xref: /dpdk/drivers/net/mlx5/mlx5_flow_verbs.c (revision d38febb08d57fec29fed27a2d12a507fc6fcdfa1)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2018 Mellanox Technologies, Ltd
3  */
4 
5 #include <netinet/in.h>
6 #include <sys/queue.h>
7 #include <stdalign.h>
8 #include <stdint.h>
9 #include <string.h>
10 
11 #include <rte_common.h>
12 #include <rte_ether.h>
13 #include <ethdev_driver.h>
14 #include <rte_flow.h>
15 #include <rte_flow_driver.h>
16 #include <rte_malloc.h>
17 #include <rte_ip.h>
18 
19 #include <mlx5_glue.h>
20 #include <mlx5_prm.h>
21 #include <mlx5_malloc.h>
22 
23 #include "mlx5_defs.h"
24 #include "mlx5.h"
25 #include "mlx5_flow.h"
26 #include "mlx5_rx.h"
27 
28 #define VERBS_SPEC_INNER(item_flags) \
29 	(!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
30 
31 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
32 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
33 	{ 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
34 };
35 
36 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
37 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
38 	{ 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
39 	{ 9, 10, 11 }, { 12, 13, 14 },
40 };
41 
42 /* Verbs specification header. */
43 struct ibv_spec_header {
44 	enum ibv_flow_spec_type type;
45 	uint16_t size;
46 };
47 
48 /**
49  * Discover the maximum number of priority available.
50  *
51  * @param[in] dev
52  *   Pointer to the Ethernet device structure.
53  *
54  * @return
55  *   number of supported flow priority on success, a negative errno
56  *   value otherwise and rte_errno is set.
57  */
58 int
59 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
60 {
61 	struct mlx5_priv *priv = dev->data->dev_private;
62 	struct {
63 		struct ibv_flow_attr attr;
64 		struct ibv_flow_spec_eth eth;
65 		struct ibv_flow_spec_action_drop drop;
66 	} flow_attr = {
67 		.attr = {
68 			.num_of_specs = 2,
69 			.port = (uint8_t)priv->dev_port,
70 		},
71 		.eth = {
72 			.type = IBV_FLOW_SPEC_ETH,
73 			.size = sizeof(struct ibv_flow_spec_eth),
74 		},
75 		.drop = {
76 			.size = sizeof(struct ibv_flow_spec_action_drop),
77 			.type = IBV_FLOW_SPEC_ACTION_DROP,
78 		},
79 	};
80 	struct ibv_flow *flow;
81 	struct mlx5_hrxq *drop = priv->drop_queue.hrxq;
82 	uint16_t vprio[] = { 8, 16 };
83 	int i;
84 	int priority = 0;
85 
86 	if (!drop->qp) {
87 		rte_errno = ENOTSUP;
88 		return -rte_errno;
89 	}
90 	for (i = 0; i != RTE_DIM(vprio); i++) {
91 		flow_attr.attr.priority = vprio[i] - 1;
92 		flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
93 		if (!flow)
94 			break;
95 		claim_zero(mlx5_glue->destroy_flow(flow));
96 		priority = vprio[i];
97 	}
98 	switch (priority) {
99 	case 8:
100 		priority = RTE_DIM(priority_map_3);
101 		break;
102 	case 16:
103 		priority = RTE_DIM(priority_map_5);
104 		break;
105 	default:
106 		rte_errno = ENOTSUP;
107 		DRV_LOG(ERR,
108 			"port %u verbs maximum priority: %d expected 8/16",
109 			dev->data->port_id, priority);
110 		return -rte_errno;
111 	}
112 	DRV_LOG(INFO, "port %u supported flow priorities:"
113 		" 0-%d for ingress or egress root table,"
114 		" 0-%d for non-root table or transfer root table.",
115 		dev->data->port_id, priority - 2,
116 		MLX5_NON_ROOT_FLOW_MAX_PRIO - 1);
117 	return priority;
118 }
119 
120 /**
121  * Adjust flow priority based on the highest layer and the request priority.
122  *
123  * @param[in] dev
124  *   Pointer to the Ethernet device structure.
125  * @param[in] priority
126  *   The rule base priority.
127  * @param[in] subpriority
128  *   The priority based on the items.
129  *
130  * @return
131  *   The new priority.
132  */
133 uint32_t
134 mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
135 				   uint32_t subpriority)
136 {
137 	uint32_t res = 0;
138 	struct mlx5_priv *priv = dev->data->dev_private;
139 
140 	switch (priv->config.flow_prio) {
141 	case RTE_DIM(priority_map_3):
142 		res = priority_map_3[priority][subpriority];
143 		break;
144 	case RTE_DIM(priority_map_5):
145 		res = priority_map_5[priority][subpriority];
146 		break;
147 	}
148 	return  res;
149 }
150 
151 /**
152  * Get Verbs flow counter by index.
153  *
154  * @param[in] dev
155  *   Pointer to the Ethernet device structure.
156  * @param[in] idx
157  *   mlx5 flow counter index in the container.
158  * @param[out] ppool
159  *   mlx5 flow counter pool in the container,
160  *
161  * @return
162  *   A pointer to the counter, NULL otherwise.
163  */
164 static struct mlx5_flow_counter *
165 flow_verbs_counter_get_by_idx(struct rte_eth_dev *dev,
166 			      uint32_t idx,
167 			      struct mlx5_flow_counter_pool **ppool)
168 {
169 	struct mlx5_priv *priv = dev->data->dev_private;
170 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
171 	struct mlx5_flow_counter_pool *pool;
172 
173 	idx = (idx - 1) & (MLX5_CNT_SHARED_OFFSET - 1);
174 	pool = cmng->pools[idx / MLX5_COUNTERS_PER_POOL];
175 	MLX5_ASSERT(pool);
176 	if (ppool)
177 		*ppool = pool;
178 	return MLX5_POOL_GET_CNT(pool, idx % MLX5_COUNTERS_PER_POOL);
179 }
180 
181 /**
182  * Create Verbs flow counter with Verbs library.
183  *
184  * @param[in] dev
185  *   Pointer to the Ethernet device structure.
186  * @param[in, out] counter
187  *   mlx5 flow counter object, contains the counter id,
188  *   handle of created Verbs flow counter is returned
189  *   in cs field (if counters are supported).
190  *
191  * @return
192  *   0 On success else a negative errno value is returned
193  *   and rte_errno is set.
194  */
195 static int
196 flow_verbs_counter_create(struct rte_eth_dev *dev,
197 			  struct mlx5_flow_counter *counter)
198 {
199 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
200 	struct mlx5_priv *priv = dev->data->dev_private;
201 	struct ibv_context *ctx = priv->sh->ctx;
202 	struct ibv_counter_set_init_attr init = {
203 			 .counter_set_id = counter->shared_info.id};
204 
205 	counter->dcs_when_free = mlx5_glue->create_counter_set(ctx, &init);
206 	if (!counter->dcs_when_free) {
207 		rte_errno = ENOTSUP;
208 		return -ENOTSUP;
209 	}
210 	return 0;
211 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
212 	struct mlx5_priv *priv = dev->data->dev_private;
213 	struct ibv_context *ctx = priv->sh->ctx;
214 	struct ibv_counters_init_attr init = {0};
215 	struct ibv_counter_attach_attr attach;
216 	int ret;
217 
218 	memset(&attach, 0, sizeof(attach));
219 	counter->dcs_when_free = mlx5_glue->create_counters(ctx, &init);
220 	if (!counter->dcs_when_free) {
221 		rte_errno = ENOTSUP;
222 		return -ENOTSUP;
223 	}
224 	attach.counter_desc = IBV_COUNTER_PACKETS;
225 	attach.index = 0;
226 	ret = mlx5_glue->attach_counters(counter->dcs_when_free, &attach, NULL);
227 	if (!ret) {
228 		attach.counter_desc = IBV_COUNTER_BYTES;
229 		attach.index = 1;
230 		ret = mlx5_glue->attach_counters
231 					(counter->dcs_when_free, &attach, NULL);
232 	}
233 	if (ret) {
234 		claim_zero(mlx5_glue->destroy_counters(counter->dcs_when_free));
235 		counter->dcs_when_free = NULL;
236 		rte_errno = ret;
237 		return -ret;
238 	}
239 	return 0;
240 #else
241 	(void)dev;
242 	(void)counter;
243 	rte_errno = ENOTSUP;
244 	return -ENOTSUP;
245 #endif
246 }
247 
248 /**
249  * Get a flow counter.
250  *
251  * @param[in] dev
252  *   Pointer to the Ethernet device structure.
253  * @param[in] shared
254  *   Indicate if this counter is shared with other flows.
255  * @param[in] id
256  *   Counter identifier.
257  *
258  * @return
259  *   Index to the counter, 0 otherwise and rte_errno is set.
