2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published
7 * by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #define pr_fmt(fmt) "iommu: " fmt
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/module.h>
26 #include <linux/slab.h>
27 #include <linux/errno.h>
28 #include <linux/iommu.h>
29 #include <linux/idr.h>
30 #include <linux/notifier.h>
31 #include <linux/err.h>
32 #include <linux/pci.h>
33 #include <linux/bitops.h>
34 #include <linux/property.h>
35 #include <trace/events/iommu.h>
37 static struct kset
*iommu_group_kset
;
38 static DEFINE_IDA(iommu_group_ida
);
40 struct iommu_callback_data
{
41 const struct iommu_ops
*ops
;
46 struct kobject
*devices_kobj
;
47 struct list_head devices
;
49 struct blocking_notifier_head notifier
;
51 void (*iommu_data_release
)(void *iommu_data
);
54 struct iommu_domain
*default_domain
;
55 struct iommu_domain
*domain
;
59 struct list_head list
;
64 struct iommu_group_attribute
{
65 struct attribute attr
;
66 ssize_t (*show
)(struct iommu_group
*group
, char *buf
);
67 ssize_t (*store
)(struct iommu_group
*group
,
68 const char *buf
, size_t count
);
71 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
72 struct iommu_group_attribute iommu_group_attr_##_name = \
73 __ATTR(_name, _mode, _show, _store)
75 #define to_iommu_group_attr(_attr) \
76 container_of(_attr, struct iommu_group_attribute, attr)
77 #define to_iommu_group(_kobj) \
78 container_of(_kobj, struct iommu_group, kobj)
80 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
82 static int __iommu_attach_device(struct iommu_domain
*domain
,
84 static int __iommu_attach_group(struct iommu_domain
*domain
,
85 struct iommu_group
*group
);
86 static void __iommu_detach_group(struct iommu_domain
*domain
,
87 struct iommu_group
*group
);
89 static ssize_t
iommu_group_attr_show(struct kobject
*kobj
,
90 struct attribute
*__attr
, char *buf
)
92 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
93 struct iommu_group
*group
= to_iommu_group(kobj
);
97 ret
= attr
->show(group
, buf
);
101 static ssize_t
iommu_group_attr_store(struct kobject
*kobj
,
102 struct attribute
*__attr
,
103 const char *buf
, size_t count
)
105 struct iommu_group_attribute
*attr
= to_iommu_group_attr(__attr
);
106 struct iommu_group
*group
= to_iommu_group(kobj
);
110 ret
= attr
->store(group
, buf
, count
);
114 static const struct sysfs_ops iommu_group_sysfs_ops
= {
115 .show
= iommu_group_attr_show
,
116 .store
= iommu_group_attr_store
,
119 static int iommu_group_create_file(struct iommu_group
*group
,
120 struct iommu_group_attribute
*attr
)
122 return sysfs_create_file(&group
->kobj
, &attr
->attr
);
125 static void iommu_group_remove_file(struct iommu_group
*group
,
126 struct iommu_group_attribute
*attr
)
128 sysfs_remove_file(&group
->kobj
, &attr
->attr
);
131 static ssize_t
iommu_group_show_name(struct iommu_group
*group
, char *buf
)
133 return sprintf(buf
, "%s\n", group
->name
);
136 static IOMMU_GROUP_ATTR(name
, S_IRUGO
, iommu_group_show_name
, NULL
);
138 static void iommu_group_release(struct kobject
*kobj
)
140 struct iommu_group
*group
= to_iommu_group(kobj
);
142 pr_debug("Releasing group %d\n", group
->id
);
144 if (group
->iommu_data_release
)
145 group
->iommu_data_release(group
->iommu_data
);
147 ida_simple_remove(&iommu_group_ida
, group
->id
);
149 if (group
->default_domain
)
150 iommu_domain_free(group
->default_domain
);
156 static struct kobj_type iommu_group_ktype
= {
157 .sysfs_ops
= &iommu_group_sysfs_ops
,
158 .release
= iommu_group_release
,
162 * iommu_group_alloc - Allocate a new group
163 * @name: Optional name to associate with group, visible in sysfs
165 * This function is called by an iommu driver to allocate a new iommu
166 * group. The iommu group represents the minimum granularity of the iommu.
167 * Upon successful return, the caller holds a reference to the supplied
168 * group in order to hold the group until devices are added. Use
169 * iommu_group_put() to release this extra reference count, allowing the
170 * group to be automatically reclaimed once it has no devices or external
173 struct iommu_group
*iommu_group_alloc(void)
175 struct iommu_group
*group
;
178 group
= kzalloc(sizeof(*group
), GFP_KERNEL
);
180 return ERR_PTR(-ENOMEM
);
182 group
->kobj
.kset
= iommu_group_kset
;
183 mutex_init(&group
->mutex
);
184 INIT_LIST_HEAD(&group
->devices
);
185 BLOCKING_INIT_NOTIFIER_HEAD(&group
->notifier
);
187 ret
= ida_simple_get(&iommu_group_ida
, 0, 0, GFP_KERNEL
);
194 ret
= kobject_init_and_add(&group
->kobj
, &iommu_group_ktype
,
195 NULL
, "%d", group
->id
);
197 ida_simple_remove(&iommu_group_ida
, group
->id
);
202 group
->devices_kobj
= kobject_create_and_add("devices", &group
->kobj
);
203 if (!group
->devices_kobj
) {
204 kobject_put(&group
->kobj
); /* triggers .release & free */
205 return ERR_PTR(-ENOMEM
);
209 * The devices_kobj holds a reference on the group kobject, so
210 * as long as that exists so will the group. We can therefore
211 * use the devices_kobj for reference counting.
