2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published
7 * by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #define pr_fmt(fmt) "iommu: " fmt
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/module.h>
26 #include <linux/slab.h>
27 #include <linux/errno.h>
28 #include <linux/iommu.h>
29 #include <linux/idr.h>
30 #include <linux/notifier.h>
31 #include <linux/err.h>
32 #include <linux/pci.h>
33 #include <linux/bitops.h>
34 #include <trace/events/iommu.h>
36 static struct kset *iommu_group_kset;
37 static struct ida iommu_group_ida;
38 static struct mutex iommu_group_mutex;
40 struct iommu_callback_data {
41 const struct iommu_ops *ops;
46 struct kobject *devices_kobj;
47 struct list_head devices;
49 struct blocking_notifier_head notifier;
51 void (*iommu_data_release)(void *iommu_data);
54 struct iommu_domain *default_domain;
58 struct list_head list;
63 struct iommu_group_attribute {
64 struct attribute attr;
65 ssize_t (*show)(struct iommu_group *group, char *buf);
66 ssize_t (*store)(struct iommu_group *group,
67 const char *buf, size_t count);
70 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
71 struct iommu_group_attribute iommu_group_attr_##_name = \
72 __ATTR(_name, _mode, _show, _store)
74 #define to_iommu_group_attr(_attr) \
75 container_of(_attr, struct iommu_group_attribute, attr)
76 #define to_iommu_group(_kobj) \
77 container_of(_kobj, struct iommu_group, kobj)
79 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
82 static ssize_t iommu_group_attr_show(struct kobject *kobj,
83 struct attribute *__attr, char *buf)
85 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
86 struct iommu_group *group = to_iommu_group(kobj);
90 ret = attr->show(group, buf);
94 static ssize_t iommu_group_attr_store(struct kobject *kobj,
95 struct attribute *__attr,
96 const char *buf, size_t count)
98 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
99 struct iommu_group *group = to_iommu_group(kobj);
103 ret = attr->store(group, buf, count);
107 static const struct sysfs_ops iommu_group_sysfs_ops = {
108 .show = iommu_group_attr_show,
109 .store = iommu_group_attr_store,
112 static int iommu_group_create_file(struct iommu_group *group,
113 struct iommu_group_attribute *attr)
115 return sysfs_create_file(&group->kobj, &attr->attr);
118 static void iommu_group_remove_file(struct iommu_group *group,
119 struct iommu_group_attribute *attr)
121 sysfs_remove_file(&group->kobj, &attr->attr);
124 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
126 return sprintf(buf, "%s\n", group->name);
129 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
131 static void iommu_group_release(struct kobject *kobj)
133 struct iommu_group *group = to_iommu_group(kobj);
135 pr_debug("Releasing group %d\n", group->id);
137 if (group->iommu_data_release)
138 group->iommu_data_release(group->iommu_data);
140 mutex_lock(&iommu_group_mutex);
141 ida_remove(&iommu_group_ida, group->id);
142 mutex_unlock(&iommu_group_mutex);
144 if (group->default_domain)
145 iommu_domain_free(group->default_domain);
151 static struct kobj_type iommu_group_ktype = {
152 .sysfs_ops = &iommu_group_sysfs_ops,
153 .release = iommu_group_release,
157 * iommu_group_alloc - Allocate a new group
158 * @name: Optional name to associate with group, visible in sysfs
160 * This function is called by an iommu driver to allocate a new iommu
161 * group. The iommu group represents the minimum granularity of the iommu.
