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_iommu_enabled(DeviceState *d)
1125{
1126    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1127    PCIDevice *dev = &proxy->pci_dev;
1128    AddressSpace *dma_as = pci_device_iommu_address_space(dev);
1129
1130    if (dma_as == &address_space_memory) {
1131        return false;
1132    }
1133
1134    return true;
1135}
1136
1137static bool virtio_pci_queue_enabled(DeviceState *d, int n)
1138{
1139    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1140    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1141
1142    if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1143        return proxy->vqs[n].enabled;
1144    }
1145
1146    return virtio_queue_enabled_legacy(vdev, n);
1147}
1148
1149static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy,
1150                                   struct virtio_pci_cap *cap)
1151{
1152    PCIDevice *dev = &proxy->pci_dev;
1153    int offset;
1154
1155    offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0,
1156                                cap->cap_len, &error_abort);
1157
1158    assert(cap->cap_len >= sizeof *cap);
1159    memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len,
1160           cap->cap_len - PCI_CAP_FLAGS);
1161
1162    return offset;
1163}
1164
1165static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr,
1166                                       unsigned size)
1167{
1168    VirtIOPCIProxy *proxy = opaque;
1169    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1170    uint32_t val = 0;
1171    int i;
1172
1173    if (vdev == NULL) {
1174        return UINT64_MAX;
1175    }
1176
1177    switch (addr) {
1178    case VIRTIO_PCI_COMMON_DFSELECT:
1179        val = proxy->dfselect;
1180        break;
1181    case VIRTIO_PCI_COMMON_DF:
1182        if (proxy->dfselect <= 1) {
1183            VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1184
1185            val = (vdev->host_features & ~vdc->legacy_features) >>
1186                (32 * proxy->dfselect);
1187        }
1188        break;
1189    case VIRTIO_PCI_COMMON_GFSELECT:
1190        val = proxy->gfselect;
1191        break;
1192    case VIRTIO_PCI_COMMON_GF:
1193        if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1194            val = proxy->guest_features[proxy->gfselect];
1195        }
1196        break;
1197    case VIRTIO_PCI_COMMON_MSIX:
1198        val = vdev->config_vector;
1199        break;
1200    case VIRTIO_PCI_COMMON_NUMQ:
1201        for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) {
1202            if (virtio_queue_get_num(vdev, i)) {
1203                val = i + 1;
1204            }
1205        }
1206        break;
1207    case VIRTIO_PCI_COMMON_STATUS:
1208        val = vdev->status;
1209        break;
1210    case VIRTIO_PCI_COMMON_CFGGENERATION:
1211        val = vdev->generation;
1212        break;
1213    case VIRTIO_PCI_COMMON_Q_SELECT:
1214        val = vdev->queue_sel;
1215        break;
1216    case VIRTIO_PCI_COMMON_Q_SIZE:
1217        val = virtio_queue_get_num(vdev, vdev->queue_sel);
1218        break;
1219    case VIRTIO_PCI_COMMON_Q_MSIX:
1220        val = virtio_queue_vector(vdev, vdev->queue_sel);
1221        break;
1222    case VIRTIO_PCI_COMMON_Q_ENABLE:
1223        val = proxy->vqs[vdev->queue_sel].enabled;
1224        break;
1225    case VIRTIO_PCI_COMMON_Q_NOFF:
1226        /* Simply map queues in order */
1227        val = vdev->queue_sel;
1228        break;
1229    case VIRTIO_PCI_COMMON_Q_DESCLO:
1230        val = proxy->vqs[vdev->queue_sel].desc[0];
1231        break;
1232    case VIRTIO_PCI_COMMON_Q_DESCHI:
1233        val = proxy->vqs[vdev->queue_sel].desc[1];
1234        break;
1235    case VIRTIO_PCI_COMMON_Q_AVAILLO:
1236        val = proxy->vqs[vdev->queue_sel].avail[0];
1237        break;
1238    case VIRTIO_PCI_COMMON_Q_AVAILHI:
1239        val = proxy->vqs[vdev->queue_sel].avail[1];
1240        break;
1241    case VIRTIO_PCI_COMMON_Q_USEDLO:
1242        val = proxy->vqs[vdev->queue_sel].used[0];
1243        break;
1244    case VIRTIO_PCI_COMMON_Q_USEDHI:
1245        val = proxy->vqs[vdev->queue_sel].used[1];
1246        break;
1247    default:
1248        val = 0;
1249    }
1250
1251    return val;
1252}
1253
1254static void virtio_pci_common_write(void *opaque, hwaddr addr,
1255                                    uint64_t val, unsigned size)
1256{
1257    VirtIOPCIProxy *proxy = opaque;
1258    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1259
1260    if (vdev == NULL) {
1261        return;
1262    }
1263
1264    switch (addr) {
1265    case VIRTIO_PCI_COMMON_DFSELECT:
1266        proxy->dfselect = val;
1267        break;
1268    case VIRTIO_PCI_COMMON_GFSELECT:
1269        proxy->gfselect = val;
1270        break;
1271    case VIRTIO_PCI_COMMON_GF:
