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 <trace/events/iommu.h>
36 static struct kset
*iommu_group_kset
;
37 static struct ida iommu_group_ida
;
38 static struct mutex iommu_group_mutex
;
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 mutex_lock(&iommu_group_mutex
);
148 ida_remove(&iommu_group_ida
, group
->id
);
149 mutex_unlock(&iommu_group_mutex
);
151 if (group
->default_domain
)
152 iommu_domain_free(group
->default_domain
);
158 static struct kobj_type iommu_group_ktype
= {
159 .sysfs_ops
= &iommu_group_sysfs_ops
,
160 .release
= iommu_group_release
,
164 * iommu_group_alloc - Allocate a new group
165 * @name: Optional name to associate with group, visible in sysfs
167 * This function is called by an iommu driver to allocate a new iommu
168 * group. The iommu group represents the minimum granularity of the iommu.
169 * Upon successful return, the caller holds a reference to the supplied
170 * group in order to hold the group until devices are added. Use
171 * iommu_group_put() to release this extra reference count, allowing the
172 * group to be automatically reclaimed once it has no devices or external
175 struct iommu_group
*iommu_group_alloc(void)
177 struct iommu_group
*group
;
180 group
= kzalloc(sizeof(*group
), GFP_KERNEL
);
182 return ERR_PTR(-ENOMEM
);
184 group
->kobj
.kset
= iommu_group_kset
;
185 mutex_init(&group
->mutex
);
186 INIT_LIST_HEAD(&group
->devices
);
187 BLOCKING_INIT_NOTIFIER_HEAD(&group
->notifier
);
189 mutex_lock(&iommu_group_mutex
);
192 if (unlikely(0 == ida_pre_get(&iommu_group_ida
, GFP_KERNEL
))) {
194 mutex_unlock(&iommu_group_mutex
);
195 return ERR_PTR(-ENOMEM
);
198 if (-EAGAIN
== ida_get_new(&iommu_group_ida
, &group
->id
))
201 mutex_unlock(&iommu_group_mutex
);
203 ret
= kobject_init_and_add(&group
->kobj
, &iommu_group_ktype
,
204 NULL
, "%d", group
->id
);
206 mutex_lock(&iommu_group_mutex
);
207 ida_remove(&iommu_group_ida
, group
->id
);
208 mutex_unlock(&iommu_group_mutex
);
213 group
->devices_kobj
= kobject_create_and_add("devices", &group
->kobj
);
214 if (!group
->devices_kobj
) {
215 kobject_put(&group
->kobj
); /* triggers .release & free */
216 return ERR_PTR(-ENOMEM
);
220 * The devices_kobj holds a reference on the group kobject, so
221 * as long as that exists so will the group. We can therefore
222 * use the devices_kobj for reference counting.
224 kobject_put(&group
->kobj
);
226 pr_debug("Allocated group %d\n", group
->id
);
230 EXPORT_SYMBOL_GPL(iommu_group_alloc
);
232 struct iommu_group
*iommu_group_get_by_id(int id
)
234 struct kobject
*group_kobj
;
235 struct iommu_group
*group
;
238 if (!iommu_group_kset
)
241 name
= kasprintf(GFP_KERNEL
, "%d", id
);
245 group_kobj
= kset_find_obj(iommu_group_kset
, name
);
251 group
= container_of(group_kobj
, struct iommu_group
, kobj
);
252 BUG_ON(group
->id
!= id
);
254 kobject_get(group
->devices_kobj
);
255 kobject_put(&group
->kobj
);
259 EXPORT_SYMBOL_GPL(iommu_group_get_by_id
);
262 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
265 * iommu drivers can store data in the group for use when doing iommu
266 * operations. This function provides a way to retrieve it. Caller
267 * should hold a group reference.
269 void *iommu_group_get_iommudata(struct iommu_group
*group
)
271 return group
->iommu_data
;
273 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata
);
276 * iommu_group_set_iommudata - set iommu_data for a group
278 * @iommu_data: new data
279 * @release: release function for iommu_data
281 * iommu drivers can store data in the group for use when doing iommu
282 * operations. This function provides a way to set the data after
283 * the group has been allocated. Caller should hold a group reference.
285 void iommu_group_set_iommudata(struct iommu_group
*group
, void *iommu_data
,
286 void (*release
)(void *iommu_data
))
288 group
->iommu_data
= iommu_data
;
289 group
->iommu_data_release
= release
;
291 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata
);
294 * iommu_group_set_name - set name for a group
298 * Allow iommu driver to set a name for a group. When set it will
299 * appear in a name attribute file under the group in sysfs.
