xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision daa02b5cddbb8e11b31d41e2bf7bb1ae64dcae2f)
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_rx.h"
27 #include "mlx5_tx.h"
28 #include "mlx5_autoconf.h"
29 
30 /**
31  * Get the interface index from device name.
32  *
33  * @param[in] dev
34  *   Pointer to Ethernet device.
35  *
36  * @return
37  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
38  */
39 unsigned int
40 mlx5_ifindex(const struct rte_eth_dev *dev)
41 {
42 	struct mlx5_priv *priv = dev->data->dev_private;
43 	unsigned int ifindex;
44 
45 	MLX5_ASSERT(priv);
46 	MLX5_ASSERT(priv->if_index);
47 	if (priv->master && priv->sh->bond.ifindex > 0)
48 		ifindex = priv->sh->bond.ifindex;
49 	else
50 		ifindex = priv->if_index;
51 	if (!ifindex)
52 		rte_errno = ENXIO;
53 	return ifindex;
54 }
55 
56 /**
57  * DPDK callback for Ethernet device configuration.
58  *
59  * @param dev
60  *   Pointer to Ethernet device structure.
61  *
62  * @return
63  *   0 on success, a negative errno value otherwise and rte_errno is set.
64  */
65 int
66 mlx5_dev_configure(struct rte_eth_dev *dev)
67 {
68 	struct mlx5_priv *priv = dev->data->dev_private;
69 	unsigned int rxqs_n = dev->data->nb_rx_queues;
70 	unsigned int txqs_n = dev->data->nb_tx_queues;
71 	const uint8_t use_app_rss_key =
72 		!!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
73 	int ret = 0;
74 
75 	if (use_app_rss_key &&
76 	    (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
77 	     MLX5_RSS_HASH_KEY_LEN)) {
78 		DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
79 			dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
80 		rte_errno = EINVAL;
81 		return -rte_errno;
82 	}
83 	priv->rss_conf.rss_key =
84 		mlx5_realloc(priv->rss_conf.rss_key, MLX5_MEM_RTE,
85 			    MLX5_RSS_HASH_KEY_LEN, 0, SOCKET_ID_ANY);
86 	if (!priv->rss_conf.rss_key) {
87 		DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
88 			dev->data->port_id, rxqs_n);
89 		rte_errno = ENOMEM;
90 		return -rte_errno;
91 	}
92 
93 	if ((dev->data->dev_conf.txmode.offloads &
94 			RTE_ETH_TX_OFFLOAD_SEND_ON_TIMESTAMP) &&
95 			rte_mbuf_dyn_tx_timestamp_register(NULL, NULL) != 0) {
96 		DRV_LOG(ERR, "port %u cannot register Tx timestamp field/flag",
97 			dev->data->port_id);
98 		return -rte_errno;
99 	}
100 	memcpy(priv->rss_conf.rss_key,
101 	       use_app_rss_key ?
102 	       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
103 	       rss_hash_default_key,
104 	       MLX5_RSS_HASH_KEY_LEN);
105 	priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
106 	priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
107 	priv->rxqs = (void *)dev->data->rx_queues;
108 	priv->txqs = (void *)dev->data->tx_queues;
109 	if (txqs_n != priv->txqs_n) {
110 		DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
111 			dev->data->port_id, priv->txqs_n, txqs_n);
112 		priv->txqs_n = txqs_n;
113 	}
114 	if (rxqs_n > priv->config.ind_table_max_size) {
115 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
116 			dev->data->port_id, rxqs_n);
117 		rte_errno = EINVAL;
118 		return -rte_errno;
119 	}
120 	if (rxqs_n != priv->rxqs_n) {
121 		DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
122 			dev->data->port_id, priv->rxqs_n, rxqs_n);
123 		priv->rxqs_n = rxqs_n;
124 	}
125 	priv->skip_default_rss_reta = 0;
126 	ret = mlx5_proc_priv_init(dev);
127 	if (ret)
128 		return ret;
129 	return 0;
130 }
131 
132 /**
133  * Configure default RSS reta.
134  *
135  * @param dev
136  *   Pointer to Ethernet device structure.
137  *
138  * @return
139  *   0 on success, a negative errno value otherwise and rte_errno is set.
