xref: /dpdk/drivers/net/mlx5/linux/mlx5_ethdev_os.c (revision 1c839246f934340e8dfb8fd71bc436f81541a587)
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 #include <mlx5_nl.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 /* Get interface index from SubFunction device name. */
133 int
134 mlx5_auxiliary_get_ifindex(const char *sf_name)
135 {
136 	char if_name[IF_NAMESIZE] = { 0 };
137 
138 	if (mlx5_auxiliary_get_child_name(sf_name, "/net",
139 					  if_name, sizeof(if_name)) != 0)
140 		return -rte_errno;
141 	return if_nametoindex(if_name);
142 }
143 
144 /**
145  * Get interface name from private structure.
146  *
147  * This is a port representor-aware version of mlx5_get_ifname_sysfs().
148  *
149  * @param[in] dev
150  *   Pointer to Ethernet device.
151  * @param[out] ifname
152  *   Interface name output buffer.
153  *
154  * @return
155  *   0 on success, a negative errno value otherwise and rte_errno is set.
156  */
157 int
158 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[MLX5_NAMESIZE])
159 {
160 	struct mlx5_priv *priv = dev->data->dev_private;
161 	unsigned int ifindex;
162 
163 	MLX5_ASSERT(priv);
164 	MLX5_ASSERT(priv->sh);
165 	if (priv->master && priv->sh->bond.ifindex > 0) {
166 		memcpy(ifname, priv->sh->bond.ifname, MLX5_NAMESIZE);
167 		return 0;
168 	}
169 	ifindex = mlx5_ifindex(dev);
170 	if (!ifindex) {
171 		if (!priv->representor)
172 			return mlx5_get_ifname_sysfs(priv->sh->ibdev_path,
173 						     *ifname);
174 		rte_errno = ENXIO;
175 		return -rte_errno;
176 	}
177 	if (if_indextoname(ifindex, &(*ifname)[0]))
178 		return 0;
179 	rte_errno = errno;
180 	return -rte_errno;
181 }
182 
183 /**
184  * Perform ifreq ioctl() on associated netdev ifname.
185  *
186  * @param[in] ifname
187  *   Pointer to netdev name.
188  * @param req
189  *   Request number to pass to ioctl().
190  * @param[out] ifr
191  *   Interface request structure output buffer.
192  *
193  * @return
194  *   0 on success, a negative errno value otherwise and rte_errno is set.
195  */
196 static int
197 mlx5_ifreq_by_ifname(const char *ifname, int req, struct ifreq *ifr)
198 {
199 	int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
200 	int ret = 0;
201 
202 	if (sock == -1) {
203 		rte_errno = errno;
204 		return -rte_errno;
205 	}
206 	rte_strscpy(ifr->ifr_name, ifname, sizeof(ifr->ifr_name));
207 	ret = ioctl(sock, req, ifr);
208 	if (ret == -1) {
209 		rte_errno = errno;
210 		goto error;
211 	}
212 	close(sock);
213 	return 0;
214 error:
215 	close(sock);
216 	return -rte_errno;
217 }
218 
219 /**
220  * Perform ifreq ioctl() on associated Ethernet device.
221  *
222  * @param[in] dev
223  *   Pointer to Ethernet device.
224  * @param req
225  *   Request number to pass to ioctl().
226  * @param[out] ifr
227  *   Interface request structure output buffer.
228  *
229  * @return
230  *   0 on success, a negative errno value otherwise and rte_errno is set.
231  */
232 static int
233 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
234 {
235 	char ifname[sizeof(ifr->ifr_name)];
236 	int ret;
237 
238 	ret = mlx5_get_ifname(dev, &ifname);
239 	if (ret)
240 		return -rte_errno;
241 	return mlx5_ifreq_by_ifname(ifname, req, ifr);
242 }
243 
244 /**
245  * Get device MTU.
246  *
247  * @param dev
248  *   Pointer to Ethernet device.
249  * @param[out] mtu
250  *   MTU value output buffer.
251  *
252  * @return
253  *   0 on success, a negative errno value otherwise and rte_errno is set.
254  */
255 int
256 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
257 {
258 	struct ifreq request;
259 	int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
260 
261 	if (ret)
262 		return ret;
263 	*mtu = request.ifr_mtu;
264 	return 0;
265 }
266 
267 /**
268  * Set device MTU.
269  *
270  * @param dev
271  *   Pointer to Ethernet device.
272  * @param mtu
273  *   MTU value to set.
274  *
275  * @return
276  *   0 on success, a negative errno value otherwise and rte_errno is set.
277  */
278 int
279 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
280 {
281 	struct ifreq request = { .ifr_mtu = mtu, };
282 
283 	return mlx5_ifreq(dev, SIOCSIFMTU, &request);
284 }
285 
286 /**
287  * Set device flags.
288  *
289  * @param dev
290  *   Pointer to Ethernet device.
291  * @param keep
292  *   Bitmask for flags that must remain untouched.
293  * @param flags
294  *   Bitmask for flags to modify.
295  *
296  * @return
297  *   0 on success, a negative errno value otherwise and rte_errno is set.
298  */
299 static int
300 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
301 {
302 	struct ifreq request;
303 	int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
304 
305 	if (ret)
306 		return ret;
307 	request.ifr_flags &= keep;
308 	request.ifr_flags |= flags & ~keep;
309 	return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
310 }
311 
312 /**
313  * Get device current raw clock counter
314  *
315  * @param dev
316  *   Pointer to Ethernet device structure.
317  * @param[out] time
318  *   Current raw clock counter of the device.
319  *
320  * @return
321  *   0 if the clock has correctly been read
322  *   The value of errno in case of error
323  */
324 int
325 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
326 {
327 	struct mlx5_priv *priv = dev->data->dev_private;
328 	struct ibv_context *ctx = priv->sh->cdev->ctx;
329 	struct ibv_values_ex values;
330 	int err = 0;
331 
332 	values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
333 	err = mlx5_glue->query_rt_values_ex(ctx, &values);
334 	if (err != 0) {
335 		DRV_LOG(WARNING, "Could not query the clock !");
336 		return err;
337 	}
338 	*clock = values.raw_clock.tv_nsec;
339 	return 0;
340 }
341 
342 /**
343  * Retrieve the master device for representor in the same switch domain.
344  *
345  * @param dev
346  *   Pointer to representor Ethernet device structure.
347  *
348  * @return
349  *   Master device structure  on success, NULL otherwise.
350  */
351 static struct rte_eth_dev *
352 mlx5_find_master_dev(struct rte_eth_dev *dev)
353 {
354 	struct mlx5_priv *priv;
355 	uint16_t port_id;
356 	uint16_t domain_id;
357 
358 	priv = dev->data->dev_private;
359 	domain_id = priv->domain_id;
360 	MLX5_ASSERT(priv->representor);
361 	MLX5_ETH_FOREACH_DEV(port_id, dev->device) {
362 		struct mlx5_priv *opriv =
363 			rte_eth_devices[port_id].data->dev_private;
364 		if (opriv &&
365 		    opriv->master &&
366 		    opriv->domain_id == domain_id &&
367 		    opriv->sh == priv->sh)
368 			return &rte_eth_devices[port_id];
369 	}
370 	return NULL;
371 }
372 
373 /**
374  * DPDK callback to retrieve physical link information.
