_FLOW_ACTION_MARK)
2134 		return rte_flow_error_set(error, EINVAL,
2135 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2136 					  "can't mark and flag in same flow");
2137 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
2138 		return rte_flow_error_set(error, EINVAL,
2139 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2140 					  "can't have 2 flag"
2141 					  " actions in same flow");
2142 	return 0;
2143 }
2144 
2145 /**
2146  * Validate MARK action.
2147  *
2148  * @param[in] dev
2149  *   Pointer to the rte_eth_dev structure.
2150  * @param[in] action
2151  *   Pointer to action.
2152  * @param[in] action_flags
2153  *   Holds the actions detected until now.
2154  * @param[in] attr
2155  *   Pointer to flow attributes
2156  * @param[out] error
2157  *   Pointer to error structure.
2158  *
2159  * @return
2160  *   0 on success, a negative errno value otherwise and rte_errno is set.
2161  */
2162 static int
2163 flow_dv_validate_action_mark(struct rte_eth_dev *dev,
2164 			     const struct rte_flow_action *action,
2165 			     uint64_t action_flags,
2166 			     const struct rte_flow_attr *attr,
2167 			     struct rte_flow_error *error)
2168 {
2169 	struct mlx5_priv *priv = dev->data->dev_private;
2170 	struct mlx5_dev_config *config = &priv->config;
2171 	const struct rte_flow_action_mark *mark = action->conf;
2172 	int ret;
2173 
2174 	/* Fall back if no extended metadata register support. */
2175 	if (config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY)
2176 		return mlx5_flow_validate_action_mark(action, action_flags,
2177 						      attr, error);
2178 	/* Extensive metadata mode requires registers. */
2179 	if (!mlx5_flow_ext_mreg_supported(dev))
2180 		return rte_flow_error_set(error, ENOTSUP,
2181 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2182 					  "no metadata registers "
2183 					  "to support mark action");
2184 	if (!priv->sh->dv_mark_mask)
2185 		return rte_flow_error_set(error, ENOTSUP,
2186 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2187 					  "extended metadata register"
2188 					  " isn't available");
2189 	ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2190 	if (ret < 0)
2191 		return ret;
2192 	MLX5_ASSERT(ret > 0);
2193 	if (!mark)
2194 		return rte_flow_error_set(error, EINVAL,
2195 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2196 					  "configuration cannot be null");
2197 	if (mark->id >= (MLX5_FLOW_MARK_MAX & priv->sh->dv_mark_mask))
2198 		return rte_flow_error_set(error, EINVAL,
2199 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
2200 					  &mark->id,
2201 					  "mark id exceeds the limit");
2202 	if (action_flags & MLX5_FLOW_ACTION_FLAG)
2203 		return rte_flow_error_set(error, EINVAL,
2204 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2205 					  "can't flag and mark in same flow");
2206 	if (action_flags & MLX5_FLOW_ACTION_MARK)
2207 		return rte_flow_error_set(error, EINVAL,
2208 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2209 					  "can't have 2 mark actions in same"
2210 					  " flow");
2211 	return 0;
2212 }
2213 
2214 /**
2215  * Validate SET_META action.
2216  *
2217  * @param[in] dev
2218  *   Pointer to the rte_eth_dev structure.
2219  * @param[in] action
2220  *   Pointer to the action structure.
2221  * @param[in] action_flags
2222  *   Holds the actions detected until now.
2223  * @param[in] attr
2224  *   Pointer to flow attributes
2225  * @param[out] error
2226  *   Pointer to error structure.
2227  *
2228  * @return
2229  *   0 on success, a negative errno value otherwise and rte_errno is set.
2230  */
2231 static int
2232 flow_dv_validate_action_set_meta(struct rte_eth_dev *dev,
2233 				 const struct rte_flow_action *action,
2234 				 uint64_t action_flags __rte_unused,
2235 				 const struct rte_flow_attr *attr,
2236 				 struct rte_flow_error *error)
2237 {
2238 	const struct rte_flow_action_set_meta *conf;
2239 	uint32_t nic_mask = UINT32_MAX;
2240 	int reg;
2241 
2242 	if (!mlx5_flow_ext_mreg_supported(dev))
2243 		return rte_flow_error_set(error, ENOTSUP,
2244 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2245 					  "extended metadata register"
2246 					  " isn't supported");
2247 	reg = flow_dv_get_metadata_reg(dev, attr, error);
2248 	if (reg < 0)
2249 		return reg;
2250 	if (reg != REG_A && reg != REG_B) {
2251 		struct mlx5_priv *priv = dev->data->dev_private;
2252 
2253 		nic_mask = priv->sh->dv_meta_mask;
2254 	}
2255 	if (!(action->conf))
2256 		return rte_flow_error_set(error, EINVAL,
2257 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2258 					  "configuration cannot be null");
2259 	conf = (const struct rte_flow_action_set_meta *)action->conf;
2260 	if (!conf->mask)
2261 		return rte_flow_error_set(error, EINVAL,
2262 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2263 					  "zero mask doesn't have any effect");
2264 	if (conf->mask & ~nic_mask)
2265 		return rte_flow_error_set(error, EINVAL,
2266 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2267 					  "meta data must be within reg C0");
2268 	return 0;
2269 }
2270 
2271 /**
2272  * Validate SET_TAG action.
2273  *
2274  * @param[in] dev
2275  *   Pointer to the rte_eth_dev structure.
2276  * @param[in] action
2277  *   Pointer to the action structure.
2278  * @param[in] action_flags
2279  *   Holds the actions detected until now.
2280  * @param[in] attr
2281  *   Pointer to flow attributes
2282  * @param[out] error
2283  *   Pointer to error structure.
2284  *
2285  * @return
2286  *   0 on success, a negative errno value otherwise and rte_errno is set.
2287  */
2288 static int
2289 flow_dv_validate_action_set_tag(struct rte_eth_dev *dev,
2290 				const struct rte_flow_action *action,
2291 				uint64_t action_flags,
2292 				const struct rte_flow_attr *attr,
2293 				struct rte_flow_error *error)
2294 {
2295 	const struct rte_flow_action_set_tag *conf;
2296 	const uint64_t terminal_action_flags =
2297 		MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_QUEUE |
2298 		MLX5_FLOW_ACTION_RSS;
2299 	int ret;
2300 
2301 	if (!mlx5_flow_ext_mreg_supported(dev))
2302 		return rte_flow_error_set(error, ENOTSUP,
2303 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2304 					  "extensive metadata register"
2305 					  " isn't supported");
2306 	if (!(action->conf))
2307 		return rte_flow_error_set(error, EINVAL,
2308 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2309 					  "configuration cannot be null");
2310 	conf = (const struct rte_flow_action_set_tag *)action->conf;
2311 	if (!conf->mask)
2312 		return rte_flow_error_set(error, EINVAL,
2313 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2314 					  "zero mask doesn't have any effect");
2315 	ret = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, conf->index, error);
2316 	if (ret < 0)
2317 		return ret;
2318 	if (!attr->transfer && attr->ingress &&
2319 	    (action_flags & terminal_action_flags))
2320 		return rte_flow_error_set(error, EINVAL,
2321 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2322 					  "set_tag has no effect"
2323 					  " with terminal actions");
2324 	return 0;
2325 }
2326 
2327 /**
2328  * Validate count action.
2329  *
2330  * @param[in] dev
2331  *   Pointer to rte_eth_dev structure.
2332  * @param[out] error
2333  *   Pointer to error structure.
2334  *
2335  * @return
2336  *   0 on success, a negative errno value otherwise and rte_errno is set.
2337  */
2338 static int
2339 flow_dv_validate_action_count(struct rte_eth_dev *dev,
2340 			      struct rte_flow_error *error)
2341 {
2342 	struct mlx5_priv *priv = dev->data->dev_private;
2343 
2344 	if (!priv->config.devx)
2345 		goto notsup_err;
2346 #ifdef HAVE_IBV_FLOW_DEVX_COUNTERS
2347 	return 0;
2348 #endif
2349 notsup_err:
2350 	return rte_flow_error_set
2351 		      (error, ENOTSUP,
2352 		       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2353 		       NULL,
2354 		       "count action not supported");
2355 }
2356 
2357 /**
2358  * Validate the L2 encap action.
2359  *
2360  * @param[in] dev
2361  *   Pointer to the rte_eth_dev structure.
2362  * @param[in] action_flags
2363  *   Holds the actions detected until now.
2364  * @param[in] action
2365  *   Pointer to the action structure.
2366  * @param[in] attr
2367  *   Pointer to flow attributes.
2368  * @param[out] error
2369  *   Pointer to error structure.
2370  *
2371  * @return
2372  *   0 on success, a negative errno value otherwise and rte_errno is set.
2373  */
2374 static int
2375 flow_dv_validate_action_l2_encap(struct rte_eth_dev *dev,
2376 				 uint64_t action_flags,
2377 				 const struct rte_flow_action *action,
2378 				 const struct rte_flow_attr *attr,
2379 				 struct rte_flow_error *error)
2380 {
2381 	const struct mlx5_priv *priv = dev->data->dev_private;
2382 
2383 	if (!(action->conf))
2384 		return rte_flow_error_set(error, EINVAL,
2385 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
2386 					  "configuration cannot be null");
2387 	if (action_flags & MLX5_FLOW_ACTION_ENCAP)
2388 		return rte_flow_error_set(error, EINVAL,
2389 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2390 					  "can only have a single encap action "
2391 					  "in a flow");
2392 	if (!attr->transfer && priv->representor)
2393 		return rte_flow_error_set(error, ENOTSUP,
2394 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2395 					  "encap action for VF representor "
2396 					  "not supported on NIC table");
2397 	return 0;
2398 }
2399 
2400 /**
2401  * Validate a decap action.
2402  *
2403  * @param[in] dev
2404  *   Pointer to the rte_eth_dev structure.
2405  * @param[in] action_flags
2406  *   Holds the actions detected until now.
2407  * @param[in] attr
2408  *   Pointer to flow attributes
2409  * @param[out] error
2410  *   Pointer to error structure.
2411  *
2412  * @return
2413  *   0 on success, a negative errno value otherwise and rte_errno is set.
2414  */
2415 static int
2416 flow_dv_validate_action_decap(struct rte_eth_dev *dev,
2417 			      uint64_t action_flags,
2418 			      const struct rte_flow_attr *attr,
2419 			      struct rte_flow_error *error)
2420 {
2421 	const struct mlx5_priv *priv = dev->data->dev_private;
2422 
2423 	if (priv->config.hca_attr.scatter_fcs_w_decap_disable &&
2424 	    !priv->config.decap_en)
2425 		return rte_flow_error_set(error, ENOTSUP,
2426 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2427 					  "decap is not enabled");
2428 	if (action_flags & MLX5_FLOW_XCAP_ACTIONS)
2429 		return rte_flow_error_set(error, ENOTSUP,
2430 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2431 					  action_flags &
2432 					  MLX5_FLOW_ACTION_DECAP ? "can only "
2433 					  "have a single decap action" : "decap "
2434 					  "after encap is not supported");
2435 	if (action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS)
2436 		return rte_flow_error_set(error, EINVAL,
2437 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2438 					  "can't have decap action after"
2439 					  " modify action");
2440 	if (attr->egress)
2441 		return rte_flow_error_set(error, ENOTSUP,
2442 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
2443 					  NULL,
2444 					  "decap action not supported for "
2445 					  "egress");
2446 	if (!attr->transfer && priv->representor)
2447 		return rte_flow_error_set(error, ENOTSUP,
2448 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2449 					  "decap action for VF representor "
2450 					  "not supported on NIC table");
2451 	return 0;
2452 }
2453 
2454 const struct rte_flow_action_raw_decap empty_decap = {.data = NULL, .size = 0,};
2455 
2456 /**
2457  * Validate the raw encap and decap actions.
2458  *
2459  * @param[in] dev
2460  *   Pointer to the rte_eth_dev structure.
2461  * @param[in] decap
2462  *   Pointer to the decap action.
2463  * @param[in] encap
2464  *   Pointer to the encap action.
2465  * @param[in] attr
2466  *   Pointer to flow attributes
2467  * @param[in/out] action_flags
2468  *   Holds the actions detected until now.
2469  * @param[out] actions_n
2470  *   pointer to the number of actions counter.
2471  * @param[out] error
2472  *   Pointer to error structure.
2473  *
2474  * @return
2475  *   0 on success, a negative errno value otherwise and rte_errno is set.
2476  */
2477 static int
2478 flow_dv_validate_action_raw_encap_decap
2479 	(struct rte_eth_dev *dev,
2480 	 const struct rte_flow_action_raw_decap *decap,
2481 	 const struct rte_flow_action_raw_encap *encap,
2482 	 const struct rte_flow_attr *attr, uint64_t *action_flags,
2483 	 int *actions_n, struct rte_flow_error *error)
2484 {
2485 	const struct mlx5_priv *priv = dev->data->dev_private;
2486 	int ret;
2487 
2488 	if (encap && (!encap->size || !encap->data))
2489 		return rte_flow_error_set(error, EINVAL,
2490 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2491 					  "raw encap data cannot be empty");
2492 	if (decap && encap) {
2493 		if (decap->size <= MLX5_ENCAPSULATION_DECISION_SIZE &&
2494 		    encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2495 			/* L3 encap. */
2496 			decap = NULL;
2497 		else if (encap->size <=
2498 			   MLX5_ENCAPSULATION_DECISION_SIZE &&
2499 			   decap->size >
2500 			   MLX5_ENCAPSULATION_DECISION_SIZE)
2501 			/* L3 decap. */
2502 			encap = NULL;
2503 		else if (encap->size >
2504 			   MLX5_ENCAPSULATION_DECISION_SIZE &&
2505 			   decap->size >
2506 			   MLX5_ENCAPSULATION_DECISION_SIZE)
2507 			/* 2 L2 actions: encap and decap. */
2508 			;
2509 		else
2510 			return rte_flow_error_set(error,
2511 				ENOTSUP,
2512 				RTE_FLOW_ERROR_TYPE_ACTION,
2513 				NULL, "unsupported too small "
2514 				"raw decap and too small raw "
2515 				"encap combination");
2516 	}
2517 	if (decap) {
2518 		ret = flow_dv_validate_action_decap(dev, *action_flags, attr,
2519 						    error);
2520 		if (ret < 0)
2521 			return ret;
2522 		*action_flags |= MLX5_FLOW_ACTION_DECAP;
2523 		++(*actions_n);
2524 	}
2525 	if (encap) {
2526 		if (encap->size <= MLX5_ENCAPSULATION_DECISION_SIZE)
2527 			return rte_flow_error_set(error, ENOTSUP,
2528 						  RTE_FLOW_ERROR_TYPE_ACTION,
2529 						  NULL,
2530 						  "small raw encap size");
2531 		if (*action_flags & MLX5_FLOW_ACTION_ENCAP)
2532 			return rte_flow_error_set(error, EINVAL,
2533 						  RTE_FLOW_ERROR_TYPE_ACTION,
2534 						  NULL,
2535 						  "more than one encap action");
2536 		if (!attr->transfer && priv->representor)
2537 			return rte_flow_error_set
2538 					(error, ENOTSUP,
2539 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2540 					 "encap action for VF representor "
2541 					 "not supported on NIC table");
2542 		*action_flags |= MLX5_FLOW_ACTION_ENCAP;
2543 		++(*actions_n);
2544 	}
2545 	return 0;
2546 }
2547 
2548 /**
2549  * Find existing encap/decap resource or create and register a new one.
2550  *
2551  * @param[in, out] dev
2552  *   Pointer to rte_eth_dev structure.
2553  * @param[in, out] resource
2554  *   Pointer to encap/decap resource.
2555  * @parm[in, out] dev_flow
2556  *   Pointer to the dev_flow.
2557  * @param[out] error
2558  *   pointer to error structure.
2559  *
2560  * @return
2561  *   0 on success otherwise -errno and errno is set.
2562  */
2563 static int
2564 flow_dv_encap_decap_resource_register
2565 			(struct rte_eth_dev *dev,
2566 			 struct mlx5_flow_dv_encap_decap_resource *resource,
2567 			 struct mlx5_flow *dev_flow,
2568 			 struct rte_flow_error *error)
2569 {
2570 	struct mlx5_priv *priv = dev->data->dev_private;
2571 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2572 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
2573 	struct mlx5dv_dr_domain *domain;
2574 	uint32_t idx = 0;
2575 	int ret;
2576 
2577 	resource->flags = dev_flow->dv.group ? 0 : 1;
2578 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2579 		domain = sh->fdb_domain;
2580 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2581 		domain = sh->rx_domain;
2582 	else
2583 		domain = sh->tx_domain;
2584 	/* Lookup a matching resource from cache. */
2585 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], sh->encaps_decaps, idx,
2586 		      cache_resource, next) {
2587 		if (resource->reformat_type == cache_resource->reformat_type &&
2588 		    resource->ft_type == cache_resource->ft_type &&
2589 		    resource->flags == cache_resource->flags &&
2590 		    resource->size == cache_resource->size &&
2591 		    !memcmp((const void *)resource->buf,
2592 			    (const void *)cache_resource->buf,
2593 			    resource->size)) {
2594 			DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d++",
2595 				(void *)cache_resource,
2596 				rte_atomic32_read(&cache_resource->refcnt));
2597 			rte_atomic32_inc(&cache_resource->refcnt);
2598 			dev_flow->handle->dvh.rix_encap_decap = idx;
2599 			dev_flow->dv.encap_decap = cache_resource;
2600 			return 0;
2601 		}
2602 	}
2603 	/* Register new encap/decap resource. */
2604 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
2605 				       &dev_flow->handle->dvh.rix_encap_decap);
2606 	if (!cache_resource)
2607 		return rte_flow_error_set(error, ENOMEM,
2608 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2609 					  "cannot allocate resource memory");
2610 	*cache_resource = *resource;
2611 	ret = mlx5_flow_os_create_flow_action_packet_reformat
2612 					(sh->ctx, domain, cache_resource,
2613 					 &cache_resource->action);
2614 	if (ret) {
2615 		mlx5_free(cache_resource);
2616 		return rte_flow_error_set(error, ENOMEM,
2617 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2618 					  NULL, "cannot create action");
2619 	}
2620 	rte_atomic32_init(&cache_resource->refcnt);
2621 	rte_atomic32_inc(&cache_resource->refcnt);
2622 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_DECAP_ENCAP], &sh->encaps_decaps,
2623 		     dev_flow->handle->dvh.rix_encap_decap, cache_resource,
2624 		     next);
2625 	dev_flow->dv.encap_decap = cache_resource;
2626 	DRV_LOG(DEBUG, "new encap/decap resource %p: refcnt %d++",
2627 		(void *)cache_resource,
2628 		rte_atomic32_read(&cache_resource->refcnt));
2629 	return 0;
2630 }
2631 
2632 /**
2633  * Find existing table jump resource or create and register a new one.
2634  *
2635  * @param[in, out] dev
2636  *   Pointer to rte_eth_dev structure.
2637  * @param[in, out] tbl
2638  *   Pointer to flow table resource.
2639  * @parm[in, out] dev_flow
2640  *   Pointer to the dev_flow.
2641  * @param[out] error
2642  *   pointer to error structure.
2643  *
2644  * @return
2645  *   0 on success otherwise -errno and errno is set.
2646  */
2647 static int
2648 flow_dv_jump_tbl_resource_register
2649 			(struct rte_eth_dev *dev __rte_unused,
2650 			 struct mlx5_flow_tbl_resource *tbl,
2651 			 struct mlx5_flow *dev_flow,
2652 			 struct rte_flow_error *error)
2653 {
2654 	struct mlx5_flow_tbl_data_entry *tbl_data =
2655 		container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
2656 	int cnt, ret;
2657 
2658 	MLX5_ASSERT(tbl);
2659 	cnt = rte_atomic32_read(&tbl_data->jump.refcnt);
2660 	if (!cnt) {
2661 		ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
2662 				(tbl->obj, &tbl_data->jump.action);
2663 		if (ret)
2664 			return rte_flow_error_set(error, ENOMEM,
2665 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2666 					NULL, "cannot create jump action");
2667 		DRV_LOG(DEBUG, "new jump table resource %p: refcnt %d++",
2668 			(void *)&tbl_data->jump, cnt);
2669 	} else {
2670 		/* old jump should not make the table ref++. */
2671 		flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
2672 		MLX5_ASSERT(tbl_data->jump.action);
2673 		DRV_LOG(DEBUG, "existed jump table resource %p: refcnt %d++",
2674 			(void *)&tbl_data->jump, cnt);
2675 	}
2676 	rte_atomic32_inc(&tbl_data->jump.refcnt);
2677 	dev_flow->handle->rix_jump = tbl_data->idx;
2678 	dev_flow->dv.jump = &tbl_data->jump;
2679 	return 0;
2680 }
2681 
2682 /**
2683  * Find existing default miss resource or create and register a new one.
2684  *
2685  * @param[in, out] dev
2686  *   Pointer to rte_eth_dev structure.
2687  * @param[out] error
2688  *   pointer to error structure.
2689  *
2690  * @return
2691  *   0 on success otherwise -errno and errno is set.
2692  */
2693 static int
2694 flow_dv_default_miss_resource_register(struct rte_eth_dev *dev,
2695 		struct rte_flow_error *error)
2696 {
2697 	struct mlx5_priv *priv = dev->data->dev_private;
2698 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2699 	struct mlx5_flow_default_miss_resource *cache_resource =
2700 			&sh->default_miss;
2701 	int cnt = rte_atomic32_read(&cache_resource->refcnt);
2702 
2703 	if (!cnt) {
2704 		MLX5_ASSERT(cache_resource->action);
2705 		cache_resource->action =
2706 		mlx5_glue->dr_create_flow_action_default_miss();
2707 		if (!cache_resource->action)
2708 			return rte_flow_error_set(error, ENOMEM,
2709 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2710 					"cannot create default miss action");
2711 		DRV_LOG(DEBUG, "new default miss resource %p: refcnt %d++",
2712 				(void *)cache_resource->action, cnt);
2713 	}
2714 	rte_atomic32_inc(&cache_resource->refcnt);
2715 	return 0;
2716 }
2717 
2718 /**
2719  * Find existing table port ID resource or create and register a new one.
2720  *
2721  * @param[in, out] dev
2722  *   Pointer to rte_eth_dev structure.
2723  * @param[in, out] resource
2724  *   Pointer to port ID action resource.
2725  * @parm[in, out] dev_flow
2726  *   Pointer to the dev_flow.
2727  * @param[out] error
2728  *   pointer to error structure.
2729  *
2730  * @return
2731  *   0 on success otherwise -errno and errno is set.
2732  */
2733 static int
2734 flow_dv_port_id_action_resource_register
2735 			(struct rte_eth_dev *dev,
2736 			 struct mlx5_flow_dv_port_id_action_resource *resource,
2737 			 struct mlx5_flow *dev_flow,
2738 			 struct rte_flow_error *error)
2739 {
2740 	struct mlx5_priv *priv = dev->data->dev_private;
2741 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2742 	struct mlx5_flow_dv_port_id_action_resource *cache_resource;
2743 	uint32_t idx = 0;
2744 	int ret;
2745 
2746 	/* Lookup a matching resource from cache. */
2747 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PORT_ID], sh->port_id_action_list,
2748 		      idx, cache_resource, next) {
2749 		if (resource->port_id == cache_resource->port_id) {
2750 			DRV_LOG(DEBUG, "port id action resource resource %p: "
2751 				"refcnt %d++",
2752 				(void *)cache_resource,
2753 				rte_atomic32_read(&cache_resource->refcnt));
2754 			rte_atomic32_inc(&cache_resource->refcnt);
2755 			dev_flow->handle->rix_port_id_action = idx;
2756 			dev_flow->dv.port_id_action = cache_resource;
2757 			return 0;
2758 		}
2759 	}
2760 	/* Register new port id action resource. */
2761 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PORT_ID],
2762 				       &dev_flow->handle->rix_port_id_action);
2763 	if (!cache_resource)
2764 		return rte_flow_error_set(error, ENOMEM,
2765 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2766 					  "cannot allocate resource memory");
2767 	*cache_resource = *resource;
2768 	ret = mlx5_flow_os_create_flow_action_dest_port
2769 				(priv->sh->fdb_domain, resource->port_id,
2770 				 &cache_resource->action);
2771 	if (ret) {
2772 		mlx5_free(cache_resource);
2773 		return rte_flow_error_set(error, ENOMEM,
2774 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2775 					  NULL, "cannot create action");
2776 	}
2777 	rte_atomic32_init(&cache_resource->refcnt);
2778 	rte_atomic32_inc(&cache_resource->refcnt);
2779 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_PORT_ID], &sh->port_id_action_list,
2780 		     dev_flow->handle->rix_port_id_action, cache_resource,
2781 		     next);
2782 	dev_flow->dv.port_id_action = cache_resource;
2783 	DRV_LOG(DEBUG, "new port id action resource %p: refcnt %d++",
2784 		(void *)cache_resource,
2785 		rte_atomic32_read(&cache_resource->refcnt));
2786 	return 0;
2787 }
2788 
2789 /**
2790  * Find existing push vlan resource or create and register a new one.
2791  *
2792  * @param [in, out] dev
2793  *   Pointer to rte_eth_dev structure.
2794  * @param[in, out] resource
2795  *   Pointer to port ID action resource.
2796  * @parm[in, out] dev_flow
2797  *   Pointer to the dev_flow.
2798  * @param[out] error
2799  *   pointer to error structure.
2800  *
2801  * @return
2802  *   0 on success otherwise -errno and errno is set.
2803  */
2804 static int
2805 flow_dv_push_vlan_action_resource_register
2806 		       (struct rte_eth_dev *dev,
2807 			struct mlx5_flow_dv_push_vlan_action_resource *resource,
2808 			struct mlx5_flow *dev_flow,
2809 			struct rte_flow_error *error)
2810 {
2811 	struct mlx5_priv *priv = dev->data->dev_private;
2812 	struct mlx5_dev_ctx_shared *sh = priv->sh;
2813 	struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
2814 	struct mlx5dv_dr_domain *domain;
2815 	uint32_t idx = 0;
2816 	int ret;
2817 
2818 	/* Lookup a matching resource from cache. */
2819 	ILIST_FOREACH(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2820 		      sh->push_vlan_action_list, idx, cache_resource, next) {
2821 		if (resource->vlan_tag == cache_resource->vlan_tag &&
2822 		    resource->ft_type == cache_resource->ft_type) {
2823 			DRV_LOG(DEBUG, "push-VLAN action resource resource %p: "
2824 				"refcnt %d++",
2825 				(void *)cache_resource,
2826 				rte_atomic32_read(&cache_resource->refcnt));
2827 			rte_atomic32_inc(&cache_resource->refcnt);
2828 			dev_flow->handle->dvh.rix_push_vlan = idx;
2829 			dev_flow->dv.push_vlan_res = cache_resource;
2830 			return 0;
2831 		}
2832 	}
2833 	/* Register new push_vlan action resource. */
2834 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2835 				       &dev_flow->handle->dvh.rix_push_vlan);
2836 	if (!cache_resource)
2837 		return rte_flow_error_set(error, ENOMEM,
2838 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2839 					  "cannot allocate resource memory");
2840 	*cache_resource = *resource;
2841 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
2842 		domain = sh->fdb_domain;
2843 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_RX)
2844 		domain = sh->rx_domain;
2845 	else
2846 		domain = sh->tx_domain;
2847 	ret = mlx5_flow_os_create_flow_action_push_vlan
2848 					(domain, resource->vlan_tag,
2849 					 &cache_resource->action);
2850 	if (ret) {
2851 		mlx5_free(cache_resource);
2852 		return rte_flow_error_set(error, ENOMEM,
2853 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2854 					  NULL, "cannot create action");
2855 	}
2856 	rte_atomic32_init(&cache_resource->refcnt);
2857 	rte_atomic32_inc(&cache_resource->refcnt);
2858 	ILIST_INSERT(sh->ipool[MLX5_IPOOL_PUSH_VLAN],
2859 		     &sh->push_vlan_action_list,
2860 		     dev_flow->handle->dvh.rix_push_vlan,
2861 		     cache_resource, next);
2862 	dev_flow->dv.push_vlan_res = cache_resource;
2863 	DRV_LOG(DEBUG, "new push vlan action resource %p: refcnt %d++",
2864 		(void *)cache_resource,
2865 		rte_atomic32_read(&cache_resource->refcnt));
2866 	return 0;
2867 }
2868 /**
2869  * Get the size of specific rte_flow_item_type hdr size
2870  *
2871  * @param[in] item_type
2872  *   Tested rte_flow_item_type.
2873  *
2874  * @return
2875  *   sizeof struct item_type, 0 if void or irrelevant.
2876  */
2877 static size_t
2878 flow_dv_get_item_hdr_len(const enum rte_flow_item_type item_type)
2879 {
2880 	size_t retval;
2881 
2882 	switch (item_type) {
2883 	case RTE_FLOW_ITEM_TYPE_ETH:
2884 		retval = sizeof(struct rte_ether_hdr);
2885 		break;
2886 	case RTE_FLOW_ITEM_TYPE_VLAN:
2887 		retval = sizeof(struct rte_vlan_hdr);
2888 		break;
2889 	case RTE_FLOW_ITEM_TYPE_IPV4:
2890 		retval = sizeof(struct rte_ipv4_hdr);
2891 		break;
2892 	case RTE_FLOW_ITEM_TYPE_IPV6:
2893 		retval = sizeof(struct rte_ipv6_hdr);
2894 		break;
2895 	case RTE_FLOW_ITEM_TYPE_UDP:
2896 		retval = sizeof(struct rte_udp_hdr);
2897 		break;
2898 	case RTE_FLOW_ITEM_TYPE_TCP:
2899 		retval = sizeof(struct rte_tcp_hdr);
2900 		break;
2901 	case RTE_FLOW_ITEM_TYPE_VXLAN:
2902 	case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
2903 		retval = sizeof(struct rte_vxlan_hdr);
2904 		break;
2905 	case RTE_FLOW_ITEM_TYPE_GRE:
2906 	case RTE_FLOW_ITEM_TYPE_NVGRE:
2907 		retval = sizeof(struct rte_gre_hdr);
2908 		break;
2909 	case RTE_FLOW_ITEM_TYPE_MPLS:
2910 		retval = sizeof(struct rte_mpls_hdr);
2911 		break;
2912 	case RTE_FLOW_ITEM_TYPE_VOID: /* Fall through. */
2913 	default:
2914 		retval = 0;
2915 		break;
2916 	}
2917 	return retval;
2918 }
2919 
2920 #define MLX5_ENCAP_IPV4_VERSION		0x40
2921 #define MLX5_ENCAP_IPV4_IHL_MIN		0x05
2922 #define MLX5_ENCAP_IPV4_TTL_DEF		0x40
2923 #define MLX5_ENCAP_IPV6_VTC_FLOW	0x60000000
2924 #define MLX5_ENCAP_IPV6_HOP_LIMIT	0xff
2925 #define MLX5_ENCAP_VXLAN_FLAGS		0x08000000
2926 #define MLX5_ENCAP_VXLAN_GPE_FLAGS	0x04
2927 
2928 /**
2929  * Convert the encap action data from list of rte_flow_item to raw buffer
2930  *
2931  * @param[in] items
2932  *   Pointer to rte_flow_item objects list.
2933  * @param[out] buf
2934  *   Pointer to the output buffer.
2935  * @param[out] size
2936  *   Pointer to the output buffer size.
2937  * @param[out] error
2938  *   Pointer to the error structure.
2939  *
2940  * @return
2941  *   0 on success, a negative errno value otherwise and rte_errno is set.
2942  */
2943 static int
2944 flow_dv_convert_encap_data(const struct rte_flow_item *items, uint8_t *buf,
2945 			   size_t *size, struct rte_flow_error *error)
2946 {
2947 	struct rte_ether_hdr *eth = NULL;
2948 	struct rte_vlan_hdr *vlan = NULL;
2949 	struct rte_ipv4_hdr *ipv4 = NULL;
2950 	struct rte_ipv6_hdr *ipv6 = NULL;
2951 	struct rte_udp_hdr *udp = NULL;
2952 	struct rte_vxlan_hdr *vxlan = NULL;
2953 	struct rte_vxlan_gpe_hdr *vxlan_gpe = NULL;
2954 	struct rte_gre_hdr *gre = NULL;
2955 	size_t len;
2956 	size_t temp_size = 0;
2957 
2958 	if (!items)
2959 		return rte_flow_error_set(error, EINVAL,
2960 					  RTE_FLOW_ERROR_TYPE_ACTION,
2961 					  NULL, "invalid empty data");
2962 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2963 		len = flow_dv_get_item_hdr_len(items->type);
2964 		if (len + temp_size > MLX5_ENCAP_MAX_LEN)
2965 			return rte_flow_error_set(error, EINVAL,
2966 						  RTE_FLOW_ERROR_TYPE_ACTION,
2967 						  (void *)items->type,
2968 						  "items total size is too big"
2969 						  " for encap action");
2970 		rte_memcpy((void *)&buf[temp_size], items->spec, len);
2971 		switch (items->type) {
2972 		case RTE_FLOW_ITEM_TYPE_ETH:
2973 			eth = (struct rte_ether_hdr *)&buf[temp_size];
2974 			break;
2975 		case RTE_FLOW_ITEM_TYPE_VLAN:
2976 			vlan = (struct rte_vlan_hdr *)&buf[temp_size];
2977 			if (!eth)
2978 				return rte_flow_error_set(error, EINVAL,
2979 						RTE_FLOW_ERROR_TYPE_ACTION,
2980 						(void *)items->type,
2981 						"eth header not found");
2982 			if (!eth->ether_type)
2983 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_VLAN);
2984 			break;
2985 		case RTE_FLOW_ITEM_TYPE_IPV4:
2986 			ipv4 = (struct rte_ipv4_hdr *)&buf[temp_size];
2987 			if (!vlan && !eth)
2988 				return rte_flow_error_set(error, EINVAL,
2989 						RTE_FLOW_ERROR_TYPE_ACTION,
2990 						(void *)items->type,
2991 						"neither eth nor vlan"
2992 						" header found");
2993 			if (vlan && !vlan->eth_proto)
2994 				vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2995 			else if (eth && !eth->ether_type)
2996 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV4);
2997 			if (!ipv4->version_ihl)
2998 				ipv4->version_ihl = MLX5_ENCAP_IPV4_VERSION |
2999 						    MLX5_ENCAP_IPV4_IHL_MIN;
3000 			if (!ipv4->time_to_live)
3001 				ipv4->time_to_live = MLX5_ENCAP_IPV4_TTL_DEF;
3002 			break;
3003 		case RTE_FLOW_ITEM_TYPE_IPV6:
3004 			ipv6 = (struct rte_ipv6_hdr *)&buf[temp_size];
3005 			if (!vlan && !eth)
3006 				return rte_flow_error_set(error, EINVAL,
3007 						RTE_FLOW_ERROR_TYPE_ACTION,
3008 						(void *)items->type,
3009 						"neither eth nor vlan"
3010 						" header found");
3011 			if (vlan && !vlan->eth_proto)
3012 				vlan->eth_proto = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3013 			else if (eth && !eth->ether_type)
3014 				eth->ether_type = RTE_BE16(RTE_ETHER_TYPE_IPV6);
3015 			if (!ipv6->vtc_flow)
3016 				ipv6->vtc_flow =
3017 					RTE_BE32(MLX5_ENCAP_IPV6_VTC_FLOW);
3018 			if (!ipv6->hop_limits)
3019 				ipv6->hop_limits = MLX5_ENCAP_IPV6_HOP_LIMIT;
3020 			break;
3021 		case RTE_FLOW_ITEM_TYPE_UDP:
3022 			udp = (struct rte_udp_hdr *)&buf[temp_size];
3023 			if (!ipv4 && !ipv6)
3024 				return rte_flow_error_set(error, EINVAL,
3025 						RTE_FLOW_ERROR_TYPE_ACTION,
3026 						(void *)items->type,
3027 						"ip header not found");
3028 			if (ipv4 && !ipv4->next_proto_id)
3029 				ipv4->next_proto_id = IPPROTO_UDP;
3030 			else if (ipv6 && !ipv6->proto)
3031 				ipv6->proto = IPPROTO_UDP;
3032 			break;
3033 		case RTE_FLOW_ITEM_TYPE_VXLAN:
3034 			vxlan = (struct rte_vxlan_hdr *)&buf[temp_size];
3035 			if (!udp)
3036 				return rte_flow_error_set(error, EINVAL,
3037 						RTE_FLOW_ERROR_TYPE_ACTION,
3038 						(void *)items->type,
3039 						"udp header not found");
3040 			if (!udp->dst_port)
3041 				udp->dst_port = RTE_BE16(MLX5_UDP_PORT_VXLAN);
3042 			if (!vxlan->vx_flags)
3043 				vxlan->vx_flags =
3044 					RTE_BE32(MLX5_ENCAP_VXLAN_FLAGS);
3045 			break;
3046 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
3047 			vxlan_gpe = (struct rte_vxlan_gpe_hdr *)&buf[temp_size];
3048 			if (!udp)
3049 				return rte_flow_error_set(error, EINVAL,
3050 						RTE_FLOW_ERROR_TYPE_ACTION,
3051 						(void *)items->type,
3052 						"udp header not found");
3053 			if (!vxlan_gpe->proto)
3054 				return rte_flow_error_set(error, EINVAL,
3055 						RTE_FLOW_ERROR_TYPE_ACTION,
3056 						(void *)items->type,
3057 						"next protocol not found");
3058 			if (!udp->dst_port)
3059 				udp->dst_port =
3060 					RTE_BE16(MLX5_UDP_PORT_VXLAN_GPE);
3061 			if (!vxlan_gpe->vx_flags)
3062 				vxlan_gpe->vx_flags =
3063 						MLX5_ENCAP_VXLAN_GPE_FLAGS;
3064 			break;
3065 		case RTE_FLOW_ITEM_TYPE_GRE:
3066 		case RTE_FLOW_ITEM_TYPE_NVGRE:
3067 			gre = (struct rte_gre_hdr *)&buf[temp_size];
3068 			if (!gre->proto)
3069 				return rte_flow_error_set(error, EINVAL,
3070 						RTE_FLOW_ERROR_TYPE_ACTION,
3071 						(void *)items->type,
3072 						"next protocol not found");
3073 			if (!ipv4 && !ipv6)
3074 				return rte_flow_error_set(error, EINVAL,
3075 						RTE_FLOW_ERROR_TYPE_ACTION,
3076 						(void *)items->type,
3077 						"ip header not found");
3078 			if (ipv4 && !ipv4->next_proto_id)
3079 				ipv4->next_proto_id = IPPROTO_GRE;
3080 			else if (ipv6 && !ipv6->proto)
3081 				ipv6->proto = IPPROTO_GRE;
3082 			break;
3083 		case RTE_FLOW_ITEM_TYPE_VOID:
3084 			break;
3085 		default:
3086 			return rte_flow_error_set(error, EINVAL,
3087 						  RTE_FLOW_ERROR_TYPE_ACTION,
3088 						  (void *)items->type,
3089 						  "unsupported item type");
3090 			break;
3091 		}
3092 		temp_size += len;
3093 	}
3094 	*size = temp_size;
3095 	return 0;
3096 }
3097 
3098 static int
3099 flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
3100 {
3101 	struct rte_ether_hdr *eth = NULL;
3102 	struct rte_vlan_hdr *vlan = NULL;
3103 	struct rte_ipv6_hdr *ipv6 = NULL;
3104 	struct rte_udp_hdr *udp = NULL;
3105 	char *next_hdr;
3106 	uint16_t proto;
3107 
3108 	eth = (struct rte_ether_hdr *)data;
3109 	next_hdr = (char *)(eth + 1);
3110 	proto = RTE_BE16(eth->ether_type);
3111 
3112 	/* VLAN skipping */
3113 	while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
3114 		vlan = (struct rte_vlan_hdr *)next_hdr;
3115 		proto = RTE_BE16(vlan->eth_proto);
3116 		next_hdr += sizeof(struct rte_vlan_hdr);
3117 	}
3118 
3119 	/* HW calculates IPv4 csum. no need to proceed */
3120 	if (proto == RTE_ETHER_TYPE_IPV4)
3121 		return 0;
3122 
3123 	/* non IPv4/IPv6 header. not supported */
3124 	if (proto != RTE_ETHER_TYPE_IPV6) {
3125 		return rte_flow_error_set(error, ENOTSUP,
3126 					  RTE_FLOW_ERROR_TYPE_ACTION,
3127 					  NULL, "Cannot offload non IPv4/IPv6");
3128 	}
3129 
3130 	ipv6 = (struct rte_ipv6_hdr *)next_hdr;
3131 
3132 	/* ignore non UDP */
3133 	if (ipv6->proto != IPPROTO_UDP)
3134 		return 0;
3135 
3136 	udp = (struct rte_udp_hdr *)(ipv6 + 1);
3137 	udp->dgram_cksum = 0;
3138 
3139 	return 0;
3140 }
3141 
3142 /**
3143  * Convert L2 encap action to DV specification.
