xref: /openbmc/qemu/hw/vfio/iommufd.c (revision 6c635326)
1 /*
2  * iommufd container backend
3  *
4  * Copyright (C) 2023 Intel Corporation.
5  * Copyright Red Hat, Inc. 2023
6  *
7  * Authors: Yi Liu <yi.l.liu@intel.com>
8  *          Eric Auger <eric.auger@redhat.com>
9  *
10  * SPDX-License-Identifier: GPL-2.0-or-later
11  */
12 
13 #include "qemu/osdep.h"
14 #include <sys/ioctl.h>
15 #include <linux/vfio.h>
16 #include <linux/iommufd.h>
17 
18 #include "hw/vfio/vfio-common.h"
19 #include "qemu/error-report.h"
20 #include "trace.h"
21 #include "qapi/error.h"
22 #include "sysemu/iommufd.h"
23 #include "hw/qdev-core.h"
24 #include "sysemu/reset.h"
25 #include "qemu/cutils.h"
26 #include "qemu/chardev_open.h"
27 #include "pci.h"
28 
29 static int iommufd_cdev_map(const VFIOContainerBase *bcontainer, hwaddr iova,
30                             ram_addr_t size, void *vaddr, bool readonly)
31 {
32     const VFIOIOMMUFDContainer *container =
33         container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
34 
35     return iommufd_backend_map_dma(container->be,
36                                    container->ioas_id,
37                                    iova, size, vaddr, readonly);
38 }
39 
40 static int iommufd_cdev_unmap(const VFIOContainerBase *bcontainer,
41                               hwaddr iova, ram_addr_t size,
42                               IOMMUTLBEntry *iotlb)
43 {
44     const VFIOIOMMUFDContainer *container =
45         container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
46 
47     /* TODO: Handle dma_unmap_bitmap with iotlb args (migration) */
48     return iommufd_backend_unmap_dma(container->be,
49                                      container->ioas_id, iova, size);
50 }
51 
52 static bool iommufd_cdev_kvm_device_add(VFIODevice *vbasedev, Error **errp)
53 {
54     return !vfio_kvm_device_add_fd(vbasedev->fd, errp);
55 }
56 
57 static void iommufd_cdev_kvm_device_del(VFIODevice *vbasedev)
58 {
59     Error *err = NULL;
60 
61     if (vfio_kvm_device_del_fd(vbasedev->fd, &err)) {
62         error_report_err(err);
63     }
64 }
65 
66 static bool iommufd_cdev_connect_and_bind(VFIODevice *vbasedev, Error **errp)
67 {
68     IOMMUFDBackend *iommufd = vbasedev->iommufd;
69     struct vfio_device_bind_iommufd bind = {
70         .argsz = sizeof(bind),
71         .flags = 0,
72     };
73 
74     if (!iommufd_backend_connect(iommufd, errp)) {
75         return false;
76     }
77 
78     /*
79      * Add device to kvm-vfio to be prepared for the tracking
80      * in KVM. Especially for some emulated devices, it requires
81      * to have kvm information in the device open.
