xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision 90260d577b61685365b7cf1a6151b9ed6b108f2d)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright 2015 6WIND S.A.
5  *   Copyright 2015 Mellanox.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of 6WIND S.A. nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stddef.h>
35 #include <assert.h>
36 #include <unistd.h>
37 #include <stdint.h>
38 #include <stdio.h>
39 #include <string.h>
40 #include <stdlib.h>
41 #include <errno.h>
42 #include <dirent.h>
43 #include <net/if.h>
44 #include <sys/ioctl.h>
45 #include <sys/socket.h>
46 #include <sys/utsname.h>
47 #include <netinet/in.h>
48 #include <linux/ethtool.h>
49 #include <linux/sockios.h>
50 #include <linux/version.h>
51 #include <fcntl.h>
52 #include <stdalign.h>
53 
54 #include <rte_atomic.h>
55 #include <rte_ethdev.h>
56 #include <rte_mbuf.h>
57 #include <rte_common.h>
58 #include <rte_interrupts.h>
59 #include <rte_alarm.h>
60 #include <rte_malloc.h>
61 
62 #include "mlx5.h"
63 #include "mlx5_rxtx.h"
64 #include "mlx5_utils.h"
65 
66 /* Add defines in case the running kernel is not the same as user headers. */
67 #ifndef ETHTOOL_GLINKSETTINGS
68 struct ethtool_link_settings {
69 	uint32_t cmd;
70 	uint32_t speed;
71 	uint8_t duplex;
72 	uint8_t port;
73 	uint8_t phy_address;
74 	uint8_t autoneg;
75 	uint8_t mdio_support;
76 	uint8_t eth_to_mdix;
77 	uint8_t eth_tp_mdix_ctrl;
78 	int8_t link_mode_masks_nwords;
79 	uint32_t reserved[8];
80 	uint32_t link_mode_masks[];
81 };
82 
83 #define ETHTOOL_GLINKSETTINGS 0x0000004c
84 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
85 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
86 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
87 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
88 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
89 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
90 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
91 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
92 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
93 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
94 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
95 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
96 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
97 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
98 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
99 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
100 #endif
101 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
102 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
103 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
104 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
105 #endif
106 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
107 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
108 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
109 #endif
110 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
111 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
112 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
113 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
114 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
115 #endif
116 #define ETHTOOL_LINK_MODE_MASK_MAX_KERNEL_NU32 (SCHAR_MAX)
117 
118 /**
119  * Return private structure associated with an Ethernet device.
120  *
121  * @param dev
122  *   Pointer to Ethernet device structure.
123  *
124  * @return
125  *   Pointer to private structure.
126  */
127 struct priv *
128 mlx5_get_priv(struct rte_eth_dev *dev)
129 {
130 	return dev->data->dev_private;
131 }
132 
133 /**
134  * Check if running as a secondary process.
135  *
136  * @return
137  *   Nonzero if running as a secondary process.
138  */
139 inline int
140 mlx5_is_secondary(void)
141 {
142 	return rte_eal_process_type() == RTE_PROC_SECONDARY;
143 }
144 
145 /**
146  * Get interface name from private structure.
147  *
148  * @param[in] priv
149  *   Pointer to private structure.
150  * @param[out] ifname
151  *   Interface name output buffer.
152  *
153  * @return
154  *   0 on success, -1 on failure and errno is set.
155  */
156 int
157 priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
158 {
159 	DIR *dir;
160 	struct dirent *dent;
161 	unsigned int dev_type = 0;
162 	unsigned int dev_port_prev = ~0u;
163 	char match[IF_NAMESIZE] = "";
164 
165 	{
166 		MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
167 
168 		dir = opendir(path);
169 		if (dir == NULL)
170 			return -1;
171 	}
172 	while ((dent = readdir(dir)) != NULL) {
173 		char *name = dent->d_name;
174 		FILE *file;
175 		unsigned int dev_port;
176 		int r;
177 
178 		if ((name[0] == '.') &&
179 		    ((name[1] == '\0') ||
180 		     ((name[1] == '.') && (name[2] == '\0'))))
181 			continue;
182 
183 		MKSTR(path, "%s/device/net/%s/%s",
184 		      priv->ctx->device->ibdev_path, name,
185 		      (dev_type ? "dev_id" : "dev_port"));
186 
187 		file = fopen(path, "rb");
188 		if (file == NULL) {
189 			if (errno != ENOENT)
190 				continue;
191 			/*
192 			 * Switch to dev_id when dev_port does not exist as
193 			 * is the case with Linux kernel versions < 3.15.
194 			 */
195 try_dev_id:
196 			match[0] = '\0';
197 			if (dev_type)
198 				break;
199 			dev_type = 1;
200 			dev_port_prev = ~0u;
201 			rewinddir(dir);
202 			continue;
203 		}
204 		r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
205 		fclose(file);
206 		if (r != 1)
207 			continue;
208 		/*
209 		 * Switch to dev_id when dev_port returns the same value for
210 		 * all ports. May happen when using a MOFED release older than
211 		 * 3.0 with a Linux kernel >= 3.15.