162 * Upon successful return, the caller holds a reference to the supplied
163 * group in order to hold the group until devices are added. Use
164 * iommu_group_put() to release this extra reference count, allowing the
165 * group to be automatically reclaimed once it has no devices or external
168 struct iommu_group *iommu_group_alloc(void)
170 struct iommu_group *group;
173 group = kzalloc(sizeof(*group), GFP_KERNEL);
175 return ERR_PTR(-ENOMEM);
177 group->kobj.kset = iommu_group_kset;
178 mutex_init(&group->mutex);
179 INIT_LIST_HEAD(&group->devices);
180 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
182 mutex_lock(&iommu_group_mutex);
185 if (unlikely(0 == ida_pre_get(&iommu_group_ida, GFP_KERNEL))) {
187 mutex_unlock(&iommu_group_mutex);
188 return ERR_PTR(-ENOMEM);
191 if (-EAGAIN == ida_get_new(&iommu_group_ida, &group->id))
194 mutex_unlock(&iommu_group_mutex);
196 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
197 NULL, "%d", group->id);
199 mutex_lock(&iommu_group_mutex);
200 ida_remove(&iommu_group_ida, group->id);
201 mutex_unlock(&iommu_group_mutex);
206 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
207 if (!group->devices_kobj) {
208 kobject_put(&group->kobj); /* triggers .release & free */
209 return ERR_PTR(-ENOMEM);
213 * The devices_kobj holds a reference on the group kobject, so
214 * as long as that exists so will the group. We can therefore
215 * use the devices_kobj for reference counting.
217 kobject_put(&group->kobj);
219 pr_debug("Allocated group %d\n", group->id);
223 EXPORT_SYMBOL_GPL(iommu_group_alloc);
225 struct iommu_group *iommu_group_get_by_id(int id)
227 struct kobject *group_kobj;
228 struct iommu_group *group;
231 if (!iommu_group_kset)
234 name = kasprintf(GFP_KERNEL, "%d", id);
238 group_kobj = kset_find_obj(iommu_group_kset, name);
244 group = container_of(group_kobj, struct iommu_group, kobj);
245 BUG_ON(group->id != id);
247 kobject_get(group->devices_kobj);
248 kobject_put(&group->kobj);
252 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
255 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
258 * iommu drivers can store data in the group for use when doing iommu
259 * operations. This function provides a way to retrieve it. Caller
260 * should hold a group reference.
262 void *iommu_group_get_iommudata(struct iommu_group *group)
264 return group->iommu_data;
266 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
269 * iommu_group_set_iommudata - set iommu_data for a group
271 * @iommu_data: new data
272 * @release: release function for iommu_data
274 * iommu drivers can store data in the group for use when doing iommu
275 * operations. This function provides a way to set the data after
276 * the group has been allocated. Caller should hold a group reference.
278 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
279 void (*release)(void *iommu_data))
281 group->iommu_data = iommu_data;
282 group->iommu_data_release = release;
284 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
287 * iommu_group_set_name - set name for a group
291 * Allow iommu driver to set a name for a group. When set it will
292 * appear in a name attribute file under the group in sysfs.
294 int iommu_group_set_name(struct iommu_group *group, const char *name)
299 iommu_group_remove_file(group, &iommu_group_attr_name);
306 group->name = kstrdup(name, GFP_KERNEL);
310 ret = iommu_group_create_file(group, &iommu_group_attr_name);
319 EXPORT_SYMBOL_GPL(iommu_group_set_name);
322 * iommu_group_add_device - add a device to an iommu group
323 * @group: the group into which to add the device (reference should be held)
326 * This function is called by an iommu driver to add a device into a
327 * group. Adding a device increments the group reference count.
329 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
332 struct iommu_device *device;
334 device = kzalloc(sizeof(*device), GFP_KERNEL);
340 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
346 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
349 sysfs_remove_link(&dev->kobj, "iommu_group");
354 ret = sysfs_create_link_nowarn(group->devices_kobj,
355 &dev->kobj, device->name);
358 if (ret == -EEXIST && i >= 0) {
360 * Account for the slim chance of collision
361 * and append an instance to the name.
