1 /*- 2 * BSD LICENSE 3 * 4 * Copyright 2015 6WIND S.A. 5 * Copyright 2015 Mellanox. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of 6WIND S.A. nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #define _GNU_SOURCE 35 36 #include <stddef.h> 37 #include <assert.h> 38 #include <inttypes.h> 39 #include <unistd.h> 40 #include <stdint.h> 41 #include <stdio.h> 42 #include <string.h> 43 #include <stdlib.h> 44 #include <errno.h> 45 #include <dirent.h> 46 #include <net/if.h> 47 #include <sys/ioctl.h> 48 #include <sys/socket.h> 49 #include <netinet/in.h> 50 #include <linux/ethtool.h> 51 #include <linux/sockios.h> 52 #include <fcntl.h> 53 #include <stdalign.h> 54 #include <sys/un.h> 55 #include <time.h> 56 57 #include <rte_atomic.h> 58 #include <rte_ethdev.h> 59 #include <rte_bus_pci.h> 60 #include <rte_mbuf.h> 61 #include <rte_common.h> 62 #include <rte_interrupts.h> 63 #include <rte_malloc.h> 64 65 #include "mlx5.h" 66 #include "mlx5_rxtx.h" 67 #include "mlx5_utils.h" 68 69 /* Supported speed values found in /usr/include/linux/ethtool.h */ 70 #ifndef HAVE_SUPPORTED_40000baseKR4_Full 71 #define SUPPORTED_40000baseKR4_Full (1 << 23) 72 #endif 73 #ifndef HAVE_SUPPORTED_40000baseCR4_Full 74 #define SUPPORTED_40000baseCR4_Full (1 << 24) 75 #endif 76 #ifndef HAVE_SUPPORTED_40000baseSR4_Full 77 #define SUPPORTED_40000baseSR4_Full (1 << 25) 78 #endif 79 #ifndef HAVE_SUPPORTED_40000baseLR4_Full 80 #define SUPPORTED_40000baseLR4_Full (1 << 26) 81 #endif 82 #ifndef HAVE_SUPPORTED_56000baseKR4_Full 83 #define SUPPORTED_56000baseKR4_Full (1 << 27) 84 #endif 85 #ifndef HAVE_SUPPORTED_56000baseCR4_Full 86 #define SUPPORTED_56000baseCR4_Full (1 << 28) 87 #endif 88 #ifndef HAVE_SUPPORTED_56000baseSR4_Full 89 #define SUPPORTED_56000baseSR4_Full (1 << 29) 90 #endif 91 #ifndef HAVE_SUPPORTED_56000baseLR4_Full 92 #define SUPPORTED_56000baseLR4_Full (1 << 30) 93 #endif 94 95 /* Add defines in case the running kernel is not the same as user headers. */ 96 #ifndef ETHTOOL_GLINKSETTINGS 97 struct ethtool_link_settings { 98 uint32_t cmd; 99 uint32_t speed; 100 uint8_t duplex; 101 uint8_t port; 102 uint8_t phy_address; 103 uint8_t autoneg; 104 uint8_t mdio_support; 105 uint8_t eth_to_mdix; 106 uint8_t eth_tp_mdix_ctrl; 107 int8_t link_mode_masks_nwords; 108 uint32_t reserved[8]; 109 uint32_t link_mode_masks[]; 110 }; 111 112 #define ETHTOOL_GLINKSETTINGS 0x0000004c 113 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5 114 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6 115 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17 116 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18 117 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19 118 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20 119 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21 120 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22 121 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23 122 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24 123 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25 124 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26 125 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27 126 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28 127 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29 128 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30 129 #endif 130 #ifndef HAVE_ETHTOOL_LINK_MODE_25G 131 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31 132 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32 133 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33 134 #endif 135 #ifndef HAVE_ETHTOOL_LINK_MODE_50G 136 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34 137 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35 138 #endif 139 #ifndef HAVE_ETHTOOL_LINK_MODE_100G 140 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36 141 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37 142 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38 143 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39 144 #endif 145 146 /** 147 * Get interface name from private structure. 