3144  *
3145  * @param[in] dev
3146  *   Pointer to rte_eth_dev structure.
3147  * @param[in] action
3148  *   Pointer to action structure.
3149  * @param[in, out] dev_flow
3150  *   Pointer to the mlx5_flow.
3151  * @param[in] transfer
3152  *   Mark if the flow is E-Switch flow.
3153  * @param[out] error
3154  *   Pointer to the error structure.
3155  *
3156  * @return
3157  *   0 on success, a negative errno value otherwise and rte_errno is set.
3158  */
3159 static int
3160 flow_dv_create_action_l2_encap(struct rte_eth_dev *dev,
3161 			       const struct rte_flow_action *action,
3162 			       struct mlx5_flow *dev_flow,
3163 			       uint8_t transfer,
3164 			       struct rte_flow_error *error)
3165 {
3166 	const struct rte_flow_item *encap_data;
3167 	const struct rte_flow_action_raw_encap *raw_encap_data;
3168 	struct mlx5_flow_dv_encap_decap_resource res = {
3169 		.reformat_type =
3170 			MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L2_TUNNEL,
3171 		.ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3172 				      MLX5DV_FLOW_TABLE_TYPE_NIC_TX,
3173 	};
3174 
3175 	if (action->type == RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
3176 		raw_encap_data =
3177 			(const struct rte_flow_action_raw_encap *)action->conf;
3178 		res.size = raw_encap_data->size;
3179 		memcpy(res.buf, raw_encap_data->data, res.size);
3180 	} else {
3181 		if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
3182 			encap_data =
3183 				((const struct rte_flow_action_vxlan_encap *)
3184 						action->conf)->definition;
3185 		else
3186 			encap_data =
3187 				((const struct rte_flow_action_nvgre_encap *)
3188 						action->conf)->definition;
3189 		if (flow_dv_convert_encap_data(encap_data, res.buf,
3190 					       &res.size, error))
3191 			return -rte_errno;
3192 	}
3193 	if (flow_dv_zero_encap_udp_csum(res.buf, error))
3194 		return -rte_errno;
3195 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3196 		return rte_flow_error_set(error, EINVAL,
3197 					  RTE_FLOW_ERROR_TYPE_ACTION,
3198 					  NULL, "can't create L2 encap action");
3199 	return 0;
3200 }
3201 
3202 /**
3203  * Convert L2 decap action to DV specification.
3204  *
3205  * @param[in] dev
3206  *   Pointer to rte_eth_dev 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_decap(struct rte_eth_dev *dev,
3219 			       struct mlx5_flow *dev_flow,
3220 			       uint8_t transfer,
3221 			       struct rte_flow_error *error)
3222 {
3223 	struct mlx5_flow_dv_encap_decap_resource res = {
3224 		.size = 0,
3225 		.reformat_type =
3226 			MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TUNNEL_TO_L2,
3227 		.ft_type = transfer ? MLX5DV_FLOW_TABLE_TYPE_FDB :
3228 				      MLX5DV_FLOW_TABLE_TYPE_NIC_RX,
3229 	};
3230 
3231 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3232 		return rte_flow_error_set(error, EINVAL,
3233 					  RTE_FLOW_ERROR_TYPE_ACTION,
3234 					  NULL, "can't create L2 decap action");
3235 	return 0;
3236 }
3237 
3238 /**
3239  * Convert raw decap/encap (L3 tunnel) action to DV specification.
3240  *
3241  * @param[in] dev
3242  *   Pointer to rte_eth_dev structure.
3243  * @param[in] action
3244  *   Pointer to action structure.
3245  * @param[in, out] dev_flow
3246  *   Pointer to the mlx5_flow.
3247  * @param[in] attr
3248  *   Pointer to the flow attributes.
3249  * @param[out] error
3250  *   Pointer to the error structure.
3251  *
3252  * @return
3253  *   0 on success, a negative errno value otherwise and rte_errno is set.
3254  */
3255 static int
3256 flow_dv_create_action_raw_encap(struct rte_eth_dev *dev,
3257 				const struct rte_flow_action *action,
3258 				struct mlx5_flow *dev_flow,
3259 				const struct rte_flow_attr *attr,
3260 				struct rte_flow_error *error)
3261 {
3262 	const struct rte_flow_action_raw_encap *encap_data;
3263 	struct mlx5_flow_dv_encap_decap_resource res;
3264 
3265 	memset(&res, 0, sizeof(res));
3266 	encap_data = (const struct rte_flow_action_raw_encap *)action->conf;
3267 	res.size = encap_data->size;
3268 	memcpy(res.buf, encap_data->data, res.size);
3269 	res.reformat_type = res.size < MLX5_ENCAPSULATION_DECISION_SIZE ?
3270 		MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L3_TUNNEL_TO_L2 :
3271 		MLX5DV_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL;
3272 	if (attr->transfer)
3273 		res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3274 	else
3275 		res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3276 					     MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3277 	if (flow_dv_encap_decap_resource_register(dev, &res, dev_flow, error))
3278 		return rte_flow_error_set(error, EINVAL,
3279 					  RTE_FLOW_ERROR_TYPE_ACTION,
3280 					  NULL, "can't create encap action");
3281 	return 0;
3282 }
3283 
3284 /**
3285  * Create action push VLAN.
3286  *
3287  * @param[in] dev
3288  *   Pointer to rte_eth_dev structure.
3289  * @param[in] attr
3290  *   Pointer to the flow attributes.
3291  * @param[in] vlan
3292  *   Pointer to the vlan to push to the Ethernet header.
3293  * @param[in, out] dev_flow
3294  *   Pointer to the mlx5_flow.
3295  * @param[out] error
3296  *   Pointer to the error structure.
3297  *
3298  * @return
3299  *   0 on success, a negative errno value otherwise and rte_errno is set.
3300  */
3301 static int
3302 flow_dv_create_action_push_vlan(struct rte_eth_dev *dev,
3303 				const struct rte_flow_attr *attr,
3304 				const struct rte_vlan_hdr *vlan,
3305 				struct mlx5_flow *dev_flow,
3306 				struct rte_flow_error *error)
3307 {
3308 	struct mlx5_flow_dv_push_vlan_action_resource res;
3309 
3310 	memset(&res, 0, sizeof(res));
3311 	res.vlan_tag =
3312 		rte_cpu_to_be_32(((uint32_t)vlan->eth_proto) << 16 |
3313 				 vlan->vlan_tci);
3314 	if (attr->transfer)
3315 		res.ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
3316 	else
3317 		res.ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
3318 					     MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
3319 	return flow_dv_push_vlan_action_resource_register
3320 					    (dev, &res, dev_flow, error);
3321 }
3322 
3323 /**
3324  * Validate the modify-header actions.
3325  *
3326  * @param[in] action_flags
3327  *   Holds the actions detected until now.
3328  * @param[in] action
3329  *   Pointer to the modify action.
3330  * @param[out] error
3331  *   Pointer to error structure.
3332  *
3333  * @return
3334  *   0 on success, a negative errno value otherwise and rte_errno is set.
3335  */
3336 static int
3337 flow_dv_validate_action_modify_hdr(const uint64_t action_flags,
3338 				   const struct rte_flow_action *action,
3339 				   struct rte_flow_error *error)
3340 {
3341 	if (action->type != RTE_FLOW_ACTION_TYPE_DEC_TTL && !action->conf)
3342 		return rte_flow_error_set(error, EINVAL,
3343 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3344 					  NULL, "action configuration not set");
3345 	if (action_flags & MLX5_FLOW_ACTION_ENCAP)
3346 		return rte_flow_error_set(error, EINVAL,
3347 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3348 					  "can't have encap action before"
3349 					  " modify action");
3350 	return 0;
3351 }
3352 
3353 /**
3354  * Validate the modify-header MAC address actions.
3355  *
3356  * @param[in] action_flags
3357  *   Holds the actions detected until now.
3358  * @param[in] action
3359  *   Pointer to the modify action.
3360  * @param[in] item_flags
3361  *   Holds the items detected.
3362  * @param[out] error
3363  *   Pointer to error structure.
3364  *
3365  * @return
3366  *   0 on success, a negative errno value otherwise and rte_errno is set.
3367  */
3368 static int
3369 flow_dv_validate_action_modify_mac(const uint64_t action_flags,
3370 				   const struct rte_flow_action *action,
3371 				   const uint64_t item_flags,
3372 				   struct rte_flow_error *error)
3373 {
3374 	int ret = 0;
3375 
3376 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3377 	if (!ret) {
3378 		if (!(item_flags & MLX5_FLOW_LAYER_L2))
3379 			return rte_flow_error_set(error, EINVAL,
3380 						  RTE_FLOW_ERROR_TYPE_ACTION,
3381 						  NULL,
3382 						  "no L2 item in pattern");
3383 	}
3384 	return ret;
3385 }
3386 
3387 /**
3388  * Validate the modify-header IPv4 address actions.
3389  *
3390  * @param[in] action_flags
3391  *   Holds the actions detected until now.
3392  * @param[in] action
3393  *   Pointer to the modify action.
3394  * @param[in] item_flags
3395  *   Holds the items detected.
3396  * @param[out] error
3397  *   Pointer to error structure.
3398  *
3399  * @return
3400  *   0 on success, a negative errno value otherwise and rte_errno is set.
3401  */
3402 static int
3403 flow_dv_validate_action_modify_ipv4(const uint64_t action_flags,
3404 				    const struct rte_flow_action *action,
3405 				    const uint64_t item_flags,
3406 				    struct rte_flow_error *error)
3407 {
3408 	int ret = 0;
3409 	uint64_t layer;
3410 
3411 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3412 	if (!ret) {
3413 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3414 				 MLX5_FLOW_LAYER_INNER_L3_IPV4 :
3415 				 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3416 		if (!(item_flags & layer))
3417 			return rte_flow_error_set(error, EINVAL,
3418 						  RTE_FLOW_ERROR_TYPE_ACTION,
3419 						  NULL,
3420 						  "no ipv4 item in pattern");
3421 	}
3422 	return ret;
3423 }
3424 
3425 /**
3426  * Validate the modify-header IPv6 address actions.
3427  *
3428  * @param[in] action_flags
3429  *   Holds the actions detected until now.
3430  * @param[in] action
3431  *   Pointer to the modify action.
3432  * @param[in] item_flags
3433  *   Holds the items detected.
3434  * @param[out] error
3435  *   Pointer to error structure.
3436  *
3437  * @return
3438  *   0 on success, a negative errno value otherwise and rte_errno is set.
3439  */
3440 static int
3441 flow_dv_validate_action_modify_ipv6(const uint64_t action_flags,
3442 				    const struct rte_flow_action *action,
3443 				    const uint64_t item_flags,
3444 				    struct rte_flow_error *error)
3445 {
3446 	int ret = 0;
3447 	uint64_t layer;
3448 
3449 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3450 	if (!ret) {
3451 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3452 				 MLX5_FLOW_LAYER_INNER_L3_IPV6 :
3453 				 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3454 		if (!(item_flags & layer))
3455 			return rte_flow_error_set(error, EINVAL,
3456 						  RTE_FLOW_ERROR_TYPE_ACTION,
3457 						  NULL,
3458 						  "no ipv6 item in pattern");
3459 	}
3460 	return ret;
3461 }
3462 
3463 /**
3464  * Validate the modify-header TP actions.
3465  *
3466  * @param[in] action_flags
3467  *   Holds the actions detected until now.
3468  * @param[in] action
3469  *   Pointer to the modify action.
3470  * @param[in] item_flags
3471  *   Holds the items detected.
3472  * @param[out] error
3473  *   Pointer to error structure.
3474  *
3475  * @return
3476  *   0 on success, a negative errno value otherwise and rte_errno is set.
3477  */
3478 static int
3479 flow_dv_validate_action_modify_tp(const uint64_t action_flags,
3480 				  const struct rte_flow_action *action,
3481 				  const uint64_t item_flags,
3482 				  struct rte_flow_error *error)
3483 {
3484 	int ret = 0;
3485 	uint64_t layer;
3486 
3487 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3488 	if (!ret) {
3489 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3490 				 MLX5_FLOW_LAYER_INNER_L4 :
3491 				 MLX5_FLOW_LAYER_OUTER_L4;
3492 		if (!(item_flags & layer))
3493 			return rte_flow_error_set(error, EINVAL,
3494 						  RTE_FLOW_ERROR_TYPE_ACTION,
3495 						  NULL, "no transport layer "
3496 						  "in pattern");
3497 	}
3498 	return ret;
3499 }
3500 
3501 /**
3502  * Validate the modify-header actions of increment/decrement
3503  * TCP Sequence-number.
3504  *
3505  * @param[in] action_flags
3506  *   Holds the actions detected until now.
3507  * @param[in] action
3508  *   Pointer to the modify action.
3509  * @param[in] item_flags
3510  *   Holds the items detected.
3511  * @param[out] error
3512  *   Pointer to error structure.
3513  *
3514  * @return
3515  *   0 on success, a negative errno value otherwise and rte_errno is set.
3516  */
3517 static int
3518 flow_dv_validate_action_modify_tcp_seq(const uint64_t action_flags,
3519 				       const struct rte_flow_action *action,
3520 				       const uint64_t item_flags,
3521 				       struct rte_flow_error *error)
3522 {
3523 	int ret = 0;
3524 	uint64_t layer;
3525 
3526 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3527 	if (!ret) {
3528 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3529 				 MLX5_FLOW_LAYER_INNER_L4_TCP :
3530 				 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3531 		if (!(item_flags & layer))
3532 			return rte_flow_error_set(error, EINVAL,
3533 						  RTE_FLOW_ERROR_TYPE_ACTION,
3534 						  NULL, "no TCP item in"
3535 						  " pattern");
3536 		if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ &&
3537 			(action_flags & MLX5_FLOW_ACTION_DEC_TCP_SEQ)) ||
3538 		    (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ &&
3539 			(action_flags & MLX5_FLOW_ACTION_INC_TCP_SEQ)))
3540 			return rte_flow_error_set(error, EINVAL,
3541 						  RTE_FLOW_ERROR_TYPE_ACTION,
3542 						  NULL,
3543 						  "cannot decrease and increase"
3544 						  " TCP sequence number"
3545 						  " at the same time");
3546 	}
3547 	return ret;
3548 }
3549 
3550 /**
3551  * Validate the modify-header actions of increment/decrement
3552  * TCP Acknowledgment number.
3553  *
3554  * @param[in] action_flags
3555  *   Holds the actions detected until now.
3556  * @param[in] action
3557  *   Pointer to the modify action.
3558  * @param[in] item_flags
3559  *   Holds the items detected.
3560  * @param[out] error
3561  *   Pointer to error structure.
3562  *
3563  * @return
3564  *   0 on success, a negative errno value otherwise and rte_errno is set.
3565  */
3566 static int
3567 flow_dv_validate_action_modify_tcp_ack(const uint64_t action_flags,
3568 				       const struct rte_flow_action *action,
3569 				       const uint64_t item_flags,
3570 				       struct rte_flow_error *error)
3571 {
3572 	int ret = 0;
3573 	uint64_t layer;
3574 
3575 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3576 	if (!ret) {
3577 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3578 				 MLX5_FLOW_LAYER_INNER_L4_TCP :
3579 				 MLX5_FLOW_LAYER_OUTER_L4_TCP;
3580 		if (!(item_flags & layer))
3581 			return rte_flow_error_set(error, EINVAL,
3582 						  RTE_FLOW_ERROR_TYPE_ACTION,
3583 						  NULL, "no TCP item in"
3584 						  " pattern");
3585 		if ((action->type == RTE_FLOW_ACTION_TYPE_INC_TCP_ACK &&
3586 			(action_flags & MLX5_FLOW_ACTION_DEC_TCP_ACK)) ||
3587 		    (action->type == RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK &&
3588 			(action_flags & MLX5_FLOW_ACTION_INC_TCP_ACK)))
3589 			return rte_flow_error_set(error, EINVAL,
3590 						  RTE_FLOW_ERROR_TYPE_ACTION,
3591 						  NULL,
3592 						  "cannot decrease and increase"
3593 						  " TCP acknowledgment number"
3594 						  " at the same time");
3595 	}
3596 	return ret;
3597 }
3598 
3599 /**
3600  * Validate the modify-header TTL actions.
3601  *
3602  * @param[in] action_flags
3603  *   Holds the actions detected until now.
3604  * @param[in] action
3605  *   Pointer to the modify action.
3606  * @param[in] item_flags
3607  *   Holds the items detected.
3608  * @param[out] error
3609  *   Pointer to error structure.
3610  *
3611  * @return
3612  *   0 on success, a negative errno value otherwise and rte_errno is set.
3613  */
3614 static int
3615 flow_dv_validate_action_modify_ttl(const uint64_t action_flags,
3616 				   const struct rte_flow_action *action,
3617 				   const uint64_t item_flags,
3618 				   struct rte_flow_error *error)
3619 {
3620 	int ret = 0;
3621 	uint64_t layer;
3622 
3623 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3624 	if (!ret) {
3625 		layer = (action_flags & MLX5_FLOW_ACTION_DECAP) ?
3626 				 MLX5_FLOW_LAYER_INNER_L3 :
3627 				 MLX5_FLOW_LAYER_OUTER_L3;
3628 		if (!(item_flags & layer))
3629 			return rte_flow_error_set(error, EINVAL,
3630 						  RTE_FLOW_ERROR_TYPE_ACTION,
3631 						  NULL,
3632 						  "no IP protocol in pattern");
3633 	}
3634 	return ret;
3635 }
3636 
3637 /**
3638  * Validate jump action.
3639  *
3640  * @param[in] action
3641  *   Pointer to the jump action.
3642  * @param[in] action_flags
3643  *   Holds the actions detected until now.
3644  * @param[in] attributes
3645  *   Pointer to flow attributes
3646  * @param[in] external
3647  *   Action belongs to flow rule created by request external to PMD.
3648  * @param[out] error
3649  *   Pointer to error structure.
3650  *
3651  * @return
3652  *   0 on success, a negative errno value otherwise and rte_errno is set.
3653  */
3654 static int
3655 flow_dv_validate_action_jump(const struct rte_flow_action *action,
3656 			     uint64_t action_flags,
3657 			     const struct rte_flow_attr *attributes,
3658 			     bool external, struct rte_flow_error *error)
3659 {
3660 	uint32_t target_group, table;
3661 	int ret = 0;
3662 
3663 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3664 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3665 		return rte_flow_error_set(error, EINVAL,
3666 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3667 					  "can't have 2 fate actions in"
3668 					  " same flow");
3669 	if (action_flags & MLX5_FLOW_ACTION_METER)
3670 		return rte_flow_error_set(error, ENOTSUP,
3671 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3672 					  "jump with meter not support");
3673 	if (!action->conf)
3674 		return rte_flow_error_set(error, EINVAL,
3675 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3676 					  NULL, "action configuration not set");
3677 	target_group =
3678 		((const struct rte_flow_action_jump *)action->conf)->group;
3679 	ret = mlx5_flow_group_to_table(attributes, external, target_group,
3680 				       true, &table, error);
3681 	if (ret)
3682 		return ret;
3683 	if (attributes->group == target_group)
3684 		return rte_flow_error_set(error, EINVAL,
3685 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3686 					  "target group must be other than"
3687 					  " the current flow group");
3688 	return 0;
3689 }
3690 
3691 /*
3692  * Validate the port_id action.
3693  *
3694  * @param[in] dev
3695  *   Pointer to rte_eth_dev structure.
3696  * @param[in] action_flags
3697  *   Bit-fields that holds the actions detected until now.
3698  * @param[in] action
3699  *   Port_id RTE action structure.
3700  * @param[in] attr
3701  *   Attributes of flow that includes this action.
3702  * @param[out] error
3703  *   Pointer to error structure.
3704  *
3705  * @return
3706  *   0 on success, a negative errno value otherwise and rte_errno is set.
3707  */
3708 static int
3709 flow_dv_validate_action_port_id(struct rte_eth_dev *dev,
3710 				uint64_t action_flags,
3711 				const struct rte_flow_action *action,
3712 				const struct rte_flow_attr *attr,
3713 				struct rte_flow_error *error)
3714 {
3715 	const struct rte_flow_action_port_id *port_id;
3716 	struct mlx5_priv *act_priv;
3717 	struct mlx5_priv *dev_priv;
3718 	uint16_t port;
3719 
3720 	if (!attr->transfer)
3721 		return rte_flow_error_set(error, ENOTSUP,
3722 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3723 					  NULL,
3724 					  "port id action is valid in transfer"
3725 					  " mode only");
3726 	if (!action || !action->conf)
3727 		return rte_flow_error_set(error, ENOTSUP,
3728 					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
3729 					  NULL,
3730 					  "port id action parameters must be"
3731 					  " specified");
3732 	if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
3733 			    MLX5_FLOW_FATE_ESWITCH_ACTIONS))
3734 		return rte_flow_error_set(error, EINVAL,
3735 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3736 					  "can have only one fate actions in"
3737 					  " a flow");
3738 	dev_priv = mlx5_dev_to_eswitch_info(dev);
3739 	if (!dev_priv)
3740 		return rte_flow_error_set(error, rte_errno,
3741 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3742 					  NULL,
3743 					  "failed to obtain E-Switch info");
3744 	port_id = action->conf;
3745 	port = port_id->original ? dev->data->port_id : port_id->id;
3746 	act_priv = mlx5_port_to_eswitch_info(port, false);
3747 	if (!act_priv)
3748 		return rte_flow_error_set
3749 				(error, rte_errno,
3750 				 RTE_FLOW_ERROR_TYPE_ACTION_CONF, port_id,
3751 				 "failed to obtain E-Switch port id for port");
3752 	if (act_priv->domain_id != dev_priv->domain_id)
3753 		return rte_flow_error_set
3754 				(error, EINVAL,
3755 				 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3756 				 "port does not belong to"
3757 				 " E-Switch being configured");
3758 	return 0;
3759 }
3760 
3761 /**
3762  * Get the maximum number of modify header actions.
3763  *
3764  * @param dev
3765  *   Pointer to rte_eth_dev structure.
3766  * @param flags
3767  *   Flags bits to check if root level.
3768  *
3769  * @return
3770  *   Max number of modify header actions device can support.
3771  */
3772 static inline unsigned int
3773 flow_dv_modify_hdr_action_max(struct rte_eth_dev *dev __rte_unused,
3774 			      uint64_t flags)
3775 {
3776 	/*
3777 	 * There's no way to directly query the max capacity from FW.
3778 	 * The maximal value on root table should be assumed to be supported.
3779 	 */
3780 	if (!(flags & MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL))
3781 		return MLX5_MAX_MODIFY_NUM;
3782 	else
3783 		return MLX5_ROOT_TBL_MODIFY_NUM;
3784 }
3785 
3786 /**
3787  * Validate the meter action.
3788  *
3789  * @param[in] dev
3790  *   Pointer to rte_eth_dev structure.
3791  * @param[in] action_flags
3792  *   Bit-fields that holds the actions detected until now.
3793  * @param[in] action
3794  *   Pointer to the meter action.
3795  * @param[in] attr
3796  *   Attributes of flow that includes this action.
3797  * @param[out] error
3798  *   Pointer to error structure.
3799  *
3800  * @return
3801  *   0 on success, a negative errno value otherwise and rte_ernno is set.
3802  */
3803 static int
3804 mlx5_flow_validate_action_meter(struct rte_eth_dev *dev,
3805 				uint64_t action_flags,
3806 				const struct rte_flow_action *action,
3807 				const struct rte_flow_attr *attr,
3808 				struct rte_flow_error *error)
3809 {
3810 	struct mlx5_priv *priv = dev->data->dev_private;
3811 	const struct rte_flow_action_meter *am = action->conf;
3812 	struct mlx5_flow_meter *fm;
3813 
3814 	if (!am)
3815 		return rte_flow_error_set(error, EINVAL,
3816 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3817 					  "meter action conf is NULL");
3818 
3819 	if (action_flags & MLX5_FLOW_ACTION_METER)
3820 		return rte_flow_error_set(error, ENOTSUP,
3821 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3822 					  "meter chaining not support");
3823 	if (action_flags & MLX5_FLOW_ACTION_JUMP)
3824 		return rte_flow_error_set(error, ENOTSUP,
3825 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3826 					  "meter with jump not support");
3827 	if (!priv->mtr_en)
3828 		return rte_flow_error_set(error, ENOTSUP,
3829 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3830 					  NULL,
3831 					  "meter action not supported");
3832 	fm = mlx5_flow_meter_find(priv, am->mtr_id);
3833 	if (!fm)
3834 		return rte_flow_error_set(error, EINVAL,
3835 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3836 					  "Meter not found");
3837 	if (fm->ref_cnt && (!(fm->transfer == attr->transfer ||
3838 	      (!fm->ingress && !attr->ingress && attr->egress) ||
3839 	      (!fm->egress && !attr->egress && attr->ingress))))
3840 		return rte_flow_error_set(error, EINVAL,
3841 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3842 					  "Flow attributes are either invalid "
3843 					  "or have a conflict with current "
3844 					  "meter attributes");
3845 	return 0;
3846 }
3847 
3848 /**
3849  * Validate the age action.
3850  *
3851  * @param[in] action_flags
3852  *   Holds the actions detected until now.
3853  * @param[in] action
3854  *   Pointer to the age action.
3855  * @param[in] dev
3856  *   Pointer to the Ethernet device structure.
3857  * @param[out] error
3858  *   Pointer to error structure.
3859  *
3860  * @return
3861  *   0 on success, a negative errno value otherwise and rte_errno is set.
3862  */
3863 static int
3864 flow_dv_validate_action_age(uint64_t action_flags,
3865 			    const struct rte_flow_action *action,
3866 			    struct rte_eth_dev *dev,
3867 			    struct rte_flow_error *error)
3868 {
3869 	struct mlx5_priv *priv = dev->data->dev_private;
3870 	const struct rte_flow_action_age *age = action->conf;
3871 
3872 	if (!priv->config.devx || priv->counter_fallback)
3873 		return rte_flow_error_set(error, ENOTSUP,
3874 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
3875 					  NULL,
3876 					  "age action not supported");
3877 	if (!(action->conf))
3878 		return rte_flow_error_set(error, EINVAL,
3879 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
3880 					  "configuration cannot be null");
3881 	if (age->timeout >= UINT16_MAX / 2 / 10)
3882 		return rte_flow_error_set(error, ENOTSUP,
3883 					  RTE_FLOW_ERROR_TYPE_ACTION, action,
3884 					  "Max age time: 3275 seconds");
3885 	if (action_flags & MLX5_FLOW_ACTION_AGE)
3886 		return rte_flow_error_set(error, EINVAL,
3887 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3888 					  "Duplicate age ctions set");
3889 	return 0;
3890 }
3891 
3892 /**
3893  * Validate the modify-header IPv4 DSCP actions.
3894  *
3895  * @param[in] action_flags
3896  *   Holds the actions detected until now.
3897  * @param[in] action
3898  *   Pointer to the modify action.
3899  * @param[in] item_flags
3900  *   Holds the items detected.
3901  * @param[out] error
3902  *   Pointer to error structure.
3903  *
3904  * @return
3905  *   0 on success, a negative errno value otherwise and rte_errno is set.
3906  */
3907 static int
3908 flow_dv_validate_action_modify_ipv4_dscp(const uint64_t action_flags,
3909 					 const struct rte_flow_action *action,
3910 					 const uint64_t item_flags,
3911 					 struct rte_flow_error *error)
3912 {
3913 	int ret = 0;
3914 
3915 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3916 	if (!ret) {
3917 		if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV4))
3918 			return rte_flow_error_set(error, EINVAL,
3919 						  RTE_FLOW_ERROR_TYPE_ACTION,
3920 						  NULL,
3921 						  "no ipv4 item in pattern");
3922 	}
3923 	return ret;
3924 }
3925 
3926 /**
3927  * Validate the modify-header IPv6 DSCP actions.
3928  *
3929  * @param[in] action_flags
3930  *   Holds the actions detected until now.
3931  * @param[in] action
3932  *   Pointer to the modify action.
3933  * @param[in] item_flags
3934  *   Holds the items detected.
3935  * @param[out] error
3936  *   Pointer to error structure.
3937  *
3938  * @return
3939  *   0 on success, a negative errno value otherwise and rte_errno is set.
3940  */
3941 static int
3942 flow_dv_validate_action_modify_ipv6_dscp(const uint64_t action_flags,
3943 					 const struct rte_flow_action *action,
3944 					 const uint64_t item_flags,
3945 					 struct rte_flow_error *error)
3946 {
3947 	int ret = 0;
3948 
3949 	ret = flow_dv_validate_action_modify_hdr(action_flags, action, error);
3950 	if (!ret) {
3951 		if (!(item_flags & MLX5_FLOW_LAYER_L3_IPV6))
3952 			return rte_flow_error_set(error, EINVAL,
3953 						  RTE_FLOW_ERROR_TYPE_ACTION,
3954 						  NULL,
3955 						  "no ipv6 item in pattern");
3956 	}
3957 	return ret;
3958 }
3959 
3960 /**
3961  * Find existing modify-header resource or create and register a new one.
3962  *
3963  * @param dev[in, out]
3964  *   Pointer to rte_eth_dev structure.
3965  * @param[in, out] resource
3966  *   Pointer to modify-header resource.
3967  * @parm[in, out] dev_flow
3968  *   Pointer to the dev_flow.
3969  * @param[out] error
3970  *   pointer to error structure.
3971  *
3972  * @return
3973  *   0 on success otherwise -errno and errno is set.
3974  */
3975 static int
3976 flow_dv_modify_hdr_resource_register
3977 			(struct rte_eth_dev *dev,
3978 			 struct mlx5_flow_dv_modify_hdr_resource *resource,
3979 			 struct mlx5_flow *dev_flow,
3980 			 struct rte_flow_error *error)
3981 {
3982 	struct mlx5_priv *priv = dev->data->dev_private;
3983 	struct mlx5_dev_ctx_shared *sh = priv->sh;
3984 	struct mlx5_flow_dv_modify_hdr_resource *cache_resource;
3985 	struct mlx5dv_dr_domain *ns;
3986 	uint32_t actions_len;
3987 	int ret;
3988 
3989 	resource->flags = dev_flow->dv.group ? 0 :
3990 			  MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
3991 	if (resource->actions_num > flow_dv_modify_hdr_action_max(dev,
3992 				    resource->flags))
3993 		return rte_flow_error_set(error, EOVERFLOW,
3994 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
3995 					  "too many modify header items");
3996 	if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_FDB)
3997 		ns = sh->fdb_domain;
3998 	else if (resource->ft_type == MLX5DV_FLOW_TABLE_TYPE_NIC_TX)
3999 		ns = sh->tx_domain;
4000 	else
4001 		ns = sh->rx_domain;
4002 	/* Lookup a matching resource from cache. */
4003 	actions_len = resource->actions_num * sizeof(resource->actions[0]);
4004 	LIST_FOREACH(cache_resource, &sh->modify_cmds, next) {
4005 		if (resource->ft_type == cache_resource->ft_type &&
4006 		    resource->actions_num == cache_resource->actions_num &&
4007 		    resource->flags == cache_resource->flags &&
4008 		    !memcmp((const void *)resource->actions,
4009 			    (const void *)cache_resource->actions,
4010 			    actions_len)) {
4011 			DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d++",
4012 				(void *)cache_resource,
4013 				rte_atomic32_read(&cache_resource->refcnt));
4014 			rte_atomic32_inc(&cache_resource->refcnt);
4015 			dev_flow->handle->dvh.modify_hdr = cache_resource;
4016 			return 0;
4017 		}
4018 	}
4019 	/* Register new modify-header resource. */
4020 	cache_resource = mlx5_malloc(MLX5_MEM_ZERO,
4021 				    sizeof(*cache_resource) + actions_len, 0,
4022 				    SOCKET_ID_ANY);
4023 	if (!cache_resource)
4024 		return rte_flow_error_set(error, ENOMEM,
4025 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4026 					  "cannot allocate resource memory");
4027 	*cache_resource = *resource;
4028 	rte_memcpy(cache_resource->actions, resource->actions, actions_len);
4029 	ret = mlx5_flow_os_create_flow_action_modify_header
4030 					(sh->ctx, ns, cache_resource,
4031 					 actions_len, &cache_resource->action);
4032 	if (ret) {
4033 		mlx5_free(cache_resource);
4034 		return rte_flow_error_set(error, ENOMEM,
4035 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4036 					  NULL, "cannot create action");
4037 	}
4038 	rte_atomic32_init(&cache_resource->refcnt);
4039 	rte_atomic32_inc(&cache_resource->refcnt);
4040 	LIST_INSERT_HEAD(&sh->modify_cmds, cache_resource, next);
4041 	dev_flow->handle->dvh.modify_hdr = cache_resource;
4042 	DRV_LOG(DEBUG, "new modify-header resource %p: refcnt %d++",
4043 		(void *)cache_resource,
4044 		rte_atomic32_read(&cache_resource->refcnt));
4045 	return 0;
4046 }
4047 
4048 /**
4049  * Get DV flow counter by index.
4050  *
4051  * @param[in] dev
4052  *   Pointer to the Ethernet device structure.
4053  * @param[in] idx
4054  *   mlx5 flow counter index in the container.
4055  * @param[out] ppool
4056  *   mlx5 flow counter pool in the container,
4057  *
4058  * @return
4059  *   Pointer to the counter, NULL otherwise.
4060  */
4061 static struct mlx5_flow_counter *
4062 flow_dv_counter_get_by_idx(struct rte_eth_dev *dev,
4063 			   uint32_t idx,
4064 			   struct mlx5_flow_counter_pool **ppool)
4065 {
4066 	struct mlx5_priv *priv = dev->data->dev_private;
4067 	struct mlx5_pools_container *cont;
4068 	struct mlx5_flow_counter_pool *pool;
4069 	uint32_t batch = 0, age = 0;
4070 
4071 	idx--;
4072 	age = MLX_CNT_IS_AGE(idx);
4073 	idx = age ? idx - MLX5_CNT_AGE_OFFSET : idx;
4074 	if (idx >= MLX5_CNT_BATCH_OFFSET) {
4075 		idx -= MLX5_CNT_BATCH_OFFSET;
4076 		batch = 1;
4077 	}
4078 	cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4079 	MLX5_ASSERT(idx / MLX5_COUNTERS_PER_POOL < cont->n);
4080 	pool = cont->pools[idx / MLX5_COUNTERS_PER_POOL];
4081 	MLX5_ASSERT(pool);
4082 	if (ppool)
4083 		*ppool = pool;
4084 	return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
4085 }
4086 
4087 /**
4088  * Check the devx counter belongs to the pool.
4089  *
4090  * @param[in] pool
4091  *   Pointer to the counter pool.
4092  * @param[in] id
4093  *   The counter devx ID.
4094  *
4095  * @return
4096  *   True if counter belongs to the pool, false otherwise.
4097  */
4098 static bool
4099 flow_dv_is_counter_in_pool(struct mlx5_flow_counter_pool *pool, int id)
4100 {
4101 	int base = (pool->min_dcs->id / MLX5_COUNTERS_PER_POOL) *
4102 		   MLX5_COUNTERS_PER_POOL;
4103 
4104 	if (id >= base && id < base + MLX5_COUNTERS_PER_POOL)
4105 		return true;
4106 	return false;
4107 }
4108 
4109 /**
4110  * Get a pool by devx counter ID.
4111  *
4112  * @param[in] cont
4113  *   Pointer to the counter container.
4114  * @param[in] id
4115  *   The counter devx ID.
4116  *
4117  * @return
4118  *   The counter pool pointer if exists, NULL otherwise,
4119  */
4120 static struct mlx5_flow_counter_pool *
4121 flow_dv_find_pool_by_id(struct mlx5_pools_container *cont, int id)
4122 {
4123 	uint32_t i;
4124 
4125 	/* Check last used pool. */
4126 	if (cont->last_pool_idx != POOL_IDX_INVALID &&
4127 	    flow_dv_is_counter_in_pool(cont->pools[cont->last_pool_idx], id))
4128 		return cont->pools[cont->last_pool_idx];
4129 	/* ID out of range means no suitable pool in the container. */
4130 	if (id > cont->max_id || id < cont->min_id)
4131 		return NULL;
4132 	/*
4133 	 * Find the pool from the end of the container, since mostly counter
4134 	 * ID is sequence increasing, and the last pool should be the needed
4135 	 * one.
4136 	 */
4137 	i = rte_atomic16_read(&cont->n_valid);
4138 	while (i--) {
4139 		struct mlx5_flow_counter_pool *pool = cont->pools[i];
4140 
4141 		if (flow_dv_is_counter_in_pool(pool, id))
4142 			return pool;
4143 	}
4144 	return NULL;
4145 }
4146 
4147 /**
4148  * Allocate a new memory for the counter values wrapped by all the needed
4149  * management.
4150  *
4151  * @param[in] dev
4152  *   Pointer to the Ethernet device structure.
4153  * @param[in] raws_n
4154  *   The raw memory areas - each one for MLX5_COUNTERS_PER_POOL counters.
4155  *
4156  * @return
4157  *   The new memory management pointer on success, otherwise NULL and rte_errno
4158  *   is set.
