1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
4 * Author: Joerg Roedel <jroedel@suse.de>
7 #define pr_fmt(fmt) "iommu: " fmt
9 #include <linux/device.h>
10 #include <linux/kernel.h>
11 #include <linux/bug.h>
12 #include <linux/types.h>
13 #include <linux/init.h>
14 #include <linux/export.h>
15 #include <linux/slab.h>
16 #include <linux/errno.h>
17 #include <linux/iommu.h>
18 #include <linux/idr.h>
19 #include <linux/notifier.h>
20 #include <linux/err.h>
21 #include <linux/pci.h>
22 #include <linux/bitops.h>
23 #include <linux/property.h>
24 #include <linux/fsl/mc.h>
25 #include <linux/module.h>
26 #include <trace/events/iommu.h>
28 static struct kset *iommu_group_kset;
29 static DEFINE_IDA(iommu_group_ida);
31 static unsigned int iommu_def_domain_type __read_mostly;
32 static bool iommu_dma_strict __read_mostly = true;
33 static u32 iommu_cmd_line __read_mostly;
37 struct kobject *devices_kobj;
38 struct list_head devices;
40 struct blocking_notifier_head notifier;
42 void (*iommu_data_release)(void *iommu_data);
45 struct iommu_domain *default_domain;
46 struct iommu_domain *domain;
50 struct list_head list;
55 struct iommu_group_attribute {
56 struct attribute attr;
57 ssize_t (*show)(struct iommu_group *group, char *buf);
58 ssize_t (*store)(struct iommu_group *group,
59 const char *buf, size_t count);
62 static const char * const iommu_group_resv_type_string[] = {
63 [IOMMU_RESV_DIRECT] = "direct",
64 [IOMMU_RESV_DIRECT_RELAXABLE] = "direct-relaxable",
65 [IOMMU_RESV_RESERVED] = "reserved",
66 [IOMMU_RESV_MSI] = "msi",
67 [IOMMU_RESV_SW_MSI] = "msi",
70 #define IOMMU_CMD_LINE_DMA_API BIT(0)
72 static void iommu_set_cmd_line_dma_api(void)
74 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
77 static bool iommu_cmd_line_dma_api(void)
79 return !!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API);
82 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
83 struct iommu_group_attribute iommu_group_attr_##_name = \
84 __ATTR(_name, _mode, _show, _store)
86 #define to_iommu_group_attr(_attr) \
87 container_of(_attr, struct iommu_group_attribute, attr)
88 #define to_iommu_group(_kobj) \
89 container_of(_kobj, struct iommu_group, kobj)
91 static LIST_HEAD(iommu_device_list);
92 static DEFINE_SPINLOCK(iommu_device_lock);
95 * Use a function instead of an array here because the domain-type is a
96 * bit-field, so an array would waste memory.
98 static const char *iommu_domain_type_str(unsigned int t)
101 case IOMMU_DOMAIN_BLOCKED:
103 case IOMMU_DOMAIN_IDENTITY:
104 return "Passthrough";
105 case IOMMU_DOMAIN_UNMANAGED:
107 case IOMMU_DOMAIN_DMA:
114 static int __init iommu_subsys_init(void)
116 bool cmd_line = iommu_cmd_line_dma_api();
119 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
120 iommu_set_default_passthrough(false);
122 iommu_set_default_translated(false);
124 if (iommu_default_passthrough() && mem_encrypt_active()) {
125 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
126 iommu_set_default_translated(false);
130 pr_info("Default domain type: %s %s\n",
131 iommu_domain_type_str(iommu_def_domain_type),
132 cmd_line ? "(set via kernel command line)" : "");
136 subsys_initcall(iommu_subsys_init);
138 int iommu_device_register(struct iommu_device *iommu)
140 spin_lock(&iommu_device_lock);
141 list_add_tail(&iommu->list, &iommu_device_list);
142 spin_unlock(&iommu_device_lock);
145 EXPORT_SYMBOL_GPL(iommu_device_register);
147 void iommu_device_unregister(struct iommu_device *iommu)
149 spin_lock(&iommu_device_lock);
150 list_del(&iommu->list);
151 spin_unlock(&iommu_device_lock);
153 EXPORT_SYMBOL_GPL(iommu_device_unregister);
155 static struct dev_iommu *dev_iommu_get(struct device *dev)
157 struct dev_iommu *param = dev->iommu;
162 param = kzalloc(sizeof(*param), GFP_KERNEL);
166 mutex_init(¶m->lock);
171 static void dev_iommu_free(struct device *dev)
173 iommu_fwspec_free(dev);
178 int iommu_probe_device(struct device *dev)
180 const struct iommu_ops *ops = dev->bus->iommu_ops;
183 WARN_ON(dev->iommu_group);
187 if (!dev_iommu_get(dev))
190 if (!try_module_get(ops->owner)) {
192 goto err_free_dev_param;
195 ret = ops->add_device(dev);
202 module_put(ops->owner);
208 void iommu_release_device(struct device *dev)
210 const struct iommu_ops *ops = dev->bus->iommu_ops;
212 if (dev->iommu_group)
213 ops->remove_device(dev);
216 module_put(ops->owner);
221 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
223 static int __iommu_attach_device(struct iommu_domain *domain,
225 static int __iommu_attach_group(struct iommu_domain *domain,
226 struct iommu_group *group);
227 static void __iommu_detach_group(struct iommu_domain *domain,
228 struct iommu_group *group);
230 static int __init iommu_set_def_domain_type(char *str)
235 ret = kstrtobool(str, &pt);
240 iommu_set_default_passthrough(true);
242 iommu_set_default_translated(true);
246 early_param("iommu.passthrough", iommu_set_def_domain_type);
248 static int __init iommu_dma_setup(char *str)
250 return kstrtobool(str, &iommu_dma_strict);
252 early_param("iommu.strict", iommu_dma_setup);
254 static ssize_t iommu_group_attr_show(struct kobject *kobj,
255 struct attribute *__attr, char *buf)
257 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
258 struct iommu_group *group = to_iommu_group(kobj);
262 ret = attr->show(group, buf);
266 static ssize_t iommu_group_attr_store(struct kobject *kobj,
267 struct attribute *__attr,
268 const char *buf, size_t count)
270 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
271 struct iommu_group *group = to_iommu_group(kobj);
275 ret = attr->store(group, buf, count);
279 static const struct sysfs_ops iommu_group_sysfs_ops = {
280 .show = iommu_group_attr_show,
281 .store = iommu_group_attr_store,
284 static int iommu_group_create_file(struct iommu_group *group,
285 struct iommu_group_attribute *attr)
287 return sysfs_create_file(&group->kobj, &attr->attr);
290 static void iommu_group_remove_file(struct iommu_group *group,
291 struct iommu_group_attribute *attr)
293 sysfs_remove_file(&group->kobj, &attr->attr);
296 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
298 return sprintf(buf, "%s\n", group->name);
302 * iommu_insert_resv_region - Insert a new region in the
303 * list of reserved regions.
304 * @new: new region to insert
305 * @regions: list of regions
307 * Elements are sorted by start address and overlapping segments
308 * of the same type are merged.