375  *
376  * @param dev
377  *   Pointer to Ethernet device structure.
378  * @param[out] link
379  *   Storage for current link status.
380  *
381  * @return
382  *   0 on success, a negative errno value otherwise and rte_errno is set.
383  */
384 static int
385 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
386 			       struct rte_eth_link *link)
387 {
388 	struct mlx5_priv *priv = dev->data->dev_private;
389 	struct ethtool_cmd edata = {
390 		.cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
391 	};
392 	struct ifreq ifr;
393 	struct rte_eth_link dev_link;
394 	int link_speed = 0;
395 	int ret;
396 
397 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
398 	if (ret) {
399 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
400 			dev->data->port_id, strerror(rte_errno));
401 		return ret;
402 	}
403 	dev_link = (struct rte_eth_link) {
404 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
405 				(ifr.ifr_flags & IFF_RUNNING)),
406 	};
407 	ifr = (struct ifreq) {
408 		.ifr_data = (void *)&edata,
409 	};
410 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
411 	if (ret) {
412 		if (ret == -ENOTSUP && priv->representor) {
413 			struct rte_eth_dev *master;
414 
415 			/*
416 			 * For representors we can try to inherit link
417 			 * settings from the master device. Actually
418 			 * link settings do not make a lot of sense
419 			 * for representors due to missing physical
420 			 * link. The old kernel drivers supported
421 			 * emulated settings query for representors,
422 			 * the new ones do not, so we have to add
423 			 * this code for compatibility issues.
424 			 */
425 			master = mlx5_find_master_dev(dev);
426 			if (master) {
427 				ifr = (struct ifreq) {
428 					.ifr_data = (void *)&edata,
429 				};
430 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
431 			}
432 		}
433 		if (ret) {
434 			DRV_LOG(WARNING,
435 				"port %u ioctl(SIOCETHTOOL,"
436 				" ETHTOOL_GSET) failed: %s",
437 				dev->data->port_id, strerror(rte_errno));
438 			return ret;
439 		}
440 	}
441 	link_speed = ethtool_cmd_speed(&edata);
442 	if (link_speed == -1)
443 		dev_link.link_speed = RTE_ETH_SPEED_NUM_UNKNOWN;
444 	else
445 		dev_link.link_speed = link_speed;
446 	priv->link_speed_capa = 0;
447 	if (edata.supported & (SUPPORTED_1000baseT_Full |
448 			       SUPPORTED_1000baseKX_Full))
449 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_1G;
450 	if (edata.supported & SUPPORTED_10000baseKR_Full)
451 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_10G;
452 	if (edata.supported & (SUPPORTED_40000baseKR4_Full |
453 			       SUPPORTED_40000baseCR4_Full |
454 			       SUPPORTED_40000baseSR4_Full |
455 			       SUPPORTED_40000baseLR4_Full))
456 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_40G;
457 	dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
458 				RTE_ETH_LINK_HALF_DUPLEX : RTE_ETH_LINK_FULL_DUPLEX);
459 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
460 			RTE_ETH_LINK_SPEED_FIXED);
461 	*link = dev_link;
462 	return 0;
463 }
464 
465 /**
466  * Retrieve physical link information (unlocked version using new ioctl).
467  *
468  * @param dev
469  *   Pointer to Ethernet device structure.
470  * @param[out] link
471  *   Storage for current link status.
472  *
473  * @return
474  *   0 on success, a negative errno value otherwise and rte_errno is set.
475  */
476 static int
477 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
478 			     struct rte_eth_link *link)
479 
480 {
481 	struct mlx5_priv *priv = dev->data->dev_private;
482 	struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
483 	struct ifreq ifr;
484 	struct rte_eth_link dev_link;
485 	struct rte_eth_dev *master = NULL;
486 	uint64_t sc;
487 	int ret;
488 
489 	ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
490 	if (ret) {
491 		DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
492 			dev->data->port_id, strerror(rte_errno));
493 		return ret;
494 	}
495 	dev_link = (struct rte_eth_link) {
496 		.link_status = ((ifr.ifr_flags & IFF_UP) &&
497 				(ifr.ifr_flags & IFF_RUNNING)),
498 	};
499 	ifr = (struct ifreq) {
500 		.ifr_data = (void *)&gcmd,
501 	};
502 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
503 	if (ret) {
504 		if (ret == -ENOTSUP && priv->representor) {
505 			/*
506 			 * For representors we can try to inherit link
507 			 * settings from the master device. Actually
508 			 * link settings do not make a lot of sense
509 			 * for representors due to missing physical
510 			 * link. The old kernel drivers supported
511 			 * emulated settings query for representors,
512 			 * the new ones do not, so we have to add
513 			 * this code for compatibility issues.
514 			 */
515 			master = mlx5_find_master_dev(dev);
516 			if (master) {
517 				ifr = (struct ifreq) {
518 					.ifr_data = (void *)&gcmd,
519 				};
520 				ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
521 			}
522 		}
523 		if (ret) {
524 			DRV_LOG(DEBUG,
525 				"port %u ioctl(SIOCETHTOOL,"
526 				" ETHTOOL_GLINKSETTINGS) failed: %s",
527 				dev->data->port_id, strerror(rte_errno));
528 			return ret;
529 		}
530 	}
531 	gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
532 
533 	alignas(struct ethtool_link_settings)
534 	uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
535 		     sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
536 	struct ethtool_link_settings *ecmd = (void *)data;
537 
538 	*ecmd = gcmd;
539 	ifr.ifr_data = (void *)ecmd;
540 	ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
541 	if (ret) {
542 		DRV_LOG(DEBUG,
543 			"port %u ioctl(SIOCETHTOOL,"
544 			"ETHTOOL_GLINKSETTINGS) failed: %s",
545 			dev->data->port_id, strerror(rte_errno));
546 		return ret;
547 	}
548 	dev_link.link_speed = (ecmd->speed == UINT32_MAX) ?
549 				RTE_ETH_SPEED_NUM_UNKNOWN : ecmd->speed;
550 	sc = ecmd->link_mode_masks[0] |
551 		((uint64_t)ecmd->link_mode_masks[1] << 32);
552 	priv->link_speed_capa = 0;
553 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
554 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
555 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_1G;
556 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
557 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
558 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
559 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_10G;
560 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
561 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
562 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_20G;
563 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
564 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
565 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
566 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
567 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_40G;
568 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
569 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
570 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
571 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
572 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_56G;
573 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
574 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
575 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
576 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_25G;
577 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
578 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
579 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_50G;
580 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
581 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
582 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
583 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
584 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_100G;
585 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT) |
586 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT)))
587 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_200G;
588 
589 	sc = ecmd->link_mode_masks[2] |
590 		((uint64_t)ecmd->link_mode_masks[3] << 32);
591 	if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT) |
592 		  MLX5_BITSHIFT
593 		       (ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT) |
594 		  MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT)))
595 		priv->link_speed_capa |= RTE_ETH_LINK_SPEED_200G;
596 	dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
597 				RTE_ETH_LINK_HALF_DUPLEX : RTE_ETH_LINK_FULL_DUPLEX);
598 	dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
599 				  RTE_ETH_LINK_SPEED_FIXED);
600 	*link = dev_link;
601 	return 0;
602 }
603 
604 /**
605  * DPDK callback to retrieve physical link information.
