xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision 487eec3401b7a1664982f39da139980a4f5b3adc)
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 <assert.h>
8 #include <inttypes.h>
9 #include <unistd.h>
10 #include <stdbool.h>
11 #include <stdint.h>
12 #include <stdio.h>
13 #include <string.h>
14 #include <stdlib.h>
15 #include <errno.h>
16 #include <dirent.h>
17 #include <net/if.h>
18 #include <sys/ioctl.h>
19 #include <sys/socket.h>
20 #include <netinet/in.h>
21 #include <linux/ethtool.h>
22 #include <linux/sockios.h>
23 #include <fcntl.h>
24 #include <stdalign.h>
25 #include <sys/un.h>
26 #include <time.h>
27 
28 #include <rte_atomic.h>
29 #include <rte_ethdev_driver.h>
30 #include <rte_bus_pci.h>
31 #include <rte_mbuf.h>
32 #include <rte_common.h>
33 #include <rte_interrupts.h>
34 #include <rte_malloc.h>
35 #include <rte_string_fns.h>
36 #include <rte_rwlock.h>
37 #include <rte_cycles.h>
38 
39 #include "mlx5.h"
40 #include "mlx5_glue.h"
41 #include "mlx5_rxtx.h"
42 #include "mlx5_utils.h"
43 
44 /* Supported speed values found in /usr/include/linux/ethtool.h */
45 #ifndef HAVE_SUPPORTED_40000baseKR4_Full
46 #define SUPPORTED_40000baseKR4_Full (1 << 23)
47 #endif
48 #ifndef HAVE_SUPPORTED_40000baseCR4_Full
49 #define SUPPORTED_40000baseCR4_Full (1 << 24)
50 #endif
51 #ifndef HAVE_SUPPORTED_40000baseSR4_Full
52 #define SUPPORTED_40000baseSR4_Full (1 << 25)
53 #endif
54 #ifndef HAVE_SUPPORTED_40000baseLR4_Full
55 #define SUPPORTED_40000baseLR4_Full (1 << 26)
56 #endif
57 #ifndef HAVE_SUPPORTED_56000baseKR4_Full
58 #define SUPPORTED_56000baseKR4_Full (1 << 27)
59 #endif
60 #ifndef HAVE_SUPPORTED_56000baseCR4_Full
61 #define SUPPORTED_56000baseCR4_Full (1 << 28)
62 #endif
63 #ifndef HAVE_SUPPORTED_56000baseSR4_Full
64 #define SUPPORTED_56000baseSR4_Full (1 << 29)
65 #endif
66 #ifndef HAVE_SUPPORTED_56000baseLR4_Full
67 #define SUPPORTED_56000baseLR4_Full (1 << 30)
68 #endif
69 
70 /* Add defines in case the running kernel is not the same as user headers. */
71 #ifndef ETHTOOL_GLINKSETTINGS
72 struct ethtool_link_settings {
73 	uint32_t cmd;
74 	uint32_t speed;
75 	uint8_t duplex;
76 	uint8_t port;
77 	uint8_t phy_address;
78 	uint8_t autoneg;
79 	uint8_t mdio_support;
80 	uint8_t eth_to_mdix;
81 	uint8_t eth_tp_mdix_ctrl;
82 	int8_t link_mode_masks_nwords;
83 	uint32_t reserved[8];
84 	uint32_t link_mode_masks[];
85 };
86 
87 #define ETHTOOL_GLINKSETTINGS 0x0000004c
88 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
89 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
90 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
91 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
92 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
93 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
94 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
95 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
96 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
97 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
98 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
99 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
100 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
101 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
102 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
103 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
104 #endif
105 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
106 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
107 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
108 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
109 #endif
110 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
111 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
112 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
113 #endif
114 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
115 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
116 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
117 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
118 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
119 #endif
120 
121 /**
122  * Get master interface name from private structure.
123  *
124  * @param[in] dev
125  *   Pointer to Ethernet device.
126  * @param[out] ifname
127  *   Interface name output buffer.
128  *
129  * @return
130  *   0 on success, a negative errno value otherwise and rte_errno is set.
131  */
132 int
133 mlx5_get_master_ifname(const char *ibdev_path, char (*ifname)[IF_NAMESIZE])
134 {
135 	DIR *dir;
136 	struct dirent *dent;
137 	unsigned int dev_type = 0;
138 	unsigned int dev_port_prev = ~0u;
139 	char match[IF_NAMESIZE] = "";
140 
141 	assert(ibdev_path);
142 	{
143 		MKSTR(path, "%s/device/net", ibdev_path);
144 
145 		dir = opendir(path);
146 		if (dir == NULL) {
147 			rte_errno = errno;
148 			return -rte_errno;
149 		}
150 	}
151 	while ((dent = readdir(dir)) != NULL) {
152 		char *name = dent->d_name;
153 		FILE *file;
154 		unsigned int dev_port;
155 		int r;
156 
157 		if ((name[0] == '.') &&
158 		    ((name[1] == '\0') ||
159 		     ((name[1] == '.') && (name[2] == '\0'))))
160 			continue;
161 
162 		MKSTR(path, "%s/device/net/%s/%s",
163 		      ibdev_path, name,
164 		      (dev_type ? "dev_id" : "dev_port"));
165 
166 		file = fopen(path, "rb");
167 		if (file == NULL) {
168 			if (errno != ENOENT)
169 				continue;
170 			/*
171 			 * Switch to dev_id when dev_port does not exist as
172 			 * is the case with Linux kernel versions < 3.15.
173 			 */
174 try_dev_id:
175 			match[0] = '\0';
176 			if (dev_type)
177 				break;
178 			dev_type = 1;
179 			dev_port_prev = ~0u;
180 			rewinddir(dir);
181 			continue;
182 		}
183 		r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
184 		fclose(file);
185 		if (r != 1)
186 			continue;
187 		/*
188 		 * Switch to dev_id when dev_port returns the same value for
189 		 * all ports. May happen when using a MOFED release older than
190 		 * 3.0 with a Linux kernel >= 3.15.
191 		 */
192 		if (dev_port == dev_port_prev)
193 			goto try_dev_id;
194 		dev_port_prev = dev_port;
195 		if (dev_port == 0)
196 			strlcpy(match, name, sizeof(match));
197 	}
198 	closedir(dir);
199 	if (match[0] == '\0') {
200 		rte_errno = ENOENT;
201 		return -rte_errno;
202 	}
203 	strncpy(*ifname, match, sizeof(*ifname));
204 	return 0;
205 }
206 
207 /**
208  * Get interface name from private structure.
209  *
210  * This is a port representor-aware version of mlx5_get_master_ifname().
211  *
212  * @param[in] dev
213  *   Pointer to Ethernet device.
214  * @param[out] ifname
215  *   Interface name output buffer.
216  *
217  * @return
218  *   0 on success, a negative errno value otherwise and rte_errno is set.
219  */
220 int
221 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[IF_NAMESIZE])
222 {
223 	struct mlx5_priv *priv = dev->data->dev_private;
224 	unsigned int ifindex;
225 
226 	assert(priv);
227 	assert(priv->sh);
228 	ifindex = mlx5_ifindex(dev);
229 	if (!ifindex) {
230 		if (!priv->representor)
231 			return mlx5_get_master_ifname(priv->sh->ibdev_path,
232 						      ifname);
233 		rte_errno = ENXIO;
234 		return -rte_errno;
235 	}
236 	if (if_indextoname(ifindex, &(*ifname)[0]))
237 		return 0;
238 	rte_errno = errno;
239 	return -rte_errno;
240 }
241 
242 /**
243  * Get the interface index from device name.
244  *
245  * @param[in] dev
246  *   Pointer to Ethernet device.
247  *
248  * @return
249  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
250  */
251 unsigned int
252 mlx5_ifindex(const struct rte_eth_dev *dev)
253 {
254 	struct mlx5_priv *priv = dev->data->dev_private;
255 	unsigned int ifindex;
256 
257 	assert(priv);
258 	assert(priv->if_index);
259 	ifindex = priv->if_index;
260 	if (!ifindex)
261 		rte_errno = ENXIO;
262 	return ifindex;
263 }
264 
265 /**
266  * Perform ifreq ioctl() on associated Ethernet device.
267  *
268  * @param[in] dev
269  *   Pointer to Ethernet device.
270  * @param req
271  *   Request number to pass to ioctl().
272  * @param[out] ifr
273  *   Interface request structure output buffer.
274  *
275  * @return
276  *   0 on success, a negative errno value otherwise and rte_errno is set.
