1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
8 * Derived from original vfio:
9 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
13 #include <linux/cdev.h>
14 #include <linux/compat.h>
15 #include <linux/device.h>
16 #include <linux/file.h>
17 #include <linux/anon_inodes.h>
19 #include <linux/idr.h>
20 #include <linux/iommu.h>
21 #include <linux/list.h>
22 #include <linux/miscdevice.h>
23 #include <linux/module.h>
24 #include <linux/mutex.h>
25 #include <linux/pci.h>
26 #include <linux/rwsem.h>
27 #include <linux/sched.h>
28 #include <linux/slab.h>
29 #include <linux/stat.h>
30 #include <linux/string.h>
31 #include <linux/uaccess.h>
32 #include <linux/vfio.h>
33 #include <linux/wait.h>
34 #include <linux/sched/signal.h>
35 #include <linux/pm_runtime.h>
36 #include <linux/interval_tree.h>
37 #include <linux/iova_bitmap.h>
40 #define DRIVER_VERSION "0.3"
42 #define DRIVER_DESC "VFIO - User Level meta-driver"
46 struct list_head group_list;
47 struct mutex group_lock; /* locks group_list */
50 struct class *device_class;
51 struct ida device_ida;
54 static DEFINE_XARRAY(vfio_device_set_xa);
55 static const struct file_operations vfio_group_fops;
57 int vfio_assign_device_set(struct vfio_device *device, void *set_id)
59 unsigned long idx = (unsigned long)set_id;
60 struct vfio_device_set *new_dev_set;
61 struct vfio_device_set *dev_set;
67 * Atomically acquire a singleton object in the xarray for this set_id
69 xa_lock(&vfio_device_set_xa);
70 dev_set = xa_load(&vfio_device_set_xa, idx);
73 xa_unlock(&vfio_device_set_xa);
75 new_dev_set = kzalloc(sizeof(*new_dev_set), GFP_KERNEL);
78 mutex_init(&new_dev_set->lock);
79 INIT_LIST_HEAD(&new_dev_set->device_list);
80 new_dev_set->set_id = set_id;
82 xa_lock(&vfio_device_set_xa);
83 dev_set = __xa_cmpxchg(&vfio_device_set_xa, idx, NULL, new_dev_set,
86 dev_set = new_dev_set;
91 if (xa_is_err(dev_set)) {
92 xa_unlock(&vfio_device_set_xa);
93 return xa_err(dev_set);
97 dev_set->device_count++;
98 xa_unlock(&vfio_device_set_xa);
99 mutex_lock(&dev_set->lock);
100 device->dev_set = dev_set;
101 list_add_tail(&device->dev_set_list, &dev_set->device_list);
102 mutex_unlock(&dev_set->lock);
105 EXPORT_SYMBOL_GPL(vfio_assign_device_set);
107 static void vfio_release_device_set(struct vfio_device *device)
109 struct vfio_device_set *dev_set = device->dev_set;
114 mutex_lock(&dev_set->lock);
115 list_del(&device->dev_set_list);
116 mutex_unlock(&dev_set->lock);
118 xa_lock(&vfio_device_set_xa);
119 if (!--dev_set->device_count) {
120 __xa_erase(&vfio_device_set_xa,
121 (unsigned long)dev_set->set_id);
122 mutex_destroy(&dev_set->lock);
125 xa_unlock(&vfio_device_set_xa);
129 * Group objects - create, release, get, put, search
131 static struct vfio_group *
132 __vfio_group_get_from_iommu(struct iommu_group *iommu_group)
134 struct vfio_group *group;
137 * group->iommu_group from the vfio.group_list cannot be NULL
138 * under the vfio.group_lock.
140 list_for_each_entry(group, &vfio.group_list, vfio_next) {
141 if (group->iommu_group == iommu_group) {
142 refcount_inc(&group->drivers);
149 static struct vfio_group *
150 vfio_group_get_from_iommu(struct iommu_group *iommu_group)
152 struct vfio_group *group;
154 mutex_lock(&vfio.group_lock);
155 group = __vfio_group_get_from_iommu(iommu_group);
156 mutex_unlock(&vfio.group_lock);
160 static void vfio_group_release(struct device *dev)
162 struct vfio_group *group = container_of(dev, struct vfio_group, dev);
164 mutex_destroy(&group->device_lock);
165 mutex_destroy(&group->group_lock);
166 WARN_ON(group->iommu_group);
167 ida_free(&vfio.group_ida, MINOR(group->dev.devt));
171 static struct vfio_group *vfio_group_alloc(struct iommu_group *iommu_group,
172 enum vfio_group_type type)
174 struct vfio_group *group;
177 group = kzalloc(sizeof(*group), GFP_KERNEL);
179 return ERR_PTR(-ENOMEM);
181 minor = ida_alloc_max(&vfio.group_ida, MINORMASK, GFP_KERNEL);
184 return ERR_PTR(minor);
187 device_initialize(&group->dev);
188 group->dev.devt = MKDEV(MAJOR(vfio.group_devt), minor);
189 group->dev.class = vfio.class;
190 group->dev.release = vfio_group_release;
191 cdev_init(&group->cdev, &vfio_group_fops);
192 group->cdev.owner = THIS_MODULE;
194 refcount_set(&group->drivers, 1);
195 mutex_init(&group->group_lock);
196 INIT_LIST_HEAD(&group->device_list);
197 mutex_init(&group->device_lock);
198 group->iommu_group = iommu_group;
199 /* put in vfio_group_release() */
200 iommu_group_ref_get(iommu_group);
202 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
207 static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group,
208 enum vfio_group_type type)
210 struct vfio_group *group;
211 struct vfio_group *ret;
214 group = vfio_group_alloc(iommu_group, type);
218 err = dev_set_name(&group->dev, "%s%d",
219 group->type == VFIO_NO_IOMMU ? "noiommu-" : "",
220 iommu_group_id(iommu_group));
226 mutex_lock(&vfio.group_lock);
228 /* Did we race creating this group? */
229 ret = __vfio_group_get_from_iommu(iommu_group);
233 err = cdev_device_add(&group->cdev, &group->dev);
239 list_add(&group->vfio_next, &vfio.group_list);
241 mutex_unlock(&vfio.group_lock);
245 mutex_unlock(&vfio.group_lock);
247 put_device(&group->dev);
251 static void vfio_device_remove_group(struct vfio_device *device)
253 struct vfio_group *group = device->group;
254 struct iommu_group *iommu_group;
256 if (group->type == VFIO_NO_IOMMU || group->type == VFIO_EMULATED_IOMMU)
257 iommu_group_remove_device(device->dev);
259 /* Pairs with vfio_create_group() / vfio_group_get_from_iommu() */
260 if (!refcount_dec_and_mutex_lock(&group->drivers, &vfio.group_lock))
262 list_del(&group->vfio_next);
265 * We could concurrently probe another driver in the group that might
266 * race vfio_device_remove_group() with vfio_get_group(), so we have to
267 * ensure that the sysfs is all cleaned up under lock otherwise the
268 * cdev_device_add() will fail due to the name aready existing.