606  *
607  * @param dev
608  *   Pointer to Ethernet device structure.
609  * @param wait_to_complete
610  *   Wait for request completion.
611  *
612  * @return
613  *   0 if link status was not updated, positive if it was, a negative errno
614  *   value otherwise and rte_errno is set.
615  */
616 int
617 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
618 {
619 	int ret;
620 	struct rte_eth_link dev_link;
621 	time_t start_time = time(NULL);
622 	int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT;
623 
624 	do {
625 		ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
626 		if (ret == -ENOTSUP)
627 			ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
628 		if (ret == 0)
629 			break;
630 		/* Handle wait to complete situation. */
631 		if ((wait_to_complete || retry) && ret == -EAGAIN) {
632 			if (abs((int)difftime(time(NULL), start_time)) <
633 			    MLX5_LINK_STATUS_TIMEOUT) {
634 				usleep(0);
635 				continue;
636 			} else {
637 				rte_errno = EBUSY;
638 				return -rte_errno;
639 			}
640 		} else if (ret < 0) {
641 			return ret;
642 		}
643 	} while (wait_to_complete || retry-- > 0);
644 	ret = !!memcmp(&dev->data->dev_link, &dev_link,
645 		       sizeof(struct rte_eth_link));
646 	dev->data->dev_link = dev_link;
647 	return ret;
648 }
649 
650 /**
651  * DPDK callback to get flow control status.
652  *
653  * @param dev
654  *   Pointer to Ethernet device structure.
655  * @param[out] fc_conf
656  *   Flow control output buffer.
657  *
658  * @return
659  *   0 on success, a negative errno value otherwise and rte_errno is set.
660  */
661 int
662 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
663 {
664 	struct ifreq ifr;
665 	struct ethtool_pauseparam ethpause = {
666 		.cmd = ETHTOOL_GPAUSEPARAM
667 	};
668 	int ret;
669 
670 	ifr.ifr_data = (void *)&ethpause;
671 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
672 	if (ret) {
673 		DRV_LOG(WARNING,
674 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
675 			" %s",
676 			dev->data->port_id, strerror(rte_errno));
677 		return ret;
678 	}
679 	fc_conf->autoneg = ethpause.autoneg;
680 	if (ethpause.rx_pause && ethpause.tx_pause)
681 		fc_conf->mode = RTE_ETH_FC_FULL;
682 	else if (ethpause.rx_pause)
683 		fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
684 	else if (ethpause.tx_pause)
685 		fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
686 	else
687 		fc_conf->mode = RTE_ETH_FC_NONE;
688 	return 0;
689 }
690 
691 /**
692  * DPDK callback to modify flow control parameters.
693  *
694  * @param dev
695  *   Pointer to Ethernet device structure.
696  * @param[in] fc_conf
697  *   Flow control parameters.
698  *
699  * @return
700  *   0 on success, a negative errno value otherwise and rte_errno is set.
701  */
702 int
703 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
704 {
705 	struct ifreq ifr;
706 	struct ethtool_pauseparam ethpause = {
707 		.cmd = ETHTOOL_SPAUSEPARAM
708 	};
709 	int ret;
710 
711 	ifr.ifr_data = (void *)&ethpause;
712 	ethpause.autoneg = fc_conf->autoneg;
713 	if (((fc_conf->mode & RTE_ETH_FC_FULL) == RTE_ETH_FC_FULL) ||
714 	    (fc_conf->mode & RTE_ETH_FC_RX_PAUSE))
715 		ethpause.rx_pause = 1;
716 	else
717 		ethpause.rx_pause = 0;
718 
719 	if (((fc_conf->mode & RTE_ETH_FC_FULL) == RTE_ETH_FC_FULL) ||
720 	    (fc_conf->mode & RTE_ETH_FC_TX_PAUSE))
721 		ethpause.tx_pause = 1;
722 	else
723 		ethpause.tx_pause = 0;
724 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
725 	if (ret) {
726 		DRV_LOG(WARNING,
727 			"port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
728 			" failed: %s",
729 			dev->data->port_id, strerror(rte_errno));
730 		return ret;
731 	}
732 	return 0;
733 }
734 
735 /**
736  * Handle asynchronous removal event for entire multiport device.
737  *
738  * @param sh
739  *   Infiniband device shared context.
740  */
741 static void
742 mlx5_dev_interrupt_device_fatal(struct mlx5_dev_ctx_shared *sh)
743 {
744 	uint32_t i;
745 
746 	for (i = 0; i < sh->max_port; ++i) {
747 		struct rte_eth_dev *dev;
748 
749 		if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
750 			/*
751 			 * Or not existing port either no
752 			 * handler installed for this port.
753 			 */
754 			continue;
755 		}
756 		dev = &rte_eth_devices[sh->port[i].ih_port_id];
757 		MLX5_ASSERT(dev);
758 		if (dev->data->dev_conf.intr_conf.rmv)
759 			rte_eth_dev_callback_process
760 				(dev, RTE_ETH_EVENT_INTR_RMV, NULL);
761 	}
762 }
763 
764 static void
765 mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg)
766 {
767 	struct mlx5_dev_ctx_shared *sh = cb_arg;
768 	uint32_t i;
769 	uint32_t if_index;
770 
771 	if (mlx5_nl_parse_link_status_update(hdr, &if_index) < 0)
772 		return;
773 	for (i = 0; i < sh->max_port; i++) {
774 		struct mlx5_dev_shared_port *port = &sh->port[i];
775 		struct rte_eth_dev *dev;
776 		struct mlx5_priv *priv;
777 
778 		if (port->nl_ih_port_id >= RTE_MAX_ETHPORTS)
779 			continue;
780 		dev = &rte_eth_devices[port->nl_ih_port_id];
781 		/* Probing may initiate an LSC before configuration is done. */
782 		if (dev->data->dev_configured &&
783 		    !dev->data->dev_conf.intr_conf.lsc)
784 			break;
785 		priv = dev->data->dev_private;
786 		if (priv->if_index == if_index) {
787 			/* Block logical LSC events. */
788 			uint16_t prev_status = dev->data->dev_link.link_status;
789 
790 			if (mlx5_link_update(dev, 0) < 0)
791 				DRV_LOG(ERR, "Failed to update link status: %s",
792 					rte_strerror(rte_errno));
793 			else if (prev_status != dev->data->dev_link.link_status)
794 				rte_eth_dev_callback_process
795 					(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
796 			break;
797 		}
798 	}
799 }
800 
801 void
802 mlx5_dev_interrupt_handler_nl(void *arg)
803 {
804 	struct mlx5_dev_ctx_shared *sh = arg;
805 	int nlsk_fd = rte_intr_fd_get(sh->intr_handle_nl);
806 
807 	if (nlsk_fd < 0)
808 		return;
809 	if (mlx5_nl_read_events(nlsk_fd, mlx5_dev_interrupt_nl_cb, sh) < 0)
810 		DRV_LOG(ERR, "Failed to process Netlink events: %s",
811 			rte_strerror(rte_errno));
812 }
813 
814 /**
815  * Handle shared asynchronous events the NIC (removal event
816  * and link status change). Supports multiport IB device.
