hw/virtio-pci.c

   1 /*
   2  * Virtio PCI Bindings
   3  *
   4  * Copyright IBM, Corp. 2007
   5  * Copyright (c) 2009 CodeSourcery
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Paul Brook        <paul@codesourcery.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include <inttypes.h>
  17
  18 #include "virtio.h"
  19 #include "virtio-blk.h"
  20 #include "virtio-net.h"
  21 #include "virtio-serial.h"
  22 #include "pci.h"
  23 #include "qemu-error.h"
  24 #include "msix.h"
  25 #include "net.h"
  26 #include "loader.h"
  27 #include "kvm.h"
  28 #include "blockdev.h"
  29 #include "virtio-pci.h"
  30
  31 /* from Linux's linux/virtio_pci.h */
  32
  33 /* A 32-bit r/o bitmask of the features supported by the host */
  34 #define VIRTIO_PCI_HOST_FEATURES        0
  35
  36 /* A 32-bit r/w bitmask of features activated by the guest */
  37 #define VIRTIO_PCI_GUEST_FEATURES       4
  38
  39 /* A 32-bit r/w PFN for the currently selected queue */
  40 #define VIRTIO_PCI_QUEUE_PFN            8
  41
  42 /* A 16-bit r/o queue size for the currently selected queue */
  43 #define VIRTIO_PCI_QUEUE_NUM            12
  44
  45 /* A 16-bit r/w queue selector */
  46 #define VIRTIO_PCI_QUEUE_SEL            14
  47
  48 /* A 16-bit r/w queue notifier */
  49 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  50
  51 /* An 8-bit device status register.  */
  52 #define VIRTIO_PCI_STATUS               18
  53
  54 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  55  * current contents of the ISR and will also clear it.  This is effectively
  56  * a read-and-acknowledge. */
  57 #define VIRTIO_PCI_ISR                  19
  58
  59 /* MSI-X registers: only enabled if MSI-X is enabled. */
  60 /* A 16-bit vector for configuration changes. */
  61 #define VIRTIO_MSI_CONFIG_VECTOR        20
  62 /* A 16-bit vector for selected queue notifications. */
  63 #define VIRTIO_MSI_QUEUE_VECTOR         22
  64
  65 /* Config space size */
  66 #define VIRTIO_PCI_CONFIG_NOMSI         20
  67 #define VIRTIO_PCI_CONFIG_MSI           24
  68 #define VIRTIO_PCI_REGION_SIZE(dev)     (msix_present(dev) ? \
  69                                          VIRTIO_PCI_CONFIG_MSI : \
  70                                          VIRTIO_PCI_CONFIG_NOMSI)
  71
  72 /* The remaining space is defined by each driver as the per-driver
  73  * configuration space */
  74 #define VIRTIO_PCI_CONFIG(dev)          (msix_enabled(dev) ? \
  75                                          VIRTIO_PCI_CONFIG_MSI : \
  76                                          VIRTIO_PCI_CONFIG_NOMSI)
  77
  78 /* Virtio ABI version, if we increment this, we break the guest driver. */
  79 #define VIRTIO_PCI_ABI_VERSION          0
  80
  81 /* How many bits to shift physical queue address written to QUEUE_PFN.
  82  * 12 is historical, and due to x86 page size. */
  83 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  84
  85 /* Flags track per-device state like workarounds for quirks in older guests. */
  86 #define VIRTIO_PCI_FLAG_BUS_MASTER_BUG  (1 << 0)
  87
  88 /* Performance improves when virtqueue kick processing is decoupled from the
  89  * vcpu thread using ioeventfd for some devices. */
  90 #define VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT 1
  91 #define VIRTIO_PCI_FLAG_USE_IOEVENTFD   (1 << VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT)
  92
  93 /* QEMU doesn't strictly need write barriers since everything runs in
  94  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  95  * KVM or if kqemu gets SMP support.
