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 #include "range.h"
  31
  32 /* from Linux's linux/virtio_pci.h */
  33
  34 /* A 32-bit r/o bitmask of the features supported by the host */
  35 #define VIRTIO_PCI_HOST_FEATURES        0
  36
  37 /* A 32-bit r/w bitmask of features activated by the guest */
  38 #define VIRTIO_PCI_GUEST_FEATURES       4
  39
  40 /* A 32-bit r/w PFN for the currently selected queue */
  41 #define VIRTIO_PCI_QUEUE_PFN            8
  42
  43 /* A 16-bit r/o queue size for the currently selected queue */
  44 #define VIRTIO_PCI_QUEUE_NUM            12
  45
  46 /* A 16-bit r/w queue selector */
  47 #define VIRTIO_PCI_QUEUE_SEL            14
  48
  49 /* A 16-bit r/w queue notifier */
  50 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  51
  52 /* An 8-bit device status register.  */
  53 #define VIRTIO_PCI_STATUS               18
  54
  55 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  56  * current contents of the ISR and will also clear it.  This is effectively
  57  * a read-and-acknowledge. */
  58 #define VIRTIO_PCI_ISR                  19
  59
  60 /* MSI-X registers: only enabled if MSI-X is enabled. */
  61 /* A 16-bit vector for configuration changes. */
  62 #define VIRTIO_MSI_CONFIG_VECTOR        20
  63 /* A 16-bit vector for selected queue notifications. */
  64 #define VIRTIO_MSI_QUEUE_VECTOR         22
  65
  66 /* Config space size */
  67 #define VIRTIO_PCI_CONFIG_NOMSI         20
  68 #define VIRTIO_PCI_CONFIG_MSI           24
  69 #define VIRTIO_PCI_REGION_SIZE(dev)     (msix_present(dev) ? \
  70                                          VIRTIO_PCI_CONFIG_MSI : \
  71                                          VIRTIO_PCI_CONFIG_NOMSI)
  72
  73 /* The remaining space is defined by each driver as the per-driver
  74  * configuration space */
  75 #define VIRTIO_PCI_CONFIG(dev)          (msix_enabled(dev) ? \
  76                                          VIRTIO_PCI_CONFIG_MSI : \
  77                                          VIRTIO_PCI_CONFIG_NOMSI)
  78
  79 /* How many bits to shift physical queue address written to QUEUE_PFN.
  80  * 12 is historical, and due to x86 page size. */
  81 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  82
  83 /* Flags track per-device state like workarounds for quirks in older guests. */
  84 #define VIRTIO_PCI_FLAG_BUS_MASTER_BUG  (1 << 0)
  85
  86 /* Performance improves when virtqueue kick processing is decoupled from the
  87  * vcpu thread using ioeventfd for some devices. */
  88 #define VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT 1
  89 #define VIRTIO_PCI_FLAG_USE_IOEVENTFD   (1 << VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT)
  90
  91 /* QEMU doesn't strictly need write barriers since everything runs in
  92  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  93  * KVM or if kqemu gets SMP support.
  94  */
  95 #define wmb() do { } while (0)
  96
  97 /* virtio device */
  98
  99 static void virtio_pci_notify(void *opaque, uint16_t vector)
 100 {
 101     VirtIOPCIProxy *proxy = opaque;
 102     if (msix_enabled(&proxy->pci_dev))
 103         msix_notify(&proxy->pci_dev, vector);
 104     else
 105         qemu_set_irq(proxy->pci_dev.irq[0], proxy->vdev->isr & 1);
 106 }
 107
 108 static void virtio_pci_save_config(void * opaque, QEMUFile *f)
 109 {
 110     VirtIOPCIProxy *proxy = opaque;
 111     pci_device_save(&proxy->pci_dev, f);
 112     msix_save(&proxy->pci_dev, f);
 113     if (msix_present(&proxy->pci_dev))
 114         qemu_put_be16(f, proxy->vdev->config_vector);
 115 }
 116
 117 static void virtio_pci_save_queue(void * opaque, int n, QEMUFile *f)
 118 {
 119     VirtIOPCIProxy *proxy = opaque;
 120     if (msix_present(&proxy->pci_dev))
 121         qemu_put_be16(f, virtio_queue_vector(proxy->vdev, n));
 122 }
 123
 124 static int virtio_pci_load_config(void * opaque, QEMUFile *f)
 125 {
 126     VirtIOPCIProxy *proxy = opaque;
 127     int ret;
 128     ret = pci_device_load(&proxy->pci_dev, f);
 129     if (ret) {
 130         return ret;
 131     }
 132     msix_load(&proxy->pci_dev, f);
 133     if (msix_present(&proxy->pci_dev)) {
 134         qemu_get_be16s(f, &proxy->vdev->config_vector);
 135     } else {
 136         proxy->vdev->config_vector = VIRTIO_NO_VECTOR;
 137     }
 138     if (proxy->vdev->config_vector != VIRTIO_NO_VECTOR) {
 139         return msix_vector_use(&proxy->pci_dev, proxy->vdev->config_vector);
 140     }
 141     return 0;
 142 }
 143
 144 static int virtio_pci_load_queue(void * opaque, int n, QEMUFile *f)
 145 {
 146     VirtIOPCIProxy *proxy = opaque;
 147     uint16_t vector;
 148     if (msix_present(&proxy->pci_dev)) {
 149         qemu_get_be16s(f, &vector);
 150     } else {
