hw/vfio/platform.c

   1 /*
   2  * vfio based device assignment support - platform devices
   3  *
   4  * Copyright Linaro Limited, 2014
   5  *
   6  * Authors:
   7  *  Kim Phillips <kim.phillips@linaro.org>
   8  *  Eric Auger <eric.auger@linaro.org>
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2.  See
  11  * the COPYING file in the top-level directory.
  12  *
  13  * Based on vfio based PCI device assignment support:
  14  *  Copyright Red Hat, Inc. 2012
  15  */
  16
  17 #include "qemu/osdep.h"
  18 #include "qapi/error.h"
  19 #include <sys/ioctl.h>
  20 #include <linux/vfio.h>
  21
  22 #include "hw/vfio/vfio-platform.h"
  23 #include "qemu/error-report.h"
  24 #include "qemu/range.h"
  25 #include "sysemu/sysemu.h"
  26 #include "exec/memory.h"
  27 #include "exec/address-spaces.h"
  28 #include "qemu/queue.h"
  29 #include "hw/sysbus.h"
  30 #include "trace.h"
  31 #include "hw/platform-bus.h"
  32 #include "sysemu/kvm.h"
  33
  34 /*
  35  * Functions used whatever the injection method
  36  */
  37
  38 static inline bool vfio_irq_is_automasked(VFIOINTp *intp)
  39 {
  40     return intp->flags & VFIO_IRQ_INFO_AUTOMASKED;
  41 }
  42
  43 /**
  44  * vfio_init_intp - allocate, initialize the IRQ struct pointer
  45  * and add it into the list of IRQs
  46  * @vbasedev: the VFIO device handle
  47  * @info: irq info struct retrieved from VFIO driver
  48  * @errp: error object
  49  */
  50 static VFIOINTp *vfio_init_intp(VFIODevice *vbasedev,
  51                                 struct vfio_irq_info info, Error **errp)
  52 {
  53     int ret;
  54     VFIOPlatformDevice *vdev =
  55         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
  56     SysBusDevice *sbdev = SYS_BUS_DEVICE(vdev);
  57     VFIOINTp *intp;
  58
  59     intp = g_malloc0(sizeof(*intp));
  60     intp->vdev = vdev;
  61     intp->pin = info.index;
  62     intp->flags = info.flags;
  63     intp->state = VFIO_IRQ_INACTIVE;
  64     intp->kvm_accel = false;
  65
  66     sysbus_init_irq(sbdev, &intp->qemuirq);
  67
  68     /* Get an eventfd for trigger */
  69     intp->interrupt = g_malloc0(sizeof(EventNotifier));
  70     ret = event_notifier_init(intp->interrupt, 0);
  71     if (ret) {
  72         g_free(intp->interrupt);
  73         g_free(intp);
  74         error_setg_errno(errp, -ret,
  75                          "failed to initialize trigger eventd notifier");
  76         return NULL;
  77     }
  78     if (vfio_irq_is_automasked(intp)) {
  79         /* Get an eventfd for resample/unmask */
  80         intp->unmask = g_malloc0(sizeof(EventNotifier));
  81         ret = event_notifier_init(intp->unmask, 0);
  82         if (ret) {
  83             g_free(intp->interrupt);
  84             g_free(intp->unmask);
  85             g_free(intp);
  86             error_setg_errno(errp, -ret,
  87                              "failed to initialize resample eventd notifier");
  88             return NULL;
  89         }
  90     }
  91
  92     QLIST_INSERT_HEAD(&vdev->intp_list, intp, next);
  93     return intp;
  94 }
  95
  96 /**
  97  * vfio_set_trigger_eventfd - set VFIO eventfd handling
  98  *
  99  * @intp: IRQ struct handle
 100  * @handler: handler to be called on eventfd signaling
 101  *
 102  * Setup VFIO signaling and attach an optional user-side handler
 103  * to the eventfd
 104  */
 105 static int vfio_set_trigger_eventfd(VFIOINTp *intp,
 106                                     eventfd_user_side_handler_t handler)
 107 {
 108     VFIODevice *vbasedev = &intp->vdev->vbasedev;
 109     struct vfio_irq_set *irq_set;
 110     int argsz, ret;
 111     int32_t *pfd;
 112
 113     argsz = sizeof(*irq_set) + sizeof(*pfd);
 114     irq_set = g_malloc0(argsz);
 115     irq_set->argsz = argsz;