270 cdev_device_del(&group->cdev, &group->dev);
272 mutex_lock(&group->group_lock);
274 * These data structures all have paired operations that can only be
275 * undone when the caller holds a live reference on the device. Since
276 * all pairs must be undone these WARN_ON's indicate some caller did not
277 * properly hold the group reference.
279 WARN_ON(!list_empty(&group->device_list));
280 WARN_ON(group->notifier.head);
283 * Revoke all users of group->iommu_group. At this point we know there
284 * are no devices active because we are unplugging the last one. Setting
285 * iommu_group to NULL blocks all new users.
287 if (group->container)
288 vfio_group_detach_container(group);
289 iommu_group = group->iommu_group;
290 group->iommu_group = NULL;
291 mutex_unlock(&group->group_lock);
292 mutex_unlock(&vfio.group_lock);
294 iommu_group_put(iommu_group);
295 put_device(&group->dev);
299 * Device objects - create, release, get, put, search
301 /* Device reference always implies a group reference */
302 static void vfio_device_put_registration(struct vfio_device *device)
304 if (refcount_dec_and_test(&device->refcount))
305 complete(&device->comp);
308 static bool vfio_device_try_get_registration(struct vfio_device *device)
310 return refcount_inc_not_zero(&device->refcount);
313 static struct vfio_device *vfio_group_get_device(struct vfio_group *group,
316 struct vfio_device *device;
318 mutex_lock(&group->device_lock);
319 list_for_each_entry(device, &group->device_list, group_next) {
320 if (device->dev == dev &&
321 vfio_device_try_get_registration(device)) {
322 mutex_unlock(&group->device_lock);
326 mutex_unlock(&group->device_lock);
333 /* Release helper called by vfio_put_device() */
334 static void vfio_device_release(struct device *dev)
336 struct vfio_device *device =
337 container_of(dev, struct vfio_device, device);
339 vfio_release_device_set(device);
340 ida_free(&vfio.device_ida, device->index);
343 * kvfree() cannot be done here due to a life cycle mess in
344 * vfio-ccw. Before the ccw part is fixed all drivers are
345 * required to support @release and call vfio_free_device()
348 device->ops->release(device);
352 * Allocate and initialize vfio_device so it can be registered to vfio
355 * Drivers should use the wrapper vfio_alloc_device() for allocation.
356 * @size is the size of the structure to be allocated, including any
357 * private data used by the driver.
359 * Driver may provide an @init callback to cover device private data.
361 * Use vfio_put_device() to release the structure after success return.
363 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
364 const struct vfio_device_ops *ops)
366 struct vfio_device *device;
369 if (WARN_ON(size < sizeof(struct vfio_device)))
370 return ERR_PTR(-EINVAL);
372 device = kvzalloc(size, GFP_KERNEL);
374 return ERR_PTR(-ENOMEM);
376 ret = vfio_init_device(device, dev, ops);
385 EXPORT_SYMBOL_GPL(_vfio_alloc_device);
388 * Initialize a vfio_device so it can be registered to vfio core.
390 * Only vfio-ccw driver should call this interface.
392 int vfio_init_device(struct vfio_device *device, struct device *dev,
393 const struct vfio_device_ops *ops)
397 ret = ida_alloc_max(&vfio.device_ida, MINORMASK, GFP_KERNEL);
399 dev_dbg(dev, "Error to alloc index\n");
404 init_completion(&device->comp);
409 ret = ops->init(device);
414 device_initialize(&device->device);
415 device->device.release = vfio_device_release;
416 device->device.class = vfio.device_class;
417 device->device.parent = device->dev;
421 vfio_release_device_set(device);
422 ida_free(&vfio.device_ida, device->index);
425 EXPORT_SYMBOL_GPL(vfio_init_device);
428 * The helper called by driver @release callback to free the device
429 * structure. Drivers which don't have private data to clean can
430 * simply use this helper as its @release.
432 void vfio_free_device(struct vfio_device *device)
436 EXPORT_SYMBOL_GPL(vfio_free_device);
438 static struct vfio_group *vfio_noiommu_group_alloc(struct device *dev,
439 enum vfio_group_type type)
441 struct iommu_group *iommu_group;
442 struct vfio_group *group;
445 iommu_group = iommu_group_alloc();
446 if (IS_ERR(iommu_group))
447 return ERR_CAST(iommu_group);
449 ret = iommu_group_set_name(iommu_group, "vfio-noiommu");
452 ret = iommu_group_add_device(iommu_group, dev);
456 group = vfio_create_group(iommu_group, type);
458 ret = PTR_ERR(group);
459 goto out_remove_device;
461 iommu_group_put(iommu_group);
465 iommu_group_remove_device(dev);
467 iommu_group_put(iommu_group);
471 static struct vfio_group *vfio_group_find_or_alloc(struct device *dev)
473 struct iommu_group *iommu_group;
474 struct vfio_group *group;
476 iommu_group = iommu_group_get(dev);
477 if (!iommu_group && vfio_noiommu) {
479 * With noiommu enabled, create an IOMMU group for devices that
480 * don't already have one, implying no IOMMU hardware/driver
481 * exists. Taint the kernel because we're about to give a DMA
482 * capable device to a user without IOMMU protection.
