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");
387 goto err_free_device;
389 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
393 goto err_remove_link;
396 ret = sysfs_create_link_nowarn(group->devices_kobj,
397 &dev->kobj, device->name);
399 if (ret == -EEXIST && i >= 0) {
401 * Account for the slim chance of collision
402 * and append an instance to the name.
405 device->name = kasprintf(GFP_KERNEL, "%s.%d",
406 kobject_name(&dev->kobj), i++);
412 kobject_get(group->devices_kobj);
414 dev->iommu_group = group;
416 iommu_group_create_direct_mappings(group, dev);
418 mutex_lock(&group->mutex);
419 list_add_tail(&device->list, &group->devices);
421 ret = __iommu_attach_device(group->domain, dev);
422 mutex_unlock(&group->mutex);
426 /* Notify any listeners about change to group. */
427 blocking_notifier_call_chain(&group->notifier,
428 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
430 trace_add_device_to_group(group->id, dev);
432 pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
437 mutex_lock(&group->mutex);
438 list_del(&device->list);
439 mutex_unlock(&group->mutex);
440 dev->iommu_group = NULL;
441 kobject_put(group->devices_kobj);
445 sysfs_remove_link(&dev->kobj, "iommu_group");
448 pr_err("Failed to add device %s to group %d: %d\n", dev_name(dev), group->id, ret);
451 EXPORT_SYMBOL_GPL(iommu_group_add_device);
454 * iommu_group_remove_device - remove a device from it's current group
455 * @dev: device to be removed
457 * This function is called by an iommu driver to remove the device from
458 * it's current group. This decrements the iommu group reference count.
460 void iommu_group_remove_device(struct device *dev)
462 struct iommu_group *group = dev->iommu_group;
463 struct iommu_device *tmp_device, *device = NULL;
465 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
467 /* Pre-notify listeners that a device is being removed. */
468 blocking_notifier_call_chain(&group->notifier,
469 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
471 mutex_lock(&group->mutex);
472 list_for_each_entry(tmp_device, &group->devices, list) {
473 if (tmp_device->dev == dev) {
475 list_del(&device->list);
479 mutex_unlock(&group->mutex);
484 sysfs_remove_link(group->devices_kobj, device->name);
485 sysfs_remove_link(&dev->kobj, "iommu_group");
487 trace_remove_device_from_group(group->id, dev);
491 dev->iommu_group = NULL;
492 kobject_put(group->devices_kobj);
494 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
496 static int iommu_group_device_count(struct iommu_group *group)
498 struct iommu_device *entry;
501 list_for_each_entry(entry, &group->devices, list)
508 * iommu_group_for_each_dev - iterate over each device in the group
510 * @data: caller opaque data to be passed to callback function
511 * @fn: caller supplied callback function
513 * This function is called by group users to iterate over group devices.
514 * Callers should hold a reference count to the group during callback.
515 * The group->mutex is held across callbacks, which will block calls to
516 * iommu_group_add/remove_device.
518 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
519 int (*fn)(struct device *, void *))
521 struct iommu_device *device;
524 list_for_each_entry(device, &group->devices, list) {
525 ret = fn(device->dev, data);
533 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
534 int (*fn)(struct device *, void *))
538 mutex_lock(&group->mutex);
539 ret = __iommu_group_for_each_dev(group, data, fn);
540 mutex_unlock(&group->mutex);
544 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
547 * iommu_group_get - Return the group for a device and increment reference
548 * @dev: get the group that this device belongs to
550 * This function is called by iommu drivers and users to get the group
551 * for the specified device. If found, the group is returned and the group
552 * reference in incremented, else NULL.
554 struct iommu_group *iommu_group_get(struct device *dev)
556 struct iommu_group *group = dev->iommu_group;
559 kobject_get(group->devices_kobj);
563 EXPORT_SYMBOL_GPL(iommu_group_get);
566 * iommu_group_ref_get - Increment reference on a group
567 * @group: the group to use, must not be NULL
569 * This function is called by iommu drivers to take additional references on an
570 * existing group. Returns the given group for convenience.
