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 #include <dlfcn.h>
27 
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 	if (priv->bond_ifindex > 0) {
155 		memcpy(ifname, priv->bond_name, IF_NAMESIZE);
156 		return 0;
157 	}
158 	ifindex = mlx5_ifindex(dev);
159 	if (!ifindex) {
160 		if (!priv->representor)
161 			return mlx5_get_ifname_sysfs(priv->sh->ibdev_path,
162 						     *ifname);
163 		rte_errno = ENXIO;
164 		return -rte_errno;
165 	}
166 	if (if_indextoname(ifindex, &(*ifname)[0]))
167 		return 0;
168 	rte_errno = errno;
169 	return -rte_errno;
170 }
171 
172 /**
173  * Perform ifreq ioctl() on associated Ethernet device.
174  *
175  * @param[in] dev
176  *   Pointer to Ethernet device.
177  * @param req
178  *   Request number to pass to ioctl().
179  * @param[out] ifr
180  *   Interface request structure output buffer.
181  *
182  * @return
183  *   0 on success, a negative errno value otherwise and rte_errno is set.
184  */
185 static int
186 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
187 {
188 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
189 	int ret = 0;
190 
191 	if (sock == -1) {
192 		rte_errno = errno;
193 		return -rte_errno;
194 	}
195 	ret = mlx5_get_ifname(dev, &ifr->ifr_name);
196 	if (ret)
197 		goto error;
198 	ret = ioctl(sock, req, ifr);
199 	if (ret == -1) {
200 		rte_errno = errno;
201 		goto error;
202 	}
203 	close(sock);
204 	return 0;
205 error:
206 	close(sock);
207 	return -rte_errno;
208 }
209 
210 /**
211  * Get device MTU.
212  *
213  * @param dev
214  *   Pointer to Ethernet device.
215  * @param[out] mtu
216  *   MTU value output buffer.
217  *
218  * @return
219  *   0 on success, a negative errno value otherwise and rte_errno is set.
220  */
221 int
222 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
223 {
224 	struct ifreq request;
225 	int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
226 
227 	if (ret)
228 		return ret;
229 	*mtu = request.ifr_mtu;
230 	return 0;
231 }
232 
233 /**
234  * Set device MTU.
235  *
236  * @param dev
237  *   Pointer to Ethernet device.
238  * @param mtu
239  *   MTU value to set.
240  *
241  * @return
242  *   0 on success, a negative errno value otherwise and rte_errno is set.
243  */
244 int
245 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
246 {
247 	struct ifreq request = { .ifr_mtu = mtu, };
248 
249 	return mlx5_ifreq(dev, SIOCSIFMTU, &request);
250 }
251 
252 /**
253  * Set device flags.
254  *
255  * @param dev
256  *   Pointer to Ethernet device.
257  * @param keep
258  *   Bitmask for flags that must remain untouched.
259  * @param flags
260  *   Bitmask for flags to modify.
261  *
262  * @return
263  *   0 on success, a negative errno value otherwise and rte_errno is set.
264  */
265 static int
266 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
267 {
268 	struct ifreq request;
269 	int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
270 
271 	if (ret)
272 		return ret;
273 	request.ifr_flags &= keep;
274 	request.ifr_flags |= flags & ~keep;
275 	return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
276 }
277 
278 /**
279  * Get device current raw clock counter
280  *
281  * @param dev
282  *   Pointer to Ethernet device structure.
283  * @param[out] time
284  *   Current raw clock counter of the device.
285  *
286  * @return
287  *   0 if the clock has correctly been read
288  *   The value of errno in case of error
289  */
290 int
291 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
292 {
293 	struct mlx5_priv *priv = dev->data->dev_private;
294 	struct ibv_context *ctx = priv->sh->ctx;
295 	struct ibv_values_ex values;
296 	int err = 0;
297 
298 	values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
299 	err = mlx5_glue->query_rt_values_ex(ctx, &values);
300 	if (err != 0) {
301 		DRV_LOG(WARNING, "Could not query the clock !");
302 		return err;
303 	}
304 	*clock = values.raw_clock.tv_nsec;
305 	return 0;
306 }
307 
308 /**
309  * Retrieve the master device for representor in the same switch domain.
310  *
311  * @param dev
312  *   Pointer to representor Ethernet device structure.
313  *
314  * @return
315  *   Master device structure  on success, NULL otherwise.
316  */
317 static struct rte_eth_dev *
318 mlx5_find_master_dev(struct rte_eth_dev *dev)
319 {
320 	struct mlx5_priv *priv;
321 	uint16_t port_id;
322 	uint16_t domain_id;
323 
324 	priv = dev->data->dev_private;
325 	domain_id = priv->domain_id;
326 	MLX5_ASSERT(priv->representor);
327 	MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
328 		struct mlx5_priv *opriv =
329 			rte_eth_devices[port_id].data->dev_private;
330 		if (opriv &&
331 		    opriv->master &&
332 		    opriv->domain_id == domain_id &&
333 		    opriv->sh == priv->sh)
334 			return &rte_eth_devices[port_id];
335 	}
336 	return NULL;
337 }
338 
339 /**
340  * DPDK callback to retrieve physical link information.
341  *
342  * @param dev
343  *   Pointer to Ethernet device structure.
344  * @param[out] link
345  *   Storage for current link status.
346  *
347  * @return
348  *   0 on success, a negative errno value otherwise and rte_errno is set.
