xref: /dpdk/drivers/net/mlx5/mlx5_ethdev.c (revision 5b590fbe09b6163a55f279bc5d4b85dce39f1d49)
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->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->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->ibdev_path, entry);
275 		file = fopen(path, "rb");
276 	} else {
277 		MKSTR(path, "%s/device/net/%s/%s",
278 		      priv->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->ibdev_path, ifname, entry);
322 
323 	file = fopen(path, "wb");
324 	if (file == NULL)
325 		return -1;
326 	ret = fwrite(buf, 1, size, file);
327 	err = errno;
328 	if (((size_t)ret < size) || (ferror(file)))
329 		ret = -1;
330 	else
331 		ret = size;
332 	fclose(file);
333 	errno = err;
334 	return ret;
335 }
336 
337 /**
338  * Get unsigned long sysfs property.
339  *
340  * @param priv
341  *   Pointer to private structure.
342  * @param[in] name
343  *   Entry name relative to sysfs path.
344  * @param[out] value
345  *   Value output buffer.
346  *
347  * @return
348  *   0 on success, -1 on failure and errno is set.
349  */
350 static int
351 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
352 {
353 	int ret;
354 	unsigned long value_ret;
355 	char value_str[32];
356 
357 	ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
358 	if (ret == -1) {
359 		DEBUG("cannot read %s value from sysfs: %s",
360 		      name, strerror(errno));
361 		return -1;
362 	}
363 	value_str[ret] = '\0';
364 	errno = 0;
365 	value_ret = strtoul(value_str, NULL, 0);
366 	if (errno) {
367 		DEBUG("invalid %s value `%s': %s", name, value_str,
368 		      strerror(errno));
369 		return -1;
370 	}
371 	*value = value_ret;
372 	return 0;
373 }
374 
375 /**
376  * Set unsigned long sysfs property.
377  *
378  * @param priv
379  *   Pointer to private structure.
380  * @param[in] name
381  *   Entry name relative to sysfs path.
382  * @param value
383  *   Value to set.
384  *
385  * @return
386  *   0 on success, -1 on failure and errno is set.
387  */
388 static int
389 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
390 {
391 	int ret;
392 	MKSTR(value_str, "%lu", value);
393 
394 	ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
395 	if (ret == -1) {
396 		DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
397 		      name, value_str, value, strerror(errno));
398 		return -1;
399 	}
400 	return 0;
401 }
402 
403 /**
404  * Perform ifreq ioctl() on associated Ethernet device.
405  *
406  * @param[in] priv
407  *   Pointer to private structure.
408  * @param req
409  *   Request number to pass to ioctl().
410  * @param[out] ifr
411  *   Interface request structure output buffer.
412  *
413  * @return
414  *   0 on success, -1 on failure and errno is set.
415  */
416 int
417 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
418 {
419 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
420 	int ret = -1;
421 
422 	if (sock == -1)
423 		return ret;
424 	if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
425 		ret = ioctl(sock, req, ifr);
426 	close(sock);
427 	return ret;
428 }
429 
430 /**
431  * Return the number of active VFs for the current device.
432  *
433  * @param[in] priv
434  *   Pointer to private structure.
435  * @param[out] num_vfs
436  *   Number of active VFs.
437  *
438  * @return
439  *   0 on success, -1 on failure and errno is set.
440  */
441 int
442 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
443 {
444 	/* The sysfs entry name depends on the operating system. */
445 	const char **name = (const char *[]){
446 		"device/sriov_numvfs",
447 		"device/mlx5_num_vfs",
448 		NULL,
449 	};
450 	int ret;
451 
452 	do {
453 		unsigned long ulong_num_vfs;
454 
455 		ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
456 		if (!ret)
457 			*num_vfs = ulong_num_vfs;
458 	} while (*(++name) && ret);
459 	return ret;
460 }
461 
462 /**
463  * Get device MTU.
464  *
465  * @param priv
466  *   Pointer to private structure.
467  * @param[out] mtu
468  *   MTU value output buffer.
469  *
470  * @return
471  *   0 on success, -1 on failure and errno is set.
472  */
473 int
474 priv_get_mtu(struct priv *priv, uint16_t *mtu)
475 {
476 	unsigned long ulong_mtu;
477 
478 	if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
479 		return -1;
480 	*mtu = ulong_mtu;
481 	return 0;
482 }
483 
484 /**
485  * Read device counter from sysfs.
486  *
487  * @param priv
488  *   Pointer to private structure.
489  * @param name
490  *   Counter name.