277  */
278 int
279 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
280 {
281 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
282 	int ret = 0;
283 
284 	if (sock == -1) {
285 		rte_errno = errno;
286 		return -rte_errno;
287 	}
288 	ret = mlx5_get_ifname(dev, &ifr->ifr_name);
289 	if (ret)
290 		goto error;
291 	ret = ioctl(sock, req, ifr);
292 	if (ret == -1) {
293 		rte_errno = errno;
294 		goto error;
295 	}
296 	close(sock);
297 	return 0;
298 error:
299 	close(sock);
300 	return -rte_errno;
301 }
302 
303 /**
304  * Get device MTU.
305  *
306  * @param dev
307  *   Pointer to Ethernet device.
308  * @param[out] mtu
309  *   MTU value output buffer.
310  *
311  * @return
312  *   0 on success, a negative errno value otherwise and rte_errno is set.
313  */
314 int
315 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
316 {
317 	struct ifreq request;
318 	int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
319 
320 	if (ret)
321 		return ret;
322 	*mtu = request.ifr_mtu;
323 	return 0;
324 }
325 
326 /**
327  * Set device MTU.
328  *
329  * @param dev
330  *   Pointer to Ethernet device.
331  * @param mtu
332  *   MTU value to set.
333  *
334  * @return
335  *   0 on success, a negative errno value otherwise and rte_errno is set.
336  */
337 static int
338 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
339 {
340 	struct ifreq request = { .ifr_mtu = mtu, };
341 
342 	return mlx5_ifreq(dev, SIOCSIFMTU, &request);
343 }
344 
345 /**
346  * Set device flags.
347  *
348  * @param dev
349  *   Pointer to Ethernet device.
350  * @param keep
351  *   Bitmask for flags that must remain untouched.
352  * @param flags
353  *   Bitmask for flags to modify.
354  *
355  * @return
356  *   0 on success, a negative errno value otherwise and rte_errno is set.
357  */
358 int
359 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
360 {
361 	struct ifreq request;
362 	int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
363 
364 	if (ret)
365 		return ret;
366 	request.ifr_flags &= keep;
367 	request.ifr_flags |= flags & ~keep;
368 	return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
369 }
370 
371 /**
372  * DPDK callback for Ethernet device configuration.
373  *
374  * @param dev
375  *   Pointer to Ethernet device structure.
376  *
377  * @return
378  *   0 on success, a negative errno value otherwise and rte_errno is set.
379  */
380 int
381 mlx5_dev_configure(struct rte_eth_dev *dev)
382 {
383 	struct mlx5_priv *priv = dev->data->dev_private;
384 	unsigned int rxqs_n = dev->data->nb_rx_queues;
385 	unsigned int txqs_n = dev->data->nb_tx_queues;
386 	unsigned int i;
387 	unsigned int j;
388 	unsigned int reta_idx_n;
389 	const uint8_t use_app_rss_key =
390 		!!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
391 	int ret = 0;
392 
393 	if (use_app_rss_key &&
394 	    (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
395 	     MLX5_RSS_HASH_KEY_LEN)) {
396 		DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
397 			dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
398 		rte_errno = EINVAL;
399 		return -rte_errno;
400 	}
401 	priv->rss_conf.rss_key =
402 		rte_realloc(priv->rss_conf.rss_key,
403 			    MLX5_RSS_HASH_KEY_LEN, 0);
404 	if (!priv->rss_conf.rss_key) {
405 		DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
406 			dev->data->port_id, rxqs_n);
407 		rte_errno = ENOMEM;
408 		return -rte_errno;
409 	}
410 	memcpy(priv->rss_conf.rss_key,
411 	       use_app_rss_key ?
412 	       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
413 	       rss_hash_default_key,
414 	       MLX5_RSS_HASH_KEY_LEN);
415 	priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
416 	priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
417 	priv->rxqs = (void *)dev->data->rx_queues;
418 	priv->txqs = (void *)dev->data->tx_queues;
419 	if (txqs_n != priv->txqs_n) {
420 		DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
421 			dev->data->port_id, priv->txqs_n, txqs_n);
422 		priv->txqs_n = txqs_n;
423 	}
424 	if (rxqs_n > priv->config.ind_table_max_size) {
425 		DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
426 			dev->data->port_id, rxqs_n);
427 		rte_errno = EINVAL;
428 		return -rte_errno;
429 	}
430 	if (rxqs_n != priv->rxqs_n) {
431 		DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
432 			dev->data->port_id, priv->rxqs_n, rxqs_n);
433 		priv->rxqs_n = rxqs_n;
434 		/*
435 		 * If the requested number of RX queues is not a power of two,
436 		 * use the maximum indirection table size for better balancing.
437 		 * The result is always rounded to the next power of two.
438 		 */
439 		reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
440 					     priv->config.ind_table_max_size :
441 					     rxqs_n));
442 		ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
443 		if (ret)
444 			return ret;
445 		/*
446 		 * When the number of RX queues is not a power of two,
447 		 * the remaining table entries are padded with reused WQs
448 		 * and hashes are not spread uniformly.
449 		 */
450 		for (i = 0, j = 0; (i != reta_idx_n); ++i) {
451 			(*priv->reta_idx)[i] = j;
452 			if (++j == rxqs_n)
453 				j = 0;
454 		}
455 	}
456 	ret = mlx5_proc_priv_init(dev);
457 	if (ret)
458 		return ret;
459 	return 0;
460 }
461 
462 /**
463  * Sets default tuning parameters.
464  *
465  * @param dev
466  *   Pointer to Ethernet device.
467  * @param[out] info
468  *   Info structure output buffer.
469  */
470 static void
471 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
472 {
473 	struct mlx5_priv *priv = dev->data->dev_private;
474 
475 	/* Minimum CPU utilization. */
476 	info->default_rxportconf.ring_size = 256;
477 	info->default_txportconf.ring_size = 256;
478 	info->default_rxportconf.burst_size = 64;
479 	info->default_txportconf.burst_size = 64;
480 	if (priv->link_speed_capa & ETH_LINK_SPEED_100G) {
481 		info->default_rxportconf.nb_queues = 16;
482 		info->default_txportconf.nb_queues = 16;
483 		if (dev->data->nb_rx_queues > 2 ||
484 		    dev->data->nb_tx_queues > 2) {
485 			/* Max Throughput. */
486 			info->default_rxportconf.ring_size = 2048;
487 			info->default_txportconf.ring_size = 2048;
488 		}
489 	} else {
490 		info->default_rxportconf.nb_queues = 8;
491 		info->default_txportconf.nb_queues = 8;
492 		if (dev->data->nb_rx_queues > 2 ||
493 		    dev->data->nb_tx_queues > 2) {
494 			/* Max Throughput. */
495 			info->default_rxportconf.ring_size = 4096;
496 			info->default_txportconf.ring_size = 4096;
497 		}
498 	}
499 }
500 
501 /**
502  * Sets tx mbuf limiting parameters.
503  *
504  * @param dev
505  *   Pointer to Ethernet device.
506  * @param[out] info
507  *   Info structure output buffer.
508  */
509 static void
510 mlx5_set_txlimit_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
511 {
512 	struct mlx5_priv *priv = dev->data->dev_private;
513 	struct mlx5_dev_config *config = &priv->config;
514 	unsigned int inlen;
515 	uint16_t nb_max;
516 
517 	inlen = (config->txq_inline_max == MLX5_ARG_UNSET) ?
518 		MLX5_SEND_DEF_INLINE_LEN :
519 		(unsigned int)config->txq_inline_max;
520 	assert(config->txq_inline_min >= 0);
521 	inlen = RTE_MAX(inlen, (unsigned int)config->txq_inline_min);
522 	inlen = RTE_MIN(inlen, MLX5_WQE_SIZE_MAX +
523 			       MLX5_ESEG_MIN_INLINE_SIZE -
524 			       MLX5_WQE_CSEG_SIZE -
525 			       MLX5_WQE_ESEG_SIZE -
526 			       MLX5_WQE_DSEG_SIZE * 2);
527 	nb_max = (MLX5_WQE_SIZE_MAX +
528 		  MLX5_ESEG_MIN_INLINE_SIZE -
529 		  MLX5_WQE_CSEG_SIZE -
530 		  MLX5_WQE_ESEG_SIZE -
531 		  MLX5_WQE_DSEG_SIZE -
532 		  inlen) / MLX5_WSEG_SIZE;
533 	info->tx_desc_lim.nb_seg_max = nb_max;
534 	info->tx_desc_lim.nb_mtu_seg_max = nb_max;
535 }
536 
537 /**
538  * DPDK callback to get information about the device.
539  *
540  * @param dev
541  *   Pointer to Ethernet device structure.
542  * @param[out] info
543  *   Info structure output buffer.
544  */
545 int
546 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
547 {
548 	struct mlx5_priv *priv = dev->data->dev_private;
549 	struct mlx5_dev_config *config = &priv->config;
550 	unsigned int max;
551 
552 	/* FIXME: we should ask the device for these values. */
553 	info->min_rx_bufsize = 32;
554 	info->max_rx_pktlen = 65536;
555 	/*
556 	 * Since we need one CQ per QP, the limit is the minimum number
557 	 * between the two values.
