drivers/net/virtio/virtio_user_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <sys/types.h>
   7 #include <unistd.h>
   8 #include <fcntl.h>
   9 #include <linux/major.h>
  10 #include <sys/stat.h>
  11 #include <sys/sysmacros.h>
  12 #include <sys/socket.h>
  13
  14 #include <rte_malloc.h>
  15 #include <rte_kvargs.h>
  16 #include <ethdev_vdev.h>
  17 #include <rte_bus_vdev.h>
  18 #include <rte_alarm.h>
  19 #include <rte_cycles.h>
  20
  21 #include "virtio_ethdev.h"
  22 #include "virtio_logs.h"
  23 #include "virtio.h"
  24 #include "virtqueue.h"
  25 #include "virtio_rxtx.h"
  26 #include "virtio_user/virtio_user_dev.h"
  27 #include "virtio_user/vhost.h"
  28
  29 #define virtio_user_get_dev(hwp) container_of(hwp, struct virtio_user_dev, hw)
  30
  31 static void
  32 virtio_user_reset_queues_packed(struct rte_eth_dev *eth_dev)
  33 {
  34         struct virtio_user_dev *dev = eth_dev->data->dev_private;
  35         struct virtio_hw *hw = &dev->hw;
  36         struct virtnet_rx *rxvq;
  37         struct virtnet_tx *txvq;
  38         uint16_t i;
  39
  40         /* Add lock to avoid queue contention. */
  41         rte_spinlock_lock(&hw->state_lock);
  42         hw->started = 0;
  43
  44         /*
  45          * Waitting for datapath to complete before resetting queues.
  46          * 1 ms should be enough for the ongoing Tx/Rx function to finish.
  47          */
  48         rte_delay_ms(1);
  49
  50         /* Vring reset for each Tx queue and Rx queue. */
  51         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
  52                 rxvq = eth_dev->data->rx_queues[i];
  53                 virtqueue_rxvq_reset_packed(rxvq->vq);
  54                 virtio_dev_rx_queue_setup_finish(eth_dev, i);
  55         }
  56
  57         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
  58                 txvq = eth_dev->data->tx_queues[i];
  59                 virtqueue_txvq_reset_packed(txvq->vq);
  60         }
  61
  62         hw->started = 1;
  63         rte_spinlock_unlock(&hw->state_lock);
  64 }
  65
  66
  67 static int
  68 virtio_user_server_reconnect(struct virtio_user_dev *dev)
  69 {
  70         int ret, connectfd, old_status;
  71         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
  72         struct virtio_hw *hw = &dev->hw;
  73         uint64_t protocol_features;
  74
  75         connectfd = accept(dev->listenfd, NULL, NULL);
  76         if (connectfd < 0)
  77                 return -1;
  78
  79         dev->vhostfd = connectfd;
  80         old_status = dev->status;
  81
  82         virtio_reset(hw);
  83
  84         virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
  85
  86         virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
  87
  88         if (dev->ops->get_features(dev, &dev->device_features) < 0) {
  89                 PMD_INIT_LOG(ERR, "get_features failed: %s",
  90                              strerror(errno));
  91                 return -1;
  92         }
  93
  94         if (dev->device_features &
  95                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
  96                 if (dev->ops->get_protocol_features(dev, &protocol_features))
  97                         return -1;
  98
  99                 /* Offer VHOST_USER_PROTOCOL_F_STATUS */
 100                 dev->protocol_features |= (1ULL << VHOST_USER_PROTOCOL_F_STATUS);
 101                 dev->protocol_features &= protocol_features;
 102
 103                 if (dev->ops->set_protocol_features(dev, dev->protocol_features))
 104                         return -1;
 105
 106                 if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
 107                         dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
 108         }
 109
 110         dev->device_features |= dev->frontend_features;
 111
 112         /* umask vhost-user unsupported features */
 113         dev->device_features &= ~(dev->unsupported_features);
 114
 115         dev->features &= dev->device_features;
 116
 117         /* For packed ring, resetting queues is required in reconnection. */
 118         if (virtio_with_packed_queue(hw) &&
 119            (old_status & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
 120                 PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
 121                                 " when packed ring reconnecting.");
 122                 virtio_user_reset_queues_packed(eth_dev);
 123         }
 124
 125         virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
 126
