1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
4 * Author: Joerg Roedel <jroedel@suse.de>
7 #define pr_fmt(fmt) "iommu: " fmt
9 #include <linux/device.h>
10 #include <linux/kernel.h>
11 #include <linux/bug.h>
12 #include <linux/types.h>
13 #include <linux/init.h>
14 #include <linux/export.h>
15 #include <linux/slab.h>
16 #include <linux/errno.h>
17 #include <linux/iommu.h>
18 #include <linux/idr.h>
19 #include <linux/notifier.h>
20 #include <linux/err.h>
21 #include <linux/pci.h>
22 #include <linux/bitops.h>
23 #include <linux/property.h>
24 #include <linux/fsl/mc.h>
25 #include <trace/events/iommu.h>
27 static struct kset
*iommu_group_kset
;
28 static DEFINE_IDA(iommu_group_ida
);
29 #ifdef CONFIG_IOMMU_DEFAULT_PASSTHROUGH
30 static unsigned int iommu_def_domain_type
= IOMMU_DOMAIN_IDENTITY
;
32 static unsigned int iommu_def_domain_type
= IOMMU_DOMAIN_DMA
;
34 static bool iommu_dma_strict __read_mostly
= true;
38 struct kobject
*devices_kobj
;
39 struct list_head devices
;
41 struct blocking_notifier_head notifier
;
43 void (*iommu_data_release
)(void *iommu_data
);
46 struct iommu_domain
*default_domain
;
47 struct iommu_domain
*domain
;
51 struct list_head list
;
56 struct iommu_group_attribute
{
57 struct attribute attr
;
58 ssize_t (*show
)(struct iommu_group
*group
, char *buf
);
59 ssize_t (*store
)(struct iommu_group
*group
,
60 const char *buf
, size_t count
);
63 static const char * const iommu_group_resv_type_string
[] = {
64 [IOMMU_RESV_DIRECT
] = "direct",
65 [IOMMU_RESV_DIRECT_RELAXABLE
] = "direct-relaxable",
66 [IOMMU_RESV_RESERVED
] = "reserved",
67 [IOMMU_RESV_MSI
] = "msi",
68 [IOMMU_RESV_SW_MSI
] = "msi",
71 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
72 struct iommu_group_attribute iommu_group_attr_##_name = \
73 __ATTR(_name, _mode, _show, _store)
75 #define to_iommu_group_attr(_attr) \
76 container_of(_attr, struct iommu_group_attribute, attr)
77 #define to_iommu_group(_kobj) \
78 container_of(_kobj, struct iommu_group, kobj)
80 static LIST_HEAD(iommu_device_list
);
81 static DEFINE_SPINLOCK(iommu_device_lock
);
83 int iommu_device_register(struct iommu_device
*iommu
)
85 spin_lock(&iommu_device_lock
);
86 list_add_tail(&iommu
->list
, &iommu_device_list
);
87 spin_unlock(&iommu_device_lock
);
92 void iommu_device_unregister(struct iommu_device
*iommu
)
94 spin_lock(&iommu_device_lock
);
95 list_del(&iommu
->list
);
96 spin_unlock(&iommu_device_lock
);
99 static struct iommu_param
*iommu_get_dev_param(struct device
*dev
)
101 struct iommu_param
*param
= dev
->iommu_param
;
106 param
= kzalloc(sizeof(*param
), GFP_KERNEL
);
110 mutex_init(¶m
->lock
);
111 dev
->iommu_param
= param
;
115 static void iommu_free_dev_param(struct device
*dev
)
117 kfree(dev
->iommu_param
);
118 dev
->iommu_param
= NULL
;
121 int iommu_probe_device(struct device
*dev
)
123 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
126 WARN_ON(dev
->iommu_group
);
130 if (!iommu_get_dev_param(dev
))
133 ret
= ops
->add_device(dev
);
135 iommu_free_dev_param(dev
);
140 void iommu_release_device(struct device
*dev
)
142 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
144 if (dev
->iommu_group
)
145 ops
->remove_device(dev
);
147 iommu_free_dev_param(dev
);
150 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
152 static int __iommu_attach_device(struct iommu_domain
*domain
,
154 static int __iommu_attach_group(struct iommu_domain
*domain
,
155 struct iommu_group
*group
);
156 static void __iommu_detach_group(struct iommu_domain
*domain
,
157 struct iommu_group
*group
);
159 static int __init
iommu_set_def_domain_type(char *str
)
164 ret
= kstrtobool(str
, &pt
);
168 iommu_def_domain_type
= pt
? IOMMU_DOMAIN_IDENTITY
: IOMMU_DOMAIN_DMA
;
171 early_param("iommu.passthrough", iommu_set_def_domain_type
);
173 static int __init
iommu_dma_setup(char *str
)
175 return kstrtobool(str
, &iommu_dma_strict
);
177 early_param("iommu.strict", iommu_dma_setup
);
179 static ssize_t
iommu_group_attr_show(struct kobject
*kobj
,
180 struct attribute
*__attr
, char *buf
)
182 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
183 struct iommu_group
*group
= to_iommu_group(kobj
);
187 ret
= attr
->show(group
, buf
);
191 static ssize_t
iommu_group_attr_store(struct kobject
*kobj
,
192 struct attribute
*__attr
,
193 const char *buf
, size_t count
)
195 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
196 struct iommu_group
*group
= to_iommu_group(kobj
);
200 ret
= attr
->store(group
, buf
, count
);
204 static const struct sysfs_ops iommu_group_sysfs_ops
= {
205 .show
= iommu_group_attr_show
,
206 .store
= iommu_group_attr_store
,
209 static int iommu_group_create_file(struct iommu_group
*group
,
210 struct iommu_group_attribute
*attr
)
212 return sysfs_create_file(&group
->kobj
, &attr
->attr
);
215 static void iommu_group_remove_file(struct iommu_group
*group
,
216 struct iommu_group_attribute
*attr
)
218 sysfs_remove_file(&group
->kobj
, &attr
->attr
);
221 static ssize_t
iommu_group_show_name(struct iommu_group
*group
, char *buf
)
223 return sprintf(buf
, "%s\n", group
->name
);
227 * iommu_insert_resv_region - Insert a new region in the
228 * list of reserved regions.
