drivers/net/mlx5/mlx5_ethdev.c

   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 #include <stddef.h>
  35 #include <assert.h>
  36 #include <unistd.h>
  37 #include <stdint.h>
  38 #include <stdio.h>
  39 #include <string.h>
  40 #include <stdlib.h>
  41 #include <errno.h>
  42 #include <dirent.h>
  43 #include <net/if.h>
  44 #include <sys/ioctl.h>
  45 #include <sys/socket.h>
  46 #include <sys/utsname.h>
  47 #include <netinet/in.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/sockios.h>
  50 #include <linux/version.h>
  51 #include <fcntl.h>
  52 #include <stdalign.h>
  53
  54 #include <rte_atomic.h>
  55 #include <rte_ethdev.h>
  56 #include <rte_mbuf.h>
  57 #include <rte_common.h>
  58 #include <rte_interrupts.h>
  59 #include <rte_alarm.h>
  60 #include <rte_malloc.h>
  61
  62 #include "mlx5.h"
  63 #include "mlx5_rxtx.h"
  64 #include "mlx5_utils.h"
  65
  66 /* Add defines in case the running kernel is not the same as user headers. */
  67 #ifndef ETHTOOL_GLINKSETTINGS
  68 struct ethtool_link_settings {
  69         uint32_t cmd;
  70         uint32_t speed;
  71         uint8_t duplex;
  72         uint8_t port;
  73         uint8_t phy_address;
  74         uint8_t autoneg;
  75         uint8_t mdio_support;
  76         uint8_t eth_to_mdix;
  77         uint8_t eth_tp_mdix_ctrl;
  78         int8_t link_mode_masks_nwords;
  79         uint32_t reserved[8];
  80         uint32_t link_mode_masks[];
  81 };
  82
  83 #define ETHTOOL_GLINKSETTINGS 0x0000004c
  84 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
  85 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
  86 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
  87 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
  88 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
  89 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
  90 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
  91 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
  92 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
  93 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
  94 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
  95 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
  96 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
  97 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
  98 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
  99 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
 100 #endif
 101 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
 102 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
 103 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
 104 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
 105 #endif
 106 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
 107 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
 108 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
 109 #endif
 110 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
 111 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
 112 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
 113 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
 114 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
 115 #endif
 116
 117 /**
 118  * Return private structure associated with an Ethernet device.
 119  *
 120  * @param dev
 121  *   Pointer to Ethernet device structure.
 122  *
 123  * @return
 124  *   Pointer to private structure.
 125  */
 126 struct priv *
 127 mlx5_get_priv(struct rte_eth_dev *dev)
 128 {
 129         return dev->data->dev_private;
 130 }
 131
 132 /**
 133  * Check if running as a secondary process.
 134  *
 135  * @return
 136  *   Nonzero if running as a secondary process.
 137  */
 138 inline int
 139 mlx5_is_secondary(void)
 140 {
 141         return rte_eal_process_type() == RTE_PROC_SECONDARY;
 142 }
 143
 144 /**
 145  * Get interface name from private structure.
 146  *
 147  * @param[in] priv
 148  *   Pointer to private structure.
 149  * @param[out] ifname
 150  *   Interface name output buffer.
 151  *
 152  * @return
 153  *   0 on success, -1 on failure and errno is set.
 154  */
 155 int
 156 priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
 157 {
 158         DIR *dir;
 159         struct dirent *dent;
 160         unsigned int dev_type = 0;
 161         unsigned int dev_port_prev = ~0u;
 162         char match[IF_NAMESIZE] = "";
 163
 164         {
 165                 MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
 166
 167                 dir = opendir(path);
 168                 if (dir == NULL)
 169                         return -1;
 170         }
 171         while ((dent = readdir(dir)) != NULL) {
 172                 char *name = dent->d_name;
 173                 FILE *file;
 174                 unsigned int dev_port;
 175                 int r;
 176
 177                 if ((name[0] == '.') &&
 178                     ((name[1] == '\0') ||
 179                      ((name[1] == '.') && (name[2] == '\0'))))
 180                         continue;
 181
 182                 MKSTR(path, "%s/device/net/%s/%s",
 183                       priv->ctx->device->ibdev_path, name,
 184                       (dev_type ? "dev_id" : "dev_port"));
 185
 186                 file = fopen(path, "rb");
 187                 if (file == NULL) {
 188                         if (errno != ENOENT)
 189                                 continue;
 190                         /*
 191                          * Switch to dev_id when dev_port does not exist as
 192                          * is the case with Linux kernel versions < 3.15.
 193                          */
 194 try_dev_id:
 195                         match[0] = '\0';
 196                         if (dev_type)
 197                                 break;
 198                         dev_type = 1;
 199                         dev_port_prev = ~0u;
 200                         rewinddir(dir);
 201                         continue;
 202                 }
 203                 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
 204                 fclose(file);
 205                 if (r != 1)
 206                         continue;
 207                 /*
 208                  * Switch to dev_id when dev_port returns the same value for
 209                  * all ports. May happen when using a MOFED release older than
 210                  * 3.0 with a Linux kernel >= 3.15.
