hw/virtio/virtio-pci.c

   1 /*
   2  * Virtio PCI Bindings
   3  *
   4  * Copyright IBM, Corp. 2007
   5  * Copyright (c) 2009 CodeSourcery
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Paul Brook        <paul@codesourcery.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  * Contributions after 2012-01-13 are licensed under the terms of the
  15  * GNU GPL, version 2 or (at your option) any later version.
  16  */
  17
  18 #include "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
  50 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
  51                                VirtIOPCIProxy *dev);
  52 static void virtio_pci_reset(DeviceState *qdev);
  53
  54 /* virtio device */
  55 /* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */
  56 static 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  */
  64 static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d)
  65 {
  66     return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
  67 }
  68
  69 static 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
  81 static 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
  92 static 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
 107 static bool virtio_pci_modern_state_needed(void *opaque)
 108 {
 109     VirtIOPCIProxy *proxy = opaque;
 110
 111     return virtio_pci_modern(proxy);
 112 }
 113
 114 static 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
 130 static const VMStateDescription vmstate_virtio_pci = {
 131     .name = "virtio_pci",
 132     .version_id = 1,
 133     .minimum_version_id = 1,
 134     .fields = (VMStateField[]) {
 135         VMSTATE_END_OF_LIST()
 136     },
 137     .subsections = (const VMStateDescription*[]) {
 138         &vmstate_virtio_pci_modern_state_sub,
 139         NULL
 140     }
 141 };
 142
 143 static bool virtio_pci_has_extra_state(DeviceState *d)
 144 {
 145     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 146
 147     return proxy->flags & VIRTIO_PCI_FLAG_MIGRATE_EXTRA;
 148 }
 149
 150 static void virtio_pci_save_extra_state(DeviceState *d, QEMUFile *f)
 151 {
 152     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 153
 154     vmstate_save_state(f, &vmstate_virtio_pci, proxy, NULL);
 155 }
 156
 157 static int virtio_pci_load_extra_state(DeviceState *d, QEMUFile *f)
 158 {
 159     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 160
 161     return vmstate_load_state(f, &vmstate_virtio_pci, proxy, 1);
 162 }
 163
 164 static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f)
 165 {
 166     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 167     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 168
 169     if (msix_present(&proxy->pci_dev))
 170         qemu_put_be16(f, virtio_queue_vector(vdev, n));
 171 }
 172
 173 static int virtio_pci_load_config(DeviceState *d, QEMUFile *f)
 174 {
 175     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 176     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 177
 178     int ret;
 179     ret = pci_device_load(&proxy->pci_dev, f);
 180     if (ret) {
 181         return ret;
 182     }
 183     msix_unuse_all_vectors(&proxy->pci_dev);
 184     msix_load(&proxy->pci_dev, f);
 185     if (msix_present(&proxy->pci_dev)) {
 186         qemu_get_be16s(f, &vdev->config_vector);
 187     } else {
 188         vdev->config_vector = VIRTIO_NO_VECTOR;
 189     }
 190     if (vdev->config_vector != VIRTIO_NO_VECTOR) {
 191         return msix_vector_use(&proxy->pci_dev, vdev->config_vector);
 192     }
 193     return 0;
 194 }
 195
 196 static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f)
 197 {
 198     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 199     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 200
 201     uint16_t vector;
 202     if (msix_present(&proxy->pci_dev)) {
 203         qemu_get_be16s(f, &vector);
 204     } else {
 205         vector = VIRTIO_NO_VECTOR;
 206     }
 207     virtio_queue_set_vector(vdev, n, vector);
 208     if (vector != VIRTIO_NO_VECTOR) {
 209         return msix_vector_use(&proxy->pci_dev, vector);
 210     }
 211
 212     return 0;
 213 }
 214
 215 static bool virtio_pci_ioeventfd_enabled(DeviceState *d)
 216 {
 217     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 218
 219     return (proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) != 0;
 220 }
 221
 222 #define QEMU_VIRTIO_PCI_QUEUE_MEM_MULT 0x1000
 223
 224 static inline int virtio_pci_queue_mem_mult(struct VirtIOPCIProxy *proxy)
 225 {
 226     return (proxy->flags & VIRTIO_PCI_FLAG_PAGE_PER_VQ) ?
 227         QEMU_VIRTIO_PCI_QUEUE_MEM_MULT : 4;
 228 }
 229
 230 static int virtio_pci_ioeventfd_assign(DeviceState *d, EventNotifier *notifier,
 231                                        int n, bool assign)
 232 {
 233     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 234     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 235     VirtQueue *vq = virtio_get_queue(vdev, n);
 236     bool legacy = virtio_pci_legacy(proxy);
 237     bool modern = virtio_pci_modern(proxy);
 238     bool fast_mmio = kvm_ioeventfd_any_length_enabled();
 239     bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
 240     MemoryRegion *modern_mr = &proxy->notify.mr;
 241     MemoryRegion *modern_notify_mr = &proxy->notify_pio.mr;
 242     MemoryRegion *legacy_mr = &proxy->bar;
 243     hwaddr modern_addr = virtio_pci_queue_mem_mult(proxy) *
 244                          virtio_get_queue_index(vq);
 245     hwaddr legacy_addr = VIRTIO_PCI_QUEUE_NOTIFY;
 246
 247     if (assign) {
 248         if (modern) {
 249             if (fast_mmio) {
 250                 memory_region_add_eventfd(modern_mr, modern_addr, 0,
 251                                           false, n, notifier);
 252             } else {
 253                 memory_region_add_eventfd(modern_mr, modern_addr, 2,
 254                                           false, n, notifier);
 255             }
 256             if (modern_pio) {
 257                 memory_region_add_eventfd(modern_notify_mr, 0, 2,
 258                                               true, n, notifier);
 259             }
 260         }
 261         if (legacy) {
 262             memory_region_add_eventfd(legacy_mr, legacy_addr, 2,
 263                                       true, n, notifier);
 264         }
 265     } else {
 266         if (modern) {
 267             if (fast_mmio) {
 268                 memory_region_del_eventfd(modern_mr, modern_addr, 0,
 269                                           false, n, notifier);
 270             } else {
 271                 memory_region_del_eventfd(modern_mr, modern_addr, 2,
 272                                           false, n, notifier);
 273             }
 274             if (modern_pio) {
 275                 memory_region_del_eventfd(modern_notify_mr, 0, 2,
 276                                           true, n, notifier);
 277             }
 278         }
 279         if (legacy) {
 280             memory_region_del_eventfd(legacy_mr, legacy_addr, 2,
 281                                       true, n, notifier);
 282         }
 283     }
 284     return 0;
 285 }
 286
 287 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 288 {
 289     virtio_bus_start_ioeventfd(&proxy->bus);
 290 }
 291
 292 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 293 {
 294     virtio_bus_stop_ioeventfd(&proxy->bus);
 295 }
 296
 297 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 298 {
 299     VirtIOPCIProxy *proxy = opaque;
 300     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 301     hwaddr pa;
 302
 303     switch (addr) {
 304     case VIRTIO_PCI_GUEST_FEATURES:
 305         /* Guest does not negotiate properly?  We have to assume nothing. */
 306         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 307             val = virtio_bus_get_vdev_bad_features(&proxy->bus);
 308         }
 309         virtio_set_features(vdev, val);
 310         break;
 311     case VIRTIO_PCI_QUEUE_PFN:
 312         pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 313         if (pa == 0) {
 314             virtio_pci_reset(DEVICE(proxy));
 315         }
 316         else
 317             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 318         break;
 319     case VIRTIO_PCI_QUEUE_SEL:
 320         if (val < VIRTIO_QUEUE_MAX)
 321             vdev->queue_sel = val;
 322         break;
 323     case VIRTIO_PCI_QUEUE_NOTIFY:
 324         if (val < VIRTIO_QUEUE_MAX) {
 325             virtio_queue_notify(vdev, val);
 326         }
 327         break;
