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  * Contributions after 2012-01-13 are licensed under the terms of the
  15  * GNU GPL, version 2 or (at your option) any later version.
  16  */
  17
  18 #include <inttypes.h>
  19
  20 #include "virtio.h"
  21 #include "virtio-blk.h"
  22 #include "virtio-net.h"
  23 #include "virtio-serial.h"
  24 #include "virtio-scsi.h"
  25 #include "pci/pci.h"
  26 #include "qemu/error-report.h"
  27 #include "pci/msi.h"
  28 #include "pci/msix.h"
  29 #include "loader.h"
  30 #include "sysemu/kvm.h"
  31 #include "sysemu/blockdev.h"
  32 #include "virtio-pci.h"
  33 #include "qemu/range.h"
  34 #include "virtio-bus.h"
  35
  36 /* from Linux's linux/virtio_pci.h */
  37
  38 /* A 32-bit r/o bitmask of the features supported by the host */
  39 #define VIRTIO_PCI_HOST_FEATURES        0
  40
  41 /* A 32-bit r/w bitmask of features activated by the guest */
  42 #define VIRTIO_PCI_GUEST_FEATURES       4
  43
  44 /* A 32-bit r/w PFN for the currently selected queue */
  45 #define VIRTIO_PCI_QUEUE_PFN            8
  46
  47 /* A 16-bit r/o queue size for the currently selected queue */
  48 #define VIRTIO_PCI_QUEUE_NUM            12
  49
  50 /* A 16-bit r/w queue selector */
  51 #define VIRTIO_PCI_QUEUE_SEL            14
  52
  53 /* A 16-bit r/w queue notifier */
  54 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  55
  56 /* An 8-bit device status register.  */
  57 #define VIRTIO_PCI_STATUS               18
  58
  59 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  60  * current contents of the ISR and will also clear it.  This is effectively
  61  * a read-and-acknowledge. */
  62 #define VIRTIO_PCI_ISR                  19
  63
  64 /* MSI-X registers: only enabled if MSI-X is enabled. */
  65 /* A 16-bit vector for configuration changes. */
  66 #define VIRTIO_MSI_CONFIG_VECTOR        20
  67 /* A 16-bit vector for selected queue notifications. */
  68 #define VIRTIO_MSI_QUEUE_VECTOR         22
  69
  70 /* Config space size */
  71 #define VIRTIO_PCI_CONFIG_NOMSI         20
  72 #define VIRTIO_PCI_CONFIG_MSI           24
  73 #define VIRTIO_PCI_REGION_SIZE(dev)     (msix_present(dev) ? \
  74                                          VIRTIO_PCI_CONFIG_MSI : \
  75                                          VIRTIO_PCI_CONFIG_NOMSI)
  76
  77 /* The remaining space is defined by each driver as the per-driver
  78  * configuration space */
  79 #define VIRTIO_PCI_CONFIG(dev)          (msix_enabled(dev) ? \
  80                                          VIRTIO_PCI_CONFIG_MSI : \
  81                                          VIRTIO_PCI_CONFIG_NOMSI)
  82
  83 /* How many bits to shift physical queue address written to QUEUE_PFN.
  84  * 12 is historical, and due to x86 page size. */
  85 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  86
  87 /* Flags track per-device state like workarounds for quirks in older guests. */
  88 #define VIRTIO_PCI_FLAG_BUS_MASTER_BUG  (1 << 0)
  89
  90 /* QEMU doesn't strictly need write barriers since everything runs in
  91  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  92  * KVM or if kqemu gets SMP support.
  93  */
  94 #define wmb() do { } while (0)
  95
  96 /* HACK for virtio to determine if it's running a big endian guest */
  97 bool virtio_is_big_endian(void);
  98
  99 /* virtio device */
 100 /* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */
 101 static inline VirtIOPCIProxy *to_virtio_pci_proxy(DeviceState *d)
 102 {
 103     return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
 104 }
 105
 106 /* DeviceState to VirtIOPCIProxy. Note: used on datapath,
 107  * be careful and test performance if you change this.
 108  */
 109 static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d)
 110 {
 111     return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
 112 }
 113
 114 static void virtio_pci_notify(DeviceState *d, uint16_t vector)
 115 {
 116     VirtIOPCIProxy *proxy = to_virtio_pci_proxy_fast(d);
 117     if (msix_enabled(&proxy->pci_dev))
 118         msix_notify(&proxy->pci_dev, vector);
 119     else
 120         qemu_set_irq(proxy->pci_dev.irq[0], proxy->vdev->isr & 1);
 121 }
 122
 123 static void virtio_pci_save_config(DeviceState *d, QEMUFile *f)
 124 {
 125     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 126     pci_device_save(&proxy->pci_dev, f);
 127     msix_save(&proxy->pci_dev, f);
 128     if (msix_present(&proxy->pci_dev))
 129         qemu_put_be16(f, proxy->vdev->config_vector);
 130 }
 131
 132 static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f)
 133 {
 134     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 135     if (msix_present(&proxy->pci_dev))
 136         qemu_put_be16(f, virtio_queue_vector(proxy->vdev, n));
 137 }
 138
 139 static int virtio_pci_load_config(DeviceState *d, QEMUFile *f)
 140 {
 141     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 142     int ret;
 143     ret = pci_device_load(&proxy->pci_dev, f);
 144     if (ret) {
 145         return ret;
 146     }
 147     msix_unuse_all_vectors(&proxy->pci_dev);
 148     msix_load(&proxy->pci_dev, f);
 149     if (msix_present(&proxy->pci_dev)) {
 150         qemu_get_be16s(f, &proxy->vdev->config_vector);
 151     } else {
 152         proxy->vdev->config_vector = VIRTIO_NO_VECTOR;
 153     }
 154     if (proxy->vdev->config_vector != VIRTIO_NO_VECTOR) {
 155         return msix_vector_use(&proxy->pci_dev, proxy->vdev->config_vector);
 156     }
 157     return 0;
 158 }
 159
 160 static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f)
 161 {
 162     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 163     uint16_t vector;
 164     if (msix_present(&proxy->pci_dev)) {
 165         qemu_get_be16s(f, &vector);
 166     } else {
 167         vector = VIRTIO_NO_VECTOR;
 168     }
 169     virtio_queue_set_vector(proxy->vdev, n, vector);
 170     if (vector != VIRTIO_NO_VECTOR) {
 171         return msix_vector_use(&proxy->pci_dev, vector);
 172     }
 173     return 0;
 174 }
 175
 176 static int virtio_pci_set_host_notifier_internal(VirtIOPCIProxy *proxy,
 177                                                  int n, bool assign, bool set_handler)
 178 {
 179     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 180     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 181     int r = 0;
 182
 183     if (assign) {
 184         r = event_notifier_init(notifier, 1);
 185         if (r < 0) {
 186             error_report("%s: unable to init event notifier: %d",
 187                          __func__, r);
 188             return r;
 189         }
 190         virtio_queue_set_host_notifier_fd_handler(vq, true, set_handler);
 191         memory_region_add_eventfd(&proxy->bar, VIRTIO_PCI_QUEUE_NOTIFY, 2,
 192                                   true, n, notifier);
 193     } else {
 194         memory_region_del_eventfd(&proxy->bar, VIRTIO_PCI_QUEUE_NOTIFY, 2,
 195                                   true, n, notifier);
 196         virtio_queue_set_host_notifier_fd_handler(vq, false, false);
 197         event_notifier_cleanup(notifier);
 198     }
 199     return r;
 200 }
 201
 202 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 203 {
 204     int n, r;
 205
 206     if (!(proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) ||
 207         proxy->ioeventfd_disabled ||
 208         proxy->ioeventfd_started) {
 209         return;
 210     }
 211
 212     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 213         if (!virtio_queue_get_num(proxy->vdev, n)) {