 127         /* Start the device */
 128         virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
 129         if (!dev->started)
 130                 return -1;
 131
 132         if (dev->queue_pairs > 1) {
 133                 ret = virtio_user_handle_mq(dev, dev->queue_pairs);
 134                 if (ret != 0) {
 135                         PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
 136                         return -1;
 137                 }
 138         }
 139         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
 140                 if (rte_intr_disable(eth_dev->intr_handle) < 0) {
 141                         PMD_DRV_LOG(ERR, "interrupt disable failed");
 142                         return -1;
 143                 }
 144                 rte_intr_callback_unregister(eth_dev->intr_handle,
 145                                              virtio_interrupt_handler,
 146                                              eth_dev);
 147                 eth_dev->intr_handle->fd = connectfd;
 148                 rte_intr_callback_register(eth_dev->intr_handle,
 149                                            virtio_interrupt_handler, eth_dev);
 150
 151                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
 152                         PMD_DRV_LOG(ERR, "interrupt enable failed");
 153                         return -1;
 154                 }
 155         }
 156         PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
 157         return 0;
 158 }
 159
 160 static void
 161 virtio_user_delayed_handler(void *param)
 162 {
 163         struct virtio_hw *hw = (struct virtio_hw *)param;
 164         struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
 165         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 166
 167         if (rte_intr_disable(eth_dev->intr_handle) < 0) {
 168                 PMD_DRV_LOG(ERR, "interrupt disable failed");
 169                 return;
 170         }
 171         rte_intr_callback_unregister(eth_dev->intr_handle,
 172                                      virtio_interrupt_handler, eth_dev);
 173         if (dev->is_server) {
 174                 if (dev->vhostfd >= 0) {
 175                         close(dev->vhostfd);
 176                         dev->vhostfd = -1;
 177                 }
 178                 eth_dev->intr_handle->fd = dev->listenfd;
 179                 rte_intr_callback_register(eth_dev->intr_handle,
 180                                            virtio_interrupt_handler, eth_dev);
 181                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
 182                         PMD_DRV_LOG(ERR, "interrupt enable failed");
 183                         return;
 184                 }
 185         }
 186 }
 187
 188 static void
 189 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
 190                      void *dst, int length)
 191 {
 192         int i;
 193         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 194
 195         if (offset == offsetof(struct virtio_net_config, mac) &&
 196             length == RTE_ETHER_ADDR_LEN) {
 197                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
 198                         ((uint8_t *)dst)[i] = dev->mac_addr[i];
 199                 return;
 200         }
 201
 202         if (offset == offsetof(struct virtio_net_config, status)) {
 203                 char buf[128];
 204
 205                 if (dev->vhostfd >= 0) {
 206                         int r;
 207                         int flags;
 208
 209                         flags = fcntl(dev->vhostfd, F_GETFL);
 210                         if (fcntl(dev->vhostfd, F_SETFL,
 211                                         flags | O_NONBLOCK) == -1) {
 212                                 PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
 213                                 return;
 214                         }
 215                         r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
 216                         if (r == 0 || (r < 0 && errno != EAGAIN)) {
 217                                 dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
 218                                 PMD_DRV_LOG(ERR, "virtio-user port %u is down",
 219                                             hw->port_id);
 220
 221                                 /* This function could be called in the process
 222                                  * of interrupt handling, callback cannot be
 223                                  * unregistered here, set an alarm to do it.
