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 struct dev_iommu *param = dev->iommu;
214 fwnode_handle_put(param->fwspec->iommu_fwnode);
215 kfree(param->fwspec);
220 static int __iommu_probe_device(struct device *dev, struct list_head *group_list)
222 const struct iommu_ops *ops = dev->bus->iommu_ops;
223 struct iommu_device *iommu_dev;
224 struct iommu_group *group;
230 if (!dev_iommu_get(dev))
233 if (!try_module_get(ops->owner)) {
238 iommu_dev = ops->probe_device(dev);
239 if (IS_ERR(iommu_dev)) {
240 ret = PTR_ERR(iommu_dev);
244 dev->iommu->iommu_dev = iommu_dev;
246 group = iommu_group_get_for_dev(dev);
248 ret = PTR_ERR(group);
251 iommu_group_put(group);
253 if (group_list && !group->default_domain && list_empty(&group->entry))
254 list_add_tail(&group->entry, group_list);
256 iommu_device_link(iommu_dev, dev);
261 ops->release_device(dev);
264 module_put(ops->owner);
272 int iommu_probe_device(struct device *dev)
274 const struct iommu_ops *ops = dev->bus->iommu_ops;
275 struct iommu_group *group;
278 ret = __iommu_probe_device(dev, NULL);
282 group = iommu_group_get(dev);
289 * Try to allocate a default domain - needs support from the
290 * IOMMU driver. There are still some drivers which don't
291 * support default domains, so the return value is not yet
294 mutex_lock(&group->mutex);
295 iommu_alloc_default_domain(group, dev);
297 if (group->default_domain) {
298 ret = __iommu_attach_device(group->default_domain, dev);
300 mutex_unlock(&group->mutex);
301 iommu_group_put(group);
306 iommu_create_device_direct_mappings(group, dev);
308 mutex_unlock(&group->mutex);
309 iommu_group_put(group);
311 if (ops->probe_finalize)
312 ops->probe_finalize(dev);
317 iommu_release_device(dev);
324 void iommu_release_device(struct device *dev)
326 const struct iommu_ops *ops = dev->bus->iommu_ops;
331 iommu_device_unlink(dev->iommu->iommu_dev, dev);
333 ops->release_device(dev);
335 iommu_group_remove_device(dev);
336 module_put(ops->owner);
340 static int __init iommu_set_def_domain_type(char *str)
345 ret = kstrtobool(str, &pt);
350 iommu_set_default_passthrough(true);
352 iommu_set_default_translated(true);
356 early_param("iommu.passthrough", iommu_set_def_domain_type);
358 static int __init iommu_dma_setup(char *str)
360 int ret = kstrtobool(str, &iommu_dma_strict);
363 iommu_cmd_line |= IOMMU_CMD_LINE_STRICT;
366 early_param("iommu.strict", iommu_dma_setup);
368 void iommu_set_dma_strict(void)
370 iommu_dma_strict = true;
371 if (iommu_def_domain_type == IOMMU_DOMAIN_DMA_FQ)
372 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
375 static ssize_t iommu_group_attr_show(struct kobject *kobj,
376 struct attribute *__attr, char *buf)
378 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
379 struct iommu_group *group = to_iommu_group(kobj);
383 ret = attr->show(group, buf);
387 static ssize_t iommu_group_attr_store(struct kobject *kobj,
388 struct attribute *__attr,
389 const char *buf, size_t count)
391 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
392 struct iommu_group *group = to_iommu_group(kobj);
396 ret = attr->store(group, buf, count);
400 static const struct sysfs_ops iommu_group_sysfs_ops = {
401 .show = iommu_group_attr_show,
402 .store = iommu_group_attr_store,
405 static int iommu_group_create_file(struct iommu_group *group,
406 struct iommu_group_attribute *attr)
408 return sysfs_create_file(&group->kobj, &attr->attr);
411 static void iommu_group_remove_file(struct iommu_group *group,
412 struct iommu_group_attribute *attr)
414 sysfs_remove_file(&group->kobj, &attr->attr);
417 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
419 return sprintf(buf, "%s\n", group->name);
423 * iommu_insert_resv_region - Insert a new region in the
424 * list of reserved regions.
425 * @new: new region to insert
426 * @regions: list of regions
428 * Elements are sorted by start address and overlapping segments
429 * of the same type are merged.
431 static int iommu_insert_resv_region(struct iommu_resv_region *new,
432 struct list_head *regions)
434 struct iommu_resv_region *iter, *tmp, *nr, *top;
437 nr = iommu_alloc_resv_region(new->start, new->length,
438 new->prot, new->type);
442 /* First add the new element based on start address sorting */
443 list_for_each_entry(iter, regions, list) {
444 if (nr->start < iter->start ||
445 (nr->start == iter->start && nr->type <= iter->type))
448 list_add_tail(&nr->list, &iter->list);
450 /* Merge overlapping segments of type nr->type in @regions, if any */
451 list_for_each_entry_safe(iter, tmp, regions, list) {
452 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
454 /* no merge needed on elements of different types than @new */
455 if (iter->type != new->type) {
456 list_move_tail(&iter->list, &stack);
460 /* look for the last stack element of same type as @iter */
461 list_for_each_entry_reverse(top, &stack, list)
462 if (top->type == iter->type)
465 list_move_tail(&iter->list, &stack);
469 top_end = top->start + top->length - 1;
471 if (iter->start > top_end + 1) {
472 list_move_tail(&iter->list, &stack);
474 top->length = max(top_end, iter_end) - top->start + 1;
475 list_del(&iter->list);
479 list_splice(&stack, regions);
484 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
485 struct list_head *group_resv_regions)
487 struct iommu_resv_region *entry;
490 list_for_each_entry(entry, dev_resv_regions, list) {
491 ret = iommu_insert_resv_region(entry, group_resv_regions);
498 int iommu_get_group_resv_regions(struct iommu_group *group,
499 struct list_head *head)
501 struct group_device *device;
504 mutex_lock(&group->mutex);
505 list_for_each_entry(device, &group->devices, list) {
506 struct list_head dev_resv_regions;
508 INIT_LIST_HEAD(&dev_resv_regions);
509 iommu_get_resv_regions(device->dev, &dev_resv_regions);
510 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
511 iommu_put_resv_regions(device->dev, &dev_resv_regions);
515 mutex_unlock(&group->mutex);
518 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
520 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
523 struct iommu_resv_region *region, *next;
524 struct list_head group_resv_regions;
527 INIT_LIST_HEAD(&group_resv_regions);
528 iommu_get_group_resv_regions(group, &group_resv_regions);
530 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
531 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
532 (long long int)region->start,
533 (long long int)(region->start +
535 iommu_group_resv_type_string[region->type]);
542 static ssize_t iommu_group_show_type(struct iommu_group *group,
545 char *type = "unknown\n";
547 mutex_lock(&group->mutex);
548 if (group->default_domain) {
549 switch (group->default_domain->type) {
550 case IOMMU_DOMAIN_BLOCKED:
553 case IOMMU_DOMAIN_IDENTITY:
556 case IOMMU_DOMAIN_UNMANAGED:
557 type = "unmanaged\n";
559 case IOMMU_DOMAIN_DMA:
562 case IOMMU_DOMAIN_DMA_FQ:
567 mutex_unlock(&group->mutex);
573 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
575 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
576 iommu_group_show_resv_regions, NULL);
578 static IOMMU_GROUP_ATTR(type, 0644, iommu_group_show_type,
579 iommu_group_store_type);
581 static void iommu_group_release(struct kobject *kobj)
583 struct iommu_group *group = to_iommu_group(kobj);
585 pr_debug("Releasing group %d\n", group->id);
587 if (group->iommu_data_release)
588 group->iommu_data_release(group->iommu_data);
590 ida_simple_remove(&iommu_group_ida, group->id);
592 if (group->default_domain)
593 iommu_domain_free(group->default_domain);
599 static struct kobj_type iommu_group_ktype = {
600 .sysfs_ops = &iommu_group_sysfs_ops,
601 .release = iommu_group_release,
605 * iommu_group_alloc - Allocate a new group
607 * This function is called by an iommu driver to allocate a new iommu
608 * group. The iommu group represents the minimum granularity of the iommu.
609 * Upon successful return, the caller holds a reference to the supplied
610 * group in order to hold the group until devices are added. Use
611 * iommu_group_put() to release this extra reference count, allowing the
612 * group to be automatically reclaimed once it has no devices or external
615 struct iommu_group *iommu_group_alloc(void)
617 struct iommu_group *group;
620 group = kzalloc(sizeof(*group), GFP_KERNEL);
622 return ERR_PTR(-ENOMEM);
624 group->kobj.kset = iommu_group_kset;
625 mutex_init(&group->mutex);
626 INIT_LIST_HEAD(&group->devices);
627 INIT_LIST_HEAD(&group->entry);
628 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
630 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
637 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
638 NULL, "%d", group->id);
640 ida_simple_remove(&iommu_group_ida, group->id);
641 kobject_put(&group->kobj);
645 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
646 if (!group->devices_kobj) {
647 kobject_put(&group->kobj); /* triggers .release & free */
648 return ERR_PTR(-ENOMEM);
652 * The devices_kobj holds a reference on the group kobject, so
653 * as long as that exists so will the group. We can therefore
654 * use the devices_kobj for reference counting.
