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);
198 kobject_put(&group->kobj);
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);
442 sysfs_remove_link(group->devices_kobj, device->name);
446 sysfs_remove_link(&dev->kobj, "iommu_group");
449 pr_err("Failed to add device %s to group %d: %d\n", dev_name(dev), group->id, ret);
452 EXPORT_SYMBOL_GPL(iommu_group_add_device);
455 * iommu_group_remove_device - remove a device from it's current group
456 * @dev: device to be removed
458 * This function is called by an iommu driver to remove the device from
459 * it's current group. This decrements the iommu group reference count.
461 void iommu_group_remove_device(struct device *dev)
463 struct iommu_group *group = dev->iommu_group;
464 struct iommu_device *tmp_device, *device = NULL;
466 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
468 /* Pre-notify listeners that a device is being removed. */
469 blocking_notifier_call_chain(&group->notifier,
470 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
472 mutex_lock(&group->mutex);
473 list_for_each_entry(tmp_device, &group->devices, list) {
474 if (tmp_device->dev == dev) {
476 list_del(&device->list);
480 mutex_unlock(&group->mutex);
485 sysfs_remove_link(group->devices_kobj, device->name);
486 sysfs_remove_link(&dev->kobj, "iommu_group");
488 trace_remove_device_from_group(group->id, dev);
492 dev->iommu_group = NULL;
493 kobject_put(group->devices_kobj);
495 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
497 static int iommu_group_device_count(struct iommu_group *group)
499 struct iommu_device *entry;
502 list_for_each_entry(entry, &group->devices, list)
509 * iommu_group_for_each_dev - iterate over each device in the group
511 * @data: caller opaque data to be passed to callback function
512 * @fn: caller supplied callback function
514 * This function is called by group users to iterate over group devices.
515 * Callers should hold a reference count to the group during callback.
516 * The group->mutex is held across callbacks, which will block calls to
517 * iommu_group_add/remove_device.
519 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
520 int (*fn)(struct device *, void *))
522 struct iommu_device *device;
525 list_for_each_entry(device, &group->devices, list) {
526 ret = fn(device->dev, data);
534 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
535 int (*fn)(struct device *, void *))
539 mutex_lock(&group->mutex);
540 ret = __iommu_group_for_each_dev(group, data, fn);
541 mutex_unlock(&group->mutex);
545 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
548 * iommu_group_get - Return the group for a device and increment reference
549 * @dev: get the group that this device belongs to
551 * This function is called by iommu drivers and users to get the group
552 * for the specified device. If found, the group is returned and the group
553 * reference in incremented, else NULL.
555 struct iommu_group *iommu_group_get(struct device *dev)
557 struct iommu_group *group = dev->iommu_group;
560 kobject_get(group->devices_kobj);
564 EXPORT_SYMBOL_GPL(iommu_group_get);
567 * iommu_group_put - Decrement group reference
568 * @group: the group to use
570 * This function is called by iommu drivers and users to release the
571 * iommu group. Once the reference count is zero, the group is released.
573 void iommu_group_put(struct iommu_group *group)
576 kobject_put(group->devices_kobj);
578 EXPORT_SYMBOL_GPL(iommu_group_put);
581 * iommu_group_register_notifier - Register a notifier for group changes
582 * @group: the group to watch
583 * @nb: notifier block to signal
585 * This function allows iommu group users to track changes in a group.
586 * See include/linux/iommu.h for actions sent via this notifier. Caller
587 * should hold a reference to the group throughout notifier registration.
589 int iommu_group_register_notifier(struct iommu_group *group,
590 struct notifier_block *nb)
592 return blocking_notifier_chain_register(&group->notifier, nb);
594 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
597 * iommu_group_unregister_notifier - Unregister a notifier
598 * @group: the group to watch
599 * @nb: notifier block to signal
601 * Unregister a previously registered group notifier block.
603 int iommu_group_unregister_notifier(struct iommu_group *group,
604 struct notifier_block *nb)
606 return blocking_notifier_chain_unregister(&group->notifier, nb);
608 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
611 * iommu_group_id - Return ID for a group
612 * @group: the group to ID
614 * Return the unique ID for the group matching the sysfs group number.