229 * @new: new region to insert
230 * @regions: list of regions
232 * The new element is sorted by address with respect to the other
233 * regions of the same type. In case it overlaps with another
234 * region of the same type, regions are merged. In case it
235 * overlaps with another region of different type, regions are
238 static int iommu_insert_resv_region(struct iommu_resv_region
*new,
239 struct list_head
*regions
)
241 struct iommu_resv_region
*region
;
242 phys_addr_t start
= new->start
;
243 phys_addr_t end
= new->start
+ new->length
- 1;
244 struct list_head
*pos
= regions
->next
;
246 while (pos
!= regions
) {
247 struct iommu_resv_region
*entry
=
248 list_entry(pos
, struct iommu_resv_region
, list
);
249 phys_addr_t a
= entry
->start
;
250 phys_addr_t b
= entry
->start
+ entry
->length
- 1;
251 int type
= entry
->type
;
255 } else if (start
> b
) {
257 } else if ((start
>= a
) && (end
<= b
)) {
258 if (new->type
== type
)
263 if (new->type
== type
) {
264 phys_addr_t new_start
= min(a
, start
);
265 phys_addr_t new_end
= max(b
, end
);
268 list_del(&entry
->list
);
269 entry
->start
= new_start
;
270 entry
->length
= new_end
- new_start
+ 1;
271 ret
= iommu_insert_resv_region(entry
, regions
);
280 region
= iommu_alloc_resv_region(new->start
, new->length
,
281 new->prot
, new->type
);
285 list_add_tail(®ion
->list
, pos
);
290 iommu_insert_device_resv_regions(struct list_head
*dev_resv_regions
,
291 struct list_head
*group_resv_regions
)
293 struct iommu_resv_region
*entry
;
296 list_for_each_entry(entry
, dev_resv_regions
, list
) {
297 ret
= iommu_insert_resv_region(entry
, group_resv_regions
);
304 int iommu_get_group_resv_regions(struct iommu_group
*group
,
305 struct list_head
*head
)
307 struct group_device
*device
;
310 mutex_lock(&group
->mutex
);
311 list_for_each_entry(device
, &group
->devices
, list
) {
312 struct list_head dev_resv_regions
;
314 INIT_LIST_HEAD(&dev_resv_regions
);
315 iommu_get_resv_regions(device
->dev
, &dev_resv_regions
);
316 ret
= iommu_insert_device_resv_regions(&dev_resv_regions
, head
);
317 iommu_put_resv_regions(device
->dev
, &dev_resv_regions
);
321 mutex_unlock(&group
->mutex
);
324 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions
);
326 static ssize_t
iommu_group_show_resv_regions(struct iommu_group
*group
,
329 struct iommu_resv_region
*region
, *next
;
330 struct list_head group_resv_regions
;
333 INIT_LIST_HEAD(&group_resv_regions
);
334 iommu_get_group_resv_regions(group
, &group_resv_regions
);
336 list_for_each_entry_safe(region
, next
, &group_resv_regions
, list
) {
337 str
+= sprintf(str
, "0x%016llx 0x%016llx %s\n",
338 (long long int)region
->start
,
339 (long long int)(region
->start
+
341 iommu_group_resv_type_string
[region
->type
]);
348 static ssize_t
iommu_group_show_type(struct iommu_group
*group
,
351 char *type
= "unknown\n";
353 if (group
->default_domain
) {
354 switch (group
->default_domain
->type
) {
355 case IOMMU_DOMAIN_BLOCKED
:
358 case IOMMU_DOMAIN_IDENTITY
:
361 case IOMMU_DOMAIN_UNMANAGED
:
362 type
= "unmanaged\n";
364 case IOMMU_DOMAIN_DMA
:
374 static IOMMU_GROUP_ATTR(name
, S_IRUGO
, iommu_group_show_name
, NULL
);
376 static IOMMU_GROUP_ATTR(reserved_regions
, 0444,
377 iommu_group_show_resv_regions
, NULL
);
379 static IOMMU_GROUP_ATTR(type
, 0444, iommu_group_show_type
, NULL
);
381 static void iommu_group_release(struct kobject
*kobj
)
383 struct iommu_group
*group
= to_iommu_group(kobj
);
385 pr_debug("Releasing group %d\n", group
->id
);
387 if (group
->iommu_data_release
)
388 group
->iommu_data_release(group
->iommu_data
);
390 ida_simple_remove(&iommu_group_ida
, group
->id
);
392 if (group
->default_domain
)
393 iommu_domain_free(group
->default_domain
);
399 static struct kobj_type iommu_group_ktype
= {
400 .sysfs_ops
= &iommu_group_sysfs_ops
,
401 .release
= iommu_group_release
,
405 * iommu_group_alloc - Allocate a new group
407 * This function is called by an iommu driver to allocate a new iommu
408 * group. The iommu group represents the minimum granularity of the iommu.
409 * Upon successful return, the caller holds a reference to the supplied
410 * group in order to hold the group until devices are added. Use
411 * iommu_group_put() to release this extra reference count, allowing the
412 * group to be automatically reclaimed once it has no devices or external
415 struct iommu_group
*iommu_group_alloc(void)
417 struct iommu_group
*group
;
420 group
= kzalloc(sizeof(*group
), GFP_KERNEL
);
422 return ERR_PTR(-ENOMEM
);
424 group
->kobj
.kset
= iommu_group_kset
;
425 mutex_init(&group
->mutex
);
426 INIT_LIST_HEAD(&group
->devices
);
427 BLOCKING_INIT_NOTIFIER_HEAD(&group
->notifier
);
429 ret
= ida_simple_get(&iommu_group_ida
, 0, 0, GFP_KERNEL
);
436 ret
= kobject_init_and_add(&group
->kobj
, &iommu_group_ktype
,
437 NULL
, "%d", group
->id
);
439 ida_simple_remove(&iommu_group_ida
, group
->id
);
444 group
->devices_kobj
= kobject_create_and_add("devices", &group
->kobj
);
445 if (!group
->devices_kobj
) {
446 kobject_put(&group
->kobj
); /* triggers .release & free */
447 return ERR_PTR(-ENOMEM
);
451 * The devices_kobj holds a reference on the group kobject, so
452 * as long as that exists so will the group. We can therefore
453 * use the devices_kobj for reference counting.
455 kobject_put(&group
->kobj
);
457 ret
= iommu_group_create_file(group
,
458 &iommu_group_attr_reserved_regions
);
462 ret
= iommu_group_create_file(group
, &iommu_group_attr_type
);
466 pr_debug("Allocated group %d\n", group
->id
);
470 EXPORT_SYMBOL_GPL(iommu_group_alloc
);
472 struct iommu_group
*iommu_group_get_by_id(int id
)
474 struct kobject
*group_kobj
;
475 struct iommu_group
*group
;
478 if (!iommu_group_kset
)
481 name
= kasprintf(GFP_KERNEL
, "%d", id
);
485 group_kobj
= kset_find_obj(iommu_group_kset
, name
);
491 group
= container_of(group_kobj
, struct iommu_group
, kobj
);
492 BUG_ON(group
->id
!= id
);
494 kobject_get(group
->devices_kobj
);
495 kobject_put(&group
->kobj
);
499 EXPORT_SYMBOL_GPL(iommu_group_get_by_id
);
502 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
505 * iommu drivers can store data in the group for use when doing iommu
506 * operations. This function provides a way to retrieve it. Caller
507 * should hold a group reference.
509 void *iommu_group_get_iommudata(struct iommu_group
*group
)
511 return group
->iommu_data
;
513 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata
);
516 * iommu_group_set_iommudata - set iommu_data for a group
518 * @iommu_data: new data
519 * @release: release function for iommu_data
521 * iommu drivers can store data in the group for use when doing iommu
522 * operations. This function provides a way to set the data after
523 * the group has been allocated. Caller should hold a group reference.
525 void iommu_group_set_iommudata(struct iommu_group
*group
, void *iommu_data
,
526 void (*release
)(void *iommu_data
))
528 group
->iommu_data
= iommu_data
;
529 group
->iommu_data_release
= release
;
531 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata
);
534 * iommu_group_set_name - set name for a group
538 * Allow iommu driver to set a name for a group. When set it will
539 * appear in a name attribute file under the group in sysfs.
541 int iommu_group_set_name(struct iommu_group
*group
, const char *name
)
546 iommu_group_remove_file(group
, &iommu_group_attr_name
);
553 group
->name
= kstrdup(name
, GFP_KERNEL
);
557 ret
= iommu_group_create_file(group
, &iommu_group_attr_name
);
566 EXPORT_SYMBOL_GPL(iommu_group_set_name
);
568 static int iommu_group_create_direct_mappings(struct iommu_group
*group
,
571 struct iommu_domain
*domain
= group
->default_domain
;
572 struct iommu_resv_region
*entry
;
573 struct list_head mappings
;
574 unsigned long pg_size
;
577 if (!domain
|| domain
->type
!= IOMMU_DOMAIN_DMA
)
580 BUG_ON(!domain
->pgsize_bitmap
);
582 pg_size
= 1UL << __ffs(domain
->pgsize_bitmap
);
583 INIT_LIST_HEAD(&mappings
);
585 iommu_get_resv_regions(dev
, &mappings
);
587 /* We need to consider overlapping regions for different devices */
588 list_for_each_entry(entry
, &mappings
, list
) {
589 dma_addr_t start
, end
, addr
;
591 if (domain
->ops
->apply_resv_region
)
592 domain
->ops
->apply_resv_region(dev
, domain
, entry
);
594 start
= ALIGN(entry
->start
, pg_size
);
595 end
= ALIGN(entry
->start
+ entry
->length
, pg_size
);
597 if (entry
->type
!= IOMMU_RESV_DIRECT
&&
598 entry
->type
!= IOMMU_RESV_DIRECT_RELAXABLE
)
601 for (addr
= start
; addr
< end
; addr
+= pg_size
) {
602 phys_addr_t phys_addr
;
604 phys_addr
= iommu_iova_to_phys(domain
, addr
);
608 ret
= iommu_map(domain
, addr
, addr
, pg_size
, entry
->prot
);
615 iommu_flush_tlb_all(domain
);
618 iommu_put_resv_regions(dev
, &mappings
);
624 * iommu_group_add_device - add a device to an iommu group
625 * @group: the group into which to add the device (reference should be held)
628 * This function is called by an iommu driver to add a device into a
629 * group. Adding a device increments the group reference count.