656 kobject_put(&group->kobj);
658 ret = iommu_group_create_file(group,
659 &iommu_group_attr_reserved_regions);
663 ret = iommu_group_create_file(group, &iommu_group_attr_type);
667 pr_debug("Allocated group %d\n", group->id);
671 EXPORT_SYMBOL_GPL(iommu_group_alloc);
673 struct iommu_group *iommu_group_get_by_id(int id)
675 struct kobject *group_kobj;
676 struct iommu_group *group;
679 if (!iommu_group_kset)
682 name = kasprintf(GFP_KERNEL, "%d", id);
686 group_kobj = kset_find_obj(iommu_group_kset, name);
692 group = container_of(group_kobj, struct iommu_group, kobj);
693 BUG_ON(group->id != id);
695 kobject_get(group->devices_kobj);
696 kobject_put(&group->kobj);
700 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
703 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
706 * iommu drivers can store data in the group for use when doing iommu
707 * operations. This function provides a way to retrieve it. Caller
708 * should hold a group reference.
710 void *iommu_group_get_iommudata(struct iommu_group *group)
712 return group->iommu_data;
714 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
717 * iommu_group_set_iommudata - set iommu_data for a group
719 * @iommu_data: new data
720 * @release: release function for iommu_data
722 * iommu drivers can store data in the group for use when doing iommu
723 * operations. This function provides a way to set the data after
724 * the group has been allocated. Caller should hold a group reference.
726 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
727 void (*release)(void *iommu_data))
729 group->iommu_data = iommu_data;
730 group->iommu_data_release = release;
732 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
735 * iommu_group_set_name - set name for a group
739 * Allow iommu driver to set a name for a group. When set it will
740 * appear in a name attribute file under the group in sysfs.
742 int iommu_group_set_name(struct iommu_group *group, const char *name)
747 iommu_group_remove_file(group, &iommu_group_attr_name);
754 group->name = kstrdup(name, GFP_KERNEL);
758 ret = iommu_group_create_file(group, &iommu_group_attr_name);
767 EXPORT_SYMBOL_GPL(iommu_group_set_name);
769 static int iommu_create_device_direct_mappings(struct iommu_group *group,
772 struct iommu_domain *domain = group->default_domain;
773 struct iommu_resv_region *entry;
774 struct list_head mappings;
775 unsigned long pg_size;
778 if (!domain || !iommu_is_dma_domain(domain))
781 BUG_ON(!domain->pgsize_bitmap);
783 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
784 INIT_LIST_HEAD(&mappings);
786 iommu_get_resv_regions(dev, &mappings);
788 /* We need to consider overlapping regions for different devices */
789 list_for_each_entry(entry, &mappings, list) {
790 dma_addr_t start, end, addr;
793 if (domain->ops->apply_resv_region)
794 domain->ops->apply_resv_region(dev, domain, entry);
796 start = ALIGN(entry->start, pg_size);
797 end = ALIGN(entry->start + entry->length, pg_size);
799 if (entry->type != IOMMU_RESV_DIRECT &&
800 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
803 for (addr = start; addr <= end; addr += pg_size) {
804 phys_addr_t phys_addr;
809 phys_addr = iommu_iova_to_phys(domain, addr);
817 ret = iommu_map(domain, addr - map_size,
818 addr - map_size, map_size,
828 iommu_flush_iotlb_all(domain);
831 iommu_put_resv_regions(dev, &mappings);
836 static bool iommu_is_attach_deferred(struct iommu_domain *domain,
839 if (domain->ops->is_attach_deferred)
840 return domain->ops->is_attach_deferred(domain, dev);
846 * iommu_group_add_device - add a device to an iommu group
847 * @group: the group into which to add the device (reference should be held)
850 * This function is called by an iommu driver to add a device into a
851 * group. Adding a device increments the group reference count.
853 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
856 struct group_device *device;
858 device = kzalloc(sizeof(*device), GFP_KERNEL);
864 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
866 goto err_free_device;
868 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
872 goto err_remove_link;
875 ret = sysfs_create_link_nowarn(group->devices_kobj,
876 &dev->kobj, device->name);
878 if (ret == -EEXIST && i >= 0) {
880 * Account for the slim chance of collision
881 * and append an instance to the name.
884 device->name = kasprintf(GFP_KERNEL, "%s.%d",
885 kobject_name(&dev->kobj), i++);
891 kobject_get(group->devices_kobj);
893 dev->iommu_group = group;
895 mutex_lock(&group->mutex);
896 list_add_tail(&device->list, &group->devices);
897 if (group->domain && !iommu_is_attach_deferred(group->domain, dev))
898 ret = __iommu_attach_device(group->domain, dev);
899 mutex_unlock(&group->mutex);
903 /* Notify any listeners about change to group. */
904 blocking_notifier_call_chain(&group->notifier,
905 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
907 trace_add_device_to_group(group->id, dev);
909 dev_info(dev, "Adding to iommu group %d\n", group->id);
914 mutex_lock(&group->mutex);
915 list_del(&device->list);
916 mutex_unlock(&group->mutex);
917 dev->iommu_group = NULL;
918 kobject_put(group->devices_kobj);
919 sysfs_remove_link(group->devices_kobj, device->name);
923 sysfs_remove_link(&dev->kobj, "iommu_group");
926 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
929 EXPORT_SYMBOL_GPL(iommu_group_add_device);
932 * iommu_group_remove_device - remove a device from it's current group
933 * @dev: device to be removed
935 * This function is called by an iommu driver to remove the device from
936 * it's current group. This decrements the iommu group reference count.
938 void iommu_group_remove_device(struct device *dev)
940 struct iommu_group *group = dev->iommu_group;
941 struct group_device *tmp_device, *device = NULL;
946 dev_info(dev, "Removing from iommu group %d\n", group->id);
948 /* Pre-notify listeners that a device is being removed. */
949 blocking_notifier_call_chain(&group->notifier,
950 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
952 mutex_lock(&group->mutex);
953 list_for_each_entry(tmp_device, &group->devices, list) {
954 if (tmp_device->dev == dev) {
956 list_del(&device->list);
960 mutex_unlock(&group->mutex);
965 sysfs_remove_link(group->devices_kobj, device->name);
966 sysfs_remove_link(&dev->kobj, "iommu_group");
968 trace_remove_device_from_group(group->id, dev);
972 dev->iommu_group = NULL;
973 kobject_put(group->devices_kobj);
975 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
977 static int iommu_group_device_count(struct iommu_group *group)
979 struct group_device *entry;
982 list_for_each_entry(entry, &group->devices, list)
988 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
989 int (*fn)(struct device *, void *))
991 struct group_device *device;
994 list_for_each_entry(device, &group->devices, list) {
995 ret = fn(device->dev, data);
1003 * iommu_group_for_each_dev - iterate over each device in the group
1005 * @data: caller opaque data to be passed to callback function
1006 * @fn: caller supplied callback function
1008 * This function is called by group users to iterate over group devices.
1009 * Callers should hold a reference count to the group during callback.
1010 * The group->mutex is held across callbacks, which will block calls to
1011 * iommu_group_add/remove_device.
1013 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
1014 int (*fn)(struct device *, void *))
1018 mutex_lock(&group->mutex);
1019 ret = __iommu_group_for_each_dev(group, data, fn);
1020 mutex_unlock(&group->mutex);
1024 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
1027 * iommu_group_get - Return the group for a device and increment reference
1028 * @dev: get the group that this device belongs to
1030 * This function is called by iommu drivers and users to get the group
1031 * for the specified device. If found, the group is returned and the group
1032 * reference in incremented, else NULL.
1034 struct iommu_group *iommu_group_get(struct device *dev)
1036 struct iommu_group *group = dev->iommu_group;
1039 kobject_get(group->devices_kobj);
1043 EXPORT_SYMBOL_GPL(iommu_group_get);
1046 * iommu_group_ref_get - Increment reference on a group
1047 * @group: the group to use, must not be NULL
1049 * This function is called by iommu drivers to take additional references on an
1050 * existing group. Returns the given group for convenience.
1052 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
1054 kobject_get(group->devices_kobj);
1057 EXPORT_SYMBOL_GPL(iommu_group_ref_get);
1060 * iommu_group_put - Decrement group reference
1061 * @group: the group to use
1063 * This function is called by iommu drivers and users to release the
1064 * iommu group. Once the reference count is zero, the group is released.
