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
);
30 static unsigned int iommu_def_domain_type __read_mostly
;
31 static bool iommu_dma_strict __read_mostly
= true;
32 static u32 iommu_cmd_line __read_mostly
;
36 struct kobject
*devices_kobj
;
37 struct list_head devices
;
39 struct blocking_notifier_head notifier
;
41 void (*iommu_data_release
)(void *iommu_data
);
44 struct iommu_domain
*default_domain
;
45 struct iommu_domain
*domain
;
49 struct list_head list
;
54 struct iommu_group_attribute
{
55 struct attribute attr
;
56 ssize_t (*show
)(struct iommu_group
*group
, char *buf
);
57 ssize_t (*store
)(struct iommu_group
*group
,
58 const char *buf
, size_t count
);
61 static const char * const iommu_group_resv_type_string
[] = {
62 [IOMMU_RESV_DIRECT
] = "direct",
63 [IOMMU_RESV_DIRECT_RELAXABLE
] = "direct-relaxable",
64 [IOMMU_RESV_RESERVED
] = "reserved",
65 [IOMMU_RESV_MSI
] = "msi",
66 [IOMMU_RESV_SW_MSI
] = "msi",
69 #define IOMMU_CMD_LINE_DMA_API BIT(0)
71 static void iommu_set_cmd_line_dma_api(void)
73 iommu_cmd_line
|= IOMMU_CMD_LINE_DMA_API
;
76 static bool iommu_cmd_line_dma_api(void)
78 return !!(iommu_cmd_line
& IOMMU_CMD_LINE_DMA_API
);
81 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
82 struct iommu_group_attribute iommu_group_attr_##_name = \
83 __ATTR(_name, _mode, _show, _store)
85 #define to_iommu_group_attr(_attr) \
86 container_of(_attr, struct iommu_group_attribute, attr)
87 #define to_iommu_group(_kobj) \
88 container_of(_kobj, struct iommu_group, kobj)
90 static LIST_HEAD(iommu_device_list
);
91 static DEFINE_SPINLOCK(iommu_device_lock
);
94 * Use a function instead of an array here because the domain-type is a
95 * bit-field, so an array would waste memory.
97 static const char *iommu_domain_type_str(unsigned int t
)
100 case IOMMU_DOMAIN_BLOCKED
:
102 case IOMMU_DOMAIN_IDENTITY
:
103 return "Passthrough";
104 case IOMMU_DOMAIN_UNMANAGED
:
106 case IOMMU_DOMAIN_DMA
:
113 static int __init
iommu_subsys_init(void)
115 bool cmd_line
= iommu_cmd_line_dma_api();
118 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH
))
119 iommu_set_default_passthrough(false);
121 iommu_set_default_translated(false);
123 if (iommu_default_passthrough() && mem_encrypt_active()) {
124 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
125 iommu_set_default_translated(false);
129 pr_info("Default domain type: %s %s\n",
130 iommu_domain_type_str(iommu_def_domain_type
),
131 cmd_line
? "(set via kernel command line)" : "");
135 subsys_initcall(iommu_subsys_init
);
137 int iommu_device_register(struct iommu_device
*iommu
)
139 spin_lock(&iommu_device_lock
);
140 list_add_tail(&iommu
->list
, &iommu_device_list
);
141 spin_unlock(&iommu_device_lock
);
145 void iommu_device_unregister(struct iommu_device
*iommu
)
147 spin_lock(&iommu_device_lock
);
148 list_del(&iommu
->list
);
149 spin_unlock(&iommu_device_lock
);
152 static struct iommu_param
*iommu_get_dev_param(struct device
*dev
)
154 struct iommu_param
*param
= dev
->iommu_param
;
159 param
= kzalloc(sizeof(*param
), GFP_KERNEL
);
163 mutex_init(¶m
->lock
);
164 dev
->iommu_param
= param
;
168 static void iommu_free_dev_param(struct device
*dev
)
170 kfree(dev
->iommu_param
);
171 dev
->iommu_param
= NULL
;
174 int iommu_probe_device(struct device
*dev
)
176 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
179 WARN_ON(dev
->iommu_group
);
183 if (!iommu_get_dev_param(dev
))
186 ret
= ops
->add_device(dev
);
188 iommu_free_dev_param(dev
);
193 void iommu_release_device(struct device
*dev
)
195 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
197 if (dev
->iommu_group
)
198 ops
->remove_device(dev
);
200 iommu_free_dev_param(dev
);
203 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
205 static int __iommu_attach_device(struct iommu_domain
*domain
,
207 static int __iommu_attach_group(struct iommu_domain
*domain
,
208 struct iommu_group
*group
);
209 static void __iommu_detach_group(struct iommu_domain
*domain
,
210 struct iommu_group
*group
);
212 static int __init
iommu_set_def_domain_type(char *str
)
217 ret
= kstrtobool(str
, &pt
);
222 iommu_set_default_passthrough(true);
224 iommu_set_default_translated(true);
228 early_param("iommu.passthrough", iommu_set_def_domain_type
);
230 static int __init
iommu_dma_setup(char *str
)
232 return kstrtobool(str
, &iommu_dma_strict
);
234 early_param("iommu.strict", iommu_dma_setup
);
236 static ssize_t
iommu_group_attr_show(struct kobject
*kobj
,
237 struct attribute
*__attr
, char *buf
)
239 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
240 struct iommu_group
*group
= to_iommu_group(kobj
);
244 ret
= attr
->show(group
, buf
);
248 static ssize_t
iommu_group_attr_store(struct kobject
*kobj
,
249 struct attribute
*__attr
,
250 const char *buf
, size_t count
)
252 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
253 struct iommu_group
*group
= to_iommu_group(kobj
);
257 ret
= attr
->store(group
, buf
, count
);
261 static const struct sysfs_ops iommu_group_sysfs_ops
= {
262 .show
= iommu_group_attr_show
,
263 .store
= iommu_group_attr_store
,
266 static int iommu_group_create_file(struct iommu_group
*group
,
267 struct iommu_group_attribute
*attr
)
269 return sysfs_create_file(&group
->kobj
, &attr
->attr
);
272 static void iommu_group_remove_file(struct iommu_group
*group
,
273 struct iommu_group_attribute
*attr
)
275 sysfs_remove_file(&group
->kobj
, &attr
->attr
);
278 static ssize_t
iommu_group_show_name(struct iommu_group
*group
, char *buf
)
280 return sprintf(buf
, "%s\n", group
->name
);
284 * iommu_insert_resv_region - Insert a new region in the
285 * list of reserved regions.
286 * @new: new region to insert
287 * @regions: list of regions
289 * Elements are sorted by start address and overlapping segments
290 * of the same type are merged.
