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 <bus_pci_driver.h> 29 #include <rte_mbuf.h> 30 #include <rte_common.h> 31 #include <rte_eal_paging.h> 32 #include <rte_interrupts.h> 33 #include <rte_malloc.h> 34 #include <rte_string_fns.h> 35 #include <rte_rwlock.h> 36 #include <rte_cycles.h> 37 38 #include <mlx5_glue.h> 39 #include <mlx5_devx_cmds.h> 40 #include <mlx5_common.h> 41 #include <mlx5_malloc.h> 42 #include <mlx5_nl.h> 43 44 #include "mlx5.h" 45 #include "mlx5_rxtx.h" 46 #include "mlx5_utils.h" 47 48 /* Supported speed values found in /usr/include/linux/ethtool.h */ 49 #ifndef HAVE_SUPPORTED_40000baseKR4_Full 50 #define SUPPORTED_40000baseKR4_Full (1 << 23) 51 #endif 52 #ifndef HAVE_SUPPORTED_40000baseCR4_Full 53 #define SUPPORTED_40000baseCR4_Full (1 << 24) 54 #endif 55 #ifndef HAVE_SUPPORTED_40000baseSR4_Full 56 #define SUPPORTED_40000baseSR4_Full (1 << 25) 57 #endif 58 #ifndef HAVE_SUPPORTED_40000baseLR4_Full 59 #define SUPPORTED_40000baseLR4_Full (1 << 26) 60 #endif 61 #ifndef HAVE_SUPPORTED_56000baseKR4_Full 62 #define SUPPORTED_56000baseKR4_Full (1 << 27) 63 #endif 64 #ifndef HAVE_SUPPORTED_56000baseCR4_Full 65 #define SUPPORTED_56000baseCR4_Full (1 << 28) 66 #endif 67 #ifndef HAVE_SUPPORTED_56000baseSR4_Full 68 #define SUPPORTED_56000baseSR4_Full (1 << 29) 69 #endif 70 #ifndef HAVE_SUPPORTED_56000baseLR4_Full 71 #define SUPPORTED_56000baseLR4_Full (1 << 30) 72 #endif 73 74 /* Add defines in case the running kernel is not the same as user headers. */ 75 #ifndef ETHTOOL_GLINKSETTINGS 76 struct ethtool_link_settings { 77 uint32_t cmd; 78 uint32_t speed; 79 uint8_t duplex; 80 uint8_t port; 81 uint8_t phy_address; 82 uint8_t autoneg; 83 uint8_t mdio_support; 84 uint8_t eth_to_mdix; 85 uint8_t eth_tp_mdix_ctrl; 86 int8_t link_mode_masks_nwords; 87 uint32_t reserved[8]; 88 uint32_t link_mode_masks[]; 89 }; 90 91 /* The kernel values can be found in /include/uapi/linux/ethtool.h */ 92 #define ETHTOOL_GLINKSETTINGS 0x0000004c 93 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5 94 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6 95 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17 96 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18 97 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19 98 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20 99 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21 100 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22 101 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23 102 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24 103 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25 104 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26 105 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27 106 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28 107 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29 108 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30 109 #endif 110 #ifndef HAVE_ETHTOOL_LINK_MODE_25G 111 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31 112 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32 113 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33 114 #endif 115 #ifndef HAVE_ETHTOOL_LINK_MODE_50G 116 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34 117 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35 118 #endif 119 #ifndef HAVE_ETHTOOL_LINK_MODE_100G 120 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36 121 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37 122 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38 123 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39 124 #endif 125 #ifndef HAVE_ETHTOOL_LINK_MODE_200G 126 #define ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT 62 127 #define ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT 63 128 #define ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT 0 /* 64 - 64 */ 129 #define ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT 1 /* 65 - 64 */ 130 #define ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT 2 /* 66 - 64 */ 131 #endif 132 133 /* Get interface index from SubFunction device name. */ 134 int 135 mlx5_auxiliary_get_ifindex(const char *sf_name) 136 { 137 char if_name[IF_NAMESIZE] = { 0 }; 138 139 if (mlx5_auxiliary_get_child_name(sf_name, "/net", 140 if_name, sizeof(if_name)) != 0) 141 return -rte_errno; 142 return if_nametoindex(if_name); 143 } 144 145 /** 146 * Get interface name from private structure. 147 * 148 * This is a port representor-aware version of mlx5_get_ifname_sysfs(). 149 * 150 * @param[in] dev 151 * Pointer to Ethernet device. 152 * @param[out] ifname 153 * Interface name output buffer. 154 * 155 * @return 156 * 0 on success, a negative errno value otherwise and rte_errno is set. 157 */ 158 int 159 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[MLX5_NAMESIZE]) 160 { 161 struct mlx5_priv *priv = dev->data->dev_private; 162 unsigned int ifindex; 163 164 MLX5_ASSERT(priv); 165 MLX5_ASSERT(priv->sh); 166 if (priv->master && priv->sh->bond.ifindex > 0) { 167 memcpy(ifname, priv->sh->bond.ifname, MLX5_NAMESIZE); 168 return 0; 169 } 170 ifindex = mlx5_ifindex(dev); 171 if (!ifindex) { 172 if (!priv->representor) 173 return mlx5_get_ifname_sysfs(priv->sh->ibdev_path, 174 *ifname); 175 rte_errno = ENXIO; 176 return -rte_errno; 177 } 178 if (if_indextoname(ifindex, &(*ifname)[0])) 179 return 0; 180 rte_errno = errno; 181 return -rte_errno; 182 } 183 184 /** 185 * Perform ifreq ioctl() on associated netdev ifname. 186 * 187 * @param[in] ifname 188 * Pointer to netdev name. 189 * @param req 190 * Request number to pass to ioctl(). 191 * @param[out] ifr 192 * Interface request structure output buffer. 193 * 194 * @return 195 * 0 on success, a negative errno value otherwise and rte_errno is set. 196 */ 197 static int 198 mlx5_ifreq_by_ifname(const char *ifname, int req, struct ifreq *ifr) 199 { 200 int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); 201 int ret = 0; 202 203 if (sock == -1) { 204 rte_errno = errno; 205 return -rte_errno; 206 } 207 rte_strscpy(ifr->ifr_name, ifname, sizeof(ifr->ifr_name)); 208 ret = ioctl(sock, req, ifr); 209 if (ret == -1) { 210 rte_errno = errno; 211 goto error; 212 } 213 close(sock); 214 return 0; 215 error: 216 close(sock); 217 return -rte_errno; 218 } 219 220 /** 221 * Perform ifreq ioctl() on associated Ethernet device. 222 * 223 * @param[in] dev 224 * Pointer to Ethernet device. 225 * @param req 226 * Request number to pass to ioctl(). 227 * @param[out] ifr 228 * Interface request structure output buffer. 229 * 230 * @return 231 * 0 on success, a negative errno value otherwise and rte_errno is set. 232 */ 233 static int 234 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr) 235 { 236 char ifname[sizeof(ifr->ifr_name)]; 237 int ret; 238 239 ret = mlx5_get_ifname(dev, &ifname); 240 if (ret) 241 return -rte_errno; 242 return mlx5_ifreq_by_ifname(ifname, req, ifr); 243 } 244 245 /** 246 * Get device MTU. 247 * 248 * @param dev 249 * Pointer to Ethernet device. 