 214             continue;
 215         }
 216
 217         r = virtio_pci_set_host_notifier_internal(proxy, n, true, true);
 218         if (r < 0) {
 219             goto assign_error;
 220         }
 221     }
 222     proxy->ioeventfd_started = true;
 223     return;
 224
 225 assign_error:
 226     while (--n >= 0) {
 227         if (!virtio_queue_get_num(proxy->vdev, n)) {
 228             continue;
 229         }
 230
 231         r = virtio_pci_set_host_notifier_internal(proxy, n, false, false);
 232         assert(r >= 0);
 233     }
 234     proxy->ioeventfd_started = false;
 235     error_report("%s: failed. Fallback to a userspace (slower).", __func__);
 236 }
 237
 238 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 239 {
 240     int r;
 241     int n;
 242
 243     if (!proxy->ioeventfd_started) {
 244         return;
 245     }
 246
 247     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 248         if (!virtio_queue_get_num(proxy->vdev, n)) {
 249             continue;
 250         }
 251
 252         r = virtio_pci_set_host_notifier_internal(proxy, n, false, false);
 253         assert(r >= 0);
 254     }
 255     proxy->ioeventfd_started = false;
 256 }
 257
 258 void virtio_pci_reset(DeviceState *d)
 259 {
 260     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 261     virtio_pci_stop_ioeventfd(proxy);
 262     virtio_reset(proxy->vdev);
 263     msix_unuse_all_vectors(&proxy->pci_dev);
 264     proxy->flags &= ~VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 265 }
 266
 267 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 268 {
 269     VirtIOPCIProxy *proxy = opaque;
 270     VirtIODevice *vdev = proxy->vdev;
 271     hwaddr pa;
 272
 273     switch (addr) {
 274     case VIRTIO_PCI_GUEST_FEATURES:
 275         /* Guest does not negotiate properly?  We have to assume nothing. */
 276         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 277             val = vdev->bad_features ? vdev->bad_features(vdev) : 0;
 278         }
 279         virtio_set_features(vdev, val);
 280         break;
 281     case VIRTIO_PCI_QUEUE_PFN:
 282         pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 283         if (pa == 0) {
 284             virtio_pci_stop_ioeventfd(proxy);
 285             virtio_reset(proxy->vdev);
 286             msix_unuse_all_vectors(&proxy->pci_dev);
 287         }
 288         else
 289             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 290         break;
 291     case VIRTIO_PCI_QUEUE_SEL:
 292         if (val < VIRTIO_PCI_QUEUE_MAX)
 293             vdev->queue_sel = val;
 294         break;
 295     case VIRTIO_PCI_QUEUE_NOTIFY:
 296         if (val < VIRTIO_PCI_QUEUE_MAX) {
 297             virtio_queue_notify(vdev, val);
 298         }
 299         break;
 300     case VIRTIO_PCI_STATUS:
 301         if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
 302             virtio_pci_stop_ioeventfd(proxy);
 303         }
 304
 305         virtio_set_status(vdev, val & 0xFF);
 306
 307         if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
 308             virtio_pci_start_ioeventfd(proxy);
 309         }
 310
 311         if (vdev->status == 0) {
 312             virtio_reset(proxy->vdev);
 313             msix_unuse_all_vectors(&proxy->pci_dev);
 314         }
 315
 316         /* Linux before 2.6.34 sets the device as OK without enabling
 317            the PCI device bus master bit. In this case we need to disable
 318            some safety checks. */
 319         if ((val & VIRTIO_CONFIG_S_DRIVER_OK) &&
 320             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 321             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 322         }
 323         break;
 324     case VIRTIO_MSI_CONFIG_VECTOR:
 325         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 326         /* Make it possible for guest to discover an error took place. */
 327         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 328             val = VIRTIO_NO_VECTOR;
 329         vdev->config_vector = val;
 330         break;
 331     case VIRTIO_MSI_QUEUE_VECTOR:
 332         msix_vector_unuse(&proxy->pci_dev,
 333                           virtio_queue_vector(vdev, vdev->queue_sel));
 334         /* Make it possible for guest to discover an error took place. */
 335         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 336             val = VIRTIO_NO_VECTOR;
 337         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 338         break;
 339     default:
 340         error_report("%s: unexpected address 0x%x value 0x%x",
 341                      __func__, addr, val);
 342         break;
 343     }
 344 }
 345
 346 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 347 {
 348     VirtIODevice *vdev = proxy->vdev;
 349     uint32_t ret = 0xFFFFFFFF;
 350
 351     switch (addr) {
 352     case VIRTIO_PCI_HOST_FEATURES:
 353         ret = proxy->host_features;
 354         break;
 355     case VIRTIO_PCI_GUEST_FEATURES:
 356         ret = vdev->guest_features;
 357         break;
 358     case VIRTIO_PCI_QUEUE_PFN:
 359         ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 360               >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 361         break;
 362     case VIRTIO_PCI_QUEUE_NUM:
 363         ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 364         break;
 365     case VIRTIO_PCI_QUEUE_SEL:
 366         ret = vdev->queue_sel;
 367         break;
 368     case VIRTIO_PCI_STATUS:
 369         ret = vdev->status;
 370         break;
 371     case VIRTIO_PCI_ISR:
 372         /* reading from the ISR also clears it. */
 373         ret = vdev->isr;
 374         vdev->isr = 0;
 375         qemu_set_irq(proxy->pci_dev.irq[0], 0);
 376         break;
 377     case VIRTIO_MSI_CONFIG_VECTOR:
 378         ret = vdev->config_vector;
 379         break;
 380     case VIRTIO_MSI_QUEUE_VECTOR:
 381         ret = virtio_queue_vector(vdev, vdev->queue_sel);
 382         break;
 383     default:
 384         break;
 385     }
 386
 387     return ret;
 388 }
 389
 390 static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr,
 391                                        unsigned size)
 392 {
 393     VirtIOPCIProxy *proxy = opaque;
 394     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 395     uint64_t val = 0;
 396     if (addr < config) {
 397         return virtio_ioport_read(proxy, addr);
 398     }
 399     addr -= config;
 400
 401     switch (size) {
 402     case 1:
 403         val = virtio_config_readb(proxy->vdev, addr);
 404         break;
 405     case 2:
 406         val = virtio_config_readw(proxy->vdev, addr);
 407         if (virtio_is_big_endian()) {
 408             val = bswap16(val);
 409         }
 410         break;
 411     case 4:
 412         val = virtio_config_readl(proxy->vdev, addr);
 413         if (virtio_is_big_endian()) {
 414             val = bswap32(val);
 415         }
 416         break;
 417     }
 418     return val;
 419 }
 420
 421 static void virtio_pci_config_write(void *opaque, hwaddr addr,
 422                                     uint64_t val, unsigned size)
 423 {
 424     VirtIOPCIProxy *proxy = opaque;
 425     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 426     if (addr < config) {
 427         virtio_ioport_write(proxy, addr, val);
 428         return;
 429     }
 430     addr -= config;
 431     /*
 432      * Virtio-PCI is odd. Ioports are LE but config space is target native
 433      * endian.
