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
  18 #include "failsafe_private.h"
  19
  20 static struct rte_eth_dev_info default_infos = {
  21         /* Max possible number of elements */
  22         .max_rx_pktlen = UINT32_MAX,
  23         .max_rx_queues = RTE_MAX_QUEUES_PER_PORT,
  24         .max_tx_queues = RTE_MAX_QUEUES_PER_PORT,
  25         .max_mac_addrs = FAILSAFE_MAX_ETHADDR,
  26         .max_hash_mac_addrs = UINT32_MAX,
  27         .max_vfs = UINT16_MAX,
  28         .max_vmdq_pools = UINT16_MAX,
  29         .rx_desc_lim = {
  30                 .nb_max = UINT16_MAX,
  31                 .nb_min = 0,
  32                 .nb_align = 1,
  33                 .nb_seg_max = UINT16_MAX,
  34                 .nb_mtu_seg_max = UINT16_MAX,
  35         },
  36         .tx_desc_lim = {
  37                 .nb_max = UINT16_MAX,
  38                 .nb_min = 0,
  39                 .nb_align = 1,
  40                 .nb_seg_max = UINT16_MAX,
  41                 .nb_mtu_seg_max = UINT16_MAX,
  42         },
  43         /*
  44          * Set of capabilities that can be verified upon
  45          * configuring a sub-device.
  46          */
  47         .rx_offload_capa =
  48                 DEV_RX_OFFLOAD_VLAN_STRIP |
  49                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  50                 DEV_RX_OFFLOAD_UDP_CKSUM |
  51                 DEV_RX_OFFLOAD_TCP_CKSUM |
  52                 DEV_RX_OFFLOAD_TCP_LRO |
  53                 DEV_RX_OFFLOAD_QINQ_STRIP |
  54                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  55                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  56                 DEV_RX_OFFLOAD_HEADER_SPLIT |
  57                 DEV_RX_OFFLOAD_VLAN_FILTER |
  58                 DEV_RX_OFFLOAD_VLAN_EXTEND |
  59                 DEV_RX_OFFLOAD_JUMBO_FRAME |
  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_SCATTER |
  77                 DEV_RX_OFFLOAD_TIMESTAMP |
  78                 DEV_RX_OFFLOAD_SECURITY,
  79         .tx_offload_capa =
  80                 DEV_TX_OFFLOAD_MULTI_SEGS |
  81                 DEV_TX_OFFLOAD_IPV4_CKSUM |
  82                 DEV_TX_OFFLOAD_UDP_CKSUM |
  83                 DEV_TX_OFFLOAD_TCP_CKSUM |
  84                 DEV_TX_OFFLOAD_TCP_TSO,
  85         .flow_type_rss_offloads =
  86                         ETH_RSS_IP |
  87                         ETH_RSS_UDP |
  88                         ETH_RSS_TCP,
  89 };
  90
  91 static int
  92 fs_dev_configure(struct rte_eth_dev *dev)
  93 {
  94         struct sub_device *sdev;
  95         uint8_t i;
  96         int ret;
  97
  98         fs_lock(dev, 0);
  99         FOREACH_SUBDEV(sdev, i, dev) {
 100                 int rmv_interrupt = 0;
 101                 int lsc_interrupt = 0;
 102                 int lsc_enabled;
 103
 104                 if (sdev->state != DEV_PROBED &&
 105                     !(PRIV(dev)->alarm_lock == 0 && sdev->state == DEV_ACTIVE))
 106                         continue;
 107
 108                 rmv_interrupt = ETH(sdev)->data->dev_flags &
 109                                 RTE_ETH_DEV_INTR_RMV;
 110                 if (rmv_interrupt) {
 111                         DEBUG("Enabling RMV interrupts for sub_device %d", i);
 112                         dev->data->dev_conf.intr_conf.rmv = 1;
 113                 } else {
 114                         DEBUG("sub_device %d does not support RMV event", i);
 115                 }
 116                 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
 117                 lsc_interrupt = lsc_enabled &&
 118                                 (ETH(sdev)->data->dev_flags &
 119                                  RTE_ETH_DEV_INTR_LSC);
 120                 if (lsc_interrupt) {
 121                         DEBUG("Enabling LSC interrupts for sub_device %d", i);
 122                         dev->data->dev_conf.intr_conf.lsc = 1;
 123                 } else if (lsc_enabled && !lsc_interrupt) {
 124                         DEBUG("Disabling LSC interrupts for sub_device %d", i);
 125                         dev->data->dev_conf.intr_conf.lsc = 0;
 126                 }
 127                 DEBUG("Configuring sub-device %d", i);
 128                 ret = rte_eth_dev_configure(PORT_ID(sdev),
 129                                         dev->data->nb_rx_queues,
 130                                         dev->data->nb_tx_queues,
 131                                         &dev->data->dev_conf);
 132                 if (ret) {
 133                         if (!fs_err(sdev, ret))
 134                                 continue;
 135                         ERROR("Could not configure sub_device %d", i);
 136                         fs_unlock(dev, 0);
 137                         return ret;
 138                 }
 139                 if (rmv_interrupt && sdev->rmv_callback == 0) {
 140                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 141                                         RTE_ETH_EVENT_INTR_RMV,