 116     irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
 117     irq_set->index = intp->pin;
 118     irq_set->start = 0;
 119     irq_set->count = 1;
 120     pfd = (int32_t *)&irq_set->data;
 121     *pfd = event_notifier_get_fd(intp->interrupt);
 122     qemu_set_fd_handler(*pfd, (IOHandler *)handler, NULL, intp);
 123     ret = ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, irq_set);
 124     if (ret < 0) {
 125         error_report("vfio: Failed to set trigger eventfd: %m");
 126         qemu_set_fd_handler(*pfd, NULL, NULL, NULL);
 127     }
 128     g_free(irq_set);
 129     return ret;
 130 }
 131
 132 /*
 133  * Functions only used when eventfds are handled on user-side
 134  * ie. without irqfd
 135  */
 136
 137 /**
 138  * vfio_mmap_set_enabled - enable/disable the fast path mode
 139  * @vdev: the VFIO platform device
 140  * @enabled: the target mmap state
 141  *
 142  * enabled = true ~ fast path = MMIO region is mmaped (no KVM TRAP);
 143  * enabled = false ~ slow path = MMIO region is trapped and region callbacks
 144  * are called; slow path enables to trap the device IRQ status register reset
 145 */
 146
 147 static void vfio_mmap_set_enabled(VFIOPlatformDevice *vdev, bool enabled)
 148 {
 149     int i;
 150
 151     for (i = 0; i < vdev->vbasedev.num_regions; i++) {
 152         vfio_region_mmaps_set_enabled(vdev->regions[i], enabled);
 153     }
 154 }
 155
 156 /**
 157  * vfio_intp_mmap_enable - timer function, restores the fast path
 158  * if there is no more active IRQ
 159  * @opaque: actually points to the VFIO platform device
 160  *
 161  * Called on mmap timer timout, this function checks whether the
 162  * IRQ is still active and if not, restores the fast path.
 163  * by construction a single eventfd is handled at a time.
 164  * if the IRQ is still active, the timer is re-programmed.
 165  */
 166 static void vfio_intp_mmap_enable(void *opaque)
 167 {
 168     VFIOINTp *tmp;
 169     VFIOPlatformDevice *vdev = (VFIOPlatformDevice *)opaque;
 170
 171     qemu_mutex_lock(&vdev->intp_mutex);
 172     QLIST_FOREACH(tmp, &vdev->intp_list, next) {
 173         if (tmp->state == VFIO_IRQ_ACTIVE) {
 174             trace_vfio_platform_intp_mmap_enable(tmp->pin);
 175             /* re-program the timer to check active status later */
 176             timer_mod(vdev->mmap_timer,
 177                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 178                           vdev->mmap_timeout);
 179             qemu_mutex_unlock(&vdev->intp_mutex);
 180             return;
 181         }
 182     }
 183     vfio_mmap_set_enabled(vdev, true);
 184     qemu_mutex_unlock(&vdev->intp_mutex);
 185 }
 186
 187 /**
 188  * vfio_intp_inject_pending_lockheld - Injects a pending IRQ
 189  * @opaque: opaque pointer, in practice the VFIOINTp handle
 190  *
 191  * The function is called on a previous IRQ completion, from
 192  * vfio_platform_eoi, while the intp_mutex is locked.
 193  * Also in such situation, the slow path already is set and
 194  * the mmap timer was already programmed.
 195  */
 196 static void vfio_intp_inject_pending_lockheld(VFIOINTp *intp)
 197 {
 198     trace_vfio_platform_intp_inject_pending_lockheld(intp->pin,
 199                               event_notifier_get_fd(intp->interrupt));
 200
 201     intp->state = VFIO_IRQ_ACTIVE;
 202
 203     /* trigger the virtual IRQ */
 204     qemu_set_irq(intp->qemuirq, 1);
 205 }
 206
 207 /**
 208  * vfio_intp_interrupt - The user-side eventfd handler
 209  * @opaque: opaque pointer which in practice is the VFIOINTp handle
 210  *
 211  * the function is entered in event handler context:
 212  * the vIRQ is injected into the guest if there is no other active
 213  * or pending IRQ.