631 int iommu_group_add_device(struct iommu_group
*group
, struct device
*dev
)
634 struct group_device
*device
;
636 device
= kzalloc(sizeof(*device
), GFP_KERNEL
);
642 ret
= sysfs_create_link(&dev
->kobj
, &group
->kobj
, "iommu_group");
644 goto err_free_device
;
646 device
->name
= kasprintf(GFP_KERNEL
, "%s", kobject_name(&dev
->kobj
));
650 goto err_remove_link
;
653 ret
= sysfs_create_link_nowarn(group
->devices_kobj
,
654 &dev
->kobj
, device
->name
);
656 if (ret
== -EEXIST
&& i
>= 0) {
658 * Account for the slim chance of collision
659 * and append an instance to the name.
662 device
->name
= kasprintf(GFP_KERNEL
, "%s.%d",
663 kobject_name(&dev
->kobj
), i
++);
669 kobject_get(group
->devices_kobj
);
671 dev
->iommu_group
= group
;
673 iommu_group_create_direct_mappings(group
, dev
);
675 mutex_lock(&group
->mutex
);
676 list_add_tail(&device
->list
, &group
->devices
);
678 ret
= __iommu_attach_device(group
->domain
, dev
);
679 mutex_unlock(&group
->mutex
);
683 /* Notify any listeners about change to group. */
684 blocking_notifier_call_chain(&group
->notifier
,
685 IOMMU_GROUP_NOTIFY_ADD_DEVICE
, dev
);
687 trace_add_device_to_group(group
->id
, dev
);
689 dev_info(dev
, "Adding to iommu group %d\n", group
->id
);
694 mutex_lock(&group
->mutex
);
695 list_del(&device
->list
);
696 mutex_unlock(&group
->mutex
);
697 dev
->iommu_group
= NULL
;
698 kobject_put(group
->devices_kobj
);
702 sysfs_remove_link(&dev
->kobj
, "iommu_group");
705 dev_err(dev
, "Failed to add to iommu group %d: %d\n", group
->id
, ret
);
708 EXPORT_SYMBOL_GPL(iommu_group_add_device
);
711 * iommu_group_remove_device - remove a device from it's current group
712 * @dev: device to be removed
714 * This function is called by an iommu driver to remove the device from
715 * it's current group. This decrements the iommu group reference count.
717 void iommu_group_remove_device(struct device
*dev
)
719 struct iommu_group
*group
= dev
->iommu_group
;
720 struct group_device
*tmp_device
, *device
= NULL
;
722 dev_info(dev
, "Removing from iommu group %d\n", group
->id
);
724 /* Pre-notify listeners that a device is being removed. */
725 blocking_notifier_call_chain(&group
->notifier
,
726 IOMMU_GROUP_NOTIFY_DEL_DEVICE
, dev
);
728 mutex_lock(&group
->mutex
);
729 list_for_each_entry(tmp_device
, &group
->devices
, list
) {
730 if (tmp_device
->dev
== dev
) {
732 list_del(&device
->list
);
736 mutex_unlock(&group
->mutex
);
741 sysfs_remove_link(group
->devices_kobj
, device
->name
);
742 sysfs_remove_link(&dev
->kobj
, "iommu_group");
744 trace_remove_device_from_group(group
->id
, dev
);
748 dev
->iommu_group
= NULL
;
749 kobject_put(group
->devices_kobj
);
751 EXPORT_SYMBOL_GPL(iommu_group_remove_device
);
753 static int iommu_group_device_count(struct iommu_group
*group
)
755 struct group_device
*entry
;
758 list_for_each_entry(entry
, &group
->devices
, list
)
765 * iommu_group_for_each_dev - iterate over each device in the group
767 * @data: caller opaque data to be passed to callback function
768 * @fn: caller supplied callback function
770 * This function is called by group users to iterate over group devices.
771 * Callers should hold a reference count to the group during callback.
772 * The group->mutex is held across callbacks, which will block calls to
773 * iommu_group_add/remove_device.
775 static int __iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
776 int (*fn
)(struct device
*, void *))
778 struct group_device
*device
;
781 list_for_each_entry(device
, &group
->devices
, list
) {
782 ret
= fn(device
->dev
, data
);
790 int iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
791 int (*fn
)(struct device
*, void *))
795 mutex_lock(&group
->mutex
);
796 ret
= __iommu_group_for_each_dev(group
, data
, fn
);
797 mutex_unlock(&group
->mutex
);
801 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev
);
804 * iommu_group_get - Return the group for a device and increment reference
805 * @dev: get the group that this device belongs to
807 * This function is called by iommu drivers and users to get the group
808 * for the specified device. If found, the group is returned and the group
809 * reference in incremented, else NULL.
811 struct iommu_group
*iommu_group_get(struct device
*dev
)
813 struct iommu_group
*group
= dev
->iommu_group
;
816 kobject_get(group
->devices_kobj
);
820 EXPORT_SYMBOL_GPL(iommu_group_get
);
823 * iommu_group_ref_get - Increment reference on a group
824 * @group: the group to use, must not be NULL
826 * This function is called by iommu drivers to take additional references on an
827 * existing group. Returns the given group for convenience.
829 struct iommu_group
*iommu_group_ref_get(struct iommu_group
*group
)
831 kobject_get(group
->devices_kobj
);
836 * iommu_group_put - Decrement group reference
837 * @group: the group to use
839 * This function is called by iommu drivers and users to release the
840 * iommu group. Once the reference count is zero, the group is released.
842 void iommu_group_put(struct iommu_group
*group
)
845 kobject_put(group
->devices_kobj
);
847 EXPORT_SYMBOL_GPL(iommu_group_put
);
850 * iommu_group_register_notifier - Register a notifier for group changes
851 * @group: the group to watch
852 * @nb: notifier block to signal
854 * This function allows iommu group users to track changes in a group.
855 * See include/linux/iommu.h for actions sent via this notifier. Caller
856 * should hold a reference to the group throughout notifier registration.
858 int iommu_group_register_notifier(struct iommu_group
*group
,
859 struct notifier_block
*nb
)
861 return blocking_notifier_chain_register(&group
->notifier
, nb
);
863 EXPORT_SYMBOL_GPL(iommu_group_register_notifier
);
866 * iommu_group_unregister_notifier - Unregister a notifier
867 * @group: the group to watch
868 * @nb: notifier block to signal
870 * Unregister a previously registered group notifier block.
872 int iommu_group_unregister_notifier(struct iommu_group
*group
,
873 struct notifier_block
*nb
)
875 return blocking_notifier_chain_unregister(&group
->notifier
, nb
);
877 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier
);
880 * iommu_register_device_fault_handler() - Register a device fault handler
882 * @handler: the fault handler
883 * @data: private data passed as argument to the handler
885 * When an IOMMU fault event is received, this handler gets called with the
886 * fault event and data as argument. The handler should return 0 on success. If
887 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
888 * complete the fault by calling iommu_page_response() with one of the following
890 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
891 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
892 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
893 * page faults if possible.
895 * Return 0 if the fault handler was installed successfully, or an error.
897 int iommu_register_device_fault_handler(struct device
*dev
,
898 iommu_dev_fault_handler_t handler
,
901 struct iommu_param
*param
= dev
->iommu_param
;
907 mutex_lock(¶m
->lock
);
908 /* Only allow one fault handler registered for each device */
909 if (param
->fault_param
) {
915 param
->fault_param
= kzalloc(sizeof(*param
->fault_param
), GFP_KERNEL
);
916 if (!param
->fault_param
) {
921 param
->fault_param
->handler
= handler
;
922 param
->fault_param
->data
= data
;
923 mutex_init(¶m
->fault_param
->lock
);
924 INIT_LIST_HEAD(¶m
->fault_param
->faults
);
927 mutex_unlock(¶m
->lock
);
931 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler
);
934 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
937 * Remove the device fault handler installed with
938 * iommu_register_device_fault_handler().