260  */
261 static uint32_t
262 flow_verbs_counter_new(struct rte_eth_dev *dev, uint32_t shared, uint32_t id)
263 {
264 	struct mlx5_priv *priv = dev->data->dev_private;
265 	struct mlx5_flow_counter_mng *cmng = &priv->sh->cmng;
266 	struct mlx5_flow_counter_pool *pool = NULL;
267 	struct mlx5_flow_counter *cnt = NULL;
268 	union mlx5_l3t_data data;
269 	uint32_t n_valid = cmng->n_valid;
270 	uint32_t pool_idx, cnt_idx;
271 	uint32_t i;
272 	int ret;
273 
274 	if (shared && !mlx5_l3t_get_entry(priv->sh->cnt_id_tbl, id, &data) &&
275 	    data.dword)
276 		return data.dword;
277 	for (pool_idx = 0; pool_idx < n_valid; ++pool_idx) {
278 		pool = cmng->pools[pool_idx];
279 		if (!pool)
280 			continue;
281 		cnt = TAILQ_FIRST(&pool->counters[0]);
282 		if (cnt)
283 			break;
284 	}
285 	if (!cnt) {
286 		struct mlx5_flow_counter_pool **pools;
287 		uint32_t size;
288 
289 		if (n_valid == cmng->n) {
290 			/* Resize the container pool array. */
291 			size = sizeof(struct mlx5_flow_counter_pool *) *
292 				     (n_valid + MLX5_CNT_CONTAINER_RESIZE);
293 			pools = mlx5_malloc(MLX5_MEM_ZERO, size, 0,
294 					    SOCKET_ID_ANY);
295 			if (!pools)
296 				return 0;
297 			if (n_valid) {
298 				memcpy(pools, cmng->pools,
299 				       sizeof(struct mlx5_flow_counter_pool *) *
300 				       n_valid);
301 				mlx5_free(cmng->pools);
302 			}
303 			cmng->pools = pools;
304 			cmng->n += MLX5_CNT_CONTAINER_RESIZE;
305 		}
306 		/* Allocate memory for new pool*/
307 		size = sizeof(*pool) + sizeof(*cnt) * MLX5_COUNTERS_PER_POOL;
308 		pool = mlx5_malloc(MLX5_MEM_ZERO, size, 0, SOCKET_ID_ANY);
309 		if (!pool)
310 			return 0;
311 		for (i = 0; i < MLX5_COUNTERS_PER_POOL; ++i) {
312 			cnt = MLX5_POOL_GET_CNT(pool, i);
313 			TAILQ_INSERT_HEAD(&pool->counters[0], cnt, next);
314 		}
315 		cnt = MLX5_POOL_GET_CNT(pool, 0);
316 		cmng->pools[n_valid] = pool;
317 		pool_idx = n_valid;
318 		cmng->n_valid++;
319 	}
320 	TAILQ_REMOVE(&pool->counters[0], cnt, next);
321 	i = MLX5_CNT_ARRAY_IDX(pool, cnt);
322 	cnt_idx = MLX5_MAKE_CNT_IDX(pool_idx, i);
323 	if (shared) {
324 		data.dword = cnt_idx;
325 		if (mlx5_l3t_set_entry(priv->sh->cnt_id_tbl, id, &data))
326 			return 0;
327 		cnt->shared_info.id = id;
328 		cnt_idx |= MLX5_CNT_SHARED_OFFSET;
329 	}
330 	/* Create counter with Verbs. */
331 	ret = flow_verbs_counter_create(dev, cnt);
332 	if (!ret) {
333 		cnt->dcs_when_active = cnt->dcs_when_free;
334 		cnt->hits = 0;
335 		cnt->bytes = 0;
336 		return cnt_idx;
337 	}
338 	TAILQ_INSERT_HEAD(&pool->counters[0], cnt, next);
339 	/* Some error occurred in Verbs library. */
340 	rte_errno = -ret;
341 	return 0;
342 }
343 
344 /**
345  * Release a flow counter.
346  *
347  * @param[in] dev
348  *   Pointer to the Ethernet device structure.
349  * @param[in] counter
350  *   Index to the counter handler.
351  */
352 static void
353 flow_verbs_counter_release(struct rte_eth_dev *dev, uint32_t counter)
354 {
355 	struct mlx5_priv *priv = dev->data->dev_private;
356 	struct mlx5_flow_counter_pool *pool;
357 	struct mlx5_flow_counter *cnt;
358 
359 	cnt = flow_verbs_counter_get_by_idx(dev, counter, &pool);
360 	if (IS_LEGACY_SHARED_CNT(counter) &&
361 	    mlx5_l3t_clear_entry(priv->sh->cnt_id_tbl, cnt->shared_info.id))
362 		return;
363 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
364 	claim_zero(mlx5_glue->destroy_counter_set
365 			((struct ibv_counter_set *)cnt->dcs_when_active));
366 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
367 	claim_zero(mlx5_glue->destroy_counters
368 				((struct ibv_counters *)cnt->dcs_when_active));
369 #endif
370 	TAILQ_INSERT_HEAD(&pool->counters[0], cnt, next);
371 }
372 
373 /**
374  * Query a flow counter via Verbs library call.
375  *
376  * @see rte_flow_query()
377  * @see rte_flow_ops
378  */
379 static int
380 flow_verbs_counter_query(struct rte_eth_dev *dev __rte_unused,
381 			 struct rte_flow *flow, void *data,
382 			 struct rte_flow_error *error)
383 {
384 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
385 	defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
386 	if (flow->counter) {
387 		struct mlx5_flow_counter_pool *pool;
388 		struct mlx5_flow_counter *cnt = flow_verbs_counter_get_by_idx
389 						(dev, flow->counter, &pool);
390 		struct rte_flow_query_count *qc = data;
391 		uint64_t counters[2] = {0, 0};
392 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
393 		struct ibv_query_counter_set_attr query_cs_attr = {
394 			.dcs_when_free = (struct ibv_counter_set *)
395 						cnt->dcs_when_active,
396 			.query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
397 		};
398 		struct ibv_counter_set_data query_out = {
399 			.out = counters,
400 			.outlen = 2 * sizeof(uint64_t),
401 		};
402 		int err = mlx5_glue->query_counter_set(&query_cs_attr,
403 						       &query_out);
404 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
405 		int err = mlx5_glue->query_counters
406 			((struct ibv_counters *)cnt->dcs_when_active, counters,
407 				RTE_DIM(counters),
408 				IBV_READ_COUNTERS_ATTR_PREFER_CACHED);
409 #endif
410 		if (err)
411 			return rte_flow_error_set
412 				(error, err,
413 				 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
414 				 NULL,
415 				 "cannot read counter");
416 		qc->hits_set = 1;
417 		qc->bytes_set = 1;
418 		qc->hits = counters[0] - cnt->hits;
419 		qc->bytes = counters[1] - cnt->bytes;
420 		if (qc->reset) {
421 			cnt->hits = counters[0];
422 			cnt->bytes = counters[1];
423 		}
424 		return 0;
425 	}
426 	return rte_flow_error_set(error, EINVAL,
427 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
428 				  NULL,
429 				  "flow does not have counter");
430 #else
431 	(void)flow;
432 	(void)data;
433 	return rte_flow_error_set(error, ENOTSUP,
434 				  RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
435 				  NULL,
436 				  "counters are not available");
437 #endif
438 }
439 
440 /**
441  * Add a verbs item specification into @p verbs.
442  *
443  * @param[out] verbs
444  *   Pointer to verbs structure.
445  * @param[in] src
446  *   Create specification.
447  * @param[in] size
448  *   Size in bytes of the specification to copy.
449  */
450 static void
451 flow_verbs_spec_add(struct mlx5_flow_verbs_workspace *verbs,
452 		    void *src, unsigned int size)
453 {
454 	void *dst;
455 
456 	if (!verbs)
457 		return;
458 	MLX5_ASSERT(verbs->specs);
459 	dst = (void *)(verbs->specs + verbs->size);
460 	memcpy(dst, src, size);
461 	++verbs->attr.num_of_specs;
462 	verbs->size += size;
463 }
464 
465 /**
466  * Convert the @p item into a Verbs specification. This function assumes that
467  * the input is valid and that there is space to insert the requested item
468  * into the flow.
469  *
470  * @param[in, out] dev_flow
471  *   Pointer to dev_flow structure.
472  * @param[in] item
473  *   Item specification.
474  * @param[in] item_flags
475  *   Parsed item flags.
476  */
477 static void
478 flow_verbs_translate_item_eth(struct mlx5_flow *dev_flow,
479 			      const struct rte_flow_item *item,
480 			      uint64_t item_flags)
481 {
482 	const struct rte_flow_item_eth *spec = item->spec;
483 	const struct rte_flow_item_eth *mask = item->mask;
484 	const unsigned int size = sizeof(struct ibv_flow_spec_eth);
485 	struct ibv_flow_spec_eth eth = {
486 		.type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
487 		.size = size,
488 	};
489 
490 	if (!mask)
491 		mask = &rte_flow_item_eth_mask;
492 	if (spec) {
493 		unsigned int i;
494 
495 		memcpy(&eth.val.dst_mac, spec->dst.addr_bytes,
496 			RTE_ETHER_ADDR_LEN);
497 		memcpy(&eth.val.src_mac, spec->src.addr_bytes,
498 			RTE_ETHER_ADDR_LEN);
499 		eth.val.ether_type = spec->type;
500 		memcpy(&eth.mask.dst_mac, mask->dst.addr_bytes,
501 			RTE_ETHER_ADDR_LEN);
502 		memcpy(&eth.mask.src_mac, mask->src.addr_bytes,
503 			RTE_ETHER_ADDR_LEN);
504 		eth.mask.ether_type = mask->type;
505 		/* Remove unwanted bits from values. */
506 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i) {
507 			eth.val.dst_mac[i] &= eth.mask.dst_mac[i];
508 			eth.val.src_mac[i] &= eth.mask.src_mac[i];
509 		}
510 		eth.val.ether_type &= eth.mask.ether_type;
511 	}
512 	flow_verbs_spec_add(&dev_flow->verbs, &eth, size);
513 }
514 
515 /**
516  * Update the VLAN tag in the Verbs Ethernet specification.
517  * This function assumes that the input is valid and there is space to add
518  * the requested item.
519  *
520  * @param[in, out] attr
521  *   Pointer to Verbs attributes structure.
522  * @param[in] eth
523  *   Verbs structure containing the VLAN information to copy.
524  */
525 static void
526 flow_verbs_item_vlan_update(struct ibv_flow_attr *attr,
527 			    struct ibv_flow_spec_eth *eth)
528 {
529 	unsigned int i;
530 	const enum ibv_flow_spec_type search = eth->type;
531 	struct ibv_spec_header *hdr = (struct ibv_spec_header *)
532 		((uint8_t *)attr + sizeof(struct ibv_flow_attr));
533 
534 	for (i = 0; i != attr->num_of_specs; ++i) {
535 		if (hdr->type == search) {
536 			struct ibv_flow_spec_eth *e =
537 				(struct ibv_flow_spec_eth *)hdr;
538 
539 			e->val.vlan_tag = eth->val.vlan_tag;
540 			e->mask.vlan_tag = eth->mask.vlan_tag;
541 			e->val.ether_type = eth->val.ether_type;
542 			e->mask.ether_type = eth->mask.ether_type;
543 			break;
544 		}
545 		hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
546 	}
547 }
548 
549 /**
550  * Convert the @p item into a Verbs specification. This function assumes that
551  * the input is valid and that there is space to insert the requested item
552  * into the flow.
553  *
554  * @param[in, out] dev_flow
555  *   Pointer to dev_flow structure.
556  * @param[in] item
557  *   Item specification.
558  * @param[in] item_flags
559  *   Parsed item flags.