 214  */
 215 static void vfio_intp_interrupt(VFIOINTp *intp)
 216 {
 217     int ret;
 218     VFIOINTp *tmp;
 219     VFIOPlatformDevice *vdev = intp->vdev;
 220     bool delay_handling = false;
 221
 222     qemu_mutex_lock(&vdev->intp_mutex);
 223     if (intp->state == VFIO_IRQ_INACTIVE) {
 224         QLIST_FOREACH(tmp, &vdev->intp_list, next) {
 225             if (tmp->state == VFIO_IRQ_ACTIVE ||
 226                 tmp->state == VFIO_IRQ_PENDING) {
 227                 delay_handling = true;
 228                 break;
 229             }
 230         }
 231     }
 232     if (delay_handling) {
 233         /*
 234          * the new IRQ gets a pending status and is pushed in
 235          * the pending queue
 236          */
 237         intp->state = VFIO_IRQ_PENDING;
 238         trace_vfio_intp_interrupt_set_pending(intp->pin);
 239         QSIMPLEQ_INSERT_TAIL(&vdev->pending_intp_queue,
 240                              intp, pqnext);
 241         ret = event_notifier_test_and_clear(intp->interrupt);
 242         qemu_mutex_unlock(&vdev->intp_mutex);
 243         return;
 244     }
 245
 246     trace_vfio_platform_intp_interrupt(intp->pin,
 247                               event_notifier_get_fd(intp->interrupt));
 248
 249     ret = event_notifier_test_and_clear(intp->interrupt);
 250     if (!ret) {
 251         error_report("Error when clearing fd=%d (ret = %d)",
 252                      event_notifier_get_fd(intp->interrupt), ret);
 253     }
 254
 255     intp->state = VFIO_IRQ_ACTIVE;
 256
 257     /* sets slow path */
 258     vfio_mmap_set_enabled(vdev, false);
 259
 260     /* trigger the virtual IRQ */
 261     qemu_set_irq(intp->qemuirq, 1);
 262
 263     /*
 264      * Schedule the mmap timer which will restore fastpath when no IRQ
 265      * is active anymore
 266      */
 267     if (vdev->mmap_timeout) {
 268         timer_mod(vdev->mmap_timer,
 269                   qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 270                       vdev->mmap_timeout);
 271     }
 272     qemu_mutex_unlock(&vdev->intp_mutex);
 273 }
 274
 275 /**
 276  * vfio_platform_eoi - IRQ completion routine
 277  * @vbasedev: the VFIO device handle
 278  *
 279  * De-asserts the active virtual IRQ and unmasks the physical IRQ
 280  * (effective for level sensitive IRQ auto-masked by the  VFIO driver).
 281  * Then it handles next pending IRQ if any.
 282  * eoi function is called on the first access to any MMIO region
 283  * after an IRQ was triggered, trapped since slow path was set.
 284  * It is assumed this access corresponds to the IRQ status
 285  * register reset. With such a mechanism, a single IRQ can be
 286  * handled at a time since there is no way to know which IRQ
 287  * was completed by the guest (we would need additional details
 288  * about the IRQ status register mask).