  96  */
  97 #define wmb() do { } while (0)
  98
  99 /* virtio device */
 100
 101 static void virtio_pci_notify(void *opaque, uint16_t vector)
 102 {
 103     VirtIOPCIProxy *proxy = opaque;
 104     if (msix_enabled(&proxy->pci_dev))
 105         msix_notify(&proxy->pci_dev, vector);
 106     else
 107         qemu_set_irq(proxy->pci_dev.irq[0], proxy->vdev->isr & 1);
 108 }
 109
 110 static void virtio_pci_save_config(void * opaque, QEMUFile *f)
 111 {
 112     VirtIOPCIProxy *proxy = opaque;
 113     pci_device_save(&proxy->pci_dev, f);
 114     msix_save(&proxy->pci_dev, f);
 115     if (msix_present(&proxy->pci_dev))
 116         qemu_put_be16(f, proxy->vdev->config_vector);
 117 }
 118
 119 static void virtio_pci_save_queue(void * opaque, int n, QEMUFile *f)
 120 {
 121     VirtIOPCIProxy *proxy = opaque;
 122     if (msix_present(&proxy->pci_dev))
 123         qemu_put_be16(f, virtio_queue_vector(proxy->vdev, n));
 124 }
 125
 126 static int virtio_pci_load_config(void * opaque, QEMUFile *f)
 127 {
 128     VirtIOPCIProxy *proxy = opaque;
 129     int ret;
 130     ret = pci_device_load(&proxy->pci_dev, f);
 131     if (ret) {
 132         return ret;
 133     }
 134     msix_load(&proxy->pci_dev, f);
 135     if (msix_present(&proxy->pci_dev)) {
 136         qemu_get_be16s(f, &proxy->vdev->config_vector);
 137     } else {
 138         proxy->vdev->config_vector = VIRTIO_NO_VECTOR;
 139     }
 140     if (proxy->vdev->config_vector != VIRTIO_NO_VECTOR) {
 141         return msix_vector_use(&proxy->pci_dev, proxy->vdev->config_vector);
 142     }
 143     return 0;
 144 }
 145
 146 static int virtio_pci_load_queue(void * opaque, int n, QEMUFile *f)
 147 {
 148     VirtIOPCIProxy *proxy = opaque;
 149     uint16_t vector;
 150     if (msix_present(&proxy->pci_dev)) {
 151         qemu_get_be16s(f, &vector);
 152     } else {
 153         vector = VIRTIO_NO_VECTOR;
 154     }
 155     virtio_queue_set_vector(proxy->vdev, n, vector);
 156     if (vector != VIRTIO_NO_VECTOR) {
 157         return msix_vector_use(&proxy->pci_dev, vector);
 158     }
 159     return 0;
 160 }
 161
 162 static int virtio_pci_set_host_notifier_internal(VirtIOPCIProxy *proxy,
 163                                                  int n, bool assign)
 164 {
 165     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 166     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 167     int r;
 168     if (assign) {
 169         r = event_notifier_init(notifier, 1);
 170         if (r < 0) {
 171             error_report("%s: unable to init event notifier: %d",
 172                          __func__, r);
 173             return r;
 174         }
 175         r = kvm_set_ioeventfd_pio_word(event_notifier_get_fd(notifier),
 176                                        proxy->addr + VIRTIO_PCI_QUEUE_NOTIFY,
 177                                        n, assign);
 178         if (r < 0) {
 179             error_report("%s: unable to map ioeventfd: %d",
 180                          __func__, r);
 181             event_notifier_cleanup(notifier);
 182         }
 183     } else {
 184         r = kvm_set_ioeventfd_pio_word(event_notifier_get_fd(notifier),
 185                                        proxy->addr + VIRTIO_PCI_QUEUE_NOTIFY,
 186                                        n, assign);
 187         if (r < 0) {
 188             error_report("%s: unable to unmap ioeventfd: %d",
 189                          __func__, r);
 190             return r;
 191         }
 192
 193         /* Handle the race condition where the guest kicked and we deassigned
 194          * before we got around to handling the kick.