572 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
574 kobject_get(group->devices_kobj);
579 * iommu_group_put - Decrement group reference
580 * @group: the group to use
582 * This function is called by iommu drivers and users to release the
583 * iommu group. Once the reference count is zero, the group is released.
585 void iommu_group_put(struct iommu_group *group)
588 kobject_put(group->devices_kobj);
590 EXPORT_SYMBOL_GPL(iommu_group_put);
593 * iommu_group_register_notifier - Register a notifier for group changes
594 * @group: the group to watch
595 * @nb: notifier block to signal
597 * This function allows iommu group users to track changes in a group.
598 * See include/linux/iommu.h for actions sent via this notifier. Caller
599 * should hold a reference to the group throughout notifier registration.
601 int iommu_group_register_notifier(struct iommu_group *group,
602 struct notifier_block *nb)
604 return blocking_notifier_chain_register(&group->notifier, nb);
606 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
609 * iommu_group_unregister_notifier - Unregister a notifier
610 * @group: the group to watch
611 * @nb: notifier block to signal
613 * Unregister a previously registered group notifier block.
615 int iommu_group_unregister_notifier(struct iommu_group *group,
616 struct notifier_block *nb)
618 return blocking_notifier_chain_unregister(&group->notifier, nb);
620 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
623 * iommu_group_id - Return ID for a group
624 * @group: the group to ID
626 * Return the unique ID for the group matching the sysfs group number.
628 int iommu_group_id(struct iommu_group *group)
632 EXPORT_SYMBOL_GPL(iommu_group_id);
634 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
635 unsigned long *devfns);
638 * To consider a PCI device isolated, we require ACS to support Source
639 * Validation, Request Redirection, Completer Redirection, and Upstream
640 * Forwarding. This effectively means that devices cannot spoof their
641 * requester ID, requests and completions cannot be redirected, and all
642 * transactions are forwarded upstream, even as it passes through a
643 * bridge where the target device is downstream.
645 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
648 * For multifunction devices which are not isolated from each other, find
649 * all the other non-isolated functions and look for existing groups. For
650 * each function, we also need to look for aliases to or from other devices
651 * that may already have a group.
653 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
654 unsigned long *devfns)
656 struct pci_dev *tmp = NULL;
657 struct iommu_group *group;
659 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
662 for_each_pci_dev(tmp) {
663 if (tmp == pdev || tmp->bus != pdev->bus ||
664 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
665 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
668 group = get_pci_alias_group(tmp, devfns);
679 * Look for aliases to or from the given device for existing groups. DMA
680 * aliases are only supported on the same bus, therefore the search
681 * space is quite small (especially since we're really only looking at pcie
682 * device, and therefore only expect multiple slots on the root complex or
683 * downstream switch ports). It's conceivable though that a pair of
684 * multifunction devices could have aliases between them that would cause a
685 * loop. To prevent this, we use a bitmap to track where we've been.
687 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
688 unsigned long *devfns)
690 struct pci_dev *tmp = NULL;
691 struct iommu_group *group;
693 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
696 group = iommu_group_get(&pdev->dev);
700 for_each_pci_dev(tmp) {
701 if (tmp == pdev || tmp->bus != pdev->bus)
704 /* We alias them or they alias us */
705 if (pci_devs_are_dma_aliases(pdev, tmp)) {
706 group = get_pci_alias_group(tmp, devfns);
712 group = get_pci_function_alias_group(tmp, devfns);
723 struct group_for_pci_data {
724 struct pci_dev *pdev;
725 struct iommu_group *group;
729 * DMA alias iterator callback, return the last seen device. Stop and return
730 * the IOMMU group if we find one along the way.
732 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
734 struct group_for_pci_data *data = opaque;
737 data->group = iommu_group_get(&pdev->dev);
739 return data->group != NULL;
743 * Generic device_group call-back function. It just allocates one
744 * iommu-group per device.