 289  */
 290 static void vfio_platform_eoi(VFIODevice *vbasedev)
 291 {
 292     VFIOINTp *intp;
 293     VFIOPlatformDevice *vdev =
 294         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
 295
 296     qemu_mutex_lock(&vdev->intp_mutex);
 297     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 298         if (intp->state == VFIO_IRQ_ACTIVE) {
 299             trace_vfio_platform_eoi(intp->pin,
 300                                 event_notifier_get_fd(intp->interrupt));
 301             intp->state = VFIO_IRQ_INACTIVE;
 302
 303             /* deassert the virtual IRQ */
 304             qemu_set_irq(intp->qemuirq, 0);
 305
 306             if (vfio_irq_is_automasked(intp)) {
 307                 /* unmasks the physical level-sensitive IRQ */
 308                 vfio_unmask_single_irqindex(vbasedev, intp->pin);
 309             }
 310
 311             /* a single IRQ can be active at a time */
 312             break;
 313         }
 314     }
 315     /* in case there are pending IRQs, handle the first one */
 316     if (!QSIMPLEQ_EMPTY(&vdev->pending_intp_queue)) {
 317         intp = QSIMPLEQ_FIRST(&vdev->pending_intp_queue);
 318         vfio_intp_inject_pending_lockheld(intp);
 319         QSIMPLEQ_REMOVE_HEAD(&vdev->pending_intp_queue, pqnext);
 320     }
 321     qemu_mutex_unlock(&vdev->intp_mutex);
 322 }
 323
 324 /**
 325  * vfio_start_eventfd_injection - starts the virtual IRQ injection using
 326  * user-side handled eventfds
 327  * @sbdev: the sysbus device handle
 328  * @irq: the qemu irq handle
 329  */
 330
 331 static void vfio_start_eventfd_injection(SysBusDevice *sbdev, qemu_irq irq)
 332 {
 333     int ret;
 334     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(sbdev);
 335     VFIOINTp *intp;
 336
 337     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 338         if (intp->qemuirq == irq) {
 339             break;
 340         }
 341     }
 342     assert(intp);
 343
 344     ret = vfio_set_trigger_eventfd(intp, vfio_intp_interrupt);
 345     if (ret) {
 346         error_report("vfio: failed to start eventfd signaling for IRQ %d: %m",
 347                      intp->pin);
 348         abort();
 349     }
 350 }
 351
 352 /*
 353  * Functions used for irqfd
 354  */
 355
 356 /**
 357  * vfio_set_resample_eventfd - sets the resamplefd for an IRQ
 358  * @intp: the IRQ struct handle
 359  * programs the VFIO driver to unmask this IRQ when the
 360  * intp->unmask eventfd is triggered
 361  */
 362 static int vfio_set_resample_eventfd(VFIOINTp *intp)
 363 {
 364     VFIODevice *vbasedev = &intp->vdev->vbasedev;
 365     struct vfio_irq_set *irq_set;
 366     int argsz, ret;
 367     int32_t *pfd;
 368
 369     argsz = sizeof(*irq_set) + sizeof(*pfd);
 370     irq_set = g_malloc0(argsz);
 371     irq_set->argsz = argsz;
 372     irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_UNMASK;
 373     irq_set->index = intp->pin;
 374     irq_set->start = 0;
 375     irq_set->count = 1;
 376     pfd = (int32_t *)&irq_set->data;
 377     *pfd = event_notifier_get_fd(intp->unmask);
 378     qemu_set_fd_handler(*pfd, NULL, NULL, NULL);
 379     ret = ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, irq_set);
 380     g_free(irq_set);
 381     if (ret < 0) {
 382         error_report("vfio: Failed to set resample eventfd: %m");
 383     }
 384     return ret;
 385 }
 386
 387 /**
 388  * vfio_start_irqfd_injection - starts the virtual IRQ injection using
 389  * irqfd
 390  *
 391  * @sbdev: the sysbus device handle
 392  * @irq: the qemu irq handle
 393  *
 394  * In case the irqfd setup fails, we fallback to userspace handled eventfd
 395  */
 396 static void vfio_start_irqfd_injection(SysBusDevice *sbdev, qemu_irq irq)
 397 {
 398     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(sbdev);
 399     VFIOINTp *intp;
 400
 401     if (!kvm_irqfds_enabled() || !kvm_resamplefds_enabled() ||
 402         !vdev->irqfd_allowed) {
 403         goto fail_irqfd;
 404     }
 405
 406     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 407         if (intp->qemuirq == irq) {
 408             break;
 409         }
 410     }
 411     assert(intp);
 412
 413     if (kvm_irqchip_add_irqfd_notifier(kvm_state, intp->interrupt,
 414                                    intp->unmask, irq) < 0) {
 415         goto fail_irqfd;
 416     }
 417
 418     if (vfio_set_trigger_eventfd(intp, NULL) < 0) {
 419         goto fail_vfio;
 420     }
 421     if (vfio_irq_is_automasked(intp)) {