1272        if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1273            proxy->guest_features[proxy->gfselect] = val;
1274            virtio_set_features(vdev,
1275                                (((uint64_t)proxy->guest_features[1]) << 32) |
1276                                proxy->guest_features[0]);
1277        }
1278        break;
1279    case VIRTIO_PCI_COMMON_MSIX:
1280        msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
1281        /* Make it possible for guest to discover an error took place. */
1282        if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1283            val = VIRTIO_NO_VECTOR;
1284        }
1285        vdev->config_vector = val;
1286        break;
1287    case VIRTIO_PCI_COMMON_STATUS:
1288        if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1289            virtio_pci_stop_ioeventfd(proxy);
1290        }
1291
1292        virtio_set_status(vdev, val & 0xFF);
1293
1294        if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
1295            virtio_pci_start_ioeventfd(proxy);
1296        }
1297
1298        if (vdev->status == 0) {
1299            virtio_pci_reset(DEVICE(proxy));
1300        }
1301
1302        break;
1303    case VIRTIO_PCI_COMMON_Q_SELECT:
1304        if (val < VIRTIO_QUEUE_MAX) {
1305            vdev->queue_sel = val;
1306        }
1307        break;
1308    case VIRTIO_PCI_COMMON_Q_SIZE:
1309        proxy->vqs[vdev->queue_sel].num = val;
1310        virtio_queue_set_num(vdev, vdev->queue_sel,
1311                             proxy->vqs[vdev->queue_sel].num);
1312        break;
1313    case VIRTIO_PCI_COMMON_Q_MSIX:
1314        msix_vector_unuse(&proxy->pci_dev,
1315                          virtio_queue_vector(vdev, vdev->queue_sel));
1316        /* Make it possible for guest to discover an error took place. */
1317        if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1318            val = VIRTIO_NO_VECTOR;
1319        }
1320        virtio_queue_set_vector(vdev, vdev->queue_sel, val);
1321        break;
1322    case VIRTIO_PCI_COMMON_Q_ENABLE:
1323        if (val == 1) {
1324            virtio_queue_set_num(vdev, vdev->queue_sel,
1325                                 proxy->vqs[vdev->queue_sel].num);
1326            virtio_queue_set_rings(vdev, vdev->queue_sel,
1327                       ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 |
1328                       proxy->vqs[vdev->queue_sel].desc[0],
1329                       ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 |
1330                       proxy->vqs[vdev->queue_sel].avail[0],
1331                       ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 |
1332                       proxy->vqs[vdev->queue_sel].used[0]);
1333            proxy->vqs[vdev->queue_sel].enabled = 1;
1334        } else {
1335            virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val);
1336        }
1337        break;
1338    case VIRTIO_PCI_COMMON_Q_DESCLO:
1339        proxy->vqs[vdev->queue_sel].desc[0] = val;
1340        break;
1341    case VIRTIO_PCI_COMMON_Q_DESCHI:
1342        proxy->vqs[vdev->queue_sel].desc[1] = val;
1343        break;
1344    case VIRTIO_PCI_COMMON_Q_AVAILLO:
1345        proxy->vqs[vdev->queue_sel].avail[0] = val;
1346        break;
1347    case VIRTIO_PCI_COMMON_Q_AVAILHI:
1348        proxy->vqs[vdev->queue_sel].avail[1] = val;
1349        break;
1350    case VIRTIO_PCI_COMMON_Q_USEDLO:
1351        proxy->vqs[vdev->queue_sel].used[0] = val;
1352        break;
1353    case VIRTIO_PCI_COMMON_Q_USEDHI:
1354        proxy->vqs[vdev->queue_sel].used[1] = val;
1355        break;
1356    default:
1357        break;
1358    }
1359}
1360
1361
1362static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr,
1363                                       unsigned size)
1364{
1365    VirtIOPCIProxy *proxy = opaque;
1366    if (virtio_bus_get_device(&proxy->bus) == NULL) {
1367        return UINT64_MAX;
1368    }
1369
1370    return 0;
1371}
1372
1373static void virtio_pci_notify_write(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 = addr / virtio_pci_queue_mem_mult(proxy);
1380
1381    if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1382        virtio_queue_notify(vdev, queue);
1383    }
1384}
1385
1386static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr,
1387                                        uint64_t val, unsigned size)
1388{
1389    VirtIOPCIProxy *proxy = opaque;
1390    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1391
1392    unsigned queue = val;
1393
1394    if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1395        virtio_queue_notify(vdev, queue);
1396    }
1397}
1398
1399static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr,
1400                                    unsigned size)
1401{
1402    VirtIOPCIProxy *proxy = opaque;
1403    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1404    uint64_t val;
1405