 151         vector = VIRTIO_NO_VECTOR;
 152     }
 153     virtio_queue_set_vector(proxy->vdev, n, vector);
 154     if (vector != VIRTIO_NO_VECTOR) {
 155         return msix_vector_use(&proxy->pci_dev, vector);
 156     }
 157     return 0;
 158 }
 159
 160 static int virtio_pci_set_host_notifier_internal(VirtIOPCIProxy *proxy,
 161                                                  int n, bool assign)
 162 {
 163     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 164     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 165     int r;
 166     if (assign) {
 167         r = event_notifier_init(notifier, 1);
 168         if (r < 0) {
 169             error_report("%s: unable to init event notifier: %d",
 170                          __func__, r);
 171             return r;
 172         }
 173         r = kvm_set_ioeventfd_pio_word(event_notifier_get_fd(notifier),
 174                                        proxy->addr + VIRTIO_PCI_QUEUE_NOTIFY,
 175                                        n, assign);
 176         if (r < 0) {
 177             error_report("%s: unable to map ioeventfd: %d",
 178                          __func__, r);
 179             event_notifier_cleanup(notifier);
 180         }
 181     } else {
 182         r = kvm_set_ioeventfd_pio_word(event_notifier_get_fd(notifier),
 183                                        proxy->addr + VIRTIO_PCI_QUEUE_NOTIFY,
 184                                        n, assign);
 185         if (r < 0) {
 186             error_report("%s: unable to unmap ioeventfd: %d",
 187                          __func__, r);
 188             return r;
 189         }
 190
 191         /* Handle the race condition where the guest kicked and we deassigned
 192          * before we got around to handling the kick.
 193          */
 194         if (event_notifier_test_and_clear(notifier)) {
 195             virtio_queue_notify_vq(vq);
 196         }
 197
 198         event_notifier_cleanup(notifier);
 199     }
 200     return r;
 201 }
 202
 203 static void virtio_pci_host_notifier_read(void *opaque)
 204 {
 205     VirtQueue *vq = opaque;
 206     EventNotifier *n = virtio_queue_get_host_notifier(vq);
 207     if (event_notifier_test_and_clear(n)) {
 208         virtio_queue_notify_vq(vq);
 209     }
 210 }
 211
 212 static void virtio_pci_set_host_notifier_fd_handler(VirtIOPCIProxy *proxy,
 213                                                     int n, bool assign)
 214 {
 215     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 216     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 217     if (assign) {
 218         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 219                             virtio_pci_host_notifier_read, NULL, vq);
 220     } else {
 221         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 222                             NULL, NULL, NULL);
 223     }
 224 }
 225
 226 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 227 {
 228     int n, r;
 229
 230     if (!(proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) ||
 231         proxy->ioeventfd_disabled ||
 232         proxy->ioeventfd_started) {
 233         return;
 234     }
 235
 236     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 237         if (!virtio_queue_get_num(proxy->vdev, n)) {
 238             continue;
 239         }
 240
 241         r = virtio_pci_set_host_notifier_internal(proxy, n, true);
 242         if (r < 0) {
 243             goto assign_error;
 244         }
 245
 246         virtio_pci_set_host_notifier_fd_handler(proxy, n, true);
 247     }
 248     proxy->ioeventfd_started = true;
 249     return;
 250
 251 assign_error:
 252     while (--n >= 0) {
 253         if (!virtio_queue_get_num(proxy->vdev, n)) {
 254             continue;
 255         }
 256
 257         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 258         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 259         assert(r >= 0);
 260     }
 261     proxy->ioeventfd_started = false;
 262     error_report("%s: failed. Fallback to a userspace (slower).", __func__);
 263 }
 264
 265 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 266 {
 267     int r;
 268     int n;
 269
 270     if (!proxy->ioeventfd_started) {
 271         return;
 272     }
 273
 274     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 275         if (!virtio_queue_get_num(proxy->vdev, n)) {
 276             continue;
 277         }
 278
 279         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 280         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 281         assert(r >= 0);
 282     }
 283     proxy->ioeventfd_started = false;
 284 }
 285
 286 static void virtio_pci_reset(DeviceState *d)