616 int iommu_group_id(struct iommu_group *group)
620 EXPORT_SYMBOL_GPL(iommu_group_id);
622 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
623 unsigned long *devfns);
626 * To consider a PCI device isolated, we require ACS to support Source
627 * Validation, Request Redirection, Completer Redirection, and Upstream
628 * Forwarding. This effectively means that devices cannot spoof their
629 * requester ID, requests and completions cannot be redirected, and all
630 * transactions are forwarded upstream, even as it passes through a
631 * bridge where the target device is downstream.
633 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
636 * For multifunction devices which are not isolated from each other, find
637 * all the other non-isolated functions and look for existing groups. For
638 * each function, we also need to look for aliases to or from other devices
639 * that may already have a group.
641 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
642 unsigned long *devfns)
644 struct pci_dev *tmp = NULL;
645 struct iommu_group *group;
647 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
650 for_each_pci_dev(tmp) {
651 if (tmp == pdev || tmp->bus != pdev->bus ||
652 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
653 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
656 group = get_pci_alias_group(tmp, devfns);
667 * Look for aliases to or from the given device for existing groups. DMA
668 * aliases are only supported on the same bus, therefore the search
669 * space is quite small (especially since we're really only looking at pcie
670 * device, and therefore only expect multiple slots on the root complex or
671 * downstream switch ports). It's conceivable though that a pair of
672 * multifunction devices could have aliases between them that would cause a
673 * loop. To prevent this, we use a bitmap to track where we've been.
675 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
676 unsigned long *devfns)
678 struct pci_dev *tmp = NULL;
679 struct iommu_group *group;
681 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
684 group = iommu_group_get(&pdev->dev);
688 for_each_pci_dev(tmp) {
689 if (tmp == pdev || tmp->bus != pdev->bus)
692 /* We alias them or they alias us */
693 if (pci_devs_are_dma_aliases(pdev, tmp)) {
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 * Generic device_group call-back function. It just allocates one
732 * iommu-group per device.
734 struct iommu_group *generic_device_group(struct device *dev)
736 struct iommu_group *group;
738 group = iommu_group_alloc();
746 * Use standard PCI bus topology, isolation features, and DMA alias quirks
747 * to find or create an IOMMU group for a device.
749 struct iommu_group *pci_device_group(struct device *dev)
751 struct pci_dev *pdev = to_pci_dev(dev);
752 struct group_for_pci_data data;
754 struct iommu_group *group = NULL;
755 u64 devfns[4] = { 0 };
757 if (WARN_ON(!dev_is_pci(dev)))
758 return ERR_PTR(-EINVAL);
761 * Find the upstream DMA alias for the device. A device must not
762 * be aliased due to topology in order to have its own IOMMU group.
763 * If we find an alias along the way that already belongs to a
766 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
772 * Continue upstream from the point of minimum IOMMU granularity
773 * due to aliases to the point where devices are protected from
774 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
777 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
781 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
786 group = iommu_group_get(&pdev->dev);
792 * Look for existing groups on device aliases. If we alias another
793 * device or another device aliases us, use the same group.
795 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
800 * Look for existing groups on non-isolated functions on the same
801 * slot and aliases of those funcions, if any. No need to clear
802 * the search bitmap, the tested devfns are still valid.
804 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
808 /* No shared group found, allocate new */
809 group = iommu_group_alloc();
817 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
818 * @dev: target device
820 * This function is intended to be called by IOMMU drivers and extended to
821 * support common, bus-defined algorithms when determining or creating the
822 * IOMMU group for a device. On success, the caller will hold a reference
823 * to the returned IOMMU group, which will already include the provided
824 * device. The reference should be released with iommu_group_put().
