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 <sys/ioctl.h>
  18 #include <linux/vfio.h>
  19
  20 #include "hw/vfio/vfio-platform.h"
  21 #include "qemu/error-report.h"
  22 #include "qemu/range.h"
  23 #include "sysemu/sysemu.h"
  24 #include "exec/memory.h"
  25 #include "qemu/queue.h"
  26 #include "hw/sysbus.h"
  27 #include "trace.h"
  28 #include "hw/platform-bus.h"
  29 #include "sysemu/kvm.h"
  30
  31 /*
  32  * Functions used whatever the injection method
  33  */
  34
  35 static inline bool vfio_irq_is_automasked(VFIOINTp *intp)
  36 {
  37     return intp->flags & VFIO_IRQ_INFO_AUTOMASKED;
  38 }
  39
  40 /**
  41  * vfio_init_intp - allocate, initialize the IRQ struct pointer
  42  * and add it into the list of IRQs
  43  * @vbasedev: the VFIO device handle
  44  * @info: irq info struct retrieved from VFIO driver
  45  */
  46 static VFIOINTp *vfio_init_intp(VFIODevice *vbasedev,
  47                                 struct vfio_irq_info info)
  48 {
  49     int ret;
  50     VFIOPlatformDevice *vdev =
  51         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
  52     SysBusDevice *sbdev = SYS_BUS_DEVICE(vdev);
  53     VFIOINTp *intp;
  54
  55     intp = g_malloc0(sizeof(*intp));
  56     intp->vdev = vdev;
  57     intp->pin = info.index;
  58     intp->flags = info.flags;
  59     intp->state = VFIO_IRQ_INACTIVE;
  60     intp->kvm_accel = false;
  61
  62     sysbus_init_irq(sbdev, &intp->qemuirq);
  63
  64     /* Get an eventfd for trigger */
  65     intp->interrupt = g_malloc0(sizeof(EventNotifier));
  66     ret = event_notifier_init(intp->interrupt, 0);
  67     if (ret) {
  68         g_free(intp->interrupt);
  69         g_free(intp);
  70         error_report("vfio: Error: trigger event_notifier_init failed ");
  71         return NULL;
  72     }
  73     if (vfio_irq_is_automasked(intp)) {
  74         /* Get an eventfd for resample/unmask */
  75         intp->unmask = g_malloc0(sizeof(EventNotifier));
  76         ret = event_notifier_init(intp->unmask, 0);
  77         if (ret) {
  78             g_free(intp->interrupt);
  79             g_free(intp->unmask);
  80             g_free(intp);
  81             error_report("vfio: Error: resamplefd event_notifier_init failed");
  82             return NULL;
  83         }
  84     }
  85
  86     QLIST_INSERT_HEAD(&vdev->intp_list, intp, next);
  87     return intp;
  88 }
  89
  90 /**
  91  * vfio_set_trigger_eventfd - set VFIO eventfd handling
  92  *
  93  * @intp: IRQ struct handle
  94  * @handler: handler to be called on eventfd signaling
  95  *
  96  * Setup VFIO signaling and attach an optional user-side handler
  97  * to the eventfd
  98  */
  99 static int vfio_set_trigger_eventfd(VFIOINTp *intp,
 100                                     eventfd_user_side_handler_t handler)
 101 {
 102     VFIODevice *vbasedev = &intp->vdev->vbasedev;
 103     struct vfio_irq_set *irq_set;
 104     int argsz, ret;
 105     int32_t *pfd;
 106
 107     argsz = sizeof(*irq_set) + sizeof(*pfd);
 108     irq_set = g_malloc0(argsz);
 109     irq_set->argsz = argsz;
 110     irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
 111     irq_set->index = intp->pin;
 112     irq_set->start = 0;
 113     irq_set->count = 1;
 114     pfd = (int32_t *)&irq_set->data;
 115     *pfd = event_notifier_get_fd(intp->interrupt);
 116     qemu_set_fd_handler(*pfd, (IOHandler *)handler, NULL, intp);
 117     ret = ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, irq_set);