212 		 */
213 		if (dev_port == dev_port_prev)
214 			goto try_dev_id;
215 		dev_port_prev = dev_port;
216 		if (dev_port == (priv->port - 1u))
217 			snprintf(match, sizeof(match), "%s", name);
218 	}
219 	closedir(dir);
220 	if (match[0] == '\0')
221 		return -1;
222 	strncpy(*ifname, match, sizeof(*ifname));
223 	return 0;
224 }
225 
226 /**
227  * Check if the counter is located on ib counters file.
228  *
229  * @param[in] cntr
230  *   Counter name.
231  *
232  * @return
233  *   1 if counter is located on ib counters file , 0 otherwise.
234  */
235 int
236 priv_is_ib_cntr(const char *cntr)
237 {
238 	if (!strcmp(cntr, "out_of_buffer"))
239 		return 1;
240 	return 0;
241 }
242 
243 /**
244  * Read from sysfs entry.
245  *
246  * @param[in] priv
247  *   Pointer to private structure.
248  * @param[in] entry
249  *   Entry name relative to sysfs path.
250  * @param[out] buf
251  *   Data output buffer.
252  * @param size
253  *   Buffer size.
254  *
255  * @return
256  *   0 on success, -1 on failure and errno is set.
257  */
258 static int
259 priv_sysfs_read(const struct priv *priv, const char *entry,
260 		char *buf, size_t size)
261 {
262 	char ifname[IF_NAMESIZE];
263 	FILE *file;
264 	int ret;
265 	int err;
266 
267 	if (priv_get_ifname(priv, &ifname))
268 		return -1;
269 
270 	if (priv_is_ib_cntr(entry)) {
271 		MKSTR(path, "%s/ports/1/hw_counters/%s",
272 		      priv->ctx->device->ibdev_path, entry);
273 		file = fopen(path, "rb");
274 	} else {
275 		MKSTR(path, "%s/device/net/%s/%s",
276 		      priv->ctx->device->ibdev_path, ifname, entry);
277 		file = fopen(path, "rb");
278 	}
279 	if (file == NULL)
280 		return -1;
281 	ret = fread(buf, 1, size, file);
282 	err = errno;
283 	if (((size_t)ret < size) && (ferror(file)))
284 		ret = -1;
285 	else
286 		ret = size;
287 	fclose(file);
288 	errno = err;
289 	return ret;
290 }
291 
292 /**
293  * Write to sysfs entry.
294  *
295  * @param[in] priv
296  *   Pointer to private structure.
297  * @param[in] entry
298  *   Entry name relative to sysfs path.
299  * @param[in] buf
300  *   Data buffer.
301  * @param size
302  *   Buffer size.
303  *
304  * @return
305  *   0 on success, -1 on failure and errno is set.
306  */
307 static int
308 priv_sysfs_write(const struct priv *priv, const char *entry,
309 		 char *buf, size_t size)
310 {
311 	char ifname[IF_NAMESIZE];
312 	FILE *file;
313 	int ret;
314 	int err;
315 
316 	if (priv_get_ifname(priv, &ifname))
317 		return -1;
318 
319 	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
320 	      ifname, entry);
321 
322 	file = fopen(path, "wb");
323 	if (file == NULL)
324 		return -1;
325 	ret = fwrite(buf, 1, size, file);
326 	err = errno;
327 	if (((size_t)ret < size) || (ferror(file)))
328 		ret = -1;
329 	else
330 		ret = size;
331 	fclose(file);
332 	errno = err;
333 	return ret;
334 }
335 
336 /**
337  * Get unsigned long sysfs property.
338  *
339  * @param priv
340  *   Pointer to private structure.
341  * @param[in] name
342  *   Entry name relative to sysfs path.
343  * @param[out] value
344  *   Value output buffer.
345  *
346  * @return
347  *   0 on success, -1 on failure and errno is set.
348  */
349 static int
350 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
351 {
352 	int ret;
353 	unsigned long value_ret;
354 	char value_str[32];
355 
356 	ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
357 	if (ret == -1) {
358 		DEBUG("cannot read %s value from sysfs: %s",
359 		      name, strerror(errno));
360 		return -1;
361 	}
362 	value_str[ret] = '\0';
363 	errno = 0;
364 	value_ret = strtoul(value_str, NULL, 0);
365 	if (errno) {
366 		DEBUG("invalid %s value `%s': %s", name, value_str,
367 		      strerror(errno));
368 		return -1;
369 	}
370 	*value = value_ret;
371 	return 0;
372 }
373 
374 /**
375  * Set unsigned long sysfs property.
376  *
377  * @param priv
378  *   Pointer to private structure.
379  * @param[in] name
380  *   Entry name relative to sysfs path.
381  * @param value
382  *   Value to set.
383  *
384  * @return
385  *   0 on success, -1 on failure and errno is set.
