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/dma-iommu.h>
11 #include <linux/kernel.h>
12 #include <linux/bits.h>
13 #include <linux/bug.h>
14 #include <linux/types.h>
15 #include <linux/init.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/errno.h>
19 #include <linux/iommu.h>
20 #include <linux/idr.h>
21 #include <linux/notifier.h>
22 #include <linux/err.h>
23 #include <linux/pci.h>
24 #include <linux/bitops.h>
25 #include <linux/property.h>
26 #include <linux/fsl/mc.h>
27 #include <linux/module.h>
28 #include <linux/cc_platform.h>
29 #include <trace/events/iommu.h>
31 static struct kset *iommu_group_kset;
32 static DEFINE_IDA(iommu_group_ida);
34 static unsigned int iommu_def_domain_type __read_mostly;
35 static bool iommu_dma_strict __read_mostly = IS_ENABLED(CONFIG_IOMMU_DEFAULT_DMA_STRICT);
36 static u32 iommu_cmd_line __read_mostly;
40 struct kobject *devices_kobj;
41 struct list_head devices;
43 struct blocking_notifier_head notifier;
45 void (*iommu_data_release)(void *iommu_data);
48 struct iommu_domain *default_domain;
49 struct iommu_domain *domain;
50 struct list_head entry;
54 struct list_head list;
59 struct iommu_group_attribute {
60 struct attribute attr;
61 ssize_t (*show)(struct iommu_group *group, char *buf);
62 ssize_t (*store)(struct iommu_group *group,
63 const char *buf, size_t count);
66 static const char * const iommu_group_resv_type_string[] = {
67 [IOMMU_RESV_DIRECT] = "direct",
68 [IOMMU_RESV_DIRECT_RELAXABLE] = "direct-relaxable",
69 [IOMMU_RESV_RESERVED] = "reserved",
70 [IOMMU_RESV_MSI] = "msi",
71 [IOMMU_RESV_SW_MSI] = "msi",
74 #define IOMMU_CMD_LINE_DMA_API BIT(0)
75 #define IOMMU_CMD_LINE_STRICT BIT(1)
77 static int iommu_alloc_default_domain(struct iommu_group *group,
79 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
81 static int __iommu_attach_device(struct iommu_domain *domain,
83 static int __iommu_attach_group(struct iommu_domain *domain,
84 struct iommu_group *group);
85 static void __iommu_detach_group(struct iommu_domain *domain,
86 struct iommu_group *group);
87 static int iommu_create_device_direct_mappings(struct iommu_group *group,
89 static struct iommu_group *iommu_group_get_for_dev(struct device *dev);
90 static ssize_t iommu_group_store_type(struct iommu_group *group,
91 const char *buf, size_t count);
93 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
94 struct iommu_group_attribute iommu_group_attr_##_name = \
95 __ATTR(_name, _mode, _show, _store)
97 #define to_iommu_group_attr(_attr) \
98 container_of(_attr, struct iommu_group_attribute, attr)
99 #define to_iommu_group(_kobj) \
100 container_of(_kobj, struct iommu_group, kobj)
102 static LIST_HEAD(iommu_device_list);
103 static DEFINE_SPINLOCK(iommu_device_lock);
106 * Use a function instead of an array here because the domain-type is a
107 * bit-field, so an array would waste memory.
109 static const char *iommu_domain_type_str(unsigned int t)
112 case IOMMU_DOMAIN_BLOCKED:
114 case IOMMU_DOMAIN_IDENTITY:
115 return "Passthrough";
116 case IOMMU_DOMAIN_UNMANAGED:
118 case IOMMU_DOMAIN_DMA:
119 case IOMMU_DOMAIN_DMA_FQ:
126 static int __init iommu_subsys_init(void)
128 if (!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API)) {
129 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
130 iommu_set_default_passthrough(false);
132 iommu_set_default_translated(false);
134 if (iommu_default_passthrough() && cc_platform_has(CC_ATTR_MEM_ENCRYPT)) {
135 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
136 iommu_set_default_translated(false);
140 if (!iommu_default_passthrough() && !iommu_dma_strict)
141 iommu_def_domain_type = IOMMU_DOMAIN_DMA_FQ;
143 pr_info("Default domain type: %s %s\n",
144 iommu_domain_type_str(iommu_def_domain_type),
145 (iommu_cmd_line & IOMMU_CMD_LINE_DMA_API) ?
146 "(set via kernel command line)" : "");
148 if (!iommu_default_passthrough())
149 pr_info("DMA domain TLB invalidation policy: %s mode %s\n",
150 iommu_dma_strict ? "strict" : "lazy",
151 (iommu_cmd_line & IOMMU_CMD_LINE_STRICT) ?
152 "(set via kernel command line)" : "");
156 subsys_initcall(iommu_subsys_init);
159 * iommu_device_register() - Register an IOMMU hardware instance
160 * @iommu: IOMMU handle for the instance
161 * @ops: IOMMU ops to associate with the instance
162 * @hwdev: (optional) actual instance device, used for fwnode lookup
164 * Return: 0 on success, or an error.
166 int iommu_device_register(struct iommu_device *iommu,
167 const struct iommu_ops *ops, struct device *hwdev)
169 /* We need to be able to take module references appropriately */
170 if (WARN_ON(is_module_address((unsigned long)ops) && !ops->owner))
175 iommu->fwnode = hwdev->fwnode;
177 spin_lock(&iommu_device_lock);
178 list_add_tail(&iommu->list, &iommu_device_list);
179 spin_unlock(&iommu_device_lock);
182 EXPORT_SYMBOL_GPL(iommu_device_register);
184 void iommu_device_unregister(struct iommu_device *iommu)
186 spin_lock(&iommu_device_lock);
187 list_del(&iommu->list);
188 spin_unlock(&iommu_device_lock);
190 EXPORT_SYMBOL_GPL(iommu_device_unregister);
192 static struct dev_iommu *dev_iommu_get(struct device *dev)
194 struct dev_iommu *param = dev->iommu;
199 param = kzalloc(sizeof(*param), GFP_KERNEL);
203 mutex_init(¶m->lock);
208 static void dev_iommu_free(struct device *dev)
210 iommu_fwspec_free(dev);
215 static int __iommu_probe_device(struct device *dev, struct list_head *group_list)
217 const struct iommu_ops *ops = dev->bus->iommu_ops;
218 struct iommu_device *iommu_dev;
219 struct iommu_group *group;
225 if (!dev_iommu_get(dev))
228 if (!try_module_get(ops->owner)) {
233 iommu_dev = ops->probe_device(dev);
234 if (IS_ERR(iommu_dev)) {
235 ret = PTR_ERR(iommu_dev);
239 dev->iommu->iommu_dev = iommu_dev;
241 group = iommu_group_get_for_dev(dev);
243 ret = PTR_ERR(group);
246 iommu_group_put(group);
248 if (group_list && !group->default_domain && list_empty(&group->entry))
249 list_add_tail(&group->entry, group_list);
251 iommu_device_link(iommu_dev, dev);
256 ops->release_device(dev);
259 module_put(ops->owner);
267 int iommu_probe_device(struct device *dev)
269 const struct iommu_ops *ops = dev->bus->iommu_ops;
270 struct iommu_group *group;
273 ret = __iommu_probe_device(dev, NULL);
277 group = iommu_group_get(dev);
284 * Try to allocate a default domain - needs support from the
285 * IOMMU driver. There are still some drivers which don't
286 * support default domains, so the return value is not yet
289 mutex_lock(&group->mutex);
290 iommu_alloc_default_domain(group, dev);
292 if (group->default_domain) {
293 ret = __iommu_attach_device(group->default_domain, dev);
295 mutex_unlock(&group->mutex);
296 iommu_group_put(group);
301 iommu_create_device_direct_mappings(group, dev);
303 mutex_unlock(&group->mutex);
304 iommu_group_put(group);
306 if (ops->probe_finalize)
307 ops->probe_finalize(dev);
312 iommu_release_device(dev);
319 void iommu_release_device(struct device *dev)
321 const struct iommu_ops *ops = dev->bus->iommu_ops;
326 iommu_device_unlink(dev->iommu->iommu_dev, dev);
328 ops->release_device(dev);
330 iommu_group_remove_device(dev);
331 module_put(ops->owner);
335 static int __init iommu_set_def_domain_type(char *str)
340 ret = kstrtobool(str, &pt);
345 iommu_set_default_passthrough(true);
347 iommu_set_default_translated(true);
351 early_param("iommu.passthrough", iommu_set_def_domain_type);
353 static int __init iommu_dma_setup(char *str)
355 int ret = kstrtobool(str, &iommu_dma_strict);
358 iommu_cmd_line |= IOMMU_CMD_LINE_STRICT;
361 early_param("iommu.strict", iommu_dma_setup);
363 void iommu_set_dma_strict(void)
365 iommu_dma_strict = true;
366 if (iommu_def_domain_type == IOMMU_DOMAIN_DMA_FQ)
367 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
370 static ssize_t iommu_group_attr_show(struct kobject *kobj,
371 struct attribute *__attr, char *buf)
373 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
374 struct iommu_group *group = to_iommu_group(kobj);
378 ret = attr->show(group, buf);
382 static ssize_t iommu_group_attr_store(struct kobject *kobj,
383 struct attribute *__attr,
384 const char *buf, size_t count)
386 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
387 struct iommu_group *group = to_iommu_group(kobj);
391 ret = attr->store(group, buf, count);
395 static const struct sysfs_ops iommu_group_sysfs_ops = {
396 .show = iommu_group_attr_show,
397 .store = iommu_group_attr_store,
400 static int iommu_group_create_file(struct iommu_group *group,
401 struct iommu_group_attribute *attr)
403 return sysfs_create_file(&group->kobj, &attr->attr);
406 static void iommu_group_remove_file(struct iommu_group *group,
407 struct iommu_group_attribute *attr)
409 sysfs_remove_file(&group->kobj, &attr->attr);
412 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
414 return sprintf(buf, "%s\n", group->name);
418 * iommu_insert_resv_region - Insert a new region in the
419 * list of reserved regions.