491  * @param[out] cntr
492  *   Counter output buffer.
493  *
494  * @return
495  *   0 on success, -1 on failure and errno is set.
496  */
497 int
498 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
499 {
500 	unsigned long ulong_ctr;
501 
502 	if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
503 		return -1;
504 	*cntr = ulong_ctr;
505 	return 0;
506 }
507 
508 /**
509  * Set device MTU.
510  *
511  * @param priv
512  *   Pointer to private structure.
513  * @param mtu
514  *   MTU value to set.
515  *
516  * @return
517  *   0 on success, -1 on failure and errno is set.
518  */
519 static int
520 priv_set_mtu(struct priv *priv, uint16_t mtu)
521 {
522 	uint16_t new_mtu;
523 
524 	if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
525 	    priv_get_mtu(priv, &new_mtu))
526 		return -1;
527 	if (new_mtu == mtu)
528 		return 0;
529 	errno = EINVAL;
530 	return -1;
531 }
532 
533 /**
534  * Set device flags.
535  *
536  * @param priv
537  *   Pointer to private structure.
538  * @param keep
539  *   Bitmask for flags that must remain untouched.
540  * @param flags
541  *   Bitmask for flags to modify.
542  *
543  * @return
544  *   0 on success, -1 on failure and errno is set.
545  */
546 int
547 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
548 {
549 	unsigned long tmp;
550 
551 	if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
552 		return -1;
553 	tmp &= keep;
554 	tmp |= (flags & (~keep));
555 	return priv_set_sysfs_ulong(priv, "flags", tmp);
556 }
557 
558 /**
559  * Ethernet device configuration.
560  *
561  * Prepare the driver for a given number of TX and RX queues.
562  *
563  * @param dev
564  *   Pointer to Ethernet device structure.
565  *
566  * @return
567  *   0 on success, errno value on failure.
568  */
569 static int
570 dev_configure(struct rte_eth_dev *dev)
571 {
572 	struct priv *priv = dev->data->dev_private;
573 	unsigned int rxqs_n = dev->data->nb_rx_queues;
574 	unsigned int txqs_n = dev->data->nb_tx_queues;
575 	unsigned int i;
576 	unsigned int j;
577 	unsigned int reta_idx_n;
578 	const uint8_t use_app_rss_key =
579 		!!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
580 
581 	if (use_app_rss_key &&
582 	    (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
583 	     rss_hash_default_key_len)) {
584 		/* MLX5 RSS only support 40bytes key. */
585 		return EINVAL;
586 	}
587 	priv->rss_conf.rss_key =
588 		rte_realloc(priv->rss_conf.rss_key,
589 			    rss_hash_default_key_len, 0);
590 	if (!priv->rss_conf.rss_key) {
591 		ERROR("cannot allocate RSS hash key memory (%u)", rxqs_n);
592 		return ENOMEM;
593 	}
594 	memcpy(priv->rss_conf.rss_key,
595 	       use_app_rss_key ?
596 	       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
597 	       rss_hash_default_key,
598 	       rss_hash_default_key_len);
599 	priv->rss_conf.rss_key_len = rss_hash_default_key_len;
600 	priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
601 	priv->rxqs = (void *)dev->data->rx_queues;
602 	priv->txqs = (void *)dev->data->tx_queues;
603 	if (txqs_n != priv->txqs_n) {
604 		INFO("%p: TX queues number update: %u -> %u",
605 		     (void *)dev, priv->txqs_n, txqs_n);
606 		priv->txqs_n = txqs_n;
607 	}
608 	if (rxqs_n > priv->ind_table_max_size) {
609 		ERROR("cannot handle this many RX queues (%u)", rxqs_n);
610 		return EINVAL;
611 	}
612 	if (rxqs_n == priv->rxqs_n)
613 		return 0;
614 	INFO("%p: RX queues number update: %u -> %u",
615 	     (void *)dev, priv->rxqs_n, rxqs_n);
616 	priv->rxqs_n = rxqs_n;
617 	/* If the requested number of RX queues is not a power of two, use the
618 	 * maximum indirection table size for better balancing.
619 	 * The result is always rounded to the next power of two. */
620 	reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
621 				     priv->ind_table_max_size :
622 				     rxqs_n));
623 	if (priv_rss_reta_index_resize(priv, reta_idx_n))
624 		return ENOMEM;
625 	/* When the number of RX queues is not a power of two, the remaining
626 	 * table entries are padded with reused WQs and hashes are not spread
627 	 * uniformly. */
628 	for (i = 0, j = 0; (i != reta_idx_n); ++i) {
629 		(*priv->reta_idx)[i] = j;
630 		if (++j == rxqs_n)
631 			j = 0;
632 	}
633 	return 0;
634 }
635 
636 /**
637  * DPDK callback for Ethernet device configuration.