746 struct iommu_group *generic_device_group(struct device *dev)
748 struct iommu_group *group;
750 group = iommu_group_alloc();
758 * Use standard PCI bus topology, isolation features, and DMA alias quirks
759 * to find or create an IOMMU group for a device.
761 struct iommu_group *pci_device_group(struct device *dev)
763 struct pci_dev *pdev = to_pci_dev(dev);
764 struct group_for_pci_data data;
766 struct iommu_group *group = NULL;
767 u64 devfns[4] = { 0 };
769 if (WARN_ON(!dev_is_pci(dev)))
770 return ERR_PTR(-EINVAL);
773 * Find the upstream DMA alias for the device. A device must not
774 * be aliased due to topology in order to have its own IOMMU group.
775 * If we find an alias along the way that already belongs to a
778 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
784 * Continue upstream from the point of minimum IOMMU granularity
785 * due to aliases to the point where devices are protected from
786 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
789 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
793 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
798 group = iommu_group_get(&pdev->dev);
804 * Look for existing groups on device aliases. If we alias another
805 * device or another device aliases us, use the same group.
807 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
812 * Look for existing groups on non-isolated functions on the same
813 * slot and aliases of those funcions, if any. No need to clear
814 * the search bitmap, the tested devfns are still valid.
816 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
820 /* No shared group found, allocate new */
821 group = iommu_group_alloc();
829 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
830 * @dev: target device
832 * This function is intended to be called by IOMMU drivers and extended to
833 * support common, bus-defined algorithms when determining or creating the
834 * IOMMU group for a device. On success, the caller will hold a reference
835 * to the returned IOMMU group, which will already include the provided
836 * device. The reference should be released with iommu_group_put().
838 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
840 const struct iommu_ops *ops = dev->bus->iommu_ops;
841 struct iommu_group *group;
844 group = iommu_group_get(dev);
848 group = ERR_PTR(-EINVAL);
850 if (ops && ops->device_group)
851 group = ops->device_group(dev);
857 * Try to allocate a default domain - needs support from the
860 if (!group->default_domain) {
861 group->default_domain = __iommu_domain_alloc(dev->bus,
864 group->domain = group->default_domain;
867 ret = iommu_group_add_device(group, dev);
869 iommu_group_put(group);
876 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
878 return group->default_domain;
881 static int add_iommu_group(struct device *dev, void *data)
883 struct iommu_callback_data *cb = data;
884 const struct iommu_ops *ops = cb->ops;
887 if (!ops->add_device)
890 WARN_ON(dev->iommu_group);
892 ret = ops->add_device(dev);
895 * We ignore -ENODEV errors for now, as they just mean that the
896 * device is not translated by an IOMMU. We still care about
897 * other errors and fail to initialize when they happen.
905 static int remove_iommu_group(struct device *dev, void *data)
907 struct iommu_callback_data *cb = data;
908 const struct iommu_ops *ops = cb->ops;
910 if (ops->remove_device && dev->iommu_group)
911 ops->remove_device(dev);
916 static int iommu_bus_notifier(struct notifier_block *nb,
917 unsigned long action, void *data)
919 struct device *dev = data;
920 const struct iommu_ops *ops = dev->bus->iommu_ops;
921 struct iommu_group *group;
922 unsigned long group_action = 0;
925 * ADD/DEL call into iommu driver ops if provided, which may
926 * result in ADD/DEL notifiers to group->notifier
928 if (action == BUS_NOTIFY_ADD_DEVICE) {
930 return ops->add_device(dev);
931 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
932 if (ops->remove_device && dev->iommu_group) {
933 ops->remove_device(dev);
939 * Remaining BUS_NOTIFYs get filtered and republished to the
940 * group, if anyone is listening
942 group = iommu_group_get(dev);
947 case BUS_NOTIFY_BIND_DRIVER:
948 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
950 case BUS_NOTIFY_BOUND_DRIVER:
951 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
953 case BUS_NOTIFY_UNBIND_DRIVER:
954 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
956 case BUS_NOTIFY_UNBOUND_DRIVER:
957 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
962 blocking_notifier_call_chain(&group->notifier,
965 iommu_group_put(group);
969 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
972 struct notifier_block *nb;
973 struct iommu_callback_data cb = {
977 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
981 nb->notifier_call = iommu_bus_notifier;
983 err = bus_register_notifier(bus, nb);
987 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
996 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
997 bus_unregister_notifier(bus, nb);
1006 * bus_set_iommu - set iommu-callbacks for the bus
1008 * @ops: the callbacks provided by the iommu-driver
1010 * This function is called by an iommu driver to set the iommu methods
1011 * used for a particular bus. Drivers for devices on that bus can use
1012 * the iommu-api after these ops are registered.
