xref: /qemu/hw/vfio/iommufd.c (revision ec7e5a90fea996f04ea24e81b680a87bc975354a)
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-device.h"
19 #include "qemu/error-report.h"
20 #include "trace.h"
21 #include "qapi/error.h"
22 #include "system/iommufd.h"
23 #include "hw/qdev-core.h"
24 #include "system/reset.h"
25 #include "qemu/cutils.h"
26 #include "qemu/chardev_open.h"
27 #include "pci.h"
28 #include "vfio-iommufd.h"
29 #include "vfio-helpers.h"
30 #include "vfio-cpr.h"
31 #include "vfio-listener.h"
32 
33 #define TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO             \
34             TYPE_HOST_IOMMU_DEVICE_IOMMUFD "-vfio"
35 
36 static int iommufd_cdev_map(const VFIOContainerBase *bcontainer, hwaddr iova,
37                             ram_addr_t size, void *vaddr, bool readonly)
38 {
39     const VFIOIOMMUFDContainer *container =
40         container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
41 
42     return iommufd_backend_map_dma(container->be,
43                                    container->ioas_id,
44                                    iova, size, vaddr, readonly);
45 }
46 
47 static int iommufd_cdev_unmap(const VFIOContainerBase *bcontainer,
48                               hwaddr iova, ram_addr_t size,
49                               IOMMUTLBEntry *iotlb, bool unmap_all)
50 {
51     const VFIOIOMMUFDContainer *container =
52         container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
53 
54     /* unmap in halves */
55     if (unmap_all) {
56         Int128 llsize = int128_rshift(int128_2_64(), 1);
57         int ret;
58 
59         ret = iommufd_backend_unmap_dma(container->be, container->ioas_id,
60                                         0, int128_get64(llsize));
61 
62         if (ret == 0) {
63             ret = iommufd_backend_unmap_dma(container->be, container->ioas_id,
64                                             int128_get64(llsize),
65                                             int128_get64(llsize));
66         }
67 
68         return ret;
69     }
70 
71     /* TODO: Handle dma_unmap_bitmap with iotlb args (migration) */
72     return iommufd_backend_unmap_dma(container->be,
73                                      container->ioas_id, iova, size);
74 }
75 
76 static bool iommufd_cdev_kvm_device_add(VFIODevice *vbasedev, Error **errp)
77 {
78     return !vfio_kvm_device_add_fd(vbasedev->fd, errp);
79 }
80 
81 static void iommufd_cdev_kvm_device_del(VFIODevice *vbasedev)
82 {
83     Error *err = NULL;
84 
85     if (vfio_kvm_device_del_fd(vbasedev->fd, &err)) {
86         error_report_err(err);
87     }
88 }
89 
90 static bool iommufd_cdev_connect_and_bind(VFIODevice *vbasedev, Error **errp)
91 {
92     IOMMUFDBackend *iommufd = vbasedev->iommufd;
93     struct vfio_device_bind_iommufd bind = {
94         .argsz = sizeof(bind),
95         .flags = 0,
96     };
97 
98     if (!iommufd_backend_connect(iommufd, errp)) {
99         return false;
100     }
101 
102     /*
103      * Add device to kvm-vfio to be prepared for the tracking
104      * in KVM. Especially for some emulated devices, it requires
105      * to have kvm information in the device open.