638  *
639  * @param dev
640  *   Pointer to Ethernet device structure.
641  *
642  * @return
643  *   0 on success, negative errno value on failure.
644  */
645 int
646 mlx5_dev_configure(struct rte_eth_dev *dev)
647 {
648 	struct priv *priv = dev->data->dev_private;
649 	int ret;
650 
651 	if (mlx5_is_secondary())
652 		return -E_RTE_SECONDARY;
653 
654 	priv_lock(priv);
655 	ret = dev_configure(dev);
656 	assert(ret >= 0);
657 	priv_unlock(priv);
658 	return -ret;
659 }
660 
661 /**
662  * DPDK callback to get information about the device.
663  *
664  * @param dev
665  *   Pointer to Ethernet device structure.
666  * @param[out] info
667  *   Info structure output buffer.
668  */
669 void
670 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
671 {
672 	struct priv *priv = mlx5_get_priv(dev);
673 	unsigned int max;
674 	char ifname[IF_NAMESIZE];
675 
676 	info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
677 
678 	priv_lock(priv);
679 	/* FIXME: we should ask the device for these values. */
680 	info->min_rx_bufsize = 32;
681 	info->max_rx_pktlen = 65536;
682 	/*
683 	 * Since we need one CQ per QP, the limit is the minimum number
684 	 * between the two values.
685 	 */
686 	max = RTE_MIN(priv->device_attr.orig_attr.max_cq,
687 		      priv->device_attr.orig_attr.max_qp);
688 	/* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
689 	if (max >= 65535)
690 		max = 65535;
691 	info->max_rx_queues = max;
692 	info->max_tx_queues = max;
693 	info->max_mac_addrs = RTE_DIM(priv->mac);
694 	info->rx_offload_capa =
695 		(priv->hw_csum ?
696 		 (DEV_RX_OFFLOAD_IPV4_CKSUM |
697 		  DEV_RX_OFFLOAD_UDP_CKSUM |
698 		  DEV_RX_OFFLOAD_TCP_CKSUM) :
699 		 0) |
700 		(priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0);
701 	if (!priv->mps)
702 		info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
703 	if (priv->hw_csum)
704 		info->tx_offload_capa |=
705 			(DEV_TX_OFFLOAD_IPV4_CKSUM |
706 			 DEV_TX_OFFLOAD_UDP_CKSUM |
707 			 DEV_TX_OFFLOAD_TCP_CKSUM);
708 	if (priv->tso)
709 		info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
710 	if (priv->tunnel_en)
711 		info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
712 					  DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
713 					  DEV_TX_OFFLOAD_GRE_TNL_TSO);
714 	if (priv_get_ifname(priv, &ifname) == 0)
715 		info->if_index = if_nametoindex(ifname);
716 	info->reta_size = priv->reta_idx_n ?
717 		priv->reta_idx_n : priv->ind_table_max_size;
718 	info->hash_key_size = priv->rss_conf.rss_key_len;
719 	info->speed_capa = priv->link_speed_capa;
720 	priv_unlock(priv);
721 }
722 
723 const uint32_t *
724 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
725 {
726 	static const uint32_t ptypes[] = {
727 		/* refers to rxq_cq_to_pkt_type() */
728 		RTE_PTYPE_L2_ETHER,
729 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
730 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
731 		RTE_PTYPE_L4_NONFRAG,
732 		RTE_PTYPE_L4_FRAG,
733 		RTE_PTYPE_L4_TCP,
734 		RTE_PTYPE_L4_UDP,
735 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
736 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
737 		RTE_PTYPE_INNER_L4_NONFRAG,
738 		RTE_PTYPE_INNER_L4_FRAG,
739 		RTE_PTYPE_INNER_L4_TCP,
740 		RTE_PTYPE_INNER_L4_UDP,
741 		RTE_PTYPE_UNKNOWN
742 	};
743 
744 	if (dev->rx_pkt_burst == mlx5_rx_burst ||
745 	    dev->rx_pkt_burst == mlx5_rx_burst_vec)
746 		return ptypes;
747 	return NULL;
748 }
749 
750 /**
751  * DPDK callback to retrieve physical link information.
752  *
753  * @param dev
754  *   Pointer to Ethernet device structure.