 195          */
 196         if (event_notifier_test_and_clear(notifier)) {
 197             virtio_queue_notify_vq(vq);
 198         }
 199
 200         event_notifier_cleanup(notifier);
 201     }
 202     return r;
 203 }
 204
 205 static void virtio_pci_host_notifier_read(void *opaque)
 206 {
 207     VirtQueue *vq = opaque;
 208     EventNotifier *n = virtio_queue_get_host_notifier(vq);
 209     if (event_notifier_test_and_clear(n)) {
 210         virtio_queue_notify_vq(vq);
 211     }
 212 }
 213
 214 static void virtio_pci_set_host_notifier_fd_handler(VirtIOPCIProxy *proxy,
 215                                                     int n, bool assign)
 216 {
 217     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 218     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 219     if (assign) {
 220         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 221                             virtio_pci_host_notifier_read, NULL, vq);
 222     } else {
 223         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 224                             NULL, NULL, NULL);
 225     }
 226 }
 227
 228 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 229 {
 230     int n, r;
 231
 232     if (!(proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) ||
 233         proxy->ioeventfd_disabled ||
 234         proxy->ioeventfd_started) {
 235         return;
 236     }
 237
 238     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 239         if (!virtio_queue_get_num(proxy->vdev, n)) {
 240             continue;
 241         }
 242
 243         r = virtio_pci_set_host_notifier_internal(proxy, n, true);
 244         if (r < 0) {
 245             goto assign_error;
 246         }
 247
 248         virtio_pci_set_host_notifier_fd_handler(proxy, n, true);
 249     }
 250     proxy->ioeventfd_started = true;
 251     return;
 252
 253 assign_error:
 254     while (--n >= 0) {
 255         if (!virtio_queue_get_num(proxy->vdev, n)) {
 256             continue;
 257         }
 258
 259         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 260         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 261         assert(r >= 0);
 262     }
 263     proxy->ioeventfd_started = false;
 264     error_report("%s: failed. Fallback to a userspace (slower).", __func__);
 265 }
 266
 267 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 268 {
 269     int r;
 270     int n;
 271
 272     if (!proxy->ioeventfd_started) {
 273         return;
 274     }
 275
 276     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 277         if (!virtio_queue_get_num(proxy->vdev, n)) {
 278             continue;
 279         }
 280
 281         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 282         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 283         assert(r >= 0);
 284     }
 285     proxy->ioeventfd_started = false;
 286 }
 287
 288 static void virtio_pci_reset(DeviceState *d)
 289 {
 290     VirtIOPCIProxy *proxy = container_of(d, VirtIOPCIProxy, pci_dev.qdev);
 291     virtio_pci_stop_ioeventfd(proxy);
 292     virtio_reset(proxy->vdev);
 293     msix_reset(&proxy->pci_dev);
 294     proxy->flags &= ~VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 295 }
 296
 297 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 298 {
 299     VirtIOPCIProxy *proxy = opaque;
 300     VirtIODevice *vdev = proxy->vdev;
 301     target_phys_addr_t pa;
 302
 303     switch (addr) {
 304     case VIRTIO_PCI_GUEST_FEATURES:
 305         /* Guest does not negotiate properly?  We have to assume nothing. */
 306         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 307             if (vdev->bad_features)
 308                 val = proxy->host_features & vdev->bad_features(vdev);
 309             else
 310                 val = 0;
 311         }
 312         if (vdev->set_features)
 313             vdev->set_features(vdev, val);
 314         vdev->guest_features = val;
 315         break;
 316     case VIRTIO_PCI_QUEUE_PFN:
 317         pa = (target_phys_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 318         if (pa == 0) {
 319             virtio_pci_stop_ioeventfd(proxy);
 320             virtio_reset(proxy->vdev);
 321             msix_unuse_all_vectors(&proxy->pci_dev);
 322         }
 323         else
 324             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 325         break;
 326     case VIRTIO_PCI_QUEUE_SEL:
 327         if (val < VIRTIO_PCI_QUEUE_MAX)
 328             vdev->queue_sel = val;
 329         break;
 330     case VIRTIO_PCI_QUEUE_NOTIFY:
 331         virtio_queue_notify(vdev, val);
 332         break;
 333     case VIRTIO_PCI_STATUS:
 334         if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
 335             virtio_pci_stop_ioeventfd(proxy);
 336         }
 337
 338         virtio_set_status(vdev, val & 0xFF);