 211                  */
 212                 if (dev_port == dev_port_prev)
 213                         goto try_dev_id;
 214                 dev_port_prev = dev_port;
 215                 if (dev_port == (priv->port - 1u))
 216                         snprintf(match, sizeof(match), "%s", name);
 217         }
 218         closedir(dir);
 219         if (match[0] == '\0')
 220                 return -1;
 221         strncpy(*ifname, match, sizeof(*ifname));
 222         return 0;
 223 }
 224
 225 /**
 226  * Check if the counter is located on ib counters file.
 227  *
 228  * @param[in] cntr
 229  *   Counter name.
 230  *
 231  * @return
 232  *   1 if counter is located on ib counters file , 0 otherwise.
 233  */
 234 int
 235 priv_is_ib_cntr(const char *cntr)
 236 {
 237         if (!strcmp(cntr, "out_of_buffer"))
 238                 return 1;
 239         return 0;
 240 }
 241
 242 /**
 243  * Read from sysfs entry.
 244  *
 245  * @param[in] priv
 246  *   Pointer to private structure.
 247  * @param[in] entry
 248  *   Entry name relative to sysfs path.
 249  * @param[out] buf
 250  *   Data output buffer.
 251  * @param size
 252  *   Buffer size.
 253  *
 254  * @return
 255  *   0 on success, -1 on failure and errno is set.
 256  */
 257 static int
 258 priv_sysfs_read(const struct priv *priv, const char *entry,
 259                 char *buf, size_t size)
 260 {
 261         char ifname[IF_NAMESIZE];
 262         FILE *file;
 263         int ret;
 264         int err;
 265
 266         if (priv_get_ifname(priv, &ifname))
 267                 return -1;
 268
 269         if (priv_is_ib_cntr(entry)) {
 270                 MKSTR(path, "%s/ports/1/hw_counters/%s",
 271                       priv->ctx->device->ibdev_path, entry);
 272                 file = fopen(path, "rb");
 273         } else {
 274                 MKSTR(path, "%s/device/net/%s/%s",
 275                       priv->ctx->device->ibdev_path, ifname, entry);
 276                 file = fopen(path, "rb");
 277         }
 278         if (file == NULL)
 279                 return -1;
 280         ret = fread(buf, 1, size, file);
 281         err = errno;
 282         if (((size_t)ret < size) && (ferror(file)))
 283                 ret = -1;
 284         else
 285                 ret = size;
 286         fclose(file);
 287         errno = err;
 288         return ret;
 289 }
 290
 291 /**
 292  * Write to sysfs entry.
 293  *
 294  * @param[in] priv
 295  *   Pointer to private structure.
 296  * @param[in] entry
 297  *   Entry name relative to sysfs path.
 298  * @param[in] buf
 299  *   Data buffer.
 300  * @param size
 301  *   Buffer size.
 302  *
 303  * @return
 304  *   0 on success, -1 on failure and errno is set.
 305  */
 306 static int
 307 priv_sysfs_write(const struct priv *priv, const char *entry,
 308                  char *buf, size_t size)
 309 {
 310         char ifname[IF_NAMESIZE];
 311         FILE *file;
 312         int ret;
 313         int err;
 314
 315         if (priv_get_ifname(priv, &ifname))
 316                 return -1;
 317
 318         MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
 319               ifname, entry);
 320
 321         file = fopen(path, "wb");
 322         if (file == NULL)
 323                 return -1;
 324         ret = fwrite(buf, 1, size, file);
 325         err = errno;
 326         if (((size_t)ret < size) || (ferror(file)))
 327                 ret = -1;
 328         else
 329                 ret = size;
 330         fclose(file);
 331         errno = err;
 332         return ret;
 333 }
 334
 335 /**
 336  * Get unsigned long sysfs property.
 337  *
 338  * @param priv
 339  *   Pointer to private structure.
 340  * @param[in] name
 341  *   Entry name relative to sysfs path.
 342  * @param[out] value
 343  *   Value output buffer.
 344  *
 345  * @return
 346  *   0 on success, -1 on failure and errno is set.
 347  */
 348 static int
 349 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
 350 {
 351         int ret;
 352         unsigned long value_ret;
 353         char value_str[32];
 354
 355         ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
 356         if (ret == -1) {
 357                 DEBUG("cannot read %s value from sysfs: %s",
 358                       name, strerror(errno));
 359                 return -1;
 360         }
 361         value_str[ret] = '\0';
 362         errno = 0;
 363         value_ret = strtoul(value_str, NULL, 0);
 364         if (errno) {
 365                 DEBUG("invalid %s value `%s': %s", name, value_str,
 366                       strerror(errno));
 367                 return -1;
 368         }
 369         *value = value_ret;
 370         return 0;
 371 }
 372
 373 /**
 374  * Set unsigned long sysfs property.
 375  *
 376  * @param priv
 377  *   Pointer to private structure.
 378  * @param[in] name
 379  *   Entry name relative to sysfs path.
 380  * @param value
 381  *   Value to set.
 382  *
 383  * @return
 384  *   0 on success, -1 on failure and errno is set.
 385  */
 386 static int
 387 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
 388 {
 389         int ret;
 390         MKSTR(value_str, "%lu", value);
 391
 392         ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
 393         if (ret == -1) {
 394                 DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
 395                       name, value_str, value, strerror(errno));
 396                 return -1;
 397         }
 398         return 0;
 399 }
 400
 401 /**
 402  * Perform ifreq ioctl() on associated Ethernet device.
 403  *
 404  * @param[in] priv
 405  *   Pointer to private structure.
