qemu/hw/virtio/virtio-pci.c
<<
>>
Prefs
   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 "qemu/osdep.h"
  19
  20#include "exec/memop.h"
  21#include "standard-headers/linux/virtio_pci.h"
  22#include "hw/boards.h"
  23#include "hw/virtio/virtio.h"
  24#include "migration/qemu-file-types.h"
  25#include "hw/pci/pci.h"
  26#include "hw/pci/pci_bus.h"
  27#include "hw/qdev-properties.h"
  28#include "qapi/error.h"
  29#include "qemu/error-report.h"
  30#include "qemu/log.h"
  31#include "qemu/module.h"
  32#include "hw/pci/msi.h"
  33#include "hw/pci/msix.h"
  34#include "hw/loader.h"
  35#include "sysemu/kvm.h"
  36#include "virtio-pci.h"
  37#include "qemu/range.h"
  38#include "hw/virtio/virtio-bus.h"
  39#include "qapi/visitor.h"
  40#include "sysemu/replay.h"
  41
  42#define VIRTIO_PCI_REGION_SIZE(dev)     VIRTIO_PCI_CONFIG_OFF(msix_present(dev))
  43
  44#undef VIRTIO_PCI_CONFIG
  45
  46/* The remaining space is defined by each driver as the per-driver
  47 * configuration space */
  48#define VIRTIO_PCI_CONFIG_SIZE(dev)     VIRTIO_PCI_CONFIG_OFF(msix_enabled(dev))
  49
  50static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
  51                               VirtIOPCIProxy *dev);
  52static void virtio_pci_reset(DeviceState *qdev);
  53
  54/* virtio device */
  55/* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */
  56static inline VirtIOPCIProxy *to_virtio_pci_proxy(DeviceState *d)
  57{
  58    return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
  59}
  60
  61/* DeviceState to VirtIOPCIProxy. Note: used on datapath,
  62 * be careful and test performance if you change this.
  63 */
  64static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d)
  65{
  66    return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
  67}
  68
  69static void virtio_pci_notify(DeviceState *d, uint16_t vector)
  70{
  71    VirtIOPCIProxy *proxy = to_virtio_pci_proxy_fast(d);
  72
  73    if (msix_enabled(&proxy->pci_dev))
  74        msix_notify(&proxy->pci_dev, vector);
  75    else {
  76        VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
  77        pci_set_irq(&proxy->pci_dev, qatomic_read(&vdev->isr) & 1);
  78    }
  79}
  80
  81static void virtio_pci_save_config(DeviceState *d, QEMUFile *f)
  82{
  83    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
  84    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
  85
  86    pci_device_save(&proxy->pci_dev, f);
  87    msix_save(&proxy->pci_dev, f);
  88    if (msix_present(&proxy->pci_dev))
  89        qemu_put_be16(f, vdev->config_vector);
  90}
  91
  92static const VMStateDescription vmstate_virtio_pci_modern_queue_state = {
  93    .name = "virtio_pci/modern_queue_state",
  94    .version_id = 1,
  95    .minimum_version_id = 1,
  96    .fields = (VMStateField[]) {
  97        VMSTATE_UINT16(num, VirtIOPCIQueue),
  98        VMSTATE_UNUSED(1), /* enabled was stored as be16 */
  99        VMSTATE_BOOL(enabled, VirtIOPCIQueue),
 100        VMSTATE_UINT32_ARRAY(desc, VirtIOPCIQueue, 2),
 101        VMSTATE_UINT32_ARRAY(avail, VirtIOPCIQueue, 2),
 102        VMSTATE_UINT32_ARRAY(used, VirtIOPCIQueue, 2),
 103        VMSTATE_END_OF_LIST()
 104    }
 105};
 106
 107static bool virtio_pci_modern_state_needed(void *opaque)
 108{
 109    VirtIOPCIProxy *proxy = opaque;
 110
 111    return virtio_pci_modern(proxy);
 112}
 113
 114static const VMStateDescription vmstate_virtio_pci_modern_state_sub = {
 115    .name = "virtio_pci/modern_state",
 116    .version_id = 1,
 117    .minimum_version_id = 1,
 118    .needed = &virtio_pci_modern_state_needed,
 119    .fields = (VMStateField[]) {
 120        VMSTATE_UINT32(dfselect, VirtIOPCIProxy),
 121        VMSTATE_UINT32(gfselect, VirtIOPCIProxy),
 122        VMSTATE_UINT32_ARRAY(guest_features, VirtIOPCIProxy, 2),
 123        VMSTATE_STRUCT_ARRAY(vqs, VirtIOPCIProxy, VIRTIO_QUEUE_MAX, 0,
 124                             vmstate_virtio_pci_modern_queue_state,
 125                             VirtIOPCIQueue),
 126        VMSTATE_END_OF_LIST()
 127    }
 128};
 129
 130static const VMStateDescription vmstate_virtio_pci = {
 131    .name = "virtio_pci",
 132    .version_id = 1,
 133    .minimum_version_id = 1,
 134    .minimum_version_id_old = 1,
 135    .fields = (VMStateField[]) {
 136        VMSTATE_END_OF_LIST()
 137    },
 138    .subsections = (const VMStateDescription*[]) {
 139        &vmstate_virtio_pci_modern_state_sub,
 140        NULL
 141    }
 142};
 143
 144static bool virtio_pci_has_extra_state(DeviceState *d)
 145{
 146    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 147
 148    return proxy->flags & VIRTIO_PCI_FLAG_MIGRATE_EXTRA;
 149}
 150
 151static void virtio_pci_save_extra_state(DeviceState *d, QEMUFile *f)
 152{
 153    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 154
 155    vmstate_save_state(f, &vmstate_virtio_pci, proxy, NULL);
 156}
 157
 158static int virtio_pci_load_extra_state(DeviceState *d, QEMUFile *f)
 159{
 160    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 161
 162    return vmstate_load_state(f, &vmstate_virtio_pci, proxy, 1);
 163}
 164
 165static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f)
 166{
 167    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 168    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 169
 170    if (msix_present(&proxy->pci_dev))
 171        qemu_put_be16(f, virtio_queue_vector(vdev, n));
 172}
 173
 174static int virtio_pci_load_config(DeviceState *d, QEMUFile *f)
 175{
 176    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 177    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 178
 179    int ret;
 180    ret = pci_device_load(&proxy->pci_dev, f);
 181    if (ret) {
 182        return ret;
 183    }
 184    msix_unuse_all_vectors(&proxy->pci_dev);
 185    msix_load(&proxy->pci_dev, f);
 186    if (msix_present(&proxy->pci_dev)) {
 187        qemu_get_be16s(f, &vdev->config_vector);
 188    } else {
 189        vdev->config_vector = VIRTIO_NO_VECTOR;
 190    }
 191    if (vdev->config_vector != VIRTIO_NO_VECTOR) {
 192        return msix_vector_use(&proxy->pci_dev, vdev->config_vector);
 193    }
 194    return 0;
 195}
 196
 197static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f)
 198{
 199    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 200    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 201
 202    uint16_t vector;
 203    if (msix_present(&proxy->pci_dev)) {
 204        qemu_get_be16s(f, &vector);
 205    } else {
 206        vector = VIRTIO_NO_VECTOR;
 207    }
 208    virtio_queue_set_vector(vdev, n, vector);
 209    if (vector != VIRTIO_NO_VECTOR) {
 210        return msix_vector_use(&proxy->pci_dev, vector);
 211    }
 212
 213    return 0;
 214}
 215
 216static bool virtio_pci_ioeventfd_enabled(DeviceState *d)
 217{
 218    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 219
 220    return (proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) != 0;
 221}
 222
 223#define QEMU_VIRTIO_PCI_QUEUE_MEM_MULT 0x1000
 224
 225static inline int virtio_pci_queue_mem_mult(struct VirtIOPCIProxy *proxy)
 226{
 227    return (proxy->flags & VIRTIO_PCI_FLAG_PAGE_PER_VQ) ?