250 * @param[out] mtu 251 * MTU value output buffer. 252 * 253 * @return 254 * 0 on success, a negative errno value otherwise and rte_errno is set. 255 */ 256 int 257 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu) 258 { 259 struct ifreq request; 260 int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request); 261 262 if (ret) 263 return ret; 264 *mtu = request.ifr_mtu; 265 return 0; 266 } 267 268 /** 269 * Set device MTU. 270 * 271 * @param dev 272 * Pointer to Ethernet device. 273 * @param mtu 274 * MTU value to set. 275 * 276 * @return 277 * 0 on success, a negative errno value otherwise and rte_errno is set. 278 */ 279 int 280 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) 281 { 282 struct ifreq request = { .ifr_mtu = mtu, }; 283 284 return mlx5_ifreq(dev, SIOCSIFMTU, &request); 285 } 286 287 /** 288 * Set device flags. 289 * 290 * @param dev 291 * Pointer to Ethernet device. 292 * @param keep 293 * Bitmask for flags that must remain untouched. 294 * @param flags 295 * Bitmask for flags to modify. 296 * 297 * @return 298 * 0 on success, a negative errno value otherwise and rte_errno is set. 299 */ 300 static int 301 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags) 302 { 303 struct ifreq request; 304 int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request); 305 306 if (ret) 307 return ret; 308 request.ifr_flags &= keep; 309 request.ifr_flags |= flags & ~keep; 310 return mlx5_ifreq(dev, SIOCSIFFLAGS, &request); 311 } 312 313 /** 314 * Get device current raw clock counter 315 * 316 * @param dev 317 * Pointer to Ethernet device structure. 318 * @param[out] time 319 * Current raw clock counter of the device. 320 * 321 * @return 322 * 0 if the clock has correctly been read 323 * The value of errno in case of error 324 */ 325 int 326 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock) 327 { 328 struct mlx5_priv *priv = dev->data->dev_private; 329 struct ibv_context *ctx = priv->sh->cdev->ctx; 330 struct ibv_values_ex values; 331 int err = 0; 332 333 values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK; 334 err = mlx5_glue->query_rt_values_ex(ctx, &values); 335 if (err != 0) { 336 DRV_LOG(WARNING, "Could not query the clock !"); 337 return err; 338 } 339 *clock = values.raw_clock.tv_nsec; 340 return 0; 341 } 342 343 /** 344 * Retrieve the master device for representor in the same switch domain. 345 * 346 * @param dev 347 * Pointer to representor Ethernet device structure. 348 * 349 * @return 350 * Master device structure on success, NULL otherwise. 351 */ 352 static struct rte_eth_dev * 353 mlx5_find_master_dev(struct rte_eth_dev *dev) 354 { 355 struct mlx5_priv *priv; 356 uint16_t port_id; 357 uint16_t domain_id; 358 359 priv = dev->data->dev_private; 360 domain_id = priv->domain_id; 361 MLX5_ASSERT(priv->representor); 362 MLX5_ETH_FOREACH_DEV(port_id, dev->device) { 363 struct mlx5_priv *opriv = 364 rte_eth_devices[port_id].data->dev_private; 365 if (opriv && 366 opriv->master && 367 opriv->domain_id == domain_id && 368 opriv->sh == priv->sh) 369 return &rte_eth_devices[port_id]; 370 } 371 return NULL; 372 } 373 374 /** 375 * DPDK callback to retrieve physical link information. 376 * 377 * @param dev 378 * Pointer to Ethernet device structure. 379 * @param[out] link 380 * Storage for current link status. 381 * 382 * @return 383 * 0 on success, a negative errno value otherwise and rte_errno is set. 384 */ 385 static int 386 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, 387 struct rte_eth_link *link) 388 { 389 struct mlx5_priv *priv = dev->data->dev_private; 390 struct ethtool_cmd edata = { 391 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */ 392 }; 393 struct ifreq ifr; 394 struct rte_eth_link dev_link; 395 int link_speed = 0; 396 int ret; 397 398 ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr); 399 if (ret) { 400 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s", 401 dev->data->port_id, strerror(rte_errno)); 402 return ret; 403 } 404 dev_link = (struct rte_eth_link) { 405 .link_status = ((ifr.ifr_flags & IFF_UP) && 406 (ifr.ifr_flags & IFF_RUNNING)), 407 }; 408 ifr = (struct ifreq) { 409 .ifr_data = (void *)&edata, 410 }; 411 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 412 if (ret) { 413 if (ret == -ENOTSUP && priv->representor) { 414 struct rte_eth_dev *master; 415 416 /* 417 * For representors we can try to inherit link 418 * settings from the master device. Actually 419 * link settings do not make a lot of sense 420 * for representors due to missing physical 421 * link. The old kernel drivers supported 422 * emulated settings query for representors, 423 * the new ones do not, so we have to add 424 * this code for compatibility issues. 425 */ 426 master = mlx5_find_master_dev(dev); 427 if (master) { 428 ifr = (struct ifreq) { 429 .ifr_data = (void *)&edata, 430 }; 431 ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr); 432 } 433 } 434 if (ret) { 435 DRV_LOG(WARNING, 436 "port %u ioctl(SIOCETHTOOL," 437 " ETHTOOL_GSET) failed: %s", 438 dev->data->port_id, strerror(rte_errno)); 439 return ret; 440 } 441 } 442 link_speed = ethtool_cmd_speed(&edata); 443 if (link_speed == -1) 444 dev_link.link_speed = RTE_ETH_SPEED_NUM_UNKNOWN; 445 else 446 dev_link.link_speed = link_speed; 447 priv->link_speed_capa = 0; 448 if (edata.supported & (SUPPORTED_1000baseT_Full | 449 SUPPORTED_1000baseKX_Full)) 450 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_1G; 451 if (edata.supported & SUPPORTED_10000baseKR_Full) 452 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_10G; 453 if (edata.supported & (SUPPORTED_40000baseKR4_Full | 454 SUPPORTED_40000baseCR4_Full | 455 SUPPORTED_40000baseSR4_Full | 456 SUPPORTED_40000baseLR4_Full)) 457 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_40G; 458 dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ? 459 RTE_ETH_LINK_HALF_DUPLEX : RTE_ETH_LINK_FULL_DUPLEX); 460 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds & 461 RTE_ETH_LINK_SPEED_FIXED); 462 *link = dev_link; 463 return 0; 464 } 465 466 /** 467 * Retrieve physical link information (unlocked version using new ioctl). 468 * 469 * @param dev 470 * Pointer to Ethernet device structure. 471 * @param[out] link 472 * Storage for current link status. 