qemu/hw/i386/kvm/pci-assign.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2007, Neocleus Corporation.
   3 *
   4 * This work is licensed under the terms of the GNU GPL, version 2.  See
   5 * the COPYING file in the top-level directory.
   6 *
   7 *
   8 *  Assign a PCI device from the host to a guest VM.
   9 *
  10 *  This implementation uses the classic device assignment interface of KVM
  11 *  and is only available on x86 hosts. It is expected to be obsoleted by VFIO
  12 *  based device assignment.
  13 *
  14 *  Adapted for KVM (qemu-kvm) by Qumranet. QEMU version was based on qemu-kvm
  15 *  revision 4144fe9d48. See its repository for the history.
  16 *
  17 *  Copyright (c) 2007, Neocleus, Alex Novik (alex@neocleus.com)
  18 *  Copyright (c) 2007, Neocleus, Guy Zana (guy@neocleus.com)
  19 *  Copyright (C) 2008, Qumranet, Amit Shah (amit.shah@qumranet.com)
  20 *  Copyright (C) 2008, Red Hat, Amit Shah (amit.shah@redhat.com)
  21 *  Copyright (C) 2008, IBM, Muli Ben-Yehuda (muli@il.ibm.com)
  22 */
  23#include "qemu/osdep.h"
  24#include "qapi/error.h"
  25#include <sys/mman.h>
  26#include "hw/hw.h"
  27#include "hw/i386/pc.h"
  28#include "qemu/error-report.h"
  29#include "ui/console.h"
  30#include "hw/loader.h"
  31#include "monitor/monitor.h"
  32#include "qemu/range.h"
  33#include "sysemu/sysemu.h"
  34#include "hw/pci/pci.h"
  35#include "hw/pci/msi.h"
  36#include "kvm_i386.h"
  37#include "hw/pci/pci-assign.h"
  38
  39#define MSIX_PAGE_SIZE 0x1000
  40
  41/* From linux/ioport.h */
  42#define IORESOURCE_IO       0x00000100  /* Resource type */
  43#define IORESOURCE_MEM      0x00000200
  44#define IORESOURCE_IRQ      0x00000400
  45#define IORESOURCE_DMA      0x00000800
  46#define IORESOURCE_PREFETCH 0x00002000  /* No side effects */
  47#define IORESOURCE_MEM_64   0x00100000
  48
  49typedef struct PCIRegion {
  50    int type;           /* Memory or port I/O */
  51    int valid;
  52    uint64_t base_addr;
  53    uint64_t size;    /* size of the region */
  54    int resource_fd;
  55} PCIRegion;
  56
  57typedef struct PCIDevRegions {
  58    uint8_t bus, dev, func; /* Bus inside domain, device and function */
  59    int irq;                /* IRQ number */
  60    uint16_t region_number; /* number of active regions */
  61
  62    /* Port I/O or MMIO Regions */
  63    PCIRegion regions[PCI_NUM_REGIONS - 1];
  64    int config_fd;
  65} PCIDevRegions;
  66
  67typedef struct AssignedDevRegion {
  68    MemoryRegion container;
  69    MemoryRegion real_iomem;
  70    union {
  71        uint8_t *r_virtbase; /* mmapped access address for memory regions */
  72        uint32_t r_baseport; /* the base guest port for I/O regions */
  73    } u;
  74    pcibus_t e_size;    /* emulated size of region in bytes */
  75    pcibus_t r_size;    /* real size of region in bytes */
  76    PCIRegion *region;
  77} AssignedDevRegion;
  78
  79#define ASSIGNED_DEVICE_PREFER_MSI_BIT  0
  80#define ASSIGNED_DEVICE_SHARE_INTX_BIT  1
  81
  82#define ASSIGNED_DEVICE_PREFER_MSI_MASK (1 << ASSIGNED_DEVICE_PREFER_MSI_BIT)
  83#define ASSIGNED_DEVICE_SHARE_INTX_MASK (1 << ASSIGNED_DEVICE_SHARE_INTX_BIT)
  84
  85typedef struct MSIXTableEntry {
  86    uint32_t addr_lo;
  87    uint32_t addr_hi;
  88    uint32_t data;
  89    uint32_t ctrl;
  90} MSIXTableEntry;
  91
  92typedef enum AssignedIRQType {
  93    ASSIGNED_IRQ_NONE = 0,
  94    ASSIGNED_IRQ_INTX_HOST_INTX,
  95    ASSIGNED_IRQ_INTX_HOST_MSI,
  96    ASSIGNED_IRQ_MSI,
  97    ASSIGNED_IRQ_MSIX
  98} AssignedIRQType;
  99
 100typedef struct AssignedDevice {
 101    PCIDevice dev;
 102    PCIHostDeviceAddress host;
 103    uint32_t dev_id;
 104    uint32_t features;
 105    int intpin;
 106    AssignedDevRegion v_addrs[PCI_NUM_REGIONS - 1];
 107    PCIDevRegions real_device;
 108    PCIINTxRoute intx_route;
 109    AssignedIRQType assigned_irq_type;
 110    struct {
 111#define ASSIGNED_DEVICE_CAP_MSI (1 << 0)
 112#define ASSIGNED_DEVICE_CAP_MSIX (1 << 1)
 113        uint32_t available;
 114#define ASSIGNED_DEVICE_MSI_ENABLED (1 << 0)
 115#define ASSIGNED_DEVICE_MSIX_ENABLED (1 << 1)
 116#define ASSIGNED_DEVICE_MSIX_MASKED (1 << 2)
 117        uint32_t state;
 118    } cap;
 119    uint8_t emulate_config_read[PCI_CONFIG_SPACE_SIZE];
 120    uint8_t emulate_config_write[PCI_CONFIG_SPACE_SIZE];
 121    int msi_virq_nr;
 122    int *msi_virq;
 123    MSIXTableEntry *msix_table;
 124    hwaddr msix_table_addr;
 125    uint16_t msix_max;
 126    MemoryRegion mmio;
 127    char *configfd_name;
 128    int32_t bootindex;
 129} AssignedDevice;
 130
 131#define TYPE_PCI_ASSIGN "kvm-pci-assign"
 132#define PCI_ASSIGN(obj) OBJECT_CHECK(AssignedDevice, (obj), TYPE_PCI_ASSIGN)
 133
 134static void assigned_dev_update_irq_routing(PCIDevice *dev);
 135
 136static void assigned_dev_load_option_rom(AssignedDevice *dev);
 137
 138static void assigned_dev_unregister_msix_mmio(AssignedDevice *dev);
 139
 140static uint64_t assigned_dev_ioport_rw(AssignedDevRegion *dev_region,
 141                                       hwaddr addr, int size,
 142                                       uint64_t *data)
 143{
 144    uint64_t val = 0;
 145    int fd = dev_region->region->resource_fd;
 146
 147    if (data) {
 148        DEBUG("pwrite data=%" PRIx64 ", size=%d, e_phys=" TARGET_FMT_plx
 149              ", addr="TARGET_FMT_plx"\n", *data, size, addr, addr);
 150        if (pwrite(fd, data, size, addr) != size) {
 151            error_report("%s - pwrite failed %s", __func__, strerror(errno));
 152        }
 153    } else {
 154        if (pread(fd, &val, size, addr) != size) {
 155            error_report("%s - pread failed %s", __func__, strerror(errno));
 156            val = (1UL << (size * 8)) - 1;
 157        }
 158        DEBUG("pread val=%" PRIx64 ", size=%d, e_phys=" TARGET_FMT_plx
 159              ", addr=" TARGET_FMT_plx "\n", val, size, addr, addr);
 160    }
 161    return val;
 162}
 163
 164static void assigned_dev_ioport_write(void *opaque, hwaddr addr,
 165                                      uint64_t data, unsigned size)
 166{
 167    assigned_dev_ioport_rw(opaque, addr, size, &data);
 168}
 169
 170static uint64_t assigned_dev_ioport_read(void *opaque,
 171                                         hwaddr addr, unsigned size)
 172{
 173    return assigned_dev_ioport_rw(opaque, addr, size, NULL);
 174}
 175
 176static uint32_t slow_bar_readb(void *opaque, hwaddr addr)
 177{
 178    AssignedDevRegion *d = opaque;
 179    uint8_t *in = d->u.r_virtbase + addr;
 180    uint32_t r;
 181
 182    r = *in;
 183    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
 184
 185    return r;
 186}
 187
 188static uint32_t slow_bar_readw(void *opaque, hwaddr addr)
 189{
 190    AssignedDevRegion *d = opaque;
 191    uint16_t *in = (uint16_t *)(d->u.r_virtbase + addr);
 192    uint32_t r;
 193
 194    r = *in;
 195    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
 196
 197    return r;
 198}
 199
 200static uint32_t slow_bar_readl(void *opaque, hwaddr addr)
 201{
 202    AssignedDevRegion *d = opaque;
 203    uint32_t *in = (uint32_t *)(d->u.r_virtbase + addr);
 204    uint32_t r;
 205
 206    r = *in;
 207    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, r);
 208
 209    return r;
 210}
 211
 212static void slow_bar_writeb(void *opaque, hwaddr addr, uint32_t val)
 213{
 214    AssignedDevRegion *d = opaque;
 215    uint8_t *out = d->u.r_virtbase + addr;
 216
 217    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%02x\n", addr, val);
 218    *out = val;
 219}
 220
 221static void slow_bar_writew(void *opaque, hwaddr addr, uint32_t val)
 222{
 223    AssignedDevRegion *d = opaque;
 224    uint16_t *out = (uint16_t *)(d->u.r_virtbase + addr);
 225
 226    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%04x\n", addr, val);
 227    *out = val;
 228}
 229
 230static void slow_bar_writel(void *opaque, hwaddr addr, uint32_t val)
 231{
 232    AssignedDevRegion *d = opaque;
 233    uint32_t *out = (uint32_t *)(d->u.r_virtbase + addr);
 234
 235    DEBUG("addr=0x" TARGET_FMT_plx " val=0x%08x\n", addr, val);
 236    *out = val;
 237}
 238
 239static const MemoryRegionOps slow_bar_ops = {
 240    .old_mmio = {
 241        .read = { slow_bar_readb, slow_bar_readw, slow_bar_readl, },
 242        .write = { slow_bar_writeb, slow_bar_writew, slow_bar_writel, },
 243    },
 244    .endianness = DEVICE_NATIVE_ENDIAN,
 245};
 246
 247static void assigned_dev_iomem_setup(PCIDevice *pci_dev, int region_num,
 248                                     pcibus_t e_size)
 249{
 250    AssignedDevice *r_dev = PCI_ASSIGN(pci_dev);
 251    AssignedDevRegion *region = &r_dev->v_addrs[region_num];
 252    PCIRegion *real_region = &r_dev->real_device.regions[region_num];
 253
 254    if (e_size > 0) {
 255        memory_region_init(&region->container, OBJECT(pci_dev),
 256                           "assigned-dev-container", e_size);
 257        memory_region_add_subregion(&region->container, 0, &region->real_iomem);
 258
 259        /* deal with MSI-X MMIO page */
 260        if (real_region->base_addr <= r_dev->msix_table_addr &&
 261                real_region->base_addr + real_region->size >