484 group = vfio_noiommu_group_alloc(dev, VFIO_NO_IOMMU);
485 if (!IS_ERR(group)) {
486 add_taint(TAINT_USER, LOCKDEP_STILL_OK);
487 dev_warn(dev, "Adding kernel taint for vfio-noiommu group on device\n");
493 return ERR_PTR(-EINVAL);
496 * VFIO always sets IOMMU_CACHE because we offer no way for userspace to
497 * restore cache coherency. It has to be checked here because it is only
498 * valid for cases where we are using iommu groups.
500 if (!device_iommu_capable(dev, IOMMU_CAP_CACHE_COHERENCY)) {
501 iommu_group_put(iommu_group);
502 return ERR_PTR(-EINVAL);
505 group = vfio_group_get_from_iommu(iommu_group);
507 group = vfio_create_group(iommu_group, VFIO_IOMMU);
509 /* The vfio_group holds a reference to the iommu_group */
510 iommu_group_put(iommu_group);
514 static int __vfio_register_dev(struct vfio_device *device,
515 struct vfio_group *group)
517 struct vfio_device *existing_device;
521 * In all cases group is the output of one of the group allocation
522 * functions and we have group->drivers incremented for us.
525 return PTR_ERR(group);
528 * If the driver doesn't specify a set then the device is added to a
529 * singleton set just for itself.
531 if (!device->dev_set)
532 vfio_assign_device_set(device, device);
534 existing_device = vfio_group_get_device(group, device->dev);
535 if (existing_device) {
537 * group->iommu_group is non-NULL because we hold the drivers
540 dev_WARN(device->dev, "Device already exists on group %d\n",
541 iommu_group_id(group->iommu_group));
542 vfio_device_put_registration(existing_device);
547 /* Our reference on group is moved to the device */
548 device->group = group;
550 ret = dev_set_name(&device->device, "vfio%d", device->index);
554 ret = device_add(&device->device);
558 /* Refcounting can't start until the driver calls register */
559 refcount_set(&device->refcount, 1);
561 mutex_lock(&group->device_lock);
562 list_add(&device->group_next, &group->device_list);
563 mutex_unlock(&group->device_lock);
567 vfio_device_remove_group(device);
571 int vfio_register_group_dev(struct vfio_device *device)
573 return __vfio_register_dev(device,
574 vfio_group_find_or_alloc(device->dev));
576 EXPORT_SYMBOL_GPL(vfio_register_group_dev);
579 * Register a virtual device without IOMMU backing. The user of this
580 * device must not be able to directly trigger unmediated DMA.
582 int vfio_register_emulated_iommu_dev(struct vfio_device *device)
584 return __vfio_register_dev(device,
585 vfio_noiommu_group_alloc(device->dev, VFIO_EMULATED_IOMMU));
587 EXPORT_SYMBOL_GPL(vfio_register_emulated_iommu_dev);
589 static struct vfio_device *vfio_device_get_from_name(struct vfio_group *group,
592 struct vfio_device *it, *device = ERR_PTR(-ENODEV);
594 mutex_lock(&group->device_lock);
595 list_for_each_entry(it, &group->device_list, group_next) {
598 if (it->ops->match) {
599 ret = it->ops->match(it, buf);
601 device = ERR_PTR(ret);
605 ret = !strcmp(dev_name(it->dev), buf);
608 if (ret && vfio_device_try_get_registration(it)) {
613 mutex_unlock(&group->device_lock);
619 * Decrement the device reference count and wait for the device to be
620 * removed. Open file descriptors for the device... */
621 void vfio_unregister_group_dev(struct vfio_device *device)
623 struct vfio_group *group = device->group;
625 bool interrupted = false;
628 vfio_device_put_registration(device);
629 rc = try_wait_for_completion(&device->comp);
631 if (device->ops->request)
632 device->ops->request(device, i++);
635 rc = wait_for_completion_timeout(&device->comp,
638 rc = wait_for_completion_interruptible_timeout(
639 &device->comp, HZ * 10);
642 dev_warn(device->dev,
643 "Device is currently in use, task"
645 "blocked until device is released",
646 current->comm, task_pid_nr(current));
651 mutex_lock(&group->device_lock);
652 list_del(&device->group_next);
653 mutex_unlock(&group->device_lock);
655 /* Balances device_add in register path */
656 device_del(&device->device);
658 vfio_device_remove_group(device);
660 EXPORT_SYMBOL_GPL(vfio_unregister_group_dev);
663 * VFIO Group fd, /dev/vfio/$GROUP
666 * VFIO_GROUP_UNSET_CONTAINER should fail if there are other users or
667 * if there was no container to unset. Since the ioctl is called on
668 * the group, we know that still exists, therefore the only valid
669 * transition here is 1->0.
671 static int vfio_group_ioctl_unset_container(struct vfio_group *group)
675 mutex_lock(&group->group_lock);
676 if (!group->container) {
680 if (group->container_users != 1) {
684 vfio_group_detach_container(group);
687 mutex_unlock(&group->group_lock);
691 static int vfio_group_ioctl_set_container(struct vfio_group *group,
694 struct vfio_container *container;
699 if (get_user(fd, arg))
706 mutex_lock(&group->group_lock);
707 if (group->container || WARN_ON(group->container_users)) {
711 if (!group->iommu_group) {
716 container = vfio_container_from_file(f.file);
719 ret = vfio_container_attach_group(container, group);
724 mutex_unlock(&group->group_lock);
729 static const struct file_operations vfio_device_fops;
731 /* true if the vfio_device has open_device() called but not close_device() */
732 bool vfio_assert_device_open(struct vfio_device *device)
734 return !WARN_ON_ONCE(!READ_ONCE(device->open_count));
737 static int vfio_device_first_open(struct vfio_device *device)
741 lockdep_assert_held(&device->dev_set->lock);
743 if (!try_module_get(device->dev->driver->owner))
747 * Here we pass the KVM pointer with the group under the lock. If the
748 * device driver will use it, it must obtain a reference and release it
749 * during close_device.
