1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
4 * Author: Joerg Roedel <jroedel@suse.de>
7 #define pr_fmt(fmt) "iommu: " fmt
9 #include <linux/device.h>
10 #include <linux/kernel.h>
11 #include <linux/bug.h>
12 #include <linux/types.h>
13 #include <linux/init.h>
14 #include <linux/export.h>
15 #include <linux/slab.h>
16 #include <linux/errno.h>
17 #include <linux/iommu.h>
18 #include <linux/idr.h>
19 #include <linux/notifier.h>
20 #include <linux/err.h>
21 #include <linux/pci.h>
22 #include <linux/bitops.h>
23 #include <linux/property.h>
24 #include <linux/fsl/mc.h>
25 #include <linux/module.h>
26 #include <trace/events/iommu.h>
28 static struct kset *iommu_group_kset;
29 static DEFINE_IDA(iommu_group_ida);
31 static unsigned int iommu_def_domain_type __read_mostly;
32 static bool iommu_dma_strict __read_mostly = true;
33 static u32 iommu_cmd_line __read_mostly;
37 struct kobject *devices_kobj;
38 struct list_head devices;
40 struct blocking_notifier_head notifier;
42 void (*iommu_data_release)(void *iommu_data);
45 struct iommu_domain *default_domain;
46 struct iommu_domain *domain;
47 struct list_head entry;
51 struct list_head list;
56 struct iommu_group_attribute {
57 struct attribute attr;
58 ssize_t (*show)(struct iommu_group *group, char *buf);
59 ssize_t (*store)(struct iommu_group *group,
60 const char *buf, size_t count);
63 static const char * const iommu_group_resv_type_string[] = {
64 [IOMMU_RESV_DIRECT] = "direct",
65 [IOMMU_RESV_DIRECT_RELAXABLE] = "direct-relaxable",
66 [IOMMU_RESV_RESERVED] = "reserved",
67 [IOMMU_RESV_MSI] = "msi",
68 [IOMMU_RESV_SW_MSI] = "msi",
71 #define IOMMU_CMD_LINE_DMA_API BIT(0)
73 static void iommu_set_cmd_line_dma_api(void)
75 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
78 static bool iommu_cmd_line_dma_api(void)
80 return !!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API);
83 static int iommu_alloc_default_domain(struct iommu_group *group,
85 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
87 static int __iommu_attach_device(struct iommu_domain *domain,
89 static int __iommu_attach_group(struct iommu_domain *domain,
90 struct iommu_group *group);
91 static void __iommu_detach_group(struct iommu_domain *domain,
92 struct iommu_group *group);
93 static int iommu_create_device_direct_mappings(struct iommu_group *group,
95 static struct iommu_group *iommu_group_get_for_dev(struct device *dev);
97 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
98 struct iommu_group_attribute iommu_group_attr_##_name = \
99 __ATTR(_name, _mode, _show, _store)
101 #define to_iommu_group_attr(_attr) \
102 container_of(_attr, struct iommu_group_attribute, attr)
103 #define to_iommu_group(_kobj) \
104 container_of(_kobj, struct iommu_group, kobj)
106 static LIST_HEAD(iommu_device_list);
107 static DEFINE_SPINLOCK(iommu_device_lock);
110 * Use a function instead of an array here because the domain-type is a
111 * bit-field, so an array would waste memory.
113 static const char *iommu_domain_type_str(unsigned int t)
116 case IOMMU_DOMAIN_BLOCKED:
118 case IOMMU_DOMAIN_IDENTITY:
119 return "Passthrough";
120 case IOMMU_DOMAIN_UNMANAGED:
122 case IOMMU_DOMAIN_DMA:
129 static int __init iommu_subsys_init(void)
131 bool cmd_line = iommu_cmd_line_dma_api();
134 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
135 iommu_set_default_passthrough(false);
137 iommu_set_default_translated(false);
139 if (iommu_default_passthrough() && mem_encrypt_active()) {
140 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
141 iommu_set_default_translated(false);
145 pr_info("Default domain type: %s %s\n",
146 iommu_domain_type_str(iommu_def_domain_type),
147 cmd_line ? "(set via kernel command line)" : "");
151 subsys_initcall(iommu_subsys_init);
153 int iommu_device_register(struct iommu_device *iommu)
155 spin_lock(&iommu_device_lock);
156 list_add_tail(&iommu->list, &iommu_device_list);
157 spin_unlock(&iommu_device_lock);
160 EXPORT_SYMBOL_GPL(iommu_device_register);
162 void iommu_device_unregister(struct iommu_device *iommu)
164 spin_lock(&iommu_device_lock);
165 list_del(&iommu->list);
166 spin_unlock(&iommu_device_lock);
168 EXPORT_SYMBOL_GPL(iommu_device_unregister);
170 static struct dev_iommu *dev_iommu_get(struct device *dev)
172 struct dev_iommu *param = dev->iommu;
177 param = kzalloc(sizeof(*param), GFP_KERNEL);
181 mutex_init(¶m->lock);
186 static void dev_iommu_free(struct device *dev)
188 iommu_fwspec_free(dev);
193 static int __iommu_probe_device(struct device *dev, struct list_head *group_list)
195 const struct iommu_ops *ops = dev->bus->iommu_ops;
196 struct iommu_device *iommu_dev;
197 struct iommu_group *group;
203 if (!dev_iommu_get(dev))
206 if (!try_module_get(ops->owner)) {
211 iommu_dev = ops->probe_device(dev);
212 if (IS_ERR(iommu_dev)) {
213 ret = PTR_ERR(iommu_dev);
217 dev->iommu->iommu_dev = iommu_dev;
219 group = iommu_group_get_for_dev(dev);
221 ret = PTR_ERR(group);
224 iommu_group_put(group);
226 if (group_list && !group->default_domain && list_empty(&group->entry))
227 list_add_tail(&group->entry, group_list);
229 iommu_device_link(iommu_dev, dev);
234 ops->release_device(dev);
237 module_put(ops->owner);
245 int iommu_probe_device(struct device *dev)
247 const struct iommu_ops *ops = dev->bus->iommu_ops;
248 struct iommu_group *group;
251 ret = __iommu_probe_device(dev, NULL);
255 group = iommu_group_get(dev);
260 * Try to allocate a default domain - needs support from the
261 * IOMMU driver. There are still some drivers which don't
262 * support default domains, so the return value is not yet
265 iommu_alloc_default_domain(group, dev);
267 if (group->default_domain)
268 ret = __iommu_attach_device(group->default_domain, dev);
270 iommu_create_device_direct_mappings(group, dev);
272 iommu_group_put(group);
277 if (ops->probe_finalize)
278 ops->probe_finalize(dev);
283 iommu_release_device(dev);
290 void iommu_release_device(struct device *dev)
292 const struct iommu_ops *ops = dev->bus->iommu_ops;
297 iommu_device_unlink(dev->iommu->iommu_dev, dev);
298 iommu_group_remove_device(dev);
300 ops->release_device(dev);
302 module_put(ops->owner);
306 static int __init iommu_set_def_domain_type(char *str)
311 ret = kstrtobool(str, &pt);
316 iommu_set_default_passthrough(true);
318 iommu_set_default_translated(true);
322 early_param("iommu.passthrough", iommu_set_def_domain_type);
324 static int __init iommu_dma_setup(char *str)
326 return kstrtobool(str, &iommu_dma_strict);
328 early_param("iommu.strict", iommu_dma_setup);
330 static ssize_t iommu_group_attr_show(struct kobject *kobj,
331 struct attribute *__attr, char *buf)
333 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
334 struct iommu_group *group = to_iommu_group(kobj);
338 ret = attr->show(group, buf);
342 static ssize_t iommu_group_attr_store(struct kobject *kobj,
343 struct attribute *__attr,
344 const char *buf, size_t count)
346 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
347 struct iommu_group *group = to_iommu_group(kobj);
351 ret = attr->store(group, buf, count);
355 static const struct sysfs_ops iommu_group_sysfs_ops = {
356 .show = iommu_group_attr_show,
357 .store = iommu_group_attr_store,
360 static int iommu_group_create_file(struct iommu_group *group,
361 struct iommu_group_attribute *attr)
363 return sysfs_create_file(&group->kobj, &attr->attr);
366 static void iommu_group_remove_file(struct iommu_group *group,
367 struct iommu_group_attribute *attr)
369 sysfs_remove_file(&group->kobj, &attr->attr);
372 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
374 return sprintf(buf, "%s\n", group->name);
378 * iommu_insert_resv_region - Insert a new region in the
379 * list of reserved regions.