386  */
387 static int
388 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
389 {
390 	int ret;
391 	MKSTR(value_str, "%lu", value);
392 
393 	ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
394 	if (ret == -1) {
395 		DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
396 		      name, value_str, value, strerror(errno));
397 		return -1;
398 	}
399 	return 0;
400 }
401 
402 /**
403  * Perform ifreq ioctl() on associated Ethernet device.
404  *
405  * @param[in] priv
406  *   Pointer to private structure.
407  * @param req
408  *   Request number to pass to ioctl().
409  * @param[out] ifr
410  *   Interface request structure output buffer.
411  *
412  * @return
413  *   0 on success, -1 on failure and errno is set.
414  */
415 int
416 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
417 {
418 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
419 	int ret = -1;
420 
421 	if (sock == -1)
422 		return ret;
423 	if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
424 		ret = ioctl(sock, req, ifr);
425 	close(sock);
426 	return ret;
427 }
428 
429 /**
430  * Return the number of active VFs for the current device.
431  *
432  * @param[in] priv
433  *   Pointer to private structure.
434  * @param[out] num_vfs
435  *   Number of active VFs.
436  *
437  * @return
438  *   0 on success, -1 on failure and errno is set.
439  */
440 int
441 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
442 {
443 	/* The sysfs entry name depends on the operating system. */
444 	const char **name = (const char *[]){
445 		"device/sriov_numvfs",
446 		"device/mlx5_num_vfs",
447 		NULL,
448 	};
449 	int ret;
450 
451 	do {
452 		unsigned long ulong_num_vfs;
453 
454 		ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
455 		if (!ret)
456 			*num_vfs = ulong_num_vfs;
457 	} while (*(++name) && ret);
458 	return ret;
459 }
460 
461 /**
462  * Get device MTU.
463  *
464  * @param priv
465  *   Pointer to private structure.
466  * @param[out] mtu
467  *   MTU value output buffer.
468  *
469  * @return
470  *   0 on success, -1 on failure and errno is set.
471  */
472 int
473 priv_get_mtu(struct priv *priv, uint16_t *mtu)
474 {
475 	unsigned long ulong_mtu;
476 
477 	if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
478 		return -1;
479 	*mtu = ulong_mtu;
480 	return 0;
481 }
482 
483 /**
484  * Read device counter from sysfs.
485  *
486  * @param priv
487  *   Pointer to private structure.
488  * @param name
489  *   Counter name.
490  * @param[out] cntr
491  *   Counter output buffer.
492  *
493  * @return
494  *   0 on success, -1 on failure and errno is set.
495  */
496 int
497 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
498 {
499 	unsigned long ulong_ctr;
500 
501 	if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
502 		return -1;
503 	*cntr = ulong_ctr;
504 	return 0;
505 }
506 
507 /**
508  * Set device MTU.
509  *
510  * @param priv
511  *   Pointer to private structure.
512  * @param mtu
513  *   MTU value to set.
514  *
515  * @return
516  *   0 on success, -1 on failure and errno is set.
517  */
518 static int
519 priv_set_mtu(struct priv *priv, uint16_t mtu)
520 {
521 	uint16_t new_mtu;
522 
523 	if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
524 	    priv_get_mtu(priv, &new_mtu))
525 		return -1;
526 	if (new_mtu == mtu)
527 		return 0;
528 	errno = EINVAL;
529 	return -1;
530 }
531 
532 /**
533  * Set device flags.
534  *
535  * @param priv
536  *   Pointer to private structure.
537  * @param keep
538  *   Bitmask for flags that must remain untouched.
539  * @param flags
540  *   Bitmask for flags to modify.
541  *
542  * @return
543  *   0 on success, -1 on failure and errno is set.
544  */
545 int
546 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
547 {
548 	unsigned long tmp;
549 
550 	if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
551 		return -1;
552 	tmp &= keep;
553 	tmp |= (flags & (~keep));
554 	return priv_set_sysfs_ulong(priv, "flags", tmp);
555 }
556 
557 /**
558  * Ethernet device configuration.
559  *
560  * Prepare the driver for a given number of TX and RX queues.
561  *
562  * @param dev
563  *   Pointer to Ethernet device structure.
564  *
565  * @return
566  *   0 on success, errno value on failure.
567  */
568 static int
569 dev_configure(struct rte_eth_dev *dev)
570 {
571 	struct priv *priv = dev->data->dev_private;
572 	unsigned int rxqs_n = dev->data->nb_rx_queues;
573 	unsigned int txqs_n = dev->data->nb_tx_queues;
574 	unsigned int i;
575 	unsigned int j;
576 	unsigned int reta_idx_n;
577 
578 	priv->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
579 	priv->rxqs = (void *)dev->data->rx_queues;
580 	priv->txqs = (void *)dev->data->tx_queues;
581 	if (txqs_n != priv->txqs_n) {
582 		INFO("%p: TX queues number update: %u -> %u",
583 		     (void *)dev, priv->txqs_n, txqs_n);
584 		priv->txqs_n = txqs_n;
585 	}
586 	if (rxqs_n > priv->ind_table_max_size) {
587 		ERROR("cannot handle this many RX queues (%u)", rxqs_n);
588 		return EINVAL;
589 	}
590 	if (rxqs_n == priv->rxqs_n)
591 		return 0;
592 	INFO("%p: RX queues number update: %u -> %u",
593 	     (void *)dev, priv->rxqs_n, rxqs_n);
594 	priv->rxqs_n = rxqs_n;
595 	/* If the requested number of RX queues is not a power of two, use the
596 	 * maximum indirection table size for better balancing.