1406    if (vdev == NULL) {
1407        return UINT64_MAX;
1408    }
1409
1410    val = qatomic_xchg(&vdev->isr, 0);
1411    pci_irq_deassert(&proxy->pci_dev);
1412    return val;
1413}
1414
1415static void virtio_pci_isr_write(void *opaque, hwaddr addr,
1416                                 uint64_t val, unsigned size)
1417{
1418}
1419
1420static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr,
1421                                       unsigned size)
1422{
1423    VirtIOPCIProxy *proxy = opaque;
1424    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1425    uint64_t val;
1426
1427    if (vdev == NULL) {
1428        return UINT64_MAX;
1429    }
1430
1431    switch (size) {
1432    case 1:
1433        val = virtio_config_modern_readb(vdev, addr);
1434        break;
1435    case 2:
1436        val = virtio_config_modern_readw(vdev, addr);
1437        break;
1438    case 4:
1439        val = virtio_config_modern_readl(vdev, addr);
1440        break;
1441    default:
1442        val = 0;
1443        break;
1444    }
1445    return val;
1446}
1447
1448static void virtio_pci_device_write(void *opaque, hwaddr addr,
1449                                    uint64_t val, unsigned size)
1450{
1451    VirtIOPCIProxy *proxy = opaque;
1452    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1453
1454    if (vdev == NULL) {
1455        return;
1456    }
1457
1458    switch (size) {
1459    case 1:
1460        virtio_config_modern_writeb(vdev, addr, val);
1461        break;
1462    case 2:
1463        virtio_config_modern_writew(vdev, addr, val);
1464        break;
1465    case 4:
1466        virtio_config_modern_writel(vdev, addr, val);
1467        break;
1468    }
1469}
1470
1471static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy,
1472                                           const char *vdev_name)
1473{
1474    static const MemoryRegionOps common_ops = {
1475        .read = virtio_pci_common_read,
1476        .write = virtio_pci_common_write,
1477        .impl = {
1478            .min_access_size = 1,
1479            .max_access_size = 4,
1480        },
1481        .endianness = DEVICE_LITTLE_ENDIAN,
1482    };
1483    static const MemoryRegionOps isr_ops = {
1484        .read = virtio_pci_isr_read,
1485        .write = virtio_pci_isr_write,
1486        .impl = {
1487            .min_access_size = 1,
1488            .max_access_size = 4,
1489        },
1490        .endianness = DEVICE_LITTLE_ENDIAN,
1491    };
1492    static const MemoryRegionOps device_ops = {
1493        .read = virtio_pci_device_read,
1494        .write = virtio_pci_device_write,
1495        .impl = {
1496            .min_access_size = 1,
1497            .max_access_size = 4,
1498        },
1499        .endianness = DEVICE_LITTLE_ENDIAN,
1500    };
1501    static const MemoryRegionOps notify_ops = {
1502        .read = virtio_pci_notify_read,
1503        .write = virtio_pci_notify_write,
1504        .impl = {
1505            .min_access_size = 1,
1506            .max_access_size = 4,
1507        },
1508        .endianness = DEVICE_LITTLE_ENDIAN,
1509    };
1510    static const MemoryRegionOps notify_pio_ops = {
1511        .read = virtio_pci_notify_read,
1512        .write = virtio_pci_notify_write_pio,
1513        .impl = {
1514            .min_access_size = 1,
1515            .max_access_size = 4,
1516        },
1517        .endianness = DEVICE_LITTLE_ENDIAN,
1518    };
1519    g_autoptr(GString) name = g_string_new(NULL);
1520
1521    g_string_printf(name, "virtio-pci-common-%s", vdev_name);
1522    memory_region_init_io(&proxy->common.mr, OBJECT(proxy),
1523                          &common_ops,
1524                          proxy,
1525                          name->str,
1526                          proxy->common.size);
1527
1528    g_string_printf(name, "virtio-pci-isr-%s", vdev_name);
1529    memory_region_init_io(&proxy->isr.mr, OBJECT(proxy),
1530                          &isr_ops,
1531                          proxy,
1532                          name->str,
1533                          proxy->isr.size);
1534
1535    g_string_printf(name, "virtio-pci-device-%s", vdev_name);
1536    memory_region_init_io(&proxy->device.mr, OBJECT(proxy),
1537                          &device_ops,
1538                          proxy,
1539                          name->str,
1540                          proxy->device.size);
1541
1542    g_string_printf(name, "virtio-pci-notify-%s", vdev_name);
1543    memory_region_init_io(&proxy->notify.mr, OBJECT(proxy),
1544                          &notify_ops,
1545                          proxy,
1546                          name->str,
1547                          proxy->notify.size);
1548
1549    g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name);
1550    memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy),
1551                          &notify_pio_ops,