420 * @new: new region to insert
421 * @regions: list of regions
423 * Elements are sorted by start address and overlapping segments
424 * of the same type are merged.
426 static int iommu_insert_resv_region(struct iommu_resv_region *new,
427 struct list_head *regions)
429 struct iommu_resv_region *iter, *tmp, *nr, *top;
432 nr = iommu_alloc_resv_region(new->start, new->length,
433 new->prot, new->type);
437 /* First add the new element based on start address sorting */
438 list_for_each_entry(iter, regions, list) {
439 if (nr->start < iter->start ||
440 (nr->start == iter->start && nr->type <= iter->type))
443 list_add_tail(&nr->list, &iter->list);
445 /* Merge overlapping segments of type nr->type in @regions, if any */
446 list_for_each_entry_safe(iter, tmp, regions, list) {
447 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
449 /* no merge needed on elements of different types than @new */
450 if (iter->type != new->type) {
451 list_move_tail(&iter->list, &stack);
455 /* look for the last stack element of same type as @iter */
456 list_for_each_entry_reverse(top, &stack, list)
457 if (top->type == iter->type)
460 list_move_tail(&iter->list, &stack);
464 top_end = top->start + top->length - 1;
466 if (iter->start > top_end + 1) {
467 list_move_tail(&iter->list, &stack);
469 top->length = max(top_end, iter_end) - top->start + 1;
470 list_del(&iter->list);
474 list_splice(&stack, regions);
479 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
480 struct list_head *group_resv_regions)
482 struct iommu_resv_region *entry;
485 list_for_each_entry(entry, dev_resv_regions, list) {
486 ret = iommu_insert_resv_region(entry, group_resv_regions);
493 int iommu_get_group_resv_regions(struct iommu_group *group,
494 struct list_head *head)
496 struct group_device *device;
499 mutex_lock(&group->mutex);
500 list_for_each_entry(device, &group->devices, list) {
501 struct list_head dev_resv_regions;
503 INIT_LIST_HEAD(&dev_resv_regions);
504 iommu_get_resv_regions(device->dev, &dev_resv_regions);
505 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
506 iommu_put_resv_regions(device->dev, &dev_resv_regions);
510 mutex_unlock(&group->mutex);
513 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
515 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
518 struct iommu_resv_region *region, *next;
519 struct list_head group_resv_regions;
522 INIT_LIST_HEAD(&group_resv_regions);
523 iommu_get_group_resv_regions(group, &group_resv_regions);
525 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
526 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
527 (long long int)region->start,
528 (long long int)(region->start +
530 iommu_group_resv_type_string[region->type]);
537 static ssize_t iommu_group_show_type(struct iommu_group *group,
540 char *type = "unknown\n";
542 mutex_lock(&group->mutex);
543 if (group->default_domain) {
544 switch (group->default_domain->type) {
545 case IOMMU_DOMAIN_BLOCKED:
548 case IOMMU_DOMAIN_IDENTITY:
551 case IOMMU_DOMAIN_UNMANAGED:
552 type = "unmanaged\n";
554 case IOMMU_DOMAIN_DMA:
557 case IOMMU_DOMAIN_DMA_FQ:
562 mutex_unlock(&group->mutex);
568 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
570 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
571 iommu_group_show_resv_regions, NULL);
573 static IOMMU_GROUP_ATTR(type, 0644, iommu_group_show_type,
574 iommu_group_store_type);
576 static void iommu_group_release(struct kobject *kobj)
578 struct iommu_group *group = to_iommu_group(kobj);
580 pr_debug("Releasing group %d\n", group->id);
582 if (group->iommu_data_release)
583 group->iommu_data_release(group->iommu_data);
585 ida_simple_remove(&iommu_group_ida, group->id);
587 if (group->default_domain)
588 iommu_domain_free(group->default_domain);
594 static struct kobj_type iommu_group_ktype = {
595 .sysfs_ops = &iommu_group_sysfs_ops,
596 .release = iommu_group_release,
600 * iommu_group_alloc - Allocate a new group
602 * This function is called by an iommu driver to allocate a new iommu
603 * group. The iommu group represents the minimum granularity of the iommu.
604 * Upon successful return, the caller holds a reference to the supplied
605 * group in order to hold the group until devices are added. Use
606 * iommu_group_put() to release this extra reference count, allowing the
607 * group to be automatically reclaimed once it has no devices or external
610 struct iommu_group *iommu_group_alloc(void)
612 struct iommu_group *group;
615 group = kzalloc(sizeof(*group), GFP_KERNEL);
617 return ERR_PTR(-ENOMEM);
619 group->kobj.kset = iommu_group_kset;
620 mutex_init(&group->mutex);
621 INIT_LIST_HEAD(&group->devices);
622 INIT_LIST_HEAD(&group->entry);
623 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
625 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
632 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
633 NULL, "%d", group->id);
635 ida_simple_remove(&iommu_group_ida, group->id);
636 kobject_put(&group->kobj);
640 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
641 if (!group->devices_kobj) {
642 kobject_put(&group->kobj); /* triggers .release & free */
643 return ERR_PTR(-ENOMEM);
647 * The devices_kobj holds a reference on the group kobject, so
648 * as long as that exists so will the group. We can therefore
649 * use the devices_kobj for reference counting.
651 kobject_put(&group->kobj);
653 ret = iommu_group_create_file(group,
654 &iommu_group_attr_reserved_regions);
658 ret = iommu_group_create_file(group, &iommu_group_attr_type);
662 pr_debug("Allocated group %d\n", group->id);
666 EXPORT_SYMBOL_GPL(iommu_group_alloc);
668 struct iommu_group *iommu_group_get_by_id(int id)
670 struct kobject *group_kobj;
671 struct iommu_group *group;
674 if (!iommu_group_kset)
677 name = kasprintf(GFP_KERNEL, "%d", id);
681 group_kobj = kset_find_obj(iommu_group_kset, name);
687 group = container_of(group_kobj, struct iommu_group, kobj);
688 BUG_ON(group->id != id);
690 kobject_get(group->devices_kobj);
691 kobject_put(&group->kobj);
695 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
698 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
701 * iommu drivers can store data in the group for use when doing iommu
702 * operations. This function provides a way to retrieve it. Caller
703 * should hold a group reference.
705 void *iommu_group_get_iommudata(struct iommu_group *group)
707 return group->iommu_data;
709 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
712 * iommu_group_set_iommudata - set iommu_data for a group
714 * @iommu_data: new data
715 * @release: release function for iommu_data
717 * iommu drivers can store data in the group for use when doing iommu
718 * operations. This function provides a way to set the data after
719 * the group has been allocated. Caller should hold a group reference.
721 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
722 void (*release)(void *iommu_data))
724 group->iommu_data = iommu_data;
725 group->iommu_data_release = release;
727 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
730 * iommu_group_set_name - set name for a group
734 * Allow iommu driver to set a name for a group. When set it will
735 * appear in a name attribute file under the group in sysfs.
737 int iommu_group_set_name(struct iommu_group *group, const char *name)
742 iommu_group_remove_file(group, &iommu_group_attr_name);
749 group->name = kstrdup(name, GFP_KERNEL);
753 ret = iommu_group_create_file(group, &iommu_group_attr_name);
762 EXPORT_SYMBOL_GPL(iommu_group_set_name);
764 static int iommu_create_device_direct_mappings(struct iommu_group *group,
767 struct iommu_domain *domain = group->default_domain;
768 struct iommu_resv_region *entry;
769 struct list_head mappings;
770 unsigned long pg_size;
773 if (!domain || !iommu_is_dma_domain(domain))
776 BUG_ON(!domain->pgsize_bitmap);
778 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
779 INIT_LIST_HEAD(&mappings);
781 iommu_get_resv_regions(dev, &mappings);
783 /* We need to consider overlapping regions for different devices */
784 list_for_each_entry(entry, &mappings, list) {
785 dma_addr_t start, end, addr;
788 if (domain->ops->apply_resv_region)
789 domain->ops->apply_resv_region(dev, domain, entry);
791 start = ALIGN(entry->start, pg_size);
792 end = ALIGN(entry->start + entry->length, pg_size);
794 if (entry->type != IOMMU_RESV_DIRECT &&
795 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
798 for (addr = start; addr <= end; addr += pg_size) {
799 phys_addr_t phys_addr;
804 phys_addr = iommu_iova_to_phys(domain, addr);
812 ret = iommu_map(domain, addr - map_size,
813 addr - map_size, map_size,
823 iommu_flush_iotlb_all(domain);
826 iommu_put_resv_regions(dev, &mappings);
831 static bool iommu_is_attach_deferred(struct iommu_domain *domain,
834 if (domain->ops->is_attach_deferred)
835 return domain->ops->is_attach_deferred(domain, dev);
841 * iommu_group_add_device - add a device to an iommu group
842 * @group: the group into which to add the device (reference should be held)
845 * This function is called by an iommu driver to add a device into a
846 * group. Adding a device increments the group reference count.