292 int iommu_insert_resv_region(struct iommu_resv_region
*new,
293 struct list_head
*regions
)
295 struct iommu_resv_region
*iter
, *tmp
, *nr
, *top
;
298 nr
= iommu_alloc_resv_region(new->start
, new->length
,
299 new->prot
, new->type
);
303 /* First add the new element based on start address sorting */
304 list_for_each_entry(iter
, regions
, list
) {
305 if (nr
->start
< iter
->start
||
306 (nr
->start
== iter
->start
&& nr
->type
<= iter
->type
))
309 list_add_tail(&nr
->list
, &iter
->list
);
311 /* Merge overlapping segments of type nr->type in @regions, if any */
312 list_for_each_entry_safe(iter
, tmp
, regions
, list
) {
313 phys_addr_t top_end
, iter_end
= iter
->start
+ iter
->length
- 1;
315 /* no merge needed on elements of different types than @new */
316 if (iter
->type
!= new->type
) {
317 list_move_tail(&iter
->list
, &stack
);
321 /* look for the last stack element of same type as @iter */
322 list_for_each_entry_reverse(top
, &stack
, list
)
323 if (top
->type
== iter
->type
)
326 list_move_tail(&iter
->list
, &stack
);
330 top_end
= top
->start
+ top
->length
- 1;
332 if (iter
->start
> top_end
+ 1) {
333 list_move_tail(&iter
->list
, &stack
);
335 top
->length
= max(top_end
, iter_end
) - top
->start
+ 1;
336 list_del(&iter
->list
);
340 list_splice(&stack
, regions
);
345 iommu_insert_device_resv_regions(struct list_head
*dev_resv_regions
,
346 struct list_head
*group_resv_regions
)
348 struct iommu_resv_region
*entry
;
351 list_for_each_entry(entry
, dev_resv_regions
, list
) {
352 ret
= iommu_insert_resv_region(entry
, group_resv_regions
);
359 int iommu_get_group_resv_regions(struct iommu_group
*group
,
360 struct list_head
*head
)
362 struct group_device
*device
;
365 mutex_lock(&group
->mutex
);
366 list_for_each_entry(device
, &group
->devices
, list
) {
367 struct list_head dev_resv_regions
;
369 INIT_LIST_HEAD(&dev_resv_regions
);
370 iommu_get_resv_regions(device
->dev
, &dev_resv_regions
);
371 ret
= iommu_insert_device_resv_regions(&dev_resv_regions
, head
);
372 iommu_put_resv_regions(device
->dev
, &dev_resv_regions
);
376 mutex_unlock(&group
->mutex
);
379 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions
);
381 static ssize_t
iommu_group_show_resv_regions(struct iommu_group
*group
,
384 struct iommu_resv_region
*region
, *next
;
385 struct list_head group_resv_regions
;
388 INIT_LIST_HEAD(&group_resv_regions
);
389 iommu_get_group_resv_regions(group
, &group_resv_regions
);
391 list_for_each_entry_safe(region
, next
, &group_resv_regions
, list
) {
392 str
+= sprintf(str
, "0x%016llx 0x%016llx %s\n",
393 (long long int)region
->start
,
394 (long long int)(region
->start
+
396 iommu_group_resv_type_string
[region
->type
]);
403 static ssize_t
iommu_group_show_type(struct iommu_group
*group
,
406 char *type
= "unknown\n";
408 if (group
->default_domain
) {
409 switch (group
->default_domain
->type
) {
410 case IOMMU_DOMAIN_BLOCKED
:
413 case IOMMU_DOMAIN_IDENTITY
:
416 case IOMMU_DOMAIN_UNMANAGED
:
417 type
= "unmanaged\n";
419 case IOMMU_DOMAIN_DMA
:
429 static IOMMU_GROUP_ATTR(name
, S_IRUGO
, iommu_group_show_name
, NULL
);
431 static IOMMU_GROUP_ATTR(reserved_regions
, 0444,
432 iommu_group_show_resv_regions
, NULL
);
434 static IOMMU_GROUP_ATTR(type
, 0444, iommu_group_show_type
, NULL
);
436 static void iommu_group_release(struct kobject
*kobj
)
438 struct iommu_group
*group
= to_iommu_group(kobj
);
440 pr_debug("Releasing group %d\n", group
->id
);
442 if (group
->iommu_data_release
)
443 group
->iommu_data_release(group
->iommu_data
);
445 ida_simple_remove(&iommu_group_ida
, group
->id
);
447 if (group
->default_domain
)
448 iommu_domain_free(group
->default_domain
);
454 static struct kobj_type iommu_group_ktype
= {
455 .sysfs_ops
= &iommu_group_sysfs_ops
,
456 .release
= iommu_group_release
,
460 * iommu_group_alloc - Allocate a new group
462 * This function is called by an iommu driver to allocate a new iommu
463 * group. The iommu group represents the minimum granularity of the iommu.
464 * Upon successful return, the caller holds a reference to the supplied
465 * group in order to hold the group until devices are added. Use
466 * iommu_group_put() to release this extra reference count, allowing the
467 * group to be automatically reclaimed once it has no devices or external
470 struct iommu_group
*iommu_group_alloc(void)
472 struct iommu_group
*group
;
475 group
= kzalloc(sizeof(*group
), GFP_KERNEL
);
477 return ERR_PTR(-ENOMEM
);
479 group
->kobj
.kset
= iommu_group_kset
;
480 mutex_init(&group
->mutex
);
481 INIT_LIST_HEAD(&group
->devices
);
482 BLOCKING_INIT_NOTIFIER_HEAD(&group
->notifier
);
484 ret
= ida_simple_get(&iommu_group_ida
, 0, 0, GFP_KERNEL
);
491 ret
= kobject_init_and_add(&group
->kobj
, &iommu_group_ktype
,
492 NULL
, "%d", group
->id
);
494 ida_simple_remove(&iommu_group_ida
, group
->id
);
495 kobject_put(&group
->kobj
);
499 group
->devices_kobj
= kobject_create_and_add("devices", &group
->kobj
);
500 if (!group
->devices_kobj
) {
501 kobject_put(&group
->kobj
); /* triggers .release & free */
502 return ERR_PTR(-ENOMEM
);
506 * The devices_kobj holds a reference on the group kobject, so
507 * as long as that exists so will the group. We can therefore
508 * use the devices_kobj for reference counting.
510 kobject_put(&group
->kobj
);
512 ret
= iommu_group_create_file(group
,
513 &iommu_group_attr_reserved_regions
);
517 ret
= iommu_group_create_file(group
, &iommu_group_attr_type
);
521 pr_debug("Allocated group %d\n", group
->id
);
525 EXPORT_SYMBOL_GPL(iommu_group_alloc
);
527 struct iommu_group
*iommu_group_get_by_id(int id
)
529 struct kobject
*group_kobj
;
530 struct iommu_group
*group
;
533 if (!iommu_group_kset
)
536 name
= kasprintf(GFP_KERNEL
, "%d", id
);
540 group_kobj
= kset_find_obj(iommu_group_kset
, name
);
546 group
= container_of(group_kobj
, struct iommu_group
, kobj
);
547 BUG_ON(group
->id
!= id
);
549 kobject_get(group
->devices_kobj
);
550 kobject_put(&group
->kobj
);
554 EXPORT_SYMBOL_GPL(iommu_group_get_by_id
);
557 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
560 * iommu drivers can store data in the group for use when doing iommu
561 * operations. This function provides a way to retrieve it. Caller
562 * should hold a group reference.
564 void *iommu_group_get_iommudata(struct iommu_group
*group
)
566 return group
->iommu_data
;
568 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata
);
571 * iommu_group_set_iommudata - set iommu_data for a group
573 * @iommu_data: new data
574 * @release: release function for iommu_data
576 * iommu drivers can store data in the group for use when doing iommu
577 * operations. This function provides a way to set the data after
578 * the group has been allocated. Caller should hold a group reference.
580 void iommu_group_set_iommudata(struct iommu_group
*group
, void *iommu_data
,
581 void (*release
)(void *iommu_data
))
583 group
->iommu_data
= iommu_data
;
584 group
->iommu_data_release
= release
;
586 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata
);
589 * iommu_group_set_name - set name for a group
593 * Allow iommu driver to set a name for a group. When set it will
594 * appear in a name attribute file under the group in sysfs.
596 int iommu_group_set_name(struct iommu_group
*group
, const char *name
)
601 iommu_group_remove_file(group
, &iommu_group_attr_name
);
608 group
->name
= kstrdup(name
, GFP_KERNEL
);
612 ret
= iommu_group_create_file(group
, &iommu_group_attr_name
);
621 EXPORT_SYMBOL_GPL(iommu_group_set_name
);
623 static int iommu_group_create_direct_mappings(struct iommu_group
*group
,
626 struct iommu_domain
*domain
= group
->default_domain
;
627 struct iommu_resv_region
*entry
;
628 struct list_head mappings
;
629 unsigned long pg_size
;
632 if (!domain
|| domain
->type
!= IOMMU_DOMAIN_DMA
)
635 BUG_ON(!domain
->pgsize_bitmap
);
637 pg_size
= 1UL << __ffs(domain
->pgsize_bitmap
);
638 INIT_LIST_HEAD(&mappings
);
640 iommu_get_resv_regions(dev
, &mappings
);
642 /* We need to consider overlapping regions for different devices */
643 list_for_each_entry(entry
, &mappings
, list
) {
644 dma_addr_t start
, end
, addr
;
646 if (domain
->ops
->apply_resv_region
)
647 domain
->ops
->apply_resv_region(dev
, domain
, entry
);
649 start
= ALIGN(entry
->start
, pg_size
);
650 end
= ALIGN(entry
->start
+ entry
->length
, pg_size
);
652 if (entry
->type
!= IOMMU_RESV_DIRECT
&&
653 entry
->type
!= IOMMU_RESV_DIRECT_RELAXABLE
)
656 for (addr
= start
; addr
< end
; addr
+= pg_size
) {
657 phys_addr_t phys_addr
;
659 phys_addr
= iommu_iova_to_phys(domain
, addr
);
663 ret
= iommu_map(domain
, addr
, addr
, pg_size
, entry
->prot
);
670 iommu_flush_tlb_all(domain
);
673 iommu_put_resv_regions(dev
, &mappings
);
679 * iommu_group_add_device - add a device to an iommu group
680 * @group: the group into which to add the device (reference should be held)
683 * This function is called by an iommu driver to add a device into a
684 * group. Adding a device increments the group reference count.