310 int iommu_insert_resv_region(struct iommu_resv_region *new,
311 struct list_head *regions)
313 struct iommu_resv_region *iter, *tmp, *nr, *top;
316 nr = iommu_alloc_resv_region(new->start, new->length,
317 new->prot, new->type);
321 /* First add the new element based on start address sorting */
322 list_for_each_entry(iter, regions, list) {
323 if (nr->start < iter->start ||
324 (nr->start == iter->start && nr->type <= iter->type))
327 list_add_tail(&nr->list, &iter->list);
329 /* Merge overlapping segments of type nr->type in @regions, if any */
330 list_for_each_entry_safe(iter, tmp, regions, list) {
331 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
333 /* no merge needed on elements of different types than @new */
334 if (iter->type != new->type) {
335 list_move_tail(&iter->list, &stack);
339 /* look for the last stack element of same type as @iter */
340 list_for_each_entry_reverse(top, &stack, list)
341 if (top->type == iter->type)
344 list_move_tail(&iter->list, &stack);
348 top_end = top->start + top->length - 1;
350 if (iter->start > top_end + 1) {
351 list_move_tail(&iter->list, &stack);
353 top->length = max(top_end, iter_end) - top->start + 1;
354 list_del(&iter->list);
358 list_splice(&stack, regions);
363 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
364 struct list_head *group_resv_regions)
366 struct iommu_resv_region *entry;
369 list_for_each_entry(entry, dev_resv_regions, list) {
370 ret = iommu_insert_resv_region(entry, group_resv_regions);
377 int iommu_get_group_resv_regions(struct iommu_group *group,
378 struct list_head *head)
380 struct group_device *device;
383 mutex_lock(&group->mutex);
384 list_for_each_entry(device, &group->devices, list) {
385 struct list_head dev_resv_regions;
387 INIT_LIST_HEAD(&dev_resv_regions);
388 iommu_get_resv_regions(device->dev, &dev_resv_regions);
389 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
390 iommu_put_resv_regions(device->dev, &dev_resv_regions);
394 mutex_unlock(&group->mutex);
397 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
399 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
402 struct iommu_resv_region *region, *next;
403 struct list_head group_resv_regions;
406 INIT_LIST_HEAD(&group_resv_regions);
407 iommu_get_group_resv_regions(group, &group_resv_regions);
409 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
410 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
411 (long long int)region->start,
412 (long long int)(region->start +
414 iommu_group_resv_type_string[region->type]);
421 static ssize_t iommu_group_show_type(struct iommu_group *group,
424 char *type = "unknown\n";
426 if (group->default_domain) {
427 switch (group->default_domain->type) {
428 case IOMMU_DOMAIN_BLOCKED:
431 case IOMMU_DOMAIN_IDENTITY:
434 case IOMMU_DOMAIN_UNMANAGED:
435 type = "unmanaged\n";
437 case IOMMU_DOMAIN_DMA:
447 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
449 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
450 iommu_group_show_resv_regions, NULL);
452 static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
454 static void iommu_group_release(struct kobject *kobj)
456 struct iommu_group *group = to_iommu_group(kobj);
458 pr_debug("Releasing group %d\n", group->id);
460 if (group->iommu_data_release)
461 group->iommu_data_release(group->iommu_data);
463 ida_simple_remove(&iommu_group_ida, group->id);
465 if (group->default_domain)
466 iommu_domain_free(group->default_domain);
472 static struct kobj_type iommu_group_ktype = {
473 .sysfs_ops = &iommu_group_sysfs_ops,
474 .release = iommu_group_release,
478 * iommu_group_alloc - Allocate a new group
480 * This function is called by an iommu driver to allocate a new iommu
481 * group. The iommu group represents the minimum granularity of the iommu.
482 * Upon successful return, the caller holds a reference to the supplied
483 * group in order to hold the group until devices are added. Use
484 * iommu_group_put() to release this extra reference count, allowing the
485 * group to be automatically reclaimed once it has no devices or external
488 struct iommu_group *iommu_group_alloc(void)
490 struct iommu_group *group;
493 group = kzalloc(sizeof(*group), GFP_KERNEL);
495 return ERR_PTR(-ENOMEM);
497 group->kobj.kset = iommu_group_kset;
498 mutex_init(&group->mutex);
499 INIT_LIST_HEAD(&group->devices);
500 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
502 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
509 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
510 NULL, "%d", group->id);
512 ida_simple_remove(&iommu_group_ida, group->id);
517 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
518 if (!group->devices_kobj) {
519 kobject_put(&group->kobj); /* triggers .release & free */
520 return ERR_PTR(-ENOMEM);
524 * The devices_kobj holds a reference on the group kobject, so
525 * as long as that exists so will the group. We can therefore
526 * use the devices_kobj for reference counting.
528 kobject_put(&group->kobj);
530 ret = iommu_group_create_file(group,
531 &iommu_group_attr_reserved_regions);
535 ret = iommu_group_create_file(group, &iommu_group_attr_type);
539 pr_debug("Allocated group %d\n", group->id);
543 EXPORT_SYMBOL_GPL(iommu_group_alloc);
545 struct iommu_group *iommu_group_get_by_id(int id)
547 struct kobject *group_kobj;
548 struct iommu_group *group;
551 if (!iommu_group_kset)
554 name = kasprintf(GFP_KERNEL, "%d", id);
558 group_kobj = kset_find_obj(iommu_group_kset, name);
564 group = container_of(group_kobj, struct iommu_group, kobj);
565 BUG_ON(group->id != id);
567 kobject_get(group->devices_kobj);
568 kobject_put(&group->kobj);
572 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
575 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
578 * iommu drivers can store data in the group for use when doing iommu
579 * operations. This function provides a way to retrieve it. Caller
580 * should hold a group reference.
582 void *iommu_group_get_iommudata(struct iommu_group *group)
584 return group->iommu_data;
586 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
589 * iommu_group_set_iommudata - set iommu_data for a group
591 * @iommu_data: new data
592 * @release: release function for iommu_data
594 * iommu drivers can store data in the group for use when doing iommu
595 * operations. This function provides a way to set the data after
596 * the group has been allocated. Caller should hold a group reference.
598 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
599 void (*release)(void *iommu_data))
601 group->iommu_data = iommu_data;
602 group->iommu_data_release = release;
604 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
607 * iommu_group_set_name - set name for a group
611 * Allow iommu driver to set a name for a group. When set it will
612 * appear in a name attribute file under the group in sysfs.
