xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision 89813a522e68076e6f50ec18b075fa57cc5ae937)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2015 6WIND S.A.
3  * Copyright 2015 Mellanox Technologies, Ltd
4  */
5 
6 #include <stddef.h>
7 #include <unistd.h>
8 #include <string.h>
9 #include <stdint.h>
10 #include <stdlib.h>
11 #include <errno.h>
12 
13 #include <ethdev_driver.h>
14 #include <rte_bus_pci.h>
15 #include <rte_mbuf.h>
16 #include <rte_common.h>
17 #include <rte_interrupts.h>
18 #include <rte_malloc.h>
19 #include <rte_string_fns.h>
20 #include <rte_rwlock.h>
21 #include <rte_cycles.h>
22 
23 #include <mlx5_malloc.h>
24 
25 #include "mlx5_rxtx.h"
26 #include "mlx5_autoconf.h"
27 
28 /**
29  * Get the interface index from device name.
30  *
31  * @param[in] dev
32  *   Pointer to Ethernet device.
33  *
34  * @return
35  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
36  */
37 unsigned int
38 mlx5_ifindex(const struct rte_eth_dev *dev)
39 {
40 	struct mlx5_priv *priv = dev->data->dev_private;
41 	unsigned int ifindex;
42 
43 	MLX5_ASSERT(priv);
44 	MLX5_ASSERT(priv->if_index);
45 	if (priv->master && priv->sh->bond.ifindex > 0)
46 		ifindex = priv->sh->bond.ifindex;
47 	else
48 		ifindex = priv->if_index;
49 	if (!ifindex)
50 		rte_errno = ENXIO;
51 	return ifindex;
52 }
53 
54 /**
55  * DPDK callback for Ethernet device configuration.
56  *
57  * @param dev
58  *   Pointer to Ethernet device structure.
59  *
60  * @return
61  *   0 on success, a negative errno value otherwise and rte_errno is set.
62  */
63 int
64 mlx5_dev_configure(struct rte_eth_dev *dev)
65 {
66 	struct mlx5_priv *priv = dev->data->dev_private;
67 	unsigned int rxqs_n = dev->data->nb_rx_queues;
68 	unsigned int txqs_n = dev->data->nb_tx_queues;
69 	const uint8_t use_app_rss_key =
70 		!!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
71 	int ret = 0;
72 
73 	if (use_app_rss_key &&
74 	    (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
75 	     MLX5_RSS_HASH_KEY_LEN)) {
76 		DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
77 			dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
78 		rte_errno = EINVAL;
79 		return -rte_errno;
80 	}
81 	priv->rss_conf.rss_key =
82 		mlx5_realloc(priv->rss_conf.rss_key, MLX5_MEM_RTE,
83 			    MLX5_RSS_HASH_KEY_LEN, 0, SOCKET_ID_ANY);
84 	if (!priv->rss_conf.rss_key) {
85 		DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
86 			dev->data->port_id, rxqs_n);
87 		rte_errno = ENOMEM;
88 		return -rte_errno;
89 	}
90 
91 	if ((dev->data->dev_conf.txmode.offloads &
92 			DEV_TX_OFFLOAD_SEND_ON_TIMESTAMP) &&
93 			rte_mbuf_dyn_tx_timestamp_register(NULL, NULL) != 0) {
94 		DRV_LOG(ERR, "port %u cannot register Tx timestamp field/flag",
95 			dev->data->port_id);
96 		return -rte_errno;
97 	}
98 	memcpy(priv->rss_conf.rss_key,
99 	       use_app_rss_key ?
100 	       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
101 	       rss_hash_default_key,
102 	       MLX5_RSS_HASH_KEY_LEN);
103 	priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
104 	priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
105 	priv->rxqs = (void *)dev->data->rx_queues;
106 	priv->txqs = (void *)dev->data->tx_queues;
107 	if (txqs_n != priv->txqs_n) {
108 		DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
109 			dev->data->port_id, priv->txqs_n, txqs_n);
110 		priv->txqs_n = txqs_n;
111 	}
112 	if (rxqs_n > priv->config.ind_table_max_size) {
113 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
114 			dev->data->port_id, rxqs_n);
115 		rte_errno = EINVAL;
116 		return -rte_errno;
117 	}
118 	if (rxqs_n != priv->rxqs_n) {
119 		DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
120 			dev->data->port_id, priv->rxqs_n, rxqs_n);
121 		priv->rxqs_n = rxqs_n;
122 	}
123 	priv->skip_default_rss_reta = 0;
124 	ret = mlx5_proc_priv_init(dev);
125 	if (ret)
126 		return ret;
127 	return 0;
128 }
129 
130 /**
131  * Configure default RSS reta.