 262                r_dev->msix_table_addr) {
 263            uint64_t offset = r_dev->msix_table_addr - real_region->base_addr;
 264
 265            memory_region_add_subregion_overlap(&region->container,
 266                                                offset,
 267                                                &r_dev->mmio,
 268                                                1);
 269        }
 270    }
 271}
 272
 273static const MemoryRegionOps assigned_dev_ioport_ops = {
 274    .read = assigned_dev_ioport_read,
 275    .write = assigned_dev_ioport_write,
 276    .endianness = DEVICE_NATIVE_ENDIAN,
 277};
 278
 279static void assigned_dev_ioport_setup(PCIDevice *pci_dev, int region_num,
 280                                      pcibus_t size)
 281{
 282    AssignedDevice *r_dev = PCI_ASSIGN(pci_dev);
 283    AssignedDevRegion *region = &r_dev->v_addrs[region_num];
 284
 285    region->e_size = size;
 286    memory_region_init(&region->container, OBJECT(pci_dev),
 287                       "assigned-dev-container", size);
 288    memory_region_init_io(&region->real_iomem, OBJECT(pci_dev),
 289                          &assigned_dev_ioport_ops, r_dev->v_addrs + region_num,
 290                          "assigned-dev-iomem", size);
 291    memory_region_add_subregion(&region->container, 0, &region->real_iomem);
 292}
 293
 294static uint32_t assigned_dev_pci_read(PCIDevice *d, int pos, int len)
 295{
 296    AssignedDevice *pci_dev = PCI_ASSIGN(d);
 297    uint32_t val;
 298    ssize_t ret;
 299    int fd = pci_dev->real_device.config_fd;
 300
 301again:
 302    ret = pread(fd, &val, len, pos);
 303    if (ret != len) {
 304        if ((ret < 0) && (errno == EINTR || errno == EAGAIN)) {
 305            goto again;
 306        }
 307
 308        hw_error("pci read failed, ret = %zd errno = %d\n", ret, errno);
 309    }
 310
 311    return val;
 312}
 313
 314static uint8_t assigned_dev_pci_read_byte(PCIDevice *d, int pos)
 315{
 316    return (uint8_t)assigned_dev_pci_read(d, pos, 1);
 317}
 318
 319static void assigned_dev_pci_write(PCIDevice *d, int pos, uint32_t val, int len)
 320{
 321    AssignedDevice *pci_dev = PCI_ASSIGN(d);
 322    ssize_t ret;
 323    int fd = pci_dev->real_device.config_fd;
 324
 325again:
 326    ret = pwrite(fd, &val, len, pos);
 327    if (ret != len) {
 328        if ((ret < 0) && (errno == EINTR || errno == EAGAIN)) {
 329            goto again;
 330        }
 331
 332        hw_error("pci write failed, ret = %zd errno = %d\n", ret, errno);
 333    }
 334}
 335
 336static void assigned_dev_emulate_config_read(AssignedDevice *dev,
 337                                             uint32_t offset, uint32_t len)
 338{
 339    memset(dev->emulate_config_read + offset, 0xff, len);
 340}
 341
 342static void assigned_dev_direct_config_read(AssignedDevice *dev,
 343                                            uint32_t offset, uint32_t len)
 344{
 345    memset(dev->emulate_config_read + offset, 0, len);
 346}
 347
 348static void assigned_dev_direct_config_write(AssignedDevice *dev,
 349                                             uint32_t offset, uint32_t len)
 350{
 351    memset(dev->emulate_config_write + offset, 0, len);
 352}
 353
 354static uint8_t pci_find_cap_offset(PCIDevice *d, uint8_t cap, uint8_t start)
 355{
 356    int id;
 357    int max_cap = 48;
 358    int pos = start ? start : PCI_CAPABILITY_LIST;
 359    int status;
 360
 361    status = assigned_dev_pci_read_byte(d, PCI_STATUS);
 362    if ((status & PCI_STATUS_CAP_LIST) == 0) {
 363        return 0;
 364    }
 365
 366    while (max_cap--) {
 367        pos = assigned_dev_pci_read_byte(d, pos);
 368        if (pos < 0x40) {
 369            break;
 370        }
 371
 372        pos &= ~3;
 373        id = assigned_dev_pci_read_byte(d, pos + PCI_CAP_LIST_ID);
 374
 375        if (id == 0xff) {
 376            break;
 377        }
 378        if (id == cap) {
 379            return pos;
 380        }
 381
 382        pos += PCI_CAP_LIST_NEXT;
 383    }
 384    return 0;
 385}
 386
 387static void assigned_dev_register_regions(PCIRegion *io_regions,
 388                                          unsigned long regions_num,
 389                                          AssignedDevice *pci_dev,
 390                                          Error **errp)
 391{
 392    uint32_t i;
 393    PCIRegion *cur_region = io_regions;
 394
 395    for (i = 0; i < regions_num; i++, cur_region++) {
 396        if (!cur_region->valid) {
 397            continue;
 398        }
 399
 400        /* handle memory io regions */
 401        if (cur_region->type & IORESOURCE_MEM) {
 402            int t = PCI_BASE_ADDRESS_SPACE_MEMORY;
 403            if (cur_region->type & IORESOURCE_PREFETCH) {
 404                t |= PCI_BASE_ADDRESS_MEM_PREFETCH;
 405            }
 406            if (cur_region->type & IORESOURCE_MEM_64) {
 407                t |= PCI_BASE_ADDRESS_MEM_TYPE_64;
 408            }
 409
 410            /* map physical memory */
 411            pci_dev->v_addrs[i].u.r_virtbase = mmap(NULL, cur_region->size,
 412                                                    PROT_WRITE | PROT_READ,
 413                                                    MAP_SHARED,
 414                                                    cur_region->resource_fd,
 415                                                    (off_t)0);
 416
 417            if (pci_dev->v_addrs[i].u.r_virtbase == MAP_FAILED) {
 418                pci_dev->v_addrs[i].u.r_virtbase = NULL;
 419                error_setg_errno(errp, errno, "Couldn't mmap 0x%" PRIx64 "!",
 420                                 cur_region->base_addr);
 421                return;
 422            }
 423
 424            pci_dev->v_addrs[i].r_size = cur_region->size;
 425            pci_dev->v_addrs[i].e_size = 0;
 426
 427            /* add offset */
 428            pci_dev->v_addrs[i].u.r_virtbase +=
 429                (cur_region->base_addr & 0xFFF);
 430
 431            if (cur_region->size & 0xFFF) {
 432                error_report("PCI region %d at address 0x%" PRIx64 " has "
 433                             "size 0x%" PRIx64 ", which is not a multiple of "
 434                             "4K.  You might experience some performance hit "
 435                             "due to that.",
 436                             i, cur_region->base_addr, cur_region->size);
 437                memory_region_init_io(&pci_dev->v_addrs[i].real_iomem,
 438                                      OBJECT(pci_dev), &slow_bar_ops,
 439                                      &pci_dev->v_addrs[i],
 440                                      "assigned-dev-slow-bar",
 441                                      cur_region->size);
 442            } else {
 443                void *virtbase = pci_dev->v_addrs[i].u.r_virtbase;
 444                char name[32];
 445                snprintf(name, sizeof(name), "%s.bar%d",
 446                         object_get_typename(OBJECT(pci_dev)), i);
 447                memory_region_init_ram_ptr(&pci_dev->v_addrs[i].real_iomem,
 448                                           OBJECT(pci_dev), name,
 449                                           cur_region->size, virtbase);
 450                vmstate_register_ram(&pci_dev->v_addrs[i].real_iomem,
 451                                     &pci_dev->dev.qdev);
 452            }
 453
 454            assigned_dev_iomem_setup(&pci_dev->dev, i, cur_region->size);
 455            pci_register_bar((PCIDevice *) pci_dev, i, t,
 456                             &pci_dev->v_addrs[i].container);
 457            continue;
 458        } else {
 459            /* handle port io regions */
 460            uint32_t val;
 461            int ret;
 462
 463            /* Test kernel support for ioport resource read/write.  Old
 464             * kernels return EIO.  New kernels only allow 1/2/4 byte reads
 465             * so should return EINVAL for a 3 byte read */
 466            ret = pread(pci_dev->v_addrs[i].region->resource_fd, &val, 3, 0);
 467            if (ret >= 0) {
 468                error_report("Unexpected return from I/O port read: %d", ret);
 469                abort();
 470            } else if (errno != EINVAL) {
 471                error_report("Kernel doesn't support ioport resource "
 472                             "access, hiding this region.");
 473                close(pci_dev->v_addrs[i].region->resource_fd);
 474                cur_region->valid = 0;
 475                continue;
 476            }
 477
 478            pci_dev->v_addrs[i].u.r_baseport = cur_region->base_addr;
 479            pci_dev->v_addrs[i].r_size = cur_region->size;
 480            pci_dev->v_addrs[i].e_size = 0;
 481
 482            assigned_dev_ioport_setup(&pci_dev->dev, i, cur_region->size);
 483            pci_register_bar((PCIDevice *) pci_dev, i,
 484                             PCI_BASE_ADDRESS_SPACE_IO,
 485                             &pci_dev->v_addrs[i].container);
 486        }
 487    }
 488
 489    /* success */
 490}
 491
 492static void get_real_id(const char *devpath, const char *idname, uint16_t *val,
 493                        Error **errp)
 494{
 495    FILE *f;
