1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
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);
96 static ssize_t iommu_group_store_type(struct iommu_group *group,
97 const char *buf, size_t count);
99 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
100 struct iommu_group_attribute iommu_group_attr_##_name = \
101 __ATTR(_name, _mode, _show, _store)
103 #define to_iommu_group_attr(_attr) \
104 container_of(_attr, struct iommu_group_attribute, attr)
105 #define to_iommu_group(_kobj) \
106 container_of(_kobj, struct iommu_group, kobj)
108 static LIST_HEAD(iommu_device_list);
109 static DEFINE_SPINLOCK(iommu_device_lock);
112 * Use a function instead of an array here because the domain-type is a
113 * bit-field, so an array would waste memory.
115 static const char *iommu_domain_type_str(unsigned int t)
118 case IOMMU_DOMAIN_BLOCKED:
120 case IOMMU_DOMAIN_IDENTITY:
121 return "Passthrough";
122 case IOMMU_DOMAIN_UNMANAGED:
124 case IOMMU_DOMAIN_DMA:
131 static int __init iommu_subsys_init(void)
133 bool cmd_line = iommu_cmd_line_dma_api();
136 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
137 iommu_set_default_passthrough(false);
139 iommu_set_default_translated(false);
141 if (iommu_default_passthrough() && mem_encrypt_active()) {
142 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
143 iommu_set_default_translated(false);
147 pr_info("Default domain type: %s %s\n",
148 iommu_domain_type_str(iommu_def_domain_type),
149 cmd_line ? "(set via kernel command line)" : "");
153 subsys_initcall(iommu_subsys_init);
155 int iommu_device_register(struct iommu_device *iommu)
157 spin_lock(&iommu_device_lock);
158 list_add_tail(&iommu->list, &iommu_device_list);
159 spin_unlock(&iommu_device_lock);
162 EXPORT_SYMBOL_GPL(iommu_device_register);
164 void iommu_device_unregister(struct iommu_device *iommu)
166 spin_lock(&iommu_device_lock);
167 list_del(&iommu->list);
168 spin_unlock(&iommu_device_lock);
170 EXPORT_SYMBOL_GPL(iommu_device_unregister);
172 static struct dev_iommu *dev_iommu_get(struct device *dev)
174 struct dev_iommu *param = dev->iommu;
179 param = kzalloc(sizeof(*param), GFP_KERNEL);
183 mutex_init(¶m->lock);
188 static void dev_iommu_free(struct device *dev)
190 iommu_fwspec_free(dev);
195 static int __iommu_probe_device(struct device *dev, struct list_head *group_list)
197 const struct iommu_ops *ops = dev->bus->iommu_ops;
198 struct iommu_device *iommu_dev;
199 struct iommu_group *group;
205 if (!dev_iommu_get(dev))
208 if (!try_module_get(ops->owner)) {
213 iommu_dev = ops->probe_device(dev);
214 if (IS_ERR(iommu_dev)) {
215 ret = PTR_ERR(iommu_dev);
219 dev->iommu->iommu_dev = iommu_dev;
221 group = iommu_group_get_for_dev(dev);
223 ret = PTR_ERR(group);
226 iommu_group_put(group);
228 if (group_list && !group->default_domain && list_empty(&group->entry))
229 list_add_tail(&group->entry, group_list);
231 iommu_device_link(iommu_dev, dev);
236 ops->release_device(dev);
239 module_put(ops->owner);
247 int iommu_probe_device(struct device *dev)
249 const struct iommu_ops *ops = dev->bus->iommu_ops;
250 struct iommu_group *group;
253 ret = __iommu_probe_device(dev, NULL);
257 group = iommu_group_get(dev);
264 * Try to allocate a default domain - needs support from the
265 * IOMMU driver. There are still some drivers which don't
266 * support default domains, so the return value is not yet
269 iommu_alloc_default_domain(group, dev);
271 if (group->default_domain) {
272 ret = __iommu_attach_device(group->default_domain, dev);
274 iommu_group_put(group);
279 iommu_create_device_direct_mappings(group, dev);
281 iommu_group_put(group);
283 if (ops->probe_finalize)
284 ops->probe_finalize(dev);
289 iommu_release_device(dev);
296 void iommu_release_device(struct device *dev)
298 const struct iommu_ops *ops = dev->bus->iommu_ops;
303 iommu_device_unlink(dev->iommu->iommu_dev, dev);
305 ops->release_device(dev);
307 iommu_group_remove_device(dev);
308 module_put(ops->owner);
312 static int __init iommu_set_def_domain_type(char *str)
317 ret = kstrtobool(str, &pt);
322 iommu_set_default_passthrough(true);
324 iommu_set_default_translated(true);
328 early_param("iommu.passthrough", iommu_set_def_domain_type);
330 static int __init iommu_dma_setup(char *str)
332 return kstrtobool(str, &iommu_dma_strict);
334 early_param("iommu.strict", iommu_dma_setup);
336 static ssize_t iommu_group_attr_show(struct kobject *kobj,
337 struct attribute *__attr, char *buf)
339 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
340 struct iommu_group *group = to_iommu_group(kobj);
344 ret = attr->show(group, buf);
348 static ssize_t iommu_group_attr_store(struct kobject *kobj,
349 struct attribute *__attr,
350 const char *buf, size_t count)
352 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
353 struct iommu_group *group = to_iommu_group(kobj);
357 ret = attr->store(group, buf, count);
361 static const struct sysfs_ops iommu_group_sysfs_ops = {
362 .show = iommu_group_attr_show,
363 .store = iommu_group_attr_store,
366 static int iommu_group_create_file(struct iommu_group *group,
367 struct iommu_group_attribute *attr)
369 return sysfs_create_file(&group->kobj, &attr->attr);
372 static void iommu_group_remove_file(struct iommu_group *group,
373 struct iommu_group_attribute *attr)
375 sysfs_remove_file(&group->kobj, &attr->attr);
378 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
380 return sprintf(buf, "%s\n", group->name);
384 * iommu_insert_resv_region - Insert a new region in the
385 * list of reserved regions.
386 * @new: new region to insert
387 * @regions: list of regions
389 * Elements are sorted by start address and overlapping segments
390 * of the same type are merged.
392 static int iommu_insert_resv_region(struct iommu_resv_region *new,
393 struct list_head *regions)
395 struct iommu_resv_region *iter, *tmp, *nr, *top;
398 nr = iommu_alloc_resv_region(new->start, new->length,
399 new->prot, new->type);
403 /* First add the new element based on start address sorting */
404 list_for_each_entry(iter, regions, list) {
405 if (nr->start < iter->start ||
406 (nr->start == iter->start && nr->type <= iter->type))
409 list_add_tail(&nr->list, &iter->list);
411 /* Merge overlapping segments of type nr->type in @regions, if any */
412 list_for_each_entry_safe(iter, tmp, regions, list) {
413 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
415 /* no merge needed on elements of different types than @new */
416 if (iter->type != new->type) {
417 list_move_tail(&iter->list, &stack);
421 /* look for the last stack element of same type as @iter */
422 list_for_each_entry_reverse(top, &stack, list)
423 if (top->type == iter->type)
426 list_move_tail(&iter->list, &stack);
430 top_end = top->start + top->length - 1;
432 if (iter->start > top_end + 1) {
433 list_move_tail(&iter->list, &stack);
435 top->length = max(top_end, iter_end) - top->start + 1;
436 list_del(&iter->list);
440 list_splice(&stack, regions);
445 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
446 struct list_head *group_resv_regions)
448 struct iommu_resv_region *entry;
451 list_for_each_entry(entry, dev_resv_regions, list) {
452 ret = iommu_insert_resv_region(entry, group_resv_regions);
459 int iommu_get_group_resv_regions(struct iommu_group *group,
460 struct list_head *head)
462 struct group_device *device;
465 mutex_lock(&group->mutex);
466 list_for_each_entry(device, &group->devices, list) {
467 struct list_head dev_resv_regions;
469 INIT_LIST_HEAD(&dev_resv_regions);
470 iommu_get_resv_regions(device->dev, &dev_resv_regions);
471 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
472 iommu_put_resv_regions(device->dev, &dev_resv_regions);
476 mutex_unlock(&group->mutex);
479 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
481 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
484 struct iommu_resv_region *region, *next;
485 struct list_head group_resv_regions;
488 INIT_LIST_HEAD(&group_resv_regions);
489 iommu_get_group_resv_regions(group, &group_resv_regions);
491 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
492 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
493 (long long int)region->start,
494 (long long int)(region->start +
496 iommu_group_resv_type_string[region->type]);
503 static ssize_t iommu_group_show_type(struct iommu_group *group,
506 char *type = "unknown\n";
508 mutex_lock(&group->mutex);
509 if (group->default_domain) {
510 switch (group->default_domain->type) {
511 case IOMMU_DOMAIN_BLOCKED:
514 case IOMMU_DOMAIN_IDENTITY:
517 case IOMMU_DOMAIN_UNMANAGED:
518 type = "unmanaged\n";
520 case IOMMU_DOMAIN_DMA:
525 mutex_unlock(&group->mutex);
531 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
533 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
534 iommu_group_show_resv_regions, NULL);
536 static IOMMU_GROUP_ATTR(type, 0644, iommu_group_show_type,
537 iommu_group_store_type);
539 static void iommu_group_release(struct kobject *kobj)
541 struct iommu_group *group = to_iommu_group(kobj);
543 pr_debug("Releasing group %d\n", group->id);
545 if (group->iommu_data_release)
546 group->iommu_data_release(group->iommu_data);
548 ida_simple_remove(&iommu_group_ida, group->id);
550 if (group->default_domain)
551 iommu_domain_free(group->default_domain);
557 static struct kobj_type iommu_group_ktype = {
558 .sysfs_ops = &iommu_group_sysfs_ops,
559 .release = iommu_group_release,
563 * iommu_group_alloc - Allocate a new group
565 * This function is called by an iommu driver to allocate a new iommu
566 * group. The iommu group represents the minimum granularity of the iommu.