132  *
133  * @param dev
134  *   Pointer to Ethernet device structure.
135  *
136  * @return
137  *   0 on success, a negative errno value otherwise and rte_errno is set.
138  */
139 int
140 mlx5_dev_configure_rss_reta(struct rte_eth_dev *dev)
141 {
142 	struct mlx5_priv *priv = dev->data->dev_private;
143 	unsigned int rxqs_n = dev->data->nb_rx_queues;
144 	unsigned int i;
145 	unsigned int j;
146 	unsigned int reta_idx_n;
147 	int ret = 0;
148 	unsigned int *rss_queue_arr = NULL;
149 	unsigned int rss_queue_n = 0;
150 
151 	if (priv->skip_default_rss_reta)
152 		return ret;
153 	rss_queue_arr = mlx5_malloc(0, rxqs_n * sizeof(unsigned int), 0,
154 				    SOCKET_ID_ANY);
155 	if (!rss_queue_arr) {
156 		DRV_LOG(ERR, "port %u cannot allocate RSS queue list (%u)",
157 			dev->data->port_id, rxqs_n);
158 		rte_errno = ENOMEM;
159 		return -rte_errno;
160 	}
161 	for (i = 0, j = 0; i < rxqs_n; i++) {
162 		struct mlx5_rxq_data *rxq_data;
163 		struct mlx5_rxq_ctrl *rxq_ctrl;
164 
165 		rxq_data = (*priv->rxqs)[i];
166 		rxq_ctrl = container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
167 		if (rxq_ctrl && rxq_ctrl->type == MLX5_RXQ_TYPE_STANDARD)
168 			rss_queue_arr[j++] = i;
169 	}
170 	rss_queue_n = j;
171 	if (rss_queue_n > priv->config.ind_table_max_size) {
172 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
173 			dev->data->port_id, rss_queue_n);
174 		rte_errno = EINVAL;
175 		mlx5_free(rss_queue_arr);
176 		return -rte_errno;
177 	}
178 	DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
179 		dev->data->port_id, priv->rxqs_n, rxqs_n);
180 	priv->rxqs_n = rxqs_n;
181 	/*
182 	 * If the requested number of RX queues is not a power of two,
183 	 * use the maximum indirection table size for better balancing.
184 	 * The result is always rounded to the next power of two.
185 	 */
186 	reta_idx_n = (1 << log2above((rss_queue_n & (rss_queue_n - 1)) ?
187 				priv->config.ind_table_max_size :
188 				rss_queue_n));
189 	ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
190 	if (ret) {
191 		mlx5_free(rss_queue_arr);
192 		return ret;
193 	}
194 	/*
195 	 * When the number of RX queues is not a power of two,
196 	 * the remaining table entries are padded with reused WQs
197 	 * and hashes are not spread uniformly.
198 	 */
199 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
200 		(*priv->reta_idx)[i] = rss_queue_arr[j];
201 		if (++j == rss_queue_n)
202 			j = 0;
203 	}
204 	mlx5_free(rss_queue_arr);
205 	return ret;
206 }
207 
208 /**
209  * Sets default tuning parameters.
210  *
211  * @param dev
212  *   Pointer to Ethernet device.
213  * @param[out] info
214  *   Info structure output buffer.