473 * 474 * @return 475 * 0 on success, a negative errno value otherwise and rte_errno is set. 476 */ 477 static int 478 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, 479 struct rte_eth_link *link) 480 481 { 482 struct mlx5_priv *priv = dev->data->dev_private; 483 struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS }; 484 struct ifreq ifr; 485 struct rte_eth_link dev_link; 486 struct rte_eth_dev *master = NULL; 487 uint64_t sc; 488 int ret; 489 490 ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr); 491 if (ret) { 492 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s", 493 dev->data->port_id, strerror(rte_errno)); 494 return ret; 495 } 496 dev_link = (struct rte_eth_link) { 497 .link_status = ((ifr.ifr_flags & IFF_UP) && 498 (ifr.ifr_flags & IFF_RUNNING)), 499 }; 500 ifr = (struct ifreq) { 501 .ifr_data = (void *)&gcmd, 502 }; 503 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 504 if (ret) { 505 if (ret == -ENOTSUP && priv->representor) { 506 /* 507 * For representors we can try to inherit link 508 * settings from the master device. Actually 509 * link settings do not make a lot of sense 510 * for representors due to missing physical 511 * link. The old kernel drivers supported 512 * emulated settings query for representors, 513 * the new ones do not, so we have to add 514 * this code for compatibility issues. 515 */ 516 master = mlx5_find_master_dev(dev); 517 if (master) { 518 ifr = (struct ifreq) { 519 .ifr_data = (void *)&gcmd, 520 }; 521 ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr); 522 } 523 } 524 if (ret) { 525 DRV_LOG(DEBUG, 526 "port %u ioctl(SIOCETHTOOL," 527 " ETHTOOL_GLINKSETTINGS) failed: %s", 528 dev->data->port_id, strerror(rte_errno)); 529 return ret; 530 } 531 } 532 gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords; 533 534 alignas(struct ethtool_link_settings) 535 uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) + 536 sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3]; 537 struct ethtool_link_settings *ecmd = (void *)data; 538 539 *ecmd = gcmd; 540 ifr.ifr_data = (void *)ecmd; 541 ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr); 542 if (ret) { 543 DRV_LOG(DEBUG, 544 "port %u ioctl(SIOCETHTOOL," 545 "ETHTOOL_GLINKSETTINGS) failed: %s", 546 dev->data->port_id, strerror(rte_errno)); 547 return ret; 548 } 549 dev_link.link_speed = (ecmd->speed == UINT32_MAX) ? 550 RTE_ETH_SPEED_NUM_UNKNOWN : ecmd->speed; 551 sc = ecmd->link_mode_masks[0] | 552 ((uint64_t)ecmd->link_mode_masks[1] << 32); 553 priv->link_speed_capa = 0; 554 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) | 555 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))) 556 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_1G; 557 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) | 558 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) | 559 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))) 560 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_10G; 561 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) | 562 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))) 563 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_20G; 564 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) | 565 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) | 566 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) | 567 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))) 568 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_40G; 569 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) | 570 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) | 571 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) | 572 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))) 573 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_56G; 574 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) | 575 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) | 576 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))) 577 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_25G; 578 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) | 579 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))) 580 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_50G; 581 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) | 582 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) | 583 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) | 584 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))) 585 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_100G; 586 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseKR4_Full_BIT) | 587 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseSR4_Full_BIT))) 588 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_200G; 589 590 sc = ecmd->link_mode_masks[2] | 591 ((uint64_t)ecmd->link_mode_masks[3] << 32); 592 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseCR4_Full_BIT) | 593 MLX5_BITSHIFT 594 (ETHTOOL_LINK_MODE_200000baseLR4_ER4_FR4_Full_BIT) | 595 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_200000baseDR4_Full_BIT))) 596 priv->link_speed_capa |= RTE_ETH_LINK_SPEED_200G; 597 dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ? 598 RTE_ETH_LINK_HALF_DUPLEX : RTE_ETH_LINK_FULL_DUPLEX); 599 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds & 600 RTE_ETH_LINK_SPEED_FIXED); 601 *link = dev_link; 602 return 0; 603 } 604 605 /** 606 * DPDK callback to retrieve physical link information. 607 * 608 * @param dev 609 * Pointer to Ethernet device structure. 610 * @param wait_to_complete 611 * Wait for request completion. 612 * 613 * @return 614 * 0 if link status was not updated, positive if it was, a negative errno 615 * value otherwise and rte_errno is set. 616 */ 617 int 618 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete) 619 { 620 int ret; 621 struct rte_eth_link dev_link; 622 time_t start_time = time(NULL); 623 int retry = MLX5_GET_LINK_STATUS_RETRY_COUNT; 624 625 do { 626 ret = mlx5_link_update_unlocked_gs(dev, &dev_link); 627 if (ret == -ENOTSUP) 628 ret = mlx5_link_update_unlocked_gset(dev, &dev_link); 629 if (ret == 0) 630 break; 631 /* Handle wait to complete situation. */ 632 if ((wait_to_complete || retry) && ret == -EAGAIN) { 633 if (abs((int)difftime(time(NULL), start_time)) < 634 MLX5_LINK_STATUS_TIMEOUT) { 635 usleep(0); 636 continue; 637 } else { 638 rte_errno = EBUSY; 639 return -rte_errno; 640 } 641 } else if (ret < 0) { 642 return ret; 643 } 644 } while (wait_to_complete || retry-- > 0); 645 ret = !!memcmp(&dev->data->dev_link, &dev_link, 646 sizeof(struct rte_eth_link)); 647 dev->data->dev_link = dev_link; 648 return ret; 649 } 650 651 /** 652 * DPDK callback to get flow control status. 653 * 654 * @param dev 655 * Pointer to Ethernet device structure. 656 * @param[out] fc_conf 657 * Flow control output buffer. 