301 int iommu_group_set_name(struct iommu_group
*group
, const char *name
)
306 iommu_group_remove_file(group
, &iommu_group_attr_name
);
313 group
->name
= kstrdup(name
, GFP_KERNEL
);
317 ret
= iommu_group_create_file(group
, &iommu_group_attr_name
);
326 EXPORT_SYMBOL_GPL(iommu_group_set_name
);
328 static int iommu_group_create_direct_mappings(struct iommu_group
*group
,
331 struct iommu_domain
*domain
= group
->default_domain
;
332 struct iommu_dm_region
*entry
;
333 struct list_head mappings
;
334 unsigned long pg_size
;
337 if (!domain
|| domain
->type
!= IOMMU_DOMAIN_DMA
)
340 BUG_ON(!domain
->ops
->pgsize_bitmap
);
342 pg_size
= 1UL << __ffs(domain
->ops
->pgsize_bitmap
);
343 INIT_LIST_HEAD(&mappings
);
345 iommu_get_dm_regions(dev
, &mappings
);
347 /* We need to consider overlapping regions for different devices */
348 list_for_each_entry(entry
, &mappings
, list
) {
349 dma_addr_t start
, end
, addr
;
351 start
= ALIGN(entry
->start
, pg_size
);
352 end
= ALIGN(entry
->start
+ entry
->length
, pg_size
);
354 for (addr
= start
; addr
< end
; addr
+= pg_size
) {
355 phys_addr_t phys_addr
;
357 phys_addr
= iommu_iova_to_phys(domain
, addr
);
361 ret
= iommu_map(domain
, addr
, addr
, pg_size
, entry
->prot
);
369 iommu_put_dm_regions(dev
, &mappings
);
375 * iommu_group_add_device - add a device to an iommu group
376 * @group: the group into which to add the device (reference should be held)
379 * This function is called by an iommu driver to add a device into a
380 * group. Adding a device increments the group reference count.
382 int iommu_group_add_device(struct iommu_group
*group
, struct device
*dev
)
385 struct iommu_device
*device
;
387 device
= kzalloc(sizeof(*device
), GFP_KERNEL
);
393 ret
= sysfs_create_link(&dev
->kobj
, &group
->kobj
, "iommu_group");
399 device
->name
= kasprintf(GFP_KERNEL
, "%s", kobject_name(&dev
->kobj
));
402 sysfs_remove_link(&dev
->kobj
, "iommu_group");
407 ret
= sysfs_create_link_nowarn(group
->devices_kobj
,
408 &dev
->kobj
, device
->name
);
411 if (ret
== -EEXIST
&& i
>= 0) {
413 * Account for the slim chance of collision
414 * and append an instance to the name.
416 device
->name
= kasprintf(GFP_KERNEL
, "%s.%d",
417 kobject_name(&dev
->kobj
), i
++);
421 sysfs_remove_link(&dev
->kobj
, "iommu_group");
426 kobject_get(group
->devices_kobj
);
428 dev
->iommu_group
= group
;
430 iommu_group_create_direct_mappings(group
, dev
);
432 mutex_lock(&group
->mutex
);
433 list_add_tail(&device
->list
, &group
->devices
);
435 __iommu_attach_device(group
->domain
, dev
);
436 mutex_unlock(&group
->mutex
);
438 /* Notify any listeners about change to group. */
439 blocking_notifier_call_chain(&group
->notifier
,
440 IOMMU_GROUP_NOTIFY_ADD_DEVICE
, dev
);
442 trace_add_device_to_group(group
->id
, dev
);
444 pr_info("Adding device %s to group %d\n", dev_name(dev
), group
->id
);
448 EXPORT_SYMBOL_GPL(iommu_group_add_device
);
451 * iommu_group_remove_device - remove a device from it's current group
452 * @dev: device to be removed
454 * This function is called by an iommu driver to remove the device from
455 * it's current group. This decrements the iommu group reference count.
457 void iommu_group_remove_device(struct device
*dev
)
459 struct iommu_group
*group
= dev
->iommu_group
;
460 struct iommu_device
*tmp_device
, *device
= NULL
;
462 pr_info("Removing device %s from group %d\n", dev_name(dev
), group
->id
);
464 /* Pre-notify listeners that a device is being removed. */
465 blocking_notifier_call_chain(&group
->notifier
,
466 IOMMU_GROUP_NOTIFY_DEL_DEVICE
, dev
);
468 mutex_lock(&group
->mutex
);
469 list_for_each_entry(tmp_device
, &group
->devices
, list
) {
470 if (tmp_device
->dev
== dev
) {
472 list_del(&device
->list
);
476 mutex_unlock(&group
->mutex
);
481 sysfs_remove_link(group
->devices_kobj
, device
->name
);
482 sysfs_remove_link(&dev
->kobj
, "iommu_group");
484 trace_remove_device_from_group(group
->id
, dev
);
488 dev
->iommu_group
= NULL
;
489 kobject_put(group
->devices_kobj
);
491 EXPORT_SYMBOL_GPL(iommu_group_remove_device
);
493 static int iommu_group_device_count(struct iommu_group
*group
)
495 struct iommu_device
*entry
;
498 list_for_each_entry(entry
, &group
->devices
, list
)
505 * iommu_group_for_each_dev - iterate over each device in the group
507 * @data: caller opaque data to be passed to callback function
508 * @fn: caller supplied callback function
510 * This function is called by group users to iterate over group devices.