 328     case VIRTIO_PCI_STATUS:
 329         if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
 330             virtio_pci_stop_ioeventfd(proxy);
 331         }
 332
 333         virtio_set_status(vdev, val & 0xFF);
 334
 335         if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
 336             virtio_pci_start_ioeventfd(proxy);
 337         }
 338
 339         if (vdev->status == 0) {
 340             virtio_pci_reset(DEVICE(proxy));
 341         }
 342
 343         /* Linux before 2.6.34 drives the device without enabling
 344            the PCI device bus master bit. Enable it automatically
 345            for the guest. This is a PCI spec violation but so is
 346            initiating DMA with bus master bit clear. */
 347         if (val == (VIRTIO_CONFIG_S_ACKNOWLEDGE | VIRTIO_CONFIG_S_DRIVER)) {
 348             pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
 349                                      proxy->pci_dev.config[PCI_COMMAND] |
 350                                      PCI_COMMAND_MASTER, 1);
 351         }
 352         break;
 353     case VIRTIO_MSI_CONFIG_VECTOR:
 354         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 355         /* Make it possible for guest to discover an error took place. */
 356         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 357             val = VIRTIO_NO_VECTOR;
 358         vdev->config_vector = val;
 359         break;
 360     case VIRTIO_MSI_QUEUE_VECTOR:
 361         msix_vector_unuse(&proxy->pci_dev,
 362                           virtio_queue_vector(vdev, vdev->queue_sel));
 363         /* Make it possible for guest to discover an error took place. */
 364         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 365             val = VIRTIO_NO_VECTOR;
 366         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 367         break;
 368     default:
 369         qemu_log_mask(LOG_GUEST_ERROR,
 370                       "%s: unexpected address 0x%x value 0x%x\n",
 371                       __func__, addr, val);
 372         break;
 373     }
 374 }
 375
 376 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 377 {
 378     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 379     uint32_t ret = 0xFFFFFFFF;
 380
 381     switch (addr) {
 382     case VIRTIO_PCI_HOST_FEATURES:
 383         ret = vdev->host_features;
 384         break;
 385     case VIRTIO_PCI_GUEST_FEATURES:
 386         ret = vdev->guest_features;
 387         break;
 388     case VIRTIO_PCI_QUEUE_PFN:
 389         ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 390               >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 391         break;
 392     case VIRTIO_PCI_QUEUE_NUM:
 393         ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 394         break;
 395     case VIRTIO_PCI_QUEUE_SEL:
 396         ret = vdev->queue_sel;
 397         break;
 398     case VIRTIO_PCI_STATUS:
 399         ret = vdev->status;
 400         break;
 401     case VIRTIO_PCI_ISR:
 402         /* reading from the ISR also clears it. */
 403         ret = qatomic_xchg(&vdev->isr, 0);
 404         pci_irq_deassert(&proxy->pci_dev);
 405         break;
 406     case VIRTIO_MSI_CONFIG_VECTOR:
 407         ret = vdev->config_vector;
 408         break;
 409     case VIRTIO_MSI_QUEUE_VECTOR:
 410         ret = virtio_queue_vector(vdev, vdev->queue_sel);
 411         break;
 412     default:
 413         break;
 414     }
 415
 416     return ret;
 417 }
 418
 419 static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr,
 420                                        unsigned size)
 421 {
 422     VirtIOPCIProxy *proxy = opaque;
 423     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 424     uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
 425     uint64_t val = 0;
 426
 427     if (vdev == NULL) {
 428         return UINT64_MAX;
 429     }
 430
 431     if (addr < config) {
 432         return virtio_ioport_read(proxy, addr);
 433     }
 434     addr -= config;
 435
 436     switch (size) {
 437     case 1:
 438         val = virtio_config_readb(vdev, addr);
 439         break;
 440     case 2:
 441         val = virtio_config_readw(vdev, addr);
 442         if (virtio_is_big_endian(vdev)) {
 443             val = bswap16(val);
 444         }
 445         break;
 446     case 4:
 447         val = virtio_config_readl(vdev, addr);
 448         if (virtio_is_big_endian(vdev)) {
 449             val = bswap32(val);
 450         }
 451         break;
 452     }
 453     return val;
 454 }
 455
 456 static void virtio_pci_config_write(void *opaque, hwaddr addr,
 457                                     uint64_t val, unsigned size)
 458 {
 459     VirtIOPCIProxy *proxy = opaque;
 460     uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
 461     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 462
 463     if (vdev == NULL) {
 464         return;
 465     }
 466
 467     if (addr < config) {
 468         virtio_ioport_write(proxy, addr, val);
 469         return;
 470     }
 471     addr -= config;
 472     /*
 473      * Virtio-PCI is odd. Ioports are LE but config space is target native
 474      * endian.
 475      */
 476     switch (size) {
 477     case 1:
 478         virtio_config_writeb(vdev, addr, val);
 479         break;
 480     case 2:
 481         if (virtio_is_big_endian(vdev)) {
 482             val = bswap16(val);
 483         }
 484         virtio_config_writew(vdev, addr, val);
 485         break;
 486     case 4:
 487         if (virtio_is_big_endian(vdev)) {
 488             val = bswap32(val);
 489         }
 490         virtio_config_writel(vdev, addr, val);
 491         break;
 492     }
 493 }
 494
 495 static const MemoryRegionOps virtio_pci_config_ops = {
 496     .read = virtio_pci_config_read,
 497     .write = virtio_pci_config_write,
 498     .impl = {
 499         .min_access_size = 1,
 500         .max_access_size = 4,
 501     },
 502     .endianness = DEVICE_LITTLE_ENDIAN,
 503 };
 504
 505 static MemoryRegion *virtio_address_space_lookup(VirtIOPCIProxy *proxy,
 506                                                  hwaddr *off, int len)
 507 {
 508     int i;
 509     VirtIOPCIRegion *reg;
 510
 511     for (i = 0; i < ARRAY_SIZE(proxy->regs); ++i) {
 512         reg = &proxy->regs[i];
 513         if (*off >= reg->offset &&
 514             *off + len <= reg->offset + reg->size) {
 515             *off -= reg->offset;
 516             return &reg->mr;
 517         }
 518     }
 519
 520     return NULL;
 521 }
 522
 523 /* Below are generic functions to do memcpy from/to an address space,
 524  * without byteswaps, with input validation.
 525  *
 526  * As regular address_space_* APIs all do some kind of byteswap at least for
 527  * some host/target combinations, we are forced to explicitly convert to a
 528  * known-endianness integer value.
 529  * It doesn't really matter which endian format to go through, so the code
 530  * below selects the endian that causes the least amount of work on the given
 531  * host.
 532  *
 533  * Note: host pointer must be aligned.
 534  */
 535 static
 536 void virtio_address_space_write(VirtIOPCIProxy *proxy, hwaddr addr,
 537                                 const uint8_t *buf, int len)
 538 {
 539     uint64_t val;
 540     MemoryRegion *mr;
 541
 542     /* address_space_* APIs assume an aligned address.
 543      * As address is under guest control, handle illegal values.
 544      */
 545     addr &= ~(len - 1);
 546
 547     mr = virtio_address_space_lookup(proxy, &addr, len);
 548     if (!mr) {
 549         return;
 550     }
 551
 552     /* Make sure caller aligned buf properly */
 553     assert(!(((uintptr_t)buf) & (len - 1)));
 554
 555     switch (len) {
 556     case 1:
 557         val = pci_get_byte(buf);
 558         break;
 559     case 2:
 560         val = pci_get_word(buf);
 561         break;
 562     case 4:
 563         val = pci_get_long(buf);
 564         break;
 565     default:
 566         /* As length is under guest control, handle illegal values. */
 567         return;
 568     }
 569     memory_region_dispatch_write(mr, addr, val, size_memop(len) | MO_LE,
 570                                  MEMTXATTRS_UNSPECIFIED);
 571 }
 572
 573 static void
 574 virtio_address_space_read(VirtIOPCIProxy *proxy, hwaddr addr,
 575                           uint8_t *buf, int len)
 576 {
 577     uint64_t val;
 578     MemoryRegion *mr;
 579
 580     /* address_space_* APIs assume an aligned address.
 581      * As address is under guest control, handle illegal values.