940 * Return 0 on success, or an error.
942 int iommu_unregister_device_fault_handler(struct device
*dev
)
944 struct iommu_param
*param
= dev
->iommu_param
;
950 mutex_lock(¶m
->lock
);
952 if (!param
->fault_param
)
955 /* we cannot unregister handler if there are pending faults */
956 if (!list_empty(¶m
->fault_param
->faults
)) {
961 kfree(param
->fault_param
);
962 param
->fault_param
= NULL
;
965 mutex_unlock(¶m
->lock
);
969 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler
);
972 * iommu_report_device_fault() - Report fault event to device driver
974 * @evt: fault event data
976 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
977 * handler. When this function fails and the fault is recoverable, it is the
978 * caller's responsibility to complete the fault.
980 * Return 0 on success, or an error.
982 int iommu_report_device_fault(struct device
*dev
, struct iommu_fault_event
*evt
)
984 struct iommu_param
*param
= dev
->iommu_param
;
985 struct iommu_fault_event
*evt_pending
= NULL
;
986 struct iommu_fault_param
*fparam
;
992 /* we only report device fault if there is a handler registered */
993 mutex_lock(¶m
->lock
);
994 fparam
= param
->fault_param
;
995 if (!fparam
|| !fparam
->handler
) {
1000 if (evt
->fault
.type
== IOMMU_FAULT_PAGE_REQ
&&
1001 (evt
->fault
.prm
.flags
& IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE
)) {
1002 evt_pending
= kmemdup(evt
, sizeof(struct iommu_fault_event
),
1008 mutex_lock(&fparam
->lock
);
1009 list_add_tail(&evt_pending
->list
, &fparam
->faults
);
1010 mutex_unlock(&fparam
->lock
);
1013 ret
= fparam
->handler(&evt
->fault
, fparam
->data
);
1014 if (ret
&& evt_pending
) {
1015 mutex_lock(&fparam
->lock
);
1016 list_del(&evt_pending
->list
);
1017 mutex_unlock(&fparam
->lock
);
1021 mutex_unlock(¶m
->lock
);
1024 EXPORT_SYMBOL_GPL(iommu_report_device_fault
);
1026 int iommu_page_response(struct device
*dev
,
1027 struct iommu_page_response
*msg
)
1031 struct iommu_fault_event
*evt
;
1032 struct iommu_fault_page_request
*prm
;
1033 struct iommu_param
*param
= dev
->iommu_param
;
1034 struct iommu_domain
*domain
= iommu_get_domain_for_dev(dev
);
1036 if (!domain
|| !domain
->ops
->page_response
)
1039 if (!param
|| !param
->fault_param
)
1042 if (msg
->version
!= IOMMU_PAGE_RESP_VERSION_1
||
1043 msg
->flags
& ~IOMMU_PAGE_RESP_PASID_VALID
)
1046 /* Only send response if there is a fault report pending */
1047 mutex_lock(¶m
->fault_param
->lock
);
1048 if (list_empty(¶m
->fault_param
->faults
)) {
1049 dev_warn_ratelimited(dev
, "no pending PRQ, drop response\n");
1053 * Check if we have a matching page request pending to respond,
1054 * otherwise return -EINVAL
1056 list_for_each_entry(evt
, ¶m
->fault_param
->faults
, list
) {
1057 prm
= &evt
->fault
.prm
;
1058 pasid_valid
= prm
->flags
& IOMMU_FAULT_PAGE_REQUEST_PASID_VALID
;
1060 if ((pasid_valid
&& prm
->pasid
!= msg
->pasid
) ||
1061 prm
->grpid
!= msg
->grpid
)
1064 /* Sanitize the reply */
1065 msg
->flags
= pasid_valid
? IOMMU_PAGE_RESP_PASID_VALID
: 0;
1067 ret
= domain
->ops
->page_response(dev
, evt
, msg
);
1068 list_del(&evt
->list
);
1074 mutex_unlock(¶m
->fault_param
->lock
);
1077 EXPORT_SYMBOL_GPL(iommu_page_response
);
1080 * iommu_group_id - Return ID for a group
1081 * @group: the group to ID
1083 * Return the unique ID for the group matching the sysfs group number.
1085 int iommu_group_id(struct iommu_group
*group
)
1089 EXPORT_SYMBOL_GPL(iommu_group_id
);
1091 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1092 unsigned long *devfns
);
1095 * To consider a PCI device isolated, we require ACS to support Source
1096 * Validation, Request Redirection, Completer Redirection, and Upstream
1097 * Forwarding. This effectively means that devices cannot spoof their
1098 * requester ID, requests and completions cannot be redirected, and all
1099 * transactions are forwarded upstream, even as it passes through a
1100 * bridge where the target device is downstream.
1102 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1105 * For multifunction devices which are not isolated from each other, find
1106 * all the other non-isolated functions and look for existing groups. For
1107 * each function, we also need to look for aliases to or from other devices
1108 * that may already have a group.
1110 static struct iommu_group
*get_pci_function_alias_group(struct pci_dev
*pdev
,
1111 unsigned long *devfns
)
1113 struct pci_dev
*tmp
= NULL
;
1114 struct iommu_group
*group
;
1116 if (!pdev
->multifunction
|| pci_acs_enabled(pdev
, REQ_ACS_FLAGS
))
1119 for_each_pci_dev(tmp
) {
1120 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
||
1121 PCI_SLOT(tmp
->devfn
) != PCI_SLOT(pdev
->devfn
) ||
1122 pci_acs_enabled(tmp
, REQ_ACS_FLAGS
))
1125 group
= get_pci_alias_group(tmp
, devfns
);
1136 * Look for aliases to or from the given device for existing groups. DMA
1137 * aliases are only supported on the same bus, therefore the search
1138 * space is quite small (especially since we're really only looking at pcie
1139 * device, and therefore only expect multiple slots on the root complex or
1140 * downstream switch ports). It's conceivable though that a pair of
1141 * multifunction devices could have aliases between them that would cause a
1142 * loop. To prevent this, we use a bitmap to track where we've been.
1144 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1145 unsigned long *devfns
)
1147 struct pci_dev
*tmp
= NULL
;
1148 struct iommu_group
*group
;
1150 if (test_and_set_bit(pdev
->devfn
& 0xff, devfns
))
1153 group
= iommu_group_get(&pdev
->dev
);
1157 for_each_pci_dev(tmp
) {
1158 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
)
1161 /* We alias them or they alias us */
1162 if (pci_devs_are_dma_aliases(pdev
, tmp
)) {
1163 group
= get_pci_alias_group(tmp
, devfns
);
1169 group
= get_pci_function_alias_group(tmp
, devfns
);
1180 struct group_for_pci_data
{
1181 struct pci_dev
*pdev
;
1182 struct iommu_group
*group
;
1186 * DMA alias iterator callback, return the last seen device. Stop and return
1187 * the IOMMU group if we find one along the way.
1189 static int get_pci_alias_or_group(struct pci_dev
*pdev
, u16 alias
, void *opaque
)
1191 struct group_for_pci_data
*data
= opaque
;
1194 data
->group
= iommu_group_get(&pdev
->dev
);
1196 return data
->group
!= NULL
;
1200 * Generic device_group call-back function. It just allocates one
1201 * iommu-group per device.
1203 struct iommu_group
*generic_device_group(struct device
*dev
)
1205 return iommu_group_alloc();
1209 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1210 * to find or create an IOMMU group for a device.
1212 struct iommu_group
*pci_device_group(struct device
*dev
)
1214 struct pci_dev
*pdev
= to_pci_dev(dev
);
1215 struct group_for_pci_data data
;
1216 struct pci_bus
*bus
;
1217 struct iommu_group
*group
= NULL
;
1218 u64 devfns
[4] = { 0 };
1220 if (WARN_ON(!dev_is_pci(dev
)))
1221 return ERR_PTR(-EINVAL
);
1224 * Find the upstream DMA alias for the device. A device must not
1225 * be aliased due to topology in order to have its own IOMMU group.