597 	 * The result is always rounded to the next power of two. */
598 	reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
599 				     priv->ind_table_max_size :
600 				     rxqs_n));
601 	if (priv_rss_reta_index_resize(priv, reta_idx_n))
602 		return ENOMEM;
603 	/* When the number of RX queues is not a power of two, the remaining
604 	 * table entries are padded with reused WQs and hashes are not spread
605 	 * uniformly. */
606 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
607 		(*priv->reta_idx)[i] = j;
608 		if (++j == rxqs_n)
609 			j = 0;
610 	}
611 	return 0;
612 }
613 
614 /**
615  * DPDK callback for Ethernet device configuration.
616  *
617  * @param dev
618  *   Pointer to Ethernet device structure.
619  *
620  * @return
621  *   0 on success, negative errno value on failure.
622  */
623 int
624 mlx5_dev_configure(struct rte_eth_dev *dev)
625 {
626 	struct priv *priv = dev->data->dev_private;
627 	int ret;
628 
629 	if (mlx5_is_secondary())
630 		return -E_RTE_SECONDARY;
631 
632 	priv_lock(priv);
633 	ret = dev_configure(dev);
634 	assert(ret >= 0);
635 	priv_unlock(priv);
636 	return -ret;
637 }
638 
639 /**
640  * DPDK callback to get information about the device.
641  *
642  * @param dev
643  *   Pointer to Ethernet device structure.
644  * @param[out] info
645  *   Info structure output buffer.
646  */
647 void
648 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
649 {
650 	struct priv *priv = mlx5_get_priv(dev);
651 	unsigned int max;
652 	char ifname[IF_NAMESIZE];
653 
654 	info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
655 
656 	priv_lock(priv);
657 	/* FIXME: we should ask the device for these values. */
658 	info->min_rx_bufsize = 32;
659 	info->max_rx_pktlen = 65536;
660 	/*
661 	 * Since we need one CQ per QP, the limit is the minimum number
662 	 * between the two values.
663 	 */
664 	max = ((priv->device_attr.max_cq > priv->device_attr.max_qp) ?
665 	       priv->device_attr.max_qp : priv->device_attr.max_cq);
666 	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
667 	if (max >= 65535)
668 		max = 65535;
669 	info->max_rx_queues = max;
670 	info->max_tx_queues = max;
671 	info->max_mac_addrs = RTE_DIM(priv->mac);
672 	info->rx_offload_capa =
673 		(priv->hw_csum ?
674 		 (DEV_RX_OFFLOAD_IPV4_CKSUM |
675 		  DEV_RX_OFFLOAD_UDP_CKSUM |
676 		  DEV_RX_OFFLOAD_TCP_CKSUM) :
677 		 0) |
678 		(priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0);
679 	if (!priv->mps)
680 		info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
681 	if (priv->hw_csum)
682 		info->tx_offload_capa |=
683 			(DEV_TX_OFFLOAD_IPV4_CKSUM |
684 			 DEV_TX_OFFLOAD_UDP_CKSUM |
685 			 DEV_TX_OFFLOAD_TCP_CKSUM);
686 	if (priv->tso)
687 		info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
688 	if (priv->tunnel_en)
689 		info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
690 					  DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
691 					  DEV_TX_OFFLOAD_GRE_TNL_TSO);
692 	if (priv_get_ifname(priv, &ifname) == 0)
693 		info->if_index = if_nametoindex(ifname);
694 	info->reta_size = priv->reta_idx_n ?
695 		priv->reta_idx_n : priv->ind_table_max_size;
696 	info->hash_key_size = ((*priv->rss_conf) ?
697 			       (*priv->rss_conf)[0]->rss_key_len :
698 			       0);
699 	info->speed_capa = priv->link_speed_capa;
700 	priv_unlock(priv);
701 }
702 
703 const uint32_t *
704 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
705 {
706 	static const uint32_t ptypes[] = {
707 		/* refers to rxq_cq_to_pkt_type() */
708 		RTE_PTYPE_L2_ETHER,
709 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
710 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
711 		RTE_PTYPE_L4_NONFRAG,
712 		RTE_PTYPE_L4_FRAG,
713 		RTE_PTYPE_L4_TCP,
714 		RTE_PTYPE_L4_UDP,
715 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
716 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
717 		RTE_PTYPE_INNER_L4_NONFRAG,
718 		RTE_PTYPE_INNER_L4_FRAG,
719 		RTE_PTYPE_INNER_L4_TCP,
720 		RTE_PTYPE_INNER_L4_UDP,
721 		RTE_PTYPE_UNKNOWN
722 	};
723 
724 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
725 	    dev->rx_pkt_burst == mlx5_rx_burst_vec)
726 		return ptypes;
727 	return NULL;
728 }
729 
730 /**
731  * DPDK callback to retrieve physical link information.