 582      */
 583     addr &= ~(len - 1);
 584
 585     mr = virtio_address_space_lookup(proxy, &addr, len);
 586     if (!mr) {
 587         return;
 588     }
 589
 590     /* Make sure caller aligned buf properly */
 591     assert(!(((uintptr_t)buf) & (len - 1)));
 592
 593     memory_region_dispatch_read(mr, addr, &val, size_memop(len) | MO_LE,
 594                                 MEMTXATTRS_UNSPECIFIED);
 595     switch (len) {
 596     case 1:
 597         pci_set_byte(buf, val);
 598         break;
 599     case 2:
 600         pci_set_word(buf, val);
 601         break;
 602     case 4:
 603         pci_set_long(buf, val);
 604         break;
 605     default:
 606         /* As length is under guest control, handle illegal values. */
 607         break;
 608     }
 609 }
 610
 611 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 612                                 uint32_t val, int len)
 613 {
 614     VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
 615     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 616     struct virtio_pci_cfg_cap *cfg;
 617
 618     pci_default_write_config(pci_dev, address, val, len);
 619
 620     if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
 621         pcie_cap_flr_write_config(pci_dev, address, val, len);
 622     }
 623
 624     if (range_covers_byte(address, len, PCI_COMMAND)) {
 625         if (!(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 626             virtio_set_disabled(vdev, true);
 627             virtio_pci_stop_ioeventfd(proxy);
 628             virtio_set_status(vdev, vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 629         } else {
 630             virtio_set_disabled(vdev, false);
 631         }
 632     }
 633
 634     if (proxy->config_cap &&
 635         ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
 636                                                                   pci_cfg_data),
 637                        sizeof cfg->pci_cfg_data)) {
 638         uint32_t off;
 639         uint32_t len;
 640
 641         cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
 642         off = le32_to_cpu(cfg->cap.offset);
 643         len = le32_to_cpu(cfg->cap.length);
 644
 645         if (len == 1 || len == 2 || len == 4) {
 646             assert(len <= sizeof cfg->pci_cfg_data);
 647             virtio_address_space_write(proxy, off, cfg->pci_cfg_data, len);
 648         }
 649     }
 650 }
 651
 652 static uint32_t virtio_read_config(PCIDevice *pci_dev,
 653                                    uint32_t address, int len)
 654 {
 655     VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
 656     struct virtio_pci_cfg_cap *cfg;
 657
 658     if (proxy->config_cap &&
 659         ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
 660                                                                   pci_cfg_data),
 661                        sizeof cfg->pci_cfg_data)) {
 662         uint32_t off;
 663         uint32_t len;
 664
 665         cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
 666         off = le32_to_cpu(cfg->cap.offset);
 667         len = le32_to_cpu(cfg->cap.length);
 668
 669         if (len == 1 || len == 2 || len == 4) {
 670             assert(len <= sizeof cfg->pci_cfg_data);
 671             virtio_address_space_read(proxy, off, cfg->pci_cfg_data, len);
 672         }
 673     }
 674
 675     return pci_default_read_config(pci_dev, address, len);
 676 }
 677
 678 static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy,
 679                                         unsigned int queue_no,
 680                                         unsigned int vector)
 681 {
 682     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 683     int ret;
 684
 685     if (irqfd->users == 0) {
 686         ret = kvm_irqchip_add_msi_route(kvm_state, vector, &proxy->pci_dev);
 687         if (ret < 0) {
 688             return ret;
 689         }
 690         irqfd->virq = ret;
 691     }
 692     irqfd->users++;
 693     return 0;
 694 }
 695
 696 static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy,
 697                                              unsigned int vector)
 698 {
 699     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 700     if (--irqfd->users == 0) {
 701         kvm_irqchip_release_virq(kvm_state, irqfd->virq);
 702     }
 703 }
 704
 705 static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy,
 706                                  unsigned int queue_no,
 707                                  unsigned int vector)
 708 {
 709     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 710     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 711     VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 712     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 713     return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, irqfd->virq);
 714 }
 715
 716 static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy,
 717                                       unsigned int queue_no,
 718                                       unsigned int vector)
 719 {
 720     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 721     VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 722     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 723     VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
 724     int ret;
 725
 726     ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n, irqfd->virq);
 727     assert(ret == 0);
 728 }
 729
 730 static int kvm_virtio_pci_vector_use(VirtIOPCIProxy *proxy, int nvqs)
 731 {
 732     PCIDevice *dev = &proxy->pci_dev;
 733     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 734     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 735     unsigned int vector;
 736     int ret, queue_no;
 737
 738     for (queue_no = 0; queue_no < nvqs; queue_no++) {
 739         if (!virtio_queue_get_num(vdev, queue_no)) {
 740             break;
 741         }
 742         vector = virtio_queue_vector(vdev, queue_no);
 743         if (vector >= msix_nr_vectors_allocated(dev)) {
 744             continue;
 745         }
 746         ret = kvm_virtio_pci_vq_vector_use(proxy, queue_no, vector);
 747         if (ret < 0) {
 748             goto undo;
 749         }
 750         /* If guest supports masking, set up irqfd now.
 751          * Otherwise, delay until unmasked in the frontend.
 752          */
 753         if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 754             ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 755             if (ret < 0) {
 756                 kvm_virtio_pci_vq_vector_release(proxy, vector);
 757                 goto undo;
 758             }
 759         }
 760     }
 761     return 0;
 762
 763 undo:
 764     while (--queue_no >= 0) {
 765         vector = virtio_queue_vector(vdev, queue_no);
 766         if (vector >= msix_nr_vectors_allocated(dev)) {
 767             continue;
 768         }
 769         if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 770             kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 771         }
 772         kvm_virtio_pci_vq_vector_release(proxy, vector);
 773     }
 774     return ret;
 775 }
 776
 777 static void kvm_virtio_pci_vector_release(VirtIOPCIProxy *proxy, int nvqs)
 778 {
 779     PCIDevice *dev = &proxy->pci_dev;
 780     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 781     unsigned int vector;
 782     int queue_no;
 783     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 784
 785     for (queue_no = 0; queue_no < nvqs; queue_no++) {
 786         if (!virtio_queue_get_num(vdev, queue_no)) {
 787             break;
 788         }
 789         vector = virtio_queue_vector(vdev, queue_no);
 790         if (vector >= msix_nr_vectors_allocated(dev)) {
 791             continue;
 792         }
 793         /* If guest supports masking, clean up irqfd now.
 794          * Otherwise, it was cleaned when masked in the frontend.
 795          */
 796         if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 797             kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 798         }
 799         kvm_virtio_pci_vq_vector_release(proxy, vector);
 800     }
 801 }
 802
 803 static int virtio_pci_vq_vector_unmask(VirtIOPCIProxy *proxy,
 804                                        unsigned int queue_no,
 805                                        unsigned int vector,
 806                                        MSIMessage msg)
 807 {
 808     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 809     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 810     VirtQueue *vq = virtio_get_queue(vdev, queue_no);
 811     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 812     VirtIOIRQFD *irqfd;
 813     int ret = 0;
 814
 815     if (proxy->vector_irqfd) {
 816         irqfd = &proxy->vector_irqfd[vector];
 817         if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) {
 818             ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg,
 819                                                &proxy->pci_dev);
 820             if (ret < 0) {
 821                 return ret;
 822             }
 823             kvm_irqchip_commit_routes(kvm_state);
 824         }
 825     }
 826
 827     /* If guest supports masking, irqfd is already setup, unmask it.
 828      * Otherwise, set it up now.
 829      */
 830     if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 831         k->guest_notifier_mask(vdev, queue_no, false);
 832         /* Test after unmasking to avoid losing events. */
 833         if (k->guest_notifier_pending &&
 834             k->guest_notifier_pending(vdev, queue_no)) {
 835             event_notifier_set(n);
 836         }
 837     } else {
 838         ret = kvm_virtio_pci_irqfd_use(proxy, queue_no, vector);
 839     }
 840     return ret;
 841 }
 842
 843 static void virtio_pci_vq_vector_mask(VirtIOPCIProxy *proxy,
 844                                              unsigned int queue_no,
 845                                              unsigned int vector)
 846 {
 847     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 848     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 849
 850     /* If guest supports masking, keep irqfd but mask it.
 851      * Otherwise, clean it up now.