1226 * If we find an alias along the way that already belongs to a
1229 if (pci_for_each_dma_alias(pdev
, get_pci_alias_or_group
, &data
))
1235 * Continue upstream from the point of minimum IOMMU granularity
1236 * due to aliases to the point where devices are protected from
1237 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1240 for (bus
= pdev
->bus
; !pci_is_root_bus(bus
); bus
= bus
->parent
) {
1244 if (pci_acs_path_enabled(bus
->self
, NULL
, REQ_ACS_FLAGS
))
1249 group
= iommu_group_get(&pdev
->dev
);
1255 * Look for existing groups on device aliases. If we alias another
1256 * device or another device aliases us, use the same group.
1258 group
= get_pci_alias_group(pdev
, (unsigned long *)devfns
);
1263 * Look for existing groups on non-isolated functions on the same
1264 * slot and aliases of those funcions, if any. No need to clear
1265 * the search bitmap, the tested devfns are still valid.
1267 group
= get_pci_function_alias_group(pdev
, (unsigned long *)devfns
);
1271 /* No shared group found, allocate new */
1272 return iommu_group_alloc();
1275 /* Get the IOMMU group for device on fsl-mc bus */
1276 struct iommu_group
*fsl_mc_device_group(struct device
*dev
)
1278 struct device
*cont_dev
= fsl_mc_cont_dev(dev
);
1279 struct iommu_group
*group
;
1281 group
= iommu_group_get(cont_dev
);
1283 group
= iommu_group_alloc();
1288 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1289 * @dev: target device
1291 * This function is intended to be called by IOMMU drivers and extended to
1292 * support common, bus-defined algorithms when determining or creating the
1293 * IOMMU group for a device. On success, the caller will hold a reference
1294 * to the returned IOMMU group, which will already include the provided
1295 * device. The reference should be released with iommu_group_put().
1297 struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
)
1299 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1300 struct iommu_group
*group
;
1303 group
= iommu_group_get(dev
);
1308 return ERR_PTR(-EINVAL
);
1310 group
= ops
->device_group(dev
);
1311 if (WARN_ON_ONCE(group
== NULL
))
1312 return ERR_PTR(-EINVAL
);
1318 * Try to allocate a default domain - needs support from the
1321 if (!group
->default_domain
) {
1322 struct iommu_domain
*dom
;
1324 dom
= __iommu_domain_alloc(dev
->bus
, iommu_def_domain_type
);
1325 if (!dom
&& iommu_def_domain_type
!= IOMMU_DOMAIN_DMA
) {
1326 dom
= __iommu_domain_alloc(dev
->bus
, IOMMU_DOMAIN_DMA
);
1329 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1330 iommu_def_domain_type
);
1334 group
->default_domain
= dom
;
1336 group
->domain
= dom
;
1338 if (dom
&& !iommu_dma_strict
) {
1340 iommu_domain_set_attr(dom
,
1341 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE
,
1346 ret
= iommu_group_add_device(group
, dev
);
1348 iommu_group_put(group
);
1349 return ERR_PTR(ret
);
1355 struct iommu_domain
*iommu_group_default_domain(struct iommu_group
*group
)
1357 return group
->default_domain
;
1360 static int add_iommu_group(struct device
*dev
, void *data
)
1362 int ret
= iommu_probe_device(dev
);
1365 * We ignore -ENODEV errors for now, as they just mean that the
1366 * device is not translated by an IOMMU. We still care about
1367 * other errors and fail to initialize when they happen.
1375 static int remove_iommu_group(struct device
*dev
, void *data
)
1377 iommu_release_device(dev
);
1382 static int iommu_bus_notifier(struct notifier_block
*nb
,
1383 unsigned long action
, void *data
)
1385 unsigned long group_action
= 0;
1386 struct device
*dev
= data
;
1387 struct iommu_group
*group
;
1390 * ADD/DEL call into iommu driver ops if provided, which may
1391 * result in ADD/DEL notifiers to group->notifier
1393 if (action
== BUS_NOTIFY_ADD_DEVICE
) {
1396 ret
= iommu_probe_device(dev
);
1397 return (ret
) ? NOTIFY_DONE
: NOTIFY_OK
;
1398 } else if (action
== BUS_NOTIFY_REMOVED_DEVICE
) {
1399 iommu_release_device(dev
);
1404 * Remaining BUS_NOTIFYs get filtered and republished to the
1405 * group, if anyone is listening
1407 group
= iommu_group_get(dev
);
1412 case BUS_NOTIFY_BIND_DRIVER
:
1413 group_action
= IOMMU_GROUP_NOTIFY_BIND_DRIVER
;
1415 case BUS_NOTIFY_BOUND_DRIVER
:
1416 group_action
= IOMMU_GROUP_NOTIFY_BOUND_DRIVER
;
1418 case BUS_NOTIFY_UNBIND_DRIVER
:
1419 group_action
= IOMMU_GROUP_NOTIFY_UNBIND_DRIVER
;
1421 case BUS_NOTIFY_UNBOUND_DRIVER
:
1422 group_action
= IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER
;
1427 blocking_notifier_call_chain(&group
->notifier
,
1430 iommu_group_put(group
);
1434 static int iommu_bus_init(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1437 struct notifier_block
*nb
;
1439 nb
= kzalloc(sizeof(struct notifier_block
), GFP_KERNEL
);
1443 nb
->notifier_call
= iommu_bus_notifier
;
1445 err
= bus_register_notifier(bus
, nb
);
1449 err
= bus_for_each_dev(bus
, NULL
, NULL
, add_iommu_group
);
1458 bus_for_each_dev(bus
, NULL
, NULL
, remove_iommu_group
);
1459 bus_unregister_notifier(bus
, nb
);
1468 * bus_set_iommu - set iommu-callbacks for the bus
1470 * @ops: the callbacks provided by the iommu-driver
1472 * This function is called by an iommu driver to set the iommu methods
1473 * used for a particular bus. Drivers for devices on that bus can use
1474 * the iommu-api after these ops are registered.
1475 * This special function is needed because IOMMUs are usually devices on
1476 * the bus itself, so the iommu drivers are not initialized when the bus
1477 * is set up. With this function the iommu-driver can set the iommu-ops
1480 int bus_set_iommu(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1484 if (bus
->iommu_ops
!= NULL
)
1487 bus
->iommu_ops
= ops
;
1489 /* Do IOMMU specific setup for this bus-type */
1490 err
= iommu_bus_init(bus
, ops
);
1492 bus
->iommu_ops
= NULL
;
1496 EXPORT_SYMBOL_GPL(bus_set_iommu
);
1498 bool iommu_present(struct bus_type
*bus
)
1500 return bus
->iommu_ops
!= NULL
;
1502 EXPORT_SYMBOL_GPL(iommu_present
);
1504 bool iommu_capable(struct bus_type
*bus
, enum iommu_cap cap
)
1506 if (!bus
->iommu_ops
|| !bus
->iommu_ops
->capable
)
1509 return bus
->iommu_ops
->capable(cap
);
1511 EXPORT_SYMBOL_GPL(iommu_capable
);
1514 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1515 * @domain: iommu domain
1516 * @handler: fault handler
1517 * @token: user data, will be passed back to the fault handler
1519 * This function should be used by IOMMU users which want to be notified
1520 * whenever an IOMMU fault happens.
1522 * The fault handler itself should return 0 on success, and an appropriate
1523 * error code otherwise.