751 mutex_lock(&device->group->group_lock);
752 ret = vfio_device_assign_container(device);
756 device->kvm = device->group->kvm;
757 if (device->ops->open_device) {
758 ret = device->ops->open_device(device);
762 vfio_device_container_register(device);
763 mutex_unlock(&device->group->group_lock);
768 vfio_device_unassign_container(device);
770 mutex_unlock(&device->group->group_lock);
771 module_put(device->dev->driver->owner);
775 static void vfio_device_last_close(struct vfio_device *device)
777 lockdep_assert_held(&device->dev_set->lock);
779 mutex_lock(&device->group->group_lock);
780 vfio_device_container_unregister(device);
781 if (device->ops->close_device)
782 device->ops->close_device(device);
784 vfio_device_unassign_container(device);
785 mutex_unlock(&device->group->group_lock);
786 module_put(device->dev->driver->owner);
789 static struct file *vfio_device_open(struct vfio_device *device)
794 mutex_lock(&device->dev_set->lock);
795 device->open_count++;
796 if (device->open_count == 1) {
797 ret = vfio_device_first_open(device);
801 mutex_unlock(&device->dev_set->lock);
804 * We can't use anon_inode_getfd() because we need to modify
805 * the f_mode flags directly to allow more than just ioctls
807 filep = anon_inode_getfile("[vfio-device]", &vfio_device_fops,
810 ret = PTR_ERR(filep);
811 goto err_close_device;
815 * TODO: add an anon_inode interface to do this.
816 * Appears to be missing by lack of need rather than
817 * explicitly prevented. Now there's need.
819 filep->f_mode |= (FMODE_PREAD | FMODE_PWRITE);
821 if (device->group->type == VFIO_NO_IOMMU)
822 dev_warn(device->dev, "vfio-noiommu device opened by user "
823 "(%s:%d)\n", current->comm, task_pid_nr(current));
825 * On success the ref of device is moved to the file and
826 * put in vfio_device_fops_release()
831 mutex_lock(&device->dev_set->lock);
832 if (device->open_count == 1)
833 vfio_device_last_close(device);
835 device->open_count--;
836 mutex_unlock(&device->dev_set->lock);
840 static int vfio_group_ioctl_get_device_fd(struct vfio_group *group,
843 struct vfio_device *device;
849 buf = strndup_user(arg, PAGE_SIZE);
853 device = vfio_device_get_from_name(group, buf);
856 return PTR_ERR(device);
858 fdno = get_unused_fd_flags(O_CLOEXEC);
864 filep = vfio_device_open(device);
866 ret = PTR_ERR(filep);
870 fd_install(fdno, filep);
876 vfio_device_put_registration(device);
880 static int vfio_group_ioctl_get_status(struct vfio_group *group,
881 struct vfio_group_status __user *arg)
883 unsigned long minsz = offsetofend(struct vfio_group_status, flags);
884 struct vfio_group_status status;
886 if (copy_from_user(&status, arg, minsz))
889 if (status.argsz < minsz)
894 mutex_lock(&group->group_lock);
895 if (!group->iommu_group) {
896 mutex_unlock(&group->group_lock);
900 if (group->container)
901 status.flags |= VFIO_GROUP_FLAGS_CONTAINER_SET |
902 VFIO_GROUP_FLAGS_VIABLE;
903 else if (!iommu_group_dma_owner_claimed(group->iommu_group))
904 status.flags |= VFIO_GROUP_FLAGS_VIABLE;
905 mutex_unlock(&group->group_lock);
907 if (copy_to_user(arg, &status, minsz))
912 static long vfio_group_fops_unl_ioctl(struct file *filep,
913 unsigned int cmd, unsigned long arg)
915 struct vfio_group *group = filep->private_data;
916 void __user *uarg = (void __user *)arg;
919 case VFIO_GROUP_GET_DEVICE_FD:
920 return vfio_group_ioctl_get_device_fd(group, uarg);
921 case VFIO_GROUP_GET_STATUS:
922 return vfio_group_ioctl_get_status(group, uarg);
923 case VFIO_GROUP_SET_CONTAINER:
924 return vfio_group_ioctl_set_container(group, uarg);
925 case VFIO_GROUP_UNSET_CONTAINER:
926 return vfio_group_ioctl_unset_container(group);
932 static int vfio_group_fops_open(struct inode *inode, struct file *filep)
934 struct vfio_group *group =
935 container_of(inode->i_cdev, struct vfio_group, cdev);
938 mutex_lock(&group->group_lock);
941 * drivers can be zero if this races with vfio_device_remove_group(), it
942 * will be stable at 0 under the group rwsem
944 if (refcount_read(&group->drivers) == 0) {
949 if (group->type == VFIO_NO_IOMMU && !capable(CAP_SYS_RAWIO)) {
955 * Do we need multiple instances of the group open? Seems not.
957 if (group->opened_file) {
961 group->opened_file = filep;
962 filep->private_data = group;
965 mutex_unlock(&group->group_lock);
969 static int vfio_group_fops_release(struct inode *inode, struct file *filep)
971 struct vfio_group *group = filep->private_data;
973 filep->private_data = NULL;
975 mutex_lock(&group->group_lock);
977 * Device FDs hold a group file reference, therefore the group release
978 * is only called when there are no open devices.
980 WARN_ON(group->notifier.head);
981 if (group->container)
982 vfio_group_detach_container(group);
983 group->opened_file = NULL;
984 mutex_unlock(&group->group_lock);
988 static const struct file_operations vfio_group_fops = {
989 .owner = THIS_MODULE,
990 .unlocked_ioctl = vfio_group_fops_unl_ioctl,
991 .compat_ioctl = compat_ptr_ioctl,
992 .open = vfio_group_fops_open,
993 .release = vfio_group_fops_release,
997 * Wrapper around pm_runtime_resume_and_get().
998 * Return error code on failure or 0 on success.