 852      */
 853     if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
 854         k->guest_notifier_mask(vdev, queue_no, true);
 855     } else {
 856         kvm_virtio_pci_irqfd_release(proxy, queue_no, vector);
 857     }
 858 }
 859
 860 static int virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector,
 861                                     MSIMessage msg)
 862 {
 863     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 864     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 865     VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
 866     int ret, index, unmasked = 0;
 867
 868     while (vq) {
 869         index = virtio_get_queue_index(vq);
 870         if (!virtio_queue_get_num(vdev, index)) {
 871             break;
 872         }
 873         if (index < proxy->nvqs_with_notifiers) {
 874             ret = virtio_pci_vq_vector_unmask(proxy, index, vector, msg);
 875             if (ret < 0) {
 876                 goto undo;
 877             }
 878             ++unmasked;
 879         }
 880         vq = virtio_vector_next_queue(vq);
 881     }
 882
 883     return 0;
 884
 885 undo:
 886     vq = virtio_vector_first_queue(vdev, vector);
 887     while (vq && unmasked >= 0) {
 888         index = virtio_get_queue_index(vq);
 889         if (index < proxy->nvqs_with_notifiers) {
 890             virtio_pci_vq_vector_mask(proxy, index, vector);
 891             --unmasked;
 892         }
 893         vq = virtio_vector_next_queue(vq);
 894     }
 895     return ret;
 896 }
 897
 898 static void virtio_pci_vector_mask(PCIDevice *dev, unsigned vector)
 899 {
 900     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 901     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 902     VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
 903     int index;
 904
 905     while (vq) {
 906         index = virtio_get_queue_index(vq);
 907         if (!virtio_queue_get_num(vdev, index)) {
 908             break;
 909         }
 910         if (index < proxy->nvqs_with_notifiers) {
 911             virtio_pci_vq_vector_mask(proxy, index, vector);
 912         }
 913         vq = virtio_vector_next_queue(vq);
 914     }
 915 }
 916
 917 static void virtio_pci_vector_poll(PCIDevice *dev,
 918                                    unsigned int vector_start,
 919                                    unsigned int vector_end)
 920 {
 921     VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
 922     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 923     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 924     int queue_no;
 925     unsigned int vector;
 926     EventNotifier *notifier;
 927     VirtQueue *vq;
 928
 929     for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
 930         if (!virtio_queue_get_num(vdev, queue_no)) {
 931             break;
 932         }
 933         vector = virtio_queue_vector(vdev, queue_no);
 934         if (vector < vector_start || vector >= vector_end ||
 935             !msix_is_masked(dev, vector)) {
 936             continue;
 937         }
 938         vq = virtio_get_queue(vdev, queue_no);
 939         notifier = virtio_queue_get_guest_notifier(vq);
 940         if (k->guest_notifier_pending) {
 941             if (k->guest_notifier_pending(vdev, queue_no)) {
 942                 msix_set_pending(dev, vector);
 943             }
 944         } else if (event_notifier_test_and_clear(notifier)) {
 945             msix_set_pending(dev, vector);
 946         }
 947     }
 948 }
 949
 950 static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign,
 951                                          bool with_irqfd)
 952 {
 953     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 954     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 955     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
 956     VirtQueue *vq = virtio_get_queue(vdev, n);
 957     EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 958
 959     if (assign) {
 960         int r = event_notifier_init(notifier, 0);
 961         if (r < 0) {
 962             return r;
 963         }
 964         virtio_queue_set_guest_notifier_fd_handler(vq, true, with_irqfd);
 965     } else {
 966         virtio_queue_set_guest_notifier_fd_handler(vq, false, with_irqfd);
 967         event_notifier_cleanup(notifier);
 968     }
 969
 970     if (!msix_enabled(&proxy->pci_dev) &&
 971         vdev->use_guest_notifier_mask &&
 972         vdc->guest_notifier_mask) {
 973         vdc->guest_notifier_mask(vdev, n, !assign);
 974     }
 975
 976     return 0;
 977 }
 978
 979 static bool virtio_pci_query_guest_notifiers(DeviceState *d)
 980 {
 981     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 982     return msix_enabled(&proxy->pci_dev);
 983 }
 984
 985 static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign)
 986 {
 987     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
 988     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
 989     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
 990     int r, n;
 991     bool with_irqfd = msix_enabled(&proxy->pci_dev) &&
 992         kvm_msi_via_irqfd_enabled();
 993
 994     nvqs = MIN(nvqs, VIRTIO_QUEUE_MAX);
 995
 996     /* When deassigning, pass a consistent nvqs value
 997      * to avoid leaking notifiers.
 998      */
 999     assert(assign || nvqs == proxy->nvqs_with_notifiers);
1000
1001     proxy->nvqs_with_notifiers = nvqs;
1002
1003     /* Must unset vector notifier while guest notifier is still assigned */
1004     if ((proxy->vector_irqfd || k->guest_notifier_mask) && !assign) {
1005         msix_unset_vector_notifiers(&proxy->pci_dev);
1006         if (proxy->vector_irqfd) {
1007             kvm_virtio_pci_vector_release(proxy, nvqs);
1008             g_free(proxy->vector_irqfd);
1009             proxy->vector_irqfd = NULL;
1010         }
1011     }
1012
1013     for (n = 0; n < nvqs; n++) {
1014         if (!virtio_queue_get_num(vdev, n)) {
1015             break;
1016         }
1017
1018         r = virtio_pci_set_guest_notifier(d, n, assign, with_irqfd);
1019         if (r < 0) {
1020             goto assign_error;
1021         }
1022     }
1023
1024     /* Must set vector notifier after guest notifier has been assigned */
1025     if ((with_irqfd || k->guest_notifier_mask) && assign) {
1026         if (with_irqfd) {
1027             proxy->vector_irqfd =
1028                 g_malloc0(sizeof(*proxy->vector_irqfd) *
1029                           msix_nr_vectors_allocated(&proxy->pci_dev));
1030             r = kvm_virtio_pci_vector_use(proxy, nvqs);
1031             if (r < 0) {
1032                 goto assign_error;
1033             }
1034         }
1035         r = msix_set_vector_notifiers(&proxy->pci_dev,
1036                                       virtio_pci_vector_unmask,
1037                                       virtio_pci_vector_mask,
1038                                       virtio_pci_vector_poll);
1039         if (r < 0) {
1040             goto notifiers_error;
1041         }
1042     }
1043
1044     return 0;
1045
1046 notifiers_error:
1047     if (with_irqfd) {
1048         assert(assign);
1049         kvm_virtio_pci_vector_release(proxy, nvqs);
1050     }
1051
1052 assign_error:
1053     /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
1054     assert(assign);
1055     while (--n >= 0) {
1056         virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd);
1057     }
1058     return r;
1059 }
1060
1061 static int virtio_pci_set_host_notifier_mr(DeviceState *d, int n,
1062                                            MemoryRegion *mr, bool assign)
1063 {
1064     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1065     int offset;
1066
1067     if (n >= VIRTIO_QUEUE_MAX || !virtio_pci_modern(proxy) ||
1068         virtio_pci_queue_mem_mult(proxy) != memory_region_size(mr)) {
1069         return -1;
1070     }
1071
1072     if (assign) {
1073         offset = virtio_pci_queue_mem_mult(proxy) * n;
1074         memory_region_add_subregion_overlap(&proxy->notify.mr, offset, mr, 1);
1075     } else {
1076         memory_region_del_subregion(&proxy->notify.mr, mr);
1077     }
1078
1079     return 0;
1080 }
1081
1082 static void virtio_pci_vmstate_change(DeviceState *d, bool running)
1083 {
1084     VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1085     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1086
1087     if (running) {
1088         /* Old QEMU versions did not set bus master enable on status write.
1089          * Detect DRIVER set and enable it.
1090          */
1091         if ((proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION) &&
1092             (vdev->status & VIRTIO_CONFIG_S_DRIVER) &&
1093             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
1094             pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
1095                                      proxy->pci_dev.config[PCI_COMMAND] |
1096                                      PCI_COMMAND_MASTER, 1);
1097         }
1098         virtio_pci_start_ioeventfd(proxy);
1099     } else {
1100         virtio_pci_stop_ioeventfd(proxy);
1101     }
1102 }
1103
1104 /*
1105  * virtio-pci: This is the PCIDevice which has a virtio-pci-bus.
1106  */
1107
1108 static int virtio_pci_query_nvectors(DeviceState *d)
1109 {
1110     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1111
1112     return proxy->nvectors;
1113 }
1114
1115 static AddressSpace *virtio_pci_get_dma_as(DeviceState *d)
1116 {
1117     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1118     PCIDevice *dev = &proxy->pci_dev;
1119
1120     return pci_get_address_space(dev);
1121 }
1122
1123 static bool virtio_pci_iommu_enabled(DeviceState *d)
1124 {
1125     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1126     PCIDevice *dev = &proxy->pci_dev;
1127     AddressSpace *dma_as = pci_device_iommu_address_space(dev);
1128