 228        QEMU_VIRTIO_PCI_QUEUE_MEM_MULT : 4;
 229}
 230
 231static int virtio_pci_ioeventfd_assign(DeviceState *d, EventNotifier *notifier,
 232                                       int n, bool assign)
 233{
 234    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 235    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 236    VirtQueue *vq = virtio_get_queue(vdev, n);
 237    bool legacy = virtio_pci_legacy(proxy);
 238    bool modern = virtio_pci_modern(proxy);
 239    bool fast_mmio = kvm_ioeventfd_any_length_enabled();
 240    bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
 241    MemoryRegion *modern_mr = &proxy->notify.mr;
 242    MemoryRegion *modern_notify_mr = &proxy->notify_pio.mr;
 243    MemoryRegion *legacy_mr = &proxy->bar;
 244    hwaddr modern_addr = virtio_pci_queue_mem_mult(proxy) *
 245                         virtio_get_queue_index(vq);
 246    hwaddr legacy_addr = VIRTIO_PCI_QUEUE_NOTIFY;
 247
 248    if (assign) {
 249        if (modern) {
 250            if (fast_mmio) {
 251                memory_region_add_eventfd(modern_mr, modern_addr, 0,
 252                                          false, n, notifier);
 253            } else {
 254                memory_region_add_eventfd(modern_mr, modern_addr, 2,
 255                                          false, n, notifier);
 256            }
 257            if (modern_pio) {
 258                memory_region_add_eventfd(modern_notify_mr, 0, 2,
 259                                              true, n, notifier);
 260            }
 261        }
 262        if (legacy) {
 263            memory_region_add_eventfd(legacy_mr, legacy_addr, 2,
 264                                      true, n, notifier);
 265        }
 266    } else {
 267        if (modern) {
 268            if (fast_mmio) {
 269                memory_region_del_eventfd(modern_mr, modern_addr, 0,
 270                                          false, n, notifier);
 271            } else {
 272                memory_region_del_eventfd(modern_mr, modern_addr, 2,
 273                                          false, n, notifier);
 274            }
 275            if (modern_pio) {
 276                memory_region_del_eventfd(modern_notify_mr, 0, 2,
 277                                          true, n, notifier);
 278            }
 279        }
 280        if (legacy) {
 281            memory_region_del_eventfd(legacy_mr, legacy_addr, 2,
 282                                      true, n, notifier);
 283        }
 284    }
 285    return 0;
 286}
 287
 288static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 289{
 290    virtio_bus_start_ioeventfd(&proxy->bus);
 291}
 292
 293static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 294{
 295    virtio_bus_stop_ioeventfd(&proxy->bus);
 296}
 297
 298static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 299{
 300    VirtIOPCIProxy *proxy = opaque;
 301    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 302    hwaddr pa;
 303
 304    switch (addr) {
 305    case VIRTIO_PCI_GUEST_FEATURES:
 306        /* Guest does not negotiate properly?  We have to assume nothing. */
 307        if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 308            val = virtio_bus_get_vdev_bad_features(&proxy->bus);
 309        }
 310        virtio_set_features(vdev, val);
 311        break;
 312    case VIRTIO_PCI_QUEUE_PFN:
 313        pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 314        if (pa == 0) {
 315            virtio_pci_reset(DEVICE(proxy));
 316        }
 317        else
 318            virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 319        break;
 320    case VIRTIO_PCI_QUEUE_SEL:
 321        if (val < VIRTIO_QUEUE_MAX)
 322            vdev->queue_sel = val;
 323        break;
 324    case VIRTIO_PCI_QUEUE_NOTIFY:
 325        if (val < VIRTIO_QUEUE_MAX) {
 326            virtio_queue_notify(vdev, val);
 327        }
 328        break;
 329    case VIRTIO_PCI_STATUS:
 330        if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
 331            virtio_pci_stop_ioeventfd(proxy);
 332        }
 333
 334        virtio_set_status(vdev, val & 0xFF);
 335
 336        if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
 337            virtio_pci_start_ioeventfd(proxy);
 338        }
 339
 340        if (vdev->status == 0) {
 341            virtio_pci_reset(DEVICE(proxy));
 342        }
 343
 344        /* Linux before 2.6.34 drives the device without enabling
 345           the PCI device bus master bit. Enable it automatically
 346           for the guest. This is a PCI spec violation but so is
 347           initiating DMA with bus master bit clear. */
 348        if (val == (VIRTIO_CONFIG_S_ACKNOWLEDGE | VIRTIO_CONFIG_S_DRIVER)) {
 349            pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
 350                                     proxy->pci_dev.config[PCI_COMMAND] |
 351                                     PCI_COMMAND_MASTER, 1);
 352        }
 353        break;
 354    case VIRTIO_MSI_CONFIG_VECTOR:
 355        msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 356        /* Make it possible for guest to discover an error took place. */
 357        if (msix_vector_use(&proxy->pci_dev, val) < 0)
 358            val = VIRTIO_NO_VECTOR;
 359        vdev->config_vector = val;
 360        break;
 361    case VIRTIO_MSI_QUEUE_VECTOR:
 362        msix_vector_unuse(&proxy->pci_dev,
 363                          virtio_queue_vector(vdev, vdev->queue_sel));
 364        /* Make it possible for guest to discover an error took place. */
 365        if (msix_vector_use(&proxy->pci_dev, val) < 0)
 366            val = VIRTIO_NO_VECTOR;
 367        virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 368        break;
 369    default:
 370        qemu_log_mask(LOG_GUEST_ERROR,
 371                      "%s: unexpected address 0x%x value 0x%x\n",
 372                      __func__, addr, val);
 373        break;
 374    }
 375}
 376
 377static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 378{
 379    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 380    uint32_t ret = 0xFFFFFFFF;
 381
 382    switch (addr) {
 383    case VIRTIO_PCI_HOST_FEATURES:
 384        ret = vdev->host_features;
 385        break;
 386    case VIRTIO_PCI_GUEST_FEATURES:
 387        ret = vdev->guest_features;
 388        break;
 389    case VIRTIO_PCI_QUEUE_PFN:
 390        ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 391              >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 392        break;
 393    case VIRTIO_PCI_QUEUE_NUM:
 394        ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 395        break;
 396    case VIRTIO_PCI_QUEUE_SEL:
 397        ret = vdev->queue_sel;
 398        break;
 399    case VIRTIO_PCI_STATUS:
 400        ret = vdev->status;
 401        break;
 402    case VIRTIO_PCI_ISR:
 403        /* reading from the ISR also clears it. */
 404        ret = qatomic_xchg(&vdev->isr, 0);
 405        pci_irq_deassert(&proxy->pci_dev);
 406        break;
 407    case VIRTIO_MSI_CONFIG_VECTOR:
 408        ret = vdev->config_vector;
 409        break;
 410    case VIRTIO_MSI_QUEUE_VECTOR:
 411        ret = virtio_queue_vector(vdev, vdev->queue_sel);
 412        break;
 413    default:
 414        break;
 415    }
 416
 417    return ret;
 418}
 419
 420static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr,
 421                                       unsigned size)
 422{
 423    VirtIOPCIProxy *proxy = opaque;
 424    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 425    uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
 426    uint64_t val = 0;
 427
 428    if (vdev == NULL) {
 429        return UINT64_MAX;
 430    }
 431
 432    if (addr < config) {
 433        return virtio_ioport_read(proxy, addr);
 434    }
 435    addr -= config;
 436
 437    switch (size) {
 438    case 1:
 439        val = virtio_config_readb(vdev, addr);
 440        break;
 441    case 2:
 442        val = virtio_config_readw(vdev, addr);
 443        if (virtio_is_big_endian(vdev)) {
 444            val = bswap16(val);
 445        }
 446        break;
 447    case 4:
 448        val = virtio_config_readl(vdev, addr);
 449        if (virtio_is_big_endian(vdev)) {
 450            val = bswap32(val);
 451        }
 452        break;
 453    }
 454    return val;
 455}
 456
 457static void virtio_pci_config_write(void *opaque, hwaddr addr,
 458                                    uint64_t val, unsigned size)
 459{
 460    VirtIOPCIProxy *proxy = opaque;
 461    uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
 462    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 463
 464    if (vdev == NULL) {
 465        return;
 466    }
 467
 468    if (addr < config) {
 469        virtio_ioport_write(proxy, addr, val);
 470        return;
 471    }
 472    addr -= config;
 473    /*
 474     * Virtio-PCI is odd. Ioports are LE but config space is target native
 475     * endian.
 476     */
 477    switch (size) {
 478    case 1:
 479        virtio_config_writeb(vdev, addr, val);
 480        break;
 481    case 2:
 482        if (virtio_is_big_endian(vdev)) {
 483            val = bswap16(val);
 484        }
 485        virtio_config_writew(vdev, addr, val);
 486        break;
 487    case 4:
 488        if (virtio_is_big_endian(vdev)) {
 489            val = bswap32(val);
 490        }
 491        virtio_config_writel(vdev, addr, val);
 492        break;
 493    }
 494}
 495
 496static const MemoryRegionOps virtio_pci_config_ops = {
 497    .read = virtio_pci_config_read,
 498    .write = virtio_pci_config_write,
 499    .impl = {
 500        .min_access_size = 1,
 501        .max_access_size = 4,
 502    },
 503    .endianness = DEVICE_LITTLE_ENDIAN,
 504};
 505
 506static MemoryRegion *virtio_address_space_lookup(VirtIOPCIProxy *proxy,
 507                                                 hwaddr *off, int len)
 508{
 509    int i;
 510    VirtIOPCIRegion *reg;
 511
 512    for (i = 0; i < ARRAY_SIZE(proxy->regs); ++i) {
 513        reg = &proxy->regs[i];
 514        if (*off >= reg->offset &&
 515            *off + len <= reg->offset + reg->size) {
 516            *off -= reg->offset;
 517            return &reg->mr;
 518        }
 519    }
 520
 521    return NULL;
 522}
 523
 524/* Below are generic functions to do memcpy from/to an address space,
 525 * without byteswaps, with input validation.
 526 *
 527 * As regular address_space_* APIs all do some kind of byteswap at least for
 528 * some host/target combinations, we are forced to explicitly convert to a
 529 * known-endianness integer value.
 530 * It doesn't really matter which endian format to go through, so the code
 531 * below selects the endian that causes the least amount of work on the given
 532 * host.
 533 *
 534 * Note: host pointer must be aligned.