614 int iommu_group_set_name(struct iommu_group *group, const char *name)
619 iommu_group_remove_file(group, &iommu_group_attr_name);
626 group->name = kstrdup(name, GFP_KERNEL);
630 ret = iommu_group_create_file(group, &iommu_group_attr_name);
639 EXPORT_SYMBOL_GPL(iommu_group_set_name);
641 static int iommu_group_create_direct_mappings(struct iommu_group *group,
644 struct iommu_domain *domain = group->default_domain;
645 struct iommu_resv_region *entry;
646 struct list_head mappings;
647 unsigned long pg_size;
650 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
653 BUG_ON(!domain->pgsize_bitmap);
655 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
656 INIT_LIST_HEAD(&mappings);
658 iommu_get_resv_regions(dev, &mappings);
660 /* We need to consider overlapping regions for different devices */
661 list_for_each_entry(entry, &mappings, list) {
662 dma_addr_t start, end, addr;
664 if (domain->ops->apply_resv_region)
665 domain->ops->apply_resv_region(dev, domain, entry);
667 start = ALIGN(entry->start, pg_size);
668 end = ALIGN(entry->start + entry->length, pg_size);
670 if (entry->type != IOMMU_RESV_DIRECT &&
671 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
674 for (addr = start; addr < end; addr += pg_size) {
675 phys_addr_t phys_addr;
677 phys_addr = iommu_iova_to_phys(domain, addr);
681 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
688 iommu_flush_tlb_all(domain);
691 iommu_put_resv_regions(dev, &mappings);
697 * iommu_group_add_device - add a device to an iommu group
698 * @group: the group into which to add the device (reference should be held)
701 * This function is called by an iommu driver to add a device into a
702 * group. Adding a device increments the group reference count.
704 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
707 struct group_device *device;
709 device = kzalloc(sizeof(*device), GFP_KERNEL);
715 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
717 goto err_free_device;
719 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
723 goto err_remove_link;
726 ret = sysfs_create_link_nowarn(group->devices_kobj,
727 &dev->kobj, device->name);
729 if (ret == -EEXIST && i >= 0) {
731 * Account for the slim chance of collision
732 * and append an instance to the name.
735 device->name = kasprintf(GFP_KERNEL, "%s.%d",
736 kobject_name(&dev->kobj), i++);
742 kobject_get(group->devices_kobj);
744 dev->iommu_group = group;
746 iommu_group_create_direct_mappings(group, dev);
748 mutex_lock(&group->mutex);
749 list_add_tail(&device->list, &group->devices);
751 ret = __iommu_attach_device(group->domain, dev);
752 mutex_unlock(&group->mutex);
756 /* Notify any listeners about change to group. */
757 blocking_notifier_call_chain(&group->notifier,
758 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
760 trace_add_device_to_group(group->id, dev);
762 dev_info(dev, "Adding to iommu group %d\n", group->id);
767 mutex_lock(&group->mutex);
768 list_del(&device->list);
769 mutex_unlock(&group->mutex);
770 dev->iommu_group = NULL;
771 kobject_put(group->devices_kobj);
772 sysfs_remove_link(group->devices_kobj, device->name);
776 sysfs_remove_link(&dev->kobj, "iommu_group");
779 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
782 EXPORT_SYMBOL_GPL(iommu_group_add_device);
785 * iommu_group_remove_device - remove a device from it's current group
786 * @dev: device to be removed
788 * This function is called by an iommu driver to remove the device from
789 * it's current group. This decrements the iommu group reference count.
791 void iommu_group_remove_device(struct device *dev)
793 struct iommu_group *group = dev->iommu_group;
794 struct group_device *tmp_device, *device = NULL;
796 dev_info(dev, "Removing from iommu group %d\n", group->id);
798 /* Pre-notify listeners that a device is being removed. */
799 blocking_notifier_call_chain(&group->notifier,
800 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
802 mutex_lock(&group->mutex);
803 list_for_each_entry(tmp_device, &group->devices, list) {
804 if (tmp_device->dev == dev) {
806 list_del(&device->list);
810 mutex_unlock(&group->mutex);
815 sysfs_remove_link(group->devices_kobj, device->name);
816 sysfs_remove_link(&dev->kobj, "iommu_group");
818 trace_remove_device_from_group(group->id, dev);
822 dev->iommu_group = NULL;
823 kobject_put(group->devices_kobj);
825 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
827 static int iommu_group_device_count(struct iommu_group *group)
829 struct group_device *entry;
832 list_for_each_entry(entry, &group->devices, list)
839 * iommu_group_for_each_dev - iterate over each device in the group
841 * @data: caller opaque data to be passed to callback function
842 * @fn: caller supplied callback function
844 * This function is called by group users to iterate over group devices.
845 * Callers should hold a reference count to the group during callback.
846 * The group->mutex is held across callbacks, which will block calls to
847 * iommu_group_add/remove_device.
849 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
850 int (*fn)(struct device *, void *))
852 struct group_device *device;
855 list_for_each_entry(device, &group->devices, list) {
856 ret = fn(device->dev, data);
864 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
865 int (*fn)(struct device *, void *))
869 mutex_lock(&group->mutex);
870 ret = __iommu_group_for_each_dev(group, data, fn);
871 mutex_unlock(&group->mutex);
875 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
878 * iommu_group_get - Return the group for a device and increment reference
879 * @dev: get the group that this device belongs to
881 * This function is called by iommu drivers and users to get the group
882 * for the specified device. If found, the group is returned and the group
883 * reference in incremented, else NULL.
885 struct iommu_group *iommu_group_get(struct device *dev)
887 struct iommu_group *group = dev->iommu_group;
890 kobject_get(group->devices_kobj);
894 EXPORT_SYMBOL_GPL(iommu_group_get);
897 * iommu_group_ref_get - Increment reference on a group
898 * @group: the group to use, must not be NULL
900 * This function is called by iommu drivers to take additional references on an
901 * existing group. Returns the given group for convenience.
903 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
905 kobject_get(group->devices_kobj);
908 EXPORT_SYMBOL_GPL(iommu_group_ref_get);
911 * iommu_group_put - Decrement group reference
912 * @group: the group to use
914 * This function is called by iommu drivers and users to release the
915 * iommu group. Once the reference count is zero, the group is released.
917 void iommu_group_put(struct iommu_group *group)
920 kobject_put(group->devices_kobj);
922 EXPORT_SYMBOL_GPL(iommu_group_put);
925 * iommu_group_register_notifier - Register a notifier for group changes
926 * @group: the group to watch
927 * @nb: notifier block to signal
929 * This function allows iommu group users to track changes in a group.
930 * See include/linux/iommu.h for actions sent via this notifier. Caller
931 * should hold a reference to the group throughout notifier registration.
933 int iommu_group_register_notifier(struct iommu_group *group,
934 struct notifier_block *nb)
936 return blocking_notifier_chain_register(&group->notifier, nb);
938 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
941 * iommu_group_unregister_notifier - Unregister a notifier
942 * @group: the group to watch
943 * @nb: notifier block to signal
945 * Unregister a previously registered group notifier block.
947 int iommu_group_unregister_notifier(struct iommu_group *group,
948 struct notifier_block *nb)
950 return blocking_notifier_chain_unregister(&group->notifier, nb);
952 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
955 * iommu_register_device_fault_handler() - Register a device fault handler
957 * @handler: the fault handler
958 * @data: private data passed as argument to the handler
960 * When an IOMMU fault event is received, this handler gets called with the
961 * fault event and data as argument. The handler should return 0 on success. If
962 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
963 * complete the fault by calling iommu_page_response() with one of the following
965 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
966 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
967 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
968 * page faults if possible.