686 int iommu_group_add_device(struct iommu_group
*group
, struct device
*dev
)
689 struct group_device
*device
;
691 device
= kzalloc(sizeof(*device
), GFP_KERNEL
);
697 ret
= sysfs_create_link(&dev
->kobj
, &group
->kobj
, "iommu_group");
699 goto err_free_device
;
701 device
->name
= kasprintf(GFP_KERNEL
, "%s", kobject_name(&dev
->kobj
));
705 goto err_remove_link
;
708 ret
= sysfs_create_link_nowarn(group
->devices_kobj
,
709 &dev
->kobj
, device
->name
);
711 if (ret
== -EEXIST
&& i
>= 0) {
713 * Account for the slim chance of collision
714 * and append an instance to the name.
717 device
->name
= kasprintf(GFP_KERNEL
, "%s.%d",
718 kobject_name(&dev
->kobj
), i
++);
724 kobject_get(group
->devices_kobj
);
726 dev
->iommu_group
= group
;
728 iommu_group_create_direct_mappings(group
, dev
);
730 mutex_lock(&group
->mutex
);
731 list_add_tail(&device
->list
, &group
->devices
);
733 ret
= __iommu_attach_device(group
->domain
, dev
);
734 mutex_unlock(&group
->mutex
);
738 /* Notify any listeners about change to group. */
739 blocking_notifier_call_chain(&group
->notifier
,
740 IOMMU_GROUP_NOTIFY_ADD_DEVICE
, dev
);
742 trace_add_device_to_group(group
->id
, dev
);
744 dev_info(dev
, "Adding to iommu group %d\n", group
->id
);
749 mutex_lock(&group
->mutex
);
750 list_del(&device
->list
);
751 mutex_unlock(&group
->mutex
);
752 dev
->iommu_group
= NULL
;
753 kobject_put(group
->devices_kobj
);
754 sysfs_remove_link(group
->devices_kobj
, device
->name
);
758 sysfs_remove_link(&dev
->kobj
, "iommu_group");
761 dev_err(dev
, "Failed to add to iommu group %d: %d\n", group
->id
, ret
);
764 EXPORT_SYMBOL_GPL(iommu_group_add_device
);
767 * iommu_group_remove_device - remove a device from it's current group
768 * @dev: device to be removed
770 * This function is called by an iommu driver to remove the device from
771 * it's current group. This decrements the iommu group reference count.
773 void iommu_group_remove_device(struct device
*dev
)
775 struct iommu_group
*group
= dev
->iommu_group
;
776 struct group_device
*tmp_device
, *device
= NULL
;
778 dev_info(dev
, "Removing from iommu group %d\n", group
->id
);
780 /* Pre-notify listeners that a device is being removed. */
781 blocking_notifier_call_chain(&group
->notifier
,
782 IOMMU_GROUP_NOTIFY_DEL_DEVICE
, dev
);
784 mutex_lock(&group
->mutex
);
785 list_for_each_entry(tmp_device
, &group
->devices
, list
) {
786 if (tmp_device
->dev
== dev
) {
788 list_del(&device
->list
);
792 mutex_unlock(&group
->mutex
);
797 sysfs_remove_link(group
->devices_kobj
, device
->name
);
798 sysfs_remove_link(&dev
->kobj
, "iommu_group");
800 trace_remove_device_from_group(group
->id
, dev
);
804 dev
->iommu_group
= NULL
;
805 kobject_put(group
->devices_kobj
);
807 EXPORT_SYMBOL_GPL(iommu_group_remove_device
);
809 static int iommu_group_device_count(struct iommu_group
*group
)
811 struct group_device
*entry
;
814 list_for_each_entry(entry
, &group
->devices
, list
)
821 * iommu_group_for_each_dev - iterate over each device in the group
823 * @data: caller opaque data to be passed to callback function
824 * @fn: caller supplied callback function
826 * This function is called by group users to iterate over group devices.
827 * Callers should hold a reference count to the group during callback.
828 * The group->mutex is held across callbacks, which will block calls to
829 * iommu_group_add/remove_device.
831 static int __iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
832 int (*fn
)(struct device
*, void *))
834 struct group_device
*device
;
837 list_for_each_entry(device
, &group
->devices
, list
) {
838 ret
= fn(device
->dev
, data
);
846 int iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
847 int (*fn
)(struct device
*, void *))
851 mutex_lock(&group
->mutex
);
852 ret
= __iommu_group_for_each_dev(group
, data
, fn
);
853 mutex_unlock(&group
->mutex
);
857 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev
);
860 * iommu_group_get - Return the group for a device and increment reference
861 * @dev: get the group that this device belongs to
863 * This function is called by iommu drivers and users to get the group
864 * for the specified device. If found, the group is returned and the group
865 * reference in incremented, else NULL.
867 struct iommu_group
*iommu_group_get(struct device
*dev
)
869 struct iommu_group
*group
= dev
->iommu_group
;
872 kobject_get(group
->devices_kobj
);
876 EXPORT_SYMBOL_GPL(iommu_group_get
);
879 * iommu_group_ref_get - Increment reference on a group
880 * @group: the group to use, must not be NULL
882 * This function is called by iommu drivers to take additional references on an
883 * existing group. Returns the given group for convenience.
885 struct iommu_group
*iommu_group_ref_get(struct iommu_group
*group
)
887 kobject_get(group
->devices_kobj
);
892 * iommu_group_put - Decrement group reference
893 * @group: the group to use
895 * This function is called by iommu drivers and users to release the
896 * iommu group. Once the reference count is zero, the group is released.
898 void iommu_group_put(struct iommu_group
*group
)
901 kobject_put(group
->devices_kobj
);
903 EXPORT_SYMBOL_GPL(iommu_group_put
);
906 * iommu_group_register_notifier - Register a notifier for group changes
907 * @group: the group to watch
908 * @nb: notifier block to signal
910 * This function allows iommu group users to track changes in a group.
911 * See include/linux/iommu.h for actions sent via this notifier. Caller
912 * should hold a reference to the group throughout notifier registration.
914 int iommu_group_register_notifier(struct iommu_group
*group
,
915 struct notifier_block
*nb
)
917 return blocking_notifier_chain_register(&group
->notifier
, nb
);
919 EXPORT_SYMBOL_GPL(iommu_group_register_notifier
);
922 * iommu_group_unregister_notifier - Unregister a notifier
923 * @group: the group to watch
924 * @nb: notifier block to signal
926 * Unregister a previously registered group notifier block.
928 int iommu_group_unregister_notifier(struct iommu_group
*group
,
929 struct notifier_block
*nb
)
931 return blocking_notifier_chain_unregister(&group
->notifier
, nb
);
933 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier
);
936 * iommu_register_device_fault_handler() - Register a device fault handler
938 * @handler: the fault handler
939 * @data: private data passed as argument to the handler
941 * When an IOMMU fault event is received, this handler gets called with the
942 * fault event and data as argument. The handler should return 0 on success. If
943 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
944 * complete the fault by calling iommu_page_response() with one of the following
946 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
947 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
948 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
949 * page faults if possible.
951 * Return 0 if the fault handler was installed successfully, or an error.
953 int iommu_register_device_fault_handler(struct device
*dev
,
954 iommu_dev_fault_handler_t handler
,
957 struct iommu_param
*param
= dev
->iommu_param
;
963 mutex_lock(¶m
->lock
);
964 /* Only allow one fault handler registered for each device */
965 if (param
->fault_param
) {
971 param
->fault_param
= kzalloc(sizeof(*param
->fault_param
), GFP_KERNEL
);
972 if (!param
->fault_param
) {
977 param
->fault_param
->handler
= handler
;
978 param
->fault_param
->data
= data
;
979 mutex_init(¶m
->fault_param
->lock
);
980 INIT_LIST_HEAD(¶m
->fault_param
->faults
);
983 mutex_unlock(¶m
->lock
);
987 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler
);
990 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
993 * Remove the device fault handler installed with
994 * iommu_register_device_fault_handler().