1129     if (dma_as == &address_space_memory) {
1130         return false;
1131     }
1132
1133     return true;
1134 }
1135
1136 static bool virtio_pci_queue_enabled(DeviceState *d, int n)
1137 {
1138     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1139     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1140
1141     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1142         return proxy->vqs[n].enabled;
1143     }
1144
1145     return virtio_queue_enabled_legacy(vdev, n);
1146 }
1147
1148 static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy,
1149                                    struct virtio_pci_cap *cap)
1150 {
1151     PCIDevice *dev = &proxy->pci_dev;
1152     int offset;
1153
1154     offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0,
1155                                 cap->cap_len, &error_abort);
1156
1157     assert(cap->cap_len >= sizeof *cap);
1158     memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len,
1159            cap->cap_len - PCI_CAP_FLAGS);
1160
1161     return offset;
1162 }
1163
1164 static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr,
1165                                        unsigned size)
1166 {
1167     VirtIOPCIProxy *proxy = opaque;
1168     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1169     uint32_t val = 0;
1170     int i;
1171
1172     if (vdev == NULL) {
1173         return UINT64_MAX;
1174     }
1175
1176     switch (addr) {
1177     case VIRTIO_PCI_COMMON_DFSELECT:
1178         val = proxy->dfselect;
1179         break;
1180     case VIRTIO_PCI_COMMON_DF:
1181         if (proxy->dfselect <= 1) {
1182             VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1183
1184             val = (vdev->host_features & ~vdc->legacy_features) >>
1185                 (32 * proxy->dfselect);
1186         }
1187         break;
1188     case VIRTIO_PCI_COMMON_GFSELECT:
1189         val = proxy->gfselect;
1190         break;
1191     case VIRTIO_PCI_COMMON_GF:
1192         if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1193             val = proxy->guest_features[proxy->gfselect];
1194         }
1195         break;
1196     case VIRTIO_PCI_COMMON_MSIX:
1197         val = vdev->config_vector;
1198         break;
1199     case VIRTIO_PCI_COMMON_NUMQ:
1200         for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) {
1201             if (virtio_queue_get_num(vdev, i)) {
1202                 val = i + 1;
1203             }
1204         }
1205         break;
1206     case VIRTIO_PCI_COMMON_STATUS:
1207         val = vdev->status;
1208         break;
1209     case VIRTIO_PCI_COMMON_CFGGENERATION:
1210         val = vdev->generation;
1211         break;
1212     case VIRTIO_PCI_COMMON_Q_SELECT:
1213         val = vdev->queue_sel;
1214         break;
1215     case VIRTIO_PCI_COMMON_Q_SIZE:
1216         val = virtio_queue_get_num(vdev, vdev->queue_sel);
1217         break;
1218     case VIRTIO_PCI_COMMON_Q_MSIX:
1219         val = virtio_queue_vector(vdev, vdev->queue_sel);
1220         break;
1221     case VIRTIO_PCI_COMMON_Q_ENABLE:
1222         val = proxy->vqs[vdev->queue_sel].enabled;
1223         break;
1224     case VIRTIO_PCI_COMMON_Q_NOFF:
1225         /* Simply map queues in order */
1226         val = vdev->queue_sel;
1227         break;
1228     case VIRTIO_PCI_COMMON_Q_DESCLO:
1229         val = proxy->vqs[vdev->queue_sel].desc[0];
1230         break;
1231     case VIRTIO_PCI_COMMON_Q_DESCHI:
1232         val = proxy->vqs[vdev->queue_sel].desc[1];
1233         break;
1234     case VIRTIO_PCI_COMMON_Q_AVAILLO:
1235         val = proxy->vqs[vdev->queue_sel].avail[0];
1236         break;
1237     case VIRTIO_PCI_COMMON_Q_AVAILHI:
1238         val = proxy->vqs[vdev->queue_sel].avail[1];
1239         break;
1240     case VIRTIO_PCI_COMMON_Q_USEDLO:
1241         val = proxy->vqs[vdev->queue_sel].used[0];
1242         break;
1243     case VIRTIO_PCI_COMMON_Q_USEDHI:
1244         val = proxy->vqs[vdev->queue_sel].used[1];
1245         break;
1246     default:
1247         val = 0;
1248     }
1249
1250     return val;
1251 }
1252
1253 static void virtio_pci_common_write(void *opaque, hwaddr addr,
1254                                     uint64_t val, unsigned size)
1255 {
1256     VirtIOPCIProxy *proxy = opaque;
1257     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1258
1259     if (vdev == NULL) {
1260         return;
1261     }
1262
1263     switch (addr) {
1264     case VIRTIO_PCI_COMMON_DFSELECT:
1265         proxy->dfselect = val;
1266         break;
1267     case VIRTIO_PCI_COMMON_GFSELECT:
1268         proxy->gfselect = val;
1269         break;
1270     case VIRTIO_PCI_COMMON_GF:
1271         if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1272             proxy->guest_features[proxy->gfselect] = val;
1273             virtio_set_features(vdev,
1274                                 (((uint64_t)proxy->guest_features[1]) << 32) |
1275                                 proxy->guest_features[0]);
1276         }
1277         break;
1278     case VIRTIO_PCI_COMMON_MSIX:
1279         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
1280         /* Make it possible for guest to discover an error took place. */
1281         if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1282             val = VIRTIO_NO_VECTOR;
1283         }
1284         vdev->config_vector = val;
1285         break;
1286     case VIRTIO_PCI_COMMON_STATUS:
1287         if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1288             virtio_pci_stop_ioeventfd(proxy);
1289         }
1290
1291         virtio_set_status(vdev, val & 0xFF);
1292
1293         if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
1294             virtio_pci_start_ioeventfd(proxy);
1295         }
1296
1297         if (vdev->status == 0) {
1298             virtio_pci_reset(DEVICE(proxy));
1299         }
1300
1301         break;
1302     case VIRTIO_PCI_COMMON_Q_SELECT:
1303         if (val < VIRTIO_QUEUE_MAX) {
1304             vdev->queue_sel = val;
1305         }
1306         break;
1307     case VIRTIO_PCI_COMMON_Q_SIZE:
1308         proxy->vqs[vdev->queue_sel].num = val;
1309         virtio_queue_set_num(vdev, vdev->queue_sel,
1310                              proxy->vqs[vdev->queue_sel].num);
1311         break;
1312     case VIRTIO_PCI_COMMON_Q_MSIX:
1313         msix_vector_unuse(&proxy->pci_dev,
1314                           virtio_queue_vector(vdev, vdev->queue_sel));
1315         /* Make it possible for guest to discover an error took place. */
1316         if (msix_vector_use(&proxy->pci_dev, val) < 0) {
1317             val = VIRTIO_NO_VECTOR;
1318         }
1319         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
1320         break;
1321     case VIRTIO_PCI_COMMON_Q_ENABLE:
1322         if (val == 1) {
1323             virtio_queue_set_num(vdev, vdev->queue_sel,
1324                                  proxy->vqs[vdev->queue_sel].num);
1325             virtio_queue_set_rings(vdev, vdev->queue_sel,
1326                        ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 |
1327                        proxy->vqs[vdev->queue_sel].desc[0],
1328                        ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 |
1329                        proxy->vqs[vdev->queue_sel].avail[0],
1330                        ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 |
1331                        proxy->vqs[vdev->queue_sel].used[0]);
1332             proxy->vqs[vdev->queue_sel].enabled = 1;
1333         } else {
1334             virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val);
1335         }
1336         break;
1337     case VIRTIO_PCI_COMMON_Q_DESCLO:
1338         proxy->vqs[vdev->queue_sel].desc[0] = val;
1339         break;
1340     case VIRTIO_PCI_COMMON_Q_DESCHI:
1341         proxy->vqs[vdev->queue_sel].desc[1] = val;
1342         break;
1343     case VIRTIO_PCI_COMMON_Q_AVAILLO:
1344         proxy->vqs[vdev->queue_sel].avail[0] = val;
1345         break;
1346     case VIRTIO_PCI_COMMON_Q_AVAILHI:
1347         proxy->vqs[vdev->queue_sel].avail[1] = val;
1348         break;
1349     case VIRTIO_PCI_COMMON_Q_USEDLO:
1350         proxy->vqs[vdev->queue_sel].used[0] = val;
1351         break;
1352     case VIRTIO_PCI_COMMON_Q_USEDHI:
1353         proxy->vqs[vdev->queue_sel].used[1] = val;
1354         break;
1355     default:
1356         break;
1357     }
1358 }
1359
1360
1361 static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr,
1362                                        unsigned size)
1363 {
1364     VirtIOPCIProxy *proxy = opaque;
1365     if (virtio_bus_get_device(&proxy->bus) == NULL) {
1366         return UINT64_MAX;
1367     }
1368
1369     return 0;
1370 }
1371
1372 static void virtio_pci_notify_write(void *opaque, hwaddr addr,
1373                                     uint64_t val, unsigned size)
1374 {
1375     VirtIOPCIProxy *proxy = opaque;
1376     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1377
1378     unsigned queue = addr / virtio_pci_queue_mem_mult(proxy);
1379
1380     if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1381         virtio_queue_notify(vdev, queue);
1382     }
1383 }
1384
1385 static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr,
1386                                         uint64_t val, unsigned size)
1387 {
1388     VirtIOPCIProxy *proxy = opaque;
1389     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1390
1391     unsigned queue = val;
1392
1393     if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1394         virtio_queue_notify(vdev, queue);
1395     }
1396 }
1397
1398 static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr,
1399                                     unsigned size)
1400 {
1401     VirtIOPCIProxy *proxy = opaque;
1402     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1403     uint64_t val;
1404
1405     if (vdev == NULL) {
1406         return UINT64_MAX;
