hw/virtio-pci.c

   1 /*
   2  * Virtio PCI Bindings
   3  *
   4  * Copyright IBM, Corp. 2007
   5  * Copyright (c) 2009 CodeSourcery
   6  *
   7  * Authors:
   8  *  Anthony Liguori   <aliguori@us.ibm.com>
   9  *  Paul Brook        <paul@codesourcery.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  * Contributions after 2012-01-13 are licensed under the terms of the
  15  * GNU GPL, version 2 or (at your option) any later version.
  16  */
  17
  18 #include <inttypes.h>
  19
  20 #include "virtio.h"
  21 #include "virtio-blk.h"
  22 #include "virtio-net.h"
  23 #include "virtio-serial.h"
  24 #include "virtio-scsi.h"
  25 #include "pci.h"
  26 #include "qemu-error.h"
  27 #include "msix.h"
  28 #include "net.h"
  29 #include "loader.h"
  30 #include "kvm.h"
  31 #include "blockdev.h"
  32 #include "virtio-pci.h"
  33 #include "range.h"
  34
  35 /* from Linux's linux/virtio_pci.h */
  36
  37 /* A 32-bit r/o bitmask of the features supported by the host */
  38 #define VIRTIO_PCI_HOST_FEATURES        0
  39
  40 /* A 32-bit r/w bitmask of features activated by the guest */
  41 #define VIRTIO_PCI_GUEST_FEATURES       4
  42
  43 /* A 32-bit r/w PFN for the currently selected queue */
  44 #define VIRTIO_PCI_QUEUE_PFN            8
  45
  46 /* A 16-bit r/o queue size for the currently selected queue */
  47 #define VIRTIO_PCI_QUEUE_NUM            12
  48
  49 /* A 16-bit r/w queue selector */
  50 #define VIRTIO_PCI_QUEUE_SEL            14
  51
  52 /* A 16-bit r/w queue notifier */
  53 #define VIRTIO_PCI_QUEUE_NOTIFY         16
  54
  55 /* An 8-bit device status register.  */
  56 #define VIRTIO_PCI_STATUS               18
  57
  58 /* An 8-bit r/o interrupt status register.  Reading the value will return the
  59  * current contents of the ISR and will also clear it.  This is effectively
  60  * a read-and-acknowledge. */
  61 #define VIRTIO_PCI_ISR                  19
  62
  63 /* MSI-X registers: only enabled if MSI-X is enabled. */
  64 /* A 16-bit vector for configuration changes. */
  65 #define VIRTIO_MSI_CONFIG_VECTOR        20
  66 /* A 16-bit vector for selected queue notifications. */
  67 #define VIRTIO_MSI_QUEUE_VECTOR         22
  68
  69 /* Config space size */
  70 #define VIRTIO_PCI_CONFIG_NOMSI         20
  71 #define VIRTIO_PCI_CONFIG_MSI           24
  72 #define VIRTIO_PCI_REGION_SIZE(dev)     (msix_present(dev) ? \
  73                                          VIRTIO_PCI_CONFIG_MSI : \
  74                                          VIRTIO_PCI_CONFIG_NOMSI)
  75
  76 /* The remaining space is defined by each driver as the per-driver
  77  * configuration space */
  78 #define VIRTIO_PCI_CONFIG(dev)          (msix_enabled(dev) ? \
  79                                          VIRTIO_PCI_CONFIG_MSI : \
  80                                          VIRTIO_PCI_CONFIG_NOMSI)
  81
  82 /* How many bits to shift physical queue address written to QUEUE_PFN.
  83  * 12 is historical, and due to x86 page size. */
  84 #define VIRTIO_PCI_QUEUE_ADDR_SHIFT    12
  85
  86 /* Flags track per-device state like workarounds for quirks in older guests. */
  87 #define VIRTIO_PCI_FLAG_BUS_MASTER_BUG  (1 << 0)
  88
  89 /* QEMU doesn't strictly need write barriers since everything runs in
  90  * lock-step.  We'll leave the calls to wmb() in though to make it obvious for
  91  * KVM or if kqemu gets SMP support.
