drivers/net/failsafe/failsafe_ops.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2017 6WIND S.A.
   3  * Copyright 2017 Mellanox.
   4  */
   5
   6 #include <stdbool.h>
   7 #include <stdint.h>
   8 #include <unistd.h>
   9
  10 #include <rte_debug.h>
  11 #include <rte_atomic.h>
  12 #include <rte_ethdev_driver.h>
  13 #include <rte_malloc.h>
  14 #include <rte_flow.h>
  15 #include <rte_cycles.h>
  16
  17 #include "failsafe_private.h"
  18
  19 static struct rte_eth_dev_info default_infos = {
  20         /* Max possible number of elements */
  21         .max_rx_pktlen = UINT32_MAX,
  22         .max_rx_queues = RTE_MAX_QUEUES_PER_PORT,
  23         .max_tx_queues = RTE_MAX_QUEUES_PER_PORT,
  24         .max_mac_addrs = FAILSAFE_MAX_ETHADDR,
  25         .max_hash_mac_addrs = UINT32_MAX,
  26         .max_vfs = UINT16_MAX,
  27         .max_vmdq_pools = UINT16_MAX,
  28         .rx_desc_lim = {
  29                 .nb_max = UINT16_MAX,
  30                 .nb_min = 0,
  31                 .nb_align = 1,
  32                 .nb_seg_max = UINT16_MAX,
  33                 .nb_mtu_seg_max = UINT16_MAX,
  34         },
  35         .tx_desc_lim = {
  36                 .nb_max = UINT16_MAX,
  37                 .nb_min = 0,
  38                 .nb_align = 1,
  39                 .nb_seg_max = UINT16_MAX,
  40                 .nb_mtu_seg_max = UINT16_MAX,
  41         },
  42         /*
  43          * Set of capabilities that can be verified upon
  44          * configuring a sub-device.
  45          */
  46         .rx_offload_capa =
  47                 DEV_RX_OFFLOAD_VLAN_STRIP |
  48                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  49                 DEV_RX_OFFLOAD_UDP_CKSUM |
  50                 DEV_RX_OFFLOAD_TCP_CKSUM |
  51                 DEV_RX_OFFLOAD_TCP_LRO |
  52                 DEV_RX_OFFLOAD_QINQ_STRIP |
  53                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  54                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  55                 DEV_RX_OFFLOAD_HEADER_SPLIT |
  56                 DEV_RX_OFFLOAD_VLAN_FILTER |
  57                 DEV_RX_OFFLOAD_VLAN_EXTEND |
  58                 DEV_RX_OFFLOAD_JUMBO_FRAME |
  59                 DEV_RX_OFFLOAD_CRC_STRIP |
  60                 DEV_RX_OFFLOAD_SCATTER |
  61                 DEV_RX_OFFLOAD_TIMESTAMP |
  62                 DEV_RX_OFFLOAD_SECURITY,
  63         .rx_queue_offload_capa =
  64                 DEV_RX_OFFLOAD_VLAN_STRIP |
  65                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  66                 DEV_RX_OFFLOAD_UDP_CKSUM |
  67                 DEV_RX_OFFLOAD_TCP_CKSUM |
  68                 DEV_RX_OFFLOAD_TCP_LRO |
  69                 DEV_RX_OFFLOAD_QINQ_STRIP |
  70                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  71                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  72                 DEV_RX_OFFLOAD_HEADER_SPLIT |
  73                 DEV_RX_OFFLOAD_VLAN_FILTER |
  74                 DEV_RX_OFFLOAD_VLAN_EXTEND |
  75                 DEV_RX_OFFLOAD_JUMBO_FRAME |
  76                 DEV_RX_OFFLOAD_CRC_STRIP |
  77                 DEV_RX_OFFLOAD_SCATTER |
  78                 DEV_RX_OFFLOAD_TIMESTAMP |
  79                 DEV_RX_OFFLOAD_SECURITY,
  80         .tx_offload_capa =
  81                 DEV_TX_OFFLOAD_MULTI_SEGS |
  82                 DEV_TX_OFFLOAD_IPV4_CKSUM |
  83                 DEV_TX_OFFLOAD_UDP_CKSUM |
  84                 DEV_TX_OFFLOAD_TCP_CKSUM,
  85         .flow_type_rss_offloads = 0x0,
  86 };
  87
  88 static int
  89 fs_dev_configure(struct rte_eth_dev *dev)
  90 {
  91         struct sub_device *sdev;
  92         uint64_t supp_tx_offloads;
  93         uint64_t tx_offloads;
  94         uint8_t i;
  95         int ret;
  96
  97         supp_tx_offloads = PRIV(dev)->infos.tx_offload_capa;
  98         tx_offloads = dev->data->dev_conf.txmode.offloads;
  99         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 100                 rte_errno = ENOTSUP;
 101                 ERROR("Some Tx offloads are not supported, "
 102                       "requested 0x%" PRIx64 " supported 0x%" PRIx64,
 103                       tx_offloads, supp_tx_offloads);
 104                 return -rte_errno;
 105         }
 106         FOREACH_SUBDEV(sdev, i, dev) {
 107                 int rmv_interrupt = 0;
 108                 int lsc_interrupt = 0;
 109                 int lsc_enabled;
 110