380 * @new: new region to insert
381 * @regions: list of regions
383 * Elements are sorted by start address and overlapping segments
384 * of the same type are merged.
386 int iommu_insert_resv_region(struct iommu_resv_region *new,
387 struct list_head *regions)
389 struct iommu_resv_region *iter, *tmp, *nr, *top;
392 nr = iommu_alloc_resv_region(new->start, new->length,
393 new->prot, new->type);
397 /* First add the new element based on start address sorting */
398 list_for_each_entry(iter, regions, list) {
399 if (nr->start < iter->start ||
400 (nr->start == iter->start && nr->type <= iter->type))
403 list_add_tail(&nr->list, &iter->list);
405 /* Merge overlapping segments of type nr->type in @regions, if any */
406 list_for_each_entry_safe(iter, tmp, regions, list) {
407 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
409 /* no merge needed on elements of different types than @new */
410 if (iter->type != new->type) {
411 list_move_tail(&iter->list, &stack);
415 /* look for the last stack element of same type as @iter */
416 list_for_each_entry_reverse(top, &stack, list)
417 if (top->type == iter->type)
420 list_move_tail(&iter->list, &stack);
424 top_end = top->start + top->length - 1;
426 if (iter->start > top_end + 1) {
427 list_move_tail(&iter->list, &stack);
429 top->length = max(top_end, iter_end) - top->start + 1;
430 list_del(&iter->list);
434 list_splice(&stack, regions);
439 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
440 struct list_head *group_resv_regions)
442 struct iommu_resv_region *entry;
445 list_for_each_entry(entry, dev_resv_regions, list) {
446 ret = iommu_insert_resv_region(entry, group_resv_regions);
453 int iommu_get_group_resv_regions(struct iommu_group *group,
454 struct list_head *head)
456 struct group_device *device;
459 mutex_lock(&group->mutex);
460 list_for_each_entry(device, &group->devices, list) {
461 struct list_head dev_resv_regions;
463 INIT_LIST_HEAD(&dev_resv_regions);
464 iommu_get_resv_regions(device->dev, &dev_resv_regions);
465 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
466 iommu_put_resv_regions(device->dev, &dev_resv_regions);
470 mutex_unlock(&group->mutex);
473 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
475 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
478 struct iommu_resv_region *region, *next;
479 struct list_head group_resv_regions;
482 INIT_LIST_HEAD(&group_resv_regions);
483 iommu_get_group_resv_regions(group, &group_resv_regions);
485 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
486 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
487 (long long int)region->start,
488 (long long int)(region->start +
490 iommu_group_resv_type_string[region->type]);
497 static ssize_t iommu_group_show_type(struct iommu_group *group,
500 char *type = "unknown\n";
502 if (group->default_domain) {
503 switch (group->default_domain->type) {
504 case IOMMU_DOMAIN_BLOCKED:
507 case IOMMU_DOMAIN_IDENTITY:
510 case IOMMU_DOMAIN_UNMANAGED:
511 type = "unmanaged\n";
513 case IOMMU_DOMAIN_DMA:
523 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
525 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
526 iommu_group_show_resv_regions, NULL);
528 static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
530 static void iommu_group_release(struct kobject *kobj)
532 struct iommu_group *group = to_iommu_group(kobj);
534 pr_debug("Releasing group %d\n", group->id);
536 if (group->iommu_data_release)
537 group->iommu_data_release(group->iommu_data);
539 ida_simple_remove(&iommu_group_ida, group->id);
541 if (group->default_domain)
542 iommu_domain_free(group->default_domain);
548 static struct kobj_type iommu_group_ktype = {
549 .sysfs_ops = &iommu_group_sysfs_ops,
550 .release = iommu_group_release,
554 * iommu_group_alloc - Allocate a new group
556 * This function is called by an iommu driver to allocate a new iommu
557 * group. The iommu group represents the minimum granularity of the iommu.
558 * Upon successful return, the caller holds a reference to the supplied
559 * group in order to hold the group until devices are added. Use
560 * iommu_group_put() to release this extra reference count, allowing the
561 * group to be automatically reclaimed once it has no devices or external
564 struct iommu_group *iommu_group_alloc(void)
566 struct iommu_group *group;
569 group = kzalloc(sizeof(*group), GFP_KERNEL);
571 return ERR_PTR(-ENOMEM);
573 group->kobj.kset = iommu_group_kset;
574 mutex_init(&group->mutex);
575 INIT_LIST_HEAD(&group->devices);
576 INIT_LIST_HEAD(&group->entry);
577 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
579 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
586 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
587 NULL, "%d", group->id);
589 ida_simple_remove(&iommu_group_ida, group->id);
590 kobject_put(&group->kobj);
594 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
595 if (!group->devices_kobj) {
596 kobject_put(&group->kobj); /* triggers .release & free */
597 return ERR_PTR(-ENOMEM);
601 * The devices_kobj holds a reference on the group kobject, so
602 * as long as that exists so will the group. We can therefore
603 * use the devices_kobj for reference counting.
605 kobject_put(&group->kobj);
607 ret = iommu_group_create_file(group,
608 &iommu_group_attr_reserved_regions);
612 ret = iommu_group_create_file(group, &iommu_group_attr_type);
616 pr_debug("Allocated group %d\n", group->id);
620 EXPORT_SYMBOL_GPL(iommu_group_alloc);
622 struct iommu_group *iommu_group_get_by_id(int id)
624 struct kobject *group_kobj;
625 struct iommu_group *group;
628 if (!iommu_group_kset)
631 name = kasprintf(GFP_KERNEL, "%d", id);
635 group_kobj = kset_find_obj(iommu_group_kset, name);
641 group = container_of(group_kobj, struct iommu_group, kobj);
642 BUG_ON(group->id != id);
644 kobject_get(group->devices_kobj);
645 kobject_put(&group->kobj);
649 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
652 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
655 * iommu drivers can store data in the group for use when doing iommu
656 * operations. This function provides a way to retrieve it. Caller
657 * should hold a group reference.
659 void *iommu_group_get_iommudata(struct iommu_group *group)
661 return group->iommu_data;
663 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
666 * iommu_group_set_iommudata - set iommu_data for a group
668 * @iommu_data: new data
669 * @release: release function for iommu_data
671 * iommu drivers can store data in the group for use when doing iommu
672 * operations. This function provides a way to set the data after
673 * the group has been allocated. Caller should hold a group reference.
