xref: /dpdk/drivers/net/mlx5/linux/mlx5_ethdev_os.c (revision fda34680eb9abf53872bde66e119f4c0288fd62f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2015 6WIND S.A.
3  * Copyright 2015 Mellanox Technologies, Ltd
4  */
5 
6 #include <stddef.h>
7 #include <inttypes.h>
8 #include <unistd.h>
9 #include <stdbool.h>
10 #include <stdint.h>
11 #include <stdio.h>
12 #include <string.h>
13 #include <stdlib.h>
14 #include <errno.h>
15 #include <dirent.h>
16 #include <net/if.h>
17 #include <sys/ioctl.h>
18 #include <sys/socket.h>
19 #include <netinet/in.h>
20 #include <linux/ethtool.h>
21 #include <linux/sockios.h>
22 #include <fcntl.h>
23 #include <stdalign.h>
24 #include <sys/un.h>
25 #include <time.h>
26 
27 #include <rte_atomic.h>
28 #include <rte_ethdev_driver.h>
29 #include <rte_bus_pci.h>
30 #include <rte_mbuf.h>
31 #include <rte_common.h>
32 #include <rte_interrupts.h>
33 #include <rte_malloc.h>
34 #include <rte_string_fns.h>
35 #include <rte_rwlock.h>
36 #include <rte_cycles.h>
37 
38 #include <mlx5_glue.h>
39 #include <mlx5_devx_cmds.h>
40 #include <mlx5_common.h>
41 #include <mlx5_malloc.h>
42 
43 #include "mlx5.h"
44 #include "mlx5_rxtx.h"
45 #include "mlx5_utils.h"
46 
47 /* Supported speed values found in /usr/include/linux/ethtool.h */
48 #ifndef HAVE_SUPPORTED_40000baseKR4_Full
49 #define SUPPORTED_40000baseKR4_Full (1 << 23)
50 #endif
51 #ifndef HAVE_SUPPORTED_40000baseCR4_Full
52 #define SUPPORTED_40000baseCR4_Full (1 << 24)
53 #endif
54 #ifndef HAVE_SUPPORTED_40000baseSR4_Full
55 #define SUPPORTED_40000baseSR4_Full (1 << 25)
56 #endif
57 #ifndef HAVE_SUPPORTED_40000baseLR4_Full
58 #define SUPPORTED_40000baseLR4_Full (1 << 26)
59 #endif
60 #ifndef HAVE_SUPPORTED_56000baseKR4_Full
61 #define SUPPORTED_56000baseKR4_Full (1 << 27)
62 #endif
63 #ifndef HAVE_SUPPORTED_56000baseCR4_Full
64 #define SUPPORTED_56000baseCR4_Full (1 << 28)
65 #endif
66 #ifndef HAVE_SUPPORTED_56000baseSR4_Full
67 #define SUPPORTED_56000baseSR4_Full (1 << 29)
68 #endif
69 #ifndef HAVE_SUPPORTED_56000baseLR4_Full
70 #define SUPPORTED_56000baseLR4_Full (1 << 30)
71 #endif
72 
73 /* Add defines in case the running kernel is not the same as user headers. */
74 #ifndef ETHTOOL_GLINKSETTINGS
75 struct ethtool_link_settings {
76 	uint32_t cmd;
77 	uint32_t speed;
78 	uint8_t duplex;
79 	uint8_t port;
80 	uint8_t phy_address;
81 	uint8_t autoneg;
82 	uint8_t mdio_support;
83 	uint8_t eth_to_mdix;
84 	uint8_t eth_tp_mdix_ctrl;
85 	int8_t link_mode_masks_nwords;
86 	uint32_t reserved[8];
87 	uint32_t link_mode_masks[];
88 };
89 
90 /* The kernel values can be found in /include/uapi/linux/ethtool.h */
91 #define ETHTOOL_GLINKSETTINGS 0x0000004c
92 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
93 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
94 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
95 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
96 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
97 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
98 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
99 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
100 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
101 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
102 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
103 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
104 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
105 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
106 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
107 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
108 #endif
109 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
110 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
111 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
112 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
113 #endif
114 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
115 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
116 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
117 #endif
118 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
119 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
120 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
121 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
122 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
123 #endif
124 #ifndef HAVE_ETHTOOL_LINK_MODE_200G
125 #define ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT 62
126 #define ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT 63
127 #define ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT 0 /* 64 - 64 */
128 #define ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT 1 /* 65 - 64 */
129 #define ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT 2 /* 66 - 64 */
130 #endif
131 
132 
133 /**
134  * Get interface name from private structure.
135  *
136  * This is a port representor-aware version of mlx5_get_ifname_sysfs().
137  *
138  * @param[in] dev
139  *   Pointer to Ethernet device.
140  * @param[out] ifname
141  *   Interface name output buffer.
142  *
143  * @return
144  *   0 on success, a negative errno value otherwise and rte_errno is set.
145  */
146 int
147 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[IF_NAMESIZE])
148 {
149 	struct mlx5_priv *priv = dev->data->dev_private;
150 	unsigned int ifindex;
151 
152 	MLX5_ASSERT(priv);
153 	MLX5_ASSERT(priv->sh);
154 	ifindex = mlx5_ifindex(dev);
155 	if (!ifindex) {
156 		if (!priv->representor)
157 			return mlx5_get_ifname_sysfs(priv->sh->ibdev_path,
158 						     *ifname);
159 		rte_errno = ENXIO;
160 		return -rte_errno;
161 	}
162 	if (if_indextoname(ifindex, &(*ifname)[0]))
163 		return 0;
164 	rte_errno = errno;
165 	return -rte_errno;
166 }
167 
168 /**
169  * Perform ifreq ioctl() on associated Ethernet device.
170  *
171  * @param[in] dev
172  *   Pointer to Ethernet device.
173  * @param req
174  *   Request number to pass to ioctl().
175  * @param[out] ifr
176  *   Interface request structure output buffer.
177  *
178  * @return
179  *   0 on success, a negative errno value otherwise and rte_errno is set.