1013 * This special function is needed because IOMMUs are usually devices on
1014 * the bus itself, so the iommu drivers are not initialized when the bus
1015 * is set up. With this function the iommu-driver can set the iommu-ops
1018 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1022 if (bus->iommu_ops != NULL)
1025 bus->iommu_ops = ops;
1027 /* Do IOMMU specific setup for this bus-type */
1028 err = iommu_bus_init(bus, ops);
1030 bus->iommu_ops = NULL;
1034 EXPORT_SYMBOL_GPL(bus_set_iommu);
1036 bool iommu_present(struct bus_type *bus)
1038 return bus->iommu_ops != NULL;
1040 EXPORT_SYMBOL_GPL(iommu_present);
1042 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1044 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1047 return bus->iommu_ops->capable(cap);
1049 EXPORT_SYMBOL_GPL(iommu_capable);
1052 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1053 * @domain: iommu domain
1054 * @handler: fault handler
1055 * @token: user data, will be passed back to the fault handler
1057 * This function should be used by IOMMU users which want to be notified
1058 * whenever an IOMMU fault happens.
1060 * The fault handler itself should return 0 on success, and an appropriate
1061 * error code otherwise.
1063 void iommu_set_fault_handler(struct iommu_domain *domain,
1064 iommu_fault_handler_t handler,
1069 domain->handler = handler;
1070 domain->handler_token = token;
1072 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1074 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1077 struct iommu_domain *domain;
1079 if (bus == NULL || bus->iommu_ops == NULL)
1082 domain = bus->iommu_ops->domain_alloc(type);
1086 domain->ops = bus->iommu_ops;
1087 domain->type = type;
1088 /* Assume all sizes by default; the driver may override this later */
1089 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1094 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1096 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1098 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1100 void iommu_domain_free(struct iommu_domain *domain)
1102 domain->ops->domain_free(domain);
1104 EXPORT_SYMBOL_GPL(iommu_domain_free);
1106 static int __iommu_attach_device(struct iommu_domain *domain,
1110 if (unlikely(domain->ops->attach_dev == NULL))
1113 ret = domain->ops->attach_dev(domain, dev);
1115 trace_attach_device_to_domain(dev);
1119 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1121 struct iommu_group *group;
1124 group = iommu_group_get(dev);
1125 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1127 return __iommu_attach_device(domain, dev);
1130 * We have a group - lock it to make sure the device-count doesn't
1131 * change while we are attaching
1133 mutex_lock(&group->mutex);
1135 if (iommu_group_device_count(group) != 1)
1138 ret = __iommu_attach_group(domain, group);
1141 mutex_unlock(&group->mutex);
1142 iommu_group_put(group);
1146 EXPORT_SYMBOL_GPL(iommu_attach_device);
1148 static void __iommu_detach_device(struct iommu_domain *domain,
1151 if (unlikely(domain->ops->detach_dev == NULL))
1154 domain->ops->detach_dev(domain, dev);
1155 trace_detach_device_from_domain(dev);
1158 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1160 struct iommu_group *group;
1162 group = iommu_group_get(dev);
1163 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1165 return __iommu_detach_device(domain, dev);
1167 mutex_lock(&group->mutex);
1168 if (iommu_group_device_count(group) != 1) {
1173 __iommu_detach_group(domain, group);
1176 mutex_unlock(&group->mutex);
1177 iommu_group_put(group);
1179 EXPORT_SYMBOL_GPL(iommu_detach_device);
1181 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1183 struct iommu_domain *domain;
1184 struct iommu_group *group;
1186 group = iommu_group_get(dev);
1187 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1191 domain = group->domain;
1193 iommu_group_put(group);
1197 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1200 * IOMMU groups are really the natrual working unit of the IOMMU, but
1201 * the IOMMU API works on domains and devices. Bridge that gap by
1202 * iterating over the devices in a group. Ideally we'd have a single
1203 * device which represents the requestor ID of the group, but we also
1204 * allow IOMMU drivers to create policy defined minimum sets, where
1205 * the physical hardware may be able to distiguish members, but we
1206 * wish to group them at a higher level (ex. untrusted multi-function
1207 * PCI devices). Thus we attach each device.
