xref: /dpdk/drivers/net/mlx5/linux/mlx5_ethdev_os.c (revision d810252857c9c72bf6c03a917fcfff1ede6af765)
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 <ethdev_driver.h>
28 #include <rte_bus_pci.h>
29 #include <rte_mbuf.h>
30 #include <rte_common.h>
31 #include <rte_interrupts.h>
32 #include <rte_malloc.h>
33 #include <rte_string_fns.h>
34 #include <rte_rwlock.h>
35 #include <rte_cycles.h>
36 
37 #include <mlx5_glue.h>
38 #include <mlx5_devx_cmds.h>
39 #include <mlx5_common.h>
40 #include <mlx5_malloc.h>
41 
42 #include "mlx5.h"
43 #include "mlx5_rxtx.h"
44 #include "mlx5_utils.h"
45 
46 /* Supported speed values found in /usr/include/linux/ethtool.h */
47 #ifndef HAVE_SUPPORTED_40000baseKR4_Full
48 #define SUPPORTED_40000baseKR4_Full (1 << 23)
49 #endif
50 #ifndef HAVE_SUPPORTED_40000baseCR4_Full
51 #define SUPPORTED_40000baseCR4_Full (1 << 24)
52 #endif
53 #ifndef HAVE_SUPPORTED_40000baseSR4_Full
54 #define SUPPORTED_40000baseSR4_Full (1 << 25)
55 #endif
56 #ifndef HAVE_SUPPORTED_40000baseLR4_Full
57 #define SUPPORTED_40000baseLR4_Full (1 << 26)
58 #endif
59 #ifndef HAVE_SUPPORTED_56000baseKR4_Full
60 #define SUPPORTED_56000baseKR4_Full (1 << 27)
61 #endif
62 #ifndef HAVE_SUPPORTED_56000baseCR4_Full
63 #define SUPPORTED_56000baseCR4_Full (1 << 28)
64 #endif
65 #ifndef HAVE_SUPPORTED_56000baseSR4_Full
66 #define SUPPORTED_56000baseSR4_Full (1 << 29)
67 #endif
68 #ifndef HAVE_SUPPORTED_56000baseLR4_Full
69 #define SUPPORTED_56000baseLR4_Full (1 << 30)
70 #endif
71 
72 /* Add defines in case the running kernel is not the same as user headers. */
73 #ifndef ETHTOOL_GLINKSETTINGS
74 struct ethtool_link_settings {
75 	uint32_t cmd;
76 	uint32_t speed;
77 	uint8_t duplex;
78 	uint8_t port;
79 	uint8_t phy_address;
80 	uint8_t autoneg;
81 	uint8_t mdio_support;
82 	uint8_t eth_to_mdix;
83 	uint8_t eth_tp_mdix_ctrl;
84 	int8_t link_mode_masks_nwords;
85 	uint32_t reserved[8];
86 	uint32_t link_mode_masks[];
87 };
88 
89 /* The kernel values can be found in /include/uapi/linux/ethtool.h */
90 #define ETHTOOL_GLINKSETTINGS 0x0000004c
91 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
92 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
93 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
94 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
95 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
96 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
97 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
98 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
99 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
100 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
101 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
102 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
103 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
104 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
105 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
106 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
107 #endif
108 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
109 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
110 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
111 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
112 #endif
113 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
114 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
115 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
116 #endif
117 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
118 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
119 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
120 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
121 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
122 #endif
123 #ifndef HAVE_ETHTOOL_LINK_MODE_200G
124 #define ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT 62
125 #define ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT 63
126 #define ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT 0 /* 64 - 64 */
127 #define ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT 1 /* 65 - 64 */
128 #define ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT 2 /* 66 - 64 */
129 #endif
130 
131 
132 /**
133  * Get interface name from private structure.
134  *
135  * This is a port representor-aware version of mlx5_get_ifname_sysfs().
136  *
137  * @param[in] dev
138  *   Pointer to Ethernet device.
139  * @param[out] ifname
140  *   Interface name output buffer.
141  *
142  * @return
143  *   0 on success, a negative errno value otherwise and rte_errno is set.
144  */
145 int
146 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[MLX5_NAMESIZE])
147 {
148 	struct mlx5_priv *priv = dev->data->dev_private;
149 	unsigned int ifindex;
150 
151 	MLX5_ASSERT(priv);
152 	MLX5_ASSERT(priv->sh);
153 	if (priv->bond_ifindex > 0) {
154 		memcpy(ifname, priv->bond_name, MLX5_NAMESIZE);
155 		return 0;
156 	}
157 	ifindex = mlx5_ifindex(dev);
158 	if (!ifindex) {
159 		if (!priv->representor)
160 			return mlx5_get_ifname_sysfs(priv->sh->ibdev_path,
161 						     *ifname);
162 		rte_errno = ENXIO;
163 		return -rte_errno;
164 	}
165 	if (if_indextoname(ifindex, &(*ifname)[0]))
166 		return 0;
167 	rte_errno = errno;
168 	return -rte_errno;
169 }
170 
171 /**
172  * Perform ifreq ioctl() on associated Ethernet device.
173  *
174  * @param[in] dev
175  *   Pointer to Ethernet device.
176  * @param req
177  *   Request number to pass to ioctl().
178  * @param[out] ifr
179  *   Interface request structure output buffer.
180  *
181  * @return
182  *   0 on success, a negative errno value otherwise and rte_errno is set.
183  */
184 static int
185 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
186 {
187 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
188 	int ret = 0;
189 
190 	if (sock == -1) {
191 		rte_errno = errno;
192 		return -rte_errno;
193 	}
194 	ret = mlx5_get_ifname(dev, &ifr->ifr_name);
195 	if (ret)
196 		goto error;
197 	ret = ioctl(sock, req, ifr);
198 	if (ret == -1) {
199 		rte_errno = errno;
200 		goto error;
201 	}
202 	close(sock);
203 	return 0;
204 error:
205 	close(sock);
206 	return -rte_errno;
207 }
208 
209 /**
210  * Get device MTU.
211  *
212  * @param dev
213  *   Pointer to Ethernet device.
214  * @param[out] mtu
215  *   MTU value output buffer.