1552                          proxy,
1553                          name->str,
1554                          proxy->notify_pio.size);
1555}
1556
1557static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy,
1558                                         VirtIOPCIRegion *region,
1559                                         struct virtio_pci_cap *cap,
1560                                         MemoryRegion *mr,
1561                                         uint8_t bar)
1562{
1563    memory_region_add_subregion(mr, region->offset, &region->mr);
1564
1565    cap->cfg_type = region->type;
1566    cap->bar = bar;
1567    cap->offset = cpu_to_le32(region->offset);
1568    cap->length = cpu_to_le32(region->size);
1569    virtio_pci_add_mem_cap(proxy, cap);
1570
1571}
1572
1573static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy,
1574                                             VirtIOPCIRegion *region,
1575                                             struct virtio_pci_cap *cap)
1576{
1577    virtio_pci_modern_region_map(proxy, region, cap,
1578                                 &proxy->modern_bar, proxy->modern_mem_bar_idx);
1579}
1580
1581static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy,
1582                                            VirtIOPCIRegion *region,
1583                                            struct virtio_pci_cap *cap)
1584{
1585    virtio_pci_modern_region_map(proxy, region, cap,
1586                                 &proxy->io_bar, proxy->modern_io_bar_idx);
1587}
1588
1589static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy,
1590                                               VirtIOPCIRegion *region)
1591{
1592    memory_region_del_subregion(&proxy->modern_bar,
1593                                &region->mr);
1594}
1595
1596static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy,
1597                                              VirtIOPCIRegion *region)
1598{
1599    memory_region_del_subregion(&proxy->io_bar,
1600                                &region->mr);
1601}
1602
1603static void virtio_pci_pre_plugged(DeviceState *d, Error **errp)
1604{
1605    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1606    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1607
1608    if (virtio_pci_modern(proxy)) {
1609        virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
1610    }
1611
1612    virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE);
1613}
1614
1615/* This is called by virtio-bus just after the device is plugged. */
1616static void virtio_pci_device_plugged(DeviceState *d, Error **errp)
1617{
1618    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1619    VirtioBusState *bus = &proxy->bus;
1620    bool legacy = virtio_pci_legacy(proxy);
1621    bool modern;
1622    bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1623    uint8_t *config;
1624    uint32_t size;
1625    VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1626
1627    /*
1628     * Virtio capabilities present without
1629     * VIRTIO_F_VERSION_1 confuses guests
1630     */
1631    if (!proxy->ignore_backend_features &&
1632            !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) {
1633        virtio_pci_disable_modern(proxy);
1634
1635        if (!legacy) {
1636            error_setg(errp, "Device doesn't support modern mode, and legacy"
1637                             " mode is disabled");
1638            error_append_hint(errp, "Set disable-legacy to off\n");
1639
1640            return;
1641        }
1642    }
1643
1644    modern = virtio_pci_modern(proxy);
1645
1646    config = proxy->pci_dev.config;
1647    if (proxy->class_code) {
1648        pci_config_set_class(config, proxy->class_code);
1649    }
1650
1651    if (legacy) {
1652        if (!virtio_legacy_allowed(vdev)) {
1653            /*
1654             * To avoid migration issues, we allow legacy mode when legacy
1655             * check is disabled in the old machine types (< 5.1).
1656             */
1657            if (virtio_legacy_check_disabled(vdev)) {
1658                warn_report("device is modern-only, but for backward "
1659                            "compatibility legacy is allowed");
1660            } else {
1661                error_setg(errp,
1662                           "device is modern-only, use disable-legacy=on");
1663                return;
1664            }
1665        }
1666        if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1667            error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by"
1668                       " neither legacy nor transitional device");
1669            return ;
1670        }
1671        /*
1672         * Legacy and transitional devices use specific subsystem IDs.
1673         * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID)
1674         * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default.