 111                 if (sdev->state != DEV_PROBED)
 112                         continue;
 113
 114                 rmv_interrupt = ETH(sdev)->data->dev_flags &
 115                                 RTE_ETH_DEV_INTR_RMV;
 116                 if (rmv_interrupt) {
 117                         DEBUG("Enabling RMV interrupts for sub_device %d", i);
 118                         dev->data->dev_conf.intr_conf.rmv = 1;
 119                 } else {
 120                         DEBUG("sub_device %d does not support RMV event", i);
 121                 }
 122                 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
 123                 lsc_interrupt = lsc_enabled &&
 124                                 (ETH(sdev)->data->dev_flags &
 125                                  RTE_ETH_DEV_INTR_LSC);
 126                 if (lsc_interrupt) {
 127                         DEBUG("Enabling LSC interrupts for sub_device %d", i);
 128                         dev->data->dev_conf.intr_conf.lsc = 1;
 129                 } else if (lsc_enabled && !lsc_interrupt) {
 130                         DEBUG("Disabling LSC interrupts for sub_device %d", i);
 131                         dev->data->dev_conf.intr_conf.lsc = 0;
 132                 }
 133                 DEBUG("Configuring sub-device %d", i);
 134                 sdev->remove = 0;
 135                 ret = rte_eth_dev_configure(PORT_ID(sdev),
 136                                         dev->data->nb_rx_queues,
 137                                         dev->data->nb_tx_queues,
 138                                         &dev->data->dev_conf);
 139                 if (ret) {
 140                         if (!fs_err(sdev, ret))
 141                                 continue;
 142                         ERROR("Could not configure sub_device %d", i);
 143                         return ret;
 144                 }
 145                 if (rmv_interrupt) {
 146                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 147                                         RTE_ETH_EVENT_INTR_RMV,
 148                                         failsafe_eth_rmv_event_callback,
 149                                         sdev);
 150                         if (ret)
 151                                 WARN("Failed to register RMV callback for sub_device %d",
 152                                      SUB_ID(sdev));
 153                 }
 154                 dev->data->dev_conf.intr_conf.rmv = 0;
 155                 if (lsc_interrupt) {
 156                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 157                                                 RTE_ETH_EVENT_INTR_LSC,
 158                                                 failsafe_eth_lsc_event_callback,
 159                                                 dev);
 160                         if (ret)
 161                                 WARN("Failed to register LSC callback for sub_device %d",
 162                                      SUB_ID(sdev));
 163                 }
 164                 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
 165                 sdev->state = DEV_ACTIVE;
 166         }
 167         if (PRIV(dev)->state < DEV_ACTIVE)
 168                 PRIV(dev)->state = DEV_ACTIVE;
 169         return 0;
 170 }
 171
 172 static int
 173 fs_dev_start(struct rte_eth_dev *dev)
 174 {
 175         struct sub_device *sdev;
 176         uint8_t i;
 177         int ret;
 178
 179         ret = failsafe_rx_intr_install(dev);
 180         if (ret)
 181                 return ret;
 182         FOREACH_SUBDEV(sdev, i, dev) {
 183                 if (sdev->state != DEV_ACTIVE)
 184                         continue;
 185                 DEBUG("Starting sub_device %d", i);
 186                 ret = rte_eth_dev_start(PORT_ID(sdev));
 187                 if (ret) {
 188                         if (!fs_err(sdev, ret))
 189                                 continue;
 190                         return ret;
 191                 }
 192                 ret = failsafe_rx_intr_install_subdevice(sdev);
 193                 if (ret) {
 194                         if (!fs_err(sdev, ret))
 195                                 continue;
 196                         rte_eth_dev_stop(PORT_ID(sdev));
 197                         return ret;
 198                 }
 199                 sdev->state = DEV_STARTED;
 200         }
 201         if (PRIV(dev)->state < DEV_STARTED)
 202                 PRIV(dev)->state = DEV_STARTED;
