2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published
7 * by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 #define pr_fmt(fmt) "iommu: " fmt
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/module.h>
26 #include <linux/slab.h>
27 #include <linux/errno.h>
28 #include <linux/iommu.h>
29 #include <linux/idr.h>
30 #include <linux/notifier.h>
31 #include <linux/err.h>
32 #include <linux/pci.h>
33 #include <linux/bitops.h>
34 #include <trace/events/iommu.h>
36 static struct kset *iommu_group_kset;
37 static struct ida iommu_group_ida;
38 static struct mutex iommu_group_mutex;
40 struct iommu_callback_data {
41 const struct iommu_ops *ops;
46 struct kobject *devices_kobj;
47 struct list_head devices;
49 struct blocking_notifier_head notifier;
51 void (*iommu_data_release)(void *iommu_data);
54 struct iommu_domain *default_domain;
55 struct iommu_domain *domain;
59 struct list_head list;
64 struct iommu_group_attribute {
65 struct attribute attr;
66 ssize_t (*show)(struct iommu_group *group, char *buf);
67 ssize_t (*store)(struct iommu_group *group,
68 const char *buf, size_t count);
71 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
72 struct iommu_group_attribute iommu_group_attr_##_name = \
73 __ATTR(_name, _mode, _show, _store)
75 #define to_iommu_group_attr(_attr) \
76 container_of(_attr, struct iommu_group_attribute, attr)
77 #define to_iommu_group(_kobj) \
78 container_of(_kobj, struct iommu_group, kobj)
80 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
82 static int __iommu_attach_device(struct iommu_domain *domain,
84 static int __iommu_attach_group(struct iommu_domain *domain,
85 struct iommu_group *group);
86 static void __iommu_detach_group(struct iommu_domain *domain,
87 struct iommu_group *group);
89 static ssize_t iommu_group_attr_show(struct kobject *kobj,
90 struct attribute *__attr, char *buf)
92 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
93 struct iommu_group *group = to_iommu_group(kobj);
97 ret = attr->show(group, buf);
101 static ssize_t iommu_group_attr_store(struct kobject *kobj,
102 struct attribute *__attr,
103 const char *buf, size_t count)
105 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
106 struct iommu_group *group = to_iommu_group(kobj);
110 ret = attr->store(group, buf, count);
114 static const struct sysfs_ops iommu_group_sysfs_ops = {
115 .show = iommu_group_attr_show,
116 .store = iommu_group_attr_store,
119 static int iommu_group_create_file(struct iommu_group *group,
120 struct iommu_group_attribute *attr)
122 return sysfs_create_file(&group->kobj, &attr->attr);
125 static void iommu_group_remove_file(struct iommu_group *group,
126 struct iommu_group_attribute *attr)
128 sysfs_remove_file(&group->kobj, &attr->attr);
131 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
133 return sprintf(buf, "%s\n", group->name);
136 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
138 static void iommu_group_release(struct kobject *kobj)
140 struct iommu_group *group = to_iommu_group(kobj);
142 pr_debug("Releasing group %d\n", group->id);
144 if (group->iommu_data_release)
145 group->iommu_data_release(group->iommu_data);
147 mutex_lock(&iommu_group_mutex);
148 ida_remove(&iommu_group_ida, group->id);
149 mutex_unlock(&iommu_group_mutex);
151 if (group->default_domain)
152 iommu_domain_free(group->default_domain);
158 static struct kobj_type iommu_group_ktype = {
159 .sysfs_ops = &iommu_group_sysfs_ops,
160 .release = iommu_group_release,
164 * iommu_group_alloc - Allocate a new group
165 * @name: Optional name to associate with group, visible in sysfs
167 * This function is called by an iommu driver to allocate a new iommu
168 * group. The iommu group represents the minimum granularity of the iommu.