 434      */
 435     switch (size) {
 436     case 1:
 437         virtio_config_writeb(proxy->vdev, addr, val);
 438         break;
 439     case 2:
 440         if (virtio_is_big_endian()) {
 441             val = bswap16(val);
 442         }
 443         virtio_config_writew(proxy->vdev, addr, val);
 444         break;
 445     case 4:
 446         if (virtio_is_big_endian()) {
 447             val = bswap32(val);
 448         }
 449         virtio_config_writel(proxy->vdev, addr, val);
 450         break;
 451     }
 452 }
 453
 454 static const MemoryRegionOps virtio_pci_config_ops = {
 455     .read = virtio_pci_config_read,
 456     .write = virtio_pci_config_write,
 457     .impl = {
 458         .min_access_size = 1,
 459         .max_access_size = 4,
 460     },
 461     .endianness = DEVICE_LITTLE_ENDIAN,
 462 };
 463
 464 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 465                                 uint32_t val, int len)
 466 {
 467     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 468
 469     pci_default_write_config(pci_dev, address, val, len);
 470
 471     if (range_covers_byte(address, len, PCI_COMMAND) &&
 472         !(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER) &&
 473         !(proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG)) {
 474         virtio_pci_stop_ioeventfd(proxy);
 475         virtio_set_status(proxy->vdev,
 476                           proxy->vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 477     }
 478 }
 479
 480 static unsigned virtio_pci_get_features(DeviceState *d)
 481 {
 482     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 483     return proxy->host_features;
 484 }
 485
 486 static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy,
 487                                         unsigned int queue_no,
 488                                         unsigned int vector,
 489                                         MSIMessage msg)
 490 {
 491     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 492     int ret;
 493
 494     if (irqfd->users == 0) {
 495         ret = kvm_irqchip_add_msi_route(kvm_state, msg);
 496         if (ret < 0) {
 497             return ret;
 498         }
 499         irqfd->virq = ret;
 500     }
 501     irqfd->users++;
 502     return 0;
 503 }
 504
 505 static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy,
 506                                              unsigned int vector)
 507 {
 508     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 509     if (--irqfd->users == 0) {
 510         kvm_irqchip_release_virq(kvm_state, irqfd->virq);
 511     }
 512 }
 513
 514 static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy,
 515                                  unsigned int queue_no,
 516                                  unsigned int vector)
 517 {
 518     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 519     VirtQueue *vq = virtio_get_queue(proxy->vdev, queue_no);
 520     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 521     int ret;
 522     ret = kvm_irqchip_add_irqfd_notifier(kvm_state, n, irqfd->virq);
 523     return ret;
 524 }
 525
 526 static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy,
 527                                       unsigned int queue_no,
 528                                       unsigned int vector)
 529 {
 530     VirtQueue *vq = virtio_get_queue(proxy->vdev, queue_no);
 531     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 532     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 533     int ret;
 534
 535     ret = kvm_irqchip_remove_irqfd_notifier(kvm_state, n, irqfd->virq);
 536     assert(ret == 0);
 537 }
 538
 539 static int kvm_virtio_pci_vector_use(VirtIOPCIProxy *proxy, int nvqs)
 540 {
 541     PCIDevice *dev = &proxy->pci_dev;
 542     VirtIODevice *vdev = proxy->vdev;
 543     unsigned int vector;
 544     int ret, queue_no;
 545     MSIMessage msg;
 546
 547     for (queue_no = 0; queue_no < nvqs; queue_no++) {
 548         if (!virtio_queue_get_num(vdev, queue_no)) {
 549             break;
 550         }
 551         vector = virtio_queue_vector(vdev, queue_no);
 552         if (vector >= msix_nr_vectors_allocated(dev)) {
 553             continue;
 554         }
 555         msg = msix_get_message(dev, vector);
 556         ret = kvm_virtio_pci_vq_vector_use(proxy, queue_no, vector, msg);
 557         if (ret < 0) {
 558             goto undo;
 559         }
 560         /* If guest supports masking, set up irqfd now.
 561          * Otherwise, delay until unmasked in the frontend.
 562          */
 563         if (proxy->vdev->guest_notifier_mask) {
 564             ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 565             if (ret < 0) {
 566                 kvm_virtio_pci_vq_vector_release(proxy, vector);
 567                 goto undo;
 568             }
 569         }
 570     }
 571     return 0;
 572
 573 undo:
 574     while (--queue_no >= 0) {
 575         vector = virtio_queue_vector(vdev, queue_no);
 576         if (vector >= msix_nr_vectors_allocated(dev)) {
 577             continue;
 578         }
 579         if (proxy->vdev->guest_notifier_mask) {
 580             kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 581         }
 582         kvm_virtio_pci_vq_vector_release(proxy, vector);
 583     }
 584     return ret;
 585 }
 586
 587 static void kvm_virtio_pci_vector_release(VirtIOPCIProxy *proxy, int nvqs)
 588 {
 589     PCIDevice *dev = &proxy->pci_dev;
 590     VirtIODevice *vdev = proxy->vdev;
 591     unsigned int vector;
 592     int queue_no;
 593
 594     for (queue_no = 0; queue_no < nvqs; queue_no++) {
 595         if (!virtio_queue_get_num(vdev, queue_no)) {
 596             break;
 597         }
 598         vector = virtio_queue_vector(vdev, queue_no);
 599         if (vector >= msix_nr_vectors_allocated(dev)) {
 600             continue;
 601         }
 602         /* If guest supports masking, clean up irqfd now.
 603          * Otherwise, it was cleaned when masked in the frontend.