826 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
828 const struct iommu_ops *ops = dev->bus->iommu_ops;
829 struct iommu_group *group;
832 group = iommu_group_get(dev);
836 group = ERR_PTR(-EINVAL);
838 if (ops && ops->device_group)
839 group = ops->device_group(dev);
845 * Try to allocate a default domain - needs support from the
848 if (!group->default_domain) {
849 group->default_domain = __iommu_domain_alloc(dev->bus,
852 group->domain = group->default_domain;
855 ret = iommu_group_add_device(group, dev);
857 iommu_group_put(group);
864 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
866 return group->default_domain;
869 static int add_iommu_group(struct device *dev, void *data)
871 struct iommu_callback_data *cb = data;
872 const struct iommu_ops *ops = cb->ops;
875 if (!ops->add_device)
878 WARN_ON(dev->iommu_group);
880 ret = ops->add_device(dev);
883 * We ignore -ENODEV errors for now, as they just mean that the
884 * device is not translated by an IOMMU. We still care about
885 * other errors and fail to initialize when they happen.
893 static int remove_iommu_group(struct device *dev, void *data)
895 struct iommu_callback_data *cb = data;
896 const struct iommu_ops *ops = cb->ops;
898 if (ops->remove_device && dev->iommu_group)
899 ops->remove_device(dev);
904 static int iommu_bus_notifier(struct notifier_block *nb,
905 unsigned long action, void *data)
907 struct device *dev = data;
908 const struct iommu_ops *ops = dev->bus->iommu_ops;
909 struct iommu_group *group;
910 unsigned long group_action = 0;
913 * ADD/DEL call into iommu driver ops if provided, which may
914 * result in ADD/DEL notifiers to group->notifier
916 if (action == BUS_NOTIFY_ADD_DEVICE) {
918 return ops->add_device(dev);
919 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
920 if (ops->remove_device && dev->iommu_group) {
921 ops->remove_device(dev);
927 * Remaining BUS_NOTIFYs get filtered and republished to the
928 * group, if anyone is listening
930 group = iommu_group_get(dev);
935 case BUS_NOTIFY_BIND_DRIVER:
936 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
938 case BUS_NOTIFY_BOUND_DRIVER:
939 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
941 case BUS_NOTIFY_UNBIND_DRIVER:
942 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
944 case BUS_NOTIFY_UNBOUND_DRIVER:
945 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
950 blocking_notifier_call_chain(&group->notifier,
953 iommu_group_put(group);
957 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
960 struct notifier_block *nb;
961 struct iommu_callback_data cb = {
965 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
969 nb->notifier_call = iommu_bus_notifier;
971 err = bus_register_notifier(bus, nb);
975 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
984 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
985 bus_unregister_notifier(bus, nb);
994 * bus_set_iommu - set iommu-callbacks for the bus
996 * @ops: the callbacks provided by the iommu-driver
998 * This function is called by an iommu driver to set the iommu methods
999 * used for a particular bus. Drivers for devices on that bus can use
1000 * the iommu-api after these ops are registered.
1001 * This special function is needed because IOMMUs are usually devices on
1002 * the bus itself, so the iommu drivers are not initialized when the bus
1003 * is set up. With this function the iommu-driver can set the iommu-ops
1006 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1010 if (bus->iommu_ops != NULL)
1013 bus->iommu_ops = ops;
1015 /* Do IOMMU specific setup for this bus-type */
1016 err = iommu_bus_init(bus, ops);
1018 bus->iommu_ops = NULL;
1022 EXPORT_SYMBOL_GPL(bus_set_iommu);
1024 bool iommu_present(struct bus_type *bus)
1026 return bus->iommu_ops != NULL;
1028 EXPORT_SYMBOL_GPL(iommu_present);
1030 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1032 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1035 return bus->iommu_ops->capable(cap);
1037 EXPORT_SYMBOL_GPL(iommu_capable);
1040 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1041 * @domain: iommu domain
1042 * @handler: fault handler
1043 * @token: user data, will be passed back to the fault handler
1045 * This function should be used by IOMMU users which want to be notified
1046 * whenever an IOMMU fault happens.
1048 * The fault handler itself should return 0 on success, and an appropriate
1049 * error code otherwise.