658 * 659 * @return 660 * 0 on success, a negative errno value otherwise and rte_errno is set. 661 */ 662 int 663 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) 664 { 665 struct ifreq ifr; 666 struct ethtool_pauseparam ethpause = { 667 .cmd = ETHTOOL_GPAUSEPARAM 668 }; 669 int ret; 670 671 ifr.ifr_data = (void *)ðpause; 672 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 673 if (ret) { 674 DRV_LOG(WARNING, 675 "port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:" 676 " %s", 677 dev->data->port_id, strerror(rte_errno)); 678 return ret; 679 } 680 fc_conf->autoneg = ethpause.autoneg; 681 if (ethpause.rx_pause && ethpause.tx_pause) 682 fc_conf->mode = RTE_ETH_FC_FULL; 683 else if (ethpause.rx_pause) 684 fc_conf->mode = RTE_ETH_FC_RX_PAUSE; 685 else if (ethpause.tx_pause) 686 fc_conf->mode = RTE_ETH_FC_TX_PAUSE; 687 else 688 fc_conf->mode = RTE_ETH_FC_NONE; 689 return 0; 690 } 691 692 /** 693 * DPDK callback to modify flow control parameters. 694 * 695 * @param dev 696 * Pointer to Ethernet device structure. 697 * @param[in] fc_conf 698 * Flow control parameters. 699 * 700 * @return 701 * 0 on success, a negative errno value otherwise and rte_errno is set. 702 */ 703 int 704 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) 705 { 706 struct ifreq ifr; 707 struct ethtool_pauseparam ethpause = { 708 .cmd = ETHTOOL_SPAUSEPARAM 709 }; 710 int ret; 711 712 ifr.ifr_data = (void *)ðpause; 713 ethpause.autoneg = fc_conf->autoneg; 714 if (((fc_conf->mode & RTE_ETH_FC_FULL) == RTE_ETH_FC_FULL) || 715 (fc_conf->mode & RTE_ETH_FC_RX_PAUSE)) 716 ethpause.rx_pause = 1; 717 else 718 ethpause.rx_pause = 0; 719 720 if (((fc_conf->mode & RTE_ETH_FC_FULL) == RTE_ETH_FC_FULL) || 721 (fc_conf->mode & RTE_ETH_FC_TX_PAUSE)) 722 ethpause.tx_pause = 1; 723 else 724 ethpause.tx_pause = 0; 725 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 726 if (ret) { 727 DRV_LOG(WARNING, 728 "port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)" 729 " failed: %s", 730 dev->data->port_id, strerror(rte_errno)); 731 return ret; 732 } 733 return 0; 734 } 735 736 /** 737 * Handle asynchronous removal event for entire multiport device. 738 * 739 * @param sh 740 * Infiniband device shared context. 741 */ 742 static void 743 mlx5_dev_interrupt_device_fatal(struct mlx5_dev_ctx_shared *sh) 744 { 745 uint32_t i; 746 747 for (i = 0; i < sh->max_port; ++i) { 748 struct rte_eth_dev *dev; 749 struct mlx5_priv *priv; 750 751 if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) { 752 /* 753 * Or not existing port either no 754 * handler installed for this port. 755 */ 756 continue; 757 } 758 dev = &rte_eth_devices[sh->port[i].ih_port_id]; 759 MLX5_ASSERT(dev); 760 priv = dev->data->dev_private; 761 MLX5_ASSERT(priv); 762 if (!priv->rmv_notified && dev->data->dev_conf.intr_conf.rmv) { 763 /* Notify driver about removal only once. */ 764 priv->rmv_notified = 1; 765 rte_eth_dev_callback_process 766 (dev, RTE_ETH_EVENT_INTR_RMV, NULL); 767 } 768 } 769 } 770 771 static void 772 mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg) 773 { 774 struct mlx5_dev_ctx_shared *sh = cb_arg; 775 uint32_t i; 776 uint32_t if_index; 777 778 if (mlx5_nl_parse_link_status_update(hdr, &if_index) < 0) 779 return; 780 for (i = 0; i < sh->max_port; i++) { 781 struct mlx5_dev_shared_port *port = &sh->port[i]; 782 struct rte_eth_dev *dev; 783 struct mlx5_priv *priv; 784 785 if (port->nl_ih_port_id >= RTE_MAX_ETHPORTS) 786 continue; 787 dev = &rte_eth_devices[port->nl_ih_port_id]; 788 /* Probing may initiate an LSC before configuration is done. */ 789 if (dev->data->dev_configured && 790 !dev->data->dev_conf.intr_conf.lsc) 791 break; 792 priv = dev->data->dev_private; 793 if (priv->if_index == if_index) { 794 /* Block logical LSC events. */ 795 uint16_t prev_status = dev->data->dev_link.link_status; 796 797 if (mlx5_link_update(dev, 0) < 0) 798 DRV_LOG(ERR, "Failed to update link status: %s", 799 rte_strerror(rte_errno)); 800 else if (prev_status != dev->data->dev_link.link_status) 801 rte_eth_dev_callback_process 802 (dev, RTE_ETH_EVENT_INTR_LSC, NULL); 803 break; 804 } 805 } 806 } 807 808 void 809 mlx5_dev_interrupt_handler_nl(void *arg) 810 { 811 struct mlx5_dev_ctx_shared *sh = arg; 812 int nlsk_fd = rte_intr_fd_get(sh->intr_handle_nl); 813 814 if (nlsk_fd < 0) 815 return; 816 if (mlx5_nl_read_events(nlsk_fd, mlx5_dev_interrupt_nl_cb, sh) < 0) 817 DRV_LOG(ERR, "Failed to process Netlink events: %s", 818 rte_strerror(rte_errno)); 819 } 820 821 /** 822 * Handle shared asynchronous events the NIC (removal event 823 * and link status change). Supports multiport IB device. 824 * 825 * @param cb_arg 826 * Callback argument. 827 */ 828 void 829 mlx5_dev_interrupt_handler(void *cb_arg) 830 { 831 struct mlx5_dev_ctx_shared *sh = cb_arg; 832 struct ibv_async_event event; 833 834 /* Read all message from the IB device and acknowledge them. */ 835 for (;;) { 836 struct rte_eth_dev *dev; 837 uint32_t tmp; 838 839 if (mlx5_glue->get_async_event(sh->cdev->ctx, &event)) { 840 if (errno == EIO) { 841 DRV_LOG(DEBUG, 842 "IBV async event queue closed on: %s", 843 sh->ibdev_name); 844 mlx5_dev_interrupt_device_fatal(sh); 845 } 846 break; 847 } 848 if (event.event_type == IBV_EVENT_DEVICE_FATAL) { 849 /* 850 * The DEVICE_FATAL event can be called by kernel 851 * twice - from mlx5 and uverbs layers, and port 852 * index is not applicable. We should notify all 853 * existing ports. 854 */ 855 mlx5_dev_interrupt_device_fatal(sh); 856 mlx5_glue->ack_async_event(&event); 857 continue; 858 } 859 /* Retrieve and check IB port index. */ 860 tmp = (uint32_t)event.element.port_num; 861 MLX5_ASSERT(tmp <= sh->max_port); 862 if (!tmp) { 863 /* Unsupported device level event. */ 864 mlx5_glue->ack_async_event(&event); 865 DRV_LOG(DEBUG, 866 "unsupported common event (type %d)", 867 event.event_type); 868 continue; 869 } 870 if (tmp > sh->max_port) { 871 /* Invalid IB port index. */ 872 mlx5_glue->ack_async_event(&event); 873 DRV_LOG(DEBUG, 874 "cannot handle an event (type %d)" 875 "due to invalid IB port index (%u)", 876 event.event_type, tmp); 877 continue; 878 } 879 if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) { 880 /* No handler installed. */ 881 mlx5_glue->ack_async_event(&event); 882 DRV_LOG(DEBUG, 883 "cannot handle an event (type %d)" 884 "due to no handler installed for port %u", 885 event.event_type, tmp); 886 continue; 887 } 888 /* Retrieve ethernet device descriptor. */ 889 tmp = sh->port[tmp - 1].ih_port_id; 890 dev = &rte_eth_devices[tmp]; 891 MLX5_ASSERT(dev); 892 DRV_LOG(DEBUG, 893 "port %u cannot handle an unknown event (type %d)", 894 dev->data->port_id, event.event_type); 895 mlx5_glue->ack_async_event(&event); 896 } 897 } 898 899 /** 900 * Handle DEVX interrupts from the NIC. 901 * This function is probably called from the DPDK host thread. 902 * 903 * @param cb_arg 904 * Callback argument. 