4159  */
4160 static struct mlx5_counter_stats_mem_mng *
4161 flow_dv_create_counter_stat_mem_mng(struct rte_eth_dev *dev, int raws_n)
4162 {
4163 	struct mlx5_priv *priv = dev->data->dev_private;
4164 	struct mlx5_dev_ctx_shared *sh = priv->sh;
4165 	struct mlx5_devx_mkey_attr mkey_attr;
4166 	struct mlx5_counter_stats_mem_mng *mem_mng;
4167 	volatile struct flow_counter_stats *raw_data;
4168 	int size = (sizeof(struct flow_counter_stats) *
4169 			MLX5_COUNTERS_PER_POOL +
4170 			sizeof(struct mlx5_counter_stats_raw)) * raws_n +
4171 			sizeof(struct mlx5_counter_stats_mem_mng);
4172 	size_t pgsize = rte_mem_page_size();
4173 	if (pgsize == (size_t)-1) {
4174 		DRV_LOG(ERR, "Failed to get mem page size");
4175 		rte_errno = ENOMEM;
4176 		return NULL;
4177 	}
4178 	uint8_t *mem = mlx5_malloc(MLX5_MEM_ZERO, size, pgsize,
4179 				  SOCKET_ID_ANY);
4180 	int i;
4181 
4182 	if (!mem) {
4183 		rte_errno = ENOMEM;
4184 		return NULL;
4185 	}
4186 	mem_mng = (struct mlx5_counter_stats_mem_mng *)(mem + size) - 1;
4187 	size = sizeof(*raw_data) * MLX5_COUNTERS_PER_POOL * raws_n;
4188 	mem_mng->umem = mlx5_glue->devx_umem_reg(sh->ctx, mem, size,
4189 						 IBV_ACCESS_LOCAL_WRITE);
4190 	if (!mem_mng->umem) {
4191 		rte_errno = errno;
4192 		mlx5_free(mem);
4193 		return NULL;
4194 	}
4195 	mkey_attr.addr = (uintptr_t)mem;
4196 	mkey_attr.size = size;
4197 	mkey_attr.umem_id = mlx5_os_get_umem_id(mem_mng->umem);
4198 	mkey_attr.pd = sh->pdn;
4199 	mkey_attr.log_entity_size = 0;
4200 	mkey_attr.pg_access = 0;
4201 	mkey_attr.klm_array = NULL;
4202 	mkey_attr.klm_num = 0;
4203 	if (priv->config.hca_attr.relaxed_ordering_write &&
4204 		priv->config.hca_attr.relaxed_ordering_read  &&
4205 		!haswell_broadwell_cpu)
4206 		mkey_attr.relaxed_ordering = 1;
4207 	mem_mng->dm = mlx5_devx_cmd_mkey_create(sh->ctx, &mkey_attr);
4208 	if (!mem_mng->dm) {
4209 		mlx5_glue->devx_umem_dereg(mem_mng->umem);
4210 		rte_errno = errno;
4211 		mlx5_free(mem);
4212 		return NULL;
4213 	}
4214 	mem_mng->raws = (struct mlx5_counter_stats_raw *)(mem + size);
4215 	raw_data = (volatile struct flow_counter_stats *)mem;
4216 	for (i = 0; i < raws_n; ++i) {
4217 		mem_mng->raws[i].mem_mng = mem_mng;
4218 		mem_mng->raws[i].data = raw_data + i * MLX5_COUNTERS_PER_POOL;
4219 	}
4220 	LIST_INSERT_HEAD(&sh->cmng.mem_mngs, mem_mng, next);
4221 	return mem_mng;
4222 }
4223 
4224 /**
4225  * Resize a counter container.
4226  *
4227  * @param[in] dev
4228  *   Pointer to the Ethernet device structure.
4229  * @param[in] batch
4230  *   Whether the pool is for counter that was allocated by batch command.
4231  * @param[in] age
4232  *   Whether the pool is for Aging counter.
4233  *
4234  * @return
4235  *   0 on success, otherwise negative errno value and rte_errno is set.
4236  */
4237 static int
4238 flow_dv_container_resize(struct rte_eth_dev *dev,
4239 				uint32_t batch, uint32_t age)
4240 {
4241 	struct mlx5_priv *priv = dev->data->dev_private;
4242 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4243 							       age);
4244 	struct mlx5_counter_stats_mem_mng *mem_mng = NULL;
4245 	void *old_pools = cont->pools;
4246 	uint32_t resize = cont->n + MLX5_CNT_CONTAINER_RESIZE;
4247 	uint32_t mem_size = sizeof(struct mlx5_flow_counter_pool *) * resize;
4248 	void *pools = mlx5_malloc(MLX5_MEM_ZERO, mem_size, 0, SOCKET_ID_ANY);
4249 
4250 	if (!pools) {
4251 		rte_errno = ENOMEM;
4252 		return -ENOMEM;
4253 	}
4254 	if (old_pools)
4255 		memcpy(pools, old_pools, cont->n *
4256 				       sizeof(struct mlx5_flow_counter_pool *));
4257 	/*
4258 	 * Fallback mode query the counter directly, no background query
4259 	 * resources are needed.
4260 	 */
4261 	if (!priv->counter_fallback) {
4262 		int i;
4263 
4264 		mem_mng = flow_dv_create_counter_stat_mem_mng(dev,
4265 			  MLX5_CNT_CONTAINER_RESIZE + MLX5_MAX_PENDING_QUERIES);
4266 		if (!mem_mng) {
4267 			mlx5_free(pools);
4268 			return -ENOMEM;
4269 		}
4270 		for (i = 0; i < MLX5_MAX_PENDING_QUERIES; ++i)
4271 			LIST_INSERT_HEAD(&priv->sh->cmng.free_stat_raws,
4272 					 mem_mng->raws +
4273 					 MLX5_CNT_CONTAINER_RESIZE +
4274 					 i, next);
4275 	}
4276 	rte_spinlock_lock(&cont->resize_sl);
4277 	cont->n = resize;
4278 	cont->mem_mng = mem_mng;
4279 	cont->pools = pools;
4280 	rte_spinlock_unlock(&cont->resize_sl);
4281 	if (old_pools)
4282 		mlx5_free(old_pools);
4283 	return 0;
4284 }
4285 
4286 /**
4287  * Query a devx flow counter.
4288  *
4289  * @param[in] dev
4290  *   Pointer to the Ethernet device structure.
4291  * @param[in] cnt
4292  *   Index to the flow counter.
4293  * @param[out] pkts
4294  *   The statistics value of packets.
4295  * @param[out] bytes
4296  *   The statistics value of bytes.
4297  *
4298  * @return
4299  *   0 on success, otherwise a negative errno value and rte_errno is set.
4300  */
4301 static inline int
4302 _flow_dv_query_count(struct rte_eth_dev *dev, uint32_t counter, uint64_t *pkts,
4303 		     uint64_t *bytes)
4304 {
4305 	struct mlx5_priv *priv = dev->data->dev_private;
4306 	struct mlx5_flow_counter_pool *pool = NULL;
4307 	struct mlx5_flow_counter *cnt;
4308 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
4309 	int offset;
4310 
4311 	cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4312 	MLX5_ASSERT(pool);
4313 	if (counter < MLX5_CNT_BATCH_OFFSET) {
4314 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4315 		if (priv->counter_fallback)
4316 			return mlx5_devx_cmd_flow_counter_query(cnt_ext->dcs, 0,
4317 					0, pkts, bytes, 0, NULL, NULL, 0);
4318 	}
4319 
4320 	rte_spinlock_lock(&pool->sl);
4321 	/*
4322 	 * The single counters allocation may allocate smaller ID than the
4323 	 * current allocated in parallel to the host reading.
4324 	 * In this case the new counter values must be reported as 0.
4325 	 */
4326 	if (unlikely(cnt_ext && cnt_ext->dcs->id < pool->raw->min_dcs_id)) {
4327 		*pkts = 0;
4328 		*bytes = 0;
4329 	} else {
4330 		offset = MLX5_CNT_ARRAY_IDX(pool, cnt);
4331 		*pkts = rte_be_to_cpu_64(pool->raw->data[offset].hits);
4332 		*bytes = rte_be_to_cpu_64(pool->raw->data[offset].bytes);
4333 	}
4334 	rte_spinlock_unlock(&pool->sl);
4335 	return 0;
4336 }
4337 
4338 /**
4339  * Create and initialize a new counter pool.
4340  *
4341  * @param[in] dev
4342  *   Pointer to the Ethernet device structure.
4343  * @param[out] dcs
4344  *   The devX counter handle.
4345  * @param[in] batch
4346  *   Whether the pool is for counter that was allocated by batch command.
4347  * @param[in] age
4348  *   Whether the pool is for counter that was allocated for aging.
4349  * @param[in/out] cont_cur
4350  *   Pointer to the container pointer, it will be update in pool resize.
4351  *
4352  * @return
4353  *   The pool container pointer on success, NULL otherwise and rte_errno is set.
4354  */
4355 static struct mlx5_flow_counter_pool *
4356 flow_dv_pool_create(struct rte_eth_dev *dev, struct mlx5_devx_obj *dcs,
4357 		    uint32_t batch, uint32_t age)
4358 {
4359 	struct mlx5_priv *priv = dev->data->dev_private;
4360 	struct mlx5_flow_counter_pool *pool;
4361 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4362 							       age);
4363 	int16_t n_valid = rte_atomic16_read(&cont->n_valid);
4364 	uint32_t size = sizeof(*pool);
4365 
4366 	if (cont->n == n_valid && flow_dv_container_resize(dev, batch, age))
4367 		return NULL;
4368 	size += MLX5_COUNTERS_PER_POOL * CNT_SIZE;
4369 	size += (batch ? 0 : MLX5_COUNTERS_PER_POOL * CNTEXT_SIZE);
4370 	size += (!age ? 0 : MLX5_COUNTERS_PER_POOL * AGE_SIZE);
4371 	pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
4372 	if (!pool) {
4373 		rte_errno = ENOMEM;
4374 		return NULL;
4375 	}
4376 	pool->min_dcs = dcs;
4377 	if (!priv->counter_fallback)
4378 		pool->raw = cont->mem_mng->raws + n_valid %
4379 						      MLX5_CNT_CONTAINER_RESIZE;
4380 	pool->raw_hw = NULL;
4381 	pool->type = 0;
4382 	pool->type |= (batch ? 0 :  CNT_POOL_TYPE_EXT);
4383 	pool->type |= (!age ? 0 :  CNT_POOL_TYPE_AGE);
4384 	pool->query_gen = 0;
4385 	rte_spinlock_init(&pool->sl);
4386 	TAILQ_INIT(&pool->counters[0]);
4387 	TAILQ_INIT(&pool->counters[1]);
4388 	TAILQ_INSERT_HEAD(&cont->pool_list, pool, next);
4389 	pool->index = n_valid;
4390 	cont->pools[n_valid] = pool;
4391 	if (!batch) {
4392 		int base = RTE_ALIGN_FLOOR(dcs->id, MLX5_COUNTERS_PER_POOL);
4393 
4394 		if (base < cont->min_id)
4395 			cont->min_id = base;
4396 		if (base > cont->max_id)
4397 			cont->max_id = base + MLX5_COUNTERS_PER_POOL - 1;
4398 		cont->last_pool_idx = pool->index;
4399 	}
4400 	/* Pool initialization must be updated before host thread access. */
4401 	rte_cio_wmb();
4402 	rte_atomic16_add(&cont->n_valid, 1);
4403 	return pool;
4404 }
4405 
4406 /**
4407  * Restore skipped counters in the pool.
4408  *
4409  * As counter pool query requires the first counter dcs
4410  * ID start with 4 alinged, if the pool counters with
4411  * min_dcs ID are not aligned with 4, the counters will
4412  * be skipped.
4413  * Once other min_dcs ID less than these skipped counter
4414  * dcs ID appears, the skipped counters will be safe to
4415  * use.
4416  * Should be called when min_dcs is updated.
4417  *
4418  * @param[in] pool
4419  *   Current counter pool.
4420  * @param[in] last_min_dcs
4421  *   Last min_dcs.
4422  */
4423 static void
4424 flow_dv_counter_restore(struct mlx5_flow_counter_pool *pool,
4425 			struct mlx5_devx_obj *last_min_dcs)
4426 {
4427 	struct mlx5_flow_counter_ext *cnt_ext;
4428 	uint32_t offset, new_offset;
4429 	uint32_t skip_cnt = 0;
4430 	uint32_t i;
4431 
4432 	if (!pool->skip_cnt)
4433 		return;
4434 	/*
4435 	 * If last min_dcs is not valid. The skipped counter may even after
4436 	 * last min_dcs, set the offset to the whole pool.
4437 	 */
4438 	if (last_min_dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4439 		offset = MLX5_COUNTERS_PER_POOL;
4440 	else
4441 		offset = last_min_dcs->id % MLX5_COUNTERS_PER_POOL;
4442 	new_offset = pool->min_dcs->id % MLX5_COUNTERS_PER_POOL;
4443 	/*
4444 	 * Check the counters from 1 to the last_min_dcs range. Counters
4445 	 * before new min_dcs indicates pool still has skipped counters.
4446 	 * Counters be skipped after new min_dcs will be ready to use.
4447 	 * Offset 0 counter must be empty or min_dcs, start from 1.
4448 	 */
4449 	for (i = 1; i < offset; i++) {
4450 		cnt_ext = MLX5_GET_POOL_CNT_EXT(pool, i);
4451 		if (cnt_ext->skipped) {
4452 			if (i > new_offset) {
4453 				cnt_ext->skipped = 0;
4454 				TAILQ_INSERT_TAIL
4455 					(&pool->counters[pool->query_gen],
4456 					 MLX5_POOL_GET_CNT(pool, i), next);
4457 			} else {
4458 				skip_cnt++;
4459 			}
4460 		}
4461 	}
4462 	if (!skip_cnt)
4463 		pool->skip_cnt = 0;
4464 }
4465 
4466 /**
4467  * Prepare a new counter and/or a new counter pool.
4468  *
4469  * @param[in] dev
4470  *   Pointer to the Ethernet device structure.
4471  * @param[out] cnt_free
4472  *   Where to put the pointer of a new counter.
4473  * @param[in] batch
4474  *   Whether the pool is for counter that was allocated by batch command.
4475  * @param[in] age
4476  *   Whether the pool is for counter that was allocated for aging.
4477  *
4478  * @return
4479  *   The counter pool pointer and @p cnt_free is set on success,
4480  *   NULL otherwise and rte_errno is set.
4481  */
4482 static struct mlx5_flow_counter_pool *
4483 flow_dv_counter_pool_prepare(struct rte_eth_dev *dev,
4484 			     struct mlx5_flow_counter **cnt_free,
4485 			     uint32_t batch, uint32_t age)
4486 {
4487 	struct mlx5_priv *priv = dev->data->dev_private;
4488 	struct mlx5_pools_container *cont;
4489 	struct mlx5_flow_counter_pool *pool;
4490 	struct mlx5_counters tmp_tq;
4491 	struct mlx5_devx_obj *last_min_dcs;
4492 	struct mlx5_devx_obj *dcs = NULL;
4493 	struct mlx5_flow_counter *cnt;
4494 	uint32_t add2other;
4495 	uint32_t i;
4496 
4497 	cont = MLX5_CNT_CONTAINER(priv->sh, batch, age);
4498 	if (!batch) {
4499 retry:
4500 		add2other = 0;
4501 		/* bulk_bitmap must be 0 for single counter allocation. */
4502 		dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0);
4503 		if (!dcs)
4504 			return NULL;
4505 		pool = flow_dv_find_pool_by_id(cont, dcs->id);
4506 		/* Check if counter belongs to exist pool ID range. */
4507 		if (!pool) {
4508 			pool = flow_dv_find_pool_by_id
4509 			       (MLX5_CNT_CONTAINER
4510 			       (priv->sh, batch, (age ^ 0x1)), dcs->id);
4511 			/*
4512 			 * Pool eixsts, counter will be added to the other
4513 			 * container, need to reallocate it later.
4514 			 */
4515 			if (pool) {
4516 				add2other = 1;
4517 			} else {
4518 				pool = flow_dv_pool_create(dev, dcs, batch,
4519 							   age);
4520 				if (!pool) {
4521 					mlx5_devx_cmd_destroy(dcs);
4522 					return NULL;
4523 				}
4524 			}
4525 		}
4526 		if ((dcs->id < pool->min_dcs->id ||
4527 		    pool->min_dcs->id &
4528 		    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1)) &&
4529 		    !(dcs->id & (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))) {
4530 			/*
4531 			 * Update the pool min_dcs only if current dcs is
4532 			 * valid and exist min_dcs is not valid or greater
4533 			 * than new dcs.
4534 			 */
4535 			last_min_dcs = pool->min_dcs;
4536 			rte_atomic64_set(&pool->a64_dcs,
4537 					 (int64_t)(uintptr_t)dcs);
4538 			/*
4539 			 * Restore any skipped counters if the new min_dcs
4540 			 * ID is smaller or min_dcs is not valid.
4541 			 */
4542 			if (dcs->id < last_min_dcs->id ||
4543 			    last_min_dcs->id &
4544 			    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1))
4545 				flow_dv_counter_restore(pool, last_min_dcs);
4546 		}
4547 		i = dcs->id % MLX5_COUNTERS_PER_POOL;
4548 		cnt = MLX5_POOL_GET_CNT(pool, i);
4549 		cnt->pool = pool;
4550 		MLX5_GET_POOL_CNT_EXT(pool, i)->dcs = dcs;
4551 		/*
4552 		 * If min_dcs is not valid, it means the new allocated dcs
4553 		 * also fail to become the valid min_dcs, just skip it.
4554 		 * Or if min_dcs is valid, and new dcs ID is smaller than
4555 		 * min_dcs, but not become the min_dcs, also skip it.
4556 		 */
4557 		if (pool->min_dcs->id &
4558 		    (MLX5_CNT_BATCH_QUERY_ID_ALIGNMENT - 1) ||
4559 		    dcs->id < pool->min_dcs->id) {
4560 			MLX5_GET_POOL_CNT_EXT(pool, i)->skipped = 1;
4561 			pool->skip_cnt = 1;
4562 			goto retry;
4563 		}
4564 		if (add2other) {
4565 			TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen],
4566 					  cnt, next);
4567 			goto retry;
4568 		}
4569 		*cnt_free = cnt;
4570 		return pool;
4571 	}
4572 	/* bulk_bitmap is in 128 counters units. */
4573 	if (priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4)
4574 		dcs = mlx5_devx_cmd_flow_counter_alloc(priv->sh->ctx, 0x4);
4575 	if (!dcs) {
4576 		rte_errno = ENODATA;
4577 		return NULL;
4578 	}
4579 	pool = flow_dv_pool_create(dev, dcs, batch, age);
4580 	if (!pool) {
4581 		mlx5_devx_cmd_destroy(dcs);
4582 		return NULL;
4583 	}
4584 	TAILQ_INIT(&tmp_tq);
4585 	for (i = 1; i < MLX5_COUNTERS_PER_POOL; ++i) {
4586 		cnt = MLX5_POOL_GET_CNT(pool, i);
4587 		cnt->pool = pool;
4588 		TAILQ_INSERT_HEAD(&tmp_tq, cnt, next);
4589 	}
4590 	rte_spinlock_lock(&cont->csl);
4591 	TAILQ_CONCAT(&cont->counters, &tmp_tq, next);
4592 	rte_spinlock_unlock(&cont->csl);
4593 	*cnt_free = MLX5_POOL_GET_CNT(pool, 0);
4594 	(*cnt_free)->pool = pool;
4595 	return pool;
4596 }
4597 
4598 /**
4599  * Search for existed shared counter.
4600  *
4601  * @param[in] dev
4602  *   Pointer to the Ethernet device structure.
4603  * @param[in] id
4604  *   The shared counter ID to search.
4605  * @param[out] ppool
4606  *   mlx5 flow counter pool in the container,
4607  *
4608  * @return
4609  *   NULL if not existed, otherwise pointer to the shared extend counter.
4610  */
4611 static struct mlx5_flow_counter_ext *
4612 flow_dv_counter_shared_search(struct rte_eth_dev *dev, uint32_t id,
4613 			      struct mlx5_flow_counter_pool **ppool)
4614 {
4615 	struct mlx5_priv *priv = dev->data->dev_private;
4616 	union mlx5_l3t_data data;
4617 	uint32_t cnt_idx;
4618 
4619 	if (mlx5_l3t_get_entry(priv->sh->cnt_id_tbl, id, &data) || !data.dword)
4620 		return NULL;
4621 	cnt_idx = data.dword;
4622 	/*
4623 	 * Shared counters don't have age info. The counter extend is after
4624 	 * the counter datat structure.
4625 	 */
4626 	return (struct mlx5_flow_counter_ext *)
4627 	       ((flow_dv_counter_get_by_idx(dev, cnt_idx, ppool)) + 1);
4628 }
4629 
4630 /**
4631  * Allocate a flow counter.
4632  *
4633  * @param[in] dev
4634  *   Pointer to the Ethernet device structure.
4635  * @param[in] shared
4636  *   Indicate if this counter is shared with other flows.
4637  * @param[in] id
4638  *   Counter identifier.
4639  * @param[in] group
4640  *   Counter flow group.
4641  * @param[in] age
4642  *   Whether the counter was allocated for aging.
4643  *
4644  * @return
4645  *   Index to flow counter on success, 0 otherwise and rte_errno is set.
4646  */
4647 static uint32_t
4648 flow_dv_counter_alloc(struct rte_eth_dev *dev, uint32_t shared, uint32_t id,
4649 		      uint16_t group, uint32_t age)
4650 {
4651 	struct mlx5_priv *priv = dev->data->dev_private;
4652 	struct mlx5_flow_counter_pool *pool = NULL;
4653 	struct mlx5_flow_counter *cnt_free = NULL;
4654 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
4655 	/*
4656 	 * Currently group 0 flow counter cannot be assigned to a flow if it is
4657 	 * not the first one in the batch counter allocation, so it is better
4658 	 * to allocate counters one by one for these flows in a separate
4659 	 * container.
4660 	 * A counter can be shared between different groups so need to take
4661 	 * shared counters from the single container.
4662 	 */
4663 	uint32_t batch = (group && !shared && !priv->counter_fallback) ? 1 : 0;
4664 	struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(priv->sh, batch,
4665 							       age);
4666 	uint32_t cnt_idx;
4667 
4668 	if (!priv->config.devx) {
4669 		rte_errno = ENOTSUP;
4670 		return 0;
4671 	}
4672 	if (shared) {
4673 		cnt_ext = flow_dv_counter_shared_search(dev, id, &pool);
4674 		if (cnt_ext) {
4675 			if (cnt_ext->ref_cnt + 1 == 0) {
4676 				rte_errno = E2BIG;
4677 				return 0;
4678 			}
4679 			cnt_ext->ref_cnt++;
4680 			cnt_idx = pool->index * MLX5_COUNTERS_PER_POOL +
4681 				  (cnt_ext->dcs->id % MLX5_COUNTERS_PER_POOL)
4682 				  + 1;
4683 			return cnt_idx;
4684 		}
4685 	}
4686 	/* Get free counters from container. */
4687 	rte_spinlock_lock(&cont->csl);
4688 	cnt_free = TAILQ_FIRST(&cont->counters);
4689 	if (cnt_free)
4690 		TAILQ_REMOVE(&cont->counters, cnt_free, next);
4691 	rte_spinlock_unlock(&cont->csl);
4692 	if (!cnt_free && !flow_dv_counter_pool_prepare(dev, &cnt_free,
4693 						       batch, age))
4694 		goto err;
4695 	pool = cnt_free->pool;
4696 	if (!batch)
4697 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt_free);
4698 	/* Create a DV counter action only in the first time usage. */
4699 	if (!cnt_free->action) {
4700 		uint16_t offset;
4701 		struct mlx5_devx_obj *dcs;
4702 		int ret;
4703 
4704 		if (batch) {
4705 			offset = MLX5_CNT_ARRAY_IDX(pool, cnt_free);
4706 			dcs = pool->min_dcs;
4707 		} else {
4708 			offset = 0;
4709 			dcs = cnt_ext->dcs;
4710 		}
4711 		ret = mlx5_flow_os_create_flow_action_count(dcs->obj, offset,
4712 							    &cnt_free->action);
4713 		if (ret) {
4714 			rte_errno = errno;
4715 			goto err;
4716 		}
4717 	}
4718 	cnt_idx = MLX5_MAKE_CNT_IDX(pool->index,
4719 				MLX5_CNT_ARRAY_IDX(pool, cnt_free));
4720 	cnt_idx += batch * MLX5_CNT_BATCH_OFFSET;
4721 	cnt_idx += age * MLX5_CNT_AGE_OFFSET;
4722 	/* Update the counter reset values. */
4723 	if (_flow_dv_query_count(dev, cnt_idx, &cnt_free->hits,
4724 				 &cnt_free->bytes))
4725 		goto err;
4726 	if (cnt_ext) {
4727 		cnt_ext->shared = shared;
4728 		cnt_ext->ref_cnt = 1;
4729 		cnt_ext->id = id;
4730 		if (shared) {
4731 			union mlx5_l3t_data data;
4732 
4733 			data.dword = cnt_idx;
4734 			if (mlx5_l3t_set_entry(priv->sh->cnt_id_tbl, id, &data))
4735 				return 0;
4736 		}
4737 	}
4738 	if (!priv->counter_fallback && !priv->sh->cmng.query_thread_on)
4739 		/* Start the asynchronous batch query by the host thread. */
4740 		mlx5_set_query_alarm(priv->sh);
4741 	return cnt_idx;
4742 err:
4743 	if (cnt_free) {
4744 		cnt_free->pool = pool;
4745 		rte_spinlock_lock(&cont->csl);
4746 		TAILQ_INSERT_TAIL(&cont->counters, cnt_free, next);
4747 		rte_spinlock_unlock(&cont->csl);
4748 	}
4749 	return 0;
4750 }
4751 
4752 /**
4753  * Get age param from counter index.
4754  *
4755  * @param[in] dev
4756  *   Pointer to the Ethernet device structure.
4757  * @param[in] counter
4758  *   Index to the counter handler.
4759  *
4760  * @return
4761  *   The aging parameter specified for the counter index.
4762  */
4763 static struct mlx5_age_param*
4764 flow_dv_counter_idx_get_age(struct rte_eth_dev *dev,
4765 				uint32_t counter)
4766 {
4767 	struct mlx5_flow_counter *cnt;
4768 	struct mlx5_flow_counter_pool *pool = NULL;
4769 
4770 	flow_dv_counter_get_by_idx(dev, counter, &pool);
4771 	counter = (counter - 1) % MLX5_COUNTERS_PER_POOL;
4772 	cnt = MLX5_POOL_GET_CNT(pool, counter);
4773 	return MLX5_CNT_TO_AGE(cnt);
4774 }
4775 
4776 /**
4777  * Remove a flow counter from aged counter list.
4778  *
4779  * @param[in] dev
4780  *   Pointer to the Ethernet device structure.
4781  * @param[in] counter
4782  *   Index to the counter handler.
4783  * @param[in] cnt
4784  *   Pointer to the counter handler.
4785  */
4786 static void
4787 flow_dv_counter_remove_from_age(struct rte_eth_dev *dev,
4788 				uint32_t counter, struct mlx5_flow_counter *cnt)
4789 {
4790 	struct mlx5_age_info *age_info;
4791 	struct mlx5_age_param *age_param;
4792 	struct mlx5_priv *priv = dev->data->dev_private;
4793 
4794 	age_info = GET_PORT_AGE_INFO(priv);
4795 	age_param = flow_dv_counter_idx_get_age(dev, counter);
4796 	if (rte_atomic16_cmpset((volatile uint16_t *)
4797 			&age_param->state,
4798 			AGE_CANDIDATE, AGE_FREE)
4799 			!= AGE_CANDIDATE) {
4800 		/**
4801 		 * We need the lock even it is age timeout,
4802 		 * since counter may still in process.
4803 		 */
4804 		rte_spinlock_lock(&age_info->aged_sl);
4805 		TAILQ_REMOVE(&age_info->aged_counters, cnt, next);
4806 		rte_spinlock_unlock(&age_info->aged_sl);
4807 	}
4808 	rte_atomic16_set(&age_param->state, AGE_FREE);
4809 }
4810 /**
4811  * Release a flow counter.
4812  *
4813  * @param[in] dev
4814  *   Pointer to the Ethernet device structure.
4815  * @param[in] counter
4816  *   Index to the counter handler.
4817  */
4818 static void
4819 flow_dv_counter_release(struct rte_eth_dev *dev, uint32_t counter)
4820 {
4821 	struct mlx5_priv *priv = dev->data->dev_private;
4822 	struct mlx5_flow_counter_pool *pool = NULL;
4823 	struct mlx5_flow_counter *cnt;
4824 	struct mlx5_flow_counter_ext *cnt_ext = NULL;
4825 
4826 	if (!counter)
4827 		return;
4828 	cnt = flow_dv_counter_get_by_idx(dev, counter, &pool);
4829 	MLX5_ASSERT(pool);
4830 	if (counter < MLX5_CNT_BATCH_OFFSET) {
4831 		cnt_ext = MLX5_CNT_TO_CNT_EXT(pool, cnt);
4832 		if (cnt_ext) {
4833 			if (--cnt_ext->ref_cnt)
4834 				return;
4835 			if (cnt_ext->shared)
4836 				mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl,
4837 						     cnt_ext->id);
4838 		}
4839 	}
4840 	if (IS_AGE_POOL(pool))
4841 		flow_dv_counter_remove_from_age(dev, counter, cnt);
4842 	cnt->pool = pool;
4843 	/*
4844 	 * Put the counter back to list to be updated in none fallback mode.
4845 	 * Currently, we are using two list alternately, while one is in query,
4846 	 * add the freed counter to the other list based on the pool query_gen
4847 	 * value. After query finishes, add counter the list to the global
4848 	 * container counter list. The list changes while query starts. In
4849 	 * this case, lock will not be needed as query callback and release
4850 	 * function both operate with the different list.
4851 	 *
4852 	 */
4853 	if (!priv->counter_fallback)
4854 		TAILQ_INSERT_TAIL(&pool->counters[pool->query_gen], cnt, next);
4855 	else
4856 		TAILQ_INSERT_TAIL(&((MLX5_CNT_CONTAINER
4857 				  (priv->sh, 0, 0))->counters),
4858 				  cnt, next);
4859 }
4860 
4861 /**
4862  * Verify the @p attributes will be correctly understood by the NIC and store
4863  * them in the @p flow if everything is correct.
4864  *
4865  * @param[in] dev
4866  *   Pointer to dev struct.
4867  * @param[in] attributes
4868  *   Pointer to flow attributes
4869  * @param[in] external
4870  *   This flow rule is created by request external to PMD.
4871  * @param[out] error
4872  *   Pointer to error structure.
4873  *
4874  * @return
4875  *   - 0 on success and non root table.
4876  *   - 1 on success and root table.
4877  *   - a negative errno value otherwise and rte_errno is set.
4878  */
4879 static int
4880 flow_dv_validate_attributes(struct rte_eth_dev *dev,
4881 			    const struct rte_flow_attr *attributes,
4882 			    bool external __rte_unused,
4883 			    struct rte_flow_error *error)
4884 {
4885 	struct mlx5_priv *priv = dev->data->dev_private;
4886 	uint32_t priority_max = priv->config.flow_prio - 1;
4887 	int ret = 0;
4888 
4889 #ifndef HAVE_MLX5DV_DR
4890 	if (attributes->group)
4891 		return rte_flow_error_set(error, ENOTSUP,
4892 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
4893 					  NULL,
4894 					  "groups are not supported");
4895 #else
4896 	uint32_t table = 0;
4897 
4898 	ret = mlx5_flow_group_to_table(attributes, external,
4899 				       attributes->group, !!priv->fdb_def_rule,
4900 				       &table, error);
4901 	if (ret)
4902 		return ret;
4903 	if (!table)
4904 		ret = MLX5DV_DR_ACTION_FLAGS_ROOT_LEVEL;
4905 #endif
4906 	if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
4907 	    attributes->priority >= priority_max)
4908 		return rte_flow_error_set(error, ENOTSUP,
4909 					  RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
4910 					  NULL,
4911 					  "priority out of range");
4912 	if (attributes->transfer) {
4913 		if (!priv->config.dv_esw_en)
4914 			return rte_flow_error_set
4915 				(error, ENOTSUP,
4916 				 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4917 				 "E-Switch dr is not supported");
4918 		if (!(priv->representor || priv->master))
4919 			return rte_flow_error_set
4920 				(error, EINVAL, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4921 				 NULL, "E-Switch configuration can only be"
4922 				 " done by a master or a representor device");
4923 		if (attributes->egress)
4924 			return rte_flow_error_set
4925 				(error, ENOTSUP,
4926 				 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attributes,
4927 				 "egress is not supported");
4928 	}
4929 	if (!(attributes->egress ^ attributes->ingress))
4930 		return rte_flow_error_set(error, ENOTSUP,
4931 					  RTE_FLOW_ERROR_TYPE_ATTR, NULL,
4932 					  "must specify exactly one of "
4933 					  "ingress or egress");
4934 	return ret;
4935 }
4936 
4937 /**
4938  * Internal validation function. For validating both actions and items.
4939  *
4940  * @param[in] dev
4941  *   Pointer to the rte_eth_dev structure.
4942  * @param[in] attr
4943  *   Pointer to the flow attributes.
4944  * @param[in] items
4945  *   Pointer to the list of items.
4946  * @param[in] actions
4947  *   Pointer to the list of actions.
4948  * @param[in] external
4949  *   This flow rule is created by request external to PMD.
4950  * @param[in] hairpin
4951  *   Number of hairpin TX actions, 0 means classic flow.
4952  * @param[out] error
4953  *   Pointer to the error structure.
4954  *
4955  * @return
4956  *   0 on success, a negative errno value otherwise and rte_errno is set.