 604          */
 605         if (proxy->vdev->guest_notifier_mask) {
 606             kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 607         }
 608         kvm_virtio_pci_vq_vector_release(proxy, vector);
 609     }
 610 }
 611
 612 static int kvm_virtio_pci_vq_vector_unmask(VirtIOPCIProxy *proxy,
 613                                         unsigned int queue_no,
 614                                         unsigned int vector,
 615                                         MSIMessage msg)
 616 {
 617     VirtQueue *vq = virtio_get_queue(proxy->vdev, queue_no);
 618     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 619     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 620     int ret = 0;
 621
 622     if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) {
 623         ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg);
 624         if (ret < 0) {
 625             return ret;
 626         }
 627     }
 628
 629     /* If guest supports masking, irqfd is already setup, unmask it.
 630      * Otherwise, set it up now.
 631      */
 632     if (proxy->vdev->guest_notifier_mask) {
 633         proxy->vdev->guest_notifier_mask(proxy->vdev, queue_no, false);
 634         /* Test after unmasking to avoid losing events. */
 635         if (proxy->vdev->guest_notifier_pending &&
 636             proxy->vdev->guest_notifier_pending(proxy->vdev, queue_no)) {
 637             event_notifier_set(n);
 638         }
 639     } else {
 640         ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 641     }
 642     return ret;
 643 }
 644
 645 static void kvm_virtio_pci_vq_vector_mask(VirtIOPCIProxy *proxy,
 646                                              unsigned int queue_no,
 647                                              unsigned int vector)
 648 {
 649     /* If guest supports masking, keep irqfd but mask it.
 650      * Otherwise, clean it up now.
 651      */
 652     if (proxy->vdev->guest_notifier_mask) {
 653         proxy->vdev->guest_notifier_mask(proxy->vdev, queue_no, true);
 654     } else {
 655         kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 656     }
 657 }
 658
 659 static int kvm_virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector,
 660                                      MSIMessage msg)
 661 {
 662     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 663     VirtIODevice *vdev = proxy->vdev;
 664     int ret, queue_no;
 665
 666     for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
 667         if (!virtio_queue_get_num(vdev, queue_no)) {
 668             break;
 669         }
 670         if (virtio_queue_vector(vdev, queue_no) != vector) {
 671             continue;
 672         }
 673         ret = kvm_virtio_pci_vq_vector_unmask(proxy, queue_no, vector, msg);
 674         if (ret < 0) {
 675             goto undo;
 676         }
 677     }
 678     return 0;
 679
 680 undo:
 681     while (--queue_no >= 0) {
 682         if (virtio_queue_vector(vdev, queue_no) != vector) {
 683             continue;
 684         }
 685         kvm_virtio_pci_vq_vector_mask(proxy, queue_no, vector);
 686     }
 687     return ret;
 688 }
 689
 690 static void kvm_virtio_pci_vector_mask(PCIDevice *dev, unsigned vector)
 691 {
 692     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 693     VirtIODevice *vdev = proxy->vdev;
 694     int queue_no;
 695
 696     for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
 697         if (!virtio_queue_get_num(vdev, queue_no)) {
 698             break;
 699         }
 700         if (virtio_queue_vector(vdev, queue_no) != vector) {
 701             continue;
 702         }
 703         kvm_virtio_pci_vq_vector_mask(proxy, queue_no, vector);
 704     }
 705 }
 706
 707 static void kvm_virtio_pci_vector_poll(PCIDevice *dev,
 708                                        unsigned int vector_start,
 709                                        unsigned int vector_end)
 710 {
 711     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 712     VirtIODevice *vdev = proxy->vdev;
 713     int queue_no;
 714     unsigned int vector;
 715     EventNotifier *notifier;
 716     VirtQueue *vq;
 717
 718     for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
 719         if (!virtio_queue_get_num(vdev, queue_no)) {
 720             break;
 721         }
 722         vector = virtio_queue_vector(vdev, queue_no);
 723         if (vector < vector_start || vector >= vector_end ||
 724             !msix_is_masked(dev, vector)) {
 725             continue;
 726         }
 727         vq = virtio_get_queue(vdev, queue_no);
 728         notifier = virtio_queue_get_guest_notifier(vq);
 729         if (vdev->guest_notifier_pending) {
 730             if (vdev->guest_notifier_pending(vdev, queue_no)) {
 731                 msix_set_pending(dev, vector);
 732             }
 733         } else if (event_notifier_test_and_clear(notifier)) {
 734             msix_set_pending(dev, vector);
 735         }
 736     }
 737 }
 738
 739 static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign,
 740                                          bool with_irqfd)
 741 {
 742     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 743     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 744     EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 745
 746     if (assign) {
 747         int r = event_notifier_init(notifier, 0);
 748         if (r < 0) {
 749             return r;
 750         }
 751         virtio_queue_set_guest_notifier_fd_handler(vq, true, with_irqfd);
 752     } else {
 753         virtio_queue_set_guest_notifier_fd_handler(vq, false, with_irqfd);
 754         event_notifier_cleanup(notifier);
 755     }
 756
 757     return 0;
 758 }
 759
 760 static bool virtio_pci_query_guest_notifiers(DeviceState *d)
 761 {
 762     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 763     return msix_enabled(&proxy->pci_dev);
 764 }
 765
 766 static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign)
 767 {
 768     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 769     VirtIODevice *vdev = proxy->vdev;
 770     int r, n;
 771     bool with_irqfd = msix_enabled(&proxy->pci_dev) &&
 772         kvm_msi_via_irqfd_enabled();
 773
 774     nvqs = MIN(nvqs, VIRTIO_PCI_QUEUE_MAX);
 775
 776     /* When deassigning, pass a consistent nvqs value
 777      * to avoid leaking notifiers.