 118     g_free(irq_set);
 119     if (ret < 0) {
 120         error_report("vfio: Failed to set trigger eventfd: %m");
 121         qemu_set_fd_handler(*pfd, NULL, NULL, NULL);
 122     }
 123     return ret;
 124 }
 125
 126 /*
 127  * Functions only used when eventfds are handled on user-side
 128  * ie. without irqfd
 129  */
 130
 131 /**
 132  * vfio_mmap_set_enabled - enable/disable the fast path mode
 133  * @vdev: the VFIO platform device
 134  * @enabled: the target mmap state
 135  *
 136  * enabled = true ~ fast path = MMIO region is mmaped (no KVM TRAP);
 137  * enabled = false ~ slow path = MMIO region is trapped and region callbacks
 138  * are called; slow path enables to trap the device IRQ status register reset
 139 */
 140
 141 static void vfio_mmap_set_enabled(VFIOPlatformDevice *vdev, bool enabled)
 142 {
 143     int i;
 144
 145     trace_vfio_platform_mmap_set_enabled(enabled);
 146
 147     for (i = 0; i < vdev->vbasedev.num_regions; i++) {
 148         VFIORegion *region = vdev->regions[i];
 149
 150         memory_region_set_enabled(&region->mmap_mem, enabled);
 151     }
 152 }
 153
 154 /**
 155  * vfio_intp_mmap_enable - timer function, restores the fast path
 156  * if there is no more active IRQ
 157  * @opaque: actually points to the VFIO platform device
 158  *
 159  * Called on mmap timer timout, this function checks whether the
 160  * IRQ is still active and if not, restores the fast path.
 161  * by construction a single eventfd is handled at a time.
 162  * if the IRQ is still active, the timer is re-programmed.
 163  */
 164 static void vfio_intp_mmap_enable(void *opaque)
 165 {
 166     VFIOINTp *tmp;
 167     VFIOPlatformDevice *vdev = (VFIOPlatformDevice *)opaque;
 168
 169     qemu_mutex_lock(&vdev->intp_mutex);
 170     QLIST_FOREACH(tmp, &vdev->intp_list, next) {
 171         if (tmp->state == VFIO_IRQ_ACTIVE) {
 172             trace_vfio_platform_intp_mmap_enable(tmp->pin);
 173             /* re-program the timer to check active status later */
 174             timer_mod(vdev->mmap_timer,
 175                       qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 176                           vdev->mmap_timeout);
 177             qemu_mutex_unlock(&vdev->intp_mutex);
 178             return;
 179         }
 180     }
 181     vfio_mmap_set_enabled(vdev, true);
 182     qemu_mutex_unlock(&vdev->intp_mutex);
 183 }
 184
 185 /**
 186  * vfio_intp_inject_pending_lockheld - Injects a pending IRQ
 187  * @opaque: opaque pointer, in practice the VFIOINTp handle
 188  *
 189  * The function is called on a previous IRQ completion, from
 190  * vfio_platform_eoi, while the intp_mutex is locked.
 191  * Also in such situation, the slow path already is set and
 192  * the mmap timer was already programmed.
 193  */
 194 static void vfio_intp_inject_pending_lockheld(VFIOINTp *intp)
 195 {
 196     trace_vfio_platform_intp_inject_pending_lockheld(intp->pin,
 197                               event_notifier_get_fd(intp->interrupt));
 198
 199     intp->state = VFIO_IRQ_ACTIVE;
 200
 201     /* trigger the virtual IRQ */
 202     qemu_set_irq(intp->qemuirq, 1);
 203 }
 204
 205 /**
 206  * vfio_intp_interrupt - The user-side eventfd handler
 207  * @opaque: opaque pointer which in practice is the VFIOINTp handle
 208  *
 209  * the function is entered in event handler context:
 210  * the vIRQ is injected into the guest if there is no other active
 211  * or pending IRQ.