 406  * @param req
 407  *   Request number to pass to ioctl().
 408  * @param[out] ifr
 409  *   Interface request structure output buffer.
 410  *
 411  * @return
 412  *   0 on success, -1 on failure and errno is set.
 413  */
 414 int
 415 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
 416 {
 417         int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 418         int ret = -1;
 419
 420         if (sock == -1)
 421                 return ret;
 422         if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
 423                 ret = ioctl(sock, req, ifr);
 424         close(sock);
 425         return ret;
 426 }
 427
 428 /**
 429  * Return the number of active VFs for the current device.
 430  *
 431  * @param[in] priv
 432  *   Pointer to private structure.
 433  * @param[out] num_vfs
 434  *   Number of active VFs.
 435  *
 436  * @return
 437  *   0 on success, -1 on failure and errno is set.
 438  */
 439 int
 440 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
 441 {
 442         /* The sysfs entry name depends on the operating system. */
 443         const char **name = (const char *[]){
 444                 "device/sriov_numvfs",
 445                 "device/mlx5_num_vfs",
 446                 NULL,
 447         };
 448         int ret;
 449
 450         do {
 451                 unsigned long ulong_num_vfs;
 452
 453                 ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
 454                 if (!ret)
 455                         *num_vfs = ulong_num_vfs;
 456         } while (*(++name) && ret);
 457         return ret;
 458 }
 459
 460 /**
 461  * Get device MTU.
 462  *
 463  * @param priv
 464  *   Pointer to private structure.
 465  * @param[out] mtu
 466  *   MTU value output buffer.
 467  *
 468  * @return
 469  *   0 on success, -1 on failure and errno is set.
 470  */
 471 int
 472 priv_get_mtu(struct priv *priv, uint16_t *mtu)
 473 {
 474         unsigned long ulong_mtu;
 475
 476         if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
 477                 return -1;
 478         *mtu = ulong_mtu;
 479         return 0;
 480 }
 481
 482 /**
 483  * Read device counter from sysfs.
 484  *
 485  * @param priv
 486  *   Pointer to private structure.
 487  * @param name
 488  *   Counter name.
 489  * @param[out] cntr
 490  *   Counter output buffer.
 491  *
 492  * @return
 493  *   0 on success, -1 on failure and errno is set.
 494  */
 495 int
 496 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
 497 {
 498         unsigned long ulong_ctr;
 499
 500         if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
 501                 return -1;
 502         *cntr = ulong_ctr;
 503         return 0;
 504 }
 505
 506 /**
 507  * Set device MTU.
 508  *
 509  * @param priv
 510  *   Pointer to private structure.
 511  * @param mtu
 512  *   MTU value to set.
 513  *
 514  * @return
 515  *   0 on success, -1 on failure and errno is set.
 516  */
 517 static int
 518 priv_set_mtu(struct priv *priv, uint16_t mtu)
 519 {
 520         uint16_t new_mtu;
 521
 522         if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
 523             priv_get_mtu(priv, &new_mtu))
 524                 return -1;
 525         if (new_mtu == mtu)
 526                 return 0;
 527         errno = EINVAL;
 528         return -1;
 529 }
 530
 531 /**
 532  * Set device flags.
 533  *
 534  * @param priv
 535  *   Pointer to private structure.
 536  * @param keep
 537  *   Bitmask for flags that must remain untouched.
 538  * @param flags
 539  *   Bitmask for flags to modify.
 540  *
 541  * @return
 542  *   0 on success, -1 on failure and errno is set.
 543  */
 544 int
 545 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
 546 {
 547         unsigned long tmp;
 548
 549         if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
 550                 return -1;
 551         tmp &= keep;
 552         tmp |= (flags & (~keep));
 553         return priv_set_sysfs_ulong(priv, "flags", tmp);
 554 }
 555
 556 /**
 557  * Ethernet device configuration.
 558  *
 559  * Prepare the driver for a given number of TX and RX queues.
 560  *
 561  * @param dev
 562  *   Pointer to Ethernet device structure.
 563  *
 564  * @return
 565  *   0 on success, errno value on failure.
 566  */
 567 static int
 568 dev_configure(struct rte_eth_dev *dev)
 569 {
 570         struct priv *priv = dev->data->dev_private;
 571         unsigned int rxqs_n = dev->data->nb_rx_queues;
 572         unsigned int txqs_n = dev->data->nb_tx_queues;
 573         unsigned int i;
 574         unsigned int j;
 575         unsigned int reta_idx_n;
 576
 577         priv->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
 578         priv->rxqs = (void *)dev->data->rx_queues;
 579         priv->txqs = (void *)dev->data->tx_queues;
 580         if (txqs_n != priv->txqs_n) {
 581                 INFO("%p: TX queues number update: %u -> %u",
 582                      (void *)dev, priv->txqs_n, txqs_n);
 583                 priv->txqs_n = txqs_n;
 584         }
 585         if (rxqs_n > priv->ind_table_max_size) {
 586                 ERROR("cannot handle this many RX queues (%u)", rxqs_n);
 587                 return EINVAL;
 588         }
 589         if (rxqs_n == priv->rxqs_n)
 590                 return 0;
 591         INFO("%p: RX queues number update: %u -> %u",
 592              (void *)dev, priv->rxqs_n, rxqs_n);
 593         priv->rxqs_n = rxqs_n;
 594         /* If the requested number of RX queues is not a power of two, use the
 595          * maximum indirection table size for better balancing.