1525 void iommu_set_fault_handler(struct iommu_domain
*domain
,
1526 iommu_fault_handler_t handler
,
1531 domain
->handler
= handler
;
1532 domain
->handler_token
= token
;
1534 EXPORT_SYMBOL_GPL(iommu_set_fault_handler
);
1536 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
1539 struct iommu_domain
*domain
;
1541 if (bus
== NULL
|| bus
->iommu_ops
== NULL
)
1544 domain
= bus
->iommu_ops
->domain_alloc(type
);
1548 domain
->ops
= bus
->iommu_ops
;
1549 domain
->type
= type
;
1550 /* Assume all sizes by default; the driver may override this later */
1551 domain
->pgsize_bitmap
= bus
->iommu_ops
->pgsize_bitmap
;
1556 struct iommu_domain
*iommu_domain_alloc(struct bus_type
*bus
)
1558 return __iommu_domain_alloc(bus
, IOMMU_DOMAIN_UNMANAGED
);
1560 EXPORT_SYMBOL_GPL(iommu_domain_alloc
);
1562 void iommu_domain_free(struct iommu_domain
*domain
)
1564 domain
->ops
->domain_free(domain
);
1566 EXPORT_SYMBOL_GPL(iommu_domain_free
);
1568 static int __iommu_attach_device(struct iommu_domain
*domain
,
1572 if ((domain
->ops
->is_attach_deferred
!= NULL
) &&
1573 domain
->ops
->is_attach_deferred(domain
, dev
))
1576 if (unlikely(domain
->ops
->attach_dev
== NULL
))
1579 ret
= domain
->ops
->attach_dev(domain
, dev
);
1581 trace_attach_device_to_domain(dev
);
1585 int iommu_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
1587 struct iommu_group
*group
;
1590 group
= iommu_group_get(dev
);
1595 * Lock the group to make sure the device-count doesn't
1596 * change while we are attaching
1598 mutex_lock(&group
->mutex
);
1600 if (iommu_group_device_count(group
) != 1)
1603 ret
= __iommu_attach_group(domain
, group
);
1606 mutex_unlock(&group
->mutex
);
1607 iommu_group_put(group
);
1611 EXPORT_SYMBOL_GPL(iommu_attach_device
);
1613 static void __iommu_detach_device(struct iommu_domain
*domain
,
1616 if ((domain
->ops
->is_attach_deferred
!= NULL
) &&
1617 domain
->ops
->is_attach_deferred(domain
, dev
))
1620 if (unlikely(domain
->ops
->detach_dev
== NULL
))
1623 domain
->ops
->detach_dev(domain
, dev
);
1624 trace_detach_device_from_domain(dev
);
1627 void iommu_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
1629 struct iommu_group
*group
;
1631 group
= iommu_group_get(dev
);
1635 mutex_lock(&group
->mutex
);
1636 if (iommu_group_device_count(group
) != 1) {
1641 __iommu_detach_group(domain
, group
);
1644 mutex_unlock(&group
->mutex
);
1645 iommu_group_put(group
);
1647 EXPORT_SYMBOL_GPL(iommu_detach_device
);
1649 struct iommu_domain
*iommu_get_domain_for_dev(struct device
*dev
)
1651 struct iommu_domain
*domain
;
1652 struct iommu_group
*group
;
1654 group
= iommu_group_get(dev
);
1658 domain
= group
->domain
;
1660 iommu_group_put(group
);
1664 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev
);
1667 * For IOMMU_DOMAIN_DMA implementations which already provide their own
1668 * guarantees that the group and its default domain are valid and correct.
1670 struct iommu_domain
*iommu_get_dma_domain(struct device
*dev
)
1672 return dev
->iommu_group
->default_domain
;
1676 * IOMMU groups are really the natural working unit of the IOMMU, but
1677 * the IOMMU API works on domains and devices. Bridge that gap by
1678 * iterating over the devices in a group. Ideally we'd have a single
1679 * device which represents the requestor ID of the group, but we also
1680 * allow IOMMU drivers to create policy defined minimum sets, where
1681 * the physical hardware may be able to distiguish members, but we
1682 * wish to group them at a higher level (ex. untrusted multi-function
1683 * PCI devices). Thus we attach each device.
1685 static int iommu_group_do_attach_device(struct device
*dev
, void *data
)
1687 struct iommu_domain
*domain
= data
;
1689 return __iommu_attach_device(domain
, dev
);
1692 static int __iommu_attach_group(struct iommu_domain
*domain
,
1693 struct iommu_group
*group
)
1697 if (group
->default_domain
&& group
->domain
!= group
->default_domain
)
1700 ret
= __iommu_group_for_each_dev(group
, domain
,
1701 iommu_group_do_attach_device
);
1703 group
->domain
= domain
;
1708 int iommu_attach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1712 mutex_lock(&group
->mutex
);
1713 ret
= __iommu_attach_group(domain
, group
);
1714 mutex_unlock(&group
->mutex
);
1718 EXPORT_SYMBOL_GPL(iommu_attach_group
);
1720 static int iommu_group_do_detach_device(struct device
*dev
, void *data
)
1722 struct iommu_domain
*domain
= data
;
1724 __iommu_detach_device(domain
, dev
);
1729 static void __iommu_detach_group(struct iommu_domain
*domain
,
1730 struct iommu_group
*group
)
1734 if (!group
->default_domain
) {
1735 __iommu_group_for_each_dev(group
, domain
,
1736 iommu_group_do_detach_device
);
1737 group
->domain
= NULL
;
1741 if (group
->domain
== group
->default_domain
)
1744 /* Detach by re-attaching to the default domain */
1745 ret
= __iommu_group_for_each_dev(group
, group
->default_domain
,
1746 iommu_group_do_attach_device
);
1750 group
->domain
= group
->default_domain
;
1753 void iommu_detach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1755 mutex_lock(&group
->mutex
);
1756 __iommu_detach_group(domain
, group
);
1757 mutex_unlock(&group
->mutex
);
1759 EXPORT_SYMBOL_GPL(iommu_detach_group
);
1761 phys_addr_t
iommu_iova_to_phys(struct iommu_domain
*domain
, dma_addr_t iova
)
1763 if (unlikely(domain
->ops
->iova_to_phys
== NULL
))
1766 return domain
->ops
->iova_to_phys(domain
, iova
);
1768 EXPORT_SYMBOL_GPL(iommu_iova_to_phys
);
1770 static size_t iommu_pgsize(struct iommu_domain
*domain
,
1771 unsigned long addr_merge
, size_t size
)
1773 unsigned int pgsize_idx
;
1776 /* Max page size that still fits into 'size' */
1777 pgsize_idx
= __fls(size
);
1779 /* need to consider alignment requirements ? */
1780 if (likely(addr_merge
)) {
1781 /* Max page size allowed by address */
1782 unsigned int align_pgsize_idx
= __ffs(addr_merge
);
1783 pgsize_idx
= min(pgsize_idx
, align_pgsize_idx
);
1786 /* build a mask of acceptable page sizes */
1787 pgsize
= (1UL << (pgsize_idx
+ 1)) - 1;
1789 /* throw away page sizes not supported by the hardware */
1790 pgsize
&= domain
->pgsize_bitmap
;
1792 /* make sure we're still sane */
1795 /* pick the biggest page */
1796 pgsize_idx
= __fls(pgsize
);
1797 pgsize
= 1UL << pgsize_idx
;
1802 int iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
1803 phys_addr_t paddr
, size_t size
, int prot
)
1805 const struct iommu_ops
*ops
= domain
->ops
;
1806 unsigned long orig_iova
= iova
;
1807 unsigned int min_pagesz
;
1808 size_t orig_size
= size
;
1809 phys_addr_t orig_paddr
= paddr
;
1812 if (unlikely(ops
->map
== NULL
||
1813 domain
->pgsize_bitmap
== 0UL))
1816 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1819 /* find out the minimum page size supported */
1820 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1823 * both the virtual address and the physical one, as well as
1824 * the size of the mapping, must be aligned (at least) to the
1825 * size of the smallest page supported by the hardware
1827 if (!IS_ALIGNED(iova
| paddr
| size
, min_pagesz
)) {
1828 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1829 iova
, &paddr
, size
, min_pagesz
);
1833 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova
, &paddr
, size
);
1836 size_t pgsize
= iommu_pgsize(domain
, iova
| paddr
, size
);
1838 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1839 iova
, &paddr
, pgsize
);
1841 ret
= ops
->map(domain
, iova
, paddr
, pgsize
, prot
);
1850 if (ops
->iotlb_sync_map
)
1851 ops
->iotlb_sync_map(domain
);
1853 /* unroll mapping in case something went wrong */
1855 iommu_unmap(domain
, orig_iova
, orig_size
- size
);
1857 trace_map(orig_iova
, orig_paddr
, orig_size
);
1861 EXPORT_SYMBOL_GPL(iommu_map
);
1863 static size_t __iommu_unmap(struct iommu_domain
*domain
,
1864 unsigned long iova
, size_t size
,
1867 const struct iommu_ops
*ops
= domain
->ops
;
1868 size_t unmapped_page
, unmapped
= 0;
1869 unsigned long orig_iova
= iova
;
1870 unsigned int min_pagesz
;
1872 if (unlikely(ops
->unmap
== NULL
||
1873 domain
->pgsize_bitmap
== 0UL))
1876 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1879 /* find out the minimum page size supported */
1880 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1883 * The virtual address, as well as the size of the mapping, must be
1884 * aligned (at least) to the size of the smallest page supported
1887 if (!IS_ALIGNED(iova
| size
, min_pagesz
)) {
1888 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1889 iova
, size
, min_pagesz
);
1893 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova
, size
);
1896 * Keep iterating until we either unmap 'size' bytes (or more)
1897 * or we hit an area that isn't mapped.