 535 */
 536static
 537void virtio_address_space_write(VirtIOPCIProxy *proxy, hwaddr addr,
 538                                const uint8_t *buf, int len)
 539{
 540    uint64_t val;
 541    MemoryRegion *mr;
 542
 543    /* address_space_* APIs assume an aligned address.
 544     * As address is under guest control, handle illegal values.
 545     */
 546    addr &= ~(len - 1);
 547
 548    mr = virtio_address_space_lookup(proxy, &addr, len);
 549    if (!mr) {
 550        return;
 551    }
 552
 553    /* Make sure caller aligned buf properly */
 554    assert(!(((uintptr_t)buf) & (len - 1)));
 555
 556    switch (len) {
 557    case 1:
 558        val = pci_get_byte(buf);
 559        break;
 560    case 2:
 561        val = pci_get_word(buf);
 562        break;
 563    case 4:
 564        val = pci_get_long(buf);
 565        break;
 566    default:
 567        /* As length is under guest control, handle illegal values. */
 568        return;
 569    }
 570    memory_region_dispatch_write(mr, addr, val, size_memop(len) | MO_LE,
 571                                 MEMTXATTRS_UNSPECIFIED);
 572}
 573
 574static void
 575virtio_address_space_read(VirtIOPCIProxy *proxy, hwaddr addr,
 576                          uint8_t *buf, int len)
 577{
 578    uint64_t val;
 579    MemoryRegion *mr;
 580
 581    /* address_space_* APIs assume an aligned address.
 582     * As address is under guest control, handle illegal values.
 583     */
 584    addr &= ~(len - 1);
 585
 586    mr = virtio_address_space_lookup(proxy, &addr, len);
 587    if (!mr) {
 588        return;
 589    }
 590
 591    /* Make sure caller aligned buf properly */
 592    assert(!(((uintptr_t)buf) & (len - 1)));
 593
 594    memory_region_dispatch_read(mr, addr, &val, size_memop(len) | MO_LE,
 595                                MEMTXATTRS_UNSPECIFIED);
 596    switch (len) {
 597    case 1:
 598        pci_set_byte(buf, val);
 599        break;
 600    case 2:
 601        pci_set_word(buf, val);
 602        break;
 603    case 4:
 604        pci_set_long(buf, val);
 605        break;
 606    default:
 607        /* As length is under guest control, handle illegal values. */
 608        break;
 609    }
 610}
 611
 612static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 613                                uint32_t val, int len)
 614{
 615    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
 616    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 617    struct virtio_pci_cfg_cap *cfg;
 618
 619    pci_default_write_config(pci_dev, address, val, len);
 620
 621    if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
 622        pcie_cap_flr_write_config(pci_dev, address, val, len);
 623    }
 624
 625    if (range_covers_byte(address, len, PCI_COMMAND)) {
 626        if (!(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 627            virtio_set_disabled(vdev, true);
 628            virtio_pci_stop_ioeventfd(proxy);
 629            virtio_set_status(vdev, vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 630        } else {
 631            virtio_set_disabled(vdev, false);
 632        }
 633    }
 634
 635    if (proxy->config_cap &&
 636        ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
 637                                                                  pci_cfg_data),
 638                       sizeof cfg->pci_cfg_data)) {
 639        uint32_t off;
 640        uint32_t len;
 641
 642        cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
 643        off = le32_to_cpu(cfg->cap.offset);
 644        len = le32_to_cpu(cfg->cap.length);
 645
 646        if (len == 1 || len == 2 || len == 4) {
 647            assert(len <= sizeof cfg->pci_cfg_data);
 648            virtio_address_space_write(proxy, off, cfg->pci_cfg_data, len);
 649        }
 650    }
 651}
 652
 653static uint32_t virtio_read_config(PCIDevice *pci_dev,
 654                                   uint32_t address, int len)
 655{
 656    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
 657    struct virtio_pci_cfg_cap *cfg;
 658
 659    if (proxy->config_cap &&
 660        ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
 661                                                                  pci_cfg_data),
 662                       sizeof cfg->pci_cfg_data)) {
 663        uint32_t off;
 664        uint32_t len;
 665
 666        cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
 667        off = le32_to_cpu(cfg->cap.offset);
 668        len = le32_to_cpu(cfg->cap.length);
 669
 670        if (len == 1 || len == 2 || len == 4) {
 671            assert(len <= sizeof cfg->pci_cfg_data);
 672            virtio_address_space_read(proxy, off, cfg->pci_cfg_data, len);
 673        }
 674    }
 675
 676    return pci_default_read_config(pci_dev, address, len);
 677}
 678
 679static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy,
 680                                        unsigned int queue_no,
 681                                        unsigned int vector)
 682{
 683    VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 684    int ret;
 685
 686    if (irqfd->users == 0) {
 687        ret = kvm_irqchip_add_msi_route(kvm_state, vector, &proxy->pci_dev);
 688        if (ret < 0) {
 689            return ret;
 690        }
 691        irqfd->virq = ret;
 692    }
 693    irqfd->users++;
 694    return 0;
 695}
 696
 697static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy,
 698                                             unsigned int vector)
 699{
 700    VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 701    if (--irqfd->users == 0) {
 702        kvm_irqchip_release_virq(kvm_state, irqfd->virq);
 703    }
 704}
 705
 706static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy,
 707                                 unsigned int queue_no,
 708                                 unsigned int vector)
 709{
 710    VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 711    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 712    VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 713    EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 714    return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, irqfd->virq);
 715}
 716
 717static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy,
 718                                      unsigned int queue_no,
 719                                      unsigned int vector)
 720{
 721    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 722    VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 723    EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 724    VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 725    int ret;
 726
 727    ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n, irqfd->virq);
 728    assert(ret == 0);
 729}
 730
 731static int kvm_virtio_pci_vector_use(VirtIOPCIProxy *proxy, int nvqs)
 732{
 733    PCIDevice *dev = &proxy->pci_dev;
 734    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 735    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 736    unsigned int vector;
 737    int ret, queue_no;
 738
 739    for (queue_no = 0; queue_no < nvqs; queue_no++) {
 740        if (!virtio_queue_get_num(vdev, queue_no)) {
 741            break;
 742        }
 743        vector = virtio_queue_vector(vdev, queue_no);
 744        if (vector >= msix_nr_vectors_allocated(dev)) {
 745            continue;
 746        }
 747        ret = kvm_virtio_pci_vq_vector_use(proxy, queue_no, vector);
 748        if (ret < 0) {
 749            goto undo;
 750        }
 751        /* If guest supports masking, set up irqfd now.
 752         * Otherwise, delay until unmasked in the frontend.
 753         */
 754        if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 755            ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 756            if (ret < 0) {
 757                kvm_virtio_pci_vq_vector_release(proxy, vector);
 758                goto undo;
 759            }
 760        }
 761    }
 762    return 0;
 763
 764undo:
 765    while (--queue_no >= 0) {
 766        vector = virtio_queue_vector(vdev, queue_no);
 767        if (vector >= msix_nr_vectors_allocated(dev)) {
 768            continue;
 769        }
 770        if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 771            kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 772        }
 773        kvm_virtio_pci_vq_vector_release(proxy, vector);
 774    }
 775    return ret;
 776}
 777
 778static void kvm_virtio_pci_vector_release(VirtIOPCIProxy *proxy, int nvqs)
 779{
 780    PCIDevice *dev = &proxy->pci_dev;
 781    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 782    unsigned int vector;
 783    int queue_no;
 784    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 785
 786    for (queue_no = 0; queue_no < nvqs; queue_no++) {
 787        if (!virtio_queue_get_num(vdev, queue_no)) {
 788            break;
 789        }
 790        vector = virtio_queue_vector(vdev, queue_no);
 791        if (vector >= msix_nr_vectors_allocated(dev)) {
 792            continue;
 793        }
 794        /* If guest supports masking, clean up irqfd now.
 795         * Otherwise, it was cleaned when masked in the frontend.
 796         */
 797        if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 798            kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 799        }
 800        kvm_virtio_pci_vq_vector_release(proxy, vector);
 801    }
 802}
 803
 804static int virtio_pci_vq_vector_unmask(VirtIOPCIProxy *proxy,
 805                                       unsigned int queue_no,
 806                                       unsigned int vector,
 807                                       MSIMessage msg)
 808{
 809    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 810    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 811    VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 812    EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 813    VirtIOIRQFD *irqfd;
 814    int ret = 0;
 815
 816    if (proxy->vector_irqfd) {
 817        irqfd = &proxy->vector_irqfd[vector];
 818        if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) {
 819            ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg,
 820                                               &proxy->pci_dev);
 821            if (ret < 0) {
 822                return ret;
 823            }
 824            kvm_irqchip_commit_routes(kvm_state);
 825        }
 826    }
 827
 828    /* If guest supports masking, irqfd is already setup, unmask it.
 829     * Otherwise, set it up now.
 830     */
 831    if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 832        k->guest_notifier_mask(vdev, queue_no, false);
 833        /* Test after unmasking to avoid losing events. */
 834        if (k->guest_notifier_pending &&
 835            k->guest_notifier_pending(vdev, queue_no)) {
 836            event_notifier_set(n);
 837        }
 838    } else {
 839        ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 840    }
 841    return ret;
 842}
 843
 844static void virtio_pci_vq_vector_mask(VirtIOPCIProxy *proxy,
 845                                             unsigned int queue_no,
 846                                             unsigned int vector)
 847{
 848    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 849    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 850
 851    /* If guest supports masking, keep irqfd but mask it.