 596          * The result is always rounded to the next power of two. */
 597         reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
 598                                      priv->ind_table_max_size :
 599                                      rxqs_n));
 600         if (priv_rss_reta_index_resize(priv, reta_idx_n))
 601                 return ENOMEM;
 602         /* When the number of RX queues is not a power of two, the remaining
 603          * table entries are padded with reused WQs and hashes are not spread
 604          * uniformly. */
 605         for (i = 0, j = 0; (i != reta_idx_n); ++i) {
 606                 (*priv->reta_idx)[i] = j;
 607                 if (++j == rxqs_n)
 608                         j = 0;
 609         }
 610         return 0;
 611 }
 612
 613 /**
 614  * DPDK callback for Ethernet device configuration.
 615  *
 616  * @param dev
 617  *   Pointer to Ethernet device structure.
 618  *
 619  * @return
 620  *   0 on success, negative errno value on failure.
 621  */
 622 int
 623 mlx5_dev_configure(struct rte_eth_dev *dev)
 624 {
 625         struct priv *priv = dev->data->dev_private;
 626         int ret;
 627
 628         if (mlx5_is_secondary())
 629                 return -E_RTE_SECONDARY;
 630
 631         priv_lock(priv);
 632         ret = dev_configure(dev);
 633         assert(ret >= 0);
 634         priv_unlock(priv);
 635         return -ret;
 636 }
 637
 638 /**
 639  * DPDK callback to get information about the device.
 640  *
 641  * @param dev
 642  *   Pointer to Ethernet device structure.
 643  * @param[out] info
 644  *   Info structure output buffer.
 645  */
 646 void
 647 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 648 {
 649         struct priv *priv = mlx5_get_priv(dev);
 650         unsigned int max;
 651         char ifname[IF_NAMESIZE];
 652
 653         info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 654
 655         priv_lock(priv);
 656         /* FIXME: we should ask the device for these values. */
 657         info->min_rx_bufsize = 32;
 658         info->max_rx_pktlen = 65536;
 659         /*
 660          * Since we need one CQ per QP, the limit is the minimum number
 661          * between the two values.
 662          */
 663         max = RTE_MIN(priv->device_attr.orig_attr.max_cq,
 664                       priv->device_attr.orig_attr.max_qp);
 665         /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
 666         if (max >= 65535)
 667                 max = 65535;
 668         info->max_rx_queues = max;
 669         info->max_tx_queues = max;
 670         info->max_mac_addrs = RTE_DIM(priv->mac);
 671         info->rx_offload_capa =
 672                 (priv->hw_csum ?
 673                  (DEV_RX_OFFLOAD_IPV4_CKSUM |
 674                   DEV_RX_OFFLOAD_UDP_CKSUM |
 675                   DEV_RX_OFFLOAD_TCP_CKSUM) :
 676                  0) |
 677                 (priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0);
 678         if (!priv->mps)
 679                 info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
 680         if (priv->hw_csum)
 681                 info->tx_offload_capa |=
 682                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
 683                          DEV_TX_OFFLOAD_UDP_CKSUM |
 684                          DEV_TX_OFFLOAD_TCP_CKSUM);
 685         if (priv->tso)
 686                 info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 687         if (priv->tunnel_en)
 688                 info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
 689                                           DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
 690                                           DEV_TX_OFFLOAD_GRE_TNL_TSO);
 691         if (priv_get_ifname(priv, &ifname) == 0)
 692                 info->if_index = if_nametoindex(ifname);
 693         info->reta_size = priv->reta_idx_n ?
 694                 priv->reta_idx_n : priv->ind_table_max_size;
 695         info->hash_key_size = ((*priv->rss_conf) ?
 696                                (*priv->rss_conf)[0]->rss_key_len :
 697                                0);
 698         info->speed_capa = priv->link_speed_capa;
 699         priv_unlock(priv);
 700 }
 701
 702 const uint32_t *
 703 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 704 {
 705         static const uint32_t ptypes[] = {
 706                 /* refers to rxq_cq_to_pkt_type() */
 707                 RTE_PTYPE_L2_ETHER,
 708                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 709                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 710                 RTE_PTYPE_L4_NONFRAG,
 711                 RTE_PTYPE_L4_FRAG,
 712                 RTE_PTYPE_L4_TCP,
 713                 RTE_PTYPE_L4_UDP,
 714                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 715                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 716                 RTE_PTYPE_INNER_L4_NONFRAG,
 717                 RTE_PTYPE_INNER_L4_FRAG,
 718                 RTE_PTYPE_INNER_L4_TCP,
 719                 RTE_PTYPE_INNER_L4_UDP,
 720                 RTE_PTYPE_UNKNOWN
 721         };
 722
 723         if (dev->rx_pkt_burst == mlx5_rx_burst ||
 724             dev->rx_pkt_burst == mlx5_rx_burst_vec)
 725                 return ptypes;
 726         return NULL;
 727 }
 728
 729 /**
 730  * DPDK callback to retrieve physical link information.
 731  *
 732  * @param dev
 733  *   Pointer to Ethernet device structure.
 734  * @param wait_to_complete
 735  *   Wait for request completion (ignored).