 778      */
 779     assert(assign || nvqs == proxy->nvqs_with_notifiers);
 780
 781     proxy->nvqs_with_notifiers = nvqs;
 782
 783     /* Must unset vector notifier while guest notifier is still assigned */
 784     if (proxy->vector_irqfd && !assign) {
 785         msix_unset_vector_notifiers(&proxy->pci_dev);
 786         kvm_virtio_pci_vector_release(proxy, nvqs);
 787         g_free(proxy->vector_irqfd);
 788         proxy->vector_irqfd = NULL;
 789     }
 790
 791     for (n = 0; n < nvqs; n++) {
 792         if (!virtio_queue_get_num(vdev, n)) {
 793             break;
 794         }
 795
 796         r = virtio_pci_set_guest_notifier(d, n, assign,
 797                                           kvm_msi_via_irqfd_enabled());
 798         if (r < 0) {
 799             goto assign_error;
 800         }
 801     }
 802
 803     /* Must set vector notifier after guest notifier has been assigned */
 804     if (with_irqfd && assign) {
 805         proxy->vector_irqfd =
 806             g_malloc0(sizeof(*proxy->vector_irqfd) *
 807                       msix_nr_vectors_allocated(&proxy->pci_dev));
 808         r = kvm_virtio_pci_vector_use(proxy, nvqs);
 809         if (r < 0) {
 810             goto assign_error;
 811         }
 812         r = msix_set_vector_notifiers(&proxy->pci_dev,
 813                                       kvm_virtio_pci_vector_unmask,
 814                                       kvm_virtio_pci_vector_mask,
 815                                       kvm_virtio_pci_vector_poll);
 816         if (r < 0) {
 817             goto notifiers_error;
 818         }
 819     }
 820
 821     return 0;
 822
 823 notifiers_error:
 824     assert(assign);
 825     kvm_virtio_pci_vector_release(proxy, nvqs);
 826
 827 assign_error:
 828     /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
 829     assert(assign);
 830     while (--n >= 0) {
 831         virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd);
 832     }
 833     return r;
 834 }
 835
 836 static int virtio_pci_set_host_notifier(DeviceState *d, int n, bool assign)
 837 {
 838     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 839
 840     /* Stop using ioeventfd for virtqueue kick if the device starts using host
 841      * notifiers.  This makes it easy to avoid stepping on each others' toes.
 842      */
 843     proxy->ioeventfd_disabled = assign;
 844     if (assign) {
 845         virtio_pci_stop_ioeventfd(proxy);
 846     }
 847     /* We don't need to start here: it's not needed because backend
 848      * currently only stops on status change away from ok,
 849      * reset, vmstop and such. If we do add code to start here,
 850      * need to check vmstate, device state etc. */
 851     return virtio_pci_set_host_notifier_internal(proxy, n, assign, false);
 852 }
 853
 854 static void virtio_pci_vmstate_change(DeviceState *d, bool running)
 855 {
 856     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 857
 858     if (running) {
 859         /* Try to find out if the guest has bus master disabled, but is
 860            in ready state. Then we have a buggy guest OS. */
 861         if ((proxy->vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 862             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 863             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 864         }
 865         virtio_pci_start_ioeventfd(proxy);
 866     } else {
 867         virtio_pci_stop_ioeventfd(proxy);
 868     }
 869 }
 870
 871 static const VirtIOBindings virtio_pci_bindings = {
 872     .notify = virtio_pci_notify,
 873     .save_config = virtio_pci_save_config,
 874     .load_config = virtio_pci_load_config,
 875     .save_queue = virtio_pci_save_queue,
 876     .load_queue = virtio_pci_load_queue,
 877     .get_features = virtio_pci_get_features,
 878     .query_guest_notifiers = virtio_pci_query_guest_notifiers,
 879     .set_host_notifier = virtio_pci_set_host_notifier,
 880     .set_guest_notifiers = virtio_pci_set_guest_notifiers,
 881     .vmstate_change = virtio_pci_vmstate_change,
 882 };
 883
 884 void virtio_init_pci(VirtIOPCIProxy *proxy, VirtIODevice *vdev)
 885 {
 886     uint8_t *config;
 887     uint32_t size;
 888
 889     proxy->vdev = vdev;
 890
 891     config = proxy->pci_dev.config;
 892
 893     if (proxy->class_code) {
 894         pci_config_set_class(config, proxy->class_code);
 895     }
 896     pci_set_word(config + PCI_SUBSYSTEM_VENDOR_ID,
 897                  pci_get_word(config + PCI_VENDOR_ID));
 898     pci_set_word(config + PCI_SUBSYSTEM_ID, vdev->device_id);
 899     config[PCI_INTERRUPT_PIN] = 1;
 900
 901     if (vdev->nvectors &&
 902         msix_init_exclusive_bar(&proxy->pci_dev, vdev->nvectors, 1)) {
 903         vdev->nvectors = 0;
 904     }
 905
 906     proxy->pci_dev.config_write = virtio_write_config;
 907
 908     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) + vdev->config_len;
 909     if (size & (size-1))
 910         size = 1 << qemu_fls(size);
 911
 912     memory_region_init_io(&proxy->bar, &virtio_pci_config_ops, proxy,
 913                           "virtio-pci", size);
 914     pci_register_bar(&proxy->pci_dev, 0, PCI_BASE_ADDRESS_SPACE_IO,
 915                      &proxy->bar);
 916
 917     if (!kvm_has_many_ioeventfds()) {
 918         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
 919     }
 920
 921     virtio_bind_device(vdev, &virtio_pci_bindings, DEVICE(proxy));
 922     proxy->host_features |= 0x1 << VIRTIO_F_NOTIFY_ON_EMPTY;
 923     proxy->host_features |= 0x1 << VIRTIO_F_BAD_FEATURE;
 924     proxy->host_features = vdev->get_features(vdev, proxy->host_features);
 925 }
 926
 927 static int virtio_blk_init_pci(PCIDevice *pci_dev)
 928 {
 929     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 930     VirtIODevice *vdev;
 931
 932     if (proxy->class_code != PCI_CLASS_STORAGE_SCSI &&
 933         proxy->class_code != PCI_CLASS_STORAGE_OTHER)
 934         proxy->class_code = PCI_CLASS_STORAGE_SCSI;
 935
 936     vdev = virtio_blk_init(&pci_dev->qdev, &proxy->blk);
 937     if (!vdev) {
 938         return -1;
 939     }
 940     vdev->nvectors = proxy->nvectors;
 941     virtio_init_pci(proxy, vdev);
 942     /* make the actual value visible */
 943     proxy->nvectors = vdev->nvectors;
 944     return 0;
 945 }
 946
 947 static void virtio_exit_pci(PCIDevice *pci_dev)
 948 {
 949     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 950
 951     memory_region_destroy(&proxy->bar);
 952     msix_uninit_exclusive_bar(pci_dev);
 953 }
 954
 955 static void virtio_blk_exit_pci(PCIDevice *pci_dev)
 956 {
 957     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 958
 959     virtio_pci_stop_ioeventfd(proxy);