 224                                  */
 225                                 rte_eal_alarm_set(1,
 226                                                   virtio_user_delayed_handler,
 227                                                   (void *)hw);
 228                         } else {
 229                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
 230                         }
 231                         if (fcntl(dev->vhostfd, F_SETFL,
 232                                         flags & ~O_NONBLOCK) == -1) {
 233                                 PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
 234                                 return;
 235                         }
 236                 } else if (dev->is_server) {
 237                         dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
 238                         if (virtio_user_server_reconnect(dev) >= 0)
 239                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
 240                 }
 241
 242                 *(uint16_t *)dst = dev->net_status;
 243         }
 244
 245         if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
 246                 *(uint16_t *)dst = dev->max_queue_pairs;
 247 }
 248
 249 static void
 250 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
 251                       const void *src, int length)
 252 {
 253         int i;
 254         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 255
 256         if ((offset == offsetof(struct virtio_net_config, mac)) &&
 257             (length == RTE_ETHER_ADDR_LEN))
 258                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
 259                         dev->mac_addr[i] = ((const uint8_t *)src)[i];
 260         else
 261                 PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
 262                             offset, length);
 263 }
 264
 265 static void
 266 virtio_user_reset(struct virtio_hw *hw)
 267 {
 268         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 269
 270         if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
 271                 virtio_user_stop_device(dev);
 272 }
 273
 274 static void
 275 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
 276 {
 277         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 278         uint8_t old_status = dev->status;
 279
 280         if (status & VIRTIO_CONFIG_STATUS_FEATURES_OK &&
 281                         ~old_status & VIRTIO_CONFIG_STATUS_FEATURES_OK)
 282                 virtio_user_dev_set_features(dev);
 283         if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
 284                 virtio_user_start_device(dev);
 285         else if (status == VIRTIO_CONFIG_STATUS_RESET)
 286                 virtio_user_reset(hw);
 287
 288         virtio_user_dev_set_status(dev, status);
 289 }
 290
 291 static uint8_t
 292 virtio_user_get_status(struct virtio_hw *hw)
 293 {
 294         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 295
 296         virtio_user_dev_update_status(dev);
 297
 298         return dev->status;
 299 }
 300
 301 static uint64_t
 302 virtio_user_get_features(struct virtio_hw *hw)
 303 {
 304         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 305
 306         /* unmask feature bits defined in vhost user protocol */
 307         return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
 308 }
 309
 310 static void
 311 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
 312 {
 313         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 314
 315         dev->features = features & dev->device_features;
 316 }
 317
 318 static int
 319 virtio_user_features_ok(struct virtio_hw *hw __rte_unused)
 320 {
 321         return 0;
 322 }
 323
 324 static uint8_t
 325 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
 326 {
 327         /* rxq interrupts and config interrupt are separated in virtio-user,
 328          * here we only report config change.
 329          */
 330         return VIRTIO_ISR_CONFIG;
 331 }
 332
 333 static uint16_t
 334 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
 335                     uint16_t vec __rte_unused)
 336 {
 337         return 0;
 338 }
 339
 340 static uint16_t
 341 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
 342                           struct virtqueue *vq __rte_unused,
 343                           uint16_t vec)
 344 {
 345         /* pretend we have done that */
 346         return vec;
 347 }
 348
 349 /* This function is to get the queue size, aka, number of descs, of a specified
 350  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
 351  * max supported queues.
 352  */
 353 static uint16_t
 354 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
 355 {
 356         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 357
 358         /* Currently, each queue has same queue size */
 359         return dev->queue_size;
 360 }
 361
 362 static void
 363 virtio_user_setup_queue_packed(struct virtqueue *vq,
 364                                struct virtio_user_dev *dev)
 365 {
 366         uint16_t queue_idx = vq->vq_queue_index;
 367         struct vring_packed *vring;
 368         uint64_t desc_addr;
 369         uint64_t avail_addr;
 370         uint64_t used_addr;
 371         uint16_t i;
 372
 373         vring  = &dev->packed_vrings[queue_idx];
 374         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
 375         avail_addr = desc_addr + vq->vq_nentries *
 376                 sizeof(struct vring_packed_desc);
 377         used_addr = RTE_ALIGN_CEIL(avail_addr +
 378                            sizeof(struct vring_packed_desc_event),
 379                            VIRTIO_VRING_ALIGN);
 380         vring->num = vq->vq_nentries;
 381         vring->desc = (void *)(uintptr_t)desc_addr;
 382         vring->driver = (void *)(uintptr_t)avail_addr;
 383         vring->device = (void *)(uintptr_t)used_addr;
 384         dev->packed_queues[queue_idx].avail_wrap_counter = true;
 385         dev->packed_queues[queue_idx].used_wrap_counter = true;
 386
 387         for (i = 0; i < vring->num; i++)
 388                 vring->desc[i].flags = 0;
 389 }
 390
 391 static void
 392 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
 393 {
 394         uint16_t queue_idx = vq->vq_queue_index;
 395         uint64_t desc_addr, avail_addr, used_addr;
 396
 397         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
 398         avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
 399         used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
 400                                                          ring[vq->vq_nentries]),
 401                                    VIRTIO_VRING_ALIGN);
 402
 403         dev->vrings[queue_idx].num = vq->vq_nentries;
 404         dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
 405         dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
 406         dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
 407 }
 408
 409 static int
 410 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
 411 {
 412         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 413
 414         if (virtio_with_packed_queue(hw))
 415                 virtio_user_setup_queue_packed(vq, dev);
 416         else
 417                 virtio_user_setup_queue_split(vq, dev);
 418
 419         return 0;
 420 }
 421
 422 static void
 423 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
 424 {
 425         /* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
 426          * correspondingly stops the ioeventfds, and reset the status of
 427          * the device.