 736  */
 737 static int
 738 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
 739 {
 740         struct priv *priv = mlx5_get_priv(dev);
 741         struct ethtool_cmd edata = {
 742                 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
 743         };
 744         struct ifreq ifr;
 745         struct rte_eth_link dev_link;
 746         int link_speed = 0;
 747
 748         /* priv_lock() is not taken to allow concurrent calls. */
 749
 750         (void)wait_to_complete;
 751         if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
 752                 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
 753                 return -1;
 754         }
 755         memset(&dev_link, 0, sizeof(dev_link));
 756         dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
 757                                 (ifr.ifr_flags & IFF_RUNNING));
 758         ifr.ifr_data = (void *)&edata;
 759         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 760                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
 761                      strerror(errno));
 762                 return -1;
 763         }
 764         link_speed = ethtool_cmd_speed(&edata);
 765         if (link_speed == -1)
 766                 dev_link.link_speed = 0;
 767         else
 768                 dev_link.link_speed = link_speed;
 769         priv->link_speed_capa = 0;
 770         if (edata.supported & SUPPORTED_Autoneg)
 771                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 772         if (edata.supported & (SUPPORTED_1000baseT_Full |
 773                                SUPPORTED_1000baseKX_Full))
 774                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 775         if (edata.supported & SUPPORTED_10000baseKR_Full)
 776                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 777         if (edata.supported & (SUPPORTED_40000baseKR4_Full |
 778                                SUPPORTED_40000baseCR4_Full |
 779                                SUPPORTED_40000baseSR4_Full |
 780                                SUPPORTED_40000baseLR4_Full))
 781                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 782         dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
 783                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 784         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 785                         ETH_LINK_SPEED_FIXED);
 786         if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
 787                 /* Link status changed. */
 788                 dev->data->dev_link = dev_link;
 789                 return 0;
 790         }
 791         /* Link status is still the same. */
 792         return -1;
 793 }
 794
 795 /**
 796  * Retrieve physical link information (unlocked version using new ioctl).
 797  *
 798  * @param dev
 799  *   Pointer to Ethernet device structure.
 800  * @param wait_to_complete
 801  *   Wait for request completion (ignored).
 802  */
 803 static int
 804 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
 805 {
 806         struct priv *priv = mlx5_get_priv(dev);
 807         struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
 808         struct ifreq ifr;
 809         struct rte_eth_link dev_link;
 810         uint64_t sc;
 811
 812         (void)wait_to_complete;
 813         if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
 814                 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
 815                 return -1;
 816         }
 817         memset(&dev_link, 0, sizeof(dev_link));
 818         dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
 819                                 (ifr.ifr_flags & IFF_RUNNING));
 820         ifr.ifr_data = (void *)&gcmd;
 821         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 822                 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
 823                       strerror(errno));
 824                 return -1;
 825         }
 826         gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
 827
 828         alignas(struct ethtool_link_settings)
 829         uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
 830                      sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
 831         struct ethtool_link_settings *ecmd = (void *)data;
 832
 833         *ecmd = gcmd;
 834         ifr.ifr_data = (void *)ecmd;
 835         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 836                 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
 837                       strerror(errno));
 838                 return -1;
 839         }
 840         dev_link.link_speed = ecmd->speed;
 841         sc = ecmd->link_mode_masks[0] |
 842                 ((uint64_t)ecmd->link_mode_masks[1] << 32);
 843         priv->link_speed_capa = 0;
 844         if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
 845                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 846         if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
 847                   ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
 848                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 849         if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
 850                   ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
 851                   ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
 852                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 853         if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
 854                   ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
 855                 priv->link_speed_capa |= ETH_LINK_SPEED_20G;
 856         if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
 857                   ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
 858                   ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
 859                   ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
 860                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 861         if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
 862                   ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
 863                   ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
 864                   ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
 865                 priv->link_speed_capa |= ETH_LINK_SPEED_56G;
 866         if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
 867                   ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
 868                   ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
 869                 priv->link_speed_capa |= ETH_LINK_SPEED_25G;
 870         if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
 871                   ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
 872                 priv->link_speed_capa |= ETH_LINK_SPEED_50G;
 873         if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
 874                   ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
 875                   ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
 876                   ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
 877                 priv->link_speed_capa |= ETH_LINK_SPEED_100G;
 878         dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
 879                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 880         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 881                                   ETH_LINK_SPEED_FIXED);
 882         if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
 883                 /* Link status changed. */
 884                 dev->data->dev_link = dev_link;
 885                 return 0;
 886         }
 887         /* Link status is still the same. */
 888         return -1;
 889 }
 890
 891 /**
 892  * DPDK callback to retrieve physical link information.
 893  *
 894  * @param dev
 895  *   Pointer to Ethernet device structure.
 896  * @param wait_to_complete
 897  *   Wait for request completion (ignored).
 898  */
 899 int
 900 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 901 {
 902         struct utsname utsname;
 903         int ver[3];
 904
 905         if (uname(&utsname) == -1 ||
 906             sscanf(utsname.release, "%d.%d.%d",
 907                    &ver[0], &ver[1], &ver[2]) != 3 ||
 908             KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
 909                 return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
 910         return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
 911 }
 912
 913 /**
 914  * DPDK callback to change the MTU.
 915  *
 916  * @param dev
 917  *   Pointer to Ethernet device structure.
 918  * @param in_mtu
 919  *   New MTU.
 920  *
 921  * @return
 922  *   0 on success, negative errno value on failure.