169 * Upon successful return, the caller holds a reference to the supplied
170 * group in order to hold the group until devices are added. Use
171 * iommu_group_put() to release this extra reference count, allowing the
172 * group to be automatically reclaimed once it has no devices or external
175 struct iommu_group *iommu_group_alloc(void)
177 struct iommu_group *group;
180 group = kzalloc(sizeof(*group), GFP_KERNEL);
182 return ERR_PTR(-ENOMEM);
184 group->kobj.kset = iommu_group_kset;
185 mutex_init(&group->mutex);
186 INIT_LIST_HEAD(&group->devices);
187 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
189 mutex_lock(&iommu_group_mutex);
192 if (unlikely(0 == ida_pre_get(&iommu_group_ida, GFP_KERNEL))) {
194 mutex_unlock(&iommu_group_mutex);
195 return ERR_PTR(-ENOMEM);
198 if (-EAGAIN == ida_get_new(&iommu_group_ida, &group->id))
201 mutex_unlock(&iommu_group_mutex);
203 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
204 NULL, "%d", group->id);
206 mutex_lock(&iommu_group_mutex);
207 ida_remove(&iommu_group_ida, group->id);
208 mutex_unlock(&iommu_group_mutex);
209 kobject_put(&group->kobj);
213 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
214 if (!group->devices_kobj) {
215 kobject_put(&group->kobj); /* triggers .release & free */
216 return ERR_PTR(-ENOMEM);
220 * The devices_kobj holds a reference on the group kobject, so
221 * as long as that exists so will the group. We can therefore
222 * use the devices_kobj for reference counting.
224 kobject_put(&group->kobj);
226 pr_debug("Allocated group %d\n", group->id);
230 EXPORT_SYMBOL_GPL(iommu_group_alloc);
232 struct iommu_group *iommu_group_get_by_id(int id)
234 struct kobject *group_kobj;
235 struct iommu_group *group;
238 if (!iommu_group_kset)
241 name = kasprintf(GFP_KERNEL, "%d", id);
245 group_kobj = kset_find_obj(iommu_group_kset, name);
251 group = container_of(group_kobj, struct iommu_group, kobj);
252 BUG_ON(group->id != id);
254 kobject_get(group->devices_kobj);
255 kobject_put(&group->kobj);
259 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
262 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
265 * iommu drivers can store data in the group for use when doing iommu
266 * operations. This function provides a way to retrieve it. Caller
267 * should hold a group reference.
269 void *iommu_group_get_iommudata(struct iommu_group *group)
271 return group->iommu_data;
273 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
276 * iommu_group_set_iommudata - set iommu_data for a group
278 * @iommu_data: new data
279 * @release: release function for iommu_data
281 * iommu drivers can store data in the group for use when doing iommu
282 * operations. This function provides a way to set the data after
283 * the group has been allocated. Caller should hold a group reference.
285 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
286 void (*release)(void *iommu_data))
288 group->iommu_data = iommu_data;
289 group->iommu_data_release = release;
291 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
294 * iommu_group_set_name - set name for a group
298 * Allow iommu driver to set a name for a group. When set it will
299 * appear in a name attribute file under the group in sysfs.
301 int iommu_group_set_name(struct iommu_group *group, const char *name)
306 iommu_group_remove_file(group, &iommu_group_attr_name);
313 group->name = kstrdup(name, GFP_KERNEL);
317 ret = iommu_group_create_file(group, &iommu_group_attr_name);
326 EXPORT_SYMBOL_GPL(iommu_group_set_name);
328 static int iommu_group_create_direct_mappings(struct iommu_group *group,
331 struct iommu_domain *domain = group->default_domain;
332 struct iommu_dm_region *entry;
333 struct list_head mappings;
334 unsigned long pg_size;
337 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
340 BUG_ON(!domain->ops->pgsize_bitmap);
342 pg_size = 1UL << __ffs(domain->ops->pgsize_bitmap);
343 INIT_LIST_HEAD(&mappings);
345 iommu_get_dm_regions(dev, &mappings);
347 /* We need to consider overlapping regions for different devices */
348 list_for_each_entry(entry, &mappings, list) {
349 dma_addr_t start, end, addr;
351 start = ALIGN(entry->start, pg_size);
352 end = ALIGN(entry->start + entry->length, pg_size);
354 for (addr = start; addr < end; addr += pg_size) {
355 phys_addr_t phys_addr;
357 phys_addr = iommu_iova_to_phys(domain, addr);
361 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
369 iommu_put_dm_regions(dev, &mappings);
375 * iommu_group_add_device - add a device to an iommu group
376 * @group: the group into which to add the device (reference should be held)
379 * This function is called by an iommu driver to add a device into a
380 * group. Adding a device increments the group reference count.