1066 void iommu_group_put(struct iommu_group *group)
1069 kobject_put(group->devices_kobj);
1071 EXPORT_SYMBOL_GPL(iommu_group_put);
1074 * iommu_group_register_notifier - Register a notifier for group changes
1075 * @group: the group to watch
1076 * @nb: notifier block to signal
1078 * This function allows iommu group users to track changes in a group.
1079 * See include/linux/iommu.h for actions sent via this notifier. Caller
1080 * should hold a reference to the group throughout notifier registration.
1082 int iommu_group_register_notifier(struct iommu_group *group,
1083 struct notifier_block *nb)
1085 return blocking_notifier_chain_register(&group->notifier, nb);
1087 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
1090 * iommu_group_unregister_notifier - Unregister a notifier
1091 * @group: the group to watch
1092 * @nb: notifier block to signal
1094 * Unregister a previously registered group notifier block.
1096 int iommu_group_unregister_notifier(struct iommu_group *group,
1097 struct notifier_block *nb)
1099 return blocking_notifier_chain_unregister(&group->notifier, nb);
1101 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
1104 * iommu_register_device_fault_handler() - Register a device fault handler
1106 * @handler: the fault handler
1107 * @data: private data passed as argument to the handler
1109 * When an IOMMU fault event is received, this handler gets called with the
1110 * fault event and data as argument. The handler should return 0 on success. If
1111 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1112 * complete the fault by calling iommu_page_response() with one of the following
1114 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1115 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1116 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1117 * page faults if possible.
1119 * Return 0 if the fault handler was installed successfully, or an error.
1121 int iommu_register_device_fault_handler(struct device *dev,
1122 iommu_dev_fault_handler_t handler,
1125 struct dev_iommu *param = dev->iommu;
1131 mutex_lock(¶m->lock);
1132 /* Only allow one fault handler registered for each device */
1133 if (param->fault_param) {
1139 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
1140 if (!param->fault_param) {
1145 param->fault_param->handler = handler;
1146 param->fault_param->data = data;
1147 mutex_init(¶m->fault_param->lock);
1148 INIT_LIST_HEAD(¶m->fault_param->faults);
1151 mutex_unlock(¶m->lock);
1155 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1158 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1161 * Remove the device fault handler installed with
1162 * iommu_register_device_fault_handler().
1164 * Return 0 on success, or an error.
1166 int iommu_unregister_device_fault_handler(struct device *dev)
1168 struct dev_iommu *param = dev->iommu;
1174 mutex_lock(¶m->lock);
1176 if (!param->fault_param)
1179 /* we cannot unregister handler if there are pending faults */
1180 if (!list_empty(¶m->fault_param->faults)) {
1185 kfree(param->fault_param);
1186 param->fault_param = NULL;
1189 mutex_unlock(¶m->lock);
1193 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1196 * iommu_report_device_fault() - Report fault event to device driver
1198 * @evt: fault event data
1200 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1201 * handler. When this function fails and the fault is recoverable, it is the
1202 * caller's responsibility to complete the fault.
1204 * Return 0 on success, or an error.
1206 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1208 struct dev_iommu *param = dev->iommu;
1209 struct iommu_fault_event *evt_pending = NULL;
1210 struct iommu_fault_param *fparam;
1216 /* we only report device fault if there is a handler registered */
1217 mutex_lock(¶m->lock);
1218 fparam = param->fault_param;
1219 if (!fparam || !fparam->handler) {
1224 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1225 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1226 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1232 mutex_lock(&fparam->lock);
1233 list_add_tail(&evt_pending->list, &fparam->faults);
1234 mutex_unlock(&fparam->lock);
1237 ret = fparam->handler(&evt->fault, fparam->data);
1238 if (ret && evt_pending) {
1239 mutex_lock(&fparam->lock);
1240 list_del(&evt_pending->list);
1241 mutex_unlock(&fparam->lock);
1245 mutex_unlock(¶m->lock);
1248 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1250 int iommu_page_response(struct device *dev,
1251 struct iommu_page_response *msg)
1255 struct iommu_fault_event *evt;
1256 struct iommu_fault_page_request *prm;
1257 struct dev_iommu *param = dev->iommu;
1258 bool has_pasid = msg->flags & IOMMU_PAGE_RESP_PASID_VALID;
1259 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1261 if (!domain || !domain->ops->page_response)
1264 if (!param || !param->fault_param)
1267 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1268 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1271 /* Only send response if there is a fault report pending */
1272 mutex_lock(¶m->fault_param->lock);
1273 if (list_empty(¶m->fault_param->faults)) {
1274 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1278 * Check if we have a matching page request pending to respond,
1279 * otherwise return -EINVAL
1281 list_for_each_entry(evt, ¶m->fault_param->faults, list) {
1282 prm = &evt->fault.prm;
1283 if (prm->grpid != msg->grpid)
1287 * If the PASID is required, the corresponding request is
1288 * matched using the group ID, the PASID valid bit and the PASID
1289 * value. Otherwise only the group ID matches request and
1292 needs_pasid = prm->flags & IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID;
1293 if (needs_pasid && (!has_pasid || msg->pasid != prm->pasid))
1296 if (!needs_pasid && has_pasid) {
1297 /* No big deal, just clear it. */
1298 msg->flags &= ~IOMMU_PAGE_RESP_PASID_VALID;
1302 ret = domain->ops->page_response(dev, evt, msg);
1303 list_del(&evt->list);
1309 mutex_unlock(¶m->fault_param->lock);
1312 EXPORT_SYMBOL_GPL(iommu_page_response);
1315 * iommu_group_id - Return ID for a group
1316 * @group: the group to ID
1318 * Return the unique ID for the group matching the sysfs group number.
1320 int iommu_group_id(struct iommu_group *group)
1324 EXPORT_SYMBOL_GPL(iommu_group_id);
1326 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1327 unsigned long *devfns);
1330 * To consider a PCI device isolated, we require ACS to support Source
1331 * Validation, Request Redirection, Completer Redirection, and Upstream
1332 * Forwarding. This effectively means that devices cannot spoof their
1333 * requester ID, requests and completions cannot be redirected, and all
1334 * transactions are forwarded upstream, even as it passes through a
1335 * bridge where the target device is downstream.
1337 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1340 * For multifunction devices which are not isolated from each other, find
1341 * all the other non-isolated functions and look for existing groups. For
1342 * each function, we also need to look for aliases to or from other devices
1343 * that may already have a group.
1345 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1346 unsigned long *devfns)
1348 struct pci_dev *tmp = NULL;
1349 struct iommu_group *group;
1351 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1354 for_each_pci_dev(tmp) {
1355 if (tmp == pdev || tmp->bus != pdev->bus ||
1356 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1357 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1360 group = get_pci_alias_group(tmp, devfns);
1371 * Look for aliases to or from the given device for existing groups. DMA
1372 * aliases are only supported on the same bus, therefore the search
1373 * space is quite small (especially since we're really only looking at pcie
1374 * device, and therefore only expect multiple slots on the root complex or
1375 * downstream switch ports). It's conceivable though that a pair of
1376 * multifunction devices could have aliases between them that would cause a
1377 * loop. To prevent this, we use a bitmap to track where we've been.
1379 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1380 unsigned long *devfns)
1382 struct pci_dev *tmp = NULL;
1383 struct iommu_group *group;
1385 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1388 group = iommu_group_get(&pdev->dev);
1392 for_each_pci_dev(tmp) {
1393 if (tmp == pdev || tmp->bus != pdev->bus)
1396 /* We alias them or they alias us */
1397 if (pci_devs_are_dma_aliases(pdev, tmp)) {
1398 group = get_pci_alias_group(tmp, devfns);
1404 group = get_pci_function_alias_group(tmp, devfns);
1415 struct group_for_pci_data {
1416 struct pci_dev *pdev;
1417 struct iommu_group *group;
1421 * DMA alias iterator callback, return the last seen device. Stop and return
1422 * the IOMMU group if we find one along the way.
1424 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1426 struct group_for_pci_data *data = opaque;
1429 data->group = iommu_group_get(&pdev->dev);
1431 return data->group != NULL;
1435 * Generic device_group call-back function. It just allocates one
1436 * iommu-group per device.
1438 struct iommu_group *generic_device_group(struct device *dev)
1440 return iommu_group_alloc();
1442 EXPORT_SYMBOL_GPL(generic_device_group);
1445 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1446 * to find or create an IOMMU group for a device.
1448 struct iommu_group *pci_device_group(struct device *dev)
1450 struct pci_dev *pdev = to_pci_dev(dev);
1451 struct group_for_pci_data data;
1452 struct pci_bus *bus;
1453 struct iommu_group *group = NULL;
1454 u64 devfns[4] = { 0 };
1456 if (WARN_ON(!dev_is_pci(dev)))
1457 return ERR_PTR(-EINVAL);
1460 * Find the upstream DMA alias for the device. A device must not
1461 * be aliased due to topology in order to have its own IOMMU group.