558 	 */
559 	max = RTE_MIN(priv->sh->device_attr.orig_attr.max_cq,
560 		      priv->sh->device_attr.orig_attr.max_qp);
561 	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
562 	if (max >= 65535)
563 		max = 65535;
564 	info->max_rx_queues = max;
565 	info->max_tx_queues = max;
566 	info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
567 	info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
568 	info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
569 				 info->rx_queue_offload_capa);
570 	info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
571 	info->if_index = mlx5_ifindex(dev);
572 	info->reta_size = priv->reta_idx_n ?
573 		priv->reta_idx_n : config->ind_table_max_size;
574 	info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
575 	info->speed_capa = priv->link_speed_capa;
576 	info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
577 	mlx5_set_default_params(dev, info);
578 	mlx5_set_txlimit_params(dev, info);
579 	info->switch_info.name = dev->data->name;
580 	info->switch_info.domain_id = priv->domain_id;
581 	info->switch_info.port_id = priv->representor_id;
582 	if (priv->representor) {
583 		uint16_t port_id;
584 
585 		if (priv->pf_bond >= 0) {
586 			/*
587 			 * Switch port ID is opaque value with driver defined
588 			 * format. Push the PF index in bonding configurations
589 			 * in upper four bits of port ID. If we get too many
590 			 * representors (more than 4K) or PFs (more than 15)
591 			 * this approach must be reconsidered.
592 			 */
593 			if ((info->switch_info.port_id >>
594 				MLX5_PORT_ID_BONDING_PF_SHIFT) ||
595 			    priv->pf_bond > MLX5_PORT_ID_BONDING_PF_MASK) {
596 				DRV_LOG(ERR, "can't update switch port ID"
597 					     " for bonding device");
598 				assert(false);
599 				return -ENODEV;
600 			}
601 			info->switch_info.port_id |=
602 				priv->pf_bond << MLX5_PORT_ID_BONDING_PF_SHIFT;
603 		}
604 		MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
605 			struct mlx5_priv *opriv =
606 				rte_eth_devices[port_id].data->dev_private;
607 
608 			if (!opriv ||
609 			    opriv->representor ||
610 			    opriv->sh != priv->sh ||
611 			    opriv->domain_id != priv->domain_id)
612 				continue;
613 			/*
614 			 * Override switch name with that of the master
615 			 * device.
616 			 */
617 			info->switch_info.name = opriv->dev_data->name;
618 			break;
619 		}
620 	}
621 	return 0;
622 }
623 
624 /**
625  * Get device current raw clock counter
626  *
627  * @param dev
628  *   Pointer to Ethernet device structure.
629  * @param[out] time
630  *   Current raw clock counter of the device.
631  *
632  * @return
633  *   0 if the clock has correctly been read
634  *   The value of errno in case of error
635  */
636 int
637 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
638 {
639 	struct mlx5_priv *priv = dev->data->dev_private;
640 	struct ibv_context *ctx = priv->sh->ctx;
641 	struct ibv_values_ex values;
642 	int err = 0;
643 
644 	values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
645 	err = mlx5_glue->query_rt_values_ex(ctx, &values);
646 	if (err != 0) {
647 		DRV_LOG(WARNING, "Could not query the clock !");
648 		return err;
649 	}
650 	*clock = values.raw_clock.tv_nsec;
651 	return 0;
652 }
653 
654 /**
655  * Get firmware version of a device.
656  *
657  * @param dev
658  *   Ethernet device port.
659  * @param fw_ver
660  *   String output allocated by caller.
661  * @param fw_size
662  *   Size of the output string, including terminating null byte.
663  *
664  * @return
665  *   0 on success, or the size of the non truncated string if too big.
666  */
667 int mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
668 {
669 	struct mlx5_priv *priv = dev->data->dev_private;
670 	struct ibv_device_attr *attr = &priv->sh->device_attr.orig_attr;
671 	size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
672 
673 	if (fw_size < size)
674 		return size;
675 	if (fw_ver != NULL)
676 		strlcpy(fw_ver, attr->fw_ver, fw_size);
677 	return 0;
678 }
679 
680 /**
681  * Get supported packet types.
682  *
683  * @param dev
684  *   Pointer to Ethernet device structure.
685  *
686  * @return
687  *   A pointer to the supported Packet types array.
688  */
689 const uint32_t *
690 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
691 {
692 	static const uint32_t ptypes[] = {
693 		/* refers to rxq_cq_to_pkt_type() */
694 		RTE_PTYPE_L2_ETHER,
695 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
696 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
697 		RTE_PTYPE_L4_NONFRAG,
698 		RTE_PTYPE_L4_FRAG,
699 		RTE_PTYPE_L4_TCP,
700 		RTE_PTYPE_L4_UDP,
701 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
702 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
703 		RTE_PTYPE_INNER_L4_NONFRAG,
704 		RTE_PTYPE_INNER_L4_FRAG,
705 		RTE_PTYPE_INNER_L4_TCP,
706 		RTE_PTYPE_INNER_L4_UDP,
707 		RTE_PTYPE_UNKNOWN
708 	};
709 
710 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
711 	    dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
712 	    dev->rx_pkt_burst == mlx5_rx_burst_vec)
713 		return ptypes;
714 	return NULL;
715 }
716 
717 /**
718  * Retrieve the master device for representor in the same switch domain.
719  *
720  * @param dev
721  *   Pointer to representor Ethernet device structure.
722  *
723  * @return
724  *   Master device structure  on success, NULL otherwise.
725  */
726 
727 static struct rte_eth_dev *
728 mlx5_find_master_dev(struct rte_eth_dev *dev)
729 {
730 	struct mlx5_priv *priv;
731 	uint16_t port_id;
732 	uint16_t domain_id;
733 
734 	priv = dev->data->dev_private;
735 	domain_id = priv->domain_id;
736 	assert(priv->representor);
737 	MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
738 		struct mlx5_priv *opriv =
739 			rte_eth_devices[port_id].data->dev_private;
740 		if (opriv &&
741 		    opriv->master &&
742 		    opriv->domain_id == domain_id &&
743 		    opriv->sh == priv->sh)
744 			return &rte_eth_devices[port_id];
745 	}
746 	return NULL;
747 }
748 
749 /**
750  * DPDK callback to retrieve physical link information.
751  *
752  * @param dev
753  *   Pointer to Ethernet device structure.
754  * @param[out] link
755  *   Storage for current link status.
756  *
757  * @return
758  *   0 on success, a negative errno value otherwise and rte_errno is set.
759  */
760 static int
761 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
762 			       struct rte_eth_link *link)
763 {
764 	struct mlx5_priv *priv = dev->data->dev_private;
765 	struct ethtool_cmd edata = {
766 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
767 	};
768 	struct ifreq ifr;
769 	struct rte_eth_link dev_link;
770 	int link_speed = 0;
771 	int ret;
772 
773 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
774 	if (ret) {
775 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
776 			dev->data->port_id, strerror(rte_errno));
777 		return ret;
778 	}
779 	dev_link = (struct rte_eth_link) {
780 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
781 				(ifr.ifr_flags & IFF_RUNNING)),
782 	};
783 	ifr = (struct ifreq) {
784 		.ifr_data = (void *)&edata,
785 	};
786 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
787 	if (ret) {
788 		if (ret == -ENOTSUP && priv->representor) {
789 			struct rte_eth_dev *master;
790 
791 			/*
792 			 * For representors we can try to inherit link
793 			 * settings from the master device. Actually
794 			 * link settings do not make a lot of sense
795 			 * for representors due to missing physical
796 			 * link. The old kernel drivers supported
797 			 * emulated settings query for representors,
798 			 * the new ones do not, so we have to add
799 			 * this code for compatibility issues.
800 			 */
801 			master = mlx5_find_master_dev(dev);
802 			if (master) {
803 				ifr = (struct ifreq) {
804 					.ifr_data = (void *)&edata,
805 				};
806 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
807 			}
808 		}
809 		if (ret) {
810 			DRV_LOG(WARNING,
811 				"port %u ioctl(SIOCETHTOOL,"
812 				" ETHTOOL_GSET) failed: %s",
813 				dev->data->port_id, strerror(rte_errno));
814 			return ret;
815 		}
816 	}
817 	link_speed = ethtool_cmd_speed(&edata);
818 	if (link_speed == -1)
819 		dev_link.link_speed = ETH_SPEED_NUM_NONE;
820 	else
821 		dev_link.link_speed = link_speed;
822 	priv->link_speed_capa = 0;
823 	if (edata.supported & SUPPORTED_Autoneg)
824 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
825 	if (edata.supported & (SUPPORTED_1000baseT_Full |
826 			       SUPPORTED_1000baseKX_Full))
827 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
828 	if (edata.supported & SUPPORTED_10000baseKR_Full)
829 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
830 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
831 			       SUPPORTED_40000baseCR4_Full |
832 			       SUPPORTED_40000baseSR4_Full |
833 			       SUPPORTED_40000baseLR4_Full))
834 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
835 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
836 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
837 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
838 			ETH_LINK_SPEED_FIXED);
839 	if (((dev_link.link_speed && !dev_link.link_status) ||
840 	     (!dev_link.link_speed && dev_link.link_status))) {
841 		rte_errno = EAGAIN;
842 		return -rte_errno;
843 	}
844 	*link = dev_link;
845 	return 0;
846 }
847 
848 /**
849  * Retrieve physical link information (unlocked version using new ioctl).