 142                                         failsafe_eth_rmv_event_callback,
 143                                         sdev);
 144                         if (ret)
 145                                 WARN("Failed to register RMV callback for sub_device %d",
 146                                      SUB_ID(sdev));
 147                         else
 148                                 sdev->rmv_callback = 1;
 149                 }
 150                 dev->data->dev_conf.intr_conf.rmv = 0;
 151                 if (lsc_interrupt && sdev->lsc_callback == 0) {
 152                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 153                                                 RTE_ETH_EVENT_INTR_LSC,
 154                                                 failsafe_eth_lsc_event_callback,
 155                                                 dev);
 156                         if (ret)
 157                                 WARN("Failed to register LSC callback for sub_device %d",
 158                                      SUB_ID(sdev));
 159                         else
 160                                 sdev->lsc_callback = 1;
 161                 }
 162                 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
 163                 sdev->state = DEV_ACTIVE;
 164         }
 165         if (PRIV(dev)->state < DEV_ACTIVE)
 166                 PRIV(dev)->state = DEV_ACTIVE;
 167         fs_unlock(dev, 0);
 168         return 0;
 169 }
 170
 171 static void
 172 fs_set_queues_state_start(struct rte_eth_dev *dev)
 173 {
 174         struct rxq *rxq;
 175         uint16_t i;
 176
 177         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 178                 rxq = dev->data->rx_queues[i];
 179                 if (!rxq->info.conf.rx_deferred_start)
 180                         dev->data->rx_queue_state[i] =
 181                                                 RTE_ETH_QUEUE_STATE_STARTED;
 182         }
 183 }
 184
 185 static int
 186 fs_dev_start(struct rte_eth_dev *dev)
 187 {
 188         struct sub_device *sdev;
 189         uint8_t i;
 190         int ret;
 191
 192         fs_lock(dev, 0);
 193         ret = failsafe_rx_intr_install(dev);
 194         if (ret) {
 195                 fs_unlock(dev, 0);
 196                 return ret;
 197         }
 198         FOREACH_SUBDEV(sdev, i, dev) {
 199                 if (sdev->state != DEV_ACTIVE)
 200                         continue;
 201                 DEBUG("Starting sub_device %d", i);
 202                 ret = rte_eth_dev_start(PORT_ID(sdev));
 203                 if (ret) {
 204                         if (!fs_err(sdev, ret))
 205                                 continue;
 206                         fs_unlock(dev, 0);
 207                         return ret;
 208                 }
 209                 ret = failsafe_rx_intr_install_subdevice(sdev);
 210                 if (ret) {
 211                         if (!fs_err(sdev, ret))
 212                                 continue;
 213                         rte_eth_dev_stop(PORT_ID(sdev));
 214                         fs_unlock(dev, 0);
 215                         return ret;
 216                 }
 217                 sdev->state = DEV_STARTED;
 218         }
 219         if (PRIV(dev)->state < DEV_STARTED) {
 220                 PRIV(dev)->state = DEV_STARTED;
 221                 fs_set_queues_state_start(dev);
 222         }
 223         fs_switch_dev(dev, NULL);
 224         fs_unlock(dev, 0);
 225         return 0;
 226 }
 227
 228 static void
 229 fs_set_queues_state_stop(struct rte_eth_dev *dev)
 230 {
 231         uint16_t i;
 232
 233         for (i = 0; i < dev->data->nb_rx_queues; i++)
 234                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 235 }
 236
 237 static void
 238 fs_dev_stop(struct rte_eth_dev *dev)
 239 {
 240         struct sub_device *sdev;
 241         uint8_t i;
 242
 243         fs_lock(dev, 0);
 244         PRIV(dev)->state = DEV_STARTED - 1;
 245         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 246                 rte_eth_dev_stop(PORT_ID(sdev));
 247                 failsafe_rx_intr_uninstall_subdevice(sdev);
 248                 sdev->state = DEV_STARTED - 1;
 249         }
 250         failsafe_rx_intr_uninstall(dev);
 251         fs_set_queues_state_stop(dev);
 252         fs_unlock(dev, 0);
 253 }
 254
 255 static int
 256 fs_dev_set_link_up(struct rte_eth_dev *dev)
 257 {
 258         struct sub_device *sdev;
 259         uint8_t i;
 260         int ret;
 261
 262         fs_lock(dev, 0);
 263         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 264                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 265                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 266                 if ((ret = fs_err(sdev, ret))) {
 267                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 268                               " with error %d", i, ret);
 269                         fs_unlock(dev, 0);