 212  */
 213 static void vfio_intp_interrupt(VFIOINTp *intp)
 214 {
 215     int ret;
 216     VFIOINTp *tmp;
 217     VFIOPlatformDevice *vdev = intp->vdev;
 218     bool delay_handling = false;
 219
 220     qemu_mutex_lock(&vdev->intp_mutex);
 221     if (intp->state == VFIO_IRQ_INACTIVE) {
 222         QLIST_FOREACH(tmp, &vdev->intp_list, next) {
 223             if (tmp->state == VFIO_IRQ_ACTIVE ||
 224                 tmp->state == VFIO_IRQ_PENDING) {
 225                 delay_handling = true;
 226                 break;
 227             }
 228         }
 229     }
 230     if (delay_handling) {
 231         /*
 232          * the new IRQ gets a pending status and is pushed in
 233          * the pending queue
 234          */
 235         intp->state = VFIO_IRQ_PENDING;
 236         trace_vfio_intp_interrupt_set_pending(intp->pin);
 237         QSIMPLEQ_INSERT_TAIL(&vdev->pending_intp_queue,
 238                              intp, pqnext);
 239         ret = event_notifier_test_and_clear(intp->interrupt);
 240         qemu_mutex_unlock(&vdev->intp_mutex);
 241         return;
 242     }
 243
 244     trace_vfio_platform_intp_interrupt(intp->pin,
 245                               event_notifier_get_fd(intp->interrupt));
 246
 247     ret = event_notifier_test_and_clear(intp->interrupt);
 248     if (!ret) {
 249         error_report("Error when clearing fd=%d (ret = %d)",
 250                      event_notifier_get_fd(intp->interrupt), ret);
 251     }
 252
 253     intp->state = VFIO_IRQ_ACTIVE;
 254
 255     /* sets slow path */
 256     vfio_mmap_set_enabled(vdev, false);
 257
 258     /* trigger the virtual IRQ */
 259     qemu_set_irq(intp->qemuirq, 1);
 260
 261     /*
 262      * Schedule the mmap timer which will restore fastpath when no IRQ
 263      * is active anymore
 264      */
 265     if (vdev->mmap_timeout) {
 266         timer_mod(vdev->mmap_timer,
 267                   qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
 268                       vdev->mmap_timeout);
 269     }
 270     qemu_mutex_unlock(&vdev->intp_mutex);
 271 }
 272
 273 /**
 274  * vfio_platform_eoi - IRQ completion routine
 275  * @vbasedev: the VFIO device handle
 276  *
 277  * De-asserts the active virtual IRQ and unmasks the physical IRQ
 278  * (effective for level sensitive IRQ auto-masked by the  VFIO driver).
 279  * Then it handles next pending IRQ if any.
 280  * eoi function is called on the first access to any MMIO region
 281  * after an IRQ was triggered, trapped since slow path was set.
 282  * It is assumed this access corresponds to the IRQ status
 283  * register reset. With such a mechanism, a single IRQ can be
 284  * handled at a time since there is no way to know which IRQ
 285  * was completed by the guest (we would need additional details
 286  * about the IRQ status register mask).
 287  */
 288 static void vfio_platform_eoi(VFIODevice *vbasedev)
 289 {
 290     VFIOINTp *intp;
 291     VFIOPlatformDevice *vdev =
 292         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
 293
 294     qemu_mutex_lock(&vdev->intp_mutex);
 295     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 296         if (intp->state == VFIO_IRQ_ACTIVE) {
 297             trace_vfio_platform_eoi(intp->pin,
 298                                 event_notifier_get_fd(intp->interrupt));
 299             intp->state = VFIO_IRQ_INACTIVE;
 300
 301             /* deassert the virtual IRQ */
 302             qemu_set_irq(intp->qemuirq, 0);
 303
 304             if (vfio_irq_is_automasked(intp)) {
 305                 /* unmasks the physical level-sensitive IRQ */
 306                 vfio_unmask_single_irqindex(vbasedev, intp->pin);
 307             }
 308
 309             /* a single IRQ can be active at a time */
 310             break;
 311         }
 312     }
 313     /* in case there are pending IRQs, handle the first one */
 314     if (!QSIMPLEQ_EMPTY(&vdev->pending_intp_queue)) {
 315         intp = QSIMPLEQ_FIRST(&vdev->pending_intp_queue);
 316         vfio_intp_inject_pending_lockheld(intp);