180  */
181 static int
182 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
183 {
184 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
185 	int ret = 0;
186 
187 	if (sock == -1) {
188 		rte_errno = errno;
189 		return -rte_errno;
190 	}
191 	ret = mlx5_get_ifname(dev, &ifr->ifr_name);
192 	if (ret)
193 		goto error;
194 	ret = ioctl(sock, req, ifr);
195 	if (ret == -1) {
196 		rte_errno = errno;
197 		goto error;
198 	}
199 	close(sock);
200 	return 0;
201 error:
202 	close(sock);
203 	return -rte_errno;
204 }
205 
206 /**
207  * Get device MTU.
208  *
209  * @param dev
210  *   Pointer to Ethernet device.
211  * @param[out] mtu
212  *   MTU value output buffer.
213  *
214  * @return
215  *   0 on success, a negative errno value otherwise and rte_errno is set.
216  */
217 int
218 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
219 {
220 	struct ifreq request;
221 	int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
222 
223 	if (ret)
224 		return ret;
225 	*mtu = request.ifr_mtu;
226 	return 0;
227 }
228 
229 /**
230  * Set device MTU.
231  *
232  * @param dev
233  *   Pointer to Ethernet device.
234  * @param mtu
235  *   MTU value to set.
236  *
237  * @return
238  *   0 on success, a negative errno value otherwise and rte_errno is set.
239  */
240 int
241 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
242 {
243 	struct ifreq request = { .ifr_mtu = mtu, };
244 
245 	return mlx5_ifreq(dev, SIOCSIFMTU, &request);
246 }
247 
248 /**
249  * Set device flags.
250  *
251  * @param dev
252  *   Pointer to Ethernet device.
253  * @param keep
254  *   Bitmask for flags that must remain untouched.
255  * @param flags
256  *   Bitmask for flags to modify.
257  *
258  * @return
259  *   0 on success, a negative errno value otherwise and rte_errno is set.
260  */
261 static int
262 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
263 {
264 	struct ifreq request;
265 	int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
266 
267 	if (ret)
268 		return ret;
269 	request.ifr_flags &= keep;
270 	request.ifr_flags |= flags & ~keep;
271 	return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
272 }
273 
274 /**
275  * Get device current raw clock counter
276  *
277  * @param dev
278  *   Pointer to Ethernet device structure.
279  * @param[out] time
280  *   Current raw clock counter of the device.
281  *
282  * @return
283  *   0 if the clock has correctly been read
284  *   The value of errno in case of error
285  */
286 int
287 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
288 {
289 	struct mlx5_priv *priv = dev->data->dev_private;
290 	struct ibv_context *ctx = priv->sh->ctx;
291 	struct ibv_values_ex values;
292 	int err = 0;
293 
294 	values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
295 	err = mlx5_glue->query_rt_values_ex(ctx, &values);
296 	if (err != 0) {
297 		DRV_LOG(WARNING, "Could not query the clock !");
298 		return err;
299 	}
300 	*clock = values.raw_clock.tv_nsec;
301 	return 0;
302 }
303 
304 /**
305  * Retrieve the master device for representor in the same switch domain.
306  *
307  * @param dev
308  *   Pointer to representor Ethernet device structure.
309  *
310  * @return
311  *   Master device structure  on success, NULL otherwise.
312  */
313 static struct rte_eth_dev *
314 mlx5_find_master_dev(struct rte_eth_dev *dev)
315 {
316 	struct mlx5_priv *priv;
317 	uint16_t port_id;
318 	uint16_t domain_id;
319 
320 	priv = dev->data->dev_private;
321 	domain_id = priv->domain_id;
322 	MLX5_ASSERT(priv->representor);
323 	MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
324 		struct mlx5_priv *opriv =
325 			rte_eth_devices[port_id].data->dev_private;
326 		if (opriv &&
327 		    opriv->master &&
328 		    opriv->domain_id == domain_id &&
329 		    opriv->sh == priv->sh)
330 			return &rte_eth_devices[port_id];
331 	}
332 	return NULL;
333 }
334 
335 /**
336  * DPDK callback to retrieve physical link information.
337  *
338  * @param dev
339  *   Pointer to Ethernet device structure.
340  * @param[out] link
341  *   Storage for current link status.
342  *
343  * @return
344  *   0 on success, a negative errno value otherwise and rte_errno is set.
345  */
346 static int
347 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
348 			       struct rte_eth_link *link)
349 {
350 	struct mlx5_priv *priv = dev->data->dev_private;
351 	struct ethtool_cmd edata = {
352 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
353 	};
354 	struct ifreq ifr;
355 	struct rte_eth_link dev_link;
356 	int link_speed = 0;
357 	int ret;
358 
359 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
360 	if (ret) {
361 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
362 			dev->data->port_id, strerror(rte_errno));
363 		return ret;
364 	}
365 	dev_link = (struct rte_eth_link) {
366 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
367 				(ifr.ifr_flags & IFF_RUNNING)),
368 	};
369 	ifr = (struct ifreq) {
370 		.ifr_data = (void *)&edata,
371 	};
372 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
373 	if (ret) {
374 		if (ret == -ENOTSUP && priv->representor) {
375 			struct rte_eth_dev *master;
376 
377 			/*
378 			 * For representors we can try to inherit link
379 			 * settings from the master device. Actually
380 			 * link settings do not make a lot of sense
381 			 * for representors due to missing physical
382 			 * link. The old kernel drivers supported
383 			 * emulated settings query for representors,
384 			 * the new ones do not, so we have to add
385 			 * this code for compatibility issues.