970 * Return 0 if the fault handler was installed successfully, or an error.
972 int iommu_register_device_fault_handler(struct device *dev,
973 iommu_dev_fault_handler_t handler,
976 struct dev_iommu *param = dev->iommu;
982 mutex_lock(¶m->lock);
983 /* Only allow one fault handler registered for each device */
984 if (param->fault_param) {
990 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
991 if (!param->fault_param) {
996 param->fault_param->handler = handler;
997 param->fault_param->data = data;
998 mutex_init(¶m->fault_param->lock);
999 INIT_LIST_HEAD(¶m->fault_param->faults);
1002 mutex_unlock(¶m->lock);
1006 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1009 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1012 * Remove the device fault handler installed with
1013 * iommu_register_device_fault_handler().
1015 * Return 0 on success, or an error.
1017 int iommu_unregister_device_fault_handler(struct device *dev)
1019 struct dev_iommu *param = dev->iommu;
1025 mutex_lock(¶m->lock);
1027 if (!param->fault_param)
1030 /* we cannot unregister handler if there are pending faults */
1031 if (!list_empty(¶m->fault_param->faults)) {
1036 kfree(param->fault_param);
1037 param->fault_param = NULL;
1040 mutex_unlock(¶m->lock);
1044 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1047 * iommu_report_device_fault() - Report fault event to device driver
1049 * @evt: fault event data
1051 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1052 * handler. When this function fails and the fault is recoverable, it is the
1053 * caller's responsibility to complete the fault.
1055 * Return 0 on success, or an error.
1057 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1059 struct dev_iommu *param = dev->iommu;
1060 struct iommu_fault_event *evt_pending = NULL;
1061 struct iommu_fault_param *fparam;
1067 /* we only report device fault if there is a handler registered */
1068 mutex_lock(¶m->lock);
1069 fparam = param->fault_param;
1070 if (!fparam || !fparam->handler) {
1075 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1076 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1077 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1083 mutex_lock(&fparam->lock);
1084 list_add_tail(&evt_pending->list, &fparam->faults);
1085 mutex_unlock(&fparam->lock);
1088 ret = fparam->handler(&evt->fault, fparam->data);
1089 if (ret && evt_pending) {
1090 mutex_lock(&fparam->lock);
1091 list_del(&evt_pending->list);
1092 mutex_unlock(&fparam->lock);
1096 mutex_unlock(¶m->lock);
1099 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1101 int iommu_page_response(struct device *dev,
1102 struct iommu_page_response *msg)
1106 struct iommu_fault_event *evt;
1107 struct iommu_fault_page_request *prm;
1108 struct dev_iommu *param = dev->iommu;
1109 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1111 if (!domain || !domain->ops->page_response)
1114 if (!param || !param->fault_param)
1117 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1118 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1121 /* Only send response if there is a fault report pending */
1122 mutex_lock(¶m->fault_param->lock);
1123 if (list_empty(¶m->fault_param->faults)) {
1124 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1128 * Check if we have a matching page request pending to respond,
1129 * otherwise return -EINVAL
1131 list_for_each_entry(evt, ¶m->fault_param->faults, list) {
1132 prm = &evt->fault.prm;
1133 pasid_valid = prm->flags & IOMMU_FAULT_PAGE_REQUEST_PASID_VALID;
1135 if ((pasid_valid && prm->pasid != msg->pasid) ||
1136 prm->grpid != msg->grpid)
1139 /* Sanitize the reply */
1140 msg->flags = pasid_valid ? IOMMU_PAGE_RESP_PASID_VALID : 0;
1142 ret = domain->ops->page_response(dev, evt, msg);
1143 list_del(&evt->list);
1149 mutex_unlock(¶m->fault_param->lock);
1152 EXPORT_SYMBOL_GPL(iommu_page_response);
1155 * iommu_group_id - Return ID for a group
1156 * @group: the group to ID
1158 * Return the unique ID for the group matching the sysfs group number.
1160 int iommu_group_id(struct iommu_group *group)
1164 EXPORT_SYMBOL_GPL(iommu_group_id);
1166 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1167 unsigned long *devfns);
1170 * To consider a PCI device isolated, we require ACS to support Source
1171 * Validation, Request Redirection, Completer Redirection, and Upstream
1172 * Forwarding. This effectively means that devices cannot spoof their
1173 * requester ID, requests and completions cannot be redirected, and all
1174 * transactions are forwarded upstream, even as it passes through a
1175 * bridge where the target device is downstream.
1177 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1180 * For multifunction devices which are not isolated from each other, find
1181 * all the other non-isolated functions and look for existing groups. For
1182 * each function, we also need to look for aliases to or from other devices
1183 * that may already have a group.
1185 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1186 unsigned long *devfns)
1188 struct pci_dev *tmp = NULL;
1189 struct iommu_group *group;
1191 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1194 for_each_pci_dev(tmp) {
1195 if (tmp == pdev || tmp->bus != pdev->bus ||
1196 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1197 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1200 group = get_pci_alias_group(tmp, devfns);
1211 * Look for aliases to or from the given device for existing groups. DMA
1212 * aliases are only supported on the same bus, therefore the search
1213 * space is quite small (especially since we're really only looking at pcie
1214 * device, and therefore only expect multiple slots on the root complex or
1215 * downstream switch ports). It's conceivable though that a pair of
1216 * multifunction devices could have aliases between them that would cause a
1217 * loop. To prevent this, we use a bitmap to track where we've been.
1219 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1220 unsigned long *devfns)
1222 struct pci_dev *tmp = NULL;
1223 struct iommu_group *group;
1225 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1228 group = iommu_group_get(&pdev->dev);
1232 for_each_pci_dev(tmp) {
1233 if (tmp == pdev || tmp->bus != pdev->bus)
1236 /* We alias them or they alias us */
1237 if (pci_devs_are_dma_aliases(pdev, tmp)) {
1238 group = get_pci_alias_group(tmp, devfns);
1244 group = get_pci_function_alias_group(tmp, devfns);
1255 struct group_for_pci_data {
1256 struct pci_dev *pdev;
1257 struct iommu_group *group;
1261 * DMA alias iterator callback, return the last seen device. Stop and return
1262 * the IOMMU group if we find one along the way.
1264 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1266 struct group_for_pci_data *data = opaque;
1269 data->group = iommu_group_get(&pdev->dev);
1271 return data->group != NULL;
1275 * Generic device_group call-back function. It just allocates one
1276 * iommu-group per device.
1278 struct iommu_group *generic_device_group(struct device *dev)
1280 return iommu_group_alloc();
1282 EXPORT_SYMBOL_GPL(generic_device_group);
1285 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1286 * to find or create an IOMMU group for a device.
1288 struct iommu_group *pci_device_group(struct device *dev)
1290 struct pci_dev *pdev = to_pci_dev(dev);
1291 struct group_for_pci_data data;
1292 struct pci_bus *bus;
1293 struct iommu_group *group = NULL;
1294 u64 devfns[4] = { 0 };
1296 if (WARN_ON(!dev_is_pci(dev)))
1297 return ERR_PTR(-EINVAL);
1300 * Find the upstream DMA alias for the device. A device must not
1301 * be aliased due to topology in order to have its own IOMMU group.