1462 * If we find an alias along the way that already belongs to a
1465 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1471 * Continue upstream from the point of minimum IOMMU granularity
1472 * due to aliases to the point where devices are protected from
1473 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1476 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1480 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1485 group = iommu_group_get(&pdev->dev);
1491 * Look for existing groups on device aliases. If we alias another
1492 * device or another device aliases us, use the same group.
1494 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1499 * Look for existing groups on non-isolated functions on the same
1500 * slot and aliases of those funcions, if any. No need to clear
1501 * the search bitmap, the tested devfns are still valid.
1503 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1507 /* No shared group found, allocate new */
1508 return iommu_group_alloc();
1510 EXPORT_SYMBOL_GPL(pci_device_group);
1512 /* Get the IOMMU group for device on fsl-mc bus */
1513 struct iommu_group *fsl_mc_device_group(struct device *dev)
1515 struct device *cont_dev = fsl_mc_cont_dev(dev);
1516 struct iommu_group *group;
1518 group = iommu_group_get(cont_dev);
1520 group = iommu_group_alloc();
1523 EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1525 static int iommu_get_def_domain_type(struct device *dev)
1527 const struct iommu_ops *ops = dev->bus->iommu_ops;
1529 if (dev_is_pci(dev) && to_pci_dev(dev)->untrusted)
1530 return IOMMU_DOMAIN_DMA;
1532 if (ops->def_domain_type)
1533 return ops->def_domain_type(dev);
1538 static int iommu_group_alloc_default_domain(struct bus_type *bus,
1539 struct iommu_group *group,
1542 struct iommu_domain *dom;
1544 dom = __iommu_domain_alloc(bus, type);
1545 if (!dom && type != IOMMU_DOMAIN_DMA) {
1546 dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
1548 pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1555 group->default_domain = dom;
1557 group->domain = dom;
1561 static int iommu_alloc_default_domain(struct iommu_group *group,
1566 if (group->default_domain)
1569 type = iommu_get_def_domain_type(dev) ? : iommu_def_domain_type;
1571 return iommu_group_alloc_default_domain(dev->bus, group, type);
1575 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1576 * @dev: target device
1578 * This function is intended to be called by IOMMU drivers and extended to
1579 * support common, bus-defined algorithms when determining or creating the
1580 * IOMMU group for a device. On success, the caller will hold a reference
1581 * to the returned IOMMU group, which will already include the provided
1582 * device. The reference should be released with iommu_group_put().
1584 static struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1586 const struct iommu_ops *ops = dev->bus->iommu_ops;
1587 struct iommu_group *group;
1590 group = iommu_group_get(dev);
1595 return ERR_PTR(-EINVAL);
1597 group = ops->device_group(dev);
1598 if (WARN_ON_ONCE(group == NULL))
1599 return ERR_PTR(-EINVAL);
1604 ret = iommu_group_add_device(group, dev);
1611 iommu_group_put(group);
1613 return ERR_PTR(ret);
1616 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1618 return group->default_domain;
1621 static int probe_iommu_group(struct device *dev, void *data)
1623 struct list_head *group_list = data;
1624 struct iommu_group *group;
1627 /* Device is probed already if in a group */
1628 group = iommu_group_get(dev);
1630 iommu_group_put(group);
1634 ret = __iommu_probe_device(dev, group_list);
1641 static int remove_iommu_group(struct device *dev, void *data)
1643 iommu_release_device(dev);
1648 static int iommu_bus_notifier(struct notifier_block *nb,
1649 unsigned long action, void *data)
1651 unsigned long group_action = 0;
1652 struct device *dev = data;
1653 struct iommu_group *group;
1656 * ADD/DEL call into iommu driver ops if provided, which may
1657 * result in ADD/DEL notifiers to group->notifier
1659 if (action == BUS_NOTIFY_ADD_DEVICE) {
1662 ret = iommu_probe_device(dev);
1663 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1664 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1665 iommu_release_device(dev);
1670 * Remaining BUS_NOTIFYs get filtered and republished to the
1671 * group, if anyone is listening
1673 group = iommu_group_get(dev);
1678 case BUS_NOTIFY_BIND_DRIVER:
1679 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1681 case BUS_NOTIFY_BOUND_DRIVER:
1682 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1684 case BUS_NOTIFY_UNBIND_DRIVER:
1685 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1687 case BUS_NOTIFY_UNBOUND_DRIVER:
1688 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1693 blocking_notifier_call_chain(&group->notifier,
1696 iommu_group_put(group);
1700 struct __group_domain_type {
1705 static int probe_get_default_domain_type(struct device *dev, void *data)
1707 struct __group_domain_type *gtype = data;
1708 unsigned int type = iommu_get_def_domain_type(dev);
1711 if (gtype->type && gtype->type != type) {
1712 dev_warn(dev, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1713 iommu_domain_type_str(type),
1714 dev_name(gtype->dev),
1715 iommu_domain_type_str(gtype->type));
1728 static void probe_alloc_default_domain(struct bus_type *bus,
1729 struct iommu_group *group)
1731 struct __group_domain_type gtype;
1733 memset(>ype, 0, sizeof(gtype));
1735 /* Ask for default domain requirements of all devices in the group */
1736 __iommu_group_for_each_dev(group, >ype,
1737 probe_get_default_domain_type);
1740 gtype.type = iommu_def_domain_type;
1742 iommu_group_alloc_default_domain(bus, group, gtype.type);
1746 static int iommu_group_do_dma_attach(struct device *dev, void *data)
1748 struct iommu_domain *domain = data;
1751 if (!iommu_is_attach_deferred(domain, dev))
1752 ret = __iommu_attach_device(domain, dev);
1757 static int __iommu_group_dma_attach(struct iommu_group *group)
1759 return __iommu_group_for_each_dev(group, group->default_domain,
1760 iommu_group_do_dma_attach);
1763 static int iommu_group_do_probe_finalize(struct device *dev, void *data)
1765 struct iommu_domain *domain = data;
1767 if (domain->ops->probe_finalize)
1768 domain->ops->probe_finalize(dev);
1773 static void __iommu_group_dma_finalize(struct iommu_group *group)
1775 __iommu_group_for_each_dev(group, group->default_domain,
1776 iommu_group_do_probe_finalize);
1779 static int iommu_do_create_direct_mappings(struct device *dev, void *data)
1781 struct iommu_group *group = data;
1783 iommu_create_device_direct_mappings(group, dev);
1788 static int iommu_group_create_direct_mappings(struct iommu_group *group)
1790 return __iommu_group_for_each_dev(group, group,
1791 iommu_do_create_direct_mappings);
1794 int bus_iommu_probe(struct bus_type *bus)
1796 struct iommu_group *group, *next;
1797 LIST_HEAD(group_list);
1801 * This code-path does not allocate the default domain when
1802 * creating the iommu group, so do it after the groups are
1805 ret = bus_for_each_dev(bus, NULL, &group_list, probe_iommu_group);
1809 list_for_each_entry_safe(group, next, &group_list, entry) {
1810 /* Remove item from the list */
1811 list_del_init(&group->entry);
1813 mutex_lock(&group->mutex);
1815 /* Try to allocate default domain */
1816 probe_alloc_default_domain(bus, group);
1818 if (!group->default_domain) {
1819 mutex_unlock(&group->mutex);
1823 iommu_group_create_direct_mappings(group);
1825 ret = __iommu_group_dma_attach(group);
1827 mutex_unlock(&group->mutex);
1832 __iommu_group_dma_finalize(group);
1838 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1840 struct notifier_block *nb;
1843 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1847 nb->notifier_call = iommu_bus_notifier;
1849 err = bus_register_notifier(bus, nb);
1853 err = bus_iommu_probe(bus);
1862 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1863 bus_unregister_notifier(bus, nb);
1872 * bus_set_iommu - set iommu-callbacks for the bus
1874 * @ops: the callbacks provided by the iommu-driver
1876 * This function is called by an iommu driver to set the iommu methods
1877 * used for a particular bus. Drivers for devices on that bus can use
1878 * the iommu-api after these ops are registered.
