xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision f8b9a3bad46702704211d416073090c0bb7cfa6e)
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 #define _GNU_SOURCE
35 
36 #include <stddef.h>
37 #include <assert.h>
38 #include <unistd.h>
39 #include <stdint.h>
40 #include <stdio.h>
41 #include <string.h>
42 #include <stdlib.h>
43 #include <errno.h>
44 #include <dirent.h>
45 #include <net/if.h>
46 #include <sys/ioctl.h>
47 #include <sys/socket.h>
48 #include <sys/utsname.h>
49 #include <netinet/in.h>
50 #include <linux/ethtool.h>
51 #include <linux/sockios.h>
52 #include <linux/version.h>
53 #include <fcntl.h>
54 #include <stdalign.h>
55 #include <sys/un.h>
56 
57 #include <rte_atomic.h>
58 #include <rte_ethdev.h>
59 #include <rte_mbuf.h>
60 #include <rte_common.h>
61 #include <rte_interrupts.h>
62 #include <rte_alarm.h>
63 #include <rte_malloc.h>
64 
65 #include "mlx5.h"
66 #include "mlx5_rxtx.h"
67 #include "mlx5_utils.h"
68 
69 /* Add defines in case the running kernel is not the same as user headers. */
70 #ifndef ETHTOOL_GLINKSETTINGS
71 struct ethtool_link_settings {
72 	uint32_t cmd;
73 	uint32_t speed;
74 	uint8_t duplex;
75 	uint8_t port;
76 	uint8_t phy_address;
77 	uint8_t autoneg;
78 	uint8_t mdio_support;
79 	uint8_t eth_to_mdix;
80 	uint8_t eth_tp_mdix_ctrl;
81 	int8_t link_mode_masks_nwords;
82 	uint32_t reserved[8];
83 	uint32_t link_mode_masks[];
84 };
85 
86 #define ETHTOOL_GLINKSETTINGS 0x0000004c
87 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
88 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
89 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
90 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
91 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
92 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
93 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
94 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
95 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
96 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
97 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
98 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
99 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
100 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
101 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
102 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
103 #endif
104 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
105 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
106 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
107 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
108 #endif
109 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
110 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
111 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
112 #endif
113 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
114 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
115 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
116 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
117 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
118 #endif
119 
120 /**
121  * Return private structure associated with an Ethernet device.
122  *
123  * @param dev
124  *   Pointer to Ethernet device structure.
125  *
126  * @return
127  *   Pointer to private structure.
128  */
129 struct priv *
130 mlx5_get_priv(struct rte_eth_dev *dev)
131 {
132 	return dev->data->dev_private;
133 }
134 
135 /**
136  * Check if running as a secondary process.
137  *
138  * @return
139  *   Nonzero if running as a secondary process.
140  */
141 inline int
142 mlx5_is_secondary(void)
143 {
144 	return rte_eal_process_type() == RTE_PROC_SECONDARY;
145 }
146 
147 /**
148  * Get interface name from private structure.
149  *
150  * @param[in] priv
151  *   Pointer to private structure.
152  * @param[out] ifname
153  *   Interface name output buffer.
154  *
155  * @return
156  *   0 on success, -1 on failure and errno is set.
157  */
158 int
159 priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
160 {
161 	DIR *dir;
162 	struct dirent *dent;
163 	unsigned int dev_type = 0;
164 	unsigned int dev_port_prev = ~0u;
165 	char match[IF_NAMESIZE] = "";
166 
167 	{
168 		MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
169 
170 		dir = opendir(path);
171 		if (dir == NULL)
172 			return -1;
173 	}
174 	while ((dent = readdir(dir)) != NULL) {
175 		char *name = dent->d_name;
176 		FILE *file;
177 		unsigned int dev_port;
178 		int r;
179 
180 		if ((name[0] == '.') &&
181 		    ((name[1] == '\0') ||
182 		     ((name[1] == '.') && (name[2] == '\0'))))
183 			continue;
184 
185 		MKSTR(path, "%s/device/net/%s/%s",
186 		      priv->ctx->device->ibdev_path, name,
187 		      (dev_type ? "dev_id" : "dev_port"));
188 
189 		file = fopen(path, "rb");
190 		if (file == NULL) {
191 			if (errno != ENOENT)
192 				continue;
193 			/*
194 			 * Switch to dev_id when dev_port does not exist as
195 			 * is the case with Linux kernel versions < 3.15.
196 			 */
197 try_dev_id:
198 			match[0] = '\0';
199 			if (dev_type)
200 				break;
201 			dev_type = 1;
202 			dev_port_prev = ~0u;
203 			rewinddir(dir);
204 			continue;
205 		}
206 		r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
207 		fclose(file);
208 		if (r != 1)
209 			continue;
210 		/*
211 		 * Switch to dev_id when dev_port returns the same value for
212 		 * all ports. May happen when using a MOFED release older than
213 		 * 3.0 with a Linux kernel >= 3.15.
214 		 */
215 		if (dev_port == dev_port_prev)
216 			goto try_dev_id;
217 		dev_port_prev = dev_port;
218 		if (dev_port == (priv->port - 1u))
219 			snprintf(match, sizeof(match), "%s", name);
220 	}
221 	closedir(dir);
222 	if (match[0] == '\0')
223 		return -1;
224 	strncpy(*ifname, match, sizeof(*ifname));
225 	return 0;
226 }
227 
228 /**
229  * Check if the counter is located on ib counters file.