817  *
818  * @param cb_arg
819  *   Callback argument.
820  */
821 void
822 mlx5_dev_interrupt_handler(void *cb_arg)
823 {
824 	struct mlx5_dev_ctx_shared *sh = cb_arg;
825 	struct ibv_async_event event;
826 
827 	/* Read all message from the IB device and acknowledge them. */
828 	for (;;) {
829 		struct rte_eth_dev *dev;
830 		uint32_t tmp;
831 
832 		if (mlx5_glue->get_async_event(sh->cdev->ctx, &event))
833 			break;
834 		/* Retrieve and check IB port index. */
835 		tmp = (uint32_t)event.element.port_num;
836 		if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
837 			/*
838 			 * The DEVICE_FATAL event is called once for
839 			 * entire device without port specifying.
840 			 * We should notify all existing ports.
841 			 */
842 			mlx5_glue->ack_async_event(&event);
843 			mlx5_dev_interrupt_device_fatal(sh);
844 			continue;
845 		}
846 		MLX5_ASSERT(tmp && (tmp <= sh->max_port));
847 		if (!tmp) {
848 			/* Unsupported device level event. */
849 			mlx5_glue->ack_async_event(&event);
850 			DRV_LOG(DEBUG,
851 				"unsupported common event (type %d)",
852 				event.event_type);
853 			continue;
854 		}
855 		if (tmp > sh->max_port) {
856 			/* Invalid IB port index. */
857 			mlx5_glue->ack_async_event(&event);
858 			DRV_LOG(DEBUG,
859 				"cannot handle an event (type %d)"
860 				"due to invalid IB port index (%u)",
861 				event.event_type, tmp);
862 			continue;
863 		}
864 		if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
865 			/* No handler installed. */
866 			mlx5_glue->ack_async_event(&event);
867 			DRV_LOG(DEBUG,
868 				"cannot handle an event (type %d)"
869 				"due to no handler installed for port %u",
870 				event.event_type, tmp);
871 			continue;
872 		}
873 		/* Retrieve ethernet device descriptor. */
874 		tmp = sh->port[tmp - 1].ih_port_id;
875 		dev = &rte_eth_devices[tmp];
876 		MLX5_ASSERT(dev);
877 		DRV_LOG(DEBUG,
878 			"port %u cannot handle an unknown event (type %d)",
879 			dev->data->port_id, event.event_type);
880 		mlx5_glue->ack_async_event(&event);
881 	}
882 }
883 
884 /*
885  * Unregister callback handler safely. The handler may be active
886  * while we are trying to unregister it, in this case code -EAGAIN
887  * is returned by rte_intr_callback_unregister(). This routine checks
888  * the return code and tries to unregister handler again.
889  *
890  * @param handle
891  *   interrupt handle
892  * @param cb_fn
893  *   pointer to callback routine
894  * @cb_arg
895  *   opaque callback parameter
896  */
897 void
898 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
899 			      rte_intr_callback_fn cb_fn, void *cb_arg)
900 {
901 	/*
902 	 * Try to reduce timeout management overhead by not calling
903 	 * the timer related routines on the first iteration. If the
904 	 * unregistering succeeds on first call there will be no
905 	 * timer calls at all.
906 	 */
907 	uint64_t twait = 0;
908 	uint64_t start = 0;
909 
910 	do {
911 		int ret;
912 
913 		ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
914 		if (ret >= 0)
915 			return;
916 		if (ret != -EAGAIN) {
917 			DRV_LOG(INFO, "failed to unregister interrupt"
918 				      " handler (error: %d)", ret);
919 			MLX5_ASSERT(false);
920 			return;
921 		}
922 		if (twait) {
923 			struct timespec onems;
924 
925 			/* Wait one millisecond and try again. */
926 			onems.tv_sec = 0;
927 			onems.tv_nsec = NS_PER_S / MS_PER_S;
928 			nanosleep(&onems, 0);
929 			/* Check whether one second elapsed. */
930 			if ((rte_get_timer_cycles() - start) <= twait)
931 				continue;
932 		} else {
933 			/*
934 			 * We get the amount of timer ticks for one second.
935 			 * If this amount elapsed it means we spent one
936 			 * second in waiting. This branch is executed once
937 			 * on first iteration.
938 			 */
939 			twait = rte_get_timer_hz();
940 			MLX5_ASSERT(twait);
941 		}
942 		/*
943 		 * Timeout elapsed, show message (once a second) and retry.
944 		 * We have no other acceptable option here, if we ignore
945 		 * the unregistering return code the handler will not
946 		 * be unregistered, fd will be closed and we may get the
947 		 * crush. Hanging and messaging in the loop seems not to be
948 		 * the worst choice.
949 		 */
950 		DRV_LOG(INFO, "Retrying to unregister interrupt handler");
951 		start = rte_get_timer_cycles();
952 	} while (true);
953 }
954 
955 /**
956  * Handle DEVX interrupts from the NIC.
957  * This function is probably called from the DPDK host thread.
958  *
959  * @param cb_arg
960  *   Callback argument.
961  */
962 void
963 mlx5_dev_interrupt_handler_devx(void *cb_arg)
964 {
965 #ifndef HAVE_IBV_DEVX_ASYNC
966 	(void)cb_arg;
967 	return;
968 #else
969 	struct mlx5_dev_ctx_shared *sh = cb_arg;
970 	union {
971 		struct mlx5dv_devx_async_cmd_hdr cmd_resp;
972 		uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
973 			    MLX5_ST_SZ_BYTES(traffic_counter) +
974 			    sizeof(struct mlx5dv_devx_async_cmd_hdr)];
975 	} out;
976 	uint8_t *buf = out.buf + sizeof(out.cmd_resp);
977 
978 	while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
979 						   &out.cmd_resp,
980 						   sizeof(out.buf)))
981 		mlx5_flow_async_pool_query_handle
982 			(sh, (uint64_t)out.cmd_resp.wr_id,
983 			 mlx5_devx_get_out_command_status(buf));
984 #endif /* HAVE_IBV_DEVX_ASYNC */
985 }
986 
987 /**
988  * DPDK callback to bring the link DOWN.
989  *
990  * @param dev
991  *   Pointer to Ethernet device structure.
992  *
993  * @return
994  *   0 on success, a negative errno value otherwise and rte_errno is set.
995  */
996 int
997 mlx5_set_link_down(struct rte_eth_dev *dev)
998 {
999 	return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
1000 }
1001 
1002 /**
1003  * DPDK callback to bring the link UP.