905 */ 906 void 907 mlx5_dev_interrupt_handler_devx(void *cb_arg) 908 { 909 #ifndef HAVE_IBV_DEVX_ASYNC 910 (void)cb_arg; 911 return; 912 #else 913 struct mlx5_dev_ctx_shared *sh = cb_arg; 914 union { 915 struct mlx5dv_devx_async_cmd_hdr cmd_resp; 916 uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) + 917 MLX5_ST_SZ_BYTES(traffic_counter) + 918 sizeof(struct mlx5dv_devx_async_cmd_hdr)]; 919 } out; 920 uint8_t *buf = out.buf + sizeof(out.cmd_resp); 921 922 while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp, 923 &out.cmd_resp, 924 sizeof(out.buf))) 925 mlx5_flow_async_pool_query_handle 926 (sh, (uint64_t)out.cmd_resp.wr_id, 927 mlx5_devx_get_out_command_status(buf)); 928 #endif /* HAVE_IBV_DEVX_ASYNC */ 929 } 930 931 /** 932 * DPDK callback to bring the link DOWN. 933 * 934 * @param dev 935 * Pointer to Ethernet device structure. 936 * 937 * @return 938 * 0 on success, a negative errno value otherwise and rte_errno is set. 939 */ 940 int 941 mlx5_set_link_down(struct rte_eth_dev *dev) 942 { 943 return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP); 944 } 945 946 /** 947 * DPDK callback to bring the link UP. 948 * 949 * @param dev 950 * Pointer to Ethernet device structure. 951 * 952 * @return 953 * 0 on success, a negative errno value otherwise and rte_errno is set. 954 */ 955 int 956 mlx5_set_link_up(struct rte_eth_dev *dev) 957 { 958 return mlx5_set_flags(dev, ~IFF_UP, IFF_UP); 959 } 960 961 /** 962 * Check if mlx5 device was removed. 963 * 964 * @param dev 965 * Pointer to Ethernet device structure. 966 * 967 * @return 968 * 1 when device is removed, otherwise 0. 969 */ 970 int 971 mlx5_is_removed(struct rte_eth_dev *dev) 972 { 973 struct ibv_device_attr device_attr; 974 struct mlx5_priv *priv = dev->data->dev_private; 975 976 if (mlx5_glue->query_device(priv->sh->cdev->ctx, &device_attr) == EIO) 977 return 1; 978 return 0; 979 } 980 981 /** 982 * Analyze gathered port parameters via sysfs to recognize master 983 * and representor devices for E-Switch configuration. 984 * 985 * @param[in] device_dir 986 * flag of presence of "device" directory under port device key. 987 * @param[inout] switch_info 988 * Port information, including port name as a number and port name 989 * type if recognized 990 * 991 * @return 992 * master and representor flags are set in switch_info according to 993 * recognized parameters (if any). 994 */ 995 static void 996 mlx5_sysfs_check_switch_info(bool device_dir, 997 struct mlx5_switch_info *switch_info) 998 { 999 switch (switch_info->name_type) { 1000 case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN: 1001 /* 1002 * Name is not recognized, assume the master, 1003 * check the device directory presence. 1004 */ 1005 switch_info->master = device_dir; 1006 break; 1007 case MLX5_PHYS_PORT_NAME_TYPE_NOTSET: 1008 /* 1009 * Name is not set, this assumes the legacy naming 1010 * schema for master, just check if there is 1011 * a device directory. 1012 */ 1013 switch_info->master = device_dir; 1014 break; 1015 case MLX5_PHYS_PORT_NAME_TYPE_UPLINK: 1016 /* New uplink naming schema recognized. */ 1017 switch_info->master = 1; 1018 break; 1019 case MLX5_PHYS_PORT_NAME_TYPE_LEGACY: 1020 /* Legacy representors naming schema. */ 1021 switch_info->representor = !device_dir; 1022 break; 1023 case MLX5_PHYS_PORT_NAME_TYPE_PFHPF: 1024 /* Fallthrough */ 1025 case MLX5_PHYS_PORT_NAME_TYPE_PFVF: 1026 /* Fallthrough */ 1027 case MLX5_PHYS_PORT_NAME_TYPE_PFSF: 1028 /* New representors naming schema. */ 1029 switch_info->representor = 1; 1030 break; 1031 default: 1032 switch_info->master = device_dir; 1033 break; 1034 } 1035 } 1036 1037 /** 1038 * Get switch information associated with network interface. 1039 * 1040 * @param ifindex 1041 * Network interface index. 1042 * @param[out] info 1043 * Switch information object, populated in case of success. 1044 * 1045 * @return 1046 * 0 on success, a negative errno value otherwise and rte_errno is set. 1047 */ 1048 int 1049 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info) 1050 { 1051 char ifname[IF_NAMESIZE]; 1052 char *port_name = NULL; 1053 size_t port_name_size = 0; 1054 FILE *file; 1055 struct mlx5_switch_info data = { 1056 .master = 0, 1057 .representor = 0, 1058 .name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET, 1059 .port_name = 0, 1060 .switch_id = 0, 1061 }; 1062 DIR *dir; 1063 bool port_switch_id_set = false; 1064 bool device_dir = false; 1065 char c; 1066 ssize_t line_size; 1067 1068 if (!if_indextoname(ifindex, ifname)) { 1069 rte_errno = errno; 1070 return -rte_errno; 1071 } 1072 1073 MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name", 1074 ifname); 1075 MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id", 1076 ifname); 1077 MKSTR(pci_device, "/sys/class/net/%s/device", 1078 ifname); 1079 1080 file = fopen(phys_port_name, "rb"); 1081 if (file != NULL) { 1082 char *tail_nl; 1083 1084 line_size = getline(&port_name, &port_name_size, file); 1085 if (line_size < 0) { 1086 fclose(file); 1087 rte_errno = errno; 1088 return -rte_errno; 1089 } else if (line_size > 0) { 1090 /* Remove tailing newline character. */ 1091 tail_nl = strchr(port_name, '\n'); 1092 if (tail_nl) 1093 *tail_nl = '\0'; 1094 mlx5_translate_port_name(port_name, &data); 1095 } 1096 free(port_name); 1097 fclose(file); 1098 } 1099 file = fopen(phys_switch_id, "rb"); 1100 if (file == NULL) { 1101 rte_errno = errno; 1102 return -rte_errno; 1103 } 1104 port_switch_id_set = 1105 fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 && 1106 c == '\n'; 1107 fclose(file); 1108 dir = opendir(pci_device); 1109 if (dir != NULL) { 1110 closedir(dir); 1111 device_dir = true; 1112 } 1113 if (port_switch_id_set) { 1114 /* We have some E-Switch configuration. */ 1115 mlx5_sysfs_check_switch_info(device_dir, &data); 1116 } 1117 *info = data; 1118 MLX5_ASSERT(!(data.master && data.representor)); 1119 if (data.master && data.representor) { 1120 DRV_LOG(ERR, "ifindex %u device is recognized as master" 1121 " and as representor", ifindex); 1122 rte_errno = ENODEV; 1123 return -rte_errno; 1124 } 1125 return 0; 1126 } 1127 1128 /** 1129 * Get bond information associated with network interface. 1130 * 1131 * @param pf_ifindex 1132 * Network interface index of bond slave interface 1133 * @param[out] ifindex 1134 * Pointer to bond ifindex. 1135 * @param[out] ifname 1136 * Pointer to bond ifname. 1137 * 1138 * @return 1139 * 0 on success, a negative errno value otherwise and rte_errno is set. 1140 */ 1141 int 1142 mlx5_sysfs_bond_info(unsigned int pf_ifindex, unsigned int *ifindex, 1143 char *ifname) 1144 { 1145 char name[IF_NAMESIZE]; 1146 FILE *file; 1147 unsigned int index; 1148 int ret; 1149 1150 if (!if_indextoname(pf_ifindex, name) || !strlen(name)) { 1151 rte_errno = errno; 1152 return -rte_errno; 1153 } 1154 MKSTR(bond_if, "/sys/class/net/%s/master/ifindex", name); 1155 /* read bond ifindex */ 1156 file = fopen(bond_if, "rb"); 1157 if (file == NULL) { 1158 rte_errno = errno; 1159 return -rte_errno; 1160 } 1161 ret = fscanf(file, "%u", &index); 1162 fclose(file); 1163 if (ret <= 0) { 1164 rte_errno = errno; 1165 return -rte_errno; 1166 } 1167 if (ifindex) 1168 *ifindex = index; 1169 1170 /* read bond device name from symbol link */ 1171 if (ifname) { 1172 if (!if_indextoname(index, ifname)) { 1173 rte_errno = errno; 1174 return -rte_errno; 1175 } 1176 } 1177 return 0; 1178 } 1179 1180 /** 1181 * DPDK callback to retrieve plug-in module EEPROM information (type and size). 