drivers/net/failsafe/failsafe_ops.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2017 6WIND S.A.
   3  * Copyright 2017 Mellanox Technologies, Ltd
   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 #include <rte_ethdev.h>
  17 #include <rte_string_fns.h>
  18
  19 #include "failsafe_private.h"
  20
  21 static int
  22 fs_dev_configure(struct rte_eth_dev *dev)
  23 {
  24         struct sub_device *sdev;
  25         uint8_t i;
  26         int ret;
  27
  28         fs_lock(dev, 0);
  29         FOREACH_SUBDEV(sdev, i, dev) {
  30                 int rmv_interrupt = 0;
  31                 int lsc_interrupt = 0;
  32                 int lsc_enabled;
  33
  34                 if (sdev->state != DEV_PROBED &&
  35                     !(PRIV(dev)->alarm_lock == 0 && sdev->state == DEV_ACTIVE))
  36                         continue;
  37
  38                 rmv_interrupt = ETH(sdev)->data->dev_flags &
  39                                 RTE_ETH_DEV_INTR_RMV;
  40                 if (rmv_interrupt) {
  41                         DEBUG("Enabling RMV interrupts for sub_device %d", i);
  42                         dev->data->dev_conf.intr_conf.rmv = 1;
  43                 } else {
  44                         DEBUG("sub_device %d does not support RMV event", i);
  45                 }
  46                 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
  47                 lsc_interrupt = lsc_enabled &&
  48                                 (ETH(sdev)->data->dev_flags &
  49                                  RTE_ETH_DEV_INTR_LSC);
  50                 if (lsc_interrupt) {
  51                         DEBUG("Enabling LSC interrupts for sub_device %d", i);
  52                         dev->data->dev_conf.intr_conf.lsc = 1;
  53                 } else if (lsc_enabled && !lsc_interrupt) {
  54                         DEBUG("Disabling LSC interrupts for sub_device %d", i);
  55                         dev->data->dev_conf.intr_conf.lsc = 0;
  56                 }
  57                 DEBUG("Configuring sub-device %d", i);
  58                 ret = rte_eth_dev_configure(PORT_ID(sdev),
  59                                         dev->data->nb_rx_queues,
  60                                         dev->data->nb_tx_queues,
  61                                         &dev->data->dev_conf);
  62                 if (ret) {
  63                         if (!fs_err(sdev, ret))
  64                                 continue;
  65                         ERROR("Could not configure sub_device %d", i);
  66                         fs_unlock(dev, 0);
  67                         return ret;
  68                 }
  69                 if (rmv_interrupt && sdev->rmv_callback == 0) {
  70                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
  71                                         RTE_ETH_EVENT_INTR_RMV,
  72                                         failsafe_eth_rmv_event_callback,
  73                                         sdev);
  74                         if (ret)
  75                                 WARN("Failed to register RMV callback for sub_device %d",
  76                                      SUB_ID(sdev));
  77                         else
  78                                 sdev->rmv_callback = 1;
  79                 }
  80                 dev->data->dev_conf.intr_conf.rmv = 0;
  81                 if (lsc_interrupt && sdev->lsc_callback == 0) {
  82                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
  83                                                 RTE_ETH_EVENT_INTR_LSC,
  84                                                 failsafe_eth_lsc_event_callback,
  85                                                 dev);
  86                         if (ret)
  87                                 WARN("Failed to register LSC callback for sub_device %d",
  88                                      SUB_ID(sdev));
  89                         else
  90                                 sdev->lsc_callback = 1;
  91                 }
  92                 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
  93                 sdev->state = DEV_ACTIVE;
  94         }
  95         if (PRIV(dev)->state < DEV_ACTIVE)
  96                 PRIV(dev)->state = DEV_ACTIVE;
  97         fs_unlock(dev, 0);
  98         return 0;
  99 }
 100
 101 static void
 102 fs_set_queues_state_start(struct rte_eth_dev *dev)
 103 {
 104         struct rxq *rxq;
 105         struct txq *txq;
 106         uint16_t i;
 107
 108         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 109                 rxq = dev->data->rx_queues[i];
 110                 if (rxq != NULL && !rxq->info.conf.rx_deferred_start)
 111                         dev->data->rx_queue_state[i] =
 112                                                 RTE_ETH_QUEUE_STATE_STARTED;
 113         }
 114         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 115                 txq = dev->data->tx_queues[i];
 116                 if (txq != NULL && !txq->info.conf.tx_deferred_start)
 117                         dev->data->tx_queue_state[i] =
 118                                                 RTE_ETH_QUEUE_STATE_STARTED;
 119         }
 120 }
 121
 122 static int
 123 fs_dev_start(struct rte_eth_dev *dev)
 124 {
 125         struct sub_device *sdev;
 126         uint8_t i;
 127         int ret;
 128
 129         fs_lock(dev, 0);
 130         ret = failsafe_rx_intr_install(dev);
 131         if (ret) {
 132                 fs_unlock(dev, 0);
 133                 return ret;
 134         }
 135         FOREACH_SUBDEV(sdev, i, dev) {
 136                 if (sdev->state != DEV_ACTIVE)
 137                         continue;
 138                 DEBUG("Starting sub_device %d", i);
 139                 ret = rte_eth_dev_start(PORT_ID(sdev));
 140                 if (ret) {
 141                         if (!fs_err(sdev, ret))
 142                                 continue;
 143                         fs_unlock(dev, 0);
 144                         return ret;
 145                 }
 146                 ret = failsafe_rx_intr_install_subdevice(sdev);
 147                 if (ret) {
 148                         if (!fs_err(sdev, ret))
 149                                 continue;
 150                         rte_eth_dev_stop(PORT_ID(sdev));
 151                         fs_unlock(dev, 0);
 152                         return ret;
 153                 }
 154                 sdev->state = DEV_STARTED;
 155         }
 156         if (PRIV(dev)->state < DEV_STARTED) {
 157                 PRIV(dev)->state = DEV_STARTED;
 158                 fs_set_queues_state_start(dev);
 159         }
 160         fs_switch_dev(dev, NULL);
 161         fs_unlock(dev, 0);
 162         return 0;
 163 }
 164
 165 static void
 166 fs_set_queues_state_stop(struct rte_eth_dev *dev)
 167 {
 168         uint16_t i;
 169
 170         for (i = 0; i < dev->data->nb_rx_queues; i++)
 171                 if (dev->data->rx_queues[i] != NULL)
 172                         dev->data->rx_queue_state[i] =
 173                                                 RTE_ETH_QUEUE_STATE_STOPPED;
 174         for (i = 0; i < dev->data->nb_tx_queues; i++)
 175                 if (dev->data->tx_queues[i] != NULL)
 176                         dev->data->tx_queue_state[i] =
 177                                                 RTE_ETH_QUEUE_STATE_STOPPED;
 178 }
 179
 180 static void
 181 fs_dev_stop(struct rte_eth_dev *dev)
 182 {