 422         if (vfio_set_resample_eventfd(intp) < 0) {
 423             goto fail_vfio;
 424         }
 425         trace_vfio_platform_start_level_irqfd_injection(intp->pin,
 426                                     event_notifier_get_fd(intp->interrupt),
 427                                     event_notifier_get_fd(intp->unmask));
 428     } else {
 429         trace_vfio_platform_start_edge_irqfd_injection(intp->pin,
 430                                     event_notifier_get_fd(intp->interrupt));
 431     }
 432
 433     intp->kvm_accel = true;
 434
 435     return;
 436 fail_vfio:
 437     kvm_irqchip_remove_irqfd_notifier(kvm_state, intp->interrupt, irq);
 438     error_report("vfio: failed to start eventfd signaling for IRQ %d: %m",
 439                  intp->pin);
 440     abort();
 441 fail_irqfd:
 442     vfio_start_eventfd_injection(sbdev, irq);
 443     return;
 444 }
 445
 446 /* VFIO skeleton */
 447
 448 static void vfio_platform_compute_needs_reset(VFIODevice *vbasedev)
 449 {
 450     vbasedev->needs_reset = true;
 451 }
 452
 453 /* not implemented yet */
 454 static int vfio_platform_hot_reset_multi(VFIODevice *vbasedev)
 455 {
 456     return -1;
 457 }
 458
 459 /**
 460  * vfio_populate_device - Allocate and populate MMIO region
 461  * and IRQ structs according to driver returned information
 462  * @vbasedev: the VFIO device handle
 463  * @errp: error object
 464  *
 465  */
 466 static int vfio_populate_device(VFIODevice *vbasedev, Error **errp)
 467 {
 468     VFIOINTp *intp, *tmp;
 469     int i, ret = -1;
 470     VFIOPlatformDevice *vdev =
 471         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
 472
 473     if (!(vbasedev->flags & VFIO_DEVICE_FLAGS_PLATFORM)) {
 474         error_setg(errp, "this isn't a platform device");
 475         return ret;
 476     }
 477
 478     vdev->regions = g_new0(VFIORegion *, vbasedev->num_regions);
 479
 480     for (i = 0; i < vbasedev->num_regions; i++) {
 481         char *name = g_strdup_printf("VFIO %s region %d\n", vbasedev->name, i);
 482
 483         vdev->regions[i] = g_new0(VFIORegion, 1);
 484         ret = vfio_region_setup(OBJECT(vdev), vbasedev,
 485                                 vdev->regions[i], i, name);
 486         g_free(name);
 487         if (ret) {
 488             error_setg_errno(errp, -ret, "failed to get region %d info", i);
 489             goto reg_error;
 490         }
 491     }
 492
 493     vdev->mmap_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
 494                                     vfio_intp_mmap_enable, vdev);
 495
 496     QSIMPLEQ_INIT(&vdev->pending_intp_queue);
 497
 498     for (i = 0; i < vbasedev->num_irqs; i++) {
 499         struct vfio_irq_info irq = { .argsz = sizeof(irq) };
 500
 501         irq.index = i;
 502         ret = ioctl(vbasedev->fd, VFIO_DEVICE_GET_IRQ_INFO, &irq);
 503         if (ret) {
 504             error_setg_errno(errp, -ret, "failed to get device irq info");
 505             goto irq_err;
 506         } else {
 507             trace_vfio_platform_populate_interrupts(irq.index,
 508                                                     irq.count,
 509                                                     irq.flags);
 510             intp = vfio_init_intp(vbasedev, irq, errp);
 511             if (!intp) {
 512                 ret = -1;
 513                 goto irq_err;
 514             }
 515         }
 516     }
 517     return 0;
 518 irq_err:
 519     timer_del(vdev->mmap_timer);
 520     QLIST_FOREACH_SAFE(intp, &vdev->intp_list, next, tmp) {
 521         QLIST_REMOVE(intp, next);
 522         g_free(intp);
 523     }
 524 reg_error:
 525     for (i = 0; i < vbasedev->num_regions; i++) {
 526         if (vdev->regions[i]) {
 527             vfio_region_finalize(vdev->regions[i]);
 528         }
 529         g_free(vdev->regions[i]);
 530     }
 531     g_free(vdev->regions);
 532     return ret;
 533 }
 534
 535 /* specialized functions for VFIO Platform devices */
 536 static VFIODeviceOps vfio_platform_ops = {
 537     .vfio_compute_needs_reset = vfio_platform_compute_needs_reset,
 538     .vfio_hot_reset_multi = vfio_platform_hot_reset_multi,
 539     .vfio_eoi = vfio_platform_eoi,
 540 };
 541
 542 /**
 543  * vfio_base_device_init - perform preliminary VFIO setup
 544  * @vbasedev: the VFIO device handle
 545  * @errp: error object
 546  *
 547  * Implement the VFIO command sequence that allows to discover
 548  * assigned device resources: group extraction, device
 549  * fd retrieval, resource query.