567 * Upon successful return, the caller holds a reference to the supplied
568 * group in order to hold the group until devices are added. Use
569 * iommu_group_put() to release this extra reference count, allowing the
570 * group to be automatically reclaimed once it has no devices or external
573 struct iommu_group *iommu_group_alloc(void)
575 struct iommu_group *group;
578 group = kzalloc(sizeof(*group), GFP_KERNEL);
580 return ERR_PTR(-ENOMEM);
582 group->kobj.kset = iommu_group_kset;
583 mutex_init(&group->mutex);
584 INIT_LIST_HEAD(&group->devices);
585 INIT_LIST_HEAD(&group->entry);
586 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
588 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
595 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
596 NULL, "%d", group->id);
598 ida_simple_remove(&iommu_group_ida, group->id);
599 kobject_put(&group->kobj);
603 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
604 if (!group->devices_kobj) {
605 kobject_put(&group->kobj); /* triggers .release & free */
606 return ERR_PTR(-ENOMEM);
610 * The devices_kobj holds a reference on the group kobject, so
611 * as long as that exists so will the group. We can therefore
612 * use the devices_kobj for reference counting.
614 kobject_put(&group->kobj);
616 ret = iommu_group_create_file(group,
617 &iommu_group_attr_reserved_regions);
621 ret = iommu_group_create_file(group, &iommu_group_attr_type);
625 pr_debug("Allocated group %d\n", group->id);
629 EXPORT_SYMBOL_GPL(iommu_group_alloc);
631 struct iommu_group *iommu_group_get_by_id(int id)
633 struct kobject *group_kobj;
634 struct iommu_group *group;
637 if (!iommu_group_kset)
640 name = kasprintf(GFP_KERNEL, "%d", id);
644 group_kobj = kset_find_obj(iommu_group_kset, name);
650 group = container_of(group_kobj, struct iommu_group, kobj);
651 BUG_ON(group->id != id);
653 kobject_get(group->devices_kobj);
654 kobject_put(&group->kobj);
658 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
661 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
664 * iommu drivers can store data in the group for use when doing iommu
665 * operations. This function provides a way to retrieve it. Caller
666 * should hold a group reference.
668 void *iommu_group_get_iommudata(struct iommu_group *group)
670 return group->iommu_data;
672 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
675 * iommu_group_set_iommudata - set iommu_data for a group
677 * @iommu_data: new data
678 * @release: release function for iommu_data
680 * iommu drivers can store data in the group for use when doing iommu
681 * operations. This function provides a way to set the data after
682 * the group has been allocated. Caller should hold a group reference.
684 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
685 void (*release)(void *iommu_data))
687 group->iommu_data = iommu_data;
688 group->iommu_data_release = release;
690 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
693 * iommu_group_set_name - set name for a group
697 * Allow iommu driver to set a name for a group. When set it will
698 * appear in a name attribute file under the group in sysfs.
700 int iommu_group_set_name(struct iommu_group *group, const char *name)
705 iommu_group_remove_file(group, &iommu_group_attr_name);
712 group->name = kstrdup(name, GFP_KERNEL);
716 ret = iommu_group_create_file(group, &iommu_group_attr_name);
725 EXPORT_SYMBOL_GPL(iommu_group_set_name);
727 static int iommu_create_device_direct_mappings(struct iommu_group *group,
730 struct iommu_domain *domain = group->default_domain;
731 struct iommu_resv_region *entry;
732 struct list_head mappings;
733 unsigned long pg_size;
736 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
739 BUG_ON(!domain->pgsize_bitmap);
741 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
742 INIT_LIST_HEAD(&mappings);
744 iommu_get_resv_regions(dev, &mappings);
746 /* We need to consider overlapping regions for different devices */
747 list_for_each_entry(entry, &mappings, list) {
748 dma_addr_t start, end, addr;
751 if (domain->ops->apply_resv_region)
752 domain->ops->apply_resv_region(dev, domain, entry);
754 start = ALIGN(entry->start, pg_size);
755 end = ALIGN(entry->start + entry->length, pg_size);
757 if (entry->type != IOMMU_RESV_DIRECT &&
758 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
761 for (addr = start; addr <= end; addr += pg_size) {
762 phys_addr_t phys_addr;
767 phys_addr = iommu_iova_to_phys(domain, addr);
775 ret = iommu_map(domain, addr - map_size,
776 addr - map_size, map_size,
786 iommu_flush_iotlb_all(domain);
789 iommu_put_resv_regions(dev, &mappings);
794 static bool iommu_is_attach_deferred(struct iommu_domain *domain,
797 if (domain->ops->is_attach_deferred)
798 return domain->ops->is_attach_deferred(domain, dev);
804 * iommu_group_add_device - add a device to an iommu group
805 * @group: the group into which to add the device (reference should be held)
808 * This function is called by an iommu driver to add a device into a
809 * group. Adding a device increments the group reference count.
811 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
814 struct group_device *device;
816 device = kzalloc(sizeof(*device), GFP_KERNEL);
822 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
824 goto err_free_device;
826 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
830 goto err_remove_link;
833 ret = sysfs_create_link_nowarn(group->devices_kobj,
834 &dev->kobj, device->name);
836 if (ret == -EEXIST && i >= 0) {
838 * Account for the slim chance of collision
839 * and append an instance to the name.
842 device->name = kasprintf(GFP_KERNEL, "%s.%d",
843 kobject_name(&dev->kobj), i++);
849 kobject_get(group->devices_kobj);
851 dev->iommu_group = group;
853 mutex_lock(&group->mutex);
854 list_add_tail(&device->list, &group->devices);
855 if (group->domain && !iommu_is_attach_deferred(group->domain, dev))
856 ret = __iommu_attach_device(group->domain, dev);
857 mutex_unlock(&group->mutex);
861 /* Notify any listeners about change to group. */
862 blocking_notifier_call_chain(&group->notifier,
863 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
865 trace_add_device_to_group(group->id, dev);
867 dev_info(dev, "Adding to iommu group %d\n", group->id);
872 mutex_lock(&group->mutex);
873 list_del(&device->list);
874 mutex_unlock(&group->mutex);
875 dev->iommu_group = NULL;
876 kobject_put(group->devices_kobj);
877 sysfs_remove_link(group->devices_kobj, device->name);
881 sysfs_remove_link(&dev->kobj, "iommu_group");
884 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
887 EXPORT_SYMBOL_GPL(iommu_group_add_device);
890 * iommu_group_remove_device - remove a device from it's current group
891 * @dev: device to be removed
893 * This function is called by an iommu driver to remove the device from
894 * it's current group. This decrements the iommu group reference count.
896 void iommu_group_remove_device(struct device *dev)
898 struct iommu_group *group = dev->iommu_group;
899 struct group_device *tmp_device, *device = NULL;
901 dev_info(dev, "Removing from iommu group %d\n", group->id);
903 /* Pre-notify listeners that a device is being removed. */
904 blocking_notifier_call_chain(&group->notifier,
905 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
907 mutex_lock(&group->mutex);
908 list_for_each_entry(tmp_device, &group->devices, list) {
909 if (tmp_device->dev == dev) {
911 list_del(&device->list);
915 mutex_unlock(&group->mutex);
920 sysfs_remove_link(group->devices_kobj, device->name);
921 sysfs_remove_link(&dev->kobj, "iommu_group");
923 trace_remove_device_from_group(group->id, dev);
927 dev->iommu_group = NULL;
928 kobject_put(group->devices_kobj);
930 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
932 static int iommu_group_device_count(struct iommu_group *group)
934 struct group_device *entry;
937 list_for_each_entry(entry, &group->devices, list)
944 * iommu_group_for_each_dev - iterate over each device in the group
946 * @data: caller opaque data to be passed to callback function
947 * @fn: caller supplied callback function
949 * This function is called by group users to iterate over group devices.