 203         fs_switch_dev(dev, NULL);
 204         return 0;
 205 }
 206
 207 static void
 208 fs_dev_stop(struct rte_eth_dev *dev)
 209 {
 210         struct sub_device *sdev;
 211         uint8_t i;
 212
 213         PRIV(dev)->state = DEV_STARTED - 1;
 214         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 215                 rte_eth_dev_stop(PORT_ID(sdev));
 216                 failsafe_rx_intr_uninstall_subdevice(sdev);
 217                 sdev->state = DEV_STARTED - 1;
 218         }
 219         failsafe_rx_intr_uninstall(dev);
 220 }
 221
 222 static int
 223 fs_dev_set_link_up(struct rte_eth_dev *dev)
 224 {
 225         struct sub_device *sdev;
 226         uint8_t i;
 227         int ret;
 228
 229         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 230                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 231                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 232                 if ((ret = fs_err(sdev, ret))) {
 233                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 234                               " with error %d", i, ret);
 235                         return ret;
 236                 }
 237         }
 238         return 0;
 239 }
 240
 241 static int
 242 fs_dev_set_link_down(struct rte_eth_dev *dev)
 243 {
 244         struct sub_device *sdev;
 245         uint8_t i;
 246         int ret;
 247
 248         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 249                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 250                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 251                 if ((ret = fs_err(sdev, ret))) {
 252                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 253                               " with error %d", i, ret);
 254                         return ret;
 255                 }
 256         }
 257         return 0;
 258 }
 259
 260 static void fs_dev_free_queues(struct rte_eth_dev *dev);
 261 static void
 262 fs_dev_close(struct rte_eth_dev *dev)
 263 {
 264         struct sub_device *sdev;
 265         uint8_t i;
 266
 267         failsafe_hotplug_alarm_cancel(dev);
 268         if (PRIV(dev)->state == DEV_STARTED)
 269                 dev->dev_ops->dev_stop(dev);
 270         PRIV(dev)->state = DEV_ACTIVE - 1;
 271         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 272                 DEBUG("Closing sub_device %d", i);
 273                 rte_eth_dev_close(PORT_ID(sdev));
 274                 sdev->state = DEV_ACTIVE - 1;
 275         }
 276         fs_dev_free_queues(dev);
 277 }
 278
 279 static bool
 280 fs_rxq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 281 {
 282         uint64_t port_offloads;
 283         uint64_t queue_supp_offloads;
 284         uint64_t port_supp_offloads;
 285
 286         port_offloads = dev->data->dev_conf.rxmode.offloads;
 287         queue_supp_offloads = PRIV(dev)->infos.rx_queue_offload_capa;
 288         port_supp_offloads = PRIV(dev)->infos.rx_offload_capa;
 289         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 290              offloads)
 291                 return false;
 292         /* Verify we have no conflict with port offloads */
 293         if ((port_offloads ^ offloads) & port_supp_offloads)
 294                 return false;
 295         return true;
 296 }
 297
 298 static void
 299 fs_rx_queue_release(void *queue)
 300 {
 301         struct rte_eth_dev *dev;
 302         struct sub_device *sdev;
 303         uint8_t i;
 304         struct rxq *rxq;
 305
 306         if (queue == NULL)
 307                 return;
 308         rxq = queue;
 309         if (rxq->event_fd > 0)
 310                 close(rxq->event_fd);
 311         dev = rxq->priv->dev;
 312         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 313                 SUBOPS(sdev, rx_queue_release)
 314                         (ETH(sdev)->data->rx_queues[rxq->qid]);
 315         dev->data->rx_queues[rxq->qid] = NULL;
 316         rte_free(rxq);
 317 }
 318
 319 static int
 320 fs_rx_queue_setup(struct rte_eth_dev *dev,
 321                 uint16_t rx_queue_id,
 322                 uint16_t nb_rx_desc,
 323                 unsigned int socket_id,
 324                 const struct rte_eth_rxconf *rx_conf,
 325                 struct rte_mempool *mb_pool)
 326 {
 327         /*
 328          * FIXME: Add a proper interface in rte_eal_interrupts for
 329          * allocating eventfd as an interrupt vector.