 852     * Otherwise, clean it up now.
 853     */ 
 854    if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 855        k->guest_notifier_mask(vdev, queue_no, true);
 856    } else {
 857        kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 858    }
 859}
 860
 861static int virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector,
 862                                    MSIMessage msg)
 863{
 864    VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 865    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 866    VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
 867    int ret, index, unmasked = 0;
 868
 869    while (vq) {
 870        index = virtio_get_queue_index(vq);
 871        if (!virtio_queue_get_num(vdev, index)) {
 872            break;
 873        }
 874        if (index < proxy->nvqs_with_notifiers) {
 875            ret = virtio_pci_vq_vector_unmask(proxy, index, vector, msg);
 876            if (ret < 0) {
 877                goto undo;
 878            }
 879            ++unmasked;
 880        }
 881        vq = virtio_vector_next_queue(vq);
 882    }
 883
 884    return 0;
 885
 886undo:
 887    vq = virtio_vector_first_queue(vdev, vector);
 888    while (vq && unmasked >= 0) {
 889        index = virtio_get_queue_index(vq);
 890        if (index < proxy->nvqs_with_notifiers) {
 891            virtio_pci_vq_vector_mask(proxy, index, vector);
 892            --unmasked;
 893        }
 894        vq = virtio_vector_next_queue(vq);
 895    }
 896    return ret;
 897}
 898
 899static void virtio_pci_vector_mask(PCIDevice *dev, unsigned vector)
 900{
 901    VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 902    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 903    VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
 904    int index;
 905
 906    while (vq) {
 907        index = virtio_get_queue_index(vq);
 908        if (!virtio_queue_get_num(vdev, index)) {
 909            break;
 910        }
 911        if (index < proxy->nvqs_with_notifiers) {
 912            virtio_pci_vq_vector_mask(proxy, index, vector);
 913        }
 914        vq = virtio_vector_next_queue(vq);
 915    }
 916}
 917
 918static void virtio_pci_vector_poll(PCIDevice *dev,
 919                                   unsigned int vector_start,
 920                                   unsigned int vector_end)
 921{
 922    VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 923    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 924    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 925    int queue_no;
 926    unsigned int vector;
 927    EventNotifier *notifier;
 928    VirtQueue *vq;
 929
 930    for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
 931        if (!virtio_queue_get_num(vdev, queue_no)) {
 932            break;
 933        }
 934        vector = virtio_queue_vector(vdev, queue_no);
 935        if (vector < vector_start || vector >= vector_end ||
 936            !msix_is_masked(dev, vector)) {
 937            continue;
 938        }
 939        vq = virtio_get_queue(vdev, queue_no);
 940        notifier = virtio_queue_get_guest_notifier(vq);
 941        if (k->guest_notifier_pending) {
 942            if (k->guest_notifier_pending(vdev, queue_no)) {
 943                msix_set_pending(dev, vector);
 944            }
 945        } else if (event_notifier_test_and_clear(notifier)) {
 946            msix_set_pending(dev, vector);
 947        }
 948    }
 949}
 950
 951static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign,
 952                                         bool with_irqfd)
 953{
 954    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 955    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 956    VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
 957    VirtQueue *vq = virtio_get_queue(vdev, n);
 958    EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 959
 960    if (assign) {
 961        int r = event_notifier_init(notifier, 0);
 962        if (r < 0) {
 963            return r;
 964        }
 965        virtio_queue_set_guest_notifier_fd_handler(vq, true, with_irqfd);
 966    } else {
 967        virtio_queue_set_guest_notifier_fd_handler(vq, false, with_irqfd);
 968        event_notifier_cleanup(notifier);
 969    }
 970
 971    if (!msix_enabled(&proxy->pci_dev) &&
 972        vdev->use_guest_notifier_mask &&
 973        vdc->guest_notifier_mask) {
 974        vdc->guest_notifier_mask(vdev, n, !assign);
 975    }
 976
 977    return 0;
 978}
 979
 980static bool virtio_pci_query_guest_notifiers(DeviceState *d)
 981{
 982    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 983    return msix_enabled(&proxy->pci_dev);
 984}
 985
 986static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign)
 987{
 988    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 989    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 990    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 991    int r, n;
 992    bool with_irqfd = msix_enabled(&proxy->pci_dev) &&
 993        kvm_msi_via_irqfd_enabled();
 994
 995    nvqs = MIN(nvqs, VIRTIO_QUEUE_MAX);
 996
 997    /* When deassigning, pass a consistent nvqs value
 998     * to avoid leaking notifiers.
 999     */
1000    assert(assign || nvqs == proxy->nvqs_with_notifiers);
1001
1002    proxy->nvqs_with_notifiers = nvqs;
1003
1004    /* Must unset vector notifier while guest notifier is still assigned */
1005    if ((proxy->vector_irqfd || k->guest_notifier_mask) && !assign) {
1006        msix_unset_vector_notifiers(&proxy->pci_dev);
1007        if (proxy->vector_irqfd) {
1008            kvm_virtio_pci_vector_release(proxy, nvqs);
1009            g_free(proxy->vector_irqfd);
1010            proxy->vector_irqfd = NULL;
1011        }
1012    }
1013
1014    for (n = 0; n < nvqs; n++) {
1015        if (!virtio_queue_get_num(vdev, n)) {
1016            break;
1017        }
1018
1019        r = virtio_pci_set_guest_notifier(d, n, assign, with_irqfd);
1020        if (r < 0) {
1021            goto assign_error;
1022        }
1023    }
1024
1025    /* Must set vector notifier after guest notifier has been assigned */
1026    if ((with_irqfd || k->guest_notifier_mask) && assign) {
1027        if (with_irqfd) {
1028            proxy->vector_irqfd =
1029                g_malloc0(sizeof(*proxy->vector_irqfd) *
1030                          msix_nr_vectors_allocated(&proxy->pci_dev));
1031            r = kvm_virtio_pci_vector_use(proxy, nvqs);
1032            if (r < 0) {
1033                goto assign_error;
1034            }
1035        }
1036        r = msix_set_vector_notifiers(&proxy->pci_dev,
1037                                      virtio_pci_vector_unmask,
1038                                      virtio_pci_vector_mask,
1039                                      virtio_pci_vector_poll);
1040        if (r < 0) {
1041            goto notifiers_error;
1042        }
1043    }
1044
1045    return 0;
1046
1047notifiers_error:
1048    if (with_irqfd) {
1049        assert(assign);
1050        kvm_virtio_pci_vector_release(proxy, nvqs);
1051    }
1052
1053assign_error:
1054    /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
1055    assert(assign);
1056    while (--n >= 0) {
1057        virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd);
1058    }
1059    return r;
1060}
1061
1062static int virtio_pci_set_host_notifier_mr(DeviceState *d, int n,
1063                                           MemoryRegion *mr, bool assign)
1064{
1065    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1066    int offset;
1067
1068    if (n >= VIRTIO_QUEUE_MAX || !virtio_pci_modern(proxy) ||
1069        virtio_pci_queue_mem_mult(proxy) != memory_region_size(mr)) {
1070        return -1;
1071    }
1072
1073    if (assign) {
1074        offset = virtio_pci_queue_mem_mult(proxy) * n;
1075        memory_region_add_subregion_overlap(&proxy->notify.mr, offset, mr, 1);
1076    } else {
1077        memory_region_del_subregion(&proxy->notify.mr, mr);
1078    }
1079
1080    return 0;
1081}
1082
1083static void virtio_pci_vmstate_change(DeviceState *d, bool running)
1084{
1085    VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1086    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1087
1088    if (running) {
1089        /* Old QEMU versions did not set bus master enable on status write.
1090         * Detect DRIVER set and enable it.
1091         */
1092        if ((proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION) &&
1093            (vdev->status & VIRTIO_CONFIG_S_DRIVER) &&
1094            !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
1095            pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
1096                                     proxy->pci_dev.config[PCI_COMMAND] |
1097                                     PCI_COMMAND_MASTER, 1);
1098        }
1099        virtio_pci_start_ioeventfd(proxy);
1100    } else {
1101        virtio_pci_stop_ioeventfd(proxy);
1102    }
1103}
1104
1105/*
1106 * virtio-pci: This is the PCIDevice which has a virtio-pci-bus.