950 * Callers should hold a reference count to the group during callback.
951 * The group->mutex is held across callbacks, which will block calls to
952 * iommu_group_add/remove_device.
954 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
955 int (*fn)(struct device *, void *))
957 struct group_device *device;
960 list_for_each_entry(device, &group->devices, list) {
961 ret = fn(device->dev, data);
969 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
970 int (*fn)(struct device *, void *))
974 mutex_lock(&group->mutex);
975 ret = __iommu_group_for_each_dev(group, data, fn);
976 mutex_unlock(&group->mutex);
980 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
983 * iommu_group_get - Return the group for a device and increment reference
984 * @dev: get the group that this device belongs to
986 * This function is called by iommu drivers and users to get the group
987 * for the specified device. If found, the group is returned and the group
988 * reference in incremented, else NULL.
990 struct iommu_group *iommu_group_get(struct device *dev)
992 struct iommu_group *group = dev->iommu_group;
995 kobject_get(group->devices_kobj);
999 EXPORT_SYMBOL_GPL(iommu_group_get);
1002 * iommu_group_ref_get - Increment reference on a group
1003 * @group: the group to use, must not be NULL
1005 * This function is called by iommu drivers to take additional references on an
1006 * existing group. Returns the given group for convenience.
1008 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
1010 kobject_get(group->devices_kobj);
1013 EXPORT_SYMBOL_GPL(iommu_group_ref_get);
1016 * iommu_group_put - Decrement group reference
1017 * @group: the group to use
1019 * This function is called by iommu drivers and users to release the
1020 * iommu group. Once the reference count is zero, the group is released.
1022 void iommu_group_put(struct iommu_group *group)
1025 kobject_put(group->devices_kobj);
1027 EXPORT_SYMBOL_GPL(iommu_group_put);
1030 * iommu_group_register_notifier - Register a notifier for group changes
1031 * @group: the group to watch
1032 * @nb: notifier block to signal
1034 * This function allows iommu group users to track changes in a group.
1035 * See include/linux/iommu.h for actions sent via this notifier. Caller
1036 * should hold a reference to the group throughout notifier registration.
1038 int iommu_group_register_notifier(struct iommu_group *group,
1039 struct notifier_block *nb)
1041 return blocking_notifier_chain_register(&group->notifier, nb);
1043 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
1046 * iommu_group_unregister_notifier - Unregister a notifier
1047 * @group: the group to watch
1048 * @nb: notifier block to signal
1050 * Unregister a previously registered group notifier block.
1052 int iommu_group_unregister_notifier(struct iommu_group *group,
1053 struct notifier_block *nb)
1055 return blocking_notifier_chain_unregister(&group->notifier, nb);
1057 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
1060 * iommu_register_device_fault_handler() - Register a device fault handler
1062 * @handler: the fault handler
1063 * @data: private data passed as argument to the handler
1065 * When an IOMMU fault event is received, this handler gets called with the
1066 * fault event and data as argument. The handler should return 0 on success. If
1067 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1068 * complete the fault by calling iommu_page_response() with one of the following
1070 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1071 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1072 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1073 * page faults if possible.
1075 * Return 0 if the fault handler was installed successfully, or an error.
1077 int iommu_register_device_fault_handler(struct device *dev,
1078 iommu_dev_fault_handler_t handler,
1081 struct dev_iommu *param = dev->iommu;
1087 mutex_lock(¶m->lock);
1088 /* Only allow one fault handler registered for each device */
1089 if (param->fault_param) {
1095 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
1096 if (!param->fault_param) {
1101 param->fault_param->handler = handler;
1102 param->fault_param->data = data;
1103 mutex_init(¶m->fault_param->lock);
1104 INIT_LIST_HEAD(¶m->fault_param->faults);
1107 mutex_unlock(¶m->lock);
1111 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1114 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1117 * Remove the device fault handler installed with
1118 * iommu_register_device_fault_handler().
1120 * Return 0 on success, or an error.
1122 int iommu_unregister_device_fault_handler(struct device *dev)
1124 struct dev_iommu *param = dev->iommu;
1130 mutex_lock(¶m->lock);
1132 if (!param->fault_param)
1135 /* we cannot unregister handler if there are pending faults */
1136 if (!list_empty(¶m->fault_param->faults)) {
1141 kfree(param->fault_param);
1142 param->fault_param = NULL;
1145 mutex_unlock(¶m->lock);
1149 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1152 * iommu_report_device_fault() - Report fault event to device driver
1154 * @evt: fault event data
1156 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1157 * handler. When this function fails and the fault is recoverable, it is the
1158 * caller's responsibility to complete the fault.
1160 * Return 0 on success, or an error.
1162 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1164 struct dev_iommu *param = dev->iommu;
1165 struct iommu_fault_event *evt_pending = NULL;
1166 struct iommu_fault_param *fparam;
1172 /* we only report device fault if there is a handler registered */
1173 mutex_lock(¶m->lock);
1174 fparam = param->fault_param;
1175 if (!fparam || !fparam->handler) {
1180 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1181 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1182 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1188 mutex_lock(&fparam->lock);
1189 list_add_tail(&evt_pending->list, &fparam->faults);
1190 mutex_unlock(&fparam->lock);
1193 ret = fparam->handler(&evt->fault, fparam->data);
1194 if (ret && evt_pending) {
1195 mutex_lock(&fparam->lock);
1196 list_del(&evt_pending->list);
1197 mutex_unlock(&fparam->lock);
1201 mutex_unlock(¶m->lock);
1204 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1206 int iommu_page_response(struct device *dev,
1207 struct iommu_page_response *msg)
1211 struct iommu_fault_event *evt;
1212 struct iommu_fault_page_request *prm;
1213 struct dev_iommu *param = dev->iommu;
1214 bool has_pasid = msg->flags & IOMMU_PAGE_RESP_PASID_VALID;
1215 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1217 if (!domain || !domain->ops->page_response)
1220 if (!param || !param->fault_param)
1223 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1224 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1227 /* Only send response if there is a fault report pending */
1228 mutex_lock(¶m->fault_param->lock);
1229 if (list_empty(¶m->fault_param->faults)) {
1230 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1234 * Check if we have a matching page request pending to respond,
1235 * otherwise return -EINVAL
1237 list_for_each_entry(evt, ¶m->fault_param->faults, list) {
1238 prm = &evt->fault.prm;
1239 if (prm->grpid != msg->grpid)
1243 * If the PASID is required, the corresponding request is
1244 * matched using the group ID, the PASID valid bit and the PASID
1245 * value. Otherwise only the group ID matches request and
1248 needs_pasid = prm->flags & IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID;
1249 if (needs_pasid && (!has_pasid || msg->pasid != prm->pasid))
1252 if (!needs_pasid && has_pasid) {
1253 /* No big deal, just clear it. */
1254 msg->flags &= ~IOMMU_PAGE_RESP_PASID_VALID;
1258 ret = domain->ops->page_response(dev, evt, msg);
1259 list_del(&evt->list);
1265 mutex_unlock(¶m->fault_param->lock);
1268 EXPORT_SYMBOL_GPL(iommu_page_response);
1271 * iommu_group_id - Return ID for a group
1272 * @group: the group to ID
1274 * Return the unique ID for the group matching the sysfs group number.
1276 int iommu_group_id(struct iommu_group *group)
1280 EXPORT_SYMBOL_GPL(iommu_group_id);
1282 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1283 unsigned long *devfns);
1286 * To consider a PCI device isolated, we require ACS to support Source
1287 * Validation, Request Redirection, Completer Redirection, and Upstream
1288 * Forwarding. This effectively means that devices cannot spoof their
1289 * requester ID, requests and completions cannot be redirected, and all
1290 * transactions are forwarded upstream, even as it passes through a
1291 * bridge where the target device is downstream.
1293 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1296 * For multifunction devices which are not isolated from each other, find
1297 * all the other non-isolated functions and look for existing groups. For
1298 * each function, we also need to look for aliases to or from other devices
1299 * that may already have a group.
