drivers/net/failsafe/failsafe_ops.c

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