1302 * If we find an alias along the way that already belongs to a
1305 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1311 * Continue upstream from the point of minimum IOMMU granularity
1312 * due to aliases to the point where devices are protected from
1313 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1316 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1320 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1325 group = iommu_group_get(&pdev->dev);
1331 * Look for existing groups on device aliases. If we alias another
1332 * device or another device aliases us, use the same group.
1334 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1339 * Look for existing groups on non-isolated functions on the same
1340 * slot and aliases of those funcions, if any. No need to clear
1341 * the search bitmap, the tested devfns are still valid.
1343 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1347 /* No shared group found, allocate new */
1348 return iommu_group_alloc();
1350 EXPORT_SYMBOL_GPL(pci_device_group);
1352 /* Get the IOMMU group for device on fsl-mc bus */
1353 struct iommu_group *fsl_mc_device_group(struct device *dev)
1355 struct device *cont_dev = fsl_mc_cont_dev(dev);
1356 struct iommu_group *group;
1358 group = iommu_group_get(cont_dev);
1360 group = iommu_group_alloc();
1363 EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1366 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1367 * @dev: target device
1369 * This function is intended to be called by IOMMU drivers and extended to
1370 * support common, bus-defined algorithms when determining or creating the
1371 * IOMMU group for a device. On success, the caller will hold a reference
1372 * to the returned IOMMU group, which will already include the provided
1373 * device. The reference should be released with iommu_group_put().
1375 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1377 const struct iommu_ops *ops = dev->bus->iommu_ops;
1378 struct iommu_group *group;
1381 group = iommu_group_get(dev);
1386 return ERR_PTR(-EINVAL);
1388 group = ops->device_group(dev);
1389 if (WARN_ON_ONCE(group == NULL))
1390 return ERR_PTR(-EINVAL);
1396 * Try to allocate a default domain - needs support from the
1399 if (!group->default_domain) {
1400 struct iommu_domain *dom;
1402 dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1403 if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
1404 dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
1407 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1408 iommu_def_domain_type);
1412 group->default_domain = dom;
1414 group->domain = dom;
1416 if (dom && !iommu_dma_strict) {
1418 iommu_domain_set_attr(dom,
1419 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1424 ret = iommu_group_add_device(group, dev);
1426 iommu_group_put(group);
1427 return ERR_PTR(ret);
1432 EXPORT_SYMBOL_GPL(iommu_group_get_for_dev);
1434 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1436 return group->default_domain;
1439 static int add_iommu_group(struct device *dev, void *data)
1441 int ret = iommu_probe_device(dev);
1444 * We ignore -ENODEV errors for now, as they just mean that the
1445 * device is not translated by an IOMMU. We still care about
1446 * other errors and fail to initialize when they happen.
1454 static int remove_iommu_group(struct device *dev, void *data)
1456 iommu_release_device(dev);
1461 static int iommu_bus_notifier(struct notifier_block *nb,
1462 unsigned long action, void *data)
1464 unsigned long group_action = 0;
1465 struct device *dev = data;
1466 struct iommu_group *group;
1469 * ADD/DEL call into iommu driver ops if provided, which may
1470 * result in ADD/DEL notifiers to group->notifier
1472 if (action == BUS_NOTIFY_ADD_DEVICE) {
1475 ret = iommu_probe_device(dev);
1476 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1477 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1478 iommu_release_device(dev);
1483 * Remaining BUS_NOTIFYs get filtered and republished to the
1484 * group, if anyone is listening
1486 group = iommu_group_get(dev);
1491 case BUS_NOTIFY_BIND_DRIVER:
1492 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1494 case BUS_NOTIFY_BOUND_DRIVER:
1495 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1497 case BUS_NOTIFY_UNBIND_DRIVER:
1498 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1500 case BUS_NOTIFY_UNBOUND_DRIVER:
1501 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1506 blocking_notifier_call_chain(&group->notifier,
1509 iommu_group_put(group);
1513 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1516 struct notifier_block *nb;
1518 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1522 nb->notifier_call = iommu_bus_notifier;
1524 err = bus_register_notifier(bus, nb);
1528 err = bus_for_each_dev(bus, NULL, NULL, add_iommu_group);
1537 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1538 bus_unregister_notifier(bus, nb);
1547 * bus_set_iommu - set iommu-callbacks for the bus
1549 * @ops: the callbacks provided by the iommu-driver
1551 * This function is called by an iommu driver to set the iommu methods
1552 * used for a particular bus. Drivers for devices on that bus can use
1553 * the iommu-api after these ops are registered.
1554 * This special function is needed because IOMMUs are usually devices on
1555 * the bus itself, so the iommu drivers are not initialized when the bus
1556 * is set up. With this function the iommu-driver can set the iommu-ops
1559 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1564 bus->iommu_ops = NULL;
1568 if (bus->iommu_ops != NULL)
1571 bus->iommu_ops = ops;
1573 /* Do IOMMU specific setup for this bus-type */
1574 err = iommu_bus_init(bus, ops);
1576 bus->iommu_ops = NULL;
1580 EXPORT_SYMBOL_GPL(bus_set_iommu);
1582 bool iommu_present(struct bus_type *bus)
1584 return bus->iommu_ops != NULL;
1586 EXPORT_SYMBOL_GPL(iommu_present);
1588 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1590 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1593 return bus->iommu_ops->capable(cap);
1595 EXPORT_SYMBOL_GPL(iommu_capable);
1598 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1599 * @domain: iommu domain
1600 * @handler: fault handler
1601 * @token: user data, will be passed back to the fault handler
1603 * This function should be used by IOMMU users which want to be notified
1604 * whenever an IOMMU fault happens.
1606 * The fault handler itself should return 0 on success, and an appropriate
1607 * error code otherwise.