 496    char name[128];
 497    long id;
 498
 499    snprintf(name, sizeof(name), "%s%s", devpath, idname);
 500    f = fopen(name, "r");
 501    if (f == NULL) {
 502        error_setg_file_open(errp, errno, name);
 503        return;
 504    }
 505    if (fscanf(f, "%li\n", &id) == 1) {
 506        *val = id;
 507    } else {
 508        error_setg(errp, "Failed to parse contents of '%s'", name);
 509    }
 510    fclose(f);
 511}
 512
 513static void get_real_vendor_id(const char *devpath, uint16_t *val,
 514                               Error **errp)
 515{
 516    get_real_id(devpath, "vendor", val, errp);
 517}
 518
 519static void get_real_device_id(const char *devpath, uint16_t *val,
 520                               Error **errp)
 521{
 522    get_real_id(devpath, "device", val, errp);
 523}
 524
 525static void get_real_device(AssignedDevice *pci_dev, Error **errp)
 526{
 527    char dir[128], name[128];
 528    int fd, r = 0;
 529    FILE *f;
 530    uint64_t start, end, size, flags;
 531    uint16_t id;
 532    PCIRegion *rp;
 533    PCIDevRegions *dev = &pci_dev->real_device;
 534    Error *local_err = NULL;
 535
 536    dev->region_number = 0;
 537
 538    snprintf(dir, sizeof(dir), "/sys/bus/pci/devices/%04x:%02x:%02x.%x/",
 539             pci_dev->host.domain, pci_dev->host.bus,
 540             pci_dev->host.slot, pci_dev->host.function);
 541
 542    snprintf(name, sizeof(name), "%sconfig", dir);
 543
 544    if (pci_dev->configfd_name && *pci_dev->configfd_name) {
 545        dev->config_fd = monitor_fd_param(cur_mon, pci_dev->configfd_name,
 546                                          &local_err);
 547        if (local_err) {
 548            error_propagate(errp, local_err);
 549            return;
 550        }
 551    } else {
 552        dev->config_fd = open(name, O_RDWR);
 553
 554        if (dev->config_fd == -1) {
 555            error_setg_file_open(errp, errno, name);
 556            return;
 557        }
 558    }
 559again:
 560    r = read(dev->config_fd, pci_dev->dev.config,
 561             pci_config_size(&pci_dev->dev));
 562    if (r < 0) {
 563        if (errno == EINTR || errno == EAGAIN) {
 564            goto again;
 565        }
 566        error_setg_errno(errp, errno, "read(\"%s\")",
 567                         (pci_dev->configfd_name && *pci_dev->configfd_name) ?
 568                         pci_dev->configfd_name : name);
 569        return;
 570    }
 571
 572    /* Restore or clear multifunction, this is always controlled by qemu */
 573    if (pci_dev->dev.cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
 574        pci_dev->dev.config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
 575    } else {
 576        pci_dev->dev.config[PCI_HEADER_TYPE] &= ~PCI_HEADER_TYPE_MULTI_FUNCTION;
 577    }
 578
 579    /* Clear host resource mapping info.  If we choose not to register a
 580     * BAR, such as might be the case with the option ROM, we can get
 581     * confusing, unwritable, residual addresses from the host here. */
 582    memset(&pci_dev->dev.config[PCI_BASE_ADDRESS_0], 0, 24);
 583    memset(&pci_dev->dev.config[PCI_ROM_ADDRESS], 0, 4);
 584
 585    snprintf(name, sizeof(name), "%sresource", dir);
 586
 587    f = fopen(name, "r");
 588    if (f == NULL) {
 589        error_setg_file_open(errp, errno, name);
 590        return;
 591    }
 592
 593    for (r = 0; r < PCI_ROM_SLOT; r++) {
 594        if (fscanf(f, "%" SCNi64 " %" SCNi64 " %" SCNi64 "\n",
 595                   &start, &end, &flags) != 3) {
 596            break;
 597        }
 598
 599        rp = dev->regions + r;
 600        rp->valid = 0;
 601        rp->resource_fd = -1;
 602        size = end - start + 1;
 603        flags &= IORESOURCE_IO | IORESOURCE_MEM | IORESOURCE_PREFETCH
 604                 | IORESOURCE_MEM_64;
 605        if (size == 0 || (flags & ~IORESOURCE_PREFETCH) == 0) {
 606            continue;
 607        }
 608        if (flags & IORESOURCE_MEM) {
 609            flags &= ~IORESOURCE_IO;
 610        } else {
 611            flags &= ~IORESOURCE_PREFETCH;
 612        }
 613        snprintf(name, sizeof(name), "%sresource%d", dir, r);
 614        fd = open(name, O_RDWR);
 615        if (fd == -1) {
 616            continue;
 617        }
 618        rp->resource_fd = fd;
 619
 620        rp->type = flags;
 621        rp->valid = 1;
 622        rp->base_addr = start;
 623        rp->size = size;
 624        pci_dev->v_addrs[r].region = rp;
 625        DEBUG("region %d size %" PRIu64 " start 0x%" PRIx64
 626              " type %d resource_fd %d\n",
 627              r, rp->size, start, rp->type, rp->resource_fd);
 628    }
 629
 630    fclose(f);
 631
 632    /* read and fill vendor ID */
 633    get_real_vendor_id(dir, &id, &local_err);
 634    if (local_err) {
 635        error_propagate(errp, local_err);
 636        return;
 637    }
 638    pci_dev->dev.config[0] = id & 0xff;
 639    pci_dev->dev.config[1] = (id & 0xff00) >> 8;
 640
 641    /* read and fill device ID */
 642    get_real_device_id(dir, &id, &local_err);
 643    if (local_err) {
 644        error_propagate(errp, local_err);
 645        return;
 646    }
 647    pci_dev->dev.config[2] = id & 0xff;
 648    pci_dev->dev.config[3] = (id & 0xff00) >> 8;
 649
 650    pci_word_test_and_clear_mask(pci_dev->emulate_config_write + PCI_COMMAND,
 651                                 PCI_COMMAND_MASTER | PCI_COMMAND_INTX_DISABLE);
 652
 653    dev->region_number = r;
 654}
 655
 656static void free_msi_virqs(AssignedDevice *dev)
 657{
 658    int i;
 659
 660    for (i = 0; i < dev->msi_virq_nr; i++) {
 661        if (dev->msi_virq[i] >= 0) {
 662            kvm_irqchip_release_virq(kvm_state, dev->msi_virq[i]);
 663            dev->msi_virq[i] = -1;
 664        }
 665    }
 666    g_free(dev->msi_virq);
 667    dev->msi_virq = NULL;
 668    dev->msi_virq_nr = 0;
 669}
 670
 671static void free_assigned_device(AssignedDevice *dev)
 672{
 673    int i;
 674
 675    if (dev->cap.available & ASSIGNED_DEVICE_CAP_MSIX) {
 676        assigned_dev_unregister_msix_mmio(dev);
 677    }
 678    for (i = 0; i < dev->real_device.region_number; i++) {
 679        PCIRegion *pci_region = &dev->real_device.regions[i];
 680        AssignedDevRegion *region = &dev->v_addrs[i];
 681
 682        if (!pci_region->valid) {
 683            continue;
 684        }
 685        if (pci_region->type & IORESOURCE_IO) {
 686            if (region->u.r_baseport) {
 687                memory_region_del_subregion(&region->container,
 688                                            &region->real_iomem);
 689            }
 690        } else if (pci_region->type & IORESOURCE_MEM) {
 691            if (region->u.r_virtbase) {
 692                memory_region_del_subregion(&region->container,
 693                                            &region->real_iomem);
 694
 695                /* Remove MSI-X table subregion */
 696                if (pci_region->base_addr <= dev->msix_table_addr &&
 697                    pci_region->base_addr + pci_region->size >
 698                    dev->msix_table_addr) {
 699                    memory_region_del_subregion(&region->container,
 700                                                &dev->mmio);
 701                }
 702                if (munmap(region->u.r_virtbase,
 703                           (pci_region->size + 0xFFF) & 0xFFFFF000)) {
 704                    error_report("Failed to unmap assigned device region: %s",
 705                                 strerror(errno));
 706                }
 707            }
 708        }
 709        if (pci_region->resource_fd >= 0) {
 710            close(pci_region->resource_fd);
 711        }
 712    }
 713
 714    if (dev->real_device.config_fd >= 0) {
 715        close(dev->real_device.config_fd);
 716    }
 717
 718    free_msi_virqs(dev);
 719}
 720
 721/* This function tries to determine the cause of the PCI assignment failure. It
 722 * always returns the cause as a dynamically allocated, human readable string.
 723 * If the function fails to determine the cause for any internal reason, then
 724 * the returned string will state that fact.
 725 */
 726static char *assign_failed_examine(const AssignedDevice *dev)
 727{
 728    char name[PATH_MAX], dir[PATH_MAX], driver[PATH_MAX] = {}, *ns;
 729    uint16_t vendor_id, device_id;
 730    int r;
 731    Error *local_err = NULL;
 732
 733    snprintf(dir, sizeof(dir), "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/",
 734            dev->host.domain, dev->host.bus, dev->host.slot,
 735            dev->host.function);
 736
 737    snprintf(name, sizeof(name), "%sdriver", dir);
 738
 739    r = readlink(name, driver, sizeof(driver));
 740    if ((r <= 0) || r >= sizeof(driver)) {
 741        goto fail;
 742    }
 743
 744    driver[r] = 0;
 745    ns = strrchr(driver, '/');
 746    if (!ns) {
 747        goto fail;
 748    }
 749
 750    ns++;
 751
 752    if ((get_real_vendor_id(dir, &vendor_id, &local_err), local_err) ||
 753        (get_real_device_id(dir, &device_id, &local_err), local_err)) {
 754        /* We're already analyzing an assignment error, so we suppress this
 755         * one just like the others above.