848 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
851 struct group_device *device;
853 device = kzalloc(sizeof(*device), GFP_KERNEL);
859 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
861 goto err_free_device;
863 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
867 goto err_remove_link;
870 ret = sysfs_create_link_nowarn(group->devices_kobj,
871 &dev->kobj, device->name);
873 if (ret == -EEXIST && i >= 0) {
875 * Account for the slim chance of collision
876 * and append an instance to the name.
879 device->name = kasprintf(GFP_KERNEL, "%s.%d",
880 kobject_name(&dev->kobj), i++);
886 kobject_get(group->devices_kobj);
888 dev->iommu_group = group;
890 mutex_lock(&group->mutex);
891 list_add_tail(&device->list, &group->devices);
892 if (group->domain && !iommu_is_attach_deferred(group->domain, dev))
893 ret = __iommu_attach_device(group->domain, dev);
894 mutex_unlock(&group->mutex);
898 /* Notify any listeners about change to group. */
899 blocking_notifier_call_chain(&group->notifier,
900 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
902 trace_add_device_to_group(group->id, dev);
904 dev_info(dev, "Adding to iommu group %d\n", group->id);
909 mutex_lock(&group->mutex);
910 list_del(&device->list);
911 mutex_unlock(&group->mutex);
912 dev->iommu_group = NULL;
913 kobject_put(group->devices_kobj);
914 sysfs_remove_link(group->devices_kobj, device->name);
918 sysfs_remove_link(&dev->kobj, "iommu_group");
921 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
924 EXPORT_SYMBOL_GPL(iommu_group_add_device);
927 * iommu_group_remove_device - remove a device from it's current group
928 * @dev: device to be removed
930 * This function is called by an iommu driver to remove the device from
931 * it's current group. This decrements the iommu group reference count.
933 void iommu_group_remove_device(struct device *dev)
935 struct iommu_group *group = dev->iommu_group;
936 struct group_device *tmp_device, *device = NULL;
941 dev_info(dev, "Removing from iommu group %d\n", group->id);
943 /* Pre-notify listeners that a device is being removed. */
944 blocking_notifier_call_chain(&group->notifier,
945 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
947 mutex_lock(&group->mutex);
948 list_for_each_entry(tmp_device, &group->devices, list) {
949 if (tmp_device->dev == dev) {
951 list_del(&device->list);
955 mutex_unlock(&group->mutex);
960 sysfs_remove_link(group->devices_kobj, device->name);
961 sysfs_remove_link(&dev->kobj, "iommu_group");
963 trace_remove_device_from_group(group->id, dev);
967 dev->iommu_group = NULL;
968 kobject_put(group->devices_kobj);
970 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
972 static int iommu_group_device_count(struct iommu_group *group)
974 struct group_device *entry;
977 list_for_each_entry(entry, &group->devices, list)
984 * iommu_group_for_each_dev - iterate over each device in the group
986 * @data: caller opaque data to be passed to callback function
987 * @fn: caller supplied callback function
989 * This function is called by group users to iterate over group devices.
990 * Callers should hold a reference count to the group during callback.
991 * The group->mutex is held across callbacks, which will block calls to
992 * iommu_group_add/remove_device.
994 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
995 int (*fn)(struct device *, void *))
997 struct group_device *device;
1000 list_for_each_entry(device, &group->devices, list) {
1001 ret = fn(device->dev, data);
1009 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
1010 int (*fn)(struct device *, void *))
1014 mutex_lock(&group->mutex);
1015 ret = __iommu_group_for_each_dev(group, data, fn);
1016 mutex_unlock(&group->mutex);
1020 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
1023 * iommu_group_get - Return the group for a device and increment reference
1024 * @dev: get the group that this device belongs to
1026 * This function is called by iommu drivers and users to get the group
1027 * for the specified device. If found, the group is returned and the group
1028 * reference in incremented, else NULL.
1030 struct iommu_group *iommu_group_get(struct device *dev)
1032 struct iommu_group *group = dev->iommu_group;
1035 kobject_get(group->devices_kobj);
1039 EXPORT_SYMBOL_GPL(iommu_group_get);
1042 * iommu_group_ref_get - Increment reference on a group
1043 * @group: the group to use, must not be NULL
1045 * This function is called by iommu drivers to take additional references on an
1046 * existing group. Returns the given group for convenience.
1048 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
1050 kobject_get(group->devices_kobj);
1053 EXPORT_SYMBOL_GPL(iommu_group_ref_get);
1056 * iommu_group_put - Decrement group reference
1057 * @group: the group to use
1059 * This function is called by iommu drivers and users to release the
1060 * iommu group. Once the reference count is zero, the group is released.
1062 void iommu_group_put(struct iommu_group *group)
1065 kobject_put(group->devices_kobj);
1067 EXPORT_SYMBOL_GPL(iommu_group_put);
1070 * iommu_group_register_notifier - Register a notifier for group changes
1071 * @group: the group to watch
1072 * @nb: notifier block to signal
1074 * This function allows iommu group users to track changes in a group.
1075 * See include/linux/iommu.h for actions sent via this notifier. Caller
1076 * should hold a reference to the group throughout notifier registration.
1078 int iommu_group_register_notifier(struct iommu_group *group,
1079 struct notifier_block *nb)
1081 return blocking_notifier_chain_register(&group->notifier, nb);
1083 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
1086 * iommu_group_unregister_notifier - Unregister a notifier
1087 * @group: the group to watch
1088 * @nb: notifier block to signal
1090 * Unregister a previously registered group notifier block.
1092 int iommu_group_unregister_notifier(struct iommu_group *group,
1093 struct notifier_block *nb)
1095 return blocking_notifier_chain_unregister(&group->notifier, nb);
1097 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
1100 * iommu_register_device_fault_handler() - Register a device fault handler
1102 * @handler: the fault handler
1103 * @data: private data passed as argument to the handler
1105 * When an IOMMU fault event is received, this handler gets called with the
1106 * fault event and data as argument. The handler should return 0 on success. If
1107 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1108 * complete the fault by calling iommu_page_response() with one of the following
1110 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1111 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1112 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1113 * page faults if possible.
1115 * Return 0 if the fault handler was installed successfully, or an error.
1117 int iommu_register_device_fault_handler(struct device *dev,
1118 iommu_dev_fault_handler_t handler,
1121 struct dev_iommu *param = dev->iommu;
1127 mutex_lock(¶m->lock);
1128 /* Only allow one fault handler registered for each device */
1129 if (param->fault_param) {
1135 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
1136 if (!param->fault_param) {
1141 param->fault_param->handler = handler;
1142 param->fault_param->data = data;
1143 mutex_init(¶m->fault_param->lock);
1144 INIT_LIST_HEAD(¶m->fault_param->faults);
1147 mutex_unlock(¶m->lock);
1151 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1154 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1157 * Remove the device fault handler installed with
1158 * iommu_register_device_fault_handler().
1160 * Return 0 on success, or an error.
1162 int iommu_unregister_device_fault_handler(struct device *dev)
1164 struct dev_iommu *param = dev->iommu;
1170 mutex_lock(¶m->lock);
1172 if (!param->fault_param)
1175 /* we cannot unregister handler if there are pending faults */
1176 if (!list_empty(¶m->fault_param->faults)) {
1181 kfree(param->fault_param);
1182 param->fault_param = NULL;
1185 mutex_unlock(¶m->lock);
1189 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1192 * iommu_report_device_fault() - Report fault event to device driver
1194 * @evt: fault event data
1196 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1197 * handler. When this function fails and the fault is recoverable, it is the
1198 * caller's responsibility to complete the fault.
1200 * Return 0 on success, or an error.
1202 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1204 struct dev_iommu *param = dev->iommu;
1205 struct iommu_fault_event *evt_pending = NULL;
1206 struct iommu_fault_param *fparam;
1212 /* we only report device fault if there is a handler registered */
1213 mutex_lock(¶m->lock);
1214 fparam = param->fault_param;
1215 if (!fparam || !fparam->handler) {
1220 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1221 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1222 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1228 mutex_lock(&fparam->lock);
1229 list_add_tail(&evt_pending->list, &fparam->faults);
1230 mutex_unlock(&fparam->lock);
1233 ret = fparam->handler(&evt->fault, fparam->data);
1234 if (ret && evt_pending) {
1235 mutex_lock(&fparam->lock);
1236 list_del(&evt_pending->list);
1237 mutex_unlock(&fparam->lock);
1241 mutex_unlock(¶m->lock);
1244 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1246 int iommu_page_response(struct device *dev,
1247 struct iommu_page_response *msg)
1251 struct iommu_fault_event *evt;
1252 struct iommu_fault_page_request *prm;
1253 struct dev_iommu *param = dev->iommu;
1254 bool has_pasid = msg->flags & IOMMU_PAGE_RESP_PASID_VALID;
1255 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1257 if (!domain || !domain->ops->page_response)
1260 if (!param || !param->fault_param)
1263 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1264 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1267 /* Only send response if there is a fault report pending */
1268 mutex_lock(¶m->fault_param->lock);
1269 if (list_empty(¶m->fault_param->faults)) {
1270 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1274 * Check if we have a matching page request pending to respond,
1275 * otherwise return -EINVAL
1277 list_for_each_entry(evt, ¶m->fault_param->faults, list) {
1278 prm = &evt->fault.prm;
1279 if (prm->grpid != msg->grpid)
1283 * If the PASID is required, the corresponding request is
1284 * matched using the group ID, the PASID valid bit and the PASID
1285 * value. Otherwise only the group ID matches request and
1288 needs_pasid = prm->flags & IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID;
1289 if (needs_pasid && (!has_pasid || msg->pasid != prm->pasid))
1292 if (!needs_pasid && has_pasid) {
1293 /* No big deal, just clear it. */
1294 msg->flags &= ~IOMMU_PAGE_RESP_PASID_VALID;
1298 ret = domain->ops->page_response(dev, evt, msg);
1299 list_del(&evt->list);
1305 mutex_unlock(¶m->fault_param->lock);
1308 EXPORT_SYMBOL_GPL(iommu_page_response);
1311 * iommu_group_id - Return ID for a group
1312 * @group: the group to ID
1314 * Return the unique ID for the group matching the sysfs group number.