755  * @param wait_to_complete
756  *   Wait for request completion (ignored).
757  */
758 static int
759 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
760 {
761 	struct priv *priv = mlx5_get_priv(dev);
762 	struct ethtool_cmd edata = {
763 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
764 	};
765 	struct ifreq ifr;
766 	struct rte_eth_link dev_link;
767 	int link_speed = 0;
768 
769 	/* priv_lock() is not taken to allow concurrent calls. */
770 
771 	(void)wait_to_complete;
772 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
773 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
774 		return -1;
775 	}
776 	memset(&dev_link, 0, sizeof(dev_link));
777 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
778 				(ifr.ifr_flags & IFF_RUNNING));
779 	ifr.ifr_data = (void *)&edata;
780 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
781 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
782 		     strerror(errno));
783 		return -1;
784 	}
785 	link_speed = ethtool_cmd_speed(&edata);
786 	if (link_speed == -1)
787 		dev_link.link_speed = 0;
788 	else
789 		dev_link.link_speed = link_speed;
790 	priv->link_speed_capa = 0;
791 	if (edata.supported & SUPPORTED_Autoneg)
792 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
793 	if (edata.supported & (SUPPORTED_1000baseT_Full |
794 			       SUPPORTED_1000baseKX_Full))
795 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
796 	if (edata.supported & SUPPORTED_10000baseKR_Full)
797 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
798 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
799 			       SUPPORTED_40000baseCR4_Full |
800 			       SUPPORTED_40000baseSR4_Full |
801 			       SUPPORTED_40000baseLR4_Full))
802 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
803 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
804 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
805 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
806 			ETH_LINK_SPEED_FIXED);
807 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
808 		/* Link status changed. */
809 		dev->data->dev_link = dev_link;
810 		return 0;
811 	}
812 	/* Link status is still the same. */
813 	return -1;
814 }
815 
816 /**
817  * Retrieve physical link information (unlocked version using new ioctl).
818  *
819  * @param dev
820  *   Pointer to Ethernet device structure.
821  * @param wait_to_complete
822  *   Wait for request completion (ignored).
823  */
824 static int
825 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
826 {
827 	struct priv *priv = mlx5_get_priv(dev);
828 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
829 	struct ifreq ifr;
830 	struct rte_eth_link dev_link;
831 	uint64_t sc;
832 
833 	(void)wait_to_complete;
834 	if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
835 		WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
836 		return -1;
837 	}
838 	memset(&dev_link, 0, sizeof(dev_link));
839 	dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
840 				(ifr.ifr_flags & IFF_RUNNING));
841 	ifr.ifr_data = (void *)&gcmd;
842 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
843 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
844 		      strerror(errno));
845 		return -1;
846 	}
847 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
848 
849 	alignas(struct ethtool_link_settings)
850 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
851 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
852 	struct ethtool_link_settings *ecmd = (void *)data;
853 
854 	*ecmd = gcmd;
855 	ifr.ifr_data = (void *)ecmd;
856 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
857 		DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
858 		      strerror(errno));
859 		return -1;
860 	}
861 	dev_link.link_speed = ecmd->speed;
862 	sc = ecmd->link_mode_masks[0] |
863 		((uint64_t)ecmd->link_mode_masks[1] << 32);
864 	priv->link_speed_capa = 0;
865 	if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
866 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
867 	if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
868 		  ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
869 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
870 	if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
871 		  ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
872 		  ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
873 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
874 	if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
875 		  ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
876 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
877 	if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
878 		  ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
879 		  ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
880 		  ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
881 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
882 	if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
883 		  ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
884 		  ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
885 		  ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
886 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
887 	if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
888 		  ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
889 		  ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
890 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
891 	if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
892 		  ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
893 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
894 	if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
895 		  ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
896 		  ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
897 		  ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
898 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
899 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
900 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
901 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
902 				  ETH_LINK_SPEED_FIXED);
903 	if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
904 		/* Link status changed. */
905 		dev->data->dev_link = dev_link;
906 		return 0;
907 	}
908 	/* Link status is still the same. */
909 	return -1;
910 }
911 
912 /**
913  * DPDK callback to retrieve physical link information.
914  *
915  * @param dev
916  *   Pointer to Ethernet device structure.
917  * @param wait_to_complete
918  *   Wait for request completion (ignored).