 428          * For modern devices, set queue desc, avail, used in PCI bar to 0,
 429          * not see any more behavior in QEMU.
 430          *
 431          * Here we just care about what information to deliver to vhost-user
 432          * or vhost-kernel. So we just close ioeventfd for now.
 433          */
 434         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 435
 436         close(dev->callfds[vq->vq_queue_index]);
 437         close(dev->kickfds[vq->vq_queue_index]);
 438 }
 439
 440 static void
 441 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
 442 {
 443         uint64_t buf = 1;
 444         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 445
 446         if (hw->cvq && (hw->cvq->vq == vq)) {
 447                 if (virtio_with_packed_queue(vq->hw))
 448                         virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
 449                 else
 450                         virtio_user_handle_cq(dev, vq->vq_queue_index);
 451                 return;
 452         }
 453
 454         if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
 455                 PMD_DRV_LOG(ERR, "failed to kick backend: %s",
 456                             strerror(errno));
 457 }
 458
 459 static int
 460 virtio_user_dev_close(struct virtio_hw *hw)
 461 {
 462         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 463
 464         virtio_user_dev_uninit(dev);
 465
 466         return 0;
 467 }
 468
 469 const struct virtio_ops virtio_user_ops = {
 470         .read_dev_cfg   = virtio_user_read_dev_config,
 471         .write_dev_cfg  = virtio_user_write_dev_config,
 472         .get_status     = virtio_user_get_status,
 473         .set_status     = virtio_user_set_status,
 474         .get_features   = virtio_user_get_features,
 475         .set_features   = virtio_user_set_features,
 476         .features_ok    = virtio_user_features_ok,
 477         .get_isr        = virtio_user_get_isr,
 478         .set_config_irq = virtio_user_set_config_irq,
 479         .set_queue_irq  = virtio_user_set_queue_irq,
 480         .get_queue_num  = virtio_user_get_queue_num,
 481         .setup_queue    = virtio_user_setup_queue,
 482         .del_queue      = virtio_user_del_queue,
 483         .notify_queue   = virtio_user_notify_queue,
 484         .dev_close      = virtio_user_dev_close,
 485 };
 486
 487 static const char *valid_args[] = {
 488 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
 489         VIRTIO_USER_ARG_QUEUES_NUM,
 490 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
 491         VIRTIO_USER_ARG_CQ_NUM,
 492 #define VIRTIO_USER_ARG_MAC            "mac"
 493         VIRTIO_USER_ARG_MAC,
 494 #define VIRTIO_USER_ARG_PATH           "path"
 495         VIRTIO_USER_ARG_PATH,
 496 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
 497         VIRTIO_USER_ARG_QUEUE_SIZE,
 498 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
 499         VIRTIO_USER_ARG_INTERFACE_NAME,
 500 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
 501         VIRTIO_USER_ARG_SERVER_MODE,
 502 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
 503         VIRTIO_USER_ARG_MRG_RXBUF,
 504 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
 505         VIRTIO_USER_ARG_IN_ORDER,
 506 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
 507         VIRTIO_USER_ARG_PACKED_VQ,
 508 #define VIRTIO_USER_ARG_SPEED          "speed"
 509         VIRTIO_USER_ARG_SPEED,
 510 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
 511         VIRTIO_USER_ARG_VECTORIZED,
 512         NULL
 513 };
 514
 515 #define VIRTIO_USER_DEF_CQ_EN   0
 516 #define VIRTIO_USER_DEF_Q_NUM   1
 517 #define VIRTIO_USER_DEF_Q_SZ    256
 518 #define VIRTIO_USER_DEF_SERVER_MODE     0
 519
 520 static int
 521 get_string_arg(const char *key __rte_unused,
 522                const char *value, void *extra_args)
 523 {
 524         if (!value || !extra_args)
 525                 return -EINVAL;