1301 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1302 unsigned long *devfns)
1304 struct pci_dev *tmp = NULL;
1305 struct iommu_group *group;
1307 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1310 for_each_pci_dev(tmp) {
1311 if (tmp == pdev || tmp->bus != pdev->bus ||
1312 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1313 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1316 group = get_pci_alias_group(tmp, devfns);
1327 * Look for aliases to or from the given device for existing groups. DMA
1328 * aliases are only supported on the same bus, therefore the search
1329 * space is quite small (especially since we're really only looking at pcie
1330 * device, and therefore only expect multiple slots on the root complex or
1331 * downstream switch ports). It's conceivable though that a pair of
1332 * multifunction devices could have aliases between them that would cause a
1333 * loop. To prevent this, we use a bitmap to track where we've been.
1335 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1336 unsigned long *devfns)
1338 struct pci_dev *tmp = NULL;
1339 struct iommu_group *group;
1341 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1344 group = iommu_group_get(&pdev->dev);
1348 for_each_pci_dev(tmp) {
1349 if (tmp == pdev || tmp->bus != pdev->bus)
1352 /* We alias them or they alias us */
1353 if (pci_devs_are_dma_aliases(pdev, tmp)) {
1354 group = get_pci_alias_group(tmp, devfns);
1360 group = get_pci_function_alias_group(tmp, devfns);
1371 struct group_for_pci_data {
1372 struct pci_dev *pdev;
1373 struct iommu_group *group;
1377 * DMA alias iterator callback, return the last seen device. Stop and return
1378 * the IOMMU group if we find one along the way.
1380 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1382 struct group_for_pci_data *data = opaque;
1385 data->group = iommu_group_get(&pdev->dev);
1387 return data->group != NULL;
1391 * Generic device_group call-back function. It just allocates one
1392 * iommu-group per device.
1394 struct iommu_group *generic_device_group(struct device *dev)
1396 return iommu_group_alloc();
1398 EXPORT_SYMBOL_GPL(generic_device_group);
1401 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1402 * to find or create an IOMMU group for a device.
1404 struct iommu_group *pci_device_group(struct device *dev)
1406 struct pci_dev *pdev = to_pci_dev(dev);
1407 struct group_for_pci_data data;
1408 struct pci_bus *bus;
1409 struct iommu_group *group = NULL;
1410 u64 devfns[4] = { 0 };
1412 if (WARN_ON(!dev_is_pci(dev)))
1413 return ERR_PTR(-EINVAL);
1416 * Find the upstream DMA alias for the device. A device must not
1417 * be aliased due to topology in order to have its own IOMMU group.
1418 * If we find an alias along the way that already belongs to a
1421 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1427 * Continue upstream from the point of minimum IOMMU granularity
1428 * due to aliases to the point where devices are protected from
1429 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1432 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1436 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1441 group = iommu_group_get(&pdev->dev);
1447 * Look for existing groups on device aliases. If we alias another
1448 * device or another device aliases us, use the same group.
1450 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1455 * Look for existing groups on non-isolated functions on the same
1456 * slot and aliases of those funcions, if any. No need to clear
1457 * the search bitmap, the tested devfns are still valid.
1459 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1463 /* No shared group found, allocate new */
1464 return iommu_group_alloc();
1466 EXPORT_SYMBOL_GPL(pci_device_group);
1468 /* Get the IOMMU group for device on fsl-mc bus */
1469 struct iommu_group *fsl_mc_device_group(struct device *dev)
1471 struct device *cont_dev = fsl_mc_cont_dev(dev);
1472 struct iommu_group *group;
1474 group = iommu_group_get(cont_dev);
1476 group = iommu_group_alloc();
1479 EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1481 static int iommu_get_def_domain_type(struct device *dev)
1483 const struct iommu_ops *ops = dev->bus->iommu_ops;
1485 if (dev_is_pci(dev) && to_pci_dev(dev)->untrusted)
1486 return IOMMU_DOMAIN_DMA;
1488 if (ops->def_domain_type)
1489 return ops->def_domain_type(dev);
1494 static int iommu_group_alloc_default_domain(struct bus_type *bus,
1495 struct iommu_group *group,
1498 struct iommu_domain *dom;
1500 dom = __iommu_domain_alloc(bus, type);
1501 if (!dom && type != IOMMU_DOMAIN_DMA) {
1502 dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
1504 pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1511 group->default_domain = dom;
1513 group->domain = dom;
1515 if (!iommu_dma_strict) {
1517 iommu_domain_set_attr(dom,
1518 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1525 static int iommu_alloc_default_domain(struct iommu_group *group,
1530 if (group->default_domain)
1533 type = iommu_get_def_domain_type(dev) ? : iommu_def_domain_type;
1535 return iommu_group_alloc_default_domain(dev->bus, group, type);
1539 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1540 * @dev: target device
1542 * This function is intended to be called by IOMMU drivers and extended to
1543 * support common, bus-defined algorithms when determining or creating the
1544 * IOMMU group for a device. On success, the caller will hold a reference
1545 * to the returned IOMMU group, which will already include the provided
1546 * device. The reference should be released with iommu_group_put().
1548 static struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1550 const struct iommu_ops *ops = dev->bus->iommu_ops;
1551 struct iommu_group *group;
1554 group = iommu_group_get(dev);
1559 return ERR_PTR(-EINVAL);
1561 group = ops->device_group(dev);
1562 if (WARN_ON_ONCE(group == NULL))
1563 return ERR_PTR(-EINVAL);
1568 ret = iommu_group_add_device(group, dev);
1575 iommu_group_put(group);
1577 return ERR_PTR(ret);
1580 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1582 return group->default_domain;
1585 static int probe_iommu_group(struct device *dev, void *data)
1587 struct list_head *group_list = data;
1588 struct iommu_group *group;
1591 /* Device is probed already if in a group */
1592 group = iommu_group_get(dev);
1594 iommu_group_put(group);
1598 ret = __iommu_probe_device(dev, group_list);
1605 static int remove_iommu_group(struct device *dev, void *data)
1607 iommu_release_device(dev);
1612 static int iommu_bus_notifier(struct notifier_block *nb,
1613 unsigned long action, void *data)
1615 unsigned long group_action = 0;
1616 struct device *dev = data;
1617 struct iommu_group *group;
1620 * ADD/DEL call into iommu driver ops if provided, which may
1621 * result in ADD/DEL notifiers to group->notifier
1623 if (action == BUS_NOTIFY_ADD_DEVICE) {
1626 ret = iommu_probe_device(dev);
1627 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1628 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1629 iommu_release_device(dev);
1634 * Remaining BUS_NOTIFYs get filtered and republished to the
1635 * group, if anyone is listening
1637 group = iommu_group_get(dev);
1642 case BUS_NOTIFY_BIND_DRIVER:
1643 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1645 case BUS_NOTIFY_BOUND_DRIVER:
1646 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1648 case BUS_NOTIFY_UNBIND_DRIVER:
1649 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1651 case BUS_NOTIFY_UNBOUND_DRIVER:
1652 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1657 blocking_notifier_call_chain(&group->notifier,
1660 iommu_group_put(group);
1664 struct __group_domain_type {
1669 static int probe_get_default_domain_type(struct device *dev, void *data)
1671 struct __group_domain_type *gtype = data;
1672 unsigned int type = iommu_get_def_domain_type(dev);
1675 if (gtype->type && gtype->type != type) {
1676 dev_warn(dev, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1677 iommu_domain_type_str(type),
1678 dev_name(gtype->dev),
1679 iommu_domain_type_str(gtype->type));
1692 static void probe_alloc_default_domain(struct bus_type *bus,
1693 struct iommu_group *group)
1695 struct __group_domain_type gtype;
1697 memset(>ype, 0, sizeof(gtype));
1699 /* Ask for default domain requirements of all devices in the group */
1700 __iommu_group_for_each_dev(group, >ype,
1701 probe_get_default_domain_type);
1704 gtype.type = iommu_def_domain_type;
1706 iommu_group_alloc_default_domain(bus, group, gtype.type);
1710 static int iommu_group_do_dma_attach(struct device *dev, void *data)
1712 struct iommu_domain *domain = data;
1715 if (!iommu_is_attach_deferred(domain, dev))
1716 ret = __iommu_attach_device(domain, dev);
1721 static int __iommu_group_dma_attach(struct iommu_group *group)
1723 return __iommu_group_for_each_dev(group, group->default_domain,
1724 iommu_group_do_dma_attach);
1727 static int iommu_group_do_probe_finalize(struct device *dev, void *data)
1729 struct iommu_domain *domain = data;
1731 if (domain->ops->probe_finalize)
1732 domain->ops->probe_finalize(dev);
1737 static void __iommu_group_dma_finalize(struct iommu_group *group)
1739 __iommu_group_for_each_dev(group, group->default_domain,
1740 iommu_group_do_probe_finalize);
1743 static int iommu_do_create_direct_mappings(struct device *dev, void *data)
1745 struct iommu_group *group = data;
1747 iommu_create_device_direct_mappings(group, dev);
1752 static int iommu_group_create_direct_mappings(struct iommu_group *group)
1754 return __iommu_group_for_each_dev(group, group,
1755 iommu_do_create_direct_mappings);
1758 int bus_iommu_probe(struct bus_type *bus)
1760 struct iommu_group *group, *next;
1761 LIST_HEAD(group_list);
1765 * This code-path does not allocate the default domain when
1766 * creating the iommu group, so do it after the groups are
1769 ret = bus_for_each_dev(bus, NULL, &group_list, probe_iommu_group);
1773 list_for_each_entry_safe(group, next, &group_list, entry) {
1774 /* Remove item from the list */
1775 list_del_init(&group->entry);
1777 mutex_lock(&group->mutex);
1779 /* Try to allocate default domain */
1780 probe_alloc_default_domain(bus, group);
1782 if (!group->default_domain) {
1783 mutex_unlock(&group->mutex);
1787 iommu_group_create_direct_mappings(group);
1789 ret = __iommu_group_dma_attach(group);
1791 mutex_unlock(&group->mutex);
1796 __iommu_group_dma_finalize(group);
1802 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1804 struct notifier_block *nb;
1807 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1811 nb->notifier_call = iommu_bus_notifier;
1813 err = bus_register_notifier(bus, nb);
1817 err = bus_iommu_probe(bus);
1826 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1827 bus_unregister_notifier(bus, nb);
1836 * bus_set_iommu - set iommu-callbacks for the bus
1838 * @ops: the callbacks provided by the iommu-driver
1840 * This function is called by an iommu driver to set the iommu methods
1841 * used for a particular bus. Drivers for devices on that bus can use
1842 * the iommu-api after these ops are registered.