732  *
733  * @param dev
734  *   Pointer to Ethernet device structure.
735  * @param wait_to_complete
736  *   Wait for request completion (ignored).
737  */
738 static int
739 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
740 {
741 	struct priv *priv = mlx5_get_priv(dev);
742 	struct ethtool_cmd edata = {
743 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
744 	};
745 	struct ifreq ifr;
746 	struct rte_eth_link dev_link;
747 	int link_speed = 0;
748 
749 	/* priv_lock() is not taken to allow concurrent calls. */
750 
751 	(void)wait_to_complete;
752 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
753 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
754 		return -1;
755 	}
756 	memset(&dev_link, 0, sizeof(dev_link));
757 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
758 				(ifr.ifr_flags & IFF_RUNNING));
759 	ifr.ifr_data = (void *)&edata;
760 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
761 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
762 		     strerror(errno));
763 		return -1;
764 	}
765 	link_speed = ethtool_cmd_speed(&edata);
766 	if (link_speed == -1)
767 		dev_link.link_speed = 0;
768 	else
769 		dev_link.link_speed = link_speed;
770 	priv->link_speed_capa = 0;
771 	if (edata.supported & SUPPORTED_Autoneg)
772 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
773 	if (edata.supported & (SUPPORTED_1000baseT_Full |
774 			       SUPPORTED_1000baseKX_Full))
775 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
776 	if (edata.supported & SUPPORTED_10000baseKR_Full)
777 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
778 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
779 			       SUPPORTED_40000baseCR4_Full |
780 			       SUPPORTED_40000baseSR4_Full |
781 			       SUPPORTED_40000baseLR4_Full))
782 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
783 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
784 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
785 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
786 			ETH_LINK_SPEED_FIXED);
787 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
788 		/* Link status changed. */
789 		dev->data->dev_link = dev_link;
790 		return 0;
791 	}
792 	/* Link status is still the same. */
793 	return -1;
794 }
795 
796 /**
797  * Retrieve physical link information (unlocked version using new ioctl).
798  *
799  * @param dev
800  *   Pointer to Ethernet device structure.
801  * @param wait_to_complete
802  *   Wait for request completion (ignored).
803  */
804 static int
805 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
806 {
807 	struct priv *priv = mlx5_get_priv(dev);
808 	alignas(struct ethtool_link_settings)
809 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
810 		     sizeof(uint32_t) *
811 		     ETHTOOL_LINK_MODE_MASK_MAX_KERNEL_NU32 * 3];
812 	struct ethtool_link_settings *ecmd = (void *)data;
813 	struct ifreq ifr;
814 	struct rte_eth_link dev_link;
815 	uint64_t sc;
816 
817 	(void)wait_to_complete;
818 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
819 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
820 		return -1;
821 	}
822 	memset(&dev_link, 0, sizeof(dev_link));
823 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
824 				(ifr.ifr_flags & IFF_RUNNING));
825 	memset(ecmd, 0, sizeof(*ecmd));
826 	ecmd->cmd = ETHTOOL_GLINKSETTINGS;
827 	ifr.ifr_data = (void *)ecmd;
828 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
829 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
830 		      strerror(errno));
831 		return -1;
832 	}
833 	ecmd->link_mode_masks_nwords = -ecmd->link_mode_masks_nwords;
834 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
835 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
836 		      strerror(errno));
837 		return -1;
838 	}
839 	dev_link.link_speed = ecmd->speed;
840 	sc = ecmd->link_mode_masks[0] |
841 		((uint64_t)ecmd->link_mode_masks[1] << 32);
842 	priv->link_speed_capa = 0;
843 	if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
844 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
845 	if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
846 		  ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
847 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
848 	if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
849 		  ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
850 		  ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
851 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
852 	if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
853 		  ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
854 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
855 	if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
856 		  ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
857 		  ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
858 		  ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
859 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
860 	if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
861 		  ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
862 		  ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
863 		  ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
864 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
865 	if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
866 		  ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
867 		  ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
868 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
869 	if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
870 		  ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
871 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
872 	if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
873 		  ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
874 		  ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
875 		  ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
876 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
877 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
878 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
879 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
880 				  ETH_LINK_SPEED_FIXED);
881 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
882 		/* Link status changed. */
883 		dev->data->dev_link = dev_link;
884 		return 0;
885 	}
886 	/* Link status is still the same. */
887 	return -1;
888 }
889 
890 /**
891  * DPDK callback to retrieve physical link information.
892  *
893  * @param dev
894  *   Pointer to Ethernet device structure.
895  * @param wait_to_complete
896  *   Wait for request completion (ignored).