1209 static int iommu_group_do_attach_device(struct device *dev, void *data)
1211 struct iommu_domain *domain = data;
1213 return __iommu_attach_device(domain, dev);
1216 static int __iommu_attach_group(struct iommu_domain *domain,
1217 struct iommu_group *group)
1221 if (group->default_domain && group->domain != group->default_domain)
1224 ret = __iommu_group_for_each_dev(group, domain,
1225 iommu_group_do_attach_device);
1227 group->domain = domain;
1232 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1236 mutex_lock(&group->mutex);
1237 ret = __iommu_attach_group(domain, group);
1238 mutex_unlock(&group->mutex);
1242 EXPORT_SYMBOL_GPL(iommu_attach_group);
1244 static int iommu_group_do_detach_device(struct device *dev, void *data)
1246 struct iommu_domain *domain = data;
1248 __iommu_detach_device(domain, dev);
1253 static void __iommu_detach_group(struct iommu_domain *domain,
1254 struct iommu_group *group)
1258 if (!group->default_domain) {
1259 __iommu_group_for_each_dev(group, domain,
1260 iommu_group_do_detach_device);
1261 group->domain = NULL;
1265 if (group->domain == group->default_domain)
1268 /* Detach by re-attaching to the default domain */
1269 ret = __iommu_group_for_each_dev(group, group->default_domain,
1270 iommu_group_do_attach_device);
1274 group->domain = group->default_domain;
1277 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1279 mutex_lock(&group->mutex);
1280 __iommu_detach_group(domain, group);
1281 mutex_unlock(&group->mutex);
1283 EXPORT_SYMBOL_GPL(iommu_detach_group);
1285 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1287 if (unlikely(domain->ops->iova_to_phys == NULL))
1290 return domain->ops->iova_to_phys(domain, iova);
1292 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1294 static size_t iommu_pgsize(struct iommu_domain *domain,
1295 unsigned long addr_merge, size_t size)
1297 unsigned int pgsize_idx;
1300 /* Max page size that still fits into 'size' */
1301 pgsize_idx = __fls(size);
1303 /* need to consider alignment requirements ? */
1304 if (likely(addr_merge)) {
1305 /* Max page size allowed by address */
1306 unsigned int align_pgsize_idx = __ffs(addr_merge);
1307 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1310 /* build a mask of acceptable page sizes */
1311 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1313 /* throw away page sizes not supported by the hardware */
1314 pgsize &= domain->pgsize_bitmap;
1316 /* make sure we're still sane */
1319 /* pick the biggest page */
1320 pgsize_idx = __fls(pgsize);
1321 pgsize = 1UL << pgsize_idx;
1326 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1327 phys_addr_t paddr, size_t size, int prot)
1329 unsigned long orig_iova = iova;
1330 unsigned int min_pagesz;
1331 size_t orig_size = size;
1332 phys_addr_t orig_paddr = paddr;
1335 if (unlikely(domain->ops->map == NULL ||
1336 domain->pgsize_bitmap == 0UL))
1339 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1342 /* find out the minimum page size supported */
1343 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1346 * both the virtual address and the physical one, as well as
1347 * the size of the mapping, must be aligned (at least) to the
1348 * size of the smallest page supported by the hardware
1350 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1351 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1352 iova, &paddr, size, min_pagesz);
1356 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1359 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1361 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1362 iova, &paddr, pgsize);
1364 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1373 /* unroll mapping in case something went wrong */
1375 iommu_unmap(domain, orig_iova, orig_size - size);