1000 static inline int vfio_device_pm_runtime_get(struct vfio_device *device)
1002 struct device *dev = device->dev;
1004 if (dev->driver && dev->driver->pm) {
1007 ret = pm_runtime_resume_and_get(dev);
1009 dev_info_ratelimited(dev,
1010 "vfio: runtime resume failed %d\n", ret);
1019 * Wrapper around pm_runtime_put().
1021 static inline void vfio_device_pm_runtime_put(struct vfio_device *device)
1023 struct device *dev = device->dev;
1025 if (dev->driver && dev->driver->pm)
1026 pm_runtime_put(dev);
1032 static int vfio_device_fops_release(struct inode *inode, struct file *filep)
1034 struct vfio_device *device = filep->private_data;
1036 mutex_lock(&device->dev_set->lock);
1037 vfio_assert_device_open(device);
1038 if (device->open_count == 1)
1039 vfio_device_last_close(device);
1040 device->open_count--;
1041 mutex_unlock(&device->dev_set->lock);
1043 vfio_device_put_registration(device);
1049 * vfio_mig_get_next_state - Compute the next step in the FSM
1050 * @cur_fsm - The current state the device is in
1051 * @new_fsm - The target state to reach
1052 * @next_fsm - Pointer to the next step to get to new_fsm
1054 * Return 0 upon success, otherwise -errno
1055 * Upon success the next step in the state progression between cur_fsm and
1056 * new_fsm will be set in next_fsm.
1058 * This breaks down requests for combination transitions into smaller steps and
1059 * returns the next step to get to new_fsm. The function may need to be called
1060 * multiple times before reaching new_fsm.
1063 int vfio_mig_get_next_state(struct vfio_device *device,
1064 enum vfio_device_mig_state cur_fsm,
1065 enum vfio_device_mig_state new_fsm,
1066 enum vfio_device_mig_state *next_fsm)
1068 enum { VFIO_DEVICE_NUM_STATES = VFIO_DEVICE_STATE_RUNNING_P2P + 1 };
1070 * The coding in this table requires the driver to implement the
1071 * following FSM arcs:
1077 * If P2P is supported then the driver must also implement these FSM
1079 * RUNNING -> RUNNING_P2P
1080 * RUNNING_P2P -> RUNNING
1081 * RUNNING_P2P -> STOP
1082 * STOP -> RUNNING_P2P
1083 * Without P2P the driver must implement:
1087 * The coding will step through multiple states for some combination
1088 * transitions; if all optional features are supported, this means the
1090 * RESUMING -> STOP -> RUNNING_P2P
1091 * RESUMING -> STOP -> RUNNING_P2P -> RUNNING
1092 * RESUMING -> STOP -> STOP_COPY
1093 * RUNNING -> RUNNING_P2P -> STOP
1094 * RUNNING -> RUNNING_P2P -> STOP -> RESUMING
1095 * RUNNING -> RUNNING_P2P -> STOP -> STOP_COPY
1096 * RUNNING_P2P -> STOP -> RESUMING
1097 * RUNNING_P2P -> STOP -> STOP_COPY
1098 * STOP -> RUNNING_P2P -> RUNNING
1099 * STOP_COPY -> STOP -> RESUMING
1100 * STOP_COPY -> STOP -> RUNNING_P2P
1101 * STOP_COPY -> STOP -> RUNNING_P2P -> RUNNING
1103 static const u8 vfio_from_fsm_table[VFIO_DEVICE_NUM_STATES][VFIO_DEVICE_NUM_STATES] = {
1104 [VFIO_DEVICE_STATE_STOP] = {
1105 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
1106 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING_P2P,
1107 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP_COPY,
1108 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RESUMING,
1109 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
1110 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1112 [VFIO_DEVICE_STATE_RUNNING] = {
1113 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_RUNNING_P2P,
1114 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING,
1115 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_RUNNING_P2P,
1116 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RUNNING_P2P,
1117 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
1118 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1120 [VFIO_DEVICE_STATE_STOP_COPY] = {
1121 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
1122 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_STOP,
1123 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP_COPY,
1124 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_STOP,
1125 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_STOP,
1126 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1128 [VFIO_DEVICE_STATE_RESUMING] = {
1129 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
1130 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_STOP,
1131 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP,
1132 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RESUMING,
1133 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_STOP,
1134 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1136 [VFIO_DEVICE_STATE_RUNNING_P2P] = {
1137 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
1138 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING,
1139 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP,
1140 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_STOP,
1141 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
1142 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1144 [VFIO_DEVICE_STATE_ERROR] = {
1145 [VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_ERROR,
1146 [VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_ERROR,
1147 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_ERROR,
1148 [VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_ERROR,
1149 [VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_ERROR,
1150 [VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
1154 static const unsigned int state_flags_table[VFIO_DEVICE_NUM_STATES] = {
1155 [VFIO_DEVICE_STATE_STOP] = VFIO_MIGRATION_STOP_COPY,
1156 [VFIO_DEVICE_STATE_RUNNING] = VFIO_MIGRATION_STOP_COPY,
1157 [VFIO_DEVICE_STATE_STOP_COPY] = VFIO_MIGRATION_STOP_COPY,
1158 [VFIO_DEVICE_STATE_RESUMING] = VFIO_MIGRATION_STOP_COPY,
1159 [VFIO_DEVICE_STATE_RUNNING_P2P] =
1160 VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P,
1161 [VFIO_DEVICE_STATE_ERROR] = ~0U,
1164 if (WARN_ON(cur_fsm >= ARRAY_SIZE(vfio_from_fsm_table) ||
1165 (state_flags_table[cur_fsm] & device->migration_flags) !=
1166 state_flags_table[cur_fsm]))
1169 if (new_fsm >= ARRAY_SIZE(vfio_from_fsm_table) ||
1170 (state_flags_table[new_fsm] & device->migration_flags) !=
1171 state_flags_table[new_fsm])
1175 * Arcs touching optional and unsupported states are skipped over. The
1176 * driver will instead see an arc from the original state to the next
1177 * logical state, as per the above comment.