 339
 340         if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
 341             virtio_pci_start_ioeventfd(proxy);
 342         }
 343
 344         if (vdev->status == 0) {
 345             virtio_reset(proxy->vdev);
 346             msix_unuse_all_vectors(&proxy->pci_dev);
 347         }
 348
 349         /* Linux before 2.6.34 sets the device as OK without enabling
 350            the PCI device bus master bit. In this case we need to disable
 351            some safety checks. */
 352         if ((val & VIRTIO_CONFIG_S_DRIVER_OK) &&
 353             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 354             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 355         }
 356         break;
 357     case VIRTIO_MSI_CONFIG_VECTOR:
 358         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 359         /* Make it possible for guest to discover an error took place. */
 360         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 361             val = VIRTIO_NO_VECTOR;
 362         vdev->config_vector = val;
 363         break;
 364     case VIRTIO_MSI_QUEUE_VECTOR:
 365         msix_vector_unuse(&proxy->pci_dev,
 366                           virtio_queue_vector(vdev, vdev->queue_sel));
 367         /* Make it possible for guest to discover an error took place. */
 368         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 369             val = VIRTIO_NO_VECTOR;
 370         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 371         break;
 372     default:
 373         error_report("%s: unexpected address 0x%x value 0x%x",
 374                      __func__, addr, val);
 375         break;
 376     }
 377 }
 378
 379 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 380 {
 381     VirtIODevice *vdev = proxy->vdev;
 382     uint32_t ret = 0xFFFFFFFF;
 383
 384     switch (addr) {
 385     case VIRTIO_PCI_HOST_FEATURES:
 386         ret = proxy->host_features;
 387         break;
 388     case VIRTIO_PCI_GUEST_FEATURES:
 389         ret = vdev->guest_features;
 390         break;
 391     case VIRTIO_PCI_QUEUE_PFN:
 392         ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 393               >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 394         break;
 395     case VIRTIO_PCI_QUEUE_NUM:
 396         ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 397         break;
 398     case VIRTIO_PCI_QUEUE_SEL:
 399         ret = vdev->queue_sel;
 400         break;
 401     case VIRTIO_PCI_STATUS:
 402         ret = vdev->status;
 403         break;
 404     case VIRTIO_PCI_ISR:
 405         /* reading from the ISR also clears it. */
 406         ret = vdev->isr;
 407         vdev->isr = 0;
 408         qemu_set_irq(proxy->pci_dev.irq[0], 0);
 409         break;
 410     case VIRTIO_MSI_CONFIG_VECTOR:
 411         ret = vdev->config_vector;
 412         break;
 413     case VIRTIO_MSI_QUEUE_VECTOR:
 414         ret = virtio_queue_vector(vdev, vdev->queue_sel);
 415         break;
 416     default:
 417         break;
 418     }
 419
 420     return ret;
 421 }
 422
 423 static uint32_t virtio_pci_config_readb(void *opaque, uint32_t addr)
 424 {
 425     VirtIOPCIProxy *proxy = opaque;
 426     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 427     addr -= proxy->addr;
 428     if (addr < config)
 429         return virtio_ioport_read(proxy, addr);
 430     addr -= config;
 431     return virtio_config_readb(proxy->vdev, addr);
 432 }
 433
 434 static uint32_t virtio_pci_config_readw(void *opaque, uint32_t addr)
 435 {
 436     VirtIOPCIProxy *proxy = opaque;
 437     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 438     addr -= proxy->addr;
 439     if (addr < config)
 440         return virtio_ioport_read(proxy, addr);
 441     addr -= config;
 442     return virtio_config_readw(proxy->vdev, addr);
 443 }
 444
 445 static uint32_t virtio_pci_config_readl(void *opaque, uint32_t addr)
 446 {
 447     VirtIOPCIProxy *proxy = opaque;
 448     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 449     addr -= proxy->addr;
 450     if (addr < config)
 451         return virtio_ioport_read(proxy, addr);
 452     addr -= config;
 453     return virtio_config_readl(proxy->vdev, addr);
 454 }
 455
 456 static void virtio_pci_config_writeb(void *opaque, uint32_t addr, uint32_t val)
 457 {
 458     VirtIOPCIProxy *proxy = opaque;
 459     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 460     addr -= proxy->addr;
 461     if (addr < config) {
 462         virtio_ioport_write(proxy, addr, val);
 463         return;
 464     }
 465     addr -= config;
 466     virtio_config_writeb(proxy->vdev, addr, val);
 467 }
 468
 469 static void virtio_pci_config_writew(void *opaque, uint32_t addr, uint32_t val)
 470 {
 471     VirtIOPCIProxy *proxy = opaque;