230  *
231  * @param[in] cntr
232  *   Counter name.
233  *
234  * @return
235  *   1 if counter is located on ib counters file , 0 otherwise.
236  */
237 int
238 priv_is_ib_cntr(const char *cntr)
239 {
240 	if (!strcmp(cntr, "out_of_buffer"))
241 		return 1;
242 	return 0;
243 }
244 
245 /**
246  * Read from sysfs entry.
247  *
248  * @param[in] priv
249  *   Pointer to private structure.
250  * @param[in] entry
251  *   Entry name relative to sysfs path.
252  * @param[out] buf
253  *   Data output buffer.
254  * @param size
255  *   Buffer size.
256  *
257  * @return
258  *   0 on success, -1 on failure and errno is set.
259  */
260 static int
261 priv_sysfs_read(const struct priv *priv, const char *entry,
262 		char *buf, size_t size)
263 {
264 	char ifname[IF_NAMESIZE];
265 	FILE *file;
266 	int ret;
267 	int err;
268 
269 	if (priv_get_ifname(priv, &ifname))
270 		return -1;
271 
272 	if (priv_is_ib_cntr(entry)) {
273 		MKSTR(path, "%s/ports/1/hw_counters/%s",
274 		      priv->ctx->device->ibdev_path, entry);
275 		file = fopen(path, "rb");
276 	} else {
277 		MKSTR(path, "%s/device/net/%s/%s",
278 		      priv->ctx->device->ibdev_path, ifname, entry);
279 		file = fopen(path, "rb");
280 	}
281 	if (file == NULL)
282 		return -1;
283 	ret = fread(buf, 1, size, file);
284 	err = errno;
285 	if (((size_t)ret < size) && (ferror(file)))
286 		ret = -1;
287 	else
288 		ret = size;
289 	fclose(file);
290 	errno = err;
291 	return ret;
292 }
293 
294 /**
295  * Write to sysfs entry.
296  *
297  * @param[in] priv
298  *   Pointer to private structure.
299  * @param[in] entry
300  *   Entry name relative to sysfs path.
301  * @param[in] buf
302  *   Data buffer.
303  * @param size
304  *   Buffer size.
305  *
306  * @return
307  *   0 on success, -1 on failure and errno is set.
308  */
309 static int
310 priv_sysfs_write(const struct priv *priv, const char *entry,
311 		 char *buf, size_t size)
312 {
313 	char ifname[IF_NAMESIZE];
314 	FILE *file;
315 	int ret;
316 	int err;
317 
318 	if (priv_get_ifname(priv, &ifname))
319 		return -1;
320 
321 	MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
322 	      ifname, entry);
323 
324 	file = fopen(path, "wb");
325 	if (file == NULL)
326 		return -1;
327 	ret = fwrite(buf, 1, size, file);
328 	err = errno;
329 	if (((size_t)ret < size) || (ferror(file)))
330 		ret = -1;
331 	else
332 		ret = size;
333 	fclose(file);
334 	errno = err;
335 	return ret;
336 }
337 
338 /**
339  * Get unsigned long sysfs property.
340  *
341  * @param priv
342  *   Pointer to private structure.
343  * @param[in] name
344  *   Entry name relative to sysfs path.
345  * @param[out] value
346  *   Value output buffer.
347  *
348  * @return
349  *   0 on success, -1 on failure and errno is set.
350  */
351 static int
352 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
353 {
354 	int ret;
355 	unsigned long value_ret;
356 	char value_str[32];
357 
358 	ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
359 	if (ret == -1) {
360 		DEBUG("cannot read %s value from sysfs: %s",
361 		      name, strerror(errno));
362 		return -1;
363 	}
364 	value_str[ret] = '\0';
365 	errno = 0;
366 	value_ret = strtoul(value_str, NULL, 0);
367 	if (errno) {
368 		DEBUG("invalid %s value `%s': %s", name, value_str,
369 		      strerror(errno));
370 		return -1;
371 	}
372 	*value = value_ret;
373 	return 0;
374 }
375 
376 /**
377  * Set unsigned long sysfs property.
378  *
379  * @param priv
380  *   Pointer to private structure.
381  * @param[in] name
382  *   Entry name relative to sysfs path.
383  * @param value
384  *   Value to set.
385  *
386  * @return
387  *   0 on success, -1 on failure and errno is set.
388  */
389 static int
390 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
391 {
392 	int ret;
393 	MKSTR(value_str, "%lu", value);
394 
395 	ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
396 	if (ret == -1) {
397 		DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
398 		      name, value_str, value, strerror(errno));
399 		return -1;
400 	}
401 	return 0;
402 }
403 
404 /**
405  * Perform ifreq ioctl() on associated Ethernet device.
406  *
407  * @param[in] priv
408  *   Pointer to private structure.
409  * @param req
410  *   Request number to pass to ioctl().
411  * @param[out] ifr
412  *   Interface request structure output buffer.
413  *
414  * @return
415  *   0 on success, -1 on failure and errno is set.
416  */
417 int
418 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
419 {
420 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
421 	int ret = -1;
422 
423 	if (sock == -1)
424 		return ret;
425 	if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
426 		ret = ioctl(sock, req, ifr);
427 	close(sock);
428 	return ret;
429 }
430 
431 /**
432  * Return the number of active VFs for the current device.