675 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
676 void (*release)(void *iommu_data))
678 group->iommu_data = iommu_data;
679 group->iommu_data_release = release;
681 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
684 * iommu_group_set_name - set name for a group
688 * Allow iommu driver to set a name for a group. When set it will
689 * appear in a name attribute file under the group in sysfs.
691 int iommu_group_set_name(struct iommu_group *group, const char *name)
696 iommu_group_remove_file(group, &iommu_group_attr_name);
703 group->name = kstrdup(name, GFP_KERNEL);
707 ret = iommu_group_create_file(group, &iommu_group_attr_name);
716 EXPORT_SYMBOL_GPL(iommu_group_set_name);
718 static int iommu_create_device_direct_mappings(struct iommu_group *group,
721 struct iommu_domain *domain = group->default_domain;
722 struct iommu_resv_region *entry;
723 struct list_head mappings;
724 unsigned long pg_size;
727 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
730 BUG_ON(!domain->pgsize_bitmap);
732 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
733 INIT_LIST_HEAD(&mappings);
735 iommu_get_resv_regions(dev, &mappings);
737 /* We need to consider overlapping regions for different devices */
738 list_for_each_entry(entry, &mappings, list) {
739 dma_addr_t start, end, addr;
741 if (domain->ops->apply_resv_region)
742 domain->ops->apply_resv_region(dev, domain, entry);
744 start = ALIGN(entry->start, pg_size);
745 end = ALIGN(entry->start + entry->length, pg_size);
747 if (entry->type != IOMMU_RESV_DIRECT &&
748 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
751 for (addr = start; addr < end; addr += pg_size) {
752 phys_addr_t phys_addr;
754 phys_addr = iommu_iova_to_phys(domain, addr);
758 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
765 iommu_flush_tlb_all(domain);
768 iommu_put_resv_regions(dev, &mappings);
773 static bool iommu_is_attach_deferred(struct iommu_domain *domain,
776 if (domain->ops->is_attach_deferred)
777 return domain->ops->is_attach_deferred(domain, dev);
783 * iommu_group_add_device - add a device to an iommu group
784 * @group: the group into which to add the device (reference should be held)
787 * This function is called by an iommu driver to add a device into a
788 * group. Adding a device increments the group reference count.
790 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
793 struct group_device *device;
795 device = kzalloc(sizeof(*device), GFP_KERNEL);
801 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
803 goto err_free_device;
805 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
809 goto err_remove_link;
812 ret = sysfs_create_link_nowarn(group->devices_kobj,
813 &dev->kobj, device->name);
815 if (ret == -EEXIST && i >= 0) {
817 * Account for the slim chance of collision
818 * and append an instance to the name.
821 device->name = kasprintf(GFP_KERNEL, "%s.%d",
822 kobject_name(&dev->kobj), i++);
828 kobject_get(group->devices_kobj);
830 dev->iommu_group = group;
832 mutex_lock(&group->mutex);
833 list_add_tail(&device->list, &group->devices);
834 if (group->domain && !iommu_is_attach_deferred(group->domain, dev))
835 ret = __iommu_attach_device(group->domain, dev);
836 mutex_unlock(&group->mutex);
840 /* Notify any listeners about change to group. */
841 blocking_notifier_call_chain(&group->notifier,
842 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
844 trace_add_device_to_group(group->id, dev);
846 dev_info(dev, "Adding to iommu group %d\n", group->id);
851 mutex_lock(&group->mutex);
852 list_del(&device->list);
853 mutex_unlock(&group->mutex);
854 dev->iommu_group = NULL;
855 kobject_put(group->devices_kobj);
856 sysfs_remove_link(group->devices_kobj, device->name);
860 sysfs_remove_link(&dev->kobj, "iommu_group");
863 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
866 EXPORT_SYMBOL_GPL(iommu_group_add_device);
869 * iommu_group_remove_device - remove a device from it's current group
870 * @dev: device to be removed
872 * This function is called by an iommu driver to remove the device from
873 * it's current group. This decrements the iommu group reference count.
875 void iommu_group_remove_device(struct device *dev)
877 struct iommu_group *group = dev->iommu_group;
878 struct group_device *tmp_device, *device = NULL;
880 dev_info(dev, "Removing from iommu group %d\n", group->id);
882 /* Pre-notify listeners that a device is being removed. */
883 blocking_notifier_call_chain(&group->notifier,
884 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
886 mutex_lock(&group->mutex);
887 list_for_each_entry(tmp_device, &group->devices, list) {
888 if (tmp_device->dev == dev) {
890 list_del(&device->list);
894 mutex_unlock(&group->mutex);
899 sysfs_remove_link(group->devices_kobj, device->name);
900 sysfs_remove_link(&dev->kobj, "iommu_group");
902 trace_remove_device_from_group(group->id, dev);
906 dev->iommu_group = NULL;
907 kobject_put(group->devices_kobj);
909 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
911 static int iommu_group_device_count(struct iommu_group *group)
913 struct group_device *entry;
916 list_for_each_entry(entry, &group->devices, list)
923 * iommu_group_for_each_dev - iterate over each device in the group
925 * @data: caller opaque data to be passed to callback function
926 * @fn: caller supplied callback function
928 * This function is called by group users to iterate over group devices.
929 * Callers should hold a reference count to the group during callback.
930 * The group->mutex is held across callbacks, which will block calls to
931 * iommu_group_add/remove_device.
933 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
934 int (*fn)(struct device *, void *))
936 struct group_device *device;
939 list_for_each_entry(device, &group->devices, list) {
940 ret = fn(device->dev, data);
948 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
949 int (*fn)(struct device *, void *))
953 mutex_lock(&group->mutex);
954 ret = __iommu_group_for_each_dev(group, data, fn);
955 mutex_unlock(&group->mutex);
959 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
962 * iommu_group_get - Return the group for a device and increment reference
963 * @dev: get the group that this device belongs to
965 * This function is called by iommu drivers and users to get the group
966 * for the specified device. If found, the group is returned and the group
967 * reference in incremented, else NULL.
969 struct iommu_group *iommu_group_get(struct device *dev)
971 struct iommu_group *group = dev->iommu_group;
974 kobject_get(group->devices_kobj);
978 EXPORT_SYMBOL_GPL(iommu_group_get);
981 * iommu_group_ref_get - Increment reference on a group
982 * @group: the group to use, must not be NULL
984 * This function is called by iommu drivers to take additional references on an
985 * existing group. Returns the given group for convenience.
987 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
989 kobject_get(group->devices_kobj);
992 EXPORT_SYMBOL_GPL(iommu_group_ref_get);
995 * iommu_group_put - Decrement group reference
996 * @group: the group to use
998 * This function is called by iommu drivers and users to release the
999 * iommu group. Once the reference count is zero, the group is released.
1001 void iommu_group_put(struct iommu_group *group)
1004 kobject_put(group->devices_kobj);
1006 EXPORT_SYMBOL_GPL(iommu_group_put);
1009 * iommu_group_register_notifier - Register a notifier for group changes
1010 * @group: the group to watch
1011 * @nb: notifier block to signal
1013 * This function allows iommu group users to track changes in a group.
1014 * See include/linux/iommu.h for actions sent via this notifier. Caller
1015 * should hold a reference to the group throughout notifier registration.
1017 int iommu_group_register_notifier(struct iommu_group *group,
1018 struct notifier_block *nb)
1020 return blocking_notifier_chain_register(&group->notifier, nb);
1022 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
1025 * iommu_group_unregister_notifier - Unregister a notifier
1026 * @group: the group to watch
1027 * @nb: notifier block to signal
1029 * Unregister a previously registered group notifier block.