511 * Callers should hold a reference count to the group during callback.
512 * The group->mutex is held across callbacks, which will block calls to
513 * iommu_group_add/remove_device.
515 static int __iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
516 int (*fn
)(struct device
*, void *))
518 struct iommu_device
*device
;
521 list_for_each_entry(device
, &group
->devices
, list
) {
522 ret
= fn(device
->dev
, data
);
530 int iommu_group_for_each_dev(struct iommu_group
*group
, void *data
,
531 int (*fn
)(struct device
*, void *))
535 mutex_lock(&group
->mutex
);
536 ret
= __iommu_group_for_each_dev(group
, data
, fn
);
537 mutex_unlock(&group
->mutex
);
541 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev
);
544 * iommu_group_get - Return the group for a device and increment reference
545 * @dev: get the group that this device belongs to
547 * This function is called by iommu drivers and users to get the group
548 * for the specified device. If found, the group is returned and the group
549 * reference in incremented, else NULL.
551 struct iommu_group
*iommu_group_get(struct device
*dev
)
553 struct iommu_group
*group
= dev
->iommu_group
;
556 kobject_get(group
->devices_kobj
);
560 EXPORT_SYMBOL_GPL(iommu_group_get
);
563 * iommu_group_put - Decrement group reference
564 * @group: the group to use
566 * This function is called by iommu drivers and users to release the
567 * iommu group. Once the reference count is zero, the group is released.
569 void iommu_group_put(struct iommu_group
*group
)
572 kobject_put(group
->devices_kobj
);
574 EXPORT_SYMBOL_GPL(iommu_group_put
);
577 * iommu_group_register_notifier - Register a notifier for group changes
578 * @group: the group to watch
579 * @nb: notifier block to signal
581 * This function allows iommu group users to track changes in a group.
582 * See include/linux/iommu.h for actions sent via this notifier. Caller
583 * should hold a reference to the group throughout notifier registration.
585 int iommu_group_register_notifier(struct iommu_group
*group
,
586 struct notifier_block
*nb
)
588 return blocking_notifier_chain_register(&group
->notifier
, nb
);
590 EXPORT_SYMBOL_GPL(iommu_group_register_notifier
);
593 * iommu_group_unregister_notifier - Unregister a notifier
594 * @group: the group to watch
595 * @nb: notifier block to signal
597 * Unregister a previously registered group notifier block.
599 int iommu_group_unregister_notifier(struct iommu_group
*group
,
600 struct notifier_block
*nb
)
602 return blocking_notifier_chain_unregister(&group
->notifier
, nb
);
604 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier
);
607 * iommu_group_id - Return ID for a group
608 * @group: the group to ID
610 * Return the unique ID for the group matching the sysfs group number.
612 int iommu_group_id(struct iommu_group
*group
)
616 EXPORT_SYMBOL_GPL(iommu_group_id
);
618 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
619 unsigned long *devfns
);
622 * To consider a PCI device isolated, we require ACS to support Source
623 * Validation, Request Redirection, Completer Redirection, and Upstream
624 * Forwarding. This effectively means that devices cannot spoof their
625 * requester ID, requests and completions cannot be redirected, and all
626 * transactions are forwarded upstream, even as it passes through a
627 * bridge where the target device is downstream.
629 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
632 * For multifunction devices which are not isolated from each other, find
633 * all the other non-isolated functions and look for existing groups. For
634 * each function, we also need to look for aliases to or from other devices
635 * that may already have a group.
637 static struct iommu_group
*get_pci_function_alias_group(struct pci_dev
*pdev
,
638 unsigned long *devfns
)
640 struct pci_dev
*tmp
= NULL
;
641 struct iommu_group
*group
;
643 if (!pdev
->multifunction
|| pci_acs_enabled(pdev
, REQ_ACS_FLAGS
))
646 for_each_pci_dev(tmp
) {
647 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
||
648 PCI_SLOT(tmp
->devfn
) != PCI_SLOT(pdev
->devfn
) ||
649 pci_acs_enabled(tmp
, REQ_ACS_FLAGS
))
652 group
= get_pci_alias_group(tmp
, devfns
);
663 * Look for aliases to or from the given device for exisiting groups. The
664 * dma_alias_devfn only supports aliases on the same bus, therefore the search
665 * space is quite small (especially since we're really only looking at pcie
666 * device, and therefore only expect multiple slots on the root complex or
667 * downstream switch ports). It's conceivable though that a pair of
668 * multifunction devices could have aliases between them that would cause a
669 * loop. To prevent this, we use a bitmap to track where we've been.