82      */
83     if (!iommufd_cdev_kvm_device_add(vbasedev, errp)) {
84         goto err_kvm_device_add;
85     }
86 
87     /* Bind device to iommufd */
88     bind.iommufd = iommufd->fd;
89     if (ioctl(vbasedev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind)) {
90         error_setg_errno(errp, errno, "error bind device fd=%d to iommufd=%d",
91                          vbasedev->fd, bind.iommufd);
92         goto err_bind;
93     }
94 
95     vbasedev->devid = bind.out_devid;
96     trace_iommufd_cdev_connect_and_bind(bind.iommufd, vbasedev->name,
97                                         vbasedev->fd, vbasedev->devid);
98     return true;
99 err_bind:
100     iommufd_cdev_kvm_device_del(vbasedev);
101 err_kvm_device_add:
102     iommufd_backend_disconnect(iommufd);
103     return false;
104 }
105 
106 static void iommufd_cdev_unbind_and_disconnect(VFIODevice *vbasedev)
107 {
108     /* Unbind is automatically conducted when device fd is closed */
109     iommufd_cdev_kvm_device_del(vbasedev);
110     iommufd_backend_disconnect(vbasedev->iommufd);
111 }
112 
113 static int iommufd_cdev_getfd(const char *sysfs_path, Error **errp)
114 {
115     ERRP_GUARD();
116     long int ret = -ENOTTY;
117     g_autofree char *path = NULL;
118     g_autofree char *vfio_dev_path = NULL;
119     g_autofree char *vfio_path = NULL;
120     DIR *dir = NULL;
121     struct dirent *dent;
122     g_autofree gchar *contents = NULL;
123     gsize length;
124     int major, minor;
125     dev_t vfio_devt;
126 
127     path = g_strdup_printf("%s/vfio-dev", sysfs_path);
128     dir = opendir(path);
129     if (!dir) {
130         error_setg_errno(errp, errno, "couldn't open directory %s", path);
131         goto out;
132     }
133 
134     while ((dent = readdir(dir))) {
135         if (!strncmp(dent->d_name, "vfio", 4)) {
136             vfio_dev_path = g_strdup_printf("%s/%s/dev", path, dent->d_name);
137             break;
138         }
139     }
140 
141     if (!vfio_dev_path) {
142         error_setg(errp, "failed to find vfio-dev/vfioX/dev");
143         goto out_close_dir;
144     }
145 
146     if (!g_file_get_contents(vfio_dev_path, &contents, &length, NULL)) {
147         error_setg(errp, "failed to load \"%s\"", vfio_dev_path);
148         goto out_close_dir;
149     }
150 
151     if (sscanf(contents, "%d:%d", &major, &minor) != 2) {
152         error_setg(errp, "failed to get major:minor for \"%s\"", vfio_dev_path);
153         goto out_close_dir;
154     }
155     vfio_devt = makedev(major, minor);
156 
157     vfio_path = g_strdup_printf("/dev/vfio/devices/%s", dent->d_name);
158     ret = open_cdev(vfio_path, vfio_devt);
159     if (ret < 0) {
160         error_setg(errp, "Failed to open %s", vfio_path);
161     }
162 
163     trace_iommufd_cdev_getfd(vfio_path, ret);
164 
165 out_close_dir:
166     closedir(dir);
167 out:
168     if (*errp) {
169         error_prepend(errp, VFIO_MSG_PREFIX, path);
170     }
171 
172     return ret;
173 }
174 
175 static int iommufd_cdev_attach_ioas_hwpt(VFIODevice *vbasedev, uint32_t id,
176                                          Error **errp)
177 {
178     int iommufd = vbasedev->iommufd->fd;
179     struct vfio_device_attach_iommufd_pt attach_data = {
180         .argsz = sizeof(attach_data),
181         .flags = 0,
182         .pt_id = id,
183     };
184 
185     /* Attach device to an IOAS or hwpt within iommufd */
186     if (ioctl(vbasedev->fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &attach_data)) {
187         error_setg_errno(errp, errno,
188                          "[iommufd=%d] error attach %s (%d) to id=%d",
189                          iommufd, vbasedev->name, vbasedev->fd, id);
190         return -errno;
191     }
192 
193     trace_iommufd_cdev_attach_ioas_hwpt(iommufd, vbasedev->name,
194                                         vbasedev->fd, id);
195     return 0;
196 }
197 
198 static bool iommufd_cdev_detach_ioas_hwpt(VFIODevice *vbasedev, Error **errp)
199 {
200     int iommufd = vbasedev->iommufd->fd;
201     struct vfio_device_detach_iommufd_pt detach_data = {
202         .argsz = sizeof(detach_data),
203         .flags = 0,
204     };
205 
206     if (ioctl(vbasedev->fd, VFIO_DEVICE_DETACH_IOMMUFD_PT, &detach_data)) {
207         error_setg_errno(errp, errno, "detach %s failed", vbasedev->name);
208         return false;
209     }
210 
211     trace_iommufd_cdev_detach_ioas_hwpt(iommufd, vbasedev->name);
212     return true;
213 }
214 
215 static bool iommufd_cdev_autodomains_get(VFIODevice *vbasedev,
216                                          VFIOIOMMUFDContainer *container,
217                                          Error **errp)
218 {
219     ERRP_GUARD();
220     IOMMUFDBackend *iommufd = vbasedev->iommufd;
221     uint32_t flags = 0;
222     VFIOIOASHwpt *hwpt;
223     uint32_t hwpt_id;
224     int ret;
225 
226     /* Try to find a domain */
227     QLIST_FOREACH(hwpt, &container->hwpt_list, next) {
228         ret = iommufd_cdev_attach_ioas_hwpt(vbasedev, hwpt->hwpt_id, errp);
229         if (ret) {
230             /* -EINVAL means the domain is incompatible with the device. */
231             if (ret == -EINVAL) {
232                 /*
233                  * It is an expected failure and it just means we will try
234                  * another domain, or create one if no existing compatible
235                  * domain is found. Hence why the error is discarded below.