996 * Return 0 on success, or an error.
998 int iommu_unregister_device_fault_handler(struct device
*dev
)
1000 struct iommu_param
*param
= dev
->iommu_param
;
1006 mutex_lock(¶m
->lock
);
1008 if (!param
->fault_param
)
1011 /* we cannot unregister handler if there are pending faults */
1012 if (!list_empty(¶m
->fault_param
->faults
)) {
1017 kfree(param
->fault_param
);
1018 param
->fault_param
= NULL
;
1021 mutex_unlock(¶m
->lock
);
1025 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler
);
1028 * iommu_report_device_fault() - Report fault event to device driver
1030 * @evt: fault event data
1032 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1033 * handler. When this function fails and the fault is recoverable, it is the
1034 * caller's responsibility to complete the fault.
1036 * Return 0 on success, or an error.
1038 int iommu_report_device_fault(struct device
*dev
, struct iommu_fault_event
*evt
)
1040 struct iommu_param
*param
= dev
->iommu_param
;
1041 struct iommu_fault_event
*evt_pending
= NULL
;
1042 struct iommu_fault_param
*fparam
;
1048 /* we only report device fault if there is a handler registered */
1049 mutex_lock(¶m
->lock
);
1050 fparam
= param
->fault_param
;
1051 if (!fparam
|| !fparam
->handler
) {
1056 if (evt
->fault
.type
== IOMMU_FAULT_PAGE_REQ
&&
1057 (evt
->fault
.prm
.flags
& IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE
)) {
1058 evt_pending
= kmemdup(evt
, sizeof(struct iommu_fault_event
),
1064 mutex_lock(&fparam
->lock
);
1065 list_add_tail(&evt_pending
->list
, &fparam
->faults
);
1066 mutex_unlock(&fparam
->lock
);
1069 ret
= fparam
->handler(&evt
->fault
, fparam
->data
);
1070 if (ret
&& evt_pending
) {
1071 mutex_lock(&fparam
->lock
);
1072 list_del(&evt_pending
->list
);
1073 mutex_unlock(&fparam
->lock
);
1077 mutex_unlock(¶m
->lock
);
1080 EXPORT_SYMBOL_GPL(iommu_report_device_fault
);
1082 int iommu_page_response(struct device
*dev
,
1083 struct iommu_page_response
*msg
)
1087 struct iommu_fault_event
*evt
;
1088 struct iommu_fault_page_request
*prm
;
1089 struct iommu_param
*param
= dev
->iommu_param
;
1090 struct iommu_domain
*domain
= iommu_get_domain_for_dev(dev
);
1092 if (!domain
|| !domain
->ops
->page_response
)
1095 if (!param
|| !param
->fault_param
)
1098 if (msg
->version
!= IOMMU_PAGE_RESP_VERSION_1
||
1099 msg
->flags
& ~IOMMU_PAGE_RESP_PASID_VALID
)
1102 /* Only send response if there is a fault report pending */
1103 mutex_lock(¶m
->fault_param
->lock
);
1104 if (list_empty(¶m
->fault_param
->faults
)) {
1105 dev_warn_ratelimited(dev
, "no pending PRQ, drop response\n");
1109 * Check if we have a matching page request pending to respond,
1110 * otherwise return -EINVAL
1112 list_for_each_entry(evt
, ¶m
->fault_param
->faults
, list
) {
1113 prm
= &evt
->fault
.prm
;
1114 pasid_valid
= prm
->flags
& IOMMU_FAULT_PAGE_REQUEST_PASID_VALID
;
1116 if ((pasid_valid
&& prm
->pasid
!= msg
->pasid
) ||
1117 prm
->grpid
!= msg
->grpid
)
1120 /* Sanitize the reply */
1121 msg
->flags
= pasid_valid
? IOMMU_PAGE_RESP_PASID_VALID
: 0;
1123 ret
= domain
->ops
->page_response(dev
, evt
, msg
);
1124 list_del(&evt
->list
);
1130 mutex_unlock(¶m
->fault_param
->lock
);
1133 EXPORT_SYMBOL_GPL(iommu_page_response
);
1136 * iommu_group_id - Return ID for a group
1137 * @group: the group to ID
1139 * Return the unique ID for the group matching the sysfs group number.
1141 int iommu_group_id(struct iommu_group
*group
)
1145 EXPORT_SYMBOL_GPL(iommu_group_id
);
1147 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1148 unsigned long *devfns
);
1151 * To consider a PCI device isolated, we require ACS to support Source
1152 * Validation, Request Redirection, Completer Redirection, and Upstream
1153 * Forwarding. This effectively means that devices cannot spoof their
1154 * requester ID, requests and completions cannot be redirected, and all
1155 * transactions are forwarded upstream, even as it passes through a
1156 * bridge where the target device is downstream.
1158 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1161 * For multifunction devices which are not isolated from each other, find
1162 * all the other non-isolated functions and look for existing groups. For
1163 * each function, we also need to look for aliases to or from other devices
1164 * that may already have a group.
1166 static struct iommu_group
*get_pci_function_alias_group(struct pci_dev
*pdev
,
1167 unsigned long *devfns
)
1169 struct pci_dev
*tmp
= NULL
;
1170 struct iommu_group
*group
;
1172 if (!pdev
->multifunction
|| pci_acs_enabled(pdev
, REQ_ACS_FLAGS
))
1175 for_each_pci_dev(tmp
) {
1176 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
||
1177 PCI_SLOT(tmp
->devfn
) != PCI_SLOT(pdev
->devfn
) ||
1178 pci_acs_enabled(tmp
, REQ_ACS_FLAGS
))
1181 group
= get_pci_alias_group(tmp
, devfns
);
1192 * Look for aliases to or from the given device for existing groups. DMA
1193 * aliases are only supported on the same bus, therefore the search
1194 * space is quite small (especially since we're really only looking at pcie
1195 * device, and therefore only expect multiple slots on the root complex or
1196 * downstream switch ports). It's conceivable though that a pair of
1197 * multifunction devices could have aliases between them that would cause a
1198 * loop. To prevent this, we use a bitmap to track where we've been.
1200 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1201 unsigned long *devfns
)
1203 struct pci_dev
*tmp
= NULL
;
1204 struct iommu_group
*group
;
1206 if (test_and_set_bit(pdev
->devfn
& 0xff, devfns
))
1209 group
= iommu_group_get(&pdev
->dev
);
1213 for_each_pci_dev(tmp
) {
1214 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
)
1217 /* We alias them or they alias us */
1218 if (pci_devs_are_dma_aliases(pdev
, tmp
)) {
1219 group
= get_pci_alias_group(tmp
, devfns
);
1225 group
= get_pci_function_alias_group(tmp
, devfns
);
1236 struct group_for_pci_data
{
1237 struct pci_dev
*pdev
;
1238 struct iommu_group
*group
;
1242 * DMA alias iterator callback, return the last seen device. Stop and return
1243 * the IOMMU group if we find one along the way.
1245 static int get_pci_alias_or_group(struct pci_dev
*pdev
, u16 alias
, void *opaque
)
1247 struct group_for_pci_data
*data
= opaque
;
1250 data
->group
= iommu_group_get(&pdev
->dev
);
1252 return data
->group
!= NULL
;
1256 * Generic device_group call-back function. It just allocates one
1257 * iommu-group per device.
1259 struct iommu_group
*generic_device_group(struct device
*dev
)
1261 return iommu_group_alloc();
1265 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1266 * to find or create an IOMMU group for a device.
1268 struct iommu_group
*pci_device_group(struct device
*dev
)
1270 struct pci_dev
*pdev
= to_pci_dev(dev
);
1271 struct group_for_pci_data data
;
1272 struct pci_bus
*bus
;
1273 struct iommu_group
*group
= NULL
;
1274 u64 devfns
[4] = { 0 };
1276 if (WARN_ON(!dev_is_pci(dev
)))
1277 return ERR_PTR(-EINVAL
);
1280 * Find the upstream DMA alias for the device. A device must not
1281 * be aliased due to topology in order to have its own IOMMU group.