1107 */
1108
1109static int virtio_pci_query_nvectors(DeviceState *d)
1110{
1111    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1112
1113    return proxy->nvectors;
1114}
1115
1116static AddressSpace *virtio_pci_get_dma_as(DeviceState *d)
1117{
1118    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1119    PCIDevice *dev = &proxy->pci_dev;
1120
1121    return pci_get_address_space(dev);
1122}
1123
1124static bool virtio_pci_queue_enabled(DeviceState *d, int n)
1125{
1126    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1127    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1128
1129    if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1130        return proxy->vqs[n].enabled;
1131    }
1132
1133    return virtio_queue_enabled_legacy(vdev, n);
1134}
1135
1136static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy,
1137                                   struct virtio_pci_cap *cap)
1138{
1139    PCIDevice *dev = &proxy->pci_dev;
1140    int offset;
1141
1142    offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0,
1143                                cap->cap_len, &error_abort);
1144
1145    assert(cap->cap_len >= sizeof *cap);
1146    memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len,
1147           cap->cap_len - PCI_CAP_FLAGS);
1148
1149    return offset;
1150}
1151
1152static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr,
1153                                       unsigned size)
1154{
1155    VirtIOPCIProxy *proxy = opaque;
1156    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1157    uint32_t val = 0;
1158    int i;
1159
1160    if (vdev == NULL) {
1161        return UINT64_MAX;
1162    }
1163
1164    switch (addr) {
1165    case VIRTIO_PCI_COMMON_DFSELECT:
1166        val = proxy->dfselect;
1167        break;
1168    case VIRTIO_PCI_COMMON_DF:
1169        if (proxy->dfselect <= 1) {
1170            VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1171
1172            val = (vdev->host_features & ~vdc->legacy_features) >>
1173                (32 * proxy->dfselect);
1174        }
1175        break;
1176    case VIRTIO_PCI_COMMON_GFSELECT:
1177        val = proxy->gfselect;
1178        break;
1179    case VIRTIO_PCI_COMMON_GF:
1180        if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1181            val = proxy->guest_features[proxy->gfselect];
1182        }
1183        break;
1184    case VIRTIO_PCI_COMMON_MSIX:
1185        val = vdev->config_vector;
1186        break;
1187    case VIRTIO_PCI_COMMON_NUMQ:
1188        for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) {
1189            if (virtio_queue_get_num(vdev, i)) {
1190                val = i + 1;
1191            }
1192        }
1193        break;
1194    case VIRTIO_PCI_COMMON_STATUS:
1195        val = vdev->status;
1196        break;
1197    case VIRTIO_PCI_COMMON_CFGGENERATION:
1198        val = vdev->generation;
1199        break;
1200    case VIRTIO_PCI_COMMON_Q_SELECT:
1201        val = vdev->queue_sel;
1202        break;
1203    case VIRTIO_PCI_COMMON_Q_SIZE:
1204        val = virtio_queue_get_num(vdev, vdev->queue_sel);
1205        break;
1206    case VIRTIO_PCI_COMMON_Q_MSIX:
1207        val = virtio_queue_vector(vdev, vdev->queue_sel);
1208        break;
1209    case VIRTIO_PCI_COMMON_Q_ENABLE:
1210        val = proxy->vqs[vdev->queue_sel].enabled;
1211        break;
1212    case VIRTIO_PCI_COMMON_Q_NOFF:
1213        /* Simply map queues in order */
1214        val = vdev->queue_sel;
1215        break;
1216    case VIRTIO_PCI_COMMON_Q_DESCLO:
1217        val = proxy->vqs[vdev->queue_sel].desc[0];
1218        break;
1219    case VIRTIO_PCI_COMMON_Q_DESCHI:
1220        val = proxy->vqs[vdev->queue_sel].desc[1];
1221        break;
1222    case VIRTIO_PCI_COMMON_Q_AVAILLO:
1223        val = proxy->vqs[vdev->queue_sel].avail[0];
1224        break;
1225    case VIRTIO_PCI_COMMON_Q_AVAILHI:
1226        val = proxy->vqs[vdev->queue_sel].avail[1];
1227        break;
1228    case VIRTIO_PCI_COMMON_Q_USEDLO:
1229        val = proxy->vqs[vdev->queue_sel].used[0];
1230        break;
1231    case VIRTIO_PCI_COMMON_Q_USEDHI:
1232        val = proxy->vqs[vdev->queue_sel].used[1];
1233        break;
1234    default:
1235        val = 0;
1236    }
1237
1238    return val;
1239}
1240
1241static void virtio_pci_common_write(void *opaque, hwaddr addr,
1242                                    uint64_t val, unsigned size)
1243{
1244    VirtIOPCIProxy *proxy = opaque;
1245    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1246
1247    if (vdev == NULL) {
1248        return;
1249    }
1250
1251    switch (addr) {
1252    case VIRTIO_PCI_COMMON_DFSELECT:
1253        proxy->dfselect = val;
1254        break;
1255    case VIRTIO_PCI_COMMON_GFSELECT:
1256        proxy->gfselect = val;
1257        break;
1258    case VIRTIO_PCI_COMMON_GF:
1259        if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1260            proxy->guest_features[proxy->gfselect] = val;
1261            virtio_set_features(vdev,
1262                                (((uint64_t)proxy->guest_features[1]) << 32) |
1263                                proxy->guest_features[0]);
1264        }
1265        break;
1266    case VIRTIO_PCI_COMMON_MSIX:
1267        msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
1268        /* Make it possible for guest to discover an error took place. */
1269        if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1270            val = VIRTIO_NO_VECTOR;
1271        }
1272        vdev->config_vector = val;
1273        break;
1274    case VIRTIO_PCI_COMMON_STATUS:
1275        if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1276            virtio_pci_stop_ioeventfd(proxy);
1277        }
1278
1279        virtio_set_status(vdev, val & 0xFF);
1280
1281        if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
1282            virtio_pci_start_ioeventfd(proxy);
1283        }
1284
1285        if (vdev->status == 0) {
1286            virtio_pci_reset(DEVICE(proxy));
1287        }
1288
1289        break;
1290    case VIRTIO_PCI_COMMON_Q_SELECT:
1291        if (val < VIRTIO_QUEUE_MAX) {
1292            vdev->queue_sel = val;
1293        }
1294        break;
1295    case VIRTIO_PCI_COMMON_Q_SIZE:
1296        proxy->vqs[vdev->queue_sel].num = val;
1297        virtio_queue_set_num(vdev, vdev->queue_sel,
1298                             proxy->vqs[vdev->queue_sel].num);
1299        break;
1300    case VIRTIO_PCI_COMMON_Q_MSIX:
1301        msix_vector_unuse(&proxy->pci_dev,
1302                          virtio_queue_vector(vdev, vdev->queue_sel));
1303        /* Make it possible for guest to discover an error took place. */
1304        if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1305            val = VIRTIO_NO_VECTOR;
1306        }
1307        virtio_queue_set_vector(vdev, vdev->queue_sel, val);
1308        break;
1309    case VIRTIO_PCI_COMMON_Q_ENABLE:
1310        if (val == 1) {
1311            virtio_queue_set_num(vdev, vdev->queue_sel,
1312                                 proxy->vqs[vdev->queue_sel].num);
1313            virtio_queue_set_rings(vdev, vdev->queue_sel,
1314                       ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 |
1315                       proxy->vqs[vdev->queue_sel].desc[0],
1316                       ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 |
1317                       proxy->vqs[vdev->queue_sel].avail[0],
1318                       ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 |
1319                       proxy->vqs[vdev->queue_sel].used[0]);
1320            proxy->vqs[vdev->queue_sel].enabled = 1;
1321        } else {
1322            virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val);
1323        }
1324        break;
1325    case VIRTIO_PCI_COMMON_Q_DESCLO:
1326        proxy->vqs[vdev->queue_sel].desc[0] = val;
1327        break;
1328    case VIRTIO_PCI_COMMON_Q_DESCHI:
1329        proxy->vqs[vdev->queue_sel].desc[1] = val;
1330        break;
1331    case VIRTIO_PCI_COMMON_Q_AVAILLO:
1332        proxy->vqs[vdev->queue_sel].avail[0] = val;
1333        break;
1334    case VIRTIO_PCI_COMMON_Q_AVAILHI:
1335        proxy->vqs[vdev->queue_sel].avail[1] = val;
1336        break;
1337    case VIRTIO_PCI_COMMON_Q_USEDLO:
1338        proxy->vqs[vdev->queue_sel].used[0] = val;
1339        break;
1340    case VIRTIO_PCI_COMMON_Q_USEDHI:
1341        proxy->vqs[vdev->queue_sel].used[1] = val;
1342        break;
1343    default:
1344        break;
1345    }
1346}
1347
1348
1349static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr,
1350                                       unsigned size)
1351{
1352    VirtIOPCIProxy *proxy = opaque;
1353    if (virtio_bus_get_device(&proxy->bus) == NULL) {
1354        return UINT64_MAX;
1355    }
1356
1357    return 0;
1358}
1359
1360static void virtio_pci_notify_write(void *opaque, hwaddr addr,
1361                                    uint64_t val, unsigned size)
1362{
1363    VirtIOPCIProxy *proxy = opaque;
1364    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1365
1366    unsigned queue = addr / virtio_pci_queue_mem_mult(proxy);
1367
1368    if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1369        virtio_queue_notify(vdev, queue);
1370    }
1371}
1372
1373static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr,
1374                                        uint64_t val, unsigned size)
1375{
1376    VirtIOPCIProxy *proxy = opaque;
1377    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1378
1379    unsigned queue = val;
1380
1381    if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1382        virtio_queue_notify(vdev, queue);
1383    }
1384}
1385
1386static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr,
1387                                    unsigned size)
1388{
1389    VirtIOPCIProxy *proxy = opaque;