671 static struct iommu_group
*get_pci_alias_group(struct pci_dev
*pdev
,
672 unsigned long *devfns
)
674 struct pci_dev
*tmp
= NULL
;
675 struct iommu_group
*group
;
677 if (test_and_set_bit(pdev
->devfn
& 0xff, devfns
))
680 group
= iommu_group_get(&pdev
->dev
);
684 for_each_pci_dev(tmp
) {
685 if (tmp
== pdev
|| tmp
->bus
!= pdev
->bus
)
688 /* We alias them or they alias us */
689 if (((pdev
->dev_flags
& PCI_DEV_FLAGS_DMA_ALIAS_DEVFN
) &&
690 pdev
->dma_alias_devfn
== tmp
->devfn
) ||
691 ((tmp
->dev_flags
& PCI_DEV_FLAGS_DMA_ALIAS_DEVFN
) &&
692 tmp
->dma_alias_devfn
== pdev
->devfn
)) {
694 group
= get_pci_alias_group(tmp
, devfns
);
700 group
= get_pci_function_alias_group(tmp
, devfns
);
711 struct group_for_pci_data
{
712 struct pci_dev
*pdev
;
713 struct iommu_group
*group
;
717 * DMA alias iterator callback, return the last seen device. Stop and return
718 * the IOMMU group if we find one along the way.
720 static int get_pci_alias_or_group(struct pci_dev
*pdev
, u16 alias
, void *opaque
)
722 struct group_for_pci_data
*data
= opaque
;
725 data
->group
= iommu_group_get(&pdev
->dev
);
727 return data
->group
!= NULL
;
731 * Use standard PCI bus topology, isolation features, and DMA alias quirks
732 * to find or create an IOMMU group for a device.
734 static struct iommu_group
*iommu_group_get_for_pci_dev(struct pci_dev
*pdev
)
736 struct group_for_pci_data data
;
738 struct iommu_group
*group
= NULL
;
739 u64 devfns
[4] = { 0 };
742 * Find the upstream DMA alias for the device. A device must not
743 * be aliased due to topology in order to have its own IOMMU group.
744 * If we find an alias along the way that already belongs to a
747 if (pci_for_each_dma_alias(pdev
, get_pci_alias_or_group
, &data
))
753 * Continue upstream from the point of minimum IOMMU granularity
754 * due to aliases to the point where devices are protected from
755 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
758 for (bus
= pdev
->bus
; !pci_is_root_bus(bus
); bus
= bus
->parent
) {
762 if (pci_acs_path_enabled(bus
->self
, NULL
, REQ_ACS_FLAGS
))
767 group
= iommu_group_get(&pdev
->dev
);
773 * Look for existing groups on device aliases. If we alias another
774 * device or another device aliases us, use the same group.
776 group
= get_pci_alias_group(pdev
, (unsigned long *)devfns
);
781 * Look for existing groups on non-isolated functions on the same
782 * slot and aliases of those funcions, if any. No need to clear
783 * the search bitmap, the tested devfns are still valid.
785 group
= get_pci_function_alias_group(pdev
, (unsigned long *)devfns
);
789 /* No shared group found, allocate new */
790 group
= iommu_group_alloc();
795 * Try to allocate a default domain - needs support from the
798 group
->default_domain
= __iommu_domain_alloc(pdev
->dev
.bus
,
800 group
->domain
= group
->default_domain
;
806 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
807 * @dev: target device
809 * This function is intended to be called by IOMMU drivers and extended to
810 * support common, bus-defined algorithms when determining or creating the
811 * IOMMU group for a device. On success, the caller will hold a reference
812 * to the returned IOMMU group, which will already include the provided
813 * device. The reference should be released with iommu_group_put().
815 struct iommu_group
*iommu_group_get_for_dev(struct device
*dev
)
817 struct iommu_group
*group
;
820 group
= iommu_group_get(dev
);
824 if (!dev_is_pci(dev
))
825 return ERR_PTR(-EINVAL
);
827 group
= iommu_group_get_for_pci_dev(to_pci_dev(dev
));
832 ret
= iommu_group_add_device(group
, dev
);
834 iommu_group_put(group
);
841 struct iommu_domain
*iommu_group_default_domain(struct iommu_group
*group
)
843 return group
->default_domain
;
846 static int add_iommu_group(struct device
*dev
, void *data
)
848 struct iommu_callback_data
*cb
= data
;
849 const struct iommu_ops
*ops
= cb
->ops
;
852 if (!ops
->add_device
)
855 WARN_ON(dev
->iommu_group
);
857 ret
= ops
->add_device(dev
);
860 * We ignore -ENODEV errors for now, as they just mean that the
861 * device is not translated by an IOMMU. We still care about
862 * other errors and fail to initialize when they happen.
