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/dma-iommu.h>
11 #include <linux/kernel.h>
12 #include <linux/bits.h>
13 #include <linux/bug.h>
14 #include <linux/types.h>
15 #include <linux/init.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/errno.h>
19 #include <linux/iommu.h>
20 #include <linux/idr.h>
21 #include <linux/notifier.h>
22 #include <linux/err.h>
23 #include <linux/pci.h>
24 #include <linux/bitops.h>
25 #include <linux/property.h>
26 #include <linux/fsl/mc.h>
27 #include <linux/module.h>
28 #include <trace/events/iommu.h>
30 static struct kset
*iommu_group_kset
;
31 static DEFINE_IDA(iommu_group_ida
);
33 static unsigned int iommu_def_domain_type __read_mostly
;
34 static bool iommu_dma_strict __read_mostly
= IS_ENABLED(CONFIG_IOMMU_DEFAULT_DMA_STRICT
);
35 static u32 iommu_cmd_line __read_mostly
;
39 struct kobject
*devices_kobj
;
40 struct list_head devices
;
42 struct blocking_notifier_head notifier
;
44 void (*iommu_data_release
)(void *iommu_data
);
47 struct iommu_domain
*default_domain
;
48 struct iommu_domain
*domain
;
49 struct list_head entry
;
53 struct list_head list
;
58 struct iommu_group_attribute
{
59 struct attribute attr
;
60 ssize_t (*show
)(struct iommu_group
*group
, char *buf
);
61 ssize_t (*store
)(struct iommu_group
*group
,
62 const char *buf
, size_t count
);
65 static const char * const iommu_group_resv_type_string
[] = {
66 [IOMMU_RESV_DIRECT
] = "direct",
67 [IOMMU_RESV_DIRECT_RELAXABLE
] = "direct-relaxable",
68 [IOMMU_RESV_RESERVED
] = "reserved",
69 [IOMMU_RESV_MSI
] = "msi",
70 [IOMMU_RESV_SW_MSI
] = "msi",
73 #define IOMMU_CMD_LINE_DMA_API BIT(0)
74 #define IOMMU_CMD_LINE_STRICT BIT(1)
76 static int iommu_alloc_default_domain(struct iommu_group
*group
,
78 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
80 static int __iommu_attach_device(struct iommu_domain
*domain
,
82 static int __iommu_attach_group(struct iommu_domain
*domain
,
83 struct iommu_group
*group
);
84 static void __iommu_detach_group(struct iommu_domain
*domain
,
85 struct iommu_group
*group
);
86 static int iommu_create_device_direct_mappings(struct iommu_group
*group
,
88 static struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
);
89 static ssize_t
iommu_group_store_type(struct iommu_group
*group
,
90 const char *buf
, size_t count
);
92 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
93 struct iommu_group_attribute iommu_group_attr_##_name = \
94 __ATTR(_name, _mode, _show, _store)
96 #define to_iommu_group_attr(_attr) \
97 container_of(_attr, struct iommu_group_attribute, attr)
98 #define to_iommu_group(_kobj) \
99 container_of(_kobj, struct iommu_group, kobj)
101 static LIST_HEAD(iommu_device_list
);
102 static DEFINE_SPINLOCK(iommu_device_lock
);
105 * Use a function instead of an array here because the domain-type is a
106 * bit-field, so an array would waste memory.
108 static const char *iommu_domain_type_str(unsigned int t
)
111 case IOMMU_DOMAIN_BLOCKED
:
113 case IOMMU_DOMAIN_IDENTITY
:
114 return "Passthrough";
115 case IOMMU_DOMAIN_UNMANAGED
:
117 case IOMMU_DOMAIN_DMA
:
118 case IOMMU_DOMAIN_DMA_FQ
:
125 static int __init
iommu_subsys_init(void)
127 if (!(iommu_cmd_line
& IOMMU_CMD_LINE_DMA_API
)) {
128 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH
))
129 iommu_set_default_passthrough(false);
131 iommu_set_default_translated(false);
133 if (iommu_default_passthrough() && mem_encrypt_active()) {
134 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
135 iommu_set_default_translated(false);
139 if (!iommu_default_passthrough() && !iommu_dma_strict
)
140 iommu_def_domain_type
= IOMMU_DOMAIN_DMA_FQ
;
142 pr_info("Default domain type: %s %s\n",
143 iommu_domain_type_str(iommu_def_domain_type
),
144 (iommu_cmd_line
& IOMMU_CMD_LINE_DMA_API
) ?
145 "(set via kernel command line)" : "");
147 if (!iommu_default_passthrough())
148 pr_info("DMA domain TLB invalidation policy: %s mode %s\n",
149 iommu_dma_strict
? "strict" : "lazy",
150 (iommu_cmd_line
& IOMMU_CMD_LINE_STRICT
) ?
151 "(set via kernel command line)" : "");
155 subsys_initcall(iommu_subsys_init
);
158 * iommu_device_register() - Register an IOMMU hardware instance
159 * @iommu: IOMMU handle for the instance
160 * @ops: IOMMU ops to associate with the instance
161 * @hwdev: (optional) actual instance device, used for fwnode lookup
163 * Return: 0 on success, or an error.
165 int iommu_device_register(struct iommu_device
*iommu
,
166 const struct iommu_ops
*ops
, struct device
*hwdev
)
168 /* We need to be able to take module references appropriately */
169 if (WARN_ON(is_module_address((unsigned long)ops
) && !ops
->owner
))
174 iommu
->fwnode
= hwdev
->fwnode
;
176 spin_lock(&iommu_device_lock
);
177 list_add_tail(&iommu
->list
, &iommu_device_list
);
178 spin_unlock(&iommu_device_lock
);
181 EXPORT_SYMBOL_GPL(iommu_device_register
);
183 void iommu_device_unregister(struct iommu_device
*iommu
)
185 spin_lock(&iommu_device_lock
);
186 list_del(&iommu
->list
);
187 spin_unlock(&iommu_device_lock
);
189 EXPORT_SYMBOL_GPL(iommu_device_unregister
);
191 static struct dev_iommu
*dev_iommu_get(struct device
*dev
)
193 struct dev_iommu
*param
= dev
->iommu
;
198 param
= kzalloc(sizeof(*param
), GFP_KERNEL
);
202 mutex_init(¶m
->lock
);
207 static void dev_iommu_free(struct device
*dev
)
209 iommu_fwspec_free(dev
);
214 static int __iommu_probe_device(struct device
*dev
, struct list_head
*group_list
)
216 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
217 struct iommu_device
*iommu_dev
;
218 struct iommu_group
*group
;
224 if (!dev_iommu_get(dev
))
227 if (!try_module_get(ops
->owner
)) {
232 iommu_dev
= ops
->probe_device(dev
);
233 if (IS_ERR(iommu_dev
)) {
234 ret
= PTR_ERR(iommu_dev
);
238 dev
->iommu
->iommu_dev
= iommu_dev
;
240 group
= iommu_group_get_for_dev(dev
);
242 ret
= PTR_ERR(group
);
245 iommu_group_put(group
);
247 if (group_list
&& !group
->default_domain
&& list_empty(&group
->entry
))
248 list_add_tail(&group
->entry
, group_list
);
250 iommu_device_link(iommu_dev
, dev
);
255 ops
->release_device(dev
);
258 module_put(ops
->owner
);
266 int iommu_probe_device(struct device
*dev
)
268 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
269 struct iommu_group
*group
;
272 ret
= __iommu_probe_device(dev
, NULL
);
276 group
= iommu_group_get(dev
);
283 * Try to allocate a default domain - needs support from the
284 * IOMMU driver. There are still some drivers which don't
285 * support default domains, so the return value is not yet
288 mutex_lock(&group
->mutex
);
289 iommu_alloc_default_domain(group
, dev
);
290 mutex_unlock(&group
->mutex
);
292 if (group
->default_domain
) {
293 ret
= __iommu_attach_device(group
->default_domain
, dev
);
295 iommu_group_put(group
);
300 iommu_create_device_direct_mappings(group
, dev
);
302 iommu_group_put(group
);
304 if (ops
->probe_finalize
)
305 ops
->probe_finalize(dev
);
310 iommu_release_device(dev
);
317 void iommu_release_device(struct device
*dev
)
319 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
324 iommu_device_unlink(dev
->iommu
->iommu_dev
, dev
);
326 ops
->release_device(dev
);
328 iommu_group_remove_device(dev
);
329 module_put(ops
->owner
);
333 static int __init
iommu_set_def_domain_type(char *str
)
338 ret
= kstrtobool(str
, &pt
);
343 iommu_set_default_passthrough(true);
345 iommu_set_default_translated(true);
349 early_param("iommu.passthrough", iommu_set_def_domain_type
);
351 static int __init
iommu_dma_setup(char *str
)
353 int ret
= kstrtobool(str
, &iommu_dma_strict
);
356 iommu_cmd_line
|= IOMMU_CMD_LINE_STRICT
;
359 early_param("iommu.strict", iommu_dma_setup
);
361 void iommu_set_dma_strict(void)
363 iommu_dma_strict
= true;
364 if (iommu_def_domain_type
== IOMMU_DOMAIN_DMA_FQ
)
365 iommu_def_domain_type
= IOMMU_DOMAIN_DMA
;
368 static ssize_t
iommu_group_attr_show(struct kobject
*kobj
,
369 struct attribute
*__attr
, char *buf
)
371 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
372 struct iommu_group
*group
= to_iommu_group(kobj
);
376 ret
= attr
->show(group
, buf
);
380 static ssize_t
iommu_group_attr_store(struct kobject
*kobj
,
381 struct attribute
*__attr
,
382 const char *buf
, size_t count
)
384 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
385 struct iommu_group
*group
= to_iommu_group(kobj
);
389 ret
= attr
->store(group
, buf
, count
);
393 static const struct sysfs_ops iommu_group_sysfs_ops
= {
394 .show
= iommu_group_attr_show
,
395 .store
= iommu_group_attr_store
,
398 static int iommu_group_create_file(struct iommu_group
*group
,
399 struct iommu_group_attribute
*attr
)
401 return sysfs_create_file(&group
->kobj
, &attr
->attr
);
404 static void iommu_group_remove_file(struct iommu_group
*group
,
405 struct iommu_group_attribute
*attr
)
407 sysfs_remove_file(&group
->kobj
, &attr
->attr
);
410 static ssize_t
iommu_group_show_name(struct iommu_group
*group
, char *buf
)
412 return sprintf(buf
, "%s\n", group
->name
);
416 * iommu_insert_resv_region - Insert a new region in the
417 * list of reserved regions.
418 * @new: new region to insert
419 * @regions: list of regions
421 * Elements are sorted by start address and overlapping segments
422 * of the same type are merged.