1051 void iommu_set_fault_handler(struct iommu_domain *domain,
1052 iommu_fault_handler_t handler,
1057 domain->handler = handler;
1058 domain->handler_token = token;
1060 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1062 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1065 struct iommu_domain *domain;
1067 if (bus == NULL || bus->iommu_ops == NULL)
1070 domain = bus->iommu_ops->domain_alloc(type);
1074 domain->ops = bus->iommu_ops;
1075 domain->type = type;
1076 /* Assume all sizes by default; the driver may override this later */
1077 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1082 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1084 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1086 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1088 void iommu_domain_free(struct iommu_domain *domain)
1090 domain->ops->domain_free(domain);
1092 EXPORT_SYMBOL_GPL(iommu_domain_free);
1094 static int __iommu_attach_device(struct iommu_domain *domain,
1098 if (unlikely(domain->ops->attach_dev == NULL))
1101 ret = domain->ops->attach_dev(domain, dev);
1103 trace_attach_device_to_domain(dev);
1107 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1109 struct iommu_group *group;
1112 group = iommu_group_get(dev);
1113 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1115 return __iommu_attach_device(domain, dev);
1118 * We have a group - lock it to make sure the device-count doesn't
1119 * change while we are attaching
1121 mutex_lock(&group->mutex);
1123 if (iommu_group_device_count(group) != 1)
1126 ret = __iommu_attach_group(domain, group);
1129 mutex_unlock(&group->mutex);
1130 iommu_group_put(group);
1134 EXPORT_SYMBOL_GPL(iommu_attach_device);
1136 static void __iommu_detach_device(struct iommu_domain *domain,
1139 if (unlikely(domain->ops->detach_dev == NULL))
1142 domain->ops->detach_dev(domain, dev);
1143 trace_detach_device_from_domain(dev);
1146 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1148 struct iommu_group *group;
1150 group = iommu_group_get(dev);
1151 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1153 return __iommu_detach_device(domain, dev);
1155 mutex_lock(&group->mutex);
1156 if (iommu_group_device_count(group) != 1) {
1161 __iommu_detach_group(domain, group);
1164 mutex_unlock(&group->mutex);
1165 iommu_group_put(group);
1167 EXPORT_SYMBOL_GPL(iommu_detach_device);
1169 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1171 struct iommu_domain *domain;
1172 struct iommu_group *group;
1174 group = iommu_group_get(dev);
1175 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1179 domain = group->domain;
1181 iommu_group_put(group);
1185 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1188 * IOMMU groups are really the natrual working unit of the IOMMU, but
1189 * the IOMMU API works on domains and devices. Bridge that gap by
1190 * iterating over the devices in a group. Ideally we'd have a single
1191 * device which represents the requestor ID of the group, but we also
1192 * allow IOMMU drivers to create policy defined minimum sets, where
1193 * the physical hardware may be able to distiguish members, but we
1194 * wish to group them at a higher level (ex. untrusted multi-function
1195 * PCI devices). Thus we attach each device.