870 static int remove_iommu_group(struct device
*dev
, void *data
)
872 struct iommu_callback_data
*cb
= data
;
873 const struct iommu_ops
*ops
= cb
->ops
;
875 if (ops
->remove_device
&& dev
->iommu_group
)
876 ops
->remove_device(dev
);
881 static int iommu_bus_notifier(struct notifier_block
*nb
,
882 unsigned long action
, void *data
)
884 struct device
*dev
= data
;
885 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
886 struct iommu_group
*group
;
887 unsigned long group_action
= 0;
890 * ADD/DEL call into iommu driver ops if provided, which may
891 * result in ADD/DEL notifiers to group->notifier
893 if (action
== BUS_NOTIFY_ADD_DEVICE
) {
895 return ops
->add_device(dev
);
896 } else if (action
== BUS_NOTIFY_REMOVED_DEVICE
) {
897 if (ops
->remove_device
&& dev
->iommu_group
) {
898 ops
->remove_device(dev
);
904 * Remaining BUS_NOTIFYs get filtered and republished to the
905 * group, if anyone is listening
907 group
= iommu_group_get(dev
);
912 case BUS_NOTIFY_BIND_DRIVER
:
913 group_action
= IOMMU_GROUP_NOTIFY_BIND_DRIVER
;
915 case BUS_NOTIFY_BOUND_DRIVER
:
916 group_action
= IOMMU_GROUP_NOTIFY_BOUND_DRIVER
;
918 case BUS_NOTIFY_UNBIND_DRIVER
:
919 group_action
= IOMMU_GROUP_NOTIFY_UNBIND_DRIVER
;
921 case BUS_NOTIFY_UNBOUND_DRIVER
:
922 group_action
= IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER
;
927 blocking_notifier_call_chain(&group
->notifier
,
930 iommu_group_put(group
);
934 static int iommu_bus_init(struct bus_type
*bus
, const struct iommu_ops
*ops
)
937 struct notifier_block
*nb
;
938 struct iommu_callback_data cb
= {
942 nb
= kzalloc(sizeof(struct notifier_block
), GFP_KERNEL
);
946 nb
->notifier_call
= iommu_bus_notifier
;
948 err
= bus_register_notifier(bus
, nb
);
952 err
= bus_for_each_dev(bus
, NULL
, &cb
, add_iommu_group
);
961 bus_for_each_dev(bus
, NULL
, &cb
, remove_iommu_group
);
962 bus_unregister_notifier(bus
, nb
);
971 * bus_set_iommu - set iommu-callbacks for the bus
973 * @ops: the callbacks provided by the iommu-driver
975 * This function is called by an iommu driver to set the iommu methods
976 * used for a particular bus. Drivers for devices on that bus can use
977 * the iommu-api after these ops are registered.
978 * This special function is needed because IOMMUs are usually devices on
979 * the bus itself, so the iommu drivers are not initialized when the bus
980 * is set up. With this function the iommu-driver can set the iommu-ops
983 int bus_set_iommu(struct bus_type
*bus
, const struct iommu_ops
*ops
)
987 if (bus
->iommu_ops
!= NULL
)
990 bus
->iommu_ops
= ops
;
992 /* Do IOMMU specific setup for this bus-type */
993 err
= iommu_bus_init(bus
, ops
);
995 bus
->iommu_ops
= NULL
;
999 EXPORT_SYMBOL_GPL(bus_set_iommu
);
1001 bool iommu_present(struct bus_type
*bus
)
1003 return bus
->iommu_ops
!= NULL
;
1005 EXPORT_SYMBOL_GPL(iommu_present
);
1007 bool iommu_capable(struct bus_type
*bus
, enum iommu_cap cap
)
1009 if (!bus
->iommu_ops
|| !bus
->iommu_ops
->capable
)
1012 return bus
->iommu_ops
->capable(cap
);
1014 EXPORT_SYMBOL_GPL(iommu_capable
);
1017 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1018 * @domain: iommu domain
1019 * @handler: fault handler
1020 * @token: user data, will be passed back to the fault handler
1022 * This function should be used by IOMMU users which want to be notified
1023 * whenever an IOMMU fault happens.
1025 * The fault handler itself should return 0 on success, and an appropriate
1026 * error code otherwise.