 472     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 473     addr -= proxy->addr;
 474     if (addr < config) {
 475         virtio_ioport_write(proxy, addr, val);
 476         return;
 477     }
 478     addr -= config;
 479     virtio_config_writew(proxy->vdev, addr, val);
 480 }
 481
 482 static void virtio_pci_config_writel(void *opaque, uint32_t addr, uint32_t val)
 483 {
 484     VirtIOPCIProxy *proxy = opaque;
 485     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 486     addr -= proxy->addr;
 487     if (addr < config) {
 488         virtio_ioport_write(proxy, addr, val);
 489         return;
 490     }
 491     addr -= config;
 492     virtio_config_writel(proxy->vdev, addr, val);
 493 }
 494
 495 static void virtio_map(PCIDevice *pci_dev, int region_num,
 496                        pcibus_t addr, pcibus_t size, int type)
 497 {
 498     VirtIOPCIProxy *proxy = container_of(pci_dev, VirtIOPCIProxy, pci_dev);
 499     VirtIODevice *vdev = proxy->vdev;
 500     unsigned config_len = VIRTIO_PCI_REGION_SIZE(pci_dev) + vdev->config_len;
 501
 502     proxy->addr = addr;
 503
 504     register_ioport_write(addr, config_len, 1, virtio_pci_config_writeb, proxy);
 505     register_ioport_write(addr, config_len, 2, virtio_pci_config_writew, proxy);
 506     register_ioport_write(addr, config_len, 4, virtio_pci_config_writel, proxy);
 507     register_ioport_read(addr, config_len, 1, virtio_pci_config_readb, proxy);
 508     register_ioport_read(addr, config_len, 2, virtio_pci_config_readw, proxy);
 509     register_ioport_read(addr, config_len, 4, virtio_pci_config_readl, proxy);
 510
 511     if (vdev->config_len)
 512         vdev->get_config(vdev, vdev->config);
 513 }
 514
 515 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 516                                 uint32_t val, int len)
 517 {
 518     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 519
 520     if (PCI_COMMAND == address) {
 521         if (!(val & PCI_COMMAND_MASTER)) {
 522             if (!(proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG)) {
 523                 virtio_pci_stop_ioeventfd(proxy);
 524                 virtio_set_status(proxy->vdev,
 525                                   proxy->vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 526             }
 527         }
 528     }
 529
 530     pci_default_write_config(pci_dev, address, val, len);
 531     msix_write_config(pci_dev, address, val, len);
 532 }
 533
 534 static unsigned virtio_pci_get_features(void *opaque)
 535 {
 536     VirtIOPCIProxy *proxy = opaque;
 537     return proxy->host_features;
 538 }
 539
 540 static void virtio_pci_guest_notifier_read(void *opaque)
 541 {
 542     VirtQueue *vq = opaque;
 543     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 544     if (event_notifier_test_and_clear(n)) {
 545         virtio_irq(vq);
 546     }
 547 }
 548
 549 static int virtio_pci_set_guest_notifier(void *opaque, int n, bool assign)
 550 {
 551     VirtIOPCIProxy *proxy = opaque;
 552     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 553     EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 554
 555     if (assign) {
 556         int r = event_notifier_init(notifier, 0);
 557         if (r < 0) {
 558             return r;
 559         }
 560         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 561                             virtio_pci_guest_notifier_read, NULL, vq);
 562     } else {
 563         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 564                             NULL, NULL, NULL);
 565         event_notifier_cleanup(notifier);
 566     }
 567
 568     return 0;
 569 }
 570
 571 static bool virtio_pci_query_guest_notifiers(void *opaque)
 572 {
 573     VirtIOPCIProxy *proxy = opaque;
 574     return msix_enabled(&proxy->pci_dev);
 575 }
 576
 577 static int virtio_pci_set_guest_notifiers(void *opaque, bool assign)
 578 {
 579     VirtIOPCIProxy *proxy = opaque;
 580     VirtIODevice *vdev = proxy->vdev;
 581     int r, n;
 582
 583     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 584         if (!virtio_queue_get_num(vdev, n)) {
 585             break;
 586         }
 587
 588         r = virtio_pci_set_guest_notifier(opaque, n, assign);
 589         if (r < 0) {
 590             goto assign_error;
 591         }
 592     }
 593
 594     return 0;
 595
 596 assign_error:
 597     /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
 598     while (--n >= 0) {
 599         virtio_pci_set_guest_notifier(opaque, n, !assign);
 600     }
 601     return r;
 602 }
 603
 604 static int virtio_pci_set_host_notifier(void *opaque, int n, bool assign)
 605 {
 606     VirtIOPCIProxy *proxy = opaque;
 607
 608     /* Stop using ioeventfd for virtqueue kick if the device starts using host
 609      * notifiers.  This makes it easy to avoid stepping on each others' toes.
