xref: /linux/drivers/iommu/iommu.c (revision 4eb7ae7a301d3586c3351e81d5c3cfe2304a1a6a)
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 <trace/events/iommu.h>
26 
27 static struct kset *iommu_group_kset;
28 static DEFINE_IDA(iommu_group_ida);
29 
30 static unsigned int iommu_def_domain_type __read_mostly;
31 static bool iommu_dma_strict __read_mostly = true;
32 static u32 iommu_cmd_line __read_mostly;
33 
34 struct iommu_group {
35 	struct kobject kobj;
36 	struct kobject *devices_kobj;
37 	struct list_head devices;
38 	struct mutex mutex;
39 	struct blocking_notifier_head notifier;
40 	void *iommu_data;
41 	void (*iommu_data_release)(void *iommu_data);
42 	char *name;
43 	int id;
44 	struct iommu_domain *default_domain;
45 	struct iommu_domain *domain;
46 };
47 
48 struct group_device {
49 	struct list_head list;
50 	struct device *dev;
51 	char *name;
52 };
53 
54 struct iommu_group_attribute {
55 	struct attribute attr;
56 	ssize_t (*show)(struct iommu_group *group, char *buf);
57 	ssize_t (*store)(struct iommu_group *group,
58 			 const char *buf, size_t count);
59 };
60 
61 static const char * const iommu_group_resv_type_string[] = {
62 	[IOMMU_RESV_DIRECT]			= "direct",
63 	[IOMMU_RESV_DIRECT_RELAXABLE]		= "direct-relaxable",
64 	[IOMMU_RESV_RESERVED]			= "reserved",
65 	[IOMMU_RESV_MSI]			= "msi",
66 	[IOMMU_RESV_SW_MSI]			= "msi",
67 };
68 
69 #define IOMMU_CMD_LINE_DMA_API		BIT(0)
70 
71 static void iommu_set_cmd_line_dma_api(void)
72 {
73 	iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
74 }
75 
76 static bool iommu_cmd_line_dma_api(void)
77 {
78 	return !!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API);
79 }
80 
81 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store)		\
82 struct iommu_group_attribute iommu_group_attr_##_name =		\
83 	__ATTR(_name, _mode, _show, _store)
84 
85 #define to_iommu_group_attr(_attr)	\
86 	container_of(_attr, struct iommu_group_attribute, attr)
87 #define to_iommu_group(_kobj)		\
88 	container_of(_kobj, struct iommu_group, kobj)
89 
90 static LIST_HEAD(iommu_device_list);
91 static DEFINE_SPINLOCK(iommu_device_lock);
92 
93 /*
94  * Use a function instead of an array here because the domain-type is a
95  * bit-field, so an array would waste memory.
96  */
97 static const char *iommu_domain_type_str(unsigned int t)
98 {
99 	switch (t) {
100 	case IOMMU_DOMAIN_BLOCKED:
101 		return "Blocked";
102 	case IOMMU_DOMAIN_IDENTITY:
103 		return "Passthrough";
104 	case IOMMU_DOMAIN_UNMANAGED:
105 		return "Unmanaged";
106 	case IOMMU_DOMAIN_DMA:
107 		return "Translated";
108 	default:
109 		return "Unknown";
110 	}
111 }
112 
113 static int __init iommu_subsys_init(void)
114 {
115 	bool cmd_line = iommu_cmd_line_dma_api();
116 
117 	if (!cmd_line) {
118 		if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
119 			iommu_set_default_passthrough(false);
120 		else
121 			iommu_set_default_translated(false);
122 
123 		if (iommu_default_passthrough() && mem_encrypt_active()) {
124 			pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
125 			iommu_set_default_translated(false);
126 		}
127 	}
128 
129 	pr_info("Default domain type: %s %s\n",
130 		iommu_domain_type_str(iommu_def_domain_type),
131 		cmd_line ? "(set via kernel command line)" : "");
132 
133 	return 0;
134 }
135 subsys_initcall(iommu_subsys_init);
136 
137 int iommu_device_register(struct iommu_device *iommu)
138 {
139 	spin_lock(&iommu_device_lock);
140 	list_add_tail(&iommu->list, &iommu_device_list);
141 	spin_unlock(&iommu_device_lock);
142 	return 0;
143 }
144 
145 void iommu_device_unregister(struct iommu_device *iommu)
146 {
147 	spin_lock(&iommu_device_lock);
148 	list_del(&iommu->list);
149 	spin_unlock(&iommu_device_lock);
150 }
151 
152 static struct iommu_param *iommu_get_dev_param(struct device *dev)