424 static int iommu_insert_resv_region(struct iommu_resv_region
*new,
425 struct list_head
*regions
)
427 struct iommu_resv_region
*iter
, *tmp
, *nr
, *top
;
430 nr
= iommu_alloc_resv_region(new->start
, new->length
,
431 new->prot
, new->type
);
435 /* First add the new element based on start address sorting */
436 list_for_each_entry(iter
, regions
, list
) {
437 if (nr
->start
< iter
->start
||
438 (nr
->start
== iter
->start
&& nr
->type
<= iter
->type
))
441 list_add_tail(&nr
->list
, &iter
->list
);
443 /* Merge overlapping segments of type nr->type in @regions, if any */
444 list_for_each_entry_safe(iter
, tmp
, regions
, list
) {
445 phys_addr_t top_end
, iter_end
= iter
->start
+ iter
->length
- 1;
447 /* no merge needed on elements of different types than @new */
448 if (iter
->type
!= new->type
) {
449 list_move_tail(&iter
->list
, &stack
);
453 /* look for the last stack element of same type as @iter */
454 list_for_each_entry_reverse(top
, &stack
, list
)
455 if (top
->type
== iter
->type
)
458 list_move_tail(&iter
->list
, &stack
);
462 top_end
= top
->start
+ top
->length
- 1;
464 if (iter
->start
> top_end
+ 1) {
465 list_move_tail(&iter
->list
, &stack
);
467 top
->length
= max(top_end
, iter_end
) - top
->start
+ 1;
468 list_del(&iter
->list
);
472 list_splice(&stack
, regions
);
477 iommu_insert_device_resv_regions(struct list_head
*dev_resv_regions
,
478 struct list_head
*group_resv_regions
)
480 struct iommu_resv_region
*entry
;
483 list_for_each_entry(entry
, dev_resv_regions
, list
) {
484 ret
= iommu_insert_resv_region(entry
, group_resv_regions
);
491 int iommu_get_group_resv_regions(struct iommu_group
*group
,
492 struct list_head
*head
)
494 struct group_device
*device
;
497 mutex_lock(&group
->mutex
);
498 list_for_each_entry(device
, &group
->devices
, list
) {
499 struct list_head dev_resv_regions
;
501 INIT_LIST_HEAD(&dev_resv_regions
);
502 iommu_get_resv_regions(device
->dev
, &dev_resv_regions
);
503 ret
= iommu_insert_device_resv_regions(&dev_resv_regions
, head
);
504 iommu_put_resv_regions(device
->dev
, &dev_resv_regions
);
508 mutex_unlock(&group
->mutex
);
511 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions
);
513 static ssize_t
iommu_group_show_resv_regions(struct iommu_group
*group
,
516 struct iommu_resv_region
*region
, *next
;
517 struct list_head group_resv_regions
;
520 INIT_LIST_HEAD(&group_resv_regions
);
521 iommu_get_group_resv_regions(group
, &group_resv_regions
);
523 list_for_each_entry_safe(region
, next
, &group_resv_regions
, list
) {
524 str
+= sprintf(str
, "0x%016llx 0x%016llx %s\n",
525 (long long int)region
->start
,
526 (long long int)(region
->start
+
528 iommu_group_resv_type_string
[region
->type
]);
535 static ssize_t
iommu_group_show_type(struct iommu_group
*group
,
538 char *type
= "unknown\n";
540 mutex_lock(&group
->mutex
);
541 if (group
->default_domain
) {
542 switch (group
->default_domain
->type
) {
543 case IOMMU_DOMAIN_BLOCKED
:
546 case IOMMU_DOMAIN_IDENTITY
:
549 case IOMMU_DOMAIN_UNMANAGED
:
550 type
= "unmanaged\n";
552 case IOMMU_DOMAIN_DMA
:
555 case IOMMU_DOMAIN_DMA_FQ
:
560 mutex_unlock(&group
->mutex
);
566 static IOMMU_GROUP_ATTR(name
, S_IRUGO
, iommu_group_show_name
, NULL
);
568 static IOMMU_GROUP_ATTR(reserved_regions
, 0444,
569 iommu_group_show_resv_regions
, NULL
);
571 static IOMMU_GROUP_ATTR(type
, 0644, iommu_group_show_type
,
572 iommu_group_store_type
);
574 static void iommu_group_release(struct kobject
*kobj
)
576 struct iommu_group
*group
= to_iommu_group(kobj
);
578 pr_debug("Releasing group %d\n", group
->id
);
580 if (group
->iommu_data_release
)
581 group
->iommu_data_release(group
->iommu_data
);
583 ida_simple_remove(&iommu_group_ida
, group
->id
);
585 if (group
->default_domain
)
586 iommu_domain_free(group
->default_domain
);
592 static struct kobj_type iommu_group_ktype
= {
593 .sysfs_ops
= &iommu_group_sysfs_ops
,
594 .release
= iommu_group_release
,
598 * iommu_group_alloc - Allocate a new group
600 * This function is called by an iommu driver to allocate a new iommu
601 * group. The iommu group represents the minimum granularity of the iommu.
602 * Upon successful return, the caller holds a reference to the supplied
603 * group in order to hold the group until devices are added. Use
604 * iommu_group_put() to release this extra reference count, allowing the
605 * group to be automatically reclaimed once it has no devices or external
608 struct iommu_group
*iommu_group_alloc(void)
610 struct iommu_group
*group
;
613 group
= kzalloc(sizeof(*group
), GFP_KERNEL
);
615 return ERR_PTR(-ENOMEM
);
617 group
->kobj
.kset
= iommu_group_kset
;
618 mutex_init(&group
->mutex
);
619 INIT_LIST_HEAD(&group
->devices
);
620 INIT_LIST_HEAD(&group
->entry
);
621 BLOCKING_INIT_NOTIFIER_HEAD(&group
->notifier
);
623 ret
= ida_simple_get(&iommu_group_ida
, 0, 0, GFP_KERNEL
);
630 ret
= kobject_init_and_add(&group
->kobj
, &iommu_group_ktype
,
631 NULL
, "%d", group
->id
);
633 ida_simple_remove(&iommu_group_ida
, group
->id
);
634 kobject_put(&group
->kobj
);
638 group
->devices_kobj
= kobject_create_and_add("devices", &group
->kobj
);
639 if (!group
->devices_kobj
) {
640 kobject_put(&group
->kobj
); /* triggers .release & free */
641 return ERR_PTR(-ENOMEM
);
645 * The devices_kobj holds a reference on the group kobject, so
646 * as long as that exists so will the group. We can therefore
647 * use the devices_kobj for reference counting.
649 kobject_put(&group
->kobj
);
651 ret
= iommu_group_create_file(group
,
652 &iommu_group_attr_reserved_regions
);
656 ret
= iommu_group_create_file(group
, &iommu_group_attr_type
);
660 pr_debug("Allocated group %d\n", group
->id
);
664 EXPORT_SYMBOL_GPL(iommu_group_alloc
);
666 struct iommu_group
*iommu_group_get_by_id(int id
)
668 struct kobject
*group_kobj
;
669 struct iommu_group
*group
;
672 if (!iommu_group_kset
)
675 name
= kasprintf(GFP_KERNEL
, "%d", id
);
679 group_kobj
= kset_find_obj(iommu_group_kset
, name
);
685 group
= container_of(group_kobj
, struct iommu_group
, kobj
);
686 BUG_ON(group
->id
!= id
);
688 kobject_get(group
->devices_kobj
);
689 kobject_put(&group
->kobj
);
693 EXPORT_SYMBOL_GPL(iommu_group_get_by_id
);
696 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
699 * iommu drivers can store data in the group for use when doing iommu
700 * operations. This function provides a way to retrieve it. Caller
701 * should hold a group reference.
703 void *iommu_group_get_iommudata(struct iommu_group
*group
)
705 return group
->iommu_data
;
707 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata
);
710 * iommu_group_set_iommudata - set iommu_data for a group
712 * @iommu_data: new data
713 * @release: release function for iommu_data
715 * iommu drivers can store data in the group for use when doing iommu
716 * operations. This function provides a way to set the data after
717 * the group has been allocated. Caller should hold a group reference.
719 void iommu_group_set_iommudata(struct iommu_group
*group
, void *iommu_data
,
720 void (*release
)(void *iommu_data
))
722 group
->iommu_data
= iommu_data
;
723 group
->iommu_data_release
= release
;
725 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata
);
728 * iommu_group_set_name - set name for a group
732 * Allow iommu driver to set a name for a group. When set it will
733 * appear in a name attribute file under the group in sysfs.
735 int iommu_group_set_name(struct iommu_group
*group
, const char *name
)
740 iommu_group_remove_file(group
, &iommu_group_attr_name
);
747 group
->name
= kstrdup(name
, GFP_KERNEL
);
751 ret
= iommu_group_create_file(group
, &iommu_group_attr_name
);
760 EXPORT_SYMBOL_GPL(iommu_group_set_name
);
762 static int iommu_create_device_direct_mappings(struct iommu_group
*group
,
765 struct iommu_domain
*domain
= group
->default_domain
;
766 struct iommu_resv_region
*entry
;
767 struct list_head mappings
;
768 unsigned long pg_size
;
771 if (!domain
|| !iommu_is_dma_domain(domain
))
774 BUG_ON(!domain
->pgsize_bitmap
);
776 pg_size
= 1UL << __ffs(domain
->pgsize_bitmap
);
777 INIT_LIST_HEAD(&mappings
);
779 iommu_get_resv_regions(dev
, &mappings
);
781 /* We need to consider overlapping regions for different devices */
782 list_for_each_entry(entry
, &mappings
, list
) {
783 dma_addr_t start
, end
, addr
;
786 if (domain
->ops
->apply_resv_region
)
787 domain
->ops
->apply_resv_region(dev
, domain
, entry
);
789 start
= ALIGN(entry
->start
, pg_size
);
790 end
= ALIGN(entry
->start
+ entry
->length
, pg_size
);
792 if (entry
->type
!= IOMMU_RESV_DIRECT
&&
793 entry
->type
!= IOMMU_RESV_DIRECT_RELAXABLE
)
796 for (addr
= start
; addr
<= end
; addr
+= pg_size
) {
797 phys_addr_t phys_addr
;
802 phys_addr
= iommu_iova_to_phys(domain
, addr
);
810 ret
= iommu_map(domain
, addr
- map_size
,
811 addr
- map_size
, map_size
,
821 iommu_flush_iotlb_all(domain
);
824 iommu_put_resv_regions(dev
, &mappings
);
829 static bool iommu_is_attach_deferred(struct iommu_domain
*domain
,
832 if (domain
->ops
->is_attach_deferred
)
833 return domain
->ops
->is_attach_deferred(domain
, dev
);
839 * iommu_group_add_device - add a device to an iommu group
840 * @group: the group into which to add the device (reference should be held)
843 * This function is called by an iommu driver to add a device into a
844 * group. Adding a device increments the group reference count.
846 int iommu_group_add_device(struct iommu_group
*group
, struct device
*dev
)
849 struct group_device
*device
;
851 device
= kzalloc(sizeof(*device
), GFP_KERNEL
);
857 ret
= sysfs_create_link(&dev
->kobj
, &group
->kobj
, "iommu_group");
859 goto err_free_device
;
861 device
->name
= kasprintf(GFP_KERNEL
, "%s", kobject_name(&dev
->kobj
));
865 goto err_remove_link
;
868 ret
= sysfs_create_link_nowarn(group
->devices_kobj
,
869 &dev
->kobj
, device
->name
);
871 if (ret
== -EEXIST
&& i
>= 0) {
873 * Account for the slim chance of collision
874 * and append an instance to the name.