140  */
141 int
142 mlx5_dev_configure_rss_reta(struct rte_eth_dev *dev)
143 {
144 	struct mlx5_priv *priv = dev->data->dev_private;
145 	unsigned int rxqs_n = dev->data->nb_rx_queues;
146 	unsigned int i;
147 	unsigned int j;
148 	unsigned int reta_idx_n;
149 	int ret = 0;
150 	unsigned int *rss_queue_arr = NULL;
151 	unsigned int rss_queue_n = 0;
152 
153 	if (priv->skip_default_rss_reta)
154 		return ret;
155 	rss_queue_arr = mlx5_malloc(0, rxqs_n * sizeof(unsigned int), 0,
156 				    SOCKET_ID_ANY);
157 	if (!rss_queue_arr) {
158 		DRV_LOG(ERR, "port %u cannot allocate RSS queue list (%u)",
159 			dev->data->port_id, rxqs_n);
160 		rte_errno = ENOMEM;
161 		return -rte_errno;
162 	}
163 	for (i = 0, j = 0; i < rxqs_n; i++) {
164 		struct mlx5_rxq_data *rxq_data;
165 		struct mlx5_rxq_ctrl *rxq_ctrl;
166 
167 		rxq_data = (*priv->rxqs)[i];
168 		rxq_ctrl = container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
169 		if (rxq_ctrl && rxq_ctrl->type == MLX5_RXQ_TYPE_STANDARD)
170 			rss_queue_arr[j++] = i;
171 	}
172 	rss_queue_n = j;
173 	if (rss_queue_n > priv->config.ind_table_max_size) {
174 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
175 			dev->data->port_id, rss_queue_n);
176 		rte_errno = EINVAL;
177 		mlx5_free(rss_queue_arr);
178 		return -rte_errno;
179 	}
180 	DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
181 		dev->data->port_id, priv->rxqs_n, rxqs_n);
182 	priv->rxqs_n = rxqs_n;
183 	/*
184 	 * If the requested number of RX queues is not a power of two,
185 	 * use the maximum indirection table size for better balancing.
186 	 * The result is always rounded to the next power of two.
187 	 */
188 	reta_idx_n = (1 << log2above((rss_queue_n & (rss_queue_n - 1)) ?
189 				priv->config.ind_table_max_size :
190 				rss_queue_n));
191 	ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
192 	if (ret) {
193 		mlx5_free(rss_queue_arr);
194 		return ret;
195 	}
196 	/*
197 	 * When the number of RX queues is not a power of two,
198 	 * the remaining table entries are padded with reused WQs
199 	 * and hashes are not spread uniformly.
200 	 */
201 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
202 		(*priv->reta_idx)[i] = rss_queue_arr[j];
203 		if (++j == rss_queue_n)
204 			j = 0;
205 	}
206 	mlx5_free(rss_queue_arr);
207 	return ret;
208 }
209 
210 /**
211  * Sets default tuning parameters.
212  *
213  * @param dev
214  *   Pointer to Ethernet device.
215  * @param[out] info
216  *   Info structure output buffer.
217  */
218 static void
219 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
220 {
221 	struct mlx5_priv *priv = dev->data->dev_private;
222 
223 	/* Minimum CPU utilization. */
224 	info->default_rxportconf.ring_size = 256;
225 	info->default_txportconf.ring_size = 256;
226 	info->default_rxportconf.burst_size = MLX5_RX_DEFAULT_BURST;
227 	info->default_txportconf.burst_size = MLX5_TX_DEFAULT_BURST;
228 	if ((priv->link_speed_capa & RTE_ETH_LINK_SPEED_200G) |
229 		(priv->link_speed_capa & RTE_ETH_LINK_SPEED_100G)) {
230 		info->default_rxportconf.nb_queues = 16;
231 		info->default_txportconf.nb_queues = 16;
232 		if (dev->data->nb_rx_queues > 2 ||
233 		    dev->data->nb_tx_queues > 2) {
234 			/* Max Throughput. */
235 			info->default_rxportconf.ring_size = 2048;
236 			info->default_txportconf.ring_size = 2048;
237 		}
238 	} else {
239 		info->default_rxportconf.nb_queues = 8;
240 		info->default_txportconf.nb_queues = 8;
241 		if (dev->data->nb_rx_queues > 2 ||
242 		    dev->data->nb_tx_queues > 2) {
243 			/* Max Throughput. */
244 			info->default_rxportconf.ring_size = 4096;
245 			info->default_txportconf.ring_size = 4096;
246 		}
247 	}
248 }
249 
250 /**
251  * Sets tx mbuf limiting parameters.