236                  */
237                 error_free(*errp);
238                 *errp = NULL;
239                 continue;
240             }
241 
242             return false;
243         } else {
244             vbasedev->hwpt = hwpt;
245             QLIST_INSERT_HEAD(&hwpt->device_list, vbasedev, hwpt_next);
246             return true;
247         }
248     }
249 
250     if (!iommufd_backend_alloc_hwpt(iommufd, vbasedev->devid,
251                                     container->ioas_id, flags,
252                                     IOMMU_HWPT_DATA_NONE, 0, NULL,
253                                     &hwpt_id, errp)) {
254         return false;
255     }
256 
257     hwpt = g_malloc0(sizeof(*hwpt));
258     hwpt->hwpt_id = hwpt_id;
259     QLIST_INIT(&hwpt->device_list);
260 
261     ret = iommufd_cdev_attach_ioas_hwpt(vbasedev, hwpt->hwpt_id, errp);
262     if (ret) {
263         iommufd_backend_free_id(container->be, hwpt->hwpt_id);
264         g_free(hwpt);
265         return false;
266     }
267 
268     vbasedev->hwpt = hwpt;
269     QLIST_INSERT_HEAD(&hwpt->device_list, vbasedev, hwpt_next);
270     QLIST_INSERT_HEAD(&container->hwpt_list, hwpt, next);
271     return true;
272 }
273 
274 static void iommufd_cdev_autodomains_put(VFIODevice *vbasedev,
275                                          VFIOIOMMUFDContainer *container)
276 {
277     VFIOIOASHwpt *hwpt = vbasedev->hwpt;
278 
279     QLIST_REMOVE(vbasedev, hwpt_next);
280     vbasedev->hwpt = NULL;
281 
282     if (QLIST_EMPTY(&hwpt->device_list)) {
283         QLIST_REMOVE(hwpt, next);
284         iommufd_backend_free_id(container->be, hwpt->hwpt_id);
285         g_free(hwpt);
286     }
287 }
288 
289 static bool iommufd_cdev_attach_container(VFIODevice *vbasedev,
290                                           VFIOIOMMUFDContainer *container,
291                                           Error **errp)
292 {
293     /* mdevs aren't physical devices and will fail with auto domains */
294     if (!vbasedev->mdev) {
295         return iommufd_cdev_autodomains_get(vbasedev, container, errp);
296     }
297 
298     return !iommufd_cdev_attach_ioas_hwpt(vbasedev, container->ioas_id, errp);
299 }
300 
301 static void iommufd_cdev_detach_container(VFIODevice *vbasedev,
302                                           VFIOIOMMUFDContainer *container)
303 {
304     Error *err = NULL;
305 
306     if (!iommufd_cdev_detach_ioas_hwpt(vbasedev, &err)) {
307         error_report_err(err);
308     }
309 
310     if (vbasedev->hwpt) {
311         iommufd_cdev_autodomains_put(vbasedev, container);
312     }
313 
314 }
315 
316 static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)
317 {
318     VFIOContainerBase *bcontainer = &container->bcontainer;
319 
320     if (!QLIST_EMPTY(&bcontainer->device_list)) {
321         return;
322     }
323     memory_listener_unregister(&bcontainer->listener);
324     iommufd_backend_free_id(container->be, container->ioas_id);
325     object_unref(container);
326 }
327 
328 static int iommufd_cdev_ram_block_discard_disable(bool state)
329 {
330     /*
331      * We support coordinated discarding of RAM via the RamDiscardManager.