4957  */
4958 static int
4959 flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
4960 		 const struct rte_flow_item items[],
4961 		 const struct rte_flow_action actions[],
4962 		 bool external, int hairpin, struct rte_flow_error *error)
4963 {
4964 	int ret;
4965 	uint64_t action_flags = 0;
4966 	uint64_t item_flags = 0;
4967 	uint64_t last_item = 0;
4968 	uint8_t next_protocol = 0xff;
4969 	uint16_t ether_type = 0;
4970 	int actions_n = 0;
4971 	uint8_t item_ipv6_proto = 0;
4972 	const struct rte_flow_item *gre_item = NULL;
4973 	const struct rte_flow_action_raw_decap *decap;
4974 	const struct rte_flow_action_raw_encap *encap;
4975 	const struct rte_flow_action_rss *rss;
4976 	const struct rte_flow_item_tcp nic_tcp_mask = {
4977 		.hdr = {
4978 			.tcp_flags = 0xFF,
4979 			.src_port = RTE_BE16(UINT16_MAX),
4980 			.dst_port = RTE_BE16(UINT16_MAX),
4981 		}
4982 	};
4983 	const struct rte_flow_item_ipv4 nic_ipv4_mask = {
4984 		.hdr = {
4985 			.src_addr = RTE_BE32(0xffffffff),
4986 			.dst_addr = RTE_BE32(0xffffffff),
4987 			.type_of_service = 0xff,
4988 			.next_proto_id = 0xff,
4989 			.time_to_live = 0xff,
4990 		},
4991 	};
4992 	const struct rte_flow_item_ipv6 nic_ipv6_mask = {
4993 		.hdr = {
4994 			.src_addr =
4995 			"\xff\xff\xff\xff\xff\xff\xff\xff"
4996 			"\xff\xff\xff\xff\xff\xff\xff\xff",
4997 			.dst_addr =
4998 			"\xff\xff\xff\xff\xff\xff\xff\xff"
4999 			"\xff\xff\xff\xff\xff\xff\xff\xff",
5000 			.vtc_flow = RTE_BE32(0xffffffff),
5001 			.proto = 0xff,
5002 			.hop_limits = 0xff,
5003 		},
5004 	};
5005 	const struct rte_flow_item_ecpri nic_ecpri_mask = {
5006 		.hdr = {
5007 			.common = {
5008 				.u32 =
5009 				RTE_BE32(((const struct rte_ecpri_common_hdr) {
5010 					.type = 0xFF,
5011 					}).u32),
5012 			},
5013 			.dummy[0] = 0xffffffff,
5014 		},
5015 	};
5016 	struct mlx5_priv *priv = dev->data->dev_private;
5017 	struct mlx5_dev_config *dev_conf = &priv->config;
5018 	uint16_t queue_index = 0xFFFF;
5019 	const struct rte_flow_item_vlan *vlan_m = NULL;
5020 	int16_t rw_act_num = 0;
5021 	uint64_t is_root;
5022 
5023 	if (items == NULL)
5024 		return -1;
5025 	ret = flow_dv_validate_attributes(dev, attr, external, error);
5026 	if (ret < 0)
5027 		return ret;
5028 	is_root = (uint64_t)ret;
5029 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
5030 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
5031 		int type = items->type;
5032 
5033 		if (!mlx5_flow_os_item_supported(type))
5034 			return rte_flow_error_set(error, ENOTSUP,
5035 						  RTE_FLOW_ERROR_TYPE_ITEM,
5036 						  NULL, "item not supported");
5037 		switch (type) {
5038 		case RTE_FLOW_ITEM_TYPE_VOID:
5039 			break;
5040 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
5041 			ret = flow_dv_validate_item_port_id
5042 					(dev, items, attr, item_flags, error);
5043 			if (ret < 0)
5044 				return ret;
5045 			last_item = MLX5_FLOW_ITEM_PORT_ID;
5046 			break;
5047 		case RTE_FLOW_ITEM_TYPE_ETH:
5048 			ret = mlx5_flow_validate_item_eth(items, item_flags,
5049 							  error);
5050 			if (ret < 0)
5051 				return ret;
5052 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
5053 					     MLX5_FLOW_LAYER_OUTER_L2;
5054 			if (items->mask != NULL && items->spec != NULL) {
5055 				ether_type =
5056 					((const struct rte_flow_item_eth *)
5057 					 items->spec)->type;
5058 				ether_type &=
5059 					((const struct rte_flow_item_eth *)
5060 					 items->mask)->type;
5061 				ether_type = rte_be_to_cpu_16(ether_type);
5062 			} else {
5063 				ether_type = 0;
5064 			}
5065 			break;
5066 		case RTE_FLOW_ITEM_TYPE_VLAN:
5067 			ret = flow_dv_validate_item_vlan(items, item_flags,
5068 							 dev, error);
5069 			if (ret < 0)
5070 				return ret;
5071 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
5072 					     MLX5_FLOW_LAYER_OUTER_VLAN;
5073 			if (items->mask != NULL && items->spec != NULL) {
5074 				ether_type =
5075 					((const struct rte_flow_item_vlan *)
5076 					 items->spec)->inner_type;
5077 				ether_type &=
5078 					((const struct rte_flow_item_vlan *)
5079 					 items->mask)->inner_type;
5080 				ether_type = rte_be_to_cpu_16(ether_type);
5081 			} else {
5082 				ether_type = 0;
5083 			}
5084 			/* Store outer VLAN mask for of_push_vlan action. */
5085 			if (!tunnel)
5086 				vlan_m = items->mask;
5087 			break;
5088 		case RTE_FLOW_ITEM_TYPE_IPV4:
5089 			mlx5_flow_tunnel_ip_check(items, next_protocol,
5090 						  &item_flags, &tunnel);
5091 			ret = mlx5_flow_validate_item_ipv4(items, item_flags,
5092 							   last_item,
5093 							   ether_type,
5094 							   &nic_ipv4_mask,
5095 							   error);
5096 			if (ret < 0)
5097 				return ret;
5098 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
5099 					     MLX5_FLOW_LAYER_OUTER_L3_IPV4;
5100 			if (items->mask != NULL &&
5101 			    ((const struct rte_flow_item_ipv4 *)
5102 			     items->mask)->hdr.next_proto_id) {
5103 				next_protocol =
5104 					((const struct rte_flow_item_ipv4 *)
5105 					 (items->spec))->hdr.next_proto_id;
5106 				next_protocol &=
5107 					((const struct rte_flow_item_ipv4 *)
5108 					 (items->mask))->hdr.next_proto_id;
5109 			} else {
5110 				/* Reset for inner layer. */
5111 				next_protocol = 0xff;
5112 			}
5113 			break;
5114 		case RTE_FLOW_ITEM_TYPE_IPV6:
5115 			mlx5_flow_tunnel_ip_check(items, next_protocol,
5116 						  &item_flags, &tunnel);
5117 			ret = mlx5_flow_validate_item_ipv6(items, item_flags,
5118 							   last_item,
5119 							   ether_type,
5120 							   &nic_ipv6_mask,
5121 							   error);
5122 			if (ret < 0)
5123 				return ret;
5124 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
5125 					     MLX5_FLOW_LAYER_OUTER_L3_IPV6;
5126 			if (items->mask != NULL &&
5127 			    ((const struct rte_flow_item_ipv6 *)
5128 			     items->mask)->hdr.proto) {
5129 				item_ipv6_proto =
5130 					((const struct rte_flow_item_ipv6 *)
5131 					 items->spec)->hdr.proto;
5132 				next_protocol =
5133 					((const struct rte_flow_item_ipv6 *)
5134 					 items->spec)->hdr.proto;
5135 				next_protocol &=
5136 					((const struct rte_flow_item_ipv6 *)
5137 					 items->mask)->hdr.proto;
5138 			} else {
5139 				/* Reset for inner layer. */
5140 				next_protocol = 0xff;
5141 			}
5142 			break;
5143 		case RTE_FLOW_ITEM_TYPE_TCP:
5144 			ret = mlx5_flow_validate_item_tcp
5145 						(items, item_flags,
5146 						 next_protocol,
5147 						 &nic_tcp_mask,
5148 						 error);
5149 			if (ret < 0)
5150 				return ret;
5151 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
5152 					     MLX5_FLOW_LAYER_OUTER_L4_TCP;
5153 			break;
5154 		case RTE_FLOW_ITEM_TYPE_UDP:
5155 			ret = mlx5_flow_validate_item_udp(items, item_flags,
5156 							  next_protocol,
5157 							  error);
5158 			if (ret < 0)
5159 				return ret;
5160 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
5161 					     MLX5_FLOW_LAYER_OUTER_L4_UDP;
5162 			break;
5163 		case RTE_FLOW_ITEM_TYPE_GRE:
5164 			ret = mlx5_flow_validate_item_gre(items, item_flags,
5165 							  next_protocol, error);
5166 			if (ret < 0)
5167 				return ret;
5168 			gre_item = items;
5169 			last_item = MLX5_FLOW_LAYER_GRE;
5170 			break;
5171 		case RTE_FLOW_ITEM_TYPE_NVGRE:
5172 			ret = mlx5_flow_validate_item_nvgre(items, item_flags,
5173 							    next_protocol,
5174 							    error);
5175 			if (ret < 0)
5176 				return ret;
5177 			last_item = MLX5_FLOW_LAYER_NVGRE;
5178 			break;
5179 		case RTE_FLOW_ITEM_TYPE_GRE_KEY:
5180 			ret = mlx5_flow_validate_item_gre_key
5181 				(items, item_flags, gre_item, error);
5182 			if (ret < 0)
5183 				return ret;
5184 			last_item = MLX5_FLOW_LAYER_GRE_KEY;
5185 			break;
5186 		case RTE_FLOW_ITEM_TYPE_VXLAN:
5187 			ret = mlx5_flow_validate_item_vxlan(items, item_flags,
5188 							    error);
5189 			if (ret < 0)
5190 				return ret;
5191 			last_item = MLX5_FLOW_LAYER_VXLAN;
5192 			break;
5193 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
5194 			ret = mlx5_flow_validate_item_vxlan_gpe(items,
5195 								item_flags, dev,
5196 								error);
5197 			if (ret < 0)
5198 				return ret;
5199 			last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
5200 			break;
5201 		case RTE_FLOW_ITEM_TYPE_GENEVE:
5202 			ret = mlx5_flow_validate_item_geneve(items,
5203 							     item_flags, dev,
5204 							     error);
5205 			if (ret < 0)
5206 				return ret;
5207 			last_item = MLX5_FLOW_LAYER_GENEVE;
5208 			break;
5209 		case RTE_FLOW_ITEM_TYPE_MPLS:
5210 			ret = mlx5_flow_validate_item_mpls(dev, items,
5211 							   item_flags,
5212 							   last_item, error);
5213 			if (ret < 0)
5214 				return ret;
5215 			last_item = MLX5_FLOW_LAYER_MPLS;
5216 			break;
5217 
5218 		case RTE_FLOW_ITEM_TYPE_MARK:
5219 			ret = flow_dv_validate_item_mark(dev, items, attr,
5220 							 error);
5221 			if (ret < 0)
5222 				return ret;
5223 			last_item = MLX5_FLOW_ITEM_MARK;
5224 			break;
5225 		case RTE_FLOW_ITEM_TYPE_META:
5226 			ret = flow_dv_validate_item_meta(dev, items, attr,
5227 							 error);
5228 			if (ret < 0)
5229 				return ret;
5230 			last_item = MLX5_FLOW_ITEM_METADATA;
5231 			break;
5232 		case RTE_FLOW_ITEM_TYPE_ICMP:
5233 			ret = mlx5_flow_validate_item_icmp(items, item_flags,
5234 							   next_protocol,
5235 							   error);
5236 			if (ret < 0)
5237 				return ret;
5238 			last_item = MLX5_FLOW_LAYER_ICMP;
5239 			break;
5240 		case RTE_FLOW_ITEM_TYPE_ICMP6:
5241 			ret = mlx5_flow_validate_item_icmp6(items, item_flags,
5242 							    next_protocol,
5243 							    error);
5244 			if (ret < 0)
5245 				return ret;
5246 			item_ipv6_proto = IPPROTO_ICMPV6;
5247 			last_item = MLX5_FLOW_LAYER_ICMP6;
5248 			break;
5249 		case RTE_FLOW_ITEM_TYPE_TAG:
5250 			ret = flow_dv_validate_item_tag(dev, items,
5251 							attr, error);
5252 			if (ret < 0)
5253 				return ret;
5254 			last_item = MLX5_FLOW_ITEM_TAG;
5255 			break;
5256 		case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
5257 		case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
5258 			break;
5259 		case RTE_FLOW_ITEM_TYPE_GTP:
5260 			ret = flow_dv_validate_item_gtp(dev, items, item_flags,
5261 							error);
5262 			if (ret < 0)
5263 				return ret;
5264 			last_item = MLX5_FLOW_LAYER_GTP;
5265 			break;
5266 		case RTE_FLOW_ITEM_TYPE_ECPRI:
5267 			/* Capacity will be checked in the translate stage. */
5268 			ret = mlx5_flow_validate_item_ecpri(items, item_flags,
5269 							    last_item,
5270 							    ether_type,
5271 							    &nic_ecpri_mask,
5272 							    error);
5273 			if (ret < 0)
5274 				return ret;
5275 			last_item = MLX5_FLOW_LAYER_ECPRI;
5276 			break;
5277 		default:
5278 			return rte_flow_error_set(error, ENOTSUP,
5279 						  RTE_FLOW_ERROR_TYPE_ITEM,
5280 						  NULL, "item not supported");
5281 		}
5282 		item_flags |= last_item;
5283 	}
5284 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5285 		int type = actions->type;
5286 
5287 		if (!mlx5_flow_os_action_supported(type))
5288 			return rte_flow_error_set(error, ENOTSUP,
5289 						  RTE_FLOW_ERROR_TYPE_ACTION,
5290 						  actions,
5291 						  "action not supported");
5292 		if (actions_n == MLX5_DV_MAX_NUMBER_OF_ACTIONS)
5293 			return rte_flow_error_set(error, ENOTSUP,
5294 						  RTE_FLOW_ERROR_TYPE_ACTION,
5295 						  actions, "too many actions");
5296 		switch (type) {
5297 		case RTE_FLOW_ACTION_TYPE_VOID:
5298 			break;
5299 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
5300 			ret = flow_dv_validate_action_port_id(dev,
5301 							      action_flags,
5302 							      actions,
5303 							      attr,
5304 							      error);
5305 			if (ret)
5306 				return ret;
5307 			action_flags |= MLX5_FLOW_ACTION_PORT_ID;
5308 			++actions_n;
5309 			break;
5310 		case RTE_FLOW_ACTION_TYPE_FLAG:
5311 			ret = flow_dv_validate_action_flag(dev, action_flags,
5312 							   attr, error);
5313 			if (ret < 0)
5314 				return ret;
5315 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5316 				/* Count all modify-header actions as one. */
5317 				if (!(action_flags &
5318 				      MLX5_FLOW_MODIFY_HDR_ACTIONS))
5319 					++actions_n;
5320 				action_flags |= MLX5_FLOW_ACTION_FLAG |
5321 						MLX5_FLOW_ACTION_MARK_EXT;
5322 			} else {
5323 				action_flags |= MLX5_FLOW_ACTION_FLAG;
5324 				++actions_n;
5325 			}
5326 			rw_act_num += MLX5_ACT_NUM_SET_MARK;
5327 			break;
5328 		case RTE_FLOW_ACTION_TYPE_MARK:
5329 			ret = flow_dv_validate_action_mark(dev, actions,
5330 							   action_flags,
5331 							   attr, error);
5332 			if (ret < 0)
5333 				return ret;
5334 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
5335 				/* Count all modify-header actions as one. */
5336 				if (!(action_flags &
5337 				      MLX5_FLOW_MODIFY_HDR_ACTIONS))
5338 					++actions_n;
5339 				action_flags |= MLX5_FLOW_ACTION_MARK |
5340 						MLX5_FLOW_ACTION_MARK_EXT;
5341 			} else {
5342 				action_flags |= MLX5_FLOW_ACTION_MARK;
5343 				++actions_n;
5344 			}
5345 			rw_act_num += MLX5_ACT_NUM_SET_MARK;
5346 			break;
5347 		case RTE_FLOW_ACTION_TYPE_SET_META:
5348 			ret = flow_dv_validate_action_set_meta(dev, actions,
5349 							       action_flags,
5350 							       attr, error);
5351 			if (ret < 0)
5352 				return ret;
5353 			/* Count all modify-header actions as one action. */
5354 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5355 				++actions_n;
5356 			action_flags |= MLX5_FLOW_ACTION_SET_META;
5357 			rw_act_num += MLX5_ACT_NUM_SET_META;
5358 			break;
5359 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
5360 			ret = flow_dv_validate_action_set_tag(dev, actions,
5361 							      action_flags,
5362 							      attr, error);
5363 			if (ret < 0)
5364 				return ret;
5365 			/* Count all modify-header actions as one action. */
5366 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5367 				++actions_n;
5368 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
5369 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5370 			break;
5371 		case RTE_FLOW_ACTION_TYPE_DROP:
5372 			ret = mlx5_flow_validate_action_drop(action_flags,
5373 							     attr, error);
5374 			if (ret < 0)
5375 				return ret;
5376 			action_flags |= MLX5_FLOW_ACTION_DROP;
5377 			++actions_n;
5378 			break;
5379 		case RTE_FLOW_ACTION_TYPE_QUEUE:
5380 			ret = mlx5_flow_validate_action_queue(actions,
5381 							      action_flags, dev,
5382 							      attr, error);
5383 			if (ret < 0)
5384 				return ret;
5385 			queue_index = ((const struct rte_flow_action_queue *)
5386 							(actions->conf))->index;
5387 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
5388 			++actions_n;
5389 			break;
5390 		case RTE_FLOW_ACTION_TYPE_RSS:
5391 			rss = actions->conf;
5392 			ret = mlx5_flow_validate_action_rss(actions,
5393 							    action_flags, dev,
5394 							    attr, item_flags,
5395 							    error);
5396 			if (ret < 0)
5397 				return ret;
5398 			if (rss != NULL && rss->queue_num)
5399 				queue_index = rss->queue[0];
5400 			action_flags |= MLX5_FLOW_ACTION_RSS;
5401 			++actions_n;
5402 			break;
5403 		case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
5404 			ret =
5405 			mlx5_flow_validate_action_default_miss(action_flags,
5406 					attr, error);
5407 			if (ret < 0)
5408 				return ret;
5409 			action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
5410 			++actions_n;
5411 			break;
5412 		case RTE_FLOW_ACTION_TYPE_COUNT:
5413 			ret = flow_dv_validate_action_count(dev, error);
5414 			if (ret < 0)
5415 				return ret;
5416 			action_flags |= MLX5_FLOW_ACTION_COUNT;
5417 			++actions_n;
5418 			break;
5419 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
5420 			if (flow_dv_validate_action_pop_vlan(dev,
5421 							     action_flags,
5422 							     actions,
5423 							     item_flags, attr,
5424 							     error))
5425 				return -rte_errno;
5426 			action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
5427 			++actions_n;
5428 			break;
5429 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
5430 			ret = flow_dv_validate_action_push_vlan(dev,
5431 								action_flags,
5432 								vlan_m,
5433 								actions, attr,
5434 								error);
5435 			if (ret < 0)
5436 				return ret;
5437 			action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
5438 			++actions_n;
5439 			break;
5440 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
5441 			ret = flow_dv_validate_action_set_vlan_pcp
5442 						(action_flags, actions, error);
5443 			if (ret < 0)
5444 				return ret;
5445 			/* Count PCP with push_vlan command. */
5446 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
5447 			break;
5448 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
5449 			ret = flow_dv_validate_action_set_vlan_vid
5450 						(item_flags, action_flags,
5451 						 actions, error);
5452 			if (ret < 0)
5453 				return ret;
5454 			/* Count VID with push_vlan command. */
5455 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
5456 			rw_act_num += MLX5_ACT_NUM_MDF_VID;
5457 			break;
5458 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
5459 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
5460 			ret = flow_dv_validate_action_l2_encap(dev,
5461 							       action_flags,
5462 							       actions, attr,
5463 							       error);
5464 			if (ret < 0)
5465 				return ret;
5466 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
5467 			++actions_n;
5468 			break;
5469 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
5470 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
5471 			ret = flow_dv_validate_action_decap(dev, action_flags,
5472 							    attr, error);
5473 			if (ret < 0)
5474 				return ret;
5475 			action_flags |= MLX5_FLOW_ACTION_DECAP;
5476 			++actions_n;
5477 			break;
5478 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
5479 			ret = flow_dv_validate_action_raw_encap_decap
5480 				(dev, NULL, actions->conf, attr, &action_flags,
5481 				 &actions_n, error);
5482 			if (ret < 0)
5483 				return ret;
5484 			break;
5485 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
5486 			decap = actions->conf;
5487 			while ((++actions)->type == RTE_FLOW_ACTION_TYPE_VOID)
5488 				;
5489 			if (actions->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
5490 				encap = NULL;
5491 				actions--;
5492 			} else {
5493 				encap = actions->conf;
5494 			}
5495 			ret = flow_dv_validate_action_raw_encap_decap
5496 					   (dev,
5497 					    decap ? decap : &empty_decap, encap,
5498 					    attr, &action_flags, &actions_n,
5499 					    error);
5500 			if (ret < 0)
5501 				return ret;
5502 			break;
5503 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
5504 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
5505 			ret = flow_dv_validate_action_modify_mac(action_flags,
5506 								 actions,
5507 								 item_flags,
5508 								 error);
5509 			if (ret < 0)
5510 				return ret;
5511 			/* Count all modify-header actions as one action. */
5512 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5513 				++actions_n;
5514 			action_flags |= actions->type ==
5515 					RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
5516 						MLX5_FLOW_ACTION_SET_MAC_SRC :
5517 						MLX5_FLOW_ACTION_SET_MAC_DST;
5518 			/*
5519 			 * Even if the source and destination MAC addresses have
5520 			 * overlap in the header with 4B alignment, the convert
5521 			 * function will handle them separately and 4 SW actions
5522 			 * will be created. And 2 actions will be added each
5523 			 * time no matter how many bytes of address will be set.
5524 			 */
5525 			rw_act_num += MLX5_ACT_NUM_MDF_MAC;
5526 			break;
5527 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
5528 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
5529 			ret = flow_dv_validate_action_modify_ipv4(action_flags,
5530 								  actions,
5531 								  item_flags,
5532 								  error);
5533 			if (ret < 0)
5534 				return ret;
5535 			/* Count all modify-header actions as one action. */
5536 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5537 				++actions_n;
5538 			action_flags |= actions->type ==
5539 					RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
5540 						MLX5_FLOW_ACTION_SET_IPV4_SRC :
5541 						MLX5_FLOW_ACTION_SET_IPV4_DST;
5542 			rw_act_num += MLX5_ACT_NUM_MDF_IPV4;
5543 			break;
5544 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
5545 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
5546 			ret = flow_dv_validate_action_modify_ipv6(action_flags,
5547 								  actions,
5548 								  item_flags,
5549 								  error);
5550 			if (ret < 0)
5551 				return ret;
5552 			if (item_ipv6_proto == IPPROTO_ICMPV6)
5553 				return rte_flow_error_set(error, ENOTSUP,
5554 					RTE_FLOW_ERROR_TYPE_ACTION,
5555 					actions,
5556 					"Can't change header "
5557 					"with ICMPv6 proto");
5558 			/* Count all modify-header actions as one action. */
5559 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5560 				++actions_n;
5561 			action_flags |= actions->type ==
5562 					RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
5563 						MLX5_FLOW_ACTION_SET_IPV6_SRC :
5564 						MLX5_FLOW_ACTION_SET_IPV6_DST;
5565 			rw_act_num += MLX5_ACT_NUM_MDF_IPV6;
5566 			break;
5567 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
5568 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
5569 			ret = flow_dv_validate_action_modify_tp(action_flags,
5570 								actions,
5571 								item_flags,
5572 								error);
5573 			if (ret < 0)
5574 				return ret;
5575 			/* Count all modify-header actions as one action. */
5576 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5577 				++actions_n;
5578 			action_flags |= actions->type ==
5579 					RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
5580 						MLX5_FLOW_ACTION_SET_TP_SRC :
5581 						MLX5_FLOW_ACTION_SET_TP_DST;
5582 			rw_act_num += MLX5_ACT_NUM_MDF_PORT;
5583 			break;
5584 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
5585 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
5586 			ret = flow_dv_validate_action_modify_ttl(action_flags,
5587 								 actions,
5588 								 item_flags,
5589 								 error);
5590 			if (ret < 0)
5591 				return ret;
5592 			/* Count all modify-header actions as one action. */
5593 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5594 				++actions_n;
5595 			action_flags |= actions->type ==
5596 					RTE_FLOW_ACTION_TYPE_SET_TTL ?
5597 						MLX5_FLOW_ACTION_SET_TTL :
5598 						MLX5_FLOW_ACTION_DEC_TTL;
5599 			rw_act_num += MLX5_ACT_NUM_MDF_TTL;
5600 			break;
5601 		case RTE_FLOW_ACTION_TYPE_JUMP:
5602 			ret = flow_dv_validate_action_jump(actions,
5603 							   action_flags,
5604 							   attr, external,
5605 							   error);
5606 			if (ret)
5607 				return ret;
5608 			++actions_n;
5609 			action_flags |= MLX5_FLOW_ACTION_JUMP;
5610 			break;
5611 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
5612 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
5613 			ret = flow_dv_validate_action_modify_tcp_seq
5614 								(action_flags,
5615 								 actions,
5616 								 item_flags,
5617 								 error);
5618 			if (ret < 0)
5619 				return ret;
5620 			/* Count all modify-header actions as one action. */
5621 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5622 				++actions_n;
5623 			action_flags |= actions->type ==
5624 					RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
5625 						MLX5_FLOW_ACTION_INC_TCP_SEQ :
5626 						MLX5_FLOW_ACTION_DEC_TCP_SEQ;
5627 			rw_act_num += MLX5_ACT_NUM_MDF_TCPSEQ;
5628 			break;
5629 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
5630 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
5631 			ret = flow_dv_validate_action_modify_tcp_ack
5632 								(action_flags,
5633 								 actions,
5634 								 item_flags,
5635 								 error);
5636 			if (ret < 0)
5637 				return ret;
5638 			/* Count all modify-header actions as one action. */
5639 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5640 				++actions_n;
5641 			action_flags |= actions->type ==
5642 					RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
5643 						MLX5_FLOW_ACTION_INC_TCP_ACK :
5644 						MLX5_FLOW_ACTION_DEC_TCP_ACK;
5645 			rw_act_num += MLX5_ACT_NUM_MDF_TCPACK;
5646 			break;
5647 		case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
5648 			break;
5649 		case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
5650 		case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
5651 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5652 			break;
5653 		case RTE_FLOW_ACTION_TYPE_METER:
5654 			ret = mlx5_flow_validate_action_meter(dev,
5655 							      action_flags,
5656 							      actions, attr,
5657 							      error);
5658 			if (ret < 0)
5659 				return ret;
5660 			action_flags |= MLX5_FLOW_ACTION_METER;
5661 			++actions_n;
5662 			/* Meter action will add one more TAG action. */
5663 			rw_act_num += MLX5_ACT_NUM_SET_TAG;
5664 			break;
5665 		case RTE_FLOW_ACTION_TYPE_AGE:
5666 			ret = flow_dv_validate_action_age(action_flags,
5667 							  actions, dev,
5668 							  error);
5669 			if (ret < 0)
5670 				return ret;
5671 			action_flags |= MLX5_FLOW_ACTION_AGE;
5672 			++actions_n;
5673 			break;
5674 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
5675 			ret = flow_dv_validate_action_modify_ipv4_dscp
5676 							 (action_flags,
5677 							  actions,
5678 							  item_flags,
5679 							  error);
5680 			if (ret < 0)
5681 				return ret;
5682 			/* Count all modify-header actions as one action. */
5683 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5684 				++actions_n;
5685 			action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
5686 			rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5687 			break;
5688 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
5689 			ret = flow_dv_validate_action_modify_ipv6_dscp
5690 								(action_flags,
5691 								 actions,
5692 								 item_flags,
5693 								 error);
5694 			if (ret < 0)
5695 				return ret;
5696 			/* Count all modify-header actions as one action. */
5697 			if (!(action_flags & MLX5_FLOW_MODIFY_HDR_ACTIONS))
5698 				++actions_n;
5699 			action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
5700 			rw_act_num += MLX5_ACT_NUM_SET_DSCP;
5701 			break;
5702 		default:
5703 			return rte_flow_error_set(error, ENOTSUP,
5704 						  RTE_FLOW_ERROR_TYPE_ACTION,
5705 						  actions,
5706 						  "action not supported");
5707 		}
5708 	}
5709 	/*
5710 	 * Validate the drop action mutual exclusion with other actions.
5711 	 * Drop action is mutually-exclusive with any other action, except for
5712 	 * Count action.
5713 	 */
5714 	if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
5715 	    (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
5716 		return rte_flow_error_set(error, EINVAL,
5717 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
5718 					  "Drop action is mutually-exclusive "
5719 					  "with any other action, except for "
5720 					  "Count action");
5721 	/* Eswitch has few restrictions on using items and actions */
5722 	if (attr->transfer) {
5723 		if (!mlx5_flow_ext_mreg_supported(dev) &&
5724 		    action_flags & MLX5_FLOW_ACTION_FLAG)
5725 			return rte_flow_error_set(error, ENOTSUP,
5726 						  RTE_FLOW_ERROR_TYPE_ACTION,
5727 						  NULL,
5728 						  "unsupported action FLAG");
5729 		if (!mlx5_flow_ext_mreg_supported(dev) &&
5730 		    action_flags & MLX5_FLOW_ACTION_MARK)
5731 			return rte_flow_error_set(error, ENOTSUP,
5732 						  RTE_FLOW_ERROR_TYPE_ACTION,
5733 						  NULL,
5734 						  "unsupported action MARK");
5735 		if (action_flags & MLX5_FLOW_ACTION_QUEUE)
5736 			return rte_flow_error_set(error, ENOTSUP,
5737 						  RTE_FLOW_ERROR_TYPE_ACTION,
5738 						  NULL,
5739 						  "unsupported action QUEUE");
5740 		if (action_flags & MLX5_FLOW_ACTION_RSS)
5741 			return rte_flow_error_set(error, ENOTSUP,
5742 						  RTE_FLOW_ERROR_TYPE_ACTION,
5743 						  NULL,
5744 						  "unsupported action RSS");
5745 		if (!(action_flags & MLX5_FLOW_FATE_ESWITCH_ACTIONS))
5746 			return rte_flow_error_set(error, EINVAL,
5747 						  RTE_FLOW_ERROR_TYPE_ACTION,
5748 						  actions,
5749 						  "no fate action is found");
5750 	} else {
5751 		if (!(action_flags & MLX5_FLOW_FATE_ACTIONS) && attr->ingress)
5752 			return rte_flow_error_set(error, EINVAL,
5753 						  RTE_FLOW_ERROR_TYPE_ACTION,
5754 						  actions,
5755 						  "no fate action is found");
5756 	}
5757 	/* Continue validation for Xcap and VLAN actions.*/
5758 	if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
5759 			     MLX5_FLOW_VLAN_ACTIONS)) &&
5760 	    (queue_index == 0xFFFF ||
5761 	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
5762 		if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
5763 		    MLX5_FLOW_XCAP_ACTIONS)
5764 			return rte_flow_error_set(error, ENOTSUP,
5765 						  RTE_FLOW_ERROR_TYPE_ACTION,
5766 						  NULL, "encap and decap "
5767 						  "combination aren't supported");
5768 		if (!attr->transfer && attr->ingress) {
5769 			if (action_flags & MLX5_FLOW_ACTION_ENCAP)
5770 				return rte_flow_error_set
5771 						(error, ENOTSUP,
5772 						 RTE_FLOW_ERROR_TYPE_ACTION,
5773 						 NULL, "encap is not supported"
5774 						 " for ingress traffic");
5775 			else if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
5776 				return rte_flow_error_set
5777 						(error, ENOTSUP,
5778 						 RTE_FLOW_ERROR_TYPE_ACTION,
5779 						 NULL, "push VLAN action not "
5780 						 "supported for ingress");
5781 			else if ((action_flags & MLX5_FLOW_VLAN_ACTIONS) ==
5782 					MLX5_FLOW_VLAN_ACTIONS)
5783 				return rte_flow_error_set
5784 						(error, ENOTSUP,
5785 						 RTE_FLOW_ERROR_TYPE_ACTION,
5786 						 NULL, "no support for "
5787 						 "multiple VLAN actions");
5788 		}
5789 	}
5790 	/* Hairpin flow will add one more TAG action. */
5791 	if (hairpin > 0)
5792 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
5793 	/* extra metadata enabled: one more TAG action will be add. */
5794 	if (dev_conf->dv_flow_en &&
5795 	    dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
5796 	    mlx5_flow_ext_mreg_supported(dev))
5797 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
5798 	if ((uint32_t)rw_act_num >
5799 			flow_dv_modify_hdr_action_max(dev, is_root)) {
5800 		return rte_flow_error_set(error, ENOTSUP,
5801 					  RTE_FLOW_ERROR_TYPE_ACTION,
5802 					  NULL, "too many header modify"
5803 					  " actions to support");
5804 	}
5805 	return 0;
5806 }
5807 
5808 /**
5809  * Internal preparation function. Allocates the DV flow size,
5810  * this size is constant.
5811  *
5812  * @param[in] dev
5813  *   Pointer to the rte_eth_dev structure.
5814  * @param[in] attr
5815  *   Pointer to the flow attributes.
5816  * @param[in] items
5817  *   Pointer to the list of items.
5818  * @param[in] actions
5819  *   Pointer to the list of actions.
5820  * @param[out] error
5821  *   Pointer to the error structure.
5822  *
5823  * @return
5824  *   Pointer to mlx5_flow object on success,
5825  *   otherwise NULL and rte_errno is set.
5826  */
5827 static struct mlx5_flow *
5828 flow_dv_prepare(struct rte_eth_dev *dev,
5829 		const struct rte_flow_attr *attr __rte_unused,
5830 		const struct rte_flow_item items[] __rte_unused,
5831 		const struct rte_flow_action actions[] __rte_unused,
5832 		struct rte_flow_error *error)
5833 {
5834 	uint32_t handle_idx = 0;
5835 	struct mlx5_flow *dev_flow;
5836 	struct mlx5_flow_handle *dev_handle;
5837 	struct mlx5_priv *priv = dev->data->dev_private;
5838 
5839 	/* In case of corrupting the memory. */
5840 	if (priv->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
5841 		rte_flow_error_set(error, ENOSPC,
5842 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5843 				   "not free temporary device flow");
5844 		return NULL;
5845 	}
5846 	dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
5847 				   &handle_idx);
5848 	if (!dev_handle) {
5849 		rte_flow_error_set(error, ENOMEM,
5850 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5851 				   "not enough memory to create flow handle");
5852 		return NULL;
5853 	}
5854 	/* No multi-thread supporting. */
5855 	dev_flow = &((struct mlx5_flow *)priv->inter_flows)[priv->flow_idx++];
5856 	dev_flow->handle = dev_handle;
5857 	dev_flow->handle_idx = handle_idx;
5858 	/*
5859 	 * In some old rdma-core releases, before continuing, a check of the
5860 	 * length of matching parameter will be done at first. It needs to use
5861 	 * the length without misc4 param. If the flow has misc4 support, then
5862 	 * the length needs to be adjusted accordingly. Each param member is
5863 	 * aligned with a 64B boundary naturally.
5864 	 */
5865 	dev_flow->dv.value.size = MLX5_ST_SZ_BYTES(fte_match_param) -
5866 				  MLX5_ST_SZ_BYTES(fte_match_set_misc4);
5867 	/*
5868 	 * The matching value needs to be cleared to 0 before using. In the
5869 	 * past, it will be automatically cleared when using rte_*alloc
5870 	 * API. The time consumption will be almost the same as before.
5871 	 */
5872 	memset(dev_flow->dv.value.buf, 0, MLX5_ST_SZ_BYTES(fte_match_param));
5873 	dev_flow->ingress = attr->ingress;
5874 	dev_flow->dv.transfer = attr->transfer;
5875 	return dev_flow;
5876 }
5877 
5878 #ifdef RTE_LIBRTE_MLX5_DEBUG
5879 /**
5880  * Sanity check for match mask and value. Similar to check_valid_spec() in
5881  * kernel driver. If unmasked bit is present in value, it returns failure.
5882  *
5883  * @param match_mask
5884  *   pointer to match mask buffer.
5885  * @param match_value
5886  *   pointer to match value buffer.
5887  *
5888  * @return
5889  *   0 if valid, -EINVAL otherwise.
5890  */
5891 static int
5892 flow_dv_check_valid_spec(void *match_mask, void *match_value)
5893 {
5894 	uint8_t *m = match_mask;
5895 	uint8_t *v = match_value;
5896 	unsigned int i;
5897 
5898 	for (i = 0; i < MLX5_ST_SZ_BYTES(fte_match_param); ++i) {
5899 		if (v[i] & ~m[i]) {
5900 			DRV_LOG(ERR,
5901 				"match_value differs from match_criteria"
5902 				" %p[%u] != %p[%u]",
5903 				match_value, i, match_mask, i);
5904 			return -EINVAL;
5905 		}
5906 	}
5907 	return 0;
5908 }
5909 #endif
5910 
5911 /**
5912  * Add match of ip_version.
5913  *
5914  * @param[in] group
5915  *   Flow group.
5916  * @param[in] headers_v
5917  *   Values header pointer.
5918  * @param[in] headers_m
5919  *   Masks header pointer.
5920  * @param[in] ip_version
5921  *   The IP version to set.
5922  */
5923 static inline void
5924 flow_dv_set_match_ip_version(uint32_t group,
5925 			     void *headers_v,
5926 			     void *headers_m,
5927 			     uint8_t ip_version)
5928 {
5929 	if (group == 0)
5930 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 0xf);
5931 	else
5932 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version,
5933 			 ip_version);
5934 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_version, ip_version);
5935 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, 0);
5936 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype, 0);
5937 }
5938 
5939 /**
5940  * Add Ethernet item to matcher and to the value.
5941  *
5942  * @param[in, out] matcher
5943  *   Flow matcher.
5944  * @param[in, out] key
5945  *   Flow matcher value.
5946  * @param[in] item
5947  *   Flow pattern to translate.
5948  * @param[in] inner
5949  *   Item is inner pattern.
5950  */
5951 static void
5952 flow_dv_translate_item_eth(void *matcher, void *key,
5953 			   const struct rte_flow_item *item, int inner,
5954 			   uint32_t group)
5955 {
5956 	const struct rte_flow_item_eth *eth_m = item->mask;
5957 	const struct rte_flow_item_eth *eth_v = item->spec;
5958 	const struct rte_flow_item_eth nic_mask = {
5959 		.dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5960 		.src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
5961 		.type = RTE_BE16(0xffff),
5962 	};
5963 	void *headers_m;
5964 	void *headers_v;
5965 	char *l24_v;
5966 	unsigned int i;
5967 
5968 	if (!eth_v)
5969 		return;
5970 	if (!eth_m)
5971 		eth_m = &nic_mask;
5972 	if (inner) {
5973 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5974 					 inner_headers);
5975 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
5976 	} else {
5977 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
5978 					 outer_headers);
5979 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
5980 	}
5981 	memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, dmac_47_16),
5982 	       &eth_m->dst, sizeof(eth_m->dst));
5983 	/* The value must be in the range of the mask. */
5984 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, dmac_47_16);
5985 	for (i = 0; i < sizeof(eth_m->dst); ++i)
5986 		l24_v[i] = eth_m->dst.addr_bytes[i] & eth_v->dst.addr_bytes[i];
5987 	memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m, smac_47_16),
5988 	       &eth_m->src, sizeof(eth_m->src));
5989 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v, smac_47_16);
5990 	/* The value must be in the range of the mask. */
5991 	for (i = 0; i < sizeof(eth_m->dst); ++i)
5992 		l24_v[i] = eth_m->src.addr_bytes[i] & eth_v->src.addr_bytes[i];
5993 	if (eth_v->type) {
5994 		/* When ethertype is present set mask for tagged VLAN. */
5995 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
5996 		/* Set value for tagged VLAN if ethertype is 802.1Q. */
5997 		if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
5998 		    eth_v->type == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
5999 			MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag,
6000 				 1);
6001 			/* Return here to avoid setting match on ethertype. */
6002 			return;
6003 		}
6004 	}
6005 	/*
6006 	 * HW supports match on one Ethertype, the Ethertype following the last
6007 	 * VLAN tag of the packet (see PRM).
6008 	 * Set match on ethertype only if ETH header is not followed by VLAN.
6009 	 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6010 	 * ethertype, and use ip_version field instead.
6011 	 * eCPRI over Ether layer will use type value 0xAEFE.
6012 	 */
6013 	if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6014 	    eth_m->type == 0xFFFF) {
6015 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6016 	} else if (eth_v->type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6017 		   eth_m->type == 0xFFFF) {
6018 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6019 	} else {
6020 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6021 			 rte_be_to_cpu_16(eth_m->type));
6022 		l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6023 				     ethertype);
6024 		*(uint16_t *)(l24_v) = eth_m->type & eth_v->type;
6025 	}
6026 }
6027 
6028 /**
6029  * Add VLAN item to matcher and to the value.
6030  *
6031  * @param[in, out] dev_flow
6032  *   Flow descriptor.
6033  * @param[in, out] matcher
6034  *   Flow matcher.
6035  * @param[in, out] key
6036  *   Flow matcher value.
6037  * @param[in] item
6038  *   Flow pattern to translate.
6039  * @param[in] inner
6040  *   Item is inner pattern.
6041  */
6042 static void
6043 flow_dv_translate_item_vlan(struct mlx5_flow *dev_flow,
6044 			    void *matcher, void *key,
6045 			    const struct rte_flow_item *item,
6046 			    int inner, uint32_t group)
6047 {
6048 	const struct rte_flow_item_vlan *vlan_m = item->mask;
6049 	const struct rte_flow_item_vlan *vlan_v = item->spec;
6050 	void *headers_m;
6051 	void *headers_v;
6052 	uint16_t tci_m;
6053 	uint16_t tci_v;
6054 
6055 	if (inner) {
6056 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6057 					 inner_headers);
6058 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6059 	} else {
6060 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6061 					 outer_headers);
6062 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6063 		/*
6064 		 * This is workaround, masks are not supported,
6065 		 * and pre-validated.
6066 		 */
6067 		if (vlan_v)
6068 			dev_flow->handle->vf_vlan.tag =
6069 					rte_be_to_cpu_16(vlan_v->tci) & 0x0fff;
6070 	}
6071 	/*
6072 	 * When VLAN item exists in flow, mark packet as tagged,
6073 	 * even if TCI is not specified.
6074 	 */
6075 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6076 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
6077 	if (!vlan_v)
6078 		return;
6079 	if (!vlan_m)
6080 		vlan_m = &rte_flow_item_vlan_mask;
6081 	tci_m = rte_be_to_cpu_16(vlan_m->tci);
6082 	tci_v = rte_be_to_cpu_16(vlan_m->tci & vlan_v->tci);
6083 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_vid, tci_m);
6084 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, tci_v);
6085 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_cfi, tci_m >> 12);
6086 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_cfi, tci_v >> 12);
6087 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, first_prio, tci_m >> 13);
6088 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, tci_v >> 13);
6089 	/*
6090 	 * HW is optimized for IPv4/IPv6. In such cases, avoid setting
6091 	 * ethertype, and use ip_version field instead.
6092 	 */
6093 	if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV4) &&
6094 	    vlan_m->inner_type == 0xFFFF) {
6095 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6096 	} else if (vlan_v->inner_type == RTE_BE16(RTE_ETHER_TYPE_IPV6) &&
6097 		   vlan_m->inner_type == 0xFFFF) {
6098 		flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6099 	} else {
6100 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ethertype,
6101 			 rte_be_to_cpu_16(vlan_m->inner_type));
6102 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
6103 			 rte_be_to_cpu_16(vlan_m->inner_type &
6104 					  vlan_v->inner_type));
6105 	}
6106 }
6107 
6108 /**
6109  * Add IPV4 item to matcher and to the value.
6110  *
6111  * @param[in, out] matcher
6112  *   Flow matcher.
6113  * @param[in, out] key
6114  *   Flow matcher value.
6115  * @param[in] item
6116  *   Flow pattern to translate.
6117  * @param[in] item_flags
6118  *   Bit-fields that holds the items detected until now.
6119  * @param[in] inner
6120  *   Item is inner pattern.
6121  * @param[in] group
6122  *   The group to insert the rule.
6123  */
6124 static void
6125 flow_dv_translate_item_ipv4(void *matcher, void *key,
6126 			    const struct rte_flow_item *item,
6127 			    const uint64_t item_flags,
6128 			    int inner, uint32_t group)
6129 {
6130 	const struct rte_flow_item_ipv4 *ipv4_m = item->mask;
6131 	const struct rte_flow_item_ipv4 *ipv4_v = item->spec;
6132 	const struct rte_flow_item_ipv4 nic_mask = {
6133 		.hdr = {
6134 			.src_addr = RTE_BE32(0xffffffff),
6135 			.dst_addr = RTE_BE32(0xffffffff),
6136 			.type_of_service = 0xff,
6137 			.next_proto_id = 0xff,
6138 			.time_to_live = 0xff,
6139 		},
6140 	};
6141 	void *headers_m;
6142 	void *headers_v;
6143 	char *l24_m;
6144 	char *l24_v;
6145 	uint8_t tos;
6146 
6147 	if (inner) {
6148 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6149 					 inner_headers);
6150 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6151 	} else {
6152 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6153 					 outer_headers);
6154 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6155 	}
6156 	flow_dv_set_match_ip_version(group, headers_v, headers_m, 4);
6157 	/*
6158 	 * On outer header (which must contains L2), or inner header with L2,
6159 	 * set cvlan_tag mask bit to mark this packet as untagged.