 270                         return ret;
 271                 }
 272         }
 273         fs_unlock(dev, 0);
 274         return 0;
 275 }
 276
 277 static int
 278 fs_dev_set_link_down(struct rte_eth_dev *dev)
 279 {
 280         struct sub_device *sdev;
 281         uint8_t i;
 282         int ret;
 283
 284         fs_lock(dev, 0);
 285         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 286                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 287                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 288                 if ((ret = fs_err(sdev, ret))) {
 289                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 290                               " with error %d", i, ret);
 291                         fs_unlock(dev, 0);
 292                         return ret;
 293                 }
 294         }
 295         fs_unlock(dev, 0);
 296         return 0;
 297 }
 298
 299 static void fs_dev_free_queues(struct rte_eth_dev *dev);
 300 static void
 301 fs_dev_close(struct rte_eth_dev *dev)
 302 {
 303         struct sub_device *sdev;
 304         uint8_t i;
 305
 306         fs_lock(dev, 0);
 307         failsafe_hotplug_alarm_cancel(dev);
 308         if (PRIV(dev)->state == DEV_STARTED)
 309                 dev->dev_ops->dev_stop(dev);
 310         PRIV(dev)->state = DEV_ACTIVE - 1;
 311         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 312                 DEBUG("Closing sub_device %d", i);
 313                 failsafe_eth_dev_unregister_callbacks(sdev);
 314                 rte_eth_dev_close(PORT_ID(sdev));
 315                 sdev->state = DEV_ACTIVE - 1;
 316         }
 317         fs_dev_free_queues(dev);
 318         fs_unlock(dev, 0);
 319 }
 320
 321 static int
 322 fs_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 323 {
 324         struct sub_device *sdev;
 325         uint8_t i;
 326         int ret;
 327         int err = 0;
 328         bool failure = true;
 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_rx_queue_stop(port_id, rx_queue_id);
 335                 ret = fs_err(sdev, ret);
 336                 if (ret) {
 337                         ERROR("Rx queue stop failed for subdevice %d", i);
 338                         err = ret;
 339                 } else {
 340                         failure = false;
 341                 }
 342         }
 343         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 344         fs_unlock(dev, 0);
 345         /* Return 0 in case of at least one successful queue stop */
 346         return (failure) ? err : 0;
 347 }
 348
 349 static int
 350 fs_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 351 {
 352         struct sub_device *sdev;
 353         uint8_t i;
 354         int ret;
 355
 356         fs_lock(dev, 0);
 357         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 358                 uint16_t port_id = ETH(sdev)->data->port_id;
 359
 360                 ret = rte_eth_dev_rx_queue_start(port_id, rx_queue_id);
 361                 ret = fs_err(sdev, ret);
 362                 if (ret) {
 363                         ERROR("Rx queue start failed for subdevice %d", i);
 364                         fs_rx_queue_stop(dev, rx_queue_id);
 365                         fs_unlock(dev, 0);
 366                         return ret;
 367                 }
 368         }
 369         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 370         fs_unlock(dev, 0);
 371         return 0;
 372 }
 373
 374 static void
 375 fs_rx_queue_release(void *queue)
 376 {
 377         struct rte_eth_dev *dev;
 378         struct sub_device *sdev;
 379         uint8_t i;
 380         struct rxq *rxq;
 381
 382         if (queue == NULL)
 383                 return;
 384         rxq = queue;
 385         dev = rxq->priv->dev;
 386         fs_lock(dev, 0);
 387         if (rxq->event_fd > 0)
 388                 close(rxq->event_fd);
 389         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 390                 if (ETH(sdev)->data->rx_queues != NULL &&
 391                     ETH(sdev)->data->rx_queues[rxq->qid] != NULL) {
 392                         SUBOPS(sdev, rx_queue_release)
 393                                 (ETH(sdev)->data->rx_queues[rxq->qid]);
 394                 }
 395         }
 396         dev->data->rx_queues[rxq->qid] = NULL;
 397         rte_free(rxq);
 398         fs_unlock(dev, 0);
 399 }
 400
 401 static int
 402 fs_rx_queue_setup(struct rte_eth_dev *dev,
 403                 uint16_t rx_queue_id,
 404                 uint16_t nb_rx_desc,
 405                 unsigned int socket_id,
 406                 const struct rte_eth_rxconf *rx_conf,
 407                 struct rte_mempool *mb_pool)
 408 {
 409         /*
 410          * FIXME: Add a proper interface in rte_eal_interrupts for
 411          * allocating eventfd as an interrupt vector.
 412          * For the time being, fake as if we are using MSIX interrupts,
 413          * this will cause rte_intr_efd_enable to allocate an eventfd for us.