 183         struct sub_device *sdev;
 184         uint8_t i;
 185
 186         fs_lock(dev, 0);
 187         PRIV(dev)->state = DEV_STARTED - 1;
 188         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 189                 rte_eth_dev_stop(PORT_ID(sdev));
 190                 failsafe_rx_intr_uninstall_subdevice(sdev);
 191                 sdev->state = DEV_STARTED - 1;
 192         }
 193         failsafe_rx_intr_uninstall(dev);
 194         fs_set_queues_state_stop(dev);
 195         fs_unlock(dev, 0);
 196 }
 197
 198 static int
 199 fs_dev_set_link_up(struct rte_eth_dev *dev)
 200 {
 201         struct sub_device *sdev;
 202         uint8_t i;
 203         int ret;
 204
 205         fs_lock(dev, 0);
 206         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 207                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 208                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 209                 if ((ret = fs_err(sdev, ret))) {
 210                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 211                               " with error %d", i, ret);
 212                         fs_unlock(dev, 0);
 213                         return ret;
 214                 }
 215         }
 216         fs_unlock(dev, 0);
 217         return 0;
 218 }
 219
 220 static int
 221 fs_dev_set_link_down(struct rte_eth_dev *dev)
 222 {
 223         struct sub_device *sdev;
 224         uint8_t i;
 225         int ret;
 226
 227         fs_lock(dev, 0);
 228         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 229                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 230                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 231                 if ((ret = fs_err(sdev, ret))) {
 232                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 233                               " with error %d", i, ret);
 234                         fs_unlock(dev, 0);
 235                         return ret;
 236                 }
 237         }
 238         fs_unlock(dev, 0);
 239         return 0;
 240 }
 241
 242 static int
 243 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 244 {
 245         struct sub_device *sdev;
 246         uint8_t i;
 247         int ret;
 248         int err = 0;
 249         bool failure = true;
 250
 251         fs_lock(dev, 0);
 252         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 253                 uint16_t port_id = ETH(sdev)->data->port_id;
 254
 255                 ret = rte_eth_dev_rx_queue_stop(port_id, rx_queue_id);
 256                 ret = fs_err(sdev, ret);
 257                 if (ret) {
 258                         ERROR("Rx queue stop failed for subdevice %d", i);
 259                         err = ret;
 260                 } else {
 261                         failure = false;
 262                 }
 263         }
 264         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 265         fs_unlock(dev, 0);
 266         /* Return 0 in case of at least one successful queue stop */
 267         return (failure) ? err : 0;
 268 }
 269
 270 static int
 271 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 272 {
 273         struct sub_device *sdev;
 274         uint8_t i;
 275         int ret;
 276
 277         fs_lock(dev, 0);
 278         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 279                 uint16_t port_id = ETH(sdev)->data->port_id;
 280
 281                 ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
 282                 ret = fs_err(sdev, ret);
 283                 if (ret) {
 284                         ERROR("Rx queue start failed for subdevice %d", i);
 285                         fs_rx_queue_stop(dev, rx_queue_id);
 286                         fs_unlock(dev, 0);
 287                         return ret;
 288                 }
 289         }
 290         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 291         fs_unlock(dev, 0);
 292         return 0;
 293 }
 294
 295 static int
 296 fs_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 297 {
 298         struct sub_device *sdev;
 299         uint8_t i;
 300         int ret;
 301         int err = 0;
 302         bool failure = true;
 303
 304         fs_lock(dev, 0);
 305         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 306                 uint16_t port_id = ETH(sdev)->data->port_id;
 307
 308                 ret = rte_eth_dev_tx_queue_stop(port_id, tx_queue_id);
 309                 ret = fs_err(sdev, ret);
 310                 if (ret) {
 311                         ERROR("Tx queue stop failed for subdevice %d", i);
 312                         err = ret;
 313                 } else {
 314                         failure = false;
 315                 }
 316         }
 317         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 318         fs_unlock(dev, 0);
 319         /* Return 0 in case of at least one successful queue stop */
 320         return (failure) ? err : 0;
 321 }
 322
 323 static int
 324 fs_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 325 {
 326         struct sub_device *sdev;
 327         uint8_t i;
 328         int ret;
 329
 330         fs_lock(dev, 0);
 331         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 332                 uint16_t port_id = ETH(sdev)->data->port_id;
 333
 334                 ret = rte_eth_dev_tx_queue_start(port_id, tx_queue_id);
 335                 ret = fs_err(sdev, ret);
 336                 if (ret) {
 337                         ERROR("Tx queue start failed for subdevice %d", i);
 338                         fs_tx_queue_stop(dev, tx_queue_id);
 339                         fs_unlock(dev, 0);
 340                         return ret;
 341                 }
 342         }
 343         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 344         fs_unlock(dev, 0);
 345         return 0;
 346 }
 347
 348 static void
 349 fs_rx_queue_release(void *queue)
 350 {
 351         struct rte_eth_dev *dev;
 352         struct sub_device *sdev;
 353         uint8_t i;
 354         struct rxq *rxq;
 355
 356         if (queue == NULL)
 357                 return;
 358         rxq = queue;
 359         dev = &rte_eth_devices[rxq->priv->data->port_id];
 360         fs_lock(dev, 0);
 361         if (rxq->event_fd >= 0)
 362                 close(rxq->event_fd);
 363         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 364                 if (ETH(sdev)->data->rx_queues != NULL &&
 365                     ETH(sdev)->data->rx_queues[rxq->qid] != NULL) {
 366                         SUBOPS(sdev, rx_queue_release)
 367                                 (ETH(sdev)->data->rx_queues[rxq->qid]);
 368                 }
 369         }
 370         dev->data->rx_queues[rxq->qid] = NULL;
 371         rte_free(rxq);
 372         fs_unlock(dev, 0);
 373 }
 374
 375 static int
 376 fs_rx_queue_setup(struct rte_eth_dev *dev,
 377                 uint16_t rx_queue_id,
 378                 uint16_t nb_rx_desc,
 379                 unsigned int socket_id,
 380                 const struct rte_eth_rxconf *rx_conf,
 381                 struct rte_mempool *mb_pool)
 382 {
 383         /*
 384          * FIXME: Add a proper interface in rte_eal_interrupts for
 385          * allocating eventfd as an interrupt vector.
 386          * For the time being, fake as if we are using MSIX interrupts,
 387          * this will cause rte_intr_efd_enable to allocate an eventfd for us.