1179 *next_fsm = vfio_from_fsm_table[cur_fsm][new_fsm];
1180 while ((state_flags_table[*next_fsm] & device->migration_flags) !=
1181 state_flags_table[*next_fsm])
1182 *next_fsm = vfio_from_fsm_table[*next_fsm][new_fsm];
1184 return (*next_fsm != VFIO_DEVICE_STATE_ERROR) ? 0 : -EINVAL;
1186 EXPORT_SYMBOL_GPL(vfio_mig_get_next_state);
1189 * Convert the drivers's struct file into a FD number and return it to userspace
1191 static int vfio_ioct_mig_return_fd(struct file *filp, void __user *arg,
1192 struct vfio_device_feature_mig_state *mig)
1197 fd = get_unused_fd_flags(O_CLOEXEC);
1204 if (copy_to_user(arg, mig, sizeof(*mig))) {
1206 goto out_put_unused;
1208 fd_install(fd, filp);
1219 vfio_ioctl_device_feature_mig_device_state(struct vfio_device *device,
1220 u32 flags, void __user *arg,
1224 offsetofend(struct vfio_device_feature_mig_state, data_fd);
1225 struct vfio_device_feature_mig_state mig;
1226 struct file *filp = NULL;
1229 if (!device->mig_ops)
1232 ret = vfio_check_feature(flags, argsz,
1233 VFIO_DEVICE_FEATURE_SET |
1234 VFIO_DEVICE_FEATURE_GET,
1239 if (copy_from_user(&mig, arg, minsz))
1242 if (flags & VFIO_DEVICE_FEATURE_GET) {
1243 enum vfio_device_mig_state curr_state;
1245 ret = device->mig_ops->migration_get_state(device,
1249 mig.device_state = curr_state;
1253 /* Handle the VFIO_DEVICE_FEATURE_SET */
1254 filp = device->mig_ops->migration_set_state(device, mig.device_state);
1255 if (IS_ERR(filp) || !filp)
1258 return vfio_ioct_mig_return_fd(filp, arg, &mig);
1261 if (copy_to_user(arg, &mig, sizeof(mig)))
1264 return PTR_ERR(filp);
1268 static int vfio_ioctl_device_feature_migration(struct vfio_device *device,
1269 u32 flags, void __user *arg,
1272 struct vfio_device_feature_migration mig = {
1273 .flags = device->migration_flags,
1277 if (!device->mig_ops)
1280 ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_GET,
1284 if (copy_to_user(arg, &mig, sizeof(mig)))
1289 /* Ranges should fit into a single kernel page */
1290 #define LOG_MAX_RANGES \
1291 (PAGE_SIZE / sizeof(struct vfio_device_feature_dma_logging_range))
1294 vfio_ioctl_device_feature_logging_start(struct vfio_device *device,
1295 u32 flags, void __user *arg,
1299 offsetofend(struct vfio_device_feature_dma_logging_control,
1301 struct vfio_device_feature_dma_logging_range __user *ranges;
1302 struct vfio_device_feature_dma_logging_control control;
1303 struct vfio_device_feature_dma_logging_range range;
1304 struct rb_root_cached root = RB_ROOT_CACHED;
1305 struct interval_tree_node *nodes;
1310 if (!device->log_ops)
1313 ret = vfio_check_feature(flags, argsz,
1314 VFIO_DEVICE_FEATURE_SET,
1319 if (copy_from_user(&control, arg, minsz))
1322 nnodes = control.num_ranges;
1326 if (nnodes > LOG_MAX_RANGES)
1329 ranges = u64_to_user_ptr(control.ranges);
1330 nodes = kmalloc_array(nnodes, sizeof(struct interval_tree_node),
1335 for (i = 0; i < nnodes; i++) {
1336 if (copy_from_user(&range, &ranges[i], sizeof(range))) {
1340 if (!IS_ALIGNED(range.iova, control.page_size) ||
1341 !IS_ALIGNED(range.length, control.page_size)) {
1346 if (check_add_overflow(range.iova, range.length, &iova_end) ||
1347 iova_end > ULONG_MAX) {
1352 nodes[i].start = range.iova;
1353 nodes[i].last = range.iova + range.length - 1;
1354 if (interval_tree_iter_first(&root, nodes[i].start,
1356 /* Range overlapping */
1360 interval_tree_insert(nodes + i, &root);
1363 ret = device->log_ops->log_start(device, &root, nnodes,
1364 &control.page_size);
1368 if (copy_to_user(arg, &control, sizeof(control))) {
1370 device->log_ops->log_stop(device);
1379 vfio_ioctl_device_feature_logging_stop(struct vfio_device *device,
1380 u32 flags, void __user *arg,
1385 if (!device->log_ops)
1388 ret = vfio_check_feature(flags, argsz,
1389 VFIO_DEVICE_FEATURE_SET, 0);
1393 return device->log_ops->log_stop(device);
1396 static int vfio_device_log_read_and_clear(struct iova_bitmap *iter,
1397 unsigned long iova, size_t length,
1400 struct vfio_device *device = opaque;
1402 return device->log_ops->log_read_and_clear(device, iova, length, iter);
1406 vfio_ioctl_device_feature_logging_report(struct vfio_device *device,
1407 u32 flags, void __user *arg,
1411 offsetofend(struct vfio_device_feature_dma_logging_report,
1413 struct vfio_device_feature_dma_logging_report report;
1414 struct iova_bitmap *iter;
1418 if (!device->log_ops)
1421 ret = vfio_check_feature(flags, argsz,
1422 VFIO_DEVICE_FEATURE_GET,
1427 if (copy_from_user(&report, arg, minsz))
1430 if (report.page_size < SZ_4K || !is_power_of_2(report.page_size))
1433 if (check_add_overflow(report.iova, report.length, &iova_end) ||
1434 iova_end > ULONG_MAX)
1437 iter = iova_bitmap_alloc(report.iova, report.length,
1439 u64_to_user_ptr(report.bitmap));
1441 return PTR_ERR(iter);
1443 ret = iova_bitmap_for_each(iter, device,
1444 vfio_device_log_read_and_clear);
1446 iova_bitmap_free(iter);
1450 static int vfio_ioctl_device_feature(struct vfio_device *device,
1451 struct vfio_device_feature __user *arg)
1453 size_t minsz = offsetofend(struct vfio_device_feature, flags);
1454 struct vfio_device_feature feature;
1456 if (copy_from_user(&feature, arg, minsz))
1459 if (feature.argsz < minsz)