349  */
350 static int
351 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
352 			       struct rte_eth_link *link)
353 {
354 	struct mlx5_priv *priv = dev->data->dev_private;
355 	struct ethtool_cmd edata = {
356 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
357 	};
358 	struct ifreq ifr;
359 	struct rte_eth_link dev_link;
360 	int link_speed = 0;
361 	int ret;
362 
363 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
364 	if (ret) {
365 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
366 			dev->data->port_id, strerror(rte_errno));
367 		return ret;
368 	}
369 	dev_link = (struct rte_eth_link) {
370 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
371 				(ifr.ifr_flags & IFF_RUNNING)),
372 	};
373 	ifr = (struct ifreq) {
374 		.ifr_data = (void *)&edata,
375 	};
376 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
377 	if (ret) {
378 		if (ret == -ENOTSUP && priv->representor) {
379 			struct rte_eth_dev *master;
380 
381 			/*
382 			 * For representors we can try to inherit link
383 			 * settings from the master device. Actually
384 			 * link settings do not make a lot of sense
385 			 * for representors due to missing physical
386 			 * link. The old kernel drivers supported
387 			 * emulated settings query for representors,
388 			 * the new ones do not, so we have to add
389 			 * this code for compatibility issues.
390 			 */
391 			master = mlx5_find_master_dev(dev);
392 			if (master) {
393 				ifr = (struct ifreq) {
394 					.ifr_data = (void *)&edata,
395 				};
396 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
397 			}
398 		}
399 		if (ret) {
400 			DRV_LOG(WARNING,
401 				"port %u ioctl(SIOCETHTOOL,"
402 				" ETHTOOL_GSET) failed: %s",
403 				dev->data->port_id, strerror(rte_errno));
404 			return ret;
405 		}
406 	}
407 	link_speed = ethtool_cmd_speed(&edata);
408 	if (link_speed == -1)
409 		dev_link.link_speed = ETH_SPEED_NUM_UNKNOWN;
410 	else
411 		dev_link.link_speed = link_speed;
412 	priv->link_speed_capa = 0;
413 	if (edata.supported & SUPPORTED_Autoneg)
414 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
415 	if (edata.supported & (SUPPORTED_1000baseT_Full |
416 			       SUPPORTED_1000baseKX_Full))
417 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
418 	if (edata.supported & SUPPORTED_10000baseKR_Full)
419 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
420 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
421 			       SUPPORTED_40000baseCR4_Full |
422 			       SUPPORTED_40000baseSR4_Full |
423 			       SUPPORTED_40000baseLR4_Full))
424 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
425 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
426 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
427 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
428 			ETH_LINK_SPEED_FIXED);
429 	*link = dev_link;
430 	return 0;
431 }
432 
433 /**
434  * Retrieve physical link information (unlocked version using new ioctl).
435  *
436  * @param dev
437  *   Pointer to Ethernet device structure.
438  * @param[out] link
439  *   Storage for current link status.
440  *
441  * @return
442  *   0 on success, a negative errno value otherwise and rte_errno is set.
443  */
444 static int
445 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
446 			     struct rte_eth_link *link)
447 
448 {
449 	struct mlx5_priv *priv = dev->data->dev_private;
450 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
451 	struct ifreq ifr;
452 	struct rte_eth_link dev_link;
453 	struct rte_eth_dev *master = NULL;
454 	uint64_t sc;
455 	int ret;
456 
457 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
458 	if (ret) {
459 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
460 			dev->data->port_id, strerror(rte_errno));
461 		return ret;
462 	}
463 	dev_link = (struct rte_eth_link) {
464 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
465 				(ifr.ifr_flags & IFF_RUNNING)),
466 	};
467 	ifr = (struct ifreq) {
468 		.ifr_data = (void *)&gcmd,
469 	};
470 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
471 	if (ret) {
472 		if (ret == -ENOTSUP && priv->representor) {
473 			/*
474 			 * For representors we can try to inherit link
475 			 * settings from the master device. Actually
476 			 * link settings do not make a lot of sense
477 			 * for representors due to missing physical
478 			 * link. The old kernel drivers supported
479 			 * emulated settings query for representors,
480 			 * the new ones do not, so we have to add
481 			 * this code for compatibility issues.
482 			 */
483 			master = mlx5_find_master_dev(dev);
484 			if (master) {
485 				ifr = (struct ifreq) {
486 					.ifr_data = (void *)&gcmd,
487 				};
488 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
489 			}
490 		}
491 		if (ret) {
492 			DRV_LOG(DEBUG,
493 				"port %u ioctl(SIOCETHTOOL,"
494 				" ETHTOOL_GLINKSETTINGS) failed: %s",
495 				dev->data->port_id, strerror(rte_errno));
496 			return ret;
497 		}
498 	}
499 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
500 
501 	alignas(struct ethtool_link_settings)
502 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
503 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
504 	struct ethtool_link_settings *ecmd = (void *)data;
505 
506 	*ecmd = gcmd;
507 	ifr.ifr_data = (void *)ecmd;
508 	ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
509 	if (ret) {
510 		DRV_LOG(DEBUG,
511 			"port %u ioctl(SIOCETHTOOL,"
512 			"ETHTOOL_GLINKSETTINGS) failed: %s",
513 			dev->data->port_id, strerror(rte_errno));
514 		return ret;
515 	}
516 	dev_link.link_speed = (ecmd->speed == UINT32_MAX) ?