213 kobject_put(&group
->kobj
);
215 pr_debug("Allocated group %d\n", group
->id
);
219 EXPORT_SYMBOL_GPL(iommu_group_alloc
);
221 struct iommu_group
*iommu_group_get_by_id(int id
)
223 struct kobject
*group_kobj
;
224 struct iommu_group
*group
;
227 if (!iommu_group_kset
)
230 name
= kasprintf(GFP_KERNEL
, "%d", id
);
234 group_kobj
= kset_find_obj(iommu_group_kset
, name
);
240 group
= container_of(group_kobj
, struct iommu_group
, kobj
);
241 BUG_ON(group
->id
!= id
);
243 kobject_get(group
->devices_kobj
);
244 kobject_put(&group
->kobj
);
248 EXPORT_SYMBOL_GPL(iommu_group_get_by_id
);
251 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
254 * iommu drivers can store data in the group for use when doing iommu
255 * operations. This function provides a way to retrieve it. Caller
256 * should hold a group reference.
258 void *iommu_group_get_iommudata(struct iommu_group
*group
)
260 return group
->iommu_data
;
262 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata
);
265 * iommu_group_set_iommudata - set iommu_data for a group
267 * @iommu_data: new data
268 * @release: release function for iommu_data
270 * iommu drivers can store data in the group for use when doing iommu
271 * operations. This function provides a way to set the data after
272 * the group has been allocated. Caller should hold a group reference.
274 void iommu_group_set_iommudata(struct iommu_group
*group
, void *iommu_data
,
275 void (*release
)(void *iommu_data
))
277 group
->iommu_data
= iommu_data
;
278 group
->iommu_data_release
= release
;
280 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata
);
283 * iommu_group_set_name - set name for a group
287 * Allow iommu driver to set a name for a group. When set it will
288 * appear in a name attribute file under the group in sysfs.
290 int iommu_group_set_name(struct iommu_group
*group
, const char *name
)
295 iommu_group_remove_file(group
, &iommu_group_attr_name
);
302 group
->name
= kstrdup(name
, GFP_KERNEL
);
306 ret
= iommu_group_create_file(group
, &iommu_group_attr_name
);
315 EXPORT_SYMBOL_GPL(iommu_group_set_name
);
317 static int iommu_group_create_direct_mappings(struct iommu_group
*group
,
320 struct iommu_domain
*domain
= group
->default_domain
;
321 struct iommu_dm_region
*entry
;
322 struct list_head mappings
;
323 unsigned long pg_size
;
326 if (!domain
|| domain
->type
!= IOMMU_DOMAIN_DMA
)
329 BUG_ON(!domain
->pgsize_bitmap
);
331 pg_size
= 1UL << __ffs(domain
->pgsize_bitmap
);
332 INIT_LIST_HEAD(&mappings
);
334 iommu_get_dm_regions(dev
, &mappings
);
336 /* We need to consider overlapping regions for different devices */
337 list_for_each_entry(entry
, &mappings
, list
) {
338 dma_addr_t start
, end
, addr
;
340 if (domain
->ops
->apply_dm_region
)
341 domain
->ops
->apply_dm_region(dev
, domain
, entry
);
343 start
= ALIGN(entry
->start
, pg_size
);
344 end
= ALIGN(entry
->start
+ entry
->length
, pg_size
);
346 for (addr
= start
; addr
< end
; addr
+= pg_size
) {
347 phys_addr_t phys_addr
;
349 phys_addr
= iommu_iova_to_phys(domain
, addr
);
353 ret
= iommu_map(domain
, addr
, addr
, pg_size
, entry
->prot
);
361 iommu_put_dm_regions(dev
, &mappings
);
367 * iommu_group_add_device - add a device to an iommu group
368 * @group: the group into which to add the device (reference should be held)
371 * This function is called by an iommu driver to add a device into a
372 * group. Adding a device increments the group reference count.
374 int iommu_group_add_device(struct iommu_group
*group
, struct device
*dev
)
377 struct iommu_device
*device
;
379 device
= kzalloc(sizeof(*device
), GFP_KERNEL
);
385 ret
= sysfs_create_link(&dev
->kobj
, &group
->kobj
, "iommu_group");
391 device
->name
= kasprintf(GFP_KERNEL
, "%s", kobject_name(&dev
->kobj
));
394 sysfs_remove_link(&dev
->kobj
, "iommu_group");
399 ret
= sysfs_create_link_nowarn(group
->devices_kobj
,
400 &dev
->kobj
, device
->name
);
403 if (ret
== -EEXIST
&& i
>= 0) {
405 * Account for the slim chance of collision
406 * and append an instance to the name.
408 device
->name
= kasprintf(GFP_KERNEL
, "%s.%d",
409 kobject_name(&dev
->kobj
), i
++);
413 sysfs_remove_link(&dev
->kobj
, "iommu_group");
418 kobject_get(group
->devices_kobj
);
420 dev
->iommu_group
= group
;
422 iommu_group_create_direct_mappings(group
, dev
);
424 mutex_lock(&group
->mutex
);
425 list_add_tail(&device
->list
, &group
->devices
);
427 __iommu_attach_device(group
->domain
, dev
);
428 mutex_unlock(&group
->mutex
);
430 /* Notify any listeners about change to group. */
431 blocking_notifier_call_chain(&group
->notifier
,
432 IOMMU_GROUP_NOTIFY_ADD_DEVICE
, dev
);
434 trace_add_device_to_group(group
->id
, dev
);
436 pr_info("Adding device %s to group %d\n", dev_name(dev
), group
->id
);
440 EXPORT_SYMBOL_GPL(iommu_group_add_device
);
443 * iommu_group_remove_device - remove a device from it's current group
444 * @dev: device to be removed
446 * This function is called by an iommu driver to remove the device from
447 * it's current group. This decrements the iommu group reference count.