386 			 */
387 			master = mlx5_find_master_dev(dev);
388 			if (master) {
389 				ifr = (struct ifreq) {
390 					.ifr_data = (void *)&edata,
391 				};
392 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
393 			}
394 		}
395 		if (ret) {
396 			DRV_LOG(WARNING,
397 				"port %u ioctl(SIOCETHTOOL,"
398 				" ETHTOOL_GSET) failed: %s",
399 				dev->data->port_id, strerror(rte_errno));
400 			return ret;
401 		}
402 	}
403 	link_speed = ethtool_cmd_speed(&edata);
404 	if (link_speed == -1)
405 		dev_link.link_speed = ETH_SPEED_NUM_NONE;
406 	else
407 		dev_link.link_speed = link_speed;
408 	priv->link_speed_capa = 0;
409 	if (edata.supported & SUPPORTED_Autoneg)
410 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
411 	if (edata.supported & (SUPPORTED_1000baseT_Full |
412 			       SUPPORTED_1000baseKX_Full))
413 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
414 	if (edata.supported & SUPPORTED_10000baseKR_Full)
415 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
416 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
417 			       SUPPORTED_40000baseCR4_Full |
418 			       SUPPORTED_40000baseSR4_Full |
419 			       SUPPORTED_40000baseLR4_Full))
420 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
421 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
422 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
423 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
424 			ETH_LINK_SPEED_FIXED);
425 	if (((dev_link.link_speed && !dev_link.link_status) ||
426 	     (!dev_link.link_speed && dev_link.link_status))) {
427 		rte_errno = EAGAIN;
428 		return -rte_errno;
429 	}
430 	*link = dev_link;
431 	return 0;
432 }
433 
434 /**
435  * Retrieve physical link information (unlocked version using new ioctl).
436  *
437  * @param dev
438  *   Pointer to Ethernet device structure.
439  * @param[out] link
440  *   Storage for current link status.
441  *
442  * @return
443  *   0 on success, a negative errno value otherwise and rte_errno is set.
444  */
445 static int
446 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
447 			     struct rte_eth_link *link)
448 
449 {
450 	struct mlx5_priv *priv = dev->data->dev_private;
451 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
452 	struct ifreq ifr;
453 	struct rte_eth_link dev_link;
454 	struct rte_eth_dev *master = NULL;
455 	uint64_t sc;
456 	int ret;
457 
458 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
459 	if (ret) {
460 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
461 			dev->data->port_id, strerror(rte_errno));
462 		return ret;
463 	}
464 	dev_link = (struct rte_eth_link) {
465 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
466 				(ifr.ifr_flags & IFF_RUNNING)),
467 	};
468 	ifr = (struct ifreq) {
469 		.ifr_data = (void *)&gcmd,
470 	};
471 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
472 	if (ret) {
473 		if (ret == -ENOTSUP && priv->representor) {
474 			/*
475 			 * For representors we can try to inherit link
476 			 * settings from the master device. Actually
477 			 * link settings do not make a lot of sense
478 			 * for representors due to missing physical
479 			 * link. The old kernel drivers supported
480 			 * emulated settings query for representors,
481 			 * the new ones do not, so we have to add
482 			 * this code for compatibility issues.
483 			 */
484 			master = mlx5_find_master_dev(dev);
485 			if (master) {
486 				ifr = (struct ifreq) {
487 					.ifr_data = (void *)&gcmd,
488 				};
489 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
490 			}
491 		}
492 		if (ret) {
493 			DRV_LOG(DEBUG,
494 				"port %u ioctl(SIOCETHTOOL,"
495 				" ETHTOOL_GLINKSETTINGS) failed: %s",
496 				dev->data->port_id, strerror(rte_errno));
497 			return ret;
498 		}
499 	}
500 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
501 
502 	alignas(struct ethtool_link_settings)
503 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
504 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
505 	struct ethtool_link_settings *ecmd = (void *)data;
506 
507 	*ecmd = gcmd;
508 	ifr.ifr_data = (void *)ecmd;
509 	ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
510 	if (ret) {
511 		DRV_LOG(DEBUG,
512 			"port %u ioctl(SIOCETHTOOL,"
513 			"ETHTOOL_GLINKSETTINGS) failed: %s",
514 			dev->data->port_id, strerror(rte_errno));
515 		return ret;
516 	}
517 	dev_link.link_speed = (ecmd->speed == UINT32_MAX) ? ETH_SPEED_NUM_NONE :
518 							    ecmd->speed;
519 	sc = ecmd->link_mode_masks[0] |
520 		((uint64_t)ecmd->link_mode_masks[1] << 32);
521 	priv->link_speed_capa = 0;
522 	if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
523 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
524 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
525 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
526 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
527 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
528 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
529 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
530 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
531 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
532 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
533 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
534 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
535 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
536 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
537 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
538 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
539 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
540 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
541 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
542 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
543 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
544 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
545 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
546 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
547 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
548 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
549 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
550 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
551 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
552 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
553 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
554 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
555 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
556 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT) |
557 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT)))
558 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
559 
560 	sc = ecmd->link_mode_masks[2] |
561 		((uint64_t)ecmd->link_mode_masks[3] << 32);
562 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT) |
563 		  MLX5_BITSHIFT
564 		       (ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT) |
565 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT)))
566 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
567 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
568 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
569 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
570 				  ETH_LINK_SPEED_FIXED);
571 	if (((dev_link.link_speed && !dev_link.link_status) ||
572 	     (!dev_link.link_speed && dev_link.link_status))) {
573 		rte_errno = EAGAIN;
574 		return -rte_errno;
575 	}
576 	*link = dev_link;
577 	return 0;
578 }
579 
580 /**
581  * DPDK callback to retrieve physical link information.
582  *
583  * @param dev
584  *   Pointer to Ethernet device structure.
585  * @param wait_to_complete
586  *   Wait for request completion.
587  *
588  * @return
589  *   0 if link status was not updated, positive if it was, a negative errno
590  *   value otherwise and rte_errno is set.