1316 int iommu_group_id(struct iommu_group *group)
1320 EXPORT_SYMBOL_GPL(iommu_group_id);
1322 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1323 unsigned long *devfns);
1326 * To consider a PCI device isolated, we require ACS to support Source
1327 * Validation, Request Redirection, Completer Redirection, and Upstream
1328 * Forwarding. This effectively means that devices cannot spoof their
1329 * requester ID, requests and completions cannot be redirected, and all
1330 * transactions are forwarded upstream, even as it passes through a
1331 * bridge where the target device is downstream.
1333 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1336 * For multifunction devices which are not isolated from each other, find
1337 * all the other non-isolated functions and look for existing groups. For
1338 * each function, we also need to look for aliases to or from other devices
1339 * that may already have a group.
1341 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1342 unsigned long *devfns)
1344 struct pci_dev *tmp = NULL;
1345 struct iommu_group *group;
1347 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1350 for_each_pci_dev(tmp) {
1351 if (tmp == pdev || tmp->bus != pdev->bus ||
1352 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1353 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1356 group = get_pci_alias_group(tmp, devfns);
1367 * Look for aliases to or from the given device for existing groups. DMA
1368 * aliases are only supported on the same bus, therefore the search
1369 * space is quite small (especially since we're really only looking at pcie
1370 * device, and therefore only expect multiple slots on the root complex or
1371 * downstream switch ports). It's conceivable though that a pair of
1372 * multifunction devices could have aliases between them that would cause a
1373 * loop. To prevent this, we use a bitmap to track where we've been.
1375 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1376 unsigned long *devfns)
1378 struct pci_dev *tmp = NULL;
1379 struct iommu_group *group;
1381 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1384 group = iommu_group_get(&pdev->dev);
1388 for_each_pci_dev(tmp) {
1389 if (tmp == pdev || tmp->bus != pdev->bus)
1392 /* We alias them or they alias us */
1393 if (pci_devs_are_dma_aliases(pdev, tmp)) {
1394 group = get_pci_alias_group(tmp, devfns);
1400 group = get_pci_function_alias_group(tmp, devfns);
1411 struct group_for_pci_data {
1412 struct pci_dev *pdev;
1413 struct iommu_group *group;
1417 * DMA alias iterator callback, return the last seen device. Stop and return
1418 * the IOMMU group if we find one along the way.
1420 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1422 struct group_for_pci_data *data = opaque;
1425 data->group = iommu_group_get(&pdev->dev);
1427 return data->group != NULL;
1431 * Generic device_group call-back function. It just allocates one
1432 * iommu-group per device.
1434 struct iommu_group *generic_device_group(struct device *dev)
1436 return iommu_group_alloc();
1438 EXPORT_SYMBOL_GPL(generic_device_group);
1441 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1442 * to find or create an IOMMU group for a device.
1444 struct iommu_group *pci_device_group(struct device *dev)
1446 struct pci_dev *pdev = to_pci_dev(dev);
1447 struct group_for_pci_data data;
1448 struct pci_bus *bus;
1449 struct iommu_group *group = NULL;
1450 u64 devfns[4] = { 0 };
1452 if (WARN_ON(!dev_is_pci(dev)))
1453 return ERR_PTR(-EINVAL);
1456 * Find the upstream DMA alias for the device. A device must not
1457 * be aliased due to topology in order to have its own IOMMU group.
1458 * If we find an alias along the way that already belongs to a
1461 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1467 * Continue upstream from the point of minimum IOMMU granularity
1468 * due to aliases to the point where devices are protected from
1469 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1472 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1476 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1481 group = iommu_group_get(&pdev->dev);
1487 * Look for existing groups on device aliases. If we alias another
1488 * device or another device aliases us, use the same group.
1490 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1495 * Look for existing groups on non-isolated functions on the same
1496 * slot and aliases of those funcions, if any. No need to clear
1497 * the search bitmap, the tested devfns are still valid.
1499 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1503 /* No shared group found, allocate new */
1504 return iommu_group_alloc();
1506 EXPORT_SYMBOL_GPL(pci_device_group);
1508 /* Get the IOMMU group for device on fsl-mc bus */
1509 struct iommu_group *fsl_mc_device_group(struct device *dev)
1511 struct device *cont_dev = fsl_mc_cont_dev(dev);
1512 struct iommu_group *group;
1514 group = iommu_group_get(cont_dev);
1516 group = iommu_group_alloc();
1519 EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1521 static int iommu_get_def_domain_type(struct device *dev)
1523 const struct iommu_ops *ops = dev->bus->iommu_ops;
1525 if (dev_is_pci(dev) && to_pci_dev(dev)->untrusted)
1526 return IOMMU_DOMAIN_DMA;
1528 if (ops->def_domain_type)
1529 return ops->def_domain_type(dev);
1534 static int iommu_group_alloc_default_domain(struct bus_type *bus,
1535 struct iommu_group *group,
1538 struct iommu_domain *dom;
1540 dom = __iommu_domain_alloc(bus, type);
1541 if (!dom && type != IOMMU_DOMAIN_DMA) {
1542 dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
1544 pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1551 group->default_domain = dom;
1553 group->domain = dom;
1557 static int iommu_alloc_default_domain(struct iommu_group *group,
1562 if (group->default_domain)
1565 type = iommu_get_def_domain_type(dev) ? : iommu_def_domain_type;
1567 return iommu_group_alloc_default_domain(dev->bus, group, type);
1571 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1572 * @dev: target device
1574 * This function is intended to be called by IOMMU drivers and extended to
1575 * support common, bus-defined algorithms when determining or creating the
1576 * IOMMU group for a device. On success, the caller will hold a reference
1577 * to the returned IOMMU group, which will already include the provided
1578 * device. The reference should be released with iommu_group_put().
1580 static struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1582 const struct iommu_ops *ops = dev->bus->iommu_ops;
1583 struct iommu_group *group;
1586 group = iommu_group_get(dev);
1591 return ERR_PTR(-EINVAL);
1593 group = ops->device_group(dev);
1594 if (WARN_ON_ONCE(group == NULL))
1595 return ERR_PTR(-EINVAL);
1600 ret = iommu_group_add_device(group, dev);
1607 iommu_group_put(group);
1609 return ERR_PTR(ret);
1612 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1614 return group->default_domain;
1617 static int probe_iommu_group(struct device *dev, void *data)
1619 struct list_head *group_list = data;
1620 struct iommu_group *group;
1623 /* Device is probed already if in a group */
1624 group = iommu_group_get(dev);
1626 iommu_group_put(group);
1630 ret = __iommu_probe_device(dev, group_list);
1637 static int remove_iommu_group(struct device *dev, void *data)
1639 iommu_release_device(dev);
1644 static int iommu_bus_notifier(struct notifier_block *nb,
1645 unsigned long action, void *data)
1647 unsigned long group_action = 0;
1648 struct device *dev = data;
1649 struct iommu_group *group;
1652 * ADD/DEL call into iommu driver ops if provided, which may
1653 * result in ADD/DEL notifiers to group->notifier
1655 if (action == BUS_NOTIFY_ADD_DEVICE) {
1658 ret = iommu_probe_device(dev);
1659 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1660 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1661 iommu_release_device(dev);
1666 * Remaining BUS_NOTIFYs get filtered and republished to the
1667 * group, if anyone is listening
1669 group = iommu_group_get(dev);
1674 case BUS_NOTIFY_BIND_DRIVER:
1675 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1677 case BUS_NOTIFY_BOUND_DRIVER:
1678 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1680 case BUS_NOTIFY_UNBIND_DRIVER:
1681 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1683 case BUS_NOTIFY_UNBOUND_DRIVER:
1684 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1689 blocking_notifier_call_chain(&group->notifier,
1692 iommu_group_put(group);
1696 struct __group_domain_type {
1701 static int probe_get_default_domain_type(struct device *dev, void *data)
1703 struct __group_domain_type *gtype = data;
1704 unsigned int type = iommu_get_def_domain_type(dev);
1707 if (gtype->type && gtype->type != type) {
1708 dev_warn(dev, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1709 iommu_domain_type_str(type),
1710 dev_name(gtype->dev),
1711 iommu_domain_type_str(gtype->type));
1724 static void probe_alloc_default_domain(struct bus_type *bus,
1725 struct iommu_group *group)
1727 struct __group_domain_type gtype;
1729 memset(>ype, 0, sizeof(gtype));
1731 /* Ask for default domain requirements of all devices in the group */
1732 __iommu_group_for_each_dev(group, >ype,
1733 probe_get_default_domain_type);
1736 gtype.type = iommu_def_domain_type;
1738 iommu_group_alloc_default_domain(bus, group, gtype.type);
1742 static int iommu_group_do_dma_attach(struct device *dev, void *data)
1744 struct iommu_domain *domain = data;
1747 if (!iommu_is_attach_deferred(domain, dev))
1748 ret = __iommu_attach_device(domain, dev);
1753 static int __iommu_group_dma_attach(struct iommu_group *group)
1755 return __iommu_group_for_each_dev(group, group->default_domain,
1756 iommu_group_do_dma_attach);
1759 static int iommu_group_do_probe_finalize(struct device *dev, void *data)
1761 struct iommu_domain *domain = data;
1763 if (domain->ops->probe_finalize)
1764 domain->ops->probe_finalize(dev);
1769 static void __iommu_group_dma_finalize(struct iommu_group *group)
1771 __iommu_group_for_each_dev(group, group->default_domain,
1772 iommu_group_do_probe_finalize);
1775 static int iommu_do_create_direct_mappings(struct device *dev, void *data)
1777 struct iommu_group *group = data;
1779 iommu_create_device_direct_mappings(group, dev);
1784 static int iommu_group_create_direct_mappings(struct iommu_group *group)
1786 return __iommu_group_for_each_dev(group, group,
1787 iommu_do_create_direct_mappings);
1790 int bus_iommu_probe(struct bus_type *bus)
1792 struct iommu_group *group, *next;
1793 LIST_HEAD(group_list);
1797 * This code-path does not allocate the default domain when
1798 * creating the iommu group, so do it after the groups are
1801 ret = bus_for_each_dev(bus, NULL, &group_list, probe_iommu_group);
1805 list_for_each_entry_safe(group, next, &group_list, entry) {
1806 /* Remove item from the list */
1807 list_del_init(&group->entry);
1809 mutex_lock(&group->mutex);
1811 /* Try to allocate default domain */
1812 probe_alloc_default_domain(bus, group);
1814 if (!group->default_domain) {
1815 mutex_unlock(&group->mutex);
1819 iommu_group_create_direct_mappings(group);
1821 ret = __iommu_group_dma_attach(group);
1823 mutex_unlock(&group->mutex);
1828 __iommu_group_dma_finalize(group);
1834 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1836 struct notifier_block *nb;
1839 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1843 nb->notifier_call = iommu_bus_notifier;
1845 err = bus_register_notifier(bus, nb);
1849 err = bus_iommu_probe(bus);
1858 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1859 bus_unregister_notifier(bus, nb);
1868 * bus_set_iommu - set iommu-callbacks for the bus
1870 * @ops: the callbacks provided by the iommu-driver
1872 * This function is called by an iommu driver to set the iommu methods
1873 * used for a particular bus. Drivers for devices on that bus can use
1874 * the iommu-api after these ops are registered.
