xref: /kernel/linux/linux-5.10/drivers/iommu/iommu.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
4 * Author: Joerg Roedel <jroedel@suse.de>
5 */
6
7#define pr_fmt(fmt)    "iommu: " fmt
8
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>
27
28static struct kset *iommu_group_kset;
29static DEFINE_IDA(iommu_group_ida);
30
31static unsigned int iommu_def_domain_type __read_mostly;
32static bool iommu_dma_strict __read_mostly = true;
33static u32 iommu_cmd_line __read_mostly;
34
35struct iommu_group {
36	struct kobject kobj;
37	struct kobject *devices_kobj;
38	struct list_head devices;
39	struct mutex mutex;
40	struct blocking_notifier_head notifier;
41	void *iommu_data;
42	void (*iommu_data_release)(void *iommu_data);
43	char *name;
44	int id;
45	struct iommu_domain *default_domain;
46	struct iommu_domain *domain;
47	struct list_head entry;
48};
49
50struct group_device {
51	struct list_head list;
52	struct device *dev;
53	char *name;
54};
55
56struct 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);
61};
62
63static 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",
69};
70
71#define IOMMU_CMD_LINE_DMA_API		BIT(0)
72
73static void iommu_set_cmd_line_dma_api(void)
74{
75	iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
76}
77
78static bool iommu_cmd_line_dma_api(void)
79{
80	return !!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API);
81}
82
83static int iommu_alloc_default_domain(struct iommu_group *group,
84				      struct device *dev);
85static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
86						 unsigned type);
87static int __iommu_attach_device(struct iommu_domain *domain,
88				 struct device *dev);
89static int __iommu_attach_group(struct iommu_domain *domain,
90				struct iommu_group *group);
91static void __iommu_detach_group(struct iommu_domain *domain,
92				 struct iommu_group *group);
93static int iommu_create_device_direct_mappings(struct iommu_group *group,
94					       struct device *dev);
95static struct iommu_group *iommu_group_get_for_dev(struct device *dev);
96
97#define IOMMU_GROUP_ATTR(_name, _mode, _show, _store)		\
98struct iommu_group_attribute iommu_group_attr_##_name =		\
99	__ATTR(_name, _mode, _show, _store)
100
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)
105
106static LIST_HEAD(iommu_device_list);
107static DEFINE_SPINLOCK(iommu_device_lock);
108
109/*
110 * Use a function instead of an array here because the domain-type is a
111 * bit-field, so an array would waste memory.
112 */
113static const char *iommu_domain_type_str(unsigned int t)
114{
115	switch (t) {
116	case IOMMU_DOMAIN_BLOCKED:
117		return "Blocked";
118	case IOMMU_DOMAIN_IDENTITY:
119		return "Passthrough";
120	case IOMMU_DOMAIN_UNMANAGED:
121		return "Unmanaged";
122	case IOMMU_DOMAIN_DMA:
123		return "Translated";
124	default:
125		return "Unknown";
126	}
127}
128
129static int __init iommu_subsys_init(void)
130{
131	bool cmd_line = iommu_cmd_line_dma_api();
132
133	if (!cmd_line) {
134		if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
135			iommu_set_default_passthrough(false);
136		else
137			iommu_set_default_translated(false);
138
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);
142		}
143	}
144
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)" : "");
148
149	return 0;
150}
151subsys_initcall(iommu_subsys_init);
152
153int iommu_device_register(struct iommu_device *iommu)
154{
155	spin_lock(&iommu_device_lock);
156	list_add_tail(&iommu->list, &iommu_device_list);
157	spin_unlock(&iommu_device_lock);
158	return 0;
159}
160EXPORT_SYMBOL_GPL(iommu_device_register);
161
162void iommu_device_unregister(struct iommu_device *iommu)
163{
164	spin_lock(&iommu_device_lock);
165	list_del(&iommu->list);
166	spin_unlock(&iommu_device_lock);
167}
168EXPORT_SYMBOL_GPL(iommu_device_unregister);
169
170static struct dev_iommu *dev_iommu_get(struct device *dev)
171{
172	struct dev_iommu *param = dev->iommu;
173
174	if (param)
175		return param;
176
177	param = kzalloc(sizeof(*param), GFP_KERNEL);
178	if (!param)
179		return NULL;
180
181	mutex_init(&param->lock);
182	dev->iommu = param;
183	return param;
184}
185
186static void dev_iommu_free(struct device *dev)
187{
188	struct dev_iommu *param = dev->iommu;
189
190	dev->iommu = NULL;
191	if (param->fwspec) {
192		fwnode_handle_put(param->fwspec->iommu_fwnode);
193		kfree(param->fwspec);
194	}
195	kfree(param);
196}
197
198static int __iommu_probe_device(struct device *dev, struct list_head *group_list)
199{
200	const struct iommu_ops *ops = dev->bus->iommu_ops;
201	struct iommu_device *iommu_dev;
202	struct iommu_group *group;
203	int ret;
204
205	if (!ops)
206		return -ENODEV;
207
208	if (!dev_iommu_get(dev))
209		return -ENOMEM;
210
211	if (!try_module_get(ops->owner)) {
212		ret = -EINVAL;
213		goto err_free;
214	}
215
216	iommu_dev = ops->probe_device(dev);
217	if (IS_ERR(iommu_dev)) {
218		ret = PTR_ERR(iommu_dev);
219		goto out_module_put;
220	}
221
222	dev->iommu->iommu_dev = iommu_dev;
223
224	group = iommu_group_get_for_dev(dev);
225	if (IS_ERR(group)) {
226		ret = PTR_ERR(group);
227		goto out_release;
228	}
229	iommu_group_put(group);
230
231	if (group_list && !group->default_domain && list_empty(&group->entry))
232		list_add_tail(&group->entry, group_list);
233
234	iommu_device_link(iommu_dev, dev);
235
236	return 0;
237
238out_release:
239	ops->release_device(dev);
240
241out_module_put:
242	module_put(ops->owner);
243
244err_free:
245	dev_iommu_free(dev);
246
247	return ret;
248}
249
250int iommu_probe_device(struct device *dev)
251{
252	const struct iommu_ops *ops = dev->bus->iommu_ops;
253	struct iommu_group *group;
254	int ret;
255
256	ret = __iommu_probe_device(dev, NULL);
257	if (ret)
258		goto err_out;
259
260	group = iommu_group_get(dev);
261	if (!group)
262		goto err_release;
263
264	/*
265	 * Try to allocate a default domain - needs support from the
266	 * IOMMU driver. There are still some drivers which don't
267	 * support default domains, so the return value is not yet
268	 * checked.
269	 */
270	iommu_alloc_default_domain(group, dev);
271
272	if (group->default_domain) {
273		ret = __iommu_attach_device(group->default_domain, dev);
274		if (ret) {
275			iommu_group_put(group);
276			goto err_release;
277		}
278	}
279
280	iommu_create_device_direct_mappings(group, dev);
281
282	iommu_group_put(group);
283
284	if (ops->probe_finalize)
285		ops->probe_finalize(dev);
286
287	return 0;
288
289err_release:
290	iommu_release_device(dev);
291
292err_out:
293	return ret;
294
295}
296
297void iommu_release_device(struct device *dev)
298{
299	const struct iommu_ops *ops = dev->bus->iommu_ops;
300
301	if (!dev->iommu)
302		return;
303
304	iommu_device_unlink(dev->iommu->iommu_dev, dev);
305
306	ops->release_device(dev);
307
308	iommu_group_remove_device(dev);
309	module_put(ops->owner);
310	dev_iommu_free(dev);
311}
312
313static int __init iommu_set_def_domain_type(char *str)
314{
315	bool pt;
316	int ret;
317
318	ret = kstrtobool(str, &pt);
319	if (ret)
320		return ret;
321
322	if (pt)
323		iommu_set_default_passthrough(true);
324	else
325		iommu_set_default_translated(true);
326
327	return 0;
328}
329early_param("iommu.passthrough", iommu_set_def_domain_type);
330
331static int __init iommu_dma_setup(char *str)
332{
333	return kstrtobool(str, &iommu_dma_strict);
334}
335early_param("iommu.strict", iommu_dma_setup);
336
337static ssize_t iommu_group_attr_show(struct kobject *kobj,
338				     struct attribute *__attr, char *buf)
339{
340	struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
341	struct iommu_group *group = to_iommu_group(kobj);
342	ssize_t ret = -EIO;
343
344	if (attr->show)
345		ret = attr->show(group, buf);
346	return ret;
347}
348
349static ssize_t iommu_group_attr_store(struct kobject *kobj,
350				      struct attribute *__attr,
351				      const char *buf, size_t count)
352{
353	struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
354	struct iommu_group *group = to_iommu_group(kobj);
355	ssize_t ret = -EIO;
356
357	if (attr->store)
358		ret = attr->store(group, buf, count);
359	return ret;
360}
361
362static const struct sysfs_ops iommu_group_sysfs_ops = {
363	.show = iommu_group_attr_show,
364	.store = iommu_group_attr_store,
365};
366
367static int iommu_group_create_file(struct iommu_group *group,
368				   struct iommu_group_attribute *attr)
369{
370	return sysfs_create_file(&group->kobj, &attr->attr);
371}
372
373static void iommu_group_remove_file(struct iommu_group *group,
374				    struct iommu_group_attribute *attr)
375{
376	sysfs_remove_file(&group->kobj, &attr->attr);
377}
378
379static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
380{
381	return sprintf(buf, "%s\n", group->name);
382}
383
384/**
385 * iommu_insert_resv_region - Insert a new region in the
386 * list of reserved regions.
387 * @new: new region to insert
388 * @regions: list of regions
389 *
390 * Elements are sorted by start address and overlapping segments
391 * of the same type are merged.