560  */
561 static void
562 flow_verbs_translate_item_vlan(struct mlx5_flow *dev_flow,
563 			       const struct rte_flow_item *item,
564 			       uint64_t item_flags)
565 {
566 	const struct rte_flow_item_vlan *spec = item->spec;
567 	const struct rte_flow_item_vlan *mask = item->mask;
568 	unsigned int size = sizeof(struct ibv_flow_spec_eth);
569 	const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
570 	struct ibv_flow_spec_eth eth = {
571 		.type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
572 		.size = size,
573 	};
574 	const uint32_t l2m = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
575 				      MLX5_FLOW_LAYER_OUTER_L2;
576 
577 	if (!mask)
578 		mask = &rte_flow_item_vlan_mask;
579 	if (spec) {
580 		eth.val.vlan_tag = spec->tci;
581 		eth.mask.vlan_tag = mask->tci;
582 		eth.val.vlan_tag &= eth.mask.vlan_tag;
583 		eth.val.ether_type = spec->inner_type;
584 		eth.mask.ether_type = mask->inner_type;
585 		eth.val.ether_type &= eth.mask.ether_type;
586 	}
587 	if (!(item_flags & l2m))
588 		flow_verbs_spec_add(&dev_flow->verbs, &eth, size);
589 	else
590 		flow_verbs_item_vlan_update(&dev_flow->verbs.attr, &eth);
591 	if (!tunnel)
592 		dev_flow->handle->vf_vlan.tag =
593 			rte_be_to_cpu_16(spec->tci) & 0x0fff;
594 }
595 
596 /**
597  * Convert the @p item into a Verbs specification. This function assumes that
598  * the input is valid and that there is space to insert the requested item
599  * into the flow.
600  *
601  * @param[in, out] dev_flow
602  *   Pointer to dev_flow structure.
603  * @param[in] item
604  *   Item specification.
605  * @param[in] item_flags
606  *   Parsed item flags.
607  */
608 static void
609 flow_verbs_translate_item_ipv4(struct mlx5_flow *dev_flow,
610 			       const struct rte_flow_item *item,
611 			       uint64_t item_flags)
612 {
613 	const struct rte_flow_item_ipv4 *spec = item->spec;
614 	const struct rte_flow_item_ipv4 *mask = item->mask;
615 	unsigned int size = sizeof(struct ibv_flow_spec_ipv4_ext);
616 	struct ibv_flow_spec_ipv4_ext ipv4 = {
617 		.type = IBV_FLOW_SPEC_IPV4_EXT | VERBS_SPEC_INNER(item_flags),
618 		.size = size,
619 	};
620 
621 	if (!mask)
622 		mask = &rte_flow_item_ipv4_mask;
623 	if (spec) {
624 		ipv4.val = (struct ibv_flow_ipv4_ext_filter){
625 			.src_ip = spec->hdr.src_addr,
626 			.dst_ip = spec->hdr.dst_addr,
627 			.proto = spec->hdr.next_proto_id,
628 			.tos = spec->hdr.type_of_service,
629 		};
630 		ipv4.mask = (struct ibv_flow_ipv4_ext_filter){
631 			.src_ip = mask->hdr.src_addr,
632 			.dst_ip = mask->hdr.dst_addr,
633 			.proto = mask->hdr.next_proto_id,
634 			.tos = mask->hdr.type_of_service,
635 		};
636 		/* Remove unwanted bits from values. */
637 		ipv4.val.src_ip &= ipv4.mask.src_ip;
638 		ipv4.val.dst_ip &= ipv4.mask.dst_ip;
639 		ipv4.val.proto &= ipv4.mask.proto;
640 		ipv4.val.tos &= ipv4.mask.tos;
641 	}
642 	flow_verbs_spec_add(&dev_flow->verbs, &ipv4, size);
643 }
644 
645 /**
646  * Convert the @p item into a Verbs specification. This function assumes that
647  * the input is valid and that there is space to insert the requested item
648  * into the flow.
649  *
650  * @param[in, out] dev_flow
651  *   Pointer to dev_flow structure.
652  * @param[in] item
653  *   Item specification.
654  * @param[in] item_flags
655  *   Parsed item flags.
656  */
657 static void
658 flow_verbs_translate_item_ipv6(struct mlx5_flow *dev_flow,
659 			       const struct rte_flow_item *item,
660 			       uint64_t item_flags)
661 {
662 	const struct rte_flow_item_ipv6 *spec = item->spec;
663 	const struct rte_flow_item_ipv6 *mask = item->mask;
664 	unsigned int size = sizeof(struct ibv_flow_spec_ipv6);
665 	struct ibv_flow_spec_ipv6 ipv6 = {
666 		.type = IBV_FLOW_SPEC_IPV6 | VERBS_SPEC_INNER(item_flags),
667 		.size = size,
668 	};
669 
670 	if (!mask)
671 		mask = &rte_flow_item_ipv6_mask;
672 	if (spec) {
673 		unsigned int i;
674 		uint32_t vtc_flow_val;
675 		uint32_t vtc_flow_mask;
676 
677 		memcpy(&ipv6.val.src_ip, spec->hdr.src_addr,
678 		       RTE_DIM(ipv6.val.src_ip));
679 		memcpy(&ipv6.val.dst_ip, spec->hdr.dst_addr,
680 		       RTE_DIM(ipv6.val.dst_ip));
681 		memcpy(&ipv6.mask.src_ip, mask->hdr.src_addr,
682 		       RTE_DIM(ipv6.mask.src_ip));
683 		memcpy(&ipv6.mask.dst_ip, mask->hdr.dst_addr,
684 		       RTE_DIM(ipv6.mask.dst_ip));
685 		vtc_flow_val = rte_be_to_cpu_32(spec->hdr.vtc_flow);
686 		vtc_flow_mask = rte_be_to_cpu_32(mask->hdr.vtc_flow);
687 		ipv6.val.flow_label =
688 			rte_cpu_to_be_32((vtc_flow_val & RTE_IPV6_HDR_FL_MASK) >>
689 					 RTE_IPV6_HDR_FL_SHIFT);
690 		ipv6.val.traffic_class = (vtc_flow_val & RTE_IPV6_HDR_TC_MASK) >>
691 					 RTE_IPV6_HDR_TC_SHIFT;
692 		ipv6.val.next_hdr = spec->hdr.proto;
693 		ipv6.mask.flow_label =
694 			rte_cpu_to_be_32((vtc_flow_mask & RTE_IPV6_HDR_FL_MASK) >>
695 					 RTE_IPV6_HDR_FL_SHIFT);
696 		ipv6.mask.traffic_class = (vtc_flow_mask & RTE_IPV6_HDR_TC_MASK) >>
697 					  RTE_IPV6_HDR_TC_SHIFT;
698 		ipv6.mask.next_hdr = mask->hdr.proto;
699 		/* Remove unwanted bits from values. */
700 		for (i = 0; i < RTE_DIM(ipv6.val.src_ip); ++i) {
701 			ipv6.val.src_ip[i] &= ipv6.mask.src_ip[i];
702 			ipv6.val.dst_ip[i] &= ipv6.mask.dst_ip[i];
703 		}
704 		ipv6.val.flow_label &= ipv6.mask.flow_label;
705 		ipv6.val.traffic_class &= ipv6.mask.traffic_class;
706 		ipv6.val.next_hdr &= ipv6.mask.next_hdr;
707 	}
708 	flow_verbs_spec_add(&dev_flow->verbs, &ipv6, size);
709 }
710 
711 /**
712  * Convert the @p item into a Verbs specification. This function assumes that
713  * the input is valid and that there is space to insert the requested item
714  * into the flow.
715  *
716  * @param[in, out] dev_flow
717  *   Pointer to dev_flow structure.
718  * @param[in] item
719  *   Item specification.
720  * @param[in] item_flags
721  *   Parsed item flags.
722  */
723 static void
724 flow_verbs_translate_item_tcp(struct mlx5_flow *dev_flow,
725 			      const struct rte_flow_item *item,
726 			      uint64_t item_flags __rte_unused)
727 {
728 	const struct rte_flow_item_tcp *spec = item->spec;
729 	const struct rte_flow_item_tcp *mask = item->mask;
730 	unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
731 	struct ibv_flow_spec_tcp_udp tcp = {
732 		.type = IBV_FLOW_SPEC_TCP | VERBS_SPEC_INNER(item_flags),
733 		.size = size,
734 	};
735 
736 	if (!mask)
737 		mask = &rte_flow_item_tcp_mask;
738 	if (spec) {
739 		tcp.val.dst_port = spec->hdr.dst_port;
740 		tcp.val.src_port = spec->hdr.src_port;
741 		tcp.mask.dst_port = mask->hdr.dst_port;
742 		tcp.mask.src_port = mask->hdr.src_port;
743 		/* Remove unwanted bits from values. */
744 		tcp.val.src_port &= tcp.mask.src_port;
745 		tcp.val.dst_port &= tcp.mask.dst_port;
746 	}
747 	flow_verbs_spec_add(&dev_flow->verbs, &tcp, size);
748 }
749 
750 /**
751  * Convert the @p item into a Verbs specification. This function assumes that
752  * the input is valid and that there is space to insert the requested item
753  * into the flow.
754  *
755  * @param[in, out] dev_flow
756  *   Pointer to dev_flow structure.
757  * @param[in] item
758  *   Item specification.
759  * @param[in] item_flags
760  *   Parsed item flags.
761  */
762 static void
763 flow_verbs_translate_item_udp(struct mlx5_flow *dev_flow,
764 			      const struct rte_flow_item *item,
765 			      uint64_t item_flags __rte_unused)
766 {
767 	const struct rte_flow_item_udp *spec = item->spec;
768 	const struct rte_flow_item_udp *mask = item->mask;
769 	unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
770 	struct ibv_flow_spec_tcp_udp udp = {
771 		.type = IBV_FLOW_SPEC_UDP | VERBS_SPEC_INNER(item_flags),
772 		.size = size,
773 	};
774 
775 	if (!mask)
776 		mask = &rte_flow_item_udp_mask;
777 	if (spec) {
778 		udp.val.dst_port = spec->hdr.dst_port;
779 		udp.val.src_port = spec->hdr.src_port;
780 		udp.mask.dst_port = mask->hdr.dst_port;
781 		udp.mask.src_port = mask->hdr.src_port;
782 		/* Remove unwanted bits from values. */
783 		udp.val.src_port &= udp.mask.src_port;
784 		udp.val.dst_port &= udp.mask.dst_port;
785 	}
786 	item++;
787 	while (item->type == RTE_FLOW_ITEM_TYPE_VOID)
788 		item++;
789 	if (!(udp.val.dst_port & udp.mask.dst_port)) {
790 		switch ((item)->type) {
791 		case RTE_FLOW_ITEM_TYPE_VXLAN:
792 			udp.val.dst_port = htons(MLX5_UDP_PORT_VXLAN);
793 			udp.mask.dst_port = 0xffff;
794 			break;
795 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
796 			udp.val.dst_port = htons(MLX5_UDP_PORT_VXLAN_GPE);
797 			udp.mask.dst_port = 0xffff;
798 			break;
799 		case RTE_FLOW_ITEM_TYPE_MPLS:
800 			udp.val.dst_port = htons(MLX5_UDP_PORT_MPLS);
801 			udp.mask.dst_port = 0xffff;
802 			break;
803 		default:
804 			break;
805 		}
806 	}
807 
808 	flow_verbs_spec_add(&dev_flow->verbs, &udp, size);
809 }
810 
811 /**
812  * Convert the @p item into a Verbs specification. This function assumes that
813  * the input is valid and that there is space to insert the requested item
814  * into the flow.