 923  */
 924 int
 925 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 926 {
 927         struct priv *priv = dev->data->dev_private;
 928         uint16_t kern_mtu;
 929         int ret = 0;
 930
 931         if (mlx5_is_secondary())
 932                 return -E_RTE_SECONDARY;
 933
 934         priv_lock(priv);
 935         ret = priv_get_mtu(priv, &kern_mtu);
 936         if (ret)
 937                 goto out;
 938         /* Set kernel interface MTU first. */
 939         ret = priv_set_mtu(priv, mtu);
 940         if (ret)
 941                 goto out;
 942         ret = priv_get_mtu(priv, &kern_mtu);
 943         if (ret)
 944                 goto out;
 945         if (kern_mtu == mtu) {
 946                 priv->mtu = mtu;
 947                 DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
 948         }
 949         priv_unlock(priv);
 950         return 0;
 951 out:
 952         ret = errno;
 953         WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
 954              strerror(ret));
 955         priv_unlock(priv);
 956         assert(ret >= 0);
 957         return -ret;
 958 }
 959
 960 /**
 961  * DPDK callback to get flow control status.
 962  *
 963  * @param dev
 964  *   Pointer to Ethernet device structure.
 965  * @param[out] fc_conf
 966  *   Flow control output buffer.
 967  *
 968  * @return
 969  *   0 on success, negative errno value on failure.
 970  */
 971 int
 972 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 973 {
 974         struct priv *priv = dev->data->dev_private;
 975         struct ifreq ifr;
 976         struct ethtool_pauseparam ethpause = {
 977                 .cmd = ETHTOOL_GPAUSEPARAM
 978         };
 979         int ret;
 980
 981         if (mlx5_is_secondary())
 982                 return -E_RTE_SECONDARY;
 983
 984         ifr.ifr_data = (void *)&ethpause;
 985         priv_lock(priv);
 986         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 987                 ret = errno;
 988                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
 989                      " failed: %s",
 990                      strerror(ret));
 991                 goto out;
 992         }
 993
 994         fc_conf->autoneg = ethpause.autoneg;
 995         if (ethpause.rx_pause && ethpause.tx_pause)
 996                 fc_conf->mode = RTE_FC_FULL;
 997         else if (ethpause.rx_pause)
 998                 fc_conf->mode = RTE_FC_RX_PAUSE;
 999         else if (ethpause.tx_pause)
1000                 fc_conf->mode = RTE_FC_TX_PAUSE;
1001         else
1002                 fc_conf->mode = RTE_FC_NONE;
1003         ret = 0;
1004
1005 out:
1006         priv_unlock(priv);
1007         assert(ret >= 0);
1008         return -ret;
1009 }
1010
1011 /**
1012  * DPDK callback to modify flow control parameters.
1013  *
1014  * @param dev
1015  *   Pointer to Ethernet device structure.
1016  * @param[in] fc_conf
1017  *   Flow control parameters.
1018  *
1019  * @return
1020  *   0 on success, negative errno value on failure.
1021  */
1022 int
1023 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1024 {
1025         struct priv *priv = dev->data->dev_private;
1026         struct ifreq ifr;
1027         struct ethtool_pauseparam ethpause = {
1028                 .cmd = ETHTOOL_SPAUSEPARAM
1029         };
1030         int ret;
1031
1032         if (mlx5_is_secondary())
1033                 return -E_RTE_SECONDARY;
1034
1035         ifr.ifr_data = (void *)&ethpause;
1036         ethpause.autoneg = fc_conf->autoneg;
1037         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1038             (fc_conf->mode & RTE_FC_RX_PAUSE))
1039                 ethpause.rx_pause = 1;
1040         else
1041                 ethpause.rx_pause = 0;
1042
1043         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1044             (fc_conf->mode & RTE_FC_TX_PAUSE))
1045                 ethpause.tx_pause = 1;
1046         else
1047                 ethpause.tx_pause = 0;
1048
1049         priv_lock(priv);
1050         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1051                 ret = errno;
1052                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1053                      " failed: %s",
1054                      strerror(ret));
1055                 goto out;
1056         }
1057         ret = 0;
1058
1059 out:
1060         priv_unlock(priv);
1061         assert(ret >= 0);
1062         return -ret;
1063 }
1064
1065 /**
1066  * Get PCI information from struct ibv_device.
1067  *
1068  * @param device
1069  *   Pointer to Ethernet device structure.
1070  * @param[out] pci_addr
1071  *   PCI bus address output buffer.
1072  *
1073  * @return
1074  *   0 on success, -1 on failure and errno is set.
1075  */
1076 int
1077 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1078                             struct rte_pci_addr *pci_addr)
1079 {
1080         FILE *file;
1081         char line[32];
1082         MKSTR(path, "%s/device/uevent", device->ibdev_path);
1083
1084         file = fopen(path, "rb");
1085         if (file == NULL)
1086                 return -1;
1087         while (fgets(line, sizeof(line), file) == line) {
1088                 size_t len = strlen(line);
1089                 int ret;
1090
1091                 /* Truncate long lines. */
1092                 if (len == (sizeof(line) - 1))
1093                         while (line[(len - 1)] != '\n') {
1094                                 ret = fgetc(file);
1095                                 if (ret == EOF)
1096                                         break;
1097                                 line[(len - 1)] = ret;
1098                         }
1099                 /* Extract information. */
1100                 if (sscanf(line,
1101                            "PCI_SLOT_NAME="
1102                            "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1103                            &pci_addr->domain,
1104                            &pci_addr->bus,
1105                            &pci_addr->devid,
1106                            &pci_addr->function) == 4) {
1107                         ret = 0;
1108                         break;
1109                 }
1110         }
1111         fclose(file);
1112         return 0;
1113 }
1114
1115 /**
1116  * Update the link status.