 330          * For the time being, fake as if we are using MSIX interrupts,
 331          * this will cause rte_intr_efd_enable to allocate an eventfd for us.
 332          */
 333         struct rte_intr_handle intr_handle = {
 334                 .type = RTE_INTR_HANDLE_VFIO_MSIX,
 335                 .efds = { -1, },
 336         };
 337         struct sub_device *sdev;
 338         struct rxq *rxq;
 339         uint8_t i;
 340         int ret;
 341
 342         rxq = dev->data->rx_queues[rx_queue_id];
 343         if (rxq != NULL) {
 344                 fs_rx_queue_release(rxq);
 345                 dev->data->rx_queues[rx_queue_id] = NULL;
 346         }
 347         /* Verify application offloads are valid for our port and queue. */
 348         if (fs_rxq_offloads_valid(dev, rx_conf->offloads) == false) {
 349                 rte_errno = ENOTSUP;
 350                 ERROR("Rx queue offloads 0x%" PRIx64
 351                       " don't match port offloads 0x%" PRIx64
 352                       " or supported offloads 0x%" PRIx64,
 353                       rx_conf->offloads,
 354                       dev->data->dev_conf.rxmode.offloads,
 355                       PRIV(dev)->infos.rx_offload_capa |
 356                       PRIV(dev)->infos.rx_queue_offload_capa);
 357                 return -rte_errno;
 358         }
 359         rxq = rte_zmalloc(NULL,
 360                           sizeof(*rxq) +
 361                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 362                           RTE_CACHE_LINE_SIZE);
 363         if (rxq == NULL)
 364                 return -ENOMEM;
 365         FOREACH_SUBDEV(sdev, i, dev)
 366                 rte_atomic64_init(&rxq->refcnt[i]);
 367         rxq->qid = rx_queue_id;
 368         rxq->socket_id = socket_id;
 369         rxq->info.mp = mb_pool;
 370         rxq->info.conf = *rx_conf;
 371         rxq->info.nb_desc = nb_rx_desc;
 372         rxq->priv = PRIV(dev);
 373         rxq->sdev = PRIV(dev)->subs;
 374         ret = rte_intr_efd_enable(&intr_handle, 1);
 375         if (ret < 0)
 376                 return ret;
 377         rxq->event_fd = intr_handle.efds[0];
 378         dev->data->rx_queues[rx_queue_id] = rxq;
 379         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 380                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 381                                 rx_queue_id,
 382                                 nb_rx_desc, socket_id,
 383                                 rx_conf, mb_pool);
 384                 if ((ret = fs_err(sdev, ret))) {
 385                         ERROR("RX queue setup failed for sub_device %d", i);
 386                         goto free_rxq;
 387                 }
 388         }
 389         return 0;
 390 free_rxq:
 391         fs_rx_queue_release(rxq);
 392         return ret;
 393 }
 394
 395 static int
 396 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
 397 {
 398         struct rxq *rxq;
 399         struct sub_device *sdev;
 400         uint8_t i;
 401         int ret;
 402         int rc = 0;
 403
 404         if (idx >= dev->data->nb_rx_queues) {
 405                 rte_errno = EINVAL;
 406                 return -rte_errno;
 407         }
 408         rxq = dev->data->rx_queues[idx];
 409         if (rxq == NULL || rxq->event_fd <= 0) {
 410                 rte_errno = EINVAL;
 411                 return -rte_errno;
 412         }
 413         /* Fail if proxy service is nor running. */
 414         if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
 415                 ERROR("failsafe interrupt services are not running");
 416                 rte_errno = EAGAIN;
 417                 return -rte_errno;
 418         }
 419         rxq->enable_events = 1;
 420         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 421                 ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
 422                 ret = fs_err(sdev, ret);
 423                 if (ret)
 424                         rc = ret;
 425         }
 426         if (rc)
 427                 rte_errno = -rc;
 428         return rc;
 429 }
 430
 431 static int
 432 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
 433 {
 434         struct rxq *rxq;
 435         struct sub_device *sdev;
 436         uint64_t u64;
 437         uint8_t i;
 438         int rc = 0;
 439         int ret;
 440
 441         if (idx >= dev->data->nb_rx_queues) {
 442                 rte_errno = EINVAL;
 443                 return -rte_errno;
 444         }
 445         rxq = dev->data->rx_queues[idx];
 446         if (rxq == NULL || rxq->event_fd <= 0) {
 447                 rte_errno = EINVAL;
 448                 return -rte_errno;
 449         }
 450         rxq->enable_events = 0;
 451         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 452                 ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