919  */
920 int
921 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
922 {
923 	struct utsname utsname;
924 	int ver[3];
925 
926 	if (uname(&utsname) == -1 ||
927 	    sscanf(utsname.release, "%d.%d.%d",
928 		   &ver[0], &ver[1], &ver[2]) != 3 ||
929 	    KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
930 		return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
931 	return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
932 }
933 
934 /**
935  * DPDK callback to change the MTU.
936  *
937  * @param dev
938  *   Pointer to Ethernet device structure.
939  * @param in_mtu
940  *   New MTU.
941  *
942  * @return
943  *   0 on success, negative errno value on failure.
944  */
945 int
946 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
947 {
948 	struct priv *priv = dev->data->dev_private;
949 	uint16_t kern_mtu;
950 	int ret = 0;
951 
952 	if (mlx5_is_secondary())
953 		return -E_RTE_SECONDARY;
954 
955 	priv_lock(priv);
956 	ret = priv_get_mtu(priv, &kern_mtu);
957 	if (ret)
958 		goto out;
959 	/* Set kernel interface MTU first. */
960 	ret = priv_set_mtu(priv, mtu);
961 	if (ret)
962 		goto out;
963 	ret = priv_get_mtu(priv, &kern_mtu);
964 	if (ret)
965 		goto out;
966 	if (kern_mtu == mtu) {
967 		priv->mtu = mtu;
968 		DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
969 	}
970 	priv_unlock(priv);
971 	return 0;
972 out:
973 	ret = errno;
974 	WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
975 	     strerror(ret));
976 	priv_unlock(priv);
977 	assert(ret >= 0);
978 	return -ret;
979 }
980 
981 /**
982  * DPDK callback to get flow control status.
983  *
984  * @param dev
985  *   Pointer to Ethernet device structure.
986  * @param[out] fc_conf
987  *   Flow control output buffer.
988  *
989  * @return
990  *   0 on success, negative errno value on failure.
991  */
992 int
993 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
994 {
995 	struct priv *priv = dev->data->dev_private;
996 	struct ifreq ifr;
997 	struct ethtool_pauseparam ethpause = {
998 		.cmd = ETHTOOL_GPAUSEPARAM
999 	};
1000 	int ret;
1001 
1002 	if (mlx5_is_secondary())
1003 		return -E_RTE_SECONDARY;
1004 
1005 	ifr.ifr_data = (void *)&ethpause;
1006 	priv_lock(priv);
1007 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1008 		ret = errno;
1009 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
1010 		     " failed: %s",
1011 		     strerror(ret));
1012 		goto out;
1013 	}
1014 
1015 	fc_conf->autoneg = ethpause.autoneg;
1016 	if (ethpause.rx_pause && ethpause.tx_pause)
1017 		fc_conf->mode = RTE_FC_FULL;
1018 	else if (ethpause.rx_pause)
1019 		fc_conf->mode = RTE_FC_RX_PAUSE;
1020 	else if (ethpause.tx_pause)
1021 		fc_conf->mode = RTE_FC_TX_PAUSE;
1022 	else
1023 		fc_conf->mode = RTE_FC_NONE;
1024 	ret = 0;
1025 
1026 out:
1027 	priv_unlock(priv);
1028 	assert(ret >= 0);
1029 	return -ret;
1030 }
1031 
1032 /**
1033  * DPDK callback to modify flow control parameters.
1034  *
1035  * @param dev
1036  *   Pointer to Ethernet device structure.
1037  * @param[in] fc_conf
1038  *   Flow control parameters.
1039  *
1040  * @return
1041  *   0 on success, negative errno value on failure.
1042  */
1043 int
1044 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1045 {
1046 	struct priv *priv = dev->data->dev_private;
1047 	struct ifreq ifr;
1048 	struct ethtool_pauseparam ethpause = {
1049 		.cmd = ETHTOOL_SPAUSEPARAM
1050 	};
1051 	int ret;
1052 
1053 	if (mlx5_is_secondary())
1054 		return -E_RTE_SECONDARY;
1055 
1056 	ifr.ifr_data = (void *)&ethpause;
1057 	ethpause.autoneg = fc_conf->autoneg;
1058 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1059 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
1060 		ethpause.rx_pause = 1;
1061 	else
1062 		ethpause.rx_pause = 0;
1063 
1064 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1065 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
1066 		ethpause.tx_pause = 1;
1067 	else
1068 		ethpause.tx_pause = 0;
1069 
1070 	priv_lock(priv);
1071 	if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1072 		ret = errno;
1073 		WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1074 		     " failed: %s",
1075 		     strerror(ret));
1076 		goto out;
1077 	}
1078 	ret = 0;
1079 
1080 out:
1081 	priv_unlock(priv);
1082 	assert(ret >= 0);
1083 	return -ret;
1084 }
1085 
1086 /**
1087  * Get PCI information from struct ibv_device.