815  *
816  * @param[in, out] dev_flow
817  *   Pointer to dev_flow structure.
818  * @param[in] item
819  *   Item specification.
820  * @param[in] item_flags
821  *   Parsed item flags.
822  */
823 static void
824 flow_verbs_translate_item_vxlan(struct mlx5_flow *dev_flow,
825 				const struct rte_flow_item *item,
826 				uint64_t item_flags __rte_unused)
827 {
828 	const struct rte_flow_item_vxlan *spec = item->spec;
829 	const struct rte_flow_item_vxlan *mask = item->mask;
830 	unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
831 	struct ibv_flow_spec_tunnel vxlan = {
832 		.type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
833 		.size = size,
834 	};
835 	union vni {
836 		uint32_t vlan_id;
837 		uint8_t vni[4];
838 	} id = { .vlan_id = 0, };
839 
840 	if (!mask)
841 		mask = &rte_flow_item_vxlan_mask;
842 	if (spec) {
843 		memcpy(&id.vni[1], spec->vni, 3);
844 		vxlan.val.tunnel_id = id.vlan_id;
845 		memcpy(&id.vni[1], mask->vni, 3);
846 		vxlan.mask.tunnel_id = id.vlan_id;
847 		/* Remove unwanted bits from values. */
848 		vxlan.val.tunnel_id &= vxlan.mask.tunnel_id;
849 	}
850 	flow_verbs_spec_add(&dev_flow->verbs, &vxlan, size);
851 }
852 
853 /**
854  * Convert the @p item into a Verbs specification. This function assumes that
855  * the input is valid and that there is space to insert the requested item
856  * into the flow.
857  *
858  * @param[in, out] dev_flow
859  *   Pointer to dev_flow structure.
860  * @param[in] item
861  *   Item specification.
862  * @param[in] item_flags
863  *   Parsed item flags.
864  */
865 static void
866 flow_verbs_translate_item_vxlan_gpe(struct mlx5_flow *dev_flow,
867 				    const struct rte_flow_item *item,
868 				    uint64_t item_flags __rte_unused)
869 {
870 	const struct rte_flow_item_vxlan_gpe *spec = item->spec;
871 	const struct rte_flow_item_vxlan_gpe *mask = item->mask;
872 	unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
873 	struct ibv_flow_spec_tunnel vxlan_gpe = {
874 		.type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
875 		.size = size,
876 	};
877 	union vni {
878 		uint32_t vlan_id;
879 		uint8_t vni[4];
880 	} id = { .vlan_id = 0, };
881 
882 	if (!mask)
883 		mask = &rte_flow_item_vxlan_gpe_mask;
884 	if (spec) {
885 		memcpy(&id.vni[1], spec->vni, 3);
886 		vxlan_gpe.val.tunnel_id = id.vlan_id;
887 		memcpy(&id.vni[1], mask->vni, 3);
888 		vxlan_gpe.mask.tunnel_id = id.vlan_id;
889 		/* Remove unwanted bits from values. */
890 		vxlan_gpe.val.tunnel_id &= vxlan_gpe.mask.tunnel_id;
891 	}
892 	flow_verbs_spec_add(&dev_flow->verbs, &vxlan_gpe, size);
893 }
894 
895 /**
896  * Update the protocol in Verbs IPv4/IPv6 spec.
897  *
898  * @param[in, out] attr
899  *   Pointer to Verbs attributes structure.
900  * @param[in] search
901  *   Specification type to search in order to update the IP protocol.
902  * @param[in] protocol
903  *   Protocol value to set if none is present in the specification.
904  */
905 static void
906 flow_verbs_item_gre_ip_protocol_update(struct ibv_flow_attr *attr,
907 				       enum ibv_flow_spec_type search,
908 				       uint8_t protocol)
909 {
910 	unsigned int i;
911 	struct ibv_spec_header *hdr = (struct ibv_spec_header *)
912 		((uint8_t *)attr + sizeof(struct ibv_flow_attr));
913 
914 	if (!attr)
915 		return;
916 	for (i = 0; i != attr->num_of_specs; ++i) {
917 		if (hdr->type == search) {
918 			union {
919 				struct ibv_flow_spec_ipv4_ext *ipv4;
920 				struct ibv_flow_spec_ipv6 *ipv6;
921 			} ip;
922 
923 			switch (search) {
924 			case IBV_FLOW_SPEC_IPV4_EXT:
925 				ip.ipv4 = (struct ibv_flow_spec_ipv4_ext *)hdr;
926 				if (!ip.ipv4->val.proto) {
927 					ip.ipv4->val.proto = protocol;
928 					ip.ipv4->mask.proto = 0xff;
929 				}
930 				break;
931 			case IBV_FLOW_SPEC_IPV6:
932 				ip.ipv6 = (struct ibv_flow_spec_ipv6 *)hdr;
933 				if (!ip.ipv6->val.next_hdr) {
934 					ip.ipv6->val.next_hdr = protocol;
935 					ip.ipv6->mask.next_hdr = 0xff;
936 				}
937 				break;
938 			default:
939 				break;
940 			}
941 			break;
942 		}
943 		hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
944 	}
945 }
946 
947 /**
948  * Convert the @p item into a Verbs specification. This function assumes that
949  * the input is valid and that there is space to insert the requested item
950  * into the flow.
951  *
952  * @param[in, out] dev_flow
953  *   Pointer to dev_flow structure.
954  * @param[in] item
955  *   Item specification.
956  * @param[in] item_flags
957  *   Parsed item flags.
958  */
959 static void
960 flow_verbs_translate_item_gre(struct mlx5_flow *dev_flow,
961 			      const struct rte_flow_item *item __rte_unused,
962 			      uint64_t item_flags)
963 {
964 	struct mlx5_flow_verbs_workspace *verbs = &dev_flow->verbs;
965 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
966 	unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
967 	struct ibv_flow_spec_tunnel tunnel = {
968 		.type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
969 		.size = size,
970 	};
971 #else
972 	const struct rte_flow_item_gre *spec = item->spec;
973 	const struct rte_flow_item_gre *mask = item->mask;
974 	unsigned int size = sizeof(struct ibv_flow_spec_gre);
975 	struct ibv_flow_spec_gre tunnel = {
976 		.type = IBV_FLOW_SPEC_GRE,
977 		.size = size,
978 	};
979 
980 	if (!mask)
981 		mask = &rte_flow_item_gre_mask;
982 	if (spec) {
983 		tunnel.val.c_ks_res0_ver = spec->c_rsvd0_ver;
984 		tunnel.val.protocol = spec->protocol;
985 		tunnel.mask.c_ks_res0_ver = mask->c_rsvd0_ver;
986 		tunnel.mask.protocol = mask->protocol;
987 		/* Remove unwanted bits from values. */
988 		tunnel.val.c_ks_res0_ver &= tunnel.mask.c_ks_res0_ver;
989 		tunnel.val.protocol &= tunnel.mask.protocol;
990 		tunnel.val.key &= tunnel.mask.key;
991 	}
992 #endif
993 	if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
994 		flow_verbs_item_gre_ip_protocol_update(&verbs->attr,
995 						       IBV_FLOW_SPEC_IPV4_EXT,
996 						       IPPROTO_GRE);
997 	else
998 		flow_verbs_item_gre_ip_protocol_update(&verbs->attr,
999 						       IBV_FLOW_SPEC_IPV6,
1000 						       IPPROTO_GRE);
1001 	flow_verbs_spec_add(verbs, &tunnel, size);
1002 }
1003 
1004 /**
1005  * Convert the @p action into a Verbs specification. This function assumes that
1006  * the input is valid and that there is space to insert the requested action
1007  * into the flow. This function also return the action that was added.
1008  *
1009  * @param[in, out] dev_flow
1010  *   Pointer to dev_flow structure.
1011  * @param[in] item
1012  *   Item specification.
1013  * @param[in] item_flags
1014  *   Parsed item flags.
1015  */
1016 static void
1017 flow_verbs_translate_item_mpls(struct mlx5_flow *dev_flow __rte_unused,
1018 			       const struct rte_flow_item *item __rte_unused,
1019 			       uint64_t item_flags __rte_unused)
1020 {
1021 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1022 	const struct rte_flow_item_mpls *spec = item->spec;
1023 	const struct rte_flow_item_mpls *mask = item->mask;
1024 	unsigned int size = sizeof(struct ibv_flow_spec_mpls);
1025 	struct ibv_flow_spec_mpls mpls = {
1026 		.type = IBV_FLOW_SPEC_MPLS,
1027 		.size = size,
1028 	};
1029 
1030 	if (!mask)
1031 		mask = &rte_flow_item_mpls_mask;
1032 	if (spec) {
1033 		memcpy(&mpls.val.label, spec, sizeof(mpls.val.label));
1034 		memcpy(&mpls.mask.label, mask, sizeof(mpls.mask.label));
1035 		/* Remove unwanted bits from values.  */
1036 		mpls.val.label &= mpls.mask.label;
1037 	}
1038 	flow_verbs_spec_add(&dev_flow->verbs, &mpls, size);
1039 #endif
1040 }
1041 
1042 /**
1043  * Convert the @p action into a Verbs specification. This function assumes that
1044  * the input is valid and that there is space to insert the requested action
1045  * into the flow.
1046  *
1047  * @param[in] dev_flow
1048  *   Pointer to mlx5_flow.
1049  * @param[in] action
1050  *   Action configuration.
1051  */
1052 static void
1053 flow_verbs_translate_action_drop
1054 	(struct mlx5_flow *dev_flow,
1055 	 const struct rte_flow_action *action __rte_unused)
1056 {
1057 	unsigned int size = sizeof(struct ibv_flow_spec_action_drop);
1058 	struct ibv_flow_spec_action_drop drop = {
1059 			.type = IBV_FLOW_SPEC_ACTION_DROP,
1060 			.size = size,
1061 	};
1062 
1063 	flow_verbs_spec_add(&dev_flow->verbs, &drop, size);
1064 }
1065 
1066 /**
1067  * Convert the @p action into a Verbs specification. This function assumes that
1068  * the input is valid and that there is space to insert the requested action
1069  * into the flow.