106      */
107     if (!iommufd_cdev_kvm_device_add(vbasedev, errp)) {
108         goto err_kvm_device_add;
109     }
110 
111     /* Bind device to iommufd */
112     bind.iommufd = iommufd->fd;
113     if (ioctl(vbasedev->fd, VFIO_DEVICE_BIND_IOMMUFD, &bind)) {
114         error_setg_errno(errp, errno, "error bind device fd=%d to iommufd=%d",
115                          vbasedev->fd, bind.iommufd);
116         goto err_bind;
117     }
118 
119     vbasedev->devid = bind.out_devid;
120     trace_iommufd_cdev_connect_and_bind(bind.iommufd, vbasedev->name,
121                                         vbasedev->fd, vbasedev->devid);
122     return true;
123 err_bind:
124     iommufd_cdev_kvm_device_del(vbasedev);
125 err_kvm_device_add:
126     iommufd_backend_disconnect(iommufd);
127     return false;
128 }
129 
130 static void iommufd_cdev_unbind_and_disconnect(VFIODevice *vbasedev)
131 {
132     /* Unbind is automatically conducted when device fd is closed */
133     iommufd_cdev_kvm_device_del(vbasedev);
134     iommufd_backend_disconnect(vbasedev->iommufd);
135 }
136 
137 static bool iommufd_hwpt_dirty_tracking(VFIOIOASHwpt *hwpt)
138 {
139     return hwpt && hwpt->hwpt_flags & IOMMU_HWPT_ALLOC_DIRTY_TRACKING;
140 }
141 
142 static int iommufd_set_dirty_page_tracking(const VFIOContainerBase *bcontainer,
143                                            bool start, Error **errp)
144 {
145     const VFIOIOMMUFDContainer *container =
146         container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
147     VFIOIOASHwpt *hwpt;
148 
149     QLIST_FOREACH(hwpt, &container->hwpt_list, next) {
150         if (!iommufd_hwpt_dirty_tracking(hwpt)) {
151             continue;
152         }
153 
154         if (!iommufd_backend_set_dirty_tracking(container->be,
155                                                 hwpt->hwpt_id, start, errp)) {
156             goto err;
157         }
158     }
159 
160     return 0;
161 
162 err:
163     QLIST_FOREACH(hwpt, &container->hwpt_list, next) {
164         if (!iommufd_hwpt_dirty_tracking(hwpt)) {
165             continue;
166         }
167         iommufd_backend_set_dirty_tracking(container->be,
168                                            hwpt->hwpt_id, !start, NULL);
169     }
170     return -EINVAL;
171 }
172 
173 static int iommufd_query_dirty_bitmap(const VFIOContainerBase *bcontainer,
174                                       VFIOBitmap *vbmap, hwaddr iova,
175                                       hwaddr size, Error **errp)
176 {
177     VFIOIOMMUFDContainer *container = container_of(bcontainer,
178                                                    VFIOIOMMUFDContainer,
179                                                    bcontainer);
180     unsigned long page_size = qemu_real_host_page_size();
181     VFIOIOASHwpt *hwpt;
182 
183     QLIST_FOREACH(hwpt, &container->hwpt_list, next) {
184         if (!iommufd_hwpt_dirty_tracking(hwpt)) {
185             continue;
186         }
187 
188         if (!iommufd_backend_get_dirty_bitmap(container->be, hwpt->hwpt_id,
189                                               iova, size, page_size,
190                                               (uint64_t *)vbmap->bitmap,
191                                               errp)) {
192             return -EINVAL;
193         }
194     }
195 
196     return 0;
197 }
198 
199 static int iommufd_cdev_getfd(const char *sysfs_path, Error **errp)
200 {
201     ERRP_GUARD();
202     long int ret = -ENOTTY;
203     g_autofree char *path = NULL;
204     g_autofree char *vfio_dev_path = NULL;
205     g_autofree char *vfio_path = NULL;
206     DIR *dir = NULL;
207     struct dirent *dent;
208     g_autofree gchar *contents = NULL;
209     gsize length;
210     int major, minor;
211     dev_t vfio_devt;
212 
213     path = g_strdup_printf("%s/vfio-dev", sysfs_path);
214     dir = opendir(path);
215     if (!dir) {
216         error_setg_errno(errp, errno, "couldn't open directory %s", path);
217         goto out;
218     }
219 
220     while ((dent = readdir(dir))) {
221         if (!strncmp(dent->d_name, "vfio", 4)) {
222             vfio_dev_path = g_strdup_printf("%s/%s/dev", path, dent->d_name);
223             break;
224         }
225     }
226 