 317         QSIMPLEQ_REMOVE_HEAD(&vdev->pending_intp_queue, pqnext);
 318     }
 319     qemu_mutex_unlock(&vdev->intp_mutex);
 320 }
 321
 322 /**
 323  * vfio_start_eventfd_injection - starts the virtual IRQ injection using
 324  * user-side handled eventfds
 325  * @sbdev: the sysbus device handle
 326  * @irq: the qemu irq handle
 327  */
 328
 329 static void vfio_start_eventfd_injection(SysBusDevice *sbdev, qemu_irq irq)
 330 {
 331     int ret;
 332     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(sbdev);
 333     VFIOINTp *intp;
 334
 335     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 336         if (intp->qemuirq == irq) {
 337             break;
 338         }
 339     }
 340     assert(intp);
 341
 342     ret = vfio_set_trigger_eventfd(intp, vfio_intp_interrupt);
 343     if (ret) {
 344         error_report("vfio: failed to start eventfd signaling for IRQ %d: %m",
 345                      intp->pin);
 346         abort();
 347     }
 348 }
 349
 350 /*
 351  * Functions used for irqfd
 352  */
 353
 354 /**
 355  * vfio_set_resample_eventfd - sets the resamplefd for an IRQ
 356  * @intp: the IRQ struct handle
 357  * programs the VFIO driver to unmask this IRQ when the
 358  * intp->unmask eventfd is triggered
 359  */
 360 static int vfio_set_resample_eventfd(VFIOINTp *intp)
 361 {
 362     VFIODevice *vbasedev = &intp->vdev->vbasedev;
 363     struct vfio_irq_set *irq_set;
 364     int argsz, ret;
 365     int32_t *pfd;
 366
 367     argsz = sizeof(*irq_set) + sizeof(*pfd);
 368     irq_set = g_malloc0(argsz);
 369     irq_set->argsz = argsz;
 370     irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_UNMASK;
 371     irq_set->index = intp->pin;
 372     irq_set->start = 0;
 373     irq_set->count = 1;
 374     pfd = (int32_t *)&irq_set->data;
 375     *pfd = event_notifier_get_fd(intp->unmask);
 376     qemu_set_fd_handler(*pfd, NULL, NULL, NULL);
 377     ret = ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, irq_set);
 378     g_free(irq_set);
 379     if (ret < 0) {
 380         error_report("vfio: Failed to set resample eventfd: %m");
 381     }
 382     return ret;
 383 }
 384
 385 /**
 386  * vfio_start_irqfd_injection - starts the virtual IRQ injection using
 387  * irqfd
 388  *
 389  * @sbdev: the sysbus device handle
 390  * @irq: the qemu irq handle
 391  *
 392  * In case the irqfd setup fails, we fallback to userspace handled eventfd
 393  */
 394 static void vfio_start_irqfd_injection(SysBusDevice *sbdev, qemu_irq irq)
 395 {
 396     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(sbdev);
 397     VFIOINTp *intp;
 398
 399     if (!kvm_irqfds_enabled() || !kvm_resamplefds_enabled() ||
 400         !vdev->irqfd_allowed) {
 401         goto fail_irqfd;
 402     }
 403
 404     QLIST_FOREACH(intp, &vdev->intp_list, next) {
 405         if (intp->qemuirq == irq) {
 406             break;
 407         }
 408     }
 409     assert(intp);
 410
 411     if (kvm_irqchip_add_irqfd_notifier(kvm_state, intp->interrupt,
 412                                    intp->unmask, irq) < 0) {
 413         goto fail_irqfd;
 414     }
 415
 416     if (vfio_set_trigger_eventfd(intp, NULL) < 0) {
 417         goto fail_vfio;
 418     }
 419     if (vfio_irq_is_automasked(intp)) {
 420         if (vfio_set_resample_eventfd(intp) < 0) {
 421             goto fail_vfio;
 422         }
 423         trace_vfio_platform_start_level_irqfd_injection(intp->pin,
 424                                     event_notifier_get_fd(intp->interrupt),
 425                                     event_notifier_get_fd(intp->unmask));
 426     } else {
 427         trace_vfio_platform_start_edge_irqfd_injection(intp->pin,
 428                                     event_notifier_get_fd(intp->interrupt));
 429     }
 430
 431     intp->kvm_accel = true;
 432
 433     return;
 434 fail_vfio:
 435     kvm_irqchip_remove_irqfd_notifier(kvm_state, intp->interrupt, irq);
 436     error_report("vfio: failed to start eventfd signaling for IRQ %d: %m",
 437                  intp->pin);
 438     abort();
 439 fail_irqfd:
 440     vfio_start_eventfd_injection(sbdev, irq);