392 */
393static int iommu_insert_resv_region(struct iommu_resv_region *new,
394				    struct list_head *regions)
395{
396	struct iommu_resv_region *iter, *tmp, *nr, *top;
397	LIST_HEAD(stack);
398
399	nr = iommu_alloc_resv_region(new->start, new->length,
400				     new->prot, new->type);
401	if (!nr)
402		return -ENOMEM;
403
404	/* First add the new element based on start address sorting */
405	list_for_each_entry(iter, regions, list) {
406		if (nr->start < iter->start ||
407		    (nr->start == iter->start && nr->type <= iter->type))
408			break;
409	}
410	list_add_tail(&nr->list, &iter->list);
411
412	/* Merge overlapping segments of type nr->type in @regions, if any */
413	list_for_each_entry_safe(iter, tmp, regions, list) {
414		phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
415
416		/* no merge needed on elements of different types than @new */
417		if (iter->type != new->type) {
418			list_move_tail(&iter->list, &stack);
419			continue;
420		}
421
422		/* look for the last stack element of same type as @iter */
423		list_for_each_entry_reverse(top, &stack, list)
424			if (top->type == iter->type)
425				goto check_overlap;
426
427		list_move_tail(&iter->list, &stack);
428		continue;
429
430check_overlap:
431		top_end = top->start + top->length - 1;
432
433		if (iter->start > top_end + 1) {
434			list_move_tail(&iter->list, &stack);
435		} else {
436			top->length = max(top_end, iter_end) - top->start + 1;
437			list_del(&iter->list);
438			kfree(iter);
439		}
440	}
441	list_splice(&stack, regions);
442	return 0;
443}
444
445static int
446iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
447				 struct list_head *group_resv_regions)
448{
449	struct iommu_resv_region *entry;
450	int ret = 0;
451
452	list_for_each_entry(entry, dev_resv_regions, list) {
453		ret = iommu_insert_resv_region(entry, group_resv_regions);
454		if (ret)
455			break;
456	}
457	return ret;
458}
459
460int iommu_get_group_resv_regions(struct iommu_group *group,
461				 struct list_head *head)
462{
463	struct group_device *device;
464	int ret = 0;
465
466	mutex_lock(&group->mutex);
467	list_for_each_entry(device, &group->devices, list) {
468		struct list_head dev_resv_regions;
469
470		INIT_LIST_HEAD(&dev_resv_regions);
471		iommu_get_resv_regions(device->dev, &dev_resv_regions);
472		ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
473		iommu_put_resv_regions(device->dev, &dev_resv_regions);
474		if (ret)
475			break;
476	}
477	mutex_unlock(&group->mutex);
478	return ret;
479}
480EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
481
482static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
483					     char *buf)
484{
485	struct iommu_resv_region *region, *next;
486	struct list_head group_resv_regions;
487	char *str = buf;
488
489	INIT_LIST_HEAD(&group_resv_regions);
490	iommu_get_group_resv_regions(group, &group_resv_regions);
491
492	list_for_each_entry_safe(region, next, &group_resv_regions, list) {
493		str += sprintf(str, "0x%016llx 0x%016llx %s\n",
494			       (long long int)region->start,
495			       (long long int)(region->start +
496						region->length - 1),
497			       iommu_group_resv_type_string[region->type]);
498		kfree(region);
499	}
500
501	return (str - buf);
502}
503
504static ssize_t iommu_group_show_type(struct iommu_group *group,
505				     char *buf)
506{
507	char *type = "unknown\n";
508
509	if (group->default_domain) {
510		switch (group->default_domain->type) {
511		case IOMMU_DOMAIN_BLOCKED:
512			type = "blocked\n";
513			break;
514		case IOMMU_DOMAIN_IDENTITY:
515			type = "identity\n";
516			break;
517		case IOMMU_DOMAIN_UNMANAGED:
518			type = "unmanaged\n";
519			break;
520		case IOMMU_DOMAIN_DMA:
521			type = "DMA\n";
522			break;
523		}
524	}
525	strcpy(buf, type);
526
527	return strlen(type);
528}
529
530static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
531
532static IOMMU_GROUP_ATTR(reserved_regions, 0444,
533			iommu_group_show_resv_regions, NULL);
534
535static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
536
537static void iommu_group_release(struct kobject *kobj)
538{
539	struct iommu_group *group = to_iommu_group(kobj);
540
541	pr_debug("Releasing group %d\n", group->id);
542
543	if (group->iommu_data_release)
544		group->iommu_data_release(group->iommu_data);
545
546	ida_simple_remove(&iommu_group_ida, group->id);
547
548	if (group->default_domain)
549		iommu_domain_free(group->default_domain);
550
551	kfree(group->name);
552	kfree(group);
553}
554
555static struct kobj_type iommu_group_ktype = {
556	.sysfs_ops = &iommu_group_sysfs_ops,
557	.release = iommu_group_release,
558};
559
560/**
561 * iommu_group_alloc - Allocate a new group
562 *
563 * This function is called by an iommu driver to allocate a new iommu
564 * group.  The iommu group represents the minimum granularity of the iommu.
565 * Upon successful return, the caller holds a reference to the supplied
566 * group in order to hold the group until devices are added.  Use
567 * iommu_group_put() to release this extra reference count, allowing the
568 * group to be automatically reclaimed once it has no devices or external
569 * references.
570 */
571struct iommu_group *iommu_group_alloc(void)
572{
573	struct iommu_group *group;
574	int ret;
575
576	group = kzalloc(sizeof(*group), GFP_KERNEL);
577	if (!group)
578		return ERR_PTR(-ENOMEM);
579
580	group->kobj.kset = iommu_group_kset;
581	mutex_init(&group->mutex);
582	INIT_LIST_HEAD(&group->devices);
583	INIT_LIST_HEAD(&group->entry);
584	BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
585
586	ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
587	if (ret < 0) {
588		kfree(group);
589		return ERR_PTR(ret);
590	}
591	group->id = ret;
592
593	ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
594				   NULL, "%d", group->id);
595	if (ret) {
596		ida_simple_remove(&iommu_group_ida, group->id);
597		kobject_put(&group->kobj);
598		return ERR_PTR(ret);
599	}
600
601	group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
602	if (!group->devices_kobj) {
603		kobject_put(&group->kobj); /* triggers .release & free */
604		return ERR_PTR(-ENOMEM);
605	}
606
607	/*
608	 * The devices_kobj holds a reference on the group kobject, so
609	 * as long as that exists so will the group.  We can therefore
610	 * use the devices_kobj for reference counting.
611	 */
612	kobject_put(&group->kobj);
613
614	ret = iommu_group_create_file(group,
615				      &iommu_group_attr_reserved_regions);
616	if (ret)
617		return ERR_PTR(ret);
618
619	ret = iommu_group_create_file(group, &iommu_group_attr_type);
620	if (ret)
621		return ERR_PTR(ret);
622
623	pr_debug("Allocated group %d\n", group->id);
624
625	return group;
626}
627EXPORT_SYMBOL_GPL(iommu_group_alloc);
628
629struct iommu_group *iommu_group_get_by_id(int id)
630{
631	struct kobject *group_kobj;
632	struct iommu_group *group;
633	const char *name;
634
635	if (!iommu_group_kset)
636		return NULL;
637
638	name = kasprintf(GFP_KERNEL, "%d", id);
639	if (!name)
640		return NULL;
641
642	group_kobj = kset_find_obj(iommu_group_kset, name);
643	kfree(name);
644
645	if (!group_kobj)
646		return NULL;
647
648	group = container_of(group_kobj, struct iommu_group, kobj);
649	BUG_ON(group->id != id);
650
651	kobject_get(group->devices_kobj);
652	kobject_put(&group->kobj);
653
654	return group;
655}
656EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
657
658/**
659 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
660 * @group: the group
661 *
662 * iommu drivers can store data in the group for use when doing iommu
663 * operations.  This function provides a way to retrieve it.  Caller
664 * should hold a group reference.
665 */
666void *iommu_group_get_iommudata(struct iommu_group *group)
667{
668	return group->iommu_data;
669}
670EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
671
672/**
673 * iommu_group_set_iommudata - set iommu_data for a group
674 * @group: the group
675 * @iommu_data: new data
676 * @release: release function for iommu_data
677 *
678 * iommu drivers can store data in the group for use when doing iommu
679 * operations.  This function provides a way to set the data after
680 * the group has been allocated.  Caller should hold a group reference.
681 */
682void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
683			       void (*release)(void *iommu_data))
684{
685	group->iommu_data = iommu_data;
686	group->iommu_data_release = release;
687}
688EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
689
690/**
691 * iommu_group_set_name - set name for a group
692 * @group: the group
693 * @name: name
694 *
695 * Allow iommu driver to set a name for a group.  When set it will
696 * appear in a name attribute file under the group in sysfs.
697 */
698int iommu_group_set_name(struct iommu_group *group, const char *name)
699{
700	int ret;
701
702	if (group->name) {
703		iommu_group_remove_file(group, &iommu_group_attr_name);
704		kfree(group->name);
705		group->name = NULL;
706		if (!name)
707			return 0;
708	}
709
710	group->name = kstrdup(name, GFP_KERNEL);
711	if (!group->name)
712		return -ENOMEM;
713
714	ret = iommu_group_create_file(group, &iommu_group_attr_name);
715	if (ret) {
716		kfree(group->name);
717		group->name = NULL;
718		return ret;
719	}
720
721	return 0;
722}
723EXPORT_SYMBOL_GPL(iommu_group_set_name);
724
725static int iommu_create_device_direct_mappings(struct iommu_group *group,
726					       struct device *dev)
727{
728	struct iommu_domain *domain = group->default_domain;
729	struct iommu_resv_region *entry;
730	struct list_head mappings;
731	unsigned long pg_size;
732	int ret = 0;
733
734	if (!domain || domain->type != IOMMU_DOMAIN_DMA)
735		return 0;
736
737	BUG_ON(!domain->pgsize_bitmap);
738
739	pg_size = 1UL << __ffs(domain->pgsize_bitmap);
740	INIT_LIST_HEAD(&mappings);
741
742	iommu_get_resv_regions(dev, &mappings);
743
744	/* We need to consider overlapping regions for different devices */
745	list_for_each_entry(entry, &mappings, list) {
746		dma_addr_t start, end, addr;
747
748		if (domain->ops->apply_resv_region)
749			domain->ops->apply_resv_region(dev, domain, entry);
750
751		start = ALIGN(entry->start, pg_size);
752		end   = ALIGN(entry->start + entry->length, pg_size);
753
754		if (entry->type != IOMMU_RESV_DIRECT &&
755		    entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
756			continue;
757
758		for (addr = start; addr < end; addr += pg_size) {
759			phys_addr_t phys_addr;
760
761			phys_addr = iommu_iova_to_phys(domain, addr);
762			if (phys_addr)
763				continue;
764
765			ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
766			if (ret)
767				goto out;
768		}
769
770	}
771
772	iommu_flush_iotlb_all(domain);
773
774out:
775	iommu_put_resv_regions(dev, &mappings);
776
777	return ret;
778}
779
780static bool iommu_is_attach_deferred(struct iommu_domain *domain,
781				     struct device *dev)
782{
783	if (domain->ops->is_attach_deferred)
784		return domain->ops->is_attach_deferred(domain, dev);
785
786	return false;
787}
788
789/**
790 * iommu_group_add_device - add a device to an iommu group
791 * @group: the group into which to add the device (reference should be held)
792 * @dev: the device
793 *
794 * This function is called by an iommu driver to add a device into a
795 * group.  Adding a device increments the group reference count.
796 */
797int iommu_group_add_device(struct iommu_group *group, struct device *dev)
798{
799	int ret, i = 0;
800	struct group_device *device;
801
802	device = kzalloc(sizeof(*device), GFP_KERNEL);
803	if (!device)
804		return -ENOMEM;
805
806	device->dev = dev;
807
808	ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
809	if (ret)
810		goto err_free_device;
811
812	device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
813rename:
814	if (!device->name) {
815		ret = -ENOMEM;
816		goto err_remove_link;
817	}
818
819	ret = sysfs_create_link_nowarn(group->devices_kobj,
820				       &dev->kobj, device->name);
821	if (ret) {
822		if (ret == -EEXIST && i >= 0) {
823			/*
824			 * Account for the slim chance of collision
825			 * and append an instance to the name.