1282 * If we find an alias along the way that already belongs to a
1285 if (pci_for_each_dma_alias(pdev
, get_pci_alias_or_group
, &data
))
1291 * Continue upstream from the point of minimum IOMMU granularity
1292 * due to aliases to the point where devices are protected from
1293 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1296 for (bus
= pdev
->bus
; !pci_is_root_bus(bus
); bus
= bus
->parent
) {
1300 if (pci_acs_path_enabled(bus
->self
, NULL
, REQ_ACS_FLAGS
))
1305 group
= iommu_group_get(&pdev
->dev
);
1311 * Look for existing groups on device aliases. If we alias another
1312 * device or another device aliases us, use the same group.
1314 group
= get_pci_alias_group(pdev
, (unsigned long *)devfns
);
1319 * Look for existing groups on non-isolated functions on the same
1320 * slot and aliases of those funcions, if any. No need to clear
1321 * the search bitmap, the tested devfns are still valid.
1323 group
= get_pci_function_alias_group(pdev
, (unsigned long *)devfns
);
1327 /* No shared group found, allocate new */
1328 return iommu_group_alloc();
1331 /* Get the IOMMU group for device on fsl-mc bus */
1332 struct iommu_group
*fsl_mc_device_group(struct device
*dev
)
1334 struct device
*cont_dev
= fsl_mc_cont_dev(dev
);
1335 struct iommu_group
*group
;
1337 group
= iommu_group_get(cont_dev
);
1339 group
= iommu_group_alloc();
1344 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1345 * @dev: target device
1347 * This function is intended to be called by IOMMU drivers and extended to
1348 * support common, bus-defined algorithms when determining or creating the
1349 * IOMMU group for a device. On success, the caller will hold a reference
1350 * to the returned IOMMU group, which will already include the provided
1351 * device. The reference should be released with iommu_group_put().
1353 struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
)
1355 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1356 struct iommu_group
*group
;
1359 group
= iommu_group_get(dev
);
1364 return ERR_PTR(-EINVAL
);
1366 group
= ops
->device_group(dev
);
1367 if (WARN_ON_ONCE(group
== NULL
))
1368 return ERR_PTR(-EINVAL
);
1374 * Try to allocate a default domain - needs support from the
1377 if (!group
->default_domain
) {
1378 struct iommu_domain
*dom
;
1380 dom
= __iommu_domain_alloc(dev
->bus
, iommu_def_domain_type
);
1381 if (!dom
&& iommu_def_domain_type
!= IOMMU_DOMAIN_DMA
) {
1382 dom
= __iommu_domain_alloc(dev
->bus
, IOMMU_DOMAIN_DMA
);
1385 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1386 iommu_def_domain_type
);
1390 group
->default_domain
= dom
;
1392 group
->domain
= dom
;
1394 if (dom
&& !iommu_dma_strict
) {
1396 iommu_domain_set_attr(dom
,
1397 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE
,
1402 ret
= iommu_group_add_device(group
, dev
);
1404 iommu_group_put(group
);
1405 return ERR_PTR(ret
);
1411 struct iommu_domain
*iommu_group_default_domain(struct iommu_group
*group
)
1413 return group
->default_domain
;
1416 static int add_iommu_group(struct device
*dev
, void *data
)
1418 int ret
= iommu_probe_device(dev
);
1421 * We ignore -ENODEV errors for now, as they just mean that the
1422 * device is not translated by an IOMMU. We still care about
1423 * other errors and fail to initialize when they happen.
1431 static int remove_iommu_group(struct device
*dev
, void *data
)
1433 iommu_release_device(dev
);
1438 static int iommu_bus_notifier(struct notifier_block
*nb
,
1439 unsigned long action
, void *data
)
1441 unsigned long group_action
= 0;
1442 struct device
*dev
= data
;
1443 struct iommu_group
*group
;
1446 * ADD/DEL call into iommu driver ops if provided, which may
1447 * result in ADD/DEL notifiers to group->notifier
1449 if (action
== BUS_NOTIFY_ADD_DEVICE
) {
1452 ret
= iommu_probe_device(dev
);
1453 return (ret
) ? NOTIFY_DONE
: NOTIFY_OK
;
1454 } else if (action
== BUS_NOTIFY_REMOVED_DEVICE
) {
1455 iommu_release_device(dev
);
1460 * Remaining BUS_NOTIFYs get filtered and republished to the
1461 * group, if anyone is listening
1463 group
= iommu_group_get(dev
);
1468 case BUS_NOTIFY_BIND_DRIVER
:
1469 group_action
= IOMMU_GROUP_NOTIFY_BIND_DRIVER
;
1471 case BUS_NOTIFY_BOUND_DRIVER
:
1472 group_action
= IOMMU_GROUP_NOTIFY_BOUND_DRIVER
;
1474 case BUS_NOTIFY_UNBIND_DRIVER
:
1475 group_action
= IOMMU_GROUP_NOTIFY_UNBIND_DRIVER
;
1477 case BUS_NOTIFY_UNBOUND_DRIVER
:
1478 group_action
= IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER
;
1483 blocking_notifier_call_chain(&group
->notifier
,
1486 iommu_group_put(group
);
1490 static int iommu_bus_init(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1493 struct notifier_block
*nb
;
1495 nb
= kzalloc(sizeof(struct notifier_block
), GFP_KERNEL
);
1499 nb
->notifier_call
= iommu_bus_notifier
;
1501 err
= bus_register_notifier(bus
, nb
);
1505 err
= bus_for_each_dev(bus
, NULL
, NULL
, add_iommu_group
);
1514 bus_for_each_dev(bus
, NULL
, NULL
, remove_iommu_group
);
1515 bus_unregister_notifier(bus
, nb
);
1524 * bus_set_iommu - set iommu-callbacks for the bus
1526 * @ops: the callbacks provided by the iommu-driver
1528 * This function is called by an iommu driver to set the iommu methods
1529 * used for a particular bus. Drivers for devices on that bus can use
1530 * the iommu-api after these ops are registered.
1531 * This special function is needed because IOMMUs are usually devices on
1532 * the bus itself, so the iommu drivers are not initialized when the bus
1533 * is set up. With this function the iommu-driver can set the iommu-ops
1536 int bus_set_iommu(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1540 if (bus
->iommu_ops
!= NULL
)
1543 bus
->iommu_ops
= ops
;
1545 /* Do IOMMU specific setup for this bus-type */
1546 err
= iommu_bus_init(bus
, ops
);
1548 bus
->iommu_ops
= NULL
;
1552 EXPORT_SYMBOL_GPL(bus_set_iommu
);
1554 bool iommu_present(struct bus_type
*bus
)
1556 return bus
->iommu_ops
!= NULL
;
1558 EXPORT_SYMBOL_GPL(iommu_present
);
1560 bool iommu_capable(struct bus_type
*bus
, enum iommu_cap cap
)
1562 if (!bus
->iommu_ops
|| !bus
->iommu_ops
->capable
)
1565 return bus
->iommu_ops
->capable(cap
);
1567 EXPORT_SYMBOL_GPL(iommu_capable
);
1570 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1571 * @domain: iommu domain
1572 * @handler: fault handler
1573 * @token: user data, will be passed back to the fault handler
1575 * This function should be used by IOMMU users which want to be notified
1576 * whenever an IOMMU fault happens.
1578 * The fault handler itself should return 0 on success, and an appropriate
1579 * error code otherwise.