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)[IF_NAMESIZE]) 159 { 160 struct priv *priv = dev->data->dev_private; 161 DIR *dir; 162 struct dirent *dent; 163 unsigned int dev_type = 0; 164 unsigned int dev_port_prev = ~0u; 165 char match[IF_NAMESIZE] = ""; 166 167 { 168 MKSTR(path, "%s/device/net", priv->ibdev_path); 169 170 dir = opendir(path); 171 if (dir == NULL) { 172 rte_errno = errno; 173 return -rte_errno; 174 } 175 } 176 while ((dent = readdir(dir)) != NULL) { 177 char *name = dent->d_name; 178 FILE *file; 179 unsigned int dev_port; 180 int r; 181 182 if ((name[0] == '.') && 183 ((name[1] == '\0') || 184 ((name[1] == '.') && (name[2] == '\0')))) 185 continue; 186 187 MKSTR(path, "%s/device/net/%s/%s", 188 priv->ibdev_path, name, 189 (dev_type ? "dev_id" : "dev_port")); 190 191 file = fopen(path, "rb"); 192 if (file == NULL) { 193 if (errno != ENOENT) 194 continue; 195 /* 196 * Switch to dev_id when dev_port does not exist as 197 * is the case with Linux kernel versions < 3.15. 198 */ 199 try_dev_id: 200 match[0] = '\0'; 201 if (dev_type) 202 break; 203 dev_type = 1; 204 dev_port_prev = ~0u; 205 rewinddir(dir); 206 continue; 207 } 208 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port); 209 fclose(file); 210 if (r != 1) 211 continue; 212 /* 213 * Switch to dev_id when dev_port returns the same value for 214 * all ports. May happen when using a MOFED release older than 215 * 3.0 with a Linux kernel >= 3.15. 216 */ 217 if (dev_port == dev_port_prev) 218 goto try_dev_id; 219 dev_port_prev = dev_port; 220 if (dev_port == (priv->port - 1u)) 221 snprintf(match, sizeof(match), "%s", name); 222 } 223 closedir(dir); 224 if (match[0] == '\0') { 225 rte_errno = ENOENT; 226 return -rte_errno; 227 } 228 strncpy(*ifname, match, sizeof(*ifname)); 229 return 0; 230 } 231 232 /** 233 * Perform ifreq ioctl() on associated Ethernet device. 234 * 235 * @param[in] dev 236 * Pointer to Ethernet device. 237 * @param req 238 * Request number to pass to ioctl(). 239 * @param[out] ifr 240 * Interface request structure output buffer. 241 * 242 * @return 243 * 0 on success, a negative errno value otherwise and rte_errno is set. 244 */ 245 int 246 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr) 247 { 248 int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); 249 int ret = 0; 250 251 if (sock == -1) { 252 rte_errno = errno; 253 return -rte_errno; 254 } 255 ret = mlx5_get_ifname(dev, &ifr->ifr_name); 256 if (ret) 257 goto error; 258 ret = ioctl(sock, req, ifr); 259 if (ret == -1) { 260 rte_errno = errno; 261 goto error; 262 } 263 close(sock); 264 return 0; 265 error: 266 close(sock); 267 return -rte_errno; 268 } 269 270 /** 271 * Get device MTU. 272 * 273 * @param dev 274 * Pointer to Ethernet device. 275 * @param[out] mtu 276 * MTU value output buffer. 277 * 278 * @return 279 * 0 on success, a negative errno value otherwise and rte_errno is set. 280 */ 281 int 282 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu) 283 { 284 struct ifreq request; 285 int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request); 286 287 if (ret) 288 return ret; 289 *mtu = request.ifr_mtu; 290 return 0; 291 } 292 293 /** 294 * Set device MTU. 295 * 296 * @param dev 297 * Pointer to Ethernet device. 298 * @param mtu 299 * MTU value to set. 300 * 301 * @return 302 * 0 on success, a negative errno value otherwise and rte_errno is set. 303 */ 304 static int 305 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) 306 { 307 struct ifreq request = { .ifr_mtu = mtu, }; 308 309 return mlx5_ifreq(dev, SIOCSIFMTU, &request); 310 } 311 312 /** 313 * Set device flags. 314 * 315 * @param dev 316 * Pointer to Ethernet device. 317 * @param keep 318 * Bitmask for flags that must remain untouched. 319 * @param flags 320 * Bitmask for flags to modify. 321 * 322 * @return 323 * 0 on success, a negative errno value otherwise and rte_errno is set. 324 */ 325 int 326 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags) 327 { 328 struct ifreq request; 329 int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request); 330 331 if (ret) 332 return ret; 333 request.ifr_flags &= keep; 334 request.ifr_flags |= flags & ~keep; 335 return mlx5_ifreq(dev, SIOCSIFFLAGS, &request); 336 } 337 338 /** 339 * DPDK callback for Ethernet device configuration. 