227     if (!vfio_dev_path) {
228         error_setg(errp, "failed to find vfio-dev/vfioX/dev");
229         goto out_close_dir;
230     }
231 
232     if (!g_file_get_contents(vfio_dev_path, &contents, &length, NULL)) {
233         error_setg(errp, "failed to load \"%s\"", vfio_dev_path);
234         goto out_close_dir;
235     }
236 
237     if (sscanf(contents, "%d:%d", &major, &minor) != 2) {
238         error_setg(errp, "failed to get major:minor for \"%s\"", vfio_dev_path);
239         goto out_close_dir;
240     }
241     vfio_devt = makedev(major, minor);
242 
243     vfio_path = g_strdup_printf("/dev/vfio/devices/%s", dent->d_name);
244     ret = open_cdev(vfio_path, vfio_devt);
245     if (ret < 0) {
246         error_setg(errp, "Failed to open %s", vfio_path);
247     }
248 
249     trace_iommufd_cdev_getfd(vfio_path, ret);
250 
251 out_close_dir:
252     closedir(dir);
253 out:
254     if (*errp) {
255         error_prepend(errp, VFIO_MSG_PREFIX, path);
256     }
257 
258     return ret;
259 }
260 
261 static int iommufd_cdev_attach_ioas_hwpt(VFIODevice *vbasedev, uint32_t id,
262                                          Error **errp)
263 {
264     int iommufd = vbasedev->iommufd->fd;
265     struct vfio_device_attach_iommufd_pt attach_data = {
266         .argsz = sizeof(attach_data),
267         .flags = 0,
268         .pt_id = id,
269     };
270 
271     /* Attach device to an IOAS or hwpt within iommufd */
272     if (ioctl(vbasedev->fd, VFIO_DEVICE_ATTACH_IOMMUFD_PT, &attach_data)) {
273         error_setg_errno(errp, errno,
274                          "[iommufd=%d] error attach %s (%d) to id=%d",
275                          iommufd, vbasedev->name, vbasedev->fd, id);
276         return -errno;
277     }
278 
279     trace_iommufd_cdev_attach_ioas_hwpt(iommufd, vbasedev->name,
280                                         vbasedev->fd, id);
281     return 0;
282 }
283 
284 static bool iommufd_cdev_detach_ioas_hwpt(VFIODevice *vbasedev, Error **errp)
285 {
286     int iommufd = vbasedev->iommufd->fd;
287     struct vfio_device_detach_iommufd_pt detach_data = {
288         .argsz = sizeof(detach_data),
289         .flags = 0,
290     };
291 
292     if (ioctl(vbasedev->fd, VFIO_DEVICE_DETACH_IOMMUFD_PT, &detach_data)) {
293         error_setg_errno(errp, errno, "detach %s failed", vbasedev->name);
294         return false;
295     }
296 
297     trace_iommufd_cdev_detach_ioas_hwpt(iommufd, vbasedev->name);
298     return true;
299 }
300 
301 static bool iommufd_cdev_autodomains_get(VFIODevice *vbasedev,
302                                          VFIOIOMMUFDContainer *container,
303                                          Error **errp)
304 {
305     ERRP_GUARD();
306     IOMMUFDBackend *iommufd = vbasedev->iommufd;
307     uint32_t type, flags = 0;
308     uint64_t hw_caps;
309     VFIOIOASHwpt *hwpt;
310     uint32_t hwpt_id;
311     int ret;
312 
313     /* Try to find a domain */
314     QLIST_FOREACH(hwpt, &container->hwpt_list, next) {
315         ret = iommufd_cdev_attach_ioas_hwpt(vbasedev, hwpt->hwpt_id, errp);
316         if (ret) {
317             /* -EINVAL means the domain is incompatible with the device. */
318             if (ret == -EINVAL) {
319                 /*
320                  * It is an expected failure and it just means we will try
321                  * another domain, or create one if no existing compatible
322                  * domain is found. Hence why the error is discarded below.
323                  */
324                 error_free(*errp);
325                 *errp = NULL;
326                 continue;
327             }
328 
329             return false;
330         } else {
331             vbasedev->hwpt = hwpt;
332             QLIST_INSERT_HEAD(&hwpt->device_list, vbasedev, hwpt_next);
333             vbasedev->iommu_dirty_tracking = iommufd_hwpt_dirty_tracking(hwpt);
334             return true;
335         }
336     }
337 
338     /*
339      * This is quite early and VFIO Migration state isn't yet fully
340      * initialized, thus rely only on IOMMU hardware capabilities as to
341      * whether IOMMU dirty tracking is going to be requested. Later
342      * vfio_migration_realize() may decide to use VF dirty tracking
343      * instead.