1581 void iommu_set_fault_handler(struct iommu_domain
*domain
,
1582 iommu_fault_handler_t handler
,
1587 domain
->handler
= handler
;
1588 domain
->handler_token
= token
;
1590 EXPORT_SYMBOL_GPL(iommu_set_fault_handler
);
1592 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
1595 struct iommu_domain
*domain
;
1597 if (bus
== NULL
|| bus
->iommu_ops
== NULL
)
1600 domain
= bus
->iommu_ops
->domain_alloc(type
);
1604 domain
->ops
= bus
->iommu_ops
;
1605 domain
->type
= type
;
1606 /* Assume all sizes by default; the driver may override this later */
1607 domain
->pgsize_bitmap
= bus
->iommu_ops
->pgsize_bitmap
;
1612 struct iommu_domain
*iommu_domain_alloc(struct bus_type
*bus
)
1614 return __iommu_domain_alloc(bus
, IOMMU_DOMAIN_UNMANAGED
);
1616 EXPORT_SYMBOL_GPL(iommu_domain_alloc
);
1618 void iommu_domain_free(struct iommu_domain
*domain
)
1620 domain
->ops
->domain_free(domain
);
1622 EXPORT_SYMBOL_GPL(iommu_domain_free
);
1624 static int __iommu_attach_device(struct iommu_domain
*domain
,
1628 if ((domain
->ops
->is_attach_deferred
!= NULL
) &&
1629 domain
->ops
->is_attach_deferred(domain
, dev
))
1632 if (unlikely(domain
->ops
->attach_dev
== NULL
))
1635 ret
= domain
->ops
->attach_dev(domain
, dev
);
1637 trace_attach_device_to_domain(dev
);
1641 int iommu_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
1643 struct iommu_group
*group
;
1646 group
= iommu_group_get(dev
);
1651 * Lock the group to make sure the device-count doesn't
1652 * change while we are attaching
1654 mutex_lock(&group
->mutex
);
1656 if (iommu_group_device_count(group
) != 1)
1659 ret
= __iommu_attach_group(domain
, group
);
1662 mutex_unlock(&group
->mutex
);
1663 iommu_group_put(group
);
1667 EXPORT_SYMBOL_GPL(iommu_attach_device
);
1669 static void __iommu_detach_device(struct iommu_domain
*domain
,
1672 if ((domain
->ops
->is_attach_deferred
!= NULL
) &&
1673 domain
->ops
->is_attach_deferred(domain
, dev
))
1676 if (unlikely(domain
->ops
->detach_dev
== NULL
))
1679 domain
->ops
->detach_dev(domain
, dev
);
1680 trace_detach_device_from_domain(dev
);
1683 void iommu_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
1685 struct iommu_group
*group
;
1687 group
= iommu_group_get(dev
);
1691 mutex_lock(&group
->mutex
);
1692 if (iommu_group_device_count(group
) != 1) {
1697 __iommu_detach_group(domain
, group
);
1700 mutex_unlock(&group
->mutex
);
1701 iommu_group_put(group
);
1703 EXPORT_SYMBOL_GPL(iommu_detach_device
);
1705 struct iommu_domain
*iommu_get_domain_for_dev(struct device
*dev
)
1707 struct iommu_domain
*domain
;
1708 struct iommu_group
*group
;
1710 group
= iommu_group_get(dev
);
1714 domain
= group
->domain
;
1716 iommu_group_put(group
);
1720 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev
);
1723 * For IOMMU_DOMAIN_DMA implementations which already provide their own
1724 * guarantees that the group and its default domain are valid and correct.
1726 struct iommu_domain
*iommu_get_dma_domain(struct device
*dev
)
1728 return dev
->iommu_group
->default_domain
;
1732 * IOMMU groups are really the natural working unit of the IOMMU, but
1733 * the IOMMU API works on domains and devices. Bridge that gap by
1734 * iterating over the devices in a group. Ideally we'd have a single
1735 * device which represents the requestor ID of the group, but we also
1736 * allow IOMMU drivers to create policy defined minimum sets, where
1737 * the physical hardware may be able to distiguish members, but we
1738 * wish to group them at a higher level (ex. untrusted multi-function
1739 * PCI devices). Thus we attach each device.
1741 static int iommu_group_do_attach_device(struct device
*dev
, void *data
)
1743 struct iommu_domain
*domain
= data
;
1745 return __iommu_attach_device(domain
, dev
);
1748 static int __iommu_attach_group(struct iommu_domain
*domain
,
1749 struct iommu_group
*group
)
1753 if (group
->default_domain
&& group
->domain
!= group
->default_domain
)
1756 ret
= __iommu_group_for_each_dev(group
, domain
,
1757 iommu_group_do_attach_device
);
1759 group
->domain
= domain
;
1764 int iommu_attach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1768 mutex_lock(&group
->mutex
);
1769 ret
= __iommu_attach_group(domain
, group
);
1770 mutex_unlock(&group
->mutex
);
1774 EXPORT_SYMBOL_GPL(iommu_attach_group
);
1776 static int iommu_group_do_detach_device(struct device
*dev
, void *data
)
1778 struct iommu_domain
*domain
= data
;
1780 __iommu_detach_device(domain
, dev
);
1785 static void __iommu_detach_group(struct iommu_domain
*domain
,
1786 struct iommu_group
*group
)
1790 if (!group
->default_domain
) {
1791 __iommu_group_for_each_dev(group
, domain
,
1792 iommu_group_do_detach_device
);
1793 group
->domain
= NULL
;
1797 if (group
->domain
== group
->default_domain
)
1800 /* Detach by re-attaching to the default domain */
1801 ret
= __iommu_group_for_each_dev(group
, group
->default_domain
,
1802 iommu_group_do_attach_device
);
1806 group
->domain
= group
->default_domain
;
1809 void iommu_detach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1811 mutex_lock(&group
->mutex
);
1812 __iommu_detach_group(domain
, group
);
1813 mutex_unlock(&group
->mutex
);
1815 EXPORT_SYMBOL_GPL(iommu_detach_group
);
1817 phys_addr_t
iommu_iova_to_phys(struct iommu_domain
*domain
, dma_addr_t iova
)
1819 if (unlikely(domain
->ops
->iova_to_phys
== NULL
))
1822 return domain
->ops
->iova_to_phys(domain
, iova
);
1824 EXPORT_SYMBOL_GPL(iommu_iova_to_phys
);
1826 static size_t iommu_pgsize(struct iommu_domain
*domain
,
1827 unsigned long addr_merge
, size_t size
)
1829 unsigned int pgsize_idx
;
1832 /* Max page size that still fits into 'size' */
1833 pgsize_idx
= __fls(size
);
1835 /* need to consider alignment requirements ? */
1836 if (likely(addr_merge
)) {
1837 /* Max page size allowed by address */
1838 unsigned int align_pgsize_idx
= __ffs(addr_merge
);
1839 pgsize_idx
= min(pgsize_idx
, align_pgsize_idx
);
1842 /* build a mask of acceptable page sizes */
1843 pgsize
= (1UL << (pgsize_idx
+ 1)) - 1;
1845 /* throw away page sizes not supported by the hardware */
1846 pgsize
&= domain
->pgsize_bitmap
;
1848 /* make sure we're still sane */
1851 /* pick the biggest page */
1852 pgsize_idx
= __fls(pgsize
);
1853 pgsize
= 1UL << pgsize_idx
;
1858 int iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
1859 phys_addr_t paddr
, size_t size
, int prot
)
1861 const struct iommu_ops
*ops
= domain
->ops
;
1862 unsigned long orig_iova
= iova
;
1863 unsigned int min_pagesz
;
1864 size_t orig_size
= size
;
1865 phys_addr_t orig_paddr
= paddr
;
1868 if (unlikely(ops
->map
== NULL
||
1869 domain
->pgsize_bitmap
== 0UL))
1872 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1875 /* find out the minimum page size supported */
1876 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1879 * both the virtual address and the physical one, as well as
1880 * the size of the mapping, must be aligned (at least) to the
1881 * size of the smallest page supported by the hardware
1883 if (!IS_ALIGNED(iova
| paddr
| size
, min_pagesz
)) {
1884 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1885 iova
, &paddr
, size
, min_pagesz
);
1889 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova
, &paddr
, size
);
1892 size_t pgsize
= iommu_pgsize(domain
, iova
| paddr
, size
);
1894 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1895 iova
, &paddr
, pgsize
);
1897 ret
= ops
->map(domain
, iova
, paddr
, pgsize
, prot
);
1906 if (ops
->iotlb_sync_map
)
1907 ops
->iotlb_sync_map(domain
);
1909 /* unroll mapping in case something went wrong */
1911 iommu_unmap(domain
, orig_iova
, orig_size
- size
);
1913 trace_map(orig_iova
, orig_paddr
, orig_size
);
1917 EXPORT_SYMBOL_GPL(iommu_map
);
1919 static size_t __iommu_unmap(struct iommu_domain
*domain
,
1920 unsigned long iova
, size_t size
,
1921 struct iommu_iotlb_gather
*iotlb_gather
)
1923 const struct iommu_ops
*ops
= domain
->ops
;
1924 size_t unmapped_page
, unmapped
= 0;
1925 unsigned long orig_iova
= iova
;
1926 unsigned int min_pagesz
;
1928 if (unlikely(ops
->unmap
== NULL
||
1929 domain
->pgsize_bitmap
== 0UL))
1932 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1935 /* find out the minimum page size supported */
1936 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1939 * The virtual address, as well as the size of the mapping, must be
1940 * aligned (at least) to the size of the smallest page supported
1943 if (!IS_ALIGNED(iova
| size
, min_pagesz
)) {
1944 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1945 iova
, size
, min_pagesz
);
1949 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova
, size
);
1952 * Keep iterating until we either unmap 'size' bytes (or more)
1953 * or we hit an area that isn't mapped.
