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