363 device->name = kasprintf(GFP_KERNEL, "%s.%d",
364 kobject_name(&dev->kobj), i++);
368 sysfs_remove_link(&dev->kobj, "iommu_group");
373 kobject_get(group->devices_kobj);
375 dev->iommu_group = group;
377 mutex_lock(&group->mutex);
378 list_add_tail(&device->list, &group->devices);
379 mutex_unlock(&group->mutex);
381 /* Notify any listeners about change to group. */
382 blocking_notifier_call_chain(&group->notifier,
383 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
385 trace_add_device_to_group(group->id, dev);
387 pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
391 EXPORT_SYMBOL_GPL(iommu_group_add_device);
394 * iommu_group_remove_device - remove a device from it's current group
395 * @dev: device to be removed
397 * This function is called by an iommu driver to remove the device from
398 * it's current group. This decrements the iommu group reference count.
400 void iommu_group_remove_device(struct device *dev)
402 struct iommu_group *group = dev->iommu_group;
403 struct iommu_device *tmp_device, *device = NULL;
405 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
407 /* Pre-notify listeners that a device is being removed. */
408 blocking_notifier_call_chain(&group->notifier,
409 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
411 mutex_lock(&group->mutex);
412 list_for_each_entry(tmp_device, &group->devices, list) {
413 if (tmp_device->dev == dev) {
415 list_del(&device->list);
419 mutex_unlock(&group->mutex);
424 sysfs_remove_link(group->devices_kobj, device->name);
425 sysfs_remove_link(&dev->kobj, "iommu_group");
427 trace_remove_device_from_group(group->id, dev);
431 dev->iommu_group = NULL;
432 kobject_put(group->devices_kobj);
434 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
437 * iommu_group_for_each_dev - iterate over each device in the group
439 * @data: caller opaque data to be passed to callback function
440 * @fn: caller supplied callback function
442 * This function is called by group users to iterate over group devices.
443 * Callers should hold a reference count to the group during callback.
444 * The group->mutex is held across callbacks, which will block calls to
445 * iommu_group_add/remove_device.
447 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
448 int (*fn)(struct device *, void *))
450 struct iommu_device *device;
453 mutex_lock(&group->mutex);
454 list_for_each_entry(device, &group->devices, list) {
455 ret = fn(device->dev, data);
459 mutex_unlock(&group->mutex);
462 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
465 * iommu_group_get - Return the group for a device and increment reference
466 * @dev: get the group that this device belongs to
468 * This function is called by iommu drivers and users to get the group
469 * for the specified device. If found, the group is returned and the group
470 * reference in incremented, else NULL.
472 struct iommu_group *iommu_group_get(struct device *dev)
474 struct iommu_group *group = dev->iommu_group;
477 kobject_get(group->devices_kobj);
481 EXPORT_SYMBOL_GPL(iommu_group_get);
484 * iommu_group_put - Decrement group reference
485 * @group: the group to use
487 * This function is called by iommu drivers and users to release the
488 * iommu group. Once the reference count is zero, the group is released.
490 void iommu_group_put(struct iommu_group *group)
493 kobject_put(group->devices_kobj);
495 EXPORT_SYMBOL_GPL(iommu_group_put);
498 * iommu_group_register_notifier - Register a notifier for group changes
499 * @group: the group to watch
500 * @nb: notifier block to signal
502 * This function allows iommu group users to track changes in a group.
503 * See include/linux/iommu.h for actions sent via this notifier. Caller
504 * should hold a reference to the group throughout notifier registration.
506 int iommu_group_register_notifier(struct iommu_group *group,
507 struct notifier_block *nb)
509 return blocking_notifier_chain_register(&group->notifier, nb);
511 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
514 * iommu_group_unregister_notifier - Unregister a notifier
515 * @group: the group to watch
516 * @nb: notifier block to signal
518 * Unregister a previously registered group notifier block.
520 int iommu_group_unregister_notifier(struct iommu_group *group,
521 struct notifier_block *nb)
523 return blocking_notifier_chain_unregister(&group->notifier, nb);
525 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
528 * iommu_group_id - Return ID for a group
529 * @group: the group to ID
531 * Return the unique ID for the group matching the sysfs group number.
