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_put - Decrement group reference
567 * @group: the group to use
569 * This function is called by iommu drivers and users to release the
570 * iommu group. Once the reference count is zero, the group is released.
572 void iommu_group_put(struct iommu_group *group)
575 kobject_put(group->devices_kobj);
577 EXPORT_SYMBOL_GPL(iommu_group_put);
580 * iommu_group_register_notifier - Register a notifier for group changes
581 * @group: the group to watch
582 * @nb: notifier block to signal
584 * This function allows iommu group users to track changes in a group.
585 * See include/linux/iommu.h for actions sent via this notifier. Caller
586 * should hold a reference to the group throughout notifier registration.
588 int iommu_group_register_notifier(struct iommu_group *group,
589 struct notifier_block *nb)
591 return blocking_notifier_chain_register(&group->notifier, nb);
593 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
596 * iommu_group_unregister_notifier - Unregister a notifier
597 * @group: the group to watch
598 * @nb: notifier block to signal
600 * Unregister a previously registered group notifier block.
602 int iommu_group_unregister_notifier(struct iommu_group *group,
603 struct notifier_block *nb)
605 return blocking_notifier_chain_unregister(&group->notifier, nb);
607 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
610 * iommu_group_id - Return ID for a group
611 * @group: the group to ID
613 * Return the unique ID for the group matching the sysfs group number.
615 int iommu_group_id(struct iommu_group *group)
619 EXPORT_SYMBOL_GPL(iommu_group_id);
621 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
622 unsigned long *devfns);
625 * To consider a PCI device isolated, we require ACS to support Source
626 * Validation, Request Redirection, Completer Redirection, and Upstream
627 * Forwarding. This effectively means that devices cannot spoof their
628 * requester ID, requests and completions cannot be redirected, and all
629 * transactions are forwarded upstream, even as it passes through a
630 * bridge where the target device is downstream.
632 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
635 * For multifunction devices which are not isolated from each other, find
636 * all the other non-isolated functions and look for existing groups. For
637 * each function, we also need to look for aliases to or from other devices
638 * that may already have a group.
640 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
641 unsigned long *devfns)
643 struct pci_dev *tmp = NULL;
644 struct iommu_group *group;
646 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
649 for_each_pci_dev(tmp) {
650 if (tmp == pdev || tmp->bus != pdev->bus ||
651 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
652 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
655 group = get_pci_alias_group(tmp, devfns);
666 * Look for aliases to or from the given device for existing groups. DMA
667 * aliases are only supported on the same bus, therefore the search
668 * space is quite small (especially since we're really only looking at pcie
669 * device, and therefore only expect multiple slots on the root complex or
670 * downstream switch ports). It's conceivable though that a pair of
671 * multifunction devices could have aliases between them that would cause a
672 * loop. To prevent this, we use a bitmap to track where we've been.
674 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
675 unsigned long *devfns)
677 struct pci_dev *tmp = NULL;
678 struct iommu_group *group;
680 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
683 group = iommu_group_get(&pdev->dev);
687 for_each_pci_dev(tmp) {
688 if (tmp == pdev || tmp->bus != pdev->bus)
691 /* We alias them or they alias us */
692 if (pci_devs_are_dma_aliases(pdev, tmp)) {
693 group = get_pci_alias_group(tmp, devfns);
699 group = get_pci_function_alias_group(tmp, devfns);
710 struct group_for_pci_data {
711 struct pci_dev *pdev;
712 struct iommu_group *group;
716 * DMA alias iterator callback, return the last seen device. Stop and return
717 * the IOMMU group if we find one along the way.
719 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
721 struct group_for_pci_data *data = opaque;
724 data->group = iommu_group_get(&pdev->dev);
726 return data->group != NULL;
730 * Generic device_group call-back function. It just allocates one
731 * iommu-group per device.
733 struct iommu_group *generic_device_group(struct device *dev)
735 struct iommu_group *group;
737 group = iommu_group_alloc();
745 * Use standard PCI bus topology, isolation features, and DMA alias quirks
746 * to find or create an IOMMU group for a device.
748 struct iommu_group *pci_device_group(struct device *dev)
750 struct pci_dev *pdev = to_pci_dev(dev);
751 struct group_for_pci_data data;
753 struct iommu_group *group = NULL;
754 u64 devfns[4] = { 0 };
756 if (WARN_ON(!dev_is_pci(dev)))
757 return ERR_PTR(-EINVAL);
760 * Find the upstream DMA alias for the device. A device must not
761 * be aliased due to topology in order to have its own IOMMU group.