 526
 527         *(char **)extra_args = strdup(value);
 528
 529         if (!*(char **)extra_args)
 530                 return -ENOMEM;
 531
 532         return 0;
 533 }
 534
 535 static int
 536 get_integer_arg(const char *key __rte_unused,
 537                 const char *value, void *extra_args)
 538 {
 539         uint64_t integer = 0;
 540         if (!value || !extra_args)
 541                 return -EINVAL;
 542         errno = 0;
 543         integer = strtoull(value, NULL, 0);
 544         /* extra_args keeps default value, it should be replaced
 545          * only in case of successful parsing of the 'value' arg
 546          */
 547         if (errno == 0)
 548                 *(uint64_t *)extra_args = integer;
 549         return -errno;
 550 }
 551
 552 static uint32_t
 553 vdpa_dynamic_major_num(void)
 554 {
 555         FILE *fp;
 556         char *line = NULL;
 557         size_t size;
 558         char name[11];
 559         bool found = false;
 560         uint32_t num;
 561
 562         fp = fopen("/proc/devices", "r");
 563         if (fp == NULL) {
 564                 PMD_INIT_LOG(ERR, "Cannot open /proc/devices: %s",
 565                              strerror(errno));
 566                 return UNNAMED_MAJOR;
 567         }
 568
 569         while (getline(&line, &size, fp) > 0) {
 570                 char *stripped = line + strspn(line, " ");
 571                 if ((sscanf(stripped, "%u %10s", &num, name) == 2) &&
 572                     (strncmp(name, "vhost-vdpa", 10) == 0)) {
 573                         found = true;
 574                         break;
 575                 }
 576         }
 577         fclose(fp);
 578         return found ? num : UNNAMED_MAJOR;
 579 }
 580
 581 static enum virtio_user_backend_type
 582 virtio_user_backend_type(const char *path)
 583 {
 584         struct stat sb;
 585
 586         if (stat(path, &sb) == -1) {
 587                 if (errno == ENOENT)
 588                         return VIRTIO_USER_BACKEND_VHOST_USER;
 589
 590                 PMD_INIT_LOG(ERR, "Stat fails: %s (%s)\n", path,
 591                              strerror(errno));
 592                 return VIRTIO_USER_BACKEND_UNKNOWN;
 593         }
 594
 595         if (S_ISSOCK(sb.st_mode)) {
 596                 return VIRTIO_USER_BACKEND_VHOST_USER;
 597         } else if (S_ISCHR(sb.st_mode)) {
 598                 if (major(sb.st_rdev) == MISC_MAJOR)
 599                         return VIRTIO_USER_BACKEND_VHOST_KERNEL;
 600                 if (major(sb.st_rdev) == vdpa_dynamic_major_num())
 601                         return VIRTIO_USER_BACKEND_VHOST_VDPA;
 602         }
 603         return VIRTIO_USER_BACKEND_UNKNOWN;
 604 }
 605
 606 static struct rte_eth_dev *
 607 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
 608 {
 609         struct rte_eth_dev *eth_dev;
 610         struct rte_eth_dev_data *data;
 611         struct virtio_hw *hw;
 612         struct virtio_user_dev *dev;
 613
 614         eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*dev));
 615         if (!eth_dev) {
 616                 PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
 617                 return NULL;
 618         }
 619
 620         data = eth_dev->data;
 621         dev = eth_dev->data->dev_private;
 622         hw = &dev->hw;
 623
 624         hw->port_id = data->port_id;
 625         dev->port_id = data->port_id;
 626         VIRTIO_OPS(hw) = &virtio_user_ops;
 627
 628         hw->intr_lsc = 1;
 629         hw->use_vec_rx = 0;
 630         hw->use_vec_tx = 0;
 631         hw->use_inorder_rx = 0;
 632         hw->use_inorder_tx = 0;
 633
 634         return eth_dev;
 635 }
 636
 637 static void
 638 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
 639 {
 640         rte_eth_dev_release_port(eth_dev);
 641 }
 642
 643 /* Dev initialization routine. Invoked once for each virtio vdev at
 644  * EAL init time, see rte_bus_probe().