517 				ETH_SPEED_NUM_UNKNOWN : ecmd->speed;
518 	sc = ecmd->link_mode_masks[0] |
519 		((uint64_t)ecmd->link_mode_masks[1] << 32);
520 	priv->link_speed_capa = 0;
521 	if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
522 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
523 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
524 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
525 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
526 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
527 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
528 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
529 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
530 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
531 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
532 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
533 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
534 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
535 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
536 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
537 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
538 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
539 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
540 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
541 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
542 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
543 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
544 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
545 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
546 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
547 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
548 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
549 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
550 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
551 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
552 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
553 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
554 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
555 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT) |
556 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT)))
557 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
558 
559 	sc = ecmd->link_mode_masks[2] |
560 		((uint64_t)ecmd->link_mode_masks[3] << 32);
561 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT) |
562 		  MLX5_BITSHIFT
563 		       (ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT) |
564 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT)))
565 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
566 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
567 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
568 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
569 				  ETH_LINK_SPEED_FIXED);
570 	*link = dev_link;
571 	return 0;
572 }
573 
574 /**
575  * DPDK callback to retrieve physical link information.
576  *
577  * @param dev
578  *   Pointer to Ethernet device structure.
579  * @param wait_to_complete
580  *   Wait for request completion.
581  *
582  * @return
583  *   0 if link status was not updated, positive if it was, a negative errno
584  *   value otherwise and rte_errno is set.
585  */
586 int
587 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
588 {
589 	int ret;
590 	struct rte_eth_link dev_link;
591 	time_t start_time = time(NULL);
592 	int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT;
593 
594 	do {
595 		ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
596 		if (ret == -ENOTSUP)
597 			ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
598 		if (ret == 0)
599 			break;
600 		/* Handle wait to complete situation. */
601 		if ((wait_to_complete || retry) && ret == -EAGAIN) {
602 			if (abs((int)difftime(time(NULL), start_time)) <
603 			    MLX5_LINK_STATUS_TIMEOUT) {
604 				usleep(0);
605 				continue;
606 			} else {
607 				rte_errno = EBUSY;
608 				return -rte_errno;
609 			}
610 		} else if (ret < 0) {
611 			return ret;
612 		}
613 	} while (wait_to_complete || retry-- > 0);
614 	ret = !!memcmp(&dev->data->dev_link, &dev_link,
615 		       sizeof(struct rte_eth_link));
616 	dev->data->dev_link = dev_link;
617 	return ret;
618 }
619 
620 /**
621  * DPDK callback to get flow control status.
622  *
623  * @param dev
624  *   Pointer to Ethernet device structure.
625  * @param[out] fc_conf
626  *   Flow control output buffer.
627  *
628  * @return
629  *   0 on success, a negative errno value otherwise and rte_errno is set.
630  */
631 int
632 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
633 {
634 	struct ifreq ifr;
635 	struct ethtool_pauseparam ethpause = {
636 		.cmd = ETHTOOL_GPAUSEPARAM
637 	};
638 	int ret;
639 
640 	ifr.ifr_data = (void *)&ethpause;
641 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
642 	if (ret) {
643 		DRV_LOG(WARNING,
644 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
645 			" %s",
646 			dev->data->port_id, strerror(rte_errno));
647 		return ret;
648 	}
649 	fc_conf->autoneg = ethpause.autoneg;
650 	if (ethpause.rx_pause && ethpause.tx_pause)
651 		fc_conf->mode = RTE_FC_FULL;
652 	else if (ethpause.rx_pause)
653 		fc_conf->mode = RTE_FC_RX_PAUSE;
654 	else if (ethpause.tx_pause)
655 		fc_conf->mode = RTE_FC_TX_PAUSE;
656 	else
657 		fc_conf->mode = RTE_FC_NONE;
658 	return 0;
659 }
660 
661 /**
662  * DPDK callback to modify flow control parameters.
663  *
664  * @param dev
665  *   Pointer to Ethernet device structure.
666  * @param[in] fc_conf
667  *   Flow control parameters.
668  *
669  * @return
670  *   0 on success, a negative errno value otherwise and rte_errno is set.
671  */
672 int
673 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
674 {
675 	struct ifreq ifr;
676 	struct ethtool_pauseparam ethpause = {
677 		.cmd = ETHTOOL_SPAUSEPARAM
678 	};
679 	int ret;
680 
681 	ifr.ifr_data = (void *)&ethpause;
682 	ethpause.autoneg = fc_conf->autoneg;
683 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
684 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
685 		ethpause.rx_pause = 1;
686 	else
687 		ethpause.rx_pause = 0;
688 
689 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
690 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
691 		ethpause.tx_pause = 1;
692 	else
693 		ethpause.tx_pause = 0;
694 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
695 	if (ret) {
696 		DRV_LOG(WARNING,
697 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
698 			" failed: %s",
699 			dev->data->port_id, strerror(rte_errno));
700 		return ret;
701 	}
702 	return 0;
703 }
704 
705 /**
706  * Handle asynchronous removal event for entire multiport device.
707  *
708  * @param sh
709  *   Infiniband device shared context.
710  */
711 static void
712 mlx5_dev_interrupt_device_fatal(struct mlx5_dev_ctx_shared *sh)
713 {
714 	uint32_t i;
715 
716 	for (i = 0; i < sh->max_port; ++i) {
717 		struct rte_eth_dev *dev;
718 
719 		if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
720 			/*
721 			 * Or not existing port either no
722 			 * handler installed for this port.
723 			 */
724 			continue;
725 		}
726 		dev = &rte_eth_devices[sh->port[i].ih_port_id];
727 		MLX5_ASSERT(dev);
728 		if (dev->data->dev_conf.intr_conf.rmv)
729 			rte_eth_dev_callback_process
730 				(dev, RTE_ETH_EVENT_INTR_RMV, NULL);
731 	}
732 }
733 
734 /**
735  * Handle shared asynchronous events the NIC (removal event
736  * and link status change). Supports multiport IB device.
737  *
738  * @param cb_arg
739  *   Callback argument.