591  */
592 int
593 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
594 {
595 	int ret;
596 	struct rte_eth_link dev_link;
597 	time_t start_time = time(NULL);
598 	int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT;
599 
600 	do {
601 		ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
602 		if (ret == -ENOTSUP)
603 			ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
604 		if (ret == 0)
605 			break;
606 		/* Handle wait to complete situation. */
607 		if ((wait_to_complete || retry) && ret == -EAGAIN) {
608 			if (abs((int)difftime(time(NULL), start_time)) <
609 			    MLX5_LINK_STATUS_TIMEOUT) {
610 				usleep(0);
611 				continue;
612 			} else {
613 				rte_errno = EBUSY;
614 				return -rte_errno;
615 			}
616 		} else if (ret < 0) {
617 			return ret;
618 		}
619 	} while (wait_to_complete || retry-- > 0);
620 	ret = !!memcmp(&dev->data->dev_link, &dev_link,
621 		       sizeof(struct rte_eth_link));
622 	dev->data->dev_link = dev_link;
623 	return ret;
624 }
625 
626 /**
627  * DPDK callback to get flow control status.
628  *
629  * @param dev
630  *   Pointer to Ethernet device structure.
631  * @param[out] fc_conf
632  *   Flow control output buffer.
633  *
634  * @return
635  *   0 on success, a negative errno value otherwise and rte_errno is set.
636  */
637 int
638 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
639 {
640 	struct ifreq ifr;
641 	struct ethtool_pauseparam ethpause = {
642 		.cmd = ETHTOOL_GPAUSEPARAM
643 	};
644 	int ret;
645 
646 	ifr.ifr_data = (void *)&ethpause;
647 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
648 	if (ret) {
649 		DRV_LOG(WARNING,
650 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
651 			" %s",
652 			dev->data->port_id, strerror(rte_errno));
653 		return ret;
654 	}
655 	fc_conf->autoneg = ethpause.autoneg;
656 	if (ethpause.rx_pause && ethpause.tx_pause)
657 		fc_conf->mode = RTE_FC_FULL;
658 	else if (ethpause.rx_pause)
659 		fc_conf->mode = RTE_FC_RX_PAUSE;
660 	else if (ethpause.tx_pause)
661 		fc_conf->mode = RTE_FC_TX_PAUSE;
662 	else
663 		fc_conf->mode = RTE_FC_NONE;
664 	return 0;
665 }
666 
667 /**
668  * DPDK callback to modify flow control parameters.
669  *
670  * @param dev
671  *   Pointer to Ethernet device structure.
672  * @param[in] fc_conf
673  *   Flow control parameters.
674  *
675  * @return
676  *   0 on success, a negative errno value otherwise and rte_errno is set.
677  */
678 int
679 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
680 {
681 	struct ifreq ifr;
682 	struct ethtool_pauseparam ethpause = {
683 		.cmd = ETHTOOL_SPAUSEPARAM
684 	};
685 	int ret;
686 
687 	ifr.ifr_data = (void *)&ethpause;
688 	ethpause.autoneg = fc_conf->autoneg;
689 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
690 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
691 		ethpause.rx_pause = 1;
692 	else
693 		ethpause.rx_pause = 0;
694 
695 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
696 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
697 		ethpause.tx_pause = 1;
698 	else
699 		ethpause.tx_pause = 0;
700 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
701 	if (ret) {
702 		DRV_LOG(WARNING,
703 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
704 			" failed: %s",
705 			dev->data->port_id, strerror(rte_errno));
706 		return ret;
707 	}
708 	return 0;
709 }
710 
711 /**
712  * Handle asynchronous removal event for entire multiport device.
713  *
714  * @param sh
715  *   Infiniband device shared context.
716  */
717 static void
718 mlx5_dev_interrupt_device_fatal(struct mlx5_dev_ctx_shared *sh)
719 {
720 	uint32_t i;
721 
722 	for (i = 0; i < sh->max_port; ++i) {
723 		struct rte_eth_dev *dev;
724 
725 		if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
726 			/*
727 			 * Or not existing port either no
728 			 * handler installed for this port.
729 			 */
730 			continue;
731 		}
732 		dev = &rte_eth_devices[sh->port[i].ih_port_id];
733 		MLX5_ASSERT(dev);
734 		if (dev->data->dev_conf.intr_conf.rmv)
735 			_rte_eth_dev_callback_process
736 				(dev, RTE_ETH_EVENT_INTR_RMV, NULL);
737 	}
738 }
739 
740 /**
741  * Handle shared asynchronous events the NIC (removal event
742  * and link status change). Supports multiport IB device.
743  *
744  * @param cb_arg
745  *   Callback argument.
746  */
747 void
748 mlx5_dev_interrupt_handler(void *cb_arg)
749 {
750 	struct mlx5_dev_ctx_shared *sh = cb_arg;
751 	struct ibv_async_event event;
752 
753 	/* Read all message from the IB device and acknowledge them. */
754 	for (;;) {
755 		struct rte_eth_dev *dev;
756 		uint32_t tmp;
757 
758 		if (mlx5_glue->get_async_event(sh->ctx, &event))
759 			break;
760 		/* Retrieve and check IB port index. */
761 		tmp = (uint32_t)event.element.port_num;
762 		if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
763 			/*
764 			 * The DEVICE_FATAL event is called once for
765 			 * entire device without port specifying.
766 			 * We should notify all existing ports.