340 * 341 * @param dev 342 * Pointer to Ethernet device structure. 343 * 344 * @return 345 * 0 on success, a negative errno value otherwise and rte_errno is set. 346 */ 347 int 348 mlx5_dev_configure(struct rte_eth_dev *dev) 349 { 350 struct priv *priv = dev->data->dev_private; 351 unsigned int rxqs_n = dev->data->nb_rx_queues; 352 unsigned int txqs_n = dev->data->nb_tx_queues; 353 unsigned int i; 354 unsigned int j; 355 unsigned int reta_idx_n; 356 const uint8_t use_app_rss_key = 357 !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key; 358 int ret = 0; 359 360 if (use_app_rss_key && 361 (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len != 362 rss_hash_default_key_len)) { 363 DRV_LOG(ERR, "port %u RSS key len must be %zu Bytes long", 364 dev->data->port_id, rss_hash_default_key_len); 365 rte_errno = EINVAL; 366 return -rte_errno; 367 } 368 priv->rss_conf.rss_key = 369 rte_realloc(priv->rss_conf.rss_key, 370 rss_hash_default_key_len, 0); 371 if (!priv->rss_conf.rss_key) { 372 DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)", 373 dev->data->port_id, rxqs_n); 374 rte_errno = ENOMEM; 375 return -rte_errno; 376 } 377 memcpy(priv->rss_conf.rss_key, 378 use_app_rss_key ? 379 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key : 380 rss_hash_default_key, 381 rss_hash_default_key_len); 382 priv->rss_conf.rss_key_len = rss_hash_default_key_len; 383 priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf; 384 priv->rxqs = (void *)dev->data->rx_queues; 385 priv->txqs = (void *)dev->data->tx_queues; 386 if (txqs_n != priv->txqs_n) { 387 DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u", 388 dev->data->port_id, priv->txqs_n, txqs_n); 389 priv->txqs_n = txqs_n; 390 } 391 if (rxqs_n > priv->ind_table_max_size) { 392 DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)", 393 dev->data->port_id, rxqs_n); 394 rte_errno = EINVAL; 395 return -rte_errno; 396 } 397 if (rxqs_n == priv->rxqs_n) 398 return 0; 399 DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u", 400 dev->data->port_id, priv->rxqs_n, rxqs_n); 401 priv->rxqs_n = rxqs_n; 402 /* If the requested number of RX queues is not a power of two, use the 403 * maximum indirection table size for better balancing. 404 * The result is always rounded to the next power of two. */ 405 reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ? 406 priv->ind_table_max_size : 407 rxqs_n)); 408 ret = mlx5_rss_reta_index_resize(dev, reta_idx_n); 409 if (ret) 410 return ret; 411 /* When the number of RX queues is not a power of two, the remaining 412 * table entries are padded with reused WQs and hashes are not spread 413 * uniformly. */ 414 for (i = 0, j = 0; (i != reta_idx_n); ++i) { 415 (*priv->reta_idx)[i] = j; 416 if (++j == rxqs_n) 417 j = 0; 418 } 419 return 0; 420 } 421 422 /** 423 * DPDK callback to get information about the device. 424 * 425 * @param dev 426 * Pointer to Ethernet device structure. 427 * @param[out] info 428 * Info structure output buffer. 429 */ 430 void 431 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info) 432 { 433 struct priv *priv = dev->data->dev_private; 434 unsigned int max; 435 char ifname[IF_NAMESIZE]; 436 437 info->pci_dev = RTE_ETH_DEV_TO_PCI(dev); 438 /* FIXME: we should ask the device for these values. */ 439 info->min_rx_bufsize = 32; 440 info->max_rx_pktlen = 65536; 441 /* 442 * Since we need one CQ per QP, the limit is the minimum number 443 * between the two values. 444 */ 445 max = RTE_MIN(priv->device_attr.orig_attr.max_cq, 446 priv->device_attr.orig_attr.max_qp); 447 /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */ 448 if (max >= 65535) 449 max = 65535; 450 info->max_rx_queues = max; 451 info->max_tx_queues = max; 452 info->max_mac_addrs = RTE_DIM(priv->mac); 453 info->rx_offload_capa = 454 (priv->hw_csum ? 