449 void iommu_group_remove_device(struct device
*dev
)
451 struct iommu_group
*group
= dev
->iommu_group
;
452 struct iommu_device
*tmp_device
, *device
= NULL
;
454 pr_info("Removing device %s from group %d\n", dev_name(dev
), group
->id
);
456 /* Pre-notify listeners that a device is being removed. */
457 blocking_notifier_call_chain(&group
->notifier
,
458 IOMMU_GROUP_NOTIFY_DEL_DEVICE
, dev
);
460 mutex_lock(&group
->mutex
);
461 list_for_each_entry(tmp_device
, &group
->devices
, list
) {
462 if (tmp_device
->dev
== dev
) {
464 list_del(&device
->list
);
468 mutex_unlock(&group
->mutex
);
473 sysfs_remove_link(group
->devices_kobj
, device
->name
);
474 sysfs_remove_link(&dev
->kobj
, "iommu_group");
476 trace_remove_device_from_group(group
->id
, dev
);
480 dev
->iommu_group
= NULL
;
481 kobject_put(group
->devices_kobj
);
483 EXPORT_SYMBOL_GPL(iommu_group_remove_device
);
485 static int iommu_group_device_count(struct iommu_group
*group
)
487 struct iommu_device
*entry
;
490 list_for_each_entry(entry
, &group
->devices
, list
)
497 * iommu_group_for_each_dev - iterate over each device in the group
499 * @data: caller opaque data to be passed to callback function
500 * @fn: caller supplied callback function
502 * This function is called by group users to iterate over group devices.
503 * Callers should hold a reference count to the group during callback.
504 * The group->mutex is held across callbacks, which will block calls to
505 * iommu_group_add/remove_device.
507 static int __iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
508 int (*fn
)(struct device
*, void *))
510 struct iommu_device
*device
;
513 list_for_each_entry(device
, &group
->devices
, list
) {
514 ret
= fn(device
->dev
, data
);
522 int iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
523 int (*fn
)(struct device
*, void *))
527 mutex_lock(&group
->mutex
);
528 ret
= __iommu_group_for_each_dev(group
, data
, fn
);
529 mutex_unlock(&group
->mutex
);
533 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev
);
536 * iommu_group_get - Return the group for a device and increment reference
537 * @dev: get the group that this device belongs to
539 * This function is called by iommu drivers and users to get the group
540 * for the specified device. If found, the group is returned and the group
541 * reference in incremented, else NULL.
543 struct iommu_group
*iommu_group_get(struct device
*dev
)
545 struct iommu_group
*group
= dev
->iommu_group
;
548 kobject_get(group
->devices_kobj
);
552 EXPORT_SYMBOL_GPL(iommu_group_get
);
555 * iommu_group_ref_get - Increment reference on a group
556 * @group: the group to use, must not be NULL
558 * This function is called by iommu drivers to take additional references on an
559 * existing group. Returns the given group for convenience.
561 struct iommu_group
*iommu_group_ref_get(struct iommu_group
*group
)
563 kobject_get(group
->devices_kobj
);
568 * iommu_group_put - Decrement group reference
569 * @group: the group to use
571 * This function is called by iommu drivers and users to release the
572 * iommu group. Once the reference count is zero, the group is released.
574 void iommu_group_put(struct iommu_group
*group
)
577 kobject_put(group
->devices_kobj
);
579 EXPORT_SYMBOL_GPL(iommu_group_put
);
582 * iommu_group_register_notifier - Register a notifier for group changes
583 * @group: the group to watch
584 * @nb: notifier block to signal
586 * This function allows iommu group users to track changes in a group.
587 * See include/linux/iommu.h for actions sent via this notifier. Caller
588 * should hold a reference to the group throughout notifier registration.
590 int iommu_group_register_notifier(struct iommu_group
*group
,
591 struct notifier_block
*nb
)
593 return blocking_notifier_chain_register(&group
->notifier
, nb
);
595 EXPORT_SYMBOL_GPL(iommu_group_register_notifier
);
598 * iommu_group_unregister_notifier - Unregister a notifier
599 * @group: the group to watch
600 * @nb: notifier block to signal
602 * Unregister a previously registered group notifier block.
604 int iommu_group_unregister_notifier(struct iommu_group
*group
,
605 struct notifier_block
*nb
)
607 return blocking_notifier_chain_unregister(&group
->notifier
, nb
);
609 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier
);
612 * iommu_group_id - Return ID for a group
613 * @group: the group to ID
615 * Return the unique ID for the group matching the sysfs group number.
617 int iommu_group_id(struct iommu_group
*group
)
621 EXPORT_SYMBOL_GPL(iommu_group_id
);
623 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
624 unsigned long *devfns
);
627 * To consider a PCI device isolated, we require ACS to support Source
628 * Validation, Request Redirection, Completer Redirection, and Upstream
629 * Forwarding. This effectively means that devices cannot spoof their
630 * requester ID, requests and completions cannot be redirected, and all
631 * transactions are forwarded upstream, even as it passes through a
632 * bridge where the target device is downstream.
634 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
637 * For multifunction devices which are not isolated from each other, find
638 * all the other non-isolated functions and look for existing groups. For
639 * each function, we also need to look for aliases to or from other devices
640 * that may already have a group.