216  *
217  * @return
218  *   0 on success, a negative errno value otherwise and rte_errno is set.
219  */
220 int
221 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
222 {
223 	struct ifreq request;
224 	int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
225 
226 	if (ret)
227 		return ret;
228 	*mtu = request.ifr_mtu;
229 	return 0;
230 }
231 
232 /**
233  * Set device MTU.
234  *
235  * @param dev
236  *   Pointer to Ethernet device.
237  * @param mtu
238  *   MTU value to set.
239  *
240  * @return
241  *   0 on success, a negative errno value otherwise and rte_errno is set.
242  */
243 int
244 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
245 {
246 	struct ifreq request = { .ifr_mtu = mtu, };
247 
248 	return mlx5_ifreq(dev, SIOCSIFMTU, &request);
249 }
250 
251 /**
252  * Set device flags.
253  *
254  * @param dev
255  *   Pointer to Ethernet device.
256  * @param keep
257  *   Bitmask for flags that must remain untouched.
258  * @param flags
259  *   Bitmask for flags to modify.
260  *
261  * @return
262  *   0 on success, a negative errno value otherwise and rte_errno is set.
263  */
264 static int
265 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
266 {
267 	struct ifreq request;
268 	int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
269 
270 	if (ret)
271 		return ret;
272 	request.ifr_flags &= keep;
273 	request.ifr_flags |= flags & ~keep;
274 	return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
275 }
276 
277 /**
278  * Get device current raw clock counter
279  *
280  * @param dev
281  *   Pointer to Ethernet device structure.
282  * @param[out] time
283  *   Current raw clock counter of the device.
284  *
285  * @return
286  *   0 if the clock has correctly been read
287  *   The value of errno in case of error
288  */
289 int
290 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
291 {
292 	struct mlx5_priv *priv = dev->data->dev_private;
293 	struct ibv_context *ctx = priv->sh->ctx;
294 	struct ibv_values_ex values;
295 	int err = 0;
296 
297 	values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
298 	err = mlx5_glue->query_rt_values_ex(ctx, &values);
299 	if (err != 0) {
300 		DRV_LOG(WARNING, "Could not query the clock !");
301 		return err;
302 	}
303 	*clock = values.raw_clock.tv_nsec;
304 	return 0;
305 }
306 
307 /**
308  * Retrieve the master device for representor in the same switch domain.
309  *
310  * @param dev
311  *   Pointer to representor Ethernet device structure.
312  *
313  * @return
314  *   Master device structure  on success, NULL otherwise.
315  */
316 static struct rte_eth_dev *
317 mlx5_find_master_dev(struct rte_eth_dev *dev)
318 {
319 	struct mlx5_priv *priv;
320 	uint16_t port_id;
321 	uint16_t domain_id;
322 
323 	priv = dev->data->dev_private;
324 	domain_id = priv->domain_id;
325 	MLX5_ASSERT(priv->representor);
326 	MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
327 		struct mlx5_priv *opriv =
328 			rte_eth_devices[port_id].data->dev_private;
329 		if (opriv &&
330 		    opriv->master &&
331 		    opriv->domain_id == domain_id &&
332 		    opriv->sh == priv->sh)
333 			return &rte_eth_devices[port_id];
334 	}
335 	return NULL;
336 }
337 
338 /**
339  * DPDK callback to retrieve physical link information.
340  *
341  * @param dev
342  *   Pointer to Ethernet device structure.
343  * @param[out] link
344  *   Storage for current link status.
345  *
346  * @return
347  *   0 on success, a negative errno value otherwise and rte_errno is set.
348  */
349 static int
350 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
351 			       struct rte_eth_link *link)
352 {
353 	struct mlx5_priv *priv = dev->data->dev_private;
354 	struct ethtool_cmd edata = {
355 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
356 	};
357 	struct ifreq ifr;
358 	struct rte_eth_link dev_link;
359 	int link_speed = 0;
360 	int ret;
361 
362 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
363 	if (ret) {
364 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
365 			dev->data->port_id, strerror(rte_errno));
366 		return ret;
367 	}
368 	dev_link = (struct rte_eth_link) {
369 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
370 				(ifr.ifr_flags & IFF_RUNNING)),
371 	};
372 	ifr = (struct ifreq) {
373 		.ifr_data = (void *)&edata,
374 	};
375 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
376 	if (ret) {
377 		if (ret == -ENOTSUP && priv->representor) {
378 			struct rte_eth_dev *master;
379 
380 			/*
381 			 * For representors we can try to inherit link
382 			 * settings from the master device. Actually
383 			 * link settings do not make a lot of sense
384 			 * for representors due to missing physical
385 			 * link. The old kernel drivers supported
386 			 * emulated settings query for representors,
387 			 * the new ones do not, so we have to add
388 			 * this code for compatibility issues.
389 			 */
390 			master = mlx5_find_master_dev(dev);
391 			if (master) {
392 				ifr = (struct ifreq) {
393 					.ifr_data = (void *)&edata,
394 				};
395 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
396 			}
397 		}
398 		if (ret) {
399 			DRV_LOG(WARNING,
400 				"port %u ioctl(SIOCETHTOOL,"
401 				" ETHTOOL_GSET) failed: %s",
402 				dev->data->port_id, strerror(rte_errno));
403 			return ret;
404 		}
405 	}
406 	link_speed = ethtool_cmd_speed(&edata);
407 	if (link_speed == -1)
408 		dev_link.link_speed = ETH_SPEED_NUM_UNKNOWN;
409 	else
410 		dev_link.link_speed = link_speed;
411 	priv->link_speed_capa = 0;
412 	if (edata.supported & SUPPORTED_Autoneg)
413 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
414 	if (edata.supported & (SUPPORTED_1000baseT_Full |
415 			       SUPPORTED_1000baseKX_Full))
416 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
417 	if (edata.supported & SUPPORTED_10000baseKR_Full)
418 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
419 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
420 			       SUPPORTED_40000baseCR4_Full |
421 			       SUPPORTED_40000baseSR4_Full |
422 			       SUPPORTED_40000baseLR4_Full))
423 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
424 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
425 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
426 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
427 			ETH_LINK_SPEED_FIXED);
428 	*link = dev_link;
429 	return 0;
430 }
431 
432 /**
433  * Retrieve physical link information (unlocked version using new ioctl).