215  */
216 static void
217 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
218 {
219 	struct mlx5_priv *priv = dev->data->dev_private;
220 
221 	/* Minimum CPU utilization. */
222 	info->default_rxportconf.ring_size = 256;
223 	info->default_txportconf.ring_size = 256;
224 	info->default_rxportconf.burst_size = MLX5_RX_DEFAULT_BURST;
225 	info->default_txportconf.burst_size = MLX5_TX_DEFAULT_BURST;
226 	if ((priv->link_speed_capa & ETH_LINK_SPEED_200G) |
227 		(priv->link_speed_capa & ETH_LINK_SPEED_100G)) {
228 		info->default_rxportconf.nb_queues = 16;
229 		info->default_txportconf.nb_queues = 16;
230 		if (dev->data->nb_rx_queues > 2 ||
231 		    dev->data->nb_tx_queues > 2) {
232 			/* Max Throughput. */
233 			info->default_rxportconf.ring_size = 2048;
234 			info->default_txportconf.ring_size = 2048;
235 		}
236 	} else {
237 		info->default_rxportconf.nb_queues = 8;
238 		info->default_txportconf.nb_queues = 8;
239 		if (dev->data->nb_rx_queues > 2 ||
240 		    dev->data->nb_tx_queues > 2) {
241 			/* Max Throughput. */
242 			info->default_rxportconf.ring_size = 4096;
243 			info->default_txportconf.ring_size = 4096;
244 		}
245 	}
246 }
247 
248 /**
249  * Sets tx mbuf limiting parameters.
250  *
251  * @param dev
252  *   Pointer to Ethernet device.
253  * @param[out] info
254  *   Info structure output buffer.
255  */
256 static void
257 mlx5_set_txlimit_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
258 {
259 	struct mlx5_priv *priv = dev->data->dev_private;
260 	struct mlx5_dev_config *config = &priv->config;
261 	unsigned int inlen;
262 	uint16_t nb_max;
263 
264 	inlen = (config->txq_inline_max == MLX5_ARG_UNSET) ?
265 		MLX5_SEND_DEF_INLINE_LEN :
266 		(unsigned int)config->txq_inline_max;
267 	MLX5_ASSERT(config->txq_inline_min >= 0);
268 	inlen = RTE_MAX(inlen, (unsigned int)config->txq_inline_min);
269 	inlen = RTE_MIN(inlen, MLX5_WQE_SIZE_MAX +
270 			       MLX5_ESEG_MIN_INLINE_SIZE -
271 			       MLX5_WQE_CSEG_SIZE -
272 			       MLX5_WQE_ESEG_SIZE -
273 			       MLX5_WQE_DSEG_SIZE * 2);
274 	nb_max = (MLX5_WQE_SIZE_MAX +
275 		  MLX5_ESEG_MIN_INLINE_SIZE -
276 		  MLX5_WQE_CSEG_SIZE -
277 		  MLX5_WQE_ESEG_SIZE -
278 		  MLX5_WQE_DSEG_SIZE -
279 		  inlen) / MLX5_WSEG_SIZE;
280 	info->tx_desc_lim.nb_seg_max = nb_max;
281 	info->tx_desc_lim.nb_mtu_seg_max = nb_max;
282 }
283 
284 /**
285  * DPDK callback to get information about the device.
286  *
287  * @param dev
288  *   Pointer to Ethernet device structure.
289  * @param[out] info
290  *   Info structure output buffer.
291  */
292 int
293 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
294 {
295 	struct mlx5_priv *priv = dev->data->dev_private;
296 	struct mlx5_dev_config *config = &priv->config;
297 	unsigned int max;
298 
299 	/* FIXME: we should ask the device for these values. */
300 	info->min_rx_bufsize = 32;
301 	info->max_rx_pktlen = 65536;
302 	info->max_lro_pkt_size = MLX5_MAX_LRO_SIZE;
303 	/*
304 	 * Since we need one CQ per QP, the limit is the minimum number
305 	 * between the two values.
306 	 */
307 	max = RTE_MIN(priv->sh->device_attr.max_cq,
308 		      priv->sh->device_attr.max_qp);
309 	/* max_rx_queues is uint16_t. */
310 	max = RTE_MIN(max, (unsigned int)UINT16_MAX);
311 	info->max_rx_queues = max;
312 	info->max_tx_queues = max;
313 	info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
314 	info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
315 	info->rx_seg_capa.max_nseg = MLX5_MAX_RXQ_NSEG;
316 	info->rx_seg_capa.multi_pools = !config->mprq.enabled;
317 	info->rx_seg_capa.offset_allowed = !config->mprq.enabled;
318 	info->rx_seg_capa.offset_align_log2 = 0;
319 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
320 				 info->rx_queue_offload_capa);
321 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
322 	info->if_index = mlx5_ifindex(dev);
323 	info->reta_size = priv->reta_idx_n ?