767 			 */
768 			mlx5_glue->ack_async_event(&event);
769 			mlx5_dev_interrupt_device_fatal(sh);
770 			continue;
771 		}
772 		MLX5_ASSERT(tmp && (tmp <= sh->max_port));
773 		if (!tmp) {
774 			/* Unsupported device level event. */
775 			mlx5_glue->ack_async_event(&event);
776 			DRV_LOG(DEBUG,
777 				"unsupported common event (type %d)",
778 				event.event_type);
779 			continue;
780 		}
781 		if (tmp > sh->max_port) {
782 			/* Invalid IB port index. */
783 			mlx5_glue->ack_async_event(&event);
784 			DRV_LOG(DEBUG,
785 				"cannot handle an event (type %d)"
786 				"due to invalid IB port index (%u)",
787 				event.event_type, tmp);
788 			continue;
789 		}
790 		if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
791 			/* No handler installed. */
792 			mlx5_glue->ack_async_event(&event);
793 			DRV_LOG(DEBUG,
794 				"cannot handle an event (type %d)"
795 				"due to no handler installed for port %u",
796 				event.event_type, tmp);
797 			continue;
798 		}
799 		/* Retrieve ethernet device descriptor. */
800 		tmp = sh->port[tmp - 1].ih_port_id;
801 		dev = &rte_eth_devices[tmp];
802 		MLX5_ASSERT(dev);
803 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
804 		     event.event_type == IBV_EVENT_PORT_ERR) &&
805 			dev->data->dev_conf.intr_conf.lsc) {
806 			mlx5_glue->ack_async_event(&event);
807 			if (mlx5_link_update(dev, 0) == -EAGAIN) {
808 				usleep(0);
809 				continue;
810 			}
811 			_rte_eth_dev_callback_process
812 				(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
813 			continue;
814 		}
815 		DRV_LOG(DEBUG,
816 			"port %u cannot handle an unknown event (type %d)",
817 			dev->data->port_id, event.event_type);
818 		mlx5_glue->ack_async_event(&event);
819 	}
820 }
821 
822 /*
823  * Unregister callback handler safely. The handler may be active
824  * while we are trying to unregister it, in this case code -EAGAIN
825  * is returned by rte_intr_callback_unregister(). This routine checks
826  * the return code and tries to unregister handler again.
827  *
828  * @param handle
829  *   interrupt handle
830  * @param cb_fn
831  *   pointer to callback routine
832  * @cb_arg
833  *   opaque callback parameter
834  */
835 void
836 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
837 			      rte_intr_callback_fn cb_fn, void *cb_arg)
838 {
839 	/*
840 	 * Try to reduce timeout management overhead by not calling
841 	 * the timer related routines on the first iteration. If the
842 	 * unregistering succeeds on first call there will be no
843 	 * timer calls at all.
844 	 */
845 	uint64_t twait = 0;
846 	uint64_t start = 0;
847 
848 	do {
849 		int ret;
850 
851 		ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
852 		if (ret >= 0)
853 			return;
854 		if (ret != -EAGAIN) {
855 			DRV_LOG(INFO, "failed to unregister interrupt"
856 				      " handler (error: %d)", ret);
857 			MLX5_ASSERT(false);
858 			return;
859 		}
860 		if (twait) {
861 			struct timespec onems;
862 
863 			/* Wait one millisecond and try again. */
864 			onems.tv_sec = 0;
865 			onems.tv_nsec = NS_PER_S / MS_PER_S;
866 			nanosleep(&onems, 0);
867 			/* Check whether one second elapsed. */
868 			if ((rte_get_timer_cycles() - start) <= twait)
869 				continue;
870 		} else {
871 			/*
872 			 * We get the amount of timer ticks for one second.
873 			 * If this amount elapsed it means we spent one
874 			 * second in waiting. This branch is executed once
875 			 * on first iteration.
876 			 */
877 			twait = rte_get_timer_hz();
878 			MLX5_ASSERT(twait);
879 		}
880 		/*
881 		 * Timeout elapsed, show message (once a second) and retry.
882 		 * We have no other acceptable option here, if we ignore
883 		 * the unregistering return code the handler will not
884 		 * be unregistered, fd will be closed and we may get the
885 		 * crush. Hanging and messaging in the loop seems not to be
886 		 * the worst choice.
887 		 */
888 		DRV_LOG(INFO, "Retrying to unregister interrupt handler");
889 		start = rte_get_timer_cycles();
890 	} while (true);
891 }
892 
893 /**
894  * Handle DEVX interrupts from the NIC.
895  * This function is probably called from the DPDK host thread.
896  *
897  * @param cb_arg
898  *   Callback argument.
899  */
900 void
901 mlx5_dev_interrupt_handler_devx(void *cb_arg)
902 {
903 #ifndef HAVE_IBV_DEVX_ASYNC
904 	(void)cb_arg;
905 	return;
906 #else
907 	struct mlx5_dev_ctx_shared *sh = cb_arg;
908 	union {
909 		struct mlx5dv_devx_async_cmd_hdr cmd_resp;
910 		uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
911 			    MLX5_ST_SZ_BYTES(traffic_counter) +
912 			    sizeof(struct mlx5dv_devx_async_cmd_hdr)];
913 	} out;
914 	uint8_t *buf = out.buf + sizeof(out.cmd_resp);
915 
916 	while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
917 						   &out.cmd_resp,
918 						   sizeof(out.buf)))
919 		mlx5_flow_async_pool_query_handle
920 			(sh, (uint64_t)out.cmd_resp.wr_id,
921 			 mlx5_devx_get_out_command_status(buf));
922 #endif /* HAVE_IBV_DEVX_ASYNC */
923 }
924 
925 /**
926  * DPDK callback to bring the link DOWN.
927  *
928  * @param dev
929  *   Pointer to Ethernet device structure.
930  *
931  * @return
932  *   0 on success, a negative errno value otherwise and rte_errno is set.
933  */
934 int
935 mlx5_set_link_down(struct rte_eth_dev *dev)
936 {
937 	return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
938 }
939 
940 /**
941  * DPDK callback to bring the link UP.
942  *
943  * @param dev
944  *   Pointer to Ethernet device structure.
945  *
946  * @return
947  *   0 on success, a negative errno value otherwise and rte_errno is set.
948  */
949 int
950 mlx5_set_link_up(struct rte_eth_dev *dev)
951 {
952 	return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
953 }
954 
955 /**
956  * Check if mlx5 device was removed.
957  *
958  * @param dev
959  *   Pointer to Ethernet device structure.
960  *
961  * @return
962  *   1 when device is removed, otherwise 0.
963  */
964 int
965 mlx5_is_removed(struct rte_eth_dev *dev)
966 {
967 	struct ibv_device_attr device_attr;
968 	struct mlx5_priv *priv = dev->data->dev_private;
969 
970 	if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
971 		return 1;
972 	return 0;
973 }
974 
975 /**
976  * Analyze gathered port parameters via sysfs to recognize master
977  * and representor devices for E-Switch configuration.
978  *
979  * @param[in] device_dir
980  *   flag of presence of "device" directory under port device key.