434  *
435  * @param dev
436  *   Pointer to Ethernet device structure.
437  * @param[out] link
438  *   Storage for current link status.
439  *
440  * @return
441  *   0 on success, a negative errno value otherwise and rte_errno is set.
442  */
443 static int
444 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
445 			     struct rte_eth_link *link)
446 
447 {
448 	struct mlx5_priv *priv = dev->data->dev_private;
449 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
450 	struct ifreq ifr;
451 	struct rte_eth_link dev_link;
452 	struct rte_eth_dev *master = NULL;
453 	uint64_t sc;
454 	int ret;
455 
456 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
457 	if (ret) {
458 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
459 			dev->data->port_id, strerror(rte_errno));
460 		return ret;
461 	}
462 	dev_link = (struct rte_eth_link) {
463 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
464 				(ifr.ifr_flags & IFF_RUNNING)),
465 	};
466 	ifr = (struct ifreq) {
467 		.ifr_data = (void *)&gcmd,
468 	};
469 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
470 	if (ret) {
471 		if (ret == -ENOTSUP && priv->representor) {
472 			/*
473 			 * For representors we can try to inherit link
474 			 * settings from the master device. Actually
475 			 * link settings do not make a lot of sense
476 			 * for representors due to missing physical
477 			 * link. The old kernel drivers supported
478 			 * emulated settings query for representors,
479 			 * the new ones do not, so we have to add
480 			 * this code for compatibility issues.
481 			 */
482 			master = mlx5_find_master_dev(dev);
483 			if (master) {
484 				ifr = (struct ifreq) {
485 					.ifr_data = (void *)&gcmd,
486 				};
487 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
488 			}
489 		}
490 		if (ret) {
491 			DRV_LOG(DEBUG,
492 				"port %u ioctl(SIOCETHTOOL,"
493 				" ETHTOOL_GLINKSETTINGS) failed: %s",
494 				dev->data->port_id, strerror(rte_errno));
495 			return ret;
496 		}
497 	}
498 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
499 
500 	alignas(struct ethtool_link_settings)
501 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
502 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
503 	struct ethtool_link_settings *ecmd = (void *)data;
504 
505 	*ecmd = gcmd;
506 	ifr.ifr_data = (void *)ecmd;
507 	ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
508 	if (ret) {
509 		DRV_LOG(DEBUG,
510 			"port %u ioctl(SIOCETHTOOL,"
511 			"ETHTOOL_GLINKSETTINGS) failed: %s",
512 			dev->data->port_id, strerror(rte_errno));
513 		return ret;
514 	}
515 	dev_link.link_speed = (ecmd->speed == UINT32_MAX) ?
516 				ETH_SPEED_NUM_UNKNOWN : ecmd->speed;
517 	sc = ecmd->link_mode_masks[0] |
518 		((uint64_t)ecmd->link_mode_masks[1] << 32);
519 	priv->link_speed_capa = 0;
520 	if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
521 		priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
522 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
523 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
524 		priv->link_speed_capa |= ETH_LINK_SPEED_1G;
525 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
526 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
527 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
528 		priv->link_speed_capa |= ETH_LINK_SPEED_10G;
529 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
530 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
531 		priv->link_speed_capa |= ETH_LINK_SPEED_20G;
532 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
533 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
534 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
535 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
536 		priv->link_speed_capa |= ETH_LINK_SPEED_40G;
537 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
538 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
539 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
540 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
541 		priv->link_speed_capa |= ETH_LINK_SPEED_56G;
542 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
543 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
544 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
545 		priv->link_speed_capa |= ETH_LINK_SPEED_25G;
546 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
547 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
548 		priv->link_speed_capa |= ETH_LINK_SPEED_50G;
549 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
550 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
551 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
552 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
553 		priv->link_speed_capa |= ETH_LINK_SPEED_100G;
554 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT) |
555 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT)))
556 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
557 
558 	sc = ecmd->link_mode_masks[2] |
559 		((uint64_t)ecmd->link_mode_masks[3] << 32);
560 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT) |
561 		  MLX5_BITSHIFT
562 		       (ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT) |
563 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT)))
564 		priv->link_speed_capa |= ETH_LINK_SPEED_200G;
565 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
566 				ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
567 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
568 				  ETH_LINK_SPEED_FIXED);
569 	*link = dev_link;
570 	return 0;
571 }
572 
573 /**
574  * DPDK callback to retrieve physical link information.
575  *
576  * @param dev
577  *   Pointer to Ethernet device structure.
578  * @param wait_to_complete
579  *   Wait for request completion.
580  *
581  * @return
582  *   0 if link status was not updated, positive if it was, a negative errno
583  *   value otherwise and rte_errno is set.
584  */
585 int
586 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
587 {
588 	int ret;
589 	struct rte_eth_link dev_link;
590 	time_t start_time = time(NULL);
591 	int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT;
592 
593 	do {
594 		ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
595 		if (ret == -ENOTSUP)
596 			ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
597 		if (ret == 0)
598 			break;
599 		/* Handle wait to complete situation. */
600 		if ((wait_to_complete || retry) && ret == -EAGAIN) {
601 			if (abs((int)difftime(time(NULL), start_time)) <
602 			    MLX5_LINK_STATUS_TIMEOUT) {
603 				usleep(0);
604 				continue;
605 			} else {
606 				rte_errno = EBUSY;
607 				return -rte_errno;
608 			}
609 		} else if (ret < 0) {
610 			return ret;
611 		}
612 	} while (wait_to_complete || retry-- > 0);
613 	ret = !!memcmp(&dev->data->dev_link, &dev_link,
614 		       sizeof(struct rte_eth_link));
615 	dev->data->dev_link = dev_link;
616 	return ret;
617 }
618 
619 /**
620  * DPDK callback to get flow control status.
621  *
622  * @param dev
623  *   Pointer to Ethernet device structure.
624  * @param[out] fc_conf
625  *   Flow control output buffer.
626  *
627  * @return
628  *   0 on success, a negative errno value otherwise and rte_errno is set.