877 device
->name
= kasprintf(GFP_KERNEL
, "%s.%d",
878 kobject_name(&dev
->kobj
), i
++);
884 kobject_get(group
->devices_kobj
);
886 dev
->iommu_group
= group
;
888 mutex_lock(&group
->mutex
);
889 list_add_tail(&device
->list
, &group
->devices
);
890 if (group
->domain
&& !iommu_is_attach_deferred(group
->domain
, dev
))
891 ret
= __iommu_attach_device(group
->domain
, dev
);
892 mutex_unlock(&group
->mutex
);
896 /* Notify any listeners about change to group. */
897 blocking_notifier_call_chain(&group
->notifier
,
898 IOMMU_GROUP_NOTIFY_ADD_DEVICE
, dev
);
900 trace_add_device_to_group(group
->id
, dev
);
902 dev_info(dev
, "Adding to iommu group %d\n", group
->id
);
907 mutex_lock(&group
->mutex
);
908 list_del(&device
->list
);
909 mutex_unlock(&group
->mutex
);
910 dev
->iommu_group
= NULL
;
911 kobject_put(group
->devices_kobj
);
912 sysfs_remove_link(group
->devices_kobj
, device
->name
);
916 sysfs_remove_link(&dev
->kobj
, "iommu_group");
919 dev_err(dev
, "Failed to add to iommu group %d: %d\n", group
->id
, ret
);
922 EXPORT_SYMBOL_GPL(iommu_group_add_device
);
925 * iommu_group_remove_device - remove a device from it's current group
926 * @dev: device to be removed
928 * This function is called by an iommu driver to remove the device from
929 * it's current group. This decrements the iommu group reference count.
931 void iommu_group_remove_device(struct device
*dev
)
933 struct iommu_group
*group
= dev
->iommu_group
;
934 struct group_device
*tmp_device
, *device
= NULL
;
939 dev_info(dev
, "Removing from iommu group %d\n", group
->id
);
941 /* Pre-notify listeners that a device is being removed. */
942 blocking_notifier_call_chain(&group
->notifier
,
943 IOMMU_GROUP_NOTIFY_DEL_DEVICE
, dev
);
945 mutex_lock(&group
->mutex
);
946 list_for_each_entry(tmp_device
, &group
->devices
, list
) {
947 if (tmp_device
->dev
== dev
) {
949 list_del(&device
->list
);
953 mutex_unlock(&group
->mutex
);
958 sysfs_remove_link(group
->devices_kobj
, device
->name
);
959 sysfs_remove_link(&dev
->kobj
, "iommu_group");
961 trace_remove_device_from_group(group
->id
, dev
);
965 dev
->iommu_group
= NULL
;
966 kobject_put(group
->devices_kobj
);
968 EXPORT_SYMBOL_GPL(iommu_group_remove_device
);
970 static int iommu_group_device_count(struct iommu_group
*group
)
972 struct group_device
*entry
;
975 list_for_each_entry(entry
, &group
->devices
, list
)
982 * iommu_group_for_each_dev - iterate over each device in the group
984 * @data: caller opaque data to be passed to callback function
985 * @fn: caller supplied callback function
987 * This function is called by group users to iterate over group devices.
988 * Callers should hold a reference count to the group during callback.
989 * The group->mutex is held across callbacks, which will block calls to
990 * iommu_group_add/remove_device.
992 static int __iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
993 int (*fn
)(struct device
*, void *))
995 struct group_device
*device
;
998 list_for_each_entry(device
, &group
->devices
, list
) {
999 ret
= fn(device
->dev
, data
);
1007 int iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
1008 int (*fn
)(struct device
*, void *))
1012 mutex_lock(&group
->mutex
);
1013 ret
= __iommu_group_for_each_dev(group
, data
, fn
);
1014 mutex_unlock(&group
->mutex
);
1018 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev
);
1021 * iommu_group_get - Return the group for a device and increment reference
1022 * @dev: get the group that this device belongs to
1024 * This function is called by iommu drivers and users to get the group
1025 * for the specified device. If found, the group is returned and the group
1026 * reference in incremented, else NULL.
1028 struct iommu_group
*iommu_group_get(struct device
*dev
)
1030 struct iommu_group
*group
= dev
->iommu_group
;
1033 kobject_get(group
->devices_kobj
);
1037 EXPORT_SYMBOL_GPL(iommu_group_get
);
1040 * iommu_group_ref_get - Increment reference on a group
1041 * @group: the group to use, must not be NULL
1043 * This function is called by iommu drivers to take additional references on an
1044 * existing group. Returns the given group for convenience.
1046 struct iommu_group
*iommu_group_ref_get(struct iommu_group
*group
)
1048 kobject_get(group
->devices_kobj
);
1051 EXPORT_SYMBOL_GPL(iommu_group_ref_get
);
1054 * iommu_group_put - Decrement group reference
1055 * @group: the group to use
1057 * This function is called by iommu drivers and users to release the
1058 * iommu group. Once the reference count is zero, the group is released.
1060 void iommu_group_put(struct iommu_group
*group
)
1063 kobject_put(group
->devices_kobj
);
1065 EXPORT_SYMBOL_GPL(iommu_group_put
);
1068 * iommu_group_register_notifier - Register a notifier for group changes
1069 * @group: the group to watch
1070 * @nb: notifier block to signal
1072 * This function allows iommu group users to track changes in a group.
1073 * See include/linux/iommu.h for actions sent via this notifier. Caller
1074 * should hold a reference to the group throughout notifier registration.
1076 int iommu_group_register_notifier(struct iommu_group
*group
,
1077 struct notifier_block
*nb
)
1079 return blocking_notifier_chain_register(&group
->notifier
, nb
);
1081 EXPORT_SYMBOL_GPL(iommu_group_register_notifier
);
1084 * iommu_group_unregister_notifier - Unregister a notifier
1085 * @group: the group to watch
1086 * @nb: notifier block to signal
1088 * Unregister a previously registered group notifier block.
1090 int iommu_group_unregister_notifier(struct iommu_group
*group
,
1091 struct notifier_block
*nb
)
1093 return blocking_notifier_chain_unregister(&group
->notifier
, nb
);
1095 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier
);
1098 * iommu_register_device_fault_handler() - Register a device fault handler
1100 * @handler: the fault handler
1101 * @data: private data passed as argument to the handler
1103 * When an IOMMU fault event is received, this handler gets called with the
1104 * fault event and data as argument. The handler should return 0 on success. If
1105 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1106 * complete the fault by calling iommu_page_response() with one of the following
1108 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1109 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1110 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1111 * page faults if possible.
1113 * Return 0 if the fault handler was installed successfully, or an error.
1115 int iommu_register_device_fault_handler(struct device
*dev
,
1116 iommu_dev_fault_handler_t handler
,
1119 struct dev_iommu
*param
= dev
->iommu
;
1125 mutex_lock(¶m
->lock
);
1126 /* Only allow one fault handler registered for each device */
1127 if (param
->fault_param
) {
1133 param
->fault_param
= kzalloc(sizeof(*param
->fault_param
), GFP_KERNEL
);
1134 if (!param
->fault_param
) {
1139 param
->fault_param
->handler
= handler
;
1140 param
->fault_param
->data
= data
;
1141 mutex_init(¶m
->fault_param
->lock
);
1142 INIT_LIST_HEAD(¶m
->fault_param
->faults
);
1145 mutex_unlock(¶m
->lock
);
1149 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler
);
1152 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1155 * Remove the device fault handler installed with
1156 * iommu_register_device_fault_handler().
1158 * Return 0 on success, or an error.
1160 int iommu_unregister_device_fault_handler(struct device
*dev
)
1162 struct dev_iommu
*param
= dev
->iommu
;
1168 mutex_lock(¶m
->lock
);
1170 if (!param
->fault_param
)
1173 /* we cannot unregister handler if there are pending faults */
1174 if (!list_empty(¶m
->fault_param
->faults
)) {
1179 kfree(param
->fault_param
);
1180 param
->fault_param
= NULL
;
1183 mutex_unlock(¶m
->lock
);
1187 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler
);
1190 * iommu_report_device_fault() - Report fault event to device driver
1192 * @evt: fault event data
1194 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1195 * handler. When this function fails and the fault is recoverable, it is the
1196 * caller's responsibility to complete the fault.
1198 * Return 0 on success, or an error.
1200 int iommu_report_device_fault(struct device
*dev
, struct iommu_fault_event
*evt
)
1202 struct dev_iommu
*param
= dev
->iommu
;
1203 struct iommu_fault_event
*evt_pending
= NULL
;
1204 struct iommu_fault_param
*fparam
;
1210 /* we only report device fault if there is a handler registered */
1211 mutex_lock(¶m
->lock
);
1212 fparam
= param
->fault_param
;
1213 if (!fparam
|| !fparam
->handler
) {
1218 if (evt
->fault
.type
== IOMMU_FAULT_PAGE_REQ
&&
1219 (evt
->fault
.prm
.flags
& IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE
)) {
1220 evt_pending
= kmemdup(evt
, sizeof(struct iommu_fault_event
),
1226 mutex_lock(&fparam
->lock
);
1227 list_add_tail(&evt_pending
->list
, &fparam
->faults
);
1228 mutex_unlock(&fparam
->lock
);
1231 ret
= fparam
->handler(&evt
->fault
, fparam
->data
);
1232 if (ret
&& evt_pending
) {
1233 mutex_lock(&fparam
->lock
);
1234 list_del(&evt_pending
->list
);
1235 mutex_unlock(&fparam
->lock
);
1239 mutex_unlock(¶m
->lock
);
1242 EXPORT_SYMBOL_GPL(iommu_report_device_fault
);
1244 int iommu_page_response(struct device
*dev
,
1245 struct iommu_page_response
*msg
)
1249 struct iommu_fault_event
*evt
;
1250 struct iommu_fault_page_request
*prm
;
1251 struct dev_iommu
*param
= dev
->iommu
;
1252 bool has_pasid
= msg
->flags
& IOMMU_PAGE_RESP_PASID_VALID
;
1253 struct iommu_domain
*domain
= iommu_get_domain_for_dev(dev
);
1255 if (!domain
|| !domain
->ops
->page_response
)
1258 if (!param
|| !param
->fault_param
)
1261 if (msg
->version
!= IOMMU_PAGE_RESP_VERSION_1
||
1262 msg
->flags
& ~IOMMU_PAGE_RESP_PASID_VALID
)
1265 /* Only send response if there is a fault report pending */
1266 mutex_lock(¶m
->fault_param
->lock
);
1267 if (list_empty(¶m
->fault_param
->faults
)) {
1268 dev_warn_ratelimited(dev
, "no pending PRQ, drop response\n");
1272 * Check if we have a matching page request pending to respond,
1273 * otherwise return -EINVAL
1275 list_for_each_entry(evt
, ¶m
->fault_param
->faults
, list
) {
1276 prm
= &evt
->fault
.prm
;
1277 if (prm
->grpid
!= msg
->grpid
)
1281 * If the PASID is required, the corresponding request is
1282 * matched using the group ID, the PASID valid bit and the PASID
1283 * value. Otherwise only the group ID matches request and
1286 needs_pasid
= prm
->flags
& IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID
;
1287 if (needs_pasid
&& (!has_pasid
|| msg
->pasid
!= prm
->pasid
))
1290 if (!needs_pasid
&& has_pasid
) {
1291 /* No big deal, just clear it. */
1292 msg
->flags
&= ~IOMMU_PAGE_RESP_PASID_VALID
;
1296 ret
= domain
->ops
->page_response(dev
, evt
, msg
);
1297 list_del(&evt
->list
);
1303 mutex_unlock(¶m
->fault_param
->lock
);
1306 EXPORT_SYMBOL_GPL(iommu_page_response
);
1309 * iommu_group_id - Return ID for a group
1310 * @group: the group to ID
1312 * Return the unique ID for the group matching the sysfs group number.