433  *
434  * @param[in] priv
435  *   Pointer to private structure.
436  * @param[out] num_vfs
437  *   Number of active VFs.
438  *
439  * @return
440  *   0 on success, -1 on failure and errno is set.
441  */
442 int
443 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
444 {
445 	/* The sysfs entry name depends on the operating system. */
446 	const char **name = (const char *[]){
447 		"device/sriov_numvfs",
448 		"device/mlx5_num_vfs",
449 		NULL,
450 	};
451 	int ret;
452 
453 	do {
454 		unsigned long ulong_num_vfs;
455 
456 		ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
457 		if (!ret)
458 			*num_vfs = ulong_num_vfs;
459 	} while (*(++name) && ret);
460 	return ret;
461 }
462 
463 /**
464  * Get device MTU.
465  *
466  * @param priv
467  *   Pointer to private structure.
468  * @param[out] mtu
469  *   MTU value output buffer.
470  *
471  * @return
472  *   0 on success, -1 on failure and errno is set.
473  */
474 int
475 priv_get_mtu(struct priv *priv, uint16_t *mtu)
476 {
477 	unsigned long ulong_mtu;
478 
479 	if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
480 		return -1;
481 	*mtu = ulong_mtu;
482 	return 0;
483 }
484 
485 /**
486  * Read device counter from sysfs.
487  *
488  * @param priv
489  *   Pointer to private structure.
490  * @param name
491  *   Counter name.
492  * @param[out] cntr
493  *   Counter output buffer.
494  *
495  * @return
496  *   0 on success, -1 on failure and errno is set.
497  */
498 int
499 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
500 {
501 	unsigned long ulong_ctr;
502 
503 	if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
504 		return -1;
505 	*cntr = ulong_ctr;
506 	return 0;
507 }
508 
509 /**
510  * Set device MTU.
511  *
512  * @param priv
513  *   Pointer to private structure.
514  * @param mtu
515  *   MTU value to set.
516  *
517  * @return
518  *   0 on success, -1 on failure and errno is set.
519  */
520 static int
521 priv_set_mtu(struct priv *priv, uint16_t mtu)
522 {
523 	uint16_t new_mtu;
524 
525 	if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
526 	    priv_get_mtu(priv, &new_mtu))
527 		return -1;
528 	if (new_mtu == mtu)
529 		return 0;
530 	errno = EINVAL;
531 	return -1;
532 }
533 
534 /**
535  * Set device flags.
536  *
537  * @param priv
538  *   Pointer to private structure.
539  * @param keep
540  *   Bitmask for flags that must remain untouched.
541  * @param flags
542  *   Bitmask for flags to modify.
543  *
544  * @return
545  *   0 on success, -1 on failure and errno is set.
546  */
547 int
548 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
549 {
550 	unsigned long tmp;
551 
552 	if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
553 		return -1;
554 	tmp &= keep;
555 	tmp |= (flags & (~keep));
556 	return priv_set_sysfs_ulong(priv, "flags", tmp);
557 }
558 
559 /**
560  * Ethernet device configuration.
561  *
562  * Prepare the driver for a given number of TX and RX queues.
563  *
564  * @param dev
565  *   Pointer to Ethernet device structure.
566  *
567  * @return
568  *   0 on success, errno value on failure.
569  */
570 static int
571 dev_configure(struct rte_eth_dev *dev)
572 {
573 	struct priv *priv = dev->data->dev_private;
574 	unsigned int rxqs_n = dev->data->nb_rx_queues;
575 	unsigned int txqs_n = dev->data->nb_tx_queues;
576 	unsigned int i;
577 	unsigned int j;
578 	unsigned int reta_idx_n;
579 
580 	priv->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
581 	priv->rxqs = (void *)dev->data->rx_queues;
582 	priv->txqs = (void *)dev->data->tx_queues;
583 	if (txqs_n != priv->txqs_n) {
584 		INFO("%p: TX queues number update: %u -> %u",
585 		     (void *)dev, priv->txqs_n, txqs_n);
586 		priv->txqs_n = txqs_n;
587 	}
588 	if (rxqs_n > priv->ind_table_max_size) {
589 		ERROR("cannot handle this many RX queues (%u)", rxqs_n);
590 		return EINVAL;
591 	}
592 	if (rxqs_n == priv->rxqs_n)
593 		return 0;
594 	INFO("%p: RX queues number update: %u -> %u",
595 	     (void *)dev, priv->rxqs_n, rxqs_n);
596 	priv->rxqs_n = rxqs_n;
597 	/* If the requested number of RX queues is not a power of two, use the
598 	 * maximum indirection table size for better balancing.
599 	 * The result is always rounded to the next power of two. */
600 	reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
601 				     priv->ind_table_max_size :
602 				     rxqs_n));
603 	if (priv_rss_reta_index_resize(priv, reta_idx_n))
604 		return ENOMEM;
605 	/* When the number of RX queues is not a power of two, the remaining
606 	 * table entries are padded with reused WQs and hashes are not spread
607 	 * uniformly. */
608 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
609 		(*priv->reta_idx)[i] = j;
610 		if (++j == rxqs_n)
611 			j = 0;
612 	}
613 	return 0;
614 }
615 
616 /**
617  * DPDK callback for Ethernet device configuration.