1879 * This special function is needed because IOMMUs are usually devices on
1880 * the bus itself, so the iommu drivers are not initialized when the bus
1881 * is set up. With this function the iommu-driver can set the iommu-ops
1884 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1889 bus->iommu_ops = NULL;
1893 if (bus->iommu_ops != NULL)
1896 bus->iommu_ops = ops;
1898 /* Do IOMMU specific setup for this bus-type */
1899 err = iommu_bus_init(bus, ops);
1901 bus->iommu_ops = NULL;
1905 EXPORT_SYMBOL_GPL(bus_set_iommu);
1907 bool iommu_present(struct bus_type *bus)
1909 return bus->iommu_ops != NULL;
1911 EXPORT_SYMBOL_GPL(iommu_present);
1913 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1915 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1918 return bus->iommu_ops->capable(cap);
1920 EXPORT_SYMBOL_GPL(iommu_capable);
1923 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1924 * @domain: iommu domain
1925 * @handler: fault handler
1926 * @token: user data, will be passed back to the fault handler
1928 * This function should be used by IOMMU users which want to be notified
1929 * whenever an IOMMU fault happens.
1931 * The fault handler itself should return 0 on success, and an appropriate
1932 * error code otherwise.
1934 void iommu_set_fault_handler(struct iommu_domain *domain,
1935 iommu_fault_handler_t handler,
1940 domain->handler = handler;
1941 domain->handler_token = token;
1943 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1945 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1948 struct iommu_domain *domain;
1950 if (bus == NULL || bus->iommu_ops == NULL)
1953 domain = bus->iommu_ops->domain_alloc(type);
1957 domain->ops = bus->iommu_ops;
1958 domain->type = type;
1959 /* Assume all sizes by default; the driver may override this later */
1960 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1962 if (iommu_is_dma_domain(domain) && iommu_get_dma_cookie(domain)) {
1963 iommu_domain_free(domain);
1969 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1971 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1973 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1975 void iommu_domain_free(struct iommu_domain *domain)
1977 iommu_put_dma_cookie(domain);
1978 domain->ops->domain_free(domain);
1980 EXPORT_SYMBOL_GPL(iommu_domain_free);
1982 static int __iommu_attach_device(struct iommu_domain *domain,
1987 if (unlikely(domain->ops->attach_dev == NULL))
1990 ret = domain->ops->attach_dev(domain, dev);
1992 trace_attach_device_to_domain(dev);
1996 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1998 struct iommu_group *group;
2001 group = iommu_group_get(dev);
2006 * Lock the group to make sure the device-count doesn't
2007 * change while we are attaching
2009 mutex_lock(&group->mutex);
2011 if (iommu_group_device_count(group) != 1)
2014 ret = __iommu_attach_group(domain, group);
2017 mutex_unlock(&group->mutex);
2018 iommu_group_put(group);
2022 EXPORT_SYMBOL_GPL(iommu_attach_device);
2024 int iommu_deferred_attach(struct device *dev, struct iommu_domain *domain)
2026 const struct iommu_ops *ops = domain->ops;
2028 if (ops->is_attach_deferred && ops->is_attach_deferred(domain, dev))
2029 return __iommu_attach_device(domain, dev);
2035 * Check flags and other user provided data for valid combinations. We also
2036 * make sure no reserved fields or unused flags are set. This is to ensure
2037 * not breaking userspace in the future when these fields or flags are used.
2039 static int iommu_check_cache_invl_data(struct iommu_cache_invalidate_info *info)
2044 if (info->version != IOMMU_CACHE_INVALIDATE_INFO_VERSION_1)
2047 mask = (1 << IOMMU_CACHE_INV_TYPE_NR) - 1;
2048 if (info->cache & ~mask)
2051 if (info->granularity >= IOMMU_INV_GRANU_NR)
2054 switch (info->granularity) {
2055 case IOMMU_INV_GRANU_ADDR:
2056 if (info->cache & IOMMU_CACHE_INV_TYPE_PASID)
2059 mask = IOMMU_INV_ADDR_FLAGS_PASID |
2060 IOMMU_INV_ADDR_FLAGS_ARCHID |
2061 IOMMU_INV_ADDR_FLAGS_LEAF;
2063 if (info->granu.addr_info.flags & ~mask)
2066 case IOMMU_INV_GRANU_PASID:
2067 mask = IOMMU_INV_PASID_FLAGS_PASID |
2068 IOMMU_INV_PASID_FLAGS_ARCHID;
2069 if (info->granu.pasid_info.flags & ~mask)
2073 case IOMMU_INV_GRANU_DOMAIN:
2074 if (info->cache & IOMMU_CACHE_INV_TYPE_DEV_IOTLB)
2081 /* Check reserved padding fields */
2082 for (i = 0; i < sizeof(info->padding); i++) {
2083 if (info->padding[i])
2090 int iommu_uapi_cache_invalidate(struct iommu_domain *domain, struct device *dev,
2093 struct iommu_cache_invalidate_info inv_info = { 0 };
2097 if (unlikely(!domain->ops->cache_invalidate))
2101 * No new spaces can be added before the variable sized union, the
2102 * minimum size is the offset to the union.
2104 minsz = offsetof(struct iommu_cache_invalidate_info, granu);
2106 /* Copy minsz from user to get flags and argsz */
2107 if (copy_from_user(&inv_info, uinfo, minsz))
2110 /* Fields before the variable size union are mandatory */
2111 if (inv_info.argsz < minsz)
2114 /* PASID and address granu require additional info beyond minsz */
2115 if (inv_info.granularity == IOMMU_INV_GRANU_PASID &&
2116 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.pasid_info))
2119 if (inv_info.granularity == IOMMU_INV_GRANU_ADDR &&
2120 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.addr_info))
2124 * User might be using a newer UAPI header which has a larger data
2125 * size, we shall support the existing flags within the current
2126 * size. Copy the remaining user data _after_ minsz but not more
2127 * than the current kernel supported size.
2129 if (copy_from_user((void *)&inv_info + minsz, uinfo + minsz,
2130 min_t(u32, inv_info.argsz, sizeof(inv_info)) - minsz))
2133 /* Now the argsz is validated, check the content */
2134 ret = iommu_check_cache_invl_data(&inv_info);
2138 return domain->ops->cache_invalidate(domain, dev, &inv_info);
2140 EXPORT_SYMBOL_GPL(iommu_uapi_cache_invalidate);
2142 static int iommu_check_bind_data(struct iommu_gpasid_bind_data *data)
2147 if (data->version != IOMMU_GPASID_BIND_VERSION_1)
2150 /* Check the range of supported formats */
2151 if (data->format >= IOMMU_PASID_FORMAT_LAST)
2154 /* Check all flags */
2155 mask = IOMMU_SVA_GPASID_VAL;
2156 if (data->flags & ~mask)
2159 /* Check reserved padding fields */
2160 for (i = 0; i < sizeof(data->padding); i++) {
2161 if (data->padding[i])
2168 static int iommu_sva_prepare_bind_data(void __user *udata,
2169 struct iommu_gpasid_bind_data *data)
2174 * No new spaces can be added before the variable sized union, the
2175 * minimum size is the offset to the union.
2177 minsz = offsetof(struct iommu_gpasid_bind_data, vendor);
2179 /* Copy minsz from user to get flags and argsz */
2180 if (copy_from_user(data, udata, minsz))
2183 /* Fields before the variable size union are mandatory */
2184 if (data->argsz < minsz)
2187 * User might be using a newer UAPI header, we shall let IOMMU vendor
2188 * driver decide on what size it needs. Since the guest PASID bind data
2189 * can be vendor specific, larger argsz could be the result of extension
2190 * for one vendor but it should not affect another vendor.
2191 * Copy the remaining user data _after_ minsz
2193 if (copy_from_user((void *)data + minsz, udata + minsz,
2194 min_t(u32, data->argsz, sizeof(*data)) - minsz))
2197 return iommu_check_bind_data(data);
2200 int iommu_uapi_sva_bind_gpasid(struct iommu_domain *domain, struct device *dev,
2203 struct iommu_gpasid_bind_data data = { 0 };
2206 if (unlikely(!domain->ops->sva_bind_gpasid))
2209 ret = iommu_sva_prepare_bind_data(udata, &data);
2213 return domain->ops->sva_bind_gpasid(domain, dev, &data);
2215 EXPORT_SYMBOL_GPL(iommu_uapi_sva_bind_gpasid);
2217 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2220 if (unlikely(!domain->ops->sva_unbind_gpasid))
2223 return domain->ops->sva_unbind_gpasid(dev, pasid);
2225 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
2227 int iommu_uapi_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2230 struct iommu_gpasid_bind_data data = { 0 };
2233 if (unlikely(!domain->ops->sva_bind_gpasid))
2236 ret = iommu_sva_prepare_bind_data(udata, &data);
2240 return iommu_sva_unbind_gpasid(domain, dev, data.hpasid);
2242 EXPORT_SYMBOL_GPL(iommu_uapi_sva_unbind_gpasid);
2244 static void __iommu_detach_device(struct iommu_domain *domain,
2247 if (iommu_is_attach_deferred(domain, dev))
2250 if (unlikely(domain->ops->detach_dev == NULL))
2253 domain->ops->detach_dev(domain, dev);
2254 trace_detach_device_from_domain(dev);
2257 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
2259 struct iommu_group *group;
2261 group = iommu_group_get(dev);
2265 mutex_lock(&group->mutex);
2266 if (iommu_group_device_count(group) != 1) {
2271 __iommu_detach_group(domain, group);
2274 mutex_unlock(&group->mutex);
2275 iommu_group_put(group);
2277 EXPORT_SYMBOL_GPL(iommu_detach_device);
2279 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
2281 struct iommu_domain *domain;
2282 struct iommu_group *group;
2284 group = iommu_group_get(dev);
2288 domain = group->domain;
2290 iommu_group_put(group);
2294 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
2297 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2298 * guarantees that the group and its default domain are valid and correct.