324 		priv->reta_idx_n : config->ind_table_max_size;
325 	info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
326 	info->speed_capa = priv->link_speed_capa;
327 	info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
328 	mlx5_set_default_params(dev, info);
329 	mlx5_set_txlimit_params(dev, info);
330 	info->switch_info.name = dev->data->name;
331 	info->switch_info.domain_id = priv->domain_id;
332 	info->switch_info.port_id = priv->representor_id;
333 	if (priv->representor) {
334 		uint16_t port_id;
335 
336 		MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
337 			struct mlx5_priv *opriv =
338 				rte_eth_devices[port_id].data->dev_private;
339 
340 			if (!opriv ||
341 			    opriv->representor ||
342 			    opriv->sh != priv->sh ||
343 			    opriv->domain_id != priv->domain_id)
344 				continue;
345 			/*
346 			 * Override switch name with that of the master
347 			 * device.
348 			 */
349 			info->switch_info.name = opriv->dev_data->name;
350 			break;
351 		}
352 	}
353 	return 0;
354 }
355 
356 /**
357  * Calculate representor ID from port switch info.
358  *
359  * Uint16 representor ID bits definition:
360  *   pf: 2
361  *   type: 2
362  *   vf/sf: 12
363  *
364  * @param info
365  *   Port switch info.
366  * @param hpf_type
367  *   Use this type if port is HPF.
368  *
369  * @return
370  *   Encoded representor ID.
371  */
372 uint16_t
373 mlx5_representor_id_encode(const struct mlx5_switch_info *info,
374 			   enum rte_eth_representor_type hpf_type)
375 {
376 	enum rte_eth_representor_type type = RTE_ETH_REPRESENTOR_VF;
377 	uint16_t repr = info->port_name;
378 
379 	if (info->representor == 0)
380 		return UINT16_MAX;
381 	if (info->name_type == MLX5_PHYS_PORT_NAME_TYPE_PFSF)
382 		type = RTE_ETH_REPRESENTOR_SF;
383 	if (info->name_type == MLX5_PHYS_PORT_NAME_TYPE_PFHPF) {
384 		type = hpf_type;
385 		repr = UINT16_MAX;
386 	}
387 	return MLX5_REPRESENTOR_ID(info->pf_num, type, repr);
388 }
389 
390 /**
391  * DPDK callback to get information about representor.
392  *
393  * Representor ID bits definition:
394  *   vf/sf: 12
395  *   type: 2
396  *   pf: 2
397  *
398  * @param dev
399  *   Pointer to Ethernet device structure.
400  * @param[out] info
401  *   Nullable info structure output buffer.
402  *
403  * @return
404  *   negative on error, or the number of representor ranges.
405  */
406 int
407 mlx5_representor_info_get(struct rte_eth_dev *dev,
408 			  struct rte_eth_representor_info *info)
409 {
410 	struct mlx5_priv *priv = dev->data->dev_private;
411 	int n_type = 4; /* Representor types, VF, HPF@VF, SF and HPF@SF. */
412 	int n_pf = 2; /* Number of PFs. */
413 	int i = 0, pf;
414 
415 	if (info == NULL)
416 		goto out;
417 	info->controller = 0;
418 	info->pf = priv->pf_bond >= 0 ? priv->pf_bond : 0;
419 	for (pf = 0; pf < n_pf; ++pf) {
420 		/* VF range. */
421 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
422 		info->ranges[i].controller = 0;
423 		info->ranges[i].pf = pf;
424 		info->ranges[i].vf = 0;
425 		info->ranges[i].id_base =
426 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
427 		info->ranges[i].id_end =
428 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
429 		snprintf(info->ranges[i].name,
430 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
431 		i++;
432 		/* HPF range of VF type. */
433 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
434 		info->ranges[i].controller = 0;
435 		info->ranges[i].pf = pf;
436 		info->ranges[i].vf = UINT16_MAX;
437 		info->ranges[i].id_base =
438 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
439 		info->ranges[i].id_end =
440 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
441 		snprintf(info->ranges[i].name,
442 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
443 		i++;
444 		/* SF range. */
445 		info->ranges[i].type = RTE_ETH_REPRESENTOR_SF;
446 		info->ranges[i].controller = 0;
447 		info->ranges[i].pf = pf;
448 		info->ranges[i].vf = 0;
449 		info->ranges[i].id_base =
450 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
451 		info->ranges[i].id_end =
452 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
453 		snprintf(info->ranges[i].name,
454 			 sizeof(info->ranges[i].name), "pf%dsf", pf);
455 		i++;
456 		/* HPF range of SF type. */
457 		info->ranges[i].type = RTE_ETH_REPRESENTOR_SF;
458 		info->ranges[i].controller = 0;
459 		info->ranges[i].pf = pf;
460 		info->ranges[i].vf = UINT16_MAX;
461 		info->ranges[i].id_base =
462 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
463 		info->ranges[i].id_end =
464 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
465 		snprintf(info->ranges[i].name,
466 			 sizeof(info->ranges[i].name), "pf%dsf", pf);
467 		i++;
468 	}
469 out:
470 	return n_type * n_pf;
471 }
472 
473 /**
474  * Get firmware version of a device.