1004  *
1005  * @param dev
1006  *   Pointer to Ethernet device structure.
1007  *
1008  * @return
1009  *   0 on success, a negative errno value otherwise and rte_errno is set.
1010  */
1011 int
1012 mlx5_set_link_up(struct rte_eth_dev *dev)
1013 {
1014 	return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
1015 }
1016 
1017 /**
1018  * Check if mlx5 device was removed.
1019  *
1020  * @param dev
1021  *   Pointer to Ethernet device structure.
1022  *
1023  * @return
1024  *   1 when device is removed, otherwise 0.
1025  */
1026 int
1027 mlx5_is_removed(struct rte_eth_dev *dev)
1028 {
1029 	struct ibv_device_attr device_attr;
1030 	struct mlx5_priv *priv = dev->data->dev_private;
1031 
1032 	if (mlx5_glue->query_device(priv->sh->cdev->ctx, &device_attr) == EIO)
1033 		return 1;
1034 	return 0;
1035 }
1036 
1037 /**
1038  * Analyze gathered port parameters via sysfs to recognize master
1039  * and representor devices for E-Switch configuration.
1040  *
1041  * @param[in] device_dir
1042  *   flag of presence of "device" directory under port device key.
1043  * @param[inout] switch_info
1044  *   Port information, including port name as a number and port name
1045  *   type if recognized
1046  *
1047  * @return
1048  *   master and representor flags are set in switch_info according to
1049  *   recognized parameters (if any).
1050  */
1051 static void
1052 mlx5_sysfs_check_switch_info(bool device_dir,
1053 			     struct mlx5_switch_info *switch_info)
1054 {
1055 	switch (switch_info->name_type) {
1056 	case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1057 		/*
1058 		 * Name is not recognized, assume the master,
1059 		 * check the device directory presence.
1060 		 */
1061 		switch_info->master = device_dir;
1062 		break;
1063 	case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1064 		/*
1065 		 * Name is not set, this assumes the legacy naming
1066 		 * schema for master, just check if there is
1067 		 * a device directory.
1068 		 */
1069 		switch_info->master = device_dir;
1070 		break;
1071 	case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1072 		/* New uplink naming schema recognized. */
1073 		switch_info->master = 1;
1074 		break;
1075 	case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1076 		/* Legacy representors naming schema. */
1077 		switch_info->representor = !device_dir;
1078 		break;
1079 	case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
1080 		/* Fallthrough */
1081 	case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1082 		/* Fallthrough */
1083 	case MLX5_PHYS_PORT_NAME_TYPE_PFSF:
1084 		/* New representors naming schema. */
1085 		switch_info->representor = 1;
1086 		break;
1087 	default:
1088 		switch_info->master = device_dir;
1089 		break;
1090 	}
1091 }
1092 
1093 /**
1094  * Get switch information associated with network interface.
1095  *
1096  * @param ifindex
1097  *   Network interface index.
1098  * @param[out] info
1099  *   Switch information object, populated in case of success.
1100  *
1101  * @return
1102  *   0 on success, a negative errno value otherwise and rte_errno is set.
1103  */
1104 int
1105 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1106 {
1107 	char ifname[IF_NAMESIZE];
1108 	char port_name[IF_NAMESIZE];
1109 	FILE *file;
1110 	struct mlx5_switch_info data = {
1111 		.master = 0,
1112 		.representor = 0,
1113 		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1114 		.port_name = 0,
1115 		.switch_id = 0,
1116 	};
1117 	DIR *dir;
1118 	bool port_switch_id_set = false;
1119 	bool device_dir = false;
1120 	char c;
1121 	int ret;
1122 
1123 	if (!if_indextoname(ifindex, ifname)) {
1124 		rte_errno = errno;
1125 		return -rte_errno;
1126 	}
1127 
1128 	MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1129 	      ifname);
1130 	MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1131 	      ifname);
1132 	MKSTR(pci_device, "/sys/class/net/%s/device",
1133 	      ifname);
1134 
1135 	file = fopen(phys_port_name, "rb");
1136 	if (file != NULL) {
1137 		ret = fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", port_name);
1138 		fclose(file);
1139 		if (ret == 1)
1140 			mlx5_translate_port_name(port_name, &data);
1141 	}
1142 	file = fopen(phys_switch_id, "rb");
1143 	if (file == NULL) {
1144 		rte_errno = errno;
1145 		return -rte_errno;
1146 	}
1147 	port_switch_id_set =
1148 		fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1149 		c == '\n';
1150 	fclose(file);
1151 	dir = opendir(pci_device);
1152 	if (dir != NULL) {
1153 		closedir(dir);
1154 		device_dir = true;
1155 	}
1156 	if (port_switch_id_set) {
1157 		/* We have some E-Switch configuration. */
1158 		mlx5_sysfs_check_switch_info(device_dir, &data);
1159 	}
1160 	*info = data;
1161 	MLX5_ASSERT(!(data.master && data.representor));
1162 	if (data.master && data.representor) {
1163 		DRV_LOG(ERR, "ifindex %u device is recognized as master"
1164 			     " and as representor", ifindex);
1165 		rte_errno = ENODEV;
1166 		return -rte_errno;
1167 	}
1168 	return 0;
1169 }
1170 
1171 /**
1172  * Get bond information associated with network interface.
1173  *
1174  * @param pf_ifindex
1175  *   Network interface index of bond slave interface
1176  * @param[out] ifindex
1177  *   Pointer to bond ifindex.
1178  * @param[out] ifname
1179  *   Pointer to bond ifname.
1180  *
1181  * @return
1182  *   0 on success, a negative errno value otherwise and rte_errno is set.
1183  */
1184 int
1185 mlx5_sysfs_bond_info(unsigned int pf_ifindex, unsigned int *ifindex,
1186 		     char *ifname)
1187 {
1188 	char name[IF_NAMESIZE];
1189 	FILE *file;
1190 	unsigned int index;
1191 	int ret;
1192 
1193 	if (!if_indextoname(pf_ifindex, name) || !strlen(name)) {
1194 		rte_errno = errno;
1195 		return -rte_errno;
1196 	}
1197 	MKSTR(bond_if, "/sys/class/net/%s/master/ifindex", name);
1198 	/* read bond ifindex */
1199 	file = fopen(bond_if, "rb");
1200 	if (file == NULL) {
1201 		rte_errno = errno;
1202 		return -rte_errno;
1203 	}
1204 	ret = fscanf(file, "%u", &index);
1205 	fclose(file);
1206 	if (ret <= 0) {
1207 		rte_errno = errno;
1208 		return -rte_errno;
1209 	}
1210 	if (ifindex)
1211 		*ifindex = index;
1212 
1213 	/* read bond device name from symbol link */
1214 	if (ifname) {
1215 		if (!if_indextoname(index, ifname)) {
1216 			rte_errno = errno;
1217 			return -rte_errno;
1218 		}
1219 	}
1220 	return 0;
1221 }
1222 
1223 /**
1224  * DPDK callback to retrieve plug-in module EEPROM information (type and size).
1225  *
1226  * @param dev
1227  *   Pointer to Ethernet device structure.