 287 {
 288     VirtIOPCIProxy *proxy = container_of(d, VirtIOPCIProxy, pci_dev.qdev);
 289     virtio_pci_stop_ioeventfd(proxy);
 290     virtio_reset(proxy->vdev);
 291     msix_reset(&proxy->pci_dev);
 292     proxy->flags &= ~VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 293 }
 294
 295 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 296 {
 297     VirtIOPCIProxy *proxy = opaque;
 298     VirtIODevice *vdev = proxy->vdev;
 299     target_phys_addr_t pa;
 300
 301     switch (addr) {
 302     case VIRTIO_PCI_GUEST_FEATURES:
 303         /* Guest does not negotiate properly?  We have to assume nothing. */
 304         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 305             if (vdev->bad_features)
 306                 val = proxy->host_features & vdev->bad_features(vdev);
 307             else
 308                 val = 0;
 309         }
 310         if (vdev->set_features)
 311             vdev->set_features(vdev, val);
 312         vdev->guest_features = val;
 313         break;
 314     case VIRTIO_PCI_QUEUE_PFN:
 315         pa = (target_phys_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 316         if (pa == 0) {
 317             virtio_pci_stop_ioeventfd(proxy);
 318             virtio_reset(proxy->vdev);
 319             msix_unuse_all_vectors(&proxy->pci_dev);
 320         }
 321         else
 322             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 323         break;
 324     case VIRTIO_PCI_QUEUE_SEL:
 325         if (val < VIRTIO_PCI_QUEUE_MAX)
 326             vdev->queue_sel = val;
 327         break;
 328     case VIRTIO_PCI_QUEUE_NOTIFY:
 329         if (val < VIRTIO_PCI_QUEUE_MAX) {
 330             virtio_queue_notify(vdev, val);
 331         }
 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     pci_default_write_config(pci_dev, address, val, len);
 521
 522     if (range_covers_byte(address, len, PCI_COMMAND) &&
 523         !(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER) &&
 524         !(proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG)) {
 525         virtio_pci_stop_ioeventfd(proxy);
 526         virtio_set_status(proxy->vdev,
 527                           proxy->vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 528     }
 529
 530     msix_write_config(pci_dev, address, val, len);
 531 }
 532
 533 static unsigned virtio_pci_get_features(void *opaque)
 534 {
 535     VirtIOPCIProxy *proxy = opaque;
 536     return proxy->host_features;
 537 }
 538
 539 static void virtio_pci_guest_notifier_read(void *opaque)
 540 {
 541     VirtQueue *vq = opaque;
 542     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 543     if (event_notifier_test_and_clear(n)) {
 544         virtio_irq(vq);
 545     }
 546 }
 547
 548 static int virtio_pci_set_guest_notifier(void *opaque, int n, bool assign)
 549 {
 550     VirtIOPCIProxy *proxy = opaque;
 551     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 552     EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 553
 554     if (assign) {
 555         int r = event_notifier_init(notifier, 0);
 556         if (r < 0) {
 557             return r;
 558         }
 559         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 560                             virtio_pci_guest_notifier_read, NULL, vq);
 561     } else {
 562         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 563                             NULL, NULL, NULL);
 564         event_notifier_cleanup(notifier);
 565     }
 566
 567     return 0;
 568 }
 569
 570 static bool virtio_pci_query_guest_notifiers(void *opaque)
 571 {
 572     VirtIOPCIProxy *proxy = opaque;
 573     return msix_enabled(&proxy->pci_dev);
 574 }
 575
 576 static int virtio_pci_set_guest_notifiers(void *opaque, bool assign)
 577 {
 578     VirtIOPCIProxy *proxy = opaque;
 579     VirtIODevice *vdev = proxy->vdev;
 580     int r, n;
 581
 582     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 583         if (!virtio_queue_get_num(vdev, n)) {
 584             break;
 585         }
 586
 587         r = virtio_pci_set_guest_notifier(opaque, n, assign);
 588         if (r < 0) {
 589             goto assign_error;
 590         }
 591     }
 592
 593     return 0;
 594
 595 assign_error:
 596     /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
 597     while (--n >= 0) {
 598         virtio_pci_set_guest_notifier(opaque, n, !assign);
 599     }
 600     return r;
 601 }
 602
 603 static int virtio_pci_set_host_notifier(void *opaque, int n, bool assign)
 604 {
 605     VirtIOPCIProxy *proxy = opaque;
 606
 607     /* Stop using ioeventfd for virtqueue kick if the device starts using host
 608      * notifiers.  This makes it easy to avoid stepping on each others' toes.