981  * @param[inout] switch_info
982  *   Port information, including port name as a number and port name
983  *   type if recognized
984  *
985  * @return
986  *   master and representor flags are set in switch_info according to
987  *   recognized parameters (if any).
988  */
989 static void
990 mlx5_sysfs_check_switch_info(bool device_dir,
991 			     struct mlx5_switch_info *switch_info)
992 {
993 	switch (switch_info->name_type) {
994 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
995 		/*
996 		 * Name is not recognized, assume the master,
997 		 * check the device directory presence.
998 		 */
999 		switch_info->master = device_dir;
1000 		break;
1001 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1002 		/*
1003 		 * Name is not set, this assumes the legacy naming
1004 		 * schema for master, just check if there is
1005 		 * a device directory.
1006 		 */
1007 		switch_info->master = device_dir;
1008 		break;
1009 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1010 		/* New uplink naming schema recognized. */
1011 		switch_info->master = 1;
1012 		break;
1013 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1014 		/* Legacy representors naming schema. */
1015 		switch_info->representor = !device_dir;
1016 		break;
1017 	case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
1018 		/* Fallthrough */
1019 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1020 		/* New representors naming schema. */
1021 		switch_info->representor = 1;
1022 		break;
1023 	}
1024 }
1025 
1026 /**
1027  * Get switch information associated with network interface.
1028  *
1029  * @param ifindex
1030  *   Network interface index.
1031  * @param[out] info
1032  *   Switch information object, populated in case of success.
1033  *
1034  * @return
1035  *   0 on success, a negative errno value otherwise and rte_errno is set.
1036  */
1037 int
1038 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1039 {
1040 	char ifname[IF_NAMESIZE];
1041 	char port_name[IF_NAMESIZE];
1042 	FILE *file;
1043 	struct mlx5_switch_info data = {
1044 		.master = 0,
1045 		.representor = 0,
1046 		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1047 		.port_name = 0,
1048 		.switch_id = 0,
1049 	};
1050 	DIR *dir;
1051 	bool port_switch_id_set = false;
1052 	bool device_dir = false;
1053 	char c;
1054 	int ret;
1055 
1056 	if (!if_indextoname(ifindex, ifname)) {
1057 		rte_errno = errno;
1058 		return -rte_errno;
1059 	}
1060 
1061 	MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1062 	      ifname);
1063 	MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1064 	      ifname);
1065 	MKSTR(pci_device, "/sys/class/net/%s/device",
1066 	      ifname);
1067 
1068 	file = fopen(phys_port_name, "rb");
1069 	if (file != NULL) {
1070 		ret = fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", port_name);
1071 		fclose(file);
1072 		if (ret == 1)
1073 			mlx5_translate_port_name(port_name, &data);
1074 	}
1075 	file = fopen(phys_switch_id, "rb");
1076 	if (file == NULL) {
1077 		rte_errno = errno;
1078 		return -rte_errno;
1079 	}
1080 	port_switch_id_set =
1081 		fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1082 		c == '\n';
1083 	fclose(file);
1084 	dir = opendir(pci_device);
1085 	if (dir != NULL) {
1086 		closedir(dir);
1087 		device_dir = true;
1088 	}
1089 	if (port_switch_id_set) {
1090 		/* We have some E-Switch configuration. */
1091 		mlx5_sysfs_check_switch_info(device_dir, &data);
1092 	}
1093 	*info = data;
1094 	MLX5_ASSERT(!(data.master && data.representor));
1095 	if (data.master && data.representor) {
1096 		DRV_LOG(ERR, "ifindex %u device is recognized as master"
1097 			     " and as representor", ifindex);
1098 		rte_errno = ENODEV;
1099 		return -rte_errno;
1100 	}
1101 	return 0;
1102 }
1103 
1104 /**
1105  * DPDK callback to retrieve plug-in module EEPROM information (type and size).
1106  *
1107  * @param dev
1108  *   Pointer to Ethernet device structure.
1109  * @param[out] modinfo
1110  *   Storage for plug-in module EEPROM information.
1111  *
1112  * @return
1113  *   0 on success, a negative errno value otherwise and rte_errno is set.
1114  */
1115 int
1116 mlx5_get_module_info(struct rte_eth_dev *dev,
1117 		     struct rte_eth_dev_module_info *modinfo)
1118 {
1119 	struct ethtool_modinfo info = {
1120 		.cmd = ETHTOOL_GMODULEINFO,
1121 	};
1122 	struct ifreq ifr = (struct ifreq) {
1123 		.ifr_data = (void *)&info,
1124 	};
1125 	int ret = 0;
1126 
1127 	if (!dev || !modinfo) {
1128 		DRV_LOG(WARNING, "missing argument, cannot get module info");
1129 		rte_errno = EINVAL;
1130 		return -rte_errno;
1131 	}
1132 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1133 	if (ret) {
1134 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1135 			dev->data->port_id, strerror(rte_errno));
1136 		return ret;
1137 	}
1138 	modinfo->type = info.type;
1139 	modinfo->eeprom_len = info.eeprom_len;
1140 	return ret;
1141 }
1142 
1143 /**
1144  * DPDK callback to retrieve plug-in module EEPROM data.
1145  *
1146  * @param dev
1147  *   Pointer to Ethernet device structure.
1148  * @param[out] info
1149  *   Storage for plug-in module EEPROM data.
1150  *
1151  * @return
1152  *   0 on success, a negative errno value otherwise and rte_errno is set.