897  */
898 int
899 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
900 {
901 	struct utsname utsname;
902 	int ver[3];
903 
904 	if (uname(&utsname) == -1 ||
905 	    sscanf(utsname.release, "%d.%d.%d",
906 		   &ver[0], &ver[1], &ver[2]) != 3 ||
907 	    KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
908 		return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
909 	return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
910 }
911 
912 /**
913  * DPDK callback to change the MTU.
914  *
915  * @param dev
916  *   Pointer to Ethernet device structure.
917  * @param in_mtu
918  *   New MTU.
919  *
920  * @return
921  *   0 on success, negative errno value on failure.
922  */
923 int
924 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
925 {
926 	struct priv *priv = dev->data->dev_private;
927 	uint16_t kern_mtu;
928 	int ret = 0;
929 
930 	if (mlx5_is_secondary())
931 		return -E_RTE_SECONDARY;
932 
933 	priv_lock(priv);
934 	ret = priv_get_mtu(priv, &kern_mtu);
935 	if (ret)
936 		goto out;
937 	/* Set kernel interface MTU first. */
938 	ret = priv_set_mtu(priv, mtu);
939 	if (ret)
940 		goto out;
941 	ret = priv_get_mtu(priv, &kern_mtu);
942 	if (ret)
943 		goto out;
944 	if (kern_mtu == mtu) {
945 		priv->mtu = mtu;
946 		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
947 	}
948 	priv_unlock(priv);
949 	return 0;
950 out:
951 	ret = errno;
952 	WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
953 	     strerror(ret));
954 	priv_unlock(priv);
955 	assert(ret >= 0);
956 	return -ret;
957 }
958 
959 /**
960  * DPDK callback to get flow control status.
961  *
962  * @param dev
963  *   Pointer to Ethernet device structure.
964  * @param[out] fc_conf
965  *   Flow control output buffer.
966  *
967  * @return
968  *   0 on success, negative errno value on failure.
969  */
970 int
971 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
972 {
973 	struct priv *priv = dev->data->dev_private;
974 	struct ifreq ifr;
975 	struct ethtool_pauseparam ethpause = {
976 		.cmd = ETHTOOL_GPAUSEPARAM
977 	};
978 	int ret;
979 
980 	if (mlx5_is_secondary())
981 		return -E_RTE_SECONDARY;
982 
983 	ifr.ifr_data = (void *)&ethpause;
984 	priv_lock(priv);
985 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
986 		ret = errno;
987 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
988 		     " failed: %s",
989 		     strerror(ret));
990 		goto out;
991 	}
992 
993 	fc_conf->autoneg = ethpause.autoneg;
994 	if (ethpause.rx_pause && ethpause.tx_pause)
995 		fc_conf->mode = RTE_FC_FULL;
996 	else if (ethpause.rx_pause)
997 		fc_conf->mode = RTE_FC_RX_PAUSE;
998 	else if (ethpause.tx_pause)
999 		fc_conf->mode = RTE_FC_TX_PAUSE;
1000 	else
1001 		fc_conf->mode = RTE_FC_NONE;
1002 	ret = 0;
1003 
1004 out:
1005 	priv_unlock(priv);
1006 	assert(ret >= 0);
1007 	return -ret;
1008 }
1009 
1010 /**
1011  * DPDK callback to modify flow control parameters.
1012  *
1013  * @param dev
1014  *   Pointer to Ethernet device structure.
1015  * @param[in] fc_conf
1016  *   Flow control parameters.
1017  *
1018  * @return
1019  *   0 on success, negative errno value on failure.
1020  */
1021 int
1022 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1023 {
1024 	struct priv *priv = dev->data->dev_private;
1025 	struct ifreq ifr;
1026 	struct ethtool_pauseparam ethpause = {
1027 		.cmd = ETHTOOL_SPAUSEPARAM
1028 	};
1029 	int ret;
1030 
1031 	if (mlx5_is_secondary())
1032 		return -E_RTE_SECONDARY;
1033 
1034 	ifr.ifr_data = (void *)&ethpause;
1035 	ethpause.autoneg = fc_conf->autoneg;
1036 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1037 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
1038 		ethpause.rx_pause = 1;
1039 	else
1040 		ethpause.rx_pause = 0;
1041 
1042 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1043 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
1044 		ethpause.tx_pause = 1;
1045 	else
1046 		ethpause.tx_pause = 0;
1047 
1048 	priv_lock(priv);
1049 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1050 		ret = errno;
1051 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1052 		     " failed: %s",
1053 		     strerror(ret));
1054 		goto out;
1055 	}
1056 	ret = 0;
1057 
1058 out:
1059 	priv_unlock(priv);
1060 	assert(ret >= 0);
1061 	return -ret;
1062 }
1063 
1064 /**
1065  * Get PCI information from struct ibv_device.
1066  *
1067  * @param device
1068  *   Pointer to Ethernet device structure.
1069  * @param[out] pci_addr
1070  *   PCI bus address output buffer.