332      */
333     return ram_block_uncoordinated_discard_disable(state);
334 }
335 
336 static bool iommufd_cdev_get_info_iova_range(VFIOIOMMUFDContainer *container,
337                                              uint32_t ioas_id, Error **errp)
338 {
339     VFIOContainerBase *bcontainer = &container->bcontainer;
340     g_autofree struct iommu_ioas_iova_ranges *info = NULL;
341     struct iommu_iova_range *iova_ranges;
342     int sz, fd = container->be->fd;
343 
344     info = g_malloc0(sizeof(*info));
345     info->size = sizeof(*info);
346     info->ioas_id = ioas_id;
347 
348     if (ioctl(fd, IOMMU_IOAS_IOVA_RANGES, info) && errno != EMSGSIZE) {
349         goto error;
350     }
351 
352     sz = info->num_iovas * sizeof(struct iommu_iova_range);
353     info = g_realloc(info, sizeof(*info) + sz);
354     info->allowed_iovas = (uintptr_t)(info + 1);
355 
356     if (ioctl(fd, IOMMU_IOAS_IOVA_RANGES, info)) {
357         goto error;
358     }
359 
360     iova_ranges = (struct iommu_iova_range *)(uintptr_t)info->allowed_iovas;
361 
362     for (int i = 0; i < info->num_iovas; i++) {
363         Range *range = g_new(Range, 1);
364 
365         range_set_bounds(range, iova_ranges[i].start, iova_ranges[i].last);
366         bcontainer->iova_ranges =
367             range_list_insert(bcontainer->iova_ranges, range);
368     }
369     bcontainer->pgsizes = info->out_iova_alignment;
370 
371     return true;
372 
373 error:
374     error_setg_errno(errp, errno, "Cannot get IOVA ranges");
375     return false;
376 }
377 
378 static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,
379                                 AddressSpace *as, Error **errp)
380 {
381     VFIOContainerBase *bcontainer;
382     VFIOIOMMUFDContainer *container;
383     VFIOAddressSpace *space;
384     struct vfio_device_info dev_info = { .argsz = sizeof(dev_info) };
385     int ret, devfd;
386     uint32_t ioas_id;
387     Error *err = NULL;
388     const VFIOIOMMUClass *iommufd_vioc =
389         VFIO_IOMMU_CLASS(object_class_by_name(TYPE_VFIO_IOMMU_IOMMUFD));
390 
391     if (vbasedev->fd < 0) {
392         devfd = iommufd_cdev_getfd(vbasedev->sysfsdev, errp);
393         if (devfd < 0) {
394             return false;
395         }
396         vbasedev->fd = devfd;
397     } else {
398         devfd = vbasedev->fd;
399     }
400 
401     if (!iommufd_cdev_connect_and_bind(vbasedev, errp)) {
402         goto err_connect_bind;
403     }
404 
405     space = vfio_get_address_space(as);
406 
407     /* try to attach to an existing container in this space */
408     QLIST_FOREACH(bcontainer, &space->containers, next) {
409         container = container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
410         if (VFIO_IOMMU_GET_CLASS(bcontainer) != iommufd_vioc ||
411             vbasedev->iommufd != container->be) {
412             continue;
413         }
414         if (!iommufd_cdev_attach_container(vbasedev, container, &err)) {
415             const char *msg = error_get_pretty(err);
416 
417             trace_iommufd_cdev_fail_attach_existing_container(msg);
418             error_free(err);
419             err = NULL;
420         } else {
421             ret = iommufd_cdev_ram_block_discard_disable(true);
422             if (ret) {
423                 error_setg(errp,
424                               "Cannot set discarding of RAM broken (%d)", ret);
425                 goto err_discard_disable;
426             }
427             goto found_container;
428         }
429     }
430 
431     /* Need to allocate a new dedicated container */
432     if (!iommufd_backend_alloc_ioas(vbasedev->iommufd, &ioas_id, errp)) {
433         goto err_alloc_ioas;
434     }
435 
436     trace_iommufd_cdev_alloc_ioas(vbasedev->iommufd->fd, ioas_id);
437 