 388          */
 389         struct rte_intr_handle intr_handle = {
 390                 .type = RTE_INTR_HANDLE_VFIO_MSIX,
 391                 .efds = { -1, },
 392         };
 393         struct sub_device *sdev;
 394         struct rxq *rxq;
 395         uint8_t i;
 396         int ret;
 397
 398         fs_lock(dev, 0);
 399         if (rx_conf->rx_deferred_start) {
 400                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 401                         if (SUBOPS(sdev, rx_queue_start) == NULL) {
 402                                 ERROR("Rx queue deferred start is not "
 403                                         "supported for subdevice %d", i);
 404                                 fs_unlock(dev, 0);
 405                                 return -EINVAL;
 406                         }
 407                 }
 408         }
 409         rxq = dev->data->rx_queues[rx_queue_id];
 410         if (rxq != NULL) {
 411                 fs_rx_queue_release(rxq);
 412                 dev->data->rx_queues[rx_queue_id] = NULL;
 413         }
 414         rxq = rte_zmalloc(NULL,
 415                           sizeof(*rxq) +
 416                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 417                           RTE_CACHE_LINE_SIZE);
 418         if (rxq == NULL) {
 419                 fs_unlock(dev, 0);
 420                 return -ENOMEM;
 421         }
 422         FOREACH_SUBDEV(sdev, i, dev)
 423                 rte_atomic64_init(&rxq->refcnt[i]);
 424         rxq->qid = rx_queue_id;
 425         rxq->socket_id = socket_id;
 426         rxq->info.mp = mb_pool;
 427         rxq->info.conf = *rx_conf;
 428         rxq->info.nb_desc = nb_rx_desc;
 429         rxq->priv = PRIV(dev);
 430         rxq->sdev = PRIV(dev)->subs;
 431         ret = rte_intr_efd_enable(&intr_handle, 1);
 432         if (ret < 0) {
 433                 fs_unlock(dev, 0);
 434                 return ret;
 435         }
 436         rxq->event_fd = intr_handle.efds[0];
 437         dev->data->rx_queues[rx_queue_id] = rxq;
 438         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 439                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 440                                 rx_queue_id,
 441                                 nb_rx_desc, socket_id,
 442                                 rx_conf, mb_pool);
 443                 if ((ret = fs_err(sdev, ret))) {
 444                         ERROR("RX queue setup failed for sub_device %d", i);
 445                         goto free_rxq;
 446                 }
 447         }
 448         fs_unlock(dev, 0);
 449         return 0;
 450 free_rxq:
 451         fs_rx_queue_release(rxq);
 452         fs_unlock(dev, 0);
 453         return ret;
 454 }
 455
 456 static int
 457 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
 458 {
 459         struct rxq *rxq;
 460         struct sub_device *sdev;
 461         uint8_t i;
 462         int ret;
 463         int rc = 0;
 464
 465         fs_lock(dev, 0);
 466         if (idx >= dev->data->nb_rx_queues) {
 467                 rc = -EINVAL;
 468                 goto unlock;
 469         }
 470         rxq = dev->data->rx_queues[idx];
 471         if (rxq == NULL || rxq->event_fd <= 0) {
 472                 rc = -EINVAL;
 473                 goto unlock;
 474         }
 475         /* Fail if proxy service is nor running. */
 476         if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
 477                 ERROR("failsafe interrupt services are not running");
 478                 rc = -EAGAIN;
 479                 goto unlock;
 480         }
 481         rxq->enable_events = 1;
 482         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 483                 ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
 484                 ret = fs_err(sdev, ret);
 485                 if (ret)
 486                         rc = ret;
 487         }
 488 unlock:
 489         fs_unlock(dev, 0);
 490         if (rc)
 491                 rte_errno = -rc;
 492         return rc;
 493 }
 494
 495 static int
 496 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
 497 {
 498         struct rxq *rxq;
 499         struct sub_device *sdev;
 500         uint64_t u64;
 501         uint8_t i;
 502         int rc = 0;
 503         int ret;
 504
 505         fs_lock(dev, 0);
 506         if (idx >= dev->data->nb_rx_queues) {
 507                 rc = -EINVAL;
 508                 goto unlock;
 509         }
 510         rxq = dev->data->rx_queues[idx];
 511         if (rxq == NULL || rxq->event_fd <= 0) {
 512                 rc = -EINVAL;
 513                 goto unlock;
 514         }
 515         rxq->enable_events = 0;
 516         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 517                 ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
 518                 ret = fs_err(sdev, ret);
 519                 if (ret)
 520                         rc = ret;
 521         }
 522         /* Clear pending events */
 523         while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
 524                 ;
 525 unlock:
 526         fs_unlock(dev, 0);
 527         if (rc)
 528                 rte_errno = -rc;
 529         return rc;
 530 }
 531
 532 static void
 533 fs_tx_queue_release(void *queue)
 534 {
 535         struct rte_eth_dev *dev;
 536         struct sub_device *sdev;
 537         uint8_t i;
 538         struct txq *txq;
 539
 540         if (queue == NULL)
 541                 return;
 542         txq = queue;
 543         dev = &rte_eth_devices[txq->priv->data->port_id];
 544         fs_lock(dev, 0);
 545         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 546                 if (ETH(sdev)->data->tx_queues != NULL &&
 547                     ETH(sdev)->data->tx_queues[txq->qid] != NULL) {
 548                         SUBOPS(sdev, tx_queue_release)
 549                                 (ETH(sdev)->data->tx_queues[txq->qid]);
 550                 }
 551         }
 552         dev->data->tx_queues[txq->qid] = NULL;
 553         rte_free(txq);
 554         fs_unlock(dev, 0);
 555 }
 556
 557 static int
 558 fs_tx_queue_setup(struct rte_eth_dev *dev,
 559                 uint16_t tx_queue_id,
 560                 uint16_t nb_tx_desc,
 561                 unsigned int socket_id,
 562                 const struct rte_eth_txconf *tx_conf)
 563 {
 564         struct sub_device *sdev;
 565         struct txq *txq;
 566         uint8_t i;
 567         int ret;
 568
 569         fs_lock(dev, 0);
 570         if (tx_conf->tx_deferred_start) {
 571                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 572                         if (SUBOPS(sdev, tx_queue_start) == NULL) {
 573                                 ERROR("Tx queue deferred start is not "
 574                                         "supported for subdevice %d", i);
 575                                 fs_unlock(dev, 0);
 576                                 return -EINVAL;
 577                         }
 578                 }
 579         }
 580         txq = dev->data->tx_queues[tx_queue_id];
 581         if (txq != NULL) {
 582                 fs_tx_queue_release(txq);
 583                 dev->data->tx_queues[tx_queue_id] = NULL;
 584         }
 585         txq = rte_zmalloc("ethdev TX queue",
 586                           sizeof(*txq) +
 587                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 588                           RTE_CACHE_LINE_SIZE);
 589         if (txq == NULL) {
 590                 fs_unlock(dev, 0);
 591                 return -ENOMEM;
 592         }
 593         FOREACH_SUBDEV(sdev, i, dev)
 594                 rte_atomic64_init(&txq->refcnt[i]);
 595         txq->qid = tx_queue_id;
 596         txq->socket_id = socket_id;
 597         txq->info.conf = *tx_conf;
 598         txq->info.nb_desc = nb_tx_desc;
 599         txq->priv = PRIV(dev);
 600         dev->data->tx_queues[tx_queue_id] = txq;
 601         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 602                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 603                                 tx_queue_id,
 604                                 nb_tx_desc, socket_id,
 605                                 tx_conf);
 606                 if ((ret = fs_err(sdev, ret))) {