826			 */
827			kfree(device->name);
828			device->name = kasprintf(GFP_KERNEL, "%s.%d",
829						 kobject_name(&dev->kobj), i++);
830			goto rename;
831		}
832		goto err_free_name;
833	}
834
835	kobject_get(group->devices_kobj);
836
837	dev->iommu_group = group;
838
839	mutex_lock(&group->mutex);
840	list_add_tail(&device->list, &group->devices);
841	if (group->domain  && !iommu_is_attach_deferred(group->domain, dev))
842		ret = __iommu_attach_device(group->domain, dev);
843	mutex_unlock(&group->mutex);
844	if (ret)
845		goto err_put_group;
846
847	/* Notify any listeners about change to group. */
848	blocking_notifier_call_chain(&group->notifier,
849				     IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
850
851	trace_add_device_to_group(group->id, dev);
852
853	dev_info(dev, "Adding to iommu group %d\n", group->id);
854
855	return 0;
856
857err_put_group:
858	mutex_lock(&group->mutex);
859	list_del(&device->list);
860	mutex_unlock(&group->mutex);
861	dev->iommu_group = NULL;
862	kobject_put(group->devices_kobj);
863	sysfs_remove_link(group->devices_kobj, device->name);
864err_free_name:
865	kfree(device->name);
866err_remove_link:
867	sysfs_remove_link(&dev->kobj, "iommu_group");
868err_free_device:
869	kfree(device);
870	dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
871	return ret;
872}
873EXPORT_SYMBOL_GPL(iommu_group_add_device);
874
875/**
876 * iommu_group_remove_device - remove a device from it's current group
877 * @dev: device to be removed
878 *
879 * This function is called by an iommu driver to remove the device from
880 * it's current group.  This decrements the iommu group reference count.
881 */
882void iommu_group_remove_device(struct device *dev)
883{
884	struct iommu_group *group = dev->iommu_group;
885	struct group_device *tmp_device, *device = NULL;
886
887	if (!group)
888		return;
889
890	dev_info(dev, "Removing from iommu group %d\n", group->id);
891
892	/* Pre-notify listeners that a device is being removed. */
893	blocking_notifier_call_chain(&group->notifier,
894				     IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
895
896	mutex_lock(&group->mutex);
897	list_for_each_entry(tmp_device, &group->devices, list) {
898		if (tmp_device->dev == dev) {
899			device = tmp_device;
900			list_del(&device->list);
901			break;
902		}
903	}
904	mutex_unlock(&group->mutex);
905
906	if (!device)
907		return;
908
909	sysfs_remove_link(group->devices_kobj, device->name);
910	sysfs_remove_link(&dev->kobj, "iommu_group");
911
912	trace_remove_device_from_group(group->id, dev);
913
914	kfree(device->name);
915	kfree(device);
916	dev->iommu_group = NULL;
917	kobject_put(group->devices_kobj);
918}
919EXPORT_SYMBOL_GPL(iommu_group_remove_device);
920
921static int iommu_group_device_count(struct iommu_group *group)
922{
923	struct group_device *entry;
924	int ret = 0;
925
926	list_for_each_entry(entry, &group->devices, list)
927		ret++;
928
929	return ret;
930}
931
932/**
933 * iommu_group_for_each_dev - iterate over each device in the group
934 * @group: the group
935 * @data: caller opaque data to be passed to callback function
936 * @fn: caller supplied callback function
937 *
938 * This function is called by group users to iterate over group devices.
939 * Callers should hold a reference count to the group during callback.
940 * The group->mutex is held across callbacks, which will block calls to
941 * iommu_group_add/remove_device.
942 */
943static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
944				      int (*fn)(struct device *, void *))
945{
946	struct group_device *device;
947	int ret = 0;
948
949	list_for_each_entry(device, &group->devices, list) {
950		ret = fn(device->dev, data);
951		if (ret)
952			break;
953	}
954	return ret;
955}
956
957
958int iommu_group_for_each_dev(struct iommu_group *group, void *data,
959			     int (*fn)(struct device *, void *))
960{
961	int ret;
962
963	mutex_lock(&group->mutex);
964	ret = __iommu_group_for_each_dev(group, data, fn);
965	mutex_unlock(&group->mutex);
966
967	return ret;
968}
969EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
970
971/**
972 * iommu_group_get - Return the group for a device and increment reference
973 * @dev: get the group that this device belongs to
974 *
975 * This function is called by iommu drivers and users to get the group
976 * for the specified device.  If found, the group is returned and the group
977 * reference in incremented, else NULL.
978 */
979struct iommu_group *iommu_group_get(struct device *dev)
980{
981	struct iommu_group *group = dev->iommu_group;
982
983	if (group)
984		kobject_get(group->devices_kobj);
985
986	return group;
987}
988EXPORT_SYMBOL_GPL(iommu_group_get);
989
990/**
991 * iommu_group_ref_get - Increment reference on a group
992 * @group: the group to use, must not be NULL
993 *
994 * This function is called by iommu drivers to take additional references on an
995 * existing group.  Returns the given group for convenience.
996 */
997struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
998{
999	kobject_get(group->devices_kobj);
1000	return group;
1001}
1002EXPORT_SYMBOL_GPL(iommu_group_ref_get);
1003
1004/**
1005 * iommu_group_put - Decrement group reference
1006 * @group: the group to use
1007 *
1008 * This function is called by iommu drivers and users to release the
1009 * iommu group.  Once the reference count is zero, the group is released.
1010 */
1011void iommu_group_put(struct iommu_group *group)
1012{
1013	if (group)
1014		kobject_put(group->devices_kobj);
1015}
1016EXPORT_SYMBOL_GPL(iommu_group_put);
1017
1018/**
1019 * iommu_group_register_notifier - Register a notifier for group changes
1020 * @group: the group to watch
1021 * @nb: notifier block to signal
1022 *
1023 * This function allows iommu group users to track changes in a group.
1024 * See include/linux/iommu.h for actions sent via this notifier.  Caller
1025 * should hold a reference to the group throughout notifier registration.
1026 */
1027int iommu_group_register_notifier(struct iommu_group *group,
1028				  struct notifier_block *nb)
1029{
1030	return blocking_notifier_chain_register(&group->notifier, nb);
1031}
1032EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
1033
1034/**
1035 * iommu_group_unregister_notifier - Unregister a notifier
1036 * @group: the group to watch
1037 * @nb: notifier block to signal
1038 *
1039 * Unregister a previously registered group notifier block.
1040 */
1041int iommu_group_unregister_notifier(struct iommu_group *group,
1042				    struct notifier_block *nb)
1043{
1044	return blocking_notifier_chain_unregister(&group->notifier, nb);
1045}
1046EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
1047
1048/**
1049 * iommu_register_device_fault_handler() - Register a device fault handler
1050 * @dev: the device
1051 * @handler: the fault handler
1052 * @data: private data passed as argument to the handler
1053 *
1054 * When an IOMMU fault event is received, this handler gets called with the
1055 * fault event and data as argument. The handler should return 0 on success. If
1056 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
1057 * complete the fault by calling iommu_page_response() with one of the following
1058 * response code:
1059 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
1060 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
1061 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
1062 *   page faults if possible.
1063 *
1064 * Return 0 if the fault handler was installed successfully, or an error.
1065 */
1066int iommu_register_device_fault_handler(struct device *dev,
1067					iommu_dev_fault_handler_t handler,
1068					void *data)
1069{
1070	struct dev_iommu *param = dev->iommu;
1071	int ret = 0;
1072
1073	if (!param)
1074		return -EINVAL;
1075
1076	mutex_lock(&param->lock);
1077	/* Only allow one fault handler registered for each device */
1078	if (param->fault_param) {
1079		ret = -EBUSY;
1080		goto done_unlock;
1081	}
1082
1083	get_device(dev);
1084	param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
1085	if (!param->fault_param) {
1086		put_device(dev);
1087		ret = -ENOMEM;
1088		goto done_unlock;
1089	}
1090	param->fault_param->handler = handler;
1091	param->fault_param->data = data;
1092	mutex_init(&param->fault_param->lock);
1093	INIT_LIST_HEAD(&param->fault_param->faults);
1094
1095done_unlock:
1096	mutex_unlock(&param->lock);
1097
1098	return ret;
1099}
1100EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
1101
1102/**
1103 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
1104 * @dev: the device
1105 *
1106 * Remove the device fault handler installed with
1107 * iommu_register_device_fault_handler().
1108 *
1109 * Return 0 on success, or an error.
1110 */
1111int iommu_unregister_device_fault_handler(struct device *dev)
1112{
1113	struct dev_iommu *param = dev->iommu;
1114	int ret = 0;
1115
1116	if (!param)
1117		return -EINVAL;
1118
1119	mutex_lock(&param->lock);
1120
1121	if (!param->fault_param)
1122		goto unlock;
1123
1124	/* we cannot unregister handler if there are pending faults */
1125	if (!list_empty(&param->fault_param->faults)) {
1126		ret = -EBUSY;
1127		goto unlock;
1128	}
1129
1130	kfree(param->fault_param);
1131	param->fault_param = NULL;
1132	put_device(dev);
1133unlock:
1134	mutex_unlock(&param->lock);
1135
1136	return ret;
1137}
1138EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1139
1140/**
1141 * iommu_report_device_fault() - Report fault event to device driver
1142 * @dev: the device
1143 * @evt: fault event data
1144 *
1145 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1146 * handler. When this function fails and the fault is recoverable, it is the
1147 * caller's responsibility to complete the fault.
1148 *
1149 * Return 0 on success, or an error.
1150 */
1151int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1152{
1153	struct dev_iommu *param = dev->iommu;
1154	struct iommu_fault_event *evt_pending = NULL;
1155	struct iommu_fault_param *fparam;
1156	int ret = 0;
1157
1158	if (!param || !evt)
1159		return -EINVAL;
1160
1161	/* we only report device fault if there is a handler registered */
1162	mutex_lock(&param->lock);
1163	fparam = param->fault_param;
1164	if (!fparam || !fparam->handler) {
1165		ret = -EINVAL;
1166		goto done_unlock;
1167	}
1168
1169	if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1170	    (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1171		evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1172				      GFP_KERNEL);
1173		if (!evt_pending) {
1174			ret = -ENOMEM;
1175			goto done_unlock;
1176		}
1177		mutex_lock(&fparam->lock);
1178		list_add_tail(&evt_pending->list, &fparam->faults);
1179		mutex_unlock(&fparam->lock);
1180	}
1181
1182	ret = fparam->handler(&evt->fault, fparam->data);
1183	if (ret && evt_pending) {
1184		mutex_lock(&fparam->lock);
1185		list_del(&evt_pending->list);
1186		mutex_unlock(&fparam->lock);
1187		kfree(evt_pending);
1188	}
1189done_unlock:
1190	mutex_unlock(&param->lock);
1191	return ret;
1192}
1193EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1194
1195int iommu_page_response(struct device *dev,
1196			struct iommu_page_response *msg)
1197{
1198	bool needs_pasid;
1199	int ret = -EINVAL;
1200	struct iommu_fault_event *evt;
1201	struct iommu_fault_page_request *prm;
1202	struct dev_iommu *param = dev->iommu;
1203	bool has_pasid = msg->flags & IOMMU_PAGE_RESP_PASID_VALID;
1204	struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1205
1206	if (!domain || !domain->ops->page_response)
1207		return -ENODEV;
1208
1209	if (!param || !param->fault_param)
1210		return -EINVAL;
1211
1212	if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1213	    msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1214		return -EINVAL;
1215
1216	/* Only send response if there is a fault report pending */
1217	mutex_lock(&param->fault_param->lock);
1218	if (list_empty(&param->fault_param->faults)) {
1219		dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1220		goto done_unlock;
1221	}
1222	/*
1223	 * Check if we have a matching page request pending to respond,
1224	 * otherwise return -EINVAL
1225	 */
1226	list_for_each_entry(evt, &param->fault_param->faults, list) {
1227		prm = &evt->fault.prm;
1228		if (prm->grpid != msg->grpid)
1229			continue;
1230
1231		/*
1232		 * If the PASID is required, the corresponding request is
1233		 * matched using the group ID, the PASID valid bit and the PASID
1234		 * value. Otherwise only the group ID matches request and
1235		 * response.