1609 void iommu_set_fault_handler(struct iommu_domain *domain,
1610 iommu_fault_handler_t handler,
1615 domain->handler = handler;
1616 domain->handler_token = token;
1618 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1620 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1623 struct iommu_domain *domain;
1625 if (bus == NULL || bus->iommu_ops == NULL)
1628 domain = bus->iommu_ops->domain_alloc(type);
1632 domain->ops = bus->iommu_ops;
1633 domain->type = type;
1634 /* Assume all sizes by default; the driver may override this later */
1635 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1640 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1642 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1644 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1646 void iommu_domain_free(struct iommu_domain *domain)
1648 domain->ops->domain_free(domain);
1650 EXPORT_SYMBOL_GPL(iommu_domain_free);
1652 static int __iommu_attach_device(struct iommu_domain *domain,
1656 if ((domain->ops->is_attach_deferred != NULL) &&
1657 domain->ops->is_attach_deferred(domain, dev))
1660 if (unlikely(domain->ops->attach_dev == NULL))
1663 ret = domain->ops->attach_dev(domain, dev);
1665 trace_attach_device_to_domain(dev);
1669 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1671 struct iommu_group *group;
1674 group = iommu_group_get(dev);
1679 * Lock the group to make sure the device-count doesn't
1680 * change while we are attaching
1682 mutex_lock(&group->mutex);
1684 if (iommu_group_device_count(group) != 1)
1687 ret = __iommu_attach_group(domain, group);
1690 mutex_unlock(&group->mutex);
1691 iommu_group_put(group);
1695 EXPORT_SYMBOL_GPL(iommu_attach_device);
1697 int iommu_cache_invalidate(struct iommu_domain *domain, struct device *dev,
1698 struct iommu_cache_invalidate_info *inv_info)
1700 if (unlikely(!domain->ops->cache_invalidate))
1703 return domain->ops->cache_invalidate(domain, dev, inv_info);
1705 EXPORT_SYMBOL_GPL(iommu_cache_invalidate);
1707 int iommu_sva_bind_gpasid(struct iommu_domain *domain,
1708 struct device *dev, struct iommu_gpasid_bind_data *data)
1710 if (unlikely(!domain->ops->sva_bind_gpasid))
1713 return domain->ops->sva_bind_gpasid(domain, dev, data);
1715 EXPORT_SYMBOL_GPL(iommu_sva_bind_gpasid);
1717 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
1720 if (unlikely(!domain->ops->sva_unbind_gpasid))
1723 return domain->ops->sva_unbind_gpasid(dev, pasid);
1725 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
1727 static void __iommu_detach_device(struct iommu_domain *domain,
1730 if ((domain->ops->is_attach_deferred != NULL) &&
1731 domain->ops->is_attach_deferred(domain, dev))
1734 if (unlikely(domain->ops->detach_dev == NULL))
1737 domain->ops->detach_dev(domain, dev);
1738 trace_detach_device_from_domain(dev);
1741 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1743 struct iommu_group *group;
1745 group = iommu_group_get(dev);
1749 mutex_lock(&group->mutex);
1750 if (iommu_group_device_count(group) != 1) {
1755 __iommu_detach_group(domain, group);
1758 mutex_unlock(&group->mutex);
1759 iommu_group_put(group);
1761 EXPORT_SYMBOL_GPL(iommu_detach_device);
1763 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1765 struct iommu_domain *domain;
1766 struct iommu_group *group;
1768 group = iommu_group_get(dev);
1772 domain = group->domain;
1774 iommu_group_put(group);
1778 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1781 * For IOMMU_DOMAIN_DMA implementations which already provide their own
1782 * guarantees that the group and its default domain are valid and correct.
1784 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
1786 return dev->iommu_group->default_domain;
1790 * IOMMU groups are really the natural working unit of the IOMMU, but
1791 * the IOMMU API works on domains and devices. Bridge that gap by
1792 * iterating over the devices in a group. Ideally we'd have a single
1793 * device which represents the requestor ID of the group, but we also
1794 * allow IOMMU drivers to create policy defined minimum sets, where
1795 * the physical hardware may be able to distiguish members, but we
1796 * wish to group them at a higher level (ex. untrusted multi-function
1797 * PCI devices). Thus we attach each device.
1799 static int iommu_group_do_attach_device(struct device *dev, void *data)
1801 struct iommu_domain *domain = data;
1803 return __iommu_attach_device(domain, dev);
1806 static int __iommu_attach_group(struct iommu_domain *domain,
1807 struct iommu_group *group)
1811 if (group->default_domain && group->domain != group->default_domain)
1814 ret = __iommu_group_for_each_dev(group, domain,
1815 iommu_group_do_attach_device);
1817 group->domain = domain;
1822 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1826 mutex_lock(&group->mutex);
1827 ret = __iommu_attach_group(domain, group);
1828 mutex_unlock(&group->mutex);
1832 EXPORT_SYMBOL_GPL(iommu_attach_group);
1834 static int iommu_group_do_detach_device(struct device *dev, void *data)
1836 struct iommu_domain *domain = data;
1838 __iommu_detach_device(domain, dev);
1843 static void __iommu_detach_group(struct iommu_domain *domain,
1844 struct iommu_group *group)
1848 if (!group->default_domain) {
1849 __iommu_group_for_each_dev(group, domain,
1850 iommu_group_do_detach_device);
1851 group->domain = NULL;
1855 if (group->domain == group->default_domain)
1858 /* Detach by re-attaching to the default domain */
1859 ret = __iommu_group_for_each_dev(group, group->default_domain,
1860 iommu_group_do_attach_device);
1864 group->domain = group->default_domain;
1867 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1869 mutex_lock(&group->mutex);
1870 __iommu_detach_group(domain, group);
1871 mutex_unlock(&group->mutex);
1873 EXPORT_SYMBOL_GPL(iommu_detach_group);
1875 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1877 if (unlikely(domain->ops->iova_to_phys == NULL))
1880 return domain->ops->iova_to_phys(domain, iova);
1882 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1884 static size_t iommu_pgsize(struct iommu_domain *domain,
1885 unsigned long addr_merge, size_t size)
1887 unsigned int pgsize_idx;
1890 /* Max page size that still fits into 'size' */
1891 pgsize_idx = __fls(size);
1893 /* need to consider alignment requirements ? */
1894 if (likely(addr_merge)) {
1895 /* Max page size allowed by address */
1896 unsigned int align_pgsize_idx = __ffs(addr_merge);
1897 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1900 /* build a mask of acceptable page sizes */
1901 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1903 /* throw away page sizes not supported by the hardware */
1904 pgsize &= domain->pgsize_bitmap;
1906 /* make sure we're still sane */
1909 /* pick the biggest page */
1910 pgsize_idx = __fls(pgsize);
1911 pgsize = 1UL << pgsize_idx;
1916 int __iommu_map(struct iommu_domain *domain, unsigned long iova,
1917 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
1919 const struct iommu_ops *ops = domain->ops;
1920 unsigned long orig_iova = iova;
1921 unsigned int min_pagesz;
1922 size_t orig_size = size;
1923 phys_addr_t orig_paddr = paddr;
1926 if (unlikely(ops->map == NULL ||
1927 domain->pgsize_bitmap == 0UL))
1930 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1933 /* find out the minimum page size supported */
1934 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1937 * both the virtual address and the physical one, as well as
1938 * the size of the mapping, must be aligned (at least) to the
1939 * size of the smallest page supported by the hardware
1941 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1942 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1943 iova, &paddr, size, min_pagesz);
1947 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1950 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1952 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1953 iova, &paddr, pgsize);
1954 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
1964 if (ops->iotlb_sync_map)
1965 ops->iotlb_sync_map(domain);
1967 /* unroll mapping in case something went wrong */
1969 iommu_unmap(domain, orig_iova, orig_size - size);
1971 trace_map(orig_iova, orig_paddr, orig_size);
1976 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1977 phys_addr_t paddr, size_t size, int prot)
1980 return __iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
1982 EXPORT_SYMBOL_GPL(iommu_map);
1984 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
1985 phys_addr_t paddr, size_t size, int prot)
1987 return __iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
1989 EXPORT_SYMBOL_GPL(iommu_map_atomic);
1991 static size_t __iommu_unmap(struct iommu_domain *domain,
1992 unsigned long iova, size_t size,
1993 struct iommu_iotlb_gather *iotlb_gather)
1995 const struct iommu_ops *ops = domain->ops;
1996 size_t unmapped_page, unmapped = 0;
1997 unsigned long orig_iova = iova;
1998 unsigned int min_pagesz;
2000 if (unlikely(ops->unmap == NULL ||
2001 domain->pgsize_bitmap == 0UL))
2004 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2007 /* find out the minimum page size supported */
2008 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2011 * The virtual address, as well as the size of the mapping, must be
2012 * aligned (at least) to the size of the smallest page supported
2015 if (!IS_ALIGNED(iova | size, min_pagesz)) {
2016 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2017 iova, size, min_pagesz);
2021 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2024 * Keep iterating until we either unmap 'size' bytes (or more)
2025 * or we hit an area that isn't mapped.