1070  *
1071  * @param[in] rss_desc
1072  *   Pointer to mlx5_flow_rss_desc.
1073  * @param[in] action
1074  *   Action configuration.
1075  */
1076 static void
1077 flow_verbs_translate_action_queue(struct mlx5_flow_rss_desc *rss_desc,
1078 				  const struct rte_flow_action *action)
1079 {
1080 	const struct rte_flow_action_queue *queue = action->conf;
1081 
1082 	rss_desc->queue[0] = queue->index;
1083 	rss_desc->queue_num = 1;
1084 }
1085 
1086 /**
1087  * Convert the @p action into a Verbs specification. This function assumes that
1088  * the input is valid and that there is space to insert the requested action
1089  * into the flow.
1090  *
1091  * @param[in] rss_desc
1092  *   Pointer to mlx5_flow_rss_desc.
1093  * @param[in] action
1094  *   Action configuration.
1095  */
1096 static void
1097 flow_verbs_translate_action_rss(struct mlx5_flow_rss_desc *rss_desc,
1098 				const struct rte_flow_action *action)
1099 {
1100 	const struct rte_flow_action_rss *rss = action->conf;
1101 	const uint8_t *rss_key;
1102 
1103 	memcpy(rss_desc->queue, rss->queue, rss->queue_num * sizeof(uint16_t));
1104 	rss_desc->queue_num = rss->queue_num;
1105 	/* NULL RSS key indicates default RSS key. */
1106 	rss_key = !rss->key ? rss_hash_default_key : rss->key;
1107 	memcpy(rss_desc->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
1108 	/*
1109 	 * rss->level and rss.types should be set in advance when expanding
1110 	 * items for RSS.
1111 	 */
1112 }
1113 
1114 /**
1115  * Convert the @p action into a Verbs specification. This function assumes that
1116  * the input is valid and that there is space to insert the requested action
1117  * into the flow.
1118  *
1119  * @param[in] dev_flow
1120  *   Pointer to mlx5_flow.
1121  * @param[in] action
1122  *   Action configuration.
1123  */
1124 static void
1125 flow_verbs_translate_action_flag
1126 	(struct mlx5_flow *dev_flow,
1127 	 const struct rte_flow_action *action __rte_unused)
1128 {
1129 	unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
1130 	struct ibv_flow_spec_action_tag tag = {
1131 		.type = IBV_FLOW_SPEC_ACTION_TAG,
1132 		.size = size,
1133 		.tag_id = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT),
1134 	};
1135 
1136 	flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
1137 }
1138 
1139 /**
1140  * Convert the @p action into a Verbs specification. This function assumes that
1141  * the input is valid and that there is space to insert the requested action
1142  * into the flow.
1143  *
1144  * @param[in] dev_flow
1145  *   Pointer to mlx5_flow.
1146  * @param[in] action
1147  *   Action configuration.
1148  */
1149 static void
1150 flow_verbs_translate_action_mark(struct mlx5_flow *dev_flow,
1151 				 const struct rte_flow_action *action)
1152 {
1153 	const struct rte_flow_action_mark *mark = action->conf;
1154 	unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
1155 	struct ibv_flow_spec_action_tag tag = {
1156 		.type = IBV_FLOW_SPEC_ACTION_TAG,
1157 		.size = size,
1158 		.tag_id = mlx5_flow_mark_set(mark->id),
1159 	};
1160 
1161 	flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
1162 }
1163 
1164 /**
1165  * Convert the @p action into a Verbs specification. This function assumes that
1166  * the input is valid and that there is space to insert the requested action
1167  * into the flow.
1168  *
1169  * @param[in] dev
1170  *   Pointer to the Ethernet device structure.
1171  * @param[in] action
1172  *   Action configuration.
1173  * @param[in] dev_flow
1174  *   Pointer to mlx5_flow.
1175  * @param[out] error
1176  *   Pointer to error structure.
1177  *
1178  * @return
1179  *   0 On success else a negative errno value is returned and rte_errno is set.
1180  */
1181 static int
1182 flow_verbs_translate_action_count(struct mlx5_flow *dev_flow,
1183 				  const struct rte_flow_action *action,
1184 				  struct rte_eth_dev *dev,
1185 				  struct rte_flow_error *error)
1186 {
1187 	const struct rte_flow_action_count *count = action->conf;
1188 	struct rte_flow *flow = dev_flow->flow;
1189 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
1190 	defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1191 	struct mlx5_flow_counter_pool *pool;
1192 	struct mlx5_flow_counter *cnt = NULL;
1193 	unsigned int size = sizeof(struct ibv_flow_spec_counter_action);
1194 	struct ibv_flow_spec_counter_action counter = {
1195 		.type = IBV_FLOW_SPEC_ACTION_COUNT,
1196 		.size = size,
1197 	};
1198 #endif
1199 
1200 	if (!flow->counter) {
1201 		flow->counter = flow_verbs_counter_new(dev, count->shared,
1202 						       count->id);
1203 		if (!flow->counter)
1204 			return rte_flow_error_set(error, rte_errno,
1205 						  RTE_FLOW_ERROR_TYPE_ACTION,
1206 						  action,
1207 						  "cannot get counter"
1208 						  " context.");
1209 	}
1210 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
1211 	cnt = flow_verbs_counter_get_by_idx(dev, flow->counter, &pool);
1212 	counter.counter_set_handle =
1213 		((struct ibv_counter_set *)cnt->dcs_when_active)->handle;
1214 	flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
1215 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1216 	cnt = flow_verbs_counter_get_by_idx(dev, flow->counter, &pool);
1217 	counter.counters = (struct ibv_counters *)cnt->dcs_when_active;
1218 	flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
1219 #endif
1220 	return 0;
1221 }
1222 
1223 /**
1224  * Internal validation function. For validating both actions and items.
1225  *
1226  * @param[in] dev
1227  *   Pointer to the Ethernet device structure.
1228  * @param[in] attr
1229  *   Pointer to the flow attributes.
1230  * @param[in] items
1231  *   Pointer to the list of items.
1232  * @param[in] actions
1233  *   Pointer to the list of actions.
1234  * @param[in] external
1235  *   This flow rule is created by request external to PMD.
1236  * @param[in] hairpin
1237  *   Number of hairpin TX actions, 0 means classic flow.
1238  * @param[out] error
1239  *   Pointer to the error structure.
1240  *
1241  * @return
1242  *   0 on success, a negative errno value otherwise and rte_errno is set.
1243  */
1244 static int
1245 flow_verbs_validate(struct rte_eth_dev *dev,
1246 		    const struct rte_flow_attr *attr,
1247 		    const struct rte_flow_item items[],
1248 		    const struct rte_flow_action actions[],
1249 		    bool external __rte_unused,
1250 		    int hairpin __rte_unused,
1251 		    struct rte_flow_error *error)
1252 {
1253 	int ret;
1254 	uint64_t action_flags = 0;
1255 	uint64_t item_flags = 0;
1256 	uint64_t last_item = 0;
1257 	uint8_t next_protocol = 0xff;
1258 	uint16_t ether_type = 0;
1259 	bool is_empty_vlan = false;
1260 	uint16_t udp_dport = 0;
1261 
1262 	if (items == NULL)
1263 		return -1;
1264 	ret = mlx5_flow_validate_attributes(dev, attr, error);
1265 	if (ret < 0)
1266 		return ret;
1267 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1268 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1269 		int ret = 0;
1270 
1271 		switch (items->type) {
1272 		case RTE_FLOW_ITEM_TYPE_VOID:
1273 			break;
1274 		case RTE_FLOW_ITEM_TYPE_ETH:
1275 			ret = mlx5_flow_validate_item_eth(items, item_flags,
1276 							  false, error);
1277 			if (ret < 0)
1278 				return ret;
1279 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1280 					     MLX5_FLOW_LAYER_OUTER_L2;
1281 			if (items->mask != NULL && items->spec != NULL) {
1282 				ether_type =
1283 					((const struct rte_flow_item_eth *)
1284 					 items->spec)->type;
1285 				ether_type &=
1286 					((const struct rte_flow_item_eth *)
1287 					 items->mask)->type;
1288 				if (ether_type == RTE_BE16(RTE_ETHER_TYPE_VLAN))
1289 					is_empty_vlan = true;
1290 				ether_type = rte_be_to_cpu_16(ether_type);
1291 			} else {
1292 				ether_type = 0;
1293 			}
1294 			break;
1295 		case RTE_FLOW_ITEM_TYPE_VLAN:
1296 			ret = mlx5_flow_validate_item_vlan(items, item_flags,
1297 							   dev, error);
1298 			if (ret < 0)
1299 				return ret;
1300 			last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1301 					      MLX5_FLOW_LAYER_INNER_VLAN) :
1302 					     (MLX5_FLOW_LAYER_OUTER_L2 |
1303 					      MLX5_FLOW_LAYER_OUTER_VLAN);
1304 			if (items->mask != NULL && items->spec != NULL) {
1305 				ether_type =
1306 					((const struct rte_flow_item_vlan *)
1307 					 items->spec)->inner_type;
1308 				ether_type &=
1309 					((const struct rte_flow_item_vlan *)
1310 					 items->mask)->inner_type;
1311 				ether_type = rte_be_to_cpu_16(ether_type);
1312 			} else {
1313 				ether_type = 0;
1314 			}
1315 			is_empty_vlan = false;
1316 			break;
1317 		case RTE_FLOW_ITEM_TYPE_IPV4:
1318 			ret = mlx5_flow_validate_item_ipv4
1319 						(items, item_flags,
1320 						 last_item, ether_type, NULL,
1321 						 MLX5_ITEM_RANGE_NOT_ACCEPTED,
1322 						 error);
1323 			if (ret < 0)
1324 				return ret;
1325 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1326 					     MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1327 			if (items->mask != NULL &&
1328 			    ((const struct rte_flow_item_ipv4 *)
1329 			     items->mask)->hdr.next_proto_id) {
1330 				next_protocol =
1331 					((const struct rte_flow_item_ipv4 *)
1332 					 (items->spec))->hdr.next_proto_id;
1333 				next_protocol &=
1334 					((const struct rte_flow_item_ipv4 *)
1335 					 (items->mask))->hdr.next_proto_id;
1336 			} else {
1337 				/* Reset for inner layer. */
1338 				next_protocol = 0xff;