 414          */
 415         struct rte_intr_handle intr_handle = {
 416                 .type = RTE_INTR_HANDLE_VFIO_MSIX,
 417                 .efds = { -1, },
 418         };
 419         struct sub_device *sdev;
 420         struct rxq *rxq;
 421         uint8_t i;
 422         int ret;
 423
 424         fs_lock(dev, 0);
 425         if (rx_conf->rx_deferred_start) {
 426                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 427                         if (SUBOPS(sdev, rx_queue_start) == NULL) {
 428                                 ERROR("Rx queue deferred start is not "
 429                                         "supported for subdevice %d", i);
 430                                 fs_unlock(dev, 0);
 431                                 return -EINVAL;
 432                         }
 433                 }
 434         }
 435         rxq = dev->data->rx_queues[rx_queue_id];
 436         if (rxq != NULL) {
 437                 fs_rx_queue_release(rxq);
 438                 dev->data->rx_queues[rx_queue_id] = NULL;
 439         }
 440         rxq = rte_zmalloc(NULL,
 441                           sizeof(*rxq) +
 442                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 443                           RTE_CACHE_LINE_SIZE);
 444         if (rxq == NULL) {
 445                 fs_unlock(dev, 0);
 446                 return -ENOMEM;
 447         }
 448         FOREACH_SUBDEV(sdev, i, dev)
 449                 rte_atomic64_init(&rxq->refcnt[i]);
 450         rxq->qid = rx_queue_id;
 451         rxq->socket_id = socket_id;
 452         rxq->info.mp = mb_pool;
 453         rxq->info.conf = *rx_conf;
 454         rxq->info.nb_desc = nb_rx_desc;
 455         rxq->priv = PRIV(dev);
 456         rxq->sdev = PRIV(dev)->subs;
 457         ret = rte_intr_efd_enable(&intr_handle, 1);
 458         if (ret < 0) {
 459                 fs_unlock(dev, 0);
 460                 return ret;
 461         }
 462         rxq->event_fd = intr_handle.efds[0];
 463         dev->data->rx_queues[rx_queue_id] = rxq;
 464         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 465                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 466                                 rx_queue_id,
 467                                 nb_rx_desc, socket_id,
 468                                 rx_conf, mb_pool);
 469                 if ((ret = fs_err(sdev, ret))) {
 470                         ERROR("RX queue setup failed for sub_device %d", i);
 471                         goto free_rxq;
 472                 }
 473         }
 474         fs_unlock(dev, 0);
 475         return 0;
 476 free_rxq:
 477         fs_rx_queue_release(rxq);
 478         fs_unlock(dev, 0);
 479         return ret;
 480 }
 481
 482 static int
 483 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
 484 {
 485         struct rxq *rxq;
 486         struct sub_device *sdev;
 487         uint8_t i;
 488         int ret;
 489         int rc = 0;
 490
 491         fs_lock(dev, 0);
 492         if (idx >= dev->data->nb_rx_queues) {
 493                 rc = -EINVAL;
 494                 goto unlock;
 495         }
 496         rxq = dev->data->rx_queues[idx];
 497         if (rxq == NULL || rxq->event_fd <= 0) {
 498                 rc = -EINVAL;
 499                 goto unlock;
 500         }
 501         /* Fail if proxy service is nor running. */
 502         if (PRIV(dev)->rxp.sstate != SS_RUNNING) {
 503                 ERROR("failsafe interrupt services are not running");
 504                 rc = -EAGAIN;
 505                 goto unlock;
 506         }
 507         rxq->enable_events = 1;
 508         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 509                 ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
 510                 ret = fs_err(sdev, ret);
 511                 if (ret)
 512                         rc = ret;
 513         }
 514 unlock:
 515         fs_unlock(dev, 0);
 516         if (rc)
 517                 rte_errno = -rc;
 518         return rc;
 519 }
 520
 521 static int
 522 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
 523 {
 524         struct rxq *rxq;
 525         struct sub_device *sdev;
 526         uint64_t u64;
 527         uint8_t i;
 528         int rc = 0;
 529         int ret;
 530
 531         fs_lock(dev, 0);
 532         if (idx >= dev->data->nb_rx_queues) {
 533                 rc = -EINVAL;
 534                 goto unlock;
 535         }
 536         rxq = dev->data->rx_queues[idx];
 537         if (rxq == NULL || rxq->event_fd <= 0) {
 538                 rc = -EINVAL;
 539                 goto unlock;
 540         }
 541         rxq->enable_events = 0;
 542         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 543                 ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
 544                 ret = fs_err(sdev, ret);
 545                 if (ret)
 546                         rc = ret;
 547         }
 548         /* Clear pending events */