475  *
476  * @param dev
477  *   Ethernet device port.
478  * @param fw_ver
479  *   String output allocated by caller.
480  * @param fw_size
481  *   Size of the output string, including terminating null byte.
482  *
483  * @return
484  *   0 on success, or the size of the non truncated string if too big.
485  */
486 int
487 mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
488 {
489 	struct mlx5_priv *priv = dev->data->dev_private;
490 	struct mlx5_dev_attr *attr = &priv->sh->device_attr;
491 	size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
492 
493 	if (fw_size < size)
494 		return size;
495 	if (fw_ver != NULL)
496 		strlcpy(fw_ver, attr->fw_ver, fw_size);
497 	return 0;
498 }
499 
500 /**
501  * Get supported packet types.
502  *
503  * @param dev
504  *   Pointer to Ethernet device structure.
505  *
506  * @return
507  *   A pointer to the supported Packet types array.
508  */
509 const uint32_t *
510 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
511 {
512 	static const uint32_t ptypes[] = {
513 		/* refers to rxq_cq_to_pkt_type() */
514 		RTE_PTYPE_L2_ETHER,
515 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
516 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
517 		RTE_PTYPE_L4_NONFRAG,
518 		RTE_PTYPE_L4_FRAG,
519 		RTE_PTYPE_L4_TCP,
520 		RTE_PTYPE_L4_UDP,
521 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
522 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
523 		RTE_PTYPE_INNER_L4_NONFRAG,
524 		RTE_PTYPE_INNER_L4_FRAG,
525 		RTE_PTYPE_INNER_L4_TCP,
526 		RTE_PTYPE_INNER_L4_UDP,
527 		RTE_PTYPE_UNKNOWN
528 	};
529 
530 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
531 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
532 	    dev->rx_pkt_burst == mlx5_rx_burst_vec ||
533 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq_vec)
534 		return ptypes;
535 	return NULL;
536 }
537 
538 /**
539  * DPDK callback to change the MTU.
540  *
541  * @param dev
542  *   Pointer to Ethernet device structure.
543  * @param in_mtu
544  *   New MTU.
545  *
546  * @return
547  *   0 on success, a negative errno value otherwise and rte_errno is set.
548  */
549 int
550 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
551 {
552 	struct mlx5_priv *priv = dev->data->dev_private;
553 	uint16_t kern_mtu = 0;
554 	int ret;
555 
556 	ret = mlx5_get_mtu(dev, &kern_mtu);
557 	if (ret)
558 		return ret;
559 	/* Set kernel interface MTU first. */
560 	ret = mlx5_set_mtu(dev, mtu);
561 	if (ret)
562 		return ret;
563 	ret = mlx5_get_mtu(dev, &kern_mtu);
564 	if (ret)
565 		return ret;
566 	if (kern_mtu == mtu) {
567 		priv->mtu = mtu;
568 		DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
569 			dev->data->port_id, mtu);
570 		return 0;
571 	}
572 	rte_errno = EAGAIN;
573 	return -rte_errno;
574 }
575 
576 /**
577  * Configure the RX function to use.
578  *
579  * @param dev
580  *   Pointer to private data structure.
581  *
582  * @return
583  *   Pointer to selected Rx burst function.