1031 int iommu_group_unregister_notifier(struct iommu_group *group,
1032 struct notifier_block *nb)
1034 return blocking_notifier_chain_unregister(&group->notifier, nb);
1036 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
1039 * iommu_register_device_fault_handler() - Register a device fault handler
1041 * @handler: the fault handler
1042 * @data: private data passed as argument to the handler
1044 * When an IOMMU fault event is received, this handler gets called with the
1045 * fault event and data as argument. The handler should return 0 on success. If
1046 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1047 * complete the fault by calling iommu_page_response() with one of the following
1049 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1050 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1051 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1052 * page faults if possible.
1054 * Return 0 if the fault handler was installed successfully, or an error.
1056 int iommu_register_device_fault_handler(struct device *dev,
1057 iommu_dev_fault_handler_t handler,
1060 struct dev_iommu *param = dev->iommu;
1066 mutex_lock(¶m->lock);
1067 /* Only allow one fault handler registered for each device */
1068 if (param->fault_param) {
1074 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
1075 if (!param->fault_param) {
1080 param->fault_param->handler = handler;
1081 param->fault_param->data = data;
1082 mutex_init(¶m->fault_param->lock);
1083 INIT_LIST_HEAD(¶m->fault_param->faults);
1086 mutex_unlock(¶m->lock);
1090 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1093 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1096 * Remove the device fault handler installed with
1097 * iommu_register_device_fault_handler().
1099 * Return 0 on success, or an error.
1101 int iommu_unregister_device_fault_handler(struct device *dev)
1103 struct dev_iommu *param = dev->iommu;
1109 mutex_lock(¶m->lock);
1111 if (!param->fault_param)
1114 /* we cannot unregister handler if there are pending faults */
1115 if (!list_empty(¶m->fault_param->faults)) {
1120 kfree(param->fault_param);
1121 param->fault_param = NULL;
1124 mutex_unlock(¶m->lock);
1128 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1131 * iommu_report_device_fault() - Report fault event to device driver
1133 * @evt: fault event data
1135 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1136 * handler. When this function fails and the fault is recoverable, it is the
1137 * caller's responsibility to complete the fault.
1139 * Return 0 on success, or an error.
1141 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1143 struct dev_iommu *param = dev->iommu;
1144 struct iommu_fault_event *evt_pending = NULL;
1145 struct iommu_fault_param *fparam;
1151 /* we only report device fault if there is a handler registered */
1152 mutex_lock(¶m->lock);
1153 fparam = param->fault_param;
1154 if (!fparam || !fparam->handler) {
1159 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1160 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1161 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1167 mutex_lock(&fparam->lock);
1168 list_add_tail(&evt_pending->list, &fparam->faults);
1169 mutex_unlock(&fparam->lock);
1172 ret = fparam->handler(&evt->fault, fparam->data);
1173 if (ret && evt_pending) {
1174 mutex_lock(&fparam->lock);
1175 list_del(&evt_pending->list);
1176 mutex_unlock(&fparam->lock);
1180 mutex_unlock(¶m->lock);
1183 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1185 int iommu_page_response(struct device *dev,
1186 struct iommu_page_response *msg)
1190 struct iommu_fault_event *evt;
1191 struct iommu_fault_page_request *prm;
1192 struct dev_iommu *param = dev->iommu;
1193 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1195 if (!domain || !domain->ops->page_response)
1198 if (!param || !param->fault_param)
1201 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1202 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1205 /* Only send response if there is a fault report pending */
1206 mutex_lock(¶m->fault_param->lock);
1207 if (list_empty(¶m->fault_param->faults)) {
1208 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1212 * Check if we have a matching page request pending to respond,
1213 * otherwise return -EINVAL
1215 list_for_each_entry(evt, ¶m->fault_param->faults, list) {
1216 prm = &evt->fault.prm;
1217 pasid_valid = prm->flags & IOMMU_FAULT_PAGE_REQUEST_PASID_VALID;
1219 if ((pasid_valid && prm->pasid != msg->pasid) ||
1220 prm->grpid != msg->grpid)
1223 /* Sanitize the reply */
1224 msg->flags = pasid_valid ? IOMMU_PAGE_RESP_PASID_VALID : 0;
1226 ret = domain->ops->page_response(dev, evt, msg);
1227 list_del(&evt->list);
1233 mutex_unlock(¶m->fault_param->lock);
1236 EXPORT_SYMBOL_GPL(iommu_page_response);
1239 * iommu_group_id - Return ID for a group
1240 * @group: the group to ID
1242 * Return the unique ID for the group matching the sysfs group number.
1244 int iommu_group_id(struct iommu_group *group)
1248 EXPORT_SYMBOL_GPL(iommu_group_id);
1250 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1251 unsigned long *devfns);
1254 * To consider a PCI device isolated, we require ACS to support Source
1255 * Validation, Request Redirection, Completer Redirection, and Upstream
1256 * Forwarding. This effectively means that devices cannot spoof their
1257 * requester ID, requests and completions cannot be redirected, and all
1258 * transactions are forwarded upstream, even as it passes through a
1259 * bridge where the target device is downstream.
1261 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1264 * For multifunction devices which are not isolated from each other, find
1265 * all the other non-isolated functions and look for existing groups. For
1266 * each function, we also need to look for aliases to or from other devices
1267 * that may already have a group.
1269 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1270 unsigned long *devfns)
1272 struct pci_dev *tmp = NULL;
1273 struct iommu_group *group;
1275 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1278 for_each_pci_dev(tmp) {
1279 if (tmp == pdev || tmp->bus != pdev->bus ||
1280 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1281 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1284 group = get_pci_alias_group(tmp, devfns);
1295 * Look for aliases to or from the given device for existing groups. DMA
1296 * aliases are only supported on the same bus, therefore the search
1297 * space is quite small (especially since we're really only looking at pcie
1298 * device, and therefore only expect multiple slots on the root complex or
1299 * downstream switch ports). It's conceivable though that a pair of
1300 * multifunction devices could have aliases between them that would cause a
1301 * loop. To prevent this, we use a bitmap to track where we've been.
1303 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1304 unsigned long *devfns)
1306 struct pci_dev *tmp = NULL;
1307 struct iommu_group *group;
1309 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1312 group = iommu_group_get(&pdev->dev);
1316 for_each_pci_dev(tmp) {
1317 if (tmp == pdev || tmp->bus != pdev->bus)
1320 /* We alias them or they alias us */
1321 if (pci_devs_are_dma_aliases(pdev, tmp)) {
1322 group = get_pci_alias_group(tmp, devfns);
1328 group = get_pci_function_alias_group(tmp, devfns);
1339 struct group_for_pci_data {
1340 struct pci_dev *pdev;
1341 struct iommu_group *group;
1345 * DMA alias iterator callback, return the last seen device. Stop and return
1346 * the IOMMU group if we find one along the way.
1348 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1350 struct group_for_pci_data *data = opaque;
1353 data->group = iommu_group_get(&pdev->dev);
1355 return data->group != NULL;
1359 * Generic device_group call-back function. It just allocates one
1360 * iommu-group per device.
1362 struct iommu_group *generic_device_group(struct device *dev)
1364 return iommu_group_alloc();
1366 EXPORT_SYMBOL_GPL(generic_device_group);
1369 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1370 * to find or create an IOMMU group for a device.