1071  *
1072  * @return
1073  *   0 on success, -1 on failure and errno is set.
1074  */
1075 int
1076 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1077 			    struct rte_pci_addr *pci_addr)
1078 {
1079 	FILE *file;
1080 	char line[32];
1081 	MKSTR(path, "%s/device/uevent", device->ibdev_path);
1082 
1083 	file = fopen(path, "rb");
1084 	if (file == NULL)
1085 		return -1;
1086 	while (fgets(line, sizeof(line), file) == line) {
1087 		size_t len = strlen(line);
1088 		int ret;
1089 
1090 		/* Truncate long lines. */
1091 		if (len == (sizeof(line) - 1))
1092 			while (line[(len - 1)] != '\n') {
1093 				ret = fgetc(file);
1094 				if (ret == EOF)
1095 					break;
1096 				line[(len - 1)] = ret;
1097 			}
1098 		/* Extract information. */
1099 		if (sscanf(line,
1100 			   "PCI_SLOT_NAME="
1101 			   "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1102 			   &pci_addr->domain,
1103 			   &pci_addr->bus,
1104 			   &pci_addr->devid,
1105 			   &pci_addr->function) == 4) {
1106 			ret = 0;
1107 			break;
1108 		}
1109 	}
1110 	fclose(file);
1111 	return 0;
1112 }
1113 
1114 /**
1115  * Link status handler.
1116  *
1117  * @param priv
1118  *   Pointer to private structure.
1119  * @param dev
1120  *   Pointer to the rte_eth_dev structure.
1121  *
1122  * @return
1123  *   Nonzero if the callback process can be called immediately.
1124  */
1125 static int
1126 priv_dev_link_status_handler(struct priv *priv, struct rte_eth_dev *dev)
1127 {
1128 	struct ibv_async_event event;
1129 	struct rte_eth_link *link = &dev->data->dev_link;
1130 	int ret = 0;
1131 
1132 	/* Read all message and acknowledge them. */
1133 	for (;;) {
1134 		if (ibv_get_async_event(priv->ctx, &event))
1135 			break;
1136 
1137 		if (event.event_type != IBV_EVENT_PORT_ACTIVE &&
1138 		    event.event_type != IBV_EVENT_PORT_ERR)
1139 			DEBUG("event type %d on port %d not handled",
1140 			      event.event_type, event.element.port_num);
1141 		ibv_ack_async_event(&event);
1142 	}
1143 	mlx5_link_update(dev, 0);
1144 	if (((link->link_speed == 0) && link->link_status) ||
1145 	    ((link->link_speed != 0) && !link->link_status)) {
1146 		if (!priv->pending_alarm) {
1147 			/* Inconsistent status, check again later. */
1148 			priv->pending_alarm = 1;
1149 			rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1150 					  mlx5_dev_link_status_handler,
1151 					  dev);
1152 		}
1153 	} else {
1154 		ret = 1;
1155 	}
1156 	return ret;
1157 }
1158 
1159 /**
1160  * Handle delayed link status event.
1161  *
1162  * @param arg
1163  *   Registered argument.
1164  */
1165 void
1166 mlx5_dev_link_status_handler(void *arg)
1167 {
1168 	struct rte_eth_dev *dev = arg;
1169 	struct priv *priv = dev->data->dev_private;
1170 	int ret;
1171 
1172 	priv_lock(priv);
1173 	assert(priv->pending_alarm == 1);
1174 	priv->pending_alarm = 0;
1175 	ret = priv_dev_link_status_handler(priv, dev);
1176 	priv_unlock(priv);
1177 	if (ret)
1178 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1179 					      NULL);
1180 }
1181 
1182 /**
1183  * Handle interrupts from the NIC.
1184  *
1185  * @param[in] intr_handle
1186  *   Interrupt handler.
1187  * @param cb_arg
1188  *   Callback argument.
1189  */
1190 void
1191 mlx5_dev_interrupt_handler(void *cb_arg)
1192 {
1193 	struct rte_eth_dev *dev = cb_arg;
1194 	struct priv *priv = dev->data->dev_private;
1195 	int ret;
1196 
1197 	priv_lock(priv);
1198 	ret = priv_dev_link_status_handler(priv, dev);
1199 	priv_unlock(priv);
1200 	if (ret)
1201 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1202 					      NULL);
1203 }
1204 
1205 /**
1206  * Uninstall interrupt handler.
1207  *
1208  * @param priv
1209  *   Pointer to private structure.
1210  * @param dev
1211  *   Pointer to the rte_eth_dev structure.
1212  */
1213 void
1214 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1215 {
1216 	if (!dev->data->dev_conf.intr_conf.lsc)
1217 		return;
1218 	rte_intr_callback_unregister(&priv->intr_handle,
1219 				     mlx5_dev_interrupt_handler,
1220 				     dev);
1221 	if (priv->pending_alarm)
1222 		rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1223 	priv->pending_alarm = 0;
1224 	priv->intr_handle.fd = 0;
1225 	priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1226 }
1227 
1228 /**
1229  * Install interrupt handler.