629  */
630 int
631 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
632 {
633 	struct ifreq ifr;
634 	struct ethtool_pauseparam ethpause = {
635 		.cmd = ETHTOOL_GPAUSEPARAM
636 	};
637 	int ret;
638 
639 	ifr.ifr_data = (void *)&ethpause;
640 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
641 	if (ret) {
642 		DRV_LOG(WARNING,
643 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
644 			" %s",
645 			dev->data->port_id, strerror(rte_errno));
646 		return ret;
647 	}
648 	fc_conf->autoneg = ethpause.autoneg;
649 	if (ethpause.rx_pause && ethpause.tx_pause)
650 		fc_conf->mode = RTE_FC_FULL;
651 	else if (ethpause.rx_pause)
652 		fc_conf->mode = RTE_FC_RX_PAUSE;
653 	else if (ethpause.tx_pause)
654 		fc_conf->mode = RTE_FC_TX_PAUSE;
655 	else
656 		fc_conf->mode = RTE_FC_NONE;
657 	return 0;
658 }
659 
660 /**
661  * DPDK callback to modify flow control parameters.
662  *
663  * @param dev
664  *   Pointer to Ethernet device structure.
665  * @param[in] fc_conf
666  *   Flow control parameters.
667  *
668  * @return
669  *   0 on success, a negative errno value otherwise and rte_errno is set.
670  */
671 int
672 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
673 {
674 	struct ifreq ifr;
675 	struct ethtool_pauseparam ethpause = {
676 		.cmd = ETHTOOL_SPAUSEPARAM
677 	};
678 	int ret;
679 
680 	ifr.ifr_data = (void *)&ethpause;
681 	ethpause.autoneg = fc_conf->autoneg;
682 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
683 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
684 		ethpause.rx_pause = 1;
685 	else
686 		ethpause.rx_pause = 0;
687 
688 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
689 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
690 		ethpause.tx_pause = 1;
691 	else
692 		ethpause.tx_pause = 0;
693 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
694 	if (ret) {
695 		DRV_LOG(WARNING,
696 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
697 			" failed: %s",
698 			dev->data->port_id, strerror(rte_errno));
699 		return ret;
700 	}
701 	return 0;
702 }
703 
704 /**
705  * Handle asynchronous removal event for entire multiport device.
706  *
707  * @param sh
708  *   Infiniband device shared context.
709  */
710 static void
711 mlx5_dev_interrupt_device_fatal(struct mlx5_dev_ctx_shared *sh)
712 {
713 	uint32_t i;
714 
715 	for (i = 0; i < sh->max_port; ++i) {
716 		struct rte_eth_dev *dev;
717 
718 		if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
719 			/*
720 			 * Or not existing port either no
721 			 * handler installed for this port.
722 			 */
723 			continue;
724 		}
725 		dev = &rte_eth_devices[sh->port[i].ih_port_id];
726 		MLX5_ASSERT(dev);
727 		if (dev->data->dev_conf.intr_conf.rmv)
728 			rte_eth_dev_callback_process
729 				(dev, RTE_ETH_EVENT_INTR_RMV, NULL);
730 	}
731 }
732 
733 /**
734  * Handle shared asynchronous events the NIC (removal event
735  * and link status change). Supports multiport IB device.
736  *
737  * @param cb_arg
738  *   Callback argument.
739  */
740 void
741 mlx5_dev_interrupt_handler(void *cb_arg)
742 {
743 	struct mlx5_dev_ctx_shared *sh = cb_arg;
744 	struct ibv_async_event event;
745 
746 	/* Read all message from the IB device and acknowledge them. */
747 	for (;;) {
748 		struct rte_eth_dev *dev;
749 		uint32_t tmp;
750 
751 		if (mlx5_glue->get_async_event(sh->ctx, &event))
752 			break;
753 		/* Retrieve and check IB port index. */
754 		tmp = (uint32_t)event.element.port_num;
755 		if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
756 			/*
757 			 * The DEVICE_FATAL event is called once for
758 			 * entire device without port specifying.
759 			 * We should notify all existing ports.
760 			 */
761 			mlx5_glue->ack_async_event(&event);
762 			mlx5_dev_interrupt_device_fatal(sh);
763 			continue;
764 		}
765 		MLX5_ASSERT(tmp && (tmp <= sh->max_port));
766 		if (!tmp) {
767 			/* Unsupported device level event. */
768 			mlx5_glue->ack_async_event(&event);
769 			DRV_LOG(DEBUG,
770 				"unsupported common event (type %d)",
771 				event.event_type);
772 			continue;
773 		}
774 		if (tmp > sh->max_port) {
775 			/* Invalid IB port index. */
776 			mlx5_glue->ack_async_event(&event);
777 			DRV_LOG(DEBUG,
778 				"cannot handle an event (type %d)"
779 				"due to invalid IB port index (%u)",
780 				event.event_type, tmp);
781 			continue;
782 		}
783 		if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
784 			/* No handler installed. */
785 			mlx5_glue->ack_async_event(&event);
786 			DRV_LOG(DEBUG,
787 				"cannot handle an event (type %d)"
788 				"due to no handler installed for port %u",
789 				event.event_type, tmp);
790 			continue;
791 		}
792 		/* Retrieve ethernet device descriptor. */
793 		tmp = sh->port[tmp - 1].ih_port_id;
794 		dev = &rte_eth_devices[tmp];
795 		MLX5_ASSERT(dev);
796 		if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
797 		     event.event_type == IBV_EVENT_PORT_ERR) &&
798 			dev->data->dev_conf.intr_conf.lsc) {
799 			mlx5_glue->ack_async_event(&event);
800 			if (mlx5_link_update(dev, 0) == -EAGAIN) {
801 				usleep(0);
802 				continue;
803 			}
804 			rte_eth_dev_callback_process
805 				(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
806 			continue;
807 		}
808 		DRV_LOG(DEBUG,
809 			"port %u cannot handle an unknown event (type %d)",
810 			dev->data->port_id, event.event_type);
811 		mlx5_glue->ack_async_event(&event);
812 	}
813 }
814 
815 /*
816  * Unregister callback handler safely. The handler may be active
817  * while we are trying to unregister it, in this case code -EAGAIN
818  * is returned by rte_intr_callback_unregister(). This routine checks
819  * the return code and tries to unregister handler again.