 960     virtio_blk_exit(proxy->vdev);
 961     virtio_exit_pci(pci_dev);
 962 }
 963
 964 static int virtio_serial_init_pci(PCIDevice *pci_dev)
 965 {
 966     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 967     VirtIODevice *vdev;
 968
 969     if (proxy->class_code != PCI_CLASS_COMMUNICATION_OTHER &&
 970         proxy->class_code != PCI_CLASS_DISPLAY_OTHER && /* qemu 0.10 */
 971         proxy->class_code != PCI_CLASS_OTHERS)          /* qemu-kvm  */
 972         proxy->class_code = PCI_CLASS_COMMUNICATION_OTHER;
 973
 974     vdev = virtio_serial_init(&pci_dev->qdev, &proxy->serial);
 975     if (!vdev) {
 976         return -1;
 977     }
 978     vdev->nvectors = proxy->nvectors == DEV_NVECTORS_UNSPECIFIED
 979                                         ? proxy->serial.max_virtserial_ports + 1
 980                                         : proxy->nvectors;
 981     virtio_init_pci(proxy, vdev);
 982     proxy->nvectors = vdev->nvectors;
 983     return 0;
 984 }
 985
 986 static void virtio_serial_exit_pci(PCIDevice *pci_dev)
 987 {
 988     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 989
 990     virtio_pci_stop_ioeventfd(proxy);
 991     virtio_serial_exit(proxy->vdev);
 992     virtio_exit_pci(pci_dev);
 993 }
 994
 995 static int virtio_net_init_pci(PCIDevice *pci_dev)
 996 {
 997     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 998     VirtIODevice *vdev;
 999
1000     vdev = virtio_net_init(&pci_dev->qdev, &proxy->nic, &proxy->net,
1001                            proxy->host_features);
1002
1003     vdev->nvectors = proxy->nvectors;
1004     virtio_init_pci(proxy, vdev);
1005
1006     /* make the actual value visible */
1007     proxy->nvectors = vdev->nvectors;
1008     return 0;
1009 }
1010
1011 static void virtio_net_exit_pci(PCIDevice *pci_dev)
1012 {
1013     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1014
1015     virtio_pci_stop_ioeventfd(proxy);
1016     virtio_net_exit(proxy->vdev);
1017     virtio_exit_pci(pci_dev);
1018 }
1019
1020 static int virtio_balloon_init_pci(PCIDevice *pci_dev)
1021 {
1022     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1023     VirtIODevice *vdev;
1024
1025     if (proxy->class_code != PCI_CLASS_OTHERS &&
1026         proxy->class_code != PCI_CLASS_MEMORY_RAM) { /* qemu < 1.1 */
1027         proxy->class_code = PCI_CLASS_OTHERS;
1028     }
1029
1030     vdev = virtio_balloon_init(&pci_dev->qdev);
1031     if (!vdev) {
1032         return -1;
1033     }
1034     virtio_init_pci(proxy, vdev);
1035     return 0;
1036 }
1037
1038 static void virtio_balloon_exit_pci(PCIDevice *pci_dev)
1039 {
1040     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1041
1042     virtio_pci_stop_ioeventfd(proxy);
1043     virtio_balloon_exit(proxy->vdev);
1044     virtio_exit_pci(pci_dev);
1045 }
1046
1047 static int virtio_rng_init_pci(PCIDevice *pci_dev)
1048 {
1049     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1050     VirtIODevice *vdev;
1051
1052     if (proxy->rng.rng == NULL) {
1053         proxy->rng.default_backend = RNG_RANDOM(object_new(TYPE_RNG_RANDOM));
1054
1055         object_property_add_child(OBJECT(pci_dev),
1056                                   "default-backend",
1057                                   OBJECT(proxy->rng.default_backend),
1058                                   NULL);
1059
1060         object_property_set_link(OBJECT(pci_dev),
1061                                  OBJECT(proxy->rng.default_backend),
1062                                  "rng", NULL);
1063     }
1064
1065     vdev = virtio_rng_init(&pci_dev->qdev, &proxy->rng);
1066     if (!vdev) {
1067         return -1;
1068     }
1069     virtio_init_pci(proxy, vdev);
1070     return 0;
1071 }
1072
1073 static void virtio_rng_exit_pci(PCIDevice *pci_dev)
1074 {
1075     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1076
1077     virtio_pci_stop_ioeventfd(proxy);
1078     virtio_rng_exit(proxy->vdev);
1079     virtio_exit_pci(pci_dev);
1080 }
1081
1082 static Property virtio_blk_properties[] = {
1083     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
1084     DEFINE_BLOCK_PROPERTIES(VirtIOPCIProxy, blk.conf),
1085     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOPCIProxy, blk.conf),
1086     DEFINE_PROP_STRING("serial", VirtIOPCIProxy, blk.serial),
1087 #ifdef __linux__
1088     DEFINE_PROP_BIT("scsi", VirtIOPCIProxy, blk.scsi, 0, true),
1089 #endif
1090     DEFINE_PROP_BIT("config-wce", VirtIOPCIProxy, blk.config_wce, 0, true),
1091     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
1092 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
1093     DEFINE_PROP_BIT("x-data-plane", VirtIOPCIProxy, blk.data_plane, 0, false),
1094 #endif
1095     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
1096     DEFINE_VIRTIO_BLK_FEATURES(VirtIOPCIProxy, host_features),
1097     DEFINE_PROP_END_OF_LIST(),
1098 };
1099
1100 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1101 {
1102     DeviceClass *dc = DEVICE_CLASS(klass);
1103     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1104
1105     k->init = virtio_blk_init_pci;
1106     k->exit = virtio_blk_exit_pci;
1107     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1108     k->device_id = PCI_DEVICE_ID_VIRTIO_BLOCK;
1109     k->revision = VIRTIO_PCI_ABI_VERSION;
1110     k->class_id = PCI_CLASS_STORAGE_SCSI;
1111     dc->reset = virtio_pci_reset;
1112     dc->props = virtio_blk_properties;
1113 }
1114
1115 static const TypeInfo virtio_blk_info = {
1116     .name          = "virtio-blk-pci",
1117     .parent        = TYPE_PCI_DEVICE,
1118     .instance_size = sizeof(VirtIOPCIProxy),
1119     .class_init    = virtio_blk_class_init,
1120 };
1121
1122 static Property virtio_net_properties[] = {
1123     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
1124     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
1125     DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
1126     DEFINE_NIC_PROPERTIES(VirtIOPCIProxy, nic),
1127     DEFINE_PROP_UINT32("x-txtimer", VirtIOPCIProxy, net.txtimer, TX_TIMER_INTERVAL),
1128     DEFINE_PROP_INT32("x-txburst", VirtIOPCIProxy, net.txburst, TX_BURST),
1129     DEFINE_PROP_STRING("tx", VirtIOPCIProxy, net.tx),
1130     DEFINE_PROP_END_OF_LIST(),
1131 };
1132
1133 static void virtio_net_class_init(ObjectClass *klass, void *data)
1134 {
1135     DeviceClass *dc = DEVICE_CLASS(klass);
1136     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1137
1138     k->init = virtio_net_init_pci;
1139     k->exit = virtio_net_exit_pci;
1140     k->romfile = "pxe-virtio.rom";
1141     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1142     k->device_id = PCI_DEVICE_ID_VIRTIO_NET;