438     container = VFIO_IOMMU_IOMMUFD(object_new(TYPE_VFIO_IOMMU_IOMMUFD));
439     container->be = vbasedev->iommufd;
440     container->ioas_id = ioas_id;
441     QLIST_INIT(&container->hwpt_list);
442 
443     bcontainer = &container->bcontainer;
444     vfio_address_space_insert(space, bcontainer);
445 
446     if (!iommufd_cdev_attach_container(vbasedev, container, errp)) {
447         goto err_attach_container;
448     }
449 
450     ret = iommufd_cdev_ram_block_discard_disable(true);
451     if (ret) {
452         goto err_discard_disable;
453     }
454 
455     if (!iommufd_cdev_get_info_iova_range(container, ioas_id, &err)) {
456         error_append_hint(&err,
457                    "Fallback to default 64bit IOVA range and 4K page size\n");
458         warn_report_err(err);
459         err = NULL;
460         bcontainer->pgsizes = qemu_real_host_page_size();
461     }
462 
463     bcontainer->listener = vfio_memory_listener;
464     memory_listener_register(&bcontainer->listener, bcontainer->space->as);
465 
466     if (bcontainer->error) {
467         error_propagate_prepend(errp, bcontainer->error,
468                                 "memory listener initialization failed: ");
469         goto err_listener_register;
470     }
471 
472     bcontainer->initialized = true;
473 
474 found_container:
475     ret = ioctl(devfd, VFIO_DEVICE_GET_INFO, &dev_info);
476     if (ret) {
477         error_setg_errno(errp, errno, "error getting device info");
478         goto err_listener_register;
479     }
480 
481     if (!vfio_cpr_register_container(bcontainer, errp)) {
482         goto err_listener_register;
483     }
484 
485     /*
486      * TODO: examine RAM_BLOCK_DISCARD stuff, should we do group level
487      * for discarding incompatibility check as well?
488      */
489     if (vbasedev->ram_block_discard_allowed) {
490         iommufd_cdev_ram_block_discard_disable(false);
491     }
492 
493     vbasedev->group = 0;
494     vbasedev->num_irqs = dev_info.num_irqs;
495     vbasedev->num_regions = dev_info.num_regions;
496     vbasedev->flags = dev_info.flags;
497     vbasedev->reset_works = !!(dev_info.flags & VFIO_DEVICE_FLAGS_RESET);
498     vbasedev->bcontainer = bcontainer;
499     QLIST_INSERT_HEAD(&bcontainer->device_list, vbasedev, container_next);
500     QLIST_INSERT_HEAD(&vfio_device_list, vbasedev, global_next);
501 
502     trace_iommufd_cdev_device_info(vbasedev->name, devfd, vbasedev->num_irqs,
503                                    vbasedev->num_regions, vbasedev->flags);
504     return true;
505 
506 err_listener_register:
507     iommufd_cdev_ram_block_discard_disable(false);
508 err_discard_disable:
509     iommufd_cdev_detach_container(vbasedev, container);
510 err_attach_container:
511     iommufd_cdev_container_destroy(container);
512 err_alloc_ioas:
513     vfio_put_address_space(space);
514     iommufd_cdev_unbind_and_disconnect(vbasedev);
515 err_connect_bind:
516     close(vbasedev->fd);
517     return false;
518 }
519 
520 static void iommufd_cdev_detach(VFIODevice *vbasedev)
521 {
522     VFIOContainerBase *bcontainer = vbasedev->bcontainer;
523     VFIOAddressSpace *space = bcontainer->space;
524     VFIOIOMMUFDContainer *container = container_of(bcontainer,
525                                                    VFIOIOMMUFDContainer,
526                                                    bcontainer);
527     QLIST_REMOVE(vbasedev, global_next);
528     QLIST_REMOVE(vbasedev, container_next);
529     vbasedev->bcontainer = NULL;
530 
531     if (!vbasedev->ram_block_discard_allowed) {
532         iommufd_cdev_ram_block_discard_disable(false);
533     }
534 
535     vfio_cpr_unregister_container(bcontainer);
536     iommufd_cdev_detach_container(vbasedev, container);
537     iommufd_cdev_container_destroy(container);