6160 	 * This should be done even if item->spec is empty.
6161 	 */
6162 	if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6163 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6164 	if (!ipv4_v)
6165 		return;
6166 	if (!ipv4_m)
6167 		ipv4_m = &nic_mask;
6168 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6169 			     dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6170 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6171 			     dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
6172 	*(uint32_t *)l24_m = ipv4_m->hdr.dst_addr;
6173 	*(uint32_t *)l24_v = ipv4_m->hdr.dst_addr & ipv4_v->hdr.dst_addr;
6174 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6175 			  src_ipv4_src_ipv6.ipv4_layout.ipv4);
6176 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6177 			  src_ipv4_src_ipv6.ipv4_layout.ipv4);
6178 	*(uint32_t *)l24_m = ipv4_m->hdr.src_addr;
6179 	*(uint32_t *)l24_v = ipv4_m->hdr.src_addr & ipv4_v->hdr.src_addr;
6180 	tos = ipv4_m->hdr.type_of_service & ipv4_v->hdr.type_of_service;
6181 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn,
6182 		 ipv4_m->hdr.type_of_service);
6183 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, tos);
6184 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp,
6185 		 ipv4_m->hdr.type_of_service >> 2);
6186 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, tos >> 2);
6187 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6188 		 ipv4_m->hdr.next_proto_id);
6189 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6190 		 ipv4_v->hdr.next_proto_id & ipv4_m->hdr.next_proto_id);
6191 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6192 		 ipv4_m->hdr.time_to_live);
6193 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6194 		 ipv4_v->hdr.time_to_live & ipv4_m->hdr.time_to_live);
6195 }
6196 
6197 /**
6198  * Add IPV6 item to matcher and to the value.
6199  *
6200  * @param[in, out] matcher
6201  *   Flow matcher.
6202  * @param[in, out] key
6203  *   Flow matcher value.
6204  * @param[in] item
6205  *   Flow pattern to translate.
6206  * @param[in] item_flags
6207  *   Bit-fields that holds the items detected until now.
6208  * @param[in] inner
6209  *   Item is inner pattern.
6210  * @param[in] group
6211  *   The group to insert the rule.
6212  */
6213 static void
6214 flow_dv_translate_item_ipv6(void *matcher, void *key,
6215 			    const struct rte_flow_item *item,
6216 			    const uint64_t item_flags,
6217 			    int inner, uint32_t group)
6218 {
6219 	const struct rte_flow_item_ipv6 *ipv6_m = item->mask;
6220 	const struct rte_flow_item_ipv6 *ipv6_v = item->spec;
6221 	const struct rte_flow_item_ipv6 nic_mask = {
6222 		.hdr = {
6223 			.src_addr =
6224 				"\xff\xff\xff\xff\xff\xff\xff\xff"
6225 				"\xff\xff\xff\xff\xff\xff\xff\xff",
6226 			.dst_addr =
6227 				"\xff\xff\xff\xff\xff\xff\xff\xff"
6228 				"\xff\xff\xff\xff\xff\xff\xff\xff",
6229 			.vtc_flow = RTE_BE32(0xffffffff),
6230 			.proto = 0xff,
6231 			.hop_limits = 0xff,
6232 		},
6233 	};
6234 	void *headers_m;
6235 	void *headers_v;
6236 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6237 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6238 	char *l24_m;
6239 	char *l24_v;
6240 	uint32_t vtc_m;
6241 	uint32_t vtc_v;
6242 	int i;
6243 	int size;
6244 
6245 	if (inner) {
6246 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6247 					 inner_headers);
6248 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6249 	} else {
6250 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6251 					 outer_headers);
6252 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6253 	}
6254 	flow_dv_set_match_ip_version(group, headers_v, headers_m, 6);
6255 	/*
6256 	 * On outer header (which must contains L2), or inner header with L2,
6257 	 * set cvlan_tag mask bit to mark this packet as untagged.
6258 	 * This should be done even if item->spec is empty.
6259 	 */
6260 	if (!inner || item_flags & MLX5_FLOW_LAYER_INNER_L2)
6261 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, cvlan_tag, 1);
6262 	if (!ipv6_v)
6263 		return;
6264 	if (!ipv6_m)
6265 		ipv6_m = &nic_mask;
6266 	size = sizeof(ipv6_m->hdr.dst_addr);
6267 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6268 			     dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6269 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6270 			     dst_ipv4_dst_ipv6.ipv6_layout.ipv6);
6271 	memcpy(l24_m, ipv6_m->hdr.dst_addr, size);
6272 	for (i = 0; i < size; ++i)
6273 		l24_v[i] = l24_m[i] & ipv6_v->hdr.dst_addr[i];
6274 	l24_m = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_m,
6275 			     src_ipv4_src_ipv6.ipv6_layout.ipv6);
6276 	l24_v = MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
6277 			     src_ipv4_src_ipv6.ipv6_layout.ipv6);
6278 	memcpy(l24_m, ipv6_m->hdr.src_addr, size);
6279 	for (i = 0; i < size; ++i)
6280 		l24_v[i] = l24_m[i] & ipv6_v->hdr.src_addr[i];
6281 	/* TOS. */
6282 	vtc_m = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow);
6283 	vtc_v = rte_be_to_cpu_32(ipv6_m->hdr.vtc_flow & ipv6_v->hdr.vtc_flow);
6284 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ecn, vtc_m >> 20);
6285 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ecn, vtc_v >> 20);
6286 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_dscp, vtc_m >> 22);
6287 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_dscp, vtc_v >> 22);
6288 	/* Label. */
6289 	if (inner) {
6290 		MLX5_SET(fte_match_set_misc, misc_m, inner_ipv6_flow_label,
6291 			 vtc_m);
6292 		MLX5_SET(fte_match_set_misc, misc_v, inner_ipv6_flow_label,
6293 			 vtc_v);
6294 	} else {
6295 		MLX5_SET(fte_match_set_misc, misc_m, outer_ipv6_flow_label,
6296 			 vtc_m);
6297 		MLX5_SET(fte_match_set_misc, misc_v, outer_ipv6_flow_label,
6298 			 vtc_v);
6299 	}
6300 	/* Protocol. */
6301 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol,
6302 		 ipv6_m->hdr.proto);
6303 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6304 		 ipv6_v->hdr.proto & ipv6_m->hdr.proto);
6305 	/* Hop limit. */
6306 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_ttl_hoplimit,
6307 		 ipv6_m->hdr.hop_limits);
6308 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_ttl_hoplimit,
6309 		 ipv6_v->hdr.hop_limits & ipv6_m->hdr.hop_limits);
6310 }
6311 
6312 /**
6313  * Add TCP item to matcher and to the value.
6314  *
6315  * @param[in, out] matcher
6316  *   Flow matcher.
6317  * @param[in, out] key
6318  *   Flow matcher value.
6319  * @param[in] item
6320  *   Flow pattern to translate.
6321  * @param[in] inner
6322  *   Item is inner pattern.
6323  */
6324 static void
6325 flow_dv_translate_item_tcp(void *matcher, void *key,
6326 			   const struct rte_flow_item *item,
6327 			   int inner)
6328 {
6329 	const struct rte_flow_item_tcp *tcp_m = item->mask;
6330 	const struct rte_flow_item_tcp *tcp_v = item->spec;
6331 	void *headers_m;
6332 	void *headers_v;
6333 
6334 	if (inner) {
6335 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6336 					 inner_headers);
6337 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6338 	} else {
6339 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6340 					 outer_headers);
6341 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6342 	}
6343 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6344 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_TCP);
6345 	if (!tcp_v)
6346 		return;
6347 	if (!tcp_m)
6348 		tcp_m = &rte_flow_item_tcp_mask;
6349 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_sport,
6350 		 rte_be_to_cpu_16(tcp_m->hdr.src_port));
6351 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
6352 		 rte_be_to_cpu_16(tcp_v->hdr.src_port & tcp_m->hdr.src_port));
6353 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_dport,
6354 		 rte_be_to_cpu_16(tcp_m->hdr.dst_port));
6355 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
6356 		 rte_be_to_cpu_16(tcp_v->hdr.dst_port & tcp_m->hdr.dst_port));
6357 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, tcp_flags,
6358 		 tcp_m->hdr.tcp_flags);
6359 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_flags,
6360 		 (tcp_v->hdr.tcp_flags & tcp_m->hdr.tcp_flags));
6361 }
6362 
6363 /**
6364  * Add UDP item to matcher and to the value.
6365  *
6366  * @param[in, out] matcher
6367  *   Flow matcher.
6368  * @param[in, out] key
6369  *   Flow matcher value.
6370  * @param[in] item
6371  *   Flow pattern to translate.
6372  * @param[in] inner
6373  *   Item is inner pattern.
6374  */
6375 static void
6376 flow_dv_translate_item_udp(void *matcher, void *key,
6377 			   const struct rte_flow_item *item,
6378 			   int inner)
6379 {
6380 	const struct rte_flow_item_udp *udp_m = item->mask;
6381 	const struct rte_flow_item_udp *udp_v = item->spec;
6382 	void *headers_m;
6383 	void *headers_v;
6384 
6385 	if (inner) {
6386 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6387 					 inner_headers);
6388 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6389 	} else {
6390 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6391 					 outer_headers);
6392 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6393 	}
6394 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6395 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
6396 	if (!udp_v)
6397 		return;
6398 	if (!udp_m)
6399 		udp_m = &rte_flow_item_udp_mask;
6400 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_sport,
6401 		 rte_be_to_cpu_16(udp_m->hdr.src_port));
6402 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
6403 		 rte_be_to_cpu_16(udp_v->hdr.src_port & udp_m->hdr.src_port));
6404 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport,
6405 		 rte_be_to_cpu_16(udp_m->hdr.dst_port));
6406 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6407 		 rte_be_to_cpu_16(udp_v->hdr.dst_port & udp_m->hdr.dst_port));
6408 }
6409 
6410 /**
6411  * Add GRE optional Key item to matcher and to the value.
6412  *
6413  * @param[in, out] matcher
6414  *   Flow matcher.
6415  * @param[in, out] key
6416  *   Flow matcher value.
6417  * @param[in] item
6418  *   Flow pattern to translate.
6419  * @param[in] inner
6420  *   Item is inner pattern.
6421  */
6422 static void
6423 flow_dv_translate_item_gre_key(void *matcher, void *key,
6424 				   const struct rte_flow_item *item)
6425 {
6426 	const rte_be32_t *key_m = item->mask;
6427 	const rte_be32_t *key_v = item->spec;
6428 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6429 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6430 	rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
6431 
6432 	/* GRE K bit must be on and should already be validated */
6433 	MLX5_SET(fte_match_set_misc, misc_m, gre_k_present, 1);
6434 	MLX5_SET(fte_match_set_misc, misc_v, gre_k_present, 1);
6435 	if (!key_v)
6436 		return;
6437 	if (!key_m)
6438 		key_m = &gre_key_default_mask;
6439 	MLX5_SET(fte_match_set_misc, misc_m, gre_key_h,
6440 		 rte_be_to_cpu_32(*key_m) >> 8);
6441 	MLX5_SET(fte_match_set_misc, misc_v, gre_key_h,
6442 		 rte_be_to_cpu_32((*key_v) & (*key_m)) >> 8);
6443 	MLX5_SET(fte_match_set_misc, misc_m, gre_key_l,
6444 		 rte_be_to_cpu_32(*key_m) & 0xFF);
6445 	MLX5_SET(fte_match_set_misc, misc_v, gre_key_l,
6446 		 rte_be_to_cpu_32((*key_v) & (*key_m)) & 0xFF);
6447 }
6448 
6449 /**
6450  * Add GRE item to matcher and to the value.
6451  *
6452  * @param[in, out] matcher
6453  *   Flow matcher.
6454  * @param[in, out] key
6455  *   Flow matcher value.
6456  * @param[in] item
6457  *   Flow pattern to translate.
6458  * @param[in] inner
6459  *   Item is inner pattern.
6460  */
6461 static void
6462 flow_dv_translate_item_gre(void *matcher, void *key,
6463 			   const struct rte_flow_item *item,
6464 			   int inner)
6465 {
6466 	const struct rte_flow_item_gre *gre_m = item->mask;
6467 	const struct rte_flow_item_gre *gre_v = item->spec;
6468 	void *headers_m;
6469 	void *headers_v;
6470 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6471 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6472 	struct {
6473 		union {
6474 			__extension__
6475 			struct {
6476 				uint16_t version:3;
6477 				uint16_t rsvd0:9;
6478 				uint16_t s_present:1;
6479 				uint16_t k_present:1;
6480 				uint16_t rsvd_bit1:1;
6481 				uint16_t c_present:1;
6482 			};
6483 			uint16_t value;
6484 		};
6485 	} gre_crks_rsvd0_ver_m, gre_crks_rsvd0_ver_v;
6486 
6487 	if (inner) {
6488 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6489 					 inner_headers);
6490 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6491 	} else {
6492 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6493 					 outer_headers);
6494 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6495 	}
6496 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6497 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_GRE);
6498 	if (!gre_v)
6499 		return;
6500 	if (!gre_m)
6501 		gre_m = &rte_flow_item_gre_mask;
6502 	MLX5_SET(fte_match_set_misc, misc_m, gre_protocol,
6503 		 rte_be_to_cpu_16(gre_m->protocol));
6504 	MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6505 		 rte_be_to_cpu_16(gre_v->protocol & gre_m->protocol));
6506 	gre_crks_rsvd0_ver_m.value = rte_be_to_cpu_16(gre_m->c_rsvd0_ver);
6507 	gre_crks_rsvd0_ver_v.value = rte_be_to_cpu_16(gre_v->c_rsvd0_ver);
6508 	MLX5_SET(fte_match_set_misc, misc_m, gre_c_present,
6509 		 gre_crks_rsvd0_ver_m.c_present);
6510 	MLX5_SET(fte_match_set_misc, misc_v, gre_c_present,
6511 		 gre_crks_rsvd0_ver_v.c_present &
6512 		 gre_crks_rsvd0_ver_m.c_present);
6513 	MLX5_SET(fte_match_set_misc, misc_m, gre_k_present,
6514 		 gre_crks_rsvd0_ver_m.k_present);
6515 	MLX5_SET(fte_match_set_misc, misc_v, gre_k_present,
6516 		 gre_crks_rsvd0_ver_v.k_present &
6517 		 gre_crks_rsvd0_ver_m.k_present);
6518 	MLX5_SET(fte_match_set_misc, misc_m, gre_s_present,
6519 		 gre_crks_rsvd0_ver_m.s_present);
6520 	MLX5_SET(fte_match_set_misc, misc_v, gre_s_present,
6521 		 gre_crks_rsvd0_ver_v.s_present &
6522 		 gre_crks_rsvd0_ver_m.s_present);
6523 }
6524 
6525 /**
6526  * Add NVGRE item to matcher and to the value.
6527  *
6528  * @param[in, out] matcher
6529  *   Flow matcher.
6530  * @param[in, out] key
6531  *   Flow matcher value.
6532  * @param[in] item
6533  *   Flow pattern to translate.
6534  * @param[in] inner
6535  *   Item is inner pattern.
6536  */
6537 static void
6538 flow_dv_translate_item_nvgre(void *matcher, void *key,
6539 			     const struct rte_flow_item *item,
6540 			     int inner)
6541 {
6542 	const struct rte_flow_item_nvgre *nvgre_m = item->mask;
6543 	const struct rte_flow_item_nvgre *nvgre_v = item->spec;
6544 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6545 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6546 	const char *tni_flow_id_m;
6547 	const char *tni_flow_id_v;
6548 	char *gre_key_m;
6549 	char *gre_key_v;
6550 	int size;
6551 	int i;
6552 
6553 	/* For NVGRE, GRE header fields must be set with defined values. */
6554 	const struct rte_flow_item_gre gre_spec = {
6555 		.c_rsvd0_ver = RTE_BE16(0x2000),
6556 		.protocol = RTE_BE16(RTE_ETHER_TYPE_TEB)
6557 	};
6558 	const struct rte_flow_item_gre gre_mask = {
6559 		.c_rsvd0_ver = RTE_BE16(0xB000),
6560 		.protocol = RTE_BE16(UINT16_MAX),
6561 	};
6562 	const struct rte_flow_item gre_item = {
6563 		.spec = &gre_spec,
6564 		.mask = &gre_mask,
6565 		.last = NULL,
6566 	};
6567 	flow_dv_translate_item_gre(matcher, key, &gre_item, inner);
6568 	if (!nvgre_v)
6569 		return;
6570 	if (!nvgre_m)
6571 		nvgre_m = &rte_flow_item_nvgre_mask;
6572 	tni_flow_id_m = (const char *)nvgre_m->tni;
6573 	tni_flow_id_v = (const char *)nvgre_v->tni;
6574 	size = sizeof(nvgre_m->tni) + sizeof(nvgre_m->flow_id);
6575 	gre_key_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, gre_key_h);
6576 	gre_key_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, gre_key_h);
6577 	memcpy(gre_key_m, tni_flow_id_m, size);
6578 	for (i = 0; i < size; ++i)
6579 		gre_key_v[i] = gre_key_m[i] & tni_flow_id_v[i];
6580 }
6581 
6582 /**
6583  * Add VXLAN item to matcher and to the value.
6584  *
6585  * @param[in, out] matcher
6586  *   Flow matcher.
6587  * @param[in, out] key
6588  *   Flow matcher value.
6589  * @param[in] item
6590  *   Flow pattern to translate.
6591  * @param[in] inner
6592  *   Item is inner pattern.
6593  */
6594 static void
6595 flow_dv_translate_item_vxlan(void *matcher, void *key,
6596 			     const struct rte_flow_item *item,
6597 			     int inner)
6598 {
6599 	const struct rte_flow_item_vxlan *vxlan_m = item->mask;
6600 	const struct rte_flow_item_vxlan *vxlan_v = item->spec;
6601 	void *headers_m;
6602 	void *headers_v;
6603 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6604 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6605 	char *vni_m;
6606 	char *vni_v;
6607 	uint16_t dport;
6608 	int size;
6609 	int i;
6610 
6611 	if (inner) {
6612 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6613 					 inner_headers);
6614 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6615 	} else {
6616 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6617 					 outer_headers);
6618 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6619 	}
6620 	dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6621 		MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6622 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6623 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6624 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6625 	}
6626 	if (!vxlan_v)
6627 		return;
6628 	if (!vxlan_m)
6629 		vxlan_m = &rte_flow_item_vxlan_mask;
6630 	size = sizeof(vxlan_m->vni);
6631 	vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, vxlan_vni);
6632 	vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, vxlan_vni);
6633 	memcpy(vni_m, vxlan_m->vni, size);
6634 	for (i = 0; i < size; ++i)
6635 		vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6636 }
6637 
6638 /**
6639  * Add VXLAN-GPE item to matcher and to the value.
6640  *
6641  * @param[in, out] matcher
6642  *   Flow matcher.
6643  * @param[in, out] key
6644  *   Flow matcher value.
6645  * @param[in] item
6646  *   Flow pattern to translate.
6647  * @param[in] inner
6648  *   Item is inner pattern.
6649  */
6650 
6651 static void
6652 flow_dv_translate_item_vxlan_gpe(void *matcher, void *key,
6653 				 const struct rte_flow_item *item, int inner)
6654 {
6655 	const struct rte_flow_item_vxlan_gpe *vxlan_m = item->mask;
6656 	const struct rte_flow_item_vxlan_gpe *vxlan_v = item->spec;
6657 	void *headers_m;
6658 	void *headers_v;
6659 	void *misc_m =
6660 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_3);
6661 	void *misc_v =
6662 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
6663 	char *vni_m;
6664 	char *vni_v;
6665 	uint16_t dport;
6666 	int size;
6667 	int i;
6668 	uint8_t flags_m = 0xff;
6669 	uint8_t flags_v = 0xc;
6670 
6671 	if (inner) {
6672 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6673 					 inner_headers);
6674 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6675 	} else {
6676 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6677 					 outer_headers);
6678 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6679 	}
6680 	dport = item->type == RTE_FLOW_ITEM_TYPE_VXLAN ?
6681 		MLX5_UDP_PORT_VXLAN : MLX5_UDP_PORT_VXLAN_GPE;
6682 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6683 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6684 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6685 	}
6686 	if (!vxlan_v)
6687 		return;
6688 	if (!vxlan_m)
6689 		vxlan_m = &rte_flow_item_vxlan_gpe_mask;
6690 	size = sizeof(vxlan_m->vni);
6691 	vni_m = MLX5_ADDR_OF(fte_match_set_misc3, misc_m, outer_vxlan_gpe_vni);
6692 	vni_v = MLX5_ADDR_OF(fte_match_set_misc3, misc_v, outer_vxlan_gpe_vni);
6693 	memcpy(vni_m, vxlan_m->vni, size);
6694 	for (i = 0; i < size; ++i)
6695 		vni_v[i] = vni_m[i] & vxlan_v->vni[i];
6696 	if (vxlan_m->flags) {
6697 		flags_m = vxlan_m->flags;
6698 		flags_v = vxlan_v->flags;
6699 	}
6700 	MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_flags, flags_m);
6701 	MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, flags_v);
6702 	MLX5_SET(fte_match_set_misc3, misc_m, outer_vxlan_gpe_next_protocol,
6703 		 vxlan_m->protocol);
6704 	MLX5_SET(fte_match_set_misc3, misc_v, outer_vxlan_gpe_next_protocol,
6705 		 vxlan_v->protocol);
6706 }
6707 
6708 /**
6709  * Add Geneve item to matcher and to the value.
6710  *
6711  * @param[in, out] matcher
6712  *   Flow matcher.
6713  * @param[in, out] key
6714  *   Flow matcher value.
6715  * @param[in] item
6716  *   Flow pattern to translate.
6717  * @param[in] inner
6718  *   Item is inner pattern.
6719  */
6720 
6721 static void
6722 flow_dv_translate_item_geneve(void *matcher, void *key,
6723 			      const struct rte_flow_item *item, int inner)
6724 {
6725 	const struct rte_flow_item_geneve *geneve_m = item->mask;
6726 	const struct rte_flow_item_geneve *geneve_v = item->spec;
6727 	void *headers_m;
6728 	void *headers_v;
6729 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6730 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6731 	uint16_t dport;
6732 	uint16_t gbhdr_m;
6733 	uint16_t gbhdr_v;
6734 	char *vni_m;
6735 	char *vni_v;
6736 	size_t size, i;
6737 
6738 	if (inner) {
6739 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6740 					 inner_headers);
6741 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
6742 	} else {
6743 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
6744 					 outer_headers);
6745 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6746 	}
6747 	dport = MLX5_UDP_PORT_GENEVE;
6748 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
6749 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
6750 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
6751 	}
6752 	if (!geneve_v)
6753 		return;
6754 	if (!geneve_m)
6755 		geneve_m = &rte_flow_item_geneve_mask;
6756 	size = sizeof(geneve_m->vni);
6757 	vni_m = MLX5_ADDR_OF(fte_match_set_misc, misc_m, geneve_vni);
6758 	vni_v = MLX5_ADDR_OF(fte_match_set_misc, misc_v, geneve_vni);
6759 	memcpy(vni_m, geneve_m->vni, size);
6760 	for (i = 0; i < size; ++i)
6761 		vni_v[i] = vni_m[i] & geneve_v->vni[i];
6762 	MLX5_SET(fte_match_set_misc, misc_m, geneve_protocol_type,
6763 		 rte_be_to_cpu_16(geneve_m->protocol));
6764 	MLX5_SET(fte_match_set_misc, misc_v, geneve_protocol_type,
6765 		 rte_be_to_cpu_16(geneve_v->protocol & geneve_m->protocol));
6766 	gbhdr_m = rte_be_to_cpu_16(geneve_m->ver_opt_len_o_c_rsvd0);
6767 	gbhdr_v = rte_be_to_cpu_16(geneve_v->ver_opt_len_o_c_rsvd0);
6768 	MLX5_SET(fte_match_set_misc, misc_m, geneve_oam,
6769 		 MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6770 	MLX5_SET(fte_match_set_misc, misc_v, geneve_oam,
6771 		 MLX5_GENEVE_OAMF_VAL(gbhdr_v) & MLX5_GENEVE_OAMF_VAL(gbhdr_m));
6772 	MLX5_SET(fte_match_set_misc, misc_m, geneve_opt_len,
6773 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6774 	MLX5_SET(fte_match_set_misc, misc_v, geneve_opt_len,
6775 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_v) &
6776 		 MLX5_GENEVE_OPTLEN_VAL(gbhdr_m));
6777 }
6778 
6779 /**
6780  * Add MPLS item to matcher and to the value.
6781  *
6782  * @param[in, out] matcher
6783  *   Flow matcher.
6784  * @param[in, out] key
6785  *   Flow matcher value.
6786  * @param[in] item
6787  *   Flow pattern to translate.
6788  * @param[in] prev_layer
6789  *   The protocol layer indicated in previous item.
6790  * @param[in] inner
6791  *   Item is inner pattern.
6792  */
6793 static void
6794 flow_dv_translate_item_mpls(void *matcher, void *key,
6795 			    const struct rte_flow_item *item,
6796 			    uint64_t prev_layer,
6797 			    int inner)
6798 {
6799 	const uint32_t *in_mpls_m = item->mask;
6800 	const uint32_t *in_mpls_v = item->spec;
6801 	uint32_t *out_mpls_m = 0;
6802 	uint32_t *out_mpls_v = 0;
6803 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
6804 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
6805 	void *misc2_m = MLX5_ADDR_OF(fte_match_param, matcher,
6806 				     misc_parameters_2);
6807 	void *misc2_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6808 	void *headers_m = MLX5_ADDR_OF(fte_match_param, matcher, outer_headers);
6809 	void *headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
6810 
6811 	switch (prev_layer) {
6812 	case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6813 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
6814 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
6815 			 MLX5_UDP_PORT_MPLS);
6816 		break;
6817 	case MLX5_FLOW_LAYER_GRE:
6818 		MLX5_SET(fte_match_set_misc, misc_m, gre_protocol, 0xffff);
6819 		MLX5_SET(fte_match_set_misc, misc_v, gre_protocol,
6820 			 RTE_ETHER_TYPE_MPLS);
6821 		break;
6822 	default:
6823 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
6824 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
6825 			 IPPROTO_MPLS);
6826 		break;
6827 	}
6828 	if (!in_mpls_v)
6829 		return;
6830 	if (!in_mpls_m)
6831 		in_mpls_m = (const uint32_t *)&rte_flow_item_mpls_mask;
6832 	switch (prev_layer) {
6833 	case MLX5_FLOW_LAYER_OUTER_L4_UDP:
6834 		out_mpls_m =
6835 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6836 						 outer_first_mpls_over_udp);
6837 		out_mpls_v =
6838 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6839 						 outer_first_mpls_over_udp);
6840 		break;
6841 	case MLX5_FLOW_LAYER_GRE:
6842 		out_mpls_m =
6843 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_m,
6844 						 outer_first_mpls_over_gre);
6845 		out_mpls_v =
6846 			(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2, misc2_v,
6847 						 outer_first_mpls_over_gre);
6848 		break;
6849 	default:
6850 		/* Inner MPLS not over GRE is not supported. */
6851 		if (!inner) {
6852 			out_mpls_m =
6853 				(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6854 							 misc2_m,
6855 							 outer_first_mpls);
6856 			out_mpls_v =
6857 				(uint32_t *)MLX5_ADDR_OF(fte_match_set_misc2,
6858 							 misc2_v,
6859 							 outer_first_mpls);
6860 		}
6861 		break;
6862 	}
6863 	if (out_mpls_m && out_mpls_v) {
6864 		*out_mpls_m = *in_mpls_m;
6865 		*out_mpls_v = *in_mpls_v & *in_mpls_m;
6866 	}
6867 }
6868 
6869 /**
6870  * Add metadata register item to matcher
6871  *
6872  * @param[in, out] matcher
6873  *   Flow matcher.
6874  * @param[in, out] key
6875  *   Flow matcher value.
6876  * @param[in] reg_type
6877  *   Type of device metadata register
6878  * @param[in] value
6879  *   Register value
6880  * @param[in] mask
6881  *   Register mask
6882  */
6883 static void
6884 flow_dv_match_meta_reg(void *matcher, void *key,
6885 		       enum modify_reg reg_type,
6886 		       uint32_t data, uint32_t mask)
6887 {
6888 	void *misc2_m =
6889 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters_2);
6890 	void *misc2_v =
6891 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters_2);
6892 	uint32_t temp;
6893 
6894 	data &= mask;
6895 	switch (reg_type) {
6896 	case REG_A:
6897 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_a, mask);
6898 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_a, data);
6899 		break;
6900 	case REG_B:
6901 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_b, mask);
6902 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_b, data);
6903 		break;
6904 	case REG_C_0:
6905 		/*
6906 		 * The metadata register C0 field might be divided into
6907 		 * source vport index and META item value, we should set
6908 		 * this field according to specified mask, not as whole one.
6909 		 */
6910 		temp = MLX5_GET(fte_match_set_misc2, misc2_m, metadata_reg_c_0);
6911 		temp |= mask;
6912 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_0, temp);
6913 		temp = MLX5_GET(fte_match_set_misc2, misc2_v, metadata_reg_c_0);
6914 		temp &= ~mask;
6915 		temp |= data;
6916 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_0, temp);
6917 		break;
6918 	case REG_C_1:
6919 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_1, mask);
6920 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_1, data);
6921 		break;
6922 	case REG_C_2:
6923 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_2, mask);
6924 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_2, data);
6925 		break;
6926 	case REG_C_3:
6927 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_3, mask);
6928 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_3, data);
6929 		break;
6930 	case REG_C_4:
6931 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_4, mask);
6932 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_4, data);
6933 		break;
6934 	case REG_C_5:
6935 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_5, mask);
6936 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_5, data);
6937 		break;
6938 	case REG_C_6:
6939 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_6, mask);
6940 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_6, data);
6941 		break;
6942 	case REG_C_7:
6943 		MLX5_SET(fte_match_set_misc2, misc2_m, metadata_reg_c_7, mask);
6944 		MLX5_SET(fte_match_set_misc2, misc2_v, metadata_reg_c_7, data);
6945 		break;
6946 	default:
6947 		MLX5_ASSERT(false);
6948 		break;
6949 	}
6950 }
6951 
6952 /**
6953  * Add MARK item to matcher
6954  *
6955  * @param[in] dev
6956  *   The device to configure through.
6957  * @param[in, out] matcher
6958  *   Flow matcher.
6959  * @param[in, out] key
6960  *   Flow matcher value.
6961  * @param[in] item
6962  *   Flow pattern to translate.
6963  */
6964 static void
6965 flow_dv_translate_item_mark(struct rte_eth_dev *dev,
6966 			    void *matcher, void *key,
6967 			    const struct rte_flow_item *item)
6968 {
6969 	struct mlx5_priv *priv = dev->data->dev_private;
6970 	const struct rte_flow_item_mark *mark;
6971 	uint32_t value;
6972 	uint32_t mask;
6973 
6974 	mark = item->mask ? (const void *)item->mask :
6975 			    &rte_flow_item_mark_mask;
6976 	mask = mark->id & priv->sh->dv_mark_mask;
6977 	mark = (const void *)item->spec;
6978 	MLX5_ASSERT(mark);
6979 	value = mark->id & priv->sh->dv_mark_mask & mask;
6980 	if (mask) {
6981 		enum modify_reg reg;
6982 
6983 		/* Get the metadata register index for the mark. */
6984 		reg = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, NULL);
6985 		MLX5_ASSERT(reg > 0);
6986 		if (reg == REG_C_0) {
6987 			struct mlx5_priv *priv = dev->data->dev_private;
6988 			uint32_t msk_c0 = priv->sh->dv_regc0_mask;
6989 			uint32_t shl_c0 = rte_bsf32(msk_c0);
6990 
6991 			mask &= msk_c0;
6992 			mask <<= shl_c0;
6993 			value <<= shl_c0;
6994 		}
6995 		flow_dv_match_meta_reg(matcher, key, reg, value, mask);
6996 	}
6997 }
6998 
6999 /**
7000  * Add META item to matcher
7001  *
7002  * @param[in] dev
7003  *   The devich to configure through.
7004  * @param[in, out] matcher
7005  *   Flow matcher.
7006  * @param[in, out] key
7007  *   Flow matcher value.
7008  * @param[in] attr
7009  *   Attributes of flow that includes this item.
7010  * @param[in] item
7011  *   Flow pattern to translate.
7012  */
7013 static void
7014 flow_dv_translate_item_meta(struct rte_eth_dev *dev,
7015 			    void *matcher, void *key,
7016 			    const struct rte_flow_attr *attr,
7017 			    const struct rte_flow_item *item)
7018 {
7019 	const struct rte_flow_item_meta *meta_m;
7020 	const struct rte_flow_item_meta *meta_v;
7021 
7022 	meta_m = (const void *)item->mask;
7023 	if (!meta_m)
7024 		meta_m = &rte_flow_item_meta_mask;
7025 	meta_v = (const void *)item->spec;
7026 	if (meta_v) {
7027 		int reg;
7028 		uint32_t value = meta_v->data;
7029 		uint32_t mask = meta_m->data;
7030 
7031 		reg = flow_dv_get_metadata_reg(dev, attr, NULL);
7032 		if (reg < 0)
7033 			return;
7034 		/*
7035 		 * In datapath code there is no endianness
7036 		 * coversions for perfromance reasons, all
7037 		 * pattern conversions are done in rte_flow.
7038 		 */
7039 		value = rte_cpu_to_be_32(value);
7040 		mask = rte_cpu_to_be_32(mask);
7041 		if (reg == REG_C_0) {
7042 			struct mlx5_priv *priv = dev->data->dev_private;
7043 			uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7044 			uint32_t shl_c0 = rte_bsf32(msk_c0);
7045 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
7046 			uint32_t shr_c0 = __builtin_clz(priv->sh->dv_meta_mask);
7047 
7048 			value >>= shr_c0;
7049 			mask >>= shr_c0;
7050 #endif
7051 			value <<= shl_c0;
7052 			mask <<= shl_c0;
7053 			MLX5_ASSERT(msk_c0);
7054 			MLX5_ASSERT(!(~msk_c0 & mask));
7055 		}
7056 		flow_dv_match_meta_reg(matcher, key, reg, value, mask);
7057 	}
7058 }
7059 
7060 /**
7061  * Add vport metadata Reg C0 item to matcher
7062  *
7063  * @param[in, out] matcher
7064  *   Flow matcher.
7065  * @param[in, out] key
7066  *   Flow matcher value.
7067  * @param[in] reg
7068  *   Flow pattern to translate.
7069  */
7070 static void
7071 flow_dv_translate_item_meta_vport(void *matcher, void *key,
7072 				  uint32_t value, uint32_t mask)
7073 {
7074 	flow_dv_match_meta_reg(matcher, key, REG_C_0, value, mask);
7075 }
7076 
7077 /**
7078  * Add tag item to matcher
7079  *
7080  * @param[in] dev
7081  *   The devich to configure through.
7082  * @param[in, out] matcher
7083  *   Flow matcher.
7084  * @param[in, out] key
7085  *   Flow matcher value.
7086  * @param[in] item
7087  *   Flow pattern to translate.
7088  */
7089 static void
7090 flow_dv_translate_mlx5_item_tag(struct rte_eth_dev *dev,
7091 				void *matcher, void *key,
7092 				const struct rte_flow_item *item)
7093 {
7094 	const struct mlx5_rte_flow_item_tag *tag_v = item->spec;
7095 	const struct mlx5_rte_flow_item_tag *tag_m = item->mask;
7096 	uint32_t mask, value;
7097 
7098 	MLX5_ASSERT(tag_v);
7099 	value = tag_v->data;
7100 	mask = tag_m ? tag_m->data : UINT32_MAX;
7101 	if (tag_v->id == REG_C_0) {
7102 		struct mlx5_priv *priv = dev->data->dev_private;
7103 		uint32_t msk_c0 = priv->sh->dv_regc0_mask;
7104 		uint32_t shl_c0 = rte_bsf32(msk_c0);
7105 
7106 		mask &= msk_c0;
7107 		mask <<= shl_c0;
7108 		value <<= shl_c0;
7109 	}
7110 	flow_dv_match_meta_reg(matcher, key, tag_v->id, value, mask);
7111 }
7112 
7113 /**
7114  * Add TAG item to matcher
7115  *
7116  * @param[in] dev
7117  *   The devich to configure through.
7118  * @param[in, out] matcher
7119  *   Flow matcher.
7120  * @param[in, out] key
7121  *   Flow matcher value.
7122  * @param[in] item
7123  *   Flow pattern to translate.
7124  */
7125 static void
7126 flow_dv_translate_item_tag(struct rte_eth_dev *dev,
7127 			   void *matcher, void *key,
7128 			   const struct rte_flow_item *item)
7129 {
7130 	const struct rte_flow_item_tag *tag_v = item->spec;
7131 	const struct rte_flow_item_tag *tag_m = item->mask;
7132 	enum modify_reg reg;
7133 
7134 	MLX5_ASSERT(tag_v);
7135 	tag_m = tag_m ? tag_m : &rte_flow_item_tag_mask;
7136 	/* Get the metadata register index for the tag. */
7137 	reg = mlx5_flow_get_reg_id(dev, MLX5_APP_TAG, tag_v->index, NULL);
7138 	MLX5_ASSERT(reg > 0);
7139 	flow_dv_match_meta_reg(matcher, key, reg, tag_v->data, tag_m->data);
7140 }
7141 
7142 /**
7143  * Add source vport match to the specified matcher.
7144  *
7145  * @param[in, out] matcher
7146  *   Flow matcher.
7147  * @param[in, out] key
7148  *   Flow matcher value.
7149  * @param[in] port
7150  *   Source vport value to match
7151  * @param[in] mask
7152  *   Mask
7153  */
7154 static void
7155 flow_dv_translate_item_source_vport(void *matcher, void *key,
7156 				    int16_t port, uint16_t mask)
7157 {
7158 	void *misc_m = MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7159 	void *misc_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7160 
7161 	MLX5_SET(fte_match_set_misc, misc_m, source_port, mask);
7162 	MLX5_SET(fte_match_set_misc, misc_v, source_port, port);
7163 }
7164 
7165 /**
7166  * Translate port-id item to eswitch match on  port-id.
7167  *
7168  * @param[in] dev
7169  *   The devich to configure through.
7170  * @param[in, out] matcher
7171  *   Flow matcher.
7172  * @param[in, out] key
7173  *   Flow matcher value.
7174  * @param[in] item
7175  *   Flow pattern to translate.
7176  *
7177  * @return
7178  *   0 on success, a negative errno value otherwise.
7179  */
7180 static int
7181 flow_dv_translate_item_port_id(struct rte_eth_dev *dev, void *matcher,
7182 			       void *key, const struct rte_flow_item *item)
7183 {
7184 	const struct rte_flow_item_port_id *pid_m = item ? item->mask : NULL;
7185 	const struct rte_flow_item_port_id *pid_v = item ? item->spec : NULL;
7186 	struct mlx5_priv *priv;
7187 	uint16_t mask, id;
7188 
7189 	mask = pid_m ? pid_m->id : 0xffff;
7190 	id = pid_v ? pid_v->id : dev->data->port_id;
7191 	priv = mlx5_port_to_eswitch_info(id, item == NULL);
7192 	if (!priv)
7193 		return -rte_errno;
7194 	/* Translate to vport field or to metadata, depending on mode. */
7195 	if (priv->vport_meta_mask)
7196 		flow_dv_translate_item_meta_vport(matcher, key,
7197 						  priv->vport_meta_tag,
7198 						  priv->vport_meta_mask);
7199 	else
7200 		flow_dv_translate_item_source_vport(matcher, key,
7201 						    priv->vport_id, mask);
7202 	return 0;
7203 }
7204 
7205 /**
7206  * Add ICMP6 item to matcher and to the value.