 453                 ret = fs_err(sdev, ret);
 454                 if (ret)
 455                         rc = ret;
 456         }
 457         /* Clear pending events */
 458         while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
 459                 ;
 460         if (rc)
 461                 rte_errno = -rc;
 462         return rc;
 463 }
 464
 465 static bool
 466 fs_txq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 467 {
 468         uint64_t port_offloads;
 469         uint64_t queue_supp_offloads;
 470         uint64_t port_supp_offloads;
 471
 472         port_offloads = dev->data->dev_conf.txmode.offloads;
 473         queue_supp_offloads = PRIV(dev)->infos.tx_queue_offload_capa;
 474         port_supp_offloads = PRIV(dev)->infos.tx_offload_capa;
 475         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 476              offloads)
 477                 return false;
 478         /* Verify we have no conflict with port offloads */
 479         if ((port_offloads ^ offloads) & port_supp_offloads)
 480                 return false;
 481         return true;
 482 }
 483
 484 static void
 485 fs_tx_queue_release(void *queue)
 486 {
 487         struct rte_eth_dev *dev;
 488         struct sub_device *sdev;
 489         uint8_t i;
 490         struct txq *txq;
 491
 492         if (queue == NULL)
 493                 return;
 494         txq = queue;
 495         dev = txq->priv->dev;
 496         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 497                 SUBOPS(sdev, tx_queue_release)
 498                         (ETH(sdev)->data->tx_queues[txq->qid]);
 499         dev->data->tx_queues[txq->qid] = NULL;
 500         rte_free(txq);
 501 }
 502
 503 static int
 504 fs_tx_queue_setup(struct rte_eth_dev *dev,
 505                 uint16_t tx_queue_id,
 506                 uint16_t nb_tx_desc,
 507                 unsigned int socket_id,
 508                 const struct rte_eth_txconf *tx_conf)
 509 {
 510         struct sub_device *sdev;
 511         struct txq *txq;
 512         uint8_t i;
 513         int ret;
 514
 515         txq = dev->data->tx_queues[tx_queue_id];
 516         if (txq != NULL) {
 517                 fs_tx_queue_release(txq);
 518                 dev->data->tx_queues[tx_queue_id] = NULL;
 519         }
 520         /*
 521          * Don't verify queue offloads for applications which
 522          * use the old API.
 523          */
 524         if (tx_conf != NULL &&
 525             (tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
 526             fs_txq_offloads_valid(dev, tx_conf->offloads) == false) {
 527                 rte_errno = ENOTSUP;
 528                 ERROR("Tx queue offloads 0x%" PRIx64
 529                       " don't match port offloads 0x%" PRIx64
 530                       " or supported offloads 0x%" PRIx64,
 531                       tx_conf->offloads,
 532                       dev->data->dev_conf.txmode.offloads,
 533                       PRIV(dev)->infos.tx_offload_capa |
 534                       PRIV(dev)->infos.tx_queue_offload_capa);
 535                 return -rte_errno;
 536         }
 537         txq = rte_zmalloc("ethdev TX queue",
 538                           sizeof(*txq) +
 539                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 540                           RTE_CACHE_LINE_SIZE);
 541         if (txq == NULL)
 542                 return -ENOMEM;
 543         FOREACH_SUBDEV(sdev, i, dev)
 544                 rte_atomic64_init(&txq->refcnt[i]);
 545         txq->qid = tx_queue_id;
 546         txq->socket_id = socket_id;
 547         txq->info.conf = *tx_conf;
 548         txq->info.nb_desc = nb_tx_desc;
 549         txq->priv = PRIV(dev);
 550         dev->data->tx_queues[tx_queue_id] = txq;
 551         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 552                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 553                                 tx_queue_id,
 554                                 nb_tx_desc, socket_id,
 555                                 tx_conf);
 556                 if ((ret = fs_err(sdev, ret))) {
 557                         ERROR("TX queue setup failed for sub_device %d", i);
 558                         goto free_txq;
 559                 }
 560         }
 561         return 0;
 562 free_txq:
 563         fs_tx_queue_release(txq);
 564         return ret;
 565 }
 566
 567 static void
 568 fs_dev_free_queues(struct rte_eth_dev *dev)
 569 {
 570         uint16_t i;
 571
 572         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 573                 fs_rx_queue_release(dev->data->rx_queues[i]);
 574                 dev->data->rx_queues[i] = NULL;
 575         }
 576         dev->data->nb_rx_queues = 0;