2300 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
2302 return dev->iommu_group->default_domain;
2306 * IOMMU groups are really the natural working unit of the IOMMU, but
2307 * the IOMMU API works on domains and devices. Bridge that gap by
2308 * iterating over the devices in a group. Ideally we'd have a single
2309 * device which represents the requestor ID of the group, but we also
2310 * allow IOMMU drivers to create policy defined minimum sets, where
2311 * the physical hardware may be able to distiguish members, but we
2312 * wish to group them at a higher level (ex. untrusted multi-function
2313 * PCI devices). Thus we attach each device.
2315 static int iommu_group_do_attach_device(struct device *dev, void *data)
2317 struct iommu_domain *domain = data;
2319 return __iommu_attach_device(domain, dev);
2322 static int __iommu_attach_group(struct iommu_domain *domain,
2323 struct iommu_group *group)
2327 if (group->default_domain && group->domain != group->default_domain)
2330 ret = __iommu_group_for_each_dev(group, domain,
2331 iommu_group_do_attach_device);
2333 group->domain = domain;
2338 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
2342 mutex_lock(&group->mutex);
2343 ret = __iommu_attach_group(domain, group);
2344 mutex_unlock(&group->mutex);
2348 EXPORT_SYMBOL_GPL(iommu_attach_group);
2350 static int iommu_group_do_detach_device(struct device *dev, void *data)
2352 struct iommu_domain *domain = data;
2354 __iommu_detach_device(domain, dev);
2359 static void __iommu_detach_group(struct iommu_domain *domain,
2360 struct iommu_group *group)
2364 if (!group->default_domain) {
2365 __iommu_group_for_each_dev(group, domain,
2366 iommu_group_do_detach_device);
2367 group->domain = NULL;
2371 if (group->domain == group->default_domain)
2374 /* Detach by re-attaching to the default domain */
2375 ret = __iommu_group_for_each_dev(group, group->default_domain,
2376 iommu_group_do_attach_device);
2380 group->domain = group->default_domain;
2383 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
2385 mutex_lock(&group->mutex);
2386 __iommu_detach_group(domain, group);
2387 mutex_unlock(&group->mutex);
2389 EXPORT_SYMBOL_GPL(iommu_detach_group);
2391 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
2393 if (domain->type == IOMMU_DOMAIN_IDENTITY)
2396 if (domain->type == IOMMU_DOMAIN_BLOCKED)
2399 return domain->ops->iova_to_phys(domain, iova);
2401 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
2403 static size_t iommu_pgsize(struct iommu_domain *domain, unsigned long iova,
2404 phys_addr_t paddr, size_t size, size_t *count)
2406 unsigned int pgsize_idx, pgsize_idx_next;
2407 unsigned long pgsizes;
2408 size_t offset, pgsize, pgsize_next;
2409 unsigned long addr_merge = paddr | iova;
2411 /* Page sizes supported by the hardware and small enough for @size */
2412 pgsizes = domain->pgsize_bitmap & GENMASK(__fls(size), 0);
2414 /* Constrain the page sizes further based on the maximum alignment */
2415 if (likely(addr_merge))
2416 pgsizes &= GENMASK(__ffs(addr_merge), 0);
2418 /* Make sure we have at least one suitable page size */
2421 /* Pick the biggest page size remaining */
2422 pgsize_idx = __fls(pgsizes);
2423 pgsize = BIT(pgsize_idx);
2427 /* Find the next biggest support page size, if it exists */
2428 pgsizes = domain->pgsize_bitmap & ~GENMASK(pgsize_idx, 0);
2432 pgsize_idx_next = __ffs(pgsizes);
2433 pgsize_next = BIT(pgsize_idx_next);
2436 * There's no point trying a bigger page size unless the virtual
2437 * and physical addresses are similarly offset within the larger page.
2439 if ((iova ^ paddr) & (pgsize_next - 1))
2442 /* Calculate the offset to the next page size alignment boundary */
2443 offset = pgsize_next - (addr_merge & (pgsize_next - 1));
2446 * If size is big enough to accommodate the larger page, reduce
2447 * the number of smaller pages.
2449 if (offset + pgsize_next <= size)
2453 *count = size >> pgsize_idx;
2457 static int __iommu_map_pages(struct iommu_domain *domain, unsigned long iova,
2458 phys_addr_t paddr, size_t size, int prot,
2459 gfp_t gfp, size_t *mapped)
2461 const struct iommu_ops *ops = domain->ops;
2462 size_t pgsize, count;
2465 pgsize = iommu_pgsize(domain, iova, paddr, size, &count);
2467 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx count %zu\n",
2468 iova, &paddr, pgsize, count);
2470 if (ops->map_pages) {
2471 ret = ops->map_pages(domain, iova, paddr, pgsize, count, prot,
2474 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
2475 *mapped = ret ? 0 : pgsize;
2481 static int __iommu_map(struct iommu_domain *domain, unsigned long iova,
2482 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2484 const struct iommu_ops *ops = domain->ops;
2485 unsigned long orig_iova = iova;
2486 unsigned int min_pagesz;
2487 size_t orig_size = size;
2488 phys_addr_t orig_paddr = paddr;
2491 if (unlikely(!(ops->map || ops->map_pages) ||
2492 domain->pgsize_bitmap == 0UL))
2495 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2498 /* find out the minimum page size supported */
2499 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2502 * both the virtual address and the physical one, as well as
2503 * the size of the mapping, must be aligned (at least) to the
2504 * size of the smallest page supported by the hardware
2506 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
2507 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2508 iova, &paddr, size, min_pagesz);
2512 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
2517 ret = __iommu_map_pages(domain, iova, paddr, size, prot, gfp,
2520 * Some pages may have been mapped, even if an error occurred,
2521 * so we should account for those so they can be unmapped.
2532 /* unroll mapping in case something went wrong */
2534 iommu_unmap(domain, orig_iova, orig_size - size);
2536 trace_map(orig_iova, orig_paddr, orig_size);
2541 static int _iommu_map(struct iommu_domain *domain, unsigned long iova,
2542 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2544 const struct iommu_ops *ops = domain->ops;
2547 ret = __iommu_map(domain, iova, paddr, size, prot, gfp);
2548 if (ret == 0 && ops->iotlb_sync_map)
2549 ops->iotlb_sync_map(domain, iova, size);
2554 int iommu_map(struct iommu_domain *domain, unsigned long iova,
2555 phys_addr_t paddr, size_t size, int prot)
2558 return _iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
2560 EXPORT_SYMBOL_GPL(iommu_map);
2562 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
2563 phys_addr_t paddr, size_t size, int prot)
2565 return _iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
2567 EXPORT_SYMBOL_GPL(iommu_map_atomic);
2569 static size_t __iommu_unmap_pages(struct iommu_domain *domain,
2570 unsigned long iova, size_t size,
2571 struct iommu_iotlb_gather *iotlb_gather)
2573 const struct iommu_ops *ops = domain->ops;
2574 size_t pgsize, count;
2576 pgsize = iommu_pgsize(domain, iova, iova, size, &count);
2577 return ops->unmap_pages ?
2578 ops->unmap_pages(domain, iova, pgsize, count, iotlb_gather) :
2579 ops->unmap(domain, iova, pgsize, iotlb_gather);
2582 static size_t __iommu_unmap(struct iommu_domain *domain,
2583 unsigned long iova, size_t size,
2584 struct iommu_iotlb_gather *iotlb_gather)
2586 const struct iommu_ops *ops = domain->ops;
2587 size_t unmapped_page, unmapped = 0;
2588 unsigned long orig_iova = iova;
2589 unsigned int min_pagesz;
2591 if (unlikely(!(ops->unmap || ops->unmap_pages) ||
2592 domain->pgsize_bitmap == 0UL))
2595 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2598 /* find out the minimum page size supported */
2599 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2602 * The virtual address, as well as the size of the mapping, must be
2603 * aligned (at least) to the size of the smallest page supported
2606 if (!IS_ALIGNED(iova | size, min_pagesz)) {
2607 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2608 iova, size, min_pagesz);
2612 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2615 * Keep iterating until we either unmap 'size' bytes (or more)
2616 * or we hit an area that isn't mapped.