1143     k->revision = VIRTIO_PCI_ABI_VERSION;
1144     k->class_id = PCI_CLASS_NETWORK_ETHERNET;
1145     dc->reset = virtio_pci_reset;
1146     dc->props = virtio_net_properties;
1147 }
1148
1149 static const TypeInfo virtio_net_info = {
1150     .name          = "virtio-net-pci",
1151     .parent        = TYPE_PCI_DEVICE,
1152     .instance_size = sizeof(VirtIOPCIProxy),
1153     .class_init    = virtio_net_class_init,
1154 };
1155
1156 static Property virtio_serial_properties[] = {
1157     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
1158     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, DEV_NVECTORS_UNSPECIFIED),
1159     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
1160     DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
1161     DEFINE_PROP_UINT32("max_ports", VirtIOPCIProxy, serial.max_virtserial_ports, 31),
1162     DEFINE_PROP_END_OF_LIST(),
1163 };
1164
1165 static void virtio_serial_class_init(ObjectClass *klass, void *data)
1166 {
1167     DeviceClass *dc = DEVICE_CLASS(klass);
1168     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1169
1170     k->init = virtio_serial_init_pci;
1171     k->exit = virtio_serial_exit_pci;
1172     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1173     k->device_id = PCI_DEVICE_ID_VIRTIO_CONSOLE;
1174     k->revision = VIRTIO_PCI_ABI_VERSION;
1175     k->class_id = PCI_CLASS_COMMUNICATION_OTHER;
1176     dc->reset = virtio_pci_reset;
1177     dc->props = virtio_serial_properties;
1178 }
1179
1180 static const TypeInfo virtio_serial_info = {
1181     .name          = "virtio-serial-pci",
1182     .parent        = TYPE_PCI_DEVICE,
1183     .instance_size = sizeof(VirtIOPCIProxy),
1184     .class_init    = virtio_serial_class_init,
1185 };
1186
1187 static Property virtio_balloon_properties[] = {
1188     DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
1189     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
1190     DEFINE_PROP_END_OF_LIST(),
1191 };
1192
1193 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
1194 {
1195     DeviceClass *dc = DEVICE_CLASS(klass);
1196     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1197
1198     k->init = virtio_balloon_init_pci;
1199     k->exit = virtio_balloon_exit_pci;
1200     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1201     k->device_id = PCI_DEVICE_ID_VIRTIO_BALLOON;
1202     k->revision = VIRTIO_PCI_ABI_VERSION;
1203     k->class_id = PCI_CLASS_OTHERS;
1204     dc->reset = virtio_pci_reset;
1205     dc->props = virtio_balloon_properties;
1206 }
1207
1208 static const TypeInfo virtio_balloon_info = {
1209     .name          = "virtio-balloon-pci",
1210     .parent        = TYPE_PCI_DEVICE,
1211     .instance_size = sizeof(VirtIOPCIProxy),
1212     .class_init    = virtio_balloon_class_init,
1213 };
1214
1215 static void virtio_rng_initfn(Object *obj)
1216 {
1217     PCIDevice *pci_dev = PCI_DEVICE(obj);
1218     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1219
1220     object_property_add_link(obj, "rng", TYPE_RNG_BACKEND,
1221                              (Object **)&proxy->rng.rng, NULL);
1222 }
1223
1224 static Property virtio_rng_properties[] = {
1225     DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
1226     /* Set a default rate limit of 2^47 bytes per minute or roughly 2TB/s.  If
1227        you have an entropy source capable of generating more entropy than this
1228        and you can pass it through via virtio-rng, then hats off to you.  Until
1229        then, this is unlimited for all practical purposes.
1230     */
1231     DEFINE_PROP_UINT64("max-bytes", VirtIOPCIProxy, rng.max_bytes, INT64_MAX),
1232     DEFINE_PROP_UINT32("period", VirtIOPCIProxy, rng.period_ms, 1 << 16),
1233     DEFINE_PROP_END_OF_LIST(),
1234 };
1235
1236 static void virtio_rng_class_init(ObjectClass *klass, void *data)
1237 {
1238     DeviceClass *dc = DEVICE_CLASS(klass);
1239     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1240
1241     k->init = virtio_rng_init_pci;
1242     k->exit = virtio_rng_exit_pci;
1243     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1244     k->device_id = PCI_DEVICE_ID_VIRTIO_RNG;
1245     k->revision = VIRTIO_PCI_ABI_VERSION;
1246     k->class_id = PCI_CLASS_OTHERS;
1247     dc->reset = virtio_pci_reset;
1248     dc->props = virtio_rng_properties;
1249 }
1250
1251 static const TypeInfo virtio_rng_info = {
1252     .name          = "virtio-rng-pci",
1253     .parent        = TYPE_PCI_DEVICE,
1254     .instance_size = sizeof(VirtIOPCIProxy),
1255     .instance_init = virtio_rng_initfn,
1256     .class_init    = virtio_rng_class_init,
1257 };
1258
1259 static int virtio_scsi_init_pci(PCIDevice *pci_dev)
1260 {
1261     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1262     VirtIODevice *vdev;
1263
1264     vdev = virtio_scsi_init(&pci_dev->qdev, &proxy->scsi);
1265     if (!vdev) {
1266         return -EINVAL;
1267     }
1268
1269     vdev->nvectors = proxy->nvectors == DEV_NVECTORS_UNSPECIFIED
1270                                         ? proxy->scsi.num_queues + 3
1271                                         : proxy->nvectors;
1272     virtio_init_pci(proxy, vdev);
1273
1274     /* make the actual value visible */
1275     proxy->nvectors = vdev->nvectors;
1276     return 0;
1277 }
1278
1279 static void virtio_scsi_exit_pci(PCIDevice *pci_dev)
1280 {
1281     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
1282
1283     virtio_scsi_exit(proxy->vdev);
1284     virtio_exit_pci(pci_dev);
1285 }
1286
1287 static Property virtio_scsi_properties[] = {
1288     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
1289     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, DEV_NVECTORS_UNSPECIFIED),
1290     DEFINE_VIRTIO_SCSI_PROPERTIES(VirtIOPCIProxy, host_features, scsi),
1291     DEFINE_PROP_END_OF_LIST(),
1292 };
1293
1294 static void virtio_scsi_class_init(ObjectClass *klass, void *data)
1295 {
1296     DeviceClass *dc = DEVICE_CLASS(klass);
1297     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1298
1299     k->init = virtio_scsi_init_pci;
1300     k->exit = virtio_scsi_exit_pci;
1301     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1302     k->device_id = PCI_DEVICE_ID_VIRTIO_SCSI;
1303     k->revision = 0x00;
1304     k->class_id = PCI_CLASS_STORAGE_SCSI;
1305     dc->reset = virtio_pci_reset;
1306     dc->props = virtio_scsi_properties;
1307 }
1308
1309 static const TypeInfo virtio_scsi_info = {
1310     .name          = "virtio-scsi-pci",
1311     .parent        = TYPE_PCI_DEVICE,
1312     .instance_size = sizeof(VirtIOPCIProxy),
1313     .class_init    = virtio_scsi_class_init,
1314 };
1315
1316 /*
1317  * virtio-pci: This is the PCIDevice which has a virtio-pci-bus.