1875 * This special function is needed because IOMMUs are usually devices on
1876 * the bus itself, so the iommu drivers are not initialized when the bus
1877 * is set up. With this function the iommu-driver can set the iommu-ops
1880 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1885 bus->iommu_ops = NULL;
1889 if (bus->iommu_ops != NULL)
1892 bus->iommu_ops = ops;
1894 /* Do IOMMU specific setup for this bus-type */
1895 err = iommu_bus_init(bus, ops);
1897 bus->iommu_ops = NULL;
1901 EXPORT_SYMBOL_GPL(bus_set_iommu);
1903 bool iommu_present(struct bus_type *bus)
1905 return bus->iommu_ops != NULL;
1907 EXPORT_SYMBOL_GPL(iommu_present);
1909 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1911 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1914 return bus->iommu_ops->capable(cap);
1916 EXPORT_SYMBOL_GPL(iommu_capable);
1919 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1920 * @domain: iommu domain
1921 * @handler: fault handler
1922 * @token: user data, will be passed back to the fault handler
1924 * This function should be used by IOMMU users which want to be notified
1925 * whenever an IOMMU fault happens.
1927 * The fault handler itself should return 0 on success, and an appropriate
1928 * error code otherwise.
1930 void iommu_set_fault_handler(struct iommu_domain *domain,
1931 iommu_fault_handler_t handler,
1936 domain->handler = handler;
1937 domain->handler_token = token;
1939 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1941 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1944 struct iommu_domain *domain;
1946 if (bus == NULL || bus->iommu_ops == NULL)
1949 domain = bus->iommu_ops->domain_alloc(type);
1953 domain->ops = bus->iommu_ops;
1954 domain->type = type;
1955 /* Assume all sizes by default; the driver may override this later */
1956 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1958 if (iommu_is_dma_domain(domain) && iommu_get_dma_cookie(domain)) {
1959 iommu_domain_free(domain);
1965 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1967 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1969 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1971 void iommu_domain_free(struct iommu_domain *domain)
1973 iommu_put_dma_cookie(domain);
1974 domain->ops->domain_free(domain);
1976 EXPORT_SYMBOL_GPL(iommu_domain_free);
1978 static int __iommu_attach_device(struct iommu_domain *domain,
1983 if (unlikely(domain->ops->attach_dev == NULL))
1986 ret = domain->ops->attach_dev(domain, dev);
1988 trace_attach_device_to_domain(dev);
1992 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1994 struct iommu_group *group;
1997 group = iommu_group_get(dev);
2002 * Lock the group to make sure the device-count doesn't
2003 * change while we are attaching
2005 mutex_lock(&group->mutex);
2007 if (iommu_group_device_count(group) != 1)
2010 ret = __iommu_attach_group(domain, group);
2013 mutex_unlock(&group->mutex);
2014 iommu_group_put(group);
2018 EXPORT_SYMBOL_GPL(iommu_attach_device);
2020 int iommu_deferred_attach(struct device *dev, struct iommu_domain *domain)
2022 const struct iommu_ops *ops = domain->ops;
2024 if (ops->is_attach_deferred && ops->is_attach_deferred(domain, dev))
2025 return __iommu_attach_device(domain, dev);
2031 * Check flags and other user provided data for valid combinations. We also
2032 * make sure no reserved fields or unused flags are set. This is to ensure
2033 * not breaking userspace in the future when these fields or flags are used.
2035 static int iommu_check_cache_invl_data(struct iommu_cache_invalidate_info *info)
2040 if (info->version != IOMMU_CACHE_INVALIDATE_INFO_VERSION_1)
2043 mask = (1 << IOMMU_CACHE_INV_TYPE_NR) - 1;
2044 if (info->cache & ~mask)
2047 if (info->granularity >= IOMMU_INV_GRANU_NR)
2050 switch (info->granularity) {
2051 case IOMMU_INV_GRANU_ADDR:
2052 if (info->cache & IOMMU_CACHE_INV_TYPE_PASID)
2055 mask = IOMMU_INV_ADDR_FLAGS_PASID |
2056 IOMMU_INV_ADDR_FLAGS_ARCHID |
2057 IOMMU_INV_ADDR_FLAGS_LEAF;
2059 if (info->granu.addr_info.flags & ~mask)
2062 case IOMMU_INV_GRANU_PASID:
2063 mask = IOMMU_INV_PASID_FLAGS_PASID |
2064 IOMMU_INV_PASID_FLAGS_ARCHID;
2065 if (info->granu.pasid_info.flags & ~mask)
2069 case IOMMU_INV_GRANU_DOMAIN:
2070 if (info->cache & IOMMU_CACHE_INV_TYPE_DEV_IOTLB)
2077 /* Check reserved padding fields */
2078 for (i = 0; i < sizeof(info->padding); i++) {
2079 if (info->padding[i])
2086 int iommu_uapi_cache_invalidate(struct iommu_domain *domain, struct device *dev,
2089 struct iommu_cache_invalidate_info inv_info = { 0 };
2093 if (unlikely(!domain->ops->cache_invalidate))
2097 * No new spaces can be added before the variable sized union, the
2098 * minimum size is the offset to the union.
2100 minsz = offsetof(struct iommu_cache_invalidate_info, granu);
2102 /* Copy minsz from user to get flags and argsz */
2103 if (copy_from_user(&inv_info, uinfo, minsz))
2106 /* Fields before the variable size union are mandatory */
2107 if (inv_info.argsz < minsz)
2110 /* PASID and address granu require additional info beyond minsz */
2111 if (inv_info.granularity == IOMMU_INV_GRANU_PASID &&
2112 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.pasid_info))
2115 if (inv_info.granularity == IOMMU_INV_GRANU_ADDR &&
2116 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.addr_info))
2120 * User might be using a newer UAPI header which has a larger data
2121 * size, we shall support the existing flags within the current
2122 * size. Copy the remaining user data _after_ minsz but not more
2123 * than the current kernel supported size.
2125 if (copy_from_user((void *)&inv_info + minsz, uinfo + minsz,
2126 min_t(u32, inv_info.argsz, sizeof(inv_info)) - minsz))
2129 /* Now the argsz is validated, check the content */
2130 ret = iommu_check_cache_invl_data(&inv_info);
2134 return domain->ops->cache_invalidate(domain, dev, &inv_info);
2136 EXPORT_SYMBOL_GPL(iommu_uapi_cache_invalidate);
2138 static int iommu_check_bind_data(struct iommu_gpasid_bind_data *data)
2143 if (data->version != IOMMU_GPASID_BIND_VERSION_1)
2146 /* Check the range of supported formats */
2147 if (data->format >= IOMMU_PASID_FORMAT_LAST)
2150 /* Check all flags */
2151 mask = IOMMU_SVA_GPASID_VAL;
2152 if (data->flags & ~mask)
2155 /* Check reserved padding fields */
2156 for (i = 0; i < sizeof(data->padding); i++) {
2157 if (data->padding[i])
2164 static int iommu_sva_prepare_bind_data(void __user *udata,
2165 struct iommu_gpasid_bind_data *data)
2170 * No new spaces can be added before the variable sized union, the
2171 * minimum size is the offset to the union.