 609      */
 610     proxy->ioeventfd_disabled = assign;
 611     if (assign) {
 612         virtio_pci_stop_ioeventfd(proxy);
 613     }
 614     /* We don't need to start here: it's not needed because backend
 615      * currently only stops on status change away from ok,
 616      * reset, vmstop and such. If we do add code to start here,
 617      * need to check vmstate, device state etc. */
 618     return virtio_pci_set_host_notifier_internal(proxy, n, assign);
 619 }
 620
 621 static void virtio_pci_vmstate_change(void *opaque, bool running)
 622 {
 623     VirtIOPCIProxy *proxy = opaque;
 624
 625     if (running) {
 626         /* Try to find out if the guest has bus master disabled, but is
 627            in ready state. Then we have a buggy guest OS. */
 628         if ((proxy->vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 629             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 630             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 631         }
 632         virtio_pci_start_ioeventfd(proxy);
 633     } else {
 634         virtio_pci_stop_ioeventfd(proxy);
 635     }
 636 }
 637
 638 static const VirtIOBindings virtio_pci_bindings = {
 639     .notify = virtio_pci_notify,
 640     .save_config = virtio_pci_save_config,
 641     .load_config = virtio_pci_load_config,
 642     .save_queue = virtio_pci_save_queue,
 643     .load_queue = virtio_pci_load_queue,
 644     .get_features = virtio_pci_get_features,
 645     .query_guest_notifiers = virtio_pci_query_guest_notifiers,
 646     .set_host_notifier = virtio_pci_set_host_notifier,
 647     .set_guest_notifiers = virtio_pci_set_guest_notifiers,
 648     .vmstate_change = virtio_pci_vmstate_change,
 649 };
 650
 651 void virtio_init_pci(VirtIOPCIProxy *proxy, VirtIODevice *vdev)
 652 {
 653     uint8_t *config;
 654     uint32_t size;
 655
 656     proxy->vdev = vdev;
 657
 658     config = proxy->pci_dev.config;
 659
 660     if (proxy->class_code) {
 661         pci_config_set_class(config, proxy->class_code);
 662     }
 663     pci_set_word(config + 0x2c, pci_get_word(config + PCI_VENDOR_ID));
 664     pci_set_word(config + 0x2e, vdev->device_id);
 665     config[0x3d] = 1;
 666
 667     if (vdev->nvectors && !msix_init(&proxy->pci_dev, vdev->nvectors, 1, 0)) {
 668         pci_register_bar(&proxy->pci_dev, 1,
 669                          msix_bar_size(&proxy->pci_dev),
 670                          PCI_BASE_ADDRESS_SPACE_MEMORY,
 671                          msix_mmio_map);
 672     } else
 673         vdev->nvectors = 0;
 674
 675     proxy->pci_dev.config_write = virtio_write_config;
 676
 677     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) + vdev->config_len;
 678     if (size & (size-1))
 679         size = 1 << qemu_fls(size);
 680
 681     pci_register_bar(&proxy->pci_dev, 0, size, PCI_BASE_ADDRESS_SPACE_IO,
 682                            virtio_map);
 683
 684     if (!kvm_has_many_ioeventfds()) {
 685         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
 686     }
 687
 688     virtio_bind_device(vdev, &virtio_pci_bindings, proxy);
 689     proxy->host_features |= 0x1 << VIRTIO_F_NOTIFY_ON_EMPTY;
 690     proxy->host_features |= 0x1 << VIRTIO_F_BAD_FEATURE;
 691     proxy->host_features = vdev->get_features(vdev, proxy->host_features);
 692 }
 693
 694 static int virtio_blk_init_pci(PCIDevice *pci_dev)
 695 {
 696     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 697     VirtIODevice *vdev;
 698
 699     if (proxy->class_code != PCI_CLASS_STORAGE_SCSI &&
 700         proxy->class_code != PCI_CLASS_STORAGE_OTHER)
 701         proxy->class_code = PCI_CLASS_STORAGE_SCSI;
 702
 703     vdev = virtio_blk_init(&pci_dev->qdev, &proxy->block,
 704                            &proxy->block_serial);
 705     if (!vdev) {
 706         return -1;
 707     }