618  *
619  * @param dev
620  *   Pointer to Ethernet device structure.
621  *
622  * @return
623  *   0 on success, negative errno value on failure.
624  */
625 int
626 mlx5_dev_configure(struct rte_eth_dev *dev)
627 {
628 	struct priv *priv = dev->data->dev_private;
629 	int ret;
630 
631 	if (mlx5_is_secondary())
632 		return -E_RTE_SECONDARY;
633 
634 	priv_lock(priv);
635 	ret = dev_configure(dev);
636 	assert(ret >= 0);
637 	priv_unlock(priv);
638 	return -ret;
639 }
640 
641 /**
642  * DPDK callback to get information about the device.
643  *
644  * @param dev
645  *   Pointer to Ethernet device structure.
646  * @param[out] info
647  *   Info structure output buffer.
648  */
649 void
650 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
651 {
652 	struct priv *priv = mlx5_get_priv(dev);
653 	unsigned int max;
654 	char ifname[IF_NAMESIZE];
655 
656 	info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
657 
658 	priv_lock(priv);
659 	/* FIXME: we should ask the device for these values. */
660 	info->min_rx_bufsize = 32;
661 	info->max_rx_pktlen = 65536;
662 	/*
663 	 * Since we need one CQ per QP, the limit is the minimum number
664 	 * between the two values.
665 	 */
666 	max = RTE_MIN(priv->device_attr.orig_attr.max_cq,
667 		      priv->device_attr.orig_attr.max_qp);
668 	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
669 	if (max >= 65535)
670 		max = 65535;
671 	info->max_rx_queues = max;
672 	info->max_tx_queues = max;
673 	info->max_mac_addrs = RTE_DIM(priv->mac);
674 	info->rx_offload_capa =
675 		(priv->hw_csum ?
676 		 (DEV_RX_OFFLOAD_IPV4_CKSUM |
677 		  DEV_RX_OFFLOAD_UDP_CKSUM |
678 		  DEV_RX_OFFLOAD_TCP_CKSUM) :
679 		 0) |
680 		(priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0);
681 	if (!priv->mps)
682 		info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
683 	if (priv->hw_csum)
684 		info->tx_offload_capa |=
685 			(DEV_TX_OFFLOAD_IPV4_CKSUM |
686 			 DEV_TX_OFFLOAD_UDP_CKSUM |
687 			 DEV_TX_OFFLOAD_TCP_CKSUM);
688 	if (priv->tso)
689 		info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
690 	if (priv->tunnel_en)
691 		info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
692 					  DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
693 					  DEV_TX_OFFLOAD_GRE_TNL_TSO);
694 	if (priv_get_ifname(priv, &ifname) == 0)
695 		info->if_index = if_nametoindex(ifname);
696 	info->reta_size = priv->reta_idx_n ?
697 		priv->reta_idx_n : priv->ind_table_max_size;
698 	info->hash_key_size = ((*priv->rss_conf) ?
699 			       (*priv->rss_conf)[0]->rss_key_len :
700 			       0);
701 	info->speed_capa = priv->link_speed_capa;
702 	priv_unlock(priv);
703 }
704 
705 const uint32_t *
706 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
707 {
708 	static const uint32_t ptypes[] = {
709 		/* refers to rxq_cq_to_pkt_type() */
710 		RTE_PTYPE_L2_ETHER,
711 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
712 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
713 		RTE_PTYPE_L4_NONFRAG,
714 		RTE_PTYPE_L4_FRAG,
715 		RTE_PTYPE_L4_TCP,
716 		RTE_PTYPE_L4_UDP,
717 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
718 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
719 		RTE_PTYPE_INNER_L4_NONFRAG,
720 		RTE_PTYPE_INNER_L4_FRAG,
721 		RTE_PTYPE_INNER_L4_TCP,
722 		RTE_PTYPE_INNER_L4_UDP,
723 		RTE_PTYPE_UNKNOWN
724 	};
725 
726 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
727 	    dev->rx_pkt_burst == mlx5_rx_burst_vec)
728 		return ptypes;
729 	return NULL;
730 }
731 
732 /**
733  * DPDK callback to retrieve physical link information.
734  *
735  * @param dev
736  *   Pointer to Ethernet device structure.
737  * @param wait_to_complete
738  *   Wait for request completion (ignored).
739  */
740 static int
741 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
742 {
743 	struct priv *priv = mlx5_get_priv(dev);
744 	struct ethtool_cmd edata = {
745 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
746 	};
747 	struct ifreq ifr;
748 	struct rte_eth_link dev_link;
749 	int link_speed = 0;
750 
751 	/* priv_lock() is not taken to allow concurrent calls. */
752 
753 	(void)wait_to_complete;
754 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
755 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
756 		return -1;
757 	}
758 	memset(&dev_link, 0, sizeof(dev_link));
759 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
760 				(ifr.ifr_flags & IFF_RUNNING));
761 	ifr.ifr_data = (void *)&edata;
762 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
763 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
764 		     strerror(errno));
765 		return -1;
766 	}
767 	link_speed = ethtool_cmd_speed(&edata);
768 	if (link_speed == -1)
769 		dev_link.link_speed = 0;
770 	else
771 		dev_link.link_speed = link_speed;
772 	priv->link_speed_capa = 0;
773 	if (edata.supported & SUPPORTED_Autoneg)
774 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
775 	if (edata.supported & (SUPPORTED_1000baseT_Full |
776 			       SUPPORTED_1000baseKX_Full))
777 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
778 	if (edata.supported & SUPPORTED_10000baseKR_Full)
779 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
780 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
781 			       SUPPORTED_40000baseCR4_Full |
782 			       SUPPORTED_40000baseSR4_Full |
783 			       SUPPORTED_40000baseLR4_Full))
784 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
785 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
786 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
787 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
788 			ETH_LINK_SPEED_FIXED);
789 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
790 		/* Link status changed. */
791 		dev->data->dev_link = dev_link;
792 		return 0;
793 	}
794 	/* Link status is still the same. */
795 	return -1;
796 }
797 
798 /**
799  * Retrieve physical link information (unlocked version using new ioctl).