1372 struct iommu_group *pci_device_group(struct device *dev)
1374 struct pci_dev *pdev = to_pci_dev(dev);
1375 struct group_for_pci_data data;
1376 struct pci_bus *bus;
1377 struct iommu_group *group = NULL;
1378 u64 devfns[4] = { 0 };
1380 if (WARN_ON(!dev_is_pci(dev)))
1381 return ERR_PTR(-EINVAL);
1384 * Find the upstream DMA alias for the device. A device must not
1385 * be aliased due to topology in order to have its own IOMMU group.
1386 * If we find an alias along the way that already belongs to a
1389 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1395 * Continue upstream from the point of minimum IOMMU granularity
1396 * due to aliases to the point where devices are protected from
1397 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1400 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1404 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1409 group = iommu_group_get(&pdev->dev);
1415 * Look for existing groups on device aliases. If we alias another
1416 * device or another device aliases us, use the same group.
1418 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1423 * Look for existing groups on non-isolated functions on the same
1424 * slot and aliases of those funcions, if any. No need to clear
1425 * the search bitmap, the tested devfns are still valid.
1427 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1431 /* No shared group found, allocate new */
1432 return iommu_group_alloc();
1434 EXPORT_SYMBOL_GPL(pci_device_group);
1436 /* Get the IOMMU group for device on fsl-mc bus */
1437 struct iommu_group *fsl_mc_device_group(struct device *dev)
1439 struct device *cont_dev = fsl_mc_cont_dev(dev);
1440 struct iommu_group *group;
1442 group = iommu_group_get(cont_dev);
1444 group = iommu_group_alloc();
1447 EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1449 static int iommu_get_def_domain_type(struct device *dev)
1451 const struct iommu_ops *ops = dev->bus->iommu_ops;
1452 unsigned int type = 0;
1454 if (ops->def_domain_type)
1455 type = ops->def_domain_type(dev);
1457 return (type == 0) ? iommu_def_domain_type : type;
1460 static int iommu_group_alloc_default_domain(struct bus_type *bus,
1461 struct iommu_group *group,
1464 struct iommu_domain *dom;
1466 dom = __iommu_domain_alloc(bus, type);
1467 if (!dom && type != IOMMU_DOMAIN_DMA) {
1468 dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
1470 pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1477 group->default_domain = dom;
1479 group->domain = dom;
1481 if (!iommu_dma_strict) {
1483 iommu_domain_set_attr(dom,
1484 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1491 static int iommu_alloc_default_domain(struct iommu_group *group,
1496 if (group->default_domain)
1499 type = iommu_get_def_domain_type(dev);
1501 return iommu_group_alloc_default_domain(dev->bus, group, type);
1505 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1506 * @dev: target device
1508 * This function is intended to be called by IOMMU drivers and extended to
1509 * support common, bus-defined algorithms when determining or creating the
1510 * IOMMU group for a device. On success, the caller will hold a reference
1511 * to the returned IOMMU group, which will already include the provided
1512 * device. The reference should be released with iommu_group_put().
1514 static struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1516 const struct iommu_ops *ops = dev->bus->iommu_ops;
1517 struct iommu_group *group;
1520 group = iommu_group_get(dev);
1525 return ERR_PTR(-EINVAL);
1527 group = ops->device_group(dev);
1528 if (WARN_ON_ONCE(group == NULL))
1529 return ERR_PTR(-EINVAL);
1534 ret = iommu_group_add_device(group, dev);
1541 iommu_group_put(group);
1543 return ERR_PTR(ret);
1546 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1548 return group->default_domain;
1551 static int probe_iommu_group(struct device *dev, void *data)
1553 struct list_head *group_list = data;
1554 struct iommu_group *group;
1557 /* Device is probed already if in a group */
1558 group = iommu_group_get(dev);
1560 iommu_group_put(group);
1564 ret = __iommu_probe_device(dev, group_list);
1571 static int remove_iommu_group(struct device *dev, void *data)
1573 iommu_release_device(dev);
1578 static int iommu_bus_notifier(struct notifier_block *nb,
1579 unsigned long action, void *data)
1581 unsigned long group_action = 0;
1582 struct device *dev = data;
1583 struct iommu_group *group;
1586 * ADD/DEL call into iommu driver ops if provided, which may
1587 * result in ADD/DEL notifiers to group->notifier
1589 if (action == BUS_NOTIFY_ADD_DEVICE) {
1592 ret = iommu_probe_device(dev);
1593 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1594 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1595 iommu_release_device(dev);
1600 * Remaining BUS_NOTIFYs get filtered and republished to the
1601 * group, if anyone is listening
1603 group = iommu_group_get(dev);
1608 case BUS_NOTIFY_BIND_DRIVER:
1609 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1611 case BUS_NOTIFY_BOUND_DRIVER:
1612 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1614 case BUS_NOTIFY_UNBIND_DRIVER:
1615 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1617 case BUS_NOTIFY_UNBOUND_DRIVER:
1618 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1623 blocking_notifier_call_chain(&group->notifier,
1626 iommu_group_put(group);
1630 struct __group_domain_type {
1635 static int probe_get_default_domain_type(struct device *dev, void *data)
1637 const struct iommu_ops *ops = dev->bus->iommu_ops;
1638 struct __group_domain_type *gtype = data;
1639 unsigned int type = 0;
1641 if (ops->def_domain_type)
1642 type = ops->def_domain_type(dev);
1645 if (gtype->type && gtype->type != type) {
1646 dev_warn(dev, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1647 iommu_domain_type_str(type),
1648 dev_name(gtype->dev),
1649 iommu_domain_type_str(gtype->type));
1662 static void probe_alloc_default_domain(struct bus_type *bus,
1663 struct iommu_group *group)
1665 struct __group_domain_type gtype;
1667 memset(>ype, 0, sizeof(gtype));
1669 /* Ask for default domain requirements of all devices in the group */
1670 __iommu_group_for_each_dev(group, >ype,
1671 probe_get_default_domain_type);
1674 gtype.type = iommu_def_domain_type;
1676 iommu_group_alloc_default_domain(bus, group, gtype.type);
1680 static int iommu_group_do_dma_attach(struct device *dev, void *data)
1682 struct iommu_domain *domain = data;
1685 if (!iommu_is_attach_deferred(domain, dev))
1686 ret = __iommu_attach_device(domain, dev);
1691 static int __iommu_group_dma_attach(struct iommu_group *group)
1693 return __iommu_group_for_each_dev(group, group->default_domain,
1694 iommu_group_do_dma_attach);
1697 static int iommu_group_do_probe_finalize(struct device *dev, void *data)
1699 struct iommu_domain *domain = data;
1701 if (domain->ops->probe_finalize)
1702 domain->ops->probe_finalize(dev);
1707 static void __iommu_group_dma_finalize(struct iommu_group *group)
1709 __iommu_group_for_each_dev(group, group->default_domain,
1710 iommu_group_do_probe_finalize);
1713 static int iommu_do_create_direct_mappings(struct device *dev, void *data)
1715 struct iommu_group *group = data;
1717 iommu_create_device_direct_mappings(group, dev);
1722 static int iommu_group_create_direct_mappings(struct iommu_group *group)
1724 return __iommu_group_for_each_dev(group, group,
1725 iommu_do_create_direct_mappings);
1728 int bus_iommu_probe(struct bus_type *bus)
1730 struct iommu_group *group, *next;
1731 LIST_HEAD(group_list);
1735 * This code-path does not allocate the default domain when
1736 * creating the iommu group, so do it after the groups are
1739 ret = bus_for_each_dev(bus, NULL, &group_list, probe_iommu_group);
1743 list_for_each_entry_safe(group, next, &group_list, entry) {
1744 /* Remove item from the list */
1745 list_del_init(&group->entry);
1747 mutex_lock(&group->mutex);
1749 /* Try to allocate default domain */
1750 probe_alloc_default_domain(bus, group);
1752 if (!group->default_domain) {
1753 mutex_unlock(&group->mutex);
1757 iommu_group_create_direct_mappings(group);
1759 ret = __iommu_group_dma_attach(group);
1761 mutex_unlock(&group->mutex);
1766 __iommu_group_dma_finalize(group);
1772 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1774 struct notifier_block *nb;
1777 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1781 nb->notifier_call = iommu_bus_notifier;
1783 err = bus_register_notifier(bus, nb);
1787 err = bus_iommu_probe(bus);
1796 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1797 bus_unregister_notifier(bus, nb);
1806 * bus_set_iommu - set iommu-callbacks for the bus
1808 * @ops: the callbacks provided by the iommu-driver
1810 * This function is called by an iommu driver to set the iommu methods
1811 * used for a particular bus. Drivers for devices on that bus can use
1812 * the iommu-api after these ops are registered.