 441     return;
 442 }
 443
 444 /* VFIO skeleton */
 445
 446 static void vfio_platform_compute_needs_reset(VFIODevice *vbasedev)
 447 {
 448     vbasedev->needs_reset = true;
 449 }
 450
 451 /* not implemented yet */
 452 static int vfio_platform_hot_reset_multi(VFIODevice *vbasedev)
 453 {
 454     return -1;
 455 }
 456
 457 /**
 458  * vfio_populate_device - Allocate and populate MMIO region
 459  * and IRQ structs according to driver returned information
 460  * @vbasedev: the VFIO device handle
 461  *
 462  */
 463 static int vfio_populate_device(VFIODevice *vbasedev)
 464 {
 465     VFIOINTp *intp, *tmp;
 466     int i, ret = -1;
 467     VFIOPlatformDevice *vdev =
 468         container_of(vbasedev, VFIOPlatformDevice, vbasedev);
 469
 470     if (!(vbasedev->flags & VFIO_DEVICE_FLAGS_PLATFORM)) {
 471         error_report("vfio: Um, this isn't a platform device");
 472         return ret;
 473     }
 474
 475     vdev->regions = g_new0(VFIORegion *, vbasedev->num_regions);
 476
 477     for (i = 0; i < vbasedev->num_regions; i++) {
 478         struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) };
 479         VFIORegion *ptr;
 480
 481         vdev->regions[i] = g_new0(VFIORegion, 1);
 482         ptr = vdev->regions[i];
 483         reg_info.index = i;
 484         ret = ioctl(vbasedev->fd, VFIO_DEVICE_GET_REGION_INFO, &reg_info);
 485         if (ret) {
 486             error_report("vfio: Error getting region %d info: %m", i);
 487             goto reg_error;
 488         }
 489         ptr->flags = reg_info.flags;
 490         ptr->size = reg_info.size;
 491         ptr->fd_offset = reg_info.offset;
 492         ptr->nr = i;
 493         ptr->vbasedev = vbasedev;
 494
 495         trace_vfio_platform_populate_regions(ptr->nr,
 496                             (unsigned long)ptr->flags,
 497                             (unsigned long)ptr->size,
 498                             ptr->vbasedev->fd,
 499                             (unsigned long)ptr->fd_offset);
 500     }
 501
 502     vdev->mmap_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
 503                                     vfio_intp_mmap_enable, vdev);
 504
 505     QSIMPLEQ_INIT(&vdev->pending_intp_queue);
 506
 507     for (i = 0; i < vbasedev->num_irqs; i++) {
 508         struct vfio_irq_info irq = { .argsz = sizeof(irq) };
 509
 510         irq.index = i;
 511         ret = ioctl(vbasedev->fd, VFIO_DEVICE_GET_IRQ_INFO, &irq);
 512         if (ret) {
 513             error_printf("vfio: error getting device %s irq info",
 514                          vbasedev->name);
 515             goto irq_err;
 516         } else {
 517             trace_vfio_platform_populate_interrupts(irq.index,
 518                                                     irq.count,
 519                                                     irq.flags);
 520             intp = vfio_init_intp(vbasedev, irq);
 521             if (!intp) {
 522                 error_report("vfio: Error installing IRQ %d up", i);
 523                 goto irq_err;
 524             }
 525         }
 526     }
 527     return 0;
 528 irq_err:
 529     timer_del(vdev->mmap_timer);
 530     QLIST_FOREACH_SAFE(intp, &vdev->intp_list, next, tmp) {
 531         QLIST_REMOVE(intp, next);
 532         g_free(intp);
 533     }
 534 reg_error:
 535     for (i = 0; i < vbasedev->num_regions; i++) {
 536         g_free(vdev->regions[i]);
 537     }
 538     g_free(vdev->regions);
 539     return ret;
 540 }
 541
 542 /* specialized functions for VFIO Platform devices */
 543 static VFIODeviceOps vfio_platform_ops = {
 544     .vfio_compute_needs_reset = vfio_platform_compute_needs_reset,
 545     .vfio_hot_reset_multi = vfio_platform_hot_reset_multi,
 546     .vfio_eoi = vfio_platform_eoi,
 547 };
 548
 549 /**
 550  * vfio_base_device_init - perform preliminary VFIO setup
 551  * @vbasedev: the VFIO device handle
 552  *
 553  * Implement the VFIO command sequence that allows to discover
 554  * assigned device resources: group extraction, device
 555  * fd retrieval, resource query.