1675         */
1676        pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus));
1677    } else {
1678        /* pure virtio-1.0 */
1679        pci_set_word(config + PCI_VENDOR_ID,
1680                     PCI_VENDOR_ID_REDHAT_QUMRANET);
1681        pci_set_word(config + PCI_DEVICE_ID,
1682                     0x1040 + virtio_bus_get_vdev_id(bus));
1683        pci_config_set_revision(config, 1);
1684    }
1685    config[PCI_INTERRUPT_PIN] = 1;
1686
1687
1688    if (modern) {
1689        struct virtio_pci_cap cap = {
1690            .cap_len = sizeof cap,
1691        };
1692        struct virtio_pci_notify_cap notify = {
1693            .cap.cap_len = sizeof notify,
1694            .notify_off_multiplier =
1695                cpu_to_le32(virtio_pci_queue_mem_mult(proxy)),
1696        };
1697        struct virtio_pci_cfg_cap cfg = {
1698            .cap.cap_len = sizeof cfg,
1699            .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG,
1700        };
1701        struct virtio_pci_notify_cap notify_pio = {
1702            .cap.cap_len = sizeof notify,
1703            .notify_off_multiplier = cpu_to_le32(0x0),
1704        };
1705
1706        struct virtio_pci_cfg_cap *cfg_mask;
1707
1708        virtio_pci_modern_regions_init(proxy, vdev->name);
1709
1710        virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap);
1711        virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap);
1712        virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap);
1713        virtio_pci_modern_mem_region_map(proxy, &proxy->notify, &notify.cap);
1714
1715        if (modern_pio) {
1716            memory_region_init(&proxy->io_bar, OBJECT(proxy),
1717                               "virtio-pci-io", 0x4);
1718
1719            pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx,
1720                             PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar);
1721
1722            virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio,
1723                                            &notify_pio.cap);
1724        }
1725
1726        pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx,
1727                         PCI_BASE_ADDRESS_SPACE_MEMORY |
1728                         PCI_BASE_ADDRESS_MEM_PREFETCH |
1729                         PCI_BASE_ADDRESS_MEM_TYPE_64,
1730                         &proxy->modern_bar);
1731
1732        proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap);
1733        cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap);
1734        pci_set_byte(&cfg_mask->cap.bar, ~0x0);
1735        pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0);
1736        pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0);
1737        pci_set_long(cfg_mask->pci_cfg_data, ~0x0);
1738    }
1739
1740    if (proxy->nvectors) {
1741        int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors,
1742                                          proxy->msix_bar_idx, NULL);
1743        if (err) {
1744            /* Notice when a system that supports MSIx can't initialize it */
1745            if (err != -ENOTSUP) {
1746                warn_report("unable to init msix vectors to %" PRIu32,
1747                            proxy->nvectors);
1748            }
1749            proxy->nvectors = 0;
1750        }
1751    }
1752
1753    proxy->pci_dev.config_write = virtio_write_config;
1754    proxy->pci_dev.config_read = virtio_read_config;
1755
1756    if (legacy) {
1757        size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev)
1758            + virtio_bus_get_vdev_config_len(bus);
1759        size = pow2ceil(size);
1760
1761        memory_region_init_io(&proxy->bar, OBJECT(proxy),
1762                              &virtio_pci_config_ops,
1763                              proxy, "virtio-pci", size);
1764
1765        pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx,
1766                         PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar);
1767    }
1768}
1769
1770static void virtio_pci_device_unplugged(DeviceState *d)
1771{
1772    VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1773    bool modern = virtio_pci_modern(proxy);
1774    bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1775
1776    virtio_pci_stop_ioeventfd(proxy);
1777
1778    if (modern) {
1779        virtio_pci_modern_mem_region_unmap(proxy, &proxy->common);
1780        virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr);
1781        virtio_pci_modern_mem_region_unmap(proxy, &proxy->device);
1782        virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify);
1783        if (modern_pio) {
1784            virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio);
1785        }
1786    }
1787}
1788
1789static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
1790{
1791    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1792    VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev);
1793    bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1794                     !pci_bus_is_root(pci_get_bus(pci_dev));
1795
1796    if (kvm_enabled() && !kvm_has_many_ioeventfds()) {
1797        proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1798    }
1799
1800    /* fd-based ioevents can't be synchronized in record/replay */
1801    if (replay_mode != REPLAY_MODE_NONE) {
1802        proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1803    }
1804
1805    /*
1806     * virtio pci bar layout used by default.
1807     * subclasses can re-arrange things if needed.