1407     }
1408
1409     val = qatomic_xchg(&vdev->isr, 0);
1410     pci_irq_deassert(&proxy->pci_dev);
1411     return val;
1412 }
1413
1414 static void virtio_pci_isr_write(void *opaque, hwaddr addr,
1415                                  uint64_t val, unsigned size)
1416 {
1417 }
1418
1419 static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr,
1420                                        unsigned size)
1421 {
1422     VirtIOPCIProxy *proxy = opaque;
1423     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1424     uint64_t val;
1425
1426     if (vdev == NULL) {
1427         return UINT64_MAX;
1428     }
1429
1430     switch (size) {
1431     case 1:
1432         val = virtio_config_modern_readb(vdev, addr);
1433         break;
1434     case 2:
1435         val = virtio_config_modern_readw(vdev, addr);
1436         break;
1437     case 4:
1438         val = virtio_config_modern_readl(vdev, addr);
1439         break;
1440     default:
1441         val = 0;
1442         break;
1443     }
1444     return val;
1445 }
1446
1447 static void virtio_pci_device_write(void *opaque, hwaddr addr,
1448                                     uint64_t val, unsigned size)
1449 {
1450     VirtIOPCIProxy *proxy = opaque;
1451     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1452
1453     if (vdev == NULL) {
1454         return;
1455     }
1456
1457     switch (size) {
1458     case 1:
1459         virtio_config_modern_writeb(vdev, addr, val);
1460         break;
1461     case 2:
1462         virtio_config_modern_writew(vdev, addr, val);
1463         break;
1464     case 4:
1465         virtio_config_modern_writel(vdev, addr, val);
1466         break;
1467     }
1468 }
1469
1470 static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy,
1471                                            const char *vdev_name)
1472 {
1473     static const MemoryRegionOps common_ops = {
1474         .read = virtio_pci_common_read,
1475         .write = virtio_pci_common_write,
1476         .impl = {
1477             .min_access_size = 1,
1478             .max_access_size = 4,
1479         },
1480         .endianness = DEVICE_LITTLE_ENDIAN,
1481     };
1482     static const MemoryRegionOps isr_ops = {
1483         .read = virtio_pci_isr_read,
1484         .write = virtio_pci_isr_write,
1485         .impl = {
1486             .min_access_size = 1,
1487             .max_access_size = 4,
1488         },
1489         .endianness = DEVICE_LITTLE_ENDIAN,
1490     };
1491     static const MemoryRegionOps device_ops = {
1492         .read = virtio_pci_device_read,
1493         .write = virtio_pci_device_write,
1494         .impl = {
1495             .min_access_size = 1,
1496             .max_access_size = 4,
1497         },
1498         .endianness = DEVICE_LITTLE_ENDIAN,
1499     };
1500     static const MemoryRegionOps notify_ops = {
1501         .read = virtio_pci_notify_read,
1502         .write = virtio_pci_notify_write,
1503         .impl = {
1504             .min_access_size = 1,
1505             .max_access_size = 4,
1506         },
1507         .endianness = DEVICE_LITTLE_ENDIAN,
1508     };
1509     static const MemoryRegionOps notify_pio_ops = {
1510         .read = virtio_pci_notify_read,
1511         .write = virtio_pci_notify_write_pio,
1512         .impl = {
1513             .min_access_size = 1,
1514             .max_access_size = 4,
1515         },
1516         .endianness = DEVICE_LITTLE_ENDIAN,
1517     };
1518     g_autoptr(GString) name = g_string_new(NULL);
1519
1520     g_string_printf(name, "virtio-pci-common-%s", vdev_name);
1521     memory_region_init_io(&proxy->common.mr, OBJECT(proxy),
1522                           &common_ops,
1523                           proxy,
1524                           name->str,
1525                           proxy->common.size);
1526
1527     g_string_printf(name, "virtio-pci-isr-%s", vdev_name);
1528     memory_region_init_io(&proxy->isr.mr, OBJECT(proxy),
1529                           &isr_ops,
1530                           proxy,
1531                           name->str,
1532                           proxy->isr.size);
1533
1534     g_string_printf(name, "virtio-pci-device-%s", vdev_name);
1535     memory_region_init_io(&proxy->device.mr, OBJECT(proxy),
1536                           &device_ops,
1537                           proxy,
1538                           name->str,
1539                           proxy->device.size);
1540
1541     g_string_printf(name, "virtio-pci-notify-%s", vdev_name);
1542     memory_region_init_io(&proxy->notify.mr, OBJECT(proxy),
1543                           &notify_ops,
1544                           proxy,
1545                           name->str,
1546                           proxy->notify.size);
1547
1548     g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name);
1549     memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy),
1550                           &notify_pio_ops,
1551                           proxy,
1552                           name->str,
1553                           proxy->notify_pio.size);
1554 }
1555
1556 static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy,
1557                                          VirtIOPCIRegion *region,
1558                                          struct virtio_pci_cap *cap,
1559                                          MemoryRegion *mr,
1560                                          uint8_t bar)
1561 {
1562     memory_region_add_subregion(mr, region->offset, &region->mr);
1563
1564     cap->cfg_type = region->type;
1565     cap->bar = bar;
1566     cap->offset = cpu_to_le32(region->offset);
1567     cap->length = cpu_to_le32(region->size);
1568     virtio_pci_add_mem_cap(proxy, cap);
1569
1570 }
1571
1572 static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy,
1573                                              VirtIOPCIRegion *region,
1574                                              struct virtio_pci_cap *cap)
1575 {
1576     virtio_pci_modern_region_map(proxy, region, cap,
1577                                  &proxy->modern_bar, proxy->modern_mem_bar_idx);
1578 }
1579
1580 static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy,
1581                                             VirtIOPCIRegion *region,
1582                                             struct virtio_pci_cap *cap)
1583 {
1584     virtio_pci_modern_region_map(proxy, region, cap,
1585                                  &proxy->io_bar, proxy->modern_io_bar_idx);
1586 }
1587
1588 static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy,
1589                                                VirtIOPCIRegion *region)
1590 {
1591     memory_region_del_subregion(&proxy->modern_bar,
1592                                 &region->mr);
1593 }
1594
1595 static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy,
1596                                               VirtIOPCIRegion *region)
1597 {
1598     memory_region_del_subregion(&proxy->io_bar,
1599                                 &region->mr);
1600 }
1601
1602 static void virtio_pci_pre_plugged(DeviceState *d, Error **errp)
1603 {
1604     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1605     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1606
1607     if (virtio_pci_modern(proxy)) {
1608         virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
1609     }
1610
1611     virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE);
1612 }
1613
1614 /* This is called by virtio-bus just after the device is plugged. */
1615 static void virtio_pci_device_plugged(DeviceState *d, Error **errp)
1616 {
1617     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1618     VirtioBusState *bus = &proxy->bus;
1619     bool legacy = virtio_pci_legacy(proxy);
1620     bool modern;
1621     bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1622     uint8_t *config;
1623     uint32_t size;
1624     VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1625
1626     /*
1627      * Virtio capabilities present without
1628      * VIRTIO_F_VERSION_1 confuses guests
1629      */
1630     if (!proxy->ignore_backend_features &&
1631             !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) {
1632         virtio_pci_disable_modern(proxy);
1633
1634         if (!legacy) {
1635             error_setg(errp, "Device doesn't support modern mode, and legacy"
1636                              " mode is disabled");
1637             error_append_hint(errp, "Set disable-legacy to off\n");
1638
1639             return;
1640         }
1641     }
1642
1643     modern = virtio_pci_modern(proxy);
1644
1645     config = proxy->pci_dev.config;
1646     if (proxy->class_code) {
1647         pci_config_set_class(config, proxy->class_code);
1648     }
1649
1650     if (legacy) {
1651         if (!virtio_legacy_allowed(vdev)) {
1652             /*
1653              * To avoid migration issues, we allow legacy mode when legacy
1654              * check is disabled in the old machine types (< 5.1).
1655              */
1656             if (virtio_legacy_check_disabled(vdev)) {
1657                 warn_report("device is modern-only, but for backward "
1658                             "compatibility legacy is allowed");
1659             } else {
1660                 error_setg(errp,
1661                            "device is modern-only, use disable-legacy=on");
1662                 return;
1663             }
1664         }
1665         if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1666             error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by"
1667                        " neither legacy nor transitional device");
1668             return ;
1669         }
1670         /*
1671          * Legacy and transitional devices use specific subsystem IDs.
1672          * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID)
1673          * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default.