1153  */
1154 int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
1155 			   struct rte_dev_eeprom_info *info)
1156 {
1157 	struct ethtool_eeprom *eeprom;
1158 	struct ifreq ifr;
1159 	int ret = 0;
1160 
1161 	if (!dev || !info) {
1162 		DRV_LOG(WARNING, "missing argument, cannot get module eeprom");
1163 		rte_errno = EINVAL;
1164 		return -rte_errno;
1165 	}
1166 	eeprom = mlx5_malloc(MLX5_MEM_ZERO,
1167 			     (sizeof(struct ethtool_eeprom) + info->length), 0,
1168 			     SOCKET_ID_ANY);
1169 	if (!eeprom) {
1170 		DRV_LOG(WARNING, "port %u cannot allocate memory for "
1171 			"eeprom data", dev->data->port_id);
1172 		rte_errno = ENOMEM;
1173 		return -rte_errno;
1174 	}
1175 	eeprom->cmd = ETHTOOL_GMODULEEEPROM;
1176 	eeprom->offset = info->offset;
1177 	eeprom->len = info->length;
1178 	ifr = (struct ifreq) {
1179 		.ifr_data = (void *)eeprom,
1180 	};
1181 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1182 	if (ret)
1183 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1184 			dev->data->port_id, strerror(rte_errno));
1185 	else
1186 		rte_memcpy(info->data, eeprom->data, info->length);
1187 	mlx5_free(eeprom);
1188 	return ret;
1189 }
1190 
1191 /**
1192  * Read device counters table.
1193  *
1194  * @param dev
1195  *   Pointer to Ethernet device.
1196  * @param[out] stats
1197  *   Counters table output buffer.
1198  *
1199  * @return
1200  *   0 on success and stats is filled, negative errno value otherwise and
1201  *   rte_errno is set.
1202  */
1203 int
1204 mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
1205 {
1206 	struct mlx5_priv *priv = dev->data->dev_private;
1207 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1208 	unsigned int i;
1209 	struct ifreq ifr;
1210 	unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t);
1211 	unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz];
1212 	struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf;
1213 	int ret;
1214 
1215 	et_stats->cmd = ETHTOOL_GSTATS;
1216 	et_stats->n_stats = xstats_ctrl->stats_n;
1217 	ifr.ifr_data = (caddr_t)et_stats;
1218 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1219 	if (ret) {
1220 		DRV_LOG(WARNING,
1221 			"port %u unable to read statistic values from device",
1222 			dev->data->port_id);
1223 		return ret;
1224 	}
1225 	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
1226 		if (xstats_ctrl->info[i].dev) {
1227 			ret = mlx5_os_read_dev_stat(priv,
1228 					    xstats_ctrl->info[i].ctr_name,
1229 					    &stats[i]);
1230 			/* return last xstats counter if fail to read. */
1231 			if (ret == 0)
1232 				xstats_ctrl->xstats[i] = stats[i];
1233 			else
1234 				stats[i] = xstats_ctrl->xstats[i];
1235 		} else {
1236 			stats[i] = (uint64_t)
1237 				et_stats->data[xstats_ctrl->dev_table_idx[i]];
1238 		}
1239 	}
1240 	return 0;
1241 }
1242 
1243 /**
1244  * Query the number of statistics provided by ETHTOOL.
1245  *
1246  * @param dev
1247  *   Pointer to Ethernet device.
1248  *
1249  * @return
1250  *   Number of statistics on success, negative errno value otherwise and
1251  *   rte_errno is set.
1252  */
1253 int
1254 mlx5_os_get_stats_n(struct rte_eth_dev *dev)
1255 {
1256 	struct ethtool_drvinfo drvinfo;
1257 	struct ifreq ifr;
1258 	int ret;
1259 
1260 	drvinfo.cmd = ETHTOOL_GDRVINFO;
1261 	ifr.ifr_data = (caddr_t)&drvinfo;
1262 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1263 	if (ret) {
1264 		DRV_LOG(WARNING, "port %u unable to query number of statistics",
1265 			dev->data->port_id);
1266 		return ret;
1267 	}
1268 	return drvinfo.n_stats;
1269 }
1270 
1271 static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
1272 	{
1273 		.dpdk_name = "rx_port_unicast_bytes",
1274 		.ctr_name = "rx_vport_unicast_bytes",
1275 	},
1276 	{
1277 		.dpdk_name = "rx_port_multicast_bytes",
1278 		.ctr_name = "rx_vport_multicast_bytes",
1279 	},
1280 	{
1281 		.dpdk_name = "rx_port_broadcast_bytes",
1282 		.ctr_name = "rx_vport_broadcast_bytes",
1283 	},
1284 	{
1285 		.dpdk_name = "rx_port_unicast_packets",
1286 		.ctr_name = "rx_vport_unicast_packets",
1287 	},
1288 	{
1289 		.dpdk_name = "rx_port_multicast_packets",
1290 		.ctr_name = "rx_vport_multicast_packets",
1291 	},
1292 	{
1293 		.dpdk_name = "rx_port_broadcast_packets",
1294 		.ctr_name = "rx_vport_broadcast_packets",
1295 	},
1296 	{
1297 		.dpdk_name = "tx_port_unicast_bytes",
1298 		.ctr_name = "tx_vport_unicast_bytes",
1299 	},
1300 	{
1301 		.dpdk_name = "tx_port_multicast_bytes",
1302 		.ctr_name = "tx_vport_multicast_bytes",
1303 	},
1304 	{
1305 		.dpdk_name = "tx_port_broadcast_bytes",
1306 		.ctr_name = "tx_vport_broadcast_bytes",
1307 	},
1308 	{
1309 		.dpdk_name = "tx_port_unicast_packets",
1310 		.ctr_name = "tx_vport_unicast_packets",
1311 	},
1312 	{
1313 		.dpdk_name = "tx_port_multicast_packets",
1314 		.ctr_name = "tx_vport_multicast_packets",
1315 	},
1316 	{
1317 		.dpdk_name = "tx_port_broadcast_packets",
1318 		.ctr_name = "tx_vport_broadcast_packets",
1319 	},
1320 	{
1321 		.dpdk_name = "rx_wqe_err",
1322 		.ctr_name = "rx_wqe_err",