2027 while (unmapped < size) {
2028 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
2030 unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
2034 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2035 iova, unmapped_page);
2037 iova += unmapped_page;
2038 unmapped += unmapped_page;
2041 trace_unmap(orig_iova, size, unmapped);
2045 size_t iommu_unmap(struct iommu_domain *domain,
2046 unsigned long iova, size_t size)
2048 struct iommu_iotlb_gather iotlb_gather;
2051 iommu_iotlb_gather_init(&iotlb_gather);
2052 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2053 iommu_tlb_sync(domain, &iotlb_gather);
2057 EXPORT_SYMBOL_GPL(iommu_unmap);
2059 size_t iommu_unmap_fast(struct iommu_domain *domain,
2060 unsigned long iova, size_t size,
2061 struct iommu_iotlb_gather *iotlb_gather)
2063 return __iommu_unmap(domain, iova, size, iotlb_gather);
2065 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2067 size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2068 struct scatterlist *sg, unsigned int nents, int prot,
2071 size_t len = 0, mapped = 0;
2076 while (i <= nents) {
2077 phys_addr_t s_phys = sg_phys(sg);
2079 if (len && s_phys != start + len) {
2080 ret = __iommu_map(domain, iova + mapped, start,
2104 /* undo mappings already done */
2105 iommu_unmap(domain, iova, mapped);
2111 size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2112 struct scatterlist *sg, unsigned int nents, int prot)
2115 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2117 EXPORT_SYMBOL_GPL(iommu_map_sg);
2119 size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2120 struct scatterlist *sg, unsigned int nents, int prot)
2122 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2124 EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2126 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
2127 phys_addr_t paddr, u64 size, int prot)
2129 if (unlikely(domain->ops->domain_window_enable == NULL))
2132 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2135 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2137 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2139 if (unlikely(domain->ops->domain_window_disable == NULL))
2142 return domain->ops->domain_window_disable(domain, wnd_nr);
2144 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2147 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2148 * @domain: the iommu domain where the fault has happened
2149 * @dev: the device where the fault has happened
2150 * @iova: the faulting address
2151 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2153 * This function should be called by the low-level IOMMU implementations
2154 * whenever IOMMU faults happen, to allow high-level users, that are
2155 * interested in such events, to know about them.
2157 * This event may be useful for several possible use cases:
2158 * - mere logging of the event
2159 * - dynamic TLB/PTE loading
2160 * - if restarting of the faulting device is required
2162 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2163 * PTE/TLB loading will one day be supported, implementations will be able
2164 * to tell whether it succeeded or not according to this return value).
2166 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2167 * (though fault handlers can also return -ENOSYS, in case they want to
2168 * elicit the default behavior of the IOMMU drivers).
2170 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2171 unsigned long iova, int flags)
2176 * if upper layers showed interest and installed a fault handler,
2179 if (domain->handler)
2180 ret = domain->handler(domain, dev, iova, flags,
2181 domain->handler_token);
2183 trace_io_page_fault(dev, iova, flags);
2186 EXPORT_SYMBOL_GPL(report_iommu_fault);
2188 static int __init iommu_init(void)
2190 iommu_group_kset = kset_create_and_add("iommu_groups",
2192 BUG_ON(!iommu_group_kset);
2194 iommu_debugfs_setup();
2198 core_initcall(iommu_init);
2200 int iommu_domain_get_attr(struct iommu_domain *domain,
2201 enum iommu_attr attr, void *data)
2203 struct iommu_domain_geometry *geometry;
2208 case DOMAIN_ATTR_GEOMETRY:
2210 *geometry = domain->geometry;
2213 case DOMAIN_ATTR_PAGING:
2215 *paging = (domain->pgsize_bitmap != 0UL);
2218 if (!domain->ops->domain_get_attr)
2221 ret = domain->ops->domain_get_attr(domain, attr, data);
2226 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2228 int iommu_domain_set_attr(struct iommu_domain *domain,
2229 enum iommu_attr attr, void *data)
2235 if (domain->ops->domain_set_attr == NULL)
2238 ret = domain->ops->domain_set_attr(domain, attr, data);
2243 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
2245 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2247 const struct iommu_ops *ops = dev->bus->iommu_ops;
2249 if (ops && ops->get_resv_regions)
2250 ops->get_resv_regions(dev, list);
2253 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2255 const struct iommu_ops *ops = dev->bus->iommu_ops;
2257 if (ops && ops->put_resv_regions)
2258 ops->put_resv_regions(dev, list);
2262 * generic_iommu_put_resv_regions - Reserved region driver helper
2263 * @dev: device for which to free reserved regions
2264 * @list: reserved region list for device
2266 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2267 * for simple reservations. Memory allocated for each reserved region will be
2268 * freed. If an IOMMU driver allocates additional resources per region, it is
2269 * going to have to implement a custom callback.