1197 static int iommu_group_do_attach_device(struct device *dev, void *data)
1199 struct iommu_domain *domain = data;
1201 return __iommu_attach_device(domain, dev);
1204 static int __iommu_attach_group(struct iommu_domain *domain,
1205 struct iommu_group *group)
1209 if (group->default_domain && group->domain != group->default_domain)
1212 ret = __iommu_group_for_each_dev(group, domain,
1213 iommu_group_do_attach_device);
1215 group->domain = domain;
1220 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1224 mutex_lock(&group->mutex);
1225 ret = __iommu_attach_group(domain, group);
1226 mutex_unlock(&group->mutex);
1230 EXPORT_SYMBOL_GPL(iommu_attach_group);
1232 static int iommu_group_do_detach_device(struct device *dev, void *data)
1234 struct iommu_domain *domain = data;
1236 __iommu_detach_device(domain, dev);
1241 static void __iommu_detach_group(struct iommu_domain *domain,
1242 struct iommu_group *group)
1246 if (!group->default_domain) {
1247 __iommu_group_for_each_dev(group, domain,
1248 iommu_group_do_detach_device);
1249 group->domain = NULL;
1253 if (group->domain == group->default_domain)
1256 /* Detach by re-attaching to the default domain */
1257 ret = __iommu_group_for_each_dev(group, group->default_domain,
1258 iommu_group_do_attach_device);
1262 group->domain = group->default_domain;
1265 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1267 mutex_lock(&group->mutex);
1268 __iommu_detach_group(domain, group);
1269 mutex_unlock(&group->mutex);
1271 EXPORT_SYMBOL_GPL(iommu_detach_group);
1273 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1275 if (unlikely(domain->ops->iova_to_phys == NULL))
1278 return domain->ops->iova_to_phys(domain, iova);
1280 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1282 static size_t iommu_pgsize(struct iommu_domain *domain,
1283 unsigned long addr_merge, size_t size)
1285 unsigned int pgsize_idx;
1288 /* Max page size that still fits into 'size' */
1289 pgsize_idx = __fls(size);
1291 /* need to consider alignment requirements ? */
1292 if (likely(addr_merge)) {
1293 /* Max page size allowed by address */
1294 unsigned int align_pgsize_idx = __ffs(addr_merge);
1295 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1298 /* build a mask of acceptable page sizes */
1299 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1301 /* throw away page sizes not supported by the hardware */
1302 pgsize &= domain->pgsize_bitmap;
1304 /* make sure we're still sane */
1307 /* pick the biggest page */
1308 pgsize_idx = __fls(pgsize);
1309 pgsize = 1UL << pgsize_idx;
1314 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1315 phys_addr_t paddr, size_t size, int prot)
1317 unsigned long orig_iova = iova;
1318 unsigned int min_pagesz;
1319 size_t orig_size = size;
1320 phys_addr_t orig_paddr = paddr;
1323 if (unlikely(domain->ops->map == NULL ||
1324 domain->pgsize_bitmap == 0UL))
1327 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1330 /* find out the minimum page size supported */
1331 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1334 * both the virtual address and the physical one, as well as
1335 * the size of the mapping, must be aligned (at least) to the
1336 * size of the smallest page supported by the hardware
1338 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1339 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1340 iova, &paddr, size, min_pagesz);
1344 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1347 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1349 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1350 iova, &paddr, pgsize);
1352 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1361 /* unroll mapping in case something went wrong */
1363 iommu_unmap(domain, orig_iova, orig_size - size);
1365 trace_map(orig_iova, orig_paddr, orig_size);
1369 EXPORT_SYMBOL_GPL(iommu_map);
1371 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1373 size_t unmapped_page, unmapped = 0;
1374 unsigned int min_pagesz;
1375 unsigned long orig_iova = iova;
1377 if (unlikely(domain->ops->unmap == NULL ||
1378 domain->pgsize_bitmap == 0UL))
1381 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1384 /* find out the minimum page size supported */
1385 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1388 * The virtual address, as well as the size of the mapping, must be
1389 * aligned (at least) to the size of the smallest page supported
1392 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1393 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1394 iova, size, min_pagesz);
1398 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1401 * Keep iterating until we either unmap 'size' bytes (or more)
1402 * or we hit an area that isn't mapped.
1404 while (unmapped < size) {
1405 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1407 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1411 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1412 iova, unmapped_page);
1414 iova += unmapped_page;
1415 unmapped += unmapped_page;
1418 trace_unmap(orig_iova, size, unmapped);
1421 EXPORT_SYMBOL_GPL(iommu_unmap);
1423 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1424 struct scatterlist *sg, unsigned int nents, int prot)
1426 struct scatterlist *s;
1428 unsigned int i, min_pagesz;
1431 if (unlikely(domain->pgsize_bitmap == 0UL))
1434 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1436 for_each_sg(sg, s, nents, i) {
1437 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1440 * We are mapping on IOMMU page boundaries, so offset within
1441 * the page must be 0. However, the IOMMU may support pages
1442 * smaller than PAGE_SIZE, so s->offset may still represent
1443 * an offset of that boundary within the CPU page.