1808     *
1809     *   region 0   --  virtio legacy io bar
1810     *   region 1   --  msi-x bar
1811     *   region 2   --  virtio modern io bar (off by default)
1812     *   region 4+5 --  virtio modern memory (64bit) bar
1813     *
1814     */
1815    proxy->legacy_io_bar_idx  = 0;
1816    proxy->msix_bar_idx       = 1;
1817    proxy->modern_io_bar_idx  = 2;
1818    proxy->modern_mem_bar_idx = 4;
1819
1820    proxy->common.offset = 0x0;
1821    proxy->common.size = 0x1000;
1822    proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG;
1823
1824    proxy->isr.offset = 0x1000;
1825    proxy->isr.size = 0x1000;
1826    proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG;
1827
1828    proxy->device.offset = 0x2000;
1829    proxy->device.size = 0x1000;
1830    proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG;
1831
1832    proxy->notify.offset = 0x3000;
1833    proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX;
1834    proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1835
1836    proxy->notify_pio.offset = 0x0;
1837    proxy->notify_pio.size = 0x4;
1838    proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1839
1840    /* subclasses can enforce modern, so do this unconditionally */
1841    memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci",
1842                       /* PCI BAR regions must be powers of 2 */
1843                       pow2ceil(proxy->notify.offset + proxy->notify.size));
1844
1845    if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) {
1846        proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
1847    }
1848
1849    if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) {
1850        error_setg(errp, "device cannot work as neither modern nor legacy mode"
1851                   " is enabled");
1852        error_append_hint(errp, "Set either disable-modern or disable-legacy"
1853                          " to off\n");
1854        return;
1855    }
1856
1857    if (pcie_port && pci_is_express(pci_dev)) {
1858        int pos;
1859        uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE;
1860
1861        pos = pcie_endpoint_cap_init(pci_dev, 0);
1862        assert(pos > 0);
1863
1864        pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0,
1865                                 PCI_PM_SIZEOF, errp);
1866        if (pos < 0) {
1867            return;
1868        }
1869
1870        pci_dev->exp.pm_cap = pos;
1871
1872        /*
1873         * Indicates that this function complies with revision 1.2 of the
1874         * PCI Power Management Interface Specification.
1875         */
1876        pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3);
1877
1878        if (proxy->flags & VIRTIO_PCI_FLAG_AER) {
1879            pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset,
1880                          PCI_ERR_SIZEOF, NULL);
1881            last_pcie_cap_offset += PCI_ERR_SIZEOF;
1882        }
1883
1884        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) {
1885            /* Init error enabling flags */
1886            pcie_cap_deverr_init(pci_dev);
1887        }
1888
1889        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) {
1890            /* Init Link Control Register */
1891            pcie_cap_lnkctl_init(pci_dev);
1892        }
1893
1894        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) {
1895            /* Init Power Management Control Register */
1896            pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL,
1897                         PCI_PM_CTRL_STATE_MASK);
1898        }
1899
1900        if (proxy->flags & VIRTIO_PCI_FLAG_ATS) {
1901            pcie_ats_init(pci_dev, last_pcie_cap_offset,
1902                          proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED);
1903            last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF;
1904        }
1905
1906        if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
1907            /* Set Function Level Reset capability bit */
1908            pcie_cap_flr_init(pci_dev);
1909        }
1910    } else {
1911        /*
1912         * make future invocations of pci_is_express() return false
1913         * and pci_config_size() return PCI_CONFIG_SPACE_SIZE.
1914         */
1915        pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS;
1916    }
1917
1918    virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
1919    if (k->realize) {
1920        k->realize(proxy, errp);
1921    }
1922}
1923
1924static void virtio_pci_exit(PCIDevice *pci_dev)
1925{
1926    VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1927    bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1928                     !pci_bus_is_root(pci_get_bus(pci_dev));
1929
1930    msix_uninit_exclusive_bar(pci_dev);
1931    if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port &&
1932        pci_is_express(pci_dev)) {
1933        pcie_aer_exit(pci_dev);
1934    }
1935}
1936
1937static void virtio_pci_reset(DeviceState *qdev)
1938{
1939    VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1940    VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
1941    PCIDevice *dev = PCI_DEVICE(qdev);
1942    int i;
1943
1944    virtio_pci_stop_ioeventfd(proxy);
1945    virtio_bus_reset(bus);
1946    msix_unuse_all_vectors(&proxy->pci_dev);
1947
1948    for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1949        proxy->vqs[i].enabled = 0;
1950        proxy->vqs[i].num = 0;
1951        proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0;
1952        proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0;
1953        proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0;
1954    }
1955
1956    if (pci_is_express(dev)) {
1957        pcie_cap_deverr_reset(dev);
1958        pcie_cap_lnkctl_reset(dev);
1959
1960        pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0);
1961    }
1962}
1963
1964static Property virtio_pci_properties[] = {
1965    DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags,
1966                    VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false),
1967    DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags,
1968                    VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true),
1969    DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags,
1970                    VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false),
1971    DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags,
1972                    VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false),
1973    DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags,
1974                    VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false),