1813 * This special function is needed because IOMMUs are usually devices on
1814 * the bus itself, so the iommu drivers are not initialized when the bus
1815 * is set up. With this function the iommu-driver can set the iommu-ops
1818 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1823 bus->iommu_ops = NULL;
1827 if (bus->iommu_ops != NULL)
1830 bus->iommu_ops = ops;
1832 /* Do IOMMU specific setup for this bus-type */
1833 err = iommu_bus_init(bus, ops);
1835 bus->iommu_ops = NULL;
1839 EXPORT_SYMBOL_GPL(bus_set_iommu);
1841 bool iommu_present(struct bus_type *bus)
1843 return bus->iommu_ops != NULL;
1845 EXPORT_SYMBOL_GPL(iommu_present);
1847 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1849 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1852 return bus->iommu_ops->capable(cap);
1854 EXPORT_SYMBOL_GPL(iommu_capable);
1857 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1858 * @domain: iommu domain
1859 * @handler: fault handler
1860 * @token: user data, will be passed back to the fault handler
1862 * This function should be used by IOMMU users which want to be notified
1863 * whenever an IOMMU fault happens.
1865 * The fault handler itself should return 0 on success, and an appropriate
1866 * error code otherwise.
1868 void iommu_set_fault_handler(struct iommu_domain *domain,
1869 iommu_fault_handler_t handler,
1874 domain->handler = handler;
1875 domain->handler_token = token;
1877 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1879 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1882 struct iommu_domain *domain;
1884 if (bus == NULL || bus->iommu_ops == NULL)
1887 domain = bus->iommu_ops->domain_alloc(type);
1891 domain->ops = bus->iommu_ops;
1892 domain->type = type;
1893 /* Assume all sizes by default; the driver may override this later */
1894 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1899 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1901 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1903 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1905 void iommu_domain_free(struct iommu_domain *domain)
1907 domain->ops->domain_free(domain);
1909 EXPORT_SYMBOL_GPL(iommu_domain_free);
1911 static int __iommu_attach_device(struct iommu_domain *domain,
1916 if (unlikely(domain->ops->attach_dev == NULL))
1919 ret = domain->ops->attach_dev(domain, dev);
1921 trace_attach_device_to_domain(dev);
1925 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1927 struct iommu_group *group;
1930 group = iommu_group_get(dev);
1935 * Lock the group to make sure the device-count doesn't
1936 * change while we are attaching
1938 mutex_lock(&group->mutex);
1940 if (iommu_group_device_count(group) != 1)
1943 ret = __iommu_attach_group(domain, group);
1946 mutex_unlock(&group->mutex);
1947 iommu_group_put(group);
1951 EXPORT_SYMBOL_GPL(iommu_attach_device);
1953 int iommu_cache_invalidate(struct iommu_domain *domain, struct device *dev,
1954 struct iommu_cache_invalidate_info *inv_info)
1956 if (unlikely(!domain->ops->cache_invalidate))
1959 return domain->ops->cache_invalidate(domain, dev, inv_info);
1961 EXPORT_SYMBOL_GPL(iommu_cache_invalidate);
1963 int iommu_sva_bind_gpasid(struct iommu_domain *domain,
1964 struct device *dev, struct iommu_gpasid_bind_data *data)
1966 if (unlikely(!domain->ops->sva_bind_gpasid))
1969 return domain->ops->sva_bind_gpasid(domain, dev, data);
1971 EXPORT_SYMBOL_GPL(iommu_sva_bind_gpasid);
1973 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
1976 if (unlikely(!domain->ops->sva_unbind_gpasid))
1979 return domain->ops->sva_unbind_gpasid(dev, pasid);
1981 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
1983 static void __iommu_detach_device(struct iommu_domain *domain,
1986 if (iommu_is_attach_deferred(domain, dev))
1989 if (unlikely(domain->ops->detach_dev == NULL))
1992 domain->ops->detach_dev(domain, dev);
1993 trace_detach_device_from_domain(dev);
1996 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1998 struct iommu_group *group;
2000 group = iommu_group_get(dev);
2004 mutex_lock(&group->mutex);
2005 if (iommu_group_device_count(group) != 1) {
2010 __iommu_detach_group(domain, group);
2013 mutex_unlock(&group->mutex);
2014 iommu_group_put(group);
2016 EXPORT_SYMBOL_GPL(iommu_detach_device);
2018 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
2020 struct iommu_domain *domain;
2021 struct iommu_group *group;
2023 group = iommu_group_get(dev);
2027 domain = group->domain;
2029 iommu_group_put(group);
2033 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
2036 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2037 * guarantees that the group and its default domain are valid and correct.
2039 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
2041 return dev->iommu_group->default_domain;
2045 * IOMMU groups are really the natural working unit of the IOMMU, but
2046 * the IOMMU API works on domains and devices. Bridge that gap by
2047 * iterating over the devices in a group. Ideally we'd have a single
2048 * device which represents the requestor ID of the group, but we also
2049 * allow IOMMU drivers to create policy defined minimum sets, where
2050 * the physical hardware may be able to distiguish members, but we
2051 * wish to group them at a higher level (ex. untrusted multi-function
2052 * PCI devices). Thus we attach each device.