 550  * Precondition: the device name must be initialized
 551  */
 552 static int vfio_base_device_init(VFIODevice *vbasedev, Error **errp)
 553 {
 554     VFIOGroup *group;
 555     VFIODevice *vbasedev_iter;
 556     char *tmp, group_path[PATH_MAX], *group_name;
 557     ssize_t len;
 558     struct stat st;
 559     int groupid;
 560     int ret;
 561
 562     /* @sysfsdev takes precedence over @host */
 563     if (vbasedev->sysfsdev) {
 564         g_free(vbasedev->name);
 565         vbasedev->name = g_path_get_basename(vbasedev->sysfsdev);
 566     } else {
 567         if (!vbasedev->name || strchr(vbasedev->name, '/')) {
 568             error_setg(errp, "wrong host device name");
 569             return -EINVAL;
 570         }
 571
 572         vbasedev->sysfsdev = g_strdup_printf("/sys/bus/platform/devices/%s",
 573                                              vbasedev->name);
 574     }
 575
 576     if (stat(vbasedev->sysfsdev, &st) < 0) {
 577         error_setg_errno(errp, errno,
 578                          "failed to get the sysfs host device file status");
 579         return -errno;
 580     }
 581
 582     tmp = g_strdup_printf("%s/iommu_group", vbasedev->sysfsdev);
 583     len = readlink(tmp, group_path, sizeof(group_path));
 584     g_free(tmp);
 585
 586     if (len < 0 || len >= sizeof(group_path)) {
 587         ret = len < 0 ? -errno : -ENAMETOOLONG;
 588         error_setg_errno(errp, -ret, "no iommu_group found");
 589         return ret;
 590     }
 591
 592     group_path[len] = 0;
 593
 594     group_name = basename(group_path);
 595     if (sscanf(group_name, "%d", &groupid) != 1) {
 596         error_setg_errno(errp, errno, "failed to read %s", group_path);
 597         return -errno;
 598     }
 599
 600     trace_vfio_platform_base_device_init(vbasedev->name, groupid);
 601
 602     group = vfio_get_group(groupid, &address_space_memory, errp);
 603     if (!group) {
 604         return -ENOENT;
 605     }
 606
 607     QLIST_FOREACH(vbasedev_iter, &group->device_list, next) {
 608         if (strcmp(vbasedev_iter->name, vbasedev->name) == 0) {
 609             error_setg(errp, "device is already attached");
 610             vfio_put_group(group);
 611             return -EBUSY;
 612         }
 613     }
 614     ret = vfio_get_device(group, vbasedev->name, vbasedev, errp);
 615     if (ret) {
 616         vfio_put_group(group);
 617         return ret;
 618     }
 619
 620     ret = vfio_populate_device(vbasedev, errp);
 621     if (ret) {
 622         vfio_put_group(group);
 623     }
 624
 625     return ret;
 626 }
 627
 628 /**
 629  * vfio_platform_realize  - the device realize function
 630  * @dev: device state pointer
 631  * @errp: error
 632  *
 633  * initialize the device, its memory regions and IRQ structures
 634  * IRQ are started separately
 635  */
 636 static void vfio_platform_realize(DeviceState *dev, Error **errp)
 637 {
 638     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(dev);
 639     SysBusDevice *sbdev = SYS_BUS_DEVICE(dev);
 640     VFIODevice *vbasedev = &vdev->vbasedev;
 641     int i, ret;
 642
 643     vbasedev->type = VFIO_DEVICE_TYPE_PLATFORM;
 644     vbasedev->dev = dev;
 645     vbasedev->ops = &vfio_platform_ops;
 646
 647     qemu_mutex_init(&vdev->intp_mutex);
 648
 649     trace_vfio_platform_realize(vbasedev->sysfsdev ?