1843 * This special function is needed because IOMMUs are usually devices on
1844 * the bus itself, so the iommu drivers are not initialized when the bus
1845 * is set up. With this function the iommu-driver can set the iommu-ops
1848 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1853 bus->iommu_ops = NULL;
1857 if (bus->iommu_ops != NULL)
1860 bus->iommu_ops = ops;
1862 /* Do IOMMU specific setup for this bus-type */
1863 err = iommu_bus_init(bus, ops);
1865 bus->iommu_ops = NULL;
1869 EXPORT_SYMBOL_GPL(bus_set_iommu);
1871 bool iommu_present(struct bus_type *bus)
1873 return bus->iommu_ops != NULL;
1875 EXPORT_SYMBOL_GPL(iommu_present);
1877 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1879 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1882 return bus->iommu_ops->capable(cap);
1884 EXPORT_SYMBOL_GPL(iommu_capable);
1887 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1888 * @domain: iommu domain
1889 * @handler: fault handler
1890 * @token: user data, will be passed back to the fault handler
1892 * This function should be used by IOMMU users which want to be notified
1893 * whenever an IOMMU fault happens.
1895 * The fault handler itself should return 0 on success, and an appropriate
1896 * error code otherwise.
1898 void iommu_set_fault_handler(struct iommu_domain *domain,
1899 iommu_fault_handler_t handler,
1904 domain->handler = handler;
1905 domain->handler_token = token;
1907 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1909 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1912 struct iommu_domain *domain;
1914 if (bus == NULL || bus->iommu_ops == NULL)
1917 domain = bus->iommu_ops->domain_alloc(type);
1921 domain->ops = bus->iommu_ops;
1922 domain->type = type;
1923 /* Assume all sizes by default; the driver may override this later */
1924 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1929 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1931 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1933 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1935 void iommu_domain_free(struct iommu_domain *domain)
1937 domain->ops->domain_free(domain);
1939 EXPORT_SYMBOL_GPL(iommu_domain_free);
1941 static int __iommu_attach_device(struct iommu_domain *domain,
1946 if (unlikely(domain->ops->attach_dev == NULL))
1949 ret = domain->ops->attach_dev(domain, dev);
1951 trace_attach_device_to_domain(dev);
1955 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1957 struct iommu_group *group;
1960 group = iommu_group_get(dev);
1965 * Lock the group to make sure the device-count doesn't
1966 * change while we are attaching
1968 mutex_lock(&group->mutex);
1970 if (iommu_group_device_count(group) != 1)
1973 ret = __iommu_attach_group(domain, group);
1976 mutex_unlock(&group->mutex);
1977 iommu_group_put(group);
1981 EXPORT_SYMBOL_GPL(iommu_attach_device);
1984 * Check flags and other user provided data for valid combinations. We also
1985 * make sure no reserved fields or unused flags are set. This is to ensure
1986 * not breaking userspace in the future when these fields or flags are used.
1988 static int iommu_check_cache_invl_data(struct iommu_cache_invalidate_info *info)
1993 if (info->version != IOMMU_CACHE_INVALIDATE_INFO_VERSION_1)
1996 mask = (1 << IOMMU_CACHE_INV_TYPE_NR) - 1;
1997 if (info->cache & ~mask)
2000 if (info->granularity >= IOMMU_INV_GRANU_NR)
2003 switch (info->granularity) {
2004 case IOMMU_INV_GRANU_ADDR:
2005 if (info->cache & IOMMU_CACHE_INV_TYPE_PASID)
2008 mask = IOMMU_INV_ADDR_FLAGS_PASID |
2009 IOMMU_INV_ADDR_FLAGS_ARCHID |
2010 IOMMU_INV_ADDR_FLAGS_LEAF;
2012 if (info->granu.addr_info.flags & ~mask)
2015 case IOMMU_INV_GRANU_PASID:
2016 mask = IOMMU_INV_PASID_FLAGS_PASID |
2017 IOMMU_INV_PASID_FLAGS_ARCHID;
2018 if (info->granu.pasid_info.flags & ~mask)
2022 case IOMMU_INV_GRANU_DOMAIN:
2023 if (info->cache & IOMMU_CACHE_INV_TYPE_DEV_IOTLB)
2030 /* Check reserved padding fields */
2031 for (i = 0; i < sizeof(info->padding); i++) {
2032 if (info->padding[i])
2039 int iommu_uapi_cache_invalidate(struct iommu_domain *domain, struct device *dev,
2042 struct iommu_cache_invalidate_info inv_info = { 0 };
2046 if (unlikely(!domain->ops->cache_invalidate))
2050 * No new spaces can be added before the variable sized union, the
2051 * minimum size is the offset to the union.
2053 minsz = offsetof(struct iommu_cache_invalidate_info, granu);
2055 /* Copy minsz from user to get flags and argsz */
2056 if (copy_from_user(&inv_info, uinfo, minsz))
2059 /* Fields before the variable size union are mandatory */
2060 if (inv_info.argsz < minsz)
2063 /* PASID and address granu require additional info beyond minsz */
2064 if (inv_info.granularity == IOMMU_INV_GRANU_PASID &&
2065 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.pasid_info))
2068 if (inv_info.granularity == IOMMU_INV_GRANU_ADDR &&
2069 inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.addr_info))
2073 * User might be using a newer UAPI header which has a larger data
2074 * size, we shall support the existing flags within the current
2075 * size. Copy the remaining user data _after_ minsz but not more
2076 * than the current kernel supported size.
2078 if (copy_from_user((void *)&inv_info + minsz, uinfo + minsz,
2079 min_t(u32, inv_info.argsz, sizeof(inv_info)) - minsz))
2082 /* Now the argsz is validated, check the content */
2083 ret = iommu_check_cache_invl_data(&inv_info);
2087 return domain->ops->cache_invalidate(domain, dev, &inv_info);
2089 EXPORT_SYMBOL_GPL(iommu_uapi_cache_invalidate);
2091 static int iommu_check_bind_data(struct iommu_gpasid_bind_data *data)
2096 if (data->version != IOMMU_GPASID_BIND_VERSION_1)
2099 /* Check the range of supported formats */
2100 if (data->format >= IOMMU_PASID_FORMAT_LAST)
2103 /* Check all flags */
2104 mask = IOMMU_SVA_GPASID_VAL;
2105 if (data->flags & ~mask)
2108 /* Check reserved padding fields */
2109 for (i = 0; i < sizeof(data->padding); i++) {
2110 if (data->padding[i])
2117 static int iommu_sva_prepare_bind_data(void __user *udata,
2118 struct iommu_gpasid_bind_data *data)
2123 * No new spaces can be added before the variable sized union, the
2124 * minimum size is the offset to the union.