1390    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1391    uint64_t val;
1392
1393    if (vdev == NULL) {
1394        return UINT64_MAX;
1395    }
1396
1397    val = qatomic_xchg(&vdev->isr, 0);
1398    pci_irq_deassert(&proxy->pci_dev);
1399    return val;
1400}
1401
1402static void virtio_pci_isr_write(void *opaque, hwaddr addr,
1403                                 uint64_t val, unsigned size)
1404{
1405}
1406
1407static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr,
1408                                       unsigned size)
1409{
1410    VirtIOPCIProxy *proxy = opaque;
1411    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1412    uint64_t val;
1413
1414    if (vdev == NULL) {
1415        return UINT64_MAX;
1416    }
1417
1418    switch (size) {
1419    case 1:
1420        val = virtio_config_modern_readb(vdev, addr);
1421        break;
1422    case 2:
1423        val = virtio_config_modern_readw(vdev, addr);
1424        break;
1425    case 4:
1426        val = virtio_config_modern_readl(vdev, addr);
1427        break;
1428    default:
1429        val = 0;
1430        break;
1431    }
1432    return val;
1433}
1434
1435static void virtio_pci_device_write(void *opaque, hwaddr addr,
1436                                    uint64_t val, unsigned size)
1437{
1438    VirtIOPCIProxy *proxy = opaque;
1439    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1440
1441    if (vdev == NULL) {
1442        return;
1443    }
1444
1445    switch (size) {
1446    case 1:
1447        virtio_config_modern_writeb(vdev, addr, val);
1448        break;
1449    case 2:
1450        virtio_config_modern_writew(vdev, addr, val);
1451        break;
1452    case 4:
1453        virtio_config_modern_writel(vdev, addr, val);
1454        break;
1455    }
1456}
1457
1458static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy,
1459                                           const char *vdev_name)
1460{
1461    static const MemoryRegionOps common_ops = {
1462        .read = virtio_pci_common_read,
1463        .write = virtio_pci_common_write,
1464        .impl = {
1465            .min_access_size = 1,
1466            .max_access_size = 4,
1467        },
1468        .endianness = DEVICE_LITTLE_ENDIAN,
1469    };
1470    static const MemoryRegionOps isr_ops = {
1471        .read = virtio_pci_isr_read,
1472        .write = virtio_pci_isr_write,
1473        .impl = {
1474            .min_access_size = 1,
1475            .max_access_size = 4,
1476        },
1477        .endianness = DEVICE_LITTLE_ENDIAN,
1478    };
1479    static const MemoryRegionOps device_ops = {
1480        .read = virtio_pci_device_read,
1481        .write = virtio_pci_device_write,
1482        .impl = {
1483            .min_access_size = 1,
1484            .max_access_size = 4,
1485        },
1486        .endianness = DEVICE_LITTLE_ENDIAN,
1487    };
1488    static const MemoryRegionOps notify_ops = {
1489        .read = virtio_pci_notify_read,
1490        .write = virtio_pci_notify_write,
1491        .impl = {
1492            .min_access_size = 1,
1493            .max_access_size = 4,
1494        },
1495        .endianness = DEVICE_LITTLE_ENDIAN,
1496    };
1497    static const MemoryRegionOps notify_pio_ops = {
1498        .read = virtio_pci_notify_read,
1499        .write = virtio_pci_notify_write_pio,
1500        .impl = {
1501            .min_access_size = 1,
1502            .max_access_size = 4,
1503        },
1504        .endianness = DEVICE_LITTLE_ENDIAN,
1505    };
1506    g_autoptr(GString) name = g_string_new(NULL);
1507
1508    g_string_printf(name, "virtio-pci-common-%s", vdev_name);
1509    memory_region_init_io(&proxy->common.mr, OBJECT(proxy),
1510                          &common_ops,
1511                          proxy,
1512                          name->str,
1513                          proxy->common.size);
1514
1515    g_string_printf(name, "virtio-pci-isr-%s", vdev_name);
1516    memory_region_init_io(&proxy->isr.mr, OBJECT(proxy),
1517                          &isr_ops,
1518                          proxy,
1519                          name->str,
1520                          proxy->isr.size);
1521
1522    g_string_printf(name, "virtio-pci-device-%s", vdev_name);
1523    memory_region_init_io(&proxy->device.mr, OBJECT(proxy),
1524                          &device_ops,
1525                          proxy,
1526                          name->str,
1527                          proxy->device.size);
1528
1529    g_string_printf(name, "virtio-pci-notify-%s", vdev_name);
1530    memory_region_init_io(&proxy->notify.mr, OBJECT(proxy),
1531                          &notify_ops,
1532                          proxy,
1533                          name->str,
1534                          proxy->notify.size);
1535
1536    g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name);
1537    memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy),
1538                          &notify_pio_ops,
1539                          proxy,
1540                          name->str,
1541                          proxy->notify_pio.size);
1542}
1543
1544static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy,
1545                                         VirtIOPCIRegion *region,
1546                                         struct virtio_pci_cap *cap,
1547                                         MemoryRegion *mr,
1548                                         uint8_t bar)
1549{
1550    memory_region_add_subregion(mr, region->offset, &region->mr);
1551
1552    cap->cfg_type = region->type;
1553    cap->bar = bar;
1554    cap->offset = cpu_to_le32(region->offset);
1555    cap->length = cpu_to_le32(region->size);
1556    virtio_pci_add_mem_cap(proxy, cap);
1557
1558}
1559
1560static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy,
1561                                             VirtIOPCIRegion *region,
1562                                             struct virtio_pci_cap *cap)
1563{
1564    virtio_pci_modern_region_map(proxy, region, cap,
1565                                 &proxy->modern_bar, proxy->modern_mem_bar_idx);
1566}
1567
1568static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy,
1569                                            VirtIOPCIRegion *region,
1570                                            struct virtio_pci_cap *cap)
1571{
1572    virtio_pci_modern_region_map(proxy, region, cap,
1573                                 &proxy->io_bar, proxy->modern_io_bar_idx);
1574}
1575
1576static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy,
1577                                               VirtIOPCIRegion *region)
1578{
1579    memory_region_del_subregion(&proxy->modern_bar,
1580                                &region->mr);
1581}
1582
1583static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy,
1584                                              VirtIOPCIRegion *region)
1585{
1586    memory_region_del_subregion(&proxy->io_bar,
1587                                &region->mr);
1588}
1589
1590static void virtio_pci_pre_plugged(DeviceState *d, Error **errp)
1591{
1592    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1593    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1594
1595    if (virtio_pci_modern(proxy)) {
1596        virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
1597    }
1598
1599    virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE);
1600}
1601
1602/* This is called by virtio-bus just after the device is plugged. */
1603static void virtio_pci_device_plugged(DeviceState *d, Error **errp)
1604{
1605    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1606    VirtioBusState *bus = &proxy->bus;
1607    bool legacy = virtio_pci_legacy(proxy);
1608    bool modern;
1609    bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1610    uint8_t *config;
1611    uint32_t size;
1612    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1613
1614    /*
1615     * Virtio capabilities present without
1616     * VIRTIO_F_VERSION_1 confuses guests
1617     */
1618    if (!proxy->ignore_backend_features &&
1619            !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) {
1620        virtio_pci_disable_modern(proxy);
1621
1622        if (!legacy) {
1623            error_setg(errp, "Device doesn't support modern mode, and legacy"
1624                             " mode is disabled");
1625            error_append_hint(errp, "Set disable-legacy to off\n");
1626
1627            return;
1628        }
1629    }
1630
1631    modern = virtio_pci_modern(proxy);
1632
1633    config = proxy->pci_dev.config;
1634    if (proxy->class_code) {
1635        pci_config_set_class(config, proxy->class_code);
1636    }
1637
1638    if (legacy) {
1639        if (!virtio_legacy_allowed(vdev)) {
1640            /*
1641             * To avoid migration issues, we allow legacy mode when legacy
1642             * check is disabled in the old machine types (< 5.1).
1643             */
1644            if (virtio_legacy_check_disabled(vdev)) {
1645                warn_report("device is modern-only, but for backward "
1646                            "compatibility legacy is allowed");
1647            } else {
1648                error_setg(errp,
1649                           "device is modern-only, use disable-legacy=on");
1650                return;
1651            }
1652        }
1653        if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1654            error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by"
1655                       " neither legacy nor transitional device");
1656            return ;
1657        }
1658        /*
1659         * Legacy and transitional devices use specific subsystem IDs.
1660         * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID)
1661         * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default.