 549         while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
 550                 ;
 551 unlock:
 552         fs_unlock(dev, 0);
 553         if (rc)
 554                 rte_errno = -rc;
 555         return rc;
 556 }
 557
 558 static void
 559 fs_tx_queue_release(void *queue)
 560 {
 561         struct rte_eth_dev *dev;
 562         struct sub_device *sdev;
 563         uint8_t i;
 564         struct txq *txq;
 565
 566         if (queue == NULL)
 567                 return;
 568         txq = queue;
 569         dev = txq->priv->dev;
 570         fs_lock(dev, 0);
 571         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 572                 if (ETH(sdev)->data->tx_queues != NULL &&
 573                     ETH(sdev)->data->tx_queues[txq->qid] != NULL) {
 574                         SUBOPS(sdev, tx_queue_release)
 575                                 (ETH(sdev)->data->tx_queues[txq->qid]);
 576                 }
 577         }
 578         dev->data->tx_queues[txq->qid] = NULL;
 579         rte_free(txq);
 580         fs_unlock(dev, 0);
 581 }
 582
 583 static int
 584 fs_tx_queue_setup(struct rte_eth_dev *dev,
 585                 uint16_t tx_queue_id,
 586                 uint16_t nb_tx_desc,
 587                 unsigned int socket_id,
 588                 const struct rte_eth_txconf *tx_conf)
 589 {
 590         struct sub_device *sdev;
 591         struct txq *txq;
 592         uint8_t i;
 593         int ret;
 594
 595         if (tx_conf->tx_deferred_start) {
 596                 ERROR("Tx queue deferred start is not supported");
 597                 return -EINVAL;
 598         }
 599
 600         fs_lock(dev, 0);
 601         txq = dev->data->tx_queues[tx_queue_id];
 602         if (txq != NULL) {
 603                 fs_tx_queue_release(txq);
 604                 dev->data->tx_queues[tx_queue_id] = NULL;
 605         }
 606         txq = rte_zmalloc("ethdev TX queue",
 607                           sizeof(*txq) +
 608                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 609                           RTE_CACHE_LINE_SIZE);
 610         if (txq == NULL) {
 611                 fs_unlock(dev, 0);
 612                 return -ENOMEM;
 613         }
 614         FOREACH_SUBDEV(sdev, i, dev)
 615                 rte_atomic64_init(&txq->refcnt[i]);
 616         txq->qid = tx_queue_id;
 617         txq->socket_id = socket_id;
 618         txq->info.conf = *tx_conf;
 619         txq->info.nb_desc = nb_tx_desc;
 620         txq->priv = PRIV(dev);
 621         dev->data->tx_queues[tx_queue_id] = txq;
 622         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 623                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 624                                 tx_queue_id,
 625                                 nb_tx_desc, socket_id,
 626                                 tx_conf);
 627                 if ((ret = fs_err(sdev, ret))) {
 628                         ERROR("TX queue setup failed for sub_device %d", i);
 629                         goto free_txq;
 630                 }
 631         }
 632         fs_unlock(dev, 0);
 633         return 0;
 634 free_txq:
 635         fs_tx_queue_release(txq);
 636         fs_unlock(dev, 0);
 637         return ret;
 638 }
 639
 640 static void
 641 fs_dev_free_queues(struct rte_eth_dev *dev)
 642 {
 643         uint16_t i;
 644
 645         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 646                 fs_rx_queue_release(dev->data->rx_queues[i]);
 647                 dev->data->rx_queues[i] = NULL;
 648         }
 649         dev->data->nb_rx_queues = 0;
 650         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 651                 fs_tx_queue_release(dev->data->tx_queues[i]);
 652                 dev->data->tx_queues[i] = NULL;
 653         }
 654         dev->data->nb_tx_queues = 0;
 655 }
 656
 657 static void
 658 fs_promiscuous_enable(struct rte_eth_dev *dev)
 659 {
 660         struct sub_device *sdev;
 661         uint8_t i;
 662
 663         fs_lock(dev, 0);
 664         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 665                 rte_eth_promiscuous_enable(PORT_ID(sdev));
 666         fs_unlock(dev, 0);
 667 }
 668
 669 static void
 670 fs_promiscuous_disable(struct rte_eth_dev *dev)
 671 {
 672         struct sub_device *sdev;
 673         uint8_t i;
 674
 675         fs_lock(dev, 0);
 676         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 677                 rte_eth_promiscuous_disable(PORT_ID(sdev));
 678         fs_unlock(dev, 0);
 679 }
 680
 681 static void
 682 fs_allmulticast_enable(struct rte_eth_dev *dev)
 683 {
 684         struct sub_device *sdev;
 685         uint8_t i;
 686
 687         fs_lock(dev, 0);
 688         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 689                 rte_eth_allmulticast_enable(PORT_ID(sdev));
 690         fs_unlock(dev, 0);
 691 }
 692
 693 static void
 694 fs_allmulticast_disable(struct rte_eth_dev *dev)