1314 int iommu_group_id(struct iommu_group
*group
)
1318 EXPORT_SYMBOL_GPL(iommu_group_id
);
1320 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1321 unsigned long *devfns
);
1324 * To consider a PCI device isolated, we require ACS to support Source
1325 * Validation, Request Redirection, Completer Redirection, and Upstream
1326 * Forwarding. This effectively means that devices cannot spoof their
1327 * requester ID, requests and completions cannot be redirected, and all
1328 * transactions are forwarded upstream, even as it passes through a
1329 * bridge where the target device is downstream.
1331 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1334 * For multifunction devices which are not isolated from each other, find
1335 * all the other non-isolated functions and look for existing groups. For
1336 * each function, we also need to look for aliases to or from other devices
1337 * that may already have a group.
1339 static struct iommu_group
*get_pci_function_alias_group(struct pci_dev
*pdev
,
1340 unsigned long *devfns
)
1342 struct pci_dev
*tmp
= NULL
;
1343 struct iommu_group
*group
;
1345 if (!pdev
->multifunction
|| pci_acs_enabled(pdev
, REQ_ACS_FLAGS
))
1348 for_each_pci_dev(tmp
) {
1349 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
||
1350 PCI_SLOT(tmp
->devfn
) != PCI_SLOT(pdev
->devfn
) ||
1351 pci_acs_enabled(tmp
, REQ_ACS_FLAGS
))
1354 group
= get_pci_alias_group(tmp
, devfns
);
1365 * Look for aliases to or from the given device for existing groups. DMA
1366 * aliases are only supported on the same bus, therefore the search
1367 * space is quite small (especially since we're really only looking at pcie
1368 * device, and therefore only expect multiple slots on the root complex or
1369 * downstream switch ports). It's conceivable though that a pair of
1370 * multifunction devices could have aliases between them that would cause a
1371 * loop. To prevent this, we use a bitmap to track where we've been.
1373 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
1374 unsigned long *devfns
)
1376 struct pci_dev
*tmp
= NULL
;
1377 struct iommu_group
*group
;
1379 if (test_and_set_bit(pdev
->devfn
& 0xff, devfns
))
1382 group
= iommu_group_get(&pdev
->dev
);
1386 for_each_pci_dev(tmp
) {
1387 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
)
1390 /* We alias them or they alias us */
1391 if (pci_devs_are_dma_aliases(pdev
, tmp
)) {
1392 group
= get_pci_alias_group(tmp
, devfns
);
1398 group
= get_pci_function_alias_group(tmp
, devfns
);
1409 struct group_for_pci_data
{
1410 struct pci_dev
*pdev
;
1411 struct iommu_group
*group
;
1415 * DMA alias iterator callback, return the last seen device. Stop and return
1416 * the IOMMU group if we find one along the way.
1418 static int get_pci_alias_or_group(struct pci_dev
*pdev
, u16 alias
, void *opaque
)
1420 struct group_for_pci_data
*data
= opaque
;
1423 data
->group
= iommu_group_get(&pdev
->dev
);
1425 return data
->group
!= NULL
;
1429 * Generic device_group call-back function. It just allocates one
1430 * iommu-group per device.
1432 struct iommu_group
*generic_device_group(struct device
*dev
)
1434 return iommu_group_alloc();
1436 EXPORT_SYMBOL_GPL(generic_device_group
);
1439 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1440 * to find or create an IOMMU group for a device.
1442 struct iommu_group
*pci_device_group(struct device
*dev
)
1444 struct pci_dev
*pdev
= to_pci_dev(dev
);
1445 struct group_for_pci_data data
;
1446 struct pci_bus
*bus
;
1447 struct iommu_group
*group
= NULL
;
1448 u64 devfns
[4] = { 0 };
1450 if (WARN_ON(!dev_is_pci(dev
)))
1451 return ERR_PTR(-EINVAL
);
1454 * Find the upstream DMA alias for the device. A device must not
1455 * be aliased due to topology in order to have its own IOMMU group.
1456 * If we find an alias along the way that already belongs to a
1459 if (pci_for_each_dma_alias(pdev
, get_pci_alias_or_group
, &data
))
1465 * Continue upstream from the point of minimum IOMMU granularity
1466 * due to aliases to the point where devices are protected from
1467 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1470 for (bus
= pdev
->bus
; !pci_is_root_bus(bus
); bus
= bus
->parent
) {
1474 if (pci_acs_path_enabled(bus
->self
, NULL
, REQ_ACS_FLAGS
))
1479 group
= iommu_group_get(&pdev
->dev
);
1485 * Look for existing groups on device aliases. If we alias another
1486 * device or another device aliases us, use the same group.
1488 group
= get_pci_alias_group(pdev
, (unsigned long *)devfns
);
1493 * Look for existing groups on non-isolated functions on the same
1494 * slot and aliases of those funcions, if any. No need to clear
1495 * the search bitmap, the tested devfns are still valid.
1497 group
= get_pci_function_alias_group(pdev
, (unsigned long *)devfns
);
1501 /* No shared group found, allocate new */
1502 return iommu_group_alloc();
1504 EXPORT_SYMBOL_GPL(pci_device_group
);
1506 /* Get the IOMMU group for device on fsl-mc bus */
1507 struct iommu_group
*fsl_mc_device_group(struct device
*dev
)
1509 struct device
*cont_dev
= fsl_mc_cont_dev(dev
);
1510 struct iommu_group
*group
;
1512 group
= iommu_group_get(cont_dev
);
1514 group
= iommu_group_alloc();
1517 EXPORT_SYMBOL_GPL(fsl_mc_device_group
);
1519 static int iommu_get_def_domain_type(struct device
*dev
)
1521 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1523 if (dev_is_pci(dev
) && to_pci_dev(dev
)->untrusted
)
1524 return IOMMU_DOMAIN_DMA
;
1526 if (ops
->def_domain_type
)
1527 return ops
->def_domain_type(dev
);
1532 static int iommu_group_alloc_default_domain(struct bus_type
*bus
,
1533 struct iommu_group
*group
,
1536 struct iommu_domain
*dom
;
1538 dom
= __iommu_domain_alloc(bus
, type
);
1539 if (!dom
&& type
!= IOMMU_DOMAIN_DMA
) {
1540 dom
= __iommu_domain_alloc(bus
, IOMMU_DOMAIN_DMA
);
1542 pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1549 group
->default_domain
= dom
;
1551 group
->domain
= dom
;
1555 static int iommu_alloc_default_domain(struct iommu_group
*group
,
1560 if (group
->default_domain
)
1563 type
= iommu_get_def_domain_type(dev
) ? : iommu_def_domain_type
;
1565 return iommu_group_alloc_default_domain(dev
->bus
, group
, type
);
1569 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1570 * @dev: target device
1572 * This function is intended to be called by IOMMU drivers and extended to
1573 * support common, bus-defined algorithms when determining or creating the
1574 * IOMMU group for a device. On success, the caller will hold a reference
1575 * to the returned IOMMU group, which will already include the provided
1576 * device. The reference should be released with iommu_group_put().
1578 static struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
)
1580 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1581 struct iommu_group
*group
;
1584 group
= iommu_group_get(dev
);
1589 return ERR_PTR(-EINVAL
);
1591 group
= ops
->device_group(dev
);
1592 if (WARN_ON_ONCE(group
== NULL
))
1593 return ERR_PTR(-EINVAL
);
1598 ret
= iommu_group_add_device(group
, dev
);
1605 iommu_group_put(group
);
1607 return ERR_PTR(ret
);
1610 struct iommu_domain
*iommu_group_default_domain(struct iommu_group
*group
)
1612 return group
->default_domain
;
1615 static int probe_iommu_group(struct device
*dev
, void *data
)
1617 struct list_head
*group_list
= data
;
1618 struct iommu_group
*group
;
1621 /* Device is probed already if in a group */
1622 group
= iommu_group_get(dev
);
1624 iommu_group_put(group
);
1628 ret
= __iommu_probe_device(dev
, group_list
);
1635 static int remove_iommu_group(struct device
*dev
, void *data
)
1637 iommu_release_device(dev
);
1642 static int iommu_bus_notifier(struct notifier_block
*nb
,
1643 unsigned long action
, void *data
)
1645 unsigned long group_action
= 0;
1646 struct device
*dev
= data
;
1647 struct iommu_group
*group
;
1650 * ADD/DEL call into iommu driver ops if provided, which may
1651 * result in ADD/DEL notifiers to group->notifier
1653 if (action
== BUS_NOTIFY_ADD_DEVICE
) {
1656 ret
= iommu_probe_device(dev
);
1657 return (ret
) ? NOTIFY_DONE
: NOTIFY_OK
;
1658 } else if (action
== BUS_NOTIFY_REMOVED_DEVICE
) {
1659 iommu_release_device(dev
);
1664 * Remaining BUS_NOTIFYs get filtered and republished to the
1665 * group, if anyone is listening
1667 group
= iommu_group_get(dev
);
1672 case BUS_NOTIFY_BIND_DRIVER
:
1673 group_action
= IOMMU_GROUP_NOTIFY_BIND_DRIVER
;
1675 case BUS_NOTIFY_BOUND_DRIVER
:
1676 group_action
= IOMMU_GROUP_NOTIFY_BOUND_DRIVER
;
1678 case BUS_NOTIFY_UNBIND_DRIVER
:
1679 group_action
= IOMMU_GROUP_NOTIFY_UNBIND_DRIVER
;
1681 case BUS_NOTIFY_UNBOUND_DRIVER
:
1682 group_action
= IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER
;
1687 blocking_notifier_call_chain(&group
->notifier
,
1690 iommu_group_put(group
);
1694 struct __group_domain_type
{
1699 static int probe_get_default_domain_type(struct device
*dev
, void *data
)
1701 struct __group_domain_type
*gtype
= data
;
1702 unsigned int type
= iommu_get_def_domain_type(dev
);
1705 if (gtype
->type
&& gtype
->type
!= type
) {
1706 dev_warn(dev
, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1707 iommu_domain_type_str(type
),
1708 dev_name(gtype
->dev
),
1709 iommu_domain_type_str(gtype
->type
));
1722 static void probe_alloc_default_domain(struct bus_type
*bus
,
1723 struct iommu_group
*group
)
1725 struct __group_domain_type gtype
;
1727 memset(>ype
, 0, sizeof(gtype
));
1729 /* Ask for default domain requirements of all devices in the group */
1730 __iommu_group_for_each_dev(group
, >ype
,
1731 probe_get_default_domain_type
);
1734 gtype
.type
= iommu_def_domain_type
;
1736 iommu_group_alloc_default_domain(bus
, group
, gtype
.type
);
1740 static int iommu_group_do_dma_attach(struct device
*dev
, void *data
)
1742 struct iommu_domain
*domain
= data
;
1745 if (!iommu_is_attach_deferred(domain
, dev
))
1746 ret
= __iommu_attach_device(domain
, dev
);
1751 static int __iommu_group_dma_attach(struct iommu_group
*group
)
1753 return __iommu_group_for_each_dev(group
, group
->default_domain
,
1754 iommu_group_do_dma_attach
);
1757 static int iommu_group_do_probe_finalize(struct device
*dev
, void *data
)
1759 struct iommu_domain
*domain
= data
;
1761 if (domain
->ops
->probe_finalize
)
1762 domain
->ops
->probe_finalize(dev
);
1767 static void __iommu_group_dma_finalize(struct iommu_group
*group
)
1769 __iommu_group_for_each_dev(group
, group
->default_domain
,
1770 iommu_group_do_probe_finalize
);
1773 static int iommu_do_create_direct_mappings(struct device
*dev
, void *data
)
1775 struct iommu_group
*group
= data
;
1777 iommu_create_device_direct_mappings(group
, dev
);
1782 static int iommu_group_create_direct_mappings(struct iommu_group
*group
)
1784 return __iommu_group_for_each_dev(group
, group
,
1785 iommu_do_create_direct_mappings
);
1788 int bus_iommu_probe(struct bus_type
*bus
)
1790 struct iommu_group
*group
, *next
;
1791 LIST_HEAD(group_list
);
1795 * This code-path does not allocate the default domain when
1796 * creating the iommu group, so do it after the groups are
1799 ret
= bus_for_each_dev(bus
, NULL
, &group_list
, probe_iommu_group
);
1803 list_for_each_entry_safe(group
, next
, &group_list
, entry
) {
1804 /* Remove item from the list */
1805 list_del_init(&group
->entry
);
1807 mutex_lock(&group
->mutex
);
1809 /* Try to allocate default domain */
1810 probe_alloc_default_domain(bus
, group
);
1812 if (!group
->default_domain
) {
1813 mutex_unlock(&group
->mutex
);
1817 iommu_group_create_direct_mappings(group
);
1819 ret
= __iommu_group_dma_attach(group
);
1821 mutex_unlock(&group
->mutex
);
1826 __iommu_group_dma_finalize(group
);
1832 static int iommu_bus_init(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1834 struct notifier_block
*nb
;
1837 nb
= kzalloc(sizeof(struct notifier_block
), GFP_KERNEL
);
1841 nb
->notifier_call
= iommu_bus_notifier
;
1843 err
= bus_register_notifier(bus
, nb
);
1847 err
= bus_iommu_probe(bus
);
1856 bus_for_each_dev(bus
, NULL
, NULL
, remove_iommu_group
);
1857 bus_unregister_notifier(bus
, nb
);
1866 * bus_set_iommu - set iommu-callbacks for the bus
1868 * @ops: the callbacks provided by the iommu-driver
1870 * This function is called by an iommu driver to set the iommu methods
1871 * used for a particular bus. Drivers for devices on that bus can use
1872 * the iommu-api after these ops are registered.