1377 trace_map(orig_iova, orig_paddr, orig_size);
1381 EXPORT_SYMBOL_GPL(iommu_map);
1383 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1385 size_t unmapped_page, unmapped = 0;
1386 unsigned int min_pagesz;
1387 unsigned long orig_iova = iova;
1389 if (unlikely(domain->ops->unmap == NULL ||
1390 domain->pgsize_bitmap == 0UL))
1393 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1396 /* find out the minimum page size supported */
1397 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1400 * The virtual address, as well as the size of the mapping, must be
1401 * aligned (at least) to the size of the smallest page supported
1404 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1405 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1406 iova, size, min_pagesz);
1410 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1413 * Keep iterating until we either unmap 'size' bytes (or more)
1414 * or we hit an area that isn't mapped.
1416 while (unmapped < size) {
1417 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1419 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1423 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1424 iova, unmapped_page);
1426 iova += unmapped_page;
1427 unmapped += unmapped_page;
1430 trace_unmap(orig_iova, size, unmapped);
1433 EXPORT_SYMBOL_GPL(iommu_unmap);
1435 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1436 struct scatterlist *sg, unsigned int nents, int prot)
1438 struct scatterlist *s;
1440 unsigned int i, min_pagesz;
1443 if (unlikely(domain->pgsize_bitmap == 0UL))
1446 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1448 for_each_sg(sg, s, nents, i) {
1449 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1452 * We are mapping on IOMMU page boundaries, so offset within
1453 * the page must be 0. However, the IOMMU may support pages
1454 * smaller than PAGE_SIZE, so s->offset may still represent
1455 * an offset of that boundary within the CPU page.
1457 if (!IS_ALIGNED(s->offset, min_pagesz))
1460 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1464 mapped += s->length;
1470 /* undo mappings already done */
1471 iommu_unmap(domain, iova, mapped);
1476 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1478 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1479 phys_addr_t paddr, u64 size, int prot)
1481 if (unlikely(domain->ops->domain_window_enable == NULL))
1484 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1487 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1489 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1491 if (unlikely(domain->ops->domain_window_disable == NULL))
1494 return domain->ops->domain_window_disable(domain, wnd_nr);
1496 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1498 static int __init iommu_init(void)
1500 iommu_group_kset = kset_create_and_add("iommu_groups",
1502 BUG_ON(!iommu_group_kset);
1506 core_initcall(iommu_init);
1508 int iommu_domain_get_attr(struct iommu_domain *domain,
1509 enum iommu_attr attr, void *data)
1511 struct iommu_domain_geometry *geometry;
1517 case DOMAIN_ATTR_GEOMETRY:
1519 *geometry = domain->geometry;
1522 case DOMAIN_ATTR_PAGING:
1524 *paging = (domain->pgsize_bitmap != 0UL);
1526 case DOMAIN_ATTR_WINDOWS:
1529 if (domain->ops->domain_get_windows != NULL)
1530 *count = domain->ops->domain_get_windows(domain);
1536 if (!domain->ops->domain_get_attr)
1539 ret = domain->ops->domain_get_attr(domain, attr, data);
1544 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1546 int iommu_domain_set_attr(struct iommu_domain *domain,
1547 enum iommu_attr attr, void *data)
1553 case DOMAIN_ATTR_WINDOWS:
1556 if (domain->ops->domain_set_windows != NULL)
1557 ret = domain->ops->domain_set_windows(domain, *count);
1563 if (domain->ops->domain_set_attr == NULL)
1566 ret = domain->ops->domain_set_attr(domain, attr, data);