1228  * @param[out] modinfo
1229  *   Storage for plug-in module EEPROM information.
1230  *
1231  * @return
1232  *   0 on success, a negative errno value otherwise and rte_errno is set.
1233  */
1234 int
1235 mlx5_get_module_info(struct rte_eth_dev *dev,
1236 		     struct rte_eth_dev_module_info *modinfo)
1237 {
1238 	struct ethtool_modinfo info = {
1239 		.cmd = ETHTOOL_GMODULEINFO,
1240 	};
1241 	struct ifreq ifr = (struct ifreq) {
1242 		.ifr_data = (void *)&info,
1243 	};
1244 	int ret = 0;
1245 
1246 	if (!dev) {
1247 		DRV_LOG(WARNING, "missing argument, cannot get module info");
1248 		rte_errno = EINVAL;
1249 		return -rte_errno;
1250 	}
1251 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1252 	if (ret) {
1253 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1254 			dev->data->port_id, strerror(rte_errno));
1255 		return ret;
1256 	}
1257 	modinfo->type = info.type;
1258 	modinfo->eeprom_len = info.eeprom_len;
1259 	return ret;
1260 }
1261 
1262 /**
1263  * DPDK callback to retrieve plug-in module EEPROM data.
1264  *
1265  * @param dev
1266  *   Pointer to Ethernet device structure.
1267  * @param[out] info
1268  *   Storage for plug-in module EEPROM data.
1269  *
1270  * @return
1271  *   0 on success, a negative errno value otherwise and rte_errno is set.
1272  */
1273 int mlx5_get_module_eeprom(struct rte_eth_dev *dev,
1274 			   struct rte_dev_eeprom_info *info)
1275 {
1276 	struct ethtool_eeprom *eeprom;
1277 	struct ifreq ifr;
1278 	int ret = 0;
1279 
1280 	if (!dev) {
1281 		DRV_LOG(WARNING, "missing argument, cannot get module eeprom");
1282 		rte_errno = EINVAL;
1283 		return -rte_errno;
1284 	}
1285 	eeprom = mlx5_malloc(MLX5_MEM_ZERO,
1286 			     (sizeof(struct ethtool_eeprom) + info->length), 0,
1287 			     SOCKET_ID_ANY);
1288 	if (!eeprom) {
1289 		DRV_LOG(WARNING, "port %u cannot allocate memory for "
1290 			"eeprom data", dev->data->port_id);
1291 		rte_errno = ENOMEM;
1292 		return -rte_errno;
1293 	}
1294 	eeprom->cmd = ETHTOOL_GMODULEEEPROM;
1295 	eeprom->offset = info->offset;
1296 	eeprom->len = info->length;
1297 	ifr = (struct ifreq) {
1298 		.ifr_data = (void *)eeprom,
1299 	};
1300 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1301 	if (ret)
1302 		DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s",
1303 			dev->data->port_id, strerror(rte_errno));
1304 	else
1305 		rte_memcpy(info->data, eeprom->data, info->length);
1306 	mlx5_free(eeprom);
1307 	return ret;
1308 }
1309 
1310 /**
1311  * Read device counters table.
1312  *
1313  * @param dev
1314  *   Pointer to Ethernet device.
1315  * @param[in] pf
1316  *   PF index in case of bonding device, -1 otherwise
1317  * @param[out] stats
1318  *   Counters table output buffer.
1319  *
1320  * @return
1321  *   0 on success and stats is filled, negative errno value otherwise and
1322  *   rte_errno is set.
1323  */
1324 static int
1325 _mlx5_os_read_dev_counters(struct rte_eth_dev *dev, int pf, uint64_t *stats)
1326 {
1327 	struct mlx5_priv *priv = dev->data->dev_private;
1328 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1329 	unsigned int i;
1330 	struct ifreq ifr;
1331 	unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t);
1332 	unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz];
1333 	struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf;
1334 	int ret;
1335 
1336 	et_stats->cmd = ETHTOOL_GSTATS;
1337 	et_stats->n_stats = xstats_ctrl->stats_n;
1338 	ifr.ifr_data = (caddr_t)et_stats;
1339 	if (pf >= 0)
1340 		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[pf].ifname,
1341 					   SIOCETHTOOL, &ifr);
1342 	else
1343 		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1344 	if (ret) {
1345 		DRV_LOG(WARNING,
1346 			"port %u unable to read statistic values from device",
1347 			dev->data->port_id);
1348 		return ret;
1349 	}
1350 	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
1351 		if (xstats_ctrl->info[i].dev)
1352 			continue;
1353 		stats[i] += (uint64_t)
1354 			    et_stats->data[xstats_ctrl->dev_table_idx[i]];
1355 	}
1356 	return 0;
1357 }
1358 
1359 /**
1360  * Read device counters.
1361  *
1362  * @param dev
1363  *   Pointer to Ethernet device.
1364  * @param[out] stats
1365  *   Counters table output buffer.
1366  *
1367  * @return
1368  *   0 on success and stats is filled, negative errno value otherwise and
1369  *   rte_errno is set.
1370  */
1371 int
1372 mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats)
1373 {
1374 	struct mlx5_priv *priv = dev->data->dev_private;
1375 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1376 	int ret = 0, i;
1377 
1378 	memset(stats, 0, sizeof(*stats) * xstats_ctrl->mlx5_stats_n);
1379 	/* Read ifreq counters. */
1380 	if (priv->master && priv->pf_bond >= 0) {
1381 		/* Sum xstats from bonding device member ports. */
1382 		for (i = 0; i < priv->sh->bond.n_port; i++) {
1383 			ret = _mlx5_os_read_dev_counters(dev, i, stats);
1384 			if (ret)
1385 				return ret;
1386 		}
1387 	} else {
1388 		ret = _mlx5_os_read_dev_counters(dev, -1, stats);
1389 	}
1390 	/* Read IB counters. */
1391 	for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) {
1392 		if (!xstats_ctrl->info[i].dev)
1393 			continue;
1394 		ret = mlx5_os_read_dev_stat(priv, xstats_ctrl->info[i].ctr_name,
1395 					    &stats[i]);
1396 		/* return last xstats counter if fail to read. */
1397 		if (ret != 0)
1398 			xstats_ctrl->xstats[i] = stats[i];
1399 		else
1400 			stats[i] = xstats_ctrl->xstats[i];
1401 	}
1402 	return ret;
1403 }
1404 
1405 /**
1406  * Query the number of statistics provided by ETHTOOL.
1407  *
1408  * @param dev
1409  *   Pointer to Ethernet device.
1410  *
1411  * @return
1412  *   Number of statistics on success, negative errno value otherwise and
1413  *   rte_errno is set.