 607                         ERROR("TX queue setup failed for sub_device %d", i);
 608                         goto free_txq;
 609                 }
 610         }
 611         fs_unlock(dev, 0);
 612         return 0;
 613 free_txq:
 614         fs_tx_queue_release(txq);
 615         fs_unlock(dev, 0);
 616         return ret;
 617 }
 618
 619 static void
 620 fs_dev_free_queues(struct rte_eth_dev *dev)
 621 {
 622         uint16_t i;
 623
 624         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 625                 fs_rx_queue_release(dev->data->rx_queues[i]);
 626                 dev->data->rx_queues[i] = NULL;
 627         }
 628         dev->data->nb_rx_queues = 0;
 629         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 630                 fs_tx_queue_release(dev->data->tx_queues[i]);
 631                 dev->data->tx_queues[i] = NULL;
 632         }
 633         dev->data->nb_tx_queues = 0;
 634 }
 635
 636 int
 637 failsafe_eth_dev_close(struct rte_eth_dev *dev)
 638 {
 639         struct sub_device *sdev;
 640         uint8_t i;
 641         int ret;
 642
 643         fs_lock(dev, 0);
 644         failsafe_hotplug_alarm_cancel(dev);
 645         if (PRIV(dev)->state == DEV_STARTED)
 646                 dev->dev_ops->dev_stop(dev);
 647         PRIV(dev)->state = DEV_ACTIVE - 1;
 648         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 649                 DEBUG("Closing sub_device %d", i);
 650                 failsafe_eth_dev_unregister_callbacks(sdev);
 651                 rte_eth_dev_close(PORT_ID(sdev));
 652                 sdev->state = DEV_ACTIVE - 1;
 653         }
 654         rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
 655                                         failsafe_eth_new_event_callback, dev);
 656         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 657                 fs_unlock(dev, 0);
 658                 return 0;
 659         }
 660         fs_dev_free_queues(dev);
 661         ret = failsafe_eal_uninit(dev);
 662         if (ret)
 663                 ERROR("Error while uninitializing sub-EAL");
 664         failsafe_args_free(dev);
 665         rte_free(PRIV(dev)->subs);
 666         rte_free(PRIV(dev)->mcast_addrs);
 667         /* mac_addrs must not be freed alone because part of dev_private */
 668         dev->data->mac_addrs = NULL;
 669         fs_unlock(dev, 0);
 670         ret = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
 671         if (ret)
 672                 ERROR("Error while destroying hotplug mutex");
 673         return 0;
 674 }
 675
 676 static int
 677 fs_promiscuous_enable(struct rte_eth_dev *dev)
 678 {
 679         struct sub_device *sdev;
 680         uint8_t i;
 681         int ret = 0;
 682
 683         fs_lock(dev, 0);
 684         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 685                 ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
 686                 ret = fs_err(sdev, ret);
 687                 if (ret != 0) {
 688                         ERROR("Promiscuous mode enable failed for subdevice %d",
 689                                 PORT_ID(sdev));
 690                         break;
 691                 }
 692         }
 693         if (ret != 0) {
 694                 /* Rollback in the case of failure */
 695                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 696                         ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
 697                         ret = fs_err(sdev, ret);
 698                         if (ret != 0)
 699                                 ERROR("Promiscuous mode disable during rollback failed for subdevice %d",
 700                                         PORT_ID(sdev));
 701                 }
 702         }
 703         fs_unlock(dev, 0);
 704
 705         return ret;
 706 }
 707
 708 static int
 709 fs_promiscuous_disable(struct rte_eth_dev *dev)
 710 {
 711         struct sub_device *sdev;
 712         uint8_t i;
 713         int ret = 0;
 714
 715         fs_lock(dev, 0);
 716         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 717                 ret = rte_eth_promiscuous_disable(PORT_ID(sdev));
 718                 ret = fs_err(sdev, ret);
 719                 if (ret != 0) {
 720                         ERROR("Promiscuous mode disable failed for subdevice %d",
 721                                 PORT_ID(sdev));
 722                         break;
 723                 }
 724         }
 725         if (ret != 0) {
 726                 /* Rollback in the case of failure */
 727                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 728                         ret = rte_eth_promiscuous_enable(PORT_ID(sdev));
 729                         ret = fs_err(sdev, ret);
 730                         if (ret != 0)
 731                                 ERROR("Promiscuous mode enable during rollback failed for subdevice %d",
 732                                         PORT_ID(sdev));
 733                 }
 734         }
 735         fs_unlock(dev, 0);
 736
 737         return ret;
 738 }
 739
 740 static int
 741 fs_allmulticast_enable(struct rte_eth_dev *dev)
 742 {
 743         struct sub_device *sdev;
 744         uint8_t i;
 745         int ret = 0;
 746
 747         fs_lock(dev, 0);
 748         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 749                 ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
 750                 ret = fs_err(sdev, ret);
 751                 if (ret != 0) {
 752                         ERROR("All-multicast mode enable failed for subdevice %d",
 753                                 PORT_ID(sdev));
 754                         break;
 755                 }
 756         }
 757         if (ret != 0) {
 758                 /* Rollback in the case of failure */
 759                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 760                         ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
 761                         ret = fs_err(sdev, ret);
 762                         if (ret != 0)
 763                                 ERROR("All-multicast mode disable during rollback failed for subdevice %d",
 764                                         PORT_ID(sdev));
 765                 }
 766         }
 767         fs_unlock(dev, 0);
 768
 769         return ret;
 770 }
 771
 772 static int
 773 fs_allmulticast_disable(struct rte_eth_dev *dev)
 774 {
 775         struct sub_device *sdev;
 776         uint8_t i;
 777         int ret = 0;
 778
 779         fs_lock(dev, 0);
 780         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 781                 ret = rte_eth_allmulticast_disable(PORT_ID(sdev));
 782                 ret = fs_err(sdev, ret);
 783                 if (ret != 0) {
 784                         ERROR("All-multicast mode disable failed for subdevice %d",
 785                                 PORT_ID(sdev));
 786                         break;
 787                 }
 788         }
 789         if (ret != 0) {
 790                 /* Rollback in the case of failure */
 791                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 792                         ret = rte_eth_allmulticast_enable(PORT_ID(sdev));
 793                         ret = fs_err(sdev, ret);
 794                         if (ret != 0)
 795                                 ERROR("All-multicast mode enable during rollback failed for subdevice %d",
 796                                         PORT_ID(sdev));
 797                 }
 798         }
 799         fs_unlock(dev, 0);
 800
 801         return ret;
 802 }
 803
 804 static int
 805 fs_link_update(struct rte_eth_dev *dev,
 806                 int wait_to_complete)
 807 {
 808         struct sub_device *sdev;
 809         uint8_t i;
 810         int ret;
 811
 812         fs_lock(dev, 0);
 813         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 814                 DEBUG("Calling link_update on sub_device %d", i);
 815                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 816                 if (ret && ret != -1 && sdev->remove == 0 &&
 817                     rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
 818                         ERROR("Link update failed for sub_device %d with error %d",