1662         */
1663        pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus));
1664    } else {
1665        /* pure virtio-1.0 */
1666        pci_set_word(config + PCI_VENDOR_ID,
1667                     PCI_VENDOR_ID_REDHAT_QUMRANET);
1668        pci_set_word(config + PCI_DEVICE_ID,
1669                     0x1040 + virtio_bus_get_vdev_id(bus));
1670        pci_config_set_revision(config, 1);
1671    }
1672    config[PCI_INTERRUPT_PIN] = 1;
1673
1674
1675    if (modern) {
1676        struct virtio_pci_cap cap = {
1677            .cap_len = sizeof cap,
1678        };
1679        struct virtio_pci_notify_cap notify = {
1680            .cap.cap_len = sizeof notify,
1681            .notify_off_multiplier =
1682                cpu_to_le32(virtio_pci_queue_mem_mult(proxy)),
1683        };
1684        struct virtio_pci_cfg_cap cfg = {
1685            .cap.cap_len = sizeof cfg,
1686            .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG,
1687        };
1688        struct virtio_pci_notify_cap notify_pio = {
1689            .cap.cap_len = sizeof notify,
1690            .notify_off_multiplier = cpu_to_le32(0x0),
1691        };
1692
1693        struct virtio_pci_cfg_cap *cfg_mask;
1694
1695        virtio_pci_modern_regions_init(proxy, vdev->name);
1696
1697        virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap);
1698        virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap);
1699        virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap);
1700        virtio_pci_modern_mem_region_map(proxy, &proxy->notify, &notify.cap);
1701
1702        if (modern_pio) {
1703            memory_region_init(&proxy->io_bar, OBJECT(proxy),
1704                               "virtio-pci-io", 0x4);
1705
1706            pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx,
1707                             PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar);
1708
1709            virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio,
1710                                            &notify_pio.cap);
1711        }
1712
1713        pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx,
1714                         PCI_BASE_ADDRESS_SPACE_MEMORY |
1715                         PCI_BASE_ADDRESS_MEM_PREFETCH |
1716                         PCI_BASE_ADDRESS_MEM_TYPE_64,
1717                         &proxy->modern_bar);
1718
1719        proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap);
1720        cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap);
1721        pci_set_byte(&cfg_mask->cap.bar, ~0x0);
1722        pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0);
1723        pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0);
1724        pci_set_long(cfg_mask->pci_cfg_data, ~0x0);
1725    }
1726
1727    if (proxy->nvectors) {
1728        int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors,
1729                                          proxy->msix_bar_idx, NULL);
1730        if (err) {
1731            /* Notice when a system that supports MSIx can't initialize it */
1732            if (err != -ENOTSUP) {
1733                warn_report("unable to init msix vectors to %" PRIu32,
1734                            proxy->nvectors);
1735            }
1736            proxy->nvectors = 0;
1737        }
1738    }
1739
1740    proxy->pci_dev.config_write = virtio_write_config;
1741    proxy->pci_dev.config_read = virtio_read_config;
1742
1743    if (legacy) {
1744        size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev)
1745            + virtio_bus_get_vdev_config_len(bus);
1746        size = pow2ceil(size);
1747
1748        memory_region_init_io(&proxy->bar, OBJECT(proxy),
1749                              &virtio_pci_config_ops,
1750                              proxy, "virtio-pci", size);
1751
1752        pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx,
1753                         PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar);
1754    }
1755}
1756
1757static void virtio_pci_device_unplugged(DeviceState *d)
1758{
1759    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1760    bool modern = virtio_pci_modern(proxy);
1761    bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1762
1763    virtio_pci_stop_ioeventfd(proxy);
1764
1765    if (modern) {
1766        virtio_pci_modern_mem_region_unmap(proxy, &proxy->common);
1767        virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr);
1768        virtio_pci_modern_mem_region_unmap(proxy, &proxy->device);
1769        virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify);
1770        if (modern_pio) {
1771            virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio);
1772        }
1773    }
1774}
1775
1776static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
1777{
1778    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1779    VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev);
1780    bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1781                     !pci_bus_is_root(pci_get_bus(pci_dev));
1782
1783    if (kvm_enabled() && !kvm_has_many_ioeventfds()) {
1784        proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1785    }
1786
1787    /* fd-based ioevents can't be synchronized in record/replay */
1788    if (replay_mode != REPLAY_MODE_NONE) {
1789        proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1790    }
1791
1792    /*
1793     * virtio pci bar layout used by default.
1794     * subclasses can re-arrange things if needed.
1795     *
1796     *   region 0   --  virtio legacy io bar
1797     *   region 1   --  msi-x bar
1798     *   region 2   --  virtio modern io bar (off by default)
1799     *   region 4+5 --  virtio modern memory (64bit) bar
1800     *
1801     */
1802    proxy->legacy_io_bar_idx  = 0;
1803    proxy->msix_bar_idx       = 1;
1804    proxy->modern_io_bar_idx  = 2;
1805    proxy->modern_mem_bar_idx = 4;
1806
1807    proxy->common.offset = 0x0;
1808    proxy->common.size = 0x1000;
1809    proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG;
1810
1811    proxy->isr.offset = 0x1000;
1812    proxy->isr.size = 0x1000;
1813    proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG;
1814
1815    proxy->device.offset = 0x2000;
1816    proxy->device.size = 0x1000;
1817    proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG;
1818
1819    proxy->notify.offset = 0x3000;
1820    proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX;
1821    proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1822
1823    proxy->notify_pio.offset = 0x0;
1824    proxy->notify_pio.size = 0x4;
1825    proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1826
1827    /* subclasses can enforce modern, so do this unconditionally */
1828    memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci",
1829                       /* PCI BAR regions must be powers of 2 */
1830                       pow2ceil(proxy->notify.offset + proxy->notify.size));
1831
1832    if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) {
1833        proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
1834    }
1835
1836    if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) {
1837        error_setg(errp, "device cannot work as neither modern nor legacy mode"
1838                   " is enabled");
1839        error_append_hint(errp, "Set either disable-modern or disable-legacy"
1840                          " to off\n");
1841        return;
1842    }
1843
1844    if (pcie_port && pci_is_express(pci_dev)) {
1845        int pos;
1846        uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE;
1847
1848        pos = pcie_endpoint_cap_init(pci_dev, 0);
1849        assert(pos > 0);
1850
1851        pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0,
1852                                 PCI_PM_SIZEOF, errp);
1853        if (pos < 0) {
1854            return;
1855        }
1856
1857        pci_dev->exp.pm_cap = pos;
1858
1859        /*
1860         * Indicates that this function complies with revision 1.2 of the
1861         * PCI Power Management Interface Specification.
1862         */
1863        pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3);
1864
1865        if (proxy->flags & VIRTIO_PCI_FLAG_AER) {
1866            pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset,
1867                          PCI_ERR_SIZEOF, NULL);
1868            last_pcie_cap_offset += PCI_ERR_SIZEOF;
1869        }
1870
1871        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) {
1872            /* Init error enabling flags */
1873            pcie_cap_deverr_init(pci_dev);
1874        }
1875
1876        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) {
1877            /* Init Link Control Register */
1878            pcie_cap_lnkctl_init(pci_dev);
1879        }
1880
1881        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) {
1882            /* Init Power Management Control Register */
1883            pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL,
1884                         PCI_PM_CTRL_STATE_MASK);
1885        }
1886
1887        if (proxy->flags & VIRTIO_PCI_FLAG_ATS) {
1888            pcie_ats_init(pci_dev, last_pcie_cap_offset,
1889                          proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED);
1890            last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF;
1891        }
1892
1893        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
1894            /* Set Function Level Reset capability bit */
1895            pcie_cap_flr_init(pci_dev);
1896        }
1897    } else {
1898        /*
1899         * make future invocations of pci_is_express() return false
1900         * and pci_config_size() return PCI_CONFIG_SPACE_SIZE.