7207  *
7208  * @param[in, out] matcher
7209  *   Flow matcher.
7210  * @param[in, out] key
7211  *   Flow matcher value.
7212  * @param[in] item
7213  *   Flow pattern to translate.
7214  * @param[in] inner
7215  *   Item is inner pattern.
7216  */
7217 static void
7218 flow_dv_translate_item_icmp6(void *matcher, void *key,
7219 			      const struct rte_flow_item *item,
7220 			      int inner)
7221 {
7222 	const struct rte_flow_item_icmp6 *icmp6_m = item->mask;
7223 	const struct rte_flow_item_icmp6 *icmp6_v = item->spec;
7224 	void *headers_m;
7225 	void *headers_v;
7226 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7227 				     misc_parameters_3);
7228 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7229 	if (inner) {
7230 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7231 					 inner_headers);
7232 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7233 	} else {
7234 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7235 					 outer_headers);
7236 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7237 	}
7238 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7239 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMPV6);
7240 	if (!icmp6_v)
7241 		return;
7242 	if (!icmp6_m)
7243 		icmp6_m = &rte_flow_item_icmp6_mask;
7244 	/*
7245 	 * Force flow only to match the non-fragmented IPv6 ICMPv6 packets.
7246 	 * If only the protocol is specified, no need to match the frag.
7247 	 */
7248 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7249 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7250 	MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_type, icmp6_m->type);
7251 	MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_type,
7252 		 icmp6_v->type & icmp6_m->type);
7253 	MLX5_SET(fte_match_set_misc3, misc3_m, icmpv6_code, icmp6_m->code);
7254 	MLX5_SET(fte_match_set_misc3, misc3_v, icmpv6_code,
7255 		 icmp6_v->code & icmp6_m->code);
7256 }
7257 
7258 /**
7259  * Add ICMP item to matcher and to the value.
7260  *
7261  * @param[in, out] matcher
7262  *   Flow matcher.
7263  * @param[in, out] key
7264  *   Flow matcher value.
7265  * @param[in] item
7266  *   Flow pattern to translate.
7267  * @param[in] inner
7268  *   Item is inner pattern.
7269  */
7270 static void
7271 flow_dv_translate_item_icmp(void *matcher, void *key,
7272 			    const struct rte_flow_item *item,
7273 			    int inner)
7274 {
7275 	const struct rte_flow_item_icmp *icmp_m = item->mask;
7276 	const struct rte_flow_item_icmp *icmp_v = item->spec;
7277 	void *headers_m;
7278 	void *headers_v;
7279 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7280 				     misc_parameters_3);
7281 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7282 	if (inner) {
7283 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7284 					 inner_headers);
7285 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7286 	} else {
7287 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7288 					 outer_headers);
7289 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7290 	}
7291 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xFF);
7292 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_ICMP);
7293 	if (!icmp_v)
7294 		return;
7295 	if (!icmp_m)
7296 		icmp_m = &rte_flow_item_icmp_mask;
7297 	/*
7298 	 * Force flow only to match the non-fragmented IPv4 ICMP packets.
7299 	 * If only the protocol is specified, no need to match the frag.
7300 	 */
7301 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, frag, 1);
7302 	MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
7303 	MLX5_SET(fte_match_set_misc3, misc3_m, icmp_type,
7304 		 icmp_m->hdr.icmp_type);
7305 	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_type,
7306 		 icmp_v->hdr.icmp_type & icmp_m->hdr.icmp_type);
7307 	MLX5_SET(fte_match_set_misc3, misc3_m, icmp_code,
7308 		 icmp_m->hdr.icmp_code);
7309 	MLX5_SET(fte_match_set_misc3, misc3_v, icmp_code,
7310 		 icmp_v->hdr.icmp_code & icmp_m->hdr.icmp_code);
7311 }
7312 
7313 /**
7314  * Add GTP item to matcher and to the value.
7315  *
7316  * @param[in, out] matcher
7317  *   Flow matcher.
7318  * @param[in, out] key
7319  *   Flow matcher value.
7320  * @param[in] item
7321  *   Flow pattern to translate.
7322  * @param[in] inner
7323  *   Item is inner pattern.
7324  */
7325 static void
7326 flow_dv_translate_item_gtp(void *matcher, void *key,
7327 			   const struct rte_flow_item *item, int inner)
7328 {
7329 	const struct rte_flow_item_gtp *gtp_m = item->mask;
7330 	const struct rte_flow_item_gtp *gtp_v = item->spec;
7331 	void *headers_m;
7332 	void *headers_v;
7333 	void *misc3_m = MLX5_ADDR_OF(fte_match_param, matcher,
7334 				     misc_parameters_3);
7335 	void *misc3_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_3);
7336 	uint16_t dport = RTE_GTPU_UDP_PORT;
7337 
7338 	if (inner) {
7339 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7340 					 inner_headers);
7341 		headers_v = MLX5_ADDR_OF(fte_match_param, key, inner_headers);
7342 	} else {
7343 		headers_m = MLX5_ADDR_OF(fte_match_param, matcher,
7344 					 outer_headers);
7345 		headers_v = MLX5_ADDR_OF(fte_match_param, key, outer_headers);
7346 	}
7347 	if (!MLX5_GET16(fte_match_set_lyr_2_4, headers_v, udp_dport)) {
7348 		MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xFFFF);
7349 		MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport, dport);
7350 	}
7351 	if (!gtp_v)
7352 		return;
7353 	if (!gtp_m)
7354 		gtp_m = &rte_flow_item_gtp_mask;
7355 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_flags,
7356 		 gtp_m->v_pt_rsv_flags);
7357 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_flags,
7358 		 gtp_v->v_pt_rsv_flags & gtp_m->v_pt_rsv_flags);
7359 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_msg_type, gtp_m->msg_type);
7360 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_msg_type,
7361 		 gtp_v->msg_type & gtp_m->msg_type);
7362 	MLX5_SET(fte_match_set_misc3, misc3_m, gtpu_teid,
7363 		 rte_be_to_cpu_32(gtp_m->teid));
7364 	MLX5_SET(fte_match_set_misc3, misc3_v, gtpu_teid,
7365 		 rte_be_to_cpu_32(gtp_v->teid & gtp_m->teid));
7366 }
7367 
7368 /**
7369  * Add eCPRI item to matcher and to the value.
7370  *
7371  * @param[in] dev
7372  *   The devich to configure through.
7373  * @param[in, out] matcher
7374  *   Flow matcher.
7375  * @param[in, out] key
7376  *   Flow matcher value.
7377  * @param[in] item
7378  *   Flow pattern to translate.
7379  * @param[in] samples
7380  *   Sample IDs to be used in the matching.
7381  */
7382 static void
7383 flow_dv_translate_item_ecpri(struct rte_eth_dev *dev, void *matcher,
7384 			     void *key, const struct rte_flow_item *item)
7385 {
7386 	struct mlx5_priv *priv = dev->data->dev_private;
7387 	const struct rte_flow_item_ecpri *ecpri_m = item->mask;
7388 	const struct rte_flow_item_ecpri *ecpri_v = item->spec;
7389 	void *misc4_m = MLX5_ADDR_OF(fte_match_param, matcher,
7390 				     misc_parameters_4);
7391 	void *misc4_v = MLX5_ADDR_OF(fte_match_param, key, misc_parameters_4);
7392 	uint32_t *samples;
7393 	void *dw_m;
7394 	void *dw_v;
7395 
7396 	if (!ecpri_v)
7397 		return;
7398 	if (!ecpri_m)
7399 		ecpri_m = &rte_flow_item_ecpri_mask;
7400 	/*
7401 	 * Maximal four DW samples are supported in a single matching now.
7402 	 * Two are used now for a eCPRI matching:
7403 	 * 1. Type: one byte, mask should be 0x00ff0000 in network order
7404 	 * 2. ID of a message: one or two bytes, mask 0xffff0000 or 0xff000000
7405 	 *    if any.
7406 	 */
7407 	if (!ecpri_m->hdr.common.u32)
7408 		return;
7409 	samples = priv->sh->fp[MLX5_FLEX_PARSER_ECPRI_0].ids;
7410 	/* Need to take the whole DW as the mask to fill the entry. */
7411 	dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7412 			    prog_sample_field_value_0);
7413 	dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7414 			    prog_sample_field_value_0);
7415 	/* Already big endian (network order) in the header. */
7416 	*(uint32_t *)dw_m = ecpri_m->hdr.common.u32;
7417 	*(uint32_t *)dw_v = ecpri_v->hdr.common.u32;
7418 	/* Sample#0, used for matching type, offset 0. */
7419 	MLX5_SET(fte_match_set_misc4, misc4_m,
7420 		 prog_sample_field_id_0, samples[0]);
7421 	/* It makes no sense to set the sample ID in the mask field. */
7422 	MLX5_SET(fte_match_set_misc4, misc4_v,
7423 		 prog_sample_field_id_0, samples[0]);
7424 	/*
7425 	 * Checking if message body part needs to be matched.
7426 	 * Some wildcard rules only matching type field should be supported.
7427 	 */
7428 	if (ecpri_m->hdr.dummy[0]) {
7429 		switch (ecpri_v->hdr.common.type) {
7430 		case RTE_ECPRI_MSG_TYPE_IQ_DATA:
7431 		case RTE_ECPRI_MSG_TYPE_RTC_CTRL:
7432 		case RTE_ECPRI_MSG_TYPE_DLY_MSR:
7433 			dw_m = MLX5_ADDR_OF(fte_match_set_misc4, misc4_m,
7434 					    prog_sample_field_value_1);
7435 			dw_v = MLX5_ADDR_OF(fte_match_set_misc4, misc4_v,
7436 					    prog_sample_field_value_1);
7437 			*(uint32_t *)dw_m = ecpri_m->hdr.dummy[0];
7438 			*(uint32_t *)dw_v = ecpri_v->hdr.dummy[0];
7439 			/* Sample#1, to match message body, offset 4. */
7440 			MLX5_SET(fte_match_set_misc4, misc4_m,
7441 				 prog_sample_field_id_1, samples[1]);
7442 			MLX5_SET(fte_match_set_misc4, misc4_v,
7443 				 prog_sample_field_id_1, samples[1]);
7444 			break;
7445 		default:
7446 			/* Others, do not match any sample ID. */
7447 			break;
7448 		}
7449 	}
7450 }
7451 
7452 static uint32_t matcher_zero[MLX5_ST_SZ_DW(fte_match_param)] = { 0 };
7453 
7454 #define HEADER_IS_ZERO(match_criteria, headers)				     \
7455 	!(memcmp(MLX5_ADDR_OF(fte_match_param, match_criteria, headers),     \
7456 		 matcher_zero, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
7457 
7458 /**
7459  * Calculate flow matcher enable bitmap.
7460  *
7461  * @param match_criteria
7462  *   Pointer to flow matcher criteria.
7463  *
7464  * @return
7465  *   Bitmap of enabled fields.
7466  */
7467 static uint8_t
7468 flow_dv_matcher_enable(uint32_t *match_criteria)
7469 {
7470 	uint8_t match_criteria_enable;
7471 
7472 	match_criteria_enable =
7473 		(!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
7474 		MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT;
7475 	match_criteria_enable |=
7476 		(!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
7477 		MLX5_MATCH_CRITERIA_ENABLE_MISC_BIT;
7478 	match_criteria_enable |=
7479 		(!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
7480 		MLX5_MATCH_CRITERIA_ENABLE_INNER_BIT;
7481 	match_criteria_enable |=
7482 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
7483 		MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
7484 	match_criteria_enable |=
7485 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_3)) <<
7486 		MLX5_MATCH_CRITERIA_ENABLE_MISC3_BIT;
7487 	match_criteria_enable |=
7488 		(!HEADER_IS_ZERO(match_criteria, misc_parameters_4)) <<
7489 		MLX5_MATCH_CRITERIA_ENABLE_MISC4_BIT;
7490 	return match_criteria_enable;
7491 }
7492 
7493 
7494 /**
7495  * Get a flow table.
7496  *
7497  * @param[in, out] dev
7498  *   Pointer to rte_eth_dev structure.
7499  * @param[in] table_id
7500  *   Table id to use.
7501  * @param[in] egress
7502  *   Direction of the table.
7503  * @param[in] transfer
7504  *   E-Switch or NIC flow.
7505  * @param[out] error
7506  *   pointer to error structure.
7507  *
7508  * @return
7509  *   Returns tables resource based on the index, NULL in case of failed.
7510  */
7511 static struct mlx5_flow_tbl_resource *
7512 flow_dv_tbl_resource_get(struct rte_eth_dev *dev,
7513 			 uint32_t table_id, uint8_t egress,
7514 			 uint8_t transfer,
7515 			 struct rte_flow_error *error)
7516 {
7517 	struct mlx5_priv *priv = dev->data->dev_private;
7518 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7519 	struct mlx5_flow_tbl_resource *tbl;
7520 	union mlx5_flow_tbl_key table_key = {
7521 		{
7522 			.table_id = table_id,
7523 			.reserved = 0,
7524 			.domain = !!transfer,
7525 			.direction = !!egress,
7526 		}
7527 	};
7528 	struct mlx5_hlist_entry *pos = mlx5_hlist_lookup(sh->flow_tbls,
7529 							 table_key.v64);
7530 	struct mlx5_flow_tbl_data_entry *tbl_data;
7531 	uint32_t idx = 0;
7532 	int ret;
7533 	void *domain;
7534 
7535 	if (pos) {
7536 		tbl_data = container_of(pos, struct mlx5_flow_tbl_data_entry,
7537 					entry);
7538 		tbl = &tbl_data->tbl;
7539 		rte_atomic32_inc(&tbl->refcnt);
7540 		return tbl;
7541 	}
7542 	tbl_data = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_JUMP], &idx);
7543 	if (!tbl_data) {
7544 		rte_flow_error_set(error, ENOMEM,
7545 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7546 				   NULL,
7547 				   "cannot allocate flow table data entry");
7548 		return NULL;
7549 	}
7550 	tbl_data->idx = idx;
7551 	tbl = &tbl_data->tbl;
7552 	pos = &tbl_data->entry;
7553 	if (transfer)
7554 		domain = sh->fdb_domain;
7555 	else if (egress)
7556 		domain = sh->tx_domain;
7557 	else
7558 		domain = sh->rx_domain;
7559 	ret = mlx5_flow_os_create_flow_tbl(domain, table_id, &tbl->obj);
7560 	if (ret) {
7561 		rte_flow_error_set(error, ENOMEM,
7562 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7563 				   NULL, "cannot create flow table object");
7564 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7565 		return NULL;
7566 	}
7567 	/*
7568 	 * No multi-threads now, but still better to initialize the reference
7569 	 * count before insert it into the hash list.
7570 	 */
7571 	rte_atomic32_init(&tbl->refcnt);
7572 	/* Jump action reference count is initialized here. */
7573 	rte_atomic32_init(&tbl_data->jump.refcnt);
7574 	pos->key = table_key.v64;
7575 	ret = mlx5_hlist_insert(sh->flow_tbls, pos);
7576 	if (ret < 0) {
7577 		rte_flow_error_set(error, -ret,
7578 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7579 				   "cannot insert flow table data entry");
7580 		mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7581 		mlx5_ipool_free(sh->ipool[MLX5_IPOOL_JUMP], idx);
7582 	}
7583 	rte_atomic32_inc(&tbl->refcnt);
7584 	return tbl;
7585 }
7586 
7587 /**
7588  * Release a flow table.
7589  *
7590  * @param[in] dev
7591  *   Pointer to rte_eth_dev structure.
7592  * @param[in] tbl
7593  *   Table resource to be released.
7594  *
7595  * @return
7596  *   Returns 0 if table was released, else return 1;
7597  */
7598 static int
7599 flow_dv_tbl_resource_release(struct rte_eth_dev *dev,
7600 			     struct mlx5_flow_tbl_resource *tbl)
7601 {
7602 	struct mlx5_priv *priv = dev->data->dev_private;
7603 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7604 	struct mlx5_flow_tbl_data_entry *tbl_data =
7605 		container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7606 
7607 	if (!tbl)
7608 		return 0;
7609 	if (rte_atomic32_dec_and_test(&tbl->refcnt)) {
7610 		struct mlx5_hlist_entry *pos = &tbl_data->entry;
7611 
7612 		mlx5_flow_os_destroy_flow_tbl(tbl->obj);
7613 		tbl->obj = NULL;
7614 		/* remove the entry from the hash list and free memory. */
7615 		mlx5_hlist_remove(sh->flow_tbls, pos);
7616 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_JUMP],
7617 				tbl_data->idx);
7618 		return 0;
7619 	}
7620 	return 1;
7621 }
7622 
7623 /**
7624  * Register the flow matcher.
7625  *
7626  * @param[in, out] dev
7627  *   Pointer to rte_eth_dev structure.
7628  * @param[in, out] matcher
7629  *   Pointer to flow matcher.
7630  * @param[in, out] key
7631  *   Pointer to flow table key.
7632  * @parm[in, out] dev_flow
7633  *   Pointer to the dev_flow.
7634  * @param[out] error
7635  *   pointer to error structure.
7636  *
7637  * @return
7638  *   0 on success otherwise -errno and errno is set.
7639  */
7640 static int
7641 flow_dv_matcher_register(struct rte_eth_dev *dev,
7642 			 struct mlx5_flow_dv_matcher *matcher,
7643 			 union mlx5_flow_tbl_key *key,
7644 			 struct mlx5_flow *dev_flow,
7645 			 struct rte_flow_error *error)
7646 {
7647 	struct mlx5_priv *priv = dev->data->dev_private;
7648 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7649 	struct mlx5_flow_dv_matcher *cache_matcher;
7650 	struct mlx5dv_flow_matcher_attr dv_attr = {
7651 		.type = IBV_FLOW_ATTR_NORMAL,
7652 		.match_mask = (void *)&matcher->mask,
7653 	};
7654 	struct mlx5_flow_tbl_resource *tbl;
7655 	struct mlx5_flow_tbl_data_entry *tbl_data;
7656 	int ret;
7657 
7658 	tbl = flow_dv_tbl_resource_get(dev, key->table_id, key->direction,
7659 				       key->domain, error);
7660 	if (!tbl)
7661 		return -rte_errno;	/* No need to refill the error info */
7662 	tbl_data = container_of(tbl, struct mlx5_flow_tbl_data_entry, tbl);
7663 	/* Lookup from cache. */
7664 	LIST_FOREACH(cache_matcher, &tbl_data->matchers, next) {
7665 		if (matcher->crc == cache_matcher->crc &&
7666 		    matcher->priority == cache_matcher->priority &&
7667 		    !memcmp((const void *)matcher->mask.buf,
7668 			    (const void *)cache_matcher->mask.buf,
7669 			    cache_matcher->mask.size)) {
7670 			DRV_LOG(DEBUG,
7671 				"%s group %u priority %hd use %s "
7672 				"matcher %p: refcnt %d++",
7673 				key->domain ? "FDB" : "NIC", key->table_id,
7674 				cache_matcher->priority,
7675 				key->direction ? "tx" : "rx",
7676 				(void *)cache_matcher,
7677 				rte_atomic32_read(&cache_matcher->refcnt));
7678 			rte_atomic32_inc(&cache_matcher->refcnt);
7679 			dev_flow->handle->dvh.matcher = cache_matcher;
7680 			/* old matcher should not make the table ref++. */
7681 			flow_dv_tbl_resource_release(dev, tbl);
7682 			return 0;
7683 		}
7684 	}
7685 	/* Register new matcher. */
7686 	cache_matcher = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*cache_matcher), 0,
7687 				    SOCKET_ID_ANY);
7688 	if (!cache_matcher) {
7689 		flow_dv_tbl_resource_release(dev, tbl);
7690 		return rte_flow_error_set(error, ENOMEM,
7691 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7692 					  "cannot allocate matcher memory");
7693 	}
7694 	*cache_matcher = *matcher;
7695 	dv_attr.match_criteria_enable =
7696 		flow_dv_matcher_enable(cache_matcher->mask.buf);
7697 	dv_attr.priority = matcher->priority;
7698 	if (key->direction)
7699 		dv_attr.flags |= IBV_FLOW_ATTR_FLAGS_EGRESS;
7700 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, tbl->obj,
7701 					       &cache_matcher->matcher_object);
7702 	if (ret) {
7703 		mlx5_free(cache_matcher);
7704 #ifdef HAVE_MLX5DV_DR
7705 		flow_dv_tbl_resource_release(dev, tbl);
7706 #endif
7707 		return rte_flow_error_set(error, ENOMEM,
7708 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7709 					  NULL, "cannot create matcher");
7710 	}
7711 	/* Save the table information */
7712 	cache_matcher->tbl = tbl;
7713 	rte_atomic32_init(&cache_matcher->refcnt);
7714 	/* only matcher ref++, table ref++ already done above in get API. */
7715 	rte_atomic32_inc(&cache_matcher->refcnt);
7716 	LIST_INSERT_HEAD(&tbl_data->matchers, cache_matcher, next);
7717 	dev_flow->handle->dvh.matcher = cache_matcher;
7718 	DRV_LOG(DEBUG, "%s group %u priority %hd new %s matcher %p: refcnt %d",
7719 		key->domain ? "FDB" : "NIC", key->table_id,
7720 		cache_matcher->priority,
7721 		key->direction ? "tx" : "rx", (void *)cache_matcher,
7722 		rte_atomic32_read(&cache_matcher->refcnt));
7723 	return 0;
7724 }
7725 
7726 /**
7727  * Find existing tag resource or create and register a new one.
7728  *
7729  * @param dev[in, out]
7730  *   Pointer to rte_eth_dev structure.
7731  * @param[in, out] tag_be24
7732  *   Tag value in big endian then R-shift 8.
7733  * @parm[in, out] dev_flow
7734  *   Pointer to the dev_flow.
7735  * @param[out] error
7736  *   pointer to error structure.
7737  *
7738  * @return
7739  *   0 on success otherwise -errno and errno is set.
7740  */
7741 static int
7742 flow_dv_tag_resource_register
7743 			(struct rte_eth_dev *dev,
7744 			 uint32_t tag_be24,
7745 			 struct mlx5_flow *dev_flow,
7746 			 struct rte_flow_error *error)
7747 {
7748 	struct mlx5_priv *priv = dev->data->dev_private;
7749 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7750 	struct mlx5_flow_dv_tag_resource *cache_resource;
7751 	struct mlx5_hlist_entry *entry;
7752 	int ret;
7753 
7754 	/* Lookup a matching resource from cache. */
7755 	entry = mlx5_hlist_lookup(sh->tag_table, (uint64_t)tag_be24);
7756 	if (entry) {
7757 		cache_resource = container_of
7758 			(entry, struct mlx5_flow_dv_tag_resource, entry);
7759 		rte_atomic32_inc(&cache_resource->refcnt);
7760 		dev_flow->handle->dvh.rix_tag = cache_resource->idx;
7761 		dev_flow->dv.tag_resource = cache_resource;
7762 		DRV_LOG(DEBUG, "cached tag resource %p: refcnt now %d++",
7763 			(void *)cache_resource,
7764 			rte_atomic32_read(&cache_resource->refcnt));
7765 		return 0;
7766 	}
7767 	/* Register new resource. */
7768 	cache_resource = mlx5_ipool_zmalloc(sh->ipool[MLX5_IPOOL_TAG],
7769 				       &dev_flow->handle->dvh.rix_tag);
7770 	if (!cache_resource)
7771 		return rte_flow_error_set(error, ENOMEM,
7772 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
7773 					  "cannot allocate resource memory");
7774 	cache_resource->entry.key = (uint64_t)tag_be24;
7775 	ret = mlx5_flow_os_create_flow_action_tag(tag_be24,
7776 						  &cache_resource->action);
7777 	if (ret) {
7778 		mlx5_free(cache_resource);
7779 		return rte_flow_error_set(error, ENOMEM,
7780 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7781 					  NULL, "cannot create action");
7782 	}
7783 	rte_atomic32_init(&cache_resource->refcnt);
7784 	rte_atomic32_inc(&cache_resource->refcnt);
7785 	if (mlx5_hlist_insert(sh->tag_table, &cache_resource->entry)) {
7786 		mlx5_flow_os_destroy_flow_action(cache_resource->action);
7787 		mlx5_free(cache_resource);
7788 		return rte_flow_error_set(error, EEXIST,
7789 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
7790 					  NULL, "cannot insert tag");
7791 	}
7792 	dev_flow->dv.tag_resource = cache_resource;
7793 	DRV_LOG(DEBUG, "new tag resource %p: refcnt now %d++",
7794 		(void *)cache_resource,
7795 		rte_atomic32_read(&cache_resource->refcnt));
7796 	return 0;
7797 }
7798 
7799 /**
7800  * Release the tag.
7801  *
7802  * @param dev
7803  *   Pointer to Ethernet device.
7804  * @param tag_idx
7805  *   Tag index.
7806  *
7807  * @return
7808  *   1 while a reference on it exists, 0 when freed.
7809  */
7810 static int
7811 flow_dv_tag_release(struct rte_eth_dev *dev,
7812 		    uint32_t tag_idx)
7813 {
7814 	struct mlx5_priv *priv = dev->data->dev_private;
7815 	struct mlx5_dev_ctx_shared *sh = priv->sh;
7816 	struct mlx5_flow_dv_tag_resource *tag;
7817 
7818 	tag = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7819 	if (!tag)
7820 		return 0;
7821 	DRV_LOG(DEBUG, "port %u tag %p: refcnt %d--",
7822 		dev->data->port_id, (void *)tag,
7823 		rte_atomic32_read(&tag->refcnt));
7824 	if (rte_atomic32_dec_and_test(&tag->refcnt)) {
7825 		claim_zero(mlx5_flow_os_destroy_flow_action(tag->action));
7826 		mlx5_hlist_remove(sh->tag_table, &tag->entry);
7827 		DRV_LOG(DEBUG, "port %u tag %p: removed",
7828 			dev->data->port_id, (void *)tag);
7829 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_TAG], tag_idx);
7830 		return 0;
7831 	}
7832 	return 1;
7833 }
7834 
7835 /**
7836  * Translate port ID action to vport.
7837  *
7838  * @param[in] dev
7839  *   Pointer to rte_eth_dev structure.
7840  * @param[in] action
7841  *   Pointer to the port ID action.
7842  * @param[out] dst_port_id
7843  *   The target port ID.
7844  * @param[out] error
7845  *   Pointer to the error structure.
7846  *
7847  * @return
7848  *   0 on success, a negative errno value otherwise and rte_errno is set.
7849  */
7850 static int
7851 flow_dv_translate_action_port_id(struct rte_eth_dev *dev,
7852 				 const struct rte_flow_action *action,
7853 				 uint32_t *dst_port_id,
7854 				 struct rte_flow_error *error)
7855 {
7856 	uint32_t port;
7857 	struct mlx5_priv *priv;
7858 	const struct rte_flow_action_port_id *conf =
7859 			(const struct rte_flow_action_port_id *)action->conf;
7860 
7861 	port = conf->original ? dev->data->port_id : conf->id;
7862 	priv = mlx5_port_to_eswitch_info(port, false);
7863 	if (!priv)
7864 		return rte_flow_error_set(error, -rte_errno,
7865 					  RTE_FLOW_ERROR_TYPE_ACTION,
7866 					  NULL,
7867 					  "No eswitch info was found for port");
7868 #ifdef HAVE_MLX5DV_DR_DEVX_PORT
7869 	/*
7870 	 * This parameter is transferred to
7871 	 * mlx5dv_dr_action_create_dest_ib_port().
7872 	 */
7873 	*dst_port_id = priv->dev_port;
7874 #else
7875 	/*
7876 	 * Legacy mode, no LAG configurations is supported.
7877 	 * This parameter is transferred to
7878 	 * mlx5dv_dr_action_create_dest_vport().
7879 	 */
7880 	*dst_port_id = priv->vport_id;
7881 #endif
7882 	return 0;
7883 }
7884 
7885 /**
7886  * Create a counter with aging configuration.
7887  *
7888  * @param[in] dev
7889  *   Pointer to rte_eth_dev structure.
7890  * @param[out] count
7891  *   Pointer to the counter action configuration.
7892  * @param[in] age
7893  *   Pointer to the aging action configuration.
7894  *
7895  * @return
7896  *   Index to flow counter on success, 0 otherwise.
7897  */
7898 static uint32_t
7899 flow_dv_translate_create_counter(struct rte_eth_dev *dev,
7900 				struct mlx5_flow *dev_flow,
7901 				const struct rte_flow_action_count *count,
7902 				const struct rte_flow_action_age *age)
7903 {
7904 	uint32_t counter;
7905 	struct mlx5_age_param *age_param;
7906 
7907 	counter = flow_dv_counter_alloc(dev,
7908 				count ? count->shared : 0,
7909 				count ? count->id : 0,
7910 				dev_flow->dv.group, !!age);
7911 	if (!counter || age == NULL)
7912 		return counter;
7913 	age_param  = flow_dv_counter_idx_get_age(dev, counter);
7914 	/*
7915 	 * The counter age accuracy may have a bit delay. Have 3/4
7916 	 * second bias on the timeount in order to let it age in time.
7917 	 */
7918 	age_param->context = age->context ? age->context :
7919 		(void *)(uintptr_t)(dev_flow->flow_idx);
7920 	/*
7921 	 * The counter age accuracy may have a bit delay. Have 3/4
7922 	 * second bias on the timeount in order to let it age in time.
7923 	 */
7924 	age_param->timeout = age->timeout * 10 - MLX5_AGING_TIME_DELAY;
7925 	/* Set expire time in unit of 0.1 sec. */
7926 	age_param->port_id = dev->data->port_id;
7927 	age_param->expire = age_param->timeout +
7928 			rte_rdtsc() / (rte_get_tsc_hz() / 10);
7929 	rte_atomic16_set(&age_param->state, AGE_CANDIDATE);
7930 	return counter;
7931 }
7932 /**
7933  * Add Tx queue matcher
7934  *
7935  * @param[in] dev
7936  *   Pointer to the dev struct.
7937  * @param[in, out] matcher
7938  *   Flow matcher.
7939  * @param[in, out] key
7940  *   Flow matcher value.
7941  * @param[in] item
7942  *   Flow pattern to translate.
7943  * @param[in] inner
7944  *   Item is inner pattern.
7945  */
7946 static void
7947 flow_dv_translate_item_tx_queue(struct rte_eth_dev *dev,
7948 				void *matcher, void *key,
7949 				const struct rte_flow_item *item)
7950 {
7951 	const struct mlx5_rte_flow_item_tx_queue *queue_m;
7952 	const struct mlx5_rte_flow_item_tx_queue *queue_v;
7953 	void *misc_m =
7954 		MLX5_ADDR_OF(fte_match_param, matcher, misc_parameters);
7955 	void *misc_v =
7956 		MLX5_ADDR_OF(fte_match_param, key, misc_parameters);
7957 	struct mlx5_txq_ctrl *txq;
7958 	uint32_t queue;
7959 
7960 
7961 	queue_m = (const void *)item->mask;
7962 	if (!queue_m)
7963 		return;
7964 	queue_v = (const void *)item->spec;
7965 	if (!queue_v)
7966 		return;
7967 	txq = mlx5_txq_get(dev, queue_v->queue);
7968 	if (!txq)
7969 		return;
7970 	queue = txq->obj->sq->id;
7971 	MLX5_SET(fte_match_set_misc, misc_m, source_sqn, queue_m->queue);
7972 	MLX5_SET(fte_match_set_misc, misc_v, source_sqn,
7973 		 queue & queue_m->queue);
7974 	mlx5_txq_release(dev, queue_v->queue);
7975 }
7976 
7977 /**
7978  * Set the hash fields according to the @p flow information.
7979  *
7980  * @param[in] dev_flow
7981  *   Pointer to the mlx5_flow.
7982  * @param[in] rss_desc
7983  *   Pointer to the mlx5_flow_rss_desc.
7984  */
7985 static void
7986 flow_dv_hashfields_set(struct mlx5_flow *dev_flow,
7987 		       struct mlx5_flow_rss_desc *rss_desc)
7988 {
7989 	uint64_t items = dev_flow->handle->layers;
7990 	int rss_inner = 0;
7991 	uint64_t rss_types = rte_eth_rss_hf_refine(rss_desc->types);
7992 
7993 	dev_flow->hash_fields = 0;
7994 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
7995 	if (rss_desc->level >= 2) {
7996 		dev_flow->hash_fields |= IBV_RX_HASH_INNER;
7997 		rss_inner = 1;
7998 	}
7999 #endif
8000 	if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV4)) ||
8001 	    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV4))) {
8002 		if (rss_types & MLX5_IPV4_LAYER_TYPES) {
8003 			if (rss_types & ETH_RSS_L3_SRC_ONLY)
8004 				dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV4;
8005 			else if (rss_types & ETH_RSS_L3_DST_ONLY)
8006 				dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV4;
8007 			else
8008 				dev_flow->hash_fields |= MLX5_IPV4_IBV_RX_HASH;
8009 		}
8010 	} else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L3_IPV6)) ||
8011 		   (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L3_IPV6))) {
8012 		if (rss_types & MLX5_IPV6_LAYER_TYPES) {
8013 			if (rss_types & ETH_RSS_L3_SRC_ONLY)
8014 				dev_flow->hash_fields |= IBV_RX_HASH_SRC_IPV6;
8015 			else if (rss_types & ETH_RSS_L3_DST_ONLY)
8016 				dev_flow->hash_fields |= IBV_RX_HASH_DST_IPV6;
8017 			else
8018 				dev_flow->hash_fields |= MLX5_IPV6_IBV_RX_HASH;
8019 		}
8020 	}
8021 	if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_UDP)) ||
8022 	    (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_UDP))) {
8023 		if (rss_types & ETH_RSS_UDP) {
8024 			if (rss_types & ETH_RSS_L4_SRC_ONLY)
8025 				dev_flow->hash_fields |=
8026 						IBV_RX_HASH_SRC_PORT_UDP;
8027 			else if (rss_types & ETH_RSS_L4_DST_ONLY)
8028 				dev_flow->hash_fields |=
8029 						IBV_RX_HASH_DST_PORT_UDP;
8030 			else
8031 				dev_flow->hash_fields |= MLX5_UDP_IBV_RX_HASH;
8032 		}
8033 	} else if ((rss_inner && (items & MLX5_FLOW_LAYER_INNER_L4_TCP)) ||
8034 		   (!rss_inner && (items & MLX5_FLOW_LAYER_OUTER_L4_TCP))) {
8035 		if (rss_types & ETH_RSS_TCP) {
8036 			if (rss_types & ETH_RSS_L4_SRC_ONLY)
8037 				dev_flow->hash_fields |=
8038 						IBV_RX_HASH_SRC_PORT_TCP;
8039 			else if (rss_types & ETH_RSS_L4_DST_ONLY)
8040 				dev_flow->hash_fields |=
8041 						IBV_RX_HASH_DST_PORT_TCP;
8042 			else
8043 				dev_flow->hash_fields |= MLX5_TCP_IBV_RX_HASH;
8044 		}
8045 	}
8046 }
8047 
8048 /**
8049  * Fill the flow with DV spec, lock free
8050  * (mutex should be acquired by caller).
8051  *
8052  * @param[in] dev
8053  *   Pointer to rte_eth_dev structure.
8054  * @param[in, out] dev_flow
8055  *   Pointer to the sub flow.
8056  * @param[in] attr
8057  *   Pointer to the flow attributes.
8058  * @param[in] items
8059  *   Pointer to the list of items.
8060  * @param[in] actions
8061  *   Pointer to the list of actions.
8062  * @param[out] error
8063  *   Pointer to the error structure.
8064  *
8065  * @return
8066  *   0 on success, a negative errno value otherwise and rte_errno is set.
8067  */
8068 static int
8069 __flow_dv_translate(struct rte_eth_dev *dev,
8070 		    struct mlx5_flow *dev_flow,
8071 		    const struct rte_flow_attr *attr,
8072 		    const struct rte_flow_item items[],
8073 		    const struct rte_flow_action actions[],
8074 		    struct rte_flow_error *error)
8075 {
8076 	struct mlx5_priv *priv = dev->data->dev_private;
8077 	struct mlx5_dev_config *dev_conf = &priv->config;
8078 	struct rte_flow *flow = dev_flow->flow;
8079 	struct mlx5_flow_handle *handle = dev_flow->handle;
8080 	struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
8081 					      priv->rss_desc)
8082 					      [!!priv->flow_nested_idx];
8083 	uint64_t item_flags = 0;
8084 	uint64_t last_item = 0;
8085 	uint64_t action_flags = 0;
8086 	uint64_t priority = attr->priority;
8087 	struct mlx5_flow_dv_matcher matcher = {
8088 		.mask = {
8089 			.size = sizeof(matcher.mask.buf) -
8090 				MLX5_ST_SZ_BYTES(fte_match_set_misc4),
8091 		},
8092 	};
8093 	int actions_n = 0;
8094 	bool actions_end = false;
8095 	union {
8096 		struct mlx5_flow_dv_modify_hdr_resource res;
8097 		uint8_t len[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
8098 			    sizeof(struct mlx5_modification_cmd) *
8099 			    (MLX5_MAX_MODIFY_NUM + 1)];
8100 	} mhdr_dummy;
8101 	struct mlx5_flow_dv_modify_hdr_resource *mhdr_res = &mhdr_dummy.res;
8102 	const struct rte_flow_action_count *count = NULL;
8103 	const struct rte_flow_action_age *age = NULL;
8104 	union flow_dv_attr flow_attr = { .attr = 0 };
8105 	uint32_t tag_be;
8106 	union mlx5_flow_tbl_key tbl_key;
8107 	uint32_t modify_action_position = UINT32_MAX;
8108 	void *match_mask = matcher.mask.buf;
8109 	void *match_value = dev_flow->dv.value.buf;
8110 	uint8_t next_protocol = 0xff;
8111 	struct rte_vlan_hdr vlan = { 0 };
8112 	uint32_t table;
8113 	int ret = 0;
8114 
8115 	mhdr_res->ft_type = attr->egress ? MLX5DV_FLOW_TABLE_TYPE_NIC_TX :
8116 					   MLX5DV_FLOW_TABLE_TYPE_NIC_RX;
8117 	ret = mlx5_flow_group_to_table(attr, dev_flow->external, attr->group,
8118 				       !!priv->fdb_def_rule, &table, error);
8119 	if (ret)
8120 		return ret;
8121 	dev_flow->dv.group = table;
8122 	if (attr->transfer)
8123 		mhdr_res->ft_type = MLX5DV_FLOW_TABLE_TYPE_FDB;
8124 	if (priority == MLX5_FLOW_PRIO_RSVD)
8125 		priority = dev_conf->flow_prio - 1;
8126 	/* number of actions must be set to 0 in case of dirty stack. */
8127 	mhdr_res->actions_num = 0;
8128 	for (; !actions_end ; actions++) {
8129 		const struct rte_flow_action_queue *queue;
8130 		const struct rte_flow_action_rss *rss;
8131 		const struct rte_flow_action *action = actions;
8132 		const uint8_t *rss_key;
8133 		const struct rte_flow_action_jump *jump_data;
8134 		const struct rte_flow_action_meter *mtr;
8135 		struct mlx5_flow_tbl_resource *tbl;
8136 		uint32_t port_id = 0;
8137 		struct mlx5_flow_dv_port_id_action_resource port_id_resource;
8138 		int action_type = actions->type;
8139 		const struct rte_flow_action *found_action = NULL;
8140 		struct mlx5_flow_meter *fm = NULL;
8141 
8142 		if (!mlx5_flow_os_action_supported(action_type))
8143 			return rte_flow_error_set(error, ENOTSUP,
8144 						  RTE_FLOW_ERROR_TYPE_ACTION,
8145 						  actions,
8146 						  "action not supported");
8147 		switch (action_type) {
8148 		case RTE_FLOW_ACTION_TYPE_VOID:
8149 			break;
8150 		case RTE_FLOW_ACTION_TYPE_PORT_ID:
8151 			if (flow_dv_translate_action_port_id(dev, action,
8152 							     &port_id, error))
8153 				return -rte_errno;
8154 			port_id_resource.port_id = port_id;
8155 			MLX5_ASSERT(!handle->rix_port_id_action);
8156 			if (flow_dv_port_id_action_resource_register
8157 			    (dev, &port_id_resource, dev_flow, error))
8158 				return -rte_errno;
8159 			dev_flow->dv.actions[actions_n++] =
8160 					dev_flow->dv.port_id_action->action;
8161 			action_flags |= MLX5_FLOW_ACTION_PORT_ID;
8162 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_PORT_ID;
8163 			break;
8164 		case RTE_FLOW_ACTION_TYPE_FLAG:
8165 			action_flags |= MLX5_FLOW_ACTION_FLAG;
8166 			dev_flow->handle->mark = 1;
8167 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
8168 				struct rte_flow_action_mark mark = {
8169 					.id = MLX5_FLOW_MARK_DEFAULT,
8170 				};
8171 
8172 				if (flow_dv_convert_action_mark(dev, &mark,
8173 								mhdr_res,
8174 								error))
8175 					return -rte_errno;
8176 				action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
8177 				break;
8178 			}
8179 			tag_be = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT);
8180 			/*
8181 			 * Only one FLAG or MARK is supported per device flow
8182 			 * right now. So the pointer to the tag resource must be
8183 			 * zero before the register process.