740  */
741 void
742 mlx5_dev_interrupt_handler(void *cb_arg)
743 {
744 	struct mlx5_dev_ctx_shared *sh = cb_arg;
745 	struct ibv_async_event event;
746 
747 	/* Read all message from the IB device and acknowledge them. */
748 	for (;;) {
749 		struct rte_eth_dev *dev;
750 		uint32_t tmp;
751 
752 		if (mlx5_glue->get_async_event(sh->ctx, &event))
753 			break;
754 		/* Retrieve and check IB port index. */
755 		tmp = (uint32_t)event.element.port_num;
756 		if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
757 			/*
758 			 * The DEVICE_FATAL event is called once for
759 			 * entire device without port specifying.
760 			 * We should notify all existing ports.
761 			 */
762 			mlx5_glue->ack_async_event(&event);
763 			mlx5_dev_interrupt_device_fatal(sh);
764 			continue;
765 		}
766 		MLX5_ASSERT(tmp && (tmp <= sh->max_port));
767 		if (!tmp) {
768 			/* Unsupported device level event. */
769 			mlx5_glue->ack_async_event(&event);
770 			DRV_LOG(DEBUG,
771 				"unsupported common event (type %d)",
772 				event.event_type);
773 			continue;
774 		}
775 		if (tmp > sh->max_port) {
776 			/* Invalid IB port index. */
777 			mlx5_glue->ack_async_event(&event);
778 			DRV_LOG(DEBUG,
779 				"cannot handle an event (type %d)"
780 				"due to invalid IB port index (%u)",
781 				event.event_type, tmp);
782 			continue;
783 		}
784 		if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
785 			/* No handler installed. */
786 			mlx5_glue->ack_async_event(&event);
787 			DRV_LOG(DEBUG,
788 				"cannot handle an event (type %d)"
789 				"due to no handler installed for port %u",
790 				event.event_type, tmp);
791 			continue;
792 		}
793 		/* Retrieve ethernet device descriptor. */
794 		tmp = sh->port[tmp - 1].ih_port_id;
795 		dev = &rte_eth_devices[tmp];
796 		MLX5_ASSERT(dev);
797 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
798 		     event.event_type == IBV_EVENT_PORT_ERR) &&
799 			dev->data->dev_conf.intr_conf.lsc) {
800 			mlx5_glue->ack_async_event(&event);
801 			if (mlx5_link_update(dev, 0) == -EAGAIN) {
802 				usleep(0);
803 				continue;
804 			}
805 			rte_eth_dev_callback_process
806 				(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
807 			continue;
808 		}
809 		DRV_LOG(DEBUG,
810 			"port %u cannot handle an unknown event (type %d)",
811 			dev->data->port_id, event.event_type);
812 		mlx5_glue->ack_async_event(&event);
813 	}
814 }
815 
816 /*
817  * Unregister callback handler safely. The handler may be active
818  * while we are trying to unregister it, in this case code -EAGAIN
819  * is returned by rte_intr_callback_unregister(). This routine checks
820  * the return code and tries to unregister handler again.
821  *
822  * @param handle
823  *   interrupt handle
824  * @param cb_fn
825  *   pointer to callback routine
826  * @cb_arg
827  *   opaque callback parameter
828  */
829 void
830 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
831 			      rte_intr_callback_fn cb_fn, void *cb_arg)
832 {
833 	/*
834 	 * Try to reduce timeout management overhead by not calling
835 	 * the timer related routines on the first iteration. If the
836 	 * unregistering succeeds on first call there will be no
837 	 * timer calls at all.
838 	 */
839 	uint64_t twait = 0;
840 	uint64_t start = 0;
841 
842 	do {
843 		int ret;
844 
845 		ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
846 		if (ret >= 0)
847 			return;
848 		if (ret != -EAGAIN) {
849 			DRV_LOG(INFO, "failed to unregister interrupt"
850 				      " handler (error: %d)", ret);
851 			MLX5_ASSERT(false);
852 			return;
853 		}
854 		if (twait) {
855 			struct timespec onems;
856 
857 			/* Wait one millisecond and try again. */
858 			onems.tv_sec = 0;
859 			onems.tv_nsec = NS_PER_S / MS_PER_S;
860 			nanosleep(&onems, 0);
861 			/* Check whether one second elapsed. */
862 			if ((rte_get_timer_cycles() - start) <= twait)
863 				continue;
864 		} else {
865 			/*
866 			 * We get the amount of timer ticks for one second.
867 			 * If this amount elapsed it means we spent one
868 			 * second in waiting. This branch is executed once
869 			 * on first iteration.
870 			 */
871 			twait = rte_get_timer_hz();
872 			MLX5_ASSERT(twait);
873 		}
874 		/*
875 		 * Timeout elapsed, show message (once a second) and retry.
876 		 * We have no other acceptable option here, if we ignore
877 		 * the unregistering return code the handler will not
878 		 * be unregistered, fd will be closed and we may get the
879 		 * crush. Hanging and messaging in the loop seems not to be
880 		 * the worst choice.
881 		 */
882 		DRV_LOG(INFO, "Retrying to unregister interrupt handler");
883 		start = rte_get_timer_cycles();
884 	} while (true);
885 }
886 
887 /**
888  * Handle DEVX interrupts from the NIC.
889  * This function is probably called from the DPDK host thread.
890  *
891  * @param cb_arg
892  *   Callback argument.