1117  *
1118  * @param priv
1119  *   Pointer to private structure.
1120  *
1121  * @return
1122  *   Zero if the callback process can be called immediately.
1123  */
1124 static int
1125 priv_link_status_update(struct priv *priv)
1126 {
1127         struct rte_eth_link *link = &priv->dev->data->dev_link;
1128
1129         mlx5_link_update(priv->dev, 0);
1130         if (((link->link_speed == 0) && link->link_status) ||
1131                 ((link->link_speed != 0) && !link->link_status)) {
1132                 /*
1133                  * Inconsistent status. Event likely occurred before the
1134                  * kernel netdevice exposes the new status.
1135                  */
1136                 if (!priv->pending_alarm) {
1137                         priv->pending_alarm = 1;
1138                         rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1139                                           mlx5_dev_link_status_handler,
1140                                           priv->dev);
1141                 }
1142                 return 1;
1143         } else if (unlikely(priv->pending_alarm)) {
1144                 /* Link interrupt occurred while alarm is already scheduled. */
1145                 priv->pending_alarm = 0;
1146                 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, priv->dev);
1147         }
1148         return 0;
1149 }
1150
1151 /**
1152  * Device status handler.
1153  *
1154  * @param priv
1155  *   Pointer to private structure.
1156  * @param events
1157  *   Pointer to event flags holder.
1158  *
1159  * @return
1160  *   Events bitmap of callback process which can be called immediately.
1161  */
1162 static uint32_t
1163 priv_dev_status_handler(struct priv *priv)
1164 {
1165         struct ibv_async_event event;
1166         uint32_t ret = 0;
1167
1168         /* Read all message and acknowledge them. */
1169         for (;;) {
1170                 if (ibv_get_async_event(priv->ctx, &event))
1171                         break;
1172                 if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1173                         event.event_type == IBV_EVENT_PORT_ERR) &&
1174                         (priv->dev->data->dev_conf.intr_conf.lsc == 1))
1175                         ret |= (1 << RTE_ETH_EVENT_INTR_LSC);
1176                 else if (event.event_type == IBV_EVENT_DEVICE_FATAL &&
1177                         priv->dev->data->dev_conf.intr_conf.rmv == 1)
1178                         ret |= (1 << RTE_ETH_EVENT_INTR_RMV);
1179                 else
1180                         DEBUG("event type %d on port %d not handled",
1181                               event.event_type, event.element.port_num);
1182                 ibv_ack_async_event(&event);
1183         }
1184         if (ret & (1 << RTE_ETH_EVENT_INTR_LSC))
1185                 if (priv_link_status_update(priv))
1186                         ret &= ~(1 << RTE_ETH_EVENT_INTR_LSC);
1187         return ret;
1188 }
1189
1190 /**
1191  * Handle delayed link status event.
1192  *
1193  * @param arg
1194  *   Registered argument.
1195  */
1196 void
1197 mlx5_dev_link_status_handler(void *arg)
1198 {
1199         struct rte_eth_dev *dev = arg;
1200         struct priv *priv = dev->data->dev_private;
1201         int ret;
1202
1203         priv_lock(priv);
1204         assert(priv->pending_alarm == 1);
1205         priv->pending_alarm = 0;
1206         ret = priv_link_status_update(priv);
1207         priv_unlock(priv);
1208         if (!ret)
1209                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1210                                               NULL);
1211 }
1212
1213 /**
1214  * Handle interrupts from the NIC.
1215  *
1216  * @param[in] intr_handle
1217  *   Interrupt handler.
1218  * @param cb_arg
1219  *   Callback argument.
1220  */
1221 void
1222 mlx5_dev_interrupt_handler(void *cb_arg)
1223 {
1224         struct rte_eth_dev *dev = cb_arg;
1225         struct priv *priv = dev->data->dev_private;
1226         uint32_t events;
1227
1228         priv_lock(priv);
1229         events = priv_dev_status_handler(priv);
1230         priv_unlock(priv);
1231         if (events & (1 << RTE_ETH_EVENT_INTR_LSC))
1232                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1233                                               NULL);
1234         if (events & (1 << RTE_ETH_EVENT_INTR_RMV))
1235                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL,
1236                                               NULL);
1237 }
1238
1239 /**
1240  * Uninstall interrupt handler.
1241  *
1242  * @param priv
1243  *   Pointer to private structure.
1244  * @param dev
1245  *   Pointer to the rte_eth_dev structure.
1246  */
1247 void
1248 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1249 {
1250         if (!dev->data->dev_conf.intr_conf.lsc &&
1251                 !dev->data->dev_conf.intr_conf.rmv)
1252                 return;
1253         rte_intr_callback_unregister(&priv->intr_handle,
1254                                      mlx5_dev_interrupt_handler,
1255                                      dev);
1256         if (priv->pending_alarm)
1257                 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1258         priv->pending_alarm = 0;
1259         priv->intr_handle.fd = 0;
1260         priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1261 }
1262
1263 /**
1264  * Install interrupt handler.