642 static struct iommu_group
*get_pci_function_alias_group(struct pci_dev
*pdev
,
643 unsigned long *devfns
)
645 struct pci_dev
*tmp
= NULL
;
646 struct iommu_group
*group
;
648 if (!pdev
->multifunction
|| pci_acs_enabled(pdev
, REQ_ACS_FLAGS
))
651 for_each_pci_dev(tmp
) {
652 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
||
653 PCI_SLOT(tmp
->devfn
) != PCI_SLOT(pdev
->devfn
) ||
654 pci_acs_enabled(tmp
, REQ_ACS_FLAGS
))
657 group
= get_pci_alias_group(tmp
, devfns
);
668 * Look for aliases to or from the given device for existing groups. DMA
669 * aliases are only supported on the same bus, therefore the search
670 * space is quite small (especially since we're really only looking at pcie
671 * device, and therefore only expect multiple slots on the root complex or
672 * downstream switch ports). It's conceivable though that a pair of
673 * multifunction devices could have aliases between them that would cause a
674 * loop. To prevent this, we use a bitmap to track where we've been.
676 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
677 unsigned long *devfns
)
679 struct pci_dev
*tmp
= NULL
;
680 struct iommu_group
*group
;
682 if (test_and_set_bit(pdev
->devfn
& 0xff, devfns
))
685 group
= iommu_group_get(&pdev
->dev
);
689 for_each_pci_dev(tmp
) {
690 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
)
693 /* We alias them or they alias us */
694 if (pci_devs_are_dma_aliases(pdev
, tmp
)) {
695 group
= get_pci_alias_group(tmp
, devfns
);
701 group
= get_pci_function_alias_group(tmp
, devfns
);
712 struct group_for_pci_data
{
713 struct pci_dev
*pdev
;
714 struct iommu_group
*group
;
718 * DMA alias iterator callback, return the last seen device. Stop and return
719 * the IOMMU group if we find one along the way.
721 static int get_pci_alias_or_group(struct pci_dev
*pdev
, u16 alias
, void *opaque
)
723 struct group_for_pci_data
*data
= opaque
;
726 data
->group
= iommu_group_get(&pdev
->dev
);
728 return data
->group
!= NULL
;
732 * Generic device_group call-back function. It just allocates one
733 * iommu-group per device.
735 struct iommu_group
*generic_device_group(struct device
*dev
)
737 struct iommu_group
*group
;
739 group
= iommu_group_alloc();
747 * Use standard PCI bus topology, isolation features, and DMA alias quirks
748 * to find or create an IOMMU group for a device.
750 struct iommu_group
*pci_device_group(struct device
*dev
)
752 struct pci_dev
*pdev
= to_pci_dev(dev
);
753 struct group_for_pci_data data
;
755 struct iommu_group
*group
= NULL
;
756 u64 devfns
[4] = { 0 };
758 if (WARN_ON(!dev_is_pci(dev
)))
759 return ERR_PTR(-EINVAL
);
762 * Find the upstream DMA alias for the device. A device must not
763 * be aliased due to topology in order to have its own IOMMU group.
764 * If we find an alias along the way that already belongs to a
767 if (pci_for_each_dma_alias(pdev
, get_pci_alias_or_group
, &data
))
773 * Continue upstream from the point of minimum IOMMU granularity
774 * due to aliases to the point where devices are protected from
775 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
778 for (bus
= pdev
->bus
; !pci_is_root_bus(bus
); bus
= bus
->parent
) {
782 if (pci_acs_path_enabled(bus
->self
, NULL
, REQ_ACS_FLAGS
))
787 group
= iommu_group_get(&pdev
->dev
);
793 * Look for existing groups on device aliases. If we alias another
794 * device or another device aliases us, use the same group.
796 group
= get_pci_alias_group(pdev
, (unsigned long *)devfns
);
801 * Look for existing groups on non-isolated functions on the same
802 * slot and aliases of those funcions, if any. No need to clear
803 * the search bitmap, the tested devfns are still valid.
805 group
= get_pci_function_alias_group(pdev
, (unsigned long *)devfns
);
809 /* No shared group found, allocate new */
810 group
= iommu_group_alloc();
818 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
819 * @dev: target device
821 * This function is intended to be called by IOMMU drivers and extended to
822 * support common, bus-defined algorithms when determining or creating the
823 * IOMMU group for a device. On success, the caller will hold a reference
824 * to the returned IOMMU group, which will already include the provided
825 * device. The reference should be released with iommu_group_put().
827 struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
)
829 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
830 struct iommu_group
*group
;
833 group
= iommu_group_get(dev
);
837 group
= ERR_PTR(-EINVAL
);
839 if (ops
&& ops
->device_group
)
840 group
= ops
->device_group(dev
);
846 * Try to allocate a default domain - needs support from the
849 if (!group
->default_domain
) {
850 group
->default_domain
= __iommu_domain_alloc(dev
->bus
,
853 group
->domain
= group
->default_domain
;
856 ret
= iommu_group_add_device(group
, dev
);
858 iommu_group_put(group
);
865 struct iommu_domain
*iommu_group_default_domain(struct iommu_group
*group
)
867 return group
->default_domain
;
870 static int add_iommu_group(struct device
*dev
, void *data
)
872 struct iommu_callback_data
*cb
= data
;
873 const struct iommu_ops
*ops
= cb
->ops
;
876 if (!ops
->add_device
)
879 WARN_ON(dev
->iommu_group
);
881 ret
= ops
->add_device(dev
);
884 * We ignore -ENODEV errors for now, as they just mean that the
885 * device is not translated by an IOMMU. We still care about
886 * other errors and fail to initialize when they happen.