893  */
894 void
895 mlx5_dev_interrupt_handler_devx(void *cb_arg)
896 {
897 #ifndef HAVE_IBV_DEVX_ASYNC
898 	(void)cb_arg;
899 	return;
900 #else
901 	struct mlx5_dev_ctx_shared *sh = cb_arg;
902 	union {
903 		struct mlx5dv_devx_async_cmd_hdr cmd_resp;
904 		uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
905 			    MLX5_ST_SZ_BYTES(traffic_counter) +
906 			    sizeof(struct mlx5dv_devx_async_cmd_hdr)];
907 	} out;
908 	uint8_t *buf = out.buf + sizeof(out.cmd_resp);
909 
910 	while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
911 						   &out.cmd_resp,
912 						   sizeof(out.buf)))
913 		mlx5_flow_async_pool_query_handle
914 			(sh, (uint64_t)out.cmd_resp.wr_id,
915 			 mlx5_devx_get_out_command_status(buf));
916 #endif /* HAVE_IBV_DEVX_ASYNC */
917 }
918 
919 /**
920  * DPDK callback to bring the link DOWN.
921  *
922  * @param dev
923  *   Pointer to Ethernet device structure.
924  *
925  * @return
926  *   0 on success, a negative errno value otherwise and rte_errno is set.
927  */
928 int
929 mlx5_set_link_down(struct rte_eth_dev *dev)
930 {
931 	return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
932 }
933 
934 /**
935  * DPDK callback to bring the link UP.
936  *
937  * @param dev
938  *   Pointer to Ethernet device structure.
939  *
940  * @return
941  *   0 on success, a negative errno value otherwise and rte_errno is set.
942  */
943 int
944 mlx5_set_link_up(struct rte_eth_dev *dev)
945 {
946 	return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
947 }
948 
949 /**
950  * Check if mlx5 device was removed.
951  *
952  * @param dev
953  *   Pointer to Ethernet device structure.
954  *
955  * @return
956  *   1 when device is removed, otherwise 0.
957  */
958 int
959 mlx5_is_removed(struct rte_eth_dev *dev)
960 {
961 	struct ibv_device_attr device_attr;
962 	struct mlx5_priv *priv = dev->data->dev_private;
963 
964 	if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
965 		return 1;
966 	return 0;
967 }
968 
969 /**
970  * Analyze gathered port parameters via sysfs to recognize master
971  * and representor devices for E-Switch configuration.
972  *
973  * @param[in] device_dir
974  *   flag of presence of "device" directory under port device key.
975  * @param[inout] switch_info
976  *   Port information, including port name as a number and port name
977  *   type if recognized
978  *
979  * @return
980  *   master and representor flags are set in switch_info according to
981  *   recognized parameters (if any).
982  */
983 static void
984 mlx5_sysfs_check_switch_info(bool device_dir,
985 			     struct mlx5_switch_info *switch_info)
986 {
987 	switch (switch_info->name_type) {
988 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
989 		/*
990 		 * Name is not recognized, assume the master,
991 		 * check the device directory presence.
992 		 */
993 		switch_info->master = device_dir;
994 		break;
995 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
996 		/*
997 		 * Name is not set, this assumes the legacy naming
998 		 * schema for master, just check if there is
999 		 * a device directory.
1000 		 */
1001 		switch_info->master = device_dir;
1002 		break;
1003 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1004 		/* New uplink naming schema recognized. */
1005 		switch_info->master = 1;
1006 		break;
1007 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1008 		/* Legacy representors naming schema. */
1009 		switch_info->representor = !device_dir;
1010 		break;
1011 	case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
1012 		/* Fallthrough */
1013 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1014 		/* New representors naming schema. */
1015 		switch_info->representor = 1;
1016 		break;
1017 	}
1018 }
1019 
1020 /**
1021  * Get switch information associated with network interface.
1022  *
1023  * @param ifindex
1024  *   Network interface index.
1025  * @param[out] info
1026  *   Switch information object, populated in case of success.
1027  *
1028  * @return
1029  *   0 on success, a negative errno value otherwise and rte_errno is set.
1030  */
1031 static int (*real_if_indextoname)(unsigned int, char *);
1032 int
1033 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1034 {
1035 	char ifname[IF_NAMESIZE];
1036 	char port_name[IF_NAMESIZE];
1037 	FILE *file;
1038 	struct mlx5_switch_info data = {
1039 		.master = 0,
1040 		.representor = 0,
1041 		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1042 		.port_name = 0,
1043 		.switch_id = 0,
1044 	};
1045 	DIR *dir;
1046 	bool port_switch_id_set = false;
1047 	bool device_dir = false;
1048 	char c;
1049 	int ret;
1050 
1051 	// for ff tools
1052 	if (!real_if_indextoname) {
1053 		real_if_indextoname = dlsym(RTLD_NEXT, "if_indextoname");
1054 		if (!real_if_indextoname) {
1055 			rte_errno = errno;
1056 			return -rte_errno;
1057 		}
1058 	}
1059 
1060 	if (!real_if_indextoname(ifindex, ifname)) {
1061 		rte_errno = errno;
1062 		return -rte_errno;
1063 	}
1064 
1065 	MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1066 	      ifname);
1067 	MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1068 	      ifname);
1069 	MKSTR(pci_device, "/sys/class/net/%s/device",
1070 	      ifname);
1071 
1072 	file = fopen(phys_port_name, "rb");
1073 	if (file != NULL) {
1074 		ret = fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", port_name);
1075 		fclose(file);
1076 		if (ret == 1)
1077 			mlx5_translate_port_name(port_name, &data);
1078 	}
1079 	file = fopen(phys_switch_id, "rb");
1080 	if (file == NULL) {
1081 		rte_errno = errno;
1082 		return -rte_errno;
1083 	}
1084 	port_switch_id_set =
1085 		fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1086 		c == '\n';
1087 	fclose(file);
1088 	dir = opendir(pci_device);
1089 	if (dir != NULL) {
1090 		closedir(dir);
1091 		device_dir = true;
1092 	}
1093 	if (port_switch_id_set) {
1094 		/* We have some E-Switch configuration. */
1095 		mlx5_sysfs_check_switch_info(device_dir, &data);
1096 	}
1097 	*info = data;
1098 	MLX5_ASSERT(!(data.master && data.representor));
1099 	if (data.master && data.representor) {
1100 		DRV_LOG(ERR, "ifindex %u device is recognized as master"
1101 			     " and as representor", ifindex);
1102 		rte_errno = ENODEV;
1103 		return -rte_errno;
1104 	}
1105 	return 0;
1106 }
1107 
1108 /**
1109  * Get bond information associated with network interface.