 708     vdev->nvectors = proxy->nvectors;
 709     virtio_init_pci(proxy, vdev);
 710     /* make the actual value visible */
 711     proxy->nvectors = vdev->nvectors;
 712     return 0;
 713 }
 714
 715 static int virtio_exit_pci(PCIDevice *pci_dev)
 716 {
 717     return msix_uninit(pci_dev);
 718 }
 719
 720 static int virtio_blk_exit_pci(PCIDevice *pci_dev)
 721 {
 722     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 723
 724     virtio_pci_stop_ioeventfd(proxy);
 725     virtio_blk_exit(proxy->vdev);
 726     blockdev_mark_auto_del(proxy->block.bs);
 727     return virtio_exit_pci(pci_dev);
 728 }
 729
 730 static int virtio_serial_init_pci(PCIDevice *pci_dev)
 731 {
 732     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 733     VirtIODevice *vdev;
 734
 735     if (proxy->class_code != PCI_CLASS_COMMUNICATION_OTHER &&
 736         proxy->class_code != PCI_CLASS_DISPLAY_OTHER && /* qemu 0.10 */
 737         proxy->class_code != PCI_CLASS_OTHERS)          /* qemu-kvm  */
 738         proxy->class_code = PCI_CLASS_COMMUNICATION_OTHER;
 739
 740     vdev = virtio_serial_init(&pci_dev->qdev, &proxy->serial);
 741     if (!vdev) {
 742         return -1;
 743     }
 744     vdev->nvectors = proxy->nvectors == DEV_NVECTORS_UNSPECIFIED
 745                                         ? proxy->serial.max_virtserial_ports + 1
 746                                         : proxy->nvectors;
 747     virtio_init_pci(proxy, vdev);
 748     proxy->nvectors = vdev->nvectors;
 749     return 0;
 750 }
 751
 752 static int virtio_serial_exit_pci(PCIDevice *pci_dev)
 753 {
 754     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 755
 756     virtio_pci_stop_ioeventfd(proxy);
 757     virtio_serial_exit(proxy->vdev);
 758     return virtio_exit_pci(pci_dev);
 759 }
 760
 761 static int virtio_net_init_pci(PCIDevice *pci_dev)
 762 {
 763     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 764     VirtIODevice *vdev;
 765
 766     vdev = virtio_net_init(&pci_dev->qdev, &proxy->nic, &proxy->net);
 767
 768     vdev->nvectors = proxy->nvectors;
 769     virtio_init_pci(proxy, vdev);
 770
 771     /* make the actual value visible */
 772     proxy->nvectors = vdev->nvectors;
 773     return 0;
 774 }
 775
 776 static int virtio_net_exit_pci(PCIDevice *pci_dev)
 777 {
 778     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 779
 780     virtio_pci_stop_ioeventfd(proxy);
 781     virtio_net_exit(proxy->vdev);
 782     return virtio_exit_pci(pci_dev);
 783 }
 784
 785 static int virtio_balloon_init_pci(PCIDevice *pci_dev)
 786 {
 787     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 788     VirtIODevice *vdev;
 789
 790     vdev = virtio_balloon_init(&pci_dev->qdev);
 791     if (!vdev) {
 792         return -1;
 793     }
 794     virtio_init_pci(proxy, vdev);
 795     return 0;
 796 }
 797
 798 static int virtio_balloon_exit_pci(PCIDevice *pci_dev)
 799 {
 800     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 801
 802     virtio_pci_stop_ioeventfd(proxy);
 803     virtio_balloon_exit(proxy->vdev);
 804     return virtio_exit_pci(pci_dev);
 805 }
 806
 807 static PCIDeviceInfo virtio_info[] = {
 808     {
 809         .qdev.name = "virtio-blk-pci",
 810         .qdev.alias = "virtio-blk",
 811         .qdev.size = sizeof(VirtIOPCIProxy),
 812         .init      = virtio_blk_init_pci,
 813         .exit      = virtio_blk_exit_pci,
 814         .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET,
 815         .device_id = PCI_DEVICE_ID_VIRTIO_BLOCK,
 816         .revision  = VIRTIO_PCI_ABI_VERSION,
 817         .class_id  = PCI_CLASS_STORAGE_SCSI,
 818         .qdev.props = (Property[]) {
 819             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 820             DEFINE_BLOCK_PROPERTIES(VirtIOPCIProxy, block),
 821             DEFINE_PROP_STRING("serial", VirtIOPCIProxy, block_serial),