 819                               i, ret);
 820                         fs_unlock(dev, 0);
 821                         return ret;
 822                 }
 823         }
 824         if (TX_SUBDEV(dev)) {
 825                 struct rte_eth_link *l1;
 826                 struct rte_eth_link *l2;
 827
 828                 l1 = &dev->data->dev_link;
 829                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 830                 if (memcmp(l1, l2, sizeof(*l1))) {
 831                         *l1 = *l2;
 832                         fs_unlock(dev, 0);
 833                         return 0;
 834                 }
 835         }
 836         fs_unlock(dev, 0);
 837         return -1;
 838 }
 839
 840 static int
 841 fs_stats_get(struct rte_eth_dev *dev,
 842              struct rte_eth_stats *stats)
 843 {
 844         struct rte_eth_stats backup;
 845         struct sub_device *sdev;
 846         uint8_t i;
 847         int ret;
 848
 849         fs_lock(dev, 0);
 850         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 851         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 852                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 853                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 854
 855                 rte_memcpy(&backup, snapshot, sizeof(backup));
 856                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 857                 if (ret) {
 858                         if (!fs_err(sdev, ret)) {
 859                                 rte_memcpy(snapshot, &backup, sizeof(backup));
 860                                 goto inc;
 861                         }
 862                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 863                                   i, ret);
 864                         *timestamp = 0;
 865                         fs_unlock(dev, 0);
 866                         return ret;
 867                 }
 868                 *timestamp = rte_rdtsc();
 869 inc:
 870                 failsafe_stats_increment(stats, snapshot);
 871         }
 872         fs_unlock(dev, 0);
 873         return 0;
 874 }
 875
 876 static int
 877 fs_stats_reset(struct rte_eth_dev *dev)
 878 {
 879         struct sub_device *sdev;
 880         uint8_t i;
 881         int ret;
 882
 883         fs_lock(dev, 0);
 884         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 885                 ret = rte_eth_stats_reset(PORT_ID(sdev));
 886                 if (ret) {
 887                         if (!fs_err(sdev, ret))
 888                                 continue;
 889
 890                         ERROR("Operation rte_eth_stats_reset failed for sub_device %d with error %d",
 891                               i, ret);
 892                         fs_unlock(dev, 0);
 893                         return ret;
 894                 }
 895                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 896         }
 897         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 898         fs_unlock(dev, 0);
 899
 900         return 0;
 901 }
 902
 903 static int
 904 __fs_xstats_count(struct rte_eth_dev *dev)
 905 {
 906         struct sub_device *sdev;
 907         int count = 0;
 908         uint8_t i;
 909         int ret;
 910
 911         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 912                 ret = rte_eth_xstats_get_names(PORT_ID(sdev), NULL, 0);
 913                 if (ret < 0)
 914                         return ret;
 915                 count += ret;
 916         }
 917
 918         return count;
 919 }
 920
 921 static int
 922 __fs_xstats_get_names(struct rte_eth_dev *dev,
 923                     struct rte_eth_xstat_name *xstats_names,
 924                     unsigned int limit)
 925 {
 926         struct sub_device *sdev;
 927         unsigned int count = 0;
 928         uint8_t i;
 929
 930         /* Caller only cares about count */
 931         if (!xstats_names)
 932                 return  __fs_xstats_count(dev);
 933
 934         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 935                 struct rte_eth_xstat_name *sub_names = xstats_names + count;
 936                 int j, r;
 937
 938                 if (count >= limit)
 939                         break;
 940
 941                 r = rte_eth_xstats_get_names(PORT_ID(sdev),
 942                                              sub_names, limit - count);
 943                 if (r < 0)
 944                         return r;
 945
 946                 /* add subN_ prefix to names */
 947                 for (j = 0; j < r; j++) {
 948                         char *xname = sub_names[j].name;
 949                         char tmp[RTE_ETH_XSTATS_NAME_SIZE];
 950
 951                         if ((xname[0] == 't' || xname[0] == 'r') &&
 952                             xname[1] == 'x' && xname[2] == '_')
 953                                 snprintf(tmp, sizeof(tmp), "%.3ssub%u_%s",
 954                                          xname, i, xname + 3);
 955                         else
 956                                 snprintf(tmp, sizeof(tmp), "sub%u_%s",
 957                                          i, xname);
 958
 959                         strlcpy(xname, tmp, RTE_ETH_XSTATS_NAME_SIZE);
 960                 }
 961                 count += r;
 962         }
 963         return count;
 964 }
 965
 966 static int
 967 fs_xstats_get_names(struct rte_eth_dev *dev,
 968                     struct rte_eth_xstat_name *xstats_names,
 969                     unsigned int limit)
 970 {
 971         int ret;
 972
 973         fs_lock(dev, 0);
 974         ret = __fs_xstats_get_names(dev, xstats_names, limit);
 975         fs_unlock(dev, 0);
 976         return ret;
 977 }
 978
 979 static int
 980 __fs_xstats_get(struct rte_eth_dev *dev,
 981               struct rte_eth_xstat *xstats,
 982               unsigned int n)
 983 {
 984         unsigned int count = 0;
 985         struct sub_device *sdev;
 986         uint8_t i;
 987         int j, ret;
 988
 989         ret = __fs_xstats_count(dev);
 990         /*
 991          * if error
 992          * or caller did not give enough space
 993          * or just querying
 994          */
 995         if (ret < 0 || ret > (int)n || xstats == NULL)
 996                 return ret;
 997
 998         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 999                 ret = rte_eth_xstats_get(PORT_ID(sdev), xstats, n);
1000                 if (ret < 0)
1001                         return ret;
1002
1003                 if (ret > (int)n)
1004                         return n + count;
1005
1006                 /* add offset to id's from sub-device */
1007                 for (j = 0; j < ret; j++)
1008                         xstats[j].id += count;
1009
1010                 xstats += ret;
1011                 n -= ret;
1012                 count += ret;
1013         }
1014
1015         return count;
1016 }
1017
1018 static int
1019 fs_xstats_get(struct rte_eth_dev *dev,
1020               struct rte_eth_xstat *xstats,
1021               unsigned int n)
1022 {
1023         int ret;
1024
1025         fs_lock(dev, 0);
1026         ret = __fs_xstats_get(dev, xstats, n);
1027         fs_unlock(dev, 0);
1028
1029         return ret;
1030 }
1031
1032
1033 static int
1034 fs_xstats_reset(struct rte_eth_dev *dev)
1035 {
1036         struct sub_device *sdev;
1037         uint8_t i;
1038         int r = 0;
1039
1040         fs_lock(dev, 0);
1041         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1042                 r = rte_eth_xstats_reset(PORT_ID(sdev));
1043                 if (r < 0)
1044                         break;
1045         }
1046         fs_unlock(dev, 0);
1047
1048         return r;
1049 }
1050
1051 static void
1052 fs_dev_merge_desc_lim(struct rte_eth_desc_lim *to,
1053                       const struct rte_eth_desc_lim *from)
1054 {
1055         to->nb_max = RTE_MIN(to->nb_max, from->nb_max);
1056         to->nb_min = RTE_MAX(to->nb_min, from->nb_min);
1057         to->nb_align = RTE_MAX(to->nb_align, from->nb_align);
1058
1059         to->nb_seg_max = RTE_MIN(to->nb_seg_max, from->nb_seg_max);
1060         to->nb_mtu_seg_max = RTE_MIN(to->nb_mtu_seg_max, from->nb_mtu_seg_max);
1061 }
1062
1063 /*
1064  * Merge the information from sub-devices.