1236		 */
1237		needs_pasid = prm->flags & IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID;
1238		if (needs_pasid && (!has_pasid || msg->pasid != prm->pasid))
1239			continue;
1240
1241		if (!needs_pasid && has_pasid) {
1242			/* No big deal, just clear it. */
1243			msg->flags &= ~IOMMU_PAGE_RESP_PASID_VALID;
1244			msg->pasid = 0;
1245		}
1246
1247		ret = domain->ops->page_response(dev, evt, msg);
1248		list_del(&evt->list);
1249		kfree(evt);
1250		break;
1251	}
1252
1253done_unlock:
1254	mutex_unlock(&param->fault_param->lock);
1255	return ret;
1256}
1257EXPORT_SYMBOL_GPL(iommu_page_response);
1258
1259/**
1260 * iommu_group_id - Return ID for a group
1261 * @group: the group to ID
1262 *
1263 * Return the unique ID for the group matching the sysfs group number.
1264 */
1265int iommu_group_id(struct iommu_group *group)
1266{
1267	return group->id;
1268}
1269EXPORT_SYMBOL_GPL(iommu_group_id);
1270
1271static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1272					       unsigned long *devfns);
1273
1274/*
1275 * To consider a PCI device isolated, we require ACS to support Source
1276 * Validation, Request Redirection, Completer Redirection, and Upstream
1277 * Forwarding.  This effectively means that devices cannot spoof their
1278 * requester ID, requests and completions cannot be redirected, and all
1279 * transactions are forwarded upstream, even as it passes through a
1280 * bridge where the target device is downstream.
1281 */
1282#define REQ_ACS_FLAGS   (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1283
1284/*
1285 * For multifunction devices which are not isolated from each other, find
1286 * all the other non-isolated functions and look for existing groups.  For
1287 * each function, we also need to look for aliases to or from other devices
1288 * that may already have a group.
1289 */
1290static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1291							unsigned long *devfns)
1292{
1293	struct pci_dev *tmp = NULL;
1294	struct iommu_group *group;
1295
1296	if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1297		return NULL;
1298
1299	for_each_pci_dev(tmp) {
1300		if (tmp == pdev || tmp->bus != pdev->bus ||
1301		    PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1302		    pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1303			continue;
1304
1305		group = get_pci_alias_group(tmp, devfns);
1306		if (group) {
1307			pci_dev_put(tmp);
1308			return group;
1309		}
1310	}
1311
1312	return NULL;
1313}
1314
1315/*
1316 * Look for aliases to or from the given device for existing groups. DMA
1317 * aliases are only supported on the same bus, therefore the search
1318 * space is quite small (especially since we're really only looking at pcie
1319 * device, and therefore only expect multiple slots on the root complex or
1320 * downstream switch ports).  It's conceivable though that a pair of
1321 * multifunction devices could have aliases between them that would cause a
1322 * loop.  To prevent this, we use a bitmap to track where we've been.
1323 */
1324static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1325					       unsigned long *devfns)
1326{
1327	struct pci_dev *tmp = NULL;
1328	struct iommu_group *group;
1329
1330	if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1331		return NULL;
1332
1333	group = iommu_group_get(&pdev->dev);
1334	if (group)
1335		return group;
1336
1337	for_each_pci_dev(tmp) {
1338		if (tmp == pdev || tmp->bus != pdev->bus)
1339			continue;
1340
1341		/* We alias them or they alias us */
1342		if (pci_devs_are_dma_aliases(pdev, tmp)) {
1343			group = get_pci_alias_group(tmp, devfns);
1344			if (group) {
1345				pci_dev_put(tmp);
1346				return group;
1347			}
1348
1349			group = get_pci_function_alias_group(tmp, devfns);
1350			if (group) {
1351				pci_dev_put(tmp);
1352				return group;
1353			}
1354		}
1355	}
1356
1357	return NULL;
1358}
1359
1360struct group_for_pci_data {
1361	struct pci_dev *pdev;
1362	struct iommu_group *group;
1363};
1364
1365/*
1366 * DMA alias iterator callback, return the last seen device.  Stop and return
1367 * the IOMMU group if we find one along the way.
1368 */
1369static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1370{
1371	struct group_for_pci_data *data = opaque;
1372
1373	data->pdev = pdev;
1374	data->group = iommu_group_get(&pdev->dev);
1375
1376	return data->group != NULL;
1377}
1378
1379/*
1380 * Generic device_group call-back function. It just allocates one
1381 * iommu-group per device.
1382 */
1383struct iommu_group *generic_device_group(struct device *dev)
1384{
1385	return iommu_group_alloc();
1386}
1387EXPORT_SYMBOL_GPL(generic_device_group);
1388
1389/*
1390 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1391 * to find or create an IOMMU group for a device.
1392 */
1393struct iommu_group *pci_device_group(struct device *dev)
1394{
1395	struct pci_dev *pdev = to_pci_dev(dev);
1396	struct group_for_pci_data data;
1397	struct pci_bus *bus;
1398	struct iommu_group *group = NULL;
1399	u64 devfns[4] = { 0 };
1400
1401	if (WARN_ON(!dev_is_pci(dev)))
1402		return ERR_PTR(-EINVAL);
1403
1404	/*
1405	 * Find the upstream DMA alias for the device.  A device must not
1406	 * be aliased due to topology in order to have its own IOMMU group.
1407	 * If we find an alias along the way that already belongs to a
1408	 * group, use it.
1409	 */
1410	if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1411		return data.group;
1412
1413	pdev = data.pdev;
1414
1415	/*
1416	 * Continue upstream from the point of minimum IOMMU granularity
1417	 * due to aliases to the point where devices are protected from
1418	 * peer-to-peer DMA by PCI ACS.  Again, if we find an existing
1419	 * group, use it.
1420	 */
1421	for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1422		if (!bus->self)
1423			continue;
1424
1425		if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1426			break;
1427
1428		pdev = bus->self;
1429
1430		group = iommu_group_get(&pdev->dev);
1431		if (group)
1432			return group;
1433	}
1434
1435	/*
1436	 * Look for existing groups on device aliases.  If we alias another
1437	 * device or another device aliases us, use the same group.
1438	 */
1439	group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1440	if (group)
1441		return group;
1442
1443	/*
1444	 * Look for existing groups on non-isolated functions on the same
1445	 * slot and aliases of those funcions, if any.  No need to clear
1446	 * the search bitmap, the tested devfns are still valid.
1447	 */
1448	group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1449	if (group)
1450		return group;
1451
1452	/* No shared group found, allocate new */
1453	return iommu_group_alloc();
1454}
1455EXPORT_SYMBOL_GPL(pci_device_group);
1456
1457/* Get the IOMMU group for device on fsl-mc bus */
1458struct iommu_group *fsl_mc_device_group(struct device *dev)
1459{
1460	struct device *cont_dev = fsl_mc_cont_dev(dev);
1461	struct iommu_group *group;
1462
1463	group = iommu_group_get(cont_dev);
1464	if (!group)
1465		group = iommu_group_alloc();
1466	return group;
1467}
1468EXPORT_SYMBOL_GPL(fsl_mc_device_group);
1469
1470static int iommu_get_def_domain_type(struct device *dev)
1471{
1472	const struct iommu_ops *ops = dev->bus->iommu_ops;
1473	unsigned int type = 0;
1474
1475	if (ops->def_domain_type)
1476		type = ops->def_domain_type(dev);
1477
1478	return (type == 0) ? iommu_def_domain_type : type;
1479}
1480
1481static int iommu_group_alloc_default_domain(struct bus_type *bus,
1482					    struct iommu_group *group,
1483					    unsigned int type)
1484{
1485	struct iommu_domain *dom;
1486
1487	dom = __iommu_domain_alloc(bus, type);
1488	if (!dom && type != IOMMU_DOMAIN_DMA) {
1489		dom = __iommu_domain_alloc(bus, IOMMU_DOMAIN_DMA);
1490		if (dom)
1491			pr_warn("Failed to allocate default IOMMU domain of type %u for group %s - Falling back to IOMMU_DOMAIN_DMA",
1492				type, group->name);
1493	}
1494
1495	if (!dom)
1496		return -ENOMEM;
1497
1498	group->default_domain = dom;
1499	if (!group->domain)
1500		group->domain = dom;
1501
1502	if (!iommu_dma_strict) {
1503		int attr = 1;
1504		iommu_domain_set_attr(dom,
1505				      DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1506				      &attr);
1507	}
1508
1509	return 0;
1510}
1511
1512static int iommu_alloc_default_domain(struct iommu_group *group,
1513				      struct device *dev)
1514{
1515	unsigned int type;
1516
1517	if (group->default_domain)
1518		return 0;
1519
1520	type = iommu_get_def_domain_type(dev);
1521
1522	return iommu_group_alloc_default_domain(dev->bus, group, type);
1523}
1524
1525/**
1526 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1527 * @dev: target device
1528 *
1529 * This function is intended to be called by IOMMU drivers and extended to
1530 * support common, bus-defined algorithms when determining or creating the
1531 * IOMMU group for a device.  On success, the caller will hold a reference
1532 * to the returned IOMMU group, which will already include the provided
1533 * device.  The reference should be released with iommu_group_put().