 556  * Precondition: the device name must be initialized
 557  */
 558 static int vfio_base_device_init(VFIODevice *vbasedev)
 559 {
 560     VFIOGroup *group;
 561     VFIODevice *vbasedev_iter;
 562     char path[PATH_MAX], iommu_group_path[PATH_MAX], *group_name;
 563     ssize_t len;
 564     struct stat st;
 565     int groupid;
 566     int ret;
 567
 568     /* name must be set prior to the call */
 569     if (!vbasedev->name || strchr(vbasedev->name, '/')) {
 570         return -EINVAL;
 571     }
 572
 573     /* Check that the host device exists */
 574     g_snprintf(path, sizeof(path), "/sys/bus/platform/devices/%s/",
 575                vbasedev->name);
 576
 577     if (stat(path, &st) < 0) {
 578         error_report("vfio: error: no such host device: %s", path);
 579         return -errno;
 580     }
 581
 582     g_strlcat(path, "iommu_group", sizeof(path));
 583     len = readlink(path, iommu_group_path, sizeof(iommu_group_path));
 584     if (len < 0 || len >= sizeof(iommu_group_path)) {
 585         error_report("vfio: error no iommu_group for device");
 586         return len < 0 ? -errno : -ENAMETOOLONG;
 587     }
 588
 589     iommu_group_path[len] = 0;
 590     group_name = basename(iommu_group_path);
 591
 592     if (sscanf(group_name, "%d", &groupid) != 1) {
 593         error_report("vfio: error reading %s: %m", path);
 594         return -errno;
 595     }
 596
 597     trace_vfio_platform_base_device_init(vbasedev->name, groupid);
 598
 599     group = vfio_get_group(groupid, &address_space_memory);
 600     if (!group) {
 601         error_report("vfio: failed to get group %d", groupid);
 602         return -ENOENT;
 603     }
 604
 605     g_snprintf(path, sizeof(path), "%s", vbasedev->name);
 606
 607     QLIST_FOREACH(vbasedev_iter, &group->device_list, next) {
 608         if (strcmp(vbasedev_iter->name, vbasedev->name) == 0) {
 609             error_report("vfio: error: device %s is already attached", path);
 610             vfio_put_group(group);
 611             return -EBUSY;
 612         }
 613     }
 614     ret = vfio_get_device(group, path, vbasedev);
 615     if (ret) {
 616         error_report("vfio: failed to get device %s", path);
 617         vfio_put_group(group);
 618         return ret;
 619     }
 620
 621     ret = vfio_populate_device(vbasedev);
 622     if (ret) {
 623         error_report("vfio: failed to populate device %s", path);
 624         vfio_put_group(group);
 625     }
 626
 627     return ret;
 628 }
 629
 630 /**
 631  * vfio_map_region - initialize the 2 memory regions for a given
 632  * MMIO region index
 633  * @vdev: the VFIO platform device handle
 634  * @nr: the index of the region
 635  *
 636  * Init the top memory region and the mmapped memory region beneath
 637  * VFIOPlatformDevice is used since VFIODevice is not a QOM Object
 638  * and could not be passed to memory region functions
 639 */
 640 static void vfio_map_region(VFIOPlatformDevice *vdev, int nr)
 641 {
 642     VFIORegion *region = vdev->regions[nr];
 643     uint64_t size = region->size;
 644     char name[64];
 645
 646     if (!size) {
 647         return;
 648     }
 649
 650     g_snprintf(name, sizeof(name), "VFIO %s region %d",
 651                vdev->vbasedev.name, nr);