382 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
385 struct iommu_device *device;
387 device = kzalloc(sizeof(*device), GFP_KERNEL);
393 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
395 goto err_free_device;
397 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
401 goto err_remove_link;
404 ret = sysfs_create_link_nowarn(group->devices_kobj,
405 &dev->kobj, device->name);
407 if (ret == -EEXIST && i >= 0) {
409 * Account for the slim chance of collision
410 * and append an instance to the name.
413 device->name = kasprintf(GFP_KERNEL, "%s.%d",
414 kobject_name(&dev->kobj), i++);
420 kobject_get(group->devices_kobj);
422 dev->iommu_group = group;
424 iommu_group_create_direct_mappings(group, dev);
426 mutex_lock(&group->mutex);
427 list_add_tail(&device->list, &group->devices);
429 ret = __iommu_attach_device(group->domain, dev);
430 mutex_unlock(&group->mutex);
434 /* Notify any listeners about change to group. */
435 blocking_notifier_call_chain(&group->notifier,
436 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
438 trace_add_device_to_group(group->id, dev);
440 pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
445 mutex_lock(&group->mutex);
446 list_del(&device->list);
447 mutex_unlock(&group->mutex);
448 dev->iommu_group = NULL;
449 kobject_put(group->devices_kobj);
450 sysfs_remove_link(group->devices_kobj, device->name);
454 sysfs_remove_link(&dev->kobj, "iommu_group");
457 pr_err("Failed to add device %s to group %d: %d\n", dev_name(dev), group->id, ret);
460 EXPORT_SYMBOL_GPL(iommu_group_add_device);
463 * iommu_group_remove_device - remove a device from it's current group
464 * @dev: device to be removed
466 * This function is called by an iommu driver to remove the device from
467 * it's current group. This decrements the iommu group reference count.
469 void iommu_group_remove_device(struct device *dev)
471 struct iommu_group *group = dev->iommu_group;
472 struct iommu_device *tmp_device, *device = NULL;
474 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
476 /* Pre-notify listeners that a device is being removed. */
477 blocking_notifier_call_chain(&group->notifier,
478 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
480 mutex_lock(&group->mutex);
481 list_for_each_entry(tmp_device, &group->devices, list) {
482 if (tmp_device->dev == dev) {
484 list_del(&device->list);
488 mutex_unlock(&group->mutex);
493 sysfs_remove_link(group->devices_kobj, device->name);
494 sysfs_remove_link(&dev->kobj, "iommu_group");
496 trace_remove_device_from_group(group->id, dev);
500 dev->iommu_group = NULL;
501 kobject_put(group->devices_kobj);
503 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
505 static int iommu_group_device_count(struct iommu_group *group)
507 struct iommu_device *entry;
510 list_for_each_entry(entry, &group->devices, list)
517 * iommu_group_for_each_dev - iterate over each device in the group
519 * @data: caller opaque data to be passed to callback function
520 * @fn: caller supplied callback function
522 * This function is called by group users to iterate over group devices.
523 * Callers should hold a reference count to the group during callback.
524 * The group->mutex is held across callbacks, which will block calls to
525 * iommu_group_add/remove_device.
527 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
528 int (*fn)(struct device *, void *))
530 struct iommu_device *device;
533 list_for_each_entry(device, &group->devices, list) {
534 ret = fn(device->dev, data);
542 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
543 int (*fn)(struct device *, void *))
547 mutex_lock(&group->mutex);
548 ret = __iommu_group_for_each_dev(group, data, fn);
549 mutex_unlock(&group->mutex);
553 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
556 * iommu_group_get - Return the group for a device and increment reference
557 * @dev: get the group that this device belongs to
559 * This function is called by iommu drivers and users to get the group
560 * for the specified device. If found, the group is returned and the group
561 * reference in incremented, else NULL.
563 struct iommu_group *iommu_group_get(struct device *dev)
565 struct iommu_group *group = dev->iommu_group;
568 kobject_get(group->devices_kobj);
572 EXPORT_SYMBOL_GPL(iommu_group_get);
575 * iommu_group_put - Decrement group reference
576 * @group: the group to use
578 * This function is called by iommu drivers and users to release the
579 * iommu group. Once the reference count is zero, the group is released.