762 * If we find an alias along the way that already belongs to a
765 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
771 * Continue upstream from the point of minimum IOMMU granularity
772 * due to aliases to the point where devices are protected from
773 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
776 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
780 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
785 group = iommu_group_get(&pdev->dev);
791 * Look for existing groups on device aliases. If we alias another
792 * device or another device aliases us, use the same group.
794 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
799 * Look for existing groups on non-isolated functions on the same
800 * slot and aliases of those funcions, if any. No need to clear
801 * the search bitmap, the tested devfns are still valid.
803 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
807 /* No shared group found, allocate new */
808 group = iommu_group_alloc();
816 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
817 * @dev: target device
819 * This function is intended to be called by IOMMU drivers and extended to
820 * support common, bus-defined algorithms when determining or creating the
821 * IOMMU group for a device. On success, the caller will hold a reference
822 * to the returned IOMMU group, which will already include the provided
823 * device. The reference should be released with iommu_group_put().
825 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
827 const struct iommu_ops *ops = dev->bus->iommu_ops;
828 struct iommu_group *group;
831 group = iommu_group_get(dev);
835 group = ERR_PTR(-EINVAL);
837 if (ops && ops->device_group)
838 group = ops->device_group(dev);
844 * Try to allocate a default domain - needs support from the
847 if (!group->default_domain) {
848 group->default_domain = __iommu_domain_alloc(dev->bus,
851 group->domain = group->default_domain;
854 ret = iommu_group_add_device(group, dev);
856 iommu_group_put(group);
863 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
865 return group->default_domain;
868 static int add_iommu_group(struct device *dev, void *data)
870 struct iommu_callback_data *cb = data;
871 const struct iommu_ops *ops = cb->ops;
874 if (!ops->add_device)
877 WARN_ON(dev->iommu_group);
879 ret = ops->add_device(dev);
882 * We ignore -ENODEV errors for now, as they just mean that the
883 * device is not translated by an IOMMU. We still care about
884 * other errors and fail to initialize when they happen.
892 static int remove_iommu_group(struct device *dev, void *data)
894 struct iommu_callback_data *cb = data;
895 const struct iommu_ops *ops = cb->ops;
897 if (ops->remove_device && dev->iommu_group)
898 ops->remove_device(dev);
903 static int iommu_bus_notifier(struct notifier_block *nb,
904 unsigned long action, void *data)
906 struct device *dev = data;
907 const struct iommu_ops *ops = dev->bus->iommu_ops;
908 struct iommu_group *group;
909 unsigned long group_action = 0;
912 * ADD/DEL call into iommu driver ops if provided, which may
913 * result in ADD/DEL notifiers to group->notifier
915 if (action == BUS_NOTIFY_ADD_DEVICE) {
917 return ops->add_device(dev);
918 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
919 if (ops->remove_device && dev->iommu_group) {
920 ops->remove_device(dev);
926 * Remaining BUS_NOTIFYs get filtered and republished to the
927 * group, if anyone is listening
929 group = iommu_group_get(dev);
934 case BUS_NOTIFY_BIND_DRIVER:
935 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
937 case BUS_NOTIFY_BOUND_DRIVER:
938 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
940 case BUS_NOTIFY_UNBIND_DRIVER:
941 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
943 case BUS_NOTIFY_UNBOUND_DRIVER:
944 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
949 blocking_notifier_call_chain(&group->notifier,
952 iommu_group_put(group);
956 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
959 struct notifier_block *nb;
960 struct iommu_callback_data cb = {
964 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
968 nb->notifier_call = iommu_bus_notifier;
970 err = bus_register_notifier(bus, nb);
974 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
983 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
984 bus_unregister_notifier(bus, nb);
993 * bus_set_iommu - set iommu-callbacks for the bus
995 * @ops: the callbacks provided by the iommu-driver
997 * This function is called by an iommu driver to set the iommu methods
998 * used for a particular bus. Drivers for devices on that bus can use
999 * the iommu-api after these ops are registered.