894 static int remove_iommu_group(struct device
*dev
, void *data
)
896 struct iommu_callback_data
*cb
= data
;
897 const struct iommu_ops
*ops
= cb
->ops
;
899 if (ops
->remove_device
&& dev
->iommu_group
)
900 ops
->remove_device(dev
);
905 static int iommu_bus_notifier(struct notifier_block
*nb
,
906 unsigned long action
, void *data
)
908 struct device
*dev
= data
;
909 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
910 struct iommu_group
*group
;
911 unsigned long group_action
= 0;
914 * ADD/DEL call into iommu driver ops if provided, which may
915 * result in ADD/DEL notifiers to group->notifier
917 if (action
== BUS_NOTIFY_ADD_DEVICE
) {
919 return ops
->add_device(dev
);
920 } else if (action
== BUS_NOTIFY_REMOVED_DEVICE
) {
921 if (ops
->remove_device
&& dev
->iommu_group
) {
922 ops
->remove_device(dev
);
928 * Remaining BUS_NOTIFYs get filtered and republished to the
929 * group, if anyone is listening
931 group
= iommu_group_get(dev
);
936 case BUS_NOTIFY_BIND_DRIVER
:
937 group_action
= IOMMU_GROUP_NOTIFY_BIND_DRIVER
;
939 case BUS_NOTIFY_BOUND_DRIVER
:
940 group_action
= IOMMU_GROUP_NOTIFY_BOUND_DRIVER
;
942 case BUS_NOTIFY_UNBIND_DRIVER
:
943 group_action
= IOMMU_GROUP_NOTIFY_UNBIND_DRIVER
;
945 case BUS_NOTIFY_UNBOUND_DRIVER
:
946 group_action
= IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER
;
951 blocking_notifier_call_chain(&group
->notifier
,
954 iommu_group_put(group
);
958 static int iommu_bus_init(struct bus_type
*bus
, const struct iommu_ops
*ops
)
961 struct notifier_block
*nb
;
962 struct iommu_callback_data cb
= {
966 nb
= kzalloc(sizeof(struct notifier_block
), GFP_KERNEL
);
970 nb
->notifier_call
= iommu_bus_notifier
;
972 err
= bus_register_notifier(bus
, nb
);
976 err
= bus_for_each_dev(bus
, NULL
, &cb
, add_iommu_group
);
985 bus_for_each_dev(bus
, NULL
, &cb
, remove_iommu_group
);
986 bus_unregister_notifier(bus
, nb
);
995 * bus_set_iommu - set iommu-callbacks for the bus
997 * @ops: the callbacks provided by the iommu-driver
999 * This function is called by an iommu driver to set the iommu methods
1000 * used for a particular bus. Drivers for devices on that bus can use
1001 * the iommu-api after these ops are registered.
1002 * This special function is needed because IOMMUs are usually devices on
1003 * the bus itself, so the iommu drivers are not initialized when the bus
1004 * is set up. With this function the iommu-driver can set the iommu-ops
1007 int bus_set_iommu(struct bus_type
*bus
, const struct iommu_ops
*ops
)
1011 if (bus
->iommu_ops
!= NULL
)
1014 bus
->iommu_ops
= ops
;
1016 /* Do IOMMU specific setup for this bus-type */
1017 err
= iommu_bus_init(bus
, ops
);
1019 bus
->iommu_ops
= NULL
;
1023 EXPORT_SYMBOL_GPL(bus_set_iommu
);
1025 bool iommu_present(struct bus_type
*bus
)
1027 return bus
->iommu_ops
!= NULL
;
1029 EXPORT_SYMBOL_GPL(iommu_present
);
1031 bool iommu_capable(struct bus_type
*bus
, enum iommu_cap cap
)
1033 if (!bus
->iommu_ops
|| !bus
->iommu_ops
->capable
)
1036 return bus
->iommu_ops
->capable(cap
);
1038 EXPORT_SYMBOL_GPL(iommu_capable
);
1041 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1042 * @domain: iommu domain
1043 * @handler: fault handler
1044 * @token: user data, will be passed back to the fault handler
1046 * This function should be used by IOMMU users which want to be notified
1047 * whenever an IOMMU fault happens.
1049 * The fault handler itself should return 0 on success, and an appropriate
1050 * error code otherwise.