2173 minsz = offsetof(struct iommu_gpasid_bind_data, vendor);
2175 /* Copy minsz from user to get flags and argsz */
2176 if (copy_from_user(data, udata, minsz))
2179 /* Fields before the variable size union are mandatory */
2180 if (data->argsz < minsz)
2183 * User might be using a newer UAPI header, we shall let IOMMU vendor
2184 * driver decide on what size it needs. Since the guest PASID bind data
2185 * can be vendor specific, larger argsz could be the result of extension
2186 * for one vendor but it should not affect another vendor.
2187 * Copy the remaining user data _after_ minsz
2189 if (copy_from_user((void *)data + minsz, udata + minsz,
2190 min_t(u32, data->argsz, sizeof(*data)) - minsz))
2193 return iommu_check_bind_data(data);
2196 int iommu_uapi_sva_bind_gpasid(struct iommu_domain *domain, struct device *dev,
2199 struct iommu_gpasid_bind_data data = { 0 };
2202 if (unlikely(!domain->ops->sva_bind_gpasid))
2205 ret = iommu_sva_prepare_bind_data(udata, &data);
2209 return domain->ops->sva_bind_gpasid(domain, dev, &data);
2211 EXPORT_SYMBOL_GPL(iommu_uapi_sva_bind_gpasid);
2213 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2216 if (unlikely(!domain->ops->sva_unbind_gpasid))
2219 return domain->ops->sva_unbind_gpasid(dev, pasid);
2221 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
2223 int iommu_uapi_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2226 struct iommu_gpasid_bind_data data = { 0 };
2229 if (unlikely(!domain->ops->sva_bind_gpasid))
2232 ret = iommu_sva_prepare_bind_data(udata, &data);
2236 return iommu_sva_unbind_gpasid(domain, dev, data.hpasid);
2238 EXPORT_SYMBOL_GPL(iommu_uapi_sva_unbind_gpasid);
2240 static void __iommu_detach_device(struct iommu_domain *domain,
2243 if (iommu_is_attach_deferred(domain, dev))
2246 if (unlikely(domain->ops->detach_dev == NULL))
2249 domain->ops->detach_dev(domain, dev);
2250 trace_detach_device_from_domain(dev);
2253 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
2255 struct iommu_group *group;
2257 group = iommu_group_get(dev);
2261 mutex_lock(&group->mutex);
2262 if (iommu_group_device_count(group) != 1) {
2267 __iommu_detach_group(domain, group);
2270 mutex_unlock(&group->mutex);
2271 iommu_group_put(group);
2273 EXPORT_SYMBOL_GPL(iommu_detach_device);
2275 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
2277 struct iommu_domain *domain;
2278 struct iommu_group *group;
2280 group = iommu_group_get(dev);
2284 domain = group->domain;
2286 iommu_group_put(group);
2290 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
2293 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2294 * guarantees that the group and its default domain are valid and correct.
2296 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
2298 return dev->iommu_group->default_domain;
2302 * IOMMU groups are really the natural working unit of the IOMMU, but
2303 * the IOMMU API works on domains and devices. Bridge that gap by
2304 * iterating over the devices in a group. Ideally we'd have a single
2305 * device which represents the requestor ID of the group, but we also
2306 * allow IOMMU drivers to create policy defined minimum sets, where
2307 * the physical hardware may be able to distiguish members, but we
2308 * wish to group them at a higher level (ex. untrusted multi-function
2309 * PCI devices). Thus we attach each device.
2311 static int iommu_group_do_attach_device(struct device *dev, void *data)
2313 struct iommu_domain *domain = data;
2315 return __iommu_attach_device(domain, dev);
2318 static int __iommu_attach_group(struct iommu_domain *domain,
2319 struct iommu_group *group)
2323 if (group->default_domain && group->domain != group->default_domain)
2326 ret = __iommu_group_for_each_dev(group, domain,
2327 iommu_group_do_attach_device);
2329 group->domain = domain;
2334 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
2338 mutex_lock(&group->mutex);
2339 ret = __iommu_attach_group(domain, group);
2340 mutex_unlock(&group->mutex);
2344 EXPORT_SYMBOL_GPL(iommu_attach_group);
2346 static int iommu_group_do_detach_device(struct device *dev, void *data)
2348 struct iommu_domain *domain = data;
2350 __iommu_detach_device(domain, dev);
2355 static void __iommu_detach_group(struct iommu_domain *domain,
2356 struct iommu_group *group)
2360 if (!group->default_domain) {
2361 __iommu_group_for_each_dev(group, domain,
2362 iommu_group_do_detach_device);
2363 group->domain = NULL;
2367 if (group->domain == group->default_domain)
2370 /* Detach by re-attaching to the default domain */
2371 ret = __iommu_group_for_each_dev(group, group->default_domain,
2372 iommu_group_do_attach_device);
2376 group->domain = group->default_domain;
2379 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
2381 mutex_lock(&group->mutex);
2382 __iommu_detach_group(domain, group);
2383 mutex_unlock(&group->mutex);
2385 EXPORT_SYMBOL_GPL(iommu_detach_group);
2387 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
2389 if (domain->type == IOMMU_DOMAIN_IDENTITY)
2392 if (domain->type == IOMMU_DOMAIN_BLOCKED)
2395 return domain->ops->iova_to_phys(domain, iova);
2397 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
2399 static size_t iommu_pgsize(struct iommu_domain *domain, unsigned long iova,
2400 phys_addr_t paddr, size_t size, size_t *count)
2402 unsigned int pgsize_idx, pgsize_idx_next;
2403 unsigned long pgsizes;
2404 size_t offset, pgsize, pgsize_next;
2405 unsigned long addr_merge = paddr | iova;
2407 /* Page sizes supported by the hardware and small enough for @size */
2408 pgsizes = domain->pgsize_bitmap & GENMASK(__fls(size), 0);
2410 /* Constrain the page sizes further based on the maximum alignment */
2411 if (likely(addr_merge))
2412 pgsizes &= GENMASK(__ffs(addr_merge), 0);
2414 /* Make sure we have at least one suitable page size */
2417 /* Pick the biggest page size remaining */
2418 pgsize_idx = __fls(pgsizes);
2419 pgsize = BIT(pgsize_idx);
2423 /* Find the next biggest support page size, if it exists */
2424 pgsizes = domain->pgsize_bitmap & ~GENMASK(pgsize_idx, 0);
2428 pgsize_idx_next = __ffs(pgsizes);
2429 pgsize_next = BIT(pgsize_idx_next);
2432 * There's no point trying a bigger page size unless the virtual
2433 * and physical addresses are similarly offset within the larger page.
2435 if ((iova ^ paddr) & (pgsize_next - 1))
2438 /* Calculate the offset to the next page size alignment boundary */
2439 offset = pgsize_next - (addr_merge & (pgsize_next - 1));
2442 * If size is big enough to accommodate the larger page, reduce
2443 * the number of smaller pages.
2445 if (offset + pgsize_next <= size)
2449 *count = size >> pgsize_idx;
2453 static int __iommu_map_pages(struct iommu_domain *domain, unsigned long iova,
2454 phys_addr_t paddr, size_t size, int prot,
2455 gfp_t gfp, size_t *mapped)
2457 const struct iommu_ops *ops = domain->ops;
2458 size_t pgsize, count;
2461 pgsize = iommu_pgsize(domain, iova, paddr, size, &count);
2463 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx count %zu\n",
2464 iova, &paddr, pgsize, count);
2466 if (ops->map_pages) {
2467 ret = ops->map_pages(domain, iova, paddr, pgsize, count, prot,
2470 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
2471 *mapped = ret ? 0 : pgsize;
2477 static int __iommu_map(struct iommu_domain *domain, unsigned long iova,
2478 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2480 const struct iommu_ops *ops = domain->ops;
2481 unsigned long orig_iova = iova;
2482 unsigned int min_pagesz;
2483 size_t orig_size = size;
2484 phys_addr_t orig_paddr = paddr;
2487 if (unlikely(!(ops->map || ops->map_pages) ||
2488 domain->pgsize_bitmap == 0UL))
2491 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2494 /* find out the minimum page size supported */
2495 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2498 * both the virtual address and the physical one, as well as
2499 * the size of the mapping, must be aligned (at least) to the
2500 * size of the smallest page supported by the hardware
2502 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
2503 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2504 iova, &paddr, size, min_pagesz);
2508 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
2513 ret = __iommu_map_pages(domain, iova, paddr, size, prot, gfp,
2516 * Some pages may have been mapped, even if an error occurred,
2517 * so we should account for those so they can be unmapped.
2528 /* unroll mapping in case something went wrong */
2530 iommu_unmap(domain, orig_iova, orig_size - size);
2532 trace_map(orig_iova, orig_paddr, orig_size);
2537 static int _iommu_map(struct iommu_domain *domain, unsigned long iova,
2538 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2540 const struct iommu_ops *ops = domain->ops;
2543 ret = __iommu_map(domain, iova, paddr, size, prot, gfp);
2544 if (ret == 0 && ops->iotlb_sync_map)
2545 ops->iotlb_sync_map(domain, iova, size);
2550 int iommu_map(struct iommu_domain *domain, unsigned long iova,
2551 phys_addr_t paddr, size_t size, int prot)
2554 return _iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
2556 EXPORT_SYMBOL_GPL(iommu_map);
2558 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
2559 phys_addr_t paddr, size_t size, int prot)
2561 return _iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
2563 EXPORT_SYMBOL_GPL(iommu_map_atomic);
2565 static size_t __iommu_unmap_pages(struct iommu_domain *domain,
2566 unsigned long iova, size_t size,
2567 struct iommu_iotlb_gather *iotlb_gather)
2569 const struct iommu_ops *ops = domain->ops;
2570 size_t pgsize, count;
2572 pgsize = iommu_pgsize(domain, iova, iova, size, &count);
2573 return ops->unmap_pages ?