344      */
345     if (!iommufd_backend_get_device_info(vbasedev->iommufd, vbasedev->devid,
346                                          &type, NULL, 0, &hw_caps, errp)) {
347         return false;
348     }
349 
350     if (hw_caps & IOMMU_HW_CAP_DIRTY_TRACKING) {
351         flags = IOMMU_HWPT_ALLOC_DIRTY_TRACKING;
352     }
353 
354     if (!iommufd_backend_alloc_hwpt(iommufd, vbasedev->devid,
355                                     container->ioas_id, flags,
356                                     IOMMU_HWPT_DATA_NONE, 0, NULL,
357                                     &hwpt_id, errp)) {
358         return false;
359     }
360 
361     hwpt = g_malloc0(sizeof(*hwpt));
362     hwpt->hwpt_id = hwpt_id;
363     hwpt->hwpt_flags = flags;
364     QLIST_INIT(&hwpt->device_list);
365 
366     ret = iommufd_cdev_attach_ioas_hwpt(vbasedev, hwpt->hwpt_id, errp);
367     if (ret) {
368         iommufd_backend_free_id(container->be, hwpt->hwpt_id);
369         g_free(hwpt);
370         return false;
371     }
372 
373     vbasedev->hwpt = hwpt;
374     vbasedev->iommu_dirty_tracking = iommufd_hwpt_dirty_tracking(hwpt);
375     QLIST_INSERT_HEAD(&hwpt->device_list, vbasedev, hwpt_next);
376     QLIST_INSERT_HEAD(&container->hwpt_list, hwpt, next);
377     container->bcontainer.dirty_pages_supported |=
378                                 vbasedev->iommu_dirty_tracking;
379     if (container->bcontainer.dirty_pages_supported &&
380         !vbasedev->iommu_dirty_tracking) {
381         warn_report("IOMMU instance for device %s doesn't support dirty tracking",
382                     vbasedev->name);
383     }
384     return true;
385 }
386 
387 static void iommufd_cdev_autodomains_put(VFIODevice *vbasedev,
388                                          VFIOIOMMUFDContainer *container)
389 {
390     VFIOIOASHwpt *hwpt = vbasedev->hwpt;
391 
392     QLIST_REMOVE(vbasedev, hwpt_next);
393     vbasedev->hwpt = NULL;
394 
395     if (QLIST_EMPTY(&hwpt->device_list)) {
396         QLIST_REMOVE(hwpt, next);
397         iommufd_backend_free_id(container->be, hwpt->hwpt_id);
398         g_free(hwpt);
399     }
400 }
401 
402 static bool iommufd_cdev_attach_container(VFIODevice *vbasedev,
403                                           VFIOIOMMUFDContainer *container,
404                                           Error **errp)
405 {
406     /* mdevs aren't physical devices and will fail with auto domains */
407     if (!vbasedev->mdev) {
408         return iommufd_cdev_autodomains_get(vbasedev, container, errp);
409     }
410 
411     return !iommufd_cdev_attach_ioas_hwpt(vbasedev, container->ioas_id, errp);
412 }
413 
414 static void iommufd_cdev_detach_container(VFIODevice *vbasedev,
415                                           VFIOIOMMUFDContainer *container)
416 {
417     Error *err = NULL;
418 
419     if (!iommufd_cdev_detach_ioas_hwpt(vbasedev, &err)) {
420         error_report_err(err);
421     }
422 
423     if (vbasedev->hwpt) {
424         iommufd_cdev_autodomains_put(vbasedev, container);
425     }
426 
427 }
428 
429 static void iommufd_cdev_container_destroy(VFIOIOMMUFDContainer *container)
430 {
431     VFIOContainerBase *bcontainer = &container->bcontainer;
432 
433     if (!QLIST_EMPTY(&bcontainer->device_list)) {
434         return;
435     }
436     vfio_cpr_unregister_container(bcontainer);
437     vfio_listener_unregister(bcontainer);
438     iommufd_backend_free_id(container->be, container->ioas_id);
439     object_unref(container);
440 }
441 
442 static int iommufd_cdev_ram_block_discard_disable(bool state)
443 {
444     /*
445      * We support coordinated discarding of RAM via the RamDiscardManager.