 645  * Returns 0 on success.
 646  */
 647 static int
 648 virtio_user_pmd_probe(struct rte_vdev_device *vdev)
 649 {
 650         struct rte_kvargs *kvlist = NULL;
 651         struct rte_eth_dev *eth_dev;
 652         struct virtio_hw *hw;
 653         struct virtio_user_dev *dev;
 654         enum virtio_user_backend_type backend_type = VIRTIO_USER_BACKEND_UNKNOWN;
 655         uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
 656         uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
 657         uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
 658         uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
 659         uint64_t mrg_rxbuf = 1;
 660         uint64_t in_order = 1;
 661         uint64_t packed_vq = 0;
 662         uint64_t vectorized = 0;
 663         char *path = NULL;
 664         char *ifname = NULL;
 665         char *mac_addr = NULL;
 666         int ret = -1;
 667
 668         RTE_BUILD_BUG_ON(offsetof(struct virtio_user_dev, hw) != 0);
 669
 670         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 671                 const char *name = rte_vdev_device_name(vdev);
 672                 eth_dev = rte_eth_dev_attach_secondary(name);
 673                 if (!eth_dev) {
 674                         PMD_INIT_LOG(ERR, "Failed to probe %s", name);
 675                         return -1;
 676                 }
 677
 678                 dev = eth_dev->data->dev_private;
 679                 hw = &dev->hw;
 680                 VIRTIO_OPS(hw) = &virtio_user_ops;
 681
 682                 if (eth_virtio_dev_init(eth_dev) < 0) {
 683                         PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
 684                         rte_eth_dev_release_port(eth_dev);
 685                         return -1;
 686                 }
 687
 688                 eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
 689                 eth_dev->device = &vdev->device;
 690                 rte_eth_dev_probing_finish(eth_dev);
 691                 return 0;
 692         }
 693
 694         kvlist = rte_kvargs_parse(rte_vdev_device_args(vdev), valid_args);
 695         if (!kvlist) {
 696                 PMD_INIT_LOG(ERR, "error when parsing param");
 697                 goto end;
 698         }
 699
 700         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
 701                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
 702                                        &get_string_arg, &path) < 0) {
 703                         PMD_INIT_LOG(ERR, "error to parse %s",
 704                                      VIRTIO_USER_ARG_PATH);
 705                         goto end;
 706                 }
 707         } else {
 708                 PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
 709                              VIRTIO_USER_ARG_PATH);
 710                 goto end;
 711         }
 712
 713         backend_type = virtio_user_backend_type(path);
 714         if (backend_type == VIRTIO_USER_BACKEND_UNKNOWN) {
 715                 PMD_INIT_LOG(ERR,
 716                              "unable to determine backend type for path %s",
 717                         path);
 718                 goto end;
 719         }
 720         PMD_INIT_LOG(INFO, "Backend type detected: %s",
 721                      virtio_user_backend_strings[backend_type]);
 722
 723         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
 724                 if (backend_type != VIRTIO_USER_BACKEND_VHOST_KERNEL) {
 725                         PMD_INIT_LOG(ERR,
 726                                 "arg %s applies only to vhost-kernel backend",
 727                                 VIRTIO_USER_ARG_INTERFACE_NAME);
 728                         goto end;
 729                 }
 730
 731                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
 732                                        &get_string_arg, &ifname) < 0) {
 733                         PMD_INIT_LOG(ERR, "error to parse %s",
 734                                      VIRTIO_USER_ARG_INTERFACE_NAME);
 735                         goto end;
 736                 }
 737         }
 738
 739         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
 740                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
 741                                        &get_string_arg, &mac_addr) < 0) {
 742                         PMD_INIT_LOG(ERR, "error to parse %s",
 743                                      VIRTIO_USER_ARG_MAC);
 744                         goto end;
 745                 }