455 (DEV_RX_OFFLOAD_IPV4_CKSUM | 456 DEV_RX_OFFLOAD_UDP_CKSUM | 457 DEV_RX_OFFLOAD_TCP_CKSUM) : 458 0) | 459 (priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0) | 460 DEV_RX_OFFLOAD_TIMESTAMP; 461 462 if (!priv->mps) 463 info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT; 464 if (priv->hw_csum) 465 info->tx_offload_capa |= 466 (DEV_TX_OFFLOAD_IPV4_CKSUM | 467 DEV_TX_OFFLOAD_UDP_CKSUM | 468 DEV_TX_OFFLOAD_TCP_CKSUM); 469 if (priv->tso) 470 info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO; 471 if (priv->tunnel_en) 472 info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM | 473 DEV_TX_OFFLOAD_VXLAN_TNL_TSO | 474 DEV_TX_OFFLOAD_GRE_TNL_TSO); 475 if (mlx5_get_ifname(dev, &ifname) == 0) 476 info->if_index = if_nametoindex(ifname); 477 info->reta_size = priv->reta_idx_n ? 478 priv->reta_idx_n : priv->ind_table_max_size; 479 info->hash_key_size = rss_hash_default_key_len; 480 info->speed_capa = priv->link_speed_capa; 481 info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK; 482 } 483 484 /** 485 * Get supported packet types. 486 * 487 * @param dev 488 * Pointer to Ethernet device structure. 489 * 490 * @return 491 * A pointer to the supported Packet types array. 492 */ 493 const uint32_t * 494 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev) 495 { 496 static const uint32_t ptypes[] = { 497 /* refers to rxq_cq_to_pkt_type() */ 498 RTE_PTYPE_L2_ETHER, 499 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN, 500 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN, 501 RTE_PTYPE_L4_NONFRAG, 502 RTE_PTYPE_L4_FRAG, 503 RTE_PTYPE_L4_TCP, 504 RTE_PTYPE_L4_UDP, 505 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN, 506 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN, 507 RTE_PTYPE_INNER_L4_NONFRAG, 508 RTE_PTYPE_INNER_L4_FRAG, 509 RTE_PTYPE_INNER_L4_TCP, 510 RTE_PTYPE_INNER_L4_UDP, 511 RTE_PTYPE_UNKNOWN 512 }; 513 514 if (dev->rx_pkt_burst == mlx5_rx_burst || 515 dev->rx_pkt_burst == mlx5_rx_burst_vec) 516 return ptypes; 517 return NULL; 518 } 519 520 /** 521 * DPDK callback to retrieve physical link information. 522 * 523 * @param dev 524 * Pointer to Ethernet device structure. 525 * @param[out] link 526 * Storage for current link status. 527 * 528 * @return 529 * 0 on success, a negative errno value otherwise and rte_errno is set. 530 */ 531 static int 532 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, 533 struct rte_eth_link *link) 534 { 535 struct priv *priv = dev->data->dev_private; 536 struct ethtool_cmd edata = { 537 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */ 538 }; 539 struct ifreq ifr; 540 struct rte_eth_link dev_link; 541 int link_speed = 0; 542 int ret; 543 544 ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr); 545 if (ret) { 546 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s", 547 dev->data->port_id, strerror(rte_errno)); 548 return ret; 549 } 550 memset(&dev_link, 0, sizeof(dev_link)); 551 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) && 552 (ifr.ifr_flags & IFF_RUNNING)); 553 ifr.ifr_data = (void *)&edata; 554 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 555 if (ret) { 556 DRV_LOG(WARNING, 557 "port %u ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s", 558 dev->data->port_id, strerror(rte_errno)); 559 return ret; 560 } 561 link_speed = ethtool_cmd_speed(&edata); 562 if (link_speed == -1) 563 dev_link.link_speed = 0; 564 else 565 dev_link.link_speed = link_speed; 566 priv->link_speed_capa = 0; 567 if (edata.supported & SUPPORTED_Autoneg) 568 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG; 569 if (edata.supported & (SUPPORTED_1000baseT_Full | 570 SUPPORTED_1000baseKX_Full)) 571 priv->link_speed_capa |= ETH_LINK_SPEED_1G; 572 if (edata.supported & SUPPORTED_10000baseKR_Full) 573 priv->link_speed_capa |= ETH_LINK_SPEED_10G; 574 if (edata.supported & (SUPPORTED_40000baseKR4_Full | 575 SUPPORTED_40000baseCR4_Full | 576 SUPPORTED_40000baseSR4_Full | 577 SUPPORTED_40000baseLR4_Full)) 578 priv->link_speed_capa |= ETH_LINK_SPEED_40G; 579 dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ? 580 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); 581 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds & 582 ETH_LINK_SPEED_FIXED); 583 if ((dev_link.link_speed && !dev_link.link_status) || 584 (!dev_link.link_speed && dev_link.link_status)) { 585 rte_errno = EAGAIN; 586 return -rte_errno; 587 } 588 *link = dev_link; 589 return 0; 590 } 591 592 /** 593 * Retrieve physical link information (unlocked version using new ioctl). 