1462 /* Check unknown flags */
1464 ~(VFIO_DEVICE_FEATURE_MASK | VFIO_DEVICE_FEATURE_SET |
1465 VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_PROBE))
1468 /* GET & SET are mutually exclusive except with PROBE */
1469 if (!(feature.flags & VFIO_DEVICE_FEATURE_PROBE) &&
1470 (feature.flags & VFIO_DEVICE_FEATURE_SET) &&
1471 (feature.flags & VFIO_DEVICE_FEATURE_GET))
1474 switch (feature.flags & VFIO_DEVICE_FEATURE_MASK) {
1475 case VFIO_DEVICE_FEATURE_MIGRATION:
1476 return vfio_ioctl_device_feature_migration(
1477 device, feature.flags, arg->data,
1478 feature.argsz - minsz);
1479 case VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE:
1480 return vfio_ioctl_device_feature_mig_device_state(
1481 device, feature.flags, arg->data,
1482 feature.argsz - minsz);
1483 case VFIO_DEVICE_FEATURE_DMA_LOGGING_START:
1484 return vfio_ioctl_device_feature_logging_start(
1485 device, feature.flags, arg->data,
1486 feature.argsz - minsz);
1487 case VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP:
1488 return vfio_ioctl_device_feature_logging_stop(
1489 device, feature.flags, arg->data,
1490 feature.argsz - minsz);
1491 case VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT:
1492 return vfio_ioctl_device_feature_logging_report(
1493 device, feature.flags, arg->data,
1494 feature.argsz - minsz);
1496 if (unlikely(!device->ops->device_feature))
1498 return device->ops->device_feature(device, feature.flags,
1500 feature.argsz - minsz);
1504 static long vfio_device_fops_unl_ioctl(struct file *filep,
1505 unsigned int cmd, unsigned long arg)
1507 struct vfio_device *device = filep->private_data;
1510 ret = vfio_device_pm_runtime_get(device);
1515 case VFIO_DEVICE_FEATURE:
1516 ret = vfio_ioctl_device_feature(device, (void __user *)arg);
1520 if (unlikely(!device->ops->ioctl))
1523 ret = device->ops->ioctl(device, cmd, arg);
1527 vfio_device_pm_runtime_put(device);
1531 static ssize_t vfio_device_fops_read(struct file *filep, char __user *buf,
1532 size_t count, loff_t *ppos)
1534 struct vfio_device *device = filep->private_data;
1536 if (unlikely(!device->ops->read))
1539 return device->ops->read(device, buf, count, ppos);
1542 static ssize_t vfio_device_fops_write(struct file *filep,
1543 const char __user *buf,
1544 size_t count, loff_t *ppos)
1546 struct vfio_device *device = filep->private_data;
1548 if (unlikely(!device->ops->write))
1551 return device->ops->write(device, buf, count, ppos);
1554 static int vfio_device_fops_mmap(struct file *filep, struct vm_area_struct *vma)
1556 struct vfio_device *device = filep->private_data;
1558 if (unlikely(!device->ops->mmap))
1561 return device->ops->mmap(device, vma);
1564 static const struct file_operations vfio_device_fops = {
1565 .owner = THIS_MODULE,
1566 .release = vfio_device_fops_release,
1567 .read = vfio_device_fops_read,
1568 .write = vfio_device_fops_write,
1569 .unlocked_ioctl = vfio_device_fops_unl_ioctl,
1570 .compat_ioctl = compat_ptr_ioctl,
1571 .mmap = vfio_device_fops_mmap,
1575 * vfio_file_iommu_group - Return the struct iommu_group for the vfio group file
1576 * @file: VFIO group file
1578 * The returned iommu_group is valid as long as a ref is held on the file. This
1579 * returns a reference on the group. This function is deprecated, only the SPAPR
1580 * path in kvm should call it.
1582 struct iommu_group *vfio_file_iommu_group(struct file *file)
1584 struct vfio_group *group = file->private_data;
1585 struct iommu_group *iommu_group = NULL;
1587 if (!IS_ENABLED(CONFIG_SPAPR_TCE_IOMMU))
1590 if (!vfio_file_is_group(file))
1593 mutex_lock(&group->group_lock);
1594 if (group->iommu_group) {
1595 iommu_group = group->iommu_group;
1596 iommu_group_ref_get(iommu_group);
1598 mutex_unlock(&group->group_lock);
1601 EXPORT_SYMBOL_GPL(vfio_file_iommu_group);
1604 * vfio_file_is_group - True if the file is usable with VFIO aPIS
1605 * @file: VFIO group file
1607 bool vfio_file_is_group(struct file *file)
1609 return file->f_op == &vfio_group_fops;
1611 EXPORT_SYMBOL_GPL(vfio_file_is_group);
1614 * vfio_file_enforced_coherent - True if the DMA associated with the VFIO file
1615 * is always CPU cache coherent
1616 * @file: VFIO group file
1618 * Enforced coherency means that the IOMMU ignores things like the PCIe no-snoop
1619 * bit in DMA transactions. A return of false indicates that the user has
1620 * rights to access additional instructions such as wbinvd on x86.
1622 bool vfio_file_enforced_coherent(struct file *file)
1624 struct vfio_group *group = file->private_data;
1627 if (!vfio_file_is_group(file))
1630 mutex_lock(&group->group_lock);
1631 if (group->container) {
1632 ret = vfio_container_ioctl_check_extension(group->container,
1636 * Since the coherency state is determined only once a container
1637 * is attached the user must do so before they can prove they
1642 mutex_unlock(&group->group_lock);
1645 EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);
1648 * vfio_file_set_kvm - Link a kvm with VFIO drivers
1649 * @file: VFIO group file
1652 * When a VFIO device is first opened the KVM will be available in
1653 * device->kvm if one was associated with the group.