1000 * This special function is needed because IOMMUs are usually devices on
1001 * the bus itself, so the iommu drivers are not initialized when the bus
1002 * is set up. With this function the iommu-driver can set the iommu-ops
1005 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1009 if (bus->iommu_ops != NULL)
1012 bus->iommu_ops = ops;
1014 /* Do IOMMU specific setup for this bus-type */
1015 err = iommu_bus_init(bus, ops);
1017 bus->iommu_ops = NULL;
1021 EXPORT_SYMBOL_GPL(bus_set_iommu);
1023 bool iommu_present(struct bus_type *bus)
1025 return bus->iommu_ops != NULL;
1027 EXPORT_SYMBOL_GPL(iommu_present);
1029 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1031 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1034 return bus->iommu_ops->capable(cap);
1036 EXPORT_SYMBOL_GPL(iommu_capable);
1039 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1040 * @domain: iommu domain
1041 * @handler: fault handler
1042 * @token: user data, will be passed back to the fault handler
1044 * This function should be used by IOMMU users which want to be notified
1045 * whenever an IOMMU fault happens.
1047 * The fault handler itself should return 0 on success, and an appropriate
1048 * error code otherwise.
1050 void iommu_set_fault_handler(struct iommu_domain *domain,
1051 iommu_fault_handler_t handler,
1056 domain->handler = handler;
1057 domain->handler_token = token;
1059 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1061 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1064 struct iommu_domain *domain;
1066 if (bus == NULL || bus->iommu_ops == NULL)
1069 domain = bus->iommu_ops->domain_alloc(type);
1073 domain->ops = bus->iommu_ops;
1074 domain->type = type;
1075 /* Assume all sizes by default; the driver may override this later */
1076 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1081 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1083 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1085 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1087 void iommu_domain_free(struct iommu_domain *domain)
1089 domain->ops->domain_free(domain);
1091 EXPORT_SYMBOL_GPL(iommu_domain_free);
1093 static int __iommu_attach_device(struct iommu_domain *domain,
1097 if (unlikely(domain->ops->attach_dev == NULL))
1100 ret = domain->ops->attach_dev(domain, dev);
1102 trace_attach_device_to_domain(dev);
1106 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1108 struct iommu_group *group;
1111 group = iommu_group_get(dev);
1112 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1114 return __iommu_attach_device(domain, dev);
1117 * We have a group - lock it to make sure the device-count doesn't
1118 * change while we are attaching
1120 mutex_lock(&group->mutex);
1122 if (iommu_group_device_count(group) != 1)
1125 ret = __iommu_attach_group(domain, group);
1128 mutex_unlock(&group->mutex);
1129 iommu_group_put(group);
1133 EXPORT_SYMBOL_GPL(iommu_attach_device);
1135 static void __iommu_detach_device(struct iommu_domain *domain,
1138 if (unlikely(domain->ops->detach_dev == NULL))
1141 domain->ops->detach_dev(domain, dev);
1142 trace_detach_device_from_domain(dev);
1145 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1147 struct iommu_group *group;
1149 group = iommu_group_get(dev);
1150 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1152 return __iommu_detach_device(domain, dev);
1154 mutex_lock(&group->mutex);
1155 if (iommu_group_device_count(group) != 1) {
1160 __iommu_detach_group(domain, group);
1163 mutex_unlock(&group->mutex);
1164 iommu_group_put(group);
1166 EXPORT_SYMBOL_GPL(iommu_detach_device);
1168 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1170 struct iommu_domain *domain;
1171 struct iommu_group *group;
1173 group = iommu_group_get(dev);
1174 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1178 domain = group->domain;
1180 iommu_group_put(group);
1184 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1187 * IOMMU groups are really the natrual working unit of the IOMMU, but
1188 * the IOMMU API works on domains and devices. Bridge that gap by
1189 * iterating over the devices in a group. Ideally we'd have a single
1190 * device which represents the requestor ID of the group, but we also
1191 * allow IOMMU drivers to create policy defined minimum sets, where
1192 * the physical hardware may be able to distiguish members, but we
1193 * wish to group them at a higher level (ex. untrusted multi-function
1194 * PCI devices). Thus we attach each device.