1674          */
1675         pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus));
1676     } else {
1677         /* pure virtio-1.0 */
1678         pci_set_word(config + PCI_VENDOR_ID,
1679                      PCI_VENDOR_ID_REDHAT_QUMRANET);
1680         pci_set_word(config + PCI_DEVICE_ID,
1681                      0x1040 + virtio_bus_get_vdev_id(bus));
1682         pci_config_set_revision(config, 1);
1683     }
1684     config[PCI_INTERRUPT_PIN] = 1;
1685
1686
1687     if (modern) {
1688         struct virtio_pci_cap cap = {
1689             .cap_len = sizeof cap,
1690         };
1691         struct virtio_pci_notify_cap notify = {
1692             .cap.cap_len = sizeof notify,
1693             .notify_off_multiplier =
1694                 cpu_to_le32(virtio_pci_queue_mem_mult(proxy)),
1695         };
1696         struct virtio_pci_cfg_cap cfg = {
1697             .cap.cap_len = sizeof cfg,
1698             .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG,
1699         };
1700         struct virtio_pci_notify_cap notify_pio = {
1701             .cap.cap_len = sizeof notify,
1702             .notify_off_multiplier = cpu_to_le32(0x0),
1703         };
1704
1705         struct virtio_pci_cfg_cap *cfg_mask;
1706
1707         virtio_pci_modern_regions_init(proxy, vdev->name);
1708
1709         virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap);
1710         virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap);
1711         virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap);
1712         virtio_pci_modern_mem_region_map(proxy, &proxy->notify, &notify.cap);
1713
1714         if (modern_pio) {
1715             memory_region_init(&proxy->io_bar, OBJECT(proxy),
1716                                "virtio-pci-io", 0x4);
1717
1718             pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx,
1719                              PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar);
1720
1721             virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio,
1722                                             &notify_pio.cap);
1723         }
1724
1725         pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx,
1726                          PCI_BASE_ADDRESS_SPACE_MEMORY |
1727                          PCI_BASE_ADDRESS_MEM_PREFETCH |
1728                          PCI_BASE_ADDRESS_MEM_TYPE_64,
1729                          &proxy->modern_bar);
1730
1731         proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap);
1732         cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap);
1733         pci_set_byte(&cfg_mask->cap.bar, ~0x0);
1734         pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0);
1735         pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0);
1736         pci_set_long(cfg_mask->pci_cfg_data, ~0x0);
1737     }
1738
1739     if (proxy->nvectors) {
1740         int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors,
1741                                           proxy->msix_bar_idx, NULL);
1742         if (err) {
1743             /* Notice when a system that supports MSIx can't initialize it */
1744             if (err != -ENOTSUP) {
1745                 warn_report("unable to init msix vectors to %" PRIu32,
1746                             proxy->nvectors);
1747             }
1748             proxy->nvectors = 0;
1749         }
1750     }
1751
1752     proxy->pci_dev.config_write = virtio_write_config;
1753     proxy->pci_dev.config_read = virtio_read_config;
1754
1755     if (legacy) {
1756         size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev)
1757             + virtio_bus_get_vdev_config_len(bus);
1758         size = pow2ceil(size);
1759
1760         memory_region_init_io(&proxy->bar, OBJECT(proxy),
1761                               &virtio_pci_config_ops,
1762                               proxy, "virtio-pci", size);
1763
1764         pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx,
1765                          PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar);
1766     }
1767 }
1768
1769 static void virtio_pci_device_unplugged(DeviceState *d)
1770 {
1771     VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1772     bool modern = virtio_pci_modern(proxy);
1773     bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1774
1775     virtio_pci_stop_ioeventfd(proxy);
1776
1777     if (modern) {
1778         virtio_pci_modern_mem_region_unmap(proxy, &proxy->common);
1779         virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr);
1780         virtio_pci_modern_mem_region_unmap(proxy, &proxy->device);
1781         virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify);
1782         if (modern_pio) {
1783             virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio);
1784         }
1785     }
1786 }
1787
1788 static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
1789 {
1790     VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1791     VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev);
1792     bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1793                      !pci_bus_is_root(pci_get_bus(pci_dev));
1794
1795     if (kvm_enabled() && !kvm_has_many_ioeventfds()) {
1796         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1797     }
1798
1799     /* fd-based ioevents can't be synchronized in record/replay */
1800     if (replay_mode != REPLAY_MODE_NONE) {
1801         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
1802     }
1803
1804     /*
1805      * virtio pci bar layout used by default.
1806      * subclasses can re-arrange things if needed.
1807      *
1808      *   region 0   --  virtio legacy io bar
1809      *   region 1   --  msi-x bar
1810      *   region 2   --  virtio modern io bar (off by default)
1811      *   region 4+5 --  virtio modern memory (64bit) bar
1812      *
1813      */
1814     proxy->legacy_io_bar_idx  = 0;
1815     proxy->msix_bar_idx       = 1;
1816     proxy->modern_io_bar_idx  = 2;
1817     proxy->modern_mem_bar_idx = 4;
1818
1819     proxy->common.offset = 0x0;
1820     proxy->common.size = 0x1000;
1821     proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG;
1822
1823     proxy->isr.offset = 0x1000;
1824     proxy->isr.size = 0x1000;
1825     proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG;
1826
1827     proxy->device.offset = 0x2000;
1828     proxy->device.size = 0x1000;
1829     proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG;
1830
1831     proxy->notify.offset = 0x3000;
1832     proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX;
1833     proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1834
1835     proxy->notify_pio.offset = 0x0;
1836     proxy->notify_pio.size = 0x4;
1837     proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
1838
1839     /* subclasses can enforce modern, so do this unconditionally */
1840     memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci",
1841                        /* PCI BAR regions must be powers of 2 */
1842                        pow2ceil(proxy->notify.offset + proxy->notify.size));
1843
1844     if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) {
1845         proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
1846     }
1847
1848     if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) {
1849         error_setg(errp, "device cannot work as neither modern nor legacy mode"
1850                    " is enabled");
1851         error_append_hint(errp, "Set either disable-modern or disable-legacy"
1852                           " to off\n");
1853         return;
1854     }
1855
1856     if (pcie_port && pci_is_express(pci_dev)) {
1857         int pos;
1858         uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE;
1859
1860         pos = pcie_endpoint_cap_init(pci_dev, 0);
1861         assert(pos > 0);
1862
1863         pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0,
1864                                  PCI_PM_SIZEOF, errp);
1865         if (pos < 0) {
1866             return;
1867         }
1868
1869         pci_dev->exp.pm_cap = pos;
1870
1871         /*
1872          * Indicates that this function complies with revision 1.2 of the
1873          * PCI Power Management Interface Specification.
1874          */
1875         pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3);
1876
1877         if (proxy->flags & VIRTIO_PCI_FLAG_AER) {
1878             pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset,
1879                           PCI_ERR_SIZEOF, NULL);
1880             last_pcie_cap_offset += PCI_ERR_SIZEOF;
1881         }
1882
1883         if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) {
1884             /* Init error enabling flags */
1885             pcie_cap_deverr_init(pci_dev);
1886         }
1887
1888         if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) {
1889             /* Init Link Control Register */
1890             pcie_cap_lnkctl_init(pci_dev);
1891         }
1892
1893         if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) {
1894             /* Init Power Management Control Register */
1895             pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL,
1896                          PCI_PM_CTRL_STATE_MASK);
1897         }
1898
1899         if (proxy->flags & VIRTIO_PCI_FLAG_ATS) {
1900             pcie_ats_init(pci_dev, last_pcie_cap_offset,
1901                           proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED);
1902             last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF;
1903         }
1904
1905         if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
1906             /* Set Function Level Reset capability bit */
1907             pcie_cap_flr_init(pci_dev);
1908         }
1909     } else {
1910         /*
1911          * make future invocations of pci_is_express() return false
1912          * and pci_config_size() return PCI_CONFIG_SPACE_SIZE.