1110  *
1111  * @param pf_ifindex
1112  *   Network interface index of bond slave interface
1113  * @param[out] ifindex
1114  *   Pointer to bond ifindex.
1115  * @param[out] ifname
1116  *   Pointer to bond ifname.
1117  *
1118  * @return
1119  *   0 on success, a negative errno value otherwise and rte_errno is set.
1120  */
1121 int
1122 mlx5_sysfs_bond_info(unsigned int pf_ifindex, unsigned int *ifindex,
1123 		     char *ifname)
1124 {
1125 	char name[IF_NAMESIZE];
1126 	FILE *file;
1127 	unsigned int index;
1128 	int ret;
1129 
1130 	if (!if_indextoname(pf_ifindex, name) || !strlen(name)) {
1131 		rte_errno = errno;
1132 		return -rte_errno;
1133 	}
1134 	MKSTR(bond_if, "/sys/class/net/%s/master/ifindex", name);
1135 	/* read bond ifindex */
1136 	file = fopen(bond_if, "rb");
1137 	if (file == NULL) {
1138 		rte_errno = errno;
1139 		return -rte_errno;
1140 	}
1141 	ret = fscanf(file, "%u", &index);
1142 	fclose(file);
1143 	if (ret <= 0) {
1144 		rte_errno = errno;
1145 		return -rte_errno;
1146 	}
1147 	if (ifindex)
1148 		*ifindex = index;
1149 
1150 	/* read bond device name from symbol link */
1151 	if (ifname) {
1152 		if (!if_indextoname(index, ifname)) {
1153 			rte_errno = errno;
1154 			return -rte_errno;
1155 		}
1156 	}
1157 	return 0;
1158 }
1159 
1160 /**
1161  * DPDK callback to retrieve plug-in module EEPROM information (type and size).
1162  *
1163  * @param dev
1164  *   Pointer to Ethernet device structure.
1165  * @param[out] modinfo
1166  *   Storage for plug-in module EEPROM information.
1167  *
1168  * @return
1169  *   0 on success, a negative errno value otherwise and rte_errno is set.
1170  */
1171 int
1172 mlx5_get_module_info(struct rte_eth_dev *dev,
1173 		     struct rte_eth_dev_module_info *modinfo)
1174 {
1175 	struct ethtool_modinfo info = {
1176 		.cmd = ETHTOOL_GMODULEINFO,
1177 	};
1178 	struct ifreq ifr = (struct ifreq) {
1179 		.ifr_data = (void *)&info,
1180 	};
1181 	int ret = 0;
1182 
1183 	if (!dev || !modinfo) {
1184 		DRV_LOG(WARNING, "missing argument, cannot get module info");
1185 		rte_errno = EINVAL;
1186 		return -rte_errno;
1187 	}
1188 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1189 	if (ret) {
1190 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1191 			dev->data->port_id, strerror(rte_errno));
1192 		return ret;
1193 	}
1194 	modinfo->type = info.type;
1195 	modinfo->eeprom_len = info.eeprom_len;
1196 	return ret;
1197 }
1198 
1199 /**
1200  * DPDK callback to retrieve plug-in module EEPROM data.
1201  *
1202  * @param dev
1203  *   Pointer to Ethernet device structure.
1204  * @param[out] info
1205  *   Storage for plug-in module EEPROM data.
1206  *
1207  * @return
1208  *   0 on success, a negative errno value otherwise and rte_errno is set.
1209  */
1210 int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
1211 			   struct rte_dev_eeprom_info *info)
1212 {
1213 	struct ethtool_eeprom *eeprom;
1214 	struct ifreq ifr;
1215 	int ret = 0;
1216 
1217 	if (!dev || !info) {
1218 		DRV_LOG(WARNING, "missing argument, cannot get module eeprom");
1219 		rte_errno = EINVAL;
1220 		return -rte_errno;
1221 	}
1222 	eeprom = mlx5_malloc(MLX5_MEM_ZERO,
1223 			     (sizeof(struct ethtool_eeprom) + info->length), 0,
1224 			     SOCKET_ID_ANY);
1225 	if (!eeprom) {
1226 		DRV_LOG(WARNING, "port %u cannot allocate memory for "
1227 			"eeprom data", dev->data->port_id);
1228 		rte_errno = ENOMEM;
1229 		return -rte_errno;
1230 	}
1231 	eeprom->cmd = ETHTOOL_GMODULEEEPROM;
1232 	eeprom->offset = info->offset;
1233 	eeprom->len = info->length;
1234 	ifr = (struct ifreq) {
1235 		.ifr_data = (void *)eeprom,
1236 	};
1237 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1238 	if (ret)
1239 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1240 			dev->data->port_id, strerror(rte_errno));
1241 	else
1242 		rte_memcpy(info->data, eeprom->data, info->length);
1243 	mlx5_free(eeprom);
1244 	return ret;
1245 }
1246 
1247 /**
1248  * Read device counters table.