2618 while (unmapped < size) {
2619 unmapped_page = __iommu_unmap_pages(domain, iova,
2625 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2626 iova, unmapped_page);
2628 iova += unmapped_page;
2629 unmapped += unmapped_page;
2632 trace_unmap(orig_iova, size, unmapped);
2636 size_t iommu_unmap(struct iommu_domain *domain,
2637 unsigned long iova, size_t size)
2639 struct iommu_iotlb_gather iotlb_gather;
2642 iommu_iotlb_gather_init(&iotlb_gather);
2643 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2644 iommu_iotlb_sync(domain, &iotlb_gather);
2648 EXPORT_SYMBOL_GPL(iommu_unmap);
2650 size_t iommu_unmap_fast(struct iommu_domain *domain,
2651 unsigned long iova, size_t size,
2652 struct iommu_iotlb_gather *iotlb_gather)
2654 return __iommu_unmap(domain, iova, size, iotlb_gather);
2656 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2658 static ssize_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2659 struct scatterlist *sg, unsigned int nents, int prot,
2662 const struct iommu_ops *ops = domain->ops;
2663 size_t len = 0, mapped = 0;
2668 while (i <= nents) {
2669 phys_addr_t s_phys = sg_phys(sg);
2671 if (len && s_phys != start + len) {
2672 ret = __iommu_map(domain, iova + mapped, start,
2693 if (ops->iotlb_sync_map)
2694 ops->iotlb_sync_map(domain, iova, mapped);
2698 /* undo mappings already done */
2699 iommu_unmap(domain, iova, mapped);
2704 ssize_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2705 struct scatterlist *sg, unsigned int nents, int prot)
2708 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2710 EXPORT_SYMBOL_GPL(iommu_map_sg);
2712 ssize_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2713 struct scatterlist *sg, unsigned int nents, int prot)
2715 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2719 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2720 * @domain: the iommu domain where the fault has happened
2721 * @dev: the device where the fault has happened
2722 * @iova: the faulting address
2723 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2725 * This function should be called by the low-level IOMMU implementations
2726 * whenever IOMMU faults happen, to allow high-level users, that are
2727 * interested in such events, to know about them.
2729 * This event may be useful for several possible use cases:
2730 * - mere logging of the event
2731 * - dynamic TLB/PTE loading
2732 * - if restarting of the faulting device is required
2734 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2735 * PTE/TLB loading will one day be supported, implementations will be able
2736 * to tell whether it succeeded or not according to this return value).
2738 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2739 * (though fault handlers can also return -ENOSYS, in case they want to
2740 * elicit the default behavior of the IOMMU drivers).
2742 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2743 unsigned long iova, int flags)
2748 * if upper layers showed interest and installed a fault handler,
2751 if (domain->handler)
2752 ret = domain->handler(domain, dev, iova, flags,
2753 domain->handler_token);
2755 trace_io_page_fault(dev, iova, flags);
2758 EXPORT_SYMBOL_GPL(report_iommu_fault);
2760 static int __init iommu_init(void)
2762 iommu_group_kset = kset_create_and_add("iommu_groups",
2764 BUG_ON(!iommu_group_kset);
2766 iommu_debugfs_setup();
2770 core_initcall(iommu_init);
2772 int iommu_enable_nesting(struct iommu_domain *domain)
2774 if (domain->type != IOMMU_DOMAIN_UNMANAGED)
2776 if (!domain->ops->enable_nesting)
2778 return domain->ops->enable_nesting(domain);
2780 EXPORT_SYMBOL_GPL(iommu_enable_nesting);
2782 int iommu_set_pgtable_quirks(struct iommu_domain *domain,
2783 unsigned long quirk)
2785 if (domain->type != IOMMU_DOMAIN_UNMANAGED)
2787 if (!domain->ops->set_pgtable_quirks)
2789 return domain->ops->set_pgtable_quirks(domain, quirk);
2791 EXPORT_SYMBOL_GPL(iommu_set_pgtable_quirks);
2793 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2795 const struct iommu_ops *ops = dev->bus->iommu_ops;
2797 if (ops && ops->get_resv_regions)
2798 ops->get_resv_regions(dev, list);
2801 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2803 const struct iommu_ops *ops = dev->bus->iommu_ops;
2805 if (ops && ops->put_resv_regions)
2806 ops->put_resv_regions(dev, list);
2810 * generic_iommu_put_resv_regions - Reserved region driver helper
2811 * @dev: device for which to free reserved regions
2812 * @list: reserved region list for device
2814 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2815 * for simple reservations. Memory allocated for each reserved region will be
2816 * freed. If an IOMMU driver allocates additional resources per region, it is
2817 * going to have to implement a custom callback.
2819 void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2821 struct iommu_resv_region *entry, *next;
2823 list_for_each_entry_safe(entry, next, list, list)
2826 EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2828 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2829 size_t length, int prot,
2830 enum iommu_resv_type type)
2832 struct iommu_resv_region *region;
2834 region = kzalloc(sizeof(*region), GFP_KERNEL);
2838 INIT_LIST_HEAD(®ion->list);
2839 region->start = start;
2840 region->length = length;
2841 region->prot = prot;
2842 region->type = type;
2845 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2847 void iommu_set_default_passthrough(bool cmd_line)
2850 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
2851 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2854 void iommu_set_default_translated(bool cmd_line)
2857 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
2858 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2861 bool iommu_default_passthrough(void)
2863 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2865 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2867 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2869 const struct iommu_ops *ops = NULL;
2870 struct iommu_device *iommu;
2872 spin_lock(&iommu_device_lock);
2873 list_for_each_entry(iommu, &iommu_device_list, list)
2874 if (iommu->fwnode == fwnode) {
2878 spin_unlock(&iommu_device_lock);
2882 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2883 const struct iommu_ops *ops)
2885 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2888 return ops == fwspec->ops ? 0 : -EINVAL;
2890 if (!dev_iommu_get(dev))
2893 /* Preallocate for the overwhelmingly common case of 1 ID */
2894 fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2898 of_node_get(to_of_node(iommu_fwnode));
2899 fwspec->iommu_fwnode = iommu_fwnode;
2901 dev_iommu_fwspec_set(dev, fwspec);
2904 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2906 void iommu_fwspec_free(struct device *dev)
2908 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2911 fwnode_handle_put(fwspec->iommu_fwnode);
2913 dev_iommu_fwspec_set(dev, NULL);
2916 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2918 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2920 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2926 new_num = fwspec->num_ids + num_ids;
2928 fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2933 dev_iommu_fwspec_set(dev, fwspec);
2936 for (i = 0; i < num_ids; i++)
2937 fwspec->ids[fwspec->num_ids + i] = ids[i];
2939 fwspec->num_ids = new_num;
2942 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2945 * Per device IOMMU features.
2947 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2949 if (dev->iommu && dev->iommu->iommu_dev) {
2950 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2952 if (ops->dev_enable_feat)
2953 return ops->dev_enable_feat(dev, feat);
2958 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2961 * The device drivers should do the necessary cleanups before calling this.
2962 * For example, before disabling the aux-domain feature, the device driver
2963 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2965 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2967 if (dev->iommu && dev->iommu->iommu_dev) {
2968 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2970 if (ops->dev_disable_feat)
2971 return ops->dev_disable_feat(dev, feat);
2976 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2978 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2980 if (dev->iommu && dev->iommu->iommu_dev) {
2981 const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2983 if (ops->dev_feat_enabled)
2984 return ops->dev_feat_enabled(dev, feat);
2989 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2992 * Aux-domain specific attach/detach.
2994 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2995 * true. Also, as long as domains are attached to a device through this
2996 * interface, any tries to call iommu_attach_device() should fail
2997 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2998 * This should make us safe against a device being attached to a guest as a
2999 * whole while there are still pasid users on it (aux and sva).
3001 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
3005 if (domain->ops->aux_attach_dev)
3006 ret = domain->ops->aux_attach_dev(domain, dev);
3009 trace_attach_device_to_domain(dev);
3013 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
3015 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
3017 if (domain->ops->aux_detach_dev) {
3018 domain->ops->aux_detach_dev(domain, dev);
3019 trace_detach_device_from_domain(dev);
3022 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
3024 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
3028 if (domain->ops->aux_get_pasid)
3029 ret = domain->ops->aux_get_pasid(domain, dev);
3033 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
3036 * iommu_sva_bind_device() - Bind a process address space to a device
3038 * @mm: the mm to bind, caller must hold a reference to it
3039 * @drvdata: opaque data pointer to pass to bind callback
3041 * Create a bond between device and address space, allowing the device to access
3042 * the mm using the returned PASID. If a bond already exists between @device and
3043 * @mm, it is returned and an additional reference is taken. Caller must call
3044 * iommu_sva_unbind_device() to release each reference.
3046 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
3047 * initialize the required SVA features.
3049 * On error, returns an ERR_PTR value.