594 * 595 * @param dev 596 * Pointer to Ethernet device structure. 597 * @param[out] link 598 * Storage for current link status. 599 * 600 * @return 601 * 0 on success, a negative errno value otherwise and rte_errno is set. 602 */ 603 static int 604 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, 605 struct rte_eth_link *link) 606 607 { 608 struct priv *priv = dev->data->dev_private; 609 struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS }; 610 struct ifreq ifr; 611 struct rte_eth_link dev_link; 612 uint64_t sc; 613 int ret; 614 615 ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr); 616 if (ret) { 617 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s", 618 dev->data->port_id, strerror(rte_errno)); 619 return ret; 620 } 621 memset(&dev_link, 0, sizeof(dev_link)); 622 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) && 623 (ifr.ifr_flags & IFF_RUNNING)); 624 ifr.ifr_data = (void *)&gcmd; 625 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 626 if (ret) { 627 DRV_LOG(DEBUG, 628 "port %u ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS)" 629 " failed: %s", 630 dev->data->port_id, strerror(rte_errno)); 631 return ret; 632 } 633 gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords; 634 635 alignas(struct ethtool_link_settings) 636 uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) + 637 sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3]; 638 struct ethtool_link_settings *ecmd = (void *)data; 639 640 *ecmd = gcmd; 641 ifr.ifr_data = (void *)ecmd; 642 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 643 if (ret) { 644 DRV_LOG(DEBUG, 645 "port %u ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS)" 646 " failed: %s", 647 dev->data->port_id, strerror(rte_errno)); 648 return ret; 649 } 650 dev_link.link_speed = ecmd->speed; 651 sc = ecmd->link_mode_masks[0] | 652 ((uint64_t)ecmd->link_mode_masks[1] << 32); 653 priv->link_speed_capa = 0; 654 if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT)) 655 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG; 656 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) | 657 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))) 658 priv->link_speed_capa |= ETH_LINK_SPEED_1G; 659 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) | 660 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) | 661 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))) 662 priv->link_speed_capa |= ETH_LINK_SPEED_10G; 663 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) | 664 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))) 665 priv->link_speed_capa |= ETH_LINK_SPEED_20G; 666 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) | 667 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) | 668 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) | 669 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))) 670 priv->link_speed_capa |= ETH_LINK_SPEED_40G; 671 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) | 672 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) | 673 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) | 674 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))) 675 priv->link_speed_capa |= ETH_LINK_SPEED_56G; 676 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) | 677 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) | 678 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))) 679 priv->link_speed_capa |= ETH_LINK_SPEED_25G; 680 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) | 681 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))) 682 priv->link_speed_capa |= ETH_LINK_SPEED_50G; 683 if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) | 684 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) | 685 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) | 686 MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))) 687 priv->link_speed_capa |= ETH_LINK_SPEED_100G; 688 dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ? 