 695 {
 696         struct sub_device *sdev;
 697         uint8_t i;
 698
 699         fs_lock(dev, 0);
 700         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 701                 rte_eth_allmulticast_disable(PORT_ID(sdev));
 702         fs_unlock(dev, 0);
 703 }
 704
 705 static int
 706 fs_link_update(struct rte_eth_dev *dev,
 707                 int wait_to_complete)
 708 {
 709         struct sub_device *sdev;
 710         uint8_t i;
 711         int ret;
 712
 713         fs_lock(dev, 0);
 714         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 715                 DEBUG("Calling link_update on sub_device %d", i);
 716                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 717                 if (ret && ret != -1 && sdev->remove == 0 &&
 718                     rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
 719                         ERROR("Link update failed for sub_device %d with error %d",
 720                               i, ret);
 721                         fs_unlock(dev, 0);
 722                         return ret;
 723                 }
 724         }
 725         if (TX_SUBDEV(dev)) {
 726                 struct rte_eth_link *l1;
 727                 struct rte_eth_link *l2;
 728
 729                 l1 = &dev->data->dev_link;
 730                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 731                 if (memcmp(l1, l2, sizeof(*l1))) {
 732                         *l1 = *l2;
 733                         fs_unlock(dev, 0);
 734                         return 0;
 735                 }
 736         }
 737         fs_unlock(dev, 0);
 738         return -1;
 739 }
 740
 741 static int
 742 fs_stats_get(struct rte_eth_dev *dev,
 743              struct rte_eth_stats *stats)
 744 {
 745         struct rte_eth_stats backup;
 746         struct sub_device *sdev;
 747         uint8_t i;
 748         int ret;
 749
 750         fs_lock(dev, 0);
 751         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 752         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 753                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 754                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 755
 756                 rte_memcpy(&backup, snapshot, sizeof(backup));
 757                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 758                 if (ret) {
 759                         if (!fs_err(sdev, ret)) {
 760                                 rte_memcpy(snapshot, &backup, sizeof(backup));
 761                                 goto inc;
 762                         }
 763                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 764                                   i, ret);
 765                         *timestamp = 0;
 766                         fs_unlock(dev, 0);
 767                         return ret;
 768                 }
 769                 *timestamp = rte_rdtsc();
 770 inc:
 771                 failsafe_stats_increment(stats, snapshot);
 772         }
 773         fs_unlock(dev, 0);
 774         return 0;
 775 }
 776
 777 static void
 778 fs_stats_reset(struct rte_eth_dev *dev)
 779 {
 780         struct sub_device *sdev;
 781         uint8_t i;
 782
 783         fs_lock(dev, 0);
 784         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 785                 rte_eth_stats_reset(PORT_ID(sdev));
 786                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 787         }
 788         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 789         fs_unlock(dev, 0);
 790 }
 791
 792 /**
 793  * Fail-safe dev_infos_get rules:
 794  *
 795  * No sub_device:
 796  *   Numerables:
 797  *      Use the maximum possible values for any field, so as not
 798  *      to impede any further configuration effort.
 799  *   Capabilities:
 800  *      Limits capabilities to those that are understood by the
 801  *      fail-safe PMD. This understanding stems from the fail-safe
 802  *      being capable of verifying that the related capability is
 803  *      expressed within the device configuration (struct rte_eth_conf).
 804  *
 805  * At least one probed sub_device:
 806  *   Numerables:
 807  *      Uses values from the active probed sub_device
 808  *      The rationale here is that if any sub_device is less capable
 809  *      (for example concerning the number of queues) than the active
 810  *      sub_device, then its subsequent configuration will fail.
 811  *      It is impossible to foresee this failure when the failing sub_device
 812  *      is supposed to be plugged-in later on, so the configuration process
 813  *      is the single point of failure and error reporting.
 814  *   Capabilities:
 815  *      Uses a logical AND of RX capabilities among
 816  *      all sub_devices and the default capabilities.
 817  *      Uses a logical AND of TX capabilities among
 818  *      the active probed sub_device and the default capabilities.