  92  */
  93 #define wmb() do { } while (0)
  94
  95 /* HACK for virtio to determine if it's running a big endian guest */
  96 bool virtio_is_big_endian(void);
  97
  98 /* virtio device */
  99
 100 static void virtio_pci_notify(void *opaque, uint16_t vector)
 101 {
 102     VirtIOPCIProxy *proxy = opaque;
 103     if (msix_enabled(&proxy->pci_dev))
 104         msix_notify(&proxy->pci_dev, vector);
 105     else
 106         qemu_set_irq(proxy->pci_dev.irq[0], proxy->vdev->isr & 1);
 107 }
 108
 109 static void virtio_pci_save_config(void * opaque, QEMUFile *f)
 110 {
 111     VirtIOPCIProxy *proxy = opaque;
 112     pci_device_save(&proxy->pci_dev, f);
 113     msix_save(&proxy->pci_dev, f);
 114     if (msix_present(&proxy->pci_dev))
 115         qemu_put_be16(f, proxy->vdev->config_vector);
 116 }
 117
 118 static void virtio_pci_save_queue(void * opaque, int n, QEMUFile *f)
 119 {
 120     VirtIOPCIProxy *proxy = opaque;
 121     if (msix_present(&proxy->pci_dev))
 122         qemu_put_be16(f, virtio_queue_vector(proxy->vdev, n));
 123 }
 124
 125 static int virtio_pci_load_config(void * opaque, QEMUFile *f)
 126 {
 127     VirtIOPCIProxy *proxy = opaque;
 128     int ret;
 129     ret = pci_device_load(&proxy->pci_dev, f);
 130     if (ret) {
 131         return ret;
 132     }
 133     msix_load(&proxy->pci_dev, f);
 134     if (msix_present(&proxy->pci_dev)) {
 135         qemu_get_be16s(f, &proxy->vdev->config_vector);
 136     } else {
 137         proxy->vdev->config_vector = VIRTIO_NO_VECTOR;
 138     }
 139     if (proxy->vdev->config_vector != VIRTIO_NO_VECTOR) {
 140         return msix_vector_use(&proxy->pci_dev, proxy->vdev->config_vector);
 141     }
 142     return 0;
 143 }
 144
 145 static int virtio_pci_load_queue(void * opaque, int n, QEMUFile *f)
 146 {
 147     VirtIOPCIProxy *proxy = opaque;
 148     uint16_t vector;
 149     if (msix_present(&proxy->pci_dev)) {
 150         qemu_get_be16s(f, &vector);
 151     } else {
 152         vector = VIRTIO_NO_VECTOR;
 153     }
 154     virtio_queue_set_vector(proxy->vdev, n, vector);
 155     if (vector != VIRTIO_NO_VECTOR) {
 156         return msix_vector_use(&proxy->pci_dev, vector);
 157     }
 158     return 0;
 159 }
 160
 161 static int virtio_pci_set_host_notifier_internal(VirtIOPCIProxy *proxy,
 162                                                  int n, bool assign)
 163 {
 164     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 165     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 166     int r = 0;
 167
 168     if (assign) {
 169         r = event_notifier_init(notifier, 1);
 170         if (r < 0) {
 171             error_report("%s: unable to init event notifier: %d",
 172                          __func__, r);
 173             return r;
 174         }
 175         memory_region_add_eventfd(&proxy->bar, VIRTIO_PCI_QUEUE_NOTIFY, 2,
 176                                   true, n, event_notifier_get_fd(notifier));
 177     } else {
 178         memory_region_del_eventfd(&proxy->bar, VIRTIO_PCI_QUEUE_NOTIFY, 2,
 179                                   true, n, event_notifier_get_fd(notifier));
 180         /* Handle the race condition where the guest kicked and we deassigned
 181          * before we got around to handling the kick.
 182          */
 183         if (event_notifier_test_and_clear(notifier)) {
 184             virtio_queue_notify_vq(vq);
 185         }
 186
 187         event_notifier_cleanup(notifier);
 188     }
 189     return r;
 190 }
 191
 192 static void virtio_pci_host_notifier_read(void *opaque)
 193 {
 194     VirtQueue *vq = opaque;
 195     EventNotifier *n = virtio_queue_get_host_notifier(vq);
 196     if (event_notifier_test_and_clear(n)) {
 197         virtio_queue_notify_vq(vq);
 198     }
 199 }
 200
 201 static void virtio_pci_set_host_notifier_fd_handler(VirtIOPCIProxy *proxy,
 202                                                     int n, bool assign)
 203 {
 204     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 205     EventNotifier *notifier = virtio_queue_get_host_notifier(vq);
 206     if (assign) {
 207         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 208                             virtio_pci_host_notifier_read, NULL, vq);
 209     } else {
 210         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 211                             NULL, NULL, NULL);
 212     }
 213 }
 214
 215 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
 216 {
 217     int n, r;
 218
 219     if (!(proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) ||
 220         proxy->ioeventfd_disabled ||
 221         proxy->ioeventfd_started) {
 222         return;