446      */
447     return ram_block_uncoordinated_discard_disable(state);
448 }
449 
450 static bool iommufd_cdev_get_info_iova_range(VFIOIOMMUFDContainer *container,
451                                              uint32_t ioas_id, Error **errp)
452 {
453     VFIOContainerBase *bcontainer = &container->bcontainer;
454     g_autofree struct iommu_ioas_iova_ranges *info = NULL;
455     struct iommu_iova_range *iova_ranges;
456     int sz, fd = container->be->fd;
457 
458     info = g_malloc0(sizeof(*info));
459     info->size = sizeof(*info);
460     info->ioas_id = ioas_id;
461 
462     if (ioctl(fd, IOMMU_IOAS_IOVA_RANGES, info) && errno != EMSGSIZE) {
463         goto error;
464     }
465 
466     sz = info->num_iovas * sizeof(struct iommu_iova_range);
467     info = g_realloc(info, sizeof(*info) + sz);
468     info->allowed_iovas = (uintptr_t)(info + 1);
469 
470     if (ioctl(fd, IOMMU_IOAS_IOVA_RANGES, info)) {
471         goto error;
472     }
473 
474     iova_ranges = (struct iommu_iova_range *)(uintptr_t)info->allowed_iovas;
475 
476     for (int i = 0; i < info->num_iovas; i++) {
477         Range *range = g_new(Range, 1);
478 
479         range_set_bounds(range, iova_ranges[i].start, iova_ranges[i].last);
480         bcontainer->iova_ranges =
481             range_list_insert(bcontainer->iova_ranges, range);
482     }
483     bcontainer->pgsizes = info->out_iova_alignment;
484 
485     return true;
486 
487 error:
488     error_setg_errno(errp, errno, "Cannot get IOVA ranges");
489     return false;
490 }
491 
492 static bool iommufd_cdev_attach(const char *name, VFIODevice *vbasedev,
493                                 AddressSpace *as, Error **errp)
494 {
495     VFIOContainerBase *bcontainer;
496     VFIOIOMMUFDContainer *container;
497     VFIOAddressSpace *space;
498     struct vfio_device_info dev_info = { .argsz = sizeof(dev_info) };
499     int ret, devfd;
500     uint32_t ioas_id;
501     Error *err = NULL;
502     const VFIOIOMMUClass *iommufd_vioc =
503         VFIO_IOMMU_CLASS(object_class_by_name(TYPE_VFIO_IOMMU_IOMMUFD));
504 
505     if (vbasedev->fd < 0) {
506         devfd = iommufd_cdev_getfd(vbasedev->sysfsdev, errp);
507         if (devfd < 0) {
508             return false;
509         }
510         vbasedev->fd = devfd;
511     } else {
512         devfd = vbasedev->fd;
513     }
514 
515     if (!iommufd_cdev_connect_and_bind(vbasedev, errp)) {
516         goto err_connect_bind;
517     }
518 
519     space = vfio_address_space_get(as);
520 
521     /* try to attach to an existing container in this space */
522     QLIST_FOREACH(bcontainer, &space->containers, next) {
523         container = container_of(bcontainer, VFIOIOMMUFDContainer, bcontainer);
524         if (VFIO_IOMMU_GET_CLASS(bcontainer) != iommufd_vioc ||
525             vbasedev->iommufd != container->be) {
526             continue;
527         }
528         if (!iommufd_cdev_attach_container(vbasedev, container, &err)) {
529             const char *msg = error_get_pretty(err);
530 
531             trace_iommufd_cdev_fail_attach_existing_container(msg);
532             error_free(err);
533             err = NULL;
534         } else {
535             ret = iommufd_cdev_ram_block_discard_disable(true);
536             if (ret) {
537                 error_setg_errno(errp, -ret,
538                                  "Cannot set discarding of RAM broken");
539                 goto err_discard_disable;
540             }
541             goto found_container;
542         }
543     }
544 
545     /* Need to allocate a new dedicated container */
546     if (!iommufd_backend_alloc_ioas(vbasedev->iommufd, &ioas_id, errp)) {