1571 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1573 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1575 const struct iommu_ops *ops = dev->bus->iommu_ops;
1577 if (ops && ops->get_dm_regions)
1578 ops->get_dm_regions(dev, list);
1581 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1583 const struct iommu_ops *ops = dev->bus->iommu_ops;
1585 if (ops && ops->put_dm_regions)
1586 ops->put_dm_regions(dev, list);
1589 /* Request that a device is direct mapped by the IOMMU */
1590 int iommu_request_dm_for_dev(struct device *dev)
1592 struct iommu_domain *dm_domain;
1593 struct iommu_group *group;
1596 /* Device must already be in a group before calling this function */
1597 group = iommu_group_get_for_dev(dev);
1599 return PTR_ERR(group);
1601 mutex_lock(&group->mutex);
1603 /* Check if the default domain is already direct mapped */
1605 if (group->default_domain &&
1606 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1609 /* Don't change mappings of existing devices */
1611 if (iommu_group_device_count(group) != 1)
1614 /* Allocate a direct mapped domain */
1616 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1620 /* Attach the device to the domain */
1621 ret = __iommu_attach_group(dm_domain, group);
1623 iommu_domain_free(dm_domain);
1627 /* Make the direct mapped domain the default for this group */
1628 if (group->default_domain)
1629 iommu_domain_free(group->default_domain);
1630 group->default_domain = dm_domain;
1632 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1636 mutex_unlock(&group->mutex);
1637 iommu_group_put(group);
1642 struct iommu_instance {
1643 struct list_head list;
1644 struct fwnode_handle *fwnode;
1645 const struct iommu_ops *ops;
1647 static LIST_HEAD(iommu_instance_list);
1648 static DEFINE_SPINLOCK(iommu_instance_lock);
1650 void iommu_register_instance(struct fwnode_handle *fwnode,
1651 const struct iommu_ops *ops)
1653 struct iommu_instance *iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1655 if (WARN_ON(!iommu))
1658 of_node_get(to_of_node(fwnode));
1659 INIT_LIST_HEAD(&iommu->list);
1660 iommu->fwnode = fwnode;
1662 spin_lock(&iommu_instance_lock);
1663 list_add_tail(&iommu->list, &iommu_instance_list);
1664 spin_unlock(&iommu_instance_lock);
1667 const struct iommu_ops *iommu_get_instance(struct fwnode_handle *fwnode)
1669 struct iommu_instance *instance;
1670 const struct iommu_ops *ops = NULL;
1672 spin_lock(&iommu_instance_lock);
1673 list_for_each_entry(instance, &iommu_instance_list, list)
1674 if (instance->fwnode == fwnode) {
1675 ops = instance->ops;
1678 spin_unlock(&iommu_instance_lock);
1682 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1683 const struct iommu_ops *ops)
1685 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1688 return ops == fwspec->ops ? 0 : -EINVAL;
1690 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1694 of_node_get(to_of_node(iommu_fwnode));
1695 fwspec->iommu_fwnode = iommu_fwnode;
1697 dev->iommu_fwspec = fwspec;
1700 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1702 void iommu_fwspec_free(struct device *dev)
1704 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1707 fwnode_handle_put(fwspec->iommu_fwnode);
1709 dev->iommu_fwspec = NULL;
1712 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1714 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1716 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1723 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1724 if (size > sizeof(*fwspec)) {
1725 fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1730 for (i = 0; i < num_ids; i++)
1731 fwspec->ids[fwspec->num_ids + i] = ids[i];
1733 fwspec->num_ids += num_ids;
1734 dev->iommu_fwspec = fwspec;
1737 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);