1975    DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy,
1976                     ignore_backend_features, false),
1977    DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags,
1978                    VIRTIO_PCI_FLAG_ATS_BIT, false),
1979    DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags,
1980                    VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true),
1981    DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags,
1982                    VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true),
1983    DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags,
1984                    VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true),
1985    DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags,
1986                    VIRTIO_PCI_FLAG_INIT_PM_BIT, true),
1987    DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags,
1988                    VIRTIO_PCI_FLAG_INIT_FLR_BIT, true),
1989    DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags,
1990                    VIRTIO_PCI_FLAG_AER_BIT, false),
1991    DEFINE_PROP_END_OF_LIST(),
1992};
1993
1994static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp)
1995{
1996    VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev);
1997    VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1998    PCIDevice *pci_dev = &proxy->pci_dev;
1999
2000    if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) &&
2001        virtio_pci_modern(proxy)) {
2002        pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2003    }
2004
2005    vpciklass->parent_dc_realize(qdev, errp);
2006}
2007
2008static void virtio_pci_class_init(ObjectClass *klass, void *data)
2009{
2010    DeviceClass *dc = DEVICE_CLASS(klass);
2011    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
2012    VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass);
2013
2014    device_class_set_props(dc, virtio_pci_properties);
2015    k->realize = virtio_pci_realize;
2016    k->exit = virtio_pci_exit;
2017    k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
2018    k->revision = VIRTIO_PCI_ABI_VERSION;
2019    k->class_id = PCI_CLASS_OTHERS;
2020    device_class_set_parent_realize(dc, virtio_pci_dc_realize,
2021                                    &vpciklass->parent_dc_realize);
2022    dc->reset = virtio_pci_reset;
2023}
2024
2025static const TypeInfo virtio_pci_info = {
2026    .name          = TYPE_VIRTIO_PCI,
2027    .parent        = TYPE_PCI_DEVICE,
2028    .instance_size = sizeof(VirtIOPCIProxy),
2029    .class_init    = virtio_pci_class_init,
2030    .class_size    = sizeof(VirtioPCIClass),
2031    .abstract      = true,
2032};
2033
2034static Property virtio_pci_generic_properties[] = {
2035    DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy,
2036                            ON_OFF_AUTO_AUTO),
2037    DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false),
2038    DEFINE_PROP_END_OF_LIST(),
2039};
2040
2041static void virtio_pci_base_class_init(ObjectClass *klass, void *data)
2042{
2043    const VirtioPCIDeviceTypeInfo *t = data;
2044    if (t->class_init) {
2045        t->class_init(klass, NULL);
2046    }
2047}
2048
2049static void virtio_pci_generic_class_init(ObjectClass *klass, void *data)
2050{
2051    DeviceClass *dc = DEVICE_CLASS(klass);
2052
2053    device_class_set_props(dc, virtio_pci_generic_properties);
2054}
2055
2056static void virtio_pci_transitional_instance_init(Object *obj)
2057{
2058    VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2059
2060    proxy->disable_legacy = ON_OFF_AUTO_OFF;
2061    proxy->disable_modern = false;
2062}
2063
2064static void virtio_pci_non_transitional_instance_init(Object *obj)
2065{
2066    VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2067
2068    proxy->disable_legacy = ON_OFF_AUTO_ON;
2069    proxy->disable_modern = false;
2070}
2071
2072void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t)
2073{
2074    char *base_name = NULL;
2075    TypeInfo base_type_info = {
2076        .name          = t->base_name,
2077        .parent        = t->parent ? t->parent : TYPE_VIRTIO_PCI,
2078        .instance_size = t->instance_size,
2079        .instance_init = t->instance_init,
2080        .class_size    = t->class_size,
2081        .abstract      = true,
2082        .interfaces    = t->interfaces,
2083    };
2084    TypeInfo generic_type_info = {
2085        .name = t->generic_name,
2086        .parent = base_type_info.name,
2087        .class_init = virtio_pci_generic_class_init,
2088        .interfaces = (InterfaceInfo[]) {
2089            { INTERFACE_PCIE_DEVICE },
2090            { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2091            { }
2092        },
2093    };
2094
2095    if (!base_type_info.name) {
2096        /* No base type -> register a single generic device type */
2097        /* use intermediate %s-base-type to add generic device props */
2098        base_name = g_strdup_printf("%s-base-type", t->generic_name);
2099        base_type_info.name = base_name;
2100        base_type_info.class_init = virtio_pci_generic_class_init;
2101
2102        generic_type_info.parent = base_name;
2103        generic_type_info.class_init = virtio_pci_base_class_init;
2104        generic_type_info.class_data = (void *)t;
2105
2106        assert(!t->non_transitional_name);
2107        assert(!t->transitional_name);
2108    } else {
2109        base_type_info.class_init = virtio_pci_base_class_init;
2110        base_type_info.class_data = (void *)t;
2111    }
2112
2113    type_register(&base_type_info);
2114    if (generic_type_info.name) {
2115        type_register(&generic_type_info);
2116    }
2117
2118    if (t->non_transitional_name) {
2119        const TypeInfo non_transitional_type_info = {
2120            .name          = t->non_transitional_name,
2121            .parent        = base_type_info.name,
2122            .instance_init = virtio_pci_non_transitional_instance_init,
2123            .interfaces = (InterfaceInfo[]) {
2124                { INTERFACE_PCIE_DEVICE },
2125                { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2126                { }
2127            },
2128        };
2129        type_register(&non_transitional_type_info);
2130    }
2131
2132    if (t->transitional_name) {
2133        const TypeInfo transitional_type_info = {
2134            .name          = t->transitional_name,
2135            .parent        = base_type_info.name,
2136            .instance_init = virtio_pci_transitional_instance_init,
2137            .interfaces = (InterfaceInfo[]) {
2138                /*
2139                 * Transitional virtio devices work only as Conventional PCI
2140                 * devices because they require PIO ports.