581 void iommu_group_put(struct iommu_group *group)
584 kobject_put(group->devices_kobj);
586 EXPORT_SYMBOL_GPL(iommu_group_put);
589 * iommu_group_register_notifier - Register a notifier for group changes
590 * @group: the group to watch
591 * @nb: notifier block to signal
593 * This function allows iommu group users to track changes in a group.
594 * See include/linux/iommu.h for actions sent via this notifier. Caller
595 * should hold a reference to the group throughout notifier registration.
597 int iommu_group_register_notifier(struct iommu_group *group,
598 struct notifier_block *nb)
600 return blocking_notifier_chain_register(&group->notifier, nb);
602 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
605 * iommu_group_unregister_notifier - Unregister a notifier
606 * @group: the group to watch
607 * @nb: notifier block to signal
609 * Unregister a previously registered group notifier block.
611 int iommu_group_unregister_notifier(struct iommu_group *group,
612 struct notifier_block *nb)
614 return blocking_notifier_chain_unregister(&group->notifier, nb);
616 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
619 * iommu_group_id - Return ID for a group
620 * @group: the group to ID
622 * Return the unique ID for the group matching the sysfs group number.
624 int iommu_group_id(struct iommu_group *group)
628 EXPORT_SYMBOL_GPL(iommu_group_id);
630 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
631 unsigned long *devfns);
634 * To consider a PCI device isolated, we require ACS to support Source
635 * Validation, Request Redirection, Completer Redirection, and Upstream
636 * Forwarding. This effectively means that devices cannot spoof their
637 * requester ID, requests and completions cannot be redirected, and all
638 * transactions are forwarded upstream, even as it passes through a
639 * bridge where the target device is downstream.
641 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
644 * For multifunction devices which are not isolated from each other, find
645 * all the other non-isolated functions and look for existing groups. For
646 * each function, we also need to look for aliases to or from other devices
647 * that may already have a group.
649 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
650 unsigned long *devfns)
652 struct pci_dev *tmp = NULL;
653 struct iommu_group *group;
655 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
658 for_each_pci_dev(tmp) {
659 if (tmp == pdev || tmp->bus != pdev->bus ||
660 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
661 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
664 group = get_pci_alias_group(tmp, devfns);
675 * Look for aliases to or from the given device for exisiting groups. The
676 * dma_alias_devfn only supports aliases on the same bus, therefore the search
677 * space is quite small (especially since we're really only looking at pcie
678 * device, and therefore only expect multiple slots on the root complex or
679 * downstream switch ports). It's conceivable though that a pair of
680 * multifunction devices could have aliases between them that would cause a
681 * loop. To prevent this, we use a bitmap to track where we've been.
683 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
684 unsigned long *devfns)
686 struct pci_dev *tmp = NULL;
687 struct iommu_group *group;
689 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
692 group = iommu_group_get(&pdev->dev);
696 for_each_pci_dev(tmp) {
697 if (tmp == pdev || tmp->bus != pdev->bus)
700 /* We alias them or they alias us */
701 if (((pdev->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
702 pdev->dma_alias_devfn == tmp->devfn) ||
703 ((tmp->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
704 tmp->dma_alias_devfn == pdev->devfn)) {
706 group = get_pci_alias_group(tmp, devfns);
712 group = get_pci_function_alias_group(tmp, devfns);
723 struct group_for_pci_data {
724 struct pci_dev *pdev;
725 struct iommu_group *group;
729 * DMA alias iterator callback, return the last seen device. Stop and return
730 * the IOMMU group if we find one along the way.
732 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
734 struct group_for_pci_data *data = opaque;
737 data->group = iommu_group_get(&pdev->dev);
739 return data->group != NULL;
743 * Generic device_group call-back function. It just allocates one
744 * iommu-group per device.
746 struct iommu_group *generic_device_group(struct device *dev)
748 struct iommu_group *group;
750 group = iommu_group_alloc();
758 * Use standard PCI bus topology, isolation features, and DMA alias quirks
759 * to find or create an IOMMU group for a device.