820  *
821  * @param handle
822  *   interrupt handle
823  * @param cb_fn
824  *   pointer to callback routine
825  * @cb_arg
826  *   opaque callback parameter
827  */
828 void
829 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
830 			      rte_intr_callback_fn cb_fn, void *cb_arg)
831 {
832 	/*
833 	 * Try to reduce timeout management overhead by not calling
834 	 * the timer related routines on the first iteration. If the
835 	 * unregistering succeeds on first call there will be no
836 	 * timer calls at all.
837 	 */
838 	uint64_t twait = 0;
839 	uint64_t start = 0;
840 
841 	do {
842 		int ret;
843 
844 		ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
845 		if (ret >= 0)
846 			return;
847 		if (ret != -EAGAIN) {
848 			DRV_LOG(INFO, "failed to unregister interrupt"
849 				      " handler (error: %d)", ret);
850 			MLX5_ASSERT(false);
851 			return;
852 		}
853 		if (twait) {
854 			struct timespec onems;
855 
856 			/* Wait one millisecond and try again. */
857 			onems.tv_sec = 0;
858 			onems.tv_nsec = NS_PER_S / MS_PER_S;
859 			nanosleep(&onems, 0);
860 			/* Check whether one second elapsed. */
861 			if ((rte_get_timer_cycles() - start) <= twait)
862 				continue;
863 		} else {
864 			/*
865 			 * We get the amount of timer ticks for one second.
866 			 * If this amount elapsed it means we spent one
867 			 * second in waiting. This branch is executed once
868 			 * on first iteration.
869 			 */
870 			twait = rte_get_timer_hz();
871 			MLX5_ASSERT(twait);
872 		}
873 		/*
874 		 * Timeout elapsed, show message (once a second) and retry.
875 		 * We have no other acceptable option here, if we ignore
876 		 * the unregistering return code the handler will not
877 		 * be unregistered, fd will be closed and we may get the
878 		 * crush. Hanging and messaging in the loop seems not to be
879 		 * the worst choice.
880 		 */
881 		DRV_LOG(INFO, "Retrying to unregister interrupt handler");
882 		start = rte_get_timer_cycles();
883 	} while (true);
884 }
885 
886 /**
887  * Handle DEVX interrupts from the NIC.
888  * This function is probably called from the DPDK host thread.
889  *
890  * @param cb_arg
891  *   Callback argument.
892  */
893 void
894 mlx5_dev_interrupt_handler_devx(void *cb_arg)
895 {
896 #ifndef HAVE_IBV_DEVX_ASYNC
897 	(void)cb_arg;
898 	return;
899 #else
900 	struct mlx5_dev_ctx_shared *sh = cb_arg;
901 	union {
902 		struct mlx5dv_devx_async_cmd_hdr cmd_resp;
903 		uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
904 			    MLX5_ST_SZ_BYTES(traffic_counter) +
905 			    sizeof(struct mlx5dv_devx_async_cmd_hdr)];
906 	} out;
907 	uint8_t *buf = out.buf + sizeof(out.cmd_resp);
908 
909 	while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
910 						   &out.cmd_resp,
911 						   sizeof(out.buf)))
912 		mlx5_flow_async_pool_query_handle
913 			(sh, (uint64_t)out.cmd_resp.wr_id,
914 			 mlx5_devx_get_out_command_status(buf));
915 #endif /* HAVE_IBV_DEVX_ASYNC */
916 }
917 
918 /**
919  * DPDK callback to bring the link DOWN.
920  *
921  * @param dev
922  *   Pointer to Ethernet device structure.
923  *
924  * @return
925  *   0 on success, a negative errno value otherwise and rte_errno is set.
926  */
927 int
928 mlx5_set_link_down(struct rte_eth_dev *dev)
929 {
930 	return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
931 }
932 
933 /**
934  * DPDK callback to bring the link UP.
935  *
936  * @param dev
937  *   Pointer to Ethernet device structure.
938  *
939  * @return
940  *   0 on success, a negative errno value otherwise and rte_errno is set.
941  */
942 int
943 mlx5_set_link_up(struct rte_eth_dev *dev)
944 {
945 	return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
946 }
947 
948 /**
949  * Check if mlx5 device was removed.
950  *
951  * @param dev
952  *   Pointer to Ethernet device structure.
953  *
954  * @return
955  *   1 when device is removed, otherwise 0.
956  */
957 int
958 mlx5_is_removed(struct rte_eth_dev *dev)
959 {
960 	struct ibv_device_attr device_attr;
961 	struct mlx5_priv *priv = dev->data->dev_private;
962 
963 	if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
964 		return 1;
965 	return 0;
966 }
967 
968 /**
969  * Analyze gathered port parameters via sysfs to recognize master
970  * and representor devices for E-Switch configuration.
971  *
972  * @param[in] device_dir
973  *   flag of presence of "device" directory under port device key.
974  * @param[inout] switch_info
975  *   Port information, including port name as a number and port name
976  *   type if recognized
977  *
978  * @return
979  *   master and representor flags are set in switch_info according to
980  *   recognized parameters (if any).
981  */
982 static void
983 mlx5_sysfs_check_switch_info(bool device_dir,
984 			     struct mlx5_switch_info *switch_info)
985 {
986 	switch (switch_info->name_type) {
987 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
988 		/*
989 		 * Name is not recognized, assume the master,
990 		 * check the device directory presence.
991 		 */
992 		switch_info->master = device_dir;
993 		break;
994 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
995 		/*
996 		 * Name is not set, this assumes the legacy naming
997 		 * schema for master, just check if there is
998 		 * a device directory.
999 		 */
1000 		switch_info->master = device_dir;
1001 		break;
1002 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1003 		/* New uplink naming schema recognized. */
1004 		switch_info->master = 1;
1005 		break;
1006 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1007 		/* Legacy representors naming schema. */
1008 		switch_info->representor = !device_dir;
1009 		break;
1010 	case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
1011 		/* Fallthrough */
1012 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1013 		/* New representors naming schema. */
1014 		switch_info->representor = 1;
1015 		break;
1016 	}
1017 }
1018 
1019 /**
1020  * Get switch information associated with network interface.
