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