 610      */
 611     proxy->ioeventfd_disabled = assign;
 612     if (assign) {
 613         virtio_pci_stop_ioeventfd(proxy);
 614     }
 615     /* We don't need to start here: it's not needed because backend
 616      * currently only stops on status change away from ok,
 617      * reset, vmstop and such. If we do add code to start here,
 618      * need to check vmstate, device state etc. */
 619     return virtio_pci_set_host_notifier_internal(proxy, n, assign);
 620 }
 621
 622 static void virtio_pci_vmstate_change(void *opaque, bool running)
 623 {
 624     VirtIOPCIProxy *proxy = opaque;
 625
 626     if (running) {
 627         /* Try to find out if the guest has bus master disabled, but is
 628            in ready state. Then we have a buggy guest OS. */
 629         if ((proxy->vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 630             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 631             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 632         }
 633         virtio_pci_start_ioeventfd(proxy);
 634     } else {
 635         virtio_pci_stop_ioeventfd(proxy);
 636     }
 637 }
 638
 639 static const VirtIOBindings virtio_pci_bindings = {
 640     .notify = virtio_pci_notify,
 641     .save_config = virtio_pci_save_config,
 642     .load_config = virtio_pci_load_config,
 643     .save_queue = virtio_pci_save_queue,
 644     .load_queue = virtio_pci_load_queue,
 645     .get_features = virtio_pci_get_features,
 646     .query_guest_notifiers = virtio_pci_query_guest_notifiers,
 647     .set_host_notifier = virtio_pci_set_host_notifier,
 648     .set_guest_notifiers = virtio_pci_set_guest_notifiers,
 649     .vmstate_change = virtio_pci_vmstate_change,
 650 };
 651
 652 void virtio_init_pci(VirtIOPCIProxy *proxy, VirtIODevice *vdev,
 653                             uint16_t vendor, uint16_t device,
 654                             uint16_t class_code, uint8_t pif)
 655 {
 656     uint8_t *config;
 657     uint32_t size;
 658
 659     proxy->vdev = vdev;
 660
 661     config = proxy->pci_dev.config;
 662     pci_config_set_vendor_id(config, vendor);
 663     pci_config_set_device_id(config, device);
 664
 665     config[0x08] = VIRTIO_PCI_ABI_VERSION;
 666
 667     config[0x09] = pif;
 668     pci_config_set_class(config, class_code);
 669
 670     config[0x2c] = vendor & 0xFF;
 671     config[0x2d] = (vendor >> 8) & 0xFF;
 672     config[0x2e] = vdev->device_id & 0xFF;
 673     config[0x2f] = (vdev->device_id >> 8) & 0xFF;
 674
 675     config[0x3d] = 1;
 676
 677     if (vdev->nvectors && !msix_init(&proxy->pci_dev, vdev->nvectors, 1, 0)) {
 678         pci_register_bar(&proxy->pci_dev, 1,
 679                          msix_bar_size(&proxy->pci_dev),
 680                          PCI_BASE_ADDRESS_SPACE_MEMORY,
 681                          msix_mmio_map);
 682     } else
 683         vdev->nvectors = 0;
 684
 685     proxy->pci_dev.config_write = virtio_write_config;
 686
 687     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) + vdev->config_len;
 688     if (size & (size-1))
 689         size = 1 << qemu_fls(size);
 690
 691     pci_register_bar(&proxy->pci_dev, 0, size, PCI_BASE_ADDRESS_SPACE_IO,
 692                            virtio_map);
 693
 694     if (!kvm_has_many_ioeventfds()) {
 695         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
 696     }
 697
 698     virtio_bind_device(vdev, &virtio_pci_bindings, proxy);
 699     proxy->host_features |= 0x1 << VIRTIO_F_NOTIFY_ON_EMPTY;
 700     proxy->host_features |= 0x1 << VIRTIO_F_BAD_FEATURE;
 701     proxy->host_features = vdev->get_features(vdev, proxy->host_features);
 702 }
 703
 704 static int virtio_blk_init_pci(PCIDevice *pci_dev)
 705 {