689 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); 690 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds & 691 ETH_LINK_SPEED_FIXED); 692 if ((dev_link.link_speed && !dev_link.link_status) || 693 (!dev_link.link_speed && dev_link.link_status)) { 694 rte_errno = EAGAIN; 695 return -rte_errno; 696 } 697 *link = dev_link; 698 return 0; 699 } 700 701 /** 702 * DPDK callback to retrieve physical link information. 703 * 704 * @param dev 705 * Pointer to Ethernet device structure. 706 * @param wait_to_complete 707 * Wait for request completion. 708 * 709 * @return 710 * 0 if link status was not updated, positive if it was, a negative errno 711 * value otherwise and rte_errno is set. 712 */ 713 int 714 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete) 715 { 716 int ret; 717 struct rte_eth_link dev_link; 718 time_t start_time = time(NULL); 719 720 do { 721 ret = mlx5_link_update_unlocked_gs(dev, &dev_link); 722 if (ret) 723 ret = mlx5_link_update_unlocked_gset(dev, &dev_link); 724 if (ret == 0) 725 break; 726 /* Handle wait to complete situation. */ 727 if (wait_to_complete && ret == -EAGAIN) { 728 if (abs((int)difftime(time(NULL), start_time)) < 729 MLX5_LINK_STATUS_TIMEOUT) { 730 usleep(0); 731 continue; 732 } else { 733 rte_errno = EBUSY; 734 return -rte_errno; 735 } 736 } else if (ret < 0) { 737 return ret; 738 } 739 } while (wait_to_complete); 740 ret = !!memcmp(&dev->data->dev_link, &dev_link, 741 sizeof(struct rte_eth_link)); 742 dev->data->dev_link = dev_link; 743 return ret; 744 } 745 746 /** 747 * DPDK callback to change the MTU. 748 * 749 * @param dev 750 * Pointer to Ethernet device structure. 751 * @param in_mtu 752 * New MTU. 753 * 754 * @return 755 * 0 on success, a negative errno value otherwise and rte_errno is set. 756 */ 757 int 758 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) 759 { 760 struct priv *priv = dev->data->dev_private; 761 uint16_t kern_mtu = 0; 762 int ret; 763 764 ret = mlx5_get_mtu(dev, &kern_mtu); 765 if (ret) 766 return ret; 767 /* Set kernel interface MTU first. */ 768 ret = mlx5_set_mtu(dev, mtu); 769 if (ret) 770 return ret; 771 ret = mlx5_get_mtu(dev, &kern_mtu); 772 if (ret) 773 return ret; 774 if (kern_mtu == mtu) { 775 priv->mtu = mtu; 776 DRV_LOG(DEBUG, "port %u adapter MTU set to %u", 777 dev->data->port_id, mtu); 778 return 0; 779 } 780 rte_errno = EAGAIN; 781 return -rte_errno; 782 } 783 784 /** 785 * DPDK callback to get flow control status. 786 * 787 * @param dev 788 * Pointer to Ethernet device structure. 789 * @param[out] fc_conf 790 * Flow control output buffer. 791 * 792 * @return 793 * 0 on success, a negative errno value otherwise and rte_errno is set. 794 */ 795 int 796 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) 797 { 798 struct ifreq ifr; 799 struct ethtool_pauseparam ethpause = { 800 .cmd = ETHTOOL_GPAUSEPARAM 801 }; 802 int ret; 803 804 ifr.ifr_data = (void *)ðpause; 805 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 806 if (ret) { 807 DRV_LOG(WARNING, 808 "port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:" 809 " %s", 810 dev->data->port_id, strerror(rte_errno)); 811 return ret; 812 } 813 fc_conf->autoneg = ethpause.autoneg; 814 if (ethpause.rx_pause && ethpause.tx_pause) 815 fc_conf->mode = RTE_FC_FULL; 816 else if (ethpause.rx_pause) 817 fc_conf->mode = RTE_FC_RX_PAUSE; 818 else if (ethpause.tx_pause) 819 fc_conf->mode = RTE_FC_TX_PAUSE; 820 else 821 fc_conf->mode = RTE_FC_NONE; 822 return 0; 823 } 824 825 /** 826 * DPDK callback to modify flow control parameters. 827 * 828 * @param dev 829 * Pointer to Ethernet device structure. 830 * @param[in] fc_conf 831 * Flow control parameters. 832 * 833 * @return 834 * 0 on success, a negative errno value otherwise and rte_errno is set. 835 */ 836 int 837 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) 838 { 839 struct ifreq ifr; 840 struct ethtool_pauseparam ethpause = { 841 .cmd = ETHTOOL_SPAUSEPARAM 842 }; 843 int ret; 844 845 ifr.ifr_data = (void *)ðpause; 846 ethpause.autoneg = fc_conf->autoneg; 847 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) || 848 (fc_conf->mode & RTE_FC_RX_PAUSE)) 849 ethpause.rx_pause = 1; 850 else 851 ethpause.rx_pause = 0; 852 853 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) || 854 (fc_conf->mode & RTE_FC_TX_PAUSE)) 855 ethpause.tx_pause = 1; 856 else 857 ethpause.tx_pause = 0; 858 ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr); 859 if (ret) { 860 DRV_LOG(WARNING, 861 "port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)" 862 " failed: %s", 863 dev->data->port_id, strerror(rte_errno)); 864 return ret; 865 } 866 return 0; 867 } 868 869 /** 870 * Get PCI information from struct ibv_device. 