1052 void iommu_set_fault_handler(struct iommu_domain
*domain
,
1053 iommu_fault_handler_t handler
,
1058 domain
->handler
= handler
;
1059 domain
->handler_token
= token
;
1061 EXPORT_SYMBOL_GPL(iommu_set_fault_handler
);
1063 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
1066 struct iommu_domain
*domain
;
1068 if (bus
== NULL
|| bus
->iommu_ops
== NULL
)
1071 domain
= bus
->iommu_ops
->domain_alloc(type
);
1075 domain
->ops
= bus
->iommu_ops
;
1076 domain
->type
= type
;
1077 /* Assume all sizes by default; the driver may override this later */
1078 domain
->pgsize_bitmap
= bus
->iommu_ops
->pgsize_bitmap
;
1083 struct iommu_domain
*iommu_domain_alloc(struct bus_type
*bus
)
1085 return __iommu_domain_alloc(bus
, IOMMU_DOMAIN_UNMANAGED
);
1087 EXPORT_SYMBOL_GPL(iommu_domain_alloc
);
1089 void iommu_domain_free(struct iommu_domain
*domain
)
1091 domain
->ops
->domain_free(domain
);
1093 EXPORT_SYMBOL_GPL(iommu_domain_free
);
1095 static int __iommu_attach_device(struct iommu_domain
*domain
,
1099 if (unlikely(domain
->ops
->attach_dev
== NULL
))
1102 ret
= domain
->ops
->attach_dev(domain
, dev
);
1104 trace_attach_device_to_domain(dev
);
1108 int iommu_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
1110 struct iommu_group
*group
;
1113 group
= iommu_group_get(dev
);
1114 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1116 return __iommu_attach_device(domain
, dev
);
1119 * We have a group - lock it to make sure the device-count doesn't
1120 * change while we are attaching
1122 mutex_lock(&group
->mutex
);
1124 if (iommu_group_device_count(group
) != 1)
1127 ret
= __iommu_attach_group(domain
, group
);
1130 mutex_unlock(&group
->mutex
);
1131 iommu_group_put(group
);
1135 EXPORT_SYMBOL_GPL(iommu_attach_device
);
1137 static void __iommu_detach_device(struct iommu_domain
*domain
,
1140 if (unlikely(domain
->ops
->detach_dev
== NULL
))
1143 domain
->ops
->detach_dev(domain
, dev
);
1144 trace_detach_device_from_domain(dev
);
1147 void iommu_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
1149 struct iommu_group
*group
;
1151 group
= iommu_group_get(dev
);
1152 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1154 return __iommu_detach_device(domain
, dev
);
1156 mutex_lock(&group
->mutex
);
1157 if (iommu_group_device_count(group
) != 1) {
1162 __iommu_detach_group(domain
, group
);
1165 mutex_unlock(&group
->mutex
);
1166 iommu_group_put(group
);
1168 EXPORT_SYMBOL_GPL(iommu_detach_device
);
1170 struct iommu_domain
*iommu_get_domain_for_dev(struct device
*dev
)
1172 struct iommu_domain
*domain
;
1173 struct iommu_group
*group
;
1175 group
= iommu_group_get(dev
);
1176 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1180 domain
= group
->domain
;
1182 iommu_group_put(group
);
1186 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev
);
1189 * IOMMU groups are really the natrual working unit of the IOMMU, but
1190 * the IOMMU API works on domains and devices. Bridge that gap by
1191 * iterating over the devices in a group. Ideally we'd have a single
1192 * device which represents the requestor ID of the group, but we also
1193 * allow IOMMU drivers to create policy defined minimum sets, where
1194 * the physical hardware may be able to distiguish members, but we
1195 * wish to group them at a higher level (ex. untrusted multi-function
1196 * PCI devices). Thus we attach each device.
1198 static int iommu_group_do_attach_device(struct device
*dev
, void *data
)
1200 struct iommu_domain
*domain
= data
;
1202 return __iommu_attach_device(domain
, dev
);
1205 static int __iommu_attach_group(struct iommu_domain
*domain
,
1206 struct iommu_group
*group
)
1210 if (group
->default_domain
&& group
->domain
!= group
->default_domain
)
1213 ret
= __iommu_group_for_each_dev(group
, domain
,
1214 iommu_group_do_attach_device
);
1216 group
->domain
= domain
;
1221 int iommu_attach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1225 mutex_lock(&group
->mutex
);
1226 ret
= __iommu_attach_group(domain
, group
);
1227 mutex_unlock(&group
->mutex
);
1231 EXPORT_SYMBOL_GPL(iommu_attach_group
);
1233 static int iommu_group_do_detach_device(struct device
*dev
, void *data
)
1235 struct iommu_domain
*domain
= data
;
1237 __iommu_detach_device(domain
, dev
);
1242 static void __iommu_detach_group(struct iommu_domain
*domain
,
1243 struct iommu_group
*group
)
1247 if (!group
->default_domain
) {
1248 __iommu_group_for_each_dev(group
, domain
,
1249 iommu_group_do_detach_device
);
1250 group
->domain
= NULL
;
1254 if (group
->domain
== group
->default_domain
)
1257 /* Detach by re-attaching to the default domain */
1258 ret
= __iommu_group_for_each_dev(group
, group
->default_domain
,
1259 iommu_group_do_attach_device
);
1263 group
->domain
= group
->default_domain
;
1266 void iommu_detach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1268 mutex_lock(&group
->mutex
);
1269 __iommu_detach_group(domain
, group
);
1270 mutex_unlock(&group
->mutex
);
1272 EXPORT_SYMBOL_GPL(iommu_detach_group
);
1274 phys_addr_t
iommu_iova_to_phys(struct iommu_domain
*domain
, dma_addr_t iova
)
1276 if (unlikely(domain