1065  *
1066  * The reported values must be the common subset of all sub devices
1067  */
1068 static void
1069 fs_dev_merge_info(struct rte_eth_dev_info *info,
1070                   const struct rte_eth_dev_info *sinfo)
1071 {
1072         info->max_rx_pktlen = RTE_MIN(info->max_rx_pktlen, sinfo->max_rx_pktlen);
1073         info->max_rx_queues = RTE_MIN(info->max_rx_queues, sinfo->max_rx_queues);
1074         info->max_tx_queues = RTE_MIN(info->max_tx_queues, sinfo->max_tx_queues);
1075         info->max_mac_addrs = RTE_MIN(info->max_mac_addrs, sinfo->max_mac_addrs);
1076         info->max_hash_mac_addrs = RTE_MIN(info->max_hash_mac_addrs,
1077                                         sinfo->max_hash_mac_addrs);
1078         info->max_vmdq_pools = RTE_MIN(info->max_vmdq_pools, sinfo->max_vmdq_pools);
1079         info->max_vfs = RTE_MIN(info->max_vfs, sinfo->max_vfs);
1080
1081         fs_dev_merge_desc_lim(&info->rx_desc_lim, &sinfo->rx_desc_lim);
1082         fs_dev_merge_desc_lim(&info->tx_desc_lim, &sinfo->tx_desc_lim);
1083
1084         info->rx_offload_capa &= sinfo->rx_offload_capa;
1085         info->tx_offload_capa &= sinfo->tx_offload_capa;
1086         info->rx_queue_offload_capa &= sinfo->rx_queue_offload_capa;
1087         info->tx_queue_offload_capa &= sinfo->tx_queue_offload_capa;
1088         info->flow_type_rss_offloads &= sinfo->flow_type_rss_offloads;
1089
1090         /*
1091          * RETA size is a GCD of RETA sizes indicated by sub-devices.
1092          * Each of these sizes is a power of 2, so use the lower one.
1093          */
1094         info->reta_size = RTE_MIN(info->reta_size, sinfo->reta_size);
1095
1096         info->hash_key_size = RTE_MIN(info->hash_key_size,
1097                                       sinfo->hash_key_size);
1098 }
1099
1100 /**
1101  * Fail-safe dev_infos_get rules:
1102  *
1103  * No sub_device:
1104  *   Numerables:
1105  *      Use the maximum possible values for any field, so as not
1106  *      to impede any further configuration effort.
1107  *   Capabilities:
1108  *      Limits capabilities to those that are understood by the
1109  *      fail-safe PMD. This understanding stems from the fail-safe
1110  *      being capable of verifying that the related capability is
1111  *      expressed within the device configuration (struct rte_eth_conf).
1112  *
1113  * At least one probed sub_device:
1114  *   Numerables:
1115  *      Uses values from the active probed sub_device
1116  *      The rationale here is that if any sub_device is less capable
1117  *      (for example concerning the number of queues) than the active
1118  *      sub_device, then its subsequent configuration will fail.
1119  *      It is impossible to foresee this failure when the failing sub_device
1120  *      is supposed to be plugged-in later on, so the configuration process
1121  *      is the single point of failure and error reporting.
1122  *   Capabilities:
1123  *      Uses a logical AND of RX capabilities among
1124  *      all sub_devices and the default capabilities.
1125  *      Uses a logical AND of TX capabilities among
1126  *      the active probed sub_device and the default capabilities.
1127  *      Uses a logical AND of device capabilities among
1128  *      all sub_devices and the default capabilities.
1129  *
1130  */
1131 static int
1132 fs_dev_infos_get(struct rte_eth_dev *dev,
1133                   struct rte_eth_dev_info *infos)
1134 {
1135         struct sub_device *sdev;
1136         uint8_t i;
1137         int ret;
1138
1139         /* Use maximum upper bounds by default */
1140         infos->max_rx_pktlen = UINT32_MAX;
1141         infos->max_rx_queues = RTE_MAX_QUEUES_PER_PORT;
1142         infos->max_tx_queues = RTE_MAX_QUEUES_PER_PORT;
1143         infos->max_mac_addrs = FAILSAFE_MAX_ETHADDR;
1144         infos->max_hash_mac_addrs = UINT32_MAX;
1145         infos->max_vfs = UINT16_MAX;
1146         infos->max_vmdq_pools = UINT16_MAX;
1147         infos->reta_size = UINT16_MAX;
1148         infos->hash_key_size = UINT8_MAX;
1149
1150         /*
1151          * Set of capabilities that can be verified upon
1152          * configuring a sub-device.