1323 	},
1324 	{
1325 		.dpdk_name = "rx_crc_errors_phy",
1326 		.ctr_name = "rx_crc_errors_phy",
1327 	},
1328 	{
1329 		.dpdk_name = "rx_in_range_len_errors_phy",
1330 		.ctr_name = "rx_in_range_len_errors_phy",
1331 	},
1332 	{
1333 		.dpdk_name = "rx_symbol_err_phy",
1334 		.ctr_name = "rx_symbol_err_phy",
1335 	},
1336 	{
1337 		.dpdk_name = "tx_errors_phy",
1338 		.ctr_name = "tx_errors_phy",
1339 	},
1340 	{
1341 		.dpdk_name = "rx_out_of_buffer",
1342 		.ctr_name = "out_of_buffer",
1343 		.dev = 1,
1344 	},
1345 	{
1346 		.dpdk_name = "tx_packets_phy",
1347 		.ctr_name = "tx_packets_phy",
1348 	},
1349 	{
1350 		.dpdk_name = "rx_packets_phy",
1351 		.ctr_name = "rx_packets_phy",
1352 	},
1353 	{
1354 		.dpdk_name = "tx_discards_phy",
1355 		.ctr_name = "tx_discards_phy",
1356 	},
1357 	{
1358 		.dpdk_name = "rx_discards_phy",
1359 		.ctr_name = "rx_discards_phy",
1360 	},
1361 	{
1362 		.dpdk_name = "tx_bytes_phy",
1363 		.ctr_name = "tx_bytes_phy",
1364 	},
1365 	{
1366 		.dpdk_name = "rx_bytes_phy",
1367 		.ctr_name = "rx_bytes_phy",
1368 	},
1369 	/* Representor only */
1370 	{
1371 		.dpdk_name = "rx_packets",
1372 		.ctr_name = "vport_rx_packets",
1373 	},
1374 	{
1375 		.dpdk_name = "rx_bytes",
1376 		.ctr_name = "vport_rx_bytes",
1377 	},
1378 	{
1379 		.dpdk_name = "tx_packets",
1380 		.ctr_name = "vport_tx_packets",
1381 	},
1382 	{
1383 		.dpdk_name = "tx_bytes",
1384 		.ctr_name = "vport_tx_bytes",
1385 	},
1386 };
1387 
1388 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init);
1389 
1390 /**
1391  * Init the structures to read device counters.
1392  *
1393  * @param dev
1394  *   Pointer to Ethernet device.
1395  */
1396 void
1397 mlx5_os_stats_init(struct rte_eth_dev *dev)
1398 {
1399 	struct mlx5_priv *priv = dev->data->dev_private;
1400 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1401 	struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl;
1402 	unsigned int i;
1403 	unsigned int j;
1404 	struct ifreq ifr;
1405 	struct ethtool_gstrings *strings = NULL;
1406 	unsigned int dev_stats_n;
1407 	unsigned int str_sz;
1408 	int ret;
1409 
1410 	/* So that it won't aggregate for each init. */
1411 	xstats_ctrl->mlx5_stats_n = 0;
1412 	ret = mlx5_os_get_stats_n(dev);
1413 	if (ret < 0) {
1414 		DRV_LOG(WARNING, "port %u no extended statistics available",
1415 			dev->data->port_id);
1416 		return;
1417 	}
1418 	dev_stats_n = ret;
1419 	/* Allocate memory to grab stat names and values. */
1420 	str_sz = dev_stats_n * ETH_GSTRING_LEN;
1421 	strings = (struct ethtool_gstrings *)
1422 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
1423 			      SOCKET_ID_ANY);
1424 	if (!strings) {
1425 		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats",
1426 		     dev->data->port_id);
1427 		return;
1428 	}
1429 	strings->cmd = ETHTOOL_GSTRINGS;
1430 	strings->string_set = ETH_SS_STATS;
1431 	strings->len = dev_stats_n;
1432 	ifr.ifr_data = (caddr_t)strings;
1433 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1434 	if (ret) {
1435 		DRV_LOG(WARNING, "port %u unable to get statistic names",
1436 			dev->data->port_id);
1437 		goto free;
1438 	}
1439 	for (i = 0; i != dev_stats_n; ++i) {
1440 		const char *curr_string = (const char *)
1441 			&strings->data[i * ETH_GSTRING_LEN];
1442 
1443 		for (j = 0; j != xstats_n; ++j) {
1444 			if (!strcmp(mlx5_counters_init[j].ctr_name,
1445 				    curr_string)) {
1446 				unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1447 
1448 				xstats_ctrl->dev_table_idx[idx] = i;
1449 				xstats_ctrl->info[idx] = mlx5_counters_init[j];
1450 				break;
1451 			}
1452 		}
1453 	}
1454 	/* Add dev counters. */
1455 	for (i = 0; i != xstats_n; ++i) {
1456 		if (mlx5_counters_init[i].dev) {
1457 			unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1458 
1459 			xstats_ctrl->info[idx] = mlx5_counters_init[i];
1460 			xstats_ctrl->hw_stats[idx] = 0;
1461 		}
1462 	}
1463 	MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS);
1464 	xstats_ctrl->stats_n = dev_stats_n;
1465 	/* Copy to base at first time. */
1466 	ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base);
1467 	if (ret)
1468 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
1469 			dev->data->port_id, strerror(rte_errno));
1470 	mlx5_os_read_dev_stat(priv, "out_of_buffer", &stats_ctrl->imissed_base);
1471 	stats_ctrl->imissed = 0;
1472 free:
1473 	mlx5_free(strings);
1474 }
1475 
1476 /**
1477  * Get MAC address by querying netdevice.
1478  *
1479  * @param[in] dev
1480  *   Pointer to Ethernet device.
1481  * @param[out] mac
1482  *   MAC address output buffer.
1483  *
1484  * @return
1485  *   0 on success, a negative errno value otherwise and rte_errno is set.
1486  */
1487 int
1488 mlx5_get_mac(struct rte_eth_dev *dev, uint8_t (*mac)[RTE_ETHER_ADDR_LEN])
1489 {
1490 	struct ifreq request;
1491 	int ret;
1492 
1493 	ret = mlx5_ifreq(dev, SIOCGIFHWADDR, &request);
1494 	if (ret)
1495 		return ret;
1496 	memcpy(mac, request.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1497 	return 0;
1498 }
1499 
1500