 706     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 707     VirtIODevice *vdev;
 708
 709     if (proxy->class_code != PCI_CLASS_STORAGE_SCSI &&
 710         proxy->class_code != PCI_CLASS_STORAGE_OTHER)
 711         proxy->class_code = PCI_CLASS_STORAGE_SCSI;
 712
 713     vdev = virtio_blk_init(&pci_dev->qdev, &proxy->block);
 714     if (!vdev) {
 715         return -1;
 716     }
 717     vdev->nvectors = proxy->nvectors;
 718     virtio_init_pci(proxy, vdev,
 719                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 720                     PCI_DEVICE_ID_VIRTIO_BLOCK,
 721                     proxy->class_code, 0x00);
 722     /* make the actual value visible */
 723     proxy->nvectors = vdev->nvectors;
 724     return 0;
 725 }
 726
 727 static int virtio_exit_pci(PCIDevice *pci_dev)
 728 {
 729     return msix_uninit(pci_dev);
 730 }
 731
 732 static int virtio_blk_exit_pci(PCIDevice *pci_dev)
 733 {
 734     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 735
 736     virtio_pci_stop_ioeventfd(proxy);
 737     virtio_blk_exit(proxy->vdev);
 738     blockdev_mark_auto_del(proxy->block.bs);
 739     return virtio_exit_pci(pci_dev);
 740 }
 741
 742 static int virtio_serial_init_pci(PCIDevice *pci_dev)
 743 {
 744     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 745     VirtIODevice *vdev;
 746
 747     if (proxy->class_code != PCI_CLASS_COMMUNICATION_OTHER &&
 748         proxy->class_code != PCI_CLASS_DISPLAY_OTHER && /* qemu 0.10 */
 749         proxy->class_code != PCI_CLASS_OTHERS)          /* qemu-kvm  */
 750         proxy->class_code = PCI_CLASS_COMMUNICATION_OTHER;
 751
 752     vdev = virtio_serial_init(&pci_dev->qdev, &proxy->serial);
 753     if (!vdev) {
 754         return -1;
 755     }
 756     vdev->nvectors = proxy->nvectors == DEV_NVECTORS_UNSPECIFIED
 757                                         ? proxy->serial.max_virtserial_ports + 1
 758                                         : proxy->nvectors;
 759     virtio_init_pci(proxy, vdev,
 760                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 761                     PCI_DEVICE_ID_VIRTIO_CONSOLE,
 762                     proxy->class_code, 0x00);
 763     proxy->nvectors = vdev->nvectors;
 764     return 0;
 765 }
 766
 767 static int virtio_serial_exit_pci(PCIDevice *pci_dev)
 768 {
 769     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 770
 771     virtio_pci_stop_ioeventfd(proxy);
 772     virtio_serial_exit(proxy->vdev);
 773     return virtio_exit_pci(pci_dev);
 774 }
 775
 776 static int virtio_net_init_pci(PCIDevice *pci_dev)
 777 {
 778     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 779     VirtIODevice *vdev;
 780
 781     vdev = virtio_net_init(&pci_dev->qdev, &proxy->nic, &proxy->net);
 782
 783     vdev->nvectors = proxy->nvectors;
 784     virtio_init_pci(proxy, vdev,
 785                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 786                     PCI_DEVICE_ID_VIRTIO_NET,
 787                     PCI_CLASS_NETWORK_ETHERNET,
 788                     0x00);
 789
 790     /* make the actual value visible */
 791     proxy->nvectors = vdev->nvectors;
 792     return 0;
 793 }
 794
 795 static int virtio_net_exit_pci(PCIDevice *pci_dev)
 796 {
 797     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 798
 799     virtio_pci_stop_ioeventfd(proxy);
 800     virtio_net_exit(proxy->vdev);
 801     return virtio_exit_pci(pci_dev);
 802 }
 803
 804 static int virtio_balloon_init_pci(PCIDevice *pci_dev)
 805 {
 806     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 807     VirtIODevice *vdev;
 808
 809     vdev = virtio_balloon_init(&pci_dev->qdev);
 810     virtio_init_pci(proxy, vdev,
 811                     PCI_VENDOR_ID_REDHAT_QUMRANET,