533 int iommu_group_id(struct iommu_group *group)
537 EXPORT_SYMBOL_GPL(iommu_group_id);
539 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
540 unsigned long *devfns);
543 * To consider a PCI device isolated, we require ACS to support Source
544 * Validation, Request Redirection, Completer Redirection, and Upstream
545 * Forwarding. This effectively means that devices cannot spoof their
546 * requester ID, requests and completions cannot be redirected, and all
547 * transactions are forwarded upstream, even as it passes through a
548 * bridge where the target device is downstream.
550 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
553 * For multifunction devices which are not isolated from each other, find
554 * all the other non-isolated functions and look for existing groups. For
555 * each function, we also need to look for aliases to or from other devices
556 * that may already have a group.
558 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
559 unsigned long *devfns)
561 struct pci_dev *tmp = NULL;
562 struct iommu_group *group;
564 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
567 for_each_pci_dev(tmp) {
568 if (tmp == pdev || tmp->bus != pdev->bus ||
569 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
570 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
573 group = get_pci_alias_group(tmp, devfns);
584 * Look for aliases to or from the given device for exisiting groups. The
585 * dma_alias_devfn only supports aliases on the same bus, therefore the search
586 * space is quite small (especially since we're really only looking at pcie
587 * device, and therefore only expect multiple slots on the root complex or
588 * downstream switch ports). It's conceivable though that a pair of
589 * multifunction devices could have aliases between them that would cause a
590 * loop. To prevent this, we use a bitmap to track where we've been.
592 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
593 unsigned long *devfns)
595 struct pci_dev *tmp = NULL;
596 struct iommu_group *group;
598 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
601 group = iommu_group_get(&pdev->dev);
605 for_each_pci_dev(tmp) {
606 if (tmp == pdev || tmp->bus != pdev->bus)
609 /* We alias them or they alias us */
610 if (((pdev->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
611 pdev->dma_alias_devfn == tmp->devfn) ||
612 ((tmp->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
613 tmp->dma_alias_devfn == pdev->devfn)) {
615 group = get_pci_alias_group(tmp, devfns);
621 group = get_pci_function_alias_group(tmp, devfns);
632 struct group_for_pci_data {
633 struct pci_dev *pdev;
634 struct iommu_group *group;
638 * DMA alias iterator callback, return the last seen device. Stop and return
639 * the IOMMU group if we find one along the way.
641 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
643 struct group_for_pci_data *data = opaque;
646 data->group = iommu_group_get(&pdev->dev);
648 return data->group != NULL;
652 * Use standard PCI bus topology, isolation features, and DMA alias quirks
653 * to find or create an IOMMU group for a device.
655 static struct iommu_group *iommu_group_get_for_pci_dev(struct pci_dev *pdev)
657 struct group_for_pci_data data;
659 struct iommu_group *group = NULL;
660 u64 devfns[4] = { 0 };
663 * Find the upstream DMA alias for the device. A device must not
664 * be aliased due to topology in order to have its own IOMMU group.
665 * If we find an alias along the way that already belongs to a
668 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
674 * Continue upstream from the point of minimum IOMMU granularity
675 * due to aliases to the point where devices are protected from
676 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
679 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
683 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
688 group = iommu_group_get(&pdev->dev);
694 * Look for existing groups on device aliases. If we alias another
695 * device or another device aliases us, use the same group.
697 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
702 * Look for existing groups on non-isolated functions on the same
703 * slot and aliases of those funcions, if any. No need to clear
704 * the search bitmap, the tested devfns are still valid.
706 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
710 /* No shared group found, allocate new */
711 group = iommu_group_alloc();
714 * Try to allocate a default domain - needs support from the
717 group->default_domain = __iommu_domain_alloc(pdev->dev.bus,
725 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
726 * @dev: target device
728 * This function is intended to be called by IOMMU drivers and extended to
729 * support common, bus-defined algorithms when determining or creating the
730 * IOMMU group for a device. On success, the caller will hold a reference
731 * to the returned IOMMU group, which will already include the provided
732 * device. The reference should be released with iommu_group_put().