1088  *
1089  * @param device
1090  *   Pointer to Ethernet device structure.
1091  * @param[out] pci_addr
1092  *   PCI bus address output buffer.
1093  *
1094  * @return
1095  *   0 on success, -1 on failure and errno is set.
1096  */
1097 int
1098 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1099 			    struct rte_pci_addr *pci_addr)
1100 {
1101 	FILE *file;
1102 	char line[32];
1103 	MKSTR(path, "%s/device/uevent", device->ibdev_path);
1104 
1105 	file = fopen(path, "rb");
1106 	if (file == NULL)
1107 		return -1;
1108 	while (fgets(line, sizeof(line), file) == line) {
1109 		size_t len = strlen(line);
1110 		int ret;
1111 
1112 		/* Truncate long lines. */
1113 		if (len == (sizeof(line) - 1))
1114 			while (line[(len - 1)] != '\n') {
1115 				ret = fgetc(file);
1116 				if (ret == EOF)
1117 					break;
1118 				line[(len - 1)] = ret;
1119 			}
1120 		/* Extract information. */
1121 		if (sscanf(line,
1122 			   "PCI_SLOT_NAME="
1123 			   "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1124 			   &pci_addr->domain,
1125 			   &pci_addr->bus,
1126 			   &pci_addr->devid,
1127 			   &pci_addr->function) == 4) {
1128 			ret = 0;
1129 			break;
1130 		}
1131 	}
1132 	fclose(file);
1133 	return 0;
1134 }
1135 
1136 /**
1137  * Update the link status.
1138  *
1139  * @param priv
1140  *   Pointer to private structure.
1141  *
1142  * @return
1143  *   Zero if the callback process can be called immediately.
1144  */
1145 static int
1146 priv_link_status_update(struct priv *priv)
1147 {
1148 	struct rte_eth_link *link = &priv->dev->data->dev_link;
1149 
1150 	mlx5_link_update(priv->dev, 0);
1151 	if (((link->link_speed == 0) && link->link_status) ||
1152 		((link->link_speed != 0) && !link->link_status)) {
1153 		/*
1154 		 * Inconsistent status. Event likely occurred before the
1155 		 * kernel netdevice exposes the new status.
1156 		 */
1157 		if (!priv->pending_alarm) {
1158 			priv->pending_alarm = 1;
1159 			rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1160 					  mlx5_dev_link_status_handler,
1161 					  priv->dev);
1162 		}
1163 		return 1;
1164 	} else if (unlikely(priv->pending_alarm)) {
1165 		/* Link interrupt occurred while alarm is already scheduled. */
1166 		priv->pending_alarm = 0;
1167 		rte_eal_alarm_cancel(mlx5_dev_link_status_handler, priv->dev);
1168 	}
1169 	return 0;
1170 }
1171 
1172 /**
1173  * Device status handler.
1174  *
1175  * @param priv
1176  *   Pointer to private structure.
1177  * @param events
1178  *   Pointer to event flags holder.
1179  *
1180  * @return
1181  *   Events bitmap of callback process which can be called immediately.
1182  */
1183 static uint32_t
1184 priv_dev_status_handler(struct priv *priv)
1185 {
1186 	struct ibv_async_event event;
1187 	uint32_t ret = 0;
1188 
1189 	/* Read all message and acknowledge them. */
1190 	for (;;) {
1191 		if (ibv_get_async_event(priv->ctx, &event))
1192 			break;
1193 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1194 			event.event_type == IBV_EVENT_PORT_ERR) &&
1195 			(priv->dev->data->dev_conf.intr_conf.lsc == 1))
1196 			ret |= (1 << RTE_ETH_EVENT_INTR_LSC);
1197 		else if (event.event_type == IBV_EVENT_DEVICE_FATAL &&
1198 			priv->dev->data->dev_conf.intr_conf.rmv == 1)
1199 			ret |= (1 << RTE_ETH_EVENT_INTR_RMV);
1200 		else
1201 			DEBUG("event type %d on port %d not handled",
1202 			      event.event_type, event.element.port_num);
1203 		ibv_ack_async_event(&event);
1204 	}
1205 	if (ret & (1 << RTE_ETH_EVENT_INTR_LSC))
1206 		if (priv_link_status_update(priv))
1207 			ret &= ~(1 << RTE_ETH_EVENT_INTR_LSC);
1208 	return ret;
1209 }
1210 
1211 /**
1212  * Handle delayed link status event.