547         goto err_alloc_ioas;
548     }
549 
550     trace_iommufd_cdev_alloc_ioas(vbasedev->iommufd->fd, ioas_id);
551 
552     container = VFIO_IOMMU_IOMMUFD(object_new(TYPE_VFIO_IOMMU_IOMMUFD));
553     container->be = vbasedev->iommufd;
554     container->ioas_id = ioas_id;
555     QLIST_INIT(&container->hwpt_list);
556 
557     bcontainer = &container->bcontainer;
558     vfio_address_space_insert(space, bcontainer);
559 
560     if (!iommufd_cdev_attach_container(vbasedev, container, errp)) {
561         goto err_attach_container;
562     }
563 
564     ret = iommufd_cdev_ram_block_discard_disable(true);
565     if (ret) {
566         error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
567         goto err_discard_disable;
568     }
569 
570     if (!iommufd_cdev_get_info_iova_range(container, ioas_id, &err)) {
571         error_append_hint(&err,
572                    "Fallback to default 64bit IOVA range and 4K page size\n");
573         warn_report_err(err);
574         err = NULL;
575         bcontainer->pgsizes = qemu_real_host_page_size();
576     }
577 
578     if (!vfio_listener_register(bcontainer, errp)) {
579         goto err_listener_register;
580     }
581 
582     if (!vfio_cpr_register_container(bcontainer, errp)) {
583         goto err_listener_register;
584     }
585 
586     bcontainer->initialized = true;
587 
588 found_container:
589     ret = ioctl(devfd, VFIO_DEVICE_GET_INFO, &dev_info);
590     if (ret) {
591         error_setg_errno(errp, errno, "error getting device info");
592         goto err_listener_register;
593     }
594 
595     if (!vfio_device_hiod_create_and_realize(vbasedev,
596                      TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO, errp)) {
597         goto err_listener_register;
598     }
599 
600     /*
601      * TODO: examine RAM_BLOCK_DISCARD stuff, should we do group level
602      * for discarding incompatibility check as well?
603      */
604     if (vbasedev->ram_block_discard_allowed) {
605         iommufd_cdev_ram_block_discard_disable(false);
606     }
607 
608     vfio_device_prepare(vbasedev, bcontainer, &dev_info);
609 
610     trace_iommufd_cdev_device_info(vbasedev->name, devfd, vbasedev->num_irqs,
611                                    vbasedev->num_regions, vbasedev->flags);
612     return true;
613 
614 err_listener_register:
615     iommufd_cdev_ram_block_discard_disable(false);
616 err_discard_disable:
617     iommufd_cdev_detach_container(vbasedev, container);
618 err_attach_container:
619     iommufd_cdev_container_destroy(container);
620 err_alloc_ioas:
621     vfio_address_space_put(space);
622     iommufd_cdev_unbind_and_disconnect(vbasedev);
623 err_connect_bind:
624     close(vbasedev->fd);
625     return false;
626 }
627 
628 static void iommufd_cdev_detach(VFIODevice *vbasedev)
629 {
630     VFIOContainerBase *bcontainer = vbasedev->bcontainer;
631     VFIOAddressSpace *space = bcontainer->space;
632     VFIOIOMMUFDContainer *container = container_of(bcontainer,
633                                                    VFIOIOMMUFDContainer,
634                                                    bcontainer);
635     vfio_device_unprepare(vbasedev);
636 
637     if (!vbasedev->ram_block_discard_allowed) {
638         iommufd_cdev_ram_block_discard_disable(false);
639     }
640 
641     object_unref(vbasedev->hiod);
642     iommufd_cdev_detach_container(vbasedev, container);
643     iommufd_cdev_container_destroy(container);
644     vfio_address_space_put(space);
645 
646     iommufd_cdev_unbind_and_disconnect(vbasedev);