1196 static int iommu_group_do_attach_device(struct device *dev, void *data)
1198 struct iommu_domain *domain = data;
1200 return __iommu_attach_device(domain, dev);
1203 static int __iommu_attach_group(struct iommu_domain *domain,
1204 struct iommu_group *group)
1208 if (group->default_domain && group->domain != group->default_domain)
1211 ret = __iommu_group_for_each_dev(group, domain,
1212 iommu_group_do_attach_device);
1214 group->domain = domain;
1219 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1223 mutex_lock(&group->mutex);
1224 ret = __iommu_attach_group(domain, group);
1225 mutex_unlock(&group->mutex);
1229 EXPORT_SYMBOL_GPL(iommu_attach_group);
1231 static int iommu_group_do_detach_device(struct device *dev, void *data)
1233 struct iommu_domain *domain = data;
1235 __iommu_detach_device(domain, dev);
1240 static void __iommu_detach_group(struct iommu_domain *domain,
1241 struct iommu_group *group)
1245 if (!group->default_domain) {
1246 __iommu_group_for_each_dev(group, domain,
1247 iommu_group_do_detach_device);
1248 group->domain = NULL;
1252 if (group->domain == group->default_domain)
1255 /* Detach by re-attaching to the default domain */
1256 ret = __iommu_group_for_each_dev(group, group->default_domain,
1257 iommu_group_do_attach_device);
1261 group->domain = group->default_domain;
1264 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1266 mutex_lock(&group->mutex);
1267 __iommu_detach_group(domain, group);
1268 mutex_unlock(&group->mutex);
1270 EXPORT_SYMBOL_GPL(iommu_detach_group);
1272 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1274 if (unlikely(domain->ops->iova_to_phys == NULL))
1277 return domain->ops->iova_to_phys(domain, iova);
1279 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1281 static size_t iommu_pgsize(struct iommu_domain *domain,
1282 unsigned long addr_merge, size_t size)
1284 unsigned int pgsize_idx;
1287 /* Max page size that still fits into 'size' */
1288 pgsize_idx = __fls(size);
1290 /* need to consider alignment requirements ? */
1291 if (likely(addr_merge)) {
1292 /* Max page size allowed by address */
1293 unsigned int align_pgsize_idx = __ffs(addr_merge);
1294 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1297 /* build a mask of acceptable page sizes */
1298 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1300 /* throw away page sizes not supported by the hardware */
1301 pgsize &= domain->pgsize_bitmap;
1303 /* make sure we're still sane */
1306 /* pick the biggest page */
1307 pgsize_idx = __fls(pgsize);
1308 pgsize = 1UL << pgsize_idx;
1313 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1314 phys_addr_t paddr, size_t size, int prot)
1316 unsigned long orig_iova = iova;
1317 unsigned int min_pagesz;
1318 size_t orig_size = size;
1319 phys_addr_t orig_paddr = paddr;
1322 if (unlikely(domain->ops->map == NULL ||
1323 domain->pgsize_bitmap == 0UL))
1326 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1329 /* find out the minimum page size supported */
1330 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1333 * both the virtual address and the physical one, as well as
1334 * the size of the mapping, must be aligned (at least) to the
1335 * size of the smallest page supported by the hardware
1337 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1338 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1339 iova, &paddr, size, min_pagesz);
1343 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1346 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1348 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1349 iova, &paddr, pgsize);
1351 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1360 /* unroll mapping in case something went wrong */
1362 iommu_unmap(domain, orig_iova, orig_size - size);
1364 trace_map(orig_iova, orig_paddr, orig_size);
1368 EXPORT_SYMBOL_GPL(iommu_map);
1370 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1372 size_t unmapped_page, unmapped = 0;
1373 unsigned int min_pagesz;
1374 unsigned long orig_iova = iova;
1376 if (unlikely(domain->ops->unmap == NULL ||
1377 domain->pgsize_bitmap == 0UL))
1380 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1383 /* find out the minimum page size supported */
1384 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1387 * The virtual address, as well as the size of the mapping, must be
1388 * aligned (at least) to the size of the smallest page supported
1391 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1392 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1393 iova, size, min_pagesz);
1397 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1400 * Keep iterating until we either unmap 'size' bytes (or more)
1401 * or we hit an area that isn't mapped.
1403 while (unmapped < size) {
1404 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1406 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1410 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1411 iova, unmapped_page);
1413 iova += unmapped_page;
1414 unmapped += unmapped_page;
1417 trace_unmap(orig_iova, size, unmapped);
1420 EXPORT_SYMBOL_GPL(iommu_unmap);
1422 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1423 struct scatterlist *sg, unsigned int nents, int prot)
1425 struct scatterlist *s;
1427 unsigned int i, min_pagesz;
1430 if (unlikely(domain->pgsize_bitmap == 0UL))
1433 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1435 for_each_sg(sg, s, nents, i) {
1436 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1439 * We are mapping on IOMMU page boundaries, so offset within
1440 * the page must be 0. However, the IOMMU may support pages
1441 * smaller than PAGE_SIZE, so s->offset may still represent
1442 * an offset of that boundary within the CPU page.