 577         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 578                 fs_tx_queue_release(dev->data->tx_queues[i]);
 579                 dev->data->tx_queues[i] = NULL;
 580         }
 581         dev->data->nb_tx_queues = 0;
 582 }
 583
 584 static void
 585 fs_promiscuous_enable(struct rte_eth_dev *dev)
 586 {
 587         struct sub_device *sdev;
 588         uint8_t i;
 589
 590         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 591                 rte_eth_promiscuous_enable(PORT_ID(sdev));
 592 }
 593
 594 static void
 595 fs_promiscuous_disable(struct rte_eth_dev *dev)
 596 {
 597         struct sub_device *sdev;
 598         uint8_t i;
 599
 600         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 601                 rte_eth_promiscuous_disable(PORT_ID(sdev));
 602 }
 603
 604 static void
 605 fs_allmulticast_enable(struct rte_eth_dev *dev)
 606 {
 607         struct sub_device *sdev;
 608         uint8_t i;
 609
 610         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 611                 rte_eth_allmulticast_enable(PORT_ID(sdev));
 612 }
 613
 614 static void
 615 fs_allmulticast_disable(struct rte_eth_dev *dev)
 616 {
 617         struct sub_device *sdev;
 618         uint8_t i;
 619
 620         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 621                 rte_eth_allmulticast_disable(PORT_ID(sdev));
 622 }
 623
 624 static int
 625 fs_link_update(struct rte_eth_dev *dev,
 626                 int wait_to_complete)
 627 {
 628         struct sub_device *sdev;
 629         uint8_t i;
 630         int ret;
 631
 632         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 633                 DEBUG("Calling link_update on sub_device %d", i);
 634                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 635                 if (ret && ret != -1 && sdev->remove == 0 &&
 636                     rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
 637                         ERROR("Link update failed for sub_device %d with error %d",
 638                               i, ret);
 639                         return ret;
 640                 }
 641         }
 642         if (TX_SUBDEV(dev)) {
 643                 struct rte_eth_link *l1;
 644                 struct rte_eth_link *l2;
 645
 646                 l1 = &dev->data->dev_link;
 647                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 648                 if (memcmp(l1, l2, sizeof(*l1))) {
 649                         *l1 = *l2;
 650                         return 0;
 651                 }
 652         }
 653         return -1;
 654 }
 655
 656 static int
 657 fs_stats_get(struct rte_eth_dev *dev,
 658              struct rte_eth_stats *stats)
 659 {
 660         struct rte_eth_stats backup;
 661         struct sub_device *sdev;
 662         uint8_t i;
 663         int ret;
 664
 665         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 666         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 667                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 668                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 669
 670                 rte_memcpy(&backup, snapshot, sizeof(backup));
 671                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 672                 if (ret) {
 673                         if (!fs_err(sdev, ret)) {
 674                                 rte_memcpy(snapshot, &backup, sizeof(backup));
 675                                 goto inc;
 676                         }
 677                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 678                                   i, ret);
 679                         *timestamp = 0;
 680                         return ret;
 681                 }
 682                 *timestamp = rte_rdtsc();
 683 inc:
 684                 failsafe_stats_increment(stats, snapshot);
 685         }
 686         return 0;
 687 }
 688
 689 static void
 690 fs_stats_reset(struct rte_eth_dev *dev)
 691 {
 692         struct sub_device *sdev;
 693         uint8_t i;
 694
 695         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 696                 rte_eth_stats_reset(PORT_ID(sdev));
 697                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 698         }
 699         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 700 }
 701
 702 /**
 703  * Fail-safe dev_infos_get rules:
 704  *
 705  * No sub_device:
 706  *   Numerables:
 707  *      Use the maximum possible values for any field, so as not
 708  *      to impede any further configuration effort.
 709  *   Capabilities:
 710  *      Limits capabilities to those that are understood by the
 711  *      fail-safe PMD. This understanding stems from the fail-safe
 712  *      being capable of verifying that the related capability is
 713  *      expressed within the device configuration (struct rte_eth_conf).