252  *
253  * @param dev
254  *   Pointer to Ethernet device.
255  * @param[out] info
256  *   Info structure output buffer.
257  */
258 static void
259 mlx5_set_txlimit_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
260 {
261 	struct mlx5_priv *priv = dev->data->dev_private;
262 	struct mlx5_dev_config *config = &priv->config;
263 	unsigned int inlen;
264 	uint16_t nb_max;
265 
266 	inlen = (config->txq_inline_max == MLX5_ARG_UNSET) ?
267 		MLX5_SEND_DEF_INLINE_LEN :
268 		(unsigned int)config->txq_inline_max;
269 	MLX5_ASSERT(config->txq_inline_min >= 0);
270 	inlen = RTE_MAX(inlen, (unsigned int)config->txq_inline_min);
271 	inlen = RTE_MIN(inlen, MLX5_WQE_SIZE_MAX +
272 			       MLX5_ESEG_MIN_INLINE_SIZE -
273 			       MLX5_WQE_CSEG_SIZE -
274 			       MLX5_WQE_ESEG_SIZE -
275 			       MLX5_WQE_DSEG_SIZE * 2);
276 	nb_max = (MLX5_WQE_SIZE_MAX +
277 		  MLX5_ESEG_MIN_INLINE_SIZE -
278 		  MLX5_WQE_CSEG_SIZE -
279 		  MLX5_WQE_ESEG_SIZE -
280 		  MLX5_WQE_DSEG_SIZE -
281 		  inlen) / MLX5_WSEG_SIZE;
282 	info->tx_desc_lim.nb_seg_max = nb_max;
283 	info->tx_desc_lim.nb_mtu_seg_max = nb_max;
284 }
285 
286 /**
287  * DPDK callback to get information about the device.
288  *
289  * @param dev
290  *   Pointer to Ethernet device structure.
291  * @param[out] info
292  *   Info structure output buffer.
293  */
294 int
295 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
296 {
297 	struct mlx5_priv *priv = dev->data->dev_private;
298 	struct mlx5_dev_config *config = &priv->config;
299 	unsigned int max;
300 
301 	/* FIXME: we should ask the device for these values. */
302 	info->min_rx_bufsize = 32;
303 	info->max_rx_pktlen = 65536;
304 	info->max_lro_pkt_size = MLX5_MAX_LRO_SIZE;
305 	/*
306 	 * Since we need one CQ per QP, the limit is the minimum number
307 	 * between the two values.
308 	 */
309 	max = RTE_MIN(priv->sh->device_attr.max_cq,
310 		      priv->sh->device_attr.max_qp);
311 	/* max_rx_queues is uint16_t. */
312 	max = RTE_MIN(max, (unsigned int)UINT16_MAX);
313 	info->max_rx_queues = max;
314 	info->max_tx_queues = max;
315 	info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
316 	info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
317 	info->rx_seg_capa.max_nseg = MLX5_MAX_RXQ_NSEG;
318 	info->rx_seg_capa.multi_pools = !config->mprq.enabled;
319 	info->rx_seg_capa.offset_allowed = !config->mprq.enabled;
320 	info->rx_seg_capa.offset_align_log2 = 0;
321 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
322 				 info->rx_queue_offload_capa);
323 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
324 	info->if_index = mlx5_ifindex(dev);
325 	info->reta_size = priv->reta_idx_n ?
326 		priv->reta_idx_n : config->ind_table_max_size;
327 	info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
328 	info->speed_capa = priv->link_speed_capa;
329 	info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
330 	mlx5_set_default_params(dev, info);
331 	mlx5_set_txlimit_params(dev, info);
332 	info->switch_info.name = dev->data->name;
333 	info->switch_info.domain_id = priv->domain_id;
334 	info->switch_info.port_id = priv->representor_id;
335 	if (priv->representor) {
336 		uint16_t port_id;
337 
338 		MLX5_ETH_FOREACH_DEV(port_id, dev->device) {
339 			struct mlx5_priv *opriv =
340 				rte_eth_devices[port_id].data->dev_private;
341 
342 			if (!opriv ||
343 			    opriv->representor ||
344 			    opriv->sh != priv->sh ||
345 			    opriv->domain_id != priv->domain_id)
346 				continue;
347 			/*
348 			 * Override switch name with that of the master
349 			 * device.