1213  *
1214  * @param arg
1215  *   Registered argument.
1216  */
1217 void
1218 mlx5_dev_link_status_handler(void *arg)
1219 {
1220 	struct rte_eth_dev *dev = arg;
1221 	struct priv *priv = dev->data->dev_private;
1222 	int ret;
1223 
1224 	priv_lock(priv);
1225 	assert(priv->pending_alarm == 1);
1226 	priv->pending_alarm = 0;
1227 	ret = priv_link_status_update(priv);
1228 	priv_unlock(priv);
1229 	if (!ret)
1230 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1231 					      NULL);
1232 }
1233 
1234 /**
1235  * Handle interrupts from the NIC.
1236  *
1237  * @param[in] intr_handle
1238  *   Interrupt handler.
1239  * @param cb_arg
1240  *   Callback argument.
1241  */
1242 void
1243 mlx5_dev_interrupt_handler(void *cb_arg)
1244 {
1245 	struct rte_eth_dev *dev = cb_arg;
1246 	struct priv *priv = dev->data->dev_private;
1247 	uint32_t events;
1248 
1249 	priv_lock(priv);
1250 	events = priv_dev_status_handler(priv);
1251 	priv_unlock(priv);
1252 	if (events & (1 << RTE_ETH_EVENT_INTR_LSC))
1253 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1254 					      NULL);
1255 	if (events & (1 << RTE_ETH_EVENT_INTR_RMV))
1256 		_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL,
1257 					      NULL);
1258 }
1259 
1260 /**
1261  * Handle interrupts from the socket.
1262  *
1263  * @param cb_arg
1264  *   Callback argument.
1265  */
1266 static void
1267 mlx5_dev_handler_socket(void *cb_arg)
1268 {
1269 	struct rte_eth_dev *dev = cb_arg;
1270 	struct priv *priv = dev->data->dev_private;
1271 
1272 	priv_lock(priv);
1273 	priv_socket_handle(priv);
1274 	priv_unlock(priv);
1275 }
1276 
1277 /**
1278  * Uninstall interrupt handler.
1279  *
1280  * @param priv
1281  *   Pointer to private structure.
1282  * @param dev
1283  *   Pointer to the rte_eth_dev structure.
1284  */
1285 void
1286 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1287 {
1288 	if (dev->data->dev_conf.intr_conf.lsc ||
1289 	    dev->data->dev_conf.intr_conf.rmv)
1290 		rte_intr_callback_unregister(&priv->intr_handle,
1291 					     mlx5_dev_interrupt_handler, dev);
1292 	if (priv->primary_socket)
1293 		rte_intr_callback_unregister(&priv->intr_handle_socket,
1294 					     mlx5_dev_handler_socket, dev);
1295 	if (priv->pending_alarm)
1296 		rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1297 	priv->pending_alarm = 0;
1298 	priv->intr_handle.fd = 0;
1299 	priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1300 	priv->intr_handle_socket.fd = 0;
1301 	priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN;
1302 }
1303 
1304 /**
1305  * Install interrupt handler.
1306  *
1307  * @param priv
1308  *   Pointer to private structure.
1309  * @param dev
1310  *   Pointer to the rte_eth_dev structure.
1311  */
1312 void
1313 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1314 {
1315 	int rc, flags;
1316 
1317 	assert(!mlx5_is_secondary());
1318 	assert(priv->ctx->async_fd > 0);
1319 	flags = fcntl(priv->ctx->async_fd, F_GETFL);
1320 	rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1321 	if (rc < 0) {
1322 		INFO("failed to change file descriptor async event queue");
1323 		dev->data->dev_conf.intr_conf.lsc = 0;
1324 		dev->data->dev_conf.intr_conf.rmv = 0;
1325 	}
1326 	if (dev->data->dev_conf.intr_conf.lsc ||
1327 	    dev->data->dev_conf.intr_conf.rmv) {
1328 		priv->intr_handle.fd = priv->ctx->async_fd;
1329 		priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1330 		rte_intr_callback_register(&priv->intr_handle,
1331 					   mlx5_dev_interrupt_handler, dev);
1332 	}
1333 
1334 	rc = priv_socket_init(priv);
1335 	if (!rc && priv->primary_socket) {
1336 		priv->intr_handle_socket.fd = priv->primary_socket;
1337 		priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT;
1338 		rte_intr_callback_register(&priv->intr_handle_socket,
1339 					   mlx5_dev_handler_socket, dev);
1340 	}
1341 }
1342 
1343 /**
1344  * Change the link state (UP / DOWN).