1230  *
1231  * @param priv
1232  *   Pointer to private structure.
1233  * @param dev
1234  *   Pointer to the rte_eth_dev structure.
1235  */
1236 void
1237 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1238 {
1239 	int rc, flags;
1240 
1241 	if (!dev->data->dev_conf.intr_conf.lsc)
1242 		return;
1243 	assert(priv->ctx->async_fd > 0);
1244 	flags = fcntl(priv->ctx->async_fd, F_GETFL);
1245 	rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1246 	if (rc < 0) {
1247 		INFO("failed to change file descriptor async event queue");
1248 		dev->data->dev_conf.intr_conf.lsc = 0;
1249 	} else {
1250 		priv->intr_handle.fd = priv->ctx->async_fd;
1251 		priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1252 		rte_intr_callback_register(&priv->intr_handle,
1253 					   mlx5_dev_interrupt_handler,
1254 					   dev);
1255 	}
1256 }
1257 
1258 /**
1259  * Change the link state (UP / DOWN).
1260  *
1261  * @param priv
1262  *   Pointer to Ethernet device structure.
1263  * @param up
1264  *   Nonzero for link up, otherwise link down.
1265  *
1266  * @return
1267  *   0 on success, errno value on failure.
1268  */
1269 static int
1270 priv_set_link(struct priv *priv, int up)
1271 {
1272 	struct rte_eth_dev *dev = priv->dev;
1273 	int err;
1274 
1275 	if (up) {
1276 		err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1277 		if (err)
1278 			return err;
1279 		priv_select_tx_function(priv);
1280 		priv_select_rx_function(priv);
1281 	} else {
1282 		err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1283 		if (err)
1284 			return err;
1285 		dev->rx_pkt_burst = removed_rx_burst;
1286 		dev->tx_pkt_burst = removed_tx_burst;
1287 	}
1288 	return 0;
1289 }
1290 
1291 /**
1292  * DPDK callback to bring the link DOWN.
1293  *
1294  * @param dev
1295  *   Pointer to Ethernet device structure.
1296  *
1297  * @return
1298  *   0 on success, errno value on failure.
1299  */
1300 int
1301 mlx5_set_link_down(struct rte_eth_dev *dev)
1302 {
1303 	struct priv *priv = dev->data->dev_private;
1304 	int err;
1305 
1306 	priv_lock(priv);
1307 	err = priv_set_link(priv, 0);
1308 	priv_unlock(priv);
1309 	return err;
1310 }
1311 
1312 /**
1313  * DPDK callback to bring the link UP.
1314  *
1315  * @param dev
1316  *   Pointer to Ethernet device structure.
1317  *
1318  * @return
1319  *   0 on success, errno value on failure.
1320  */
1321 int
1322 mlx5_set_link_up(struct rte_eth_dev *dev)
1323 {
1324 	struct priv *priv = dev->data->dev_private;
1325 	int err;
1326 
1327 	priv_lock(priv);
1328 	err = priv_set_link(priv, 1);
1329 	priv_unlock(priv);
1330 	return err;
1331 }
1332 
1333 /**
1334  * Configure the TX function to use.
1335  *
1336  * @param priv
1337  *   Pointer to private structure.
1338  */
1339 void
1340 priv_select_tx_function(struct priv *priv)
1341 {
1342 	priv->dev->tx_pkt_burst = mlx5_tx_burst;
1343 	/* Select appropriate TX function. */
1344 	if (priv->mps == MLX5_MPW_ENHANCED) {
1345 		if (priv_check_vec_tx_support(priv) > 0) {
1346 			if (priv_check_raw_vec_tx_support(priv) > 0)
1347 				priv->dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1348 			else
1349 				priv->dev->tx_pkt_burst = mlx5_tx_burst_vec;
1350 			DEBUG("selected Enhanced MPW TX vectorized function");
1351 		} else {
1352 			priv->dev->tx_pkt_burst = mlx5_tx_burst_empw;
1353 			DEBUG("selected Enhanced MPW TX function");
1354 		}
1355 	} else if (priv->mps && priv->txq_inline) {
1356 		priv->dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1357 		DEBUG("selected MPW inline TX function");
1358 	} else if (priv->mps) {
1359 		priv->dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1360 		DEBUG("selected MPW TX function");
1361 	}
1362 }
1363 
1364 /**
1365  * Configure the RX function to use.
1366  *
1367  * @param priv
1368  *   Pointer to private structure.
1369  */
1370 void
1371 priv_select_rx_function(struct priv *priv)
1372 {
1373 	if (priv_check_vec_rx_support(priv) > 0) {
1374 		priv->dev->rx_pkt_burst = mlx5_rx_burst_vec;
1375 		DEBUG("selected RX vectorized function");
1376 	} else {
1377 		priv->dev->rx_pkt_burst = mlx5_rx_burst;
1378 	}
1379 }
1380