850  *
851  * @param dev
852  *   Pointer to Ethernet device structure.
853  * @param[out] link
854  *   Storage for current link status.
855  *
856  * @return
857  *   0 on success, a negative errno value otherwise and rte_errno is set.
858  */
859 static int
860 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
861 			     struct rte_eth_link *link)
862 
863 {
864 	struct mlx5_priv *priv = dev->data->dev_private;
865 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
866 	struct ifreq ifr;
867 	struct rte_eth_link dev_link;
868 	struct rte_eth_dev *master = NULL;
869 	uint64_t sc;
870 	int ret;
871 
872 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
873 	if (ret) {
874 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
875 			dev->data->port_id, strerror(rte_errno));
876 		return ret;
877 	}
878 	dev_link = (struct rte_eth_link) {
879 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
880 				(ifr.ifr_flags & IFF_RUNNING)),
881 	};
882 	ifr = (struct ifreq) {
883 		.ifr_data = (void *)&gcmd,
884 	};
885 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
886 	if (ret) {
887 		if (ret == -ENOTSUP && priv->representor) {
888 			/*
889 			 * For representors we can try to inherit link
890 			 * settings from the master device. Actually
891 			 * link settings do not make a lot of sense
892 			 * for representors due to missing physical
893 			 * link. The old kernel drivers supported
894 			 * emulated settings query for representors,
895 			 * the new ones do not, so we have to add
896 			 * this code for compatibility issues.
897 			 */
898 			master = mlx5_find_master_dev(dev);
899 			if (master) {
900 				ifr = (struct ifreq) {
901 					.ifr_data = (void *)&gcmd,
902 				};
903 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
904 			}
905 		}
906 		if (ret) {
907 			DRV_LOG(DEBUG,
908 				"port %u ioctl(SIOCETHTOOL,"
909 				" ETHTOOL_GLINKSETTINGS) failed: %s",
910 				dev->data->port_id, strerror(rte_errno));
911 			return ret;
912 		}
913 
914 	}
915 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
916 
917 	alignas(struct ethtool_link_settings)
918 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
919 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
920 	struct ethtool_link_settings *ecmd = (void *)data;
921 
922 	*ecmd = gcmd;
923 	ifr.ifr_data = (void *)ecmd;
924 	ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
925 	if (ret) {
926 		DRV_LOG(DEBUG,
927 			"port %u ioctl(SIOCETHTOOL,"
928 			"ETHTOOL_GLINKSETTINGS) failed: %s",
929 			dev->data->port_id, strerror(rte_errno));
930 		return ret;
931 	}
932 	dev_link.link_speed = (ecmd->speed == UINT32_MAX) ? ETH_SPEED_NUM_NONE :
933 							    ecmd->speed;
934 	sc = ecmd->link_mode_masks[0] |
935 		((uint64_t)ecmd->link_mode_masks[1] << 32);
936 	priv->link_speed_capa = 0;
937 	if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
938 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
939 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
940 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
941 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
942 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
943 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
944 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
945 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
946 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
947 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
948 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
949 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
950 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
951 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
952 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
953 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
954 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
955 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
956 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
957 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
958 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
959 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
960 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
961 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
962 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
963 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
964 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
965 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
966 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
967 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
968 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
969 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
970 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
971 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
972 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
973 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
974 				  ETH_LINK_SPEED_FIXED);
975 	if (((dev_link.link_speed && !dev_link.link_status) ||
976 	     (!dev_link.link_speed && dev_link.link_status))) {
977 		rte_errno = EAGAIN;
978 		return -rte_errno;
979 	}
980 	*link = dev_link;
981 	return 0;
982 }
983 
984 /**
985  * DPDK callback to retrieve physical link information.
986  *
987  * @param dev
988  *   Pointer to Ethernet device structure.
989  * @param wait_to_complete
990  *   Wait for request completion.
991  *
992  * @return
993  *   0 if link status was not updated, positive if it was, a negative errno
994  *   value otherwise and rte_errno is set.
995  */
996 int
997 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
998 {
999 	int ret;
1000 	struct rte_eth_link dev_link;
1001 	time_t start_time = time(NULL);
1002 	int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT;
1003 
1004 	do {
1005 		ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
1006 		if (ret == -ENOTSUP)
1007 			ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
1008 		if (ret == 0)
1009 			break;
1010 		/* Handle wait to complete situation. */
1011 		if ((wait_to_complete || retry) && ret == -EAGAIN) {
1012 			if (abs((int)difftime(time(NULL), start_time)) <
1013 			    MLX5_LINK_STATUS_TIMEOUT) {
1014 				usleep(0);
1015 				continue;
1016 			} else {
1017 				rte_errno = EBUSY;
1018 				return -rte_errno;
1019 			}
1020 		} else if (ret < 0) {
1021 			return ret;
1022 		}
1023 	} while (wait_to_complete || retry-- > 0);
1024 	ret = !!memcmp(&dev->data->dev_link, &dev_link,
1025 		       sizeof(struct rte_eth_link));
1026 	dev->data->dev_link = dev_link;
1027 	return ret;
1028 }
1029 
1030 /**
1031  * DPDK callback to change the MTU.
1032  *
1033  * @param dev
1034  *   Pointer to Ethernet device structure.
1035  * @param in_mtu
1036  *   New MTU.
1037  *
1038  * @return
1039  *   0 on success, a negative errno value otherwise and rte_errno is set.
1040  */
1041 int
1042 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1043 {
1044 	struct mlx5_priv *priv = dev->data->dev_private;
1045 	uint16_t kern_mtu = 0;
1046 	int ret;
1047 
1048 	ret = mlx5_get_mtu(dev, &kern_mtu);
1049 	if (ret)
1050 		return ret;
1051 	/* Set kernel interface MTU first. */
1052 	ret = mlx5_set_mtu(dev, mtu);
1053 	if (ret)
1054 		return ret;
1055 	ret = mlx5_get_mtu(dev, &kern_mtu);
1056 	if (ret)
1057 		return ret;
1058 	if (kern_mtu == mtu) {
1059 		priv->mtu = mtu;
1060 		DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
1061 			dev->data->port_id, mtu);
1062 		return 0;
1063 	}
1064 	rte_errno = EAGAIN;
1065 	return -rte_errno;
1066 }
1067 
1068 /**
1069  * DPDK callback to get flow control status.
1070  *
1071  * @param dev
1072  *   Pointer to Ethernet device structure.
1073  * @param[out] fc_conf
1074  *   Flow control output buffer.
1075  *
1076  * @return
1077  *   0 on success, a negative errno value otherwise and rte_errno is set.
1078  */
1079 int
1080 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1081 {
1082 	struct ifreq ifr;
1083 	struct ethtool_pauseparam ethpause = {
1084 		.cmd = ETHTOOL_GPAUSEPARAM
1085 	};
1086 	int ret;
1087 
1088 	ifr.ifr_data = (void *)&ethpause;
1089 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1090 	if (ret) {
1091 		DRV_LOG(WARNING,
1092 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
1093 			" %s",
1094 			dev->data->port_id, strerror(rte_errno));
1095 		return ret;
1096 	}
1097 	fc_conf->autoneg = ethpause.autoneg;
1098 	if (ethpause.rx_pause && ethpause.tx_pause)
1099 		fc_conf->mode = RTE_FC_FULL;
1100 	else if (ethpause.rx_pause)
1101 		fc_conf->mode = RTE_FC_RX_PAUSE;
1102 	else if (ethpause.tx_pause)
1103 		fc_conf->mode = RTE_FC_TX_PAUSE;
1104 	else
1105 		fc_conf->mode = RTE_FC_NONE;
1106 	return 0;
1107 }
1108 
1109 /**
1110  * DPDK callback to modify flow control parameters.
1111  *
1112  * @param dev
1113  *   Pointer to Ethernet device structure.
1114  * @param[in] fc_conf
1115  *   Flow control parameters.
1116  *
1117  * @return
1118  *   0 on success, a negative errno value otherwise and rte_errno is set.