1873 * This special function is needed because IOMMUs are usually devices on
1874 * the bus itself, so the iommu drivers are not initialized when the bus
1875 * is set up. With this function the iommu-driver can set the iommu-ops
1878 int bus_set_iommu(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1883 bus
->iommu_ops
= NULL
;
1887 if (bus
->iommu_ops
!= NULL
)
1890 bus
->iommu_ops
= ops
;
1892 /* Do IOMMU specific setup for this bus-type */
1893 err
= iommu_bus_init(bus
, ops
);
1895 bus
->iommu_ops
= NULL
;
1899 EXPORT_SYMBOL_GPL(bus_set_iommu
);
1901 bool iommu_present(struct bus_type
*bus
)
1903 return bus
->iommu_ops
!= NULL
;
1905 EXPORT_SYMBOL_GPL(iommu_present
);
1907 bool iommu_capable(struct bus_type
*bus
, enum iommu_cap cap
)
1909 if (!bus
->iommu_ops
|| !bus
->iommu_ops
->capable
)
1912 return bus
->iommu_ops
->capable(cap
);
1914 EXPORT_SYMBOL_GPL(iommu_capable
);
1917 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1918 * @domain: iommu domain
1919 * @handler: fault handler
1920 * @token: user data, will be passed back to the fault handler
1922 * This function should be used by IOMMU users which want to be notified
1923 * whenever an IOMMU fault happens.
1925 * The fault handler itself should return 0 on success, and an appropriate
1926 * error code otherwise.
1928 void iommu_set_fault_handler(struct iommu_domain
*domain
,
1929 iommu_fault_handler_t handler
,
1934 domain
->handler
= handler
;
1935 domain
->handler_token
= token
;
1937 EXPORT_SYMBOL_GPL(iommu_set_fault_handler
);
1939 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
1942 struct iommu_domain
*domain
;
1944 if (bus
== NULL
|| bus
->iommu_ops
== NULL
)
1947 domain
= bus
->iommu_ops
->domain_alloc(type
);
1951 domain
->ops
= bus
->iommu_ops
;
1952 domain
->type
= type
;
1953 /* Assume all sizes by default; the driver may override this later */
1954 domain
->pgsize_bitmap
= bus
->iommu_ops
->pgsize_bitmap
;
1956 /* Temporarily avoid -EEXIST while drivers still get their own cookies */
1957 if (iommu_is_dma_domain(domain
) && !domain
->iova_cookie
&& iommu_get_dma_cookie(domain
)) {
1958 iommu_domain_free(domain
);
1964 struct iommu_domain
*iommu_domain_alloc(struct bus_type
*bus
)
1966 return __iommu_domain_alloc(bus
, IOMMU_DOMAIN_UNMANAGED
);
1968 EXPORT_SYMBOL_GPL(iommu_domain_alloc
);
1970 void iommu_domain_free(struct iommu_domain
*domain
)
1972 iommu_put_dma_cookie(domain
);
1973 domain
->ops
->domain_free(domain
);
1975 EXPORT_SYMBOL_GPL(iommu_domain_free
);
1977 static int __iommu_attach_device(struct iommu_domain
*domain
,
1982 if (unlikely(domain
->ops
->attach_dev
== NULL
))
1985 ret
= domain
->ops
->attach_dev(domain
, dev
);
1987 trace_attach_device_to_domain(dev
);
1991 int iommu_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
1993 struct iommu_group
*group
;
1996 group
= iommu_group_get(dev
);
2001 * Lock the group to make sure the device-count doesn't
2002 * change while we are attaching
2004 mutex_lock(&group
->mutex
);
2006 if (iommu_group_device_count(group
) != 1)
2009 ret
= __iommu_attach_group(domain
, group
);
2012 mutex_unlock(&group
->mutex
);
2013 iommu_group_put(group
);
2017 EXPORT_SYMBOL_GPL(iommu_attach_device
);
2019 int iommu_deferred_attach(struct device
*dev
, struct iommu_domain
*domain
)
2021 const struct iommu_ops
*ops
= domain
->ops
;
2023 if (ops
->is_attach_deferred
&& ops
->is_attach_deferred(domain
, dev
))
2024 return __iommu_attach_device(domain
, dev
);
2030 * Check flags and other user provided data for valid combinations. We also
2031 * make sure no reserved fields or unused flags are set. This is to ensure
2032 * not breaking userspace in the future when these fields or flags are used.
2034 static int iommu_check_cache_invl_data(struct iommu_cache_invalidate_info
*info
)
2039 if (info
->version
!= IOMMU_CACHE_INVALIDATE_INFO_VERSION_1
)
2042 mask
= (1 << IOMMU_CACHE_INV_TYPE_NR
) - 1;
2043 if (info
->cache
& ~mask
)
2046 if (info
->granularity
>= IOMMU_INV_GRANU_NR
)
2049 switch (info
->granularity
) {
2050 case IOMMU_INV_GRANU_ADDR
:
2051 if (info
->cache
& IOMMU_CACHE_INV_TYPE_PASID
)
2054 mask
= IOMMU_INV_ADDR_FLAGS_PASID
|
2055 IOMMU_INV_ADDR_FLAGS_ARCHID
|
2056 IOMMU_INV_ADDR_FLAGS_LEAF
;
2058 if (info
->granu
.addr_info
.flags
& ~mask
)
2061 case IOMMU_INV_GRANU_PASID
:
2062 mask
= IOMMU_INV_PASID_FLAGS_PASID
|
2063 IOMMU_INV_PASID_FLAGS_ARCHID
;
2064 if (info
->granu
.pasid_info
.flags
& ~mask
)
2068 case IOMMU_INV_GRANU_DOMAIN
:
2069 if (info
->cache
& IOMMU_CACHE_INV_TYPE_DEV_IOTLB
)
2076 /* Check reserved padding fields */
2077 for (i
= 0; i
< sizeof(info
->padding
); i
++) {
2078 if (info
->padding
[i
])
2085 int iommu_uapi_cache_invalidate(struct iommu_domain
*domain
, struct device
*dev
,
2088 struct iommu_cache_invalidate_info inv_info
= { 0 };
2092 if (unlikely(!domain
->ops
->cache_invalidate
))
2096 * No new spaces can be added before the variable sized union, the
2097 * minimum size is the offset to the union.
2099 minsz
= offsetof(struct iommu_cache_invalidate_info
, granu
);
2101 /* Copy minsz from user to get flags and argsz */
2102 if (copy_from_user(&inv_info
, uinfo
, minsz
))
2105 /* Fields before the variable size union are mandatory */
2106 if (inv_info
.argsz
< minsz
)
2109 /* PASID and address granu require additional info beyond minsz */
2110 if (inv_info
.granularity
== IOMMU_INV_GRANU_PASID
&&
2111 inv_info
.argsz
< offsetofend(struct iommu_cache_invalidate_info
, granu
.pasid_info
))
2114 if (inv_info
.granularity
== IOMMU_INV_GRANU_ADDR
&&
2115 inv_info
.argsz
< offsetofend(struct iommu_cache_invalidate_info
, granu
.addr_info
))
2119 * User might be using a newer UAPI header which has a larger data
2120 * size, we shall support the existing flags within the current
2121 * size. Copy the remaining user data _after_ minsz but not more
2122 * than the current kernel supported size.
2124 if (copy_from_user((void *)&inv_info
+ minsz
, uinfo
+ minsz
,
2125 min_t(u32
, inv_info
.argsz
, sizeof(inv_info
)) - minsz
))
2128 /* Now the argsz is validated, check the content */
2129 ret
= iommu_check_cache_invl_data(&inv_info
);
2133 return domain
->ops
->cache_invalidate(domain
, dev
, &inv_info
);
2135 EXPORT_SYMBOL_GPL(iommu_uapi_cache_invalidate
);
2137 static int iommu_check_bind_data(struct iommu_gpasid_bind_data
*data
)
2142 if (data
->version
!= IOMMU_GPASID_BIND_VERSION_1
)
2145 /* Check the range of supported formats */
2146 if (data
->format
>= IOMMU_PASID_FORMAT_LAST
)
2149 /* Check all flags */
2150 mask
= IOMMU_SVA_GPASID_VAL
;
2151 if (data
->flags
& ~mask
)
2154 /* Check reserved padding fields */
2155 for (i
= 0; i
< sizeof(data
->padding
); i
++) {
2156 if (data
->padding
[i
])
2163 static int iommu_sva_prepare_bind_data(void __user
*udata
,
2164 struct iommu_gpasid_bind_data
*data
)
2169 * No new spaces can be added before the variable sized union, the
2170 * minimum size is the offset to the union.
2172 minsz
= offsetof(struct iommu_gpasid_bind_data
, vendor
);
2174 /* Copy minsz from user to get flags and argsz */
2175 if (copy_from_user(data
, udata
, minsz
))
2178 /* Fields before the variable size union are mandatory */
2179 if (data
->argsz
< minsz
)
2182 * User might be using a newer UAPI header, we shall let IOMMU vendor
2183 * driver decide on what size it needs. Since the guest PASID bind data
2184 * can be vendor specific, larger argsz could be the result of extension
2185 * for one vendor but it should not affect another vendor.