800  *
801  * @param dev
802  *   Pointer to Ethernet device structure.
803  * @param wait_to_complete
804  *   Wait for request completion (ignored).
805  */
806 static int
807 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
808 {
809 	struct priv *priv = mlx5_get_priv(dev);
810 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
811 	struct ifreq ifr;
812 	struct rte_eth_link dev_link;
813 	uint64_t sc;
814 
815 	(void)wait_to_complete;
816 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
817 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
818 		return -1;
819 	}
820 	memset(&dev_link, 0, sizeof(dev_link));
821 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
822 				(ifr.ifr_flags & IFF_RUNNING));
823 	ifr.ifr_data = (void *)&gcmd;
824 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
825 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
826 		      strerror(errno));
827 		return -1;
828 	}
829 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
830 
831 	alignas(struct ethtool_link_settings)
832 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
833 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
834 	struct ethtool_link_settings *ecmd = (void *)data;
835 
836 	*ecmd = gcmd;
837 	ifr.ifr_data = (void *)ecmd;
838 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
839 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
840 		      strerror(errno));
841 		return -1;
842 	}
843 	dev_link.link_speed = ecmd->speed;
844 	sc = ecmd->link_mode_masks[0] |
845 		((uint64_t)ecmd->link_mode_masks[1] << 32);
846 	priv->link_speed_capa = 0;
847 	if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
848 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
849 	if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
850 		  ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
851 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
852 	if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
853 		  ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
854 		  ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
855 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
856 	if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
857 		  ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
858 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
859 	if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
860 		  ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
861 		  ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
862 		  ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
863 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
864 	if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
865 		  ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
866 		  ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
867 		  ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
868 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
869 	if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
870 		  ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
871 		  ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
872 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
873 	if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
874 		  ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
875 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
876 	if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
877 		  ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
878 		  ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
879 		  ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
880 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
881 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
882 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
883 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
884 				  ETH_LINK_SPEED_FIXED);
885 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
886 		/* Link status changed. */
887 		dev->data->dev_link = dev_link;
888 		return 0;
889 	}
890 	/* Link status is still the same. */
891 	return -1;
892 }
893 
894 /**
895  * DPDK callback to retrieve physical link information.
896  *
897  * @param dev
898  *   Pointer to Ethernet device structure.
899  * @param wait_to_complete
900  *   Wait for request completion (ignored).
901  */
902 int
903 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
904 {
905 	struct utsname utsname;
906 	int ver[3];
907 
908 	if (uname(&utsname) == -1 ||
909 	    sscanf(utsname.release, "%d.%d.%d",
910 		   &ver[0], &ver[1], &ver[2]) != 3 ||
911 	    KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
912 		return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
913 	return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
914 }
915 
916 /**
917  * DPDK callback to change the MTU.
918  *
919  * @param dev
920  *   Pointer to Ethernet device structure.
921  * @param in_mtu
922  *   New MTU.
923  *
924  * @return
925  *   0 on success, negative errno value on failure.
926  */
927 int
928 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
929 {
930 	struct priv *priv = dev->data->dev_private;
931 	uint16_t kern_mtu;
932 	int ret = 0;
933 
934 	if (mlx5_is_secondary())
935 		return -E_RTE_SECONDARY;
936 
937 	priv_lock(priv);
938 	ret = priv_get_mtu(priv, &kern_mtu);
939 	if (ret)
940 		goto out;
941 	/* Set kernel interface MTU first. */
942 	ret = priv_set_mtu(priv, mtu);
943 	if (ret)
944 		goto out;
945 	ret = priv_get_mtu(priv, &kern_mtu);
946 	if (ret)
947 		goto out;
948 	if (kern_mtu == mtu) {
949 		priv->mtu = mtu;
950 		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
951 	}
952 	priv_unlock(priv);
953 	return 0;
954 out:
955 	ret = errno;
956 	WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
957 	     strerror(ret));
958 	priv_unlock(priv);
959 	assert(ret >= 0);
960 	return -ret;
961 }
962 
963 /**
964  * DPDK callback to get flow control status.
965  *
966  * @param dev
967  *   Pointer to Ethernet device structure.
968  * @param[out] fc_conf
969  *   Flow control output buffer.
970  *
971  * @return
972  *   0 on success, negative errno value on failure.