1955 while (unmapped
< size
) {
1956 size_t pgsize
= iommu_pgsize(domain
, iova
, size
- unmapped
);
1958 unmapped_page
= ops
->unmap(domain
, iova
, pgsize
, iotlb_gather
);
1962 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1963 iova
, unmapped_page
);
1965 iova
+= unmapped_page
;
1966 unmapped
+= unmapped_page
;
1969 trace_unmap(orig_iova
, size
, unmapped
);
1973 size_t iommu_unmap(struct iommu_domain
*domain
,
1974 unsigned long iova
, size_t size
)
1976 struct iommu_iotlb_gather iotlb_gather
;
1979 iommu_iotlb_gather_init(&iotlb_gather
);
1980 ret
= __iommu_unmap(domain
, iova
, size
, &iotlb_gather
);
1981 iommu_tlb_sync(domain
, &iotlb_gather
);
1985 EXPORT_SYMBOL_GPL(iommu_unmap
);
1987 size_t iommu_unmap_fast(struct iommu_domain
*domain
,
1988 unsigned long iova
, size_t size
,
1989 struct iommu_iotlb_gather
*iotlb_gather
)
1991 return __iommu_unmap(domain
, iova
, size
, iotlb_gather
);
1993 EXPORT_SYMBOL_GPL(iommu_unmap_fast
);
1995 size_t iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
1996 struct scatterlist
*sg
, unsigned int nents
, int prot
)
1998 size_t len
= 0, mapped
= 0;
2003 while (i
<= nents
) {
2004 phys_addr_t s_phys
= sg_phys(sg
);
2006 if (len
&& s_phys
!= start
+ len
) {
2007 ret
= iommu_map(domain
, iova
+ mapped
, start
, len
, prot
);
2029 /* undo mappings already done */
2030 iommu_unmap(domain
, iova
, mapped
);
2035 EXPORT_SYMBOL_GPL(iommu_map_sg
);
2037 int iommu_domain_window_enable(struct iommu_domain
*domain
, u32 wnd_nr
,
2038 phys_addr_t paddr
, u64 size
, int prot
)
2040 if (unlikely(domain
->ops
->domain_window_enable
== NULL
))
2043 return domain
->ops
->domain_window_enable(domain
, wnd_nr
, paddr
, size
,
2046 EXPORT_SYMBOL_GPL(iommu_domain_window_enable
);
2048 void iommu_domain_window_disable(struct iommu_domain
*domain
, u32 wnd_nr
)
2050 if (unlikely(domain
->ops
->domain_window_disable
== NULL
))
2053 return domain
->ops
->domain_window_disable(domain
, wnd_nr
);
2055 EXPORT_SYMBOL_GPL(iommu_domain_window_disable
);
2058 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2059 * @domain: the iommu domain where the fault has happened
2060 * @dev: the device where the fault has happened
2061 * @iova: the faulting address
2062 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2064 * This function should be called by the low-level IOMMU implementations
2065 * whenever IOMMU faults happen, to allow high-level users, that are
2066 * interested in such events, to know about them.
2068 * This event may be useful for several possible use cases:
2069 * - mere logging of the event
2070 * - dynamic TLB/PTE loading
2071 * - if restarting of the faulting device is required
2073 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2074 * PTE/TLB loading will one day be supported, implementations will be able
2075 * to tell whether it succeeded or not according to this return value).
2077 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2078 * (though fault handlers can also return -ENOSYS, in case they want to
2079 * elicit the default behavior of the IOMMU drivers).
2081 int report_iommu_fault(struct iommu_domain
*domain
, struct device
*dev
,
2082 unsigned long iova
, int flags
)
2087 * if upper layers showed interest and installed a fault handler,
2090 if (domain
->handler
)
2091 ret
= domain
->handler(domain
, dev
, iova
, flags
,
2092 domain
->handler_token
);
2094 trace_io_page_fault(dev
, iova
, flags
);
2097 EXPORT_SYMBOL_GPL(report_iommu_fault
);
2099 static int __init
iommu_init(void)
2101 iommu_group_kset
= kset_create_and_add("iommu_groups",
2103 BUG_ON(!iommu_group_kset
);
2105 iommu_debugfs_setup();
2109 core_initcall(iommu_init
);
2111 int iommu_domain_get_attr(struct iommu_domain
*domain
,
2112 enum iommu_attr attr
, void *data
)
2114 struct iommu_domain_geometry
*geometry
;
2119 case DOMAIN_ATTR_GEOMETRY
:
2121 *geometry
= domain
->geometry
;
2124 case DOMAIN_ATTR_PAGING
:
2126 *paging
= (domain
->pgsize_bitmap
!= 0UL);
2129 if (!domain
->ops
->domain_get_attr
)
2132 ret
= domain
->ops
->domain_get_attr(domain
, attr
, data
);
2137 EXPORT_SYMBOL_GPL(iommu_domain_get_attr
);
2139 int iommu_domain_set_attr(struct iommu_domain
*domain
,
2140 enum iommu_attr attr
, void *data
)
2146 if (domain
->ops
->domain_set_attr
== NULL
)
2149 ret
= domain
->ops
->domain_set_attr(domain
, attr
, data
);
2154 EXPORT_SYMBOL_GPL(iommu_domain_set_attr
);
2156 void iommu_get_resv_regions(struct device
*dev
, struct list_head
*list
)
2158 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2160 if (ops
&& ops
->get_resv_regions
)
2161 ops
->get_resv_regions(dev
, list
);
2164 void iommu_put_resv_regions(struct device
*dev
, struct list_head
*list
)
2166 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2168 if (ops
&& ops
->put_resv_regions
)
2169 ops
->put_resv_regions(dev
, list
);
2172 struct iommu_resv_region
*iommu_alloc_resv_region(phys_addr_t start
,
2173 size_t length
, int prot
,
2174 enum iommu_resv_type type
)
2176 struct iommu_resv_region
*region
;
2178 region
= kzalloc(sizeof(*region
), GFP_KERNEL
);
2182 INIT_LIST_HEAD(®ion
->list
);
2183 region
->start
= start
;
2184 region
->length
= length
;
2185 region
->prot
= prot
;
2186 region
->type
= type
;
2191 request_default_domain_for_dev(struct device
*dev
, unsigned long type
)
2193 struct iommu_domain
*domain
;
2194 struct iommu_group
*group
;
2197 /* Device must already be in a group before calling this function */
2198 group
= iommu_group_get(dev
);
2202 mutex_lock(&group
->mutex
);
2205 if (group
->default_domain
&& group
->default_domain
->type
== type
)
2208 /* Don't change mappings of existing devices */
2210 if (iommu_group_device_count(group
) != 1)
2214 domain
= __iommu_domain_alloc(dev
->bus
, type
);
2218 /* Attach the device to the domain */
2219 ret
= __iommu_attach_group(domain
, group
);
2221 iommu_domain_free(domain
);
2225 /* Make the domain the default for this group */
2226 if (group
->default_domain
)
2227 iommu_domain_free(group
->default_domain
);
2228 group
->default_domain
= domain
;
2230 iommu_group_create_direct_mappings(group
, dev
);
2232 dev_info(dev
, "Using iommu %s mapping\n",
2233 type
== IOMMU_DOMAIN_DMA
? "dma" : "direct");
2237 mutex_unlock(&group
->mutex
);
2238 iommu_group_put(group
);
2243 /* Request that a device is direct mapped by the IOMMU */
2244 int iommu_request_dm_for_dev(struct device
*dev
)
2246 return request_default_domain_for_dev(dev
, IOMMU_DOMAIN_IDENTITY
);
2249 /* Request that a device can't be direct mapped by the IOMMU */
2250 int iommu_request_dma_domain_for_dev(struct device
*dev
)
2252 return request_default_domain_for_dev(dev
, IOMMU_DOMAIN_DMA
);
2255 void iommu_set_default_passthrough(bool cmd_line
)
2258 iommu_set_cmd_line_dma_api();
2260 iommu_def_domain_type
= IOMMU_DOMAIN_IDENTITY
;
2263 void iommu_set_default_translated(bool cmd_line
)
2266 iommu_set_cmd_line_dma_api();
2268 iommu_def_domain_type
= IOMMU_DOMAIN_DMA
;
2271 bool iommu_default_passthrough(void)
2273 return iommu_def_domain_type
== IOMMU_DOMAIN_IDENTITY
;
2275 EXPORT_SYMBOL_GPL(iommu_default_passthrough
);
2277 const struct iommu_ops