2054 static int iommu_group_do_attach_device(struct device *dev, void *data)
2056 struct iommu_domain *domain = data;
2058 return __iommu_attach_device(domain, dev);
2061 static int __iommu_attach_group(struct iommu_domain *domain,
2062 struct iommu_group *group)
2066 if (group->default_domain && group->domain != group->default_domain)
2069 ret = __iommu_group_for_each_dev(group, domain,
2070 iommu_group_do_attach_device);
2072 group->domain = domain;
2077 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
2081 mutex_lock(&group->mutex);
2082 ret = __iommu_attach_group(domain, group);
2083 mutex_unlock(&group->mutex);
2087 EXPORT_SYMBOL_GPL(iommu_attach_group);
2089 static int iommu_group_do_detach_device(struct device *dev, void *data)
2091 struct iommu_domain *domain = data;
2093 __iommu_detach_device(domain, dev);
2098 static void __iommu_detach_group(struct iommu_domain *domain,
2099 struct iommu_group *group)
2103 if (!group->default_domain) {
2104 __iommu_group_for_each_dev(group, domain,
2105 iommu_group_do_detach_device);
2106 group->domain = NULL;
2110 if (group->domain == group->default_domain)
2113 /* Detach by re-attaching to the default domain */
2114 ret = __iommu_group_for_each_dev(group, group->default_domain,
2115 iommu_group_do_attach_device);
2119 group->domain = group->default_domain;
2122 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
2124 mutex_lock(&group->mutex);
2125 __iommu_detach_group(domain, group);
2126 mutex_unlock(&group->mutex);
2128 EXPORT_SYMBOL_GPL(iommu_detach_group);
2130 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
2132 if (unlikely(domain->ops->iova_to_phys == NULL))
2135 return domain->ops->iova_to_phys(domain, iova);
2137 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
2139 static size_t iommu_pgsize(struct iommu_domain *domain,
2140 unsigned long addr_merge, size_t size)
2142 unsigned int pgsize_idx;
2145 /* Max page size that still fits into 'size' */
2146 pgsize_idx = __fls(size);
2148 /* need to consider alignment requirements ? */
2149 if (likely(addr_merge)) {
2150 /* Max page size allowed by address */
2151 unsigned int align_pgsize_idx = __ffs(addr_merge);
2152 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
2155 /* build a mask of acceptable page sizes */
2156 pgsize = (1UL << (pgsize_idx + 1)) - 1;
2158 /* throw away page sizes not supported by the hardware */
2159 pgsize &= domain->pgsize_bitmap;
2161 /* make sure we're still sane */
2164 /* pick the biggest page */
2165 pgsize_idx = __fls(pgsize);
2166 pgsize = 1UL << pgsize_idx;
2171 int __iommu_map(struct iommu_domain *domain, unsigned long iova,
2172 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2174 const struct iommu_ops *ops = domain->ops;
2175 unsigned long orig_iova = iova;
2176 unsigned int min_pagesz;
2177 size_t orig_size = size;
2178 phys_addr_t orig_paddr = paddr;
2181 if (unlikely(ops->map == NULL ||
2182 domain->pgsize_bitmap == 0UL))
2185 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2188 /* find out the minimum page size supported */
2189 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2192 * both the virtual address and the physical one, as well as
2193 * the size of the mapping, must be aligned (at least) to the
2194 * size of the smallest page supported by the hardware
2196 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
2197 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2198 iova, &paddr, size, min_pagesz);
2202 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
2205 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
2207 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
2208 iova, &paddr, pgsize);
2209 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
2219 if (ops->iotlb_sync_map)
2220 ops->iotlb_sync_map(domain);
2222 /* unroll mapping in case something went wrong */
2224 iommu_unmap(domain, orig_iova, orig_size - size);
2226 trace_map(orig_iova, orig_paddr, orig_size);
2231 int iommu_map(struct iommu_domain *domain, unsigned long iova,
2232 phys_addr_t paddr, size_t size, int prot)
2235 return __iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
2237 EXPORT_SYMBOL_GPL(iommu_map);
2239 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
2240 phys_addr_t paddr, size_t size, int prot)
2242 return __iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
2244 EXPORT_SYMBOL_GPL(iommu_map_atomic);
2246 static size_t __iommu_unmap(struct iommu_domain *domain,
2247 unsigned long iova, size_t size,
2248 struct iommu_iotlb_gather *iotlb_gather)
2250 const struct iommu_ops *ops = domain->ops;
2251 size_t unmapped_page, unmapped = 0;
2252 unsigned long orig_iova = iova;
2253 unsigned int min_pagesz;
2255 if (unlikely(ops->unmap == NULL ||
2256 domain->pgsize_bitmap == 0UL))
2259 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2262 /* find out the minimum page size supported */
2263 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2266 * The virtual address, as well as the size of the mapping, must be
2267 * aligned (at least) to the size of the smallest page supported
2270 if (!IS_ALIGNED(iova | size, min_pagesz)) {
2271 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2272 iova, size, min_pagesz);
2276 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2279 * Keep iterating until we either unmap 'size' bytes (or more)
2280 * or we hit an area that isn't mapped.
2282 while (unmapped < size) {
2283 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
2285 unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
2289 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2290 iova, unmapped_page);
2292 iova += unmapped_page;
2293 unmapped += unmapped_page;
2296 trace_unmap(orig_iova, size, unmapped);
2300 size_t iommu_unmap(struct iommu_domain *domain,
2301 unsigned long iova, size_t size)
2303 struct iommu_iotlb_gather iotlb_gather;
2306 iommu_iotlb_gather_init(&iotlb_gather);
2307 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2308 iommu_tlb_sync(domain, &iotlb_gather);
2312 EXPORT_SYMBOL_GPL(iommu_unmap);
2314 size_t iommu_unmap_fast(struct iommu_domain *domain,
2315 unsigned long iova, size_t size,
2316 struct iommu_iotlb_gather *iotlb_gather)
2318 return __iommu_unmap(domain, iova, size, iotlb_gather);
2320 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2322 size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2323 struct scatterlist *sg, unsigned int nents, int prot,
2326 size_t len = 0, mapped = 0;
2331 while (i <= nents) {
2332 phys_addr_t s_phys = sg_phys(sg);
2334 if (len && s_phys != start + len) {
2335 ret = __iommu_map(domain, iova + mapped, start,
2359 /* undo mappings already done */
2360 iommu_unmap(domain, iova, mapped);
2366 size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2367 struct scatterlist *sg, unsigned int nents, int prot)
2370 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2372 EXPORT_SYMBOL_GPL(iommu_map_sg);
2374 size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2375 struct scatterlist *sg, unsigned int nents, int prot)
2377 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2379 EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2381 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
2382 phys_addr_t paddr, u64 size, int prot)
2384 if (unlikely(domain->ops->domain_window_enable == NULL))
2387 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2390 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2392 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2394 if (unlikely(domain->ops->domain_window_disable == NULL))
2397 return domain->ops->domain_window_disable(domain, wnd_nr);
2399 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2402 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2403 * @domain: the iommu domain where the fault has happened
2404 * @dev: the device where the fault has happened
2405 * @iova: the faulting address
2406 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2408 * This function should be called by the low-level IOMMU implementations
2409 * whenever IOMMU faults happen, to allow high-level users, that are
2410 * interested in such events, to know about them.
2412 * This event may be useful for several possible use cases:
2413 * - mere logging of the event
2414 * - dynamic TLB/PTE loading
2415 * - if restarting of the faulting device is required
2417 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2418 * PTE/TLB loading will one day be supported, implementations will be able
2419 * to tell whether it succeeded or not according to this return value).
2421 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2422 * (though fault handlers can also return -ENOSYS, in case they want to
2423 * elicit the default behavior of the IOMMU drivers).
2425 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2426 unsigned long iova, int flags)
2431 * if upper layers showed interest and installed a fault handler,
2434 if (domain->handler)
2435 ret = domain->handler(domain, dev, iova, flags,
2436 domain->handler_token);
2438 trace_io_page_fault(dev, iova, flags);
2441 EXPORT_SYMBOL_GPL(report_iommu_fault);
2443 static int __init iommu_init(void)
2445 iommu_group_kset = kset_create_and_add("iommu_groups",
2447 BUG_ON(!iommu_group_kset);
2449 iommu_debugfs_setup();
2453 core_initcall(iommu_init);
2455 int iommu_domain_get_attr(struct iommu_domain *domain,
2456 enum iommu_attr attr, void *data)
2458 struct iommu_domain_geometry *geometry;
2463 case DOMAIN_ATTR_GEOMETRY:
2465 *geometry = domain->geometry;
2468 case DOMAIN_ATTR_PAGING:
2470 *paging = (domain->pgsize_bitmap != 0UL);
2473 if (!domain->ops->domain_get_attr)
2476 ret = domain->ops->domain_get_attr(domain, attr, data);
2481 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2483 int iommu_domain_set_attr(struct iommu_domain *domain,
2484 enum iommu_attr attr, void *data)
2490 if (domain->ops->domain_set_attr == NULL)
2493 ret = domain->ops->domain_set_attr(domain, attr, data);
2498 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
2500 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2502 const struct iommu_ops *ops = dev->bus->iommu_ops;
2504 if (ops && ops->get_resv_regions)
2505 ops->get_resv_regions(dev, list);
2508 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2510 const struct iommu_ops *ops = dev->bus->iommu_ops;
2512 if (ops && ops->put_resv_regions)
2513 ops->put_resv_regions(dev, list);
2517 * generic_iommu_put_resv_regions - Reserved region driver helper
2518 * @dev: device for which to free reserved regions
2519 * @list: reserved region list for device
2521 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2522 * for simple reservations. Memory allocated for each reserved region will be
2523 * freed. If an IOMMU driver allocates additional resources per region, it is
2524 * going to have to implement a custom callback.