1021  *
1022  * @param ifindex
1023  *   Network interface index.
1024  * @param[out] info
1025  *   Switch information object, populated in case of success.
1026  *
1027  * @return
1028  *   0 on success, a negative errno value otherwise and rte_errno is set.
1029  */
1030 int
1031 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1032 {
1033 	char ifname[IF_NAMESIZE];
1034 	char port_name[IF_NAMESIZE];
1035 	FILE *file;
1036 	struct mlx5_switch_info data = {
1037 		.master = 0,
1038 		.representor = 0,
1039 		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1040 		.port_name = 0,
1041 		.switch_id = 0,
1042 	};
1043 	DIR *dir;
1044 	bool port_switch_id_set = false;
1045 	bool device_dir = false;
1046 	char c;
1047 	int ret;
1048 
1049 	if (!if_indextoname(ifindex, ifname)) {
1050 		rte_errno = errno;
1051 		return -rte_errno;
1052 	}
1053 
1054 	MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1055 	      ifname);
1056 	MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1057 	      ifname);
1058 	MKSTR(pci_device, "/sys/class/net/%s/device",
1059 	      ifname);
1060 
1061 	file = fopen(phys_port_name, "rb");
1062 	if (file != NULL) {
1063 		ret = fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", port_name);
1064 		fclose(file);
1065 		if (ret == 1)
1066 			mlx5_translate_port_name(port_name, &data);
1067 	}
1068 	file = fopen(phys_switch_id, "rb");
1069 	if (file == NULL) {
1070 		rte_errno = errno;
1071 		return -rte_errno;
1072 	}
1073 	port_switch_id_set =
1074 		fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1075 		c == '\n';
1076 	fclose(file);
1077 	dir = opendir(pci_device);
1078 	if (dir != NULL) {
1079 		closedir(dir);
1080 		device_dir = true;
1081 	}
1082 	if (port_switch_id_set) {
1083 		/* We have some E-Switch configuration. */
1084 		mlx5_sysfs_check_switch_info(device_dir, &data);
1085 	}
1086 	*info = data;
1087 	MLX5_ASSERT(!(data.master && data.representor));
1088 	if (data.master && data.representor) {
1089 		DRV_LOG(ERR, "ifindex %u device is recognized as master"
1090 			     " and as representor", ifindex);
1091 		rte_errno = ENODEV;
1092 		return -rte_errno;
1093 	}
1094 	return 0;
1095 }
1096 
1097 /**
1098  * Get bond information associated with network interface.
1099  *
1100  * @param pf_ifindex
1101  *   Network interface index of bond slave interface
1102  * @param[out] ifindex
1103  *   Pointer to bond ifindex.
1104  * @param[out] ifname
1105  *   Pointer to bond ifname.
1106  *
1107  * @return
1108  *   0 on success, a negative errno value otherwise and rte_errno is set.
1109  */
1110 int
1111 mlx5_sysfs_bond_info(unsigned int pf_ifindex, unsigned int *ifindex,
1112 		     char *ifname)
1113 {
1114 	char name[IF_NAMESIZE];
1115 	FILE *file;
1116 	unsigned int index;
1117 	int ret;
1118 
1119 	if (!if_indextoname(pf_ifindex, name) || !strlen(name)) {
1120 		rte_errno = errno;
1121 		return -rte_errno;
1122 	}
1123 	MKSTR(bond_if, "/sys/class/net/%s/master/ifindex", name);
1124 	/* read bond ifindex */
1125 	file = fopen(bond_if, "rb");
1126 	if (file == NULL) {
1127 		rte_errno = errno;
1128 		return -rte_errno;
1129 	}
1130 	ret = fscanf(file, "%u", &index);
1131 	fclose(file);
1132 	if (ret <= 0) {
1133 		rte_errno = errno;
1134 		return -rte_errno;
1135 	}
1136 	if (ifindex)
1137 		*ifindex = index;
1138 
1139 	/* read bond device name from symbol link */
1140 	if (ifname) {
1141 		if (!if_indextoname(index, ifname)) {
1142 			rte_errno = errno;
1143 			return -rte_errno;
1144 		}
1145 	}
1146 	return 0;
1147 }
1148 
1149 /**
1150  * DPDK callback to retrieve plug-in module EEPROM information (type and size).
1151  *
1152  * @param dev
1153  *   Pointer to Ethernet device structure.
1154  * @param[out] modinfo
1155  *   Storage for plug-in module EEPROM information.
1156  *
1157  * @return
1158  *   0 on success, a negative errno value otherwise and rte_errno is set.
1159  */
1160 int
1161 mlx5_get_module_info(struct rte_eth_dev *dev,
1162 		     struct rte_eth_dev_module_info *modinfo)
1163 {
1164 	struct ethtool_modinfo info = {
1165 		.cmd = ETHTOOL_GMODULEINFO,
1166 	};
1167 	struct ifreq ifr = (struct ifreq) {
1168 		.ifr_data = (void *)&info,
1169 	};
1170 	int ret = 0;
1171 
1172 	if (!dev || !modinfo) {
1173 		DRV_LOG(WARNING, "missing argument, cannot get module info");
1174 		rte_errno = EINVAL;
1175 		return -rte_errno;
1176 	}
1177 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1178 	if (ret) {
1179 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1180 			dev->data->port_id, strerror(rte_errno));
1181 		return ret;
1182 	}
1183 	modinfo->type = info.type;
1184 	modinfo->eeprom_len = info.eeprom_len;
1185 	return ret;
1186 }
1187 
1188 /**
1189  * DPDK callback to retrieve plug-in module EEPROM data.
1190  *
1191  * @param dev
1192  *   Pointer to Ethernet device structure.
1193  * @param[out] info
1194  *   Storage for plug-in module EEPROM data.
1195  *
1196  * @return
1197  *   0 on success, a negative errno value otherwise and rte_errno is set.