 756         */
 757        error_free(local_err);
 758        goto fail;
 759    }
 760
 761    return g_strdup_printf(
 762        "*** The driver '%s' is occupying your device %04x:%02x:%02x.%x.\n"
 763        "***\n"
 764        "*** You can try the following commands to free it:\n"
 765        "***\n"
 766        "*** $ echo \"%04x %04x\" > /sys/bus/pci/drivers/pci-stub/new_id\n"
 767        "*** $ echo \"%04x:%02x:%02x.%x\" > /sys/bus/pci/drivers/%s/unbind\n"
 768        "*** $ echo \"%04x:%02x:%02x.%x\" > /sys/bus/pci/drivers/"
 769        "pci-stub/bind\n"
 770        "*** $ echo \"%04x %04x\" > /sys/bus/pci/drivers/pci-stub/remove_id\n"
 771        "***\n",
 772        ns, dev->host.domain, dev->host.bus, dev->host.slot,
 773        dev->host.function, vendor_id, device_id,
 774        dev->host.domain, dev->host.bus, dev->host.slot, dev->host.function,
 775        ns, dev->host.domain, dev->host.bus, dev->host.slot,
 776        dev->host.function, vendor_id, device_id);
 777
 778fail:
 779    return g_strdup("Couldn't find out why.\n");
 780}
 781
 782static void assign_device(AssignedDevice *dev, Error **errp)
 783{
 784    uint32_t flags = KVM_DEV_ASSIGN_ENABLE_IOMMU;
 785    int r;
 786
 787    /* Only pass non-zero PCI segment to capable module */
 788    if (!kvm_check_extension(kvm_state, KVM_CAP_PCI_SEGMENT) &&
 789        dev->host.domain) {
 790        error_setg(errp, "Can't assign device inside non-zero PCI segment "
 791                   "as this KVM module doesn't support it.");
 792        return;
 793    }
 794
 795    if (!kvm_check_extension(kvm_state, KVM_CAP_IOMMU)) {
 796        error_setg(errp, "No IOMMU found.  Unable to assign device \"%s\"",
 797                   dev->dev.qdev.id);
 798        return;
 799    }
 800
 801    if (dev->features & ASSIGNED_DEVICE_SHARE_INTX_MASK &&
 802        kvm_has_intx_set_mask()) {
 803        flags |= KVM_DEV_ASSIGN_PCI_2_3;
 804    }
 805
 806    r = kvm_device_pci_assign(kvm_state, &dev->host, flags, &dev->dev_id);
 807    if (r < 0) {
 808        switch (r) {
 809        case -EBUSY: {
 810            char *cause;
 811
 812            cause = assign_failed_examine(dev);
 813            error_setg_errno(errp, -r, "Failed to assign device \"%s\"",
 814                             dev->dev.qdev.id);
 815            error_append_hint(errp, "%s", cause);
 816            g_free(cause);
 817            break;
 818        }
 819        default:
 820            error_setg_errno(errp, -r, "Failed to assign device \"%s\"",
 821                             dev->dev.qdev.id);
 822            break;
 823        }
 824    }
 825}
 826
 827static void verify_irqchip_in_kernel(Error **errp)
 828{
 829    if (kvm_irqchip_in_kernel()) {
 830        return;
 831    }
 832    error_setg(errp, "pci-assign requires KVM with in-kernel irqchip enabled");
 833}
 834
 835static int assign_intx(AssignedDevice *dev, Error **errp)
 836{
 837    AssignedIRQType new_type;
 838    PCIINTxRoute intx_route;
 839    bool intx_host_msi;
 840    int r;
 841    Error *local_err = NULL;
 842
 843    /* Interrupt PIN 0 means don't use INTx */
 844    if (assigned_dev_pci_read_byte(&dev->dev, PCI_INTERRUPT_PIN) == 0) {
 845        pci_device_set_intx_routing_notifier(&dev->dev, NULL);
 846        return 0;
 847    }
 848
 849    verify_irqchip_in_kernel(&local_err);
 850    if (local_err) {
 851        error_propagate(errp, local_err);
 852        return -ENOTSUP;
 853    }
 854
 855    pci_device_set_intx_routing_notifier(&dev->dev,
 856                                         assigned_dev_update_irq_routing);
 857
 858    intx_route = pci_device_route_intx_to_irq(&dev->dev, dev->intpin);
 859    assert(intx_route.mode != PCI_INTX_INVERTED);
 860
 861    if (!pci_intx_route_changed(&dev->intx_route, &intx_route)) {
 862        return 0;
 863    }
 864
 865    switch (dev->assigned_irq_type) {
 866    case ASSIGNED_IRQ_INTX_HOST_INTX:
 867    case ASSIGNED_IRQ_INTX_HOST_MSI:
 868        intx_host_msi = dev->assigned_irq_type == ASSIGNED_IRQ_INTX_HOST_MSI;
 869        r = kvm_device_intx_deassign(kvm_state, dev->dev_id, intx_host_msi);
 870        break;
 871    case ASSIGNED_IRQ_MSI:
 872        r = kvm_device_msi_deassign(kvm_state, dev->dev_id);
 873        break;
 874    case ASSIGNED_IRQ_MSIX:
 875        r = kvm_device_msix_deassign(kvm_state, dev->dev_id);
 876        break;
 877    default:
 878        r = 0;
 879        break;
 880    }
 881    if (r) {
 882        perror("assign_intx: deassignment of previous interrupt failed");
 883    }
 884    dev->assigned_irq_type = ASSIGNED_IRQ_NONE;
 885
 886    if (intx_route.mode == PCI_INTX_DISABLED) {
 887        dev->intx_route = intx_route;
 888        return 0;
 889    }
 890
 891retry:
 892    if (dev->features & ASSIGNED_DEVICE_PREFER_MSI_MASK &&
 893        dev->cap.available & ASSIGNED_DEVICE_CAP_MSI) {
 894        intx_host_msi = true;
 895        new_type = ASSIGNED_IRQ_INTX_HOST_MSI;
 896    } else {
 897        intx_host_msi = false;
 898        new_type = ASSIGNED_IRQ_INTX_HOST_INTX;
 899    }
 900
 901    r = kvm_device_intx_assign(kvm_state, dev->dev_id, intx_host_msi,
 902                               intx_route.irq);
 903    if (r < 0) {
 904        if (r == -EIO && !(dev->features & ASSIGNED_DEVICE_PREFER_MSI_MASK) &&
 905            dev->cap.available & ASSIGNED_DEVICE_CAP_MSI) {
 906            /* Retry with host-side MSI. There might be an IRQ conflict and
 907             * either the kernel or the device doesn't support sharing. */
 908            error_report("Host-side INTx sharing not supported, "
 909                         "using MSI instead");
 910            error_printf("Some devices do not work properly in this mode.\n");
 911            dev->features |= ASSIGNED_DEVICE_PREFER_MSI_MASK;
 912            goto retry;
 913        }
 914        error_setg_errno(errp, -r, "Failed to assign irq for \"%s\"",
 915                         dev->dev.qdev.id);
 916        error_append_hint(errp, "Perhaps you are assigning a device "
 917                          "that shares an IRQ with another device?\n");
 918        return r;
 919    }
 920
 921    dev->intx_route = intx_route;
 922    dev->assigned_irq_type = new_type;
 923    return r;
 924}
 925
 926static void deassign_device(AssignedDevice *dev)
 927{
 928    int r;
 929
 930    r = kvm_device_pci_deassign(kvm_state, dev->dev_id);
 931    assert(r == 0);
 932}
 933
 934/* The pci config space got updated. Check if irq numbers have changed
 935 * for our devices
 936 */
 937static void assigned_dev_update_irq_routing(PCIDevice *dev)
 938{
 939    AssignedDevice *assigned_dev = PCI_ASSIGN(dev);
 940    Error *err = NULL;
 941    int r;
 942
 943    r = assign_intx(assigned_dev, &err);
 944    if (r < 0) {
 945        error_report_err(err);
 946        err = NULL;
 947        qdev_unplug(&dev->qdev, &err);
 948        assert(!err);
 949    }
 950}
 951
 952static void assigned_dev_update_msi(PCIDevice *pci_dev)
 953{
 954    AssignedDevice *assigned_dev = PCI_ASSIGN(pci_dev);
 955    uint8_t ctrl_byte = pci_get_byte(pci_dev->config + pci_dev->msi_cap +
 956                                     PCI_MSI_FLAGS);
 957    int r;
 958
 959    /* Some guests gratuitously disable MSI even if they're not using it,
 960     * try to catch this by only deassigning irqs if the guest is using
 961     * MSI or intends to start. */
 962    if (assigned_dev->assigned_irq_type == ASSIGNED_IRQ_MSI ||
 963        (ctrl_byte & PCI_MSI_FLAGS_ENABLE)) {
 964        r = kvm_device_msi_deassign(kvm_state, assigned_dev->dev_id);
 965        /* -ENXIO means no assigned irq */
 966        if (r && r != -ENXIO) {
 967            perror("assigned_dev_update_msi: deassign irq");
 968        }
 969
 970        free_msi_virqs(assigned_dev);
 971
 972        assigned_dev->assigned_irq_type = ASSIGNED_IRQ_NONE;
 973        pci_device_set_intx_routing_notifier(pci_dev, NULL);
 974    }
 975
 976    if (ctrl_byte & PCI_MSI_FLAGS_ENABLE) {
 977        MSIMessage msg = msi_get_message(pci_dev, 0);
 978        int virq;
 979
 980        virq = kvm_irqchip_add_msi_route(kvm_state, msg, pci_dev);
 981        if (virq < 0) {
 982            perror("assigned_dev_update_msi: kvm_irqchip_add_msi_route");
 983            return;
 984        }
 985
 986        assigned_dev->msi_virq = g_malloc(sizeof(*assigned_dev->msi_virq));
 987        assigned_dev->msi_virq_nr = 1;
 988        assigned_dev->msi_virq[0] = virq;
 989        if (kvm_device_msi_assign(kvm_state, assigned_dev->dev_id, virq) < 0) {
 990            perror("assigned_dev_update_msi: kvm_device_msi_assign");
 991        }
 992
 993        assigned_dev->intx_route.mode = PCI_INTX_DISABLED;
 994        assigned_dev->intx_route.irq = -1;
 995        assigned_dev->assigned_irq_type = ASSIGNED_IRQ_MSI;
 996    } else {
 997        Error *local_err = NULL;
 998
 999        assign_intx(assigned_dev, &local_err);
1000        if (local_err) {
1001            error_report_err(local_err);
1002        }
1003    }
1004}
1005
1006static void assigned_dev_update_msi_msg(PCIDevice *pci_dev)
1007{
1008    AssignedDevice *assigned_dev = PCI_ASSIGN(pci_dev);
1009    uint8_t ctrl_byte = pci_get_byte(pci_dev->config + pci_dev->msi_cap +
1010                                     PCI_MSI_FLAGS);
1011
1012    if (assigned_dev->assigned_irq_type != ASSIGNED_IRQ_MSI ||
1013        !(ctrl_byte & PCI_MSI_FLAGS_ENABLE)) {
1014        return;
1015    }
1016
1017    kvm_irqchip_update_msi_route(kvm_state, assigned_dev->msi_virq[0],
1018                                 msi_get_message(pci_dev, 0), pci_dev);
1019}
1020
1021static bool assigned_dev_msix_masked(MSIXTableEntry *entry)
1022{
1023    return (entry->ctrl & cpu_to_le32(0x1)) != 0;
1024}
1025
1026/*
1027 * When MSI-X is first enabled the vector table typically has all the
1028 * vectors masked, so we can't use that as the obvious test to figure out
1029 * how many vectors to initially enable.  Instead we look at the data field
1030 * because this is what worked for pci-assign for a long time.  This makes
1031 * sure the physical MSI-X state tracks the guest's view, which is important
1032 * for some VF/PF and PF/fw communication channels.