1182 * 1183 * @param dev 1184 * Pointer to Ethernet device structure. 1185 * @param[out] modinfo 1186 * Storage for plug-in module EEPROM information. 1187 * 1188 * @return 1189 * 0 on success, a negative errno value otherwise and rte_errno is set. 1190 */ 1191 int 1192 mlx5_get_module_info(struct rte_eth_dev *dev, 1193 struct rte_eth_dev_module_info *modinfo) 1194 { 1195 struct ethtool_modinfo info = { 1196 .cmd = ETHTOOL_GMODULEINFO, 1197 }; 1198 struct ifreq ifr = (struct ifreq) { 1199 .ifr_data = (void *)&info, 1200 }; 1201 int ret = 0; 1202 1203 if (!dev) { 1204 DRV_LOG(WARNING, "missing argument, cannot get module info"); 1205 rte_errno = EINVAL; 1206 return -rte_errno; 1207 } 1208 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1209 if (ret) { 1210 DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s", 1211 dev->data->port_id, strerror(rte_errno)); 1212 return ret; 1213 } 1214 modinfo->type = info.type; 1215 modinfo->eeprom_len = info.eeprom_len; 1216 return ret; 1217 } 1218 1219 /** 1220 * DPDK callback to retrieve plug-in module EEPROM data. 1221 * 1222 * @param dev 1223 * Pointer to Ethernet device structure. 1224 * @param[out] info 1225 * Storage for plug-in module EEPROM data. 1226 * 1227 * @return 1228 * 0 on success, a negative errno value otherwise and rte_errno is set. 1229 */ 1230 int mlx5_get_module_eeprom(struct rte_eth_dev *dev, 1231 struct rte_dev_eeprom_info *info) 1232 { 1233 struct ethtool_eeprom *eeprom; 1234 struct ifreq ifr; 1235 int ret = 0; 1236 1237 if (!dev) { 1238 DRV_LOG(WARNING, "missing argument, cannot get module eeprom"); 1239 rte_errno = EINVAL; 1240 return -rte_errno; 1241 } 1242 eeprom = mlx5_malloc(MLX5_MEM_ZERO, 1243 (sizeof(struct ethtool_eeprom) + info->length), 0, 1244 SOCKET_ID_ANY); 1245 if (!eeprom) { 1246 DRV_LOG(WARNING, "port %u cannot allocate memory for " 1247 "eeprom data", dev->data->port_id); 1248 rte_errno = ENOMEM; 1249 return -rte_errno; 1250 } 1251 eeprom->cmd = ETHTOOL_GMODULEEEPROM; 1252 eeprom->offset = info->offset; 1253 eeprom->len = info->length; 1254 ifr = (struct ifreq) { 1255 .ifr_data = (void *)eeprom, 1256 }; 1257 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1258 if (ret) 1259 DRV_LOG(WARNING, "port %u ioctl(SIOCETHTOOL) failed: %s", 1260 dev->data->port_id, strerror(rte_errno)); 1261 else 1262 rte_memcpy(info->data, eeprom->data, info->length); 1263 mlx5_free(eeprom); 1264 return ret; 1265 } 1266 1267 /** 1268 * Read device counters table. 1269 * 1270 * @param dev 1271 * Pointer to Ethernet device. 1272 * @param[in] pf 1273 * PF index in case of bonding device, -1 otherwise 1274 * @param[out] stats 1275 * Counters table output buffer. 1276 * 1277 * @return 1278 * 0 on success and stats is filled, negative errno value otherwise and 1279 * rte_errno is set. 1280 */ 1281 static int 1282 _mlx5_os_read_dev_counters(struct rte_eth_dev *dev, int pf, uint64_t *stats) 1283 { 1284 struct mlx5_priv *priv = dev->data->dev_private; 1285 struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl; 1286 unsigned int i; 1287 struct ifreq ifr; 1288 unsigned int stats_sz = xstats_ctrl->stats_n * sizeof(uint64_t); 1289 unsigned char et_stat_buf[sizeof(struct ethtool_stats) + stats_sz]; 1290 struct ethtool_stats *et_stats = (struct ethtool_stats *)et_stat_buf; 1291 int ret; 1292 1293 et_stats->cmd = ETHTOOL_GSTATS; 1294 et_stats->n_stats = xstats_ctrl->stats_n; 1295 ifr.ifr_data = (caddr_t)et_stats; 1296 if (pf >= 0) 1297 ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[pf].ifname, 1298 SIOCETHTOOL, &ifr); 1299 else 1300 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1301 if (ret) { 1302 DRV_LOG(WARNING, 1303 "port %u unable to read statistic values from device", 1304 dev->data->port_id); 1305 return ret; 1306 } 1307 for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) { 1308 if (xstats_ctrl->info[i].dev) 1309 continue; 1310 stats[i] += (uint64_t) 1311 et_stats->data[xstats_ctrl->dev_table_idx[i]]; 1312 } 1313 return 0; 1314 } 1315 1316 /** 1317 * Read device counters. 1318 * 1319 * @param dev 1320 * Pointer to Ethernet device. 1321 * @param[out] stats 1322 * Counters table output buffer. 1323 * 1324 * @return 1325 * 0 on success and stats is filled, negative errno value otherwise and 1326 * rte_errno is set. 1327 */ 1328 int 1329 mlx5_os_read_dev_counters(struct rte_eth_dev *dev, uint64_t *stats) 1330 { 1331 struct mlx5_priv *priv = dev->data->dev_private; 1332 struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl; 1333 int ret = 0, i; 1334 1335 memset(stats, 0, sizeof(*stats) * xstats_ctrl->mlx5_stats_n); 1336 /* Read ifreq counters. */ 1337 if (priv->master && priv->pf_bond >= 0) { 1338 /* Sum xstats from bonding device member ports. */ 1339 for (i = 0; i < priv->sh->bond.n_port; i++) { 1340 ret = _mlx5_os_read_dev_counters(dev, i, stats); 1341 if (ret) 1342 return ret; 1343 } 1344 } else { 1345 ret = _mlx5_os_read_dev_counters(dev, -1, stats); 1346 if (ret) 1347 return ret; 1348 } 1349 /* Read IB counters. */ 1350 for (i = 0; i != xstats_ctrl->mlx5_stats_n; ++i) { 1351 if (!xstats_ctrl->info[i].dev) 1352 continue; 1353 /* return last xstats counter if fail to read. */ 1354 if (mlx5_os_read_dev_stat(priv, xstats_ctrl->info[i].ctr_name, 1355 &stats[i]) == 0) 1356 xstats_ctrl->xstats[i] = stats[i]; 1357 else 1358 stats[i] = xstats_ctrl->xstats[i]; 1359 } 1360 return ret; 1361 } 1362 1363 /** 1364 * Query the number of statistics provided by ETHTOOL. 1365 * 1366 * @param dev 1367 * Pointer to Ethernet device. 