 223     }
 224
 225     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 226         if (!virtio_queue_get_num(proxy->vdev, n)) {
 227             continue;
 228         }
 229
 230         r = virtio_pci_set_host_notifier_internal(proxy, n, true);
 231         if (r < 0) {
 232             goto assign_error;
 233         }
 234
 235         virtio_pci_set_host_notifier_fd_handler(proxy, n, true);
 236     }
 237     proxy->ioeventfd_started = true;
 238     return;
 239
 240 assign_error:
 241     while (--n >= 0) {
 242         if (!virtio_queue_get_num(proxy->vdev, n)) {
 243             continue;
 244         }
 245
 246         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 247         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 248         assert(r >= 0);
 249     }
 250     proxy->ioeventfd_started = false;
 251     error_report("%s: failed. Fallback to a userspace (slower).", __func__);
 252 }
 253
 254 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
 255 {
 256     int r;
 257     int n;
 258
 259     if (!proxy->ioeventfd_started) {
 260         return;
 261     }
 262
 263     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 264         if (!virtio_queue_get_num(proxy->vdev, n)) {
 265             continue;
 266         }
 267
 268         virtio_pci_set_host_notifier_fd_handler(proxy, n, false);
 269         r = virtio_pci_set_host_notifier_internal(proxy, n, false);
 270         assert(r >= 0);
 271     }
 272     proxy->ioeventfd_started = false;
 273 }
 274
 275 void virtio_pci_reset(DeviceState *d)
 276 {
 277     VirtIOPCIProxy *proxy = container_of(d, VirtIOPCIProxy, pci_dev.qdev);
 278     virtio_pci_stop_ioeventfd(proxy);
 279     virtio_reset(proxy->vdev);
 280     msix_reset(&proxy->pci_dev);
 281     proxy->flags &= ~VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 282 }
 283
 284 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
 285 {
 286     VirtIOPCIProxy *proxy = opaque;
 287     VirtIODevice *vdev = proxy->vdev;
 288     target_phys_addr_t pa;
 289
 290     switch (addr) {
 291     case VIRTIO_PCI_GUEST_FEATURES:
 292         /* Guest does not negotiate properly?  We have to assume nothing. */
 293         if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
 294             val = vdev->bad_features ? vdev->bad_features(vdev) : 0;
 295         }
 296         virtio_set_features(vdev, val);
 297         break;
 298     case VIRTIO_PCI_QUEUE_PFN:
 299         pa = (target_phys_addr_t)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 300         if (pa == 0) {
 301             virtio_pci_stop_ioeventfd(proxy);
 302             virtio_reset(proxy->vdev);
 303             msix_unuse_all_vectors(&proxy->pci_dev);
 304         }
 305         else
 306             virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
 307         break;
 308     case VIRTIO_PCI_QUEUE_SEL:
 309         if (val < VIRTIO_PCI_QUEUE_MAX)
 310             vdev->queue_sel = val;
 311         break;
 312     case VIRTIO_PCI_QUEUE_NOTIFY:
 313         if (val < VIRTIO_PCI_QUEUE_MAX) {
 314             virtio_queue_notify(vdev, val);
 315         }
 316         break;
 317     case VIRTIO_PCI_STATUS:
 318         if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
 319             virtio_pci_stop_ioeventfd(proxy);
 320         }
 321
 322         virtio_set_status(vdev, val & 0xFF);
 323
 324         if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
 325             virtio_pci_start_ioeventfd(proxy);
 326         }
 327
 328         if (vdev->status == 0) {
 329             virtio_reset(proxy->vdev);
 330             msix_unuse_all_vectors(&proxy->pci_dev);
 331         }
 332
 333         /* Linux before 2.6.34 sets the device as OK without enabling
 334            the PCI device bus master bit. In this case we need to disable
 335            some safety checks. */
 336         if ((val & VIRTIO_CONFIG_S_DRIVER_OK) &&
 337             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 338             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 339         }
 340         break;
 341     case VIRTIO_MSI_CONFIG_VECTOR:
 342         msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
 343         /* Make it possible for guest to discover an error took place. */
 344         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 345             val = VIRTIO_NO_VECTOR;
 346         vdev->config_vector = val;
 347         break;
 348     case VIRTIO_MSI_QUEUE_VECTOR:
 349         msix_vector_unuse(&proxy->pci_dev,
 350                           virtio_queue_vector(vdev, vdev->queue_sel));
 351         /* Make it possible for guest to discover an error took place. */