 714  *
 715  * At least one probed sub_device:
 716  *   Numerables:
 717  *      Uses values from the active probed sub_device
 718  *      The rationale here is that if any sub_device is less capable
 719  *      (for example concerning the number of queues) than the active
 720  *      sub_device, then its subsequent configuration will fail.
 721  *      It is impossible to foresee this failure when the failing sub_device
 722  *      is supposed to be plugged-in later on, so the configuration process
 723  *      is the single point of failure and error reporting.
 724  *   Capabilities:
 725  *      Uses a logical AND of RX capabilities among
 726  *      all sub_devices and the default capabilities.
 727  *      Uses a logical AND of TX capabilities among
 728  *      the active probed sub_device and the default capabilities.
 729  *
 730  */
 731 static void
 732 fs_dev_infos_get(struct rte_eth_dev *dev,
 733                   struct rte_eth_dev_info *infos)
 734 {
 735         struct sub_device *sdev;
 736         uint8_t i;
 737
 738         sdev = TX_SUBDEV(dev);
 739         if (sdev == NULL) {
 740                 DEBUG("No probed device, using default infos");
 741                 rte_memcpy(&PRIV(dev)->infos, &default_infos,
 742                            sizeof(default_infos));
 743         } else {
 744                 uint64_t rx_offload_capa;
 745                 uint64_t rxq_offload_capa;
 746
 747                 rx_offload_capa = default_infos.rx_offload_capa;
 748                 rxq_offload_capa = default_infos.rx_queue_offload_capa;
 749                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 750                         rte_eth_dev_info_get(PORT_ID(sdev),
 751                                         &PRIV(dev)->infos);
 752                         rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
 753                         rxq_offload_capa &=
 754                                         PRIV(dev)->infos.rx_queue_offload_capa;
 755                 }
 756                 sdev = TX_SUBDEV(dev);
 757                 rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
 758                 PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
 759                 PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
 760                 PRIV(dev)->infos.tx_offload_capa &=
 761                                         default_infos.tx_offload_capa;
 762                 PRIV(dev)->infos.tx_queue_offload_capa &=
 763                                         default_infos.tx_queue_offload_capa;
 764                 PRIV(dev)->infos.flow_type_rss_offloads &=
 765                                         default_infos.flow_type_rss_offloads;
 766         }
 767         rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
 768 }
 769
 770 static const uint32_t *
 771 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 772 {
 773         struct sub_device *sdev;
 774         struct rte_eth_dev *edev;
 775
 776         sdev = TX_SUBDEV(dev);
 777         if (sdev == NULL)
 778                 return NULL;
 779         edev = ETH(sdev);
 780         /* ENOTSUP: counts as no supported ptypes */
 781         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL)
 782                 return NULL;
 783         /*
 784          * The API does not permit to do a clean AND of all ptypes,
 785          * It is also incomplete by design and we do not really care
 786          * to have a best possible value in this context.
 787          * We just return the ptypes of the device of highest
 788          * priority, usually the PREFERRED device.
 789          */
 790         return SUBOPS(sdev, dev_supported_ptypes_get)(edev);
 791 }
 792
 793 static int
 794 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 795 {
 796         struct sub_device *sdev;
 797         uint8_t i;
 798         int ret;
 799
 800         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 801                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
 802                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
 803                 if ((ret = fs_err(sdev, ret))) {
 804                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
 805                               i, ret);
 806                         return ret;
 807                 }
 808         }
 809         return 0;
 810 }
 811
 812 static int
 813 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 814 {
 815         struct sub_device *sdev;
 816         uint8_t i;
 817         int ret;
 818
 819         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 820                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
 821                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
 822                 if ((ret = fs_err(sdev, ret))) {
 823                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
 824                               " with error %d", i, ret);
 825                         return ret;
 826                 }
 827         }
 828         return 0;
 829 }
 830
 831 static int
 832 fs_flow_ctrl_get(struct rte_eth_dev *dev,
 833                 struct rte_eth_fc_conf *fc_conf)
 834 {
 835         struct sub_device *sdev;
 836
 837         sdev = TX_SUBDEV(dev);
 838         if (sdev == NULL)
 839                 return 0;
 840         if (SUBOPS(sdev, flow_ctrl_get) == NULL)
 841                 return -ENOTSUP;
 842         return SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
 843 }
 844
 845 static int
 846 fs_flow_ctrl_set(struct rte_eth_dev *dev,
 847                 struct rte_eth_fc_conf *fc_conf)
 848 {
 849         struct sub_device *sdev;
 850         uint8_t i;
 851         int ret;
 852
 853         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 854                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
 855                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
 856                 if ((ret = fs_err(sdev, ret))) {
 857                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
 858                               " with error %d", i, ret);
 859                         return ret;
 860                 }
 861         }
 862         return 0;
 863 }
 864
 865 static void
 866 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 867 {
 868         struct sub_device *sdev;
 869         uint8_t i;
 870
 871         /* No check: already done within the rte_eth_dev_mac_addr_remove
 872          * call for the fail-safe device.