734 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
736 struct iommu_group *group;
739 group = iommu_group_get(dev);
743 if (!dev_is_pci(dev))
744 return ERR_PTR(-EINVAL);
746 group = iommu_group_get_for_pci_dev(to_pci_dev(dev));
751 ret = iommu_group_add_device(group, dev);
753 iommu_group_put(group);
760 static int add_iommu_group(struct device *dev, void *data)
762 struct iommu_callback_data *cb = data;
763 const struct iommu_ops *ops = cb->ops;
765 if (!ops->add_device)
768 WARN_ON(dev->iommu_group);
770 return ops->add_device(dev);
773 static int remove_iommu_group(struct device *dev, void *data)
775 struct iommu_callback_data *cb = data;
776 const struct iommu_ops *ops = cb->ops;
778 if (ops->remove_device && dev->iommu_group)
779 ops->remove_device(dev);
784 static int iommu_bus_notifier(struct notifier_block *nb,
785 unsigned long action, void *data)
787 struct device *dev = data;
788 const struct iommu_ops *ops = dev->bus->iommu_ops;
789 struct iommu_group *group;
790 unsigned long group_action = 0;
793 * ADD/DEL call into iommu driver ops if provided, which may
794 * result in ADD/DEL notifiers to group->notifier
796 if (action == BUS_NOTIFY_ADD_DEVICE) {
798 return ops->add_device(dev);
799 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
800 if (ops->remove_device && dev->iommu_group) {
801 ops->remove_device(dev);
807 * Remaining BUS_NOTIFYs get filtered and republished to the
808 * group, if anyone is listening
810 group = iommu_group_get(dev);
815 case BUS_NOTIFY_BIND_DRIVER:
816 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
818 case BUS_NOTIFY_BOUND_DRIVER:
819 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
821 case BUS_NOTIFY_UNBIND_DRIVER:
822 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
824 case BUS_NOTIFY_UNBOUND_DRIVER:
825 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
830 blocking_notifier_call_chain(&group->notifier,
833 iommu_group_put(group);
837 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
840 struct notifier_block *nb;
841 struct iommu_callback_data cb = {
845 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
849 nb->notifier_call = iommu_bus_notifier;
851 err = bus_register_notifier(bus, nb);
855 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
864 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
865 bus_unregister_notifier(bus, nb);
874 * bus_set_iommu - set iommu-callbacks for the bus
876 * @ops: the callbacks provided by the iommu-driver
878 * This function is called by an iommu driver to set the iommu methods
879 * used for a particular bus. Drivers for devices on that bus can use
880 * the iommu-api after these ops are registered.
881 * This special function is needed because IOMMUs are usually devices on
882 * the bus itself, so the iommu drivers are not initialized when the bus
883 * is set up. With this function the iommu-driver can set the iommu-ops
886 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
890 if (bus->iommu_ops != NULL)
893 bus->iommu_ops = ops;
895 /* Do IOMMU specific setup for this bus-type */
896 err = iommu_bus_init(bus, ops);
898 bus->iommu_ops = NULL;
902 EXPORT_SYMBOL_GPL(bus_set_iommu);
904 bool iommu_present(struct bus_type *bus)
906 return bus->iommu_ops != NULL;
908 EXPORT_SYMBOL_GPL(iommu_present);
910 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
912 if (!bus->iommu_ops || !bus->iommu_ops->capable)
915 return bus->iommu_ops->capable(cap);
917 EXPORT_SYMBOL_GPL(iommu_capable);
920 * iommu_set_fault_handler() - set a fault handler for an iommu domain
921 * @domain: iommu domain
922 * @handler: fault handler
923 * @token: user data, will be passed back to the fault handler
925 * This function should be used by IOMMU users which want to be notified
926 * whenever an IOMMU fault happens.