 352         if (msix_vector_use(&proxy->pci_dev, val) < 0)
 353             val = VIRTIO_NO_VECTOR;
 354         virtio_queue_set_vector(vdev, vdev->queue_sel, val);
 355         break;
 356     default:
 357         error_report("%s: unexpected address 0x%x value 0x%x",
 358                      __func__, addr, val);
 359         break;
 360     }
 361 }
 362
 363 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
 364 {
 365     VirtIODevice *vdev = proxy->vdev;
 366     uint32_t ret = 0xFFFFFFFF;
 367
 368     switch (addr) {
 369     case VIRTIO_PCI_HOST_FEATURES:
 370         ret = proxy->host_features;
 371         break;
 372     case VIRTIO_PCI_GUEST_FEATURES:
 373         ret = vdev->guest_features;
 374         break;
 375     case VIRTIO_PCI_QUEUE_PFN:
 376         ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
 377               >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
 378         break;
 379     case VIRTIO_PCI_QUEUE_NUM:
 380         ret = virtio_queue_get_num(vdev, vdev->queue_sel);
 381         break;
 382     case VIRTIO_PCI_QUEUE_SEL:
 383         ret = vdev->queue_sel;
 384         break;
 385     case VIRTIO_PCI_STATUS:
 386         ret = vdev->status;
 387         break;
 388     case VIRTIO_PCI_ISR:
 389         /* reading from the ISR also clears it. */
 390         ret = vdev->isr;
 391         vdev->isr = 0;
 392         qemu_set_irq(proxy->pci_dev.irq[0], 0);
 393         break;
 394     case VIRTIO_MSI_CONFIG_VECTOR:
 395         ret = vdev->config_vector;
 396         break;
 397     case VIRTIO_MSI_QUEUE_VECTOR:
 398         ret = virtio_queue_vector(vdev, vdev->queue_sel);
 399         break;
 400     default:
 401         break;
 402     }
 403
 404     return ret;
 405 }
 406
 407 static uint32_t virtio_pci_config_readb(void *opaque, uint32_t addr)
 408 {
 409     VirtIOPCIProxy *proxy = opaque;
 410     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 411     if (addr < config)
 412         return virtio_ioport_read(proxy, addr);
 413     addr -= config;
 414     return virtio_config_readb(proxy->vdev, addr);
 415 }
 416
 417 static uint32_t virtio_pci_config_readw(void *opaque, uint32_t addr)
 418 {
 419     VirtIOPCIProxy *proxy = opaque;
 420     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 421     uint16_t val;
 422     if (addr < config)
 423         return virtio_ioport_read(proxy, addr);
 424     addr -= config;
 425     val = virtio_config_readw(proxy->vdev, addr);
 426     if (virtio_is_big_endian()) {
 427         /*
 428          * virtio is odd, ioports are LE but config space is target native
 429          * endian. However, in qemu, all PIO is LE, so we need to re-swap
 430          * on BE targets
 431          */
 432         val = bswap16(val);
 433     }
 434     return val;
 435 }
 436
 437 static uint32_t virtio_pci_config_readl(void *opaque, uint32_t addr)
 438 {
 439     VirtIOPCIProxy *proxy = opaque;
 440     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 441     uint32_t val;
 442     if (addr < config)
 443         return virtio_ioport_read(proxy, addr);
 444     addr -= config;
 445     val = virtio_config_readl(proxy->vdev, addr);
 446     if (virtio_is_big_endian()) {
 447         val = bswap32(val);
 448     }
 449     return val;
 450 }
 451
 452 static void virtio_pci_config_writeb(void *opaque, uint32_t addr, uint32_t val)
 453 {
 454     VirtIOPCIProxy *proxy = opaque;
 455     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 456     if (addr < config) {
 457         virtio_ioport_write(proxy, addr, val);
 458         return;
 459     }
 460     addr -= config;
 461     virtio_config_writeb(proxy->vdev, addr, val);
 462 }
 463
 464 static void virtio_pci_config_writew(void *opaque, uint32_t addr, uint32_t val)
 465 {
 466     VirtIOPCIProxy *proxy = opaque;
 467     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 468     if (addr < config) {
 469         virtio_ioport_write(proxy, addr, val);
 470         return;
 471     }
 472     addr -= config;
 473     if (virtio_is_big_endian()) {
 474         val = bswap16(val);
 475     }
 476     virtio_config_writew(proxy->vdev, addr, val);
 477 }
 478
 479 static void virtio_pci_config_writel(void *opaque, uint32_t addr, uint32_t val)
 480 {
 481     VirtIOPCIProxy *proxy = opaque;
 482     uint32_t config = VIRTIO_PCI_CONFIG(&proxy->pci_dev);
 483     if (addr < config) {
 484         virtio_ioport_write(proxy, addr, val);
 485         return;
 486     }
 487     addr -= config;
 488     if (virtio_is_big_endian()) {
 489         val = bswap32(val);
 490     }
 491     virtio_config_writel(proxy->vdev, addr, val);