 873          */
 874         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 875                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
 876                                 &dev->data->mac_addrs[index]);
 877         PRIV(dev)->mac_addr_pool[index] = 0;
 878 }
 879
 880 static int
 881 fs_mac_addr_add(struct rte_eth_dev *dev,
 882                 struct ether_addr *mac_addr,
 883                 uint32_t index,
 884                 uint32_t vmdq)
 885 {
 886         struct sub_device *sdev;
 887         int ret;
 888         uint8_t i;
 889
 890         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
 891         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 892                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
 893                 if ((ret = fs_err(sdev, ret))) {
 894                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
 895                               PRIu8 " with error %d", i, ret);
 896                         return ret;
 897                 }
 898         }
 899         if (index >= PRIV(dev)->nb_mac_addr) {
 900                 DEBUG("Growing mac_addrs array");
 901                 PRIV(dev)->nb_mac_addr = index;
 902         }
 903         PRIV(dev)->mac_addr_pool[index] = vmdq;
 904         return 0;
 905 }
 906
 907 static void
 908 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 909 {
 910         struct sub_device *sdev;
 911         uint8_t i;
 912
 913         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 914                 rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
 915 }
 916
 917 static int
 918 fs_filter_ctrl(struct rte_eth_dev *dev,
 919                 enum rte_filter_type type,
 920                 enum rte_filter_op op,
 921                 void *arg)
 922 {
 923         struct sub_device *sdev;
 924         uint8_t i;
 925         int ret;
 926
 927         if (type == RTE_ETH_FILTER_GENERIC &&
 928             op == RTE_ETH_FILTER_GET) {
 929                 *(const void **)arg = &fs_flow_ops;
 930                 return 0;
 931         }
 932         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 933                 DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
 934                 ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
 935                 if ((ret = fs_err(sdev, ret))) {
 936                         ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
 937                               " with error %d", i, ret);
 938                         return ret;
 939                 }
 940         }
 941         return 0;
 942 }
 943
 944 const struct eth_dev_ops failsafe_ops = {
 945         .dev_configure = fs_dev_configure,
 946         .dev_start = fs_dev_start,
 947         .dev_stop = fs_dev_stop,
 948         .dev_set_link_down = fs_dev_set_link_down,
 949         .dev_set_link_up = fs_dev_set_link_up,
 950         .dev_close = fs_dev_close,
 951         .promiscuous_enable = fs_promiscuous_enable,
 952         .promiscuous_disable = fs_promiscuous_disable,
 953         .allmulticast_enable = fs_allmulticast_enable,
 954         .allmulticast_disable = fs_allmulticast_disable,
 955         .link_update = fs_link_update,
 956         .stats_get = fs_stats_get,
 957         .stats_reset = fs_stats_reset,
 958         .dev_infos_get = fs_dev_infos_get,
 959         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
 960         .mtu_set = fs_mtu_set,
 961         .vlan_filter_set = fs_vlan_filter_set,
 962         .rx_queue_setup = fs_rx_queue_setup,
 963         .tx_queue_setup = fs_tx_queue_setup,
 964         .rx_queue_release = fs_rx_queue_release,
 965         .tx_queue_release = fs_tx_queue_release,
 966         .rx_queue_intr_enable = fs_rx_intr_enable,
 967         .rx_queue_intr_disable = fs_rx_intr_disable,
 968         .flow_ctrl_get = fs_flow_ctrl_get,
 969         .flow_ctrl_set = fs_flow_ctrl_set,
 970         .mac_addr_remove = fs_mac_addr_remove,
 971         .mac_addr_add = fs_mac_addr_add,
 972         .mac_addr_set = fs_mac_addr_set,
 973         .filter_ctrl = fs_filter_ctrl,
 974 };