647     close(vbasedev->fd);
648 }
649 
650 static VFIODevice *iommufd_cdev_pci_find_by_devid(__u32 devid)
651 {
652     VFIODevice *vbasedev_iter;
653     const VFIOIOMMUClass *iommufd_vioc =
654         VFIO_IOMMU_CLASS(object_class_by_name(TYPE_VFIO_IOMMU_IOMMUFD));
655 
656     QLIST_FOREACH(vbasedev_iter, &vfio_device_list, global_next) {
657         if (VFIO_IOMMU_GET_CLASS(vbasedev_iter->bcontainer) != iommufd_vioc) {
658             continue;
659         }
660         if (devid == vbasedev_iter->devid) {
661             return vbasedev_iter;
662         }
663     }
664     return NULL;
665 }
666 
667 static VFIOPCIDevice *
668 iommufd_cdev_dep_get_realized_vpdev(struct vfio_pci_dependent_device *dep_dev,
669                                     VFIODevice *reset_dev)
670 {
671     VFIODevice *vbasedev_tmp;
672 
673     if (dep_dev->devid == reset_dev->devid ||
674         dep_dev->devid == VFIO_PCI_DEVID_OWNED) {
675         return NULL;
676     }
677 
678     vbasedev_tmp = iommufd_cdev_pci_find_by_devid(dep_dev->devid);
679     if (!vbasedev_tmp || !vbasedev_tmp->dev->realized ||
680         vbasedev_tmp->type != VFIO_DEVICE_TYPE_PCI) {
681         return NULL;
682     }
683 
684     return container_of(vbasedev_tmp, VFIOPCIDevice, vbasedev);
685 }
686 
687 static int iommufd_cdev_pci_hot_reset(VFIODevice *vbasedev, bool single)
688 {
689     VFIOPCIDevice *vdev = container_of(vbasedev, VFIOPCIDevice, vbasedev);
690     struct vfio_pci_hot_reset_info *info = NULL;
691     struct vfio_pci_dependent_device *devices;
692     struct vfio_pci_hot_reset *reset;
693     int ret, i;
694     bool multi = false;
695 
696     trace_vfio_pci_hot_reset(vdev->vbasedev.name, single ? "one" : "multi");
697 
698     if (!single) {
699         vfio_pci_pre_reset(vdev);
700     }
701     vdev->vbasedev.needs_reset = false;
702 
703     ret = vfio_pci_get_pci_hot_reset_info(vdev, &info);
704 
705     if (ret) {
706         goto out_single;
707     }
708 
709     assert(info->flags & VFIO_PCI_HOT_RESET_FLAG_DEV_ID);
710 
711     devices = &info->devices[0];
712 
713     if (!(info->flags & VFIO_PCI_HOT_RESET_FLAG_DEV_ID_OWNED)) {
714         if (!vdev->has_pm_reset) {
715             for (i = 0; i < info->count; i++) {
716                 if (devices[i].devid == VFIO_PCI_DEVID_NOT_OWNED) {
717                     error_report("vfio: Cannot reset device %s, "
718                                  "depends on device %04x:%02x:%02x.%x "
719                                  "which is not owned.",
720                                  vdev->vbasedev.name, devices[i].segment,
721                                  devices[i].bus, PCI_SLOT(devices[i].devfn),
722                                  PCI_FUNC(devices[i].devfn));
723                 }
724             }
725         }
726         ret = -EPERM;
727         goto out_single;
728     }
729 
730     trace_vfio_pci_hot_reset_has_dep_devices(vdev->vbasedev.name);
731 
732     for (i = 0; i < info->count; i++) {
733         VFIOPCIDevice *tmp;
734 
735         trace_iommufd_cdev_pci_hot_reset_dep_devices(devices[i].segment,
736                                                      devices[i].bus,
737                                                      PCI_SLOT(devices[i].devfn),
738                                                      PCI_FUNC(devices[i].devfn),
739                                                      devices[i].devid);
740 
741         /*
742          * If a VFIO cdev device is resettable, all the dependent devices
743          * are either bound to same iommufd or within same iommu_groups as
744          * one of the iommufd bound devices.