 492 }
 493
 494 static const MemoryRegionPortio virtio_portio[] = {
 495     { 0, 0x10000, 1, .write = virtio_pci_config_writeb, },
 496     { 0, 0x10000, 2, .write = virtio_pci_config_writew, },
 497     { 0, 0x10000, 4, .write = virtio_pci_config_writel, },
 498     { 0, 0x10000, 1, .read = virtio_pci_config_readb, },
 499     { 0, 0x10000, 2, .read = virtio_pci_config_readw, },
 500     { 0, 0x10000, 4, .read = virtio_pci_config_readl, },
 501     PORTIO_END_OF_LIST()
 502 };
 503
 504 static const MemoryRegionOps virtio_pci_config_ops = {
 505     .old_portio = virtio_portio,
 506     .endianness = DEVICE_LITTLE_ENDIAN,
 507 };
 508
 509 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
 510                                 uint32_t val, int len)
 511 {
 512     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 513
 514     pci_default_write_config(pci_dev, address, val, len);
 515
 516     if (range_covers_byte(address, len, PCI_COMMAND) &&
 517         !(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER) &&
 518         !(proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG)) {
 519         virtio_pci_stop_ioeventfd(proxy);
 520         virtio_set_status(proxy->vdev,
 521                           proxy->vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
 522     }
 523
 524     msix_write_config(pci_dev, address, val, len);
 525 }
 526
 527 static unsigned virtio_pci_get_features(void *opaque)
 528 {
 529     VirtIOPCIProxy *proxy = opaque;
 530     return proxy->host_features;
 531 }
 532
 533 static void virtio_pci_guest_notifier_read(void *opaque)
 534 {
 535     VirtQueue *vq = opaque;
 536     EventNotifier *n = virtio_queue_get_guest_notifier(vq);
 537     if (event_notifier_test_and_clear(n)) {
 538         virtio_irq(vq);
 539     }
 540 }
 541
 542 static int virtio_pci_set_guest_notifier(void *opaque, int n, bool assign)
 543 {
 544     VirtIOPCIProxy *proxy = opaque;
 545     VirtQueue *vq = virtio_get_queue(proxy->vdev, n);
 546     EventNotifier *notifier = virtio_queue_get_guest_notifier(vq);
 547
 548     if (assign) {
 549         int r = event_notifier_init(notifier, 0);
 550         if (r < 0) {
 551             return r;
 552         }
 553         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 554                             virtio_pci_guest_notifier_read, NULL, vq);
 555     } else {
 556         qemu_set_fd_handler(event_notifier_get_fd(notifier),
 557                             NULL, NULL, NULL);
 558         event_notifier_cleanup(notifier);
 559     }
 560
 561     return 0;
 562 }
 563
 564 static bool virtio_pci_query_guest_notifiers(void *opaque)
 565 {
 566     VirtIOPCIProxy *proxy = opaque;
 567     return msix_enabled(&proxy->pci_dev);
 568 }
 569
 570 static int virtio_pci_set_guest_notifiers(void *opaque, bool assign)
 571 {
 572     VirtIOPCIProxy *proxy = opaque;
 573     VirtIODevice *vdev = proxy->vdev;
 574     int r, n;
 575
 576     for (n = 0; n < VIRTIO_PCI_QUEUE_MAX; n++) {
 577         if (!virtio_queue_get_num(vdev, n)) {
 578             break;
 579         }
 580
 581         r = virtio_pci_set_guest_notifier(opaque, n, assign);
 582         if (r < 0) {
 583             goto assign_error;
 584         }
 585     }
 586
 587     return 0;
 588
 589 assign_error:
 590     /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
 591     while (--n >= 0) {
 592         virtio_pci_set_guest_notifier(opaque, n, !assign);
 593     }
 594     return r;
 595 }
 596
 597 static int virtio_pci_set_host_notifier(void *opaque, int n, bool assign)
 598 {
 599     VirtIOPCIProxy *proxy = opaque;
 600
 601     /* Stop using ioeventfd for virtqueue kick if the device starts using host
 602      * notifiers.  This makes it easy to avoid stepping on each others' toes.
 603      */
 604     proxy->ioeventfd_disabled = assign;
 605     if (assign) {
 606         virtio_pci_stop_ioeventfd(proxy);
 607     }
 608     /* We don't need to start here: it's not needed because backend
 609      * currently only stops on status change away from ok,
 610      * reset, vmstop and such. If we do add code to start here,
 611      * need to check vmstate, device state etc. */
 612     return virtio_pci_set_host_notifier_internal(proxy, n, assign);
 613 }
 614
 615 static void virtio_pci_vmstate_change(void *opaque, bool running)
 616 {
 617     VirtIOPCIProxy *proxy = opaque;
 618
 619     if (running) {
 620         /* Try to find out if the guest has bus master disabled, but is
 621            in ready state. Then we have a buggy guest OS. */
 622         if ((proxy->vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 623             !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