2271 void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2273 struct iommu_resv_region *entry, *next;
2275 list_for_each_entry_safe(entry, next, list, list)
2278 EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2280 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2281 size_t length, int prot,
2282 enum iommu_resv_type type)
2284 struct iommu_resv_region *region;
2286 region = kzalloc(sizeof(*region), GFP_KERNEL);
2290 INIT_LIST_HEAD(®ion->list);
2291 region->start = start;
2292 region->length = length;
2293 region->prot = prot;
2294 region->type = type;
2297 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2300 request_default_domain_for_dev(struct device *dev, unsigned long type)
2302 struct iommu_domain *domain;
2303 struct iommu_group *group;
2306 /* Device must already be in a group before calling this function */
2307 group = iommu_group_get(dev);
2311 mutex_lock(&group->mutex);
2314 if (group->default_domain && group->default_domain->type == type)
2317 /* Don't change mappings of existing devices */
2319 if (iommu_group_device_count(group) != 1)
2323 domain = __iommu_domain_alloc(dev->bus, type);
2327 /* Attach the device to the domain */
2328 ret = __iommu_attach_group(domain, group);
2330 iommu_domain_free(domain);
2334 /* Make the domain the default for this group */
2335 if (group->default_domain)
2336 iommu_domain_free(group->default_domain);
2337 group->default_domain = domain;
2339 iommu_group_create_direct_mappings(group, dev);
2341 dev_info(dev, "Using iommu %s mapping\n",
2342 type == IOMMU_DOMAIN_DMA ? "dma" : "direct");
2346 mutex_unlock(&group->mutex);
2347 iommu_group_put(group);
2352 /* Request that a device is direct mapped by the IOMMU */
2353 int iommu_request_dm_for_dev(struct device *dev)
2355 return request_default_domain_for_dev(dev, IOMMU_DOMAIN_IDENTITY);
2358 /* Request that a device can't be direct mapped by the IOMMU */
2359 int iommu_request_dma_domain_for_dev(struct device *dev)
2361 return request_default_domain_for_dev(dev, IOMMU_DOMAIN_DMA);
2364 void iommu_set_default_passthrough(bool cmd_line)
2367 iommu_set_cmd_line_dma_api();
2369 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2372 void iommu_set_default_translated(bool cmd_line)
2375 iommu_set_cmd_line_dma_api();
2377 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2380 bool iommu_default_passthrough(void)
2382 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2384 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2386 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2388 const struct iommu_ops *ops = NULL;
2389 struct iommu_device *iommu;
2391 spin_lock(&iommu_device_lock);
2392 list_for_each_entry(iommu, &iommu_device_list, list)
2393 if (iommu->fwnode == fwnode) {
2397 spin_unlock(&iommu_device_lock);
2401 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2402 const struct iommu_ops *ops)
2404 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2407 return ops == fwspec->ops ? 0 : -EINVAL;
2409 if (!dev_iommu_get(dev))
2412 /* Preallocate for the overwhelmingly common case of 1 ID */
2413 fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2417 of_node_get(to_of_node(iommu_fwnode));
2418 fwspec->iommu_fwnode = iommu_fwnode;
2420 dev_iommu_fwspec_set(dev, fwspec);
2423 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2425 void iommu_fwspec_free(struct device *dev)
2427 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2430 fwnode_handle_put(fwspec->iommu_fwnode);
2432 dev_iommu_fwspec_set(dev, NULL);
2435 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2437 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2439 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2445 new_num = fwspec->num_ids + num_ids;
2447 fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2452 dev_iommu_fwspec_set(dev, fwspec);
2455 for (i = 0; i < num_ids; i++)
2456 fwspec->ids[fwspec->num_ids + i] = ids[i];
2458 fwspec->num_ids = new_num;
2461 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2464 * Per device IOMMU features.
2466 bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2468 const struct iommu_ops *ops = dev->bus->iommu_ops;
2470 if (ops && ops->dev_has_feat)
2471 return ops->dev_has_feat(dev, feat);
2475 EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2477 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2479 const struct iommu_ops *ops = dev->bus->iommu_ops;
2481 if (ops && ops->dev_enable_feat)
2482 return ops->dev_enable_feat(dev, feat);
2486 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2489 * The device drivers should do the necessary cleanups before calling this.
2490 * For example, before disabling the aux-domain feature, the device driver
2491 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2493 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2495 const struct iommu_ops *ops = dev->bus->iommu_ops;
2497 if (ops && ops->dev_disable_feat)
2498 return ops->dev_disable_feat(dev, feat);
2502 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2504 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2506 const struct iommu_ops *ops = dev->bus->iommu_ops;
2508 if (ops && ops->dev_feat_enabled)
2509 return ops->dev_feat_enabled(dev, feat);
2513 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2516 * Aux-domain specific attach/detach.
2518 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2519 * true. Also, as long as domains are attached to a device through this
2520 * interface, any tries to call iommu_attach_device() should fail
2521 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2522 * This should make us safe against a device being attached to a guest as a
2523 * whole while there are still pasid users on it (aux and sva).
2525 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2529 if (domain->ops->aux_attach_dev)
2530 ret = domain->ops->aux_attach_dev(domain, dev);
2533 trace_attach_device_to_domain(dev);
2537 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2539 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2541 if (domain->ops->aux_detach_dev) {
2542 domain->ops->aux_detach_dev(domain, dev);
2543 trace_detach_device_from_domain(dev);
2546 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2548 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2552 if (domain->ops->aux_get_pasid)
2553 ret = domain->ops->aux_get_pasid(domain, dev);
2557 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
2560 * iommu_sva_bind_device() - Bind a process address space to a device
2562 * @mm: the mm to bind, caller must hold a reference to it
2564 * Create a bond between device and address space, allowing the device to access
2565 * the mm using the returned PASID. If a bond already exists between @device and
2566 * @mm, it is returned and an additional reference is taken. Caller must call
2567 * iommu_sva_unbind_device() to release each reference.
2569 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2570 * initialize the required SVA features.
2572 * On error, returns an ERR_PTR value.
2575 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2577 struct iommu_group *group;
2578 struct iommu_sva *handle = ERR_PTR(-EINVAL);
2579 const struct iommu_ops *ops = dev->bus->iommu_ops;
2581 if (!ops || !ops->sva_bind)
2582 return ERR_PTR(-ENODEV);
2584 group = iommu_group_get(dev);
2586 return ERR_PTR(-ENODEV);
2588 /* Ensure device count and domain don't change while we're binding */
2589 mutex_lock(&group->mutex);
2592 * To keep things simple, SVA currently doesn't support IOMMU groups
2593 * with more than one device. Existing SVA-capable systems are not
2594 * affected by the problems that required IOMMU groups (lack of ACS
2595 * isolation, device ID aliasing and other hardware issues).
2597 if (iommu_group_device_count(group) != 1)
2600 handle = ops->sva_bind(dev, mm, drvdata);
2603 mutex_unlock(&group->mutex);
2604 iommu_group_put(group);
2608 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
2611 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2612 * @handle: the handle returned by iommu_sva_bind_device()
2614 * Put reference to a bond between device and address space. The device should
2615 * not be issuing any more transaction for this PASID. All outstanding page
2616 * requests for this PASID must have been flushed to the IOMMU.
2618 * Returns 0 on success, or an error value
2620 void iommu_sva_unbind_device(struct iommu_sva *handle)
2622 struct iommu_group *group;
2623 struct device *dev = handle->dev;
2624 const struct iommu_ops *ops = dev->bus->iommu_ops;
2626 if (!ops || !ops->sva_unbind)
2629 group = iommu_group_get(dev);
2633 mutex_lock(&group->mutex);
2634 ops->sva_unbind(handle);
2635 mutex_unlock(&group->mutex);
2637 iommu_group_put(group);
2639 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
2641 int iommu_sva_set_ops(struct iommu_sva *handle,
2642 const struct iommu_sva_ops *sva_ops)
2644 if (handle->ops && handle->ops != sva_ops)
2647 handle->ops = sva_ops;
2650 EXPORT_SYMBOL_GPL(iommu_sva_set_ops);
2652 int iommu_sva_get_pasid(struct iommu_sva *handle)
2654 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
2656 if (!ops || !ops->sva_get_pasid)
2657 return IOMMU_PASID_INVALID;
2659 return ops->sva_get_pasid(handle);
2661 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);