1445 if (!IS_ALIGNED(s->offset, min_pagesz))
1448 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1452 mapped += s->length;
1458 /* undo mappings already done */
1459 iommu_unmap(domain, iova, mapped);
1464 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1466 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1467 phys_addr_t paddr, u64 size, int prot)
1469 if (unlikely(domain->ops->domain_window_enable == NULL))
1472 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1475 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1477 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1479 if (unlikely(domain->ops->domain_window_disable == NULL))
1482 return domain->ops->domain_window_disable(domain, wnd_nr);
1484 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1486 static int __init iommu_init(void)
1488 iommu_group_kset = kset_create_and_add("iommu_groups",
1490 BUG_ON(!iommu_group_kset);
1494 core_initcall(iommu_init);
1496 int iommu_domain_get_attr(struct iommu_domain *domain,
1497 enum iommu_attr attr, void *data)
1499 struct iommu_domain_geometry *geometry;
1505 case DOMAIN_ATTR_GEOMETRY:
1507 *geometry = domain->geometry;
1510 case DOMAIN_ATTR_PAGING:
1512 *paging = (domain->pgsize_bitmap != 0UL);
1514 case DOMAIN_ATTR_WINDOWS:
1517 if (domain->ops->domain_get_windows != NULL)
1518 *count = domain->ops->domain_get_windows(domain);
1524 if (!domain->ops->domain_get_attr)
1527 ret = domain->ops->domain_get_attr(domain, attr, data);
1532 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1534 int iommu_domain_set_attr(struct iommu_domain *domain,
1535 enum iommu_attr attr, void *data)
1541 case DOMAIN_ATTR_WINDOWS:
1544 if (domain->ops->domain_set_windows != NULL)
1545 ret = domain->ops->domain_set_windows(domain, *count);
1551 if (domain->ops->domain_set_attr == NULL)
1554 ret = domain->ops->domain_set_attr(domain, attr, data);
1559 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1561 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1563 const struct iommu_ops *ops = dev->bus->iommu_ops;
1565 if (ops && ops->get_dm_regions)
1566 ops->get_dm_regions(dev, list);
1569 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1571 const struct iommu_ops *ops = dev->bus->iommu_ops;
1573 if (ops && ops->put_dm_regions)
1574 ops->put_dm_regions(dev, list);
1577 /* Request that a device is direct mapped by the IOMMU */
1578 int iommu_request_dm_for_dev(struct device *dev)
1580 struct iommu_domain *dm_domain;
1581 struct iommu_group *group;
1584 /* Device must already be in a group before calling this function */
1585 group = iommu_group_get(dev);
1589 mutex_lock(&group->mutex);
1591 /* Check if the default domain is already direct mapped */
1593 if (group->default_domain &&
1594 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1597 /* Don't change mappings of existing devices */
1599 if (iommu_group_device_count(group) != 1)
1602 /* Allocate a direct mapped domain */
1604 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1608 /* Attach the device to the domain */
1609 ret = __iommu_attach_group(dm_domain, group);
1611 iommu_domain_free(dm_domain);
1615 /* Make the direct mapped domain the default for this group */
1616 if (group->default_domain)
1617 iommu_domain_free(group->default_domain);
1618 group->default_domain = dm_domain;
1620 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1624 mutex_unlock(&group->mutex);
1625 iommu_group_put(group);
1630 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1631 const struct iommu_ops *ops)
1633 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1636 return ops == fwspec->ops ? 0 : -EINVAL;
1638 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1642 of_node_get(to_of_node(iommu_fwnode));
1643 fwspec->iommu_fwnode = iommu_fwnode;
1645 dev->iommu_fwspec = fwspec;
1648 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1650 void iommu_fwspec_free(struct device *dev)
1652 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1655 fwnode_handle_put(fwspec->iommu_fwnode);
1657 dev->iommu_fwspec = NULL;
1660 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1662 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1664 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1671 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1672 if (size > sizeof(*fwspec)) {
1673 fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1678 for (i = 0; i < num_ids; i++)
1679 fwspec->ids[fwspec->num_ids + i] = ids[i];
1681 fwspec->num_ids += num_ids;
1682 dev->iommu_fwspec = fwspec;
1685 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);