8184 			 */
8185 			MLX5_ASSERT(!handle->dvh.rix_tag);
8186 			if (flow_dv_tag_resource_register(dev, tag_be,
8187 							  dev_flow, error))
8188 				return -rte_errno;
8189 			MLX5_ASSERT(dev_flow->dv.tag_resource);
8190 			dev_flow->dv.actions[actions_n++] =
8191 					dev_flow->dv.tag_resource->action;
8192 			break;
8193 		case RTE_FLOW_ACTION_TYPE_MARK:
8194 			action_flags |= MLX5_FLOW_ACTION_MARK;
8195 			dev_flow->handle->mark = 1;
8196 			if (dev_conf->dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
8197 				const struct rte_flow_action_mark *mark =
8198 					(const struct rte_flow_action_mark *)
8199 						actions->conf;
8200 
8201 				if (flow_dv_convert_action_mark(dev, mark,
8202 								mhdr_res,
8203 								error))
8204 					return -rte_errno;
8205 				action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
8206 				break;
8207 			}
8208 			/* Fall-through */
8209 		case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
8210 			/* Legacy (non-extensive) MARK action. */
8211 			tag_be = mlx5_flow_mark_set
8212 			      (((const struct rte_flow_action_mark *)
8213 			       (actions->conf))->id);
8214 			MLX5_ASSERT(!handle->dvh.rix_tag);
8215 			if (flow_dv_tag_resource_register(dev, tag_be,
8216 							  dev_flow, error))
8217 				return -rte_errno;
8218 			MLX5_ASSERT(dev_flow->dv.tag_resource);
8219 			dev_flow->dv.actions[actions_n++] =
8220 					dev_flow->dv.tag_resource->action;
8221 			break;
8222 		case RTE_FLOW_ACTION_TYPE_SET_META:
8223 			if (flow_dv_convert_action_set_meta
8224 				(dev, mhdr_res, attr,
8225 				 (const struct rte_flow_action_set_meta *)
8226 				  actions->conf, error))
8227 				return -rte_errno;
8228 			action_flags |= MLX5_FLOW_ACTION_SET_META;
8229 			break;
8230 		case RTE_FLOW_ACTION_TYPE_SET_TAG:
8231 			if (flow_dv_convert_action_set_tag
8232 				(dev, mhdr_res,
8233 				 (const struct rte_flow_action_set_tag *)
8234 				  actions->conf, error))
8235 				return -rte_errno;
8236 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8237 			break;
8238 		case RTE_FLOW_ACTION_TYPE_DROP:
8239 			action_flags |= MLX5_FLOW_ACTION_DROP;
8240 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
8241 			break;
8242 		case RTE_FLOW_ACTION_TYPE_QUEUE:
8243 			queue = actions->conf;
8244 			rss_desc->queue_num = 1;
8245 			rss_desc->queue[0] = queue->index;
8246 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
8247 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
8248 			break;
8249 		case RTE_FLOW_ACTION_TYPE_RSS:
8250 			rss = actions->conf;
8251 			memcpy(rss_desc->queue, rss->queue,
8252 			       rss->queue_num * sizeof(uint16_t));
8253 			rss_desc->queue_num = rss->queue_num;
8254 			/* NULL RSS key indicates default RSS key. */
8255 			rss_key = !rss->key ? rss_hash_default_key : rss->key;
8256 			memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
8257 			/*
8258 			 * rss->level and rss.types should be set in advance
8259 			 * when expanding items for RSS.
8260 			 */
8261 			action_flags |= MLX5_FLOW_ACTION_RSS;
8262 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
8263 			break;
8264 		case RTE_FLOW_ACTION_TYPE_AGE:
8265 		case RTE_FLOW_ACTION_TYPE_COUNT:
8266 			if (!dev_conf->devx) {
8267 				return rte_flow_error_set
8268 					      (error, ENOTSUP,
8269 					       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8270 					       NULL,
8271 					       "count action not supported");
8272 			}
8273 			/* Save information first, will apply later. */
8274 			if (actions->type == RTE_FLOW_ACTION_TYPE_COUNT)
8275 				count = action->conf;
8276 			else
8277 				age = action->conf;
8278 			action_flags |= MLX5_FLOW_ACTION_COUNT;
8279 			break;
8280 		case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
8281 			dev_flow->dv.actions[actions_n++] =
8282 						priv->sh->pop_vlan_action;
8283 			action_flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
8284 			break;
8285 		case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
8286 			if (!(action_flags &
8287 			      MLX5_FLOW_ACTION_OF_SET_VLAN_VID))
8288 				flow_dev_get_vlan_info_from_items(items, &vlan);
8289 			vlan.eth_proto = rte_be_to_cpu_16
8290 			     ((((const struct rte_flow_action_of_push_vlan *)
8291 						   actions->conf)->ethertype));
8292 			found_action = mlx5_flow_find_action
8293 					(actions + 1,
8294 					 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID);
8295 			if (found_action)
8296 				mlx5_update_vlan_vid_pcp(found_action, &vlan);
8297 			found_action = mlx5_flow_find_action
8298 					(actions + 1,
8299 					 RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP);
8300 			if (found_action)
8301 				mlx5_update_vlan_vid_pcp(found_action, &vlan);
8302 			if (flow_dv_create_action_push_vlan
8303 					    (dev, attr, &vlan, dev_flow, error))
8304 				return -rte_errno;
8305 			dev_flow->dv.actions[actions_n++] =
8306 					dev_flow->dv.push_vlan_res->action;
8307 			action_flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
8308 			break;
8309 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
8310 			/* of_vlan_push action handled this action */
8311 			MLX5_ASSERT(action_flags &
8312 				    MLX5_FLOW_ACTION_OF_PUSH_VLAN);
8313 			break;
8314 		case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
8315 			if (action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN)
8316 				break;
8317 			flow_dev_get_vlan_info_from_items(items, &vlan);
8318 			mlx5_update_vlan_vid_pcp(actions, &vlan);
8319 			/* If no VLAN push - this is a modify header action */
8320 			if (flow_dv_convert_action_modify_vlan_vid
8321 						(mhdr_res, actions, error))
8322 				return -rte_errno;
8323 			action_flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
8324 			break;
8325 		case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
8326 		case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
8327 			if (flow_dv_create_action_l2_encap(dev, actions,
8328 							   dev_flow,
8329 							   attr->transfer,
8330 							   error))
8331 				return -rte_errno;
8332 			dev_flow->dv.actions[actions_n++] =
8333 					dev_flow->dv.encap_decap->action;
8334 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
8335 			break;
8336 		case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
8337 		case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
8338 			if (flow_dv_create_action_l2_decap(dev, dev_flow,
8339 							   attr->transfer,
8340 							   error))
8341 				return -rte_errno;
8342 			dev_flow->dv.actions[actions_n++] =
8343 					dev_flow->dv.encap_decap->action;
8344 			action_flags |= MLX5_FLOW_ACTION_DECAP;
8345 			break;
8346 		case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
8347 			/* Handle encap with preceding decap. */
8348 			if (action_flags & MLX5_FLOW_ACTION_DECAP) {
8349 				if (flow_dv_create_action_raw_encap
8350 					(dev, actions, dev_flow, attr, error))
8351 					return -rte_errno;
8352 				dev_flow->dv.actions[actions_n++] =
8353 					dev_flow->dv.encap_decap->action;
8354 			} else {
8355 				/* Handle encap without preceding decap. */
8356 				if (flow_dv_create_action_l2_encap
8357 				    (dev, actions, dev_flow, attr->transfer,
8358 				     error))
8359 					return -rte_errno;
8360 				dev_flow->dv.actions[actions_n++] =
8361 					dev_flow->dv.encap_decap->action;
8362 			}
8363 			action_flags |= MLX5_FLOW_ACTION_ENCAP;
8364 			break;
8365 		case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
8366 			while ((++action)->type == RTE_FLOW_ACTION_TYPE_VOID)
8367 				;
8368 			if (action->type != RTE_FLOW_ACTION_TYPE_RAW_ENCAP) {
8369 				if (flow_dv_create_action_l2_decap
8370 				    (dev, dev_flow, attr->transfer, error))
8371 					return -rte_errno;
8372 				dev_flow->dv.actions[actions_n++] =
8373 					dev_flow->dv.encap_decap->action;
8374 			}
8375 			/* If decap is followed by encap, handle it at encap. */
8376 			action_flags |= MLX5_FLOW_ACTION_DECAP;
8377 			break;
8378 		case RTE_FLOW_ACTION_TYPE_JUMP:
8379 			jump_data = action->conf;
8380 			ret = mlx5_flow_group_to_table(attr, dev_flow->external,
8381 						       jump_data->group,
8382 						       !!priv->fdb_def_rule,
8383 						       &table, error);
8384 			if (ret)
8385 				return ret;
8386 			tbl = flow_dv_tbl_resource_get(dev, table,
8387 						       attr->egress,
8388 						       attr->transfer, error);
8389 			if (!tbl)
8390 				return rte_flow_error_set
8391 						(error, errno,
8392 						 RTE_FLOW_ERROR_TYPE_ACTION,
8393 						 NULL,
8394 						 "cannot create jump action.");
8395 			if (flow_dv_jump_tbl_resource_register
8396 			    (dev, tbl, dev_flow, error)) {
8397 				flow_dv_tbl_resource_release(dev, tbl);
8398 				return rte_flow_error_set
8399 						(error, errno,
8400 						 RTE_FLOW_ERROR_TYPE_ACTION,
8401 						 NULL,
8402 						 "cannot create jump action.");
8403 			}
8404 			dev_flow->dv.actions[actions_n++] =
8405 					dev_flow->dv.jump->action;
8406 			action_flags |= MLX5_FLOW_ACTION_JUMP;
8407 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_JUMP;
8408 			break;
8409 		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
8410 		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
8411 			if (flow_dv_convert_action_modify_mac
8412 					(mhdr_res, actions, error))
8413 				return -rte_errno;
8414 			action_flags |= actions->type ==
8415 					RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
8416 					MLX5_FLOW_ACTION_SET_MAC_SRC :
8417 					MLX5_FLOW_ACTION_SET_MAC_DST;
8418 			break;
8419 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
8420 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
8421 			if (flow_dv_convert_action_modify_ipv4
8422 					(mhdr_res, actions, error))
8423 				return -rte_errno;
8424 			action_flags |= actions->type ==
8425 					RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
8426 					MLX5_FLOW_ACTION_SET_IPV4_SRC :
8427 					MLX5_FLOW_ACTION_SET_IPV4_DST;
8428 			break;
8429 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
8430 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
8431 			if (flow_dv_convert_action_modify_ipv6
8432 					(mhdr_res, actions, error))
8433 				return -rte_errno;
8434 			action_flags |= actions->type ==
8435 					RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
8436 					MLX5_FLOW_ACTION_SET_IPV6_SRC :
8437 					MLX5_FLOW_ACTION_SET_IPV6_DST;
8438 			break;
8439 		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
8440 		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
8441 			if (flow_dv_convert_action_modify_tp
8442 					(mhdr_res, actions, items,
8443 					 &flow_attr, dev_flow, !!(action_flags &
8444 					 MLX5_FLOW_ACTION_DECAP), error))
8445 				return -rte_errno;
8446 			action_flags |= actions->type ==
8447 					RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
8448 					MLX5_FLOW_ACTION_SET_TP_SRC :
8449 					MLX5_FLOW_ACTION_SET_TP_DST;
8450 			break;
8451 		case RTE_FLOW_ACTION_TYPE_DEC_TTL:
8452 			if (flow_dv_convert_action_modify_dec_ttl
8453 					(mhdr_res, items, &flow_attr, dev_flow,
8454 					 !!(action_flags &
8455 					 MLX5_FLOW_ACTION_DECAP), error))
8456 				return -rte_errno;
8457 			action_flags |= MLX5_FLOW_ACTION_DEC_TTL;
8458 			break;
8459 		case RTE_FLOW_ACTION_TYPE_SET_TTL:
8460 			if (flow_dv_convert_action_modify_ttl
8461 					(mhdr_res, actions, items, &flow_attr,
8462 					 dev_flow, !!(action_flags &
8463 					 MLX5_FLOW_ACTION_DECAP), error))
8464 				return -rte_errno;
8465 			action_flags |= MLX5_FLOW_ACTION_SET_TTL;
8466 			break;
8467 		case RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ:
8468 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_SEQ:
8469 			if (flow_dv_convert_action_modify_tcp_seq
8470 					(mhdr_res, actions, error))
8471 				return -rte_errno;
8472 			action_flags |= actions->type ==
8473 					RTE_FLOW_ACTION_TYPE_INC_TCP_SEQ ?
8474 					MLX5_FLOW_ACTION_INC_TCP_SEQ :
8475 					MLX5_FLOW_ACTION_DEC_TCP_SEQ;
8476 			break;
8477 
8478 		case RTE_FLOW_ACTION_TYPE_INC_TCP_ACK:
8479 		case RTE_FLOW_ACTION_TYPE_DEC_TCP_ACK:
8480 			if (flow_dv_convert_action_modify_tcp_ack
8481 					(mhdr_res, actions, error))
8482 				return -rte_errno;
8483 			action_flags |= actions->type ==
8484 					RTE_FLOW_ACTION_TYPE_INC_TCP_ACK ?
8485 					MLX5_FLOW_ACTION_INC_TCP_ACK :
8486 					MLX5_FLOW_ACTION_DEC_TCP_ACK;
8487 			break;
8488 		case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
8489 			if (flow_dv_convert_action_set_reg
8490 					(mhdr_res, actions, error))
8491 				return -rte_errno;
8492 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8493 			break;
8494 		case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
8495 			if (flow_dv_convert_action_copy_mreg
8496 					(dev, mhdr_res, actions, error))
8497 				return -rte_errno;
8498 			action_flags |= MLX5_FLOW_ACTION_SET_TAG;
8499 			break;
8500 		case MLX5_RTE_FLOW_ACTION_TYPE_DEFAULT_MISS:
8501 			action_flags |= MLX5_FLOW_ACTION_DEFAULT_MISS;
8502 			dev_flow->handle->fate_action =
8503 					MLX5_FLOW_FATE_DEFAULT_MISS;
8504 			break;
8505 		case RTE_FLOW_ACTION_TYPE_METER:
8506 			mtr = actions->conf;
8507 			if (!flow->meter) {
8508 				fm = mlx5_flow_meter_attach(priv, mtr->mtr_id,
8509 							    attr, error);
8510 				if (!fm)
8511 					return rte_flow_error_set(error,
8512 						rte_errno,
8513 						RTE_FLOW_ERROR_TYPE_ACTION,
8514 						NULL,
8515 						"meter not found "
8516 						"or invalid parameters");
8517 				flow->meter = fm->idx;
8518 			}
8519 			/* Set the meter action. */
8520 			if (!fm) {
8521 				fm = mlx5_ipool_get(priv->sh->ipool
8522 						[MLX5_IPOOL_MTR], flow->meter);
8523 				if (!fm)
8524 					return rte_flow_error_set(error,
8525 						rte_errno,
8526 						RTE_FLOW_ERROR_TYPE_ACTION,
8527 						NULL,
8528 						"meter not found "
8529 						"or invalid parameters");
8530 			}
8531 			dev_flow->dv.actions[actions_n++] =
8532 				fm->mfts->meter_action;
8533 			action_flags |= MLX5_FLOW_ACTION_METER;
8534 			break;
8535 		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
8536 			if (flow_dv_convert_action_modify_ipv4_dscp(mhdr_res,
8537 							      actions, error))
8538 				return -rte_errno;
8539 			action_flags |= MLX5_FLOW_ACTION_SET_IPV4_DSCP;
8540 			break;
8541 		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
8542 			if (flow_dv_convert_action_modify_ipv6_dscp(mhdr_res,
8543 							      actions, error))
8544 				return -rte_errno;
8545 			action_flags |= MLX5_FLOW_ACTION_SET_IPV6_DSCP;
8546 			break;
8547 		case RTE_FLOW_ACTION_TYPE_END:
8548 			actions_end = true;
8549 			if (mhdr_res->actions_num) {
8550 				/* create modify action if needed. */
8551 				if (flow_dv_modify_hdr_resource_register
8552 					(dev, mhdr_res, dev_flow, error))
8553 					return -rte_errno;
8554 				dev_flow->dv.actions[modify_action_position] =
8555 					handle->dvh.modify_hdr->action;
8556 			}
8557 			if (action_flags & MLX5_FLOW_ACTION_COUNT) {
8558 				flow->counter =
8559 					flow_dv_translate_create_counter(dev,
8560 						dev_flow, count, age);
8561 
8562 				if (!flow->counter)
8563 					return rte_flow_error_set
8564 						(error, rte_errno,
8565 						RTE_FLOW_ERROR_TYPE_ACTION,
8566 						NULL,
8567 						"cannot create counter"
8568 						" object.");
8569 				dev_flow->dv.actions[actions_n++] =
8570 					  (flow_dv_counter_get_by_idx(dev,
8571 					  flow->counter, NULL))->action;
8572 			}
8573 			break;
8574 		default:
8575 			break;
8576 		}
8577 		if (mhdr_res->actions_num &&
8578 		    modify_action_position == UINT32_MAX)
8579 			modify_action_position = actions_n++;
8580 	}
8581 	dev_flow->dv.actions_n = actions_n;
8582 	dev_flow->act_flags = action_flags;
8583 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
8584 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
8585 		int item_type = items->type;
8586 
8587 		if (!mlx5_flow_os_item_supported(item_type))
8588 			return rte_flow_error_set(error, ENOTSUP,
8589 						  RTE_FLOW_ERROR_TYPE_ITEM,
8590 						  NULL, "item not supported");
8591 		switch (item_type) {
8592 		case RTE_FLOW_ITEM_TYPE_PORT_ID:
8593 			flow_dv_translate_item_port_id(dev, match_mask,
8594 						       match_value, items);
8595 			last_item = MLX5_FLOW_ITEM_PORT_ID;
8596 			break;
8597 		case RTE_FLOW_ITEM_TYPE_ETH:
8598 			flow_dv_translate_item_eth(match_mask, match_value,
8599 						   items, tunnel,
8600 						   dev_flow->dv.group);
8601 			matcher.priority = action_flags &
8602 					MLX5_FLOW_ACTION_DEFAULT_MISS &&
8603 					!dev_flow->external ?
8604 					MLX5_PRIORITY_MAP_L3 :
8605 					MLX5_PRIORITY_MAP_L2;
8606 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
8607 					     MLX5_FLOW_LAYER_OUTER_L2;
8608 			break;
8609 		case RTE_FLOW_ITEM_TYPE_VLAN:
8610 			flow_dv_translate_item_vlan(dev_flow,
8611 						    match_mask, match_value,
8612 						    items, tunnel,
8613 						    dev_flow->dv.group);
8614 			matcher.priority = MLX5_PRIORITY_MAP_L2;
8615 			last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
8616 					      MLX5_FLOW_LAYER_INNER_VLAN) :
8617 					     (MLX5_FLOW_LAYER_OUTER_L2 |
8618 					      MLX5_FLOW_LAYER_OUTER_VLAN);
8619 			break;
8620 		case RTE_FLOW_ITEM_TYPE_IPV4:
8621 			mlx5_flow_tunnel_ip_check(items, next_protocol,
8622 						  &item_flags, &tunnel);
8623 			flow_dv_translate_item_ipv4(match_mask, match_value,
8624 						    items, item_flags, tunnel,
8625 						    dev_flow->dv.group);
8626 			matcher.priority = MLX5_PRIORITY_MAP_L3;
8627 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
8628 					     MLX5_FLOW_LAYER_OUTER_L3_IPV4;
8629 			if (items->mask != NULL &&
8630 			    ((const struct rte_flow_item_ipv4 *)
8631 			     items->mask)->hdr.next_proto_id) {
8632 				next_protocol =
8633 					((const struct rte_flow_item_ipv4 *)
8634 					 (items->spec))->hdr.next_proto_id;
8635 				next_protocol &=
8636 					((const struct rte_flow_item_ipv4 *)
8637 					 (items->mask))->hdr.next_proto_id;
8638 			} else {
8639 				/* Reset for inner layer. */
8640 				next_protocol = 0xff;
8641 			}
8642 			break;
8643 		case RTE_FLOW_ITEM_TYPE_IPV6:
8644 			mlx5_flow_tunnel_ip_check(items, next_protocol,
8645 						  &item_flags, &tunnel);
8646 			flow_dv_translate_item_ipv6(match_mask, match_value,
8647 						    items, item_flags, tunnel,
8648 						    dev_flow->dv.group);
8649 			matcher.priority = MLX5_PRIORITY_MAP_L3;
8650 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
8651 					     MLX5_FLOW_LAYER_OUTER_L3_IPV6;
8652 			if (items->mask != NULL &&
8653 			    ((const struct rte_flow_item_ipv6 *)
8654 			     items->mask)->hdr.proto) {
8655 				next_protocol =
8656 					((const struct rte_flow_item_ipv6 *)
8657 					 items->spec)->hdr.proto;
8658 				next_protocol &=
8659 					((const struct rte_flow_item_ipv6 *)
8660 					 items->mask)->hdr.proto;
8661 			} else {
8662 				/* Reset for inner layer. */
8663 				next_protocol = 0xff;
8664 			}
8665 			break;
8666 		case RTE_FLOW_ITEM_TYPE_TCP:
8667 			flow_dv_translate_item_tcp(match_mask, match_value,
8668 						   items, tunnel);
8669 			matcher.priority = MLX5_PRIORITY_MAP_L4;
8670 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
8671 					     MLX5_FLOW_LAYER_OUTER_L4_TCP;
8672 			break;
8673 		case RTE_FLOW_ITEM_TYPE_UDP:
8674 			flow_dv_translate_item_udp(match_mask, match_value,
8675 						   items, tunnel);
8676 			matcher.priority = MLX5_PRIORITY_MAP_L4;
8677 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
8678 					     MLX5_FLOW_LAYER_OUTER_L4_UDP;
8679 			break;
8680 		case RTE_FLOW_ITEM_TYPE_GRE:
8681 			flow_dv_translate_item_gre(match_mask, match_value,
8682 						   items, tunnel);
8683 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8684 			last_item = MLX5_FLOW_LAYER_GRE;
8685 			break;
8686 		case RTE_FLOW_ITEM_TYPE_GRE_KEY:
8687 			flow_dv_translate_item_gre_key(match_mask,
8688 						       match_value, items);
8689 			last_item = MLX5_FLOW_LAYER_GRE_KEY;
8690 			break;
8691 		case RTE_FLOW_ITEM_TYPE_NVGRE:
8692 			flow_dv_translate_item_nvgre(match_mask, match_value,
8693 						     items, tunnel);
8694 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8695 			last_item = MLX5_FLOW_LAYER_GRE;
8696 			break;
8697 		case RTE_FLOW_ITEM_TYPE_VXLAN:
8698 			flow_dv_translate_item_vxlan(match_mask, match_value,
8699 						     items, tunnel);
8700 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8701 			last_item = MLX5_FLOW_LAYER_VXLAN;
8702 			break;
8703 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
8704 			flow_dv_translate_item_vxlan_gpe(match_mask,
8705 							 match_value, items,
8706 							 tunnel);
8707 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8708 			last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
8709 			break;
8710 		case RTE_FLOW_ITEM_TYPE_GENEVE:
8711 			flow_dv_translate_item_geneve(match_mask, match_value,
8712 						      items, tunnel);
8713 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8714 			last_item = MLX5_FLOW_LAYER_GENEVE;
8715 			break;
8716 		case RTE_FLOW_ITEM_TYPE_MPLS:
8717 			flow_dv_translate_item_mpls(match_mask, match_value,
8718 						    items, last_item, tunnel);
8719 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8720 			last_item = MLX5_FLOW_LAYER_MPLS;
8721 			break;
8722 		case RTE_FLOW_ITEM_TYPE_MARK:
8723 			flow_dv_translate_item_mark(dev, match_mask,
8724 						    match_value, items);
8725 			last_item = MLX5_FLOW_ITEM_MARK;
8726 			break;
8727 		case RTE_FLOW_ITEM_TYPE_META:
8728 			flow_dv_translate_item_meta(dev, match_mask,
8729 						    match_value, attr, items);
8730 			last_item = MLX5_FLOW_ITEM_METADATA;
8731 			break;
8732 		case RTE_FLOW_ITEM_TYPE_ICMP:
8733 			flow_dv_translate_item_icmp(match_mask, match_value,
8734 						    items, tunnel);
8735 			last_item = MLX5_FLOW_LAYER_ICMP;
8736 			break;
8737 		case RTE_FLOW_ITEM_TYPE_ICMP6:
8738 			flow_dv_translate_item_icmp6(match_mask, match_value,
8739 						      items, tunnel);
8740 			last_item = MLX5_FLOW_LAYER_ICMP6;
8741 			break;
8742 		case RTE_FLOW_ITEM_TYPE_TAG:
8743 			flow_dv_translate_item_tag(dev, match_mask,
8744 						   match_value, items);
8745 			last_item = MLX5_FLOW_ITEM_TAG;
8746 			break;
8747 		case MLX5_RTE_FLOW_ITEM_TYPE_TAG:
8748 			flow_dv_translate_mlx5_item_tag(dev, match_mask,
8749 							match_value, items);
8750 			last_item = MLX5_FLOW_ITEM_TAG;
8751 			break;
8752 		case MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE:
8753 			flow_dv_translate_item_tx_queue(dev, match_mask,
8754 							match_value,
8755 							items);
8756 			last_item = MLX5_FLOW_ITEM_TX_QUEUE;
8757 			break;
8758 		case RTE_FLOW_ITEM_TYPE_GTP:
8759 			flow_dv_translate_item_gtp(match_mask, match_value,
8760 						   items, tunnel);
8761 			matcher.priority = MLX5_TUNNEL_PRIO_GET(rss_desc);
8762 			last_item = MLX5_FLOW_LAYER_GTP;
8763 			break;
8764 		case RTE_FLOW_ITEM_TYPE_ECPRI:
8765 			if (!mlx5_flex_parser_ecpri_exist(dev)) {
8766 				/* Create it only the first time to be used. */
8767 				ret = mlx5_flex_parser_ecpri_alloc(dev);
8768 				if (ret)
8769 					return rte_flow_error_set
8770 						(error, -ret,
8771 						RTE_FLOW_ERROR_TYPE_ITEM,
8772 						NULL,
8773 						"cannot create eCPRI parser");
8774 			}
8775 			/* Adjust the length matcher and device flow value. */
8776 			matcher.mask.size = MLX5_ST_SZ_BYTES(fte_match_param);
8777 			dev_flow->dv.value.size =
8778 					MLX5_ST_SZ_BYTES(fte_match_param);
8779 			flow_dv_translate_item_ecpri(dev, match_mask,
8780 						     match_value, items);
8781 			/* No other protocol should follow eCPRI layer. */
8782 			last_item = MLX5_FLOW_LAYER_ECPRI;
8783 			break;
8784 		default:
8785 			break;
8786 		}
8787 		item_flags |= last_item;
8788 	}
8789 	/*
8790 	 * When E-Switch mode is enabled, we have two cases where we need to
8791 	 * set the source port manually.
8792 	 * The first one, is in case of Nic steering rule, and the second is
8793 	 * E-Switch rule where no port_id item was found. In both cases
8794 	 * the source port is set according the current port in use.
8795 	 */
8796 	if (!(item_flags & MLX5_FLOW_ITEM_PORT_ID) &&
8797 	    (priv->representor || priv->master)) {
8798 		if (flow_dv_translate_item_port_id(dev, match_mask,
8799 						   match_value, NULL))
8800 			return -rte_errno;
8801 	}
8802 #ifdef RTE_LIBRTE_MLX5_DEBUG
8803 	MLX5_ASSERT(!flow_dv_check_valid_spec(matcher.mask.buf,
8804 					      dev_flow->dv.value.buf));
8805 #endif
8806 	/*
8807 	 * Layers may be already initialized from prefix flow if this dev_flow
8808 	 * is the suffix flow.
8809 	 */
8810 	handle->layers |= item_flags;
8811 	if (action_flags & MLX5_FLOW_ACTION_RSS)
8812 		flow_dv_hashfields_set(dev_flow, rss_desc);
8813 	/* Register matcher. */
8814 	matcher.crc = rte_raw_cksum((const void *)matcher.mask.buf,
8815 				    matcher.mask.size);
8816 	matcher.priority = mlx5_flow_adjust_priority(dev, priority,
8817 						     matcher.priority);
8818 	/* reserved field no needs to be set to 0 here. */
8819 	tbl_key.domain = attr->transfer;
8820 	tbl_key.direction = attr->egress;
8821 	tbl_key.table_id = dev_flow->dv.group;
8822 	if (flow_dv_matcher_register(dev, &matcher, &tbl_key, dev_flow, error))
8823 		return -rte_errno;
8824 	return 0;
8825 }
8826 
8827 /**
8828  * Apply the flow to the NIC, lock free,
8829  * (mutex should be acquired by caller).
8830  *
8831  * @param[in] dev
8832  *   Pointer to the Ethernet device structure.
8833  * @param[in, out] flow
8834  *   Pointer to flow structure.
8835  * @param[out] error
8836  *   Pointer to error structure.
8837  *
8838  * @return
8839  *   0 on success, a negative errno value otherwise and rte_errno is set.
8840  */
8841 static int
8842 __flow_dv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
8843 		struct rte_flow_error *error)
8844 {
8845 	struct mlx5_flow_dv_workspace *dv;
8846 	struct mlx5_flow_handle *dh;
8847 	struct mlx5_flow_handle_dv *dv_h;
8848 	struct mlx5_flow *dev_flow;
8849 	struct mlx5_priv *priv = dev->data->dev_private;
8850 	uint32_t handle_idx;
8851 	int n;
8852 	int err;
8853 	int idx;
8854 
8855 	for (idx = priv->flow_idx - 1; idx >= priv->flow_nested_idx; idx--) {
8856 		dev_flow = &((struct mlx5_flow *)priv->inter_flows)[idx];
8857 		dv = &dev_flow->dv;
8858 		dh = dev_flow->handle;
8859 		dv_h = &dh->dvh;
8860 		n = dv->actions_n;
8861 		if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8862 			if (dv->transfer) {
8863 				dv->actions[n++] = priv->sh->esw_drop_action;
8864 			} else {
8865 				struct mlx5_hrxq *drop_hrxq;
8866 				drop_hrxq = mlx5_hrxq_drop_new(dev);
8867 				if (!drop_hrxq) {
8868 					rte_flow_error_set
8869 						(error, errno,
8870 						 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8871 						 NULL,
8872 						 "cannot get drop hash queue");
8873 					goto error;
8874 				}
8875 				/*
8876 				 * Drop queues will be released by the specify
8877 				 * mlx5_hrxq_drop_release() function. Assign
8878 				 * the special index to hrxq to mark the queue
8879 				 * has been allocated.
8880 				 */
8881 				dh->rix_hrxq = UINT32_MAX;
8882 				dv->actions[n++] = drop_hrxq->action;
8883 			}
8884 		} else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8885 			struct mlx5_hrxq *hrxq;
8886 			uint32_t hrxq_idx;
8887 			struct mlx5_flow_rss_desc *rss_desc =
8888 				&((struct mlx5_flow_rss_desc *)priv->rss_desc)
8889 				[!!priv->flow_nested_idx];
8890 
8891 			MLX5_ASSERT(rss_desc->queue_num);
8892 			hrxq_idx = mlx5_hrxq_get(dev, rss_desc->key,
8893 						 MLX5_RSS_HASH_KEY_LEN,
8894 						 dev_flow->hash_fields,
8895 						 rss_desc->queue,
8896 						 rss_desc->queue_num);
8897 			if (!hrxq_idx) {
8898 				hrxq_idx = mlx5_hrxq_new
8899 						(dev, rss_desc->key,
8900 						MLX5_RSS_HASH_KEY_LEN,
8901 						dev_flow->hash_fields,
8902 						rss_desc->queue,
8903 						rss_desc->queue_num,
8904 						!!(dh->layers &
8905 						MLX5_FLOW_LAYER_TUNNEL));
8906 			}
8907 			hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
8908 					      hrxq_idx);
8909 			if (!hrxq) {
8910 				rte_flow_error_set
8911 					(error, rte_errno,
8912 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8913 					 "cannot get hash queue");
8914 				goto error;
8915 			}
8916 			dh->rix_hrxq = hrxq_idx;
8917 			dv->actions[n++] = hrxq->action;
8918 		} else if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS) {
8919 			if (flow_dv_default_miss_resource_register
8920 					(dev, error)) {
8921 				rte_flow_error_set
8922 					(error, rte_errno,
8923 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
8924 					 "cannot create default miss resource");
8925 				goto error_default_miss;
8926 			}
8927 			dh->rix_default_fate =  MLX5_FLOW_FATE_DEFAULT_MISS;
8928 			dv->actions[n++] = priv->sh->default_miss.action;
8929 		}
8930 		err = mlx5_flow_os_create_flow(dv_h->matcher->matcher_object,
8931 					       (void *)&dv->value, n,
8932 					       dv->actions, &dh->drv_flow);
8933 		if (err) {
8934 			rte_flow_error_set(error, errno,
8935 					   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
8936 					   NULL,
8937 					   "hardware refuses to create flow");
8938 			goto error;
8939 		}
8940 		if (priv->vmwa_context &&
8941 		    dh->vf_vlan.tag && !dh->vf_vlan.created) {
8942 			/*
8943 			 * The rule contains the VLAN pattern.
8944 			 * For VF we are going to create VLAN
8945 			 * interface to make hypervisor set correct
8946 			 * e-Switch vport context.
8947 			 */
8948 			mlx5_vlan_vmwa_acquire(dev, &dh->vf_vlan);
8949 		}
8950 	}
8951 	return 0;
8952 error:
8953 	if (dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
8954 		flow_dv_default_miss_resource_release(dev);
8955 error_default_miss:
8956 	err = rte_errno; /* Save rte_errno before cleanup. */
8957 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
8958 		       handle_idx, dh, next) {
8959 		/* hrxq is union, don't clear it if the flag is not set. */
8960 		if (dh->rix_hrxq) {
8961 			if (dh->fate_action == MLX5_FLOW_FATE_DROP) {
8962 				mlx5_hrxq_drop_release(dev);
8963 				dh->rix_hrxq = 0;
8964 			} else if (dh->fate_action == MLX5_FLOW_FATE_QUEUE) {
8965 				mlx5_hrxq_release(dev, dh->rix_hrxq);
8966 				dh->rix_hrxq = 0;
8967 			}
8968 		}
8969 		if (dh->vf_vlan.tag && dh->vf_vlan.created)
8970 			mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
8971 	}
8972 	rte_errno = err; /* Restore rte_errno. */
8973 	return -rte_errno;
8974 }
8975 
8976 /**
8977  * Release the flow matcher.
8978  *
8979  * @param dev
8980  *   Pointer to Ethernet device.
8981  * @param handle
8982  *   Pointer to mlx5_flow_handle.
8983  *
8984  * @return
8985  *   1 while a reference on it exists, 0 when freed.
8986  */
8987 static int
8988 flow_dv_matcher_release(struct rte_eth_dev *dev,
8989 			struct mlx5_flow_handle *handle)
8990 {
8991 	struct mlx5_flow_dv_matcher *matcher = handle->dvh.matcher;
8992 
8993 	MLX5_ASSERT(matcher->matcher_object);
8994 	DRV_LOG(DEBUG, "port %u matcher %p: refcnt %d--",
8995 		dev->data->port_id, (void *)matcher,
8996 		rte_atomic32_read(&matcher->refcnt));
8997 	if (rte_atomic32_dec_and_test(&matcher->refcnt)) {
8998 		claim_zero(mlx5_flow_os_destroy_flow_matcher
8999 			   (matcher->matcher_object));
9000 		LIST_REMOVE(matcher, next);
9001 		/* table ref-- in release interface. */
9002 		flow_dv_tbl_resource_release(dev, matcher->tbl);
9003 		mlx5_free(matcher);
9004 		DRV_LOG(DEBUG, "port %u matcher %p: removed",
9005 			dev->data->port_id, (void *)matcher);
9006 		return 0;
9007 	}
9008 	return 1;
9009 }
9010 
9011 /**
9012  * Release an encap/decap resource.
9013  *
9014  * @param dev
9015  *   Pointer to Ethernet device.
9016  * @param handle
9017  *   Pointer to mlx5_flow_handle.
9018  *
9019  * @return
9020  *   1 while a reference on it exists, 0 when freed.
9021  */
9022 static int
9023 flow_dv_encap_decap_resource_release(struct rte_eth_dev *dev,
9024 				     struct mlx5_flow_handle *handle)
9025 {
9026 	struct mlx5_priv *priv = dev->data->dev_private;
9027 	uint32_t idx = handle->dvh.rix_encap_decap;
9028 	struct mlx5_flow_dv_encap_decap_resource *cache_resource;
9029 
9030 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
9031 			 idx);
9032 	if (!cache_resource)
9033 		return 0;
9034 	MLX5_ASSERT(cache_resource->action);
9035 	DRV_LOG(DEBUG, "encap/decap resource %p: refcnt %d--",
9036 		(void *)cache_resource,
9037 		rte_atomic32_read(&cache_resource->refcnt));
9038 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9039 		claim_zero(mlx5_flow_os_destroy_flow_action
9040 						(cache_resource->action));
9041 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP],
9042 			     &priv->sh->encaps_decaps, idx,
9043 			     cache_resource, next);
9044 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_DECAP_ENCAP], idx);
9045 		DRV_LOG(DEBUG, "encap/decap resource %p: removed",
9046 			(void *)cache_resource);
9047 		return 0;
9048 	}
9049 	return 1;
9050 }
9051 
9052 /**
9053  * Release an jump to table action resource.