1534 */
1535static struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1536{
1537	const struct iommu_ops *ops = dev->bus->iommu_ops;
1538	struct iommu_group *group;
1539	int ret;
1540
1541	group = iommu_group_get(dev);
1542	if (group)
1543		return group;
1544
1545	if (!ops)
1546		return ERR_PTR(-EINVAL);
1547
1548	group = ops->device_group(dev);
1549	if (WARN_ON_ONCE(group == NULL))
1550		return ERR_PTR(-EINVAL);
1551
1552	if (IS_ERR(group))
1553		return group;
1554
1555	ret = iommu_group_add_device(group, dev);
1556	if (ret)
1557		goto out_put_group;
1558
1559	return group;
1560
1561out_put_group:
1562	iommu_group_put(group);
1563
1564	return ERR_PTR(ret);
1565}
1566
1567struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1568{
1569	return group->default_domain;
1570}
1571
1572static int probe_iommu_group(struct device *dev, void *data)
1573{
1574	struct list_head *group_list = data;
1575	struct iommu_group *group;
1576	int ret;
1577
1578	/* Device is probed already if in a group */
1579	group = iommu_group_get(dev);
1580	if (group) {
1581		iommu_group_put(group);
1582		return 0;
1583	}
1584
1585	ret = __iommu_probe_device(dev, group_list);
1586	if (ret == -ENODEV)
1587		ret = 0;
1588
1589	return ret;
1590}
1591
1592static int remove_iommu_group(struct device *dev, void *data)
1593{
1594	iommu_release_device(dev);
1595
1596	return 0;
1597}
1598
1599static int iommu_bus_notifier(struct notifier_block *nb,
1600			      unsigned long action, void *data)
1601{
1602	unsigned long group_action = 0;
1603	struct device *dev = data;
1604	struct iommu_group *group;
1605
1606	/*
1607	 * ADD/DEL call into iommu driver ops if provided, which may
1608	 * result in ADD/DEL notifiers to group->notifier
1609	 */
1610	if (action == BUS_NOTIFY_ADD_DEVICE) {
1611		int ret;
1612
1613		ret = iommu_probe_device(dev);
1614		return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1615	} else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1616		iommu_release_device(dev);
1617		return NOTIFY_OK;
1618	}
1619
1620	/*
1621	 * Remaining BUS_NOTIFYs get filtered and republished to the
1622	 * group, if anyone is listening
1623	 */
1624	group = iommu_group_get(dev);
1625	if (!group)
1626		return 0;
1627
1628	switch (action) {
1629	case BUS_NOTIFY_BIND_DRIVER:
1630		group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1631		break;
1632	case BUS_NOTIFY_BOUND_DRIVER:
1633		group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1634		break;
1635	case BUS_NOTIFY_UNBIND_DRIVER:
1636		group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1637		break;
1638	case BUS_NOTIFY_UNBOUND_DRIVER:
1639		group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1640		break;
1641	}
1642
1643	if (group_action)
1644		blocking_notifier_call_chain(&group->notifier,
1645					     group_action, dev);
1646
1647	iommu_group_put(group);
1648	return 0;
1649}
1650
1651struct __group_domain_type {
1652	struct device *dev;
1653	unsigned int type;
1654};
1655
1656static int probe_get_default_domain_type(struct device *dev, void *data)
1657{
1658	const struct iommu_ops *ops = dev->bus->iommu_ops;
1659	struct __group_domain_type *gtype = data;
1660	unsigned int type = 0;
1661
1662	if (ops->def_domain_type)
1663		type = ops->def_domain_type(dev);
1664
1665	if (type) {
1666		if (gtype->type && gtype->type != type) {
1667			dev_warn(dev, "Device needs domain type %s, but device %s in the same iommu group requires type %s - using default\n",
1668				 iommu_domain_type_str(type),
1669				 dev_name(gtype->dev),
1670				 iommu_domain_type_str(gtype->type));
1671			gtype->type = 0;
1672		}
1673
1674		if (!gtype->dev) {
1675			gtype->dev  = dev;
1676			gtype->type = type;
1677		}
1678	}
1679
1680	return 0;
1681}
1682
1683static void probe_alloc_default_domain(struct bus_type *bus,
1684				       struct iommu_group *group)
1685{
1686	struct __group_domain_type gtype;
1687
1688	memset(&gtype, 0, sizeof(gtype));
1689
1690	/* Ask for default domain requirements of all devices in the group */
1691	__iommu_group_for_each_dev(group, &gtype,
1692				   probe_get_default_domain_type);
1693
1694	if (!gtype.type)
1695		gtype.type = iommu_def_domain_type;
1696
1697	iommu_group_alloc_default_domain(bus, group, gtype.type);
1698
1699}
1700
1701static int iommu_group_do_dma_attach(struct device *dev, void *data)
1702{
1703	struct iommu_domain *domain = data;
1704	int ret = 0;
1705
1706	if (!iommu_is_attach_deferred(domain, dev))
1707		ret = __iommu_attach_device(domain, dev);
1708
1709	return ret;
1710}
1711
1712static int __iommu_group_dma_attach(struct iommu_group *group)
1713{
1714	return __iommu_group_for_each_dev(group, group->default_domain,
1715					  iommu_group_do_dma_attach);
1716}
1717
1718static int iommu_group_do_probe_finalize(struct device *dev, void *data)
1719{
1720	struct iommu_domain *domain = data;
1721
1722	if (domain->ops->probe_finalize)
1723		domain->ops->probe_finalize(dev);
1724
1725	return 0;
1726}
1727
1728static void __iommu_group_dma_finalize(struct iommu_group *group)
1729{
1730	__iommu_group_for_each_dev(group, group->default_domain,
1731				   iommu_group_do_probe_finalize);
1732}
1733
1734static int iommu_do_create_direct_mappings(struct device *dev, void *data)
1735{
1736	struct iommu_group *group = data;
1737
1738	iommu_create_device_direct_mappings(group, dev);
1739
1740	return 0;
1741}
1742
1743static int iommu_group_create_direct_mappings(struct iommu_group *group)
1744{
1745	return __iommu_group_for_each_dev(group, group,
1746					  iommu_do_create_direct_mappings);
1747}
1748
1749int bus_iommu_probe(struct bus_type *bus)
1750{
1751	struct iommu_group *group, *next;
1752	LIST_HEAD(group_list);
1753	int ret;
1754
1755	/*
1756	 * This code-path does not allocate the default domain when
1757	 * creating the iommu group, so do it after the groups are
1758	 * created.
1759	 */
1760	ret = bus_for_each_dev(bus, NULL, &group_list, probe_iommu_group);
1761	if (ret)
1762		return ret;
1763
1764	list_for_each_entry_safe(group, next, &group_list, entry) {
1765		/* Remove item from the list */
1766		list_del_init(&group->entry);
1767
1768		mutex_lock(&group->mutex);
1769
1770		/* Try to allocate default domain */
1771		probe_alloc_default_domain(bus, group);
1772
1773		if (!group->default_domain) {
1774			mutex_unlock(&group->mutex);
1775			continue;
1776		}
1777
1778		iommu_group_create_direct_mappings(group);
1779
1780		ret = __iommu_group_dma_attach(group);
1781
1782		mutex_unlock(&group->mutex);
1783
1784		if (ret)
1785			break;
1786
1787		__iommu_group_dma_finalize(group);
1788	}
1789
1790	return ret;
1791}
1792
1793static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1794{
1795	struct notifier_block *nb;
1796	int err;
1797
1798	nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1799	if (!nb)
1800		return -ENOMEM;
1801
1802	nb->notifier_call = iommu_bus_notifier;
1803
1804	err = bus_register_notifier(bus, nb);
1805	if (err)
1806		goto out_free;
1807
1808	err = bus_iommu_probe(bus);
1809	if (err)
1810		goto out_err;
1811
1812
1813	return 0;
1814
1815out_err:
1816	/* Clean up */
1817	bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1818	bus_unregister_notifier(bus, nb);
1819
1820out_free:
1821	kfree(nb);
1822
1823	return err;
1824}
1825
1826/**
1827 * bus_set_iommu - set iommu-callbacks for the bus
1828 * @bus: bus.
1829 * @ops: the callbacks provided by the iommu-driver
1830 *
1831 * This function is called by an iommu driver to set the iommu methods
1832 * used for a particular bus. Drivers for devices on that bus can use
1833 * the iommu-api after these ops are registered.
1834 * This special function is needed because IOMMUs are usually devices on
1835 * the bus itself, so the iommu drivers are not initialized when the bus
1836 * is set up. With this function the iommu-driver can set the iommu-ops
1837 * afterwards.
1838 */
1839int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1840{
1841	int err;
1842
1843	if (ops == NULL) {
1844		bus->iommu_ops = NULL;
1845		return 0;
1846	}
1847
1848	if (bus->iommu_ops != NULL)
1849		return -EBUSY;
1850
1851	bus->iommu_ops = ops;
1852
1853	/* Do IOMMU specific setup for this bus-type */
1854	err = iommu_bus_init(bus, ops);
1855	if (err)
1856		bus->iommu_ops = NULL;
1857
1858	return err;
1859}
1860EXPORT_SYMBOL_GPL(bus_set_iommu);
1861
1862bool iommu_present(struct bus_type *bus)
1863{
1864	return bus->iommu_ops != NULL;
1865}
1866EXPORT_SYMBOL_GPL(iommu_present);
1867
1868bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1869{
1870	if (!bus->iommu_ops || !bus->iommu_ops->capable)
1871		return false;
1872
1873	return bus->iommu_ops->capable(cap);
1874}
1875EXPORT_SYMBOL_GPL(iommu_capable);
1876
1877/**
1878 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1879 * @domain: iommu domain
1880 * @handler: fault handler
1881 * @token: user data, will be passed back to the fault handler
1882 *
1883 * This function should be used by IOMMU users which want to be notified
1884 * whenever an IOMMU fault happens.
1885 *
1886 * The fault handler itself should return 0 on success, and an appropriate
1887 * error code otherwise.
1888 */
1889void iommu_set_fault_handler(struct iommu_domain *domain,
1890					iommu_fault_handler_t handler,
1891					void *token)
1892{
1893	BUG_ON(!domain);
1894
1895	domain->handler = handler;
1896	domain->handler_token = token;
1897}
1898EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1899
1900static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1901						 unsigned type)
1902{
1903	struct iommu_domain *domain;
1904
1905	if (bus == NULL || bus->iommu_ops == NULL)
1906		return NULL;
1907
1908	domain = bus->iommu_ops->domain_alloc(type);
1909	if (!domain)
1910		return NULL;
1911
1912	domain->ops  = bus->iommu_ops;
1913	domain->type = type;
1914	/* Assume all sizes by default; the driver may override this later */
1915	domain->pgsize_bitmap  = bus->iommu_ops->pgsize_bitmap;
1916
1917	return domain;
1918}
1919
1920struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1921{
1922	return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1923}
1924EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1925
1926void iommu_domain_free(struct iommu_domain *domain)
1927{
1928	domain->ops->domain_free(domain);
1929}
1930EXPORT_SYMBOL_GPL(iommu_domain_free);
1931
1932static int __iommu_attach_device(struct iommu_domain *domain,
1933				 struct device *dev)
1934{
1935	int ret;
1936
1937	if (unlikely(domain->ops->attach_dev == NULL))
1938		return -ENODEV;
1939
1940	ret = domain->ops->attach_dev(domain, dev);
1941	if (!ret)
1942		trace_attach_device_to_domain(dev);
1943	return ret;
1944}
1945
1946int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1947{
1948	struct iommu_group *group;
1949	int ret;
1950
1951	group = iommu_group_get(dev);
1952	if (!group)
1953		return -ENODEV;
1954
1955	/*
1956	 * Lock the group to make sure the device-count doesn't
1957	 * change while we are attaching
1958	 */
1959	mutex_lock(&group->mutex);
1960	ret = -EINVAL;
1961	if (iommu_group_device_count(group) != 1)
1962		goto out_unlock;
1963
1964	ret = __iommu_attach_group(domain, group);
1965
1966out_unlock:
1967	mutex_unlock(&group->mutex);
1968	iommu_group_put(group);
1969
1970	return ret;
1971}
1972EXPORT_SYMBOL_GPL(iommu_attach_device);
1973
1974/*
1975 * Check flags and other user provided data for valid combinations. We also
1976 * make sure no reserved fields or unused flags are set. This is to ensure
1977 * not breaking userspace in the future when these fields or flags are used.