761 struct iommu_group *pci_device_group(struct device *dev)
763 struct pci_dev *pdev = to_pci_dev(dev);
764 struct group_for_pci_data data;
766 struct iommu_group *group = NULL;
767 u64 devfns[4] = { 0 };
769 if (WARN_ON(!dev_is_pci(dev)))
770 return ERR_PTR(-EINVAL);
773 * Find the upstream DMA alias for the device. A device must not
774 * be aliased due to topology in order to have its own IOMMU group.
775 * If we find an alias along the way that already belongs to a
778 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
784 * Continue upstream from the point of minimum IOMMU granularity
785 * due to aliases to the point where devices are protected from
786 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
789 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
793 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
798 group = iommu_group_get(&pdev->dev);
804 * Look for existing groups on device aliases. If we alias another
805 * device or another device aliases us, use the same group.
807 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
812 * Look for existing groups on non-isolated functions on the same
813 * slot and aliases of those funcions, if any. No need to clear
814 * the search bitmap, the tested devfns are still valid.
816 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
820 /* No shared group found, allocate new */
821 group = iommu_group_alloc();
829 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
830 * @dev: target device
832 * This function is intended to be called by IOMMU drivers and extended to
833 * support common, bus-defined algorithms when determining or creating the
834 * IOMMU group for a device. On success, the caller will hold a reference
835 * to the returned IOMMU group, which will already include the provided
836 * device. The reference should be released with iommu_group_put().
838 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
840 const struct iommu_ops *ops = dev->bus->iommu_ops;
841 struct iommu_group *group;
844 group = iommu_group_get(dev);
848 group = ERR_PTR(-EINVAL);
850 if (ops && ops->device_group)
851 group = ops->device_group(dev);
857 * Try to allocate a default domain - needs support from the
860 if (!group->default_domain) {
861 group->default_domain = __iommu_domain_alloc(dev->bus,
864 group->domain = group->default_domain;
867 ret = iommu_group_add_device(group, dev);
869 iommu_group_put(group);
876 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
878 return group->default_domain;
881 static int add_iommu_group(struct device *dev, void *data)
883 struct iommu_callback_data *cb = data;
884 const struct iommu_ops *ops = cb->ops;
887 if (!ops->add_device)
890 WARN_ON(dev->iommu_group);
892 ret = ops->add_device(dev);
895 * We ignore -ENODEV errors for now, as they just mean that the
896 * device is not translated by an IOMMU. We still care about
897 * other errors and fail to initialize when they happen.
905 static int remove_iommu_group(struct device *dev, void *data)
907 struct iommu_callback_data *cb = data;
908 const struct iommu_ops *ops = cb->ops;
910 if (ops->remove_device && dev->iommu_group)
911 ops->remove_device(dev);
916 static int iommu_bus_notifier(struct notifier_block *nb,
917 unsigned long action, void *data)
919 struct device *dev = data;
920 const struct iommu_ops *ops = dev->bus->iommu_ops;
921 struct iommu_group *group;
922 unsigned long group_action = 0;
925 * ADD/DEL call into iommu driver ops if provided, which may
926 * result in ADD/DEL notifiers to group->notifier
928 if (action == BUS_NOTIFY_ADD_DEVICE) {
930 return ops->add_device(dev);
931 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
932 if (ops->remove_device && dev->iommu_group) {
933 ops->remove_device(dev);
939 * Remaining BUS_NOTIFYs get filtered and republished to the
940 * group, if anyone is listening
942 group = iommu_group_get(dev);
947 case BUS_NOTIFY_BIND_DRIVER:
948 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
950 case BUS_NOTIFY_BOUND_DRIVER:
951 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
953 case BUS_NOTIFY_UNBIND_DRIVER:
954 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
956 case BUS_NOTIFY_UNBOUND_DRIVER:
957 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
962 blocking_notifier_call_chain(&group->notifier,
965 iommu_group_put(group);
969 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
972 struct notifier_block *nb;
973 struct iommu_callback_data cb = {
977 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
981 nb->notifier_call = iommu_bus_notifier;
983 err = bus_register_notifier(bus, nb);
987 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
996 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
997 bus_unregister_notifier(bus, nb);
1006 * bus_set_iommu - set iommu-callbacks for the bus
1008 * @ops: the callbacks provided by the iommu-driver
1010 * This function is called by an iommu driver to set the iommu methods
1011 * used for a particular bus. Drivers for devices on that bus can use
1012 * the iommu-api after these ops are registered.