2186 * Copy the remaining user data _after_ minsz
2188 if (copy_from_user((void *)data
+ minsz
, udata
+ minsz
,
2189 min_t(u32
, data
->argsz
, sizeof(*data
)) - minsz
))
2192 return iommu_check_bind_data(data
);
2195 int iommu_uapi_sva_bind_gpasid(struct iommu_domain
*domain
, struct device
*dev
,
2198 struct iommu_gpasid_bind_data data
= { 0 };
2201 if (unlikely(!domain
->ops
->sva_bind_gpasid
))
2204 ret
= iommu_sva_prepare_bind_data(udata
, &data
);
2208 return domain
->ops
->sva_bind_gpasid(domain
, dev
, &data
);
2210 EXPORT_SYMBOL_GPL(iommu_uapi_sva_bind_gpasid
);
2212 int iommu_sva_unbind_gpasid(struct iommu_domain
*domain
, struct device
*dev
,
2215 if (unlikely(!domain
->ops
->sva_unbind_gpasid
))
2218 return domain
->ops
->sva_unbind_gpasid(dev
, pasid
);
2220 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid
);
2222 int iommu_uapi_sva_unbind_gpasid(struct iommu_domain
*domain
, struct device
*dev
,
2225 struct iommu_gpasid_bind_data data
= { 0 };
2228 if (unlikely(!domain
->ops
->sva_bind_gpasid
))
2231 ret
= iommu_sva_prepare_bind_data(udata
, &data
);
2235 return iommu_sva_unbind_gpasid(domain
, dev
, data
.hpasid
);
2237 EXPORT_SYMBOL_GPL(iommu_uapi_sva_unbind_gpasid
);
2239 static void __iommu_detach_device(struct iommu_domain
*domain
,
2242 if (iommu_is_attach_deferred(domain
, dev
))
2245 if (unlikely(domain
->ops
->detach_dev
== NULL
))
2248 domain
->ops
->detach_dev(domain
, dev
);
2249 trace_detach_device_from_domain(dev
);
2252 void iommu_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
2254 struct iommu_group
*group
;
2256 group
= iommu_group_get(dev
);
2260 mutex_lock(&group
->mutex
);
2261 if (iommu_group_device_count(group
) != 1) {
2266 __iommu_detach_group(domain
, group
);
2269 mutex_unlock(&group
->mutex
);
2270 iommu_group_put(group
);
2272 EXPORT_SYMBOL_GPL(iommu_detach_device
);
2274 struct iommu_domain
*iommu_get_domain_for_dev(struct device
*dev
)
2276 struct iommu_domain
*domain
;
2277 struct iommu_group
*group
;
2279 group
= iommu_group_get(dev
);
2283 domain
= group
->domain
;
2285 iommu_group_put(group
);
2289 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev
);
2292 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2293 * guarantees that the group and its default domain are valid and correct.
2295 struct iommu_domain
*iommu_get_dma_domain(struct device
*dev
)
2297 return dev
->iommu_group
->default_domain
;
2301 * IOMMU groups are really the natural working unit of the IOMMU, but
2302 * the IOMMU API works on domains and devices. Bridge that gap by
2303 * iterating over the devices in a group. Ideally we'd have a single
2304 * device which represents the requestor ID of the group, but we also
2305 * allow IOMMU drivers to create policy defined minimum sets, where
2306 * the physical hardware may be able to distiguish members, but we
2307 * wish to group them at a higher level (ex. untrusted multi-function
2308 * PCI devices). Thus we attach each device.
2310 static int iommu_group_do_attach_device(struct device
*dev
, void *data
)
2312 struct iommu_domain
*domain
= data
;
2314 return __iommu_attach_device(domain
, dev
);
2317 static int __iommu_attach_group(struct iommu_domain
*domain
,
2318 struct iommu_group
*group
)
2322 if (group
->default_domain
&& group
->domain
!= group
->default_domain
)
2325 ret
= __iommu_group_for_each_dev(group
, domain
,
2326 iommu_group_do_attach_device
);
2328 group
->domain
= domain
;
2333 int iommu_attach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
2337 mutex_lock(&group
->mutex
);
2338 ret
= __iommu_attach_group(domain
, group
);
2339 mutex_unlock(&group
->mutex
);
2343 EXPORT_SYMBOL_GPL(iommu_attach_group
);
2345 static int iommu_group_do_detach_device(struct device
*dev
, void *data
)
2347 struct iommu_domain
*domain
= data
;
2349 __iommu_detach_device(domain
, dev
);
2354 static void __iommu_detach_group(struct iommu_domain
*domain
,
2355 struct iommu_group
*group
)
2359 if (!group
->default_domain
) {
2360 __iommu_group_for_each_dev(group
, domain
,
2361 iommu_group_do_detach_device
);
2362 group
->domain
= NULL
;
2366 if (group
->domain
== group
->default_domain
)
2369 /* Detach by re-attaching to the default domain */
2370 ret
= __iommu_group_for_each_dev(group
, group
->default_domain
,
2371 iommu_group_do_attach_device
);
2375 group
->domain
= group
->default_domain
;
2378 void iommu_detach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
2380 mutex_lock(&group
->mutex
);
2381 __iommu_detach_group(domain
, group
);
2382 mutex_unlock(&group
->mutex
);
2384 EXPORT_SYMBOL_GPL(iommu_detach_group
);
2386 phys_addr_t
iommu_iova_to_phys(struct iommu_domain
*domain
, dma_addr_t iova
)
2388 if (domain
->type
== IOMMU_DOMAIN_IDENTITY
)
2391 if (domain
->type
== IOMMU_DOMAIN_BLOCKED
)
2394 return domain
->ops
->iova_to_phys(domain
, iova
);
2396 EXPORT_SYMBOL_GPL(iommu_iova_to_phys
);
2398 static size_t iommu_pgsize(struct iommu_domain
*domain
, unsigned long iova
,
2399 phys_addr_t paddr
, size_t size
, size_t *count
)
2401 unsigned int pgsize_idx
, pgsize_idx_next
;
2402 unsigned long pgsizes
;
2403 size_t offset
, pgsize
, pgsize_next
;
2404 unsigned long addr_merge
= paddr
| iova
;
2406 /* Page sizes supported by the hardware and small enough for @size */
2407 pgsizes
= domain
->pgsize_bitmap
& GENMASK(__fls(size
), 0);
2409 /* Constrain the page sizes further based on the maximum alignment */
2410 if (likely(addr_merge
))
2411 pgsizes
&= GENMASK(__ffs(addr_merge
), 0);
2413 /* Make sure we have at least one suitable page size */
2416 /* Pick the biggest page size remaining */
2417 pgsize_idx
= __fls(pgsizes
);
2418 pgsize
= BIT(pgsize_idx
);
2422 /* Find the next biggest support page size, if it exists */
2423 pgsizes
= domain
->pgsize_bitmap
& ~GENMASK(pgsize_idx
, 0);
2427 pgsize_idx_next
= __ffs(pgsizes
);
2428 pgsize_next
= BIT(pgsize_idx_next
);
2431 * There's no point trying a bigger page size unless the virtual
2432 * and physical addresses are similarly offset within the larger page.
2434 if ((iova
^ paddr
) & (pgsize_next
- 1))
2437 /* Calculate the offset to the next page size alignment boundary */
2438 offset
= pgsize_next
- (addr_merge
& (pgsize_next
- 1));
2441 * If size is big enough to accommodate the larger page, reduce
2442 * the number of smaller pages.
2444 if (offset
+ pgsize_next
<= size
)
2448 *count
= size
>> pgsize_idx
;
2452 static int __iommu_map_pages(struct iommu_domain
*domain
, unsigned long iova
,
2453 phys_addr_t paddr
, size_t size
, int prot
,
2454 gfp_t gfp
, size_t *mapped
)
2456 const struct iommu_ops
*ops
= domain
->ops
;
2457 size_t pgsize
, count
;
2460 pgsize
= iommu_pgsize(domain
, iova
, paddr
, size
, &count
);
2462 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx count %zu\n",
2463 iova
, &paddr
, pgsize
, count
);
2465 if (ops
->map_pages
) {
2466 ret
= ops
->map_pages(domain
, iova
, paddr
, pgsize
, count
, prot
,
2469 ret
= ops
->map(domain
, iova
, paddr
, pgsize
, prot
, gfp
);
2470 *mapped
= ret
? 0 : pgsize
;
2476 static int __iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
2477 phys_addr_t paddr
, size_t size
, int prot
, gfp_t gfp
)
2479 const struct iommu_ops
*ops
= domain
->ops
;
2480 unsigned long orig_iova
= iova
;
2481 unsigned int min_pagesz
;
2482 size_t orig_size
= size
;
2483 phys_addr_t orig_paddr
= paddr
;
2486 if (unlikely(!(ops
->map
|| ops
->map_pages
) ||
2487 domain
->pgsize_bitmap
== 0UL))
2490 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
2493 /* find out the minimum page size supported */
2494 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
2497 * both the virtual address and the physical one, as well as
2498 * the size of the mapping, must be aligned (at least) to the
2499 * size of the smallest page supported by the hardware
2501 if (!IS_ALIGNED(iova
| paddr
| size
, min_pagesz
)) {
2502 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2503 iova
, &paddr
, size
, min_pagesz
);
2507 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova
, &paddr
, size
);
2512 ret
= __iommu_map_pages(domain
, iova
, paddr
, size
, prot
, gfp
,
2515 * Some pages may have been mapped, even if an error occurred,
2516 * so we should account for those so they can be unmapped.
2527 /* unroll mapping in case something went wrong */
2529 iommu_unmap(domain
, orig_iova
, orig_size
- size
);
2531 trace_map(orig_iova
, orig_paddr
, orig_size
);
2536 static int _iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
2537 phys_addr_t paddr
, size_t size
, int prot
, gfp_t gfp
)
2539 const struct iommu_ops
*ops
= domain
->ops
;
2542 ret
= __iommu_map(domain
, iova
, paddr
, size
, prot
, gfp
);
2543 if (ret
== 0 && ops
->iotlb_sync_map
)
2544 ops
->iotlb_sync_map(domain
, iova
, size
);
2549 int iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
2550 phys_addr_t paddr
, size_t size
, int prot
)
2553 return _iommu_map(domain
, iova
, paddr
, size
, prot
, GFP_KERNEL
);
2555 EXPORT_SYMBOL_GPL(iommu_map
);
2557 int iommu_map_atomic(struct iommu_domain
*domain
, unsigned long iova
,
2558 phys_addr_t paddr
, size_t size
, int prot
)
2560 return _iommu_map(domain
, iova
, paddr
, size
, prot
, GFP_ATOMIC
);
2562 EXPORT_SYMBOL_GPL(iommu_map_atomic
);
2564 static size_t __iommu_unmap_pages(struct iommu_domain
*domain
,
2565 unsigned long iova
, size_t size
,
2566 struct iommu_iotlb_gather
*iotlb_gather
)
2568 const struct iommu_ops
*ops
= domain
->ops
;
2569 size_t pgsize
, count
;
2571 pgsize
= iommu_pgsize(domain
, iova
, iova
, size
, &count
);
2572 return ops
->unmap_pages
?