871 * 872 * @param device 873 * Pointer to Ethernet device structure. 874 * @param[out] pci_addr 875 * PCI bus address output buffer. 876 * 877 * @return 878 * 0 on success, a negative errno value otherwise and rte_errno is set. 879 */ 880 int 881 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device, 882 struct rte_pci_addr *pci_addr) 883 { 884 FILE *file; 885 char line[32]; 886 MKSTR(path, "%s/device/uevent", device->ibdev_path); 887 888 file = fopen(path, "rb"); 889 if (file == NULL) { 890 rte_errno = errno; 891 return -rte_errno; 892 } 893 while (fgets(line, sizeof(line), file) == line) { 894 size_t len = strlen(line); 895 int ret; 896 897 /* Truncate long lines. */ 898 if (len == (sizeof(line) - 1)) 899 while (line[(len - 1)] != '\n') { 900 ret = fgetc(file); 901 if (ret == EOF) 902 break; 903 line[(len - 1)] = ret; 904 } 905 /* Extract information. */ 906 if (sscanf(line, 907 "PCI_SLOT_NAME=" 908 "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n", 909 &pci_addr->domain, 910 &pci_addr->bus, 911 &pci_addr->devid, 912 &pci_addr->function) == 4) { 913 ret = 0; 914 break; 915 } 916 } 917 fclose(file); 918 return 0; 919 } 920 921 /** 922 * Device status handler. 923 * 924 * @param dev 925 * Pointer to Ethernet device. 926 * @param events 927 * Pointer to event flags holder. 928 * 929 * @return 930 * Events bitmap of callback process which can be called immediately. 931 */ 932 static uint32_t 933 mlx5_dev_status_handler(struct rte_eth_dev *dev) 934 { 935 struct priv *priv = dev->data->dev_private; 936 struct ibv_async_event event; 937 uint32_t ret = 0; 938 939 if (mlx5_link_update(dev, 0) == -EAGAIN) { 940 usleep(0); 941 return 0; 942 } 943 /* Read all message and acknowledge them. */ 944 for (;;) { 945 if (ibv_get_async_event(priv->ctx, &event)) 946 break; 947 if ((event.event_type == IBV_EVENT_PORT_ACTIVE || 948 event.event_type == IBV_EVENT_PORT_ERR) && 949 (dev->data->dev_conf.intr_conf.lsc == 1)) 950 ret |= (1 << RTE_ETH_EVENT_INTR_LSC); 951 else if (event.event_type == IBV_EVENT_DEVICE_FATAL && 952 dev->data->dev_conf.intr_conf.rmv == 1) 953 ret |= (1 << RTE_ETH_EVENT_INTR_RMV); 954 else 955 DRV_LOG(DEBUG, 956 "port %u event type %d on not handled", 957 dev->data->port_id, event.event_type); 958 ibv_ack_async_event(&event); 959 } 960 return ret; 961 } 962 963 /** 964 * Handle interrupts from the NIC. 965 * 966 * @param[in] intr_handle 967 * Interrupt handler. 968 * @param cb_arg 969 * Callback argument. 970 */ 971 void 972 mlx5_dev_interrupt_handler(void *cb_arg) 973 { 974 struct rte_eth_dev *dev = cb_arg; 975 uint32_t events; 976 977 events = mlx5_dev_status_handler(dev); 978 if (events & (1 << RTE_ETH_EVENT_INTR_LSC)) 979 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL, 980 NULL); 981 if (events & (1 << RTE_ETH_EVENT_INTR_RMV)) 982 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL, 983 NULL); 984 } 985 986 /** 987 * Handle interrupts from the socket. 988 * 989 * @param cb_arg 990 * Callback argument. 991 */ 992 static void 993 mlx5_dev_handler_socket(void *cb_arg) 994 { 995 struct rte_eth_dev *dev = cb_arg; 996 997 mlx5_socket_handle(dev); 998 } 999 1000 /** 1001 * Uninstall interrupt handler. 1002 * 1003 * @param dev 1004 * Pointer to Ethernet device. 1005 */ 1006 void 1007 mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev) 1008 { 1009 struct priv *priv = dev->data->dev_private; 1010 1011 if (dev->data->dev_conf.intr_conf.lsc || 1012 dev->data->dev_conf.intr_conf.rmv) 1013 rte_intr_callback_unregister(&priv->intr_handle, 1014 mlx5_dev_interrupt_handler, dev); 1015 if (priv->primary_socket) 1016 rte_intr_callback_unregister(&priv->intr_handle_socket, 1017 mlx5_dev_handler_socket, dev); 1018 priv->intr_handle.fd = 0; 1019 priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; 1020 priv->intr_handle_socket.fd = 0; 1021 priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN; 1022 } 1023 1024 /** 1025 * Install interrupt handler. 