 624             proxy->flags |= VIRTIO_PCI_FLAG_BUS_MASTER_BUG;
 625         }
 626         virtio_pci_start_ioeventfd(proxy);
 627     } else {
 628         virtio_pci_stop_ioeventfd(proxy);
 629     }
 630 }
 631
 632 static const VirtIOBindings virtio_pci_bindings = {
 633     .notify = virtio_pci_notify,
 634     .save_config = virtio_pci_save_config,
 635     .load_config = virtio_pci_load_config,
 636     .save_queue = virtio_pci_save_queue,
 637     .load_queue = virtio_pci_load_queue,
 638     .get_features = virtio_pci_get_features,
 639     .query_guest_notifiers = virtio_pci_query_guest_notifiers,
 640     .set_host_notifier = virtio_pci_set_host_notifier,
 641     .set_guest_notifiers = virtio_pci_set_guest_notifiers,
 642     .vmstate_change = virtio_pci_vmstate_change,
 643 };
 644
 645 void virtio_init_pci(VirtIOPCIProxy *proxy, VirtIODevice *vdev)
 646 {
 647     uint8_t *config;
 648     uint32_t size;
 649
 650     proxy->vdev = vdev;
 651
 652     config = proxy->pci_dev.config;
 653
 654     if (proxy->class_code) {
 655         pci_config_set_class(config, proxy->class_code);
 656     }
 657     pci_set_word(config + PCI_SUBSYSTEM_VENDOR_ID,
 658                  pci_get_word(config + PCI_VENDOR_ID));
 659     pci_set_word(config + PCI_SUBSYSTEM_ID, vdev->device_id);
 660     config[PCI_INTERRUPT_PIN] = 1;
 661
 662     memory_region_init(&proxy->msix_bar, "virtio-msix", 4096);
 663     if (vdev->nvectors && !msix_init(&proxy->pci_dev, vdev->nvectors,
 664                                      &proxy->msix_bar, 1, 0)) {
 665         pci_register_bar(&proxy->pci_dev, 1, PCI_BASE_ADDRESS_SPACE_MEMORY,
 666                          &proxy->msix_bar);
 667     } else
 668         vdev->nvectors = 0;
 669
 670     proxy->pci_dev.config_write = virtio_write_config;
 671
 672     size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev) + vdev->config_len;
 673     if (size & (size-1))
 674         size = 1 << qemu_fls(size);
 675
 676     memory_region_init_io(&proxy->bar, &virtio_pci_config_ops, proxy,
 677                           "virtio-pci", size);
 678     pci_register_bar(&proxy->pci_dev, 0, PCI_BASE_ADDRESS_SPACE_IO,
 679                      &proxy->bar);
 680
 681     if (!kvm_has_many_ioeventfds()) {
 682         proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
 683     }
 684
 685     virtio_bind_device(vdev, &virtio_pci_bindings, proxy);
 686     proxy->host_features |= 0x1 << VIRTIO_F_NOTIFY_ON_EMPTY;
 687     proxy->host_features |= 0x1 << VIRTIO_F_BAD_FEATURE;
 688     proxy->host_features = vdev->get_features(vdev, proxy->host_features);
 689 }
 690
 691 static int virtio_blk_init_pci(PCIDevice *pci_dev)
 692 {
 693     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 694     VirtIODevice *vdev;
 695
 696     if (proxy->class_code != PCI_CLASS_STORAGE_SCSI &&
 697         proxy->class_code != PCI_CLASS_STORAGE_OTHER)
 698         proxy->class_code = PCI_CLASS_STORAGE_SCSI;
 699
 700     vdev = virtio_blk_init(&pci_dev->qdev, &proxy->block,
 701                            &proxy->block_serial);
 702     if (!vdev) {
 703         return -1;
 704     }
 705     vdev->nvectors = proxy->nvectors;
 706     virtio_init_pci(proxy, vdev);
 707     /* make the actual value visible */
 708     proxy->nvectors = vdev->nvectors;
 709     return 0;
 710 }
 711
 712 static int virtio_exit_pci(PCIDevice *pci_dev)
 713 {
 714     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 715     int r;
 716
 717     memory_region_destroy(&proxy->bar);
 718     r = msix_uninit(pci_dev, &proxy->msix_bar);
 719     memory_region_destroy(&proxy->msix_bar);
 720     return r;
 721 }
 722
 723 static int virtio_blk_exit_pci(PCIDevice *pci_dev)
 724 {
 725     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 726
 727     virtio_pci_stop_ioeventfd(proxy);
 728     virtio_blk_exit(proxy->vdev);
 729     blockdev_mark_auto_del(proxy->block.bs);
 730     return virtio_exit_pci(pci_dev);
 731 }
 732
 733 static int virtio_serial_init_pci(PCIDevice *pci_dev)
 734 {
 735     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 736     VirtIODevice *vdev;
 737
 738     if (proxy->class_code != PCI_CLASS_COMMUNICATION_OTHER &&
 739         proxy->class_code != PCI_CLASS_DISPLAY_OTHER && /* qemu 0.10 */
 740         proxy->class_code != PCI_CLASS_OTHERS)          /* qemu-kvm  */
 741         proxy->class_code = PCI_CLASS_COMMUNICATION_OTHER;
 742
 743     vdev = virtio_serial_init(&pci_dev->qdev, &proxy->serial);
 744     if (!vdev) {
 745         return -1;
 746     }
 747     vdev->nvectors = proxy->nvectors == DEV_NVECTORS_UNSPECIFIED