2526 void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2528 struct iommu_resv_region *entry, *next;
2530 list_for_each_entry_safe(entry, next, list, list)
2533 EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2535 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2536 size_t length, int prot,
2537 enum iommu_resv_type type)
2539 struct iommu_resv_region *region;
2541 region = kzalloc(sizeof(*region), GFP_KERNEL);
2545 INIT_LIST_HEAD(®ion->list);
2546 region->start = start;
2547 region->length = length;
2548 region->prot = prot;
2549 region->type = type;
2552 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2554 void iommu_set_default_passthrough(bool cmd_line)
2557 iommu_set_cmd_line_dma_api();
2559 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2562 void iommu_set_default_translated(bool cmd_line)
2565 iommu_set_cmd_line_dma_api();
2567 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2570 bool iommu_default_passthrough(void)
2572 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2574 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2576 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2578 const struct iommu_ops *ops = NULL;
2579 struct iommu_device *iommu;
2581 spin_lock(&iommu_device_lock);
2582 list_for_each_entry(iommu, &iommu_device_list, list)
2583 if (iommu->fwnode == fwnode) {
2587 spin_unlock(&iommu_device_lock);
2591 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2592 const struct iommu_ops *ops)
2594 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2597 return ops == fwspec->ops ? 0 : -EINVAL;
2599 if (!dev_iommu_get(dev))
2602 /* Preallocate for the overwhelmingly common case of 1 ID */
2603 fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2607 of_node_get(to_of_node(iommu_fwnode));
2608 fwspec->iommu_fwnode = iommu_fwnode;
2610 dev_iommu_fwspec_set(dev, fwspec);
2613 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2615 void iommu_fwspec_free(struct device *dev)
2617 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2620 fwnode_handle_put(fwspec->iommu_fwnode);
2622 dev_iommu_fwspec_set(dev, NULL);
2625 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2627 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2629 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2635 new_num = fwspec->num_ids + num_ids;
2637 fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2642 dev_iommu_fwspec_set(dev, fwspec);
2645 for (i = 0; i < num_ids; i++)
2646 fwspec->ids[fwspec->num_ids + i] = ids[i];
2648 fwspec->num_ids = new_num;
2651 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2654 * Per device IOMMU features.
2656 bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2658 const struct iommu_ops *ops = dev->bus->iommu_ops;
2660 if (ops && ops->dev_has_feat)
2661 return ops->dev_has_feat(dev, feat);
2665 EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2667 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2669 const struct iommu_ops *ops = dev->bus->iommu_ops;
2671 if (ops && ops->dev_enable_feat)
2672 return ops->dev_enable_feat(dev, feat);
2676 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2679 * The device drivers should do the necessary cleanups before calling this.
2680 * For example, before disabling the aux-domain feature, the device driver
2681 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2683 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2685 const struct iommu_ops *ops = dev->bus->iommu_ops;
2687 if (ops && ops->dev_disable_feat)
2688 return ops->dev_disable_feat(dev, feat);
2692 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2694 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2696 const struct iommu_ops *ops = dev->bus->iommu_ops;
2698 if (ops && ops->dev_feat_enabled)
2699 return ops->dev_feat_enabled(dev, feat);
2703 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2706 * Aux-domain specific attach/detach.
2708 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2709 * true. Also, as long as domains are attached to a device through this
2710 * interface, any tries to call iommu_attach_device() should fail
2711 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2712 * This should make us safe against a device being attached to a guest as a
2713 * whole while there are still pasid users on it (aux and sva).
2715 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2719 if (domain->ops->aux_attach_dev)
2720 ret = domain->ops->aux_attach_dev(domain, dev);
2723 trace_attach_device_to_domain(dev);
2727 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2729 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2731 if (domain->ops->aux_detach_dev) {
2732 domain->ops->aux_detach_dev(domain, dev);
2733 trace_detach_device_from_domain(dev);
2736 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2738 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2742 if (domain->ops->aux_get_pasid)
2743 ret = domain->ops->aux_get_pasid(domain, dev);
2747 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
2750 * iommu_sva_bind_device() - Bind a process address space to a device
2752 * @mm: the mm to bind, caller must hold a reference to it
2754 * Create a bond between device and address space, allowing the device to access
2755 * the mm using the returned PASID. If a bond already exists between @device and
2756 * @mm, it is returned and an additional reference is taken. Caller must call
2757 * iommu_sva_unbind_device() to release each reference.
2759 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2760 * initialize the required SVA features.
2762 * On error, returns an ERR_PTR value.
2765 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2767 struct iommu_group *group;
2768 struct iommu_sva *handle = ERR_PTR(-EINVAL);
2769 const struct iommu_ops *ops = dev->bus->iommu_ops;
2771 if (!ops || !ops->sva_bind)
2772 return ERR_PTR(-ENODEV);
2774 group = iommu_group_get(dev);
2776 return ERR_PTR(-ENODEV);
2778 /* Ensure device count and domain don't change while we're binding */
2779 mutex_lock(&group->mutex);
2782 * To keep things simple, SVA currently doesn't support IOMMU groups
2783 * with more than one device. Existing SVA-capable systems are not
2784 * affected by the problems that required IOMMU groups (lack of ACS
2785 * isolation, device ID aliasing and other hardware issues).
2787 if (iommu_group_device_count(group) != 1)
2790 handle = ops->sva_bind(dev, mm, drvdata);
2793 mutex_unlock(&group->mutex);
2794 iommu_group_put(group);
2798 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
2801 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2802 * @handle: the handle returned by iommu_sva_bind_device()
2804 * Put reference to a bond between device and address space. The device should
2805 * not be issuing any more transaction for this PASID. All outstanding page
2806 * requests for this PASID must have been flushed to the IOMMU.
2808 * Returns 0 on success, or an error value
2810 void iommu_sva_unbind_device(struct iommu_sva *handle)
2812 struct iommu_group *group;
2813 struct device *dev = handle->dev;
2814 const struct iommu_ops *ops = dev->bus->iommu_ops;
2816 if (!ops || !ops->sva_unbind)
2819 group = iommu_group_get(dev);
2823 mutex_lock(&group->mutex);
2824 ops->sva_unbind(handle);
2825 mutex_unlock(&group->mutex);
2827 iommu_group_put(group);
2829 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
2831 int iommu_sva_get_pasid(struct iommu_sva *handle)
2833 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
2835 if (!ops || !ops->sva_get_pasid)
2836 return IOMMU_PASID_INVALID;
2838 return ops->sva_get_pasid(handle);
2840 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);