3052 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
3054 struct iommu_group *group;
3055 struct iommu_sva *handle = ERR_PTR(-EINVAL);
3056 const struct iommu_ops *ops = dev->bus->iommu_ops;
3058 if (!ops || !ops->sva_bind)
3059 return ERR_PTR(-ENODEV);
3061 group = iommu_group_get(dev);
3063 return ERR_PTR(-ENODEV);
3065 /* Ensure device count and domain don't change while we're binding */
3066 mutex_lock(&group->mutex);
3069 * To keep things simple, SVA currently doesn't support IOMMU groups
3070 * with more than one device. Existing SVA-capable systems are not
3071 * affected by the problems that required IOMMU groups (lack of ACS
3072 * isolation, device ID aliasing and other hardware issues).
3074 if (iommu_group_device_count(group) != 1)
3077 handle = ops->sva_bind(dev, mm, drvdata);
3080 mutex_unlock(&group->mutex);
3081 iommu_group_put(group);
3085 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
3088 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
3089 * @handle: the handle returned by iommu_sva_bind_device()
3091 * Put reference to a bond between device and address space. The device should
3092 * not be issuing any more transaction for this PASID. All outstanding page
3093 * requests for this PASID must have been flushed to the IOMMU.
3095 void iommu_sva_unbind_device(struct iommu_sva *handle)
3097 struct iommu_group *group;
3098 struct device *dev = handle->dev;
3099 const struct iommu_ops *ops = dev->bus->iommu_ops;
3101 if (!ops || !ops->sva_unbind)
3104 group = iommu_group_get(dev);
3108 mutex_lock(&group->mutex);
3109 ops->sva_unbind(handle);
3110 mutex_unlock(&group->mutex);
3112 iommu_group_put(group);
3114 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
3116 u32 iommu_sva_get_pasid(struct iommu_sva *handle)
3118 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
3120 if (!ops || !ops->sva_get_pasid)
3121 return IOMMU_PASID_INVALID;
3123 return ops->sva_get_pasid(handle);
3125 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
3128 * Changes the default domain of an iommu group that has *only* one device
3130 * @group: The group for which the default domain should be changed
3131 * @prev_dev: The device in the group (this is used to make sure that the device
3132 * hasn't changed after the caller has called this function)
3133 * @type: The type of the new default domain that gets associated with the group
3135 * Returns 0 on success and error code on failure
3138 * 1. Presently, this function is called only when user requests to change the
3139 * group's default domain type through /sys/kernel/iommu_groups/<grp_id>/type
3140 * Please take a closer look if intended to use for other purposes.
3142 static int iommu_change_dev_def_domain(struct iommu_group *group,
3143 struct device *prev_dev, int type)
3145 struct iommu_domain *prev_dom;
3146 struct group_device *grp_dev;
3147 int ret, dev_def_dom;
3150 mutex_lock(&group->mutex);
3152 if (group->default_domain != group->domain) {
3153 dev_err_ratelimited(prev_dev, "Group not assigned to default domain\n");
3159 * iommu group wasn't locked while acquiring device lock in
3160 * iommu_group_store_type(). So, make sure that the device count hasn't
3161 * changed while acquiring device lock.
3163 * Changing default domain of an iommu group with two or more devices
3164 * isn't supported because there could be a potential deadlock. Consider
3165 * the following scenario. T1 is trying to acquire device locks of all
3166 * the devices in the group and before it could acquire all of them,
3167 * there could be another thread T2 (from different sub-system and use
3168 * case) that has already acquired some of the device locks and might be
3169 * waiting for T1 to release other device locks.
3171 if (iommu_group_device_count(group) != 1) {
3172 dev_err_ratelimited(prev_dev, "Cannot change default domain: Group has more than one device\n");
3177 /* Since group has only one device */
3178 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3181 if (prev_dev != dev) {
3182 dev_err_ratelimited(prev_dev, "Cannot change default domain: Device has been changed\n");
3187 prev_dom = group->default_domain;
3193 dev_def_dom = iommu_get_def_domain_type(dev);
3196 * If the user hasn't requested any specific type of domain and
3197 * if the device supports both the domains, then default to the
3198 * domain the device was booted with
3200 type = dev_def_dom ? : iommu_def_domain_type;
3201 } else if (dev_def_dom && type != dev_def_dom) {
3202 dev_err_ratelimited(prev_dev, "Device cannot be in %s domain\n",
3203 iommu_domain_type_str(type));
3209 * Switch to a new domain only if the requested domain type is different
3210 * from the existing default domain type
3212 if (prev_dom->type == type) {
3217 /* We can bring up a flush queue without tearing down the domain */
3218 if (type == IOMMU_DOMAIN_DMA_FQ && prev_dom->type == IOMMU_DOMAIN_DMA) {
3219 ret = iommu_dma_init_fq(prev_dom);
3221 prev_dom->type = IOMMU_DOMAIN_DMA_FQ;
3225 /* Sets group->default_domain to the newly allocated domain */
3226 ret = iommu_group_alloc_default_domain(dev->bus, group, type);
3230 ret = iommu_create_device_direct_mappings(group, dev);
3232 goto free_new_domain;
3234 ret = __iommu_attach_device(group->default_domain, dev);
3236 goto free_new_domain;
3238 group->domain = group->default_domain;
3241 * Release the mutex here because ops->probe_finalize() call-back of
3242 * some vendor IOMMU drivers calls arm_iommu_attach_device() which
3243 * in-turn might call back into IOMMU core code, where it tries to take
3244 * group->mutex, resulting in a deadlock.
3246 mutex_unlock(&group->mutex);
3248 /* Make sure dma_ops is appropriatley set */
3249 iommu_group_do_probe_finalize(dev, group->default_domain);
3250 iommu_domain_free(prev_dom);
3254 iommu_domain_free(group->default_domain);
3255 group->default_domain = prev_dom;
3256 group->domain = prev_dom;
3259 mutex_unlock(&group->mutex);
3265 * Changing the default domain through sysfs requires the users to unbind the
3266 * drivers from the devices in the iommu group, except for a DMA -> DMA-FQ
3267 * transition. Return failure if this isn't met.
3269 * We need to consider the race between this and the device release path.
3270 * device_lock(dev) is used here to guarantee that the device release path
3271 * will not be entered at the same time.
3273 static ssize_t iommu_group_store_type(struct iommu_group *group,
3274 const char *buf, size_t count)
3276 struct group_device *grp_dev;
3280 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
3283 if (WARN_ON(!group))
3286 if (sysfs_streq(buf, "identity"))
3287 req_type = IOMMU_DOMAIN_IDENTITY;
3288 else if (sysfs_streq(buf, "DMA"))
3289 req_type = IOMMU_DOMAIN_DMA;
3290 else if (sysfs_streq(buf, "DMA-FQ"))
3291 req_type = IOMMU_DOMAIN_DMA_FQ;
3292 else if (sysfs_streq(buf, "auto"))
3298 * Lock/Unlock the group mutex here before device lock to
3299 * 1. Make sure that the iommu group has only one device (this is a
3300 * prerequisite for step 2)
3301 * 2. Get struct *dev which is needed to lock device
3303 mutex_lock(&group->mutex);
3304 if (iommu_group_device_count(group) != 1) {
3305 mutex_unlock(&group->mutex);
3306 pr_err_ratelimited("Cannot change default domain: Group has more than one device\n");
3310 /* Since group has only one device */
3311 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3316 * Don't hold the group mutex because taking group mutex first and then
3317 * the device lock could potentially cause a deadlock as below. Assume
3318 * two threads T1 and T2. T1 is trying to change default domain of an
3319 * iommu group and T2 is trying to hot unplug a device or release [1] VF
3320 * of a PCIe device which is in the same iommu group. T1 takes group
3321 * mutex and before it could take device lock assume T2 has taken device
3322 * lock and is yet to take group mutex. Now, both the threads will be
3323 * waiting for the other thread to release lock. Below, lock order was
3326 * mutex_lock(&group->mutex);
3327 * iommu_change_dev_def_domain();
3328 * mutex_unlock(&group->mutex);
3329 * device_unlock(dev);
3331 * [1] Typical device release path
3332 * device_lock() from device/driver core code
3334 * -> iommu_bus_notifier()
3335 * -> iommu_release_device()
3336 * -> ops->release_device() vendor driver calls back iommu core code
3337 * -> mutex_lock() from iommu core code
3339 mutex_unlock(&group->mutex);
3341 /* Check if the device in the group still has a driver bound to it */
3343 if (device_is_bound(dev) && !(req_type == IOMMU_DOMAIN_DMA_FQ &&
3344 group->default_domain->type == IOMMU_DOMAIN_DMA)) {
3345 pr_err_ratelimited("Device is still bound to driver\n");
3350 ret = iommu_change_dev_def_domain(group, dev, req_type);