1913          */
1914         pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS;
1915     }
1916
1917     virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
1918     if (k->realize) {
1919         k->realize(proxy, errp);
1920     }
1921 }
1922
1923 static void virtio_pci_exit(PCIDevice *pci_dev)
1924 {
1925     VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
1926     bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
1927                      !pci_bus_is_root(pci_get_bus(pci_dev));
1928
1929     msix_uninit_exclusive_bar(pci_dev);
1930     if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port &&
1931         pci_is_express(pci_dev)) {
1932         pcie_aer_exit(pci_dev);
1933     }
1934 }
1935
1936 static void virtio_pci_reset(DeviceState *qdev)
1937 {
1938     VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1939     VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
1940     PCIDevice *dev = PCI_DEVICE(qdev);
1941     int i;
1942
1943     virtio_pci_stop_ioeventfd(proxy);
1944     virtio_bus_reset(bus);
1945     msix_unuse_all_vectors(&proxy->pci_dev);
1946
1947     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
1948         proxy->vqs[i].enabled = 0;
1949         proxy->vqs[i].num = 0;
1950         proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0;
1951         proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0;
1952         proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0;
1953     }
1954
1955     if (pci_is_express(dev)) {
1956         pcie_cap_deverr_reset(dev);
1957         pcie_cap_lnkctl_reset(dev);
1958
1959         pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0);
1960     }
1961 }
1962
1963 static Property virtio_pci_properties[] = {
1964     DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags,
1965                     VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false),
1966     DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags,
1967                     VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true),
1968     DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags,
1969                     VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false),
1970     DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags,
1971                     VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false),
1972     DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags,
1973                     VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false),
1974     DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy,
1975                      ignore_backend_features, false),
1976     DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags,
1977                     VIRTIO_PCI_FLAG_ATS_BIT, false),
1978     DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags,
1979                     VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true),
1980     DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags,
1981                     VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true),
1982     DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags,
1983                     VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true),
1984     DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags,
1985                     VIRTIO_PCI_FLAG_INIT_PM_BIT, true),
1986     DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags,
1987                     VIRTIO_PCI_FLAG_INIT_FLR_BIT, true),
1988     DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags,
1989                     VIRTIO_PCI_FLAG_AER_BIT, false),
1990     DEFINE_PROP_END_OF_LIST(),
1991 };
1992
1993 static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp)
1994 {
1995     VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev);
1996     VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
1997     PCIDevice *pci_dev = &proxy->pci_dev;
1998
1999     if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) &&
2000         virtio_pci_modern(proxy)) {
2001         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2002     }
2003
2004     vpciklass->parent_dc_realize(qdev, errp);
2005 }
2006
2007 static void virtio_pci_class_init(ObjectClass *klass, void *data)
2008 {
2009     DeviceClass *dc = DEVICE_CLASS(klass);
2010     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
2011     VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass);
2012
2013     device_class_set_props(dc, virtio_pci_properties);
2014     k->realize = virtio_pci_realize;
2015     k->exit = virtio_pci_exit;
2016     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
2017     k->revision = VIRTIO_PCI_ABI_VERSION;
2018     k->class_id = PCI_CLASS_OTHERS;
2019     device_class_set_parent_realize(dc, virtio_pci_dc_realize,
2020                                     &vpciklass->parent_dc_realize);
2021     dc->reset = virtio_pci_reset;
2022 }
2023
2024 static const TypeInfo virtio_pci_info = {
2025     .name          = TYPE_VIRTIO_PCI,
2026     .parent        = TYPE_PCI_DEVICE,
2027     .instance_size = sizeof(VirtIOPCIProxy),
2028     .class_init    = virtio_pci_class_init,
2029     .class_size    = sizeof(VirtioPCIClass),
2030     .abstract      = true,
2031 };
2032
2033 static Property virtio_pci_generic_properties[] = {
2034     DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy,
2035                             ON_OFF_AUTO_AUTO),
2036     DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false),
2037     DEFINE_PROP_END_OF_LIST(),
2038 };
2039
2040 static void virtio_pci_base_class_init(ObjectClass *klass, void *data)
2041 {
2042     const VirtioPCIDeviceTypeInfo *t = data;
2043     if (t->class_init) {
2044         t->class_init(klass, NULL);
2045     }
2046 }
2047
2048 static void virtio_pci_generic_class_init(ObjectClass *klass, void *data)
2049 {
2050     DeviceClass *dc = DEVICE_CLASS(klass);
2051
2052     device_class_set_props(dc, virtio_pci_generic_properties);
2053 }
2054
2055 static void virtio_pci_transitional_instance_init(Object *obj)
2056 {
2057     VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2058
2059     proxy->disable_legacy = ON_OFF_AUTO_OFF;
2060     proxy->disable_modern = false;
2061 }
2062
2063 static void virtio_pci_non_transitional_instance_init(Object *obj)
2064 {
2065     VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2066
2067     proxy->disable_legacy = ON_OFF_AUTO_ON;
2068     proxy->disable_modern = false;
2069 }
2070
2071 void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t)
2072 {
2073     char *base_name = NULL;
2074     TypeInfo base_type_info = {
2075         .name          = t->base_name,
2076         .parent        = t->parent ? t->parent : TYPE_VIRTIO_PCI,
2077         .instance_size = t->instance_size,
2078         .instance_init = t->instance_init,
2079         .class_size    = t->class_size,
2080         .abstract      = true,
2081         .interfaces    = t->interfaces,
2082     };
2083     TypeInfo generic_type_info = {
2084         .name = t->generic_name,
2085         .parent = base_type_info.name,
2086         .class_init = virtio_pci_generic_class_init,
2087         .interfaces = (InterfaceInfo[]) {
2088             { INTERFACE_PCIE_DEVICE },
2089             { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2090             { }
2091         },
2092     };
2093
2094     if (!base_type_info.name) {
2095         /* No base type -> register a single generic device type */
2096         /* use intermediate %s-base-type to add generic device props */
2097         base_name = g_strdup_printf("%s-base-type", t->generic_name);
2098         base_type_info.name = base_name;
2099         base_type_info.class_init = virtio_pci_generic_class_init;
2100
2101         generic_type_info.parent = base_name;
2102         generic_type_info.class_init = virtio_pci_base_class_init;
2103         generic_type_info.class_data = (void *)t;
2104
2105         assert(!t->non_transitional_name);
2106         assert(!t->transitional_name);
2107     } else {
2108         base_type_info.class_init = virtio_pci_base_class_init;
2109         base_type_info.class_data = (void *)t;
2110     }
2111
2112     type_register(&base_type_info);
2113     if (generic_type_info.name) {
2114         type_register(&generic_type_info);
2115     }
2116
2117     if (t->non_transitional_name) {
2118         const TypeInfo non_transitional_type_info = {
2119             .name          = t->non_transitional_name,
2120             .parent        = base_type_info.name,
2121             .instance_init = virtio_pci_non_transitional_instance_init,
2122             .interfaces = (InterfaceInfo[]) {
2123                 { INTERFACE_PCIE_DEVICE },
2124                 { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2125                 { }
2126             },
2127         };
2128         type_register(&non_transitional_type_info);
2129     }
2130
2131     if (t->transitional_name) {
2132         const TypeInfo transitional_type_info = {
2133             .name          = t->transitional_name,
2134             .parent        = base_type_info.name,
2135             .instance_init = virtio_pci_transitional_instance_init,
2136             .interfaces = (InterfaceInfo[]) {
2137                 /*
2138                  * Transitional virtio devices work only as Conventional PCI
2139                  * devices because they require PIO ports.
2140                  */
2141                 { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2142                 { }
2143             },
2144         };
2145         type_register(&transitional_type_info);
2146     }
2147     g_free(base_name);
2148 }
2149
2150 unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues)
2151 {
2152     /*
2153      * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted
2154      * virtqueue buffers can handle their completion. When a different vCPU
2155      * handles completion it may need to IPI the vCPU that submitted the
2156      * request and this adds overhead.
2157      *
2158      * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in
2159      * guests with very many vCPUs and a device that is only used by a few
2160      * vCPUs. Unfortunately optimizing that case requires manual pinning inside
2161      * the guest, so those users might as well manually set the number of
2162      * queues. There is no upper limit that can be applied automatically and
2163      * doing so arbitrarily would result in a sudden performance drop once the
2164      * threshold number of vCPUs is exceeded.
2165      */
2166     unsigned num_queues = current_machine->smp.cpus;
2167
2168     /*
2169      * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the
2170      * config change interrupt and the fixed virtqueues must be taken into
2171      * account too.
2172      */
2173     num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues);
2174
2175     /*
2176      * There is a limit to how many virtqueues a device can have.
2177      */
2178     return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues);
2179 }
2180
2181 /* virtio-pci-bus */
2182
2183 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
2184                                VirtIOPCIProxy *dev)
2185 {
2186     DeviceState *qdev = DEVICE(dev);
2187     char virtio_bus_name[] = "virtio-bus";
2188
2189     qbus_init(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev, virtio_bus_name);
2190 }
2191
2192 static void virtio_pci_bus_class_init(ObjectClass *klass, void *data)
2193 {
2194     BusClass *bus_class = BUS_CLASS(klass);
2195     VirtioBusClass *k = VIRTIO_BUS_CLASS(klass);
2196     bus_class->max_dev = 1;
2197     k->notify = virtio_pci_notify;
2198     k->save_config = virtio_pci_save_config;
2199     k->load_config = virtio_pci_load_config;
2200     k->save_queue = virtio_pci_save_queue;
2201     k->load_queue = virtio_pci_load_queue;
2202     k->save_extra_state = virtio_pci_save_extra_state;
2203     k->load_extra_state = virtio_pci_load_extra_state;
2204     k->has_extra_state = virtio_pci_has_extra_state;
2205     k->query_guest_notifiers = virtio_pci_query_guest_notifiers;
2206     k->set_guest_notifiers = virtio_pci_set_guest_notifiers;
2207     k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr;
2208     k->vmstate_change = virtio_pci_vmstate_change;
2209     k->pre_plugged = virtio_pci_pre_plugged;
2210     k->device_plugged = virtio_pci_device_plugged;
2211     k->device_unplugged = virtio_pci_device_unplugged;
2212     k->query_nvectors = virtio_pci_query_nvectors;
2213     k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled;
2214     k->ioeventfd_assign = virtio_pci_ioeventfd_assign;
2215     k->get_dma_as = virtio_pci_get_dma_as;
2216     k->iommu_enabled = virtio_pci_iommu_enabled;
2217     k->queue_enabled = virtio_pci_queue_enabled;
2218 }
2219
2220 static const TypeInfo virtio_pci_bus_info = {
2221     .name          = TYPE_VIRTIO_PCI_BUS,
2222     .parent        = TYPE_VIRTIO_BUS,
2223     .instance_size = sizeof(VirtioPCIBusState),
2224     .class_size    = sizeof(VirtioPCIBusClass),
2225     .class_init    = virtio_pci_bus_class_init,
2226 };
2227
2228 static void virtio_pci_register_types(void)
2229 {
2230     /* Base types: */
2231     type_register_static(&virtio_pci_bus_info);
2232     type_register_static(&virtio_pci_info);
2233 }
2234
2235 type_init(virtio_pci_register_types)
2236