*iommu_ops_from_fwnode(struct fwnode_handle
*fwnode
)
2279 const struct iommu_ops
*ops
= NULL
;
2280 struct iommu_device
*iommu
;
2282 spin_lock(&iommu_device_lock
);
2283 list_for_each_entry(iommu
, &iommu_device_list
, list
)
2284 if (iommu
->fwnode
== fwnode
) {
2288 spin_unlock(&iommu_device_lock
);
2292 int iommu_fwspec_init(struct device
*dev
, struct fwnode_handle
*iommu_fwnode
,
2293 const struct iommu_ops
*ops
)
2295 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2298 return ops
== fwspec
->ops
? 0 : -EINVAL
;
2300 fwspec
= kzalloc(sizeof(*fwspec
), GFP_KERNEL
);
2304 of_node_get(to_of_node(iommu_fwnode
));
2305 fwspec
->iommu_fwnode
= iommu_fwnode
;
2307 dev_iommu_fwspec_set(dev
, fwspec
);
2310 EXPORT_SYMBOL_GPL(iommu_fwspec_init
);
2312 void iommu_fwspec_free(struct device
*dev
)
2314 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2317 fwnode_handle_put(fwspec
->iommu_fwnode
);
2319 dev_iommu_fwspec_set(dev
, NULL
);
2322 EXPORT_SYMBOL_GPL(iommu_fwspec_free
);
2324 int iommu_fwspec_add_ids(struct device
*dev
, u32
*ids
, int num_ids
)
2326 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2333 size
= offsetof(struct iommu_fwspec
, ids
[fwspec
->num_ids
+ num_ids
]);
2334 if (size
> sizeof(*fwspec
)) {
2335 fwspec
= krealloc(fwspec
, size
, GFP_KERNEL
);
2339 dev_iommu_fwspec_set(dev
, fwspec
);
2342 for (i
= 0; i
< num_ids
; i
++)
2343 fwspec
->ids
[fwspec
->num_ids
+ i
] = ids
[i
];
2345 fwspec
->num_ids
+= num_ids
;
2348 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids
);
2351 * Per device IOMMU features.
2353 bool iommu_dev_has_feature(struct device
*dev
, enum iommu_dev_features feat
)
2355 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2357 if (ops
&& ops
->dev_has_feat
)
2358 return ops
->dev_has_feat(dev
, feat
);
2362 EXPORT_SYMBOL_GPL(iommu_dev_has_feature
);
2364 int iommu_dev_enable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2366 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2368 if (ops
&& ops
->dev_enable_feat
)
2369 return ops
->dev_enable_feat(dev
, feat
);
2373 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature
);
2376 * The device drivers should do the necessary cleanups before calling this.
2377 * For example, before disabling the aux-domain feature, the device driver
2378 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2380 int iommu_dev_disable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2382 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2384 if (ops
&& ops
->dev_disable_feat
)
2385 return ops
->dev_disable_feat(dev
, feat
);
2389 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature
);
2391 bool iommu_dev_feature_enabled(struct device
*dev
, enum iommu_dev_features feat
)
2393 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2395 if (ops
&& ops
->dev_feat_enabled
)
2396 return ops
->dev_feat_enabled(dev
, feat
);
2400 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled
);
2403 * Aux-domain specific attach/detach.
2405 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2406 * true. Also, as long as domains are attached to a device through this
2407 * interface, any tries to call iommu_attach_device() should fail
2408 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2409 * This should make us safe against a device being attached to a guest as a
2410 * whole while there are still pasid users on it (aux and sva).
2412 int iommu_aux_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
2416 if (domain
->ops
->aux_attach_dev
)
2417 ret
= domain
->ops
->aux_attach_dev(domain
, dev
);
2420 trace_attach_device_to_domain(dev
);
2424 EXPORT_SYMBOL_GPL(iommu_aux_attach_device
);
2426 void iommu_aux_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
2428 if (domain
->ops
->aux_detach_dev
) {
2429 domain
->ops
->aux_detach_dev(domain
, dev
);
2430 trace_detach_device_from_domain(dev
);
2433 EXPORT_SYMBOL_GPL(iommu_aux_detach_device
);
2435 int iommu_aux_get_pasid(struct iommu_domain
*domain
, struct device
*dev
)
2439 if (domain
->ops
->aux_get_pasid
)
2440 ret
= domain
->ops
->aux_get_pasid(domain
, dev
);
2444 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid
);
2447 * iommu_sva_bind_device() - Bind a process address space to a device
2449 * @mm: the mm to bind, caller must hold a reference to it
2451 * Create a bond between device and address space, allowing the device to access
2452 * the mm using the returned PASID. If a bond already exists between @device and
2453 * @mm, it is returned and an additional reference is taken. Caller must call
2454 * iommu_sva_unbind_device() to release each reference.
2456 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2457 * initialize the required SVA features.
2459 * On error, returns an ERR_PTR value.
2462 iommu_sva_bind_device(struct device
*dev
, struct mm_struct
*mm
, void *drvdata
)
2464 struct iommu_group
*group
;
2465 struct iommu_sva
*handle
= ERR_PTR(-EINVAL
);
2466 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2468 if (!ops
|| !ops
->sva_bind
)
2469 return ERR_PTR(-ENODEV
);
2471 group
= iommu_group_get(dev
);
2473 return ERR_PTR(-ENODEV
);
2475 /* Ensure device count and domain don't change while we're binding */
2476 mutex_lock(&group
->mutex
);
2479 * To keep things simple, SVA currently doesn't support IOMMU groups
2480 * with more than one device. Existing SVA-capable systems are not
2481 * affected by the problems that required IOMMU groups (lack of ACS
2482 * isolation, device ID aliasing and other hardware issues).
2484 if (iommu_group_device_count(group
) != 1)
2487 handle
= ops
->sva_bind(dev
, mm
, drvdata
);
2490 mutex_unlock(&group
->mutex
);
2491 iommu_group_put(group
);
2495 EXPORT_SYMBOL_GPL(iommu_sva_bind_device
);
2498 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2499 * @handle: the handle returned by iommu_sva_bind_device()
2501 * Put reference to a bond between device and address space. The device should
2502 * not be issuing any more transaction for this PASID. All outstanding page
2503 * requests for this PASID must have been flushed to the IOMMU.
2505 * Returns 0 on success, or an error value
2507 void iommu_sva_unbind_device(struct iommu_sva
*handle
)
2509 struct iommu_group
*group
;
2510 struct device
*dev
= handle
->dev
;
2511 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2513 if (!ops
|| !ops
->sva_unbind
)
2516 group
= iommu_group_get(dev
);
2520 mutex_lock(&group
->mutex
);
2521 ops
->sva_unbind(handle
);
2522 mutex_unlock(&group
->mutex
);
2524 iommu_group_put(group
);
2526 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device
);
2528 int iommu_sva_set_ops(struct iommu_sva
*handle
,
2529 const struct iommu_sva_ops
*sva_ops
)
2531 if (handle
->ops
&& handle
->ops
!= sva_ops
)
2534 handle
->ops
= sva_ops
;
2537 EXPORT_SYMBOL_GPL(iommu_sva_set_ops
);
2539 int iommu_sva_get_pasid(struct iommu_sva
*handle
)
2541 const struct iommu_ops
*ops
= handle
->dev
->bus
->iommu_ops
;
2543 if (!ops
|| !ops
->sva_get_pasid
)
2544 return IOMMU_PASID_INVALID
;
2546 return ops
->sva_get_pasid(handle
);
2548 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid
);