 746         }
 747
 748         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
 749                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
 750                                        &get_integer_arg, &queue_size) < 0) {
 751                         PMD_INIT_LOG(ERR, "error to parse %s",
 752                                      VIRTIO_USER_ARG_QUEUE_SIZE);
 753                         goto end;
 754                 }
 755         }
 756
 757         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
 758                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
 759                                        &get_integer_arg, &queues) < 0) {
 760                         PMD_INIT_LOG(ERR, "error to parse %s",
 761                                      VIRTIO_USER_ARG_QUEUES_NUM);
 762                         goto end;
 763                 }
 764         }
 765
 766         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
 767                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
 768                                        &get_integer_arg, &server_mode) < 0) {
 769                         PMD_INIT_LOG(ERR, "error to parse %s",
 770                                      VIRTIO_USER_ARG_SERVER_MODE);
 771                         goto end;
 772                 }
 773         }
 774
 775         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
 776                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
 777                                        &get_integer_arg, &cq) < 0) {
 778                         PMD_INIT_LOG(ERR, "error to parse %s",
 779                                      VIRTIO_USER_ARG_CQ_NUM);
 780                         goto end;
 781                 }
 782         } else if (queues > 1) {
 783                 cq = 1;
 784         }
 785
 786         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
 787                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
 788                                        &get_integer_arg, &packed_vq) < 0) {
 789                         PMD_INIT_LOG(ERR, "error to parse %s",
 790                                      VIRTIO_USER_ARG_PACKED_VQ);
 791                         goto end;
 792                 }
 793         }
 794
 795         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
 796                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
 797                                        &get_integer_arg, &vectorized) < 0) {
 798                         PMD_INIT_LOG(ERR, "error to parse %s",
 799                                      VIRTIO_USER_ARG_VECTORIZED);
 800                         goto end;
 801                 }
 802         }
 803
 804         if (queues > 1 && cq == 0) {
 805                 PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
 806                 goto end;
 807         }
 808
 809         if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
 810                 PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
 811                         VIRTIO_USER_ARG_QUEUES_NUM, queues,
 812                         VIRTIO_MAX_VIRTQUEUE_PAIRS);
 813                 goto end;
 814         }
 815
 816         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
 817                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
 818                                        &get_integer_arg, &mrg_rxbuf) < 0) {
 819                         PMD_INIT_LOG(ERR, "error to parse %s",
 820                                      VIRTIO_USER_ARG_MRG_RXBUF);
 821                         goto end;
 822                 }
 823         }
 824
 825         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
 826                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
 827                                        &get_integer_arg, &in_order) < 0) {
 828                         PMD_INIT_LOG(ERR, "error to parse %s",
 829                                      VIRTIO_USER_ARG_IN_ORDER);
 830                         goto end;
 831                 }
 832         }
 833
 834         eth_dev = virtio_user_eth_dev_alloc(vdev);
 835         if (!eth_dev) {
 836                 PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
 837                 goto end;
 838         }
 839
 840         dev = eth_dev->data->dev_private;
 841         hw = &dev->hw;
 842         if (virtio_user_dev_init(dev, path, queues, cq,
 843                          queue_size, mac_addr, &ifname, server_mode,
 844                          mrg_rxbuf, in_order, packed_vq, backend_type) < 0) {