1198  */
1199 int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
1200 			   struct rte_dev_eeprom_info *info)
1201 {
1202 	struct ethtool_eeprom *eeprom;
1203 	struct ifreq ifr;
1204 	int ret = 0;
1205 
1206 	if (!dev || !info) {
1207 		DRV_LOG(WARNING, "missing argument, cannot get module eeprom");
1208 		rte_errno = EINVAL;
1209 		return -rte_errno;
1210 	}
1211 	eeprom = mlx5_malloc(MLX5_MEM_ZERO,
1212 			     (sizeof(struct ethtool_eeprom) + info->length), 0,
1213 			     SOCKET_ID_ANY);
1214 	if (!eeprom) {
1215 		DRV_LOG(WARNING, "port %u cannot allocate memory for "
1216 			"eeprom data", dev->data->port_id);
1217 		rte_errno = ENOMEM;
1218 		return -rte_errno;
1219 	}
1220 	eeprom->cmd = ETHTOOL_GMODULEEEPROM;
1221 	eeprom->offset = info->offset;
1222 	eeprom->len = info->length;
1223 	ifr = (struct ifreq) {
1224 		.ifr_data = (void *)eeprom,
1225 	};
1226 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1227 	if (ret)
1228 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1229 			dev->data->port_id, strerror(rte_errno));
1230 	else
1231 		rte_memcpy(info->data, eeprom->data, info->length);
1232 	mlx5_free(eeprom);
1233 	return ret;
1234 }
1235 
1236 /**
1237  * Read device counters table.
1238  *
1239  * @param dev
1240  *   Pointer to Ethernet device.
1241  * @param[out] stats
1242  *   Counters table output buffer.
1243  *
1244  * @return
1245  *   0 on success and stats is filled, negative errno value otherwise and
1246  *   rte_errno is set.
1247  */
1248 int
1249 mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
1250 {
1251 	struct mlx5_priv *priv = dev->data->dev_private;
1252 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1253 	unsigned int i;
1254 	struct ifreq ifr;
1255 	unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t);
1256 	unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz];
1257 	struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf;
1258 	int ret;
1259 
1260 	et_stats->cmd = ETHTOOL_GSTATS;
1261 	et_stats->n_stats = xstats_ctrl->stats_n;
1262 	ifr.ifr_data = (caddr_t)et_stats;
1263 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1264 	if (ret) {
1265 		DRV_LOG(WARNING,
1266 			"port %u unable to read statistic values from device",
1267 			dev->data->port_id);
1268 		return ret;
1269 	}
1270 	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
1271 		if (xstats_ctrl->info[i].dev) {
1272 			ret = mlx5_os_read_dev_stat(priv,
1273 					    xstats_ctrl->info[i].ctr_name,
1274 					    &stats[i]);
1275 			/* return last xstats counter if fail to read. */
1276 			if (ret == 0)
1277 				xstats_ctrl->xstats[i] = stats[i];
1278 			else
1279 				stats[i] = xstats_ctrl->xstats[i];
1280 		} else {
1281 			stats[i] = (uint64_t)
1282 				et_stats->data[xstats_ctrl->dev_table_idx[i]];
1283 		}
1284 	}
1285 	return 0;
1286 }
1287 
1288 /**
1289  * Query the number of statistics provided by ETHTOOL.
1290  *
1291  * @param dev
1292  *   Pointer to Ethernet device.
1293  *
1294  * @return
1295  *   Number of statistics on success, negative errno value otherwise and
1296  *   rte_errno is set.
1297  */
1298 int
1299 mlx5_os_get_stats_n(struct rte_eth_dev *dev)
1300 {
1301 	struct ethtool_drvinfo drvinfo;
1302 	struct ifreq ifr;
1303 	int ret;
1304 
1305 	drvinfo.cmd = ETHTOOL_GDRVINFO;
1306 	ifr.ifr_data = (caddr_t)&drvinfo;
1307 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1308 	if (ret) {
1309 		DRV_LOG(WARNING, "port %u unable to query number of statistics",
1310 			dev->data->port_id);
1311 		return ret;
1312 	}
1313 	return drvinfo.n_stats;
1314 }
1315 
1316 static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
1317 	{
1318 		.dpdk_name = "rx_unicast_bytes",
1319 		.ctr_name = "rx_vport_unicast_bytes",
1320 	},
1321 	{
1322 		.dpdk_name = "rx_multicast_bytes",
1323 		.ctr_name = "rx_vport_multicast_bytes",
1324 	},
1325 	{
1326 		.dpdk_name = "rx_broadcast_bytes",
1327 		.ctr_name = "rx_vport_broadcast_bytes",
1328 	},
1329 	{
1330 		.dpdk_name = "rx_unicast_packets",
1331 		.ctr_name = "rx_vport_unicast_packets",
1332 	},
1333 	{
1334 		.dpdk_name = "rx_multicast_packets",
1335 		.ctr_name = "rx_vport_multicast_packets",
1336 	},
1337 	{
1338 		.dpdk_name = "rx_broadcast_packets",
1339 		.ctr_name = "rx_vport_broadcast_packets",
1340 	},
1341 	{
1342 		.dpdk_name = "tx_unicast_bytes",
1343 		.ctr_name = "tx_vport_unicast_bytes",
1344 	},
1345 	{
1346 		.dpdk_name = "tx_multicast_bytes",
1347 		.ctr_name = "tx_vport_multicast_bytes",
1348 	},
1349 	{
1350 		.dpdk_name = "tx_broadcast_bytes",
1351 		.ctr_name = "tx_vport_broadcast_bytes",
1352 	},
1353 	{
1354 		.dpdk_name = "tx_unicast_packets",
1355 		.ctr_name = "tx_vport_unicast_packets",
1356 	},
1357 	{
1358 		.dpdk_name = "tx_multicast_packets",
1359 		.ctr_name = "tx_vport_multicast_packets",
1360 	},
1361 	{
1362 		.dpdk_name = "tx_broadcast_packets",
1363 		.ctr_name = "tx_vport_broadcast_packets",
1364 	},
1365 	{
1366 		.dpdk_name = "rx_wqe_errors",
1367 		.ctr_name = "rx_wqe_err",
1368 	},
1369 	{
1370 		.dpdk_name = "rx_phy_crc_errors",