2126 minsz = offsetof(struct iommu_gpasid_bind_data, vendor);
2128 /* Copy minsz from user to get flags and argsz */
2129 if (copy_from_user(data, udata, minsz))
2132 /* Fields before the variable size union are mandatory */
2133 if (data->argsz < minsz)
2136 * User might be using a newer UAPI header, we shall let IOMMU vendor
2137 * driver decide on what size it needs. Since the guest PASID bind data
2138 * can be vendor specific, larger argsz could be the result of extension
2139 * for one vendor but it should not affect another vendor.
2140 * Copy the remaining user data _after_ minsz
2142 if (copy_from_user((void *)data + minsz, udata + minsz,
2143 min_t(u32, data->argsz, sizeof(*data)) - minsz))
2146 return iommu_check_bind_data(data);
2149 int iommu_uapi_sva_bind_gpasid(struct iommu_domain *domain, struct device *dev,
2152 struct iommu_gpasid_bind_data data = { 0 };
2155 if (unlikely(!domain->ops->sva_bind_gpasid))
2158 ret = iommu_sva_prepare_bind_data(udata, &data);
2162 return domain->ops->sva_bind_gpasid(domain, dev, &data);
2164 EXPORT_SYMBOL_GPL(iommu_uapi_sva_bind_gpasid);
2166 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2169 if (unlikely(!domain->ops->sva_unbind_gpasid))
2172 return domain->ops->sva_unbind_gpasid(dev, pasid);
2174 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
2176 int iommu_uapi_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2179 struct iommu_gpasid_bind_data data = { 0 };
2182 if (unlikely(!domain->ops->sva_bind_gpasid))
2185 ret = iommu_sva_prepare_bind_data(udata, &data);
2189 return iommu_sva_unbind_gpasid(domain, dev, data.hpasid);
2191 EXPORT_SYMBOL_GPL(iommu_uapi_sva_unbind_gpasid);
2193 static void __iommu_detach_device(struct iommu_domain *domain,
2196 if (iommu_is_attach_deferred(domain, dev))
2199 if (unlikely(domain->ops->detach_dev == NULL))
2202 domain->ops->detach_dev(domain, dev);
2203 trace_detach_device_from_domain(dev);
2206 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
2208 struct iommu_group *group;
2210 group = iommu_group_get(dev);
2214 mutex_lock(&group->mutex);
2215 if (iommu_group_device_count(group) != 1) {
2220 __iommu_detach_group(domain, group);
2223 mutex_unlock(&group->mutex);
2224 iommu_group_put(group);
2226 EXPORT_SYMBOL_GPL(iommu_detach_device);
2228 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
2230 struct iommu_domain *domain;
2231 struct iommu_group *group;
2233 group = iommu_group_get(dev);
2237 domain = group->domain;
2239 iommu_group_put(group);
2243 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
2246 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2247 * guarantees that the group and its default domain are valid and correct.
2249 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
2251 return dev->iommu_group->default_domain;
2255 * IOMMU groups are really the natural working unit of the IOMMU, but
2256 * the IOMMU API works on domains and devices. Bridge that gap by
2257 * iterating over the devices in a group. Ideally we'd have a single
2258 * device which represents the requestor ID of the group, but we also
2259 * allow IOMMU drivers to create policy defined minimum sets, where
2260 * the physical hardware may be able to distiguish members, but we
2261 * wish to group them at a higher level (ex. untrusted multi-function
2262 * PCI devices). Thus we attach each device.
2264 static int iommu_group_do_attach_device(struct device *dev, void *data)
2266 struct iommu_domain *domain = data;
2268 return __iommu_attach_device(domain, dev);
2271 static int __iommu_attach_group(struct iommu_domain *domain,
2272 struct iommu_group *group)
2276 if (group->default_domain && group->domain != group->default_domain)
2279 ret = __iommu_group_for_each_dev(group, domain,
2280 iommu_group_do_attach_device);
2282 group->domain = domain;
2287 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
2291 mutex_lock(&group->mutex);
2292 ret = __iommu_attach_group(domain, group);
2293 mutex_unlock(&group->mutex);
2297 EXPORT_SYMBOL_GPL(iommu_attach_group);
2299 static int iommu_group_do_detach_device(struct device *dev, void *data)
2301 struct iommu_domain *domain = data;
2303 __iommu_detach_device(domain, dev);
2308 static void __iommu_detach_group(struct iommu_domain *domain,
2309 struct iommu_group *group)
2313 if (!group->default_domain) {
2314 __iommu_group_for_each_dev(group, domain,
2315 iommu_group_do_detach_device);
2316 group->domain = NULL;
2320 if (group->domain == group->default_domain)
2323 /* Detach by re-attaching to the default domain */
2324 ret = __iommu_group_for_each_dev(group, group->default_domain,
2325 iommu_group_do_attach_device);
2329 group->domain = group->default_domain;
2332 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
2334 mutex_lock(&group->mutex);
2335 __iommu_detach_group(domain, group);
2336 mutex_unlock(&group->mutex);
2338 EXPORT_SYMBOL_GPL(iommu_detach_group);
2340 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
2342 if (unlikely(domain->ops->iova_to_phys == NULL))
2345 return domain->ops->iova_to_phys(domain, iova);
2347 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
2349 static size_t iommu_pgsize(struct iommu_domain *domain,
2350 unsigned long addr_merge, size_t size)
2352 unsigned int pgsize_idx;
2355 /* Max page size that still fits into 'size' */
2356 pgsize_idx = __fls(size);
2358 /* need to consider alignment requirements ? */
2359 if (likely(addr_merge)) {
2360 /* Max page size allowed by address */
2361 unsigned int align_pgsize_idx = __ffs(addr_merge);
2362 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
2365 /* build a mask of acceptable page sizes */
2366 pgsize = (1UL << (pgsize_idx + 1)) - 1;
2368 /* throw away page sizes not supported by the hardware */
2369 pgsize &= domain->pgsize_bitmap;
2371 /* make sure we're still sane */
2374 /* pick the biggest page */
2375 pgsize_idx = __fls(pgsize);
2376 pgsize = 1UL << pgsize_idx;
2381 static int __iommu_map(struct iommu_domain *domain, unsigned long iova,
2382 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2384 const struct iommu_ops *ops = domain->ops;
2385 unsigned long orig_iova = iova;
2386 unsigned int min_pagesz;
2387 size_t orig_size = size;
2388 phys_addr_t orig_paddr = paddr;
2391 if (unlikely(ops->map == NULL ||
2392 domain->pgsize_bitmap == 0UL))
2395 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2398 /* find out the minimum page size supported */
2399 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2402 * both the virtual address and the physical one, as well as
2403 * the size of the mapping, must be aligned (at least) to the
2404 * size of the smallest page supported by the hardware
2406 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
2407 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2408 iova, &paddr, size, min_pagesz);
2412 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
2415 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
2417 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
2418 iova, &paddr, pgsize);
2419 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
2429 if (ops->iotlb_sync_map)
2430 ops->iotlb_sync_map(domain);
2432 /* unroll mapping in case something went wrong */
2434 iommu_unmap(domain, orig_iova, orig_size - size);
2436 trace_map(orig_iova, orig_paddr, orig_size);
2441 int iommu_map(struct iommu_domain *domain, unsigned long iova,
2442 phys_addr_t paddr, size_t size, int prot)
2445 return __iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
2447 EXPORT_SYMBOL_GPL(iommu_map);
2449 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
2450 phys_addr_t paddr, size_t size, int prot)
2452 return __iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
2454 EXPORT_SYMBOL_GPL(iommu_map_atomic);
2456 static size_t __iommu_unmap(struct iommu_domain *domain,
2457 unsigned long iova, size_t size,
2458 struct iommu_iotlb_gather *iotlb_gather)
2460 const struct iommu_ops *ops = domain->ops;
2461 size_t unmapped_page, unmapped = 0;
2462 unsigned long orig_iova = iova;
2463 unsigned int min_pagesz;
2465 if (unlikely(ops->unmap == NULL ||
2466 domain->pgsize_bitmap == 0UL))
2469 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2472 /* find out the minimum page size supported */
2473 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2476 * The virtual address, as well as the size of the mapping, must be
2477 * aligned (at least) to the size of the smallest page supported
2480 if (!IS_ALIGNED(iova | size, min_pagesz)) {
2481 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2482 iova, size, min_pagesz);
2486 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2489 * Keep iterating until we either unmap 'size' bytes (or more)
2490 * or we hit an area that isn't mapped.