1368 * 1369 * @return 1370 * Number of statistics on success, negative errno value otherwise and 1371 * rte_errno is set. 1372 */ 1373 int 1374 mlx5_os_get_stats_n(struct rte_eth_dev *dev) 1375 { 1376 struct mlx5_priv *priv = dev->data->dev_private; 1377 struct ethtool_drvinfo drvinfo; 1378 struct ifreq ifr; 1379 int ret; 1380 1381 drvinfo.cmd = ETHTOOL_GDRVINFO; 1382 ifr.ifr_data = (caddr_t)&drvinfo; 1383 if (priv->master && priv->pf_bond >= 0) 1384 /* Bonding PF. */ 1385 ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname, 1386 SIOCETHTOOL, &ifr); 1387 else 1388 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1389 if (ret) { 1390 DRV_LOG(WARNING, "port %u unable to query number of statistics", 1391 dev->data->port_id); 1392 return ret; 1393 } 1394 return drvinfo.n_stats; 1395 } 1396 1397 static const struct mlx5_counter_ctrl mlx5_counters_init[] = { 1398 { 1399 .dpdk_name = "rx_unicast_bytes", 1400 .ctr_name = "rx_vport_unicast_bytes", 1401 }, 1402 { 1403 .dpdk_name = "rx_multicast_bytes", 1404 .ctr_name = "rx_vport_multicast_bytes", 1405 }, 1406 { 1407 .dpdk_name = "rx_broadcast_bytes", 1408 .ctr_name = "rx_vport_broadcast_bytes", 1409 }, 1410 { 1411 .dpdk_name = "rx_unicast_packets", 1412 .ctr_name = "rx_vport_unicast_packets", 1413 }, 1414 { 1415 .dpdk_name = "rx_multicast_packets", 1416 .ctr_name = "rx_vport_multicast_packets", 1417 }, 1418 { 1419 .dpdk_name = "rx_broadcast_packets", 1420 .ctr_name = "rx_vport_broadcast_packets", 1421 }, 1422 { 1423 .dpdk_name = "tx_unicast_bytes", 1424 .ctr_name = "tx_vport_unicast_bytes", 1425 }, 1426 { 1427 .dpdk_name = "tx_multicast_bytes", 1428 .ctr_name = "tx_vport_multicast_bytes", 1429 }, 1430 { 1431 .dpdk_name = "tx_broadcast_bytes", 1432 .ctr_name = "tx_vport_broadcast_bytes", 1433 }, 1434 { 1435 .dpdk_name = "tx_unicast_packets", 1436 .ctr_name = "tx_vport_unicast_packets", 1437 }, 1438 { 1439 .dpdk_name = "tx_multicast_packets", 1440 .ctr_name = "tx_vport_multicast_packets", 1441 }, 1442 { 1443 .dpdk_name = "tx_broadcast_packets", 1444 .ctr_name = "tx_vport_broadcast_packets", 1445 }, 1446 { 1447 .dpdk_name = "rx_wqe_errors", 1448 .ctr_name = "rx_wqe_err", 1449 }, 1450 { 1451 .dpdk_name = "rx_phy_crc_errors", 1452 .ctr_name = "rx_crc_errors_phy", 1453 }, 1454 { 1455 .dpdk_name = "rx_phy_in_range_len_errors", 1456 .ctr_name = "rx_in_range_len_errors_phy", 1457 }, 1458 { 1459 .dpdk_name = "rx_phy_symbol_errors", 1460 .ctr_name = "rx_symbol_err_phy", 1461 }, 1462 { 1463 .dpdk_name = "tx_phy_errors", 1464 .ctr_name = "tx_errors_phy", 1465 }, 1466 { 1467 .dpdk_name = "rx_out_of_buffer", 1468 .ctr_name = "out_of_buffer", 1469 .dev = 1, 1470 }, 1471 { 1472 .dpdk_name = "tx_phy_packets", 1473 .ctr_name = "tx_packets_phy", 1474 }, 1475 { 1476 .dpdk_name = "rx_phy_packets", 1477 .ctr_name = "rx_packets_phy", 1478 }, 1479 { 1480 .dpdk_name = "tx_phy_discard_packets", 1481 .ctr_name = "tx_discards_phy", 1482 }, 1483 { 1484 .dpdk_name = "rx_phy_discard_packets", 1485 .ctr_name = "rx_discards_phy", 1486 }, 1487 { 1488 .dpdk_name = "rx_prio0_buf_discard_packets", 1489 .ctr_name = "rx_prio0_buf_discard", 1490 }, 1491 { 1492 .dpdk_name = "rx_prio1_buf_discard_packets", 1493 .ctr_name = "rx_prio1_buf_discard", 1494 }, 1495 { 1496 .dpdk_name = "rx_prio2_buf_discard_packets", 1497 .ctr_name = "rx_prio2_buf_discard", 1498 }, 1499 { 1500 .dpdk_name = "rx_prio3_buf_discard_packets", 1501 .ctr_name = "rx_prio3_buf_discard", 1502 }, 1503 { 1504 .dpdk_name = "rx_prio4_buf_discard_packets", 1505 .ctr_name = "rx_prio4_buf_discard", 1506 }, 1507 { 1508 .dpdk_name = "rx_prio5_buf_discard_packets", 1509 .ctr_name = "rx_prio5_buf_discard", 1510 }, 1511 { 1512 .dpdk_name = "rx_prio6_buf_discard_packets", 1513 .ctr_name = "rx_prio6_buf_discard", 1514 }, 1515 { 1516 .dpdk_name = "rx_prio7_buf_discard_packets", 1517 .ctr_name = "rx_prio7_buf_discard", 1518 }, 1519 { 1520 .dpdk_name = "rx_prio0_cong_discard_packets", 1521 .ctr_name = "rx_prio0_cong_discard", 1522 }, 1523 { 1524 .dpdk_name = "rx_prio1_cong_discard_packets", 1525 .ctr_name = "rx_prio1_cong_discard", 1526 }, 1527 { 1528 .dpdk_name = "rx_prio2_cong_discard_packets", 1529 .ctr_name = "rx_prio2_cong_discard", 1530 }, 1531 { 1532 .dpdk_name = "rx_prio3_cong_discard_packets", 1533 .ctr_name = "rx_prio3_cong_discard", 1534 }, 1535 { 1536 .dpdk_name = "rx_prio4_cong_discard_packets", 1537 .ctr_name = "rx_prio4_cong_discard", 1538 }, 1539 { 1540 .dpdk_name = "rx_prio5_cong_discard_packets", 1541 .ctr_name = "rx_prio5_cong_discard", 1542 }, 1543 { 1544 .dpdk_name = "rx_prio6_cong_discard_packets", 1545 .ctr_name = "rx_prio6_cong_discard", 1546 }, 1547 { 1548 .dpdk_name = "rx_prio7_cong_discard_packets", 1549 .ctr_name = "rx_prio7_cong_discard", 1550 }, 1551 { 1552 .dpdk_name = "tx_phy_bytes", 1553 .ctr_name = "tx_bytes_phy", 1554 }, 1555 { 1556 .dpdk_name = "rx_phy_bytes", 1557 .ctr_name = "rx_bytes_phy", 1558 }, 1559 /* Representor only */ 1560 { 1561 .dpdk_name = "rx_vport_packets", 1562 .ctr_name = "vport_rx_packets", 1563 }, 1564 { 1565 .dpdk_name = "rx_vport_bytes", 1566 .ctr_name = "vport_rx_bytes", 1567 }, 1568 { 1569 .dpdk_name = "tx_vport_packets", 1570 .ctr_name = "vport_tx_packets", 1571 }, 1572 { 1573 .dpdk_name = "tx_vport_bytes", 1574 .ctr_name = "vport_tx_bytes", 1575 }, 1576 }; 1577 1578 static const unsigned int xstats_n = RTE_DIM(mlx5_counters_init); 1579 1580 /** 1581 * Init the structures to read device counters. 1582 * 1583 * @param dev 1584 * Pointer to Ethernet device. 1585 */ 1586 void 1587 mlx5_os_stats_init(struct rte_eth_dev *dev) 1588 { 1589 struct mlx5_priv *priv = dev->data->dev_private; 1590 struct mlx5_xstats_ctrl *xstats_ctrl = &priv->xstats_ctrl; 1591 struct mlx5_stats_ctrl *stats_ctrl = &priv->stats_ctrl; 1592 unsigned int i; 1593 unsigned int j; 1594 struct ifreq ifr; 1595 struct ethtool_gstrings *strings = NULL; 1596 unsigned int dev_stats_n; 1597 unsigned int str_sz; 1598 int ret; 1599 1600 /* So that it won't aggregate for each init. */ 1601 xstats_ctrl->mlx5_stats_n = 0; 1602 ret = mlx5_os_get_stats_n(dev); 1603 if (ret < 0) { 1604 DRV_LOG(WARNING, "port %u no extended statistics available", 1605 dev->data->port_id); 1606 return; 1607 } 1608 dev_stats_n = ret; 1609 /* Allocate memory to grab stat names and values. */ 1610 str_sz = dev_stats_n * ETH_GSTRING_LEN; 1611 strings = (struct ethtool_gstrings *) 1612 mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0, 1613 SOCKET_ID_ANY); 1614 if (!strings) { 1615 DRV_LOG(WARNING, "port %u unable to allocate memory for xstats", 1616 dev->data->port_id); 1617 return; 1618 } 1619 strings->cmd = ETHTOOL_GSTRINGS; 1620 strings->string_set = ETH_SS_STATS; 1621 strings->len = dev_stats_n; 1622 ifr.ifr_data = (caddr_t)strings; 1623 if (priv->master && priv->pf_bond >= 0) 1624 /* Bonding master. */ 1625 ret = mlx5_ifreq_by_ifname(priv->sh->bond.ports[0].ifname, 1626 SIOCETHTOOL, &ifr); 1627 else 1628 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1629 if (ret) { 1630 DRV_LOG(WARNING, "port %u unable to get statistic names", 1631 dev->data->port_id); 1632 goto free; 1633 } 1634 for (i = 0; i != dev_stats_n; ++i) { 1635 const char *curr_string = (const char *) 1636 &strings->data[i * ETH_GSTRING_LEN]; 1637 1638 for (j = 0; j != xstats_n; ++j) { 1639 if (!strcmp(mlx5_counters_init[j].ctr_name, 1640 curr_string)) { 1641 unsigned int idx = xstats_ctrl->mlx5_stats_n++; 1642 1643 xstats_ctrl->dev_table_idx[idx] = i; 1644 xstats_ctrl->info[idx] = mlx5_counters_init[j]; 1645 break; 1646 } 1647 } 1648 } 1649 /* Add dev counters. */ 1650 MLX5_ASSERT(xstats_ctrl->mlx5_stats_n <= MLX5_MAX_XSTATS); 1651 for (i = 0; i != xstats_n; ++i) { 1652 if (mlx5_counters_init[i].dev) { 1653 unsigned int idx = xstats_ctrl->mlx5_stats_n++; 1654 1655 xstats_ctrl->info[idx] = mlx5_counters_init[i]; 1656 xstats_ctrl->hw_stats[idx] = 0; 1657 } 1658 } 1659 xstats_ctrl->stats_n = dev_stats_n; 1660 /* Copy to base at first time. */ 1661 ret = mlx5_os_read_dev_counters(dev, xstats_ctrl->base); 1662 if (ret) 1663 DRV_LOG(ERR, "port %u cannot read device counters: %s", 1664 dev->data->port_id, strerror(rte_errno)); 1665 mlx5_os_read_dev_stat(priv, "out_of_buffer", &stats_ctrl->imissed_base); 1666 stats_ctrl->imissed = 0; 1667 free: 1668 mlx5_free(strings); 1669 } 1670 1671 /** 1672 * Get MAC address by querying netdevice. 