 650                                 vbasedev->sysfsdev : vbasedev->name,
 651                                 vdev->compat);
 652
 653     ret = vfio_base_device_init(vbasedev, errp);
 654     if (ret) {
 655         goto out;
 656     }
 657
 658     if (!vdev->compat) {
 659         GError *gerr = NULL;
 660         gchar *contents;
 661         gsize length;
 662         char *path;
 663
 664         path = g_strdup_printf("%s/of_node/compatible", vbasedev->sysfsdev);
 665         if (!g_file_get_contents(path, &contents, &length, &gerr)) {
 666             error_setg(errp, "%s", gerr->message);
 667             g_error_free(gerr);
 668             g_free(path);
 669             return;
 670         }
 671         g_free(path);
 672         vdev->compat = contents;
 673         for (vdev->num_compat = 0; length; vdev->num_compat++) {
 674             size_t skip = strlen(contents) + 1;
 675             contents += skip;
 676             length -= skip;
 677         }
 678     }
 679
 680     for (i = 0; i < vbasedev->num_regions; i++) {
 681         if (vfio_region_mmap(vdev->regions[i])) {
 682             error_report("%s mmap unsupported. Performance may be slow",
 683                          memory_region_name(vdev->regions[i]->mem));
 684         }
 685         sysbus_init_mmio(sbdev, vdev->regions[i]->mem);
 686     }
 687 out:
 688     if (!ret) {
 689         return;
 690     }
 691
 692     if (vdev->vbasedev.name) {
 693         error_prepend(errp, ERR_PREFIX, vdev->vbasedev.name);
 694     } else {
 695         error_prepend(errp, "vfio error: ");
 696     }
 697 }
 698
 699 static const VMStateDescription vfio_platform_vmstate = {
 700     .name = TYPE_VFIO_PLATFORM,
 701     .unmigratable = 1,
 702 };
 703
 704 static Property vfio_platform_dev_properties[] = {
 705     DEFINE_PROP_STRING("host", VFIOPlatformDevice, vbasedev.name),
 706     DEFINE_PROP_STRING("sysfsdev", VFIOPlatformDevice, vbasedev.sysfsdev),
 707     DEFINE_PROP_BOOL("x-no-mmap", VFIOPlatformDevice, vbasedev.no_mmap, false),
 708     DEFINE_PROP_UINT32("mmap-timeout-ms", VFIOPlatformDevice,
 709                        mmap_timeout, 1100),
 710     DEFINE_PROP_BOOL("x-irqfd", VFIOPlatformDevice, irqfd_allowed, true),
 711     DEFINE_PROP_END_OF_LIST(),
 712 };
 713
 714 static void vfio_platform_class_init(ObjectClass *klass, void *data)
 715 {
 716     DeviceClass *dc = DEVICE_CLASS(klass);
 717     SysBusDeviceClass *sbc = SYS_BUS_DEVICE_CLASS(klass);
 718
 719     dc->realize = vfio_platform_realize;
 720     dc->props = vfio_platform_dev_properties;
 721     dc->vmsd = &vfio_platform_vmstate;
 722     dc->desc = "VFIO-based platform device assignment";
 723     sbc->connect_irq_notifier = vfio_start_irqfd_injection;
 724     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 725     /* Supported by TYPE_VIRT_MACHINE */
 726     dc->user_creatable = true;
 727 }
 728
 729 static const TypeInfo vfio_platform_dev_info = {
 730     .name = TYPE_VFIO_PLATFORM,
 731     .parent = TYPE_SYS_BUS_DEVICE,
 732     .instance_size = sizeof(VFIOPlatformDevice),
 733     .class_init = vfio_platform_class_init,
 734     .class_size = sizeof(VFIOPlatformDeviceClass),
 735 };
 736
 737 static void register_vfio_platform_dev_type(void)
 738 {
 739     type_register_static(&vfio_platform_dev_info);
 740 }
 741
 742 type_init(register_vfio_platform_dev_type)