1033 */
1034static bool assigned_dev_msix_skipped(MSIXTableEntry *entry)
1035{
1036    return !entry->data;
1037}
1038
1039static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev)
1040{
1041    AssignedDevice *adev = PCI_ASSIGN(pci_dev);
1042    uint16_t entries_nr = 0;
1043    int i, r = 0;
1044    MSIXTableEntry *entry = adev->msix_table;
1045    MSIMessage msg;
1046
1047    /* Get the usable entry number for allocating */
1048    for (i = 0; i < adev->msix_max; i++, entry++) {
1049        if (assigned_dev_msix_skipped(entry)) {
1050            continue;
1051        }
1052        entries_nr++;
1053    }
1054
1055    DEBUG("MSI-X entries: %d\n", entries_nr);
1056
1057    /* It's valid to enable MSI-X with all entries masked */
1058    if (!entries_nr) {
1059        return 0;
1060    }
1061
1062    r = kvm_device_msix_init_vectors(kvm_state, adev->dev_id, entries_nr);
1063    if (r != 0) {
1064        error_report("fail to set MSI-X entry number for MSIX! %s",
1065                     strerror(-r));
1066        return r;
1067    }
1068
1069    free_msi_virqs(adev);
1070
1071    adev->msi_virq_nr = adev->msix_max;
1072    adev->msi_virq = g_malloc(adev->msix_max * sizeof(*adev->msi_virq));
1073
1074    entry = adev->msix_table;
1075    for (i = 0; i < adev->msix_max; i++, entry++) {
1076        adev->msi_virq[i] = -1;
1077
1078        if (assigned_dev_msix_skipped(entry)) {
1079            continue;
1080        }
1081
1082        msg.address = entry->addr_lo | ((uint64_t)entry->addr_hi << 32);
1083        msg.data = entry->data;
1084        r = kvm_irqchip_add_msi_route(kvm_state, msg, pci_dev);
1085        if (r < 0) {
1086            return r;
1087        }
1088        adev->msi_virq[i] = r;
1089
1090        DEBUG("MSI-X vector %d, gsi %d, addr %08x_%08x, data %08x\n", i,
1091              r, entry->addr_hi, entry->addr_lo, entry->data);
1092
1093        r = kvm_device_msix_set_vector(kvm_state, adev->dev_id, i,
1094                                       adev->msi_virq[i]);
1095        if (r) {
1096            error_report("fail to set MSI-X entry! %s", strerror(-r));
1097            break;
1098        }
1099    }
1100
1101    return r;
1102}
1103
1104static void assigned_dev_update_msix(PCIDevice *pci_dev)
1105{
1106    AssignedDevice *assigned_dev = PCI_ASSIGN(pci_dev);
1107    uint16_t ctrl_word = pci_get_word(pci_dev->config + pci_dev->msix_cap +
1108                                      PCI_MSIX_FLAGS);
1109    int r;
1110
1111    /* Some guests gratuitously disable MSIX even if they're not using it,
1112     * try to catch this by only deassigning irqs if the guest is using
1113     * MSIX or intends to start. */
1114    if ((assigned_dev->assigned_irq_type == ASSIGNED_IRQ_MSIX) ||
1115        (ctrl_word & PCI_MSIX_FLAGS_ENABLE)) {
1116        r = kvm_device_msix_deassign(kvm_state, assigned_dev->dev_id);
1117        /* -ENXIO means no assigned irq */
1118        if (r && r != -ENXIO) {
1119            perror("assigned_dev_update_msix: deassign irq");
1120        }
1121
1122        free_msi_virqs(assigned_dev);
1123
1124        assigned_dev->assigned_irq_type = ASSIGNED_IRQ_NONE;
1125        pci_device_set_intx_routing_notifier(pci_dev, NULL);
1126    }
1127
1128    if (ctrl_word & PCI_MSIX_FLAGS_ENABLE) {
1129        if (assigned_dev_update_msix_mmio(pci_dev) < 0) {
1130            perror("assigned_dev_update_msix_mmio");
1131            return;
1132        }
1133
1134        if (assigned_dev->msi_virq_nr > 0) {
1135            if (kvm_device_msix_assign(kvm_state, assigned_dev->dev_id) < 0) {
1136                perror("assigned_dev_enable_msix: assign irq");
1137                return;
1138            }
1139        }
1140        assigned_dev->intx_route.mode = PCI_INTX_DISABLED;
1141        assigned_dev->intx_route.irq = -1;
1142        assigned_dev->assigned_irq_type = ASSIGNED_IRQ_MSIX;
1143    } else {
1144        Error *local_err = NULL;
1145
1146        assign_intx(assigned_dev, &local_err);
1147        if (local_err) {
1148            error_report_err(local_err);
1149        }
1150    }
1151}
1152
1153static uint32_t assigned_dev_pci_read_config(PCIDevice *pci_dev,
1154                                             uint32_t address, int len)
1155{
1156    AssignedDevice *assigned_dev = PCI_ASSIGN(pci_dev);
1157    uint32_t virt_val = pci_default_read_config(pci_dev, address, len);
1158    uint32_t real_val, emulate_mask, full_emulation_mask;
1159
1160    emulate_mask = 0;
1161    memcpy(&emulate_mask, assigned_dev->emulate_config_read + address, len);
1162    emulate_mask = le32_to_cpu(emulate_mask);
1163
1164    full_emulation_mask = 0xffffffff >> (32 - len * 8);
1165
1166    if (emulate_mask != full_emulation_mask) {
1167        real_val = assigned_dev_pci_read(pci_dev, address, len);
1168        return (virt_val & emulate_mask) | (real_val & ~emulate_mask);
1169    } else {
1170        return virt_val;
1171    }
1172}
1173
1174static void assigned_dev_pci_write_config(PCIDevice *pci_dev, uint32_t address,
1175                                          uint32_t val, int len)
1176{
1177    AssignedDevice *assigned_dev = PCI_ASSIGN(pci_dev);
1178    uint16_t old_cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1179    uint32_t emulate_mask, full_emulation_mask;
1180    int ret;
1181
1182    pci_default_write_config(pci_dev, address, val, len);
1183
1184    if (kvm_has_intx_set_mask() &&
1185        range_covers_byte(address, len, PCI_COMMAND + 1)) {
1186        bool intx_masked = (pci_get_word(pci_dev->config + PCI_COMMAND) &
1187                            PCI_COMMAND_INTX_DISABLE);
1188
1189        if (intx_masked != !!(old_cmd & PCI_COMMAND_INTX_DISABLE)) {
1190            ret = kvm_device_intx_set_mask(kvm_state, assigned_dev->dev_id,
1191                                           intx_masked);
1192            if (ret) {
1193                perror("assigned_dev_pci_write_config: set intx mask");
1194            }
1195        }
1196    }
1197    if (assigned_dev->cap.available & ASSIGNED_DEVICE_CAP_MSI) {
1198        if (range_covers_byte(address, len,
1199                              pci_dev->msi_cap + PCI_MSI_FLAGS)) {
1200            assigned_dev_update_msi(pci_dev);
1201        } else if (ranges_overlap(address, len, /* 32bit MSI only */
1202                                  pci_dev->msi_cap + PCI_MSI_ADDRESS_LO, 6)) {
1203            assigned_dev_update_msi_msg(pci_dev);
1204        }
1205    }
1206    if (assigned_dev->cap.available & ASSIGNED_DEVICE_CAP_MSIX) {
1207        if (range_covers_byte(address, len,
1208                              pci_dev->msix_cap + PCI_MSIX_FLAGS + 1)) {
1209            assigned_dev_update_msix(pci_dev);
1210        }
1211    }
1212
1213    emulate_mask = 0;
1214    memcpy(&emulate_mask, assigned_dev->emulate_config_write + address, len);
1215    emulate_mask = le32_to_cpu(emulate_mask);
1216
1217    full_emulation_mask = 0xffffffff >> (32 - len * 8);
1218
1219    if (emulate_mask != full_emulation_mask) {
1220        if (emulate_mask) {
1221            val &= ~emulate_mask;
1222            val |= assigned_dev_pci_read(pci_dev, address, len) & emulate_mask;
1223        }
1224        assigned_dev_pci_write(pci_dev, address, val, len);
1225    }
1226}
1227
1228static void assigned_dev_setup_cap_read(AssignedDevice *dev, uint32_t offset,
1229                                        uint32_t len)
1230{
1231    assigned_dev_direct_config_read(dev, offset, len);
1232    assigned_dev_emulate_config_read(dev, offset + PCI_CAP_LIST_NEXT, 1);
1233}
1234
1235static int assigned_device_pci_cap_init(PCIDevice *pci_dev, Error **errp)
1236{
1237    AssignedDevice *dev = PCI_ASSIGN(pci_dev);
1238    PCIRegion *pci_region = dev->real_device.regions;
1239    int ret, pos;
1240    Error *local_err = NULL;
1241
1242    /* Clear initial capabilities pointer and status copied from hw */
1243    pci_set_byte(pci_dev->config + PCI_CAPABILITY_LIST, 0);