1978 */
1979static int iommu_check_cache_invl_data(struct iommu_cache_invalidate_info *info)
1980{
1981	u32 mask;
1982	int i;
1983
1984	if (info->version != IOMMU_CACHE_INVALIDATE_INFO_VERSION_1)
1985		return -EINVAL;
1986
1987	mask = (1 << IOMMU_CACHE_INV_TYPE_NR) - 1;
1988	if (info->cache & ~mask)
1989		return -EINVAL;
1990
1991	if (info->granularity >= IOMMU_INV_GRANU_NR)
1992		return -EINVAL;
1993
1994	switch (info->granularity) {
1995	case IOMMU_INV_GRANU_ADDR:
1996		if (info->cache & IOMMU_CACHE_INV_TYPE_PASID)
1997			return -EINVAL;
1998
1999		mask = IOMMU_INV_ADDR_FLAGS_PASID |
2000			IOMMU_INV_ADDR_FLAGS_ARCHID |
2001			IOMMU_INV_ADDR_FLAGS_LEAF;
2002
2003		if (info->granu.addr_info.flags & ~mask)
2004			return -EINVAL;
2005		break;
2006	case IOMMU_INV_GRANU_PASID:
2007		mask = IOMMU_INV_PASID_FLAGS_PASID |
2008			IOMMU_INV_PASID_FLAGS_ARCHID;
2009		if (info->granu.pasid_info.flags & ~mask)
2010			return -EINVAL;
2011
2012		break;
2013	case IOMMU_INV_GRANU_DOMAIN:
2014		if (info->cache & IOMMU_CACHE_INV_TYPE_DEV_IOTLB)
2015			return -EINVAL;
2016		break;
2017	default:
2018		return -EINVAL;
2019	}
2020
2021	/* Check reserved padding fields */
2022	for (i = 0; i < sizeof(info->padding); i++) {
2023		if (info->padding[i])
2024			return -EINVAL;
2025	}
2026
2027	return 0;
2028}
2029
2030int iommu_uapi_cache_invalidate(struct iommu_domain *domain, struct device *dev,
2031				void __user *uinfo)
2032{
2033	struct iommu_cache_invalidate_info inv_info = { 0 };
2034	u32 minsz;
2035	int ret;
2036
2037	if (unlikely(!domain->ops->cache_invalidate))
2038		return -ENODEV;
2039
2040	/*
2041	 * No new spaces can be added before the variable sized union, the
2042	 * minimum size is the offset to the union.
2043	 */
2044	minsz = offsetof(struct iommu_cache_invalidate_info, granu);
2045
2046	/* Copy minsz from user to get flags and argsz */
2047	if (copy_from_user(&inv_info, uinfo, minsz))
2048		return -EFAULT;
2049
2050	/* Fields before the variable size union are mandatory */
2051	if (inv_info.argsz < minsz)
2052		return -EINVAL;
2053
2054	/* PASID and address granu require additional info beyond minsz */
2055	if (inv_info.granularity == IOMMU_INV_GRANU_PASID &&
2056	    inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.pasid_info))
2057		return -EINVAL;
2058
2059	if (inv_info.granularity == IOMMU_INV_GRANU_ADDR &&
2060	    inv_info.argsz < offsetofend(struct iommu_cache_invalidate_info, granu.addr_info))
2061		return -EINVAL;
2062
2063	/*
2064	 * User might be using a newer UAPI header which has a larger data
2065	 * size, we shall support the existing flags within the current
2066	 * size. Copy the remaining user data _after_ minsz but not more
2067	 * than the current kernel supported size.
2068	 */
2069	if (copy_from_user((void *)&inv_info + minsz, uinfo + minsz,
2070			   min_t(u32, inv_info.argsz, sizeof(inv_info)) - minsz))
2071		return -EFAULT;
2072
2073	/* Now the argsz is validated, check the content */
2074	ret = iommu_check_cache_invl_data(&inv_info);
2075	if (ret)
2076		return ret;
2077
2078	return domain->ops->cache_invalidate(domain, dev, &inv_info);
2079}
2080EXPORT_SYMBOL_GPL(iommu_uapi_cache_invalidate);
2081
2082static int iommu_check_bind_data(struct iommu_gpasid_bind_data *data)
2083{
2084	u64 mask;
2085	int i;
2086
2087	if (data->version != IOMMU_GPASID_BIND_VERSION_1)
2088		return -EINVAL;
2089
2090	/* Check the range of supported formats */
2091	if (data->format >= IOMMU_PASID_FORMAT_LAST)
2092		return -EINVAL;
2093
2094	/* Check all flags */
2095	mask = IOMMU_SVA_GPASID_VAL;
2096	if (data->flags & ~mask)
2097		return -EINVAL;
2098
2099	/* Check reserved padding fields */
2100	for (i = 0; i < sizeof(data->padding); i++) {
2101		if (data->padding[i])
2102			return -EINVAL;
2103	}
2104
2105	return 0;
2106}
2107
2108static int iommu_sva_prepare_bind_data(void __user *udata,
2109				       struct iommu_gpasid_bind_data *data)
2110{
2111	u32 minsz;
2112
2113	/*
2114	 * No new spaces can be added before the variable sized union, the
2115	 * minimum size is the offset to the union.
2116	 */
2117	minsz = offsetof(struct iommu_gpasid_bind_data, vendor);
2118
2119	/* Copy minsz from user to get flags and argsz */
2120	if (copy_from_user(data, udata, minsz))
2121		return -EFAULT;
2122
2123	/* Fields before the variable size union are mandatory */
2124	if (data->argsz < minsz)
2125		return -EINVAL;
2126	/*
2127	 * User might be using a newer UAPI header, we shall let IOMMU vendor
2128	 * driver decide on what size it needs. Since the guest PASID bind data
2129	 * can be vendor specific, larger argsz could be the result of extension
2130	 * for one vendor but it should not affect another vendor.
2131	 * Copy the remaining user data _after_ minsz
2132	 */
2133	if (copy_from_user((void *)data + minsz, udata + minsz,
2134			   min_t(u32, data->argsz, sizeof(*data)) - minsz))
2135		return -EFAULT;
2136
2137	return iommu_check_bind_data(data);
2138}
2139
2140int iommu_uapi_sva_bind_gpasid(struct iommu_domain *domain, struct device *dev,
2141			       void __user *udata)
2142{
2143	struct iommu_gpasid_bind_data data = { 0 };
2144	int ret;
2145
2146	if (unlikely(!domain->ops->sva_bind_gpasid))
2147		return -ENODEV;
2148
2149	ret = iommu_sva_prepare_bind_data(udata, &data);
2150	if (ret)
2151		return ret;
2152
2153	return domain->ops->sva_bind_gpasid(domain, dev, &data);
2154}
2155EXPORT_SYMBOL_GPL(iommu_uapi_sva_bind_gpasid);
2156
2157int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2158			     ioasid_t pasid)
2159{
2160	if (unlikely(!domain->ops->sva_unbind_gpasid))
2161		return -ENODEV;
2162
2163	return domain->ops->sva_unbind_gpasid(dev, pasid);
2164}
2165EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
2166
2167int iommu_uapi_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
2168				 void __user *udata)
2169{
2170	struct iommu_gpasid_bind_data data = { 0 };
2171	int ret;
2172
2173	if (unlikely(!domain->ops->sva_bind_gpasid))
2174		return -ENODEV;
2175
2176	ret = iommu_sva_prepare_bind_data(udata, &data);
2177	if (ret)
2178		return ret;
2179
2180	return iommu_sva_unbind_gpasid(domain, dev, data.hpasid);
2181}
2182EXPORT_SYMBOL_GPL(iommu_uapi_sva_unbind_gpasid);
2183
2184static void __iommu_detach_device(struct iommu_domain *domain,
2185				  struct device *dev)
2186{
2187	if (iommu_is_attach_deferred(domain, dev))
2188		return;
2189
2190	if (unlikely(domain->ops->detach_dev == NULL))
2191		return;
2192
2193	domain->ops->detach_dev(domain, dev);
2194	trace_detach_device_from_domain(dev);
2195}
2196
2197void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
2198{
2199	struct iommu_group *group;
2200
2201	group = iommu_group_get(dev);
2202	if (!group)
2203		return;
2204
2205	mutex_lock(&group->mutex);
2206	if (iommu_group_device_count(group) != 1) {
2207		WARN_ON(1);
2208		goto out_unlock;
2209	}
2210
2211	__iommu_detach_group(domain, group);
2212
2213out_unlock:
2214	mutex_unlock(&group->mutex);
2215	iommu_group_put(group);
2216}
2217EXPORT_SYMBOL_GPL(iommu_detach_device);
2218
2219struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
2220{
2221	struct iommu_domain *domain;
2222	struct iommu_group *group;
2223
2224	group = iommu_group_get(dev);
2225	if (!group)
2226		return NULL;
2227
2228	domain = group->domain;
2229
2230	iommu_group_put(group);
2231
2232	return domain;
2233}
2234EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
2235
2236/*
2237 * For IOMMU_DOMAIN_DMA implementations which already provide their own
2238 * guarantees that the group and its default domain are valid and correct.
2239 */
2240struct iommu_domain *iommu_get_dma_domain(struct device *dev)
2241{
2242	return dev->iommu_group->default_domain;
2243}
2244
2245/*
2246 * IOMMU groups are really the natural working unit of the IOMMU, but
2247 * the IOMMU API works on domains and devices.  Bridge that gap by
2248 * iterating over the devices in a group.  Ideally we'd have a single
2249 * device which represents the requestor ID of the group, but we also
2250 * allow IOMMU drivers to create policy defined minimum sets, where
2251 * the physical hardware may be able to distiguish members, but we
2252 * wish to group them at a higher level (ex. untrusted multi-function
2253 * PCI devices).  Thus we attach each device.