1339 			}
1340 			break;
1341 		case RTE_FLOW_ITEM_TYPE_IPV6:
1342 			ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1343 							   last_item,
1344 							   ether_type, NULL,
1345 							   error);
1346 			if (ret < 0)
1347 				return ret;
1348 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1349 					     MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1350 			if (items->mask != NULL &&
1351 			    ((const struct rte_flow_item_ipv6 *)
1352 			     items->mask)->hdr.proto) {
1353 				next_protocol =
1354 					((const struct rte_flow_item_ipv6 *)
1355 					 items->spec)->hdr.proto;
1356 				next_protocol &=
1357 					((const struct rte_flow_item_ipv6 *)
1358 					 items->mask)->hdr.proto;
1359 			} else {
1360 				/* Reset for inner layer. */
1361 				next_protocol = 0xff;
1362 			}
1363 			break;
1364 		case RTE_FLOW_ITEM_TYPE_UDP:
1365 			ret = mlx5_flow_validate_item_udp(items, item_flags,
1366 							  next_protocol,
1367 							  error);
1368 			const struct rte_flow_item_udp *spec = items->spec;
1369 			const struct rte_flow_item_udp *mask = items->mask;
1370 			if (!mask)
1371 				mask = &rte_flow_item_udp_mask;
1372 			if (spec != NULL)
1373 				udp_dport = rte_be_to_cpu_16
1374 						(spec->hdr.dst_port &
1375 						 mask->hdr.dst_port);
1376 
1377 			if (ret < 0)
1378 				return ret;
1379 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1380 					     MLX5_FLOW_LAYER_OUTER_L4_UDP;
1381 			break;
1382 		case RTE_FLOW_ITEM_TYPE_TCP:
1383 			ret = mlx5_flow_validate_item_tcp
1384 						(items, item_flags,
1385 						 next_protocol,
1386 						 &rte_flow_item_tcp_mask,
1387 						 error);
1388 			if (ret < 0)
1389 				return ret;
1390 			last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1391 					     MLX5_FLOW_LAYER_OUTER_L4_TCP;
1392 			break;
1393 		case RTE_FLOW_ITEM_TYPE_VXLAN:
1394 			ret = mlx5_flow_validate_item_vxlan(dev, udp_dport,
1395 							    items, item_flags,
1396 							    attr, error);
1397 			if (ret < 0)
1398 				return ret;
1399 			last_item = MLX5_FLOW_LAYER_VXLAN;
1400 			break;
1401 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1402 			ret = mlx5_flow_validate_item_vxlan_gpe(items,
1403 								item_flags,
1404 								dev, error);
1405 			if (ret < 0)
1406 				return ret;
1407 			last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
1408 			break;
1409 		case RTE_FLOW_ITEM_TYPE_GRE:
1410 			ret = mlx5_flow_validate_item_gre(items, item_flags,
1411 							  next_protocol, error);
1412 			if (ret < 0)
1413 				return ret;
1414 			last_item = MLX5_FLOW_LAYER_GRE;
1415 			break;
1416 		case RTE_FLOW_ITEM_TYPE_MPLS:
1417 			ret = mlx5_flow_validate_item_mpls(dev, items,
1418 							   item_flags,
1419 							   last_item, error);
1420 			if (ret < 0)
1421 				return ret;
1422 			last_item = MLX5_FLOW_LAYER_MPLS;
1423 			break;
1424 		case RTE_FLOW_ITEM_TYPE_ICMP:
1425 		case RTE_FLOW_ITEM_TYPE_ICMP6:
1426 			return rte_flow_error_set(error, ENOTSUP,
1427 						  RTE_FLOW_ERROR_TYPE_ITEM,
1428 						  NULL, "ICMP/ICMP6 "
1429 						  "item not supported");
1430 		default:
1431 			return rte_flow_error_set(error, ENOTSUP,
1432 						  RTE_FLOW_ERROR_TYPE_ITEM,
1433 						  NULL, "item not supported");
1434 		}
1435 		item_flags |= last_item;
1436 	}
1437 	if (is_empty_vlan)
1438 		return rte_flow_error_set(error, ENOTSUP,
1439 						 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
1440 		    "VLAN matching without vid specification is not supported");
1441 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1442 		switch (actions->type) {
1443 		case RTE_FLOW_ACTION_TYPE_VOID:
1444 			break;
1445 		case RTE_FLOW_ACTION_TYPE_FLAG:
1446 			ret = mlx5_flow_validate_action_flag(action_flags,
1447 							     attr,
1448 							     error);
1449 			if (ret < 0)
1450 				return ret;
1451 			action_flags |= MLX5_FLOW_ACTION_FLAG;
1452 			break;
1453 		case RTE_FLOW_ACTION_TYPE_MARK:
1454 			ret = mlx5_flow_validate_action_mark(actions,
1455 							     action_flags,
1456 							     attr,
1457 							     error);
1458 			if (ret < 0)
1459 				return ret;
1460 			action_flags |= MLX5_FLOW_ACTION_MARK;
1461 			break;
1462 		case RTE_FLOW_ACTION_TYPE_DROP:
1463 			ret = mlx5_flow_validate_action_drop(action_flags,
1464 							     attr,
1465 							     error);
1466 			if (ret < 0)
1467 				return ret;
1468 			action_flags |= MLX5_FLOW_ACTION_DROP;
1469 			break;
1470 		case RTE_FLOW_ACTION_TYPE_QUEUE:
1471 			ret = mlx5_flow_validate_action_queue(actions,
1472 							      action_flags, dev,
1473 							      attr,
1474 							      error);
1475 			if (ret < 0)
1476 				return ret;
1477 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
1478 			break;
1479 		case RTE_FLOW_ACTION_TYPE_RSS:
1480 			ret = mlx5_flow_validate_action_rss(actions,
1481 							    action_flags, dev,
1482 							    attr, item_flags,
1483 							    error);
1484 			if (ret < 0)
1485 				return ret;
1486 			action_flags |= MLX5_FLOW_ACTION_RSS;
1487 			break;
1488 		case RTE_FLOW_ACTION_TYPE_COUNT:
1489 			ret = mlx5_flow_validate_action_count(dev, attr, error);
1490 			if (ret < 0)
1491 				return ret;
1492 			action_flags |= MLX5_FLOW_ACTION_COUNT;
1493 			break;
1494 		default:
1495 			return rte_flow_error_set(error, ENOTSUP,
1496 						  RTE_FLOW_ERROR_TYPE_ACTION,
1497 						  actions,
1498 						  "action not supported");
1499 		}
1500 	}
1501 	/*
1502 	 * Validate the drop action mutual exclusion with other actions.
1503 	 * Drop action is mutually-exclusive with any other action, except for
1504 	 * Count action.
1505 	 */
1506 	if ((action_flags & MLX5_FLOW_ACTION_DROP) &&
1507 	    (action_flags & ~(MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_COUNT)))
1508 		return rte_flow_error_set(error, EINVAL,
1509 					  RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1510 					  "Drop action is mutually-exclusive "
1511 					  "with any other action, except for "
1512 					  "Count action");
1513 	if (!(action_flags & MLX5_FLOW_FATE_ACTIONS))
1514 		return rte_flow_error_set(error, EINVAL,
1515 					  RTE_FLOW_ERROR_TYPE_ACTION, actions,
1516 					  "no fate action is found");
1517 	return 0;
1518 }
1519 
1520 /**
1521  * Calculate the required bytes that are needed for the action part of the verbs
1522  * flow.
1523  *
1524  * @param[in] actions
1525  *   Pointer to the list of actions.
1526  *
1527  * @return
1528  *   The size of the memory needed for all actions.
1529  */
1530 static int
1531 flow_verbs_get_actions_size(const struct rte_flow_action actions[])
1532 {
1533 	int size = 0;
1534 
1535 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1536 		switch (actions->type) {
1537 		case RTE_FLOW_ACTION_TYPE_VOID:
1538 			break;
1539 		case RTE_FLOW_ACTION_TYPE_FLAG:
1540 			size += sizeof(struct ibv_flow_spec_action_tag);
1541 			break;
1542 		case RTE_FLOW_ACTION_TYPE_MARK:
1543 			size += sizeof(struct ibv_flow_spec_action_tag);
1544 			break;
1545 		case RTE_FLOW_ACTION_TYPE_DROP:
1546 			size += sizeof(struct ibv_flow_spec_action_drop);
1547 			break;
1548 		case RTE_FLOW_ACTION_TYPE_QUEUE:
1549 			break;
1550 		case RTE_FLOW_ACTION_TYPE_RSS:
1551 			break;
1552 		case RTE_FLOW_ACTION_TYPE_COUNT:
1553 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
1554 	defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1555 			size += sizeof(struct ibv_flow_spec_counter_action);
1556 #endif
1557 			break;
1558 		default:
1559 			break;
1560 		}
1561 	}
1562 	return size;
1563 }
1564 
1565 /**
1566  * Calculate the required bytes that are needed for the item part of the verbs
1567  * flow.
1568  *
1569  * @param[in] items
1570  *   Pointer to the list of items.
1571  *
1572  * @return
1573  *   The size of the memory needed for all items.
1574  */
1575 static int
1576 flow_verbs_get_items_size(const struct rte_flow_item items[])
1577 {
1578 	int size = 0;
1579 
1580 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1581 		switch (items->type) {
1582 		case RTE_FLOW_ITEM_TYPE_VOID:
1583 			break;
1584 		case RTE_FLOW_ITEM_TYPE_ETH:
1585 			size += sizeof(struct ibv_flow_spec_eth);
1586 			break;
1587 		case RTE_FLOW_ITEM_TYPE_VLAN:
1588 			size += sizeof(struct ibv_flow_spec_eth);
1589 			break;
1590 		case RTE_FLOW_ITEM_TYPE_IPV4:
1591 			size += sizeof(struct ibv_flow_spec_ipv4_ext);
1592 			break;
1593 		case RTE_FLOW_ITEM_TYPE_IPV6:
1594 			size += sizeof(struct ibv_flow_spec_ipv6);
1595 			break;
1596 		case RTE_FLOW_ITEM_TYPE_UDP:
1597 			size += sizeof(struct ibv_flow_spec_tcp_udp);
1598 			break;
1599 		case RTE_FLOW_ITEM_TYPE_TCP:
1600 			size += sizeof(struct ibv_flow_spec_tcp_udp);
1601 			break;
1602 		case RTE_FLOW_ITEM_TYPE_VXLAN:
1603 			size += sizeof(struct ibv_flow_spec_tunnel);
1604 			break;
1605 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1606 			size += sizeof(struct ibv_flow_spec_tunnel);
1607 			break;
1608 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1609 		case RTE_FLOW_ITEM_TYPE_GRE:
1610 			size += sizeof(struct ibv_flow_spec_gre);
1611 			break;
1612 		case RTE_FLOW_ITEM_TYPE_MPLS:
1613 			size += sizeof(struct ibv_flow_spec_mpls);
1614 			break;
1615 #else
1616 		case RTE_FLOW_ITEM_TYPE_GRE:
1617 			size += sizeof(struct ibv_flow_spec_tunnel);
1618 			break;
1619 #endif
1620 		default:
1621 			break;
1622 		}
1623 	}
1624 	return size;
1625 }
1626 
1627 /**
1628  * Internal preparation function. Allocate mlx5_flow with the required size.