1119  */
1120 int
1121 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1122 {
1123 	struct ifreq ifr;
1124 	struct ethtool_pauseparam ethpause = {
1125 		.cmd = ETHTOOL_SPAUSEPARAM
1126 	};
1127 	int ret;
1128 
1129 	ifr.ifr_data = (void *)&ethpause;
1130 	ethpause.autoneg = fc_conf->autoneg;
1131 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1132 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
1133 		ethpause.rx_pause = 1;
1134 	else
1135 		ethpause.rx_pause = 0;
1136 
1137 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1138 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
1139 		ethpause.tx_pause = 1;
1140 	else
1141 		ethpause.tx_pause = 0;
1142 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1143 	if (ret) {
1144 		DRV_LOG(WARNING,
1145 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1146 			" failed: %s",
1147 			dev->data->port_id, strerror(rte_errno));
1148 		return ret;
1149 	}
1150 	return 0;
1151 }
1152 
1153 /**
1154  * Get PCI information by sysfs device path.
1155  *
1156  * @param dev_path
1157  *   Pointer to device sysfs folder name.
1158  * @param[out] pci_addr
1159  *   PCI bus address output buffer.
1160  *
1161  * @return
1162  *   0 on success, a negative errno value otherwise and rte_errno is set.
1163  */
1164 int
1165 mlx5_dev_to_pci_addr(const char *dev_path,
1166 		     struct rte_pci_addr *pci_addr)
1167 {
1168 	FILE *file;
1169 	char line[32];
1170 	MKSTR(path, "%s/device/uevent", dev_path);
1171 
1172 	file = fopen(path, "rb");
1173 	if (file == NULL) {
1174 		rte_errno = errno;
1175 		return -rte_errno;
1176 	}
1177 	while (fgets(line, sizeof(line), file) == line) {
1178 		size_t len = strlen(line);
1179 		int ret;
1180 
1181 		/* Truncate long lines. */
1182 		if (len == (sizeof(line) - 1))
1183 			while (line[(len - 1)] != '\n') {
1184 				ret = fgetc(file);
1185 				if (ret == EOF)
1186 					break;
1187 				line[(len - 1)] = ret;
1188 			}
1189 		/* Extract information. */
1190 		if (sscanf(line,
1191 			   "PCI_SLOT_NAME="
1192 			   "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1193 			   &pci_addr->domain,
1194 			   &pci_addr->bus,
1195 			   &pci_addr->devid,
1196 			   &pci_addr->function) == 4) {
1197 			ret = 0;
1198 			break;
1199 		}
1200 	}
1201 	fclose(file);
1202 	return 0;
1203 }
1204 
1205 /**
1206  * Handle asynchronous removal event for entire multiport device.
1207  *
1208  * @param sh
1209  *   Infiniband device shared context.
1210  */
1211 static void
1212 mlx5_dev_interrupt_device_fatal(struct mlx5_ibv_shared *sh)
1213 {
1214 	uint32_t i;
1215 
1216 	for (i = 0; i < sh->max_port; ++i) {
1217 		struct rte_eth_dev *dev;
1218 
1219 		if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
1220 			/*
1221 			 * Or not existing port either no
1222 			 * handler installed for this port.
1223 			 */
1224 			continue;
1225 		}
1226 		dev = &rte_eth_devices[sh->port[i].ih_port_id];
1227 		assert(dev);
1228 		if (dev->data->dev_conf.intr_conf.rmv)
1229 			_rte_eth_dev_callback_process
1230 				(dev, RTE_ETH_EVENT_INTR_RMV, NULL);
1231 	}
1232 }
1233 
1234 /**
1235  * Handle shared asynchronous events the NIC (removal event
1236  * and link status change). Supports multiport IB device.
1237  *
1238  * @param cb_arg
1239  *   Callback argument.
1240  */
1241 void
1242 mlx5_dev_interrupt_handler(void *cb_arg)
1243 {
1244 	struct mlx5_ibv_shared *sh = cb_arg;
1245 	struct ibv_async_event event;
1246 
1247 	/* Read all message from the IB device and acknowledge them. */
1248 	for (;;) {
1249 		struct rte_eth_dev *dev;
1250 		uint32_t tmp;
1251 
1252 		if (mlx5_glue->get_async_event(sh->ctx, &event))
1253 			break;
1254 		/* Retrieve and check IB port index. */
1255 		tmp = (uint32_t)event.element.port_num;
1256 		if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
1257 			/*
1258 			 * The DEVICE_FATAL event is called once for
1259 			 * entire device without port specifying.
1260 			 * We should notify all existing ports.
1261 			 */
1262 			mlx5_glue->ack_async_event(&event);
1263 			mlx5_dev_interrupt_device_fatal(sh);
1264 			continue;
1265 		}
1266 		assert(tmp && (tmp <= sh->max_port));
1267 		if (!tmp) {
1268 			/* Unsupported devive level event. */
1269 			mlx5_glue->ack_async_event(&event);
1270 			DRV_LOG(DEBUG,
1271 				"unsupported common event (type %d)",
1272 				event.event_type);
1273 			continue;
1274 		}
1275 		if (tmp > sh->max_port) {
1276 			/* Invalid IB port index. */
1277 			mlx5_glue->ack_async_event(&event);
1278 			DRV_LOG(DEBUG,
1279 				"cannot handle an event (type %d)"
1280 				"due to invalid IB port index (%u)",
1281 				event.event_type, tmp);
1282 			continue;
1283 		}
1284 		if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
1285 			/* No handler installed. */
1286 			mlx5_glue->ack_async_event(&event);
1287 			DRV_LOG(DEBUG,
1288 				"cannot handle an event (type %d)"
1289 				"due to no handler installed for port %u",
1290 				event.event_type, tmp);
1291 			continue;
1292 		}
1293 		/* Retrieve ethernet device descriptor. */
1294 		tmp = sh->port[tmp - 1].ih_port_id;
1295 		dev = &rte_eth_devices[tmp];
1296 		assert(dev);
1297 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1298 		     event.event_type == IBV_EVENT_PORT_ERR) &&
1299 			dev->data->dev_conf.intr_conf.lsc) {
1300 			mlx5_glue->ack_async_event(&event);
1301 			if (mlx5_link_update(dev, 0) == -EAGAIN) {
1302 				usleep(0);
1303 				continue;
1304 			}
1305 			_rte_eth_dev_callback_process
1306 				(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
1307 			continue;
1308 		}
1309 		DRV_LOG(DEBUG,
1310 			"port %u cannot handle an unknown event (type %d)",
1311 			dev->data->port_id, event.event_type);
1312 		mlx5_glue->ack_async_event(&event);
1313 	}
1314 }
1315 
1316 /*
1317  * Unregister callback handler safely. The handler may be active
1318  * while we are trying to unregister it, in this case code -EAGAIN
1319  * is returned by rte_intr_callback_unregister(). This routine checks
1320  * the return code and tries to unregister handler again.
1321  *
1322  * @param handle
1323  *   interrupt handle
1324  * @param cb_fn
1325  *   pointer to callback routine
1326  * @cb_arg
1327  *   opaque callback parameter
1328  */
1329 void
1330 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
1331 			      rte_intr_callback_fn cb_fn, void *cb_arg)
1332 {
1333 	/*
1334 	 * Try to reduce timeout management overhead by not calling
1335 	 * the timer related routines on the first iteration. If the
1336 	 * unregistering succeeds on first call there will be no
1337 	 * timer calls at all.
1338 	 */
1339 	uint64_t twait = 0;
1340 	uint64_t start = 0;
1341 
1342 	do {
1343 		int ret;
1344 
1345 		ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
1346 		if (ret >= 0)
1347 			return;
1348 		if (ret != -EAGAIN) {
1349 			DRV_LOG(INFO, "failed to unregister interrupt"
1350 				      " handler (error: %d)", ret);
1351 			assert(false);
1352 			return;
1353 		}
1354 		if (twait) {
1355 			struct timespec onems;
1356 
1357 			/* Wait one millisecond and try again. */
1358 			onems.tv_sec = 0;
1359 			onems.tv_nsec = NS_PER_S / MS_PER_S;
1360 			nanosleep(&onems, 0);
1361 			/* Check whether one second elapsed. */
1362 			if ((rte_get_timer_cycles() - start) <= twait)
1363 				continue;
1364 		} else {
1365 			/*
1366 			 * We get the amount of timer ticks for one second.
1367 			 * If this amount elapsed it means we spent one
1368 			 * second in waiting. This branch is executed once
1369 			 * on first iteration.
1370 			 */
1371 			twait = rte_get_timer_hz();
1372 			assert(twait);
1373 		}
1374 		/*
1375 		 * Timeout elapsed, show message (once a second) and retry.
1376 		 * We have no other acceptable option here, if we ignore
1377 		 * the unregistering return code the handler will not
1378 		 * be unregistered, fd will be closed and we may get the
1379 		 * crush. Hanging and messaging in the loop seems not to be
1380 		 * the worst choice.