538     vfio_put_address_space(space);
539 
540     iommufd_cdev_unbind_and_disconnect(vbasedev);
541     close(vbasedev->fd);
542 }
543 
544 static VFIODevice *iommufd_cdev_pci_find_by_devid(__u32 devid)
545 {
546     VFIODevice *vbasedev_iter;
547     const VFIOIOMMUClass *iommufd_vioc =
548         VFIO_IOMMU_CLASS(object_class_by_name(TYPE_VFIO_IOMMU_IOMMUFD));
549 
550     QLIST_FOREACH(vbasedev_iter, &vfio_device_list, global_next) {
551         if (VFIO_IOMMU_GET_CLASS(vbasedev_iter->bcontainer) != iommufd_vioc) {
552             continue;
553         }
554         if (devid == vbasedev_iter->devid) {
555             return vbasedev_iter;
556         }
557     }
558     return NULL;
559 }
560 
561 static VFIOPCIDevice *
562 iommufd_cdev_dep_get_realized_vpdev(struct vfio_pci_dependent_device *dep_dev,
563                                     VFIODevice *reset_dev)
564 {
565     VFIODevice *vbasedev_tmp;
566 
567     if (dep_dev->devid == reset_dev->devid ||
568         dep_dev->devid == VFIO_PCI_DEVID_OWNED) {
569         return NULL;
570     }
571 
572     vbasedev_tmp = iommufd_cdev_pci_find_by_devid(dep_dev->devid);
573     if (!vbasedev_tmp || !vbasedev_tmp->dev->realized ||
574         vbasedev_tmp->type != VFIO_DEVICE_TYPE_PCI) {
575         return NULL;
576     }
577 
578     return container_of(vbasedev_tmp, VFIOPCIDevice, vbasedev);
579 }
580 
581 static int iommufd_cdev_pci_hot_reset(VFIODevice *vbasedev, bool single)
582 {
583     VFIOPCIDevice *vdev = container_of(vbasedev, VFIOPCIDevice, vbasedev);
584     struct vfio_pci_hot_reset_info *info = NULL;
585     struct vfio_pci_dependent_device *devices;
586     struct vfio_pci_hot_reset *reset;
587     int ret, i;
588     bool multi = false;
589 
590     trace_vfio_pci_hot_reset(vdev->vbasedev.name, single ? "one" : "multi");
591 
592     if (!single) {
593         vfio_pci_pre_reset(vdev);
594     }
595     vdev->vbasedev.needs_reset = false;
596 
597     ret = vfio_pci_get_pci_hot_reset_info(vdev, &info);
598 
599     if (ret) {
600         goto out_single;
601     }
602 
603     assert(info->flags & VFIO_PCI_HOT_RESET_FLAG_DEV_ID);
604 
605     devices = &info->devices[0];
606 
607     if (!(info->flags & VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED)) {
608         if (!vdev->has_pm_reset) {
609             for (i = 0; i < info->count; i++) {
610                 if (devices[i].devid == VFIO_PCI_DEVID_NOT_OWNED) {
611                     error_report("vfio: Cannot reset device %s, "
612                                  "depends on device %04x:%02x:%02x.%x "
613                                  "which is not owned.",
614                                  vdev->vbasedev.name, devices[i].segment,
615                                  devices[i].bus, PCI_SLOT(devices[i].devfn),
616                                  PCI_FUNC(devices[i].devfn));
617                 }
618             }
619         }
620         ret = -EPERM;
621         goto out_single;
622     }
623 
624     trace_vfio_pci_hot_reset_has_dep_devices(vdev->vbasedev.name);
625 
626     for (i = 0; i < info->count; i++) {
627         VFIOPCIDevice *tmp;
628 
629         trace_iommufd_cdev_pci_hot_reset_dep_devices(devices[i].segment,
630                                                      devices[i].bus,
631                                                      PCI_SLOT(devices[i].devfn),
632                                                      PCI_FUNC(devices[i].devfn),
633                                                      devices[i].devid);
634 
635         /*
636          * If a VFIO cdev device is resettable, all the dependent devices
637          * are either bound to same iommufd or within same iommu_groups as
638          * one of the iommufd bound devices.