1244    pci_set_word(pci_dev->config + PCI_STATUS,
1245                 pci_get_word(pci_dev->config + PCI_STATUS) &
1246                 ~PCI_STATUS_CAP_LIST);
1247
1248    /* Expose MSI capability
1249     * MSI capability is the 1st capability in capability config */
1250    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSI, 0);
1251    if (pos != 0 && kvm_check_extension(kvm_state, KVM_CAP_ASSIGN_DEV_IRQ)) {
1252        verify_irqchip_in_kernel(&local_err);
1253        if (local_err) {
1254            error_propagate(errp, local_err);
1255            return -ENOTSUP;
1256        }
1257        dev->cap.available |= ASSIGNED_DEVICE_CAP_MSI;
1258        /* Only 32-bit/no-mask currently supported */
1259        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_MSI, pos, 10,
1260                                  &local_err);
1261        if (ret < 0) {
1262            error_propagate(errp, local_err);
1263            return ret;
1264        }
1265        pci_dev->msi_cap = pos;
1266
1267        pci_set_word(pci_dev->config + pos + PCI_MSI_FLAGS,
1268                     pci_get_word(pci_dev->config + pos + PCI_MSI_FLAGS) &
1269                     PCI_MSI_FLAGS_QMASK);
1270        pci_set_long(pci_dev->config + pos + PCI_MSI_ADDRESS_LO, 0);
1271        pci_set_word(pci_dev->config + pos + PCI_MSI_DATA_32, 0);
1272
1273        /* Set writable fields */
1274        pci_set_word(pci_dev->wmask + pos + PCI_MSI_FLAGS,
1275                     PCI_MSI_FLAGS_QSIZE | PCI_MSI_FLAGS_ENABLE);
1276        pci_set_long(pci_dev->wmask + pos + PCI_MSI_ADDRESS_LO, 0xfffffffc);
1277        pci_set_word(pci_dev->wmask + pos + PCI_MSI_DATA_32, 0xffff);
1278    }
1279    /* Expose MSI-X capability */
1280    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_MSIX, 0);
1281    if (pos != 0 && kvm_device_msix_supported(kvm_state)) {
1282        int bar_nr;
1283        uint32_t msix_table_entry;
1284        uint16_t msix_max;
1285
1286        verify_irqchip_in_kernel(&local_err);
1287        if (local_err) {
1288            error_propagate(errp, local_err);
1289            return -ENOTSUP;
1290        }
1291        dev->cap.available |= ASSIGNED_DEVICE_CAP_MSIX;
1292        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_MSIX, pos, 12,
1293                                  &local_err);
1294        if (ret < 0) {
1295            error_propagate(errp, local_err);
1296            return ret;
1297        }
1298        pci_dev->msix_cap = pos;
1299
1300        msix_max = (pci_get_word(pci_dev->config + pos + PCI_MSIX_FLAGS) &
1301                    PCI_MSIX_FLAGS_QSIZE) + 1;
1302        msix_max = MIN(msix_max, KVM_MAX_MSIX_PER_DEV);
1303        pci_set_word(pci_dev->config + pos + PCI_MSIX_FLAGS, msix_max - 1);
1304
1305        /* Only enable and function mask bits are writable */
1306        pci_set_word(pci_dev->wmask + pos + PCI_MSIX_FLAGS,
1307                     PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
1308
1309        msix_table_entry = pci_get_long(pci_dev->config + pos + PCI_MSIX_TABLE);
1310        bar_nr = msix_table_entry & PCI_MSIX_FLAGS_BIRMASK;
1311        msix_table_entry &= ~PCI_MSIX_FLAGS_BIRMASK;
1312        dev->msix_table_addr = pci_region[bar_nr].base_addr + msix_table_entry;
1313        dev->msix_max = msix_max;
1314    }
1315
1316    /* Minimal PM support, nothing writable, device appears to NAK changes */
1317    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_PM, 0);
1318    if (pos) {
1319        uint16_t pmc;
1320
1321        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_PM, pos, PCI_PM_SIZEOF,
1322                                  &local_err);
1323        if (ret < 0) {
1324            error_propagate(errp, local_err);
1325            return ret;
1326        }
1327
1328        assigned_dev_setup_cap_read(dev, pos, PCI_PM_SIZEOF);
1329
1330        pmc = pci_get_word(pci_dev->config + pos + PCI_CAP_FLAGS);
1331        pmc &= (PCI_PM_CAP_VER_MASK | PCI_PM_CAP_DSI);
1332        pci_set_word(pci_dev->config + pos + PCI_CAP_FLAGS, pmc);
1333
1334        /* assign_device will bring the device up to D0, so we don't need
1335         * to worry about doing that ourselves here. */
1336        pci_set_word(pci_dev->config + pos + PCI_PM_CTRL,
1337                     PCI_PM_CTRL_NO_SOFT_RESET);
1338
1339        pci_set_byte(pci_dev->config + pos + PCI_PM_PPB_EXTENSIONS, 0);
1340        pci_set_byte(pci_dev->config + pos + PCI_PM_DATA_REGISTER, 0);
1341    }
1342
1343    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_EXP, 0);
1344    if (pos) {
1345        uint8_t version, size = 0;
1346        uint16_t type, devctl, lnksta;
1347        uint32_t devcap, lnkcap;
1348
1349        version = pci_get_byte(pci_dev->config + pos + PCI_EXP_FLAGS);
1350        version &= PCI_EXP_FLAGS_VERS;
1351        if (version == 1) {
1352            size = 0x14;
1353        } else if (version == 2) {
1354            /*
1355             * Check for non-std size, accept reduced size to 0x34,
1356             * which is what bcm5761 implemented, violating the
1357             * PCIe v3.0 spec that regs should exist and be read as 0,
1358             * not optionally provided and shorten the struct size.
1359             */
1360            size = MIN(0x3c, PCI_CONFIG_SPACE_SIZE - pos);
1361            if (size < 0x34) {
1362                error_setg(errp, "Invalid size PCIe cap-id 0x%x",
1363                           PCI_CAP_ID_EXP);
1364                return -EINVAL;
1365            } else if (size != 0x3c) {
1366                error_report("WARNING, %s: PCIe cap-id 0x%x has "
1367                             "non-standard size 0x%x; std size should be 0x3c",
1368                             __func__, PCI_CAP_ID_EXP, size);
1369            }
1370        } else if (version == 0) {
1371            uint16_t vid, did;
1372            vid = pci_get_word(pci_dev->config + PCI_VENDOR_ID);
1373            did = pci_get_word(pci_dev->config + PCI_DEVICE_ID);
1374            if (vid == PCI_VENDOR_ID_INTEL && did == 0x10ed) {
1375                /*
1376                 * quirk for Intel 82599 VF with invalid PCIe capability
1377                 * version, should really be version 2 (same as PF)
1378                 */
1379                size = 0x3c;
1380            }
1381        }
1382
1383        if (size == 0) {
1384            error_setg(errp, "Unsupported PCI express capability version %d",
1385                       version);
1386            return -EINVAL;
1387        }
1388
1389        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_EXP, pos, size,
1390                                  &local_err);
1391        if (ret < 0) {
1392            error_propagate(errp, local_err);
1393            return ret;
1394        }
1395
1396        assigned_dev_setup_cap_read(dev, pos, size);
1397
1398        type = pci_get_word(pci_dev->config + pos + PCI_EXP_FLAGS);
1399        type = (type & PCI_EXP_FLAGS_TYPE) >> 4;
1400        if (type != PCI_EXP_TYPE_ENDPOINT &&
1401            type != PCI_EXP_TYPE_LEG_END && type != PCI_EXP_TYPE_RC_END) {
1402            error_setg(errp, "Device assignment only supports endpoint "
1403                       "assignment, device type %d", type);
1404            return -EINVAL;
1405        }
1406
1407        /* capabilities, pass existing read-only copy
1408         * PCI_EXP_FLAGS_IRQ: updated by hardware, should be direct read */
1409
1410        /* device capabilities: hide FLR */
1411        devcap = pci_get_long(pci_dev->config + pos + PCI_EXP_DEVCAP);
1412        devcap &= ~PCI_EXP_DEVCAP_FLR;
1413        pci_set_long(pci_dev->config + pos + PCI_EXP_DEVCAP, devcap);
1414
1415        /* device control: clear all error reporting enable bits, leaving
1416         *                 only a few host values.  Note, these are
1417         *                 all writable, but not passed to hw.