1414  */
1415 int
1416 mlx5_os_get_stats_n(struct rte_eth_dev *dev)
1417 {
1418 	struct mlx5_priv *priv = dev->data->dev_private;
1419 	struct ethtool_drvinfo drvinfo;
1420 	struct ifreq ifr;
1421 	int ret;
1422 
1423 	drvinfo.cmd = ETHTOOL_GDRVINFO;
1424 	ifr.ifr_data = (caddr_t)&drvinfo;
1425 	if (priv->master && priv->pf_bond >= 0)
1426 		/* Bonding PF. */
1427 		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname,
1428 					   SIOCETHTOOL, &ifr);
1429 	else
1430 		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1431 	if (ret) {
1432 		DRV_LOG(WARNING, "port %u unable to query number of statistics",
1433 			dev->data->port_id);
1434 		return ret;
1435 	}
1436 	return drvinfo.n_stats;
1437 }
1438 
1439 static const struct mlx5_counter_ctrl mlx5_counters_init[] = {
1440 	{
1441 		.dpdk_name = "rx_unicast_bytes",
1442 		.ctr_name = "rx_vport_unicast_bytes",
1443 	},
1444 	{
1445 		.dpdk_name = "rx_multicast_bytes",
1446 		.ctr_name = "rx_vport_multicast_bytes",
1447 	},
1448 	{
1449 		.dpdk_name = "rx_broadcast_bytes",
1450 		.ctr_name = "rx_vport_broadcast_bytes",
1451 	},
1452 	{
1453 		.dpdk_name = "rx_unicast_packets",
1454 		.ctr_name = "rx_vport_unicast_packets",
1455 	},
1456 	{
1457 		.dpdk_name = "rx_multicast_packets",
1458 		.ctr_name = "rx_vport_multicast_packets",
1459 	},
1460 	{
1461 		.dpdk_name = "rx_broadcast_packets",
1462 		.ctr_name = "rx_vport_broadcast_packets",
1463 	},
1464 	{
1465 		.dpdk_name = "tx_unicast_bytes",
1466 		.ctr_name = "tx_vport_unicast_bytes",
1467 	},
1468 	{
1469 		.dpdk_name = "tx_multicast_bytes",
1470 		.ctr_name = "tx_vport_multicast_bytes",
1471 	},
1472 	{
1473 		.dpdk_name = "tx_broadcast_bytes",
1474 		.ctr_name = "tx_vport_broadcast_bytes",
1475 	},
1476 	{
1477 		.dpdk_name = "tx_unicast_packets",
1478 		.ctr_name = "tx_vport_unicast_packets",
1479 	},
1480 	{
1481 		.dpdk_name = "tx_multicast_packets",
1482 		.ctr_name = "tx_vport_multicast_packets",
1483 	},
1484 	{
1485 		.dpdk_name = "tx_broadcast_packets",
1486 		.ctr_name = "tx_vport_broadcast_packets",
1487 	},
1488 	{
1489 		.dpdk_name = "rx_wqe_errors",
1490 		.ctr_name = "rx_wqe_err",
1491 	},
1492 	{
1493 		.dpdk_name = "rx_phy_crc_errors",
1494 		.ctr_name = "rx_crc_errors_phy",
1495 	},
1496 	{
1497 		.dpdk_name = "rx_phy_in_range_len_errors",
1498 		.ctr_name = "rx_in_range_len_errors_phy",
1499 	},
1500 	{
1501 		.dpdk_name = "rx_phy_symbol_errors",
1502 		.ctr_name = "rx_symbol_err_phy",
1503 	},
1504 	{
1505 		.dpdk_name = "tx_phy_errors",
1506 		.ctr_name = "tx_errors_phy",
1507 	},
1508 	{
1509 		.dpdk_name = "rx_out_of_buffer",
1510 		.ctr_name = "out_of_buffer",
1511 		.dev = 1,
1512 	},
1513 	{
1514 		.dpdk_name = "tx_phy_packets",
1515 		.ctr_name = "tx_packets_phy",
1516 	},
1517 	{
1518 		.dpdk_name = "rx_phy_packets",
1519 		.ctr_name = "rx_packets_phy",
1520 	},
1521 	{
1522 		.dpdk_name = "tx_phy_discard_packets",
1523 		.ctr_name = "tx_discards_phy",
1524 	},
1525 	{
1526 		.dpdk_name = "rx_phy_discard_packets",
1527 		.ctr_name = "rx_discards_phy",
1528 	},
1529 	{
1530 		.dpdk_name = "tx_phy_bytes",
1531 		.ctr_name = "tx_bytes_phy",
1532 	},
1533 	{
1534 		.dpdk_name = "rx_phy_bytes",
1535 		.ctr_name = "rx_bytes_phy",
1536 	},
1537 	/* Representor only */
1538 	{
1539 		.dpdk_name = "rx_vport_packets",
1540 		.ctr_name = "vport_rx_packets",
1541 	},
1542 	{
1543 		.dpdk_name = "rx_vport_bytes",
1544 		.ctr_name = "vport_rx_bytes",
1545 	},
1546 	{
1547 		.dpdk_name = "tx_vport_packets",
1548 		.ctr_name = "vport_tx_packets",
1549 	},
1550 	{
1551 		.dpdk_name = "tx_vport_bytes",
1552 		.ctr_name = "vport_tx_bytes",
1553 	},
1554 };
1555 
1556 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init);
1557 
1558 /**
1559  * Init the structures to read device counters.
1560  *
1561  * @param dev
1562  *   Pointer to Ethernet device.
1563  */
1564 void
1565 mlx5_os_stats_init(struct rte_eth_dev *dev)
1566 {
1567 	struct mlx5_priv *priv = dev->data->dev_private;
1568 	struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl;
1569 	struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl;
1570 	unsigned int i;
1571 	unsigned int j;
1572 	struct ifreq ifr;
1573 	struct ethtool_gstrings *strings = NULL;
1574 	unsigned int dev_stats_n;
1575 	unsigned int str_sz;
1576 	int ret;
1577 
1578 	/* So that it won't aggregate for each init. */
1579 	xstats_ctrl->mlx5_stats_n = 0;
1580 	ret = mlx5_os_get_stats_n(dev);
1581 	if (ret < 0) {
1582 		DRV_LOG(WARNING, "port %u no extended statistics available",
1583 			dev->data->port_id);
1584 		return;
1585 	}
1586 	dev_stats_n = ret;
1587 	/* Allocate memory to grab stat names and values. */
1588 	str_sz = dev_stats_n * ETH_GSTRING_LEN;
1589 	strings = (struct ethtool_gstrings *)
1590 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
1591 			      SOCKET_ID_ANY);
1592 	if (!strings) {
1593 		DRV_LOG(WARNING, "port %u unable to allocate memory for xstats",
1594 		     dev->data->port_id);
1595 		return;
1596 	}
1597 	strings->cmd = ETHTOOL_GSTRINGS;
1598 	strings->string_set = ETH_SS_STATS;
1599 	strings->len = dev_stats_n;
1600 	ifr.ifr_data = (caddr_t)strings;
1601 	if (priv->master && priv->pf_bond >= 0)
1602 		/* Bonding master. */
1603 		ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname,
1604 					   SIOCETHTOOL, &ifr);
1605 	else
1606 		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1607 	if (ret) {
1608 		DRV_LOG(WARNING, "port %u unable to get statistic names",
1609 			dev->data->port_id);
1610 		goto free;
1611 	}
1612 	for (i = 0; i != dev_stats_n; ++i) {
1613 		const char *curr_string = (const char *)
1614 			&strings->data[i * ETH_GSTRING_LEN];
1615 
1616 		for (j = 0; j != xstats_n; ++j) {
1617 			if (!strcmp(mlx5_counters_init[j].ctr_name,
1618 				    curr_string)) {
1619 				unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1620 
1621 				xstats_ctrl->dev_table_idx[idx] = i;
1622 				xstats_ctrl->info[idx] = mlx5_counters_init[j];
1623 				break;
1624 			}
1625 		}
1626 	}
1627 	/* Add dev counters. */
1628 	for (i = 0; i != xstats_n; ++i) {
1629 		if (mlx5_counters_init[i].dev) {
1630 			unsigned int idx = xstats_ctrl->mlx5_stats_n++;
1631 
1632 			xstats_ctrl->info[idx] = mlx5_counters_init[i];
1633 			xstats_ctrl->hw_stats[idx] = 0;
1634 		}
1635 	}
1636 	MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS);
1637 	xstats_ctrl->stats_n = dev_stats_n;
1638 	/* Copy to base at first time. */
1639 	ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base);
1640 	if (ret)
1641 		DRV_LOG(ERR, "port %u cannot read device counters: %s",
1642 			dev->data->port_id, strerror(rte_errno));
1643 	mlx5_os_read_dev_stat(priv, "out_of_buffer", &stats_ctrl->imissed_base);
1644 	stats_ctrl->imissed = 0;
1645 free:
1646 	mlx5_free(strings);
1647 }
1648 
1649 /**
1650  * Get MAC address by querying netdevice.