1444 if (!IS_ALIGNED(s->offset, min_pagesz))
1447 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1451 mapped += s->length;
1457 /* undo mappings already done */
1458 iommu_unmap(domain, iova, mapped);
1463 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1465 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1466 phys_addr_t paddr, u64 size, int prot)
1468 if (unlikely(domain->ops->domain_window_enable == NULL))
1471 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1474 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1476 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1478 if (unlikely(domain->ops->domain_window_disable == NULL))
1481 return domain->ops->domain_window_disable(domain, wnd_nr);
1483 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1485 static int __init iommu_init(void)
1487 iommu_group_kset = kset_create_and_add("iommu_groups",
1489 BUG_ON(!iommu_group_kset);
1493 core_initcall(iommu_init);
1495 int iommu_domain_get_attr(struct iommu_domain *domain,
1496 enum iommu_attr attr, void *data)
1498 struct iommu_domain_geometry *geometry;
1504 case DOMAIN_ATTR_GEOMETRY:
1506 *geometry = domain->geometry;
1509 case DOMAIN_ATTR_PAGING:
1511 *paging = (domain->pgsize_bitmap != 0UL);
1513 case DOMAIN_ATTR_WINDOWS:
1516 if (domain->ops->domain_get_windows != NULL)
1517 *count = domain->ops->domain_get_windows(domain);
1523 if (!domain->ops->domain_get_attr)
1526 ret = domain->ops->domain_get_attr(domain, attr, data);
1531 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1533 int iommu_domain_set_attr(struct iommu_domain *domain,
1534 enum iommu_attr attr, void *data)
1540 case DOMAIN_ATTR_WINDOWS:
1543 if (domain->ops->domain_set_windows != NULL)
1544 ret = domain->ops->domain_set_windows(domain, *count);
1550 if (domain->ops->domain_set_attr == NULL)
1553 ret = domain->ops->domain_set_attr(domain, attr, data);
1558 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1560 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1562 const struct iommu_ops *ops = dev->bus->iommu_ops;
1564 if (ops && ops->get_dm_regions)
1565 ops->get_dm_regions(dev, list);
1568 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1570 const struct iommu_ops *ops = dev->bus->iommu_ops;
1572 if (ops && ops->put_dm_regions)
1573 ops->put_dm_regions(dev, list);
1576 /* Request that a device is direct mapped by the IOMMU */
1577 int iommu_request_dm_for_dev(struct device *dev)
1579 struct iommu_domain *dm_domain;
1580 struct iommu_group *group;
1583 /* Device must already be in a group before calling this function */
1584 group = iommu_group_get_for_dev(dev);
1586 return PTR_ERR(group);
1588 mutex_lock(&group->mutex);
1590 /* Check if the default domain is already direct mapped */
1592 if (group->default_domain &&
1593 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1596 /* Don't change mappings of existing devices */
1598 if (iommu_group_device_count(group) != 1)
1601 /* Allocate a direct mapped domain */
1603 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1607 /* Attach the device to the domain */
1608 ret = __iommu_attach_group(dm_domain, group);
1610 iommu_domain_free(dm_domain);
1614 /* Make the direct mapped domain the default for this group */
1615 if (group->default_domain)
1616 iommu_domain_free(group->default_domain);
1617 group->default_domain = dm_domain;
1619 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1623 mutex_unlock(&group->mutex);
1624 iommu_group_put(group);
1629 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1630 const struct iommu_ops *ops)
1632 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1635 return ops == fwspec->ops ? 0 : -EINVAL;
1637 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1641 of_node_get(to_of_node(iommu_fwnode));
1642 fwspec->iommu_fwnode = iommu_fwnode;
1644 dev->iommu_fwspec = fwspec;
1647 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1649 void iommu_fwspec_free(struct device *dev)
1651 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1654 fwnode_handle_put(fwspec->iommu_fwnode);
1656 dev->iommu_fwspec = NULL;
1659 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1661 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1663 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1670 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1671 if (size > sizeof(*fwspec)) {
1672 fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1677 for (i = 0; i < num_ids; i++)
1678 fwspec->ids[fwspec->num_ids + i] = ids[i];
1680 fwspec->num_ids += num_ids;
1681 dev->iommu_fwspec = fwspec;
1684 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);