1418         */
1419        devctl = pci_get_word(pci_dev->config + pos + PCI_EXP_DEVCTL);
1420        devctl = (devctl & (PCI_EXP_DEVCTL_READRQ | PCI_EXP_DEVCTL_PAYLOAD)) |
1421                  PCI_EXP_DEVCTL_RELAX_EN | PCI_EXP_DEVCTL_NOSNOOP_EN;
1422        pci_set_word(pci_dev->config + pos + PCI_EXP_DEVCTL, devctl);
1423        devctl = PCI_EXP_DEVCTL_BCR_FLR | PCI_EXP_DEVCTL_AUX_PME;
1424        pci_set_word(pci_dev->wmask + pos + PCI_EXP_DEVCTL, ~devctl);
1425
1426        /* Clear device status */
1427        pci_set_word(pci_dev->config + pos + PCI_EXP_DEVSTA, 0);
1428
1429        /* Link capabilities, expose links and latencues, clear reporting */
1430        lnkcap = pci_get_long(pci_dev->config + pos + PCI_EXP_LNKCAP);
1431        lnkcap &= (PCI_EXP_LNKCAP_SLS | PCI_EXP_LNKCAP_MLW |
1432                   PCI_EXP_LNKCAP_ASPMS | PCI_EXP_LNKCAP_L0SEL |
1433                   PCI_EXP_LNKCAP_L1EL);
1434        pci_set_long(pci_dev->config + pos + PCI_EXP_LNKCAP, lnkcap);
1435
1436        /* Link control, pass existing read-only copy.  Should be writable? */
1437
1438        /* Link status, only expose current speed and width */
1439        lnksta = pci_get_word(pci_dev->config + pos + PCI_EXP_LNKSTA);
1440        lnksta &= (PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
1441        pci_set_word(pci_dev->config + pos + PCI_EXP_LNKSTA, lnksta);
1442
1443        if (version >= 2) {
1444            /* Slot capabilities, control, status - not needed for endpoints */
1445            pci_set_long(pci_dev->config + pos + PCI_EXP_SLTCAP, 0);
1446            pci_set_word(pci_dev->config + pos + PCI_EXP_SLTCTL, 0);
1447            pci_set_word(pci_dev->config + pos + PCI_EXP_SLTSTA, 0);
1448
1449            /* Root control, capabilities, status - not needed for endpoints */
1450            pci_set_word(pci_dev->config + pos + PCI_EXP_RTCTL, 0);
1451            pci_set_word(pci_dev->config + pos + PCI_EXP_RTCAP, 0);
1452            pci_set_long(pci_dev->config + pos + PCI_EXP_RTSTA, 0);
1453
1454            /* Device capabilities/control 2, pass existing read-only copy */
1455            /* Link control 2, pass existing read-only copy */
1456        }
1457    }
1458
1459    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_PCIX, 0);
1460    if (pos) {
1461        uint16_t cmd;
1462        uint32_t status;
1463
1464        /* Only expose the minimum, 8 byte capability */
1465        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_PCIX, pos, 8,
1466                                  &local_err);
1467        if (ret < 0) {
1468            error_propagate(errp, local_err);
1469            return ret;
1470        }
1471
1472        assigned_dev_setup_cap_read(dev, pos, 8);
1473
1474        /* Command register, clear upper bits, including extended modes */
1475        cmd = pci_get_word(pci_dev->config + pos + PCI_X_CMD);
1476        cmd &= (PCI_X_CMD_DPERR_E | PCI_X_CMD_ERO | PCI_X_CMD_MAX_READ |
1477                PCI_X_CMD_MAX_SPLIT);
1478        pci_set_word(pci_dev->config + pos + PCI_X_CMD, cmd);
1479
1480        /* Status register, update with emulated PCI bus location, clear
1481         * error bits, leave the rest. */
1482        status = pci_get_long(pci_dev->config + pos + PCI_X_STATUS);
1483        status &= ~(PCI_X_STATUS_BUS | PCI_X_STATUS_DEVFN);
1484        status |= pci_requester_id(pci_dev);
1485        status &= ~(PCI_X_STATUS_SPL_DISC | PCI_X_STATUS_UNX_SPL |
1486                    PCI_X_STATUS_SPL_ERR);
1487        pci_set_long(pci_dev->config + pos + PCI_X_STATUS, status);
1488    }
1489
1490    pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_VPD, 0);
1491    if (pos) {
1492        /* Direct R/W passthrough */
1493        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_VPD, pos, 8,
1494                                  &local_err);
1495        if (ret < 0) {
1496            error_propagate(errp, local_err);
1497            return ret;
1498        }
1499
1500        assigned_dev_setup_cap_read(dev, pos, 8);
1501
1502        /* direct write for cap content */
1503        assigned_dev_direct_config_write(dev, pos + 2, 6);
1504    }
1505
1506    /* Devices can have multiple vendor capabilities, get them all */
1507    for (pos = 0; (pos = pci_find_cap_offset(pci_dev, PCI_CAP_ID_VNDR, pos));
1508        pos += PCI_CAP_LIST_NEXT) {
1509        uint8_t len = pci_get_byte(pci_dev->config + pos + PCI_CAP_FLAGS);
1510        /* Direct R/W passthrough */
1511        ret = pci_add_capability2(pci_dev, PCI_CAP_ID_VNDR, pos, len,
1512                                  &local_err);
1513        if (ret < 0) {
1514            error_propagate(errp, local_err);
1515            return ret;
1516        }
1517
1518        assigned_dev_setup_cap_read(dev, pos, len);
1519
1520        /* direct write for cap content */
1521        assigned_dev_direct_config_write(dev, pos + 2, len - 2);
1522    }
1523
1524    /* If real and virtual capability list status bits differ, virtualize the
1525     * access. */
1526    if ((pci_get_word(pci_dev->config + PCI_STATUS) & PCI_STATUS_CAP_LIST) !=
1527        (assigned_dev_pci_read_byte(pci_dev, PCI_STATUS) &
1528         PCI_STATUS_CAP_LIST)) {
1529        dev->emulate_config_read[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
1530    }
1531
1532    return 0;
1533}
1534
1535static uint64_t
1536assigned_dev_msix_mmio_read(void *opaque, hwaddr addr,
1537                            unsigned size)
1538{
1539    AssignedDevice *adev = opaque;
1540    uint64_t val;
1541
1542    memcpy(&val, (void *)((uint8_t *)adev->msix_table + addr), size);
1543
1544    return val;
1545}
1546
1547static void assigned_dev_msix_mmio_write(void *opaque, hwaddr addr,
1548                                         uint64_t val, unsigned size)
1549{
1550    AssignedDevice *adev = opaque;
1551    PCIDevice *pdev = &adev->dev;
1552    uint16_t ctrl;
1553    MSIXTableEntry orig;
1554    int i = addr >> 4;
1555
1556    if (i >= adev->msix_max) {
1557        return; /* Drop write */
1558    }
1559
1560    ctrl = pci_get_word(pdev->config + pdev->msix_cap + PCI_MSIX_FLAGS);
1561
1562    DEBUG("write to MSI-X table offset 0x%lx, val 0x%lx\n", addr, val);
1563
1564    if (ctrl & PCI_MSIX_FLAGS_ENABLE) {
1565        orig = adev->msix_table[i];
1566    }
1567
1568    memcpy((uint8_t *)adev->msix_table + addr, &val, size);
1569
1570    if (ctrl & PCI_MSIX_FLAGS_ENABLE) {
1571        MSIXTableEntry *entry = &adev->msix_table[i];
1572
1573        if (!assigned_dev_msix_masked(&orig) &&
1574            assigned_dev_msix_masked(entry)) {
1575            /*
1576             * Vector masked, disable it
1577             *
1578             * XXX It's not clear if we can or should actually attempt
1579             * to mask or disable the interrupt.  KVM doesn't have
1580             * support for pending bits and kvm_assign_set_msix_entry
1581             * doesn't modify the device hardware mask.  Interrupts
1582             * while masked are simply not injected to the guest, so
1583             * are lost.  Can we get away with always injecting an
1584             * interrupt on unmask?
1585             */
1586        } else if (assigned_dev_msix_masked(&orig) &&
1587                   !assigned_dev_msix_masked(entry)) {
1588            /* Vector unmasked */
1589            if (i >= adev->msi_virq_nr || adev->msi_virq[i] < 0) {
1590                /* Previously unassigned vector, start from scratch */
1591                assigned_dev_update_msix(pdev);
1592                return;
1593            } else {
1594                /* Update an existing, previously masked vector */
1595                MSIMessage msg;
1596                int ret;
1597
1598                msg.address = entry->addr_lo |
1599                    ((uint64_t)entry->addr_hi << 32);
1600                msg.data = entry->data;
1601
1602                ret = kvm_irqchip_update_msi_route(kvm_state,
1603                                                   adev->msi_virq[i], msg,
1604                                                   pdev);
1605                if (ret) {
1606                    error_report("Error updating irq routing entry (%d)", ret);
1607                }
1608            }
1609        }
1610    }
1611}
1612
1613static const MemoryRegionOps assigned_dev_msix_mmio_ops = {
1614    .read = assigned_dev_msix_mmio_read,
1615    .write = assigned_dev_msix_mmio_write,
1616    .endianness = DEVICE_NATIVE_ENDIAN,
1617    .valid = {
1618        .min_access_size = 4,
1619        .max_access_size = 8,
1620    },
1621    .impl = {
1622        .min_access_size = 4,
1623        .max_access_size = 8,
1624    },
1625};
1626
1627static void assigned_dev_msix_reset(AssignedDevice *dev)
1628{
1629    MSIXTableEntry *entry;
1630    int i;
1631
1632    if (!dev->msix_table) {
1633        return;
1634    }
1635
1636    memset(dev->msix_table, 0, MSIX_PAGE_SIZE);
1637
1638    for (i = 0, entry = dev->msix_table; i < dev->msix_max; i++, entry++) {
1639        entry->ctrl = cpu_to_le32(0x1); /* Masked */
1640    }
1641}
1642
1643static void assigned_dev_register_msix_mmio(AssignedDevice *dev, Error **errp)
1644{
1645    dev->msix_table = mmap(NULL, MSIX_PAGE_SIZE, PROT_READ|PROT_WRITE,
1646                           MAP_ANONYMOUS|MAP_PRIVATE, 0, 0);
1647    if (dev->msix_table == MAP_FAILED) {
1648        error_setg_errno(errp, errno, "failed to allocate msix_table");
1649        dev->msix_table = NULL;
1650        return;
1651    }
1652
1653    assigned_dev_msix_reset(dev);
1654
1655    memory_region_init_io(&dev->mmio, OBJECT(dev), &assigned_dev_msix_mmio_ops,
1656                          dev, "assigned-dev-msix", MSIX_PAGE_SIZE);
1657}
1658
1659static void assigned_dev_unregister_msix_mmio(AssignedDevice *dev)
1660{
1661    if (!dev->msix_table) {
1662        return;
1663    }
1664
1665    if (munmap(dev->msix_table, MSIX_PAGE_SIZE) == -1) {
1666        error_report("error unmapping msix_table! %s", strerror(errno));
1667    }
1668    dev->msix_table = NULL;
1669}
1670
1671static const VMStateDescription vmstate_assigned_device = {
1672    .name = "pci-assign",
1673    .unmigratable = 1,
1674};
1675
1676static void reset_assigned_device(DeviceState *dev)
1677{
1678    PCIDevice *pci_dev = PCI_DEVICE(dev);
1679    AssignedDevice *adev = PCI_ASSIGN(pci_dev);
1680    char reset_file[64];
1681    const char reset[] = "1";
1682    int fd, ret;
1683
1684    /*
1685     * If a guest is reset without being shutdown, MSI/MSI-X can still
1686     * be running.  We want to return the device to a known state on
1687     * reset, so disable those here.  We especially do not want MSI-X
1688     * enabled since it lives in MMIO space, which is about to get
1689     * disabled.