1381 		 */
1382 		DRV_LOG(INFO, "Retrying to unregister interrupt handler");
1383 		start = rte_get_timer_cycles();
1384 	} while (true);
1385 }
1386 
1387 /**
1388  * Handle DEVX interrupts from the NIC.
1389  * This function is probably called from the DPDK host thread.
1390  *
1391  * @param cb_arg
1392  *   Callback argument.
1393  */
1394 void
1395 mlx5_dev_interrupt_handler_devx(void *cb_arg)
1396 {
1397 #ifndef HAVE_IBV_DEVX_ASYNC
1398 	(void)cb_arg;
1399 	return;
1400 #else
1401 	struct mlx5_ibv_shared *sh = cb_arg;
1402 	union {
1403 		struct mlx5dv_devx_async_cmd_hdr cmd_resp;
1404 		uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
1405 			    MLX5_ST_SZ_BYTES(traffic_counter) +
1406 			    sizeof(struct mlx5dv_devx_async_cmd_hdr)];
1407 	} out;
1408 	uint8_t *buf = out.buf + sizeof(out.cmd_resp);
1409 
1410 	while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
1411 						   &out.cmd_resp,
1412 						   sizeof(out.buf)))
1413 		mlx5_flow_async_pool_query_handle
1414 			(sh, (uint64_t)out.cmd_resp.wr_id,
1415 			 mlx5_devx_get_out_command_status(buf));
1416 #endif /* HAVE_IBV_DEVX_ASYNC */
1417 }
1418 
1419 /**
1420  * Uninstall shared asynchronous device events handler.
1421  * This function is implemented to support event sharing
1422  * between multiple ports of single IB device.
1423  *
1424  * @param dev
1425  *   Pointer to Ethernet device.
1426  */
1427 static void
1428 mlx5_dev_shared_handler_uninstall(struct rte_eth_dev *dev)
1429 {
1430 	struct mlx5_priv *priv = dev->data->dev_private;
1431 	struct mlx5_ibv_shared *sh = priv->sh;
1432 
1433 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1434 		return;
1435 	pthread_mutex_lock(&sh->intr_mutex);
1436 	assert(priv->ibv_port);
1437 	assert(priv->ibv_port <= sh->max_port);
1438 	assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1439 	if (sh->port[priv->ibv_port - 1].ih_port_id >= RTE_MAX_ETHPORTS)
1440 		goto exit;
1441 	assert(sh->port[priv->ibv_port - 1].ih_port_id ==
1442 					(uint32_t)dev->data->port_id);
1443 	assert(sh->intr_cnt);
1444 	sh->port[priv->ibv_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
1445 	if (!sh->intr_cnt || --sh->intr_cnt)
1446 		goto exit;
1447 	mlx5_intr_callback_unregister(&sh->intr_handle,
1448 				     mlx5_dev_interrupt_handler, sh);
1449 	sh->intr_handle.fd = 0;
1450 	sh->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1451 exit:
1452 	pthread_mutex_unlock(&sh->intr_mutex);
1453 }
1454 
1455 /**
1456  * Uninstall devx shared asynchronous device events handler.
1457  * This function is implemeted to support event sharing
1458  * between multiple ports of single IB device.
1459  *
1460  * @param dev
1461  *   Pointer to Ethernet device.
1462  */
1463 static void
1464 mlx5_dev_shared_handler_devx_uninstall(struct rte_eth_dev *dev)
1465 {
1466 	struct mlx5_priv *priv = dev->data->dev_private;
1467 	struct mlx5_ibv_shared *sh = priv->sh;
1468 
1469 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1470 		return;
1471 	pthread_mutex_lock(&sh->intr_mutex);
1472 	assert(priv->ibv_port);
1473 	assert(priv->ibv_port <= sh->max_port);
1474 	assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1475 	if (sh->port[priv->ibv_port - 1].devx_ih_port_id >= RTE_MAX_ETHPORTS)
1476 		goto exit;
1477 	assert(sh->port[priv->ibv_port - 1].devx_ih_port_id ==
1478 					(uint32_t)dev->data->port_id);
1479 	sh->port[priv->ibv_port - 1].devx_ih_port_id = RTE_MAX_ETHPORTS;
1480 	if (!sh->devx_intr_cnt || --sh->devx_intr_cnt)
1481 		goto exit;
1482 	if (sh->intr_handle_devx.fd) {
1483 		rte_intr_callback_unregister(&sh->intr_handle_devx,
1484 					     mlx5_dev_interrupt_handler_devx,
1485 					     sh);
1486 		sh->intr_handle_devx.fd = 0;
1487 		sh->intr_handle_devx.type = RTE_INTR_HANDLE_UNKNOWN;
1488 	}
1489 	if (sh->devx_comp) {
1490 		mlx5_glue->devx_destroy_cmd_comp(sh->devx_comp);
1491 		sh->devx_comp = NULL;
1492 	}
1493 exit:
1494 	pthread_mutex_unlock(&sh->intr_mutex);
1495 }
1496 
1497 /**
1498  * Install shared asynchronous device events handler.
1499  * This function is implemented to support event sharing
1500  * between multiple ports of single IB device.
1501  *
1502  * @param dev
1503  *   Pointer to Ethernet device.
1504  */
1505 static void
1506 mlx5_dev_shared_handler_install(struct rte_eth_dev *dev)
1507 {
1508 	struct mlx5_priv *priv = dev->data->dev_private;
1509 	struct mlx5_ibv_shared *sh = priv->sh;
1510 	int ret;
1511 	int flags;
1512 
1513 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1514 		return;
1515 	pthread_mutex_lock(&sh->intr_mutex);
1516 	assert(priv->ibv_port);
1517 	assert(priv->ibv_port <= sh->max_port);
1518 	assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1519 	if (sh->port[priv->ibv_port - 1].ih_port_id < RTE_MAX_ETHPORTS) {
1520 		/* The handler is already installed for this port. */
1521 		assert(sh->intr_cnt);
1522 		goto exit;
1523 	}
1524 	if (sh->intr_cnt) {
1525 		sh->port[priv->ibv_port - 1].ih_port_id =
1526 						(uint32_t)dev->data->port_id;
1527 		sh->intr_cnt++;
1528 		goto exit;
1529 	}
1530 	/* No shared handler installed. */
1531 	assert(sh->ctx->async_fd > 0);
1532 	flags = fcntl(sh->ctx->async_fd, F_GETFL);
1533 	ret = fcntl(sh->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1534 	if (ret) {
1535 		DRV_LOG(INFO, "failed to change file descriptor async event"
1536 			" queue");
1537 		/* Indicate there will be no interrupts. */
1538 		dev->data->dev_conf.intr_conf.lsc = 0;
1539 		dev->data->dev_conf.intr_conf.rmv = 0;
1540 	} else {
1541 		sh->intr_handle.fd = sh->ctx->async_fd;
1542 		sh->intr_handle.type = RTE_INTR_HANDLE_EXT;
1543 		rte_intr_callback_register(&sh->intr_handle,
1544 					   mlx5_dev_interrupt_handler, sh);
1545 		sh->intr_cnt++;
1546 		sh->port[priv->ibv_port - 1].ih_port_id =
1547 						(uint32_t)dev->data->port_id;
1548 	}
1549 exit:
1550 	pthread_mutex_unlock(&sh->intr_mutex);
1551 }
1552 
1553 /**
1554  * Install devx shared asyncronous device events handler.
1555  * This function is implemeted to support event sharing
1556  * between multiple ports of single IB device.
1557  *
1558  * @param dev
1559  *   Pointer to Ethernet device.