2254 */
2255static int iommu_group_do_attach_device(struct device *dev, void *data)
2256{
2257	struct iommu_domain *domain = data;
2258
2259	return __iommu_attach_device(domain, dev);
2260}
2261
2262static int __iommu_attach_group(struct iommu_domain *domain,
2263				struct iommu_group *group)
2264{
2265	int ret;
2266
2267	if (group->default_domain && group->domain != group->default_domain)
2268		return -EBUSY;
2269
2270	ret = __iommu_group_for_each_dev(group, domain,
2271					 iommu_group_do_attach_device);
2272	if (ret == 0)
2273		group->domain = domain;
2274
2275	return ret;
2276}
2277
2278int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
2279{
2280	int ret;
2281
2282	mutex_lock(&group->mutex);
2283	ret = __iommu_attach_group(domain, group);
2284	mutex_unlock(&group->mutex);
2285
2286	return ret;
2287}
2288EXPORT_SYMBOL_GPL(iommu_attach_group);
2289
2290static int iommu_group_do_detach_device(struct device *dev, void *data)
2291{
2292	struct iommu_domain *domain = data;
2293
2294	__iommu_detach_device(domain, dev);
2295
2296	return 0;
2297}
2298
2299static void __iommu_detach_group(struct iommu_domain *domain,
2300				 struct iommu_group *group)
2301{
2302	int ret;
2303
2304	if (!group->default_domain) {
2305		__iommu_group_for_each_dev(group, domain,
2306					   iommu_group_do_detach_device);
2307		group->domain = NULL;
2308		return;
2309	}
2310
2311	if (group->domain == group->default_domain)
2312		return;
2313
2314	/* Detach by re-attaching to the default domain */
2315	ret = __iommu_group_for_each_dev(group, group->default_domain,
2316					 iommu_group_do_attach_device);
2317	if (ret != 0)
2318		WARN_ON(1);
2319	else
2320		group->domain = group->default_domain;
2321}
2322
2323void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
2324{
2325	mutex_lock(&group->mutex);
2326	__iommu_detach_group(domain, group);
2327	mutex_unlock(&group->mutex);
2328}
2329EXPORT_SYMBOL_GPL(iommu_detach_group);
2330
2331phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
2332{
2333	if (unlikely(domain->ops->iova_to_phys == NULL))
2334		return 0;
2335
2336	return domain->ops->iova_to_phys(domain, iova);
2337}
2338EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
2339
2340static size_t iommu_pgsize(struct iommu_domain *domain,
2341			   unsigned long addr_merge, size_t size)
2342{
2343	unsigned int pgsize_idx;
2344	size_t pgsize;
2345
2346	/* Max page size that still fits into 'size' */
2347	pgsize_idx = __fls(size);
2348
2349	/* need to consider alignment requirements ? */
2350	if (likely(addr_merge)) {
2351		/* Max page size allowed by address */
2352		unsigned int align_pgsize_idx = __ffs(addr_merge);
2353		pgsize_idx = min(pgsize_idx, align_pgsize_idx);
2354	}
2355
2356	/* build a mask of acceptable page sizes */
2357	pgsize = (1UL << (pgsize_idx + 1)) - 1;
2358
2359	/* throw away page sizes not supported by the hardware */
2360	pgsize &= domain->pgsize_bitmap;
2361
2362	/* make sure we're still sane */
2363	BUG_ON(!pgsize);
2364
2365	/* pick the biggest page */
2366	pgsize_idx = __fls(pgsize);
2367	pgsize = 1UL << pgsize_idx;
2368
2369	return pgsize;
2370}
2371
2372static int __iommu_map(struct iommu_domain *domain, unsigned long iova,
2373		       phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2374{
2375	const struct iommu_ops *ops = domain->ops;
2376	unsigned long orig_iova = iova;
2377	unsigned int min_pagesz;
2378	size_t orig_size = size;
2379	phys_addr_t orig_paddr = paddr;
2380	int ret = 0;
2381
2382	if (unlikely(ops->map == NULL ||
2383		     domain->pgsize_bitmap == 0UL))
2384		return -ENODEV;
2385
2386	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2387		return -EINVAL;
2388
2389	/* find out the minimum page size supported */
2390	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2391
2392	/*
2393	 * both the virtual address and the physical one, as well as
2394	 * the size of the mapping, must be aligned (at least) to the
2395	 * size of the smallest page supported by the hardware
2396	 */
2397	if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
2398		pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
2399		       iova, &paddr, size, min_pagesz);
2400		return -EINVAL;
2401	}
2402
2403	pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
2404
2405	while (size) {
2406		size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
2407
2408		pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
2409			 iova, &paddr, pgsize);
2410		ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
2411
2412		if (ret)
2413			break;
2414
2415		iova += pgsize;
2416		paddr += pgsize;
2417		size -= pgsize;
2418	}
2419
2420	/* unroll mapping in case something went wrong */
2421	if (ret)
2422		iommu_unmap(domain, orig_iova, orig_size - size);
2423	else
2424		trace_map(orig_iova, orig_paddr, orig_size);
2425
2426	return ret;
2427}
2428
2429static int _iommu_map(struct iommu_domain *domain, unsigned long iova,
2430		      phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
2431{
2432	const struct iommu_ops *ops = domain->ops;
2433	int ret;
2434
2435	ret = __iommu_map(domain, iova, paddr, size, prot, gfp);
2436	if (ret == 0 && ops->iotlb_sync_map)
2437		ops->iotlb_sync_map(domain);
2438
2439	return ret;
2440}
2441
2442int iommu_map(struct iommu_domain *domain, unsigned long iova,
2443	      phys_addr_t paddr, size_t size, int prot)
2444{
2445	might_sleep();
2446	return _iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
2447}
2448EXPORT_SYMBOL_GPL(iommu_map);
2449
2450int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
2451	      phys_addr_t paddr, size_t size, int prot)
2452{
2453	return _iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
2454}
2455EXPORT_SYMBOL_GPL(iommu_map_atomic);
2456
2457static size_t __iommu_unmap(struct iommu_domain *domain,
2458			    unsigned long iova, size_t size,
2459			    struct iommu_iotlb_gather *iotlb_gather)
2460{
2461	const struct iommu_ops *ops = domain->ops;
2462	size_t unmapped_page, unmapped = 0;
2463	unsigned long orig_iova = iova;
2464	unsigned int min_pagesz;
2465
2466	if (unlikely(ops->unmap == NULL ||
2467		     domain->pgsize_bitmap == 0UL))
2468		return 0;
2469
2470	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
2471		return 0;
2472
2473	/* find out the minimum page size supported */
2474	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
2475
2476	/*
2477	 * The virtual address, as well as the size of the mapping, must be
2478	 * aligned (at least) to the size of the smallest page supported
2479	 * by the hardware
2480	 */
2481	if (!IS_ALIGNED(iova | size, min_pagesz)) {
2482		pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
2483		       iova, size, min_pagesz);
2484		return 0;
2485	}
2486
2487	pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
2488
2489	/*
2490	 * Keep iterating until we either unmap 'size' bytes (or more)
2491	 * or we hit an area that isn't mapped.
2492	 */
2493	while (unmapped < size) {
2494		size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
2495
2496		unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
2497		if (!unmapped_page)
2498			break;
2499
2500		pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2501			 iova, unmapped_page);
2502
2503		iova += unmapped_page;
2504		unmapped += unmapped_page;
2505	}
2506
2507	trace_unmap(orig_iova, size, unmapped);
2508	return unmapped;
2509}
2510
2511size_t iommu_unmap(struct iommu_domain *domain,
2512		   unsigned long iova, size_t size)
2513{
2514	struct iommu_iotlb_gather iotlb_gather;
2515	size_t ret;
2516
2517	iommu_iotlb_gather_init(&iotlb_gather);
2518	ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2519	iommu_iotlb_sync(domain, &iotlb_gather);
2520
2521	return ret;
2522}
2523EXPORT_SYMBOL_GPL(iommu_unmap);
2524
2525size_t iommu_unmap_fast(struct iommu_domain *domain,
2526			unsigned long iova, size_t size,
2527			struct iommu_iotlb_gather *iotlb_gather)
2528{
2529	return __iommu_unmap(domain, iova, size, iotlb_gather);
2530}
2531EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2532
2533static size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2534			     struct scatterlist *sg, unsigned int nents, int prot,
2535			     gfp_t gfp)
2536{
2537	const struct iommu_ops *ops = domain->ops;
2538	size_t len = 0, mapped = 0;
2539	phys_addr_t start;
2540	unsigned int i = 0;
2541	int ret;
2542
2543	while (i <= nents) {
2544		phys_addr_t s_phys = sg_phys(sg);
2545
2546		if (len && s_phys != start + len) {
2547			ret = __iommu_map(domain, iova + mapped, start,
2548					len, prot, gfp);
2549
2550			if (ret)
2551				goto out_err;
2552
2553			mapped += len;
2554			len = 0;
2555		}
2556
2557		if (len) {
2558			len += sg->length;
2559		} else {
2560			len = sg->length;
2561			start = s_phys;
2562		}
2563
2564		if (++i < nents)
2565			sg = sg_next(sg);
2566	}
2567
2568	if (ops->iotlb_sync_map)
2569		ops->iotlb_sync_map(domain);
2570	return mapped;
2571
2572out_err:
2573	/* undo mappings already done */
2574	iommu_unmap(domain, iova, mapped);
2575
2576	return 0;
2577
2578}
2579
2580size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2581		    struct scatterlist *sg, unsigned int nents, int prot)
2582{
2583	might_sleep();
2584	return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2585}
2586EXPORT_SYMBOL_GPL(iommu_map_sg);
2587
2588size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2589		    struct scatterlist *sg, unsigned int nents, int prot)
2590{
2591	return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2592}
2593EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2594
2595int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
2596			       phys_addr_t paddr, u64 size, int prot)
2597{
2598	if (unlikely(domain->ops->domain_window_enable == NULL))
2599		return -ENODEV;
2600
2601	return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2602						 prot);
2603}
2604EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2605
2606void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2607{
2608	if (unlikely(domain->ops->domain_window_disable == NULL))
2609		return;
2610
2611	return domain->ops->domain_window_disable(domain, wnd_nr);
2612}
2613EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2614
2615/**
2616 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2617 * @domain: the iommu domain where the fault has happened
2618 * @dev: the device where the fault has happened
2619 * @iova: the faulting address
2620 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2621 *
2622 * This function should be called by the low-level IOMMU implementations
2623 * whenever IOMMU faults happen, to allow high-level users, that are
2624 * interested in such events, to know about them.
2625 *
2626 * This event may be useful for several possible use cases:
2627 * - mere logging of the event
2628 * - dynamic TLB/PTE loading
2629 * - if restarting of the faulting device is required
2630 *
2631 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2632 * PTE/TLB loading will one day be supported, implementations will be able
2633 * to tell whether it succeeded or not according to this return value).
2634 *
2635 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2636 * (though fault handlers can also return -ENOSYS, in case they want to
2637 * elicit the default behavior of the IOMMU drivers).
2638 */
2639int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2640		       unsigned long iova, int flags)
2641{
2642	int ret = -ENOSYS;
2643
2644	/*
2645	 * if upper layers showed interest and installed a fault handler,
2646	 * invoke it.