1013 * This special function is needed because IOMMUs are usually devices on
1014 * the bus itself, so the iommu drivers are not initialized when the bus
1015 * is set up. With this function the iommu-driver can set the iommu-ops
1018 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1022 if (bus->iommu_ops != NULL)
1025 bus->iommu_ops = ops;
1027 /* Do IOMMU specific setup for this bus-type */
1028 err = iommu_bus_init(bus, ops);
1030 bus->iommu_ops = NULL;
1034 EXPORT_SYMBOL_GPL(bus_set_iommu);
1036 bool iommu_present(struct bus_type *bus)
1038 return bus->iommu_ops != NULL;
1040 EXPORT_SYMBOL_GPL(iommu_present);
1042 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1044 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1047 return bus->iommu_ops->capable(cap);
1049 EXPORT_SYMBOL_GPL(iommu_capable);
1052 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1053 * @domain: iommu domain
1054 * @handler: fault handler
1055 * @token: user data, will be passed back to the fault handler
1057 * This function should be used by IOMMU users which want to be notified
1058 * whenever an IOMMU fault happens.
1060 * The fault handler itself should return 0 on success, and an appropriate
1061 * error code otherwise.
1063 void iommu_set_fault_handler(struct iommu_domain *domain,
1064 iommu_fault_handler_t handler,
1069 domain->handler = handler;
1070 domain->handler_token = token;
1072 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1074 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1077 struct iommu_domain *domain;
1079 if (bus == NULL || bus->iommu_ops == NULL)
1082 domain = bus->iommu_ops->domain_alloc(type);
1086 domain->ops = bus->iommu_ops;
1087 domain->type = type;
1092 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1094 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1096 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1098 void iommu_domain_free(struct iommu_domain *domain)
1100 domain->ops->domain_free(domain);
1102 EXPORT_SYMBOL_GPL(iommu_domain_free);
1104 static int __iommu_attach_device(struct iommu_domain *domain,
1108 if (unlikely(domain->ops->attach_dev == NULL))
1111 ret = domain->ops->attach_dev(domain, dev);
1113 trace_attach_device_to_domain(dev);
1117 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1119 struct iommu_group *group;
1122 group = iommu_group_get(dev);
1123 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1125 return __iommu_attach_device(domain, dev);
1128 * We have a group - lock it to make sure the device-count doesn't
1129 * change while we are attaching
1131 mutex_lock(&group->mutex);
1133 if (iommu_group_device_count(group) != 1)
1136 ret = __iommu_attach_group(domain, group);
1139 mutex_unlock(&group->mutex);
1140 iommu_group_put(group);
1144 EXPORT_SYMBOL_GPL(iommu_attach_device);
1146 static void __iommu_detach_device(struct iommu_domain *domain,
1149 if (unlikely(domain->ops->detach_dev == NULL))
1152 domain->ops->detach_dev(domain, dev);
1153 trace_detach_device_from_domain(dev);
1156 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1158 struct iommu_group *group;
1160 group = iommu_group_get(dev);
1161 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1163 return __iommu_detach_device(domain, dev);
1165 mutex_lock(&group->mutex);
1166 if (iommu_group_device_count(group) != 1) {
1171 __iommu_detach_group(domain, group);
1174 mutex_unlock(&group->mutex);
1175 iommu_group_put(group);
1177 EXPORT_SYMBOL_GPL(iommu_detach_device);
1179 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1181 struct iommu_domain *domain;
1182 struct iommu_group *group;
1184 group = iommu_group_get(dev);
1185 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1189 domain = group->domain;
1191 iommu_group_put(group);
1195 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1198 * IOMMU groups are really the natrual working unit of the IOMMU, but
1199 * the IOMMU API works on domains and devices. Bridge that gap by
1200 * iterating over the devices in a group. Ideally we'd have a single
1201 * device which represents the requestor ID of the group, but we also
1202 * allow IOMMU drivers to create policy defined minimum sets, where
1203 * the physical hardware may be able to distiguish members, but we
1204 * wish to group them at a higher level (ex. untrusted multi-function
1205 * PCI devices). Thus we attach each device.