928 * The fault handler itself should return 0 on success, and an appropriate
929 * error code otherwise.
931 void iommu_set_fault_handler(struct iommu_domain *domain,
932 iommu_fault_handler_t handler,
937 domain->handler = handler;
938 domain->handler_token = token;
940 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
942 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
945 struct iommu_domain *domain;
947 if (bus == NULL || bus->iommu_ops == NULL)
950 domain = bus->iommu_ops->domain_alloc(type);
954 domain->ops = bus->iommu_ops;
960 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
962 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
964 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
966 void iommu_domain_free(struct iommu_domain *domain)
968 domain->ops->domain_free(domain);
970 EXPORT_SYMBOL_GPL(iommu_domain_free);
972 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
975 if (unlikely(domain->ops->attach_dev == NULL))
978 ret = domain->ops->attach_dev(domain, dev);
980 trace_attach_device_to_domain(dev);
983 EXPORT_SYMBOL_GPL(iommu_attach_device);
985 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
987 if (unlikely(domain->ops->detach_dev == NULL))
990 domain->ops->detach_dev(domain, dev);
991 trace_detach_device_from_domain(dev);
993 EXPORT_SYMBOL_GPL(iommu_detach_device);
996 * IOMMU groups are really the natrual working unit of the IOMMU, but
997 * the IOMMU API works on domains and devices. Bridge that gap by
998 * iterating over the devices in a group. Ideally we'd have a single
999 * device which represents the requestor ID of the group, but we also
1000 * allow IOMMU drivers to create policy defined minimum sets, where
1001 * the physical hardware may be able to distiguish members, but we
1002 * wish to group them at a higher level (ex. untrusted multi-function
1003 * PCI devices). Thus we attach each device.
1005 static int iommu_group_do_attach_device(struct device *dev, void *data)
1007 struct iommu_domain *domain = data;
1009 return iommu_attach_device(domain, dev);
1012 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1014 return iommu_group_for_each_dev(group, domain,
1015 iommu_group_do_attach_device);
1017 EXPORT_SYMBOL_GPL(iommu_attach_group);
1019 static int iommu_group_do_detach_device(struct device *dev, void *data)
1021 struct iommu_domain *domain = data;
1023 iommu_detach_device(domain, dev);
1028 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1030 iommu_group_for_each_dev(group, domain, iommu_group_do_detach_device);
1032 EXPORT_SYMBOL_GPL(iommu_detach_group);
1034 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1036 if (unlikely(domain->ops->iova_to_phys == NULL))
1039 return domain->ops->iova_to_phys(domain, iova);
1041 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1043 static size_t iommu_pgsize(struct iommu_domain *domain,
1044 unsigned long addr_merge, size_t size)
1046 unsigned int pgsize_idx;
1049 /* Max page size that still fits into 'size' */
1050 pgsize_idx = __fls(size);
1052 /* need to consider alignment requirements ? */
1053 if (likely(addr_merge)) {
1054 /* Max page size allowed by address */
1055 unsigned int align_pgsize_idx = __ffs(addr_merge);
1056 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1059 /* build a mask of acceptable page sizes */
1060 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1062 /* throw away page sizes not supported by the hardware */
1063 pgsize &= domain->ops->pgsize_bitmap;
1065 /* make sure we're still sane */
1068 /* pick the biggest page */
1069 pgsize_idx = __fls(pgsize);
1070 pgsize = 1UL << pgsize_idx;
1075 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1076 phys_addr_t paddr, size_t size, int prot)
1078 unsigned long orig_iova = iova;
1079 unsigned int min_pagesz;
1080 size_t orig_size = size;
1083 if (unlikely(domain->ops->map == NULL ||
1084 domain->ops->pgsize_bitmap == 0UL))
1087 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1090 /* find out the minimum page size supported */
1091 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1094 * both the virtual address and the physical one, as well as
1095 * the size of the mapping, must be aligned (at least) to the
1096 * size of the smallest page supported by the hardware
1098 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1099 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1100 iova, &paddr, size, min_pagesz);
1104 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1107 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1109 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1110 iova, &paddr, pgsize);
1112 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1121 /* unroll mapping in case something went wrong */
1123 iommu_unmap(domain, orig_iova, orig_size - size);
1125 trace_map(orig_iova, paddr, orig_size);
1129 EXPORT_SYMBOL_GPL(iommu_map);
1131 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1133 size_t unmapped_page, unmapped = 0;
1134 unsigned int min_pagesz;
1135 unsigned long orig_iova = iova;
1137 if (unlikely(domain->ops->unmap == NULL ||
1138 domain->ops->pgsize_bitmap == 0UL))
1141 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1144 /* find out the minimum page size supported */
1145 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1148 * The virtual address, as well as the size of the mapping, must be
1149 * aligned (at least) to the size of the smallest page supported
1152 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1153 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1154 iova, size, min_pagesz);
1158 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1161 * Keep iterating until we either unmap 'size' bytes (or more)
1162 * or we hit an area that isn't mapped.