639          */
640         assert(devices[i].devid != VFIO_PCI_DEVID_NOT_OWNED);
641 
642         tmp = iommufd_cdev_dep_get_realized_vpdev(&devices[i], &vdev->vbasedev);
643         if (!tmp) {
644             continue;
645         }
646 
647         if (single) {
648             ret = -EINVAL;
649             goto out_single;
650         }
651         vfio_pci_pre_reset(tmp);
652         tmp->vbasedev.needs_reset = false;
653         multi = true;
654     }
655 
656     if (!single && !multi) {
657         ret = -EINVAL;
658         goto out_single;
659     }
660 
661     /* Use zero length array for hot reset with iommufd backend */
662     reset = g_malloc0(sizeof(*reset));
663     reset->argsz = sizeof(*reset);
664 
665      /* Bus reset! */
666     ret = ioctl(vdev->vbasedev.fd, VFIO_DEVICE_PCI_HOT_RESET, reset);
667     g_free(reset);
668     if (ret) {
669         ret = -errno;
670     }
671 
672     trace_vfio_pci_hot_reset_result(vdev->vbasedev.name,
673                                     ret ? strerror(errno) : "Success");
674 
675     /* Re-enable INTx on affected devices */
676     for (i = 0; i < info->count; i++) {
677         VFIOPCIDevice *tmp;
678 
679         tmp = iommufd_cdev_dep_get_realized_vpdev(&devices[i], &vdev->vbasedev);
680         if (!tmp) {
681             continue;
682         }
683         vfio_pci_post_reset(tmp);
684     }
685 out_single:
686     if (!single) {
687         vfio_pci_post_reset(vdev);
688     }
689     g_free(info);
690 
691     return ret;
692 }
693 
694 static void vfio_iommu_iommufd_class_init(ObjectClass *klass, void *data)
695 {
696     VFIOIOMMUClass *vioc = VFIO_IOMMU_CLASS(klass);
697 
698     vioc->hiod_typename = TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO;
699 
700     vioc->dma_map = iommufd_cdev_map;
701     vioc->dma_unmap = iommufd_cdev_unmap;
702     vioc->attach_device = iommufd_cdev_attach;
703     vioc->detach_device = iommufd_cdev_detach;
704     vioc->pci_hot_reset = iommufd_cdev_pci_hot_reset;
705 };
706 
707 static bool hiod_iommufd_vfio_realize(HostIOMMUDevice *hiod, void *opaque,
708                                       Error **errp)
709 {
710     VFIODevice *vdev = opaque;
711     HostIOMMUDeviceCaps *caps = &hiod->caps;
712     enum iommu_hw_info_type type;
713     union {
714         struct iommu_hw_info_vtd vtd;
715     } data;
716     uint64_t hw_caps;
717 
718     hiod->agent = opaque;
719 
720     if (!iommufd_backend_get_device_info(vdev->iommufd, vdev->devid,
721                                          &type, &data, sizeof(data),
722                                          &hw_caps, errp)) {
723         return false;
724     }
725 
726     hiod->name = g_strdup(vdev->name);
727     caps->type = type;
728 
729     return true;
730 }
731 
732 static GList *
733 hiod_iommufd_vfio_get_iova_ranges(HostIOMMUDevice *hiod)
734 {
735     VFIODevice *vdev = hiod->agent;
736 
737     g_assert(vdev);
738     return vfio_container_get_iova_ranges(vdev->bcontainer);
739 }
740 
741 static uint64_t
742 hiod_iommufd_vfio_get_page_size_mask(HostIOMMUDevice *hiod)
743 {
744     VFIODevice *vdev = hiod->agent;
745 
746     g_assert(vdev);
747     return vfio_container_get_page_size_mask(vdev->bcontainer);
748 }
749 
750 
751 static void hiod_iommufd_vfio_class_init(ObjectClass *oc, void *data)
752 {
753     HostIOMMUDeviceClass *hiodc = HOST_IOMMU_DEVICE_CLASS(oc);
754 
755     hiodc->realize = hiod_iommufd_vfio_realize;
756     hiodc->get_iova_ranges = hiod_iommufd_vfio_get_iova_ranges;
757     hiodc->get_page_size_mask = hiod_iommufd_vfio_get_page_size_mask;
758 };
759 
760 static const TypeInfo types[] = {
761     {
762         .name = TYPE_VFIO_IOMMU_IOMMUFD,
763         .parent = TYPE_VFIO_IOMMU,
764         .instance_size = sizeof(VFIOIOMMUFDContainer),
765         .class_init = vfio_iommu_iommufd_class_init,
766     }, {
767         .name = TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO,
768         .parent = TYPE_HOST_IOMMU_DEVICE_IOMMUFD,
769         .class_init = hiod_iommufd_vfio_class_init,
770     }
771 };
772 
773 DEFINE_TYPES(types)
774