1207 static int iommu_group_do_attach_device(struct device *dev, void *data)
1209 struct iommu_domain *domain = data;
1211 return __iommu_attach_device(domain, dev);
1214 static int __iommu_attach_group(struct iommu_domain *domain,
1215 struct iommu_group *group)
1219 if (group->default_domain && group->domain != group->default_domain)
1222 ret = __iommu_group_for_each_dev(group, domain,
1223 iommu_group_do_attach_device);
1225 group->domain = domain;
1230 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1234 mutex_lock(&group->mutex);
1235 ret = __iommu_attach_group(domain, group);
1236 mutex_unlock(&group->mutex);
1240 EXPORT_SYMBOL_GPL(iommu_attach_group);
1242 static int iommu_group_do_detach_device(struct device *dev, void *data)
1244 struct iommu_domain *domain = data;
1246 __iommu_detach_device(domain, dev);
1251 static void __iommu_detach_group(struct iommu_domain *domain,
1252 struct iommu_group *group)
1256 if (!group->default_domain) {
1257 __iommu_group_for_each_dev(group, domain,
1258 iommu_group_do_detach_device);
1259 group->domain = NULL;
1263 if (group->domain == group->default_domain)
1266 /* Detach by re-attaching to the default domain */
1267 ret = __iommu_group_for_each_dev(group, group->default_domain,
1268 iommu_group_do_attach_device);
1272 group->domain = group->default_domain;
1275 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1277 mutex_lock(&group->mutex);
1278 __iommu_detach_group(domain, group);
1279 mutex_unlock(&group->mutex);
1281 EXPORT_SYMBOL_GPL(iommu_detach_group);
1283 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1285 if (unlikely(domain->ops->iova_to_phys == NULL))
1288 return domain->ops->iova_to_phys(domain, iova);
1290 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1292 static size_t iommu_pgsize(struct iommu_domain *domain,
1293 unsigned long addr_merge, size_t size)
1295 unsigned int pgsize_idx;
1298 /* Max page size that still fits into 'size' */
1299 pgsize_idx = __fls(size);
1301 /* need to consider alignment requirements ? */
1302 if (likely(addr_merge)) {
1303 /* Max page size allowed by address */
1304 unsigned int align_pgsize_idx = __ffs(addr_merge);
1305 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1308 /* build a mask of acceptable page sizes */
1309 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1311 /* throw away page sizes not supported by the hardware */
1312 pgsize &= domain->ops->pgsize_bitmap;
1314 /* make sure we're still sane */
1317 /* pick the biggest page */
1318 pgsize_idx = __fls(pgsize);
1319 pgsize = 1UL << pgsize_idx;
1324 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1325 phys_addr_t paddr, size_t size, int prot)
1327 unsigned long orig_iova = iova;
1328 unsigned int min_pagesz;
1329 size_t orig_size = size;
1332 if (unlikely(domain->ops->map == NULL ||
1333 domain->ops->pgsize_bitmap == 0UL))
1336 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1339 /* find out the minimum page size supported */
1340 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1343 * both the virtual address and the physical one, as well as
1344 * the size of the mapping, must be aligned (at least) to the
1345 * size of the smallest page supported by the hardware
1347 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1348 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1349 iova, &paddr, size, min_pagesz);
1353 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1356 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1358 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1359 iova, &paddr, pgsize);
1361 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1370 /* unroll mapping in case something went wrong */
1372 iommu_unmap(domain, orig_iova, orig_size - size);
1374 trace_map(orig_iova, paddr, orig_size);
1378 EXPORT_SYMBOL_GPL(iommu_map);
1380 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1382 size_t unmapped_page, unmapped = 0;
1383 unsigned int min_pagesz;
1384 unsigned long orig_iova = iova;
1386 if (unlikely(domain->ops->unmap == NULL ||
1387 domain->ops->pgsize_bitmap == 0UL))
1390 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1393 /* find out the minimum page size supported */
1394 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1397 * The virtual address, as well as the size of the mapping, must be
1398 * aligned (at least) to the size of the smallest page supported
1401 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1402 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1403 iova, size, min_pagesz);
1407 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1410 * Keep iterating until we either unmap 'size' bytes (or more)
1411 * or we hit an area that isn't mapped.