745          */
746         assert(devices[i].devid != VFIO_PCI_DEVID_NOT_OWNED);
747 
748         tmp = iommufd_cdev_dep_get_realized_vpdev(&devices[i], &vdev->vbasedev);
749         if (!tmp) {
750             continue;
751         }
752 
753         if (single) {
754             ret = -EINVAL;
755             goto out_single;
756         }
757         vfio_pci_pre_reset(tmp);
758         tmp->vbasedev.needs_reset = false;
759         multi = true;
760     }
761 
762     if (!single && !multi) {
763         ret = -EINVAL;
764         goto out_single;
765     }
766 
767     /* Use zero length array for hot reset with iommufd backend */
768     reset = g_malloc0(sizeof(*reset));
769     reset->argsz = sizeof(*reset);
770 
771      /* Bus reset! */
772     ret = ioctl(vdev->vbasedev.fd, VFIO_DEVICE_PCI_HOT_RESET, reset);
773     g_free(reset);
774     if (ret) {
775         ret = -errno;
776     }
777 
778     trace_vfio_pci_hot_reset_result(vdev->vbasedev.name,
779                                     ret ? strerror(errno) : "Success");
780 
781     /* Re-enable INTx on affected devices */
782     for (i = 0; i < info->count; i++) {
783         VFIOPCIDevice *tmp;
784 
785         tmp = iommufd_cdev_dep_get_realized_vpdev(&devices[i], &vdev->vbasedev);
786         if (!tmp) {
787             continue;
788         }
789         vfio_pci_post_reset(tmp);
790     }
791 out_single:
792     if (!single) {
793         vfio_pci_post_reset(vdev);
794     }
795     g_free(info);
796 
797     return ret;
798 }
799 
800 static void vfio_iommu_iommufd_class_init(ObjectClass *klass, const void *data)
801 {
802     VFIOIOMMUClass *vioc = VFIO_IOMMU_CLASS(klass);
803 
804     vioc->dma_map = iommufd_cdev_map;
805     vioc->dma_unmap = iommufd_cdev_unmap;
806     vioc->attach_device = iommufd_cdev_attach;
807     vioc->detach_device = iommufd_cdev_detach;
808     vioc->pci_hot_reset = iommufd_cdev_pci_hot_reset;
809     vioc->set_dirty_page_tracking = iommufd_set_dirty_page_tracking;
810     vioc->query_dirty_bitmap = iommufd_query_dirty_bitmap;
811 };
812 
813 static bool hiod_iommufd_vfio_realize(HostIOMMUDevice *hiod, void *opaque,
814                                       Error **errp)
815 {
816     VFIODevice *vdev = opaque;
817     HostIOMMUDeviceCaps *caps = &hiod->caps;
818     enum iommu_hw_info_type type;
819     union {
820         struct iommu_hw_info_vtd vtd;
821     } data;
822     uint64_t hw_caps;
823 
824     hiod->agent = opaque;
825 
826     if (!iommufd_backend_get_device_info(vdev->iommufd, vdev->devid,
827                                          &type, &data, sizeof(data),
828                                          &hw_caps, errp)) {
829         return false;
830     }
831 
832     hiod->name = g_strdup(vdev->name);
833     caps->type = type;
834     caps->hw_caps = hw_caps;
835 
836     return true;
837 }
838 
839 static GList *
840 hiod_iommufd_vfio_get_iova_ranges(HostIOMMUDevice *hiod)
841 {
842     VFIODevice *vdev = hiod->agent;
843 
844     g_assert(vdev);
845     return vfio_container_get_iova_ranges(vdev->bcontainer);
846 }
847 
848 static uint64_t
849 hiod_iommufd_vfio_get_page_size_mask(HostIOMMUDevice *hiod)
850 {
851     VFIODevice *vdev = hiod->agent;
852 
853     g_assert(vdev);
854     return vfio_container_get_page_size_mask(vdev->bcontainer);
855 }
856 
857 
858 static void hiod_iommufd_vfio_class_init(ObjectClass *oc, const void *data)
859 {
860     HostIOMMUDeviceClass *hiodc = HOST_IOMMU_DEVICE_CLASS(oc);
861 
862     hiodc->realize = hiod_iommufd_vfio_realize;
863     hiodc->get_iova_ranges = hiod_iommufd_vfio_get_iova_ranges;
864     hiodc->get_page_size_mask = hiod_iommufd_vfio_get_page_size_mask;
865 };
866 
867 static const TypeInfo types[] = {
868     {
869         .name = TYPE_VFIO_IOMMU_IOMMUFD,
870         .parent = TYPE_VFIO_IOMMU,
871         .instance_size = sizeof(VFIOIOMMUFDContainer),
872         .class_init = vfio_iommu_iommufd_class_init,
873     }, {
874         .name = TYPE_HOST_IOMMU_DEVICE_IOMMUFD_VFIO,
875         .parent = TYPE_HOST_IOMMU_DEVICE_IOMMUFD,
876         .class_init = hiod_iommufd_vfio_class_init,
877     }
878 };
879 
880 DEFINE_TYPES(types)
881