1651  *
1652  * @param[in] dev
1653  *   Pointer to Ethernet device.
1654  * @param[out] mac
1655  *   MAC address output buffer.
1656  *
1657  * @return
1658  *   0 on success, a negative errno value otherwise and rte_errno is set.
1659  */
1660 int
1661 mlx5_get_mac(struct rte_eth_dev *dev, uint8_t (*mac)[RTE_ETHER_ADDR_LEN])
1662 {
1663 	struct ifreq request;
1664 	int ret;
1665 
1666 	ret = mlx5_ifreq(dev, SIOCGIFHWADDR, &request);
1667 	if (ret)
1668 		return ret;
1669 	memcpy(mac, request.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1670 	return 0;
1671 }
1672 
1673 /*
1674  * Query dropless_rq private flag value provided by ETHTOOL.
1675  *
1676  * @param dev
1677  *   Pointer to Ethernet device.
1678  *
1679  * @return
1680  *   - 0 on success, flag is not set.
1681  *   - 1 on success, flag is set.
1682  *   - negative errno value otherwise and rte_errno is set.
1683  */
1684 int mlx5_get_flag_dropless_rq(struct rte_eth_dev *dev)
1685 {
1686 	struct {
1687 		struct ethtool_sset_info hdr;
1688 		uint32_t buf[1];
1689 	} sset_info;
1690 	struct ethtool_drvinfo drvinfo;
1691 	struct ifreq ifr;
1692 	struct ethtool_gstrings *strings = NULL;
1693 	struct ethtool_value flags;
1694 	const int32_t flag_len = sizeof(flags.data) * CHAR_BIT;
1695 	int32_t str_sz;
1696 	int32_t len;
1697 	int32_t i;
1698 	int ret;
1699 
1700 	sset_info.hdr.cmd = ETHTOOL_GSSET_INFO;
1701 	sset_info.hdr.reserved = 0;
1702 	sset_info.hdr.sset_mask = 1ULL << ETH_SS_PRIV_FLAGS;
1703 	ifr.ifr_data = (caddr_t)&sset_info;
1704 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1705 	if (!ret) {
1706 		const uint32_t *sset_lengths = sset_info.hdr.data;
1707 
1708 		len = sset_info.hdr.sset_mask ? sset_lengths[0] : 0;
1709 	} else if (ret == -EOPNOTSUPP) {
1710 		drvinfo.cmd = ETHTOOL_GDRVINFO;
1711 		ifr.ifr_data = (caddr_t)&drvinfo;
1712 		ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1713 		if (ret) {
1714 			DRV_LOG(WARNING, "port %u cannot get the driver info",
1715 				dev->data->port_id);
1716 			goto exit;
1717 		}
1718 		len = *(uint32_t *)((char *)&drvinfo +
1719 			offsetof(struct ethtool_drvinfo, n_priv_flags));
1720 	} else {
1721 		DRV_LOG(WARNING, "port %u cannot get the sset info",
1722 			dev->data->port_id);
1723 		goto exit;
1724 	}
1725 	if (!len) {
1726 		DRV_LOG(WARNING, "port %u does not have private flag",
1727 			dev->data->port_id);
1728 		rte_errno = EOPNOTSUPP;
1729 		ret = -rte_errno;
1730 		goto exit;
1731 	} else if (len > flag_len) {
1732 		DRV_LOG(WARNING, "port %u maximal private flags number is %d",
1733 			dev->data->port_id, flag_len);
1734 		len = flag_len;
1735 	}
1736 	str_sz = ETH_GSTRING_LEN * len;
1737 	strings = (struct ethtool_gstrings *)
1738 		  mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0,
1739 			      SOCKET_ID_ANY);
1740 	if (!strings) {
1741 		DRV_LOG(WARNING, "port %u unable to allocate memory for"
1742 			" private flags", dev->data->port_id);
1743 		rte_errno = ENOMEM;
1744 		ret = -rte_errno;
1745 		goto exit;
1746 	}
1747 	strings->cmd = ETHTOOL_GSTRINGS;
1748 	strings->string_set = ETH_SS_PRIV_FLAGS;
1749 	strings->len = len;
1750 	ifr.ifr_data = (caddr_t)strings;
1751 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1752 	if (ret) {
1753 		DRV_LOG(WARNING, "port %u unable to get private flags strings",
1754 			dev->data->port_id);
1755 		goto exit;
1756 	}
1757 	for (i = 0; i < len; i++) {
1758 		strings->data[(i + 1) * ETH_GSTRING_LEN - 1] = 0;
1759 		if (!strcmp((const char *)strings->data + i * ETH_GSTRING_LEN,
1760 			     "dropless_rq"))
1761 			break;
1762 	}
1763 	if (i == len) {
1764 		DRV_LOG(WARNING, "port %u does not support dropless_rq",
1765 			dev->data->port_id);
1766 		rte_errno = EOPNOTSUPP;
1767 		ret = -rte_errno;
1768 		goto exit;
1769 	}
1770 	flags.cmd = ETHTOOL_GPFLAGS;
1771 	ifr.ifr_data = (caddr_t)&flags;
1772 	ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1773 	if (ret) {
1774 		DRV_LOG(WARNING, "port %u unable to get private flags status",
1775 			dev->data->port_id);
1776 		goto exit;
1777 	}
1778 	ret = !!(flags.data & (1U << i));
1779 exit:
1780 	mlx5_free(strings);
1781 	return ret;
1782 }
1783