2647	 */
2648	if (domain->handler)
2649		ret = domain->handler(domain, dev, iova, flags,
2650						domain->handler_token);
2651
2652	trace_io_page_fault(dev, iova, flags);
2653	return ret;
2654}
2655EXPORT_SYMBOL_GPL(report_iommu_fault);
2656
2657static int __init iommu_init(void)
2658{
2659	iommu_group_kset = kset_create_and_add("iommu_groups",
2660					       NULL, kernel_kobj);
2661	BUG_ON(!iommu_group_kset);
2662
2663	iommu_debugfs_setup();
2664
2665	return 0;
2666}
2667core_initcall(iommu_init);
2668
2669int iommu_domain_get_attr(struct iommu_domain *domain,
2670			  enum iommu_attr attr, void *data)
2671{
2672	struct iommu_domain_geometry *geometry;
2673	bool *paging;
2674	int ret = 0;
2675
2676	switch (attr) {
2677	case DOMAIN_ATTR_GEOMETRY:
2678		geometry  = data;
2679		*geometry = domain->geometry;
2680
2681		break;
2682	case DOMAIN_ATTR_PAGING:
2683		paging  = data;
2684		*paging = (domain->pgsize_bitmap != 0UL);
2685		break;
2686	default:
2687		if (!domain->ops->domain_get_attr)
2688			return -EINVAL;
2689
2690		ret = domain->ops->domain_get_attr(domain, attr, data);
2691	}
2692
2693	return ret;
2694}
2695EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2696
2697int iommu_domain_set_attr(struct iommu_domain *domain,
2698			  enum iommu_attr attr, void *data)
2699{
2700	int ret = 0;
2701
2702	switch (attr) {
2703	default:
2704		if (domain->ops->domain_set_attr == NULL)
2705			return -EINVAL;
2706
2707		ret = domain->ops->domain_set_attr(domain, attr, data);
2708	}
2709
2710	return ret;
2711}
2712EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
2713
2714void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2715{
2716	const struct iommu_ops *ops = dev->bus->iommu_ops;
2717
2718	if (ops && ops->get_resv_regions)
2719		ops->get_resv_regions(dev, list);
2720}
2721
2722void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2723{
2724	const struct iommu_ops *ops = dev->bus->iommu_ops;
2725
2726	if (ops && ops->put_resv_regions)
2727		ops->put_resv_regions(dev, list);
2728}
2729
2730/**
2731 * generic_iommu_put_resv_regions - Reserved region driver helper
2732 * @dev: device for which to free reserved regions
2733 * @list: reserved region list for device
2734 *
2735 * IOMMU drivers can use this to implement their .put_resv_regions() callback
2736 * for simple reservations. Memory allocated for each reserved region will be
2737 * freed. If an IOMMU driver allocates additional resources per region, it is
2738 * going to have to implement a custom callback.
2739 */
2740void generic_iommu_put_resv_regions(struct device *dev, struct list_head *list)
2741{
2742	struct iommu_resv_region *entry, *next;
2743
2744	list_for_each_entry_safe(entry, next, list, list)
2745		kfree(entry);
2746}
2747EXPORT_SYMBOL(generic_iommu_put_resv_regions);
2748
2749struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2750						  size_t length, int prot,
2751						  enum iommu_resv_type type)
2752{
2753	struct iommu_resv_region *region;
2754
2755	region = kzalloc(sizeof(*region), GFP_KERNEL);
2756	if (!region)
2757		return NULL;
2758
2759	INIT_LIST_HEAD(&region->list);
2760	region->start = start;
2761	region->length = length;
2762	region->prot = prot;
2763	region->type = type;
2764	return region;
2765}
2766EXPORT_SYMBOL_GPL(iommu_alloc_resv_region);
2767
2768void iommu_set_default_passthrough(bool cmd_line)
2769{
2770	if (cmd_line)
2771		iommu_set_cmd_line_dma_api();
2772
2773	iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2774}
2775
2776void iommu_set_default_translated(bool cmd_line)
2777{
2778	if (cmd_line)
2779		iommu_set_cmd_line_dma_api();
2780
2781	iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2782}
2783
2784bool iommu_default_passthrough(void)
2785{
2786	return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2787}
2788EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2789
2790const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2791{
2792	const struct iommu_ops *ops = NULL;
2793	struct iommu_device *iommu;
2794
2795	spin_lock(&iommu_device_lock);
2796	list_for_each_entry(iommu, &iommu_device_list, list)
2797		if (iommu->fwnode == fwnode) {
2798			ops = iommu->ops;
2799			break;
2800		}
2801	spin_unlock(&iommu_device_lock);
2802	return ops;
2803}
2804
2805int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2806		      const struct iommu_ops *ops)
2807{
2808	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2809
2810	if (fwspec)
2811		return ops == fwspec->ops ? 0 : -EINVAL;
2812
2813	if (!dev_iommu_get(dev))
2814		return -ENOMEM;
2815
2816	/* Preallocate for the overwhelmingly common case of 1 ID */
2817	fwspec = kzalloc(struct_size(fwspec, ids, 1), GFP_KERNEL);
2818	if (!fwspec)
2819		return -ENOMEM;
2820
2821	of_node_get(to_of_node(iommu_fwnode));
2822	fwspec->iommu_fwnode = iommu_fwnode;
2823	fwspec->ops = ops;
2824	dev_iommu_fwspec_set(dev, fwspec);
2825	return 0;
2826}
2827EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2828
2829void iommu_fwspec_free(struct device *dev)
2830{
2831	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2832
2833	if (fwspec) {
2834		fwnode_handle_put(fwspec->iommu_fwnode);
2835		kfree(fwspec);
2836		dev_iommu_fwspec_set(dev, NULL);
2837	}
2838}
2839EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2840
2841int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2842{
2843	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2844	int i, new_num;
2845
2846	if (!fwspec)
2847		return -EINVAL;
2848
2849	new_num = fwspec->num_ids + num_ids;
2850	if (new_num > 1) {
2851		fwspec = krealloc(fwspec, struct_size(fwspec, ids, new_num),
2852				  GFP_KERNEL);
2853		if (!fwspec)
2854			return -ENOMEM;
2855
2856		dev_iommu_fwspec_set(dev, fwspec);
2857	}
2858
2859	for (i = 0; i < num_ids; i++)
2860		fwspec->ids[fwspec->num_ids + i] = ids[i];
2861
2862	fwspec->num_ids = new_num;
2863	return 0;
2864}
2865EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2866
2867/*
2868 * Per device IOMMU features.
2869 */
2870bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2871{
2872	const struct iommu_ops *ops = dev->bus->iommu_ops;
2873
2874	if (ops && ops->dev_has_feat)
2875		return ops->dev_has_feat(dev, feat);
2876
2877	return false;
2878}
2879EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2880
2881int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2882{
2883	if (dev->iommu && dev->iommu->iommu_dev) {
2884		const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2885
2886		if (ops->dev_enable_feat)
2887			return ops->dev_enable_feat(dev, feat);
2888	}
2889
2890	return -ENODEV;
2891}
2892EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2893
2894/*
2895 * The device drivers should do the necessary cleanups before calling this.
2896 * For example, before disabling the aux-domain feature, the device driver
2897 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2898 */
2899int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2900{
2901	if (dev->iommu && dev->iommu->iommu_dev) {
2902		const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2903
2904		if (ops->dev_disable_feat)
2905			return ops->dev_disable_feat(dev, feat);
2906	}
2907
2908	return -EBUSY;
2909}
2910EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2911
2912bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2913{
2914	if (dev->iommu && dev->iommu->iommu_dev) {
2915		const struct iommu_ops *ops = dev->iommu->iommu_dev->ops;
2916
2917		if (ops->dev_feat_enabled)
2918			return ops->dev_feat_enabled(dev, feat);
2919	}
2920
2921	return false;
2922}
2923EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2924
2925/*
2926 * Aux-domain specific attach/detach.
2927 *
2928 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2929 * true. Also, as long as domains are attached to a device through this
2930 * interface, any tries to call iommu_attach_device() should fail
2931 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2932 * This should make us safe against a device being attached to a guest as a
2933 * whole while there are still pasid users on it (aux and sva).
2934 */
2935int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2936{
2937	int ret = -ENODEV;
2938
2939	if (domain->ops->aux_attach_dev)
2940		ret = domain->ops->aux_attach_dev(domain, dev);
2941
2942	if (!ret)
2943		trace_attach_device_to_domain(dev);
2944
2945	return ret;
2946}
2947EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2948
2949void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2950{
2951	if (domain->ops->aux_detach_dev) {
2952		domain->ops->aux_detach_dev(domain, dev);
2953		trace_detach_device_from_domain(dev);
2954	}
2955}
2956EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2957
2958int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2959{
2960	int ret = -ENODEV;
2961
2962	if (domain->ops->aux_get_pasid)
2963		ret = domain->ops->aux_get_pasid(domain, dev);
2964
2965	return ret;
2966}
2967EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
2968
2969/**
2970 * iommu_sva_bind_device() - Bind a process address space to a device
2971 * @dev: the device
2972 * @mm: the mm to bind, caller must hold a reference to it
2973 *
2974 * Create a bond between device and address space, allowing the device to access
2975 * the mm using the returned PASID. If a bond already exists between @device and
2976 * @mm, it is returned and an additional reference is taken. Caller must call
2977 * iommu_sva_unbind_device() to release each reference.
2978 *
2979 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2980 * initialize the required SVA features.
2981 *
2982 * On error, returns an ERR_PTR value.
2983 */
2984struct iommu_sva *
2985iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2986{
2987	struct iommu_group *group;
2988	struct iommu_sva *handle = ERR_PTR(-EINVAL);
2989	const struct iommu_ops *ops = dev->bus->iommu_ops;
2990
2991	if (!ops || !ops->sva_bind)
2992		return ERR_PTR(-ENODEV);
2993
2994	group = iommu_group_get(dev);
2995	if (!group)
2996		return ERR_PTR(-ENODEV);
2997
2998	/* Ensure device count and domain don't change while we're binding */
2999	mutex_lock(&group->mutex);
3000
3001	/*
3002	 * To keep things simple, SVA currently doesn't support IOMMU groups
3003	 * with more than one device. Existing SVA-capable systems are not
3004	 * affected by the problems that required IOMMU groups (lack of ACS
3005	 * isolation, device ID aliasing and other hardware issues).
3006	 */
3007	if (iommu_group_device_count(group) != 1)
3008		goto out_unlock;
3009
3010	handle = ops->sva_bind(dev, mm, drvdata);
3011
3012out_unlock:
3013	mutex_unlock(&group->mutex);
3014	iommu_group_put(group);
3015
3016	return handle;
3017}
3018EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
3019
3020/**
3021 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
3022 * @handle: the handle returned by iommu_sva_bind_device()
3023 *
3024 * Put reference to a bond between device and address space. The device should
3025 * not be issuing any more transaction for this PASID. All outstanding page
3026 * requests for this PASID must have been flushed to the IOMMU.
3027 *
3028 * Returns 0 on success, or an error value
3029 */
3030void iommu_sva_unbind_device(struct iommu_sva *handle)
3031{
3032	struct iommu_group *group;
3033	struct device *dev = handle->dev;
3034	const struct iommu_ops *ops = dev->bus->iommu_ops;
3035
3036	if (!ops || !ops->sva_unbind)
3037		return;
3038
3039	group = iommu_group_get(dev);
3040	if (!group)
3041		return;
3042
3043	mutex_lock(&group->mutex);
3044	ops->sva_unbind(handle);
3045	mutex_unlock(&group->mutex);
3046
3047	iommu_group_put(group);
3048}
3049EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
3050
3051u32 iommu_sva_get_pasid(struct iommu_sva *handle)
3052{
3053	const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
3054
3055	if (!ops || !ops->sva_get_pasid)
3056		return IOMMU_PASID_INVALID;
3057
3058	return ops->sva_get_pasid(handle);
3059}
3060EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
3061