1899 while (unmapped
< size
) {
1900 size_t pgsize
= iommu_pgsize(domain
, iova
, size
- unmapped
);
1902 unmapped_page
= ops
->unmap(domain
, iova
, pgsize
);
1906 if (sync
&& ops
->iotlb_range_add
)
1907 ops
->iotlb_range_add(domain
, iova
, pgsize
);
1909 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1910 iova
, unmapped_page
);
1912 iova
+= unmapped_page
;
1913 unmapped
+= unmapped_page
;
1916 if (sync
&& ops
->iotlb_sync
)
1917 ops
->iotlb_sync(domain
);
1919 trace_unmap(orig_iova
, size
, unmapped
);
1923 size_t iommu_unmap(struct iommu_domain
*domain
,
1924 unsigned long iova
, size_t size
)
1926 return __iommu_unmap(domain
, iova
, size
, true);
1928 EXPORT_SYMBOL_GPL(iommu_unmap
);
1930 size_t iommu_unmap_fast(struct iommu_domain
*domain
,
1931 unsigned long iova
, size_t size
)
1933 return __iommu_unmap(domain
, iova
, size
, false);
1935 EXPORT_SYMBOL_GPL(iommu_unmap_fast
);
1937 size_t iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
1938 struct scatterlist
*sg
, unsigned int nents
, int prot
)
1940 size_t len
= 0, mapped
= 0;
1945 while (i
<= nents
) {
1946 phys_addr_t s_phys
= sg_phys(sg
);
1948 if (len
&& s_phys
!= start
+ len
) {
1949 ret
= iommu_map(domain
, iova
+ mapped
, start
, len
, prot
);
1971 /* undo mappings already done */
1972 iommu_unmap(domain
, iova
, mapped
);
1977 EXPORT_SYMBOL_GPL(iommu_map_sg
);
1979 int iommu_domain_window_enable(struct iommu_domain
*domain
, u32 wnd_nr
,
1980 phys_addr_t paddr
, u64 size
, int prot
)
1982 if (unlikely(domain
->ops
->domain_window_enable
== NULL
))
1985 return domain
->ops
->domain_window_enable(domain
, wnd_nr
, paddr
, size
,
1988 EXPORT_SYMBOL_GPL(iommu_domain_window_enable
);
1990 void iommu_domain_window_disable(struct iommu_domain
*domain
, u32 wnd_nr
)
1992 if (unlikely(domain
->ops
->domain_window_disable
== NULL
))
1995 return domain
->ops
->domain_window_disable(domain
, wnd_nr
);
1997 EXPORT_SYMBOL_GPL(iommu_domain_window_disable
);
2000 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2001 * @domain: the iommu domain where the fault has happened
2002 * @dev: the device where the fault has happened
2003 * @iova: the faulting address
2004 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2006 * This function should be called by the low-level IOMMU implementations
2007 * whenever IOMMU faults happen, to allow high-level users, that are
2008 * interested in such events, to know about them.
2010 * This event may be useful for several possible use cases:
2011 * - mere logging of the event
2012 * - dynamic TLB/PTE loading
2013 * - if restarting of the faulting device is required
2015 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2016 * PTE/TLB loading will one day be supported, implementations will be able
2017 * to tell whether it succeeded or not according to this return value).
2019 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2020 * (though fault handlers can also return -ENOSYS, in case they want to
2021 * elicit the default behavior of the IOMMU drivers).
2023 int report_iommu_fault(struct iommu_domain
*domain
, struct device
*dev
,
2024 unsigned long iova
, int flags
)
2029 * if upper layers showed interest and installed a fault handler,
2032 if (domain
->handler
)
2033 ret
= domain
->handler(domain
, dev
, iova
, flags
,
2034 domain
->handler_token
);
2036 trace_io_page_fault(dev
, iova
, flags
);
2039 EXPORT_SYMBOL_GPL(report_iommu_fault
);
2041 static int __init
iommu_init(void)
2043 iommu_group_kset
= kset_create_and_add("iommu_groups",
2045 BUG_ON(!iommu_group_kset
);
2047 iommu_debugfs_setup();
2051 core_initcall(iommu_init
);
2053 int iommu_domain_get_attr(struct iommu_domain
*domain
,
2054 enum iommu_attr attr
, void *data
)
2056 struct iommu_domain_geometry
*geometry
;
2061 case DOMAIN_ATTR_GEOMETRY
:
2063 *geometry
= domain
->geometry
;
2066 case DOMAIN_ATTR_PAGING
:
2068 *paging
= (domain
->pgsize_bitmap
!= 0UL);
2071 if (!domain
->ops
->domain_get_attr
)
2074 ret
= domain
->ops
->domain_get_attr(domain
, attr
, data
);
2079 EXPORT_SYMBOL_GPL(iommu_domain_get_attr
);
2081 int iommu_domain_set_attr(struct iommu_domain
*domain
,
2082 enum iommu_attr attr
, void *data
)
2088 if (domain
->ops
->domain_set_attr
== NULL
)
2091 ret
= domain
->ops
->domain_set_attr(domain
, attr
, data
);
2096 EXPORT_SYMBOL_GPL(iommu_domain_set_attr
);
2098 void iommu_get_resv_regions(struct device
*dev
, struct list_head
*list
)
2100 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2102 if (ops
&& ops
->get_resv_regions
)
2103 ops
->get_resv_regions(dev
, list
);
2106 void iommu_put_resv_regions(struct device
*dev
, struct list_head
*list
)
2108 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2110 if (ops
&& ops
->put_resv_regions
)
2111 ops
->put_resv_regions(dev
, list
);
2114 struct iommu_resv_region
*iommu_alloc_resv_region(phys_addr_t start
,
2115 size_t length
, int prot
,
2116 enum iommu_resv_type type
)
2118 struct iommu_resv_region
*region
;
2120 region
= kzalloc(sizeof(*region
), GFP_KERNEL
);
2124 INIT_LIST_HEAD(®ion
->list
);
2125 region
->start
= start
;
2126 region
->length
= length
;
2127 region
->prot
= prot
;
2128 region
->type
= type
;
2133 request_default_domain_for_dev(struct device
*dev
, unsigned long type
)
2135 struct iommu_domain
*domain
;
2136 struct iommu_group
*group
;
2139 /* Device must already be in a group before calling this function */
2140 group
= iommu_group_get(dev
);
2144 mutex_lock(&group
->mutex
);
2146 /* Check if the default domain is already direct mapped */
2148 if (group
->default_domain
&& group
->default_domain
->type
== type
)
2151 /* Don't change mappings of existing devices */
2153 if (iommu_group_device_count(group
) != 1)
2156 /* Allocate a direct mapped domain */
2158 domain
= __iommu_domain_alloc(dev
->bus
, type
);
2162 /* Attach the device to the domain */
2163 ret
= __iommu_attach_group(domain
, group
);
2165 iommu_domain_free(domain
);
2169 iommu_group_create_direct_mappings(group
, dev
);
2171 /* Make the direct mapped domain the default for this group */
2172 if (group
->default_domain
)
2173 iommu_domain_free(group
->default_domain
);
2174 group
->default_domain
= domain
;
2176 dev_info(dev
, "Using iommu %s mapping\n",
2177 type
== IOMMU_DOMAIN_DMA
? "dma" : "direct");
2181 mutex_unlock(&group
->mutex
);
2182 iommu_group_put(group
);
2187 /* Request that a device is direct mapped by the IOMMU */
2188 int iommu_request_dm_for_dev(struct device
*dev
)
2190 return request_default_domain_for_dev(dev
, IOMMU_DOMAIN_IDENTITY
);
2193 /* Request that a device can't be direct mapped by the IOMMU */
2194 int iommu_request_dma_domain_for_dev(struct device
*dev
)
2196 return request_default_domain_for_dev(dev
, IOMMU_DOMAIN_DMA
);
2199 const struct iommu_ops
*iommu_ops_from_fwnode(struct fwnode_handle
*fwnode
)
2201 const struct iommu_ops
*ops
= NULL
;
2202 struct iommu_device
*iommu
;
2204 spin_lock(&iommu_device_lock
);