1026 * 1027 * @param dev 1028 * Pointer to Ethernet device. 1029 */ 1030 void 1031 mlx5_dev_interrupt_handler_install(struct rte_eth_dev *dev) 1032 { 1033 struct priv *priv = dev->data->dev_private; 1034 int ret; 1035 int flags; 1036 1037 assert(priv->ctx->async_fd > 0); 1038 flags = fcntl(priv->ctx->async_fd, F_GETFL); 1039 ret = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK); 1040 if (ret) { 1041 DRV_LOG(INFO, 1042 "port %u failed to change file descriptor async event" 1043 " queue", 1044 dev->data->port_id); 1045 dev->data->dev_conf.intr_conf.lsc = 0; 1046 dev->data->dev_conf.intr_conf.rmv = 0; 1047 } 1048 if (dev->data->dev_conf.intr_conf.lsc || 1049 dev->data->dev_conf.intr_conf.rmv) { 1050 priv->intr_handle.fd = priv->ctx->async_fd; 1051 priv->intr_handle.type = RTE_INTR_HANDLE_EXT; 1052 rte_intr_callback_register(&priv->intr_handle, 1053 mlx5_dev_interrupt_handler, dev); 1054 } 1055 ret = mlx5_socket_init(dev); 1056 if (ret) 1057 DRV_LOG(ERR, "port %u cannot initialise socket: %s", 1058 dev->data->port_id, strerror(rte_errno)); 1059 else if (priv->primary_socket) { 1060 priv->intr_handle_socket.fd = priv->primary_socket; 1061 priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT; 1062 rte_intr_callback_register(&priv->intr_handle_socket, 1063 mlx5_dev_handler_socket, dev); 1064 } 1065 } 1066 1067 /** 1068 * DPDK callback to bring the link DOWN. 1069 * 1070 * @param dev 1071 * Pointer to Ethernet device structure. 1072 * 1073 * @return 1074 * 0 on success, a negative errno value otherwise and rte_errno is set. 1075 */ 1076 int 1077 mlx5_set_link_down(struct rte_eth_dev *dev) 1078 { 1079 return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP); 1080 } 1081 1082 /** 1083 * DPDK callback to bring the link UP. 1084 * 1085 * @param dev 1086 * Pointer to Ethernet device structure. 1087 * 1088 * @return 1089 * 0 on success, a negative errno value otherwise and rte_errno is set. 1090 */ 1091 int 1092 mlx5_set_link_up(struct rte_eth_dev *dev) 1093 { 1094 return mlx5_set_flags(dev, ~IFF_UP, IFF_UP); 1095 } 1096 1097 /** 1098 * Configure the TX function to use. 1099 * 1100 * @param dev 1101 * Pointer to rte_eth_dev structure. 1102 * 1103 * @return 1104 * Pointer to selected Tx burst function. 1105 */ 1106 eth_tx_burst_t 1107 mlx5_select_tx_function(struct rte_eth_dev *dev) 1108 { 1109 struct priv *priv = dev->data->dev_private; 1110 eth_tx_burst_t tx_pkt_burst = mlx5_tx_burst; 1111 1112 /* Select appropriate TX function. */ 1113 if (priv->mps == MLX5_MPW_ENHANCED) { 1114 if (mlx5_check_vec_tx_support(dev) > 0) { 1115 if (mlx5_check_raw_vec_tx_support(dev) > 0) 1116 tx_pkt_burst = mlx5_tx_burst_raw_vec; 1117 else 1118 tx_pkt_burst = mlx5_tx_burst_vec; 1119 DRV_LOG(DEBUG, 1120 "port %u selected enhanced MPW Tx vectorized" 1121 " function", 1122 dev->data->port_id); 1123 } else { 1124 tx_pkt_burst = mlx5_tx_burst_empw; 1125 DRV_LOG(DEBUG, 1126 "port %u selected enhanced MPW Tx function", 1127 dev->data->port_id); 1128 } 1129 } else if (priv->mps && priv->txq_inline) { 1130 tx_pkt_burst = mlx5_tx_burst_mpw_inline; 1131 DRV_LOG(DEBUG, "port %u selected MPW inline Tx function", 1132 dev->data->port_id); 1133 } else if (priv->mps) { 1134 tx_pkt_burst = mlx5_tx_burst_mpw; 1135 DRV_LOG(DEBUG, "port %u selected MPW Tx function", 1136 dev->data->port_id); 1137 } 1138 return tx_pkt_burst; 1139 } 1140 1141 /** 1142 * Configure the RX function to use. 1143 * 1144 * @param dev 1145 * Pointer to rte_eth_dev structure. 1146 * 1147 * @return 1148 * Pointer to selected Rx burst function. 1149 */ 1150 eth_rx_burst_t 1151 mlx5_select_rx_function(struct rte_eth_dev *dev) 1152 { 1153 eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst; 1154 1155 assert(dev != NULL); 1156 if (mlx5_check_vec_rx_support(dev) > 0) { 1157 rx_pkt_burst = mlx5_rx_burst_vec; 1158 DRV_LOG(DEBUG, "port %u selected Rx vectorized function", 1159 dev->data->port_id); 1160 } 1161 return rx_pkt_burst; 1162 } 1163