584  */
585 eth_rx_burst_t
586 mlx5_select_rx_function(struct rte_eth_dev *dev)
587 {
588 	eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
589 
590 	MLX5_ASSERT(dev != NULL);
591 	if (mlx5_check_vec_rx_support(dev) > 0) {
592 		if (mlx5_mprq_enabled(dev)) {
593 			rx_pkt_burst = mlx5_rx_burst_mprq_vec;
594 			DRV_LOG(DEBUG, "port %u selected vectorized"
595 				" MPRQ Rx function", dev->data->port_id);
596 		} else {
597 			rx_pkt_burst = mlx5_rx_burst_vec;
598 			DRV_LOG(DEBUG, "port %u selected vectorized"
599 				" SPRQ Rx function", dev->data->port_id);
600 		}
601 	} else if (mlx5_mprq_enabled(dev)) {
602 		rx_pkt_burst = mlx5_rx_burst_mprq;
603 		DRV_LOG(DEBUG, "port %u selected MPRQ Rx function",
604 			dev->data->port_id);
605 	} else {
606 		DRV_LOG(DEBUG, "port %u selected SPRQ Rx function",
607 			dev->data->port_id);
608 	}
609 	return rx_pkt_burst;
610 }
611 
612 /**
613  * Get the E-Switch parameters by port id.
614  *
615  * @param[in] port
616  *   Device port id.
617  * @param[in] valid
618  *   Device port id is valid, skip check. This flag is useful
619  *   when trials are performed from probing and device is not
620  *   flagged as valid yet (in attaching process).
621  * @param[out] es_domain_id
622  *   E-Switch domain id.
623  * @param[out] es_port_id
624  *   The port id of the port in the E-Switch.
625  *
626  * @return
627  *   pointer to device private data structure containing data needed
628  *   on success, NULL otherwise and rte_errno is set.
629  */
630 struct mlx5_priv *
631 mlx5_port_to_eswitch_info(uint16_t port, bool valid)
632 {
633 	struct rte_eth_dev *dev;
634 	struct mlx5_priv *priv;
635 
636 	if (port >= RTE_MAX_ETHPORTS) {
637 		rte_errno = EINVAL;
638 		return NULL;
639 	}
640 	if (!valid && !rte_eth_dev_is_valid_port(port)) {
641 		rte_errno = ENODEV;
642 		return NULL;
643 	}
644 	dev = &rte_eth_devices[port];
645 	priv = dev->data->dev_private;
646 	if (!(priv->representor || priv->master)) {
647 		rte_errno = EINVAL;
648 		return NULL;
649 	}
650 	return priv;
651 }
652 
653 /**
654  * Get the E-Switch parameters by device instance.
655  *
656  * @param[in] port
657  *   Device port id.
658  * @param[out] es_domain_id
659  *   E-Switch domain id.
660  * @param[out] es_port_id
661  *   The port id of the port in the E-Switch.
662  *
663  * @return
664  *   pointer to device private data structure containing data needed
665  *   on success, NULL otherwise and rte_errno is set.
666  */
667 struct mlx5_priv *
668 mlx5_dev_to_eswitch_info(struct rte_eth_dev *dev)
669 {
670 	struct mlx5_priv *priv;
671 
672 	priv = dev->data->dev_private;
673 	if (!(priv->representor || priv->master)) {
674 		rte_errno = EINVAL;
675 		return NULL;
676 	}
677 	return priv;
678 }
679 
680 /**
681  * DPDK callback to retrieve hairpin capabilities.
682  *
683  * @param dev
684  *   Pointer to Ethernet device structure.
685  * @param[out] cap
686  *   Storage for hairpin capability data.
687  *
688  * @return
689  *   0 on success, a negative errno value otherwise and rte_errno is set.
690  */
691 int
692 mlx5_hairpin_cap_get(struct rte_eth_dev *dev, struct rte_eth_hairpin_cap *cap)
693 {
694 	struct mlx5_priv *priv = dev->data->dev_private;
695 	struct mlx5_dev_config *config = &priv->config;
696 
697 	if (!priv->sh->devx || !config->dest_tir || !config->dv_flow_en) {
698 		rte_errno = ENOTSUP;
699 		return -rte_errno;
700 	}
701 	cap->max_nb_queues = UINT16_MAX;
702 	cap->max_rx_2_tx = 1;
703 	cap->max_tx_2_rx = 1;
704 	cap->max_nb_desc = 8192;
705 	return 0;
706 }
707