 748                                         ? proxy->serial.max_virtserial_ports + 1
 749                                         : proxy->nvectors;
 750     virtio_init_pci(proxy, vdev);
 751     proxy->nvectors = vdev->nvectors;
 752     return 0;
 753 }
 754
 755 static int virtio_serial_exit_pci(PCIDevice *pci_dev)
 756 {
 757     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 758
 759     virtio_pci_stop_ioeventfd(proxy);
 760     virtio_serial_exit(proxy->vdev);
 761     return virtio_exit_pci(pci_dev);
 762 }
 763
 764 static int virtio_net_init_pci(PCIDevice *pci_dev)
 765 {
 766     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 767     VirtIODevice *vdev;
 768
 769     vdev = virtio_net_init(&pci_dev->qdev, &proxy->nic, &proxy->net);
 770
 771     vdev->nvectors = proxy->nvectors;
 772     virtio_init_pci(proxy, vdev);
 773
 774     /* make the actual value visible */
 775     proxy->nvectors = vdev->nvectors;
 776     return 0;
 777 }
 778
 779 static int virtio_net_exit_pci(PCIDevice *pci_dev)
 780 {
 781     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 782
 783     virtio_pci_stop_ioeventfd(proxy);
 784     virtio_net_exit(proxy->vdev);
 785     return virtio_exit_pci(pci_dev);
 786 }
 787
 788 static int virtio_balloon_init_pci(PCIDevice *pci_dev)
 789 {
 790     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 791     VirtIODevice *vdev;
 792
 793     if (proxy->class_code != PCI_CLASS_OTHERS &&
 794         proxy->class_code != PCI_CLASS_MEMORY_RAM) { /* qemu < 1.1 */
 795         proxy->class_code = PCI_CLASS_OTHERS;
 796     }
 797
 798     vdev = virtio_balloon_init(&pci_dev->qdev);
 799     if (!vdev) {
 800         return -1;
 801     }
 802     virtio_init_pci(proxy, vdev);
 803     return 0;
 804 }
 805
 806 static int virtio_balloon_exit_pci(PCIDevice *pci_dev)
 807 {
 808     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 809
 810     virtio_pci_stop_ioeventfd(proxy);
 811     virtio_balloon_exit(proxy->vdev);
 812     return virtio_exit_pci(pci_dev);
 813 }
 814
 815 static Property virtio_blk_properties[] = {
 816     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 817     DEFINE_BLOCK_PROPERTIES(VirtIOPCIProxy, block),
 818     DEFINE_PROP_STRING("serial", VirtIOPCIProxy, block_serial),
 819     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 820     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
 821     DEFINE_VIRTIO_BLK_FEATURES(VirtIOPCIProxy, host_features),
 822     DEFINE_PROP_END_OF_LIST(),
 823 };
 824
 825 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 826 {
 827     DeviceClass *dc = DEVICE_CLASS(klass);
 828     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 829
 830     k->init = virtio_blk_init_pci;
 831     k->exit = virtio_blk_exit_pci;
 832     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 833     k->device_id = PCI_DEVICE_ID_VIRTIO_BLOCK;
 834     k->revision = VIRTIO_PCI_ABI_VERSION;
 835     k->class_id = PCI_CLASS_STORAGE_SCSI;
 836     dc->reset = virtio_pci_reset;
 837     dc->props = virtio_blk_properties;
 838 }
 839
 840 static TypeInfo virtio_blk_info = {
 841     .name          = "virtio-blk-pci",
 842     .parent        = TYPE_PCI_DEVICE,
 843     .instance_size = sizeof(VirtIOPCIProxy),
 844     .class_init    = virtio_blk_class_init,
 845 };
 846
 847 static Property virtio_net_properties[] = {
 848     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
 849     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
 850     DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
 851     DEFINE_NIC_PROPERTIES(VirtIOPCIProxy, nic),
 852     DEFINE_PROP_UINT32("x-txtimer", VirtIOPCIProxy, net.txtimer, TX_TIMER_INTERVAL),
 853     DEFINE_PROP_INT32("x-txburst", VirtIOPCIProxy, net.txburst, TX_BURST),
 854     DEFINE_PROP_STRING("tx", VirtIOPCIProxy, net.tx),
 855     DEFINE_PROP_END_OF_LIST(),
 856 };
 857
 858 static void virtio_net_class_init(ObjectClass *klass, void *data)
 859 {
 860     DeviceClass *dc = DEVICE_CLASS(klass);
 861     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 862
 863     k->init = virtio_net_init_pci;
 864     k->exit = virtio_net_exit_pci;
 865     k->romfile = "pxe-virtio.rom";
 866     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 867     k->device_id = PCI_DEVICE_ID_VIRTIO_NET;
 868     k->revision = VIRTIO_PCI_ABI_VERSION;
 869     k->class_id = PCI_CLASS_NETWORK_ETHERNET;
 870     dc->reset = virtio_pci_reset;
 871     dc->props = virtio_net_properties;
 872 }
 873
 874 static TypeInfo virtio_net_info = {
 875     .name          = "virtio-net-pci",
 876     .parent        = TYPE_PCI_DEVICE,