2573 ops
->unmap_pages(domain
, iova
, pgsize
, count
, iotlb_gather
) :
2574 ops
->unmap(domain
, iova
, pgsize
, iotlb_gather
);
2577 static size_t __iommu_unmap(struct iommu_domain
*domain
,
2578 unsigned long iova
, size_t size
,
2579 struct iommu_iotlb_gather
*iotlb_gather
)
2581 const struct iommu_ops
*ops
= domain
->ops
;
2582 size_t unmapped_page
, unmapped
= 0;
2583 unsigned long orig_iova
= iova
;
2584 unsigned int min_pagesz
;
2586 if (unlikely(!(ops
->unmap
|| ops
->unmap_pages
) ||
2587 domain
->pgsize_bitmap
== 0UL))
2590 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
2593 /* find out the minimum page size supported */
2594 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
2597 * The virtual address, as well as the size of the mapping, must be
2598 * aligned (at least) to the size of the smallest page supported
2601 if (!IS_ALIGNED(iova
| size
, min_pagesz
)) {
2602 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2603 iova
, size
, min_pagesz
);
2607 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova
, size
);
2610 * Keep iterating until we either unmap 'size' bytes (or more)
2611 * or we hit an area that isn't mapped.
2613 while (unmapped
< size
) {
2614 unmapped_page
= __iommu_unmap_pages(domain
, iova
,
2620 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2621 iova
, unmapped_page
);
2623 iova
+= unmapped_page
;
2624 unmapped
+= unmapped_page
;
2627 trace_unmap(orig_iova
, size
, unmapped
);
2631 size_t iommu_unmap(struct iommu_domain
*domain
,
2632 unsigned long iova
, size_t size
)
2634 struct iommu_iotlb_gather iotlb_gather
;
2637 iommu_iotlb_gather_init(&iotlb_gather
);
2638 ret
= __iommu_unmap(domain
, iova
, size
, &iotlb_gather
);
2639 iommu_iotlb_sync(domain
, &iotlb_gather
);
2643 EXPORT_SYMBOL_GPL(iommu_unmap
);
2645 size_t iommu_unmap_fast(struct iommu_domain
*domain
,
2646 unsigned long iova
, size_t size
,
2647 struct iommu_iotlb_gather
*iotlb_gather
)
2649 return __iommu_unmap(domain
, iova
, size
, iotlb_gather
);
2651 EXPORT_SYMBOL_GPL(iommu_unmap_fast
);
2653 static ssize_t
__iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
2654 struct scatterlist
*sg
, unsigned int nents
, int prot
,
2657 const struct iommu_ops
*ops
= domain
->ops
;
2658 size_t len
= 0, mapped
= 0;
2663 while (i
<= nents
) {
2664 phys_addr_t s_phys
= sg_phys(sg
);
2666 if (len
&& s_phys
!= start
+ len
) {
2667 ret
= __iommu_map(domain
, iova
+ mapped
, start
,
2688 if (ops
->iotlb_sync_map
)
2689 ops
->iotlb_sync_map(domain
, iova
, mapped
);
2693 /* undo mappings already done */
2694 iommu_unmap(domain
, iova
, mapped
);
2699 ssize_t
iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
2700 struct scatterlist
*sg
, unsigned int nents
, int prot
)
2703 return __iommu_map_sg(domain
, iova
, sg
, nents
, prot
, GFP_KERNEL
);
2705 EXPORT_SYMBOL_GPL(iommu_map_sg
);
2707 ssize_t
iommu_map_sg_atomic(struct iommu_domain
*domain
, unsigned long iova
,
2708 struct scatterlist
*sg
, unsigned int nents
, int prot
)
2710 return __iommu_map_sg(domain
, iova
, sg
, nents
, prot
, GFP_ATOMIC
);
2714 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2715 * @domain: the iommu domain where the fault has happened
2716 * @dev: the device where the fault has happened
2717 * @iova: the faulting address
2718 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2720 * This function should be called by the low-level IOMMU implementations
2721 * whenever IOMMU faults happen, to allow high-level users, that are
2722 * interested in such events, to know about them.
2724 * This event may be useful for several possible use cases:
2725 * - mere logging of the event
2726 * - dynamic TLB/PTE loading
2727 * - if restarting of the faulting device is required
2729 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2730 * PTE/TLB loading will one day be supported, implementations will be able
2731 * to tell whether it succeeded or not according to this return value).
2733 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2734 * (though fault handlers can also return -ENOSYS, in case they want to
2735 * elicit the default behavior of the IOMMU drivers).
2737 int report_iommu_fault(struct iommu_domain
*domain
, struct device
*dev
,
2738 unsigned long iova
, int flags
)
2743 * if upper layers showed interest and installed a fault handler,
2746 if (domain
->handler
)
2747 ret
= domain
->handler(domain
, dev
, iova
, flags
,
2748 domain
->handler_token
);
2750 trace_io_page_fault(dev
, iova
, flags
);
2753 EXPORT_SYMBOL_GPL(report_iommu_fault
);
2755 static int __init
iommu_init(void)
2757 iommu_group_kset
= kset_create_and_add("iommu_groups",
2759 BUG_ON(!iommu_group_kset
);
2761 iommu_debugfs_setup();
2765 core_initcall(iommu_init
);
2767 int iommu_enable_nesting(struct iommu_domain
*domain
)
2769 if (domain
->type
!= IOMMU_DOMAIN_UNMANAGED
)
2771 if (!domain
->ops
->enable_nesting
)
2773 return domain
->ops
->enable_nesting(domain
);
2775 EXPORT_SYMBOL_GPL(iommu_enable_nesting
);
2777 int iommu_set_pgtable_quirks(struct iommu_domain
*domain
,
2778 unsigned long quirk
)
2780 if (domain
->type
!= IOMMU_DOMAIN_UNMANAGED
)
2782 if (!domain
->ops
->set_pgtable_quirks
)
2784 return domain
->ops
->set_pgtable_quirks(domain
, quirk
);
2786 EXPORT_SYMBOL_GPL(iommu_set_pgtable_quirks
);
2788 void iommu_get_resv_regions(struct device
*dev
, struct list_head
*list
)
2790 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2792 if (ops
&& ops
->get_resv_regions
)
2793 ops
->get_resv_regions(dev
, list
);
2796 void iommu_put_resv_regions(struct device
*dev
, struct list_head
*list
)
2798 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
2800 if (ops
&& ops
->put_resv_regions
)
2801 ops
->put_resv_regions(dev
, list
);
2805 * generic_iommu_put_resv_regions - Reserved region driver helper
2806 * @dev: device for which to free reserved regions
2807 * @list: reserved region list for device
2809 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2810 * for simple reservations. Memory allocated for each reserved region will be
2811 * freed. If an IOMMU driver allocates additional resources per region, it is
2812 * going to have to implement a custom callback.
2814 void generic_iommu_put_resv_regions(struct device
*dev
, struct list_head
*list
)
2816 struct iommu_resv_region
*entry
, *next
;
2818 list_for_each_entry_safe(entry
, next
, list
, list
)
2821 EXPORT_SYMBOL(generic_iommu_put_resv_regions
);
2823 struct iommu_resv_region
*iommu_alloc_resv_region(phys_addr_t start
,
2824 size_t length
, int prot
,
2825 enum iommu_resv_type type
)
2827 struct iommu_resv_region
*region
;
2829 region
= kzalloc(sizeof(*region
), GFP_KERNEL
);
2833 INIT_LIST_HEAD(®ion
->list
);
2834 region
->start
= start
;
2835 region
->length
= length
;
2836 region
->prot
= prot
;
2837 region
->type
= type
;
2840 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region
);
2842 void iommu_set_default_passthrough(bool cmd_line
)
2845 iommu_cmd_line
|= IOMMU_CMD_LINE_DMA_API
;
2846 iommu_def_domain_type
= IOMMU_DOMAIN_IDENTITY
;
2849 void iommu_set_default_translated(bool cmd_line
)
2852 iommu_cmd_line
|= IOMMU_CMD_LINE_DMA_API
;
2853 iommu_def_domain_type
= IOMMU_DOMAIN_DMA
;
2856 bool iommu_default_passthrough(void)
2858 return iommu_def_domain_type
== IOMMU_DOMAIN_IDENTITY
;
2860 EXPORT_SYMBOL_GPL(iommu_default_passthrough
);
2862 const struct iommu_ops
*iommu_ops_from_fwnode(struct fwnode_handle
*fwnode
)
2864 const struct iommu_ops
*ops
= NULL
;
2865 struct iommu_device
*iommu
;
2867 spin_lock(&iommu_device_lock
);
2868 list_for_each_entry(iommu
, &iommu_device_list
, list
)
2869 if (iommu
->fwnode
== fwnode
) {
2873 spin_unlock(&iommu_device_lock
);
2877 int iommu_fwspec_init(struct device
*dev
, struct fwnode_handle
*iommu_fwnode
,
2878 const struct iommu_ops
*ops
)
2880 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2883 return ops
== fwspec
->ops
? 0 : -EINVAL
;
2885 if (!dev_iommu_get(dev
))
2888 /* Preallocate for the overwhelmingly common case of 1 ID */
2889 fwspec
= kzalloc(struct_size(fwspec
, ids
, 1), GFP_KERNEL
);
2893 of_node_get(to_of_node(iommu_fwnode
));
2894 fwspec
->iommu_fwnode
= iommu_fwnode
;
2896 dev_iommu_fwspec_set(dev
, fwspec
);
2899 EXPORT_SYMBOL_GPL(iommu_fwspec_init
);
2901 void iommu_fwspec_free(struct device
*dev
)
2903 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2906 fwnode_handle_put(fwspec
->iommu_fwnode
);
2908 dev_iommu_fwspec_set(dev
, NULL
);
2911 EXPORT_SYMBOL_GPL(iommu_fwspec_free
);
2913 int iommu_fwspec_add_ids(struct device
*dev
, u32
*ids
, int num_ids
)
2915 struct iommu_fwspec
*fwspec
= dev_iommu_fwspec_get(dev
);
2921 new_num
= fwspec
->num_ids
+ num_ids
;
2923 fwspec
= krealloc(fwspec
, struct_size(fwspec
, ids
, new_num
),
2928 dev_iommu_fwspec_set(dev
, fwspec
);
2931 for (i
= 0; i
< num_ids
; i
++)
2932 fwspec
->ids
[fwspec
->num_ids
+ i
] = ids
[i
];
2934 fwspec
->num_ids
= new_num
;
2937 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids
);
2940 * Per device IOMMU features.
2942 int iommu_dev_enable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2944 if (dev
->iommu
&& dev
->iommu
->iommu_dev
) {
2945 const struct iommu_ops
*ops
= dev
->iommu
->iommu_dev
->ops
;
2947 if (ops
->dev_enable_feat
)
2948 return ops
->dev_enable_feat(dev
, feat
);
2953 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature
);
2956 * The device drivers should do the necessary cleanups before calling this.
2957 * For example, before disabling the aux-domain feature, the device driver
2958 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2960 int iommu_dev_disable_feature(struct device
*dev
, enum iommu_dev_features feat
)
2962 if (dev
->iommu
&& dev
->iommu
->iommu_dev
) {
2963 const struct iommu_ops
*ops
= dev
->iommu
->iommu_dev
->ops
;
2965 if (ops
->dev_disable_feat
)
2966 return ops
->dev_disable_feat(dev
, feat
);
2971 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature
);
2973 bool iommu_dev_feature_enabled(struct device
*dev
, enum iommu_dev_features feat
)
2975 if (dev
->iommu
&& dev
->iommu
->iommu_dev
) {
2976 const struct iommu_ops
*ops
= dev
->iommu
->iommu_dev
->ops
;
2978 if (ops
->dev_feat_enabled
)
2979 return ops
->dev_feat_enabled(dev
, feat
);
2984 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled
);
2987 * Aux-domain specific attach/detach.