 819  *
 820  */
 821 static void
 822 fs_dev_infos_get(struct rte_eth_dev *dev,
 823                   struct rte_eth_dev_info *infos)
 824 {
 825         struct sub_device *sdev;
 826         uint8_t i;
 827
 828         sdev = TX_SUBDEV(dev);
 829         if (sdev == NULL) {
 830                 DEBUG("No probed device, using default infos");
 831                 rte_memcpy(&PRIV(dev)->infos, &default_infos,
 832                            sizeof(default_infos));
 833         } else {
 834                 uint64_t rx_offload_capa;
 835                 uint64_t rxq_offload_capa;
 836                 uint64_t rss_hf_offload_capa;
 837
 838                 rx_offload_capa = default_infos.rx_offload_capa;
 839                 rxq_offload_capa = default_infos.rx_queue_offload_capa;
 840                 rss_hf_offload_capa = default_infos.flow_type_rss_offloads;
 841                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 842                         rte_eth_dev_info_get(PORT_ID(sdev),
 843                                         &PRIV(dev)->infos);
 844                         rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
 845                         rxq_offload_capa &=
 846                                         PRIV(dev)->infos.rx_queue_offload_capa;
 847                         rss_hf_offload_capa &=
 848                                         PRIV(dev)->infos.flow_type_rss_offloads;
 849                 }
 850                 sdev = TX_SUBDEV(dev);
 851                 rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
 852                 PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
 853                 PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
 854                 PRIV(dev)->infos.flow_type_rss_offloads = rss_hf_offload_capa;
 855                 PRIV(dev)->infos.tx_offload_capa &=
 856                                         default_infos.tx_offload_capa;
 857                 PRIV(dev)->infos.tx_queue_offload_capa &=
 858                                         default_infos.tx_queue_offload_capa;
 859         }
 860         rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
 861 }
 862
 863 static const uint32_t *
 864 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 865 {
 866         struct sub_device *sdev;
 867         struct rte_eth_dev *edev;
 868         const uint32_t *ret;
 869
 870         fs_lock(dev, 0);
 871         sdev = TX_SUBDEV(dev);
 872         if (sdev == NULL) {
 873                 ret = NULL;
 874                 goto unlock;
 875         }
 876         edev = ETH(sdev);
 877         /* ENOTSUP: counts as no supported ptypes */
 878         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL) {
 879                 ret = NULL;
 880                 goto unlock;
 881         }
 882         /*
 883          * The API does not permit to do a clean AND of all ptypes,
 884          * It is also incomplete by design and we do not really care
 885          * to have a best possible value in this context.
 886          * We just return the ptypes of the device of highest
 887          * priority, usually the PREFERRED device.
 888          */
 889         ret = SUBOPS(sdev, dev_supported_ptypes_get)(edev);
 890 unlock:
 891         fs_unlock(dev, 0);
 892         return ret;
 893 }
 894
 895 static int
 896 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 897 {
 898         struct sub_device *sdev;
 899         uint8_t i;
 900         int ret;
 901
 902         fs_lock(dev, 0);
 903         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 904                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
 905                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
 906                 if ((ret = fs_err(sdev, ret))) {
 907                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
 908                               i, ret);
 909                         fs_unlock(dev, 0);
 910                         return ret;
 911                 }
 912         }
 913         fs_unlock(dev, 0);
 914         return 0;
 915 }
 916
 917 static int
 918 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 919 {
 920         struct sub_device *sdev;
 921         uint8_t i;
 922         int ret;
 923
 924         fs_lock(dev, 0);
 925         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 926                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
 927                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
 928                 if ((ret = fs_err(sdev, ret))) {
 929                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
 930                               " with error %d", i, ret);
 931                         fs_unlock(dev, 0);
 932                         return ret;
 933                 }
 934         }
 935         fs_unlock(dev, 0);
 936         return 0;
 937 }
 938
 939 static int
 940 fs_flow_ctrl_get(struct rte_eth_dev *dev,
 941                 struct rte_eth_fc_conf *fc_conf)
 942 {
 943         struct sub_device *sdev;
 944         int ret;
 945
 946         fs_lock(dev, 0);
 947         sdev = TX_SUBDEV(dev);
 948         if (sdev == NULL) {
 949                 ret = 0;
 950                 goto unlock;
 951         }
 952         if (SUBOPS(sdev, flow_ctrl_get) == NULL) {
 953                 ret = -ENOTSUP;
 954                 goto unlock;
 955         }
 956         ret = SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
 957 unlock:
 958         fs_unlock(dev, 0);
 959         return ret;
 960 }
 961
 962 static int
 963 fs_flow_ctrl_set(struct rte_eth_dev *dev,
 964                 struct rte_eth_fc_conf *fc_conf)
 965 {
 966         struct sub_device *sdev;
 967         uint8_t i;
 968         int ret;
 969
 970         fs_lock(dev, 0);
 971         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 972                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
 973                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
 974                 if ((ret = fs_err(sdev, ret))) {
 975                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
 976                               " with error %d", i, ret);
 977                         fs_unlock(dev, 0);
 978                         return ret;
 979                 }
 980         }
 981         fs_unlock(dev, 0);
 982         return 0;
 983 }
 984
 985 static void
 986 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 987 {
 988         struct sub_device *sdev;
 989         uint8_t i;
 990
 991         fs_lock(dev, 0);
 992         /* No check: already done within the rte_eth_dev_mac_addr_remove
 993          * call for the fail-safe device.