1560  */
1561 static void
1562 mlx5_dev_shared_handler_devx_install(struct rte_eth_dev *dev)
1563 {
1564 	struct mlx5_priv *priv = dev->data->dev_private;
1565 	struct mlx5_ibv_shared *sh = priv->sh;
1566 
1567 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1568 		return;
1569 	pthread_mutex_lock(&sh->intr_mutex);
1570 	assert(priv->ibv_port);
1571 	assert(priv->ibv_port <= sh->max_port);
1572 	assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1573 	if (sh->port[priv->ibv_port - 1].devx_ih_port_id < RTE_MAX_ETHPORTS) {
1574 		/* The handler is already installed for this port. */
1575 		assert(sh->devx_intr_cnt);
1576 		goto exit;
1577 	}
1578 	if (sh->devx_intr_cnt) {
1579 		sh->devx_intr_cnt++;
1580 		sh->port[priv->ibv_port - 1].devx_ih_port_id =
1581 					(uint32_t)dev->data->port_id;
1582 		goto exit;
1583 	}
1584 	if (priv->config.devx) {
1585 #ifndef HAVE_IBV_DEVX_ASYNC
1586 		goto exit;
1587 #else
1588 		sh->devx_comp = mlx5_glue->devx_create_cmd_comp(sh->ctx);
1589 		if (sh->devx_comp) {
1590 			int flags = fcntl(sh->devx_comp->fd, F_GETFL);
1591 			int ret = fcntl(sh->devx_comp->fd, F_SETFL,
1592 				    flags | O_NONBLOCK);
1593 
1594 			if (ret) {
1595 				DRV_LOG(INFO, "failed to change file descriptor"
1596 					" devx async event queue");
1597 			} else {
1598 				sh->intr_handle_devx.fd = sh->devx_comp->fd;
1599 				sh->intr_handle_devx.type = RTE_INTR_HANDLE_EXT;
1600 				rte_intr_callback_register
1601 					(&sh->intr_handle_devx,
1602 					 mlx5_dev_interrupt_handler_devx, sh);
1603 				sh->devx_intr_cnt++;
1604 				sh->port[priv->ibv_port - 1].devx_ih_port_id =
1605 						(uint32_t)dev->data->port_id;
1606 			}
1607 		}
1608 #endif /* HAVE_IBV_DEVX_ASYNC */
1609 	}
1610 exit:
1611 	pthread_mutex_unlock(&sh->intr_mutex);
1612 }
1613 
1614 /**
1615  * Uninstall interrupt handler.
1616  *
1617  * @param dev
1618  *   Pointer to Ethernet device.
1619  */
1620 void
1621 mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev)
1622 {
1623 	mlx5_dev_shared_handler_uninstall(dev);
1624 }
1625 
1626 /**
1627  * Install interrupt handler.
1628  *
1629  * @param dev
1630  *   Pointer to Ethernet device.
1631  */
1632 void
1633 mlx5_dev_interrupt_handler_install(struct rte_eth_dev *dev)
1634 {
1635 	mlx5_dev_shared_handler_install(dev);
1636 }
1637 
1638 /**
1639  * Devx uninstall interrupt handler.
1640  *
1641  * @param dev
1642  *   Pointer to Ethernet device.
1643  */
1644 void
1645 mlx5_dev_interrupt_handler_devx_uninstall(struct rte_eth_dev *dev)
1646 {
1647 	mlx5_dev_shared_handler_devx_uninstall(dev);
1648 }
1649 
1650 /**
1651  * Devx install interrupt handler.
1652  *
1653  * @param dev
1654  *   Pointer to Ethernet device.
1655  */
1656 void
1657 mlx5_dev_interrupt_handler_devx_install(struct rte_eth_dev *dev)
1658 {
1659 	mlx5_dev_shared_handler_devx_install(dev);
1660 }
1661 
1662 /**
1663  * DPDK callback to bring the link DOWN.
1664  *
1665  * @param dev
1666  *   Pointer to Ethernet device structure.
1667  *
1668  * @return
1669  *   0 on success, a negative errno value otherwise and rte_errno is set.
1670  */
1671 int
1672 mlx5_set_link_down(struct rte_eth_dev *dev)
1673 {
1674 	return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
1675 }
1676 
1677 /**
1678  * DPDK callback to bring the link UP.
1679  *
1680  * @param dev
1681  *   Pointer to Ethernet device structure.
1682  *
1683  * @return
1684  *   0 on success, a negative errno value otherwise and rte_errno is set.
1685  */
1686 int
1687 mlx5_set_link_up(struct rte_eth_dev *dev)
1688 {
1689 	return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
1690 }
1691 
1692 /**
1693  * Configure the RX function to use.
1694  *
1695  * @param dev
1696  *   Pointer to private data structure.
1697  *
1698  * @return
1699  *   Pointer to selected Rx burst function.
1700  */
1701 eth_rx_burst_t
1702 mlx5_select_rx_function(struct rte_eth_dev *dev)
1703 {
1704 	eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
1705 
1706 	assert(dev != NULL);
1707 	if (mlx5_check_vec_rx_support(dev) > 0) {
1708 		rx_pkt_burst = mlx5_rx_burst_vec;
1709 		DRV_LOG(DEBUG, "port %u selected Rx vectorized function",
1710 			dev->data->port_id);
1711 	} else if (mlx5_mprq_enabled(dev)) {
1712 		rx_pkt_burst = mlx5_rx_burst_mprq;
1713 	}
1714 	return rx_pkt_burst;
1715 }
1716 
1717 /**
1718  * Check if mlx5 device was removed.
1719  *
1720  * @param dev
1721  *   Pointer to Ethernet device structure.
1722  *
1723  * @return
1724  *   1 when device is removed, otherwise 0.
1725  */
1726 int
1727 mlx5_is_removed(struct rte_eth_dev *dev)
1728 {
1729 	struct ibv_device_attr device_attr;
1730 	struct mlx5_priv *priv = dev->data->dev_private;
1731 
1732 	if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
1733 		return 1;
1734 	return 0;
1735 }
1736 
1737 /**
1738  * Get the E-Switch parameters by port id.
1739  *
1740  * @param[in] port
1741  *   Device port id.
1742  * @param[out] es_domain_id
1743  *   E-Switch domain id.
1744  * @param[out] es_port_id
1745  *   The port id of the port in the E-Switch.
1746  *
1747  * @return
1748  *   pointer to device private data structure containing data needed
1749  *   on success, NULL otherwise and rte_errno is set.
1750  */
1751 struct mlx5_priv *
1752 mlx5_port_to_eswitch_info(uint16_t port)
1753 {
1754 	struct rte_eth_dev *dev;
1755 	struct mlx5_priv *priv;
1756 
1757 	if (port >= RTE_MAX_ETHPORTS) {
1758 		rte_errno = EINVAL;
1759 		return NULL;
1760 	}
1761 	if (!rte_eth_dev_is_valid_port(port)) {
1762 		rte_errno = ENODEV;
1763 		return NULL;
1764 	}
1765 	dev = &rte_eth_devices[port];
1766 	priv = dev->data->dev_private;
1767 	if (!(priv->representor || priv->master)) {
1768 		rte_errno = EINVAL;
1769 		return NULL;
1770 	}
1771 	return priv;
1772 }
1773 
1774 /**
1775  * Get the E-Switch parameters by device instance.
1776  *
1777  * @param[in] port
1778  *   Device port id.
1779  * @param[out] es_domain_id
1780  *   E-Switch domain id.
1781  * @param[out] es_port_id
1782  *   The port id of the port in the E-Switch.
1783  *
1784  * @return
1785  *   pointer to device private data structure containing data needed
1786  *   on success, NULL otherwise and rte_errno is set.
1787  */
1788 struct mlx5_priv *
1789 mlx5_dev_to_eswitch_info(struct rte_eth_dev *dev)
1790 {
1791 	struct mlx5_priv *priv;
1792 
1793 	priv = dev->data->dev_private;
1794 	if (!(priv->representor || priv->master)) {
1795 		rte_errno = EINVAL;
1796 		return NULL;
1797 	}
1798 	return priv;
1799 }
1800 
1801 /**
1802  * Get switch information associated with network interface.
1803  *
1804  * @param ifindex
1805  *   Network interface index.
1806  * @param[out] info
1807  *   Switch information object, populated in case of success.
1808  *
1809  * @return
1810  *   0 on success, a negative errno value otherwise and rte_errno is set.
1811  */
1812 int
1813 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1814 {
1815 	char ifname[IF_NAMESIZE];
1816 	char port_name[IF_NAMESIZE];
1817 	FILE *file;
1818 	struct mlx5_switch_info data = {
1819 		.master = 0,
1820 		.representor = 0,
1821 		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1822 		.port_name = 0,
1823 		.switch_id = 0,
1824 	};
1825 	DIR *dir;
1826 	bool port_switch_id_set = false;
1827 	bool device_dir = false;
1828 	char c;
1829 	int ret;
1830 
1831 	if (!if_indextoname(ifindex, ifname)) {
1832 		rte_errno = errno;
1833 		return -rte_errno;
1834 	}
1835 
1836 	MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1837 	      ifname);
1838 	MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1839 	      ifname);
1840 	MKSTR(pci_device, "/sys/class/net/%s/device",
1841 	      ifname);
1842 
1843 	file = fopen(phys_port_name, "rb");
1844 	if (file != NULL) {
1845 		ret = fscanf(file, "%s", port_name);
1846 		fclose(file);
1847 		if (ret == 1)
1848 			mlx5_translate_port_name(port_name, &data);
1849 	}
1850 	file = fopen(phys_switch_id, "rb");
1851 	if (file == NULL) {
1852 		rte_errno = errno;
1853 		return -rte_errno;
1854 	}
1855 	port_switch_id_set =
1856 		fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1857 		c == '\n';
1858 	fclose(file);
1859 	dir = opendir(pci_device);
1860 	if (dir != NULL) {
1861 		closedir(dir);
1862 		device_dir = true;
1863 	}
1864 	if (port_switch_id_set) {
1865 		/* We have some E-Switch configuration. */
1866 		mlx5_sysfs_check_switch_info(device_dir, &data);
1867 	}
1868 	*info = data;
1869 	assert(!(data.master && data.representor));
1870 	if (data.master && data.representor) {
1871 		DRV_LOG(ERR, "ifindex %u device is recognized as master"
1872 			     " and as representor", ifindex);
1873 		rte_errno = ENODEV;
1874 		return -rte_errno;
1875 	}
1876 	return 0;
1877 }
1878 
1879 /**
1880  * Analyze gathered port parameters via Netlink to recognize master
1881  * and representor devices for E-Switch configuration.