1153          */
1154         infos->rx_offload_capa =
1155                 DEV_RX_OFFLOAD_VLAN_STRIP |
1156                 DEV_RX_OFFLOAD_IPV4_CKSUM |
1157                 DEV_RX_OFFLOAD_UDP_CKSUM |
1158                 DEV_RX_OFFLOAD_TCP_CKSUM |
1159                 DEV_RX_OFFLOAD_TCP_LRO |
1160                 DEV_RX_OFFLOAD_QINQ_STRIP |
1161                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1162                 DEV_RX_OFFLOAD_MACSEC_STRIP |
1163                 DEV_RX_OFFLOAD_HEADER_SPLIT |
1164                 DEV_RX_OFFLOAD_VLAN_FILTER |
1165                 DEV_RX_OFFLOAD_VLAN_EXTEND |
1166                 DEV_RX_OFFLOAD_JUMBO_FRAME |
1167                 DEV_RX_OFFLOAD_SCATTER |
1168                 DEV_RX_OFFLOAD_TIMESTAMP |
1169                 DEV_RX_OFFLOAD_SECURITY;
1170
1171         infos->rx_queue_offload_capa =
1172                 DEV_RX_OFFLOAD_VLAN_STRIP |
1173                 DEV_RX_OFFLOAD_IPV4_CKSUM |
1174                 DEV_RX_OFFLOAD_UDP_CKSUM |
1175                 DEV_RX_OFFLOAD_TCP_CKSUM |
1176                 DEV_RX_OFFLOAD_TCP_LRO |
1177                 DEV_RX_OFFLOAD_QINQ_STRIP |
1178                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1179                 DEV_RX_OFFLOAD_MACSEC_STRIP |
1180                 DEV_RX_OFFLOAD_HEADER_SPLIT |
1181                 DEV_RX_OFFLOAD_VLAN_FILTER |
1182                 DEV_RX_OFFLOAD_VLAN_EXTEND |
1183                 DEV_RX_OFFLOAD_JUMBO_FRAME |
1184                 DEV_RX_OFFLOAD_SCATTER |
1185                 DEV_RX_OFFLOAD_TIMESTAMP |
1186                 DEV_RX_OFFLOAD_SECURITY;
1187
1188         infos->tx_offload_capa =
1189                 DEV_TX_OFFLOAD_MULTI_SEGS |
1190                 DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1191                 DEV_TX_OFFLOAD_IPV4_CKSUM |
1192                 DEV_TX_OFFLOAD_UDP_CKSUM |
1193                 DEV_TX_OFFLOAD_TCP_CKSUM |
1194                 DEV_TX_OFFLOAD_TCP_TSO;
1195
1196         infos->flow_type_rss_offloads =
1197                 ETH_RSS_IP |
1198                 ETH_RSS_UDP |
1199                 ETH_RSS_TCP;
1200         infos->dev_capa =
1201                 RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
1202                 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
1203
1204         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
1205                 struct rte_eth_dev_info sub_info;
1206
1207                 ret = rte_eth_dev_info_get(PORT_ID(sdev), &sub_info);
1208                 ret = fs_err(sdev, ret);
1209                 if (ret != 0)
1210                         return ret;
1211
1212                 fs_dev_merge_info(infos, &sub_info);
1213         }
1214
1215         return 0;
1216 }
1217
1218 static const uint32_t *
1219 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1220 {
1221         struct sub_device *sdev;
1222         struct rte_eth_dev *edev;
1223         const uint32_t *ret;
1224
1225         fs_lock(dev, 0);
1226         sdev = TX_SUBDEV(dev);
1227         if (sdev == NULL) {
1228                 ret = NULL;
1229                 goto unlock;
1230         }
1231         edev = ETH(sdev);
1232         /* ENOTSUP: counts as no supported ptypes */
1233         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
1234                 ret = NULL;
1235                 goto unlock;
1236         }
1237         /*
1238          * The API does not permit to do a clean AND of all ptypes,
1239          * It is also incomplete by design and we do not really care
1240          * to have a best possible value in this context.
1241          * We just return the ptypes of the device of highest
1242          * priority, usually the PREFERRED device.
1243          */
1244         ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev);
1245 unlock:
1246         fs_unlock(dev, 0);
1247         return ret;
1248 }
1249
1250 static int
1251 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1252 {
1253         struct sub_device *sdev;
1254         uint8_t i;
1255         int ret;
1256
1257         fs_lock(dev, 0);
1258         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1259                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
1260                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
1261                 if ((ret = fs_err(sdev, ret))) {
1262                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
1263                               i, ret);
1264                         fs_unlock(dev, 0);
1265                         return ret;
1266                 }
1267         }
1268         fs_unlock(dev, 0);
1269         return 0;
1270 }
1271
1272 static int
1273 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1274 {
1275         struct sub_device *sdev;
1276         uint8_t i;
1277         int ret;
1278
1279         fs_lock(dev, 0);
1280         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1281                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
1282                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
1283                 if ((ret = fs_err(sdev, ret))) {
1284                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
1285                               " with error %d", i, ret);
1286                         fs_unlock(dev, 0);
1287                         return ret;
1288                 }
1289         }
1290         fs_unlock(dev, 0);
1291         return 0;
1292 }
1293
1294 static int
1295 fs_flow_ctrl_get(struct rte_eth_dev *dev,
1296                 struct rte_eth_fc_conf *fc_conf)
1297 {
1298         struct sub_device *sdev;
1299         int ret;
1300
1301         fs_lock(dev, 0);
1302         sdev = TX_SUBDEV(dev);
1303         if (sdev == NULL) {
1304                 ret = 0;
1305                 goto unlock;
1306         }
1307         if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
1308                 ret = -ENOTSUP;
1309                 goto unlock;
1310         }
1311         ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
1312 unlock:
1313         fs_unlock(dev, 0);
1314         return ret;
1315 }
1316
1317 static int
1318 fs_flow_ctrl_set(struct rte_eth_dev *dev,
1319                 struct rte_eth_fc_conf *fc_conf)
1320 {
1321         struct sub_device *sdev;
1322         uint8_t i;
1323         int ret;
1324
1325         fs_lock(dev, 0);
1326         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1327                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
1328                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
1329                 if ((ret = fs_err(sdev, ret))) {
1330                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
1331                               " with error %d", i, ret);
1332                         fs_unlock(dev, 0);
1333                         return ret;
1334                 }
1335         }
1336         fs_unlock(dev, 0);
1337         return 0;
1338 }
1339
1340 static void
1341 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1342 {
1343         struct sub_device *sdev;
1344         uint8_t i;
1345
1346         fs_lock(dev, 0);
1347         /* No check: already done within the rte_eth_dev_mac_addr_remove
1348          * call for the fail-safe device.