 812                     PCI_DEVICE_ID_VIRTIO_BALLOON,
 813                     PCI_CLASS_MEMORY_RAM,
 814                     0x00);
 815     return 0;
 816 }
 817
 818 static PCIDeviceInfo virtio_info[] = {
 819     {
 820         .qdev.name = "virtio-blk-pci",
 821         .qdev.alias = "virtio-blk",
 822         .qdev.size = sizeof(VirtIOPCIProxy),
 823         .init      = virtio_blk_init_pci,
 824         .exit      = virtio_blk_exit_pci,
 825         .qdev.props = (Property[]) {
 826             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 827             DEFINE_BLOCK_PROPERTIES(VirtIOPCIProxy, block),
 828             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 829                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 830             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
 831             DEFINE_VIRTIO_BLK_FEATURES(VirtIOPCIProxy, host_features),
 832             DEFINE_PROP_END_OF_LIST(),
 833         },
 834         .qdev.reset = virtio_pci_reset,
 835     },{
 836         .qdev.name  = "virtio-net-pci",
 837         .qdev.alias = "virtio-net",
 838         .qdev.size  = sizeof(VirtIOPCIProxy),
 839         .init       = virtio_net_init_pci,
 840         .exit       = virtio_net_exit_pci,
 841         .romfile    = "pxe-virtio.rom",
 842         .qdev.props = (Property[]) {
 843             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 844                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
 845             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
 846             DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
 847             DEFINE_NIC_PROPERTIES(VirtIOPCIProxy, nic),
 848             DEFINE_PROP_UINT32("x-txtimer", VirtIOPCIProxy,
 849                                net.txtimer, TX_TIMER_INTERVAL),
 850             DEFINE_PROP_INT32("x-txburst", VirtIOPCIProxy,
 851                               net.txburst, TX_BURST),
 852             DEFINE_PROP_STRING("tx", VirtIOPCIProxy, net.tx),
 853             DEFINE_PROP_END_OF_LIST(),
 854         },
 855         .qdev.reset = virtio_pci_reset,
 856     },{
 857         .qdev.name = "virtio-serial-pci",
 858         .qdev.alias = "virtio-serial",
 859         .qdev.size = sizeof(VirtIOPCIProxy),
 860         .init      = virtio_serial_init_pci,
 861         .exit      = virtio_serial_exit_pci,
 862         .qdev.props = (Property[]) {
 863             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 864                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 865             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
 866                                DEV_NVECTORS_UNSPECIFIED),
 867             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 868             DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 869             DEFINE_PROP_UINT32("max_ports", VirtIOPCIProxy,
 870                                serial.max_virtserial_ports, 31),
 871             DEFINE_PROP_END_OF_LIST(),
 872         },
 873         .qdev.reset = virtio_pci_reset,
 874     },{
 875         .qdev.name = "virtio-balloon-pci",
 876         .qdev.alias = "virtio-balloon",
 877         .qdev.size = sizeof(VirtIOPCIProxy),
 878         .init      = virtio_balloon_init_pci,
 879         .exit      = virtio_exit_pci,
 880         .qdev.props = (Property[]) {
 881             DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 882             DEFINE_PROP_END_OF_LIST(),
 883         },
 884         .qdev.reset = virtio_pci_reset,
 885     },{
 886         /* end of list */
 887     }
 888 };
 889
 890 static void virtio_pci_register_devices(void)
 891 {
 892     pci_qdev_register_many(virtio_info);
 893 }
 894
 895 device_init(virtio_pci_register_devices)