1882  *
1883  * @param[in] num_vf_set
1884  *   flag of presence of number of VFs port attribute.
1885  * @param[inout] switch_info
1886  *   Port information, including port name as a number and port name
1887  *   type if recognized
1888  *
1889  * @return
1890  *   master and representor flags are set in switch_info according to
1891  *   recognized parameters (if any).
1892  */
1893 void
1894 mlx5_nl_check_switch_info(bool num_vf_set,
1895 			  struct mlx5_switch_info *switch_info)
1896 {
1897 	switch (switch_info->name_type) {
1898 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1899 		/*
1900 		 * Name is not recognized, assume the master,
1901 		 * check the number of VFs key presence.
1902 		 */
1903 		switch_info->master = num_vf_set;
1904 		break;
1905 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1906 		/*
1907 		 * Name is not set, this assumes the legacy naming
1908 		 * schema for master, just check if there is a
1909 		 * number of VFs key.
1910 		 */
1911 		switch_info->master = num_vf_set;
1912 		break;
1913 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1914 		/* New uplink naming schema recognized. */
1915 		switch_info->master = 1;
1916 		break;
1917 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1918 		/* Legacy representors naming schema. */
1919 		switch_info->representor = !num_vf_set;
1920 		break;
1921 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1922 		/* New representors naming schema. */
1923 		switch_info->representor = 1;
1924 		break;
1925 	}
1926 }
1927 
1928 /**
1929  * Analyze gathered port parameters via sysfs to recognize master
1930  * and representor devices for E-Switch configuration.
1931  *
1932  * @param[in] device_dir
1933  *   flag of presence of "device" directory under port device key.
1934  * @param[inout] switch_info
1935  *   Port information, including port name as a number and port name
1936  *   type if recognized
1937  *
1938  * @return
1939  *   master and representor flags are set in switch_info according to
1940  *   recognized parameters (if any).
1941  */
1942 void
1943 mlx5_sysfs_check_switch_info(bool device_dir,
1944 			     struct mlx5_switch_info *switch_info)
1945 {
1946 	switch (switch_info->name_type) {
1947 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1948 		/*
1949 		 * Name is not recognized, assume the master,
1950 		 * check the device directory presence.
1951 		 */
1952 		switch_info->master = device_dir;
1953 		break;
1954 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1955 		/*
1956 		 * Name is not set, this assumes the legacy naming
1957 		 * schema for master, just check if there is
1958 		 * a device directory.
1959 		 */
1960 		switch_info->master = device_dir;
1961 		break;
1962 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1963 		/* New uplink naming schema recognized. */
1964 		switch_info->master = 1;
1965 		break;
1966 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1967 		/* Legacy representors naming schema. */
1968 		switch_info->representor = !device_dir;
1969 		break;
1970 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1971 		/* New representors naming schema. */
1972 		switch_info->representor = 1;
1973 		break;
1974 	}
1975 }
1976 
1977 /**
1978  * Extract port name, as a number, from sysfs or netlink information.
1979  *
1980  * @param[in] port_name_in
1981  *   String representing the port name.
1982  * @param[out] port_info_out
1983  *   Port information, including port name as a number and port name
1984  *   type if recognized
1985  *
1986  * @return
1987  *   port_name field set according to recognized name format.
1988  */
1989 void
1990 mlx5_translate_port_name(const char *port_name_in,
1991 			 struct mlx5_switch_info *port_info_out)
1992 {
1993 	char pf_c1, pf_c2, vf_c1, vf_c2;
1994 	char *end;
1995 	int sc_items;
1996 
1997 	/*
1998 	 * Check for port-name as a string of the form pf0vf0
1999 	 * (support kernel ver >= 5.0 or OFED ver >= 4.6).
2000 	 */
2001 	sc_items = sscanf(port_name_in, "%c%c%d%c%c%d",
2002 			  &pf_c1, &pf_c2, &port_info_out->pf_num,
2003 			  &vf_c1, &vf_c2, &port_info_out->port_name);
2004 	if (sc_items == 6 &&
2005 	    pf_c1 == 'p' && pf_c2 == 'f' &&
2006 	    vf_c1 == 'v' && vf_c2 == 'f') {
2007 		port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_PFVF;
2008 		return;
2009 	}
2010 	/*
2011 	 * Check for port-name as a string of the form p0
2012 	 * (support kernel ver >= 5.0, or OFED ver >= 4.6).
2013 	 */
2014 	sc_items = sscanf(port_name_in, "%c%d",
2015 			  &pf_c1, &port_info_out->port_name);
2016 	if (sc_items == 2 && pf_c1 == 'p') {
2017 		port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UPLINK;
2018 		return;
2019 	}
2020 	/* Check for port-name as a number (support kernel ver < 5.0 */
2021 	errno = 0;
2022 	port_info_out->port_name = strtol(port_name_in, &end, 0);
2023 	if (!errno &&
2024 	    (size_t)(end - port_name_in) == strlen(port_name_in)) {
2025 		port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_LEGACY;
2026 		return;
2027 	}
2028 	port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN;
2029 	return;
2030 }
2031 
2032 /**
2033  * DPDK callback to retrieve plug-in module EEPROM information (type and size).
2034  *
2035  * @param dev
2036  *   Pointer to Ethernet device structure.
2037  * @param[out] modinfo
2038  *   Storage for plug-in module EEPROM information.
2039  *
2040  * @return
2041  *   0 on success, a negative errno value otherwise and rte_errno is set.
2042  */
2043 int
2044 mlx5_get_module_info(struct rte_eth_dev *dev,
2045 		     struct rte_eth_dev_module_info *modinfo)
2046 {
2047 	struct ethtool_modinfo info = {
2048 		.cmd = ETHTOOL_GMODULEINFO,
2049 	};
2050 	struct ifreq ifr = (struct ifreq) {
2051 		.ifr_data = (void *)&info,
2052 	};
2053 	int ret = 0;
2054 
2055 	if (!dev || !modinfo) {
2056 		DRV_LOG(WARNING, "missing argument, cannot get module info");
2057 		rte_errno = EINVAL;
2058 		return -rte_errno;
2059 	}
2060 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
2061 	if (ret) {
2062 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
2063 			dev->data->port_id, strerror(rte_errno));
2064 		return ret;
2065 	}
2066 	modinfo->type = info.type;
2067 	modinfo->eeprom_len = info.eeprom_len;
2068 	return ret;
2069 }
2070 
2071 /**
2072  * DPDK callback to retrieve plug-in module EEPROM data.
2073  *
2074  * @param dev
2075  *   Pointer to Ethernet device structure.
2076  * @param[out] info
2077  *   Storage for plug-in module EEPROM data.
2078  *
2079  * @return
2080  *   0 on success, a negative errno value otherwise and rte_errno is set.
2081  */
2082 int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
2083 			   struct rte_dev_eeprom_info *info)
2084 {
2085 	struct ethtool_eeprom *eeprom;
2086 	struct ifreq ifr;
2087 	int ret = 0;
2088 
2089 	if (!dev || !info) {
2090 		DRV_LOG(WARNING, "missing argument, cannot get module eeprom");
2091 		rte_errno = EINVAL;
2092 		return -rte_errno;
2093 	}
2094 	eeprom = rte_calloc(__func__, 1,
2095 			    (sizeof(struct ethtool_eeprom) + info->length), 0);
2096 	if (!eeprom) {
2097 		DRV_LOG(WARNING, "port %u cannot allocate memory for "
2098 			"eeprom data", dev->data->port_id);
2099 		rte_errno = ENOMEM;
2100 		return -rte_errno;
2101 	}
2102 	eeprom->cmd = ETHTOOL_GMODULEEEPROM;
2103 	eeprom->offset = info->offset;
2104 	eeprom->len = info->length;
2105 	ifr = (struct ifreq) {
2106 		.ifr_data = (void *)eeprom,
2107 	};
2108 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
2109 	if (ret)
2110 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
2111 			dev->data->port_id, strerror(rte_errno));
2112 	else
2113 		rte_memcpy(info->data, eeprom->data, info->length);
2114 	rte_free(eeprom);
2115 	return ret;
2116 }
2117