973  */
974 int
975 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
976 {
977 	struct priv *priv = dev->data->dev_private;
978 	struct ifreq ifr;
979 	struct ethtool_pauseparam ethpause = {
980 		.cmd = ETHTOOL_GPAUSEPARAM
981 	};
982 	int ret;
983 
984 	if (mlx5_is_secondary())
985 		return -E_RTE_SECONDARY;
986 
987 	ifr.ifr_data = (void *)&ethpause;
988 	priv_lock(priv);
989 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
990 		ret = errno;
991 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
992 		     " failed: %s",
993 		     strerror(ret));
994 		goto out;
995 	}
996 
997 	fc_conf->autoneg = ethpause.autoneg;
998 	if (ethpause.rx_pause && ethpause.tx_pause)
999 		fc_conf->mode = RTE_FC_FULL;
1000 	else if (ethpause.rx_pause)
1001 		fc_conf->mode = RTE_FC_RX_PAUSE;
1002 	else if (ethpause.tx_pause)
1003 		fc_conf->mode = RTE_FC_TX_PAUSE;
1004 	else
1005 		fc_conf->mode = RTE_FC_NONE;
1006 	ret = 0;
1007 
1008 out:
1009 	priv_unlock(priv);
1010 	assert(ret >= 0);
1011 	return -ret;
1012 }
1013 
1014 /**
1015  * DPDK callback to modify flow control parameters.
1016  *
1017  * @param dev
1018  *   Pointer to Ethernet device structure.
1019  * @param[in] fc_conf
1020  *   Flow control parameters.
1021  *
1022  * @return
1023  *   0 on success, negative errno value on failure.
1024  */
1025 int
1026 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1027 {
1028 	struct priv *priv = dev->data->dev_private;
1029 	struct ifreq ifr;
1030 	struct ethtool_pauseparam ethpause = {
1031 		.cmd = ETHTOOL_SPAUSEPARAM
1032 	};
1033 	int ret;
1034 
1035 	if (mlx5_is_secondary())
1036 		return -E_RTE_SECONDARY;
1037 
1038 	ifr.ifr_data = (void *)&ethpause;
1039 	ethpause.autoneg = fc_conf->autoneg;
1040 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1041 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
1042 		ethpause.rx_pause = 1;
1043 	else
1044 		ethpause.rx_pause = 0;
1045 
1046 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1047 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
1048 		ethpause.tx_pause = 1;
1049 	else
1050 		ethpause.tx_pause = 0;
1051 
1052 	priv_lock(priv);
1053 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1054 		ret = errno;
1055 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1056 		     " failed: %s",
1057 		     strerror(ret));
1058 		goto out;
1059 	}
1060 	ret = 0;
1061 
1062 out:
1063 	priv_unlock(priv);
1064 	assert(ret >= 0);
1065 	return -ret;
1066 }
1067 
1068 /**
1069  * Get PCI information from struct ibv_device.
1070  *
1071  * @param device
1072  *   Pointer to Ethernet device structure.
1073  * @param[out] pci_addr
1074  *   PCI bus address output buffer.
1075  *
1076  * @return
1077  *   0 on success, -1 on failure and errno is set.
1078  */
1079 int
1080 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1081 			    struct rte_pci_addr *pci_addr)
1082 {
1083 	FILE *file;
1084 	char line[32];
1085 	MKSTR(path, "%s/device/uevent", device->ibdev_path);
1086 
1087 	file = fopen(path, "rb");
1088 	if (file == NULL)
1089 		return -1;
1090 	while (fgets(line, sizeof(line), file) == line) {
1091 		size_t len = strlen(line);
1092 		int ret;
1093 
1094 		/* Truncate long lines. */
1095 		if (len == (sizeof(line) - 1))
1096 			while (line[(len - 1)] != '\n') {
1097 				ret = fgetc(file);
1098 				if (ret == EOF)
1099 					break;
1100 				line[(len - 1)] = ret;
1101 			}
1102 		/* Extract information. */
1103 		if (sscanf(line,
1104 			   "PCI_SLOT_NAME="
1105 			   "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1106 			   &pci_addr->domain,
1107 			   &pci_addr->bus,
1108 			   &pci_addr->devid,
1109 			   &pci_addr->function) == 4) {
1110 			ret = 0;
1111 			break;
1112 		}
1113 	}
1114 	fclose(file);
1115 	return 0;
1116 }
1117 
1118 /**
1119  * Update the link status.
1120  *
1121  * @param priv
1122  *   Pointer to private structure.
1123  *
1124  * @return
1125  *   Zero if the callback process can be called immediately.
1126  */
1127 static int
1128 priv_link_status_update(struct priv *priv)
1129 {
1130 	struct rte_eth_link *link = &priv->dev->data->dev_link;
1131 
1132 	mlx5_link_update(priv->dev, 0);
1133 	if (((link->link_speed == 0) && link->link_status) ||
1134 		((link->link_speed != 0) && !link->link_status)) {
1135 		/*
1136 		 * Inconsistent status. Event likely occurred before the
1137 		 * kernel netdevice exposes the new status.
1138 		 */
1139 		if (!priv->pending_alarm) {
1140 			priv->pending_alarm = 1;
1141 			rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1142 					  mlx5_dev_link_status_handler,
1143 					  priv->dev);
1144 		}
1145 		return 1;
1146 	} else if (unlikely(priv->pending_alarm)) {
1147 		/* Link interrupt occurred while alarm is already scheduled. */
1148 		priv->pending_alarm = 0;
1149 		rte_eal_alarm_cancel(mlx5_dev_link_status_handler, priv->dev);
1150 	}
1151 	return 0;
1152 }
1153 
1154 /**
1155  * Device status handler.