1690     */
1691    if (adev->assigned_irq_type == ASSIGNED_IRQ_MSIX) {
1692        uint16_t ctrl = pci_get_word(pci_dev->config +
1693                                     pci_dev->msix_cap + PCI_MSIX_FLAGS);
1694
1695        pci_set_word(pci_dev->config + pci_dev->msix_cap + PCI_MSIX_FLAGS,
1696                     ctrl & ~PCI_MSIX_FLAGS_ENABLE);
1697        assigned_dev_update_msix(pci_dev);
1698    } else if (adev->assigned_irq_type == ASSIGNED_IRQ_MSI) {
1699        uint8_t ctrl = pci_get_byte(pci_dev->config +
1700                                    pci_dev->msi_cap + PCI_MSI_FLAGS);
1701
1702        pci_set_byte(pci_dev->config + pci_dev->msi_cap + PCI_MSI_FLAGS,
1703                     ctrl & ~PCI_MSI_FLAGS_ENABLE);
1704        assigned_dev_update_msi(pci_dev);
1705    }
1706
1707    snprintf(reset_file, sizeof(reset_file),
1708             "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/reset",
1709             adev->host.domain, adev->host.bus, adev->host.slot,
1710             adev->host.function);
1711
1712    /*
1713     * Issue a device reset via pci-sysfs.  Note that we use write(2) here
1714     * and ignore the return value because some kernels have a bug that
1715     * returns 0 rather than bytes written on success, sending us into an
1716     * infinite retry loop using other write mechanisms.
1717     */
1718    fd = open(reset_file, O_WRONLY);
1719    if (fd != -1) {
1720        ret = write(fd, reset, strlen(reset));
1721        (void)ret;
1722        close(fd);
1723    }
1724
1725    /*
1726     * When a 0 is written to the bus master register, the device is logically
1727     * disconnected from the PCI bus. This avoids further DMA transfers.
1728     */
1729    assigned_dev_pci_write_config(pci_dev, PCI_COMMAND, 0, 1);
1730}
1731
1732static void assigned_realize(struct PCIDevice *pci_dev, Error **errp)
1733{
1734    AssignedDevice *dev = PCI_ASSIGN(pci_dev);
1735    uint8_t e_intx;
1736    int r;
1737    Error *local_err = NULL;
1738
1739    if (!kvm_enabled()) {
1740        error_setg(&local_err, "pci-assign requires KVM support");
1741        goto exit_with_error;
1742    }
1743
1744    if (!dev->host.domain && !dev->host.bus && !dev->host.slot &&
1745        !dev->host.function) {
1746        error_setg(&local_err, "no host device specified");
1747        goto exit_with_error;
1748    }
1749
1750    /*
1751     * Set up basic config space access control. Will be further refined during
1752     * device initialization.
1753     */
1754    assigned_dev_emulate_config_read(dev, 0, PCI_CONFIG_SPACE_SIZE);
1755    assigned_dev_direct_config_read(dev, PCI_STATUS, 2);
1756    assigned_dev_direct_config_read(dev, PCI_REVISION_ID, 1);
1757    assigned_dev_direct_config_read(dev, PCI_CLASS_PROG, 3);
1758    assigned_dev_direct_config_read(dev, PCI_CACHE_LINE_SIZE, 1);
1759    assigned_dev_direct_config_read(dev, PCI_LATENCY_TIMER, 1);
1760    assigned_dev_direct_config_read(dev, PCI_BIST, 1);
1761    assigned_dev_direct_config_read(dev, PCI_CARDBUS_CIS, 4);
1762    assigned_dev_direct_config_read(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1763    assigned_dev_direct_config_read(dev, PCI_SUBSYSTEM_ID, 2);
1764    assigned_dev_direct_config_read(dev, PCI_CAPABILITY_LIST + 1, 7);
1765    assigned_dev_direct_config_read(dev, PCI_MIN_GNT, 1);
1766    assigned_dev_direct_config_read(dev, PCI_MAX_LAT, 1);
1767    memcpy(dev->emulate_config_write, dev->emulate_config_read,
1768           sizeof(dev->emulate_config_read));
1769
1770    get_real_device(dev, &local_err);
1771    if (local_err) {
1772        goto out;
1773    }
1774
1775    if (assigned_device_pci_cap_init(pci_dev, &local_err) < 0) {
1776        goto out;
1777    }
1778
1779    /* intercept MSI-X entry page in the MMIO */
1780    if (dev->cap.available & ASSIGNED_DEVICE_CAP_MSIX) {
1781        assigned_dev_register_msix_mmio(dev, &local_err);
1782        if (local_err) {
1783            goto out;
1784        }
1785    }
1786
1787    /* handle real device's MMIO/PIO BARs */
1788    assigned_dev_register_regions(dev->real_device.regions,
1789                                  dev->real_device.region_number, dev,
1790                                  &local_err);
1791    if (local_err) {
1792        goto out;
1793    }
1794
1795    /* handle interrupt routing */
1796    e_intx = dev->dev.config[PCI_INTERRUPT_PIN] - 1;
1797    dev->intpin = e_intx;
1798    dev->intx_route.mode = PCI_INTX_DISABLED;
1799    dev->intx_route.irq = -1;
1800
1801    /* assign device to guest */
1802    assign_device(dev, &local_err);
1803    if (local_err) {
1804        goto out;
1805    }
1806
1807    /* assign legacy INTx to the device */
1808    r = assign_intx(dev, &local_err);
1809    if (r < 0) {
1810        goto assigned_out;
1811    }
1812
1813    assigned_dev_load_option_rom(dev);
1814
1815    return;
1816
1817assigned_out:
1818    deassign_device(dev);
1819
1820out:
1821    free_assigned_device(dev);
1822
1823exit_with_error:
1824    assert(local_err);
1825    error_propagate(errp, local_err);
1826}
1827
1828static void assigned_exitfn(struct PCIDevice *pci_dev)
1829{
1830    AssignedDevice *dev = PCI_ASSIGN(pci_dev);
1831
1832    deassign_device(dev);
1833    free_assigned_device(dev);
1834}
1835
1836static void assigned_dev_instance_init(Object *obj)
1837{
1838    PCIDevice *pci_dev = PCI_DEVICE(obj);
1839    AssignedDevice *d = PCI_ASSIGN(pci_dev);
1840
1841    device_add_bootindex_property(obj, &d->bootindex,
1842                                  "bootindex", NULL,
1843                                  &pci_dev->qdev, NULL);
1844}
1845
1846static Property assigned_dev_properties[] = {
1847    DEFINE_PROP_PCI_HOST_DEVADDR("host", AssignedDevice, host),
1848    DEFINE_PROP_BIT("prefer_msi", AssignedDevice, features,
1849                    ASSIGNED_DEVICE_PREFER_MSI_BIT, false),
1850    DEFINE_PROP_BIT("share_intx", AssignedDevice, features,
1851                    ASSIGNED_DEVICE_SHARE_INTX_BIT, true),
1852    DEFINE_PROP_STRING("configfd", AssignedDevice, configfd_name),
1853    DEFINE_PROP_END_OF_LIST(),
1854};
1855
1856static void assign_class_init(ObjectClass *klass, void *data)
1857{
1858    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
1859    DeviceClass *dc = DEVICE_CLASS(klass);
1860
1861    k->realize      = assigned_realize;
1862    k->exit         = assigned_exitfn;
1863    k->config_read  = assigned_dev_pci_read_config;
1864    k->config_write = assigned_dev_pci_write_config;
1865    dc->props       = assigned_dev_properties;
1866    dc->vmsd        = &vmstate_assigned_device;
1867    dc->reset       = reset_assigned_device;
1868    set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1869    dc->desc        = "KVM-based PCI passthrough";
1870}
1871
1872static const TypeInfo assign_info = {
1873    .name               = TYPE_PCI_ASSIGN,
1874    .parent             = TYPE_PCI_DEVICE,
1875    .instance_size      = sizeof(AssignedDevice),
1876    .class_init         = assign_class_init,
1877    .instance_init      = assigned_dev_instance_init,
1878};
1879
1880static void assign_register_types(void)
1881{
1882    type_register_static(&assign_info);
1883}
1884
1885type_init(assign_register_types)
1886
1887static void assigned_dev_load_option_rom(AssignedDevice *dev)
1888{
1889    int size = 0;
1890
1891    pci_assign_dev_load_option_rom(&dev->dev, OBJECT(dev), &size,
1892                                   dev->host.domain, dev->host.bus,
1893                                   dev->host.slot, dev->host.function);
1894
1895    if (!size) {
1896        error_report("pci-assign: Invalid ROM.");
1897    }
1898}
1899