2492 while (unmapped < size) {
2493 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
2495 unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
2499 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2500 iova, unmapped_page);
2502 iova += unmapped_page;
2503 unmapped += unmapped_page;
2506 trace_unmap(orig_iova, size, unmapped);
2510 size_t iommu_unmap(struct iommu_domain *domain,
2511 unsigned long iova, size_t size)
2513 struct iommu_iotlb_gather iotlb_gather;
2516 iommu_iotlb_gather_init(&iotlb_gather);
2517 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2518 iommu_iotlb_sync(domain, &iotlb_gather);
2522 EXPORT_SYMBOL_GPL(iommu_unmap);
2524 size_t iommu_unmap_fast(struct iommu_domain *domain,
2525 unsigned long iova, size_t size,
2526 struct iommu_iotlb_gather *iotlb_gather)
2528 return __iommu_unmap(domain, iova, size, iotlb_gather);
2530 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2532 static size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2533 struct scatterlist *sg, unsigned int nents, int prot,
2536 size_t len = 0, mapped = 0;
2541 while (i <= nents) {
2542 phys_addr_t s_phys = sg_phys(sg);
2544 if (len && s_phys != start + len) {
2545 ret = __iommu_map(domain, iova + mapped, start,
2569 /* undo mappings already done */
2570 iommu_unmap(domain, iova, mapped);
2576 size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2577 struct scatterlist *sg, unsigned int nents, int prot)
2580 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2582 EXPORT_SYMBOL_GPL(iommu_map_sg);
2584 size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2585 struct scatterlist *sg, unsigned int nents, int prot)
2587 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2589 EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2591 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
2592 phys_addr_t paddr, u64 size, int prot)
2594 if (unlikely(domain->ops->domain_window_enable == NULL))
2597 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2600 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2602 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2604 if (unlikely(domain->ops->domain_window_disable == NULL))
2607 return domain->ops->domain_window_disable(domain, wnd_nr);
2609 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2612 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2613 * @domain: the iommu domain where the fault has happened
2614 * @dev: the device where the fault has happened
2615 * @iova: the faulting address
2616 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2618 * This function should be called by the low-level IOMMU implementations
2619 * whenever IOMMU faults happen, to allow high-level users, that are
2620 * interested in such events, to know about them.
2622 * This event may be useful for several possible use cases:
2623 * - mere logging of the event
2624 * - dynamic TLB/PTE loading
2625 * - if restarting of the faulting device is required
2627 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2628 * PTE/TLB loading will one day be supported, implementations will be able
2629 * to tell whether it succeeded or not according to this return value).
2631 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2632 * (though fault handlers can also return -ENOSYS, in case they want to
2633 * elicit the default behavior of the IOMMU drivers).
2635 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2636 unsigned long iova, int flags)
2641 * if upper layers showed interest and installed a fault handler,
2644 if (domain->handler)
2645 ret = domain->handler(domain, dev, iova, flags,
2646 domain->handler_token);
2648 trace_io_page_fault(dev, iova, flags);
2651 EXPORT_SYMBOL_GPL(report_iommu_fault);
2653 static int __init iommu_init(void)
2655 iommu_group_kset = kset_create_and_add("iommu_groups",
2657 BUG_ON(!iommu_group_kset);
2659 iommu_debugfs_setup();
2663 core_initcall(iommu_init);
2665 int iommu_domain_get_attr(struct iommu_domain *domain,
2666 enum iommu_attr attr, void *data)
2668 struct iommu_domain_geometry *geometry;
2673 case DOMAIN_ATTR_GEOMETRY:
2675 *geometry = domain->geometry;
2678 case DOMAIN_ATTR_PAGING:
2680 *paging = (domain->pgsize_bitmap != 0UL);
2683 if (!domain->ops->domain_get_attr)
2686 ret = domain->ops->domain_get_attr(domain, attr, data);
2691 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2693 int iommu_domain_set_attr(struct iommu_domain *domain,
2694 enum iommu_attr attr, void *data)
2700 if (domain->ops->domain_set_attr == NULL)
2703 ret = domain->ops->domain_set_attr(domain, attr, data);
2708 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
2710 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2712 const struct iommu_ops *ops = dev->bus->iommu_ops;
2714 if (ops && ops->get_resv_regions)
2715 ops->get_resv_regions(dev, list);
2718 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2720 const struct iommu_ops *ops = dev->bus->iommu_ops;
2722 if (ops && ops->put_resv_regions)
2723 ops->put_resv_regions(dev, list);
2727 * generic_iommu_put_resv_regions - Reserved region driver helper
2728 * @dev: device for which to free reserved regions
2729 * @list: reserved region list for device
2731 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2732 * for simple reservations. Memory allocated for each reserved region will be
2733 * freed. If an IOMMU driver allocates additional resources per region, it is
2734 * going to have to implement a custom callback.
2736 void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2738 struct iommu_resv_region *entry, *next;
2740 list_for_each_entry_safe(entry, next, list, list)
2743 EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2745 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2746 size_t length, int prot,
2747 enum iommu_resv_type type)
2749 struct iommu_resv_region *region;
2751 region = kzalloc(sizeof(*region), GFP_KERNEL);
2755 INIT_LIST_HEAD(®ion->list);
2756 region->start = start;
2757 region->length = length;
2758 region->prot = prot;
2759 region->type = type;
2762 EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2764 void iommu_set_default_passthrough(bool cmd_line)
2767 iommu_set_cmd_line_dma_api();
2769 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2772 void iommu_set_default_translated(bool cmd_line)
2775 iommu_set_cmd_line_dma_api();
2777 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2780 bool iommu_default_passthrough(void)
2782 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2784 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2786 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2788 const struct iommu_ops *ops = NULL;
2789 struct iommu_device *iommu;
2791 spin_lock(&iommu_device_lock);
2792 list_for_each_entry(iommu, &iommu_device_list, list)
2793 if (iommu->fwnode == fwnode) {
2797 spin_unlock(&iommu_device_lock);
2801 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2802 const struct iommu_ops *ops)
2804 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2807 return ops == fwspec->ops ? 0 : -EINVAL;
2809 if (!dev_iommu_get(dev))
2812 /* Preallocate for the overwhelmingly common case of 1 ID */
2813 fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2817 of_node_get(to_of_node(iommu_fwnode));
2818 fwspec->iommu_fwnode = iommu_fwnode;
2820 dev_iommu_fwspec_set(dev, fwspec);
2823 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2825 void iommu_fwspec_free(struct device *dev)
2827 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2830 fwnode_handle_put(fwspec->iommu_fwnode);
2832 dev_iommu_fwspec_set(dev, NULL);
2835 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2837 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2839 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2845 new_num = fwspec->num_ids + num_ids;
2847 fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2852 dev_iommu_fwspec_set(dev, fwspec);
2855 for (i = 0; i < num_ids; i++)
2856 fwspec->ids[fwspec->num_ids + i] = ids[i];
2858 fwspec->num_ids = new_num;
2861 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2864 * Per device IOMMU features.
2866 bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2868 const struct iommu_ops *ops = dev->bus->iommu_ops;
2870 if (ops && ops->dev_has_feat)
2871 return ops->dev_has_feat(dev, feat);
2875 EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2877 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2879 const struct iommu_ops *ops = dev->bus->iommu_ops;
2881 if (ops && ops->dev_enable_feat)
2882 return ops->dev_enable_feat(dev, feat);
2886 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2889 * The device drivers should do the necessary cleanups before calling this.
2890 * For example, before disabling the aux-domain feature, the device driver
2891 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2893 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2895 const struct iommu_ops *ops = dev->bus->iommu_ops;
2897 if (ops && ops->dev_disable_feat)
2898 return ops->dev_disable_feat(dev, feat);
2902 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2904 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2906 const struct iommu_ops *ops = dev->bus->iommu_ops;
2908 if (ops && ops->dev_feat_enabled)
2909 return ops->dev_feat_enabled(dev, feat);
2913 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2916 * Aux-domain specific attach/detach.
2918 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2919 * true. Also, as long as domains are attached to a device through this
2920 * interface, any tries to call iommu_attach_device() should fail
2921 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2922 * This should make us safe against a device being attached to a guest as a
2923 * whole while there are still pasid users on it (aux and sva).
2925 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2929 if (domain->ops->aux_attach_dev)
2930 ret = domain->ops->aux_attach_dev(domain, dev);
2933 trace_attach_device_to_domain(dev);
2937 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2939 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2941 if (domain->ops->aux_detach_dev) {
2942 domain->ops->aux_detach_dev(domain, dev);
2943 trace_detach_device_from_domain(dev);
2946 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2948 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2952 if (domain->ops->aux_get_pasid)
2953 ret = domain->ops->aux_get_pasid(domain, dev);
2957 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
2960 * iommu_sva_bind_device() - Bind a process address space to a device
2962 * @mm: the mm to bind, caller must hold a reference to it
2964 * Create a bond between device and address space, allowing the device to access
2965 * the mm using the returned PASID. If a bond already exists between @device and
2966 * @mm, it is returned and an additional reference is taken. Caller must call
2967 * iommu_sva_unbind_device() to release each reference.