1028 void iommu_set_fault_handler(struct iommu_domain
*domain
,
1029 iommu_fault_handler_t handler
,
1034 domain
->handler
= handler
;
1035 domain
->handler_token
= token
;
1037 EXPORT_SYMBOL_GPL(iommu_set_fault_handler
);
1039 static struct iommu_domain
*__iommu_domain_alloc(struct bus_type
*bus
,
1042 struct iommu_domain
*domain
;
1044 if (bus
== NULL
|| bus
->iommu_ops
== NULL
)
1047 domain
= bus
->iommu_ops
->domain_alloc(type
);
1051 domain
->ops
= bus
->iommu_ops
;
1052 domain
->type
= type
;
1057 struct iommu_domain
*iommu_domain_alloc(struct bus_type
*bus
)
1059 return __iommu_domain_alloc(bus
, IOMMU_DOMAIN_UNMANAGED
);
1061 EXPORT_SYMBOL_GPL(iommu_domain_alloc
);
1063 void iommu_domain_free(struct iommu_domain
*domain
)
1065 domain
->ops
->domain_free(domain
);
1067 EXPORT_SYMBOL_GPL(iommu_domain_free
);
1069 static int __iommu_attach_device(struct iommu_domain
*domain
,
1073 if (unlikely(domain
->ops
->attach_dev
== NULL
))
1076 ret
= domain
->ops
->attach_dev(domain
, dev
);
1078 trace_attach_device_to_domain(dev
);
1082 int iommu_attach_device(struct iommu_domain
*domain
, struct device
*dev
)
1084 struct iommu_group
*group
;
1087 group
= iommu_group_get(dev
);
1088 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1090 return __iommu_attach_device(domain
, dev
);
1093 * We have a group - lock it to make sure the device-count doesn't
1094 * change while we are attaching
1096 mutex_lock(&group
->mutex
);
1098 if (iommu_group_device_count(group
) != 1)
1101 ret
= __iommu_attach_group(domain
, group
);
1104 mutex_unlock(&group
->mutex
);
1105 iommu_group_put(group
);
1109 EXPORT_SYMBOL_GPL(iommu_attach_device
);
1111 static void __iommu_detach_device(struct iommu_domain
*domain
,
1114 if (unlikely(domain
->ops
->detach_dev
== NULL
))
1117 domain
->ops
->detach_dev(domain
, dev
);
1118 trace_detach_device_from_domain(dev
);
1121 void iommu_detach_device(struct iommu_domain
*domain
, struct device
*dev
)
1123 struct iommu_group
*group
;
1125 group
= iommu_group_get(dev
);
1126 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1128 return __iommu_detach_device(domain
, dev
);
1130 mutex_lock(&group
->mutex
);
1131 if (iommu_group_device_count(group
) != 1) {
1136 __iommu_detach_group(domain
, group
);
1139 mutex_unlock(&group
->mutex
);
1140 iommu_group_put(group
);
1142 EXPORT_SYMBOL_GPL(iommu_detach_device
);
1144 struct iommu_domain
*iommu_get_domain_for_dev(struct device
*dev
)
1146 struct iommu_domain
*domain
;
1147 struct iommu_group
*group
;
1149 group
= iommu_group_get(dev
);
1150 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1154 domain
= group
->domain
;
1156 iommu_group_put(group
);
1160 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev
);
1163 * IOMMU groups are really the natrual working unit of the IOMMU, but
1164 * the IOMMU API works on domains and devices. Bridge that gap by
1165 * iterating over the devices in a group. Ideally we'd have a single
1166 * device which represents the requestor ID of the group, but we also
1167 * allow IOMMU drivers to create policy defined minimum sets, where
1168 * the physical hardware may be able to distiguish members, but we
1169 * wish to group them at a higher level (ex. untrusted multi-function
1170 * PCI devices). Thus we attach each device.
1172 static int iommu_group_do_attach_device(struct device
*dev
, void *data
)
1174 struct iommu_domain
*domain
= data
;
1176 return __iommu_attach_device(domain
, dev
);
1179 static int __iommu_attach_group(struct iommu_domain
*domain
,
1180 struct iommu_group
*group
)
1184 if (group
->default_domain
&& group
->domain
!= group
->default_domain
)
1187 ret
= __iommu_group_for_each_dev(group
, domain
,
1188 iommu_group_do_attach_device
);
1190 group
->domain
= domain
;
1195 int iommu_attach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1199 mutex_lock(&group
->mutex
);
1200 ret
= __iommu_attach_group(domain
, group
);
1201 mutex_unlock(&group
->mutex
);
1205 EXPORT_SYMBOL_GPL(iommu_attach_group
);
1207 static int iommu_group_do_detach_device(struct device
*dev
, void *data
)
1209 struct iommu_domain
*domain
= data
;
1211 __iommu_detach_device(domain
, dev
);
1216 static void __iommu_detach_group(struct iommu_domain
*domain
,
1217 struct iommu_group
*group
)
1221 if (!group
->default_domain
) {
1222 __iommu_group_for_each_dev(group
, domain
,
1223 iommu_group_do_detach_device
);
1224 group
->domain
= NULL
;
1228 if (group
->domain
== group
->default_domain
)
1231 /* Detach by re-attaching to the default domain */
1232 ret
= __iommu_group_for_each_dev(group
, group
->default_domain
,
1233 iommu_group_do_attach_device
);
1237 group
->domain