350 			 */
351 			info->switch_info.name = opriv->dev_data->name;
352 			break;
353 		}
354 	}
355 	return 0;
356 }
357 
358 /**
359  * Calculate representor ID from port switch info.
360  *
361  * Uint16 representor ID bits definition:
362  *   pf: 2
363  *   type: 2
364  *   vf/sf: 12
365  *
366  * @param info
367  *   Port switch info.
368  * @param hpf_type
369  *   Use this type if port is HPF.
370  *
371  * @return
372  *   Encoded representor ID.
373  */
374 uint16_t
375 mlx5_representor_id_encode(const struct mlx5_switch_info *info,
376 			   enum rte_eth_representor_type hpf_type)
377 {
378 	enum rte_eth_representor_type type = RTE_ETH_REPRESENTOR_VF;
379 	uint16_t repr = info->port_name;
380 
381 	if (info->representor == 0)
382 		return UINT16_MAX;
383 	if (info->name_type == MLX5_PHYS_PORT_NAME_TYPE_PFSF)
384 		type = RTE_ETH_REPRESENTOR_SF;
385 	if (info->name_type == MLX5_PHYS_PORT_NAME_TYPE_PFHPF) {
386 		type = hpf_type;
387 		repr = UINT16_MAX;
388 	}
389 	return MLX5_REPRESENTOR_ID(info->pf_num, type, repr);
390 }
391 
392 /**
393  * DPDK callback to get information about representor.
394  *
395  * Representor ID bits definition:
396  *   vf/sf: 12
397  *   type: 2
398  *   pf: 2
399  *
400  * @param dev
401  *   Pointer to Ethernet device structure.
402  * @param[out] info
403  *   Nullable info structure output buffer.
404  *
405  * @return
406  *   negative on error, or the number of representor ranges.
407  */
408 int
409 mlx5_representor_info_get(struct rte_eth_dev *dev,
410 			  struct rte_eth_representor_info *info)
411 {
412 	struct mlx5_priv *priv = dev->data->dev_private;
413 	int n_type = 4; /* Representor types, VF, HPF@VF, SF and HPF@SF. */
414 	int n_pf = 2; /* Number of PFs. */
415 	int i = 0, pf;
416 	int n_entries;
417 
418 	if (info == NULL)
419 		goto out;
420 
421 	n_entries = n_type * n_pf;
422 	if ((uint32_t)n_entries > info->nb_ranges_alloc)
423 		n_entries = info->nb_ranges_alloc;
424 
425 	info->controller = 0;
426 	info->pf = priv->pf_bond >= 0 ? priv->pf_bond : 0;
427 	for (pf = 0; pf < n_pf; ++pf) {
428 		/* VF range. */
429 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
430 		info->ranges[i].controller = 0;
431 		info->ranges[i].pf = pf;
432 		info->ranges[i].vf = 0;
433 		info->ranges[i].id_base =
434 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
435 		info->ranges[i].id_end =
436 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
437 		snprintf(info->ranges[i].name,
438 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
439 		i++;
440 		if (i == n_entries)
441 			break;
442 		/* HPF range of VF type. */
443 		info->ranges[i].type = RTE_ETH_REPRESENTOR_VF;
444 		info->ranges[i].controller = 0;
445 		info->ranges[i].pf = pf;
446 		info->ranges[i].vf = UINT16_MAX;
447 		info->ranges[i].id_base =
448 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
449 		info->ranges[i].id_end =
450 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
451 		snprintf(info->ranges[i].name,
452 			 sizeof(info->ranges[i].name), "pf%dvf", pf);
453 		i++;
454 		if (i == n_entries)
455 			break;
456 		/* SF range. */
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 = 0;
461 		info->ranges[i].id_base =
462 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, 0);
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 		if (i == n_entries)
469 			break;
470 		/* HPF range of SF type. */
471 		info->ranges[i].type = RTE_ETH_REPRESENTOR_SF;
472 		info->ranges[i].controller = 0;
473 		info->ranges[i].pf = pf;
474 		info->ranges[i].vf = UINT16_MAX;
475 		info->ranges[i].id_base =
476 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
477 		info->ranges[i].id_end =
478 			MLX5_REPRESENTOR_ID(pf, info->ranges[i].type, -1);
479 		snprintf(info->ranges[i].name,
480 			 sizeof(info->ranges[i].name), "pf%dsf", pf);
481 		i++;
482 		if (i == n_entries)
483 			break;
484 	}
485 	info->nb_ranges = i;
486 out:
487 	return n_type * n_pf;
488 }
489 
490 /**
491  * Get firmware version of a device.