1655 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
1657 struct vfio_group *group = file->private_data;
1659 if (!vfio_file_is_group(file))
1662 mutex_lock(&group->group_lock);
1664 mutex_unlock(&group->group_lock);
1666 EXPORT_SYMBOL_GPL(vfio_file_set_kvm);
1669 * vfio_file_has_dev - True if the VFIO file is a handle for device
1670 * @file: VFIO file to check
1671 * @device: Device that must be part of the file
1673 * Returns true if given file has permission to manipulate the given device.
1675 bool vfio_file_has_dev(struct file *file, struct vfio_device *device)
1677 struct vfio_group *group = file->private_data;
1679 if (!vfio_file_is_group(file))
1682 return group == device->group;
1684 EXPORT_SYMBOL_GPL(vfio_file_has_dev);
1687 * Sub-module support
1690 * Helper for managing a buffer of info chain capabilities, allocate or
1691 * reallocate a buffer with additional @size, filling in @id and @version
1692 * of the capability. A pointer to the new capability is returned.
1694 * NB. The chain is based at the head of the buffer, so new entries are
1695 * added to the tail, vfio_info_cap_shift() should be called to fixup the
1696 * next offsets prior to copying to the user buffer.
1698 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
1699 size_t size, u16 id, u16 version)
1702 struct vfio_info_cap_header *header, *tmp;
1704 buf = krealloc(caps->buf, caps->size + size, GFP_KERNEL);
1709 return ERR_PTR(-ENOMEM);
1713 header = buf + caps->size;
1715 /* Eventually copied to user buffer, zero */
1716 memset(header, 0, size);
1719 header->version = version;
1721 /* Add to the end of the capability chain */
1722 for (tmp = buf; tmp->next; tmp = buf + tmp->next)
1725 tmp->next = caps->size;
1730 EXPORT_SYMBOL_GPL(vfio_info_cap_add);
1732 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset)
1734 struct vfio_info_cap_header *tmp;
1735 void *buf = (void *)caps->buf;
1737 for (tmp = buf; tmp->next; tmp = buf + tmp->next - offset)
1738 tmp->next += offset;
1740 EXPORT_SYMBOL(vfio_info_cap_shift);
1742 int vfio_info_add_capability(struct vfio_info_cap *caps,
1743 struct vfio_info_cap_header *cap, size_t size)
1745 struct vfio_info_cap_header *header;
1747 header = vfio_info_cap_add(caps, size, cap->id, cap->version);
1749 return PTR_ERR(header);
1751 memcpy(header + 1, cap + 1, size - sizeof(*header));
1755 EXPORT_SYMBOL(vfio_info_add_capability);
1757 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr, int num_irqs,
1758 int max_irq_type, size_t *data_size)
1760 unsigned long minsz;
1763 minsz = offsetofend(struct vfio_irq_set, count);
1765 if ((hdr->argsz < minsz) || (hdr->index >= max_irq_type) ||
1766 (hdr->count >= (U32_MAX - hdr->start)) ||
1767 (hdr->flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
1768 VFIO_IRQ_SET_ACTION_TYPE_MASK)))
1774 if (hdr->start >= num_irqs || hdr->start + hdr->count > num_irqs)
1777 switch (hdr->flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
1778 case VFIO_IRQ_SET_DATA_NONE:
1781 case VFIO_IRQ_SET_DATA_BOOL:
1782 size = sizeof(uint8_t);
1784 case VFIO_IRQ_SET_DATA_EVENTFD:
1785 size = sizeof(int32_t);
1792 if (hdr->argsz - minsz < hdr->count * size)
1798 *data_size = hdr->count * size;
1803 EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
1806 * Module/class support
1808 static char *vfio_devnode(struct device *dev, umode_t *mode)
1810 return kasprintf(GFP_KERNEL, "vfio/%s", dev_name(dev));
1813 static int __init vfio_init(void)
1817 ida_init(&vfio.group_ida);
1818 ida_init(&vfio.device_ida);
1819 mutex_init(&vfio.group_lock);
1820 INIT_LIST_HEAD(&vfio.group_list);
1822 ret = vfio_container_init();
1826 /* /dev/vfio/$GROUP */
1827 vfio.class = class_create(THIS_MODULE, "vfio");
1828 if (IS_ERR(vfio.class)) {
1829 ret = PTR_ERR(vfio.class);
1830 goto err_group_class;
1833 vfio.class->devnode = vfio_devnode;
1835 /* /sys/class/vfio-dev/vfioX */
1836 vfio.device_class = class_create(THIS_MODULE, "vfio-dev");
1837 if (IS_ERR(vfio.device_class)) {
1838 ret = PTR_ERR(vfio.device_class);
1842 ret = alloc_chrdev_region(&vfio.group_devt, 0, MINORMASK + 1, "vfio");
1844 goto err_alloc_chrdev;
1846 pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
1850 class_destroy(vfio.device_class);
1851 vfio.device_class = NULL;
1853 class_destroy(vfio.class);
1856 vfio_container_cleanup();
1860 static void __exit vfio_cleanup(void)
1862 WARN_ON(!list_empty(&vfio.group_list));
1864 ida_destroy(&vfio.device_ida);
1865 ida_destroy(&vfio.group_ida);
1866 unregister_chrdev_region(vfio.group_devt, MINORMASK + 1);
1867 class_destroy(vfio.device_class);
1868 vfio.device_class = NULL;
1869 class_destroy(vfio.class);
1870 vfio_container_cleanup();
1872 xa_destroy(&vfio_device_set_xa);
1875 module_init(vfio_init);
1876 module_exit(vfio_cleanup);
1878 MODULE_VERSION(DRIVER_VERSION);
1879 MODULE_LICENSE("GPL v2");
1880 MODULE_AUTHOR(DRIVER_AUTHOR);
1881 MODULE_DESCRIPTION(DRIVER_DESC);
1882 MODULE_ALIAS_MISCDEV(VFIO_MINOR);
1883 MODULE_ALIAS("devname:vfio/vfio");
1884 MODULE_SOFTDEP("post: vfio_iommu_type1 vfio_iommu_spapr_tce");