->ops
->iova_to_phys
== NULL
))
1279 return domain
->ops
->iova_to_phys(domain
, iova
);
1281 EXPORT_SYMBOL_GPL(iommu_iova_to_phys
);
1283 static size_t iommu_pgsize(struct iommu_domain
*domain
,
1284 unsigned long addr_merge
, size_t size
)
1286 unsigned int pgsize_idx
;
1289 /* Max page size that still fits into 'size' */
1290 pgsize_idx
= __fls(size
);
1292 /* need to consider alignment requirements ? */
1293 if (likely(addr_merge
)) {
1294 /* Max page size allowed by address */
1295 unsigned int align_pgsize_idx
= __ffs(addr_merge
);
1296 pgsize_idx
= min(pgsize_idx
, align_pgsize_idx
);
1299 /* build a mask of acceptable page sizes */
1300 pgsize
= (1UL << (pgsize_idx
+ 1)) - 1;
1302 /* throw away page sizes not supported by the hardware */
1303 pgsize
&= domain
->pgsize_bitmap
;
1305 /* make sure we're still sane */
1308 /* pick the biggest page */
1309 pgsize_idx
= __fls(pgsize
);
1310 pgsize
= 1UL << pgsize_idx
;
1315 int iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
1316 phys_addr_t paddr
, size_t size
, int prot
)
1318 unsigned long orig_iova
= iova
;
1319 unsigned int min_pagesz
;
1320 size_t orig_size
= size
;
1321 phys_addr_t orig_paddr
= paddr
;
1324 if (unlikely(domain
->ops
->map
== NULL
||
1325 domain
->pgsize_bitmap
== 0UL))
1328 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1331 /* find out the minimum page size supported */
1332 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1335 * both the virtual address and the physical one, as well as
1336 * the size of the mapping, must be aligned (at least) to the
1337 * size of the smallest page supported by the hardware
1339 if (!IS_ALIGNED(iova
| paddr
| size
, min_pagesz
)) {
1340 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1341 iova
, &paddr
, size
, min_pagesz
);
1345 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova
, &paddr
, size
);
1348 size_t pgsize
= iommu_pgsize(domain
, iova
| paddr
, size
);
1350 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1351 iova
, &paddr
, pgsize
);
1353 ret
= domain
->ops
->map(domain
, iova
, paddr
, pgsize
, prot
);
1362 /* unroll mapping in case something went wrong */
1364 iommu_unmap(domain
, orig_iova
, orig_size
- size
);
1366 trace_map(orig_iova
, orig_paddr
, orig_size
);
1370 EXPORT_SYMBOL_GPL(iommu_map
);
1372 size_t iommu_unmap(struct iommu_domain
*domain
, unsigned long iova
, size_t size
)
1374 size_t unmapped_page
, unmapped
= 0;
1375 unsigned int min_pagesz
;
1376 unsigned long orig_iova
= iova
;
1378 if (unlikely(domain
->ops
->unmap
== NULL
||
1379 domain
->pgsize_bitmap
== 0UL))
1382 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1385 /* find out the minimum page size supported */
1386 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1389 * The virtual address, as well as the size of the mapping, must be
1390 * aligned (at least) to the size of the smallest page supported
1393 if (!IS_ALIGNED(iova
| size
, min_pagesz
)) {
1394 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1395 iova
, size
, min_pagesz
);
1399 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova
, size
);
1402 * Keep iterating until we either unmap 'size' bytes (or more)
1403 * or we hit an area that isn't mapped.
1405 while (unmapped
< size
) {
1406 size_t pgsize
= iommu_pgsize(domain
, iova
, size
- unmapped
);
1408 unmapped_page
= domain
->ops
->unmap(domain
, iova
, pgsize
);
1412 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1413 iova
, unmapped_page
);
1415 iova
+= unmapped_page
;
1416 unmapped
+= unmapped_page
;
1419 trace_unmap(orig_iova
, size
, unmapped
);
1422 EXPORT_SYMBOL_GPL(iommu_unmap
);
1424 size_t default_iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
1425 struct scatterlist
*sg
, unsigned int nents
, int prot
)
1427 struct scatterlist
*s
;
1429 unsigned int i
, min_pagesz
;
1432 if (unlikely(domain
->pgsize_bitmap
== 0UL))
1435 min_pagesz
= 1 << __ffs(domain
->pgsize_bitmap
);
1437 for_each_sg(sg
, s
, nents
, i
) {
1438 phys_addr_t phys
= page_to_phys(sg_page(s
)) + s
->offset
;
1441 * We are mapping on IOMMU page boundaries, so offset within
1442 * the page must be 0. However, the IOMMU may support pages
1443 * smaller than PAGE_SIZE, so s->offset may still represent
1444 * an offset of that boundary within the CPU page.
1446 if (!IS_ALIGNED(s
->offset
, min_pagesz
))
1449 ret
= iommu_map(domain
, iova
+ mapped
, phys
, s
->length
, prot
);
1453 mapped
+= s
->length
;
1459 /* undo mappings already done */
1460 iommu_unmap(domain
, iova
, mapped
);
1465 EXPORT_SYMBOL_GPL(default_iommu_map_sg
);
1467 int iommu_domain_window_enable(struct iommu_domain
*domain
, u32 wnd_nr
,
1468 phys_addr_t paddr
, u64 size
, int prot
)
1470 if (unlikely(domain
->ops
->domain_window_enable
== NULL
))
1473 return domain
->ops
->domain_window_enable(domain
, wnd_nr
, paddr
, size
,
1476 EXPORT_SYMBOL_GPL(iommu_domain_window_enable
);
1478 void iommu_domain_window_disable(struct iommu_domain
*domain
, u32 wnd_nr
)
1480 if (unlikely(domain
->ops
->domain_window_disable
== NULL
))
1483 return domain
->ops
->domain_window_disable(domain
, wnd_nr
);