2574 ops->unmap_pages(domain, iova, pgsize, count, iotlb_gather) :
2575 ops->unmap(domain, iova, pgsize, iotlb_gather);
2578 static size_t __iommu_unmap(struct iommu_domain *domain,
2579 unsigned long iova, size_t size,
2580 struct iommu_iotlb_gather *iotlb_gather)
2582 const struct iommu_ops *ops = domain->ops;
2583 size_t unmapped_page, unmapped = 0;
2584 unsigned long orig_iova = iova;
2585 unsigned int min_pagesz;
2587 if (unlikely(!(ops->unmap || ops->unmap_pages) ||
2588 domain->pgsize_bitmap == 0UL))
2591 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2594 /* find out the minimum page size supported */
2595 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2598 * The virtual address, as well as the size of the mapping, must be
2599 * aligned (at least) to the size of the smallest page supported
2602 if (!IS_ALIGNED(iova | size, min_pagesz)) {
2603 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2604 iova, size, min_pagesz);
2608 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2611 * Keep iterating until we either unmap 'size' bytes (or more)
2612 * or we hit an area that isn't mapped.
2614 while (unmapped < size) {
2615 unmapped_page = __iommu_unmap_pages(domain, iova,
2621 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2622 iova, unmapped_page);
2624 iova += unmapped_page;
2625 unmapped += unmapped_page;
2628 trace_unmap(orig_iova, size, unmapped);
2632 size_t iommu_unmap(struct iommu_domain *domain,
2633 unsigned long iova, size_t size)
2635 struct iommu_iotlb_gather iotlb_gather;
2638 iommu_iotlb_gather_init(&iotlb_gather);
2639 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2640 iommu_iotlb_sync(domain, &iotlb_gather);
2644 EXPORT_SYMBOL_GPL(iommu_unmap);
2646 size_t iommu_unmap_fast(struct iommu_domain *domain,
2647 unsigned long iova, size_t size,
2648 struct iommu_iotlb_gather *iotlb_gather)
2650 return __iommu_unmap(domain, iova, size, iotlb_gather);
2652 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2654 static ssize_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2655 struct scatterlist *sg, unsigned int nents, int prot,
2658 const struct iommu_ops *ops = domain->ops;
2659 size_t len = 0, mapped = 0;
2664 while (i <= nents) {
2665 phys_addr_t s_phys = sg_phys(sg);
2667 if (len && s_phys != start + len) {
2668 ret = __iommu_map(domain, iova + mapped, start,
2689 if (ops->iotlb_sync_map)
2690 ops->iotlb_sync_map(domain, iova, mapped);
2694 /* undo mappings already done */
2695 iommu_unmap(domain, iova, mapped);
2700 ssize_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2701 struct scatterlist *sg, unsigned int nents, int prot)
2704 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2706 EXPORT_SYMBOL_GPL(iommu_map_sg);
2708 ssize_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2709 struct scatterlist *sg, unsigned int nents, int prot)
2711 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2715 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2716 * @domain: the iommu domain where the fault has happened
2717 * @dev: the device where the fault has happened
2718 * @iova: the faulting address
2719 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2721 * This function should be called by the low-level IOMMU implementations
2722 * whenever IOMMU faults happen, to allow high-level users, that are
2723 * interested in such events, to know about them.
2725 * This event may be useful for several possible use cases:
2726 * - mere logging of the event
2727 * - dynamic TLB/PTE loading
2728 * - if restarting of the faulting device is required
2730 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2731 * PTE/TLB loading will one day be supported, implementations will be able
2732 * to tell whether it succeeded or not according to this return value).
2734 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2735 * (though fault handlers can also return -ENOSYS, in case they want to
2736 * elicit the default behavior of the IOMMU drivers).
2738 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2739 unsigned long iova, int flags)
2744 * if upper layers showed interest and installed a fault handler,
2747 if (domain->handler)
2748 ret = domain->handler(domain, dev, iova, flags,
2749 domain->handler_token);
2751 trace_io_page_fault(dev, iova, flags);
2754 EXPORT_SYMBOL_GPL(report_iommu_fault);
2756 static int __init iommu_init(void)
2758 iommu_group_kset = kset_create_and_add("iommu_groups",
2760 BUG_ON(!iommu_group_kset);
2762 iommu_debugfs_setup();
2766 core_initcall(iommu_init);
2768 int iommu_enable_nesting(struct iommu_domain *domain)
2770 if (domain->type != IOMMU_DOMAIN_UNMANAGED)
2772 if (!domain->ops->enable_nesting)
2774 return domain->ops->enable_nesting(domain);
2776 EXPORT_SYMBOL_GPL(iommu_enable_nesting);
2778 int iommu_set_pgtable_quirks(struct iommu_domain *domain,
2779 unsigned long quirk)
2781 if (domain->type != IOMMU_DOMAIN_UNMANAGED)
2783 if (!domain->ops->set_pgtable_quirks)
2785 return domain->ops->set_pgtable_quirks(domain, quirk);
2787 EXPORT_SYMBOL_GPL(iommu_set_pgtable_quirks);
2789 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2791 const struct iommu_ops *ops = dev->bus->iommu_ops;
2793 if (ops && ops->get_resv_regions)
2794 ops->get_resv_regions(dev, list);
2797 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2799 const struct iommu_ops *ops = dev->bus->iommu_ops;
2801 if (ops && ops->put_resv_regions)
2802 ops->put_resv_regions(dev, list);
2806 * generic_iommu_put_resv_regions - Reserved region driver helper
2807 * @dev: device for which to free reserved regions
2808 * @list: reserved region list for device
2810 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2811 * for simple reservations. Memory allocated for each reserved region will be
2812 * freed. If an IOMMU driver allocates additional resources per region, it is
2813 * going to have to implement a custom callback.
2815 void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2817 struct iommu_resv_region *entry, *next;
2819 list_for_each_entry_safe(entry, next, list, list)
2822 EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2824 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2825 size_t length, int prot,
2826 enum iommu_resv_type type)
2828 struct iommu_resv_region *region;
2830 region = kzalloc(sizeof(*region), GFP_KERNEL);
2834 INIT_LIST_HEAD(®ion->list);
2835 region->start = start;
2836 region->length = length;
2837 region->prot = prot;
2838 region->type = type;
2841 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2843 void iommu_set_default_passthrough(bool cmd_line)
2846 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
2847 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2850 void iommu_set_default_translated(bool cmd_line)
2853 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
2854 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2857 bool iommu_default_passthrough(void)
2859 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2861 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2863 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2865 const struct iommu_ops *ops = NULL;
2866 struct iommu_device *iommu;
2868 spin_lock(&iommu_device_lock);
2869 list_for_each_entry(iommu, &iommu_device_list, list)
2870 if (iommu->fwnode == fwnode) {
2874 spin_unlock(&iommu_device_lock);
2878 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2879 const struct iommu_ops *ops)
2881 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2884 return ops == fwspec->ops ? 0 : -EINVAL;
2886 if (!dev_iommu_get(dev))
2889 /* Preallocate for the overwhelmingly common case of 1 ID */
2890 fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2894 of_node_get(to_of_node(iommu_fwnode));
2895 fwspec->iommu_fwnode = iommu_fwnode;
2897 dev_iommu_fwspec_set(dev, fwspec);
2900 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2902 void iommu_fwspec_free(struct device *dev)
2904 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2907 fwnode_handle_put(fwspec->iommu_fwnode);
2909 dev_iommu_fwspec_set(dev, NULL);
2912 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2914 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2916 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2922 new_num = fwspec->num_ids + num_ids;
2924 fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2929 dev_iommu_fwspec_set(dev, fwspec);
2932 for (i = 0; i < num_ids; i++)
2933 fwspec->ids[fwspec->num_ids + i] = ids[i];
2935 fwspec->num_ids = new_num;
2938 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2941 * Per device IOMMU features.
2943 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2945 if (dev->iommu && dev->iommu->iommu_dev) {
2946 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2948 if (ops->dev_enable_feat)
2949 return ops->dev_enable_feat(dev, feat);
2954 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2957 * The device drivers should do the necessary cleanups before calling this.
2958 * For example, before disabling the aux-domain feature, the device driver
2959 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2961 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2963 if (dev->iommu && dev->iommu->iommu_dev) {
2964 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2966 if (ops->dev_disable_feat)
2967 return ops->dev_disable_feat(dev, feat);
2972 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2974 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2976 if (dev->iommu && dev->iommu->iommu_dev) {
2977 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2979 if (ops->dev_feat_enabled)
2980 return ops->dev_feat_enabled(dev, feat);
2985 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2988 * Aux-domain specific attach/detach.
2990 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2991 * true. Also, as long as domains are attached to a device through this
2992 * interface, any tries to call iommu_attach_device() should fail
2993 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2994 * This should make us safe against a device being attached to a guest as a
2995 * whole while there are still pasid users on it (aux and sva).