492  *
493  * @param dev
494  *   Ethernet device port.
495  * @param fw_ver
496  *   String output allocated by caller.
497  * @param fw_size
498  *   Size of the output string, including terminating null byte.
499  *
500  * @return
501  *   0 on success, or the size of the non truncated string if too big.
502  */
503 int
504 mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
505 {
506 	struct mlx5_priv *priv = dev->data->dev_private;
507 	struct mlx5_dev_attr *attr = &priv->sh->device_attr;
508 	size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
509 
510 	if (fw_size < size)
511 		return size;
512 	if (fw_ver != NULL)
513 		strlcpy(fw_ver, attr->fw_ver, fw_size);
514 	return 0;
515 }
516 
517 /**
518  * Get supported packet types.
519  *
520  * @param dev
521  *   Pointer to Ethernet device structure.
522  *
523  * @return
524  *   A pointer to the supported Packet types array.
525  */
526 const uint32_t *
527 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
528 {
529 	static const uint32_t ptypes[] = {
530 		/* refers to rxq_cq_to_pkt_type() */
531 		RTE_PTYPE_L2_ETHER,
532 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
533 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
534 		RTE_PTYPE_L4_NONFRAG,
535 		RTE_PTYPE_L4_FRAG,
536 		RTE_PTYPE_L4_TCP,
537 		RTE_PTYPE_L4_UDP,
538 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
539 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
540 		RTE_PTYPE_INNER_L4_NONFRAG,
541 		RTE_PTYPE_INNER_L4_FRAG,
542 		RTE_PTYPE_INNER_L4_TCP,
543 		RTE_PTYPE_INNER_L4_UDP,
544 		RTE_PTYPE_UNKNOWN
545 	};
546 
547 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
548 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
549 	    dev->rx_pkt_burst == mlx5_rx_burst_vec ||
550 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq_vec)
551 		return ptypes;
552 	return NULL;
553 }
554 
555 /**
556  * DPDK callback to change the MTU.
557  *
558  * @param dev
559  *   Pointer to Ethernet device structure.
560  * @param in_mtu
561  *   New MTU.
562  *
563  * @return
564  *   0 on success, a negative errno value otherwise and rte_errno is set.
565  */
566 int
567 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
568 {
569 	struct mlx5_priv *priv = dev->data->dev_private;
570 	uint16_t kern_mtu = 0;
571 	int ret;
572 
573 	ret = mlx5_get_mtu(dev, &kern_mtu);
574 	if (ret)
575 		return ret;
576 	/* Set kernel interface MTU first. */
577 	ret = mlx5_set_mtu(dev, mtu);
578 	if (ret)
579 		return ret;
580 	ret = mlx5_get_mtu(dev, &kern_mtu);
581 	if (ret)
582 		return ret;
583 	if (kern_mtu == mtu) {
584 		priv->mtu = mtu;
585 		DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
586 			dev->data->port_id, mtu);
587 		return 0;
588 	}
589 	rte_errno = EAGAIN;
590 	return -rte_errno;
591 }
592 
593 /**
594  * Configure the RX function to use.
595  *
596  * @param dev
597  *   Pointer to private data structure.
598  *
599  * @return
600  *   Pointer to selected Rx burst function.