1156  *
1157  * @param priv
1158  *   Pointer to private structure.
1159  * @param events
1160  *   Pointer to event flags holder.
1161  *
1162  * @return
1163  *   Events bitmap of callback process which can be called immediately.
1164  */
1165 static uint32_t
1166 priv_dev_status_handler(struct priv *priv)
1167 {
1168 	struct ibv_async_event event;
1169 	uint32_t ret = 0;
1170 
1171 	/* Read all message and acknowledge them. */
1172 	for (;;) {
1173 		if (ibv_get_async_event(priv->ctx, &event))
1174 			break;
1175 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1176 			event.event_type == IBV_EVENT_PORT_ERR) &&
1177 			(priv->dev->data->dev_conf.intr_conf.lsc == 1))
1178 			ret |= (1 << RTE_ETH_EVENT_INTR_LSC);
1179 		else if (event.event_type == IBV_EVENT_DEVICE_FATAL &&
1180 			priv->dev->data->dev_conf.intr_conf.rmv == 1)
1181 			ret |= (1 << RTE_ETH_EVENT_INTR_RMV);
1182 		else
1183 			DEBUG("event type %d on port %d not handled",
1184 			      event.event_type, event.element.port_num);
1185 		ibv_ack_async_event(&event);
1186 	}
1187 	if (ret & (1 << RTE_ETH_EVENT_INTR_LSC))
1188 		if (priv_link_status_update(priv))
1189 			ret &= ~(1 << RTE_ETH_EVENT_INTR_LSC);
1190 	return ret;
1191 }
1192 
1193 /**
1194  * Handle delayed link status event.
1195  *
1196  * @param arg
1197  *   Registered argument.
1198  */
1199 void
1200 mlx5_dev_link_status_handler(void *arg)
1201 {
1202 	struct rte_eth_dev *dev = arg;
1203 	struct priv *priv = dev->data->dev_private;
1204 	int ret;
1205 
1206 	priv_lock(priv);
1207 	assert(priv->pending_alarm == 1);
1208 	priv->pending_alarm = 0;
1209 	ret = priv_link_status_update(priv);
1210 	priv_unlock(priv);
1211 	if (!ret)
1212 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1213 					      NULL);
1214 }
1215 
1216 /**
1217  * Handle interrupts from the NIC.
1218  *
1219  * @param[in] intr_handle
1220  *   Interrupt handler.
1221  * @param cb_arg
1222  *   Callback argument.
1223  */
1224 void
1225 mlx5_dev_interrupt_handler(void *cb_arg)
1226 {
1227 	struct rte_eth_dev *dev = cb_arg;
1228 	struct priv *priv = dev->data->dev_private;
1229 	uint32_t events;
1230 
1231 	priv_lock(priv);
1232 	events = priv_dev_status_handler(priv);
1233 	priv_unlock(priv);
1234 	if (events & (1 << RTE_ETH_EVENT_INTR_LSC))
1235 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1236 					      NULL);
1237 	if (events & (1 << RTE_ETH_EVENT_INTR_RMV))
1238 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL,
1239 					      NULL);
1240 }
1241 
1242 /**
1243  * Handle interrupts from the socket.
1244  *
1245  * @param cb_arg
1246  *   Callback argument.
1247  */
1248 static void
1249 mlx5_dev_handler_socket(void *cb_arg)
1250 {
1251 	struct rte_eth_dev *dev = cb_arg;
1252 	struct priv *priv = dev->data->dev_private;
1253 
1254 	priv_lock(priv);
1255 	priv_socket_handle(priv);
1256 	priv_unlock(priv);
1257 }
1258 
1259 /**
1260  * Uninstall interrupt handler.
1261  *
1262  * @param priv
1263  *   Pointer to private structure.
1264  * @param dev
1265  *   Pointer to the rte_eth_dev structure.
1266  */
1267 void
1268 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1269 {
1270 	if (dev->data->dev_conf.intr_conf.lsc ||
1271 	    dev->data->dev_conf.intr_conf.rmv)
1272 		rte_intr_callback_unregister(&priv->intr_handle,
1273 					     mlx5_dev_interrupt_handler, dev);
1274 	if (priv->primary_socket)
1275 		rte_intr_callback_unregister(&priv->intr_handle_socket,
1276 					     mlx5_dev_handler_socket, dev);
1277 	if (priv->pending_alarm)
1278 		rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1279 	priv->pending_alarm = 0;
1280 	priv->intr_handle.fd = 0;
1281 	priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1282 	priv->intr_handle_socket.fd = 0;
1283 	priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN;
1284 }
1285 
1286 /**
1287  * Install interrupt handler.
1288  *
1289  * @param priv
1290  *   Pointer to private structure.
1291  * @param dev
1292  *   Pointer to the rte_eth_dev structure.