= group
->default_domain
;
1240 void iommu_detach_group(struct iommu_domain
*domain
, struct iommu_group
*group
)
1242 mutex_lock(&group
->mutex
);
1243 __iommu_detach_group(domain
, group
);
1244 mutex_unlock(&group
->mutex
);
1246 EXPORT_SYMBOL_GPL(iommu_detach_group
);
1248 phys_addr_t
iommu_iova_to_phys(struct iommu_domain
*domain
, dma_addr_t iova
)
1250 if (unlikely(domain
->ops
->iova_to_phys
== NULL
))
1253 return domain
->ops
->iova_to_phys(domain
, iova
);
1255 EXPORT_SYMBOL_GPL(iommu_iova_to_phys
);
1257 static size_t iommu_pgsize(struct iommu_domain
*domain
,
1258 unsigned long addr_merge
, size_t size
)
1260 unsigned int pgsize_idx
;
1263 /* Max page size that still fits into 'size' */
1264 pgsize_idx
= __fls(size
);
1266 /* need to consider alignment requirements ? */
1267 if (likely(addr_merge
)) {
1268 /* Max page size allowed by address */
1269 unsigned int align_pgsize_idx
= __ffs(addr_merge
);
1270 pgsize_idx
= min(pgsize_idx
, align_pgsize_idx
);
1273 /* build a mask of acceptable page sizes */
1274 pgsize
= (1UL << (pgsize_idx
+ 1)) - 1;
1276 /* throw away page sizes not supported by the hardware */
1277 pgsize
&= domain
->ops
->pgsize_bitmap
;
1279 /* make sure we're still sane */
1282 /* pick the biggest page */
1283 pgsize_idx
= __fls(pgsize
);
1284 pgsize
= 1UL << pgsize_idx
;
1289 int iommu_map(struct iommu_domain
*domain
, unsigned long iova
,
1290 phys_addr_t paddr
, size_t size
, int prot
)
1292 unsigned long orig_iova
= iova
;
1293 unsigned int min_pagesz
;
1294 size_t orig_size
= size
;
1297 if (unlikely(domain
->ops
->map
== NULL
||
1298 domain
->ops
->pgsize_bitmap
== 0UL))
1301 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1304 /* find out the minimum page size supported */
1305 min_pagesz
= 1 << __ffs(domain
->ops
->pgsize_bitmap
);
1308 * both the virtual address and the physical one, as well as
1309 * the size of the mapping, must be aligned (at least) to the
1310 * size of the smallest page supported by the hardware
1312 if (!IS_ALIGNED(iova
| paddr
| size
, min_pagesz
)) {
1313 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1314 iova
, &paddr
, size
, min_pagesz
);
1318 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova
, &paddr
, size
);
1321 size_t pgsize
= iommu_pgsize(domain
, iova
| paddr
, size
);
1323 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1324 iova
, &paddr
, pgsize
);
1326 ret
= domain
->ops
->map(domain
, iova
, paddr
, pgsize
, prot
);
1335 /* unroll mapping in case something went wrong */
1337 iommu_unmap(domain
, orig_iova
, orig_size
- size
);
1339 trace_map(orig_iova
, paddr
, orig_size
);
1343 EXPORT_SYMBOL_GPL(iommu_map
);
1345 size_t iommu_unmap(struct iommu_domain
*domain
, unsigned long iova
, size_t size
)
1347 size_t unmapped_page
, unmapped
= 0;
1348 unsigned int min_pagesz
;
1349 unsigned long orig_iova
= iova
;
1351 if (unlikely(domain
->ops
->unmap
== NULL
||
1352 domain
->ops
->pgsize_bitmap
== 0UL))
1355 if (unlikely(!(domain
->type
& __IOMMU_DOMAIN_PAGING
)))
1358 /* find out the minimum page size supported */
1359 min_pagesz
= 1 << __ffs(domain
->ops
->pgsize_bitmap
);
1362 * The virtual address, as well as the size of the mapping, must be
1363 * aligned (at least) to the size of the smallest page supported
1366 if (!IS_ALIGNED(iova
| size
, min_pagesz
)) {
1367 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1368 iova
, size
, min_pagesz
);
1372 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova
, size
);
1375 * Keep iterating until we either unmap 'size' bytes (or more)
1376 * or we hit an area that isn't mapped.
1378 while (unmapped
< size
) {
1379 size_t pgsize
= iommu_pgsize(domain
, iova
, size
- unmapped
);
1381 unmapped_page
= domain
->ops
->unmap(domain
, iova
, pgsize
);
1385 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1386 iova
, unmapped_page
);
1388 iova
+= unmapped_page
;
1389 unmapped
+= unmapped_page
;
1392 trace_unmap(orig_iova
, size
, unmapped
);
1395 EXPORT_SYMBOL_GPL(iommu_unmap
);
1397 size_t default_iommu_map_sg(struct iommu_domain
*domain
, unsigned long iova
,
1398 struct scatterlist
*sg
, unsigned int nents
, int prot
)
1400 struct scatterlist
*s
;
1402 unsigned int i
, min_pagesz
;
1405 if (unlikely(domain
->ops
->pgsize_bitmap
== 0UL))
1408 min_pagesz
= 1 << __ffs(domain
->ops
->pgsize_bitmap
);
1410 for_each_sg(sg
, s
, nents
, i
) {
1411 phys_addr_t phys
= sg_phys(s
);
1414 * We are mapping on IOMMU page boundaries, so offset within
1415 * the page must be 0. However, the IOMMU may support pages
1416 * smaller than PAGE_SIZE, so s->offset may still represent
1417 * an offset of that boundary within the CPU page.