601  */
602 eth_rx_burst_t
603 mlx5_select_rx_function(struct rte_eth_dev *dev)
604 {
605 	eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
606 
607 	MLX5_ASSERT(dev != NULL);
608 	if (mlx5_check_vec_rx_support(dev) > 0) {
609 		if (mlx5_mprq_enabled(dev)) {
610 			rx_pkt_burst = mlx5_rx_burst_mprq_vec;
611 			DRV_LOG(DEBUG, "port %u selected vectorized"
612 				" MPRQ Rx function", dev->data->port_id);
613 		} else {
614 			rx_pkt_burst = mlx5_rx_burst_vec;
615 			DRV_LOG(DEBUG, "port %u selected vectorized"
616 				" SPRQ Rx function", dev->data->port_id);
617 		}
618 	} else if (mlx5_mprq_enabled(dev)) {
619 		rx_pkt_burst = mlx5_rx_burst_mprq;
620 		DRV_LOG(DEBUG, "port %u selected MPRQ Rx function",
621 			dev->data->port_id);
622 	} else {
623 		DRV_LOG(DEBUG, "port %u selected SPRQ Rx function",
624 			dev->data->port_id);
625 	}
626 	return rx_pkt_burst;
627 }
628 
629 /**
630  * Get the E-Switch parameters by port id.
631  *
632  * @param[in] port
633  *   Device port id.
634  * @param[in] valid
635  *   Device port id is valid, skip check. This flag is useful
636  *   when trials are performed from probing and device is not
637  *   flagged as valid yet (in attaching process).
638  * @param[out] es_domain_id
639  *   E-Switch domain id.
640  * @param[out] es_port_id
641  *   The port id of the port in the E-Switch.
642  *
643  * @return
644  *   pointer to device private data structure containing data needed
645  *   on success, NULL otherwise and rte_errno is set.
646  */
647 struct mlx5_priv *
648 mlx5_port_to_eswitch_info(uint16_t port, bool valid)
649 {
650 	struct rte_eth_dev *dev;
651 	struct mlx5_priv *priv;
652 
653 	if (port >= RTE_MAX_ETHPORTS) {
654 		rte_errno = EINVAL;
655 		return NULL;
656 	}
657 	if (!valid && !rte_eth_dev_is_valid_port(port)) {
658 		rte_errno = ENODEV;
659 		return NULL;
660 	}
661 	dev = &rte_eth_devices[port];
662 	priv = dev->data->dev_private;
663 	if (!(priv->representor || priv->master)) {
664 		rte_errno = EINVAL;
665 		return NULL;
666 	}
667 	return priv;
668 }
669 
670 /**
671  * Get the E-Switch parameters by device instance.
672  *
673  * @param[in] port
674  *   Device port id.
675  * @param[out] es_domain_id
676  *   E-Switch domain id.
677  * @param[out] es_port_id
678  *   The port id of the port in the E-Switch.
679  *
680  * @return
681  *   pointer to device private data structure containing data needed
682  *   on success, NULL otherwise and rte_errno is set.
683  */
684 struct mlx5_priv *
685 mlx5_dev_to_eswitch_info(struct rte_eth_dev *dev)
686 {
687 	struct mlx5_priv *priv;
688 
689 	priv = dev->data->dev_private;
690 	if (!(priv->representor || priv->master)) {
691 		rte_errno = EINVAL;
692 		return NULL;
693 	}
694 	return priv;
695 }
696 
697 /**
698  * DPDK callback to retrieve hairpin capabilities.
699  *
700  * @param dev
701  *   Pointer to Ethernet device structure.
702  * @param[out] cap
703  *   Storage for hairpin capability data.
704  *
705  * @return
706  *   0 on success, a negative errno value otherwise and rte_errno is set.
707  */
708 int
709 mlx5_hairpin_cap_get(struct rte_eth_dev *dev, struct rte_eth_hairpin_cap *cap)
710 {
711 	struct mlx5_priv *priv = dev->data->dev_private;
712 	struct mlx5_dev_config *config = &priv->config;
713 
714 	if (!priv->sh->devx || !config->dest_tir || !config->dv_flow_en) {
715 		rte_errno = ENOTSUP;
716 		return -rte_errno;
717 	}
718 	cap->max_nb_queues = UINT16_MAX;
719 	cap->max_rx_2_tx = 1;
720 	cap->max_tx_2_rx = 1;
721 	cap->max_nb_desc = 8192;
722 	return 0;
723 }
724