 822             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 823                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 824             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
 825             DEFINE_VIRTIO_BLK_FEATURES(VirtIOPCIProxy, host_features),
 826             DEFINE_PROP_END_OF_LIST(),
 827         },
 828         .qdev.reset = virtio_pci_reset,
 829     },{
 830         .qdev.name  = "virtio-net-pci",
 831         .qdev.alias = "virtio-net",
 832         .qdev.size  = sizeof(VirtIOPCIProxy),
 833         .init       = virtio_net_init_pci,
 834         .exit       = virtio_net_exit_pci,
 835         .romfile    = "pxe-virtio.rom",
 836         .vendor_id  = PCI_VENDOR_ID_REDHAT_QUMRANET,
 837         .device_id  = PCI_DEVICE_ID_VIRTIO_NET,
 838         .revision   = VIRTIO_PCI_ABI_VERSION,
 839         .class_id   = PCI_CLASS_NETWORK_ETHERNET,
 840         .qdev.props = (Property[]) {
 841             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 842                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
 843             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
 844             DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
 845             DEFINE_NIC_PROPERTIES(VirtIOPCIProxy, nic),
 846             DEFINE_PROP_UINT32("x-txtimer", VirtIOPCIProxy,
 847                                net.txtimer, TX_TIMER_INTERVAL),
 848             DEFINE_PROP_INT32("x-txburst", VirtIOPCIProxy,
 849                               net.txburst, TX_BURST),
 850             DEFINE_PROP_STRING("tx", VirtIOPCIProxy, net.tx),
 851             DEFINE_PROP_END_OF_LIST(),
 852         },
 853         .qdev.reset = virtio_pci_reset,
 854     },{
 855         .qdev.name = "virtio-serial-pci",
 856         .qdev.alias = "virtio-serial",
 857         .qdev.size = sizeof(VirtIOPCIProxy),
 858         .init      = virtio_serial_init_pci,
 859         .exit      = virtio_serial_exit_pci,
 860         .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET,
 861         .device_id = PCI_DEVICE_ID_VIRTIO_CONSOLE,
 862         .revision  = VIRTIO_PCI_ABI_VERSION,
 863         .class_id  = PCI_CLASS_COMMUNICATION_OTHER,
 864         .qdev.props = (Property[]) {
 865             DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
 866                             VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 867             DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
 868                                DEV_NVECTORS_UNSPECIFIED),
 869             DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 870             DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 871             DEFINE_PROP_UINT32("max_ports", VirtIOPCIProxy,
 872                                serial.max_virtserial_ports, 31),
 873             DEFINE_PROP_END_OF_LIST(),
 874         },
 875         .qdev.reset = virtio_pci_reset,
 876     },{
 877         .qdev.name = "virtio-balloon-pci",
 878         .qdev.alias = "virtio-balloon",
 879         .qdev.size = sizeof(VirtIOPCIProxy),
 880         .init      = virtio_balloon_init_pci,
 881         .exit      = virtio_balloon_exit_pci,
 882         .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET,
 883         .device_id = PCI_DEVICE_ID_VIRTIO_BALLOON,
 884         .revision  = VIRTIO_PCI_ABI_VERSION,
 885         .class_id  = PCI_CLASS_MEMORY_RAM,
 886         .qdev.props = (Property[]) {
 887             DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 888             DEFINE_PROP_END_OF_LIST(),
 889         },
 890         .qdev.reset = virtio_pci_reset,
 891     },{
 892         /* end of list */
 893     }
 894 };
 895
 896 static void virtio_pci_register_devices(void)
 897 {
 898     pci_qdev_register_many(virtio_info);
 899 }
 900
 901 device_init(virtio_pci_register_devices)