1345  *
1346  * @param priv
1347  *   Pointer to private data structure.
1348  * @param dev
1349  *   Pointer to rte_eth_dev structure.
1350  * @param up
1351  *   Nonzero for link up, otherwise link down.
1352  *
1353  * @return
1354  *   0 on success, errno value on failure.
1355  */
1356 static int
1357 priv_dev_set_link(struct priv *priv, struct rte_eth_dev *dev, int up)
1358 {
1359 	int err;
1360 
1361 	if (up) {
1362 		err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1363 		if (err)
1364 			return err;
1365 		priv_dev_select_tx_function(priv, dev);
1366 		priv_dev_select_rx_function(priv, dev);
1367 	} else {
1368 		err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1369 		if (err)
1370 			return err;
1371 		dev->rx_pkt_burst = removed_rx_burst;
1372 		dev->tx_pkt_burst = removed_tx_burst;
1373 	}
1374 	return 0;
1375 }
1376 
1377 /**
1378  * DPDK callback to bring the link DOWN.
1379  *
1380  * @param dev
1381  *   Pointer to Ethernet device structure.
1382  *
1383  * @return
1384  *   0 on success, errno value on failure.
1385  */
1386 int
1387 mlx5_set_link_down(struct rte_eth_dev *dev)
1388 {
1389 	struct priv *priv = dev->data->dev_private;
1390 	int err;
1391 
1392 	priv_lock(priv);
1393 	err = priv_dev_set_link(priv, dev, 0);
1394 	priv_unlock(priv);
1395 	return err;
1396 }
1397 
1398 /**
1399  * DPDK callback to bring the link UP.
1400  *
1401  * @param dev
1402  *   Pointer to Ethernet device structure.
1403  *
1404  * @return
1405  *   0 on success, errno value on failure.
1406  */
1407 int
1408 mlx5_set_link_up(struct rte_eth_dev *dev)
1409 {
1410 	struct priv *priv = dev->data->dev_private;
1411 	int err;
1412 
1413 	priv_lock(priv);
1414 	err = priv_dev_set_link(priv, dev, 1);
1415 	priv_unlock(priv);
1416 	return err;
1417 }
1418 
1419 /**
1420  * Configure the TX function to use.
1421  *
1422  * @param priv
1423  *   Pointer to private data structure.
1424  * @param dev
1425  *   Pointer to rte_eth_dev structure.
1426  */
1427 void
1428 priv_dev_select_tx_function(struct priv *priv, struct rte_eth_dev *dev)
1429 {
1430 	assert(priv != NULL);
1431 	assert(dev != NULL);
1432 	dev->tx_pkt_burst = mlx5_tx_burst;
1433 	/* Select appropriate TX function. */
1434 	if (priv->mps == MLX5_MPW_ENHANCED) {
1435 		if (priv_check_vec_tx_support(priv) > 0) {
1436 			if (priv_check_raw_vec_tx_support(priv) > 0)
1437 				dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1438 			else
1439 				dev->tx_pkt_burst = mlx5_tx_burst_vec;
1440 			DEBUG("selected Enhanced MPW TX vectorized function");
1441 		} else {
1442 			dev->tx_pkt_burst = mlx5_tx_burst_empw;
1443 			DEBUG("selected Enhanced MPW TX function");
1444 		}
1445 	} else if (priv->mps && priv->txq_inline) {
1446 		dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1447 		DEBUG("selected MPW inline TX function");
1448 	} else if (priv->mps) {
1449 		dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1450 		DEBUG("selected MPW TX function");
1451 	}
1452 }
1453 
1454 /**
1455  * Configure the RX function to use.
1456  *
1457  * @param priv
1458  *   Pointer to private data structure.
1459  * @param dev
1460  *   Pointer to rte_eth_dev structure.
1461  */
1462 void
1463 priv_dev_select_rx_function(struct priv *priv, struct rte_eth_dev *dev)
1464 {
1465 	assert(priv != NULL);
1466 	assert(dev != NULL);
1467 	if (priv_check_vec_rx_support(priv) > 0) {
1468 		dev->rx_pkt_burst = mlx5_rx_burst_vec;
1469 		DEBUG("selected RX vectorized function");
1470 	} else {
1471 		dev->rx_pkt_burst = mlx5_rx_burst;
1472 	}
1473 }
1474