2205 list_for_each_entry(iommu
, &iommu_device_list
, list
)
2206 if (iommu
->fwnode
== fwnode
) {
2210 spin_unlock(&iommu_device_lock
);
2214 int iommu_fwspec_init(struct device
*dev
, struct fwnode_handle
*iommu_fwnode
,
2215 const struct iommu_ops
*ops
)
2217 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2220 return ops
== fwspec
->ops
? 0 : -EINVAL
;
2222 fwspec
= kzalloc(sizeof(*fwspec
), GFP_KERNEL
);
2226 of_node_get(to_of_node(iommu_fwnode
));
2227 fwspec
->iommu_fwnode
= iommu_fwnode
;
2229 dev_iommu_fwspec_set(dev
, fwspec
);
2232 EXPORT_SYMBOL_GPL(iommu_fwspec_init
);
2234 void iommu_fwspec_free(struct device
*dev
)
2236 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2239 fwnode_handle_put(fwspec
->iommu_fwnode
);
2241 dev_iommu_fwspec_set(dev
, NULL
);
2244 EXPORT_SYMBOL_GPL(iommu_fwspec_free
);
2246 int iommu_fwspec_add_ids(struct device
*dev
, u32
*ids
, int num_ids
)
2248 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2255 size
= offsetof(struct iommu_fwspec
, ids
[fwspec
->num_ids
+ num_ids
]);
2256 if (size
> sizeof(*fwspec
)) {
2257 fwspec
= krealloc(fwspec
, size
, GFP_KERNEL
);
2261 dev_iommu_fwspec_set(dev
, fwspec
);
2264 for (i
= 0; i
< num_ids
; i
++)
2265 fwspec
->ids
[fwspec
->num_ids
+ i
] = ids
[i
];
2267 fwspec
->num_ids
+= num_ids
;
2270 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids
);
2273 * Per device IOMMU features.
2275 bool iommu_dev_has_feature(struct device
*dev
, enum iommu_dev_features feat
)
2277 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2279 if (ops
&& ops
->dev_has_feat
)
2280 return ops
->dev_has_feat(dev
, feat
);
2284 EXPORT_SYMBOL_GPL(iommu_dev_has_feature
);
2286 int iommu_dev_enable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2288 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2290 if (ops
&& ops
->dev_enable_feat
)
2291 return ops
->dev_enable_feat(dev
, feat
);
2295 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature
);
2298 * The device drivers should do the necessary cleanups before calling this.
2299 * For example, before disabling the aux-domain feature, the device driver
2300 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2302 int iommu_dev_disable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2304 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2306 if (ops
&& ops
->dev_disable_feat
)
2307 return ops
->dev_disable_feat(dev
, feat
);
2311 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature
);
2313 bool iommu_dev_feature_enabled(struct device
*dev
, enum iommu_dev_features feat
)
2315 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2317 if (ops
&& ops
->dev_feat_enabled
)
2318 return ops
->dev_feat_enabled(dev
, feat
);
2322 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled
);
2325 * Aux-domain specific attach/detach.
2327 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2328 * true. Also, as long as domains are attached to a device through this
2329 * interface, any tries to call iommu_attach_device() should fail
2330 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2331 * This should make us safe against a device being attached to a guest as a
2332 * whole while there are still pasid users on it (aux and sva).
2334 int iommu_aux_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
2338 if (domain
->ops
->aux_attach_dev
)
2339 ret
= domain
->ops
->aux_attach_dev(domain
, dev
);
2342 trace_attach_device_to_domain(dev
);
2346 EXPORT_SYMBOL_GPL(iommu_aux_attach_device
);
2348 void iommu_aux_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
2350 if (domain
->ops
->aux_detach_dev
) {
2351 domain
->ops
->aux_detach_dev(domain
, dev
);
2352 trace_detach_device_from_domain(dev
);
2355 EXPORT_SYMBOL_GPL(iommu_aux_detach_device
);
2357 int iommu_aux_get_pasid(struct iommu_domain
*domain
, struct device
*dev
)
2361 if (domain
->ops
->aux_get_pasid
)
2362 ret
= domain
->ops
->aux_get_pasid(domain
, dev
);
2366 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid
);
2369 * iommu_sva_bind_device() - Bind a process address space to a device
2371 * @mm: the mm to bind, caller must hold a reference to it
2373 * Create a bond between device and address space, allowing the device to access
2374 * the mm using the returned PASID. If a bond already exists between @device and
2375 * @mm, it is returned and an additional reference is taken. Caller must call
2376 * iommu_sva_unbind_device() to release each reference.
2378 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2379 * initialize the required SVA features.
2381 * On error, returns an ERR_PTR value.
2384 iommu_sva_bind_device(struct device
*dev
, struct mm_struct
*mm
, void *drvdata
)
2386 struct iommu_group
*group
;
2387 struct iommu_sva
*handle
= ERR_PTR(-EINVAL
);
2388 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2390 if (!ops
|| !ops
->sva_bind
)
2391 return ERR_PTR(-ENODEV
);
2393 group
= iommu_group_get(dev
);
2395 return ERR_PTR(-ENODEV
);
2397 /* Ensure device count and domain don't change while we're binding */
2398 mutex_lock(&group
->mutex
);
2401 * To keep things simple, SVA currently doesn't support IOMMU groups
2402 * with more than one device. Existing SVA-capable systems are not
2403 * affected by the problems that required IOMMU groups (lack of ACS
2404 * isolation, device ID aliasing and other hardware issues).
2406 if (iommu_group_device_count(group
) != 1)
2409 handle
= ops
->sva_bind(dev
, mm
, drvdata
);
2412 mutex_unlock(&group
->mutex
);
2413 iommu_group_put(group
);
2417 EXPORT_SYMBOL_GPL(iommu_sva_bind_device
);
2420 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2421 * @handle: the handle returned by iommu_sva_bind_device()
2423 * Put reference to a bond between device and address space. The device should
2424 * not be issuing any more transaction for this PASID. All outstanding page
2425 * requests for this PASID must have been flushed to the IOMMU.
2427 * Returns 0 on success, or an error value
2429 void iommu_sva_unbind_device(struct iommu_sva
*handle
)
2431 struct iommu_group
*group
;
2432 struct device
*dev
= handle
->dev
;
2433 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2435 if (!ops
|| !ops
->sva_unbind
)
2438 group
= iommu_group_get(dev
);
2442 mutex_lock(&group
->mutex
);
2443 ops
->sva_unbind(handle
);
2444 mutex_unlock(&group
->mutex
);
2446 iommu_group_put(group
);
2448 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device
);
2450 int iommu_sva_set_ops(struct iommu_sva
*handle
,
2451 const struct iommu_sva_ops
*sva_ops
)
2453 if (handle
->ops
&& handle
->ops
!= sva_ops
)
2456 handle
->ops
= sva_ops
;
2459 EXPORT_SYMBOL_GPL(iommu_sva_set_ops
);
2461 int iommu_sva_get_pasid(struct iommu_sva
*handle
)
2463 const struct iommu_ops
*ops
= handle
->dev
->bus
->iommu_ops
;
2465 if (!ops
|| !ops
->sva_get_pasid
)
2466 return IOMMU_PASID_INVALID
;
2468 return ops
->sva_get_pasid(handle
);
2470 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid
);