1371 		.ctr_name = "rx_crc_errors_phy",
1372 	},
1373 	{
1374 		.dpdk_name = "rx_phy_in_range_len_errors",
1375 		.ctr_name = "rx_in_range_len_errors_phy",
1376 	},
1377 	{
1378 		.dpdk_name = "rx_phy_symbol_errors",
1379 		.ctr_name = "rx_symbol_err_phy",
1380 	},
1381 	{
1382 		.dpdk_name = "tx_phy_errors",
1383 		.ctr_name = "tx_errors_phy",
1384 	},
1385 	{
1386 		.dpdk_name = "rx_out_of_buffer",
1387 		.ctr_name = "out_of_buffer",
1388 		.dev = 1,
1389 	},
1390 	{
1391 		.dpdk_name = "tx_phy_packets",
1392 		.ctr_name = "tx_packets_phy",
1393 	},
1394 	{
1395 		.dpdk_name = "rx_phy_packets",
1396 		.ctr_name = "rx_packets_phy",
1397 	},
1398 	{
1399 		.dpdk_name = "tx_phy_discard_packets",
1400 		.ctr_name = "tx_discards_phy",
1401 	},
1402 	{
1403 		.dpdk_name = "rx_phy_discard_packets",
1404 		.ctr_name = "rx_discards_phy",
1405 	},
1406 	{
1407 		.dpdk_name = "tx_phy_bytes",
1408 		.ctr_name = "tx_bytes_phy",
1409 	},
1410 	{
1411 		.dpdk_name = "rx_phy_bytes",
1412 		.ctr_name = "rx_bytes_phy",
1413 	},
1414 	/* Representor only */
1415 	{
1416 		.dpdk_name = "rx_vport_packets",
1417 		.ctr_name = "vport_rx_packets",
1418 	},
1419 	{
1420 		.dpdk_name = "rx_vport_bytes",
1421 		.ctr_name = "vport_rx_bytes",
1422 	},
1423 	{
1424 		.dpdk_name = "tx_vport_packets",
1425 		.ctr_name = "vport_tx_packets",
1426 	},
1427 	{
1428 		.dpdk_name = "tx_vport_bytes",
1429 		.ctr_name = "vport_tx_bytes",
1430 	},
1431 };
1432 
1433 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init);
1434 
1435 /**
1436  * Init the structures to read device counters.
1437  *
1438  * @param dev
1439  *   Pointer to Ethernet device.
1440  */
1441 void
1442 mlx5_os_stats_init(struct rte_eth_dev *dev)
1443 {
1444 	struct mlx5_priv *priv = dev->data->dev_private;
1445 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1446 	struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl;
1447 	unsigned int i;
1448 	unsigned int j;
1449 	struct ifreq ifr;
1450 	struct ethtool_gstrings *strings = NULL;
1451 	unsigned int dev_stats_n;
1452 	unsigned int str_sz;
1453 	int ret;
1454 
1455 	/* So that it won't aggregate for each init. */
1456 	xstats_ctrl->mlx5_stats_n = 0;
1457 	ret = mlx5_os_get_stats_n(dev);
1458 	if (ret < 0) {
1459 		DRV_LOG(WARNING, "port %u no extended statistics available",
1460 			dev->data->port_id);
1461 		return;
1462 	}
1463 	dev_stats_n = ret;
1464 	/* Allocate memory to grab stat names and values. */
1465 	str_sz = dev_stats_n * ETH_GSTRING_LEN;
1466 	strings = (struct ethtool_gstrings *)
1467 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
1468 			      SOCKET_ID_ANY);
1469 	if (!strings) {
1470 		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats",
1471 		     dev->data->port_id);
1472 		return;
1473 	}
1474 	strings->cmd = ETHTOOL_GSTRINGS;
1475 	strings->string_set = ETH_SS_STATS;
1476 	strings->len = dev_stats_n;
1477 	ifr.ifr_data = (caddr_t)strings;
1478 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1479 	if (ret) {
1480 		DRV_LOG(WARNING, "port %u unable to get statistic names",
1481 			dev->data->port_id);
1482 		goto free;
1483 	}
1484 	for (i = 0; i != dev_stats_n; ++i) {
1485 		const char *curr_string = (const char *)
1486 			&strings->data[i * ETH_GSTRING_LEN];
1487 
1488 		for (j = 0; j != xstats_n; ++j) {
1489 			if (!strcmp(mlx5_counters_init[j].ctr_name,
1490 				    curr_string)) {
1491 				unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1492 
1493 				xstats_ctrl->dev_table_idx[idx] = i;
1494 				xstats_ctrl->info[idx] = mlx5_counters_init[j];
1495 				break;
1496 			}
1497 		}
1498 	}
1499 	/* Add dev counters. */
1500 	for (i = 0; i != xstats_n; ++i) {
1501 		if (mlx5_counters_init[i].dev) {
1502 			unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1503 
1504 			xstats_ctrl->info[idx] = mlx5_counters_init[i];
1505 			xstats_ctrl->hw_stats[idx] = 0;
1506 		}
1507 	}
1508 	MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS);
1509 	xstats_ctrl->stats_n = dev_stats_n;
1510 	/* Copy to base at first time. */
1511 	ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base);
1512 	if (ret)
1513 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
1514 			dev->data->port_id, strerror(rte_errno));
1515 	mlx5_os_read_dev_stat(priv, "out_of_buffer", &stats_ctrl->imissed_base);
1516 	stats_ctrl->imissed = 0;
1517 free:
1518 	mlx5_free(strings);
1519 }
1520 
1521 /**
1522  * Get MAC address by querying netdevice.
1523  *
1524  * @param[in] dev
1525  *   Pointer to Ethernet device.
1526  * @param[out] mac
1527  *   MAC address output buffer.
1528  *
1529  * @return
1530  *   0 on success, a negative errno value otherwise and rte_errno is set.
1531  */
1532 int
1533 mlx5_get_mac(struct rte_eth_dev *dev, uint8_t (*mac)[RTE_ETHER_ADDR_LEN])
1534 {
1535 	struct ifreq request;
1536 	int ret;
1537 
1538 	ret = mlx5_ifreq(dev, SIOCGIFHWADDR, &request);
1539 	if (ret)
1540 		return ret;
1541 	memcpy(mac, request.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1542 	return 0;
1543 }
1544 
1545