9054  *
9055  * @param dev
9056  *   Pointer to Ethernet device.
9057  * @param handle
9058  *   Pointer to mlx5_flow_handle.
9059  *
9060  * @return
9061  *   1 while a reference on it exists, 0 when freed.
9062  */
9063 static int
9064 flow_dv_jump_tbl_resource_release(struct rte_eth_dev *dev,
9065 				  struct mlx5_flow_handle *handle)
9066 {
9067 	struct mlx5_priv *priv = dev->data->dev_private;
9068 	struct mlx5_flow_dv_jump_tbl_resource *cache_resource;
9069 	struct mlx5_flow_tbl_data_entry *tbl_data;
9070 
9071 	tbl_data = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_JUMP],
9072 			     handle->rix_jump);
9073 	if (!tbl_data)
9074 		return 0;
9075 	cache_resource = &tbl_data->jump;
9076 	MLX5_ASSERT(cache_resource->action);
9077 	DRV_LOG(DEBUG, "jump table resource %p: refcnt %d--",
9078 		(void *)cache_resource,
9079 		rte_atomic32_read(&cache_resource->refcnt));
9080 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9081 		claim_zero(mlx5_flow_os_destroy_flow_action
9082 						(cache_resource->action));
9083 		/* jump action memory free is inside the table release. */
9084 		flow_dv_tbl_resource_release(dev, &tbl_data->tbl);
9085 		DRV_LOG(DEBUG, "jump table resource %p: removed",
9086 			(void *)cache_resource);
9087 		return 0;
9088 	}
9089 	return 1;
9090 }
9091 
9092 /**
9093  * Release a default miss resource.
9094  *
9095  * @param dev
9096  *   Pointer to Ethernet device.
9097  * @return
9098  *   1 while a reference on it exists, 0 when freed.
9099  */
9100 static int
9101 flow_dv_default_miss_resource_release(struct rte_eth_dev *dev)
9102 {
9103 	struct mlx5_priv *priv = dev->data->dev_private;
9104 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9105 	struct mlx5_flow_default_miss_resource *cache_resource =
9106 			&sh->default_miss;
9107 
9108 	MLX5_ASSERT(cache_resource->action);
9109 	DRV_LOG(DEBUG, "default miss resource %p: refcnt %d--",
9110 			(void *)cache_resource->action,
9111 			rte_atomic32_read(&cache_resource->refcnt));
9112 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9113 		claim_zero(mlx5_glue->destroy_flow_action
9114 				(cache_resource->action));
9115 		DRV_LOG(DEBUG, "default miss resource %p: removed",
9116 				(void *)cache_resource->action);
9117 		return 0;
9118 	}
9119 	return 1;
9120 }
9121 
9122 /**
9123  * Release a modify-header resource.
9124  *
9125  * @param handle
9126  *   Pointer to mlx5_flow_handle.
9127  *
9128  * @return
9129  *   1 while a reference on it exists, 0 when freed.
9130  */
9131 static int
9132 flow_dv_modify_hdr_resource_release(struct mlx5_flow_handle *handle)
9133 {
9134 	struct mlx5_flow_dv_modify_hdr_resource *cache_resource =
9135 							handle->dvh.modify_hdr;
9136 
9137 	MLX5_ASSERT(cache_resource->action);
9138 	DRV_LOG(DEBUG, "modify-header resource %p: refcnt %d--",
9139 		(void *)cache_resource,
9140 		rte_atomic32_read(&cache_resource->refcnt));
9141 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9142 		claim_zero(mlx5_flow_os_destroy_flow_action
9143 						(cache_resource->action));
9144 		LIST_REMOVE(cache_resource, next);
9145 		mlx5_free(cache_resource);
9146 		DRV_LOG(DEBUG, "modify-header resource %p: removed",
9147 			(void *)cache_resource);
9148 		return 0;
9149 	}
9150 	return 1;
9151 }
9152 
9153 /**
9154  * Release port ID action resource.
9155  *
9156  * @param dev
9157  *   Pointer to Ethernet device.
9158  * @param handle
9159  *   Pointer to mlx5_flow_handle.
9160  *
9161  * @return
9162  *   1 while a reference on it exists, 0 when freed.
9163  */
9164 static int
9165 flow_dv_port_id_action_resource_release(struct rte_eth_dev *dev,
9166 					struct mlx5_flow_handle *handle)
9167 {
9168 	struct mlx5_priv *priv = dev->data->dev_private;
9169 	struct mlx5_flow_dv_port_id_action_resource *cache_resource;
9170 	uint32_t idx = handle->rix_port_id_action;
9171 
9172 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
9173 					idx);
9174 	if (!cache_resource)
9175 		return 0;
9176 	MLX5_ASSERT(cache_resource->action);
9177 	DRV_LOG(DEBUG, "port ID action resource %p: refcnt %d--",
9178 		(void *)cache_resource,
9179 		rte_atomic32_read(&cache_resource->refcnt));
9180 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9181 		claim_zero(mlx5_flow_os_destroy_flow_action
9182 						(cache_resource->action));
9183 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PORT_ID],
9184 			     &priv->sh->port_id_action_list, idx,
9185 			     cache_resource, next);
9186 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PORT_ID], idx);
9187 		DRV_LOG(DEBUG, "port id action resource %p: removed",
9188 			(void *)cache_resource);
9189 		return 0;
9190 	}
9191 	return 1;
9192 }
9193 
9194 /**
9195  * Release push vlan action resource.
9196  *
9197  * @param dev
9198  *   Pointer to Ethernet device.
9199  * @param handle
9200  *   Pointer to mlx5_flow_handle.
9201  *
9202  * @return
9203  *   1 while a reference on it exists, 0 when freed.
9204  */
9205 static int
9206 flow_dv_push_vlan_action_resource_release(struct rte_eth_dev *dev,
9207 					  struct mlx5_flow_handle *handle)
9208 {
9209 	struct mlx5_priv *priv = dev->data->dev_private;
9210 	uint32_t idx = handle->dvh.rix_push_vlan;
9211 	struct mlx5_flow_dv_push_vlan_action_resource *cache_resource;
9212 
9213 	cache_resource = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
9214 					idx);
9215 	if (!cache_resource)
9216 		return 0;
9217 	MLX5_ASSERT(cache_resource->action);
9218 	DRV_LOG(DEBUG, "push VLAN action resource %p: refcnt %d--",
9219 		(void *)cache_resource,
9220 		rte_atomic32_read(&cache_resource->refcnt));
9221 	if (rte_atomic32_dec_and_test(&cache_resource->refcnt)) {
9222 		claim_zero(mlx5_flow_os_destroy_flow_action
9223 						(cache_resource->action));
9224 		ILIST_REMOVE(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN],
9225 			     &priv->sh->push_vlan_action_list, idx,
9226 			     cache_resource, next);
9227 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_PUSH_VLAN], idx);
9228 		DRV_LOG(DEBUG, "push vlan action resource %p: removed",
9229 			(void *)cache_resource);
9230 		return 0;
9231 	}
9232 	return 1;
9233 }
9234 
9235 /**
9236  * Release the fate resource.
9237  *
9238  * @param dev
9239  *   Pointer to Ethernet device.
9240  * @param handle
9241  *   Pointer to mlx5_flow_handle.
9242  */
9243 static void
9244 flow_dv_fate_resource_release(struct rte_eth_dev *dev,
9245 			       struct mlx5_flow_handle *handle)
9246 {
9247 	if (!handle->rix_fate)
9248 		return;
9249 	switch (handle->fate_action) {
9250 	case MLX5_FLOW_FATE_DROP:
9251 		mlx5_hrxq_drop_release(dev);
9252 		break;
9253 	case MLX5_FLOW_FATE_QUEUE:
9254 		mlx5_hrxq_release(dev, handle->rix_hrxq);
9255 		break;
9256 	case MLX5_FLOW_FATE_JUMP:
9257 		flow_dv_jump_tbl_resource_release(dev, handle);
9258 		break;
9259 	case MLX5_FLOW_FATE_PORT_ID:
9260 		flow_dv_port_id_action_resource_release(dev, handle);
9261 		break;
9262 	case MLX5_FLOW_FATE_DEFAULT_MISS:
9263 		flow_dv_default_miss_resource_release(dev);
9264 		break;
9265 	default:
9266 		DRV_LOG(DEBUG, "Incorrect fate action:%d", handle->fate_action);
9267 		break;
9268 	}
9269 	handle->rix_fate = 0;
9270 }
9271 
9272 /**
9273  * Remove the flow from the NIC but keeps it in memory.
9274  * Lock free, (mutex should be acquired by caller).
9275  *
9276  * @param[in] dev
9277  *   Pointer to Ethernet device.
9278  * @param[in, out] flow
9279  *   Pointer to flow structure.
9280  */
9281 static void
9282 __flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
9283 {
9284 	struct mlx5_flow_handle *dh;
9285 	uint32_t handle_idx;
9286 	struct mlx5_priv *priv = dev->data->dev_private;
9287 
9288 	if (!flow)
9289 		return;
9290 	handle_idx = flow->dev_handles;
9291 	while (handle_idx) {
9292 		dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
9293 				    handle_idx);
9294 		if (!dh)
9295 			return;
9296 		if (dh->drv_flow) {
9297 			claim_zero(mlx5_flow_os_destroy_flow(dh->drv_flow));
9298 			dh->drv_flow = NULL;
9299 		}
9300 		if (dh->fate_action == MLX5_FLOW_FATE_DROP ||
9301 		    dh->fate_action == MLX5_FLOW_FATE_QUEUE ||
9302 		    dh->fate_action == MLX5_FLOW_FATE_DEFAULT_MISS)
9303 			flow_dv_fate_resource_release(dev, dh);
9304 		if (dh->vf_vlan.tag && dh->vf_vlan.created)
9305 			mlx5_vlan_vmwa_release(dev, &dh->vf_vlan);
9306 		handle_idx = dh->next.next;
9307 	}
9308 }
9309 
9310 /**
9311  * Remove the flow from the NIC and the memory.
9312  * Lock free, (mutex should be acquired by caller).
9313  *
9314  * @param[in] dev
9315  *   Pointer to the Ethernet device structure.
9316  * @param[in, out] flow
9317  *   Pointer to flow structure.
9318  */
9319 static void
9320 __flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
9321 {
9322 	struct mlx5_flow_handle *dev_handle;
9323 	struct mlx5_priv *priv = dev->data->dev_private;
9324 
9325 	if (!flow)
9326 		return;
9327 	__flow_dv_remove(dev, flow);
9328 	if (flow->counter) {
9329 		flow_dv_counter_release(dev, flow->counter);
9330 		flow->counter = 0;
9331 	}
9332 	if (flow->meter) {
9333 		struct mlx5_flow_meter *fm;
9334 
9335 		fm = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MTR],
9336 				    flow->meter);
9337 		if (fm)
9338 			mlx5_flow_meter_detach(fm);
9339 		flow->meter = 0;
9340 	}
9341 	while (flow->dev_handles) {
9342 		uint32_t tmp_idx = flow->dev_handles;
9343 
9344 		dev_handle = mlx5_ipool_get(priv->sh->ipool
9345 					    [MLX5_IPOOL_MLX5_FLOW], tmp_idx);
9346 		if (!dev_handle)
9347 			return;
9348 		flow->dev_handles = dev_handle->next.next;
9349 		if (dev_handle->dvh.matcher)
9350 			flow_dv_matcher_release(dev, dev_handle);
9351 		if (dev_handle->dvh.rix_encap_decap)
9352 			flow_dv_encap_decap_resource_release(dev, dev_handle);
9353 		if (dev_handle->dvh.modify_hdr)
9354 			flow_dv_modify_hdr_resource_release(dev_handle);
9355 		if (dev_handle->dvh.rix_push_vlan)
9356 			flow_dv_push_vlan_action_resource_release(dev,
9357 								  dev_handle);
9358 		if (dev_handle->dvh.rix_tag)
9359 			flow_dv_tag_release(dev,
9360 					    dev_handle->dvh.rix_tag);
9361 		flow_dv_fate_resource_release(dev, dev_handle);
9362 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
9363 			   tmp_idx);
9364 	}
9365 }
9366 
9367 /**
9368  * Query a dv flow  rule for its statistics via devx.
9369  *
9370  * @param[in] dev
9371  *   Pointer to Ethernet device.
9372  * @param[in] flow
9373  *   Pointer to the sub flow.
9374  * @param[out] data
9375  *   data retrieved by the query.
9376  * @param[out] error
9377  *   Perform verbose error reporting if not NULL.
9378  *
9379  * @return
9380  *   0 on success, a negative errno value otherwise and rte_errno is set.
9381  */
9382 static int
9383 flow_dv_query_count(struct rte_eth_dev *dev, struct rte_flow *flow,
9384 		    void *data, struct rte_flow_error *error)
9385 {
9386 	struct mlx5_priv *priv = dev->data->dev_private;
9387 	struct rte_flow_query_count *qc = data;
9388 
9389 	if (!priv->config.devx)
9390 		return rte_flow_error_set(error, ENOTSUP,
9391 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9392 					  NULL,
9393 					  "counters are not supported");
9394 	if (flow->counter) {
9395 		uint64_t pkts, bytes;
9396 		struct mlx5_flow_counter *cnt;
9397 
9398 		cnt = flow_dv_counter_get_by_idx(dev, flow->counter,
9399 						 NULL);
9400 		int err = _flow_dv_query_count(dev, flow->counter, &pkts,
9401 					       &bytes);
9402 
9403 		if (err)
9404 			return rte_flow_error_set(error, -err,
9405 					RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9406 					NULL, "cannot read counters");
9407 		qc->hits_set = 1;
9408 		qc->bytes_set = 1;
9409 		qc->hits = pkts - cnt->hits;
9410 		qc->bytes = bytes - cnt->bytes;
9411 		if (qc->reset) {
9412 			cnt->hits = pkts;
9413 			cnt->bytes = bytes;
9414 		}
9415 		return 0;
9416 	}
9417 	return rte_flow_error_set(error, EINVAL,
9418 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9419 				  NULL,
9420 				  "counters are not available");
9421 }
9422 
9423 /**
9424  * Query a flow.
9425  *
9426  * @see rte_flow_query()
9427  * @see rte_flow_ops
9428  */
9429 static int
9430 flow_dv_query(struct rte_eth_dev *dev,
9431 	      struct rte_flow *flow __rte_unused,
9432 	      const struct rte_flow_action *actions __rte_unused,
9433 	      void *data __rte_unused,
9434 	      struct rte_flow_error *error __rte_unused)
9435 {
9436 	int ret = -EINVAL;
9437 
9438 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
9439 		switch (actions->type) {
9440 		case RTE_FLOW_ACTION_TYPE_VOID:
9441 			break;
9442 		case RTE_FLOW_ACTION_TYPE_COUNT:
9443 			ret = flow_dv_query_count(dev, flow, data, error);
9444 			break;
9445 		default:
9446 			return rte_flow_error_set(error, ENOTSUP,
9447 						  RTE_FLOW_ERROR_TYPE_ACTION,
9448 						  actions,
9449 						  "action not supported");
9450 		}
9451 	}
9452 	return ret;
9453 }
9454 
9455 /**
9456  * Destroy the meter table set.
9457  * Lock free, (mutex should be acquired by caller).
9458  *
9459  * @param[in] dev
9460  *   Pointer to Ethernet device.
9461  * @param[in] tbl
9462  *   Pointer to the meter table set.
9463  *
9464  * @return
9465  *   Always 0.
9466  */
9467 static int
9468 flow_dv_destroy_mtr_tbl(struct rte_eth_dev *dev,
9469 			struct mlx5_meter_domains_infos *tbl)
9470 {
9471 	struct mlx5_priv *priv = dev->data->dev_private;
9472 	struct mlx5_meter_domains_infos *mtd =
9473 				(struct mlx5_meter_domains_infos *)tbl;
9474 
9475 	if (!mtd || !priv->config.dv_flow_en)
9476 		return 0;
9477 	if (mtd->ingress.policer_rules[RTE_MTR_DROPPED])
9478 		claim_zero(mlx5_flow_os_destroy_flow
9479 			   (mtd->ingress.policer_rules[RTE_MTR_DROPPED]));
9480 	if (mtd->egress.policer_rules[RTE_MTR_DROPPED])
9481 		claim_zero(mlx5_flow_os_destroy_flow
9482 			   (mtd->egress.policer_rules[RTE_MTR_DROPPED]));
9483 	if (mtd->transfer.policer_rules[RTE_MTR_DROPPED])
9484 		claim_zero(mlx5_flow_os_destroy_flow
9485 			   (mtd->transfer.policer_rules[RTE_MTR_DROPPED]));
9486 	if (mtd->egress.color_matcher)
9487 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9488 			   (mtd->egress.color_matcher));
9489 	if (mtd->egress.any_matcher)
9490 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9491 			   (mtd->egress.any_matcher));
9492 	if (mtd->egress.tbl)
9493 		flow_dv_tbl_resource_release(dev, mtd->egress.tbl);
9494 	if (mtd->egress.sfx_tbl)
9495 		flow_dv_tbl_resource_release(dev, mtd->egress.sfx_tbl);
9496 	if (mtd->ingress.color_matcher)
9497 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9498 			   (mtd->ingress.color_matcher));
9499 	if (mtd->ingress.any_matcher)
9500 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9501 			   (mtd->ingress.any_matcher));
9502 	if (mtd->ingress.tbl)
9503 		flow_dv_tbl_resource_release(dev, mtd->ingress.tbl);
9504 	if (mtd->ingress.sfx_tbl)
9505 		flow_dv_tbl_resource_release(dev, mtd->ingress.sfx_tbl);
9506 	if (mtd->transfer.color_matcher)
9507 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9508 			   (mtd->transfer.color_matcher));
9509 	if (mtd->transfer.any_matcher)
9510 		claim_zero(mlx5_flow_os_destroy_flow_matcher
9511 			   (mtd->transfer.any_matcher));
9512 	if (mtd->transfer.tbl)
9513 		flow_dv_tbl_resource_release(dev, mtd->transfer.tbl);
9514 	if (mtd->transfer.sfx_tbl)
9515 		flow_dv_tbl_resource_release(dev, mtd->transfer.sfx_tbl);
9516 	if (mtd->drop_actn)
9517 		claim_zero(mlx5_flow_os_destroy_flow_action(mtd->drop_actn));
9518 	mlx5_free(mtd);
9519 	return 0;
9520 }
9521 
9522 /* Number of meter flow actions, count and jump or count and drop. */
9523 #define METER_ACTIONS 2
9524 
9525 /**
9526  * Create specify domain meter table and suffix table.
9527  *
9528  * @param[in] dev
9529  *   Pointer to Ethernet device.
9530  * @param[in,out] mtb
9531  *   Pointer to DV meter table set.
9532  * @param[in] egress
9533  *   Table attribute.
9534  * @param[in] transfer
9535  *   Table attribute.
9536  * @param[in] color_reg_c_idx
9537  *   Reg C index for color match.
9538  *
9539  * @return
9540  *   0 on success, -1 otherwise and rte_errno is set.
9541  */
9542 static int
9543 flow_dv_prepare_mtr_tables(struct rte_eth_dev *dev,
9544 			   struct mlx5_meter_domains_infos *mtb,
9545 			   uint8_t egress, uint8_t transfer,
9546 			   uint32_t color_reg_c_idx)
9547 {
9548 	struct mlx5_priv *priv = dev->data->dev_private;
9549 	struct mlx5_dev_ctx_shared *sh = priv->sh;
9550 	struct mlx5_flow_dv_match_params mask = {
9551 		.size = sizeof(mask.buf),
9552 	};
9553 	struct mlx5_flow_dv_match_params value = {
9554 		.size = sizeof(value.buf),
9555 	};
9556 	struct mlx5dv_flow_matcher_attr dv_attr = {
9557 		.type = IBV_FLOW_ATTR_NORMAL,
9558 		.priority = 0,
9559 		.match_criteria_enable = 0,
9560 		.match_mask = (void *)&mask,
9561 	};
9562 	void *actions[METER_ACTIONS];
9563 	struct mlx5_meter_domain_info *dtb;
9564 	struct rte_flow_error error;
9565 	int i = 0;
9566 	int ret;
9567 
9568 	if (transfer)
9569 		dtb = &mtb->transfer;
9570 	else if (egress)
9571 		dtb = &mtb->egress;
9572 	else
9573 		dtb = &mtb->ingress;
9574 	/* Create the meter table with METER level. */
9575 	dtb->tbl = flow_dv_tbl_resource_get(dev, MLX5_FLOW_TABLE_LEVEL_METER,
9576 					    egress, transfer, &error);
9577 	if (!dtb->tbl) {
9578 		DRV_LOG(ERR, "Failed to create meter policer table.");
9579 		return -1;
9580 	}
9581 	/* Create the meter suffix table with SUFFIX level. */
9582 	dtb->sfx_tbl = flow_dv_tbl_resource_get(dev,
9583 					    MLX5_FLOW_TABLE_LEVEL_SUFFIX,
9584 					    egress, transfer, &error);
9585 	if (!dtb->sfx_tbl) {
9586 		DRV_LOG(ERR, "Failed to create meter suffix table.");
9587 		return -1;
9588 	}
9589 	/* Create matchers, Any and Color. */
9590 	dv_attr.priority = 3;
9591 	dv_attr.match_criteria_enable = 0;
9592 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
9593 					       &dtb->any_matcher);
9594 	if (ret) {
9595 		DRV_LOG(ERR, "Failed to create meter"
9596 			     " policer default matcher.");
9597 		goto error_exit;
9598 	}
9599 	dv_attr.priority = 0;
9600 	dv_attr.match_criteria_enable =
9601 				1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT;
9602 	flow_dv_match_meta_reg(mask.buf, value.buf, color_reg_c_idx,
9603 			       rte_col_2_mlx5_col(RTE_COLORS), UINT8_MAX);
9604 	ret = mlx5_flow_os_create_flow_matcher(sh->ctx, &dv_attr, dtb->tbl->obj,
9605 					       &dtb->color_matcher);
9606 	if (ret) {
9607 		DRV_LOG(ERR, "Failed to create meter policer color matcher.");
9608 		goto error_exit;
9609 	}
9610 	if (mtb->count_actns[RTE_MTR_DROPPED])
9611 		actions[i++] = mtb->count_actns[RTE_MTR_DROPPED];
9612 	actions[i++] = mtb->drop_actn;
9613 	/* Default rule: lowest priority, match any, actions: drop. */
9614 	ret = mlx5_flow_os_create_flow(dtb->any_matcher, (void *)&value, i,
9615 				       actions,
9616 				       &dtb->policer_rules[RTE_MTR_DROPPED]);
9617 	if (ret) {
9618 		DRV_LOG(ERR, "Failed to create meter policer drop rule.");
9619 		goto error_exit;
9620 	}
9621 	return 0;
9622 error_exit:
9623 	return -1;
9624 }
9625 
9626 /**
9627  * Create the needed meter and suffix tables.
9628  * Lock free, (mutex should be acquired by caller).
9629  *
9630  * @param[in] dev
9631  *   Pointer to Ethernet device.
9632  * @param[in] fm
9633  *   Pointer to the flow meter.
9634  *
9635  * @return
9636  *   Pointer to table set on success, NULL otherwise and rte_errno is set.
9637  */
9638 static struct mlx5_meter_domains_infos *
9639 flow_dv_create_mtr_tbl(struct rte_eth_dev *dev,
9640 		       const struct mlx5_flow_meter *fm)
9641 {
9642 	struct mlx5_priv *priv = dev->data->dev_private;
9643 	struct mlx5_meter_domains_infos *mtb;
9644 	int ret;
9645 	int i;
9646 
9647 	if (!priv->mtr_en) {
9648 		rte_errno = ENOTSUP;
9649 		return NULL;
9650 	}
9651 	mtb = mlx5_malloc(MLX5_MEM_ZERO, sizeof(*mtb), 0, SOCKET_ID_ANY);
9652 	if (!mtb) {
9653 		DRV_LOG(ERR, "Failed to allocate memory for meter.");
9654 		return NULL;
9655 	}
9656 	/* Create meter count actions */
9657 	for (i = 0; i <= RTE_MTR_DROPPED; i++) {
9658 		struct mlx5_flow_counter *cnt;
9659 		if (!fm->policer_stats.cnt[i])
9660 			continue;
9661 		cnt = flow_dv_counter_get_by_idx(dev,
9662 		      fm->policer_stats.cnt[i], NULL);
9663 		mtb->count_actns[i] = cnt->action;
9664 	}
9665 	/* Create drop action. */
9666 	ret = mlx5_flow_os_create_flow_action_drop(&mtb->drop_actn);
9667 	if (ret) {
9668 		DRV_LOG(ERR, "Failed to create drop action.");
9669 		goto error_exit;
9670 	}
9671 	/* Egress meter table. */
9672 	ret = flow_dv_prepare_mtr_tables(dev, mtb, 1, 0, priv->mtr_color_reg);
9673 	if (ret) {
9674 		DRV_LOG(ERR, "Failed to prepare egress meter table.");
9675 		goto error_exit;
9676 	}
9677 	/* Ingress meter table. */
9678 	ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 0, priv->mtr_color_reg);
9679 	if (ret) {
9680 		DRV_LOG(ERR, "Failed to prepare ingress meter table.");
9681 		goto error_exit;
9682 	}
9683 	/* FDB meter table. */
9684 	if (priv->config.dv_esw_en) {
9685 		ret = flow_dv_prepare_mtr_tables(dev, mtb, 0, 1,
9686 						 priv->mtr_color_reg);
9687 		if (ret) {
9688 			DRV_LOG(ERR, "Failed to prepare fdb meter table.");
9689 			goto error_exit;
9690 		}
9691 	}
9692 	return mtb;
9693 error_exit:
9694 	flow_dv_destroy_mtr_tbl(dev, mtb);
9695 	return NULL;
9696 }
9697 
9698 /**
9699  * Destroy domain policer rule.
9700  *
9701  * @param[in] dt
9702  *   Pointer to domain table.
9703  */
9704 static void
9705 flow_dv_destroy_domain_policer_rule(struct mlx5_meter_domain_info *dt)
9706 {
9707 	int i;
9708 
9709 	for (i = 0; i < RTE_MTR_DROPPED; i++) {
9710 		if (dt->policer_rules[i]) {
9711 			claim_zero(mlx5_flow_os_destroy_flow
9712 				   (dt->policer_rules[i]));
9713 			dt->policer_rules[i] = NULL;
9714 		}
9715 	}
9716 	if (dt->jump_actn) {
9717 		claim_zero(mlx5_flow_os_destroy_flow_action(dt->jump_actn));
9718 		dt->jump_actn = NULL;
9719 	}
9720 }
9721 
9722 /**
9723  * Destroy policer rules.
9724  *
9725  * @param[in] dev
9726  *   Pointer to Ethernet device.
9727  * @param[in] fm
9728  *   Pointer to flow meter structure.
9729  * @param[in] attr
9730  *   Pointer to flow attributes.
9731  *
9732  * @return
9733  *   Always 0.
9734  */
9735 static int
9736 flow_dv_destroy_policer_rules(struct rte_eth_dev *dev __rte_unused,
9737 			      const struct mlx5_flow_meter *fm,
9738 			      const struct rte_flow_attr *attr)
9739 {
9740 	struct mlx5_meter_domains_infos *mtb = fm ? fm->mfts : NULL;
9741 
9742 	if (!mtb)
9743 		return 0;
9744 	if (attr->egress)
9745 		flow_dv_destroy_domain_policer_rule(&mtb->egress);
9746 	if (attr->ingress)
9747 		flow_dv_destroy_domain_policer_rule(&mtb->ingress);
9748 	if (attr->transfer)
9749 		flow_dv_destroy_domain_policer_rule(&mtb->transfer);
9750 	return 0;
9751 }
9752 
9753 /**
9754  * Create specify domain meter policer rule.
9755  *
9756  * @param[in] fm
9757  *   Pointer to flow meter structure.
9758  * @param[in] mtb
9759  *   Pointer to DV meter table set.
9760  * @param[in] mtr_reg_c
9761  *   Color match REG_C.
9762  *
9763  * @return
9764  *   0 on success, -1 otherwise.
9765  */
9766 static int
9767 flow_dv_create_policer_forward_rule(struct mlx5_flow_meter *fm,
9768 				    struct mlx5_meter_domain_info *dtb,
9769 				    uint8_t mtr_reg_c)
9770 {
9771 	struct mlx5_flow_dv_match_params matcher = {
9772 		.size = sizeof(matcher.buf),
9773 	};
9774 	struct mlx5_flow_dv_match_params value = {
9775 		.size = sizeof(value.buf),
9776 	};
9777 	struct mlx5_meter_domains_infos *mtb = fm->mfts;
9778 	void *actions[METER_ACTIONS];
9779 	int i;
9780 	int ret = 0;
9781 
9782 	/* Create jump action. */
9783 	if (!dtb->jump_actn)
9784 		ret = mlx5_flow_os_create_flow_action_dest_flow_tbl
9785 				(dtb->sfx_tbl->obj, &dtb->jump_actn);
9786 	if (ret) {
9787 		DRV_LOG(ERR, "Failed to create policer jump action.");
9788 		goto error;
9789 	}
9790 	for (i = 0; i < RTE_MTR_DROPPED; i++) {
9791 		int j = 0;
9792 
9793 		flow_dv_match_meta_reg(matcher.buf, value.buf, mtr_reg_c,
9794 				       rte_col_2_mlx5_col(i), UINT8_MAX);
9795 		if (mtb->count_actns[i])
9796 			actions[j++] = mtb->count_actns[i];
9797 		if (fm->action[i] == MTR_POLICER_ACTION_DROP)
9798 			actions[j++] = mtb->drop_actn;
9799 		else
9800 			actions[j++] = dtb->jump_actn;
9801 		ret = mlx5_flow_os_create_flow(dtb->color_matcher,
9802 					       (void *)&value, j, actions,
9803 					       &dtb->policer_rules[i]);
9804 		if (ret) {
9805 			DRV_LOG(ERR, "Failed to create policer rule.");
9806 			goto error;
9807 		}
9808 	}
9809 	return 0;
9810 error:
9811 	rte_errno = errno;
9812 	return -1;
9813 }
9814 
9815 /**
9816  * Create policer rules.
9817  *
9818  * @param[in] dev
9819  *   Pointer to Ethernet device.
9820  * @param[in] fm
9821  *   Pointer to flow meter structure.
9822  * @param[in] attr
9823  *   Pointer to flow attributes.
9824  *
9825  * @return
9826  *   0 on success, -1 otherwise.
9827  */
9828 static int
9829 flow_dv_create_policer_rules(struct rte_eth_dev *dev,
9830 			     struct mlx5_flow_meter *fm,
9831 			     const struct rte_flow_attr *attr)
9832 {
9833 	struct mlx5_priv *priv = dev->data->dev_private;
9834 	struct mlx5_meter_domains_infos *mtb = fm->mfts;
9835 	int ret;
9836 
9837 	if (attr->egress) {
9838 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->egress,
9839 						priv->mtr_color_reg);
9840 		if (ret) {
9841 			DRV_LOG(ERR, "Failed to create egress policer.");
9842 			goto error;
9843 		}
9844 	}
9845 	if (attr->ingress) {
9846 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->ingress,
9847 						priv->mtr_color_reg);
9848 		if (ret) {
9849 			DRV_LOG(ERR, "Failed to create ingress policer.");
9850 			goto error;
9851 		}
9852 	}
9853 	if (attr->transfer) {
9854 		ret = flow_dv_create_policer_forward_rule(fm, &mtb->transfer,
9855 						priv->mtr_color_reg);
9856 		if (ret) {
9857 			DRV_LOG(ERR, "Failed to create transfer policer.");
9858 			goto error;
9859 		}
9860 	}
9861 	return 0;
9862 error:
9863 	flow_dv_destroy_policer_rules(dev, fm, attr);
9864 	return -1;
9865 }
9866 
9867 /**
9868  * Query a devx counter.
9869  *
9870  * @param[in] dev
9871  *   Pointer to the Ethernet device structure.
9872  * @param[in] cnt
9873  *   Index to the flow counter.
9874  * @param[in] clear
9875  *   Set to clear the counter statistics.
9876  * @param[out] pkts
9877  *   The statistics value of packets.
9878  * @param[out] bytes
9879  *   The statistics value of bytes.
9880  *
9881  * @return
9882  *   0 on success, otherwise return -1.
9883  */
9884 static int
9885 flow_dv_counter_query(struct rte_eth_dev *dev, uint32_t counter, bool clear,
9886 		      uint64_t *pkts, uint64_t *bytes)
9887 {
9888 	struct mlx5_priv *priv = dev->data->dev_private;
9889 	struct mlx5_flow_counter *cnt;
9890 	uint64_t inn_pkts, inn_bytes;
9891 	int ret;
9892 
9893 	if (!priv->config.devx)
9894 		return -1;
9895 
9896 	ret = _flow_dv_query_count(dev, counter, &inn_pkts, &inn_bytes);
9897 	if (ret)
9898 		return -1;
9899 	cnt = flow_dv_counter_get_by_idx(dev, counter, NULL);
9900 	*pkts = inn_pkts - cnt->hits;
9901 	*bytes = inn_bytes - cnt->bytes;
9902 	if (clear) {
9903 		cnt->hits = inn_pkts;
9904 		cnt->bytes = inn_bytes;
9905 	}
9906 	return 0;
9907 }
9908 
9909 /**
9910  * Get aged-out flows.
9911  *
9912  * @param[in] dev
9913  *   Pointer to the Ethernet device structure.
9914  * @param[in] context
9915  *   The address of an array of pointers to the aged-out flows contexts.
9916  * @param[in] nb_contexts
9917  *   The length of context array pointers.
9918  * @param[out] error
9919  *   Perform verbose error reporting if not NULL. Initialized in case of
9920  *   error only.
9921  *
9922  * @return
9923  *   how many contexts get in success, otherwise negative errno value.
9924  *   if nb_contexts is 0, return the amount of all aged contexts.
9925  *   if nb_contexts is not 0 , return the amount of aged flows reported
9926  *   in the context array.
9927  * @note: only stub for now
9928  */
9929 static int
9930 flow_get_aged_flows(struct rte_eth_dev *dev,
9931 		    void **context,
9932 		    uint32_t nb_contexts,
9933 		    struct rte_flow_error *error)
9934 {
9935 	struct mlx5_priv *priv = dev->data->dev_private;
9936 	struct mlx5_age_info *age_info;
9937 	struct mlx5_age_param *age_param;
9938 	struct mlx5_flow_counter *counter;
9939 	int nb_flows = 0;
9940 
9941 	if (nb_contexts && !context)
9942 		return rte_flow_error_set(error, EINVAL,
9943 					  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
9944 					  NULL,
9945 					  "Should assign at least one flow or"
9946 					  " context to get if nb_contexts != 0");
9947 	age_info = GET_PORT_AGE_INFO(priv);
9948 	rte_spinlock_lock(&age_info->aged_sl);
9949 	TAILQ_FOREACH(counter, &age_info->aged_counters, next) {
9950 		nb_flows++;
9951 		if (nb_contexts) {
9952 			age_param = MLX5_CNT_TO_AGE(counter);
9953 			context[nb_flows - 1] = age_param->context;
9954 			if (!(--nb_contexts))
9955 				break;
9956 		}
9957 	}
9958 	rte_spinlock_unlock(&age_info->aged_sl);
9959 	MLX5_AGE_SET(age_info, MLX5_AGE_TRIGGER);
9960 	return nb_flows;
9961 }
9962 
9963 /*
9964  * Mutex-protected thunk to lock-free  __flow_dv_translate().
9965  */
9966 static int
9967 flow_dv_translate(struct rte_eth_dev *dev,
9968 		  struct mlx5_flow *dev_flow,
9969 		  const struct rte_flow_attr *attr,
9970 		  const struct rte_flow_item items[],
9971 		  const struct rte_flow_action actions[],
9972 		  struct rte_flow_error *error)
9973 {
9974 	int ret;
9975 
9976 	flow_dv_shared_lock(dev);
9977 	ret = __flow_dv_translate(dev, dev_flow, attr, items, actions, error);
9978 	flow_dv_shared_unlock(dev);
9979 	return ret;
9980 }
9981 
9982 /*
9983  * Mutex-protected thunk to lock-free  __flow_dv_apply().
9984  */
9985 static int
9986 flow_dv_apply(struct rte_eth_dev *dev,
9987 	      struct rte_flow *flow,
9988 	      struct rte_flow_error *error)
9989 {
9990 	int ret;
9991 
9992 	flow_dv_shared_lock(dev);
9993 	ret = __flow_dv_apply(dev, flow, error);
9994 	flow_dv_shared_unlock(dev);
9995 	return ret;
9996 }
9997 
9998 /*
9999  * Mutex-protected thunk to lock-free __flow_dv_remove().
10000  */
10001 static void
10002 flow_dv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
10003 {
10004 	flow_dv_shared_lock(dev);
10005 	__flow_dv_remove(dev, flow);
10006 	flow_dv_shared_unlock(dev);
10007 }
10008 
10009 /*
10010  * Mutex-protected thunk to lock-free __flow_dv_destroy().
10011  */
10012 static void
10013 flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
10014 {
10015 	flow_dv_shared_lock(dev);
10016 	__flow_dv_destroy(dev, flow);
10017 	flow_dv_shared_unlock(dev);
10018 }
10019 
10020 /*
10021  * Mutex-protected thunk to lock-free flow_dv_counter_alloc().
10022  */
10023 static uint32_t
10024 flow_dv_counter_allocate(struct rte_eth_dev *dev)
10025 {
10026 	uint32_t cnt;
10027 
10028 	flow_dv_shared_lock(dev);
10029 	cnt = flow_dv_counter_alloc(dev, 0, 0, 1, 0);
10030 	flow_dv_shared_unlock(dev);
10031 	return cnt;
10032 }
10033 
10034 /*
10035  * Mutex-protected thunk to lock-free flow_dv_counter_release().
10036  */
10037 static void
10038 flow_dv_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
10039 {
10040 	flow_dv_shared_lock(dev);
10041 	flow_dv_counter_release(dev, cnt);
10042 	flow_dv_shared_unlock(dev);
10043 }
10044 
10045 const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = {
10046 	.validate = flow_dv_validate,
10047 	.prepare = flow_dv_prepare,
10048 	.translate = flow_dv_translate,
10049 	.apply = flow_dv_apply,
10050 	.remove = flow_dv_remove,
10051 	.destroy = flow_dv_destroy,
10052 	.query = flow_dv_query,
10053 	.create_mtr_tbls = flow_dv_create_mtr_tbl,
10054 	.destroy_mtr_tbls = flow_dv_destroy_mtr_tbl,
10055 	.create_policer_rules = flow_dv_create_policer_rules,
10056 	.destroy_policer_rules = flow_dv_destroy_policer_rules,
10057 	.counter_alloc = flow_dv_counter_allocate,
10058 	.counter_free = flow_dv_counter_free,
10059 	.counter_query = flow_dv_counter_query,
10060 	.get_aged_flows = flow_get_aged_flows,
10061 };
10062 
10063 #endif /* HAVE_IBV_FLOW_DV_SUPPORT */
10064 
10065