 845                 PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
 846                 virtio_user_eth_dev_free(eth_dev);
 847                 goto end;
 848         }
 849
 850         /* previously called by pci probing for physical dev */
 851         if (eth_virtio_dev_init(eth_dev) < 0) {
 852                 PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
 853                 virtio_user_eth_dev_free(eth_dev);
 854                 goto end;
 855         }
 856
 857         if (vectorized) {
 858                 if (packed_vq) {
 859 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
 860                         hw->use_vec_rx = 1;
 861                         hw->use_vec_tx = 1;
 862 #else
 863                         PMD_INIT_LOG(INFO,
 864                                 "building environment do not support packed ring vectorized");
 865 #endif
 866                 } else {
 867                         hw->use_vec_rx = 1;
 868                 }
 869         }
 870
 871         rte_eth_dev_probing_finish(eth_dev);
 872         ret = 0;
 873
 874 end:
 875         if (kvlist)
 876                 rte_kvargs_free(kvlist);
 877         if (path)
 878                 free(path);
 879         if (mac_addr)
 880                 free(mac_addr);
 881         if (ifname)
 882                 free(ifname);
 883         return ret;
 884 }
 885
 886 static int
 887 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
 888 {
 889         const char *name;
 890         struct rte_eth_dev *eth_dev;
 891
 892         if (!vdev)
 893                 return -EINVAL;
 894
 895         name = rte_vdev_device_name(vdev);
 896         PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
 897         eth_dev = rte_eth_dev_allocated(name);
 898         /* Port has already been released by close. */
 899         if (!eth_dev)
 900                 return 0;
 901
 902         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 903                 return rte_eth_dev_release_port(eth_dev);
 904
 905         /* make sure the device is stopped, queues freed */
 906         return rte_eth_dev_close(eth_dev->data->port_id);
 907 }
 908
 909 static int virtio_user_pmd_dma_map(struct rte_vdev_device *vdev, void *addr,
 910                 uint64_t iova, size_t len)
 911 {
 912         const char *name;
 913         struct rte_eth_dev *eth_dev;
 914         struct virtio_user_dev *dev;
 915
 916         if (!vdev)
 917                 return -EINVAL;
 918
 919         name = rte_vdev_device_name(vdev);
 920         eth_dev = rte_eth_dev_allocated(name);
 921         /* Port has already been released by close. */
 922         if (!eth_dev)
 923                 return 0;
 924
 925         dev = eth_dev->data->dev_private;
 926
 927         if (dev->ops->dma_map)
 928                 return dev->ops->dma_map(dev, addr, iova, len);
 929
 930         return 0;
 931 }
 932
 933 static int virtio_user_pmd_dma_unmap(struct rte_vdev_device *vdev, void *addr,
 934                 uint64_t iova, size_t len)
 935 {
 936         const char *name;
 937         struct rte_eth_dev *eth_dev;
 938         struct virtio_user_dev *dev;
 939
 940         if (!vdev)
 941                 return -EINVAL;
 942
 943         name = rte_vdev_device_name(vdev);
 944         eth_dev = rte_eth_dev_allocated(name);
 945         /* Port has already been released by close. */
 946         if (!eth_dev)
 947                 return 0;
 948
 949         dev = eth_dev->data->dev_private;
 950
 951         if (dev->ops->dma_unmap)
 952                 return dev->ops->dma_unmap(dev, addr, iova, len);
 953
 954         return 0;
 955 }
 956
 957 static struct rte_vdev_driver virtio_user_driver = {
 958         .probe = virtio_user_pmd_probe,
 959         .remove = virtio_user_pmd_remove,
 960         .dma_map = virtio_user_pmd_dma_map,
 961         .dma_unmap = virtio_user_pmd_dma_unmap,
 962         .drv_flags = RTE_VDEV_DRV_NEED_IOVA_AS_VA,
 963 };
 964
 965 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
 966 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
 967 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
 968         "path=<path> "
 969         "mac=<mac addr> "
 970         "cq=<int> "
 971         "queue_size=<int> "
 972         "queues=<int> "
 973         "iface=<string> "
 974         "server=<0|1> "
 975         "mrg_rxbuf=<0|1> "
 976         "in_order=<0|1> "
 977         "packed_vq=<0|1> "
 978         "speed=<int> "
 979         "vectorized=<0|1>");