1249  *
1250  * @param dev
1251  *   Pointer to Ethernet device.
1252  * @param[out] stats
1253  *   Counters table output buffer.
1254  *
1255  * @return
1256  *   0 on success and stats is filled, negative errno value otherwise and
1257  *   rte_errno is set.
1258  */
1259 int
1260 mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
1261 {
1262 	struct mlx5_priv *priv = dev->data->dev_private;
1263 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1264 	unsigned int i;
1265 	struct ifreq ifr;
1266 	unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t);
1267 	unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz];
1268 	struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf;
1269 	int ret;
1270 
1271 	et_stats->cmd = ETHTOOL_GSTATS;
1272 	et_stats->n_stats = xstats_ctrl->stats_n;
1273 	ifr.ifr_data = (caddr_t)et_stats;
1274 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1275 	if (ret) {
1276 		DRV_LOG(WARNING,
1277 			"port %u unable to read statistic values from device",
1278 			dev->data->port_id);
1279 		return ret;
1280 	}
1281 	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
1282 		if (xstats_ctrl->info[i].dev) {
1283 			ret = mlx5_os_read_dev_stat(priv,
1284 					    xstats_ctrl->info[i].ctr_name,
1285 					    &stats[i]);
1286 			/* return last xstats counter if fail to read. */
1287 			if (ret == 0)
1288 				xstats_ctrl->xstats[i] = stats[i];
1289 			else
1290 				stats[i] = xstats_ctrl->xstats[i];
1291 		} else {
1292 			stats[i] = (uint64_t)
1293 				et_stats->data[xstats_ctrl->dev_table_idx[i]];
1294 		}
1295 	}
1296 	return 0;
1297 }
1298 
1299 /**
1300  * Query the number of statistics provided by ETHTOOL.
1301  *
1302  * @param dev
1303  *   Pointer to Ethernet device.
1304  *
1305  * @return
1306  *   Number of statistics on success, negative errno value otherwise and
1307  *   rte_errno is set.
1308  */
1309 int
1310 mlx5_os_get_stats_n(struct rte_eth_dev *dev)
1311 {
1312 	struct ethtool_drvinfo drvinfo;
1313 	struct ifreq ifr;
1314 	int ret;
1315 
1316 	drvinfo.cmd = ETHTOOL_GDRVINFO;
1317 	ifr.ifr_data = (caddr_t)&drvinfo;
1318 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1319 	if (ret) {
1320 		DRV_LOG(WARNING, "port %u unable to query number of statistics",
1321 			dev->data->port_id);
1322 		return ret;
1323 	}
1324 	return drvinfo.n_stats;
1325 }
1326 
1327 static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
1328 	{
1329 		.dpdk_name = "rx_unicast_bytes",
1330 		.ctr_name = "rx_vport_unicast_bytes",
1331 	},
1332 	{
1333 		.dpdk_name = "rx_multicast_bytes",
1334 		.ctr_name = "rx_vport_multicast_bytes",
1335 	},
1336 	{
1337 		.dpdk_name = "rx_broadcast_bytes",
1338 		.ctr_name = "rx_vport_broadcast_bytes",
1339 	},
1340 	{
1341 		.dpdk_name = "rx_unicast_packets",
1342 		.ctr_name = "rx_vport_unicast_packets",
1343 	},
1344 	{
1345 		.dpdk_name = "rx_multicast_packets",
1346 		.ctr_name = "rx_vport_multicast_packets",
1347 	},
1348 	{
1349 		.dpdk_name = "rx_broadcast_packets",
1350 		.ctr_name = "rx_vport_broadcast_packets",
1351 	},
1352 	{
1353 		.dpdk_name = "tx_unicast_bytes",
1354 		.ctr_name = "tx_vport_unicast_bytes",
1355 	},
1356 	{
1357 		.dpdk_name = "tx_multicast_bytes",
1358 		.ctr_name = "tx_vport_multicast_bytes",
1359 	},
1360 	{
1361 		.dpdk_name = "tx_broadcast_bytes",
1362 		.ctr_name = "tx_vport_broadcast_bytes",
1363 	},
1364 	{
1365 		.dpdk_name = "tx_unicast_packets",
1366 		.ctr_name = "tx_vport_unicast_packets",
1367 	},
1368 	{
1369 		.dpdk_name = "tx_multicast_packets",
1370 		.ctr_name = "tx_vport_multicast_packets",
1371 	},
1372 	{
1373 		.dpdk_name = "tx_broadcast_packets",
1374 		.ctr_name = "tx_vport_broadcast_packets",
1375 	},
1376 	{
1377 		.dpdk_name = "rx_wqe_errors",
1378 		.ctr_name = "rx_wqe_err",
1379 	},
1380 	{
1381 		.dpdk_name = "rx_phy_crc_errors",
1382 		.ctr_name = "rx_crc_errors_phy",
1383 	},
1384 	{
1385 		.dpdk_name = "rx_phy_in_range_len_errors",
1386 		.ctr_name = "rx_in_range_len_errors_phy",
1387 	},
1388 	{
1389 		.dpdk_name = "rx_phy_symbol_errors",
1390 		.ctr_name = "rx_symbol_err_phy",
1391 	},
1392 	{
1393 		.dpdk_name = "tx_phy_errors",
1394 		.ctr_name = "tx_errors_phy",
1395 	},
1396 	{
1397 		.dpdk_name = "rx_out_of_buffer",
1398 		.ctr_name = "out_of_buffer",
1399 		.dev = 1,
1400 	},
1401 	{
1402 		.dpdk_name = "tx_phy_packets",
1403 		.ctr_name = "tx_packets_phy",
1404 	},
1405 	{
1406 		.dpdk_name = "rx_phy_packets",
1407 		.ctr_name = "rx_packets_phy",
1408 	},
1409 	{
1410 		.dpdk_name = "tx_phy_discard_packets",
1411 		.ctr_name = "tx_discards_phy",
1412 	},
1413 	{
1414 		.dpdk_name = "rx_phy_discard_packets",
1415 		.ctr_name = "rx_discards_phy",
1416 	},
1417 	{
1418 		.dpdk_name = "tx_phy_bytes",
1419 		.ctr_name = "tx_bytes_phy",
1420 	},
1421 	{
1422 		.dpdk_name = "rx_phy_bytes",
1423 		.ctr_name = "rx_bytes_phy",
1424 	},
1425 	/* Representor only */
1426 	{
1427 		.dpdk_name = "rx_vport_packets",
1428 		.ctr_name = "vport_rx_packets",
1429 	},
1430 	{
1431 		.dpdk_name = "rx_vport_bytes",
1432 		.ctr_name = "vport_rx_bytes",
1433 	},
1434 	{
1435 		.dpdk_name = "tx_vport_packets",
1436 		.ctr_name = "vport_tx_packets",
1437 	},
1438 	{
1439 		.dpdk_name = "tx_vport_bytes",
1440 		.ctr_name = "vport_tx_bytes",
1441 	},
1442 };
1443 
1444 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init);
1445 
1446 /**
1447  * Init the structures to read device counters.