2997 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
3001 if (domain->ops->aux_attach_dev)
3002 ret = domain->ops->aux_attach_dev(domain, dev);
3005 trace_attach_device_to_domain(dev);
3009 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
3011 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
3013 if (domain->ops->aux_detach_dev) {
3014 domain->ops->aux_detach_dev(domain, dev);
3015 trace_detach_device_from_domain(dev);
3018 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
3020 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
3024 if (domain->ops->aux_get_pasid)
3025 ret = domain->ops->aux_get_pasid(domain, dev);
3029 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
3032 * iommu_sva_bind_device() - Bind a process address space to a device
3034 * @mm: the mm to bind, caller must hold a reference to it
3036 * Create a bond between device and address space, allowing the device to access
3037 * the mm using the returned PASID. If a bond already exists between @device and
3038 * @mm, it is returned and an additional reference is taken. Caller must call
3039 * iommu_sva_unbind_device() to release each reference.
3041 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
3042 * initialize the required SVA features.
3044 * On error, returns an ERR_PTR value.
3047 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
3049 struct iommu_group *group;
3050 struct iommu_sva *handle = ERR_PTR(-EINVAL);
3051 const struct iommu_ops *ops = dev->bus->iommu_ops;
3053 if (!ops || !ops->sva_bind)
3054 return ERR_PTR(-ENODEV);
3056 group = iommu_group_get(dev);
3058 return ERR_PTR(-ENODEV);
3060 /* Ensure device count and domain don't change while we're binding */
3061 mutex_lock(&group->mutex);
3064 * To keep things simple, SVA currently doesn't support IOMMU groups
3065 * with more than one device. Existing SVA-capable systems are not
3066 * affected by the problems that required IOMMU groups (lack of ACS
3067 * isolation, device ID aliasing and other hardware issues).
3069 if (iommu_group_device_count(group) != 1)
3072 handle = ops->sva_bind(dev, mm, drvdata);
3075 mutex_unlock(&group->mutex);
3076 iommu_group_put(group);
3080 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
3083 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
3084 * @handle: the handle returned by iommu_sva_bind_device()
3086 * Put reference to a bond between device and address space. The device should
3087 * not be issuing any more transaction for this PASID. All outstanding page
3088 * requests for this PASID must have been flushed to the IOMMU.
3090 void iommu_sva_unbind_device(struct iommu_sva *handle)
3092 struct iommu_group *group;
3093 struct device *dev = handle->dev;
3094 const struct iommu_ops *ops = dev->bus->iommu_ops;
3096 if (!ops || !ops->sva_unbind)
3099 group = iommu_group_get(dev);
3103 mutex_lock(&group->mutex);
3104 ops->sva_unbind(handle);
3105 mutex_unlock(&group->mutex);
3107 iommu_group_put(group);
3109 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
3111 u32 iommu_sva_get_pasid(struct iommu_sva *handle)
3113 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
3115 if (!ops || !ops->sva_get_pasid)
3116 return IOMMU_PASID_INVALID;
3118 return ops->sva_get_pasid(handle);
3120 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
3123 * Changes the default domain of an iommu group that has *only* one device
3125 * @group: The group for which the default domain should be changed
3126 * @prev_dev: The device in the group (this is used to make sure that the device
3127 * hasn't changed after the caller has called this function)
3128 * @type: The type of the new default domain that gets associated with the group
3130 * Returns 0 on success and error code on failure
3133 * 1. Presently, this function is called only when user requests to change the
3134 * group's default domain type through /sys/kernel/iommu_groups/<grp_id>/type
3135 * Please take a closer look if intended to use for other purposes.
3137 static int iommu_change_dev_def_domain(struct iommu_group *group,
3138 struct device *prev_dev, int type)
3140 struct iommu_domain *prev_dom;
3141 struct group_device *grp_dev;
3142 int ret, dev_def_dom;
3145 mutex_lock(&group->mutex);
3147 if (group->default_domain != group->domain) {
3148 dev_err_ratelimited(prev_dev, "Group not assigned to default domain\n");
3154 * iommu group wasn't locked while acquiring device lock in
3155 * iommu_group_store_type(). So, make sure that the device count hasn't
3156 * changed while acquiring device lock.
3158 * Changing default domain of an iommu group with two or more devices
3159 * isn't supported because there could be a potential deadlock. Consider
3160 * the following scenario. T1 is trying to acquire device locks of all
3161 * the devices in the group and before it could acquire all of them,
3162 * there could be another thread T2 (from different sub-system and use
3163 * case) that has already acquired some of the device locks and might be
3164 * waiting for T1 to release other device locks.
3166 if (iommu_group_device_count(group) != 1) {
3167 dev_err_ratelimited(prev_dev, "Cannot change default domain: Group has more than one device\n");
3172 /* Since group has only one device */
3173 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3176 if (prev_dev != dev) {
3177 dev_err_ratelimited(prev_dev, "Cannot change default domain: Device has been changed\n");
3182 prev_dom = group->default_domain;
3188 dev_def_dom = iommu_get_def_domain_type(dev);
3191 * If the user hasn't requested any specific type of domain and
3192 * if the device supports both the domains, then default to the
3193 * domain the device was booted with
3195 type = dev_def_dom ? : iommu_def_domain_type;
3196 } else if (dev_def_dom && type != dev_def_dom) {
3197 dev_err_ratelimited(prev_dev, "Device cannot be in %s domain\n",
3198 iommu_domain_type_str(type));
3204 * Switch to a new domain only if the requested domain type is different
3205 * from the existing default domain type
3207 if (prev_dom->type == type) {
3212 /* We can bring up a flush queue without tearing down the domain */
3213 if (type == IOMMU_DOMAIN_DMA_FQ && prev_dom->type == IOMMU_DOMAIN_DMA) {
3214 ret = iommu_dma_init_fq(prev_dom);
3216 prev_dom->type = IOMMU_DOMAIN_DMA_FQ;
3220 /* Sets group->default_domain to the newly allocated domain */
3221 ret = iommu_group_alloc_default_domain(dev->bus, group, type);
3225 ret = iommu_create_device_direct_mappings(group, dev);
3227 goto free_new_domain;
3229 ret = __iommu_attach_device(group->default_domain, dev);
3231 goto free_new_domain;
3233 group->domain = group->default_domain;
3236 * Release the mutex here because ops->probe_finalize() call-back of
3237 * some vendor IOMMU drivers calls arm_iommu_attach_device() which
3238 * in-turn might call back into IOMMU core code, where it tries to take
3239 * group->mutex, resulting in a deadlock.
3241 mutex_unlock(&group->mutex);
3243 /* Make sure dma_ops is appropriatley set */
3244 iommu_group_do_probe_finalize(dev, group->default_domain);
3245 iommu_domain_free(prev_dom);
3249 iommu_domain_free(group->default_domain);
3250 group->default_domain = prev_dom;
3251 group->domain = prev_dom;
3254 mutex_unlock(&group->mutex);
3260 * Changing the default domain through sysfs requires the users to unbind the
3261 * drivers from the devices in the iommu group, except for a DMA -> DMA-FQ
3262 * transition. Return failure if this isn't met.
3264 * We need to consider the race between this and the device release path.
3265 * device_lock(dev) is used here to guarantee that the device release path
3266 * will not be entered at the same time.
3268 static ssize_t iommu_group_store_type(struct iommu_group *group,
3269 const char *buf, size_t count)
3271 struct group_device *grp_dev;
3275 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
3278 if (WARN_ON(!group))
3281 if (sysfs_streq(buf, "identity"))
3282 req_type = IOMMU_DOMAIN_IDENTITY;
3283 else if (sysfs_streq(buf, "DMA"))
3284 req_type = IOMMU_DOMAIN_DMA;
3285 else if (sysfs_streq(buf, "DMA-FQ"))
3286 req_type = IOMMU_DOMAIN_DMA_FQ;
3287 else if (sysfs_streq(buf, "auto"))
3293 * Lock/Unlock the group mutex here before device lock to
3294 * 1. Make sure that the iommu group has only one device (this is a
3295 * prerequisite for step 2)
3296 * 2. Get struct *dev which is needed to lock device
3298 mutex_lock(&group->mutex);
3299 if (iommu_group_device_count(group) != 1) {
3300 mutex_unlock(&group->mutex);
3301 pr_err_ratelimited("Cannot change default domain: Group has more than one device\n");
3305 /* Since group has only one device */
3306 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3311 * Don't hold the group mutex because taking group mutex first and then
3312 * the device lock could potentially cause a deadlock as below. Assume
3313 * two threads T1 and T2. T1 is trying to change default domain of an
3314 * iommu group and T2 is trying to hot unplug a device or release [1] VF
3315 * of a PCIe device which is in the same iommu group. T1 takes group
3316 * mutex and before it could take device lock assume T2 has taken device
3317 * lock and is yet to take group mutex. Now, both the threads will be
3318 * waiting for the other thread to release lock. Below, lock order was
3321 * mutex_lock(&group->mutex);
3322 * iommu_change_dev_def_domain();
3323 * mutex_unlock(&group->mutex);
3324 * device_unlock(dev);
3326 * [1] Typical device release path
3327 * device_lock() from device/driver core code
3329 * -> iommu_bus_notifier()
3330 * -> iommu_release_device()
3331 * -> ops->release_device() vendor driver calls back iommu core code
3332 * -> mutex_lock() from iommu core code
3334 mutex_unlock(&group->mutex);
3336 /* Check if the device in the group still has a driver bound to it */
3338 if (device_is_bound(dev) && !(req_type == IOMMU_DOMAIN_DMA_FQ &&
3339 group->default_domain->type == IOMMU_DOMAIN_DMA)) {
3340 pr_err_ratelimited("Device is still bound to driver\n");
3345 ret = iommu_change_dev_def_domain(group, dev, req_type);