2969 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2970 * initialize the required SVA features.
2972 * On error, returns an ERR_PTR value.
2975 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2977 struct iommu_group *group;
2978 struct iommu_sva *handle = ERR_PTR(-EINVAL);
2979 const struct iommu_ops *ops = dev->bus->iommu_ops;
2981 if (!ops || !ops->sva_bind)
2982 return ERR_PTR(-ENODEV);
2984 group = iommu_group_get(dev);
2986 return ERR_PTR(-ENODEV);
2988 /* Ensure device count and domain don't change while we're binding */
2989 mutex_lock(&group->mutex);
2992 * To keep things simple, SVA currently doesn't support IOMMU groups
2993 * with more than one device. Existing SVA-capable systems are not
2994 * affected by the problems that required IOMMU groups (lack of ACS
2995 * isolation, device ID aliasing and other hardware issues).
2997 if (iommu_group_device_count(group) != 1)
3000 handle = ops->sva_bind(dev, mm, drvdata);
3003 mutex_unlock(&group->mutex);
3004 iommu_group_put(group);
3008 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
3011 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
3012 * @handle: the handle returned by iommu_sva_bind_device()
3014 * Put reference to a bond between device and address space. The device should
3015 * not be issuing any more transaction for this PASID. All outstanding page
3016 * requests for this PASID must have been flushed to the IOMMU.
3018 void iommu_sva_unbind_device(struct iommu_sva *handle)
3020 struct iommu_group *group;
3021 struct device *dev = handle->dev;
3022 const struct iommu_ops *ops = dev->bus->iommu_ops;
3024 if (!ops || !ops->sva_unbind)
3027 group = iommu_group_get(dev);
3031 mutex_lock(&group->mutex);
3032 ops->sva_unbind(handle);
3033 mutex_unlock(&group->mutex);
3035 iommu_group_put(group);
3037 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
3039 u32 iommu_sva_get_pasid(struct iommu_sva *handle)
3041 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
3043 if (!ops || !ops->sva_get_pasid)
3044 return IOMMU_PASID_INVALID;
3046 return ops->sva_get_pasid(handle);
3048 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
3051 * Changes the default domain of an iommu group that has *only* one device
3053 * @group: The group for which the default domain should be changed
3054 * @prev_dev: The device in the group (this is used to make sure that the device
3055 * hasn't changed after the caller has called this function)
3056 * @type: The type of the new default domain that gets associated with the group
3058 * Returns 0 on success and error code on failure
3061 * 1. Presently, this function is called only when user requests to change the
3062 * group's default domain type through /sys/kernel/iommu_groups/<grp_id>/type
3063 * Please take a closer look if intended to use for other purposes.
3065 static int iommu_change_dev_def_domain(struct iommu_group *group,
3066 struct device *prev_dev, int type)
3068 struct iommu_domain *prev_dom;
3069 struct group_device *grp_dev;
3070 int ret, dev_def_dom;
3076 mutex_lock(&group->mutex);
3078 if (group->default_domain != group->domain) {
3079 dev_err_ratelimited(prev_dev, "Group not assigned to default domain\n");
3085 * iommu group wasn't locked while acquiring device lock in
3086 * iommu_group_store_type(). So, make sure that the device count hasn't
3087 * changed while acquiring device lock.
3089 * Changing default domain of an iommu group with two or more devices
3090 * isn't supported because there could be a potential deadlock. Consider
3091 * the following scenario. T1 is trying to acquire device locks of all
3092 * the devices in the group and before it could acquire all of them,
3093 * there could be another thread T2 (from different sub-system and use
3094 * case) that has already acquired some of the device locks and might be
3095 * waiting for T1 to release other device locks.
3097 if (iommu_group_device_count(group) != 1) {
3098 dev_err_ratelimited(prev_dev, "Cannot change default domain: Group has more than one device\n");
3103 /* Since group has only one device */
3104 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3107 if (prev_dev != dev) {
3108 dev_err_ratelimited(prev_dev, "Cannot change default domain: Device has been changed\n");
3113 prev_dom = group->default_domain;
3119 dev_def_dom = iommu_get_def_domain_type(dev);
3122 * If the user hasn't requested any specific type of domain and
3123 * if the device supports both the domains, then default to the
3124 * domain the device was booted with
3126 type = dev_def_dom ? : iommu_def_domain_type;
3127 } else if (dev_def_dom && type != dev_def_dom) {
3128 dev_err_ratelimited(prev_dev, "Device cannot be in %s domain\n",
3129 iommu_domain_type_str(type));
3135 * Switch to a new domain only if the requested domain type is different
3136 * from the existing default domain type
3138 if (prev_dom->type == type) {
3143 /* Sets group->default_domain to the newly allocated domain */
3144 ret = iommu_group_alloc_default_domain(dev->bus, group, type);
3148 ret = iommu_create_device_direct_mappings(group, dev);
3150 goto free_new_domain;
3152 ret = __iommu_attach_device(group->default_domain, dev);
3154 goto free_new_domain;
3156 group->domain = group->default_domain;
3159 * Release the mutex here because ops->probe_finalize() call-back of
3160 * some vendor IOMMU drivers calls arm_iommu_attach_device() which
3161 * in-turn might call back into IOMMU core code, where it tries to take
3162 * group->mutex, resulting in a deadlock.
3164 mutex_unlock(&group->mutex);
3166 /* Make sure dma_ops is appropriatley set */
3167 iommu_group_do_probe_finalize(dev, group->default_domain);
3168 iommu_domain_free(prev_dom);
3172 iommu_domain_free(group->default_domain);
3173 group->default_domain = prev_dom;
3174 group->domain = prev_dom;
3177 mutex_unlock(&group->mutex);
3183 * Changing the default domain through sysfs requires the users to ubind the
3184 * drivers from the devices in the iommu group. Return failure if this doesn't
3187 * We need to consider the race between this and the device release path.
3188 * device_lock(dev) is used here to guarantee that the device release path
3189 * will not be entered at the same time.
3191 static ssize_t iommu_group_store_type(struct iommu_group *group,
3192 const char *buf, size_t count)
3194 struct group_device *grp_dev;
3198 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
3201 if (WARN_ON(!group))
3204 if (sysfs_streq(buf, "identity"))
3205 req_type = IOMMU_DOMAIN_IDENTITY;
3206 else if (sysfs_streq(buf, "DMA"))
3207 req_type = IOMMU_DOMAIN_DMA;
3208 else if (sysfs_streq(buf, "auto"))
3214 * Lock/Unlock the group mutex here before device lock to
3215 * 1. Make sure that the iommu group has only one device (this is a
3216 * prerequisite for step 2)
3217 * 2. Get struct *dev which is needed to lock device
3219 mutex_lock(&group->mutex);
3220 if (iommu_group_device_count(group) != 1) {
3221 mutex_unlock(&group->mutex);
3222 pr_err_ratelimited("Cannot change default domain: Group has more than one device\n");
3226 /* Since group has only one device */
3227 grp_dev = list_first_entry(&group->devices, struct group_device, list);
3232 * Don't hold the group mutex because taking group mutex first and then
3233 * the device lock could potentially cause a deadlock as below. Assume
3234 * two threads T1 and T2. T1 is trying to change default domain of an
3235 * iommu group and T2 is trying to hot unplug a device or release [1] VF
3236 * of a PCIe device which is in the same iommu group. T1 takes group
3237 * mutex and before it could take device lock assume T2 has taken device
3238 * lock and is yet to take group mutex. Now, both the threads will be
3239 * waiting for the other thread to release lock. Below, lock order was
3242 * mutex_lock(&group->mutex);
3243 * iommu_change_dev_def_domain();
3244 * mutex_unlock(&group->mutex);
3245 * device_unlock(dev);
3247 * [1] Typical device release path
3248 * device_lock() from device/driver core code
3250 * -> iommu_bus_notifier()
3251 * -> iommu_release_device()
3252 * -> ops->release_device() vendor driver calls back iommu core code
3253 * -> mutex_lock() from iommu core code
3255 mutex_unlock(&group->mutex);
3257 /* Check if the device in the group still has a driver bound to it */
3259 if (device_is_bound(dev)) {
3260 pr_err_ratelimited("Device is still bound to driver\n");
3265 ret = iommu_change_dev_def_domain(group, dev, req_type);