2141                 */
2142                { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2143                { }
2144            },
2145        };
2146        type_register(&transitional_type_info);
2147    }
2148    g_free(base_name);
2149}
2150
2151unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues)
2152{
2153    /*
2154     * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted
2155     * virtqueue buffers can handle their completion. When a different vCPU
2156     * handles completion it may need to IPI the vCPU that submitted the
2157     * request and this adds overhead.
2158     *
2159     * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in
2160     * guests with very many vCPUs and a device that is only used by a few
2161     * vCPUs. Unfortunately optimizing that case requires manual pinning inside
2162     * the guest, so those users might as well manually set the number of
2163     * queues. There is no upper limit that can be applied automatically and
2164     * doing so arbitrarily would result in a sudden performance drop once the
2165     * threshold number of vCPUs is exceeded.
2166     */
2167    unsigned num_queues = current_machine->smp.cpus;
2168
2169    /*
2170     * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the
2171     * config change interrupt and the fixed virtqueues must be taken into
2172     * account too.
2173     */
2174    num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues);
2175
2176    /*
2177     * There is a limit to how many virtqueues a device can have.
2178     */
2179    return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues);
2180}
2181
2182/* virtio-pci-bus */
2183
2184static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
2185                               VirtIOPCIProxy *dev)
2186{
2187    DeviceState *qdev = DEVICE(dev);
2188    char virtio_bus_name[] = "virtio-bus";
2189
2190    qbus_init(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev, virtio_bus_name);
2191}
2192
2193static void virtio_pci_bus_class_init(ObjectClass *klass, void *data)
2194{
2195    BusClass *bus_class = BUS_CLASS(klass);
2196    VirtioBusClass *k = VIRTIO_BUS_CLASS(klass);
2197    bus_class->max_dev = 1;
2198    k->notify = virtio_pci_notify;
2199    k->save_config = virtio_pci_save_config;
2200    k->load_config = virtio_pci_load_config;
2201    k->save_queue = virtio_pci_save_queue;
2202    k->load_queue = virtio_pci_load_queue;
2203    k->save_extra_state = virtio_pci_save_extra_state;
2204    k->load_extra_state = virtio_pci_load_extra_state;
2205    k->has_extra_state = virtio_pci_has_extra_state;
2206    k->query_guest_notifiers = virtio_pci_query_guest_notifiers;
2207    k->set_guest_notifiers = virtio_pci_set_guest_notifiers;
2208    k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr;
2209    k->vmstate_change = virtio_pci_vmstate_change;
2210    k->pre_plugged = virtio_pci_pre_plugged;
2211    k->device_plugged = virtio_pci_device_plugged;
2212    k->device_unplugged = virtio_pci_device_unplugged;
2213    k->query_nvectors = virtio_pci_query_nvectors;
2214    k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled;
2215    k->ioeventfd_assign = virtio_pci_ioeventfd_assign;
2216    k->get_dma_as = virtio_pci_get_dma_as;
2217    k->iommu_enabled = virtio_pci_iommu_enabled;
2218    k->queue_enabled = virtio_pci_queue_enabled;
2219}
2220
2221static const TypeInfo virtio_pci_bus_info = {
2222    .name          = TYPE_VIRTIO_PCI_BUS,
2223    .parent        = TYPE_VIRTIO_BUS,
2224    .instance_size = sizeof(VirtioPCIBusState),
2225    .class_size    = sizeof(VirtioPCIBusClass),
2226    .class_init    = virtio_pci_bus_class_init,
2227};
2228
2229static void virtio_pci_register_types(void)
2230{
2231    /* Base types: */
2232    type_register_static(&virtio_pci_bus_info);
2233    type_register_static(&virtio_pci_info);
2234}
2235
2236type_init(virtio_pci_register_types)
2237
2238