1164 while (unmapped < size) {
1165 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1167 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1171 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1172 iova, unmapped_page);
1174 iova += unmapped_page;
1175 unmapped += unmapped_page;
1178 trace_unmap(orig_iova, size, unmapped);
1181 EXPORT_SYMBOL_GPL(iommu_unmap);
1183 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1184 struct scatterlist *sg, unsigned int nents, int prot)
1186 struct scatterlist *s;
1188 unsigned int i, min_pagesz;
1191 if (unlikely(domain->ops->pgsize_bitmap == 0UL))
1194 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1196 for_each_sg(sg, s, nents, i) {
1197 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1200 * We are mapping on IOMMU page boundaries, so offset within
1201 * the page must be 0. However, the IOMMU may support pages
1202 * smaller than PAGE_SIZE, so s->offset may still represent
1203 * an offset of that boundary within the CPU page.
1205 if (!IS_ALIGNED(s->offset, min_pagesz))
1208 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1212 mapped += s->length;
1218 /* undo mappings already done */
1219 iommu_unmap(domain, iova, mapped);
1224 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1226 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1227 phys_addr_t paddr, u64 size, int prot)
1229 if (unlikely(domain->ops->domain_window_enable == NULL))
1232 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1235 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1237 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1239 if (unlikely(domain->ops->domain_window_disable == NULL))
1242 return domain->ops->domain_window_disable(domain, wnd_nr);
1244 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1246 static int __init iommu_init(void)
1248 iommu_group_kset = kset_create_and_add("iommu_groups",
1250 ida_init(&iommu_group_ida);
1251 mutex_init(&iommu_group_mutex);
1253 BUG_ON(!iommu_group_kset);
1257 arch_initcall(iommu_init);
1259 int iommu_domain_get_attr(struct iommu_domain *domain,
1260 enum iommu_attr attr, void *data)
1262 struct iommu_domain_geometry *geometry;
1268 case DOMAIN_ATTR_GEOMETRY:
1270 *geometry = domain->geometry;
1273 case DOMAIN_ATTR_PAGING:
1275 *paging = (domain->ops->pgsize_bitmap != 0UL);
1277 case DOMAIN_ATTR_WINDOWS:
1280 if (domain->ops->domain_get_windows != NULL)
1281 *count = domain->ops->domain_get_windows(domain);
1287 if (!domain->ops->domain_get_attr)
1290 ret = domain->ops->domain_get_attr(domain, attr, data);
1295 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1297 int iommu_domain_set_attr(struct iommu_domain *domain,
1298 enum iommu_attr attr, void *data)
1304 case DOMAIN_ATTR_WINDOWS:
1307 if (domain->ops->domain_set_windows != NULL)
1308 ret = domain->ops->domain_set_windows(domain, *count);
1314 if (domain->ops->domain_set_attr == NULL)
1317 ret = domain->ops->domain_set_attr(domain, attr, data);
1322 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);