1293  */
1294 void
1295 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1296 {
1297 	int rc, flags;
1298 
1299 	assert(!mlx5_is_secondary());
1300 	assert(priv->ctx->async_fd > 0);
1301 	flags = fcntl(priv->ctx->async_fd, F_GETFL);
1302 	rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1303 	if (rc < 0) {
1304 		INFO("failed to change file descriptor async event queue");
1305 		dev->data->dev_conf.intr_conf.lsc = 0;
1306 		dev->data->dev_conf.intr_conf.rmv = 0;
1307 	}
1308 	if (dev->data->dev_conf.intr_conf.lsc ||
1309 	    dev->data->dev_conf.intr_conf.rmv) {
1310 		priv->intr_handle.fd = priv->ctx->async_fd;
1311 		priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1312 		rte_intr_callback_register(&priv->intr_handle,
1313 					   mlx5_dev_interrupt_handler, dev);
1314 	}
1315 
1316 	rc = priv_socket_init(priv);
1317 	if (!rc && priv->primary_socket) {
1318 		priv->intr_handle_socket.fd = priv->primary_socket;
1319 		priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT;
1320 		rte_intr_callback_register(&priv->intr_handle_socket,
1321 					   mlx5_dev_handler_socket, dev);
1322 	}
1323 }
1324 
1325 /**
1326  * Change the link state (UP / DOWN).
1327  *
1328  * @param priv
1329  *   Pointer to private data structure.
1330  * @param dev
1331  *   Pointer to rte_eth_dev structure.
1332  * @param up
1333  *   Nonzero for link up, otherwise link down.
1334  *
1335  * @return
1336  *   0 on success, errno value on failure.
1337  */
1338 static int
1339 priv_dev_set_link(struct priv *priv, struct rte_eth_dev *dev, int up)
1340 {
1341 	int err;
1342 
1343 	if (up) {
1344 		err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1345 		if (err)
1346 			return err;
1347 		priv_dev_select_tx_function(priv, dev);
1348 		priv_dev_select_rx_function(priv, dev);
1349 	} else {
1350 		err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1351 		if (err)
1352 			return err;
1353 		dev->rx_pkt_burst = removed_rx_burst;
1354 		dev->tx_pkt_burst = removed_tx_burst;
1355 	}
1356 	return 0;
1357 }
1358 
1359 /**
1360  * DPDK callback to bring the link DOWN.
1361  *
1362  * @param dev
1363  *   Pointer to Ethernet device structure.
1364  *
1365  * @return
1366  *   0 on success, errno value on failure.
1367  */
1368 int
1369 mlx5_set_link_down(struct rte_eth_dev *dev)
1370 {
1371 	struct priv *priv = dev->data->dev_private;
1372 	int err;
1373 
1374 	priv_lock(priv);
1375 	err = priv_dev_set_link(priv, dev, 0);
1376 	priv_unlock(priv);
1377 	return err;
1378 }
1379 
1380 /**
1381  * DPDK callback to bring the link UP.
1382  *
1383  * @param dev
1384  *   Pointer to Ethernet device structure.
1385  *
1386  * @return
1387  *   0 on success, errno value on failure.
1388  */
1389 int
1390 mlx5_set_link_up(struct rte_eth_dev *dev)
1391 {
1392 	struct priv *priv = dev->data->dev_private;
1393 	int err;
1394 
1395 	priv_lock(priv);
1396 	err = priv_dev_set_link(priv, dev, 1);
1397 	priv_unlock(priv);
1398 	return err;
1399 }
1400 
1401 /**
1402  * Configure the TX function to use.
1403  *
1404  * @param priv
1405  *   Pointer to private data structure.
1406  * @param dev
1407  *   Pointer to rte_eth_dev structure.
1408  */
1409 void
1410 priv_dev_select_tx_function(struct priv *priv, struct rte_eth_dev *dev)
1411 {
1412 	assert(priv != NULL);
1413 	assert(dev != NULL);
1414 	dev->tx_pkt_burst = mlx5_tx_burst;
1415 	/* Select appropriate TX function. */
1416 	if (priv->mps == MLX5_MPW_ENHANCED) {
1417 		if (priv_check_vec_tx_support(priv) > 0) {
1418 			if (priv_check_raw_vec_tx_support(priv) > 0)
1419 				dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1420 			else
1421 				dev->tx_pkt_burst = mlx5_tx_burst_vec;
1422 			DEBUG("selected Enhanced MPW TX vectorized function");
1423 		} else {
1424 			dev->tx_pkt_burst = mlx5_tx_burst_empw;
1425 			DEBUG("selected Enhanced MPW TX function");
1426 		}
1427 	} else if (priv->mps && priv->txq_inline) {
1428 		dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1429 		DEBUG("selected MPW inline TX function");
1430 	} else if (priv->mps) {
1431 		dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1432 		DEBUG("selected MPW TX function");
1433 	}
1434 }
1435 
1436 /**
1437  * Configure the RX function to use.
1438  *
1439  * @param priv
1440  *   Pointer to private data structure.
1441  * @param dev
1442  *   Pointer to rte_eth_dev structure.
1443  */
1444 void
1445 priv_dev_select_rx_function(struct priv *priv, struct rte_eth_dev *dev)
1446 {
1447 	assert(priv != NULL);
1448 	assert(dev != NULL);
1449 	if (priv_check_vec_rx_support(priv) > 0) {
1450 		dev->rx_pkt_burst = mlx5_rx_burst_vec;
1451 		DEBUG("selected RX vectorized function");
1452 	} else {
1453 		dev->rx_pkt_burst = mlx5_rx_burst;
1454 	}
1455 }
1456