2989 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2990 * true. Also, as long as domains are attached to a device through this
2991 * interface, any tries to call iommu_attach_device() should fail
2992 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2993 * This should make us safe against a device being attached to a guest as a
2994 * whole while there are still pasid users on it (aux and sva).
2996 int iommu_aux_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
3000 if (domain
->ops
->aux_attach_dev
)
3001 ret
= domain
->ops
->aux_attach_dev(domain
, dev
);
3004 trace_attach_device_to_domain(dev
);
3008 EXPORT_SYMBOL_GPL(iommu_aux_attach_device
);
3010 void iommu_aux_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
3012 if (domain
->ops
->aux_detach_dev
) {
3013 domain
->ops
->aux_detach_dev(domain
, dev
);
3014 trace_detach_device_from_domain(dev
);
3017 EXPORT_SYMBOL_GPL(iommu_aux_detach_device
);
3019 int iommu_aux_get_pasid(struct iommu_domain
*domain
, struct device
*dev
)
3023 if (domain
->ops
->aux_get_pasid
)
3024 ret
= domain
->ops
->aux_get_pasid(domain
, dev
);
3028 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid
);
3031 * iommu_sva_bind_device() - Bind a process address space to a device
3033 * @mm: the mm to bind, caller must hold a reference to it
3035 * Create a bond between device and address space, allowing the device to access
3036 * the mm using the returned PASID. If a bond already exists between @device and
3037 * @mm, it is returned and an additional reference is taken. Caller must call
3038 * iommu_sva_unbind_device() to release each reference.
3040 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
3041 * initialize the required SVA features.
3043 * On error, returns an ERR_PTR value.
3046 iommu_sva_bind_device(struct device
*dev
, struct mm_struct
*mm
, void *drvdata
)
3048 struct iommu_group
*group
;
3049 struct iommu_sva
*handle
= ERR_PTR(-EINVAL
);
3050 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
3052 if (!ops
|| !ops
->sva_bind
)
3053 return ERR_PTR(-ENODEV
);
3055 group
= iommu_group_get(dev
);
3057 return ERR_PTR(-ENODEV
);
3059 /* Ensure device count and domain don't change while we're binding */
3060 mutex_lock(&group
->mutex
);
3063 * To keep things simple, SVA currently doesn't support IOMMU groups
3064 * with more than one device. Existing SVA-capable systems are not
3065 * affected by the problems that required IOMMU groups (lack of ACS
3066 * isolation, device ID aliasing and other hardware issues).
3068 if (iommu_group_device_count(group
) != 1)
3071 handle
= ops
->sva_bind(dev
, mm
, drvdata
);
3074 mutex_unlock(&group
->mutex
);
3075 iommu_group_put(group
);
3079 EXPORT_SYMBOL_GPL(iommu_sva_bind_device
);
3082 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
3083 * @handle: the handle returned by iommu_sva_bind_device()
3085 * Put reference to a bond between device and address space. The device should
3086 * not be issuing any more transaction for this PASID. All outstanding page
3087 * requests for this PASID must have been flushed to the IOMMU.
3089 void iommu_sva_unbind_device(struct iommu_sva
*handle
)
3091 struct iommu_group
*group
;
3092 struct device
*dev
= handle
->dev
;
3093 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
3095 if (!ops
|| !ops
->sva_unbind
)
3098 group
= iommu_group_get(dev
);
3102 mutex_lock(&group
->mutex
);
3103 ops
->sva_unbind(handle
);
3104 mutex_unlock(&group
->mutex
);
3106 iommu_group_put(group
);
3108 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device
);
3110 u32
iommu_sva_get_pasid(struct iommu_sva
*handle
)
3112 const struct iommu_ops
*ops
= handle
->dev
->bus
->iommu_ops
;
3114 if (!ops
|| !ops
->sva_get_pasid
)
3115 return IOMMU_PASID_INVALID
;
3117 return ops
->sva_get_pasid(handle
);
3119 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid
);
3122 * Changes the default domain of an iommu group that has *only* one device
3124 * @group: The group for which the default domain should be changed
3125 * @prev_dev: The device in the group (this is used to make sure that the device
3126 * hasn't changed after the caller has called this function)
3127 * @type: The type of the new default domain that gets associated with the group
3129 * Returns 0 on success and error code on failure
3132 * 1. Presently, this function is called only when user requests to change the
3133 * group's default domain type through /sys/kernel/iommu_groups/<grp_id>/type
3134 * Please take a closer look if intended to use for other purposes.
3136 static int iommu_change_dev_def_domain(struct iommu_group
*group
,
3137 struct device
*prev_dev
, int type
)
3139 struct iommu_domain
*prev_dom
;
3140 struct group_device
*grp_dev
;
3141 int ret
, dev_def_dom
;
3144 mutex_lock(&group
->mutex
);
3146 if (group
->default_domain
!= group
->domain
) {
3147 dev_err_ratelimited(prev_dev
, "Group not assigned to default domain\n");
3153 * iommu group wasn't locked while acquiring device lock in
3154 * iommu_group_store_type(). So, make sure that the device count hasn't
3155 * changed while acquiring device lock.
3157 * Changing default domain of an iommu group with two or more devices
3158 * isn't supported because there could be a potential deadlock. Consider
3159 * the following scenario. T1 is trying to acquire device locks of all
3160 * the devices in the group and before it could acquire all of them,
3161 * there could be another thread T2 (from different sub-system and use
3162 * case) that has already acquired some of the device locks and might be
3163 * waiting for T1 to release other device locks.
3165 if (iommu_group_device_count(group
) != 1) {
3166 dev_err_ratelimited(prev_dev
, "Cannot change default domain: Group has more than one device\n");
3171 /* Since group has only one device */
3172 grp_dev
= list_first_entry(&group
->devices
, struct group_device
, list
);
3175 if (prev_dev
!= dev
) {
3176 dev_err_ratelimited(prev_dev
, "Cannot change default domain: Device has been changed\n");
3181 prev_dom
= group
->default_domain
;
3187 dev_def_dom
= iommu_get_def_domain_type(dev
);
3190 * If the user hasn't requested any specific type of domain and
3191 * if the device supports both the domains, then default to the
3192 * domain the device was booted with
3194 type
= dev_def_dom
? : iommu_def_domain_type
;
3195 } else if (dev_def_dom
&& type
!= dev_def_dom
) {
3196 dev_err_ratelimited(prev_dev
, "Device cannot be in %s domain\n",
3197 iommu_domain_type_str(type
));
3203 * Switch to a new domain only if the requested domain type is different
3204 * from the existing default domain type
3206 if (prev_dom
->type
== type
) {
3211 /* We can bring up a flush queue without tearing down the domain */
3212 if (type
== IOMMU_DOMAIN_DMA_FQ
&& prev_dom
->type
== IOMMU_DOMAIN_DMA
) {
3213 ret
= iommu_dma_init_fq(prev_dom
);
3215 prev_dom
->type
= IOMMU_DOMAIN_DMA_FQ
;
3219 /* Sets group->default_domain to the newly allocated domain */
3220 ret
= iommu_group_alloc_default_domain(dev
->bus
, group
, type
);
3224 ret
= iommu_create_device_direct_mappings(group
, dev
);
3226 goto free_new_domain
;
3228 ret
= __iommu_attach_device(group
->default_domain
, dev
);
3230 goto free_new_domain
;
3232 group
->domain
= group
->default_domain
;
3235 * Release the mutex here because ops->probe_finalize() call-back of
3236 * some vendor IOMMU drivers calls arm_iommu_attach_device() which
3237 * in-turn might call back into IOMMU core code, where it tries to take
3238 * group->mutex, resulting in a deadlock.
3240 mutex_unlock(&group
->mutex
);
3242 /* Make sure dma_ops is appropriatley set */
3243 iommu_group_do_probe_finalize(dev
, group
->default_domain
);
3244 iommu_domain_free(prev_dom
);
3248 iommu_domain_free(group
->default_domain
);
3249 group
->default_domain
= prev_dom
;
3250 group
->domain
= prev_dom
;
3253 mutex_unlock(&group
->mutex
);
3259 * Changing the default domain through sysfs requires the users to unbind the
3260 * drivers from the devices in the iommu group, except for a DMA -> DMA-FQ
3261 * transition. Return failure if this isn't met.
3263 * We need to consider the race between this and the device release path.
3264 * device_lock(dev) is used here to guarantee that the device release path
3265 * will not be entered at the same time.
3267 static ssize_t
iommu_group_store_type(struct iommu_group
*group
,
3268 const char *buf
, size_t count
)
3270 struct group_device
*grp_dev
;
3274 if (!capable(CAP_SYS_ADMIN
) || !capable(CAP_SYS_RAWIO
))
3277 if (WARN_ON(!group
))
3280 if (sysfs_streq(buf
, "identity"))
3281 req_type
= IOMMU_DOMAIN_IDENTITY
;
3282 else if (sysfs_streq(buf
, "DMA"))
3283 req_type
= IOMMU_DOMAIN_DMA
;
3284 else if (sysfs_streq(buf
, "DMA-FQ"))
3285 req_type
= IOMMU_DOMAIN_DMA_FQ
;
3286 else if (sysfs_streq(buf
, "auto"))
3292 * Lock/Unlock the group mutex here before device lock to
3293 * 1. Make sure that the iommu group has only one device (this is a
3294 * prerequisite for step 2)
3295 * 2. Get struct *dev which is needed to lock device
3297 mutex_lock(&group
->mutex
);
3298 if (iommu_group_device_count(group
) != 1) {
3299 mutex_unlock(&group
->mutex
);
3300 pr_err_ratelimited("Cannot change default domain: Group has more than one device\n");
3304 /* Since group has only one device */
3305 grp_dev
= list_first_entry(&group
->devices
, struct group_device
, list
);
3310 * Don't hold the group mutex because taking group mutex first and then
3311 * the device lock could potentially cause a deadlock as below. Assume
3312 * two threads T1 and T2. T1 is trying to change default domain of an
3313 * iommu group and T2 is trying to hot unplug a device or release [1] VF
3314 * of a PCIe device which is in the same iommu group. T1 takes group
3315 * mutex and before it could take device lock assume T2 has taken device
3316 * lock and is yet to take group mutex. Now, both the threads will be
3317 * waiting for the other thread to release lock. Below, lock order was
3320 * mutex_lock(&group->mutex);
3321 * iommu_change_dev_def_domain();
3322 * mutex_unlock(&group->mutex);
3323 * device_unlock(dev);
3325 * [1] Typical device release path
3326 * device_lock() from device/driver core code
3328 * -> iommu_bus_notifier()
3329 * -> iommu_release_device()
3330 * -> ops->release_device() vendor driver calls back iommu core code
3331 * -> mutex_lock() from iommu core code
3333 mutex_unlock(&group
->mutex
);
3335 /* Check if the device in the group still has a driver bound to it */
3337 if (device_is_bound(dev
) && !(req_type
== IOMMU_DOMAIN_DMA_FQ
&&
3338 group
->default_domain
->type
== IOMMU_DOMAIN_DMA
)) {
3339 pr_err_ratelimited("Device is still bound to driver\n");
3344 ret
= iommu_change_dev_def_domain(group
, dev
, req_type
);