1318  */
1319
1320 /* This is called by virtio-bus just after the device is plugged. */
1321 static void virtio_pci_device_plugged(DeviceState *d)
1322 {
1323     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1324     VirtioBusState *bus = &proxy->bus;
1325     uint8_t *config;
1326     uint32_t size;
1327
1328     proxy->vdev = bus->vdev;
1329
1330     config = proxy->pci_dev.config;
1331     if (proxy->class_code) {
1332         pci_config_set_class(config, proxy->class_code);
1333     }
1334     pci_set_word(config + PCI_SUBSYSTEM_VENDOR_ID,
1335                  pci_get_word(config + PCI_VENDOR_ID));
1336     pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus));
1337     config[PCI_INTERRUPT_PIN] = 1;
1338
1339     if (proxy->nvectors &&
1340         msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors, 1)) {
1341         proxy->nvectors = 0;
1342     }
1343
1344     proxy->pci_dev.config_write = virtio_write_config;
1345
1346     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev)
1347          + virtio_bus_get_vdev_config_len(bus);
1348     if (size & (size - 1)) {
1349         size = 1 << qemu_fls(size);
1350     }
1351
1352     memory_region_init_io(&proxy->bar, &virtio_pci_config_ops, proxy,
1353                           "virtio-pci", size);
1354     pci_register_bar(&proxy->pci_dev, 0, PCI_BASE_ADDRESS_SPACE_IO,
1355                      &proxy->bar);
1356
1357     if (!kvm_has_many_ioeventfds()) {
1358         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1359     }
1360
1361     proxy->host_features |= 0x1 << VIRTIO_F_NOTIFY_ON_EMPTY;
1362     proxy->host_features |= 0x1 << VIRTIO_F_BAD_FEATURE;
1363     proxy->host_features = virtio_bus_get_vdev_features(bus,
1364                                                       proxy->host_features);
1365 }
1366
1367 /* This is called by virtio-bus just before the device is unplugged. */
1368 static void virtio_pci_device_unplug(DeviceState *d)
1369 {
1370     VirtIOPCIProxy *dev = VIRTIO_PCI(d);
1371     virtio_pci_stop_ioeventfd(dev);
1372 }
1373
1374 static int virtio_pci_init(PCIDevice *pci_dev)
1375 {
1376     VirtIOPCIProxy *dev = VIRTIO_PCI(pci_dev);
1377     VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev);
1378     virtio_pci_bus_new(&dev->bus, dev);
1379     if (k->init != NULL) {
1380         return k->init(dev);
1381     }
1382     return 0;
1383 }
1384
1385 static void virtio_pci_exit(PCIDevice *pci_dev)
1386 {
1387     VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1388     VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
1389     BusState *qbus = BUS(&proxy->bus);
1390     virtio_bus_destroy_device(bus);
1391     qbus_free(qbus);
1392     virtio_exit_pci(pci_dev);
1393 }
1394
1395 /*
1396  * This will be renamed virtio_pci_reset at the end of the series.
1397  * virtio_pci_reset is still in use at this moment.
1398  */
1399 static void virtio_pci_rst(DeviceState *qdev)
1400 {
1401     VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1402     VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
1403     virtio_pci_stop_ioeventfd(proxy);
1404     virtio_bus_reset(bus);
1405     msix_unuse_all_vectors(&proxy->pci_dev);
1406     proxy->flags &= ~VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
1407 }
1408
1409 static void virtio_pci_class_init(ObjectClass *klass, void *data)
1410 {
1411     DeviceClass *dc = DEVICE_CLASS(klass);
1412     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1413
1414     k->init = virtio_pci_init;
1415     k->exit = virtio_pci_exit;
1416     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
1417     k->revision = VIRTIO_PCI_ABI_VERSION;
1418     k->class_id = PCI_CLASS_OTHERS;
1419     dc->reset = virtio_pci_rst;
1420 }
1421
1422 static const TypeInfo virtio_pci_info = {
1423     .name          = TYPE_VIRTIO_PCI,
1424     .parent        = TYPE_PCI_DEVICE,
1425     .instance_size = sizeof(VirtIOPCIProxy),
1426     .class_init    = virtio_pci_class_init,
1427     .class_size    = sizeof(VirtioPCIClass),
1428     .abstract      = true,
1429 };
1430
1431 /* virtio-pci-bus */
1432
1433 void virtio_pci_bus_new(VirtioBusState *bus, VirtIOPCIProxy *dev)
1434 {
1435     DeviceState *qdev = DEVICE(dev);
1436     BusState *qbus;
1437     qbus_create_inplace((BusState *)bus, TYPE_VIRTIO_PCI_BUS, qdev, NULL);
1438     qbus = BUS(bus);
1439     qbus->allow_hotplug = 0;
1440 }
1441
1442 static void virtio_pci_bus_class_init(ObjectClass *klass, void *data)
1443 {
1444     BusClass *bus_class = BUS_CLASS(klass);
1445     VirtioBusClass *k = VIRTIO_BUS_CLASS(klass);
1446     bus_class->max_dev = 1;
1447     k->notify = virtio_pci_notify;
1448     k->save_config = virtio_pci_save_config;
1449     k->load_config = virtio_pci_load_config;
1450     k->save_queue = virtio_pci_save_queue;
1451     k->load_queue = virtio_pci_load_queue;
1452     k->get_features = virtio_pci_get_features;
1453     k->query_guest_notifiers = virtio_pci_query_guest_notifiers;
1454     k->set_host_notifier = virtio_pci_set_host_notifier;
1455     k->set_guest_notifiers = virtio_pci_set_guest_notifiers;
1456     k->vmstate_change = virtio_pci_vmstate_change;
1457     k->device_plugged = virtio_pci_device_plugged;
1458     k->device_unplug = virtio_pci_device_unplug;
1459 }
1460
1461 static const TypeInfo virtio_pci_bus_info = {
1462     .name          = TYPE_VIRTIO_PCI_BUS,
1463     .parent        = TYPE_VIRTIO_BUS,
1464     .instance_size = sizeof(VirtioPCIBusState),
1465     .class_init    = virtio_pci_bus_class_init,
1466 };
1467
1468 static void virtio_pci_register_types(void)
1469 {
1470     type_register_static(&virtio_blk_info);
1471     type_register_static(&virtio_net_info);
1472     type_register_static(&virtio_serial_info);
1473     type_register_static(&virtio_balloon_info);
1474     type_register_static(&virtio_scsi_info);
1475     type_register_static(&virtio_rng_info);
1476     type_register_static(&virtio_pci_bus_info);
1477     type_register_static(&virtio_pci_info);
1478 }
1479
1480 type_init(virtio_pci_register_types)