 994          */
 995         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 996                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
 997                                 &dev->data->mac_addrs[index]);
 998         PRIV(dev)->mac_addr_pool[index] = 0;
 999         fs_unlock(dev, 0);
1000 }
1001
1002 static int
1003 fs_mac_addr_add(struct rte_eth_dev *dev,
1004                 struct ether_addr *mac_addr,
1005                 uint32_t index,
1006                 uint32_t vmdq)
1007 {
1008         struct sub_device *sdev;
1009         int ret;
1010         uint8_t i;
1011
1012         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
1013         fs_lock(dev, 0);
1014         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1015                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
1016                 if ((ret = fs_err(sdev, ret))) {
1017                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
1018                               PRIu8 " with error %d", i, ret);
1019                         fs_unlock(dev, 0);
1020                         return ret;
1021                 }
1022         }
1023         if (index >= PRIV(dev)->nb_mac_addr) {
1024                 DEBUG("Growing mac_addrs array");
1025                 PRIV(dev)->nb_mac_addr = index;
1026         }
1027         PRIV(dev)->mac_addr_pool[index] = vmdq;
1028         fs_unlock(dev, 0);
1029         return 0;
1030 }
1031
1032 static int
1033 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1034 {
1035         struct sub_device *sdev;
1036         uint8_t i;
1037         int ret;
1038
1039         fs_lock(dev, 0);
1040         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1041                 ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
1042                 ret = fs_err(sdev, ret);
1043                 if (ret) {
1044                         ERROR("Operation rte_eth_dev_mac_addr_set failed for sub_device %d with error %d",
1045                                 i, ret);
1046                         fs_unlock(dev, 0);
1047                         return ret;
1048                 }
1049         }
1050         fs_unlock(dev, 0);
1051
1052         return 0;
1053 }
1054
1055 static int
1056 fs_rss_hash_update(struct rte_eth_dev *dev,
1057                         struct rte_eth_rss_conf *rss_conf)
1058 {
1059         struct sub_device *sdev;
1060         uint8_t i;
1061         int ret;
1062
1063         fs_lock(dev, 0);
1064         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1065                 ret = rte_eth_dev_rss_hash_update(PORT_ID(sdev), rss_conf);
1066                 ret = fs_err(sdev, ret);
1067                 if (ret) {
1068                         ERROR("Operation rte_eth_dev_rss_hash_update"
1069                                 " failed for sub_device %d with error %d",
1070                                 i, ret);
1071                         fs_unlock(dev, 0);
1072                         return ret;
1073                 }
1074         }
1075         fs_unlock(dev, 0);
1076
1077         return 0;
1078 }
1079
1080 static int
1081 fs_filter_ctrl(struct rte_eth_dev *dev,
1082                 enum rte_filter_type type,
1083                 enum rte_filter_op op,
1084                 void *arg)
1085 {
1086         struct sub_device *sdev;
1087         uint8_t i;
1088         int ret;
1089
1090         if (type == RTE_ETH_FILTER_GENERIC &&
1091             op == RTE_ETH_FILTER_GET) {
1092                 *(const void **)arg = &fs_flow_ops;
1093                 return 0;
1094         }
1095         fs_lock(dev, 0);
1096         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
1097                 DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
1098                 ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
1099                 if ((ret = fs_err(sdev, ret))) {
1100                         ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
1101                               " with error %d", i, ret);
1102                         fs_unlock(dev, 0);
1103                         return ret;
1104                 }
1105         }
1106         fs_unlock(dev, 0);
1107         return 0;
1108 }
1109
1110 const struct eth_dev_ops failsafe_ops = {
1111         .dev_configure = fs_dev_configure,
1112         .dev_start = fs_dev_start,
1113         .dev_stop = fs_dev_stop,
1114         .dev_set_link_down = fs_dev_set_link_down,
1115         .dev_set_link_up = fs_dev_set_link_up,
1116         .dev_close = fs_dev_close,
1117         .promiscuous_enable = fs_promiscuous_enable,
1118         .promiscuous_disable = fs_promiscuous_disable,
1119         .allmulticast_enable = fs_allmulticast_enable,
1120         .allmulticast_disable = fs_allmulticast_disable,
1121         .link_update = fs_link_update,
1122         .stats_get = fs_stats_get,
1123         .stats_reset = fs_stats_reset,
1124         .dev_infos_get = fs_dev_infos_get,
1125         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
1126         .mtu_set = fs_mtu_set,
1127         .vlan_filter_set = fs_vlan_filter_set,
1128         .rx_queue_start = fs_rx_queue_start,
1129         .rx_queue_stop = fs_rx_queue_stop,
1130         .rx_queue_setup = fs_rx_queue_setup,
1131         .tx_queue_setup = fs_tx_queue_setup,
1132         .rx_queue_release = fs_rx_queue_release,
1133         .tx_queue_release = fs_tx_queue_release,
1134         .rx_queue_intr_enable = fs_rx_intr_enable,
1135         .rx_queue_intr_disable = fs_rx_intr_disable,
1136         .flow_ctrl_get = fs_flow_ctrl_get,
1137         .flow_ctrl_set = fs_flow_ctrl_set,
1138         .mac_addr_remove = fs_mac_addr_remove,
1139         .mac_addr_add = fs_mac_addr_add,
1140         .mac_addr_set = fs_mac_addr_set,
1141         .rss_hash_update = fs_rss_hash_update,
1142         .filter_ctrl = fs_filter_ctrl,
1143 };