 877     .instance_size = sizeof(VirtIOPCIProxy),
 878     .class_init    = virtio_net_class_init,
 879 };
 880
 881 static Property virtio_serial_properties[] = {
 882     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags, VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, true),
 883     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, DEV_NVECTORS_UNSPECIFIED),
 884     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 885     DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 886     DEFINE_PROP_UINT32("max_ports", VirtIOPCIProxy, serial.max_virtserial_ports, 31),
 887     DEFINE_PROP_END_OF_LIST(),
 888 };
 889
 890 static void virtio_serial_class_init(ObjectClass *klass, void *data)
 891 {
 892     DeviceClass *dc = DEVICE_CLASS(klass);
 893     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 894
 895     k->init = virtio_serial_init_pci;
 896     k->exit = virtio_serial_exit_pci;
 897     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 898     k->device_id = PCI_DEVICE_ID_VIRTIO_CONSOLE;
 899     k->revision = VIRTIO_PCI_ABI_VERSION;
 900     k->class_id = PCI_CLASS_COMMUNICATION_OTHER;
 901     dc->reset = virtio_pci_reset;
 902     dc->props = virtio_serial_properties;
 903 }
 904
 905 static TypeInfo virtio_serial_info = {
 906     .name          = "virtio-serial-pci",
 907     .parent        = TYPE_PCI_DEVICE,
 908     .instance_size = sizeof(VirtIOPCIProxy),
 909     .class_init    = virtio_serial_class_init,
 910 };
 911
 912 static Property virtio_balloon_properties[] = {
 913     DEFINE_VIRTIO_COMMON_FEATURES(VirtIOPCIProxy, host_features),
 914     DEFINE_PROP_HEX32("class", VirtIOPCIProxy, class_code, 0),
 915     DEFINE_PROP_END_OF_LIST(),
 916 };
 917
 918 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
 919 {
 920     DeviceClass *dc = DEVICE_CLASS(klass);
 921     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 922
 923     k->init = virtio_balloon_init_pci;
 924     k->exit = virtio_balloon_exit_pci;
 925     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 926     k->device_id = PCI_DEVICE_ID_VIRTIO_BALLOON;
 927     k->revision = VIRTIO_PCI_ABI_VERSION;
 928     k->class_id = PCI_CLASS_OTHERS;
 929     dc->reset = virtio_pci_reset;
 930     dc->props = virtio_balloon_properties;
 931 }
 932
 933 static TypeInfo virtio_balloon_info = {
 934     .name          = "virtio-balloon-pci",
 935     .parent        = TYPE_PCI_DEVICE,
 936     .instance_size = sizeof(VirtIOPCIProxy),
 937     .class_init    = virtio_balloon_class_init,
 938 };
 939
 940 static int virtio_scsi_init_pci(PCIDevice *pci_dev)
 941 {
 942     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 943     VirtIODevice *vdev;
 944
 945     vdev = virtio_scsi_init(&pci_dev->qdev, &proxy->scsi);
 946     if (!vdev) {
 947         return -EINVAL;
 948     }
 949
 950     vdev->nvectors = proxy->nvectors;
 951     virtio_init_pci(proxy, vdev);
 952
 953     /* make the actual value visible */
 954     proxy->nvectors = vdev->nvectors;
 955     return 0;
 956 }
 957
 958 static int virtio_scsi_exit_pci(PCIDevice *pci_dev)
 959 {
 960     VirtIOPCIProxy *proxy = DO_UPCAST(VirtIOPCIProxy, pci_dev, pci_dev);
 961
 962     virtio_scsi_exit(proxy->vdev);
 963     return virtio_exit_pci(pci_dev);
 964 }
 965
 966 static Property virtio_scsi_properties[] = {
 967     DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 2),
 968     DEFINE_VIRTIO_SCSI_PROPERTIES(VirtIOPCIProxy, host_features, scsi),
 969     DEFINE_PROP_END_OF_LIST(),
 970 };
 971
 972 static void virtio_scsi_class_init(ObjectClass *klass, void *data)
 973 {
 974     DeviceClass *dc = DEVICE_CLASS(klass);
 975     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
 976
 977     k->init = virtio_scsi_init_pci;
 978     k->exit = virtio_scsi_exit_pci;
 979     k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
 980     k->device_id = PCI_DEVICE_ID_VIRTIO_SCSI;
 981     k->revision = 0x00;
 982     k->class_id = PCI_CLASS_STORAGE_SCSI;
 983     dc->reset = virtio_pci_reset;
 984     dc->props = virtio_scsi_properties;
 985 }
 986
 987 static TypeInfo virtio_scsi_info = {
 988     .name          = "virtio-scsi-pci",
 989     .parent        = TYPE_PCI_DEVICE,
 990     .instance_size = sizeof(VirtIOPCIProxy),
 991     .class_init    = virtio_scsi_class_init,
 992 };
 993
 994 static void virtio_pci_register_types(void)
 995 {
 996     type_register_static(&virtio_blk_info);
 997     type_register_static(&virtio_net_info);
 998     type_register_static(&virtio_serial_info);
 999     type_register_static(&virtio_balloon_info);
1000     type_register_static(&virtio_scsi_info);
1001 }
1002
1003 type_init(virtio_pci_register_types)