1349          */
1350         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
1351                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
1352                                 &dev->data->mac_addrs[index]);
1353         PRIV(dev)->mac_addr_pool[index] = 0;
1354         fs_unlock(dev, 0);
1355 }
1356
1357 static int
1358 fs_mac_addr_add(struct rte_eth_dev *dev,
1359                 struct rte_ether_addr *mac_addr,
1360                 uint32_t index,
1361                 uint32_t vmdq)
1362 {
1363         struct sub_device *sdev;
1364         int ret;
1365         uint8_t i;
1366
1367         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1368         fs_lock(dev, 0);
1369         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1370                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1371                 if ((ret = fs_err(sdev, ret))) {
1372                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1373                               PRIu8 " with error %d", i, ret);
1374                         fs_unlock(dev, 0);
1375                         return ret;
1376                 }
1377         }
1378         if (index >= PRIV(dev)->nb_mac_addr) {
1379                 DEBUG("Growing mac_addrs array");
1380                 PRIV(dev)->nb_mac_addr = index;
1381         }
1382         PRIV(dev)->mac_addr_pool[index] = vmdq;
1383         fs_unlock(dev, 0);
1384         return 0;
1385 }
1386
1387 static int
1388 fs_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1389 {
1390         struct sub_device *sdev;
1391         uint8_t i;
1392         int ret;
1393
1394         fs_lock(dev, 0);
1395         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1396                 ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1397                 ret = fs_err(sdev, ret);
1398                 if (ret) {
1399                         ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1400                                 i, ret);
1401                         fs_unlock(dev, 0);
1402                         return ret;
1403                 }
1404         }
1405         fs_unlock(dev, 0);
1406
1407         return 0;
1408 }
1409
1410 static int
1411 fs_set_mc_addr_list(struct rte_eth_dev *dev,
1412                     struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1413 {
1414         struct sub_device *sdev;
1415         uint8_t i;
1416         int ret;
1417         void *mcast_addrs;
1418
1419         fs_lock(dev, 0);
1420
1421         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1422                 ret = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1423                                                    mc_addr_set, nb_mc_addr);
1424                 if (ret != 0) {
1425                         ERROR("Operation rte_eth_dev_set_mc_addr_list failed for sub_device %d with error %d",
1426                               i, ret);
1427                         goto rollback;
1428                 }
1429         }
1430
1431         mcast_addrs = rte_realloc(PRIV(dev)->mcast_addrs,
1432                 nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]), 0);
1433         if (mcast_addrs == NULL && nb_mc_addr > 0) {
1434                 ret = -ENOMEM;
1435                 goto rollback;
1436         }
1437         rte_memcpy(mcast_addrs, mc_addr_set,
1438                    nb_mc_addr * sizeof(PRIV(dev)->mcast_addrs[0]));
1439         PRIV(dev)->nb_mcast_addr = nb_mc_addr;
1440         PRIV(dev)->mcast_addrs = mcast_addrs;
1441
1442         fs_unlock(dev, 0);
1443         return 0;
1444
1445 rollback:
1446         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1447                 int rc = rte_eth_dev_set_mc_addr_list(PORT_ID(sdev),
1448                         PRIV(dev)->mcast_addrs, PRIV(dev)->nb_mcast_addr);
1449                 if (rc != 0) {
1450                         ERROR("Multicast MAC address list rollback for sub_device %d failed with error %d",
1451                               i, rc);
1452                 }
1453         }
1454
1455         fs_unlock(dev, 0);
1456         return ret;
1457 }
1458
1459 static int
1460 fs_rss_hash_update(struct rte_eth_dev *dev,
1461                         struct rte_eth_rss_conf *rss_conf)
1462 {
1463         struct sub_device *sdev;
1464         uint8_t i;
1465         int ret;
1466
1467         fs_lock(dev, 0);
1468         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1469                 ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1470                 ret = fs_err(sdev, ret);
1471                 if (ret) {
1472                         ERROR("Operation rte_eth_dev_rss_hash_update"
1473                                 " failed for sub_device %d with error %d",
1474                                 i, ret);
1475                         fs_unlock(dev, 0);
1476                         return ret;
1477                 }
1478         }
1479         fs_unlock(dev, 0);
1480
1481         return 0;
1482 }
1483
1484 static int
1485 fs_filter_ctrl(struct rte_eth_dev *dev __rte_unused,
1486                 enum rte_filter_type type,
1487                 enum rte_filter_op op,
1488                 void *arg)
1489 {
1490         if (type == RTE_ETH_FILTER_GENERIC &&
1491             op == RTE_ETH_FILTER_GET) {
1492                 *(const void **)arg = &fs_flow_ops;
1493                 return 0;
1494         }
1495         return -ENOTSUP;
1496 }
1497
1498 const struct eth_dev_ops failsafe_ops = {
1499         .dev_configure = fs_dev_configure,
1500         .dev_start = fs_dev_start,
1501         .dev_stop = fs_dev_stop,
1502         .dev_set_link_down = fs_dev_set_link_down,
1503         .dev_set_link_up = fs_dev_set_link_up,
1504         .dev_close = failsafe_eth_dev_close,
1505         .promiscuous_enable = fs_promiscuous_enable,
1506         .promiscuous_disable = fs_promiscuous_disable,
1507         .allmulticast_enable = fs_allmulticast_enable,
1508         .allmulticast_disable = fs_allmulticast_disable,
1509         .link_update = fs_link_update,
1510         .stats_get = fs_stats_get,
1511         .stats_reset = fs_stats_reset,
1512         .xstats_get = fs_xstats_get,
1513         .xstats_get_names = fs_xstats_get_names,
1514         .xstats_reset = fs_xstats_reset,
1515         .dev_infos_get = fs_dev_infos_get,
1516         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1517         .mtu_set = fs_mtu_set,
1518         .vlan_filter_set = fs_vlan_filter_set,
1519         .rx_queue_start = fs_rx_queue_start,
1520         .rx_queue_stop = fs_rx_queue_stop,
1521         .tx_queue_start = fs_tx_queue_start,
1522         .tx_queue_stop = fs_tx_queue_stop,
1523         .rx_queue_setup = fs_rx_queue_setup,
1524         .tx_queue_setup = fs_tx_queue_setup,
1525         .rx_queue_release = fs_rx_queue_release,
1526         .tx_queue_release = fs_tx_queue_release,
1527         .rx_queue_intr_enable = fs_rx_intr_enable,
1528         .rx_queue_intr_disable = fs_rx_intr_disable,
1529         .flow_ctrl_get = fs_flow_ctrl_get,
1530         .flow_ctrl_set = fs_flow_ctrl_set,
1531         .mac_addr_remove = fs_mac_addr_remove,
1532         .mac_addr_add = fs_mac_addr_add,
1533         .mac_addr_set = fs_mac_addr_set,
1534         .set_mc_addr_list = fs_set_mc_addr_list,
1535         .rss_hash_update = fs_rss_hash_update,
1536         .filter_ctrl = fs_filter_ctrl,
1537 };