 652
 653     /* A "slow" read/write mapping underlies all regions */
 654     memory_region_init_io(&region->mem, OBJECT(vdev), &vfio_region_ops,
 655                           region, name, size);
 656
 657     g_strlcat(name, " mmap", sizeof(name));
 658
 659     if (vfio_mmap_region(OBJECT(vdev), region, &region->mem,
 660                          &region->mmap_mem, &region->mmap, size, 0, name)) {
 661         error_report("%s unsupported. Performance may be slow", name);
 662     }
 663 }
 664
 665 /**
 666  * vfio_platform_realize  - the device realize function
 667  * @dev: device state pointer
 668  * @errp: error
 669  *
 670  * initialize the device, its memory regions and IRQ structures
 671  * IRQ are started separately
 672  */
 673 static void vfio_platform_realize(DeviceState *dev, Error **errp)
 674 {
 675     VFIOPlatformDevice *vdev = VFIO_PLATFORM_DEVICE(dev);
 676     SysBusDevice *sbdev = SYS_BUS_DEVICE(dev);
 677     VFIODevice *vbasedev = &vdev->vbasedev;
 678     int i, ret;
 679
 680     vbasedev->type = VFIO_DEVICE_TYPE_PLATFORM;
 681     vbasedev->ops = &vfio_platform_ops;
 682
 683     trace_vfio_platform_realize(vbasedev->name, vdev->compat);
 684
 685     ret = vfio_base_device_init(vbasedev);
 686     if (ret) {
 687         error_setg(errp, "vfio: vfio_base_device_init failed for %s",
 688                    vbasedev->name);
 689         return;
 690     }
 691
 692     for (i = 0; i < vbasedev->num_regions; i++) {
 693         vfio_map_region(vdev, i);
 694         sysbus_init_mmio(sbdev, &vdev->regions[i]->mem);
 695     }
 696 }
 697
 698 static const VMStateDescription vfio_platform_vmstate = {
 699     .name = TYPE_VFIO_PLATFORM,
 700     .unmigratable = 1,
 701 };
 702
 703 static Property vfio_platform_dev_properties[] = {
 704     DEFINE_PROP_STRING("host", VFIOPlatformDevice, vbasedev.name),
 705     DEFINE_PROP_BOOL("x-no-mmap", VFIOPlatformDevice, vbasedev.no_mmap, false),
 706     DEFINE_PROP_UINT32("mmap-timeout-ms", VFIOPlatformDevice,
 707                        mmap_timeout, 1100),
 708     DEFINE_PROP_BOOL("x-irqfd", VFIOPlatformDevice, irqfd_allowed, true),
 709     DEFINE_PROP_END_OF_LIST(),
 710 };
 711
 712 static void vfio_platform_class_init(ObjectClass *klass, void *data)
 713 {
 714     DeviceClass *dc = DEVICE_CLASS(klass);
 715     SysBusDeviceClass *sbc = SYS_BUS_DEVICE_CLASS(klass);
 716
 717     dc->realize = vfio_platform_realize;
 718     dc->props = vfio_platform_dev_properties;
 719     dc->vmsd = &vfio_platform_vmstate;
 720     dc->desc = "VFIO-based platform device assignment";
 721     sbc->connect_irq_notifier = vfio_start_irqfd_injection;
 722     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 723 }
 724
 725 static const TypeInfo vfio_platform_dev_info = {
 726     .name = TYPE_VFIO_PLATFORM,
 727     .parent = TYPE_SYS_BUS_DEVICE,
 728     .instance_size = sizeof(VFIOPlatformDevice),
 729     .class_init = vfio_platform_class_init,
 730     .class_size = sizeof(VFIOPlatformDeviceClass),
 731     .abstract   = true,
 732 };
 733
 734 static void register_vfio_platform_dev_type(void)
 735 {
 736     type_register_static(&vfio_platform_dev_info);
 737 }
 738
 739 type_init(register_vfio_platform_dev_type)