1413 while (unmapped < size) {
1414 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1416 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1420 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1421 iova, unmapped_page);
1423 iova += unmapped_page;
1424 unmapped += unmapped_page;
1427 trace_unmap(orig_iova, size, unmapped);
1430 EXPORT_SYMBOL_GPL(iommu_unmap);
1432 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1433 struct scatterlist *sg, unsigned int nents, int prot)
1435 struct scatterlist *s;
1437 unsigned int i, min_pagesz;
1440 if (unlikely(domain->ops->pgsize_bitmap == 0UL))
1443 min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1445 for_each_sg(sg, s, nents, i) {
1446 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1449 * We are mapping on IOMMU page boundaries, so offset within
1450 * the page must be 0. However, the IOMMU may support pages
1451 * smaller than PAGE_SIZE, so s->offset may still represent
1452 * an offset of that boundary within the CPU page.
1454 if (!IS_ALIGNED(s->offset, min_pagesz))
1457 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1461 mapped += s->length;
1467 /* undo mappings already done */
1468 iommu_unmap(domain, iova, mapped);
1473 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1475 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1476 phys_addr_t paddr, u64 size, int prot)
1478 if (unlikely(domain->ops->domain_window_enable == NULL))
1481 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1484 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1486 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1488 if (unlikely(domain->ops->domain_window_disable == NULL))
1491 return domain->ops->domain_window_disable(domain, wnd_nr);
1493 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1495 static int __init iommu_init(void)
1497 iommu_group_kset = kset_create_and_add("iommu_groups",
1499 ida_init(&iommu_group_ida);
1500 mutex_init(&iommu_group_mutex);
1502 BUG_ON(!iommu_group_kset);
1506 core_initcall(iommu_init);
1508 int iommu_domain_get_attr(struct iommu_domain *domain,
1509 enum iommu_attr attr, void *data)
1511 struct iommu_domain_geometry *geometry;
1517 case DOMAIN_ATTR_GEOMETRY:
1519 *geometry = domain->geometry;
1522 case DOMAIN_ATTR_PAGING:
1524 *paging = (domain->ops->pgsize_bitmap != 0UL);
1526 case DOMAIN_ATTR_WINDOWS:
1529 if (domain->ops->domain_get_windows != NULL)
1530 *count = domain->ops->domain_get_windows(domain);
1536 if (!domain->ops->domain_get_attr)
1539 ret = domain->ops->domain_get_attr(domain, attr, data);
1544 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1546 int iommu_domain_set_attr(struct iommu_domain *domain,
1547 enum iommu_attr attr, void *data)
1553 case DOMAIN_ATTR_WINDOWS:
1556 if (domain->ops->domain_set_windows != NULL)
1557 ret = domain->ops->domain_set_windows(domain, *count);
1563 if (domain->ops->domain_set_attr == NULL)
1566 ret = domain->ops->domain_set_attr(domain, attr, data);
1571 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1573 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1575 const struct iommu_ops *ops = dev->bus->iommu_ops;
1577 if (ops && ops->get_dm_regions)
1578 ops->get_dm_regions(dev, list);
1581 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1583 const struct iommu_ops *ops = dev->bus->iommu_ops;
1585 if (ops && ops->put_dm_regions)
1586 ops->put_dm_regions(dev, list);
1589 /* Request that a device is direct mapped by the IOMMU */
1590 int iommu_request_dm_for_dev(struct device *dev)
1592 struct iommu_domain *dm_domain;
1593 struct iommu_group *group;
1596 /* Device must already be in a group before calling this function */
1597 group = iommu_group_get(dev);
1601 mutex_lock(&group->mutex);
1603 /* Check if the default domain is already direct mapped */
1605 if (group->default_domain &&
1606 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1609 /* Don't change mappings of existing devices */
1611 if (iommu_group_device_count(group) != 1)
1614 /* Allocate a direct mapped domain */
1616 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1620 /* Attach the device to the domain */
1621 ret = __iommu_attach_group(dm_domain, group);
1623 iommu_domain_free(dm_domain);
1627 /* Make the direct mapped domain the default for this group */
1628 if (group->default_domain)
1629 iommu_domain_free(group->default_domain);
1630 group->default_domain = dm_domain;
1632 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1636 mutex_unlock(&group->mutex);
1637 iommu_group_put(group);