1629  * The required size is calculate based on the actions and items. This function
1630  * also returns the detected actions and items for later use.
1631  *
1632  * @param[in] dev
1633  *   Pointer to Ethernet device.
1634  * @param[in] attr
1635  *   Pointer to the flow attributes.
1636  * @param[in] items
1637  *   Pointer to the list of items.
1638  * @param[in] actions
1639  *   Pointer to the list of actions.
1640  * @param[out] error
1641  *   Pointer to the error structure.
1642  *
1643  * @return
1644  *   Pointer to mlx5_flow object on success, otherwise NULL and rte_errno
1645  *   is set.
1646  */
1647 static struct mlx5_flow *
1648 flow_verbs_prepare(struct rte_eth_dev *dev,
1649 		   const struct rte_flow_attr *attr __rte_unused,
1650 		   const struct rte_flow_item items[],
1651 		   const struct rte_flow_action actions[],
1652 		   struct rte_flow_error *error)
1653 {
1654 	size_t size = 0;
1655 	uint32_t handle_idx = 0;
1656 	struct mlx5_flow *dev_flow;
1657 	struct mlx5_flow_handle *dev_handle;
1658 	struct mlx5_priv *priv = dev->data->dev_private;
1659 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1660 
1661 	MLX5_ASSERT(wks);
1662 	size += flow_verbs_get_actions_size(actions);
1663 	size += flow_verbs_get_items_size(items);
1664 	if (size > MLX5_VERBS_MAX_SPEC_ACT_SIZE) {
1665 		rte_flow_error_set(error, E2BIG,
1666 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1667 				   "Verbs spec/action size too large");
1668 		return NULL;
1669 	}
1670 	/* In case of corrupting the memory. */
1671 	if (wks->flow_idx >= MLX5_NUM_MAX_DEV_FLOWS) {
1672 		rte_flow_error_set(error, ENOSPC,
1673 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1674 				   "not free temporary device flow");
1675 		return NULL;
1676 	}
1677 	dev_handle = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
1678 				   &handle_idx);
1679 	if (!dev_handle) {
1680 		rte_flow_error_set(error, ENOMEM,
1681 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1682 				   "not enough memory to create flow handle");
1683 		return NULL;
1684 	}
1685 	MLX5_ASSERT(wks->flow_idx + 1 < RTE_DIM(wks->flows));
1686 	dev_flow = &wks->flows[wks->flow_idx++];
1687 	dev_flow->handle = dev_handle;
1688 	dev_flow->handle_idx = handle_idx;
1689 	/* Memcpy is used, only size needs to be cleared to 0. */
1690 	dev_flow->verbs.size = 0;
1691 	dev_flow->verbs.attr.num_of_specs = 0;
1692 	dev_flow->ingress = attr->ingress;
1693 	dev_flow->hash_fields = 0;
1694 	/* Need to set transfer attribute: not supported in Verbs mode. */
1695 	return dev_flow;
1696 }
1697 
1698 /**
1699  * Fill the flow with verb spec.
1700  *
1701  * @param[in] dev
1702  *   Pointer to Ethernet device.
1703  * @param[in, out] dev_flow
1704  *   Pointer to the mlx5 flow.
1705  * @param[in] attr
1706  *   Pointer to the flow attributes.
1707  * @param[in] items
1708  *   Pointer to the list of items.
1709  * @param[in] actions
1710  *   Pointer to the list of actions.
1711  * @param[out] error
1712  *   Pointer to the error structure.
1713  *
1714  * @return
1715  *   0 on success, else a negative errno value otherwise and rte_errno is set.
1716  */
1717 static int
1718 flow_verbs_translate(struct rte_eth_dev *dev,
1719 		     struct mlx5_flow *dev_flow,
1720 		     const struct rte_flow_attr *attr,
1721 		     const struct rte_flow_item items[],
1722 		     const struct rte_flow_action actions[],
1723 		     struct rte_flow_error *error)
1724 {
1725 	uint64_t item_flags = 0;
1726 	uint64_t action_flags = 0;
1727 	uint64_t priority = attr->priority;
1728 	uint32_t subpriority = 0;
1729 	struct mlx5_priv *priv = dev->data->dev_private;
1730 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1731 	struct mlx5_flow_rss_desc *rss_desc;
1732 
1733 	MLX5_ASSERT(wks);
1734 	rss_desc = &wks->rss_desc;
1735 	if (priority == MLX5_FLOW_LOWEST_PRIO_INDICATOR)
1736 		priority = priv->config.flow_prio - 1;
1737 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1738 		int ret;
1739 
1740 		switch (actions->type) {
1741 		case RTE_FLOW_ACTION_TYPE_VOID:
1742 			break;
1743 		case RTE_FLOW_ACTION_TYPE_FLAG:
1744 			flow_verbs_translate_action_flag(dev_flow, actions);
1745 			action_flags |= MLX5_FLOW_ACTION_FLAG;
1746 			dev_flow->handle->mark = 1;
1747 			break;
1748 		case RTE_FLOW_ACTION_TYPE_MARK:
1749 			flow_verbs_translate_action_mark(dev_flow, actions);
1750 			action_flags |= MLX5_FLOW_ACTION_MARK;
1751 			dev_flow->handle->mark = 1;
1752 			break;
1753 		case RTE_FLOW_ACTION_TYPE_DROP:
1754 			flow_verbs_translate_action_drop(dev_flow, actions);
1755 			action_flags |= MLX5_FLOW_ACTION_DROP;
1756 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_DROP;
1757 			break;
1758 		case RTE_FLOW_ACTION_TYPE_QUEUE:
1759 			flow_verbs_translate_action_queue(rss_desc, actions);
1760 			action_flags |= MLX5_FLOW_ACTION_QUEUE;
1761 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
1762 			break;
1763 		case RTE_FLOW_ACTION_TYPE_RSS:
1764 			flow_verbs_translate_action_rss(rss_desc, actions);
1765 			action_flags |= MLX5_FLOW_ACTION_RSS;
1766 			dev_flow->handle->fate_action = MLX5_FLOW_FATE_QUEUE;
1767 			break;
1768 		case RTE_FLOW_ACTION_TYPE_COUNT:
1769 			ret = flow_verbs_translate_action_count(dev_flow,
1770 								actions,
1771 								dev, error);
1772 			if (ret < 0)
1773 				return ret;
1774 			action_flags |= MLX5_FLOW_ACTION_COUNT;
1775 			break;
1776 		default:
1777 			return rte_flow_error_set(error, ENOTSUP,
1778 						  RTE_FLOW_ERROR_TYPE_ACTION,
1779 						  actions,
1780 						  "action not supported");
1781 		}
1782 	}
1783 	dev_flow->act_flags = action_flags;
1784 	for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1785 		int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1786 
1787 		switch (items->type) {
1788 		case RTE_FLOW_ITEM_TYPE_VOID:
1789 			break;
1790 		case RTE_FLOW_ITEM_TYPE_ETH:
1791 			flow_verbs_translate_item_eth(dev_flow, items,
1792 						      item_flags);
1793 			subpriority = MLX5_PRIORITY_MAP_L2;
1794 			item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1795 					       MLX5_FLOW_LAYER_OUTER_L2;
1796 			break;
1797 		case RTE_FLOW_ITEM_TYPE_VLAN:
1798 			flow_verbs_translate_item_vlan(dev_flow, items,
1799 						       item_flags);
1800 			subpriority = MLX5_PRIORITY_MAP_L2;
1801 			item_flags |= tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1802 						MLX5_FLOW_LAYER_INNER_VLAN) :
1803 					       (MLX5_FLOW_LAYER_OUTER_L2 |
1804 						MLX5_FLOW_LAYER_OUTER_VLAN);
1805 			break;
1806 		case RTE_FLOW_ITEM_TYPE_IPV4:
1807 			flow_verbs_translate_item_ipv4(dev_flow, items,
1808 						       item_flags);
1809 			subpriority = MLX5_PRIORITY_MAP_L3;
1810 			dev_flow->hash_fields |=
1811 				mlx5_flow_hashfields_adjust
1812 					(rss_desc, tunnel,
1813 					 MLX5_IPV4_LAYER_TYPES,
1814 					 MLX5_IPV4_IBV_RX_HASH);
1815 			item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1816 					       MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1817 			break;
1818 		case RTE_FLOW_ITEM_TYPE_IPV6:
1819 			flow_verbs_translate_item_ipv6(dev_flow, items,
1820 						       item_flags);
1821 			subpriority = MLX5_PRIORITY_MAP_L3;
1822 			dev_flow->hash_fields |=
1823 				mlx5_flow_hashfields_adjust
1824 					(rss_desc, tunnel,
1825 					 MLX5_IPV6_LAYER_TYPES,
1826 					 MLX5_IPV6_IBV_RX_HASH);
1827 			item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1828 					       MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1829 			break;
1830 		case RTE_FLOW_ITEM_TYPE_TCP:
1831 			flow_verbs_translate_item_tcp(dev_flow, items,
1832 						      item_flags);
1833 			subpriority = MLX5_PRIORITY_MAP_L4;
1834 			if (dev_flow->hash_fields != 0)
1835 				dev_flow->hash_fields |=
1836 					mlx5_flow_hashfields_adjust
1837 					(rss_desc, tunnel, ETH_RSS_TCP,
1838 					 (IBV_RX_HASH_SRC_PORT_TCP |
1839 					  IBV_RX_HASH_DST_PORT_TCP));
1840 			item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1841 					       MLX5_FLOW_LAYER_OUTER_L4_TCP;
1842 			break;
1843 		case RTE_FLOW_ITEM_TYPE_UDP:
1844 			flow_verbs_translate_item_udp(dev_flow, items,
1845 						      item_flags);
1846 			subpriority = MLX5_PRIORITY_MAP_L4;
1847 			if (dev_flow->hash_fields != 0)
1848 				dev_flow->hash_fields |=
1849 					mlx5_flow_hashfields_adjust
1850 					(rss_desc, tunnel, ETH_RSS_UDP,
1851 					 (IBV_RX_HASH_SRC_PORT_UDP |
1852 					  IBV_RX_HASH_DST_PORT_UDP));
1853 			item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1854 					       MLX5_FLOW_LAYER_OUTER_L4_UDP;
1855 			break;
1856 		case RTE_FLOW_ITEM_TYPE_VXLAN:
1857 			flow_verbs_translate_item_vxlan(dev_flow, items,
1858 							item_flags);
1859 			subpriority = MLX5_TUNNEL_PRIO_GET(rss_desc);
1860 			item_flags |= MLX5_FLOW_LAYER_VXLAN;
1861 			break;
1862 		case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1863 			flow_verbs_translate_item_vxlan_gpe(dev_flow, items,
1864 							    item_flags);
1865 			subpriority = MLX5_TUNNEL_PRIO_GET(rss_desc);
1866 			item_flags |= MLX5_FLOW_LAYER_VXLAN_GPE;
1867 			break;
1868 		case RTE_FLOW_ITEM_TYPE_GRE:
1869 			flow_verbs_translate_item_gre(dev_flow, items,
1870 						      item_flags);
1871 			subpriority = MLX5_TUNNEL_PRIO_GET(rss_desc);
1872 			item_flags |= MLX5_FLOW_LAYER_GRE;
1873 			break;
1874 		case RTE_FLOW_ITEM_TYPE_MPLS:
1875 			flow_verbs_translate_item_mpls(dev_flow, items,
1876 						       item_flags);
1877 			subpriority = MLX5_TUNNEL_PRIO_GET(rss_desc);
1878 			item_flags |= MLX5_FLOW_LAYER_MPLS;
1879 			break;
1880 		default:
1881 			return rte_flow_error_set(error, ENOTSUP,
1882 						  RTE_FLOW_ERROR_TYPE_ITEM,
1883 						  NULL, "item not supported");
1884 		}
1885 	}
1886 	dev_flow->handle->layers = item_flags;
1887 	/* Other members of attr will be ignored. */
1888 	dev_flow->verbs.attr.priority =
1889 		mlx5_flow_adjust_priority(dev, priority, subpriority);
1890 	dev_flow->verbs.attr.port = (uint8_t)priv->dev_port;
1891 	return 0;
1892 }
1893 
1894 /**
1895  * Remove the flow from the NIC but keeps it in memory.