1485 EXPORT_SYMBOL_GPL(iommu_domain_window_disable
);
1487 static int __init
iommu_init(void)
1489 iommu_group_kset
= kset_create_and_add("iommu_groups",
1491 BUG_ON(!iommu_group_kset
);
1495 core_initcall(iommu_init
);
1497 int iommu_domain_get_attr(struct iommu_domain
*domain
,
1498 enum iommu_attr attr
, void *data
)
1500 struct iommu_domain_geometry
*geometry
;
1506 case DOMAIN_ATTR_GEOMETRY
:
1508 *geometry
= domain
->geometry
;
1511 case DOMAIN_ATTR_PAGING
:
1513 *paging
= (domain
->pgsize_bitmap
!= 0UL);
1515 case DOMAIN_ATTR_WINDOWS
:
1518 if (domain
->ops
->domain_get_windows
!= NULL
)
1519 *count
= domain
->ops
->domain_get_windows(domain
);
1525 if (!domain
->ops
->domain_get_attr
)
1528 ret
= domain
->ops
->domain_get_attr(domain
, attr
, data
);
1533 EXPORT_SYMBOL_GPL(iommu_domain_get_attr
);
1535 int iommu_domain_set_attr(struct iommu_domain
*domain
,
1536 enum iommu_attr attr
, void *data
)
1542 case DOMAIN_ATTR_WINDOWS
:
1545 if (domain
->ops
->domain_set_windows
!= NULL
)
1546 ret
= domain
->ops
->domain_set_windows(domain
, *count
);
1552 if (domain
->ops
->domain_set_attr
== NULL
)
1555 ret
= domain
->ops
->domain_set_attr(domain
, attr
, data
);
1560 EXPORT_SYMBOL_GPL(iommu_domain_set_attr
);
1562 void iommu_get_dm_regions(struct device
*dev
, struct list_head
*list
)
1564 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1566 if (ops
&& ops
->get_dm_regions
)
1567 ops
->get_dm_regions(dev
, list
);
1570 void iommu_put_dm_regions(struct device
*dev
, struct list_head
*list
)
1572 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1574 if (ops
&& ops
->put_dm_regions
)
1575 ops
->put_dm_regions(dev
, list
);
1578 /* Request that a device is direct mapped by the IOMMU */
1579 int iommu_request_dm_for_dev(struct device
*dev
)
1581 struct iommu_domain
*dm_domain
;
1582 struct iommu_group
*group
;
1585 /* Device must already be in a group before calling this function */
1586 group
= iommu_group_get_for_dev(dev
);
1588 return PTR_ERR(group
);
1590 mutex_lock(&group
->mutex
);
1592 /* Check if the default domain is already direct mapped */
1594 if (group
->default_domain
&&
1595 group
->default_domain
->type
== IOMMU_DOMAIN_IDENTITY
)
1598 /* Don't change mappings of existing devices */
1600 if (iommu_group_device_count(group
) != 1)
1603 /* Allocate a direct mapped domain */
1605 dm_domain
= __iommu_domain_alloc(dev
->bus
, IOMMU_DOMAIN_IDENTITY
);
1609 /* Attach the device to the domain */
1610 ret
= __iommu_attach_group(dm_domain
, group
);
1612 iommu_domain_free(dm_domain
);
1616 /* Make the direct mapped domain the default for this group */
1617 if (group
->default_domain
)
1618 iommu_domain_free(group
->default_domain
);
1619 group
->default_domain
= dm_domain
;
1621 pr_info("Using direct mapping for device %s\n", dev_name(dev
));
1625 mutex_unlock(&group
->mutex
);
1626 iommu_group_put(group
);
1631 struct iommu_instance
{
1632 struct list_head list
;
1633 struct fwnode_handle
*fwnode
;
1634 const struct iommu_ops
*ops
;
1636 static LIST_HEAD(iommu_instance_list
);
1637 static DEFINE_SPINLOCK(iommu_instance_lock
);
1639 void iommu_register_instance(struct fwnode_handle
*fwnode
,
1640 const struct iommu_ops
*ops
)
1642 struct iommu_instance
*iommu
= kzalloc(sizeof(*iommu
), GFP_KERNEL
);
1644 if (WARN_ON(!iommu
))
1647 of_node_get(to_of_node(fwnode
));
1648 INIT_LIST_HEAD(&iommu
->list
);
1649 iommu
->fwnode
= fwnode
;
1651 spin_lock(&iommu_instance_lock
);
1652 list_add_tail(&iommu
->list
, &iommu_instance_list
);
1653 spin_unlock(&iommu_instance_lock
);
1656 const struct iommu_ops
*iommu_get_instance(struct fwnode_handle
*fwnode
)
1658 struct iommu_instance
*instance
;
1659 const struct iommu_ops
*ops
= NULL
;
1661 spin_lock(&iommu_instance_lock
);
1662 list_for_each_entry(instance
, &iommu_instance_list
, list
)
1663 if (instance
->fwnode
== fwnode
) {
1664 ops
= instance
->ops
;
1667 spin_unlock(&iommu_instance_lock
);
1671 int iommu_fwspec_init(struct device
*dev
, struct fwnode_handle
*iommu_fwnode
,
1672 const struct iommu_ops
*ops
)
1674 struct iommu_fwspec
*fwspec
= dev
->iommu_fwspec
;
1677 return ops
== fwspec
->ops
? 0 : -EINVAL
;
1679 fwspec
= kzalloc(sizeof(*fwspec
), GFP_KERNEL
);
1683 of_node_get(to_of_node(iommu_fwnode
));
1684 fwspec
->iommu_fwnode
= iommu_fwnode
;
1686 dev
->iommu_fwspec
= fwspec
;
1689 EXPORT_SYMBOL_GPL(iommu_fwspec_init
);
1691 void iommu_fwspec_free(struct device
*dev
)
1693 struct iommu_fwspec
*fwspec
= dev
->iommu_fwspec
;
1696 fwnode_handle_put(fwspec
->iommu_fwnode
);
1698 dev
->iommu_fwspec
= NULL
;
1701 EXPORT_SYMBOL_GPL(iommu_fwspec_free
);
1703 int iommu_fwspec_add_ids(struct device
*dev
, u32
*ids
, int num_ids
)
1705 struct iommu_fwspec
*fwspec
= dev
->iommu_fwspec
;
1712 size
= offsetof(struct iommu_fwspec
, ids
[fwspec
->num_ids
+ num_ids
]);
1713 if (size
> sizeof(*fwspec
)) {
1714 fwspec
= krealloc(dev
->iommu_fwspec
, size
, GFP_KERNEL
);
1719 for (i
= 0; i
< num_ids
; i
++)
1720 fwspec
->ids
[fwspec
->num_ids
+ i
] = ids
[i
];
1722 fwspec
->num_ids
+= num_ids
;
1723 dev
->iommu_fwspec
= fwspec
;
1726 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids
);