1419 if (!IS_ALIGNED(s
->offset
, min_pagesz
))
1422 ret
= iommu_map(domain
, iova
+ mapped
, phys
, s
->length
, prot
);
1426 mapped
+= s
->length
;
1432 /* undo mappings already done */
1433 iommu_unmap(domain
, iova
, mapped
);
1438 EXPORT_SYMBOL_GPL(default_iommu_map_sg
);
1440 int iommu_domain_window_enable(struct iommu_domain
*domain
, u32 wnd_nr
,
1441 phys_addr_t paddr
, u64 size
, int prot
)
1443 if (unlikely(domain
->ops
->domain_window_enable
== NULL
))
1446 return domain
->ops
->domain_window_enable(domain
, wnd_nr
, paddr
, size
,
1449 EXPORT_SYMBOL_GPL(iommu_domain_window_enable
);
1451 void iommu_domain_window_disable(struct iommu_domain
*domain
, u32 wnd_nr
)
1453 if (unlikely(domain
->ops
->domain_window_disable
== NULL
))
1456 return domain
->ops
->domain_window_disable(domain
, wnd_nr
);
1458 EXPORT_SYMBOL_GPL(iommu_domain_window_disable
);
1460 static int __init
iommu_init(void)
1462 iommu_group_kset
= kset_create_and_add("iommu_groups",
1464 ida_init(&iommu_group_ida
);
1465 mutex_init(&iommu_group_mutex
);
1467 BUG_ON(!iommu_group_kset
);
1471 core_initcall(iommu_init
);
1473 int iommu_domain_get_attr(struct iommu_domain
*domain
,
1474 enum iommu_attr attr
, void *data
)
1476 struct iommu_domain_geometry
*geometry
;
1482 case DOMAIN_ATTR_GEOMETRY
:
1484 *geometry
= domain
->geometry
;
1487 case DOMAIN_ATTR_PAGING
:
1489 *paging
= (domain
->ops
->pgsize_bitmap
!= 0UL);
1491 case DOMAIN_ATTR_WINDOWS
:
1494 if (domain
->ops
->domain_get_windows
!= NULL
)
1495 *count
= domain
->ops
->domain_get_windows(domain
);
1501 if (!domain
->ops
->domain_get_attr
)
1504 ret
= domain
->ops
->domain_get_attr(domain
, attr
, data
);
1509 EXPORT_SYMBOL_GPL(iommu_domain_get_attr
);
1511 int iommu_domain_set_attr(struct iommu_domain
*domain
,
1512 enum iommu_attr attr
, void *data
)
1518 case DOMAIN_ATTR_WINDOWS
:
1521 if (domain
->ops
->domain_set_windows
!= NULL
)
1522 ret
= domain
->ops
->domain_set_windows(domain
, *count
);
1528 if (domain
->ops
->domain_set_attr
== NULL
)
1531 ret
= domain
->ops
->domain_set_attr(domain
, attr
, data
);
1536 EXPORT_SYMBOL_GPL(iommu_domain_set_attr
);
1538 void iommu_get_dm_regions(struct device
*dev
, struct list_head
*list
)
1540 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1542 if (ops
&& ops
->get_dm_regions
)
1543 ops
->get_dm_regions(dev
, list
);
1546 void iommu_put_dm_regions(struct device
*dev
, struct list_head
*list
)
1548 const struct iommu_ops
*ops
= dev
->bus
->iommu_ops
;
1550 if (ops
&& ops
->put_dm_regions
)
1551 ops
->put_dm_regions(dev
, list
);
1554 /* Request that a device is direct mapped by the IOMMU */
1555 int iommu_request_dm_for_dev(struct device
*dev
)
1557 struct iommu_domain
*dm_domain
;
1558 struct iommu_group
*group
;
1561 /* Device must already be in a group before calling this function */
1562 group
= iommu_group_get_for_dev(dev
);
1564 return PTR_ERR(group
);
1566 mutex_lock(&group
->mutex
);
1568 /* Check if the default domain is already direct mapped */
1570 if (group
->default_domain
&&
1571 group
->default_domain
->type
== IOMMU_DOMAIN_IDENTITY
)
1574 /* Don't change mappings of existing devices */
1576 if (iommu_group_device_count(group
) != 1)
1579 /* Allocate a direct mapped domain */
1581 dm_domain
= __iommu_domain_alloc(dev
->bus
, IOMMU_DOMAIN_IDENTITY
);
1585 /* Attach the device to the domain */
1586 ret
= __iommu_attach_group(dm_domain
, group
);
1588 iommu_domain_free(dm_domain
);
1592 /* Make the direct mapped domain the default for this group */
1593 if (group
->default_domain
)
1594 iommu_domain_free(group
->default_domain
);
1595 group
->default_domain
= dm_domain
;
1597 pr_info("Using direct mapping for device %s\n", dev_name(dev
));
1601 mutex_unlock(&group
->mutex
);
1602 iommu_group_put(group
);