1901         */
1902        pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS;
1903    }
1904
1905    virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
1906    if (k->realize) {
1907        k->realize(proxy, errp);
1908    }
1909}
1910
1911static void virtio_pci_exit(PCIDevice *pci_dev)
1912{
1913    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1914    bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1915                     !pci_bus_is_root(pci_get_bus(pci_dev));
1916
1917    msix_uninit_exclusive_bar(pci_dev);
1918    if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port &&
1919        pci_is_express(pci_dev)) {
1920        pcie_aer_exit(pci_dev);
1921    }
1922}
1923
1924static void virtio_pci_reset(DeviceState *qdev)
1925{
1926    VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1927    VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
1928    PCIDevice *dev = PCI_DEVICE(qdev);
1929    int i;
1930
1931    virtio_pci_stop_ioeventfd(proxy);
1932    virtio_bus_reset(bus);
1933    msix_unuse_all_vectors(&proxy->pci_dev);
1934
1935    for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1936        proxy->vqs[i].enabled = 0;
1937        proxy->vqs[i].num = 0;
1938        proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0;
1939        proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0;
1940        proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0;
1941    }
1942
1943    if (pci_is_express(dev)) {
1944        pcie_cap_deverr_reset(dev);
1945        pcie_cap_lnkctl_reset(dev);
1946
1947        pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0);
1948    }
1949}
1950
1951static Property virtio_pci_properties[] = {
1952    DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags,
1953                    VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false),
1954    DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags,
1955                    VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true),
1956    DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags,
1957                    VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false),
1958    DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags,
1959                    VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false),
1960    DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags,
1961                    VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false),
1962    DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy,
1963                     ignore_backend_features, false),
1964    DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags,
1965                    VIRTIO_PCI_FLAG_ATS_BIT, false),
1966    DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags,
1967                    VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true),
1968    DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags,
1969                    VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true),
1970    DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags,
1971                    VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true),
1972    DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags,
1973                    VIRTIO_PCI_FLAG_INIT_PM_BIT, true),
1974    DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags,
1975                    VIRTIO_PCI_FLAG_INIT_FLR_BIT, true),
1976    DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags,
1977                    VIRTIO_PCI_FLAG_AER_BIT, false),
1978    DEFINE_PROP_END_OF_LIST(),
1979};
1980
1981static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp)
1982{
1983    VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev);
1984    VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1985    PCIDevice *pci_dev = &proxy->pci_dev;
1986
1987    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) &&
1988        virtio_pci_modern(proxy)) {
1989        pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1990    }
1991
1992    vpciklass->parent_dc_realize(qdev, errp);
1993}
1994
1995static void virtio_pci_class_init(ObjectClass *klass, void *data)
1996{
1997    DeviceClass *dc = DEVICE_CLASS(klass);
1998    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1999    VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass);
2000
2001    device_class_set_props(dc, virtio_pci_properties);
2002    k->realize = virtio_pci_realize;
2003    k->exit = virtio_pci_exit;
2004    k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
2005    k->revision = VIRTIO_PCI_ABI_VERSION;
2006    k->class_id = PCI_CLASS_OTHERS;
2007    device_class_set_parent_realize(dc, virtio_pci_dc_realize,
2008                                    &vpciklass->parent_dc_realize);
2009    dc->reset = virtio_pci_reset;
2010}
2011
2012static const TypeInfo virtio_pci_info = {
2013    .name          = TYPE_VIRTIO_PCI,
2014    .parent        = TYPE_PCI_DEVICE,
2015    .instance_size = sizeof(VirtIOPCIProxy),
2016    .class_init    = virtio_pci_class_init,
2017    .class_size    = sizeof(VirtioPCIClass),
2018    .abstract      = true,
2019};
2020
2021static Property virtio_pci_generic_properties[] = {
2022    DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy,
2023                            ON_OFF_AUTO_AUTO),
2024    DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false),
2025    DEFINE_PROP_END_OF_LIST(),
2026};
2027
2028static void virtio_pci_base_class_init(ObjectClass *klass, void *data)
2029{
2030    const VirtioPCIDeviceTypeInfo *t = data;
2031    if (t->class_init) {
2032        t->class_init(klass, NULL);
2033    }
2034}
2035
2036static void virtio_pci_generic_class_init(ObjectClass *klass, void *data)
2037{
2038    DeviceClass *dc = DEVICE_CLASS(klass);
2039
2040    device_class_set_props(dc, virtio_pci_generic_properties);
2041}
2042
2043static void virtio_pci_transitional_instance_init(Object *obj)
2044{
2045    VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2046
2047    proxy->disable_legacy = ON_OFF_AUTO_OFF;
2048    proxy->disable_modern = false;
2049}
2050
2051static void virtio_pci_non_transitional_instance_init(Object *obj)
2052{
2053    VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2054
2055    proxy->disable_legacy = ON_OFF_AUTO_ON;
2056    proxy->disable_modern = false;
2057}
2058
2059void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t)
2060{
2061    char *base_name = NULL;
2062    TypeInfo base_type_info = {
2063        .name          = t->base_name,
2064        .parent        = t->parent ? t->parent : TYPE_VIRTIO_PCI,
2065        .instance_size = t->instance_size,
2066        .instance_init = t->instance_init,
2067        .class_size    = t->class_size,
2068        .abstract      = true,
2069        .interfaces    = t->interfaces,
2070    };
2071    TypeInfo generic_type_info = {
2072        .name = t->generic_name,
2073        .parent = base_type_info.name,
2074        .class_init = virtio_pci_generic_class_init,
2075        .interfaces = (InterfaceInfo[]) {
2076            { INTERFACE_PCIE_DEVICE },
2077            { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2078            { }
2079        },
2080    };
2081
2082    if (!base_type_info.name) {
2083        /* No base type -> register a single generic device type */
2084        /* use intermediate %s-base-type to add generic device props */
2085        base_name = g_strdup_printf("%s-base-type", t->generic_name);
2086        base_type_info.name = base_name;
2087        base_type_info.class_init = virtio_pci_generic_class_init;
2088
2089        generic_type_info.parent = base_name;
2090        generic_type_info.class_init = virtio_pci_base_class_init;
2091        generic_type_info.class_data = (void *)t;
2092
2093        assert(!t->non_transitional_name);
2094        assert(!t->transitional_name);
2095    } else {
2096        base_type_info.class_init = virtio_pci_base_class_init;
2097        base_type_info.class_data = (void *)t;
2098    }
2099
2100    type_register(&base_type_info);
2101    if (generic_type_info.name) {
2102        type_register(&generic_type_info);
2103    }
2104
2105    if (t->non_transitional_name) {
2106        const TypeInfo non_transitional_type_info = {
2107            .name          = t->non_transitional_name,
2108            .parent        = base_type_info.name,
2109            .instance_init = virtio_pci_non_transitional_instance_init,
2110            .interfaces = (InterfaceInfo[]) {
2111                { INTERFACE_PCIE_DEVICE },
2112                { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2113                { }
2114            },
2115        };
2116        type_register(&non_transitional_type_info);
2117    }
2118
2119    if (t->transitional_name) {
2120        const TypeInfo transitional_type_info = {
2121            .name          = t->transitional_name,
2122            .parent        = base_type_info.name,
2123            .instance_init = virtio_pci_transitional_instance_init,
2124            .interfaces = (InterfaceInfo[]) {
2125                /*
2126                 * Transitional virtio devices work only as Conventional PCI
2127                 * devices because they require PIO ports.
2128                 */
2129                { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2130                { }
2131            },
2132        };
2133        type_register(&transitional_type_info);
2134    }
2135    g_free(base_name);
2136}
2137
2138unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues)
2139{
2140    /*
2141     * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted
2142     * virtqueue buffers can handle their completion. When a different vCPU
2143     * handles completion it may need to IPI the vCPU that submitted the
2144     * request and this adds overhead.
2145     *
2146     * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in
2147     * guests with very many vCPUs and a device that is only used by a few
2148     * vCPUs. Unfortunately optimizing that case requires manual pinning inside
2149     * the guest, so those users might as well manually set the number of
2150     * queues. There is no upper limit that can be applied automatically and
2151     * doing so arbitrarily would result in a sudden performance drop once the
2152     * threshold number of vCPUs is exceeded.
2153     */
2154    unsigned num_queues = current_machine->smp.cpus;
2155
2156    /*
2157     * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the
2158     * config change interrupt and the fixed virtqueues must be taken into
2159     * account too.
2160     */
2161    num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues);
2162
2163    /*
2164     * There is a limit to how many virtqueues a device can have.
2165     */
2166    return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues);
2167}
2168
2169/* virtio-pci-bus */
2170
2171static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
2172                               VirtIOPCIProxy *dev)
2173{
2174    DeviceState *qdev = DEVICE(dev);
2175    char virtio_bus_name[] = "virtio-bus";
2176
2177    qbus_create_inplace(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev,
2178                        virtio_bus_name);
2179}
2180
2181static void virtio_pci_bus_class_init(ObjectClass *klass, void *data)
2182{
2183    BusClass *bus_class = BUS_CLASS(klass);
2184    VirtioBusClass *k = VIRTIO_BUS_CLASS(klass);
2185    bus_class->max_dev = 1;
2186    k->notify = virtio_pci_notify;
2187    k->save_config = virtio_pci_save_config;
2188    k->load_config = virtio_pci_load_config;
2189    k->save_queue = virtio_pci_save_queue;
2190    k->load_queue = virtio_pci_load_queue;
2191    k->save_extra_state = virtio_pci_save_extra_state;
2192    k->load_extra_state = virtio_pci_load_extra_state;
2193    k->has_extra_state = virtio_pci_has_extra_state;
2194    k->query_guest_notifiers = virtio_pci_query_guest_notifiers;
2195    k->set_guest_notifiers = virtio_pci_set_guest_notifiers;
2196    k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr;
2197    k->vmstate_change = virtio_pci_vmstate_change;
2198    k->pre_plugged = virtio_pci_pre_plugged;
2199    k->device_plugged = virtio_pci_device_plugged;
2200    k->device_unplugged = virtio_pci_device_unplugged;
2201    k->query_nvectors = virtio_pci_query_nvectors;
2202    k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled;
2203    k->ioeventfd_assign = virtio_pci_ioeventfd_assign;
2204    k->get_dma_as = virtio_pci_get_dma_as;
2205    k->queue_enabled = virtio_pci_queue_enabled;
2206}
2207
2208static const TypeInfo virtio_pci_bus_info = {
2209    .name          = TYPE_VIRTIO_PCI_BUS,
2210    .parent        = TYPE_VIRTIO_BUS,
2211    .instance_size = sizeof(VirtioPCIBusState),
2212    .class_size    = sizeof(VirtioPCIBusClass),
2213    .class_init    = virtio_pci_bus_class_init,
2214};
2215
2216static void virtio_pci_register_types(void)
2217{
2218    /* Base types: */
2219    type_register_static(&virtio_pci_bus_info);
2220    type_register_static(&virtio_pci_info);
2221}
2222
2223type_init(virtio_pci_register_types)
2224
2225