1896  *
1897  * @param[in] dev
1898  *   Pointer to the Ethernet device structure.
1899  * @param[in, out] flow
1900  *   Pointer to flow structure.
1901  */
1902 static void
1903 flow_verbs_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1904 {
1905 	struct mlx5_priv *priv = dev->data->dev_private;
1906 	struct mlx5_flow_handle *handle;
1907 	uint32_t handle_idx;
1908 
1909 	if (!flow)
1910 		return;
1911 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
1912 		       handle_idx, handle, next) {
1913 		if (handle->drv_flow) {
1914 			claim_zero(mlx5_glue->destroy_flow(handle->drv_flow));
1915 			handle->drv_flow = NULL;
1916 		}
1917 		/* hrxq is union, don't touch it only the flag is set. */
1918 		if (handle->rix_hrxq &&
1919 		    handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
1920 			mlx5_hrxq_release(dev, handle->rix_hrxq);
1921 			handle->rix_hrxq = 0;
1922 		}
1923 		if (handle->vf_vlan.tag && handle->vf_vlan.created)
1924 			mlx5_vlan_vmwa_release(dev, &handle->vf_vlan);
1925 	}
1926 }
1927 
1928 /**
1929  * Remove the flow from the NIC and the memory.
1930  *
1931  * @param[in] dev
1932  *   Pointer to the Ethernet device structure.
1933  * @param[in, out] flow
1934  *   Pointer to flow structure.
1935  */
1936 static void
1937 flow_verbs_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1938 {
1939 	struct mlx5_priv *priv = dev->data->dev_private;
1940 	struct mlx5_flow_handle *handle;
1941 
1942 	if (!flow)
1943 		return;
1944 	flow_verbs_remove(dev, flow);
1945 	while (flow->dev_handles) {
1946 		uint32_t tmp_idx = flow->dev_handles;
1947 
1948 		handle = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
1949 				   tmp_idx);
1950 		if (!handle)
1951 			return;
1952 		flow->dev_handles = handle->next.next;
1953 		mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],
1954 			   tmp_idx);
1955 	}
1956 	if (flow->counter) {
1957 		flow_verbs_counter_release(dev, flow->counter);
1958 		flow->counter = 0;
1959 	}
1960 }
1961 
1962 /**
1963  * Apply the flow to the NIC.
1964  *
1965  * @param[in] dev
1966  *   Pointer to the Ethernet device structure.
1967  * @param[in, out] flow
1968  *   Pointer to flow structure.
1969  * @param[out] error
1970  *   Pointer to error structure.
1971  *
1972  * @return
1973  *   0 on success, a negative errno value otherwise and rte_errno is set.
1974  */
1975 static int
1976 flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1977 		 struct rte_flow_error *error)
1978 {
1979 	struct mlx5_priv *priv = dev->data->dev_private;
1980 	struct mlx5_flow_handle *handle;
1981 	struct mlx5_flow *dev_flow;
1982 	struct mlx5_hrxq *hrxq;
1983 	uint32_t dev_handles;
1984 	int err;
1985 	int idx;
1986 	struct mlx5_flow_workspace *wks = mlx5_flow_get_thread_workspace();
1987 
1988 	MLX5_ASSERT(wks);
1989 	for (idx = wks->flow_idx - 1; idx >= 0; idx--) {
1990 		dev_flow = &wks->flows[idx];
1991 		handle = dev_flow->handle;
1992 		if (handle->fate_action == MLX5_FLOW_FATE_DROP) {
1993 			MLX5_ASSERT(priv->drop_queue.hrxq);
1994 			hrxq = priv->drop_queue.hrxq;
1995 		} else {
1996 			uint32_t hrxq_idx;
1997 			struct mlx5_flow_rss_desc *rss_desc = &wks->rss_desc;
1998 
1999 			MLX5_ASSERT(rss_desc->queue_num);
2000 			rss_desc->key_len = MLX5_RSS_HASH_KEY_LEN;
2001 			rss_desc->hash_fields = dev_flow->hash_fields;
2002 			rss_desc->tunnel = !!(handle->layers &
2003 					      MLX5_FLOW_LAYER_TUNNEL);
2004 			rss_desc->shared_rss = 0;
2005 			hrxq_idx = mlx5_hrxq_get(dev, rss_desc);
2006 			hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
2007 					      hrxq_idx);
2008 			if (!hrxq) {
2009 				rte_flow_error_set
2010 					(error, rte_errno,
2011 					 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2012 					 "cannot get hash queue");
2013 				goto error;
2014 			}
2015 			handle->rix_hrxq = hrxq_idx;
2016 		}
2017 		MLX5_ASSERT(hrxq);
2018 		handle->drv_flow = mlx5_glue->create_flow
2019 					(hrxq->qp, &dev_flow->verbs.attr);
2020 		if (!handle->drv_flow) {
2021 			rte_flow_error_set(error, errno,
2022 					   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2023 					   NULL,
2024 					   "hardware refuses to create flow");
2025 			goto error;
2026 		}
2027 		if (priv->vmwa_context &&
2028 		    handle->vf_vlan.tag && !handle->vf_vlan.created) {
2029 			/*
2030 			 * The rule contains the VLAN pattern.
2031 			 * For VF we are going to create VLAN
2032 			 * interface to make hypervisor set correct
2033 			 * e-Switch vport context.
2034 			 */
2035 			mlx5_vlan_vmwa_acquire(dev, &handle->vf_vlan);
2036 		}
2037 	}
2038 	return 0;
2039 error:
2040 	err = rte_errno; /* Save rte_errno before cleanup. */
2041 	SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2042 		       dev_handles, handle, next) {
2043 		/* hrxq is union, don't touch it only the flag is set. */
2044 		if (handle->rix_hrxq &&
2045 		    handle->fate_action == MLX5_FLOW_FATE_QUEUE) {
2046 			mlx5_hrxq_release(dev, handle->rix_hrxq);
2047 			handle->rix_hrxq = 0;
2048 		}
2049 		if (handle->vf_vlan.tag && handle->vf_vlan.created)
2050 			mlx5_vlan_vmwa_release(dev, &handle->vf_vlan);
2051 	}
2052 	rte_errno = err; /* Restore rte_errno. */
2053 	return -rte_errno;
2054 }
2055 
2056 /**
2057  * Query a flow.
2058  *
2059  * @see rte_flow_query()
2060  * @see rte_flow_ops
2061  */
2062 static int
2063 flow_verbs_query(struct rte_eth_dev *dev,
2064 		 struct rte_flow *flow,
2065 		 const struct rte_flow_action *actions,
2066 		 void *data,
2067 		 struct rte_flow_error *error)
2068 {
2069 	int ret = -EINVAL;
2070 
2071 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2072 		switch (actions->type) {
2073 		case RTE_FLOW_ACTION_TYPE_VOID:
2074 			break;
2075 		case RTE_FLOW_ACTION_TYPE_COUNT:
2076 			ret = flow_verbs_counter_query(dev, flow, data, error);
2077 			break;
2078 		default:
2079 			return rte_flow_error_set(error, ENOTSUP,
2080 						  RTE_FLOW_ERROR_TYPE_ACTION,
2081 						  actions,
2082 						  "action not supported");
2083 		}
2084 	}
2085 	return ret;
2086 }
2087 
2088 static int
2089 flow_verbs_sync_domain(struct rte_eth_dev *dev, uint32_t domains,
2090 		       uint32_t flags)
2091 {
2092 	RTE_SET_USED(dev);
2093 	RTE_SET_USED(domains);
2094 	RTE_SET_USED(flags);
2095 
2096 	return 0;
2097 }
2098 
2099 const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
2100 	.validate = flow_verbs_validate,
2101 	.prepare = flow_verbs_prepare,
2102 	.translate = flow_verbs_translate,
2103 	.apply = flow_verbs_apply,
2104 	.remove = flow_verbs_remove,
2105 	.destroy = flow_verbs_destroy,
2106 	.query = flow_verbs_query,
2107 	.sync_domain = flow_verbs_sync_domain,
2108 };
2109