1265  *
1266  * @param priv
1267  *   Pointer to private structure.
1268  * @param dev
1269  *   Pointer to the rte_eth_dev structure.
1270  */
1271 void
1272 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1273 {
1274         int rc, flags;
1275
1276         if (!dev->data->dev_conf.intr_conf.lsc &&
1277                 !dev->data->dev_conf.intr_conf.rmv)
1278                 return;
1279         assert(priv->ctx->async_fd > 0);
1280         flags = fcntl(priv->ctx->async_fd, F_GETFL);
1281         rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1282         if (rc < 0) {
1283                 INFO("failed to change file descriptor async event queue");
1284                 dev->data->dev_conf.intr_conf.lsc = 0;
1285                 dev->data->dev_conf.intr_conf.rmv = 0;
1286         } else {
1287                 priv->intr_handle.fd = priv->ctx->async_fd;
1288                 priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1289                 rte_intr_callback_register(&priv->intr_handle,
1290                                            mlx5_dev_interrupt_handler,
1291                                            dev);
1292         }
1293 }
1294
1295 /**
1296  * Change the link state (UP / DOWN).
1297  *
1298  * @param priv
1299  *   Pointer to private data structure.
1300  * @param dev
1301  *   Pointer to rte_eth_dev structure.
1302  * @param up
1303  *   Nonzero for link up, otherwise link down.
1304  *
1305  * @return
1306  *   0 on success, errno value on failure.
1307  */
1308 static int
1309 priv_dev_set_link(struct priv *priv, struct rte_eth_dev *dev, int up)
1310 {
1311         int err;
1312
1313         if (up) {
1314                 err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1315                 if (err)
1316                         return err;
1317                 priv_dev_select_tx_function(priv, dev);
1318                 priv_dev_select_rx_function(priv, dev);
1319         } else {
1320                 err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1321                 if (err)
1322                         return err;
1323                 dev->rx_pkt_burst = removed_rx_burst;
1324                 dev->tx_pkt_burst = removed_tx_burst;
1325         }
1326         return 0;
1327 }
1328
1329 /**
1330  * DPDK callback to bring the link DOWN.
1331  *
1332  * @param dev
1333  *   Pointer to Ethernet device structure.
1334  *
1335  * @return
1336  *   0 on success, errno value on failure.
1337  */
1338 int
1339 mlx5_set_link_down(struct rte_eth_dev *dev)
1340 {
1341         struct priv *priv = dev->data->dev_private;
1342         int err;
1343
1344         priv_lock(priv);
1345         err = priv_dev_set_link(priv, dev, 0);
1346         priv_unlock(priv);
1347         return err;
1348 }
1349
1350 /**
1351  * DPDK callback to bring the link UP.
1352  *
1353  * @param dev
1354  *   Pointer to Ethernet device structure.
1355  *
1356  * @return
1357  *   0 on success, errno value on failure.
1358  */
1359 int
1360 mlx5_set_link_up(struct rte_eth_dev *dev)
1361 {
1362         struct priv *priv = dev->data->dev_private;
1363         int err;
1364
1365         priv_lock(priv);
1366         err = priv_dev_set_link(priv, dev, 1);
1367         priv_unlock(priv);
1368         return err;
1369 }
1370
1371 /**
1372  * Configure the TX function to use.
1373  *
1374  * @param priv
1375  *   Pointer to private data structure.
1376  * @param dev
1377  *   Pointer to rte_eth_dev structure.
1378  */
1379 void
1380 priv_dev_select_tx_function(struct priv *priv, struct rte_eth_dev *dev)
1381 {
1382         assert(priv != NULL);
1383         assert(dev != NULL);
1384         dev->tx_pkt_burst = mlx5_tx_burst;
1385         /* Select appropriate TX function. */
1386         if (priv->mps == MLX5_MPW_ENHANCED) {
1387                 if (priv_check_vec_tx_support(priv) > 0) {
1388                         if (priv_check_raw_vec_tx_support(priv) > 0)
1389                                 dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1390                         else
1391                                 dev->tx_pkt_burst = mlx5_tx_burst_vec;
1392                         DEBUG("selected Enhanced MPW TX vectorized function");
1393                 } else {
1394                         dev->tx_pkt_burst = mlx5_tx_burst_empw;
1395                         DEBUG("selected Enhanced MPW TX function");
1396                 }
1397         } else if (priv->mps && priv->txq_inline) {
1398                 dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1399                 DEBUG("selected MPW inline TX function");
1400         } else if (priv->mps) {
1401                 dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1402                 DEBUG("selected MPW TX function");
1403         }
1404 }
1405
1406 /**
1407  * Configure the RX function to use.
1408  *
1409  * @param priv
1410  *   Pointer to private data structure.
1411  * @param dev
1412  *   Pointer to rte_eth_dev structure.
1413  */
1414 void
1415 priv_dev_select_rx_function(struct priv *priv, struct rte_eth_dev *dev)
1416 {
1417         assert(priv != NULL);
1418         assert(dev != NULL);
1419         if (priv_check_vec_rx_support(priv) > 0) {
1420                 dev->rx_pkt_burst = mlx5_rx_burst_vec;
1421                 DEBUG("selected RX vectorized function");
1422         } else {
1423                 dev->rx_pkt_burst = mlx5_rx_burst;
1424         }
1425 }