1673 * 1674 * @param[in] dev 1675 * Pointer to Ethernet device. 1676 * @param[out] mac 1677 * MAC address output buffer. 1678 * 1679 * @return 1680 * 0 on success, a negative errno value otherwise and rte_errno is set. 1681 */ 1682 int 1683 mlx5_get_mac(struct rte_eth_dev *dev, uint8_t (*mac)[RTE_ETHER_ADDR_LEN]) 1684 { 1685 struct ifreq request; 1686 int ret; 1687 1688 ret = mlx5_ifreq(dev, SIOCGIFHWADDR, &request); 1689 if (ret) 1690 return ret; 1691 memcpy(mac, request.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN); 1692 return 0; 1693 } 1694 1695 /* 1696 * Query dropless_rq private flag value provided by ETHTOOL. 1697 * 1698 * @param dev 1699 * Pointer to Ethernet device. 1700 * 1701 * @return 1702 * - 0 on success, flag is not set. 1703 * - 1 on success, flag is set. 1704 * - negative errno value otherwise and rte_errno is set. 1705 */ 1706 int mlx5_get_flag_dropless_rq(struct rte_eth_dev *dev) 1707 { 1708 struct ethtool_sset_info *sset_info = NULL; 1709 struct ethtool_drvinfo drvinfo; 1710 struct ifreq ifr; 1711 struct ethtool_gstrings *strings = NULL; 1712 struct ethtool_value flags; 1713 const int32_t flag_len = sizeof(flags.data) * CHAR_BIT; 1714 int32_t str_sz; 1715 int32_t len; 1716 int32_t i; 1717 int ret; 1718 1719 sset_info = mlx5_malloc(0, sizeof(struct ethtool_sset_info) + 1720 sizeof(uint32_t), 0, SOCKET_ID_ANY); 1721 if (sset_info == NULL) { 1722 rte_errno = ENOMEM; 1723 return -rte_errno; 1724 } 1725 sset_info->cmd = ETHTOOL_GSSET_INFO; 1726 sset_info->reserved = 0; 1727 sset_info->sset_mask = 1ULL << ETH_SS_PRIV_FLAGS; 1728 ifr.ifr_data = (caddr_t)&sset_info; 1729 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1730 if (!ret) { 1731 const uint32_t *sset_lengths = sset_info->data; 1732 1733 len = sset_info->sset_mask ? sset_lengths[0] : 0; 1734 } else if (ret == -EOPNOTSUPP) { 1735 drvinfo.cmd = ETHTOOL_GDRVINFO; 1736 ifr.ifr_data = (caddr_t)&drvinfo; 1737 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1738 if (ret) { 1739 DRV_LOG(WARNING, "port %u cannot get the driver info", 1740 dev->data->port_id); 1741 goto exit; 1742 } 1743 len = *(uint32_t *)((char *)&drvinfo + 1744 offsetof(struct ethtool_drvinfo, n_priv_flags)); 1745 } else { 1746 DRV_LOG(WARNING, "port %u cannot get the sset info", 1747 dev->data->port_id); 1748 goto exit; 1749 } 1750 if (!len) { 1751 DRV_LOG(WARNING, "port %u does not have private flag", 1752 dev->data->port_id); 1753 rte_errno = EOPNOTSUPP; 1754 ret = -rte_errno; 1755 goto exit; 1756 } else if (len > flag_len) { 1757 DRV_LOG(WARNING, "port %u maximal private flags number is %d", 1758 dev->data->port_id, flag_len); 1759 len = flag_len; 1760 } 1761 str_sz = ETH_GSTRING_LEN * len; 1762 strings = (struct ethtool_gstrings *) 1763 mlx5_malloc(0, str_sz + sizeof(struct ethtool_gstrings), 0, 1764 SOCKET_ID_ANY); 1765 if (!strings) { 1766 DRV_LOG(WARNING, "port %u unable to allocate memory for" 1767 " private flags", dev->data->port_id); 1768 rte_errno = ENOMEM; 1769 ret = -rte_errno; 1770 goto exit; 1771 } 1772 strings->cmd = ETHTOOL_GSTRINGS; 1773 strings->string_set = ETH_SS_PRIV_FLAGS; 1774 strings->len = len; 1775 ifr.ifr_data = (caddr_t)strings; 1776 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1777 if (ret) { 1778 DRV_LOG(WARNING, "port %u unable to get private flags strings", 1779 dev->data->port_id); 1780 goto exit; 1781 } 1782 for (i = 0; i < len; i++) { 1783 strings->data[(i + 1) * ETH_GSTRING_LEN - 1] = 0; 1784 if (!strcmp((const char *)strings->data + i * ETH_GSTRING_LEN, 1785 "dropless_rq")) 1786 break; 1787 } 1788 if (i == len) { 1789 DRV_LOG(WARNING, "port %u does not support dropless_rq", 1790 dev->data->port_id); 1791 rte_errno = EOPNOTSUPP; 1792 ret = -rte_errno; 1793 goto exit; 1794 } 1795 flags.cmd = ETHTOOL_GPFLAGS; 1796 ifr.ifr_data = (caddr_t)&flags; 1797 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 1798 if (ret) { 1799 DRV_LOG(WARNING, "port %u unable to get private flags status", 1800 dev->data->port_id); 1801 goto exit; 1802 } 1803 ret = !!(flags.data & (1U << i)); 1804 exit: 1805 mlx5_free(strings); 1806 mlx5_free(sset_info); 1807 return ret; 1808 } 1809 1810 /** 1811 * Unmaps HCA PCI BAR from the current process address space. 1812 * 1813 * @param dev 1814 * Pointer to Ethernet device structure. 1815 */ 1816 void mlx5_txpp_unmap_hca_bar(struct rte_eth_dev *dev) 1817 { 1818 struct mlx5_proc_priv *ppriv = dev->process_private; 1819 1820 if (ppriv && ppriv->hca_bar) { 1821 rte_mem_unmap(ppriv->hca_bar, MLX5_ST_SZ_BYTES(initial_seg)); 1822 ppriv->hca_bar = NULL; 1823 } 1824 } 1825 1826 /** 1827 * Maps HCA PCI BAR to the current process address space. 1828 * Stores pointer in the process private structure allowing 1829 * to read internal and real time counter directly from the HW. 1830 * 1831 * @param dev 1832 * Pointer to Ethernet device structure. 1833 * 1834 * @return 1835 * 0 on success and not NULL pointer to mapped area in process structure. 1836 * negative otherwise and NULL pointer 1837 */ 1838 int mlx5_txpp_map_hca_bar(struct rte_eth_dev *dev) 1839 { 1840 struct mlx5_proc_priv *ppriv = dev->process_private; 1841 char pci_addr[PCI_PRI_STR_SIZE] = { 0 }; 1842 void *base, *expected = NULL; 1843 int fd, ret; 1844 1845 if (!ppriv) { 1846 rte_errno = ENOMEM; 1847 return -rte_errno; 1848 } 1849 if (ppriv->hca_bar) 1850 return 0; 1851 ret = mlx5_dev_to_pci_str(dev->device, pci_addr, sizeof(pci_addr)); 1852 if (ret < 0) 1853 return -rte_errno; 1854 /* Open PCI device resource 0 - HCA initialize segment */ 1855 MKSTR(name, "/sys/bus/pci/devices/%s/resource0", pci_addr); 1856 fd = open(name, O_RDWR | O_SYNC); 1857 if (fd == -1) { 1858 rte_errno = ENOTSUP; 1859 return -ENOTSUP; 1860 } 1861 base = rte_mem_map(NULL, MLX5_ST_SZ_BYTES(initial_seg), 1862 RTE_PROT_READ, RTE_MAP_SHARED, fd, 0); 1863 close(fd); 1864 if (!base) { 1865 rte_errno = ENOTSUP; 1866 return -ENOTSUP; 1867 } 1868 /* Check there is no concurrent mapping in other thread. */ 1869 if (!__atomic_compare_exchange_n(&ppriv->hca_bar, &expected, 1870 base, false, 1871 __ATOMIC_RELAXED, __ATOMIC_RELAXED)) 1872 rte_mem_unmap(base, MLX5_ST_SZ_BYTES(initial_seg)); 1873 return 0; 1874 } 1875 1876