1448  *
1449  * @param dev
1450  *   Pointer to Ethernet device.
1451  */
1452 void
1453 mlx5_os_stats_init(struct rte_eth_dev *dev)
1454 {
1455 	struct mlx5_priv *priv = dev->data->dev_private;
1456 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1457 	struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl;
1458 	unsigned int i;
1459 	unsigned int j;
1460 	struct ifreq ifr;
1461 	struct ethtool_gstrings *strings = NULL;
1462 	unsigned int dev_stats_n;
1463 	unsigned int str_sz;
1464 	int ret;
1465 
1466 	/* So that it won't aggregate for each init. */
1467 	xstats_ctrl->mlx5_stats_n = 0;
1468 	ret = mlx5_os_get_stats_n(dev);
1469 	if (ret < 0) {
1470 		DRV_LOG(WARNING, "port %u no extended statistics available",
1471 			dev->data->port_id);
1472 		return;
1473 	}
1474 	dev_stats_n = ret;
1475 	/* Allocate memory to grab stat names and values. */
1476 	str_sz = dev_stats_n * ETH_GSTRING_LEN;
1477 	strings = (struct ethtool_gstrings *)
1478 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
1479 			      SOCKET_ID_ANY);
1480 	if (!strings) {
1481 		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats",
1482 		     dev->data->port_id);
1483 		return;
1484 	}
1485 	strings->cmd = ETHTOOL_GSTRINGS;
1486 	strings->string_set = ETH_SS_STATS;
1487 	strings->len = dev_stats_n;
1488 	ifr.ifr_data = (caddr_t)strings;
1489 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1490 	if (ret) {
1491 		DRV_LOG(WARNING, "port %u unable to get statistic names",
1492 			dev->data->port_id);
1493 		goto free;
1494 	}
1495 	for (i = 0; i != dev_stats_n; ++i) {
1496 		const char *curr_string = (const char *)
1497 			&strings->data[i * ETH_GSTRING_LEN];
1498 
1499 		for (j = 0; j != xstats_n; ++j) {
1500 			if (!strcmp(mlx5_counters_init[j].ctr_name,
1501 				    curr_string)) {
1502 				unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1503 
1504 				xstats_ctrl->dev_table_idx[idx] = i;
1505 				xstats_ctrl->info[idx] = mlx5_counters_init[j];
1506 				break;
1507 			}
1508 		}
1509 	}
1510 	/* Add dev counters. */
1511 	for (i = 0; i != xstats_n; ++i) {
1512 		if (mlx5_counters_init[i].dev) {
1513 			unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1514 
1515 			xstats_ctrl->info[idx] = mlx5_counters_init[i];
1516 			xstats_ctrl->hw_stats[idx] = 0;
1517 		}
1518 	}
1519 	MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS);
1520 	xstats_ctrl->stats_n = dev_stats_n;
1521 	/* Copy to base at first time. */
1522 	ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base);
1523 	if (ret)
1524 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
1525 			dev->data->port_id, strerror(rte_errno));
1526 	mlx5_os_read_dev_stat(priv, "out_of_buffer", &stats_ctrl->imissed_base);
1527 	stats_ctrl->imissed = 0;
1528 free:
1529 	mlx5_free(strings);
1530 }
1531 
1532 /**
1533  * Get MAC address by querying netdevice.
1534  *
1535  * @param[in] dev
1536  *   Pointer to Ethernet device.
1537  * @param[out] mac
1538  *   MAC address output buffer.
1539  *
1540  * @return
1541  *   0 on success, a negative errno value otherwise and rte_errno is set.
1542  */
1543 int
1544 mlx5_get_mac(struct rte_eth_dev *dev, uint8_t (*mac)[RTE_ETHER_ADDR_LEN])
1545 {
1546 	struct ifreq request;
1547 	int ret;
1548 
1549 	ret = mlx5_ifreq(dev, SIOCGIFHWADDR, &request);
1550 	if (ret)
1551 		return ret;
1552 	memcpy(mac, request.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1553 	return 0;
1554 }
1555 
1556