153 {
154 	struct iommu_param *param = dev->iommu_param;
155 
156 	if (param)
157 		return param;
158 
159 	param = kzalloc(sizeof(*param), GFP_KERNEL);
160 	if (!param)
161 		return NULL;
162 
163 	mutex_init(&param->lock);
164 	dev->iommu_param = param;
165 	return param;
166 }
167 
168 static void iommu_free_dev_param(struct device *dev)
169 {
170 	kfree(dev->iommu_param);
171 	dev->iommu_param = NULL;
172 }
173 
174 int iommu_probe_device(struct device *dev)
175 {
176 	const struct iommu_ops *ops = dev->bus->iommu_ops;
177 	int ret;
178 
179 	WARN_ON(dev->iommu_group);
180 	if (!ops)
181 		return -EINVAL;
182 
183 	if (!iommu_get_dev_param(dev))
184 		return -ENOMEM;
185 
186 	ret = ops->add_device(dev);
187 	if (ret)
188 		iommu_free_dev_param(dev);
189 
190 	return ret;
191 }
192 
193 void iommu_release_device(struct device *dev)
194 {
195 	const struct iommu_ops *ops = dev->bus->iommu_ops;
196 
197 	if (dev->iommu_group)
198 		ops->remove_device(dev);
199 
200 	iommu_free_dev_param(dev);
201 }
202 
203 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
204 						 unsigned type);
205 static int __iommu_attach_device(struct iommu_domain *domain,
206 				 struct device *dev);
207 static int __iommu_attach_group(struct iommu_domain *domain,
208 				struct iommu_group *group);
209 static void __iommu_detach_group(struct iommu_domain *domain,
210 				 struct iommu_group *group);
211 
212 static int __init iommu_set_def_domain_type(char *str)
213 {
214 	bool pt;
215 	int ret;
216 
217 	ret = kstrtobool(str, &pt);
218 	if (ret)
219 		return ret;
220 
221 	if (pt)
222 		iommu_set_default_passthrough(true);
223 	else
224 		iommu_set_default_translated(true);
225 
226 	return 0;
227 }
228 early_param("iommu.passthrough", iommu_set_def_domain_type);
229 
230 static int __init iommu_dma_setup(char *str)
231 {
232 	return kstrtobool(str, &iommu_dma_strict);
233 }
234 early_param("iommu.strict", iommu_dma_setup);
235 
236 static ssize_t iommu_group_attr_show(struct kobject *kobj,
237 				     struct attribute *__attr, char *buf)
238 {
239 	struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
240 	struct iommu_group *group = to_iommu_group(kobj);
241 	ssize_t ret = -EIO;
242 
243 	if (attr->show)
244 		ret = attr->show(group, buf);
245 	return ret;
246 }
247 
248 static ssize_t iommu_group_attr_store(struct kobject *kobj,
249 				      struct attribute *__attr,
250 				      const char *buf, size_t count)
251 {
252 	struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
253 	struct iommu_group *group = to_iommu_group(kobj);
254 	ssize_t ret = -EIO;
255 
256 	if (attr->store)
257 		ret = attr->store(group, buf, count);
258 	return ret;
259 }
260 
261 static const struct sysfs_ops iommu_group_sysfs_ops = {
262 	.show = iommu_group_attr_show,
263 	.store = iommu_group_attr_store,
264 };
265 
266 static int iommu_group_create_file(struct iommu_group *group,
267 				   struct iommu_group_attribute *attr)
268 {
269 	return sysfs_create_file(&group->kobj, &attr->attr);
270 }
271 
272 static void iommu_group_remove_file(struct iommu_group *group,
273 				    struct iommu_group_attribute *attr)
274 {
275 	sysfs_remove_file(&group->kobj, &attr->attr);
276 }
277 
278 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
279 {
280 	return sprintf(buf, "%s\n", group->name);
281 }
282 
283 /**
284  * iommu_insert_resv_region - Insert a new region in the
285  * list of reserved regions.
286  * @new: new region to insert
287  * @regions: list of regions
288  *
289  * Elements are sorted by start address and overlapping segments
290  * of the same type are merged.
291  */
292 int iommu_insert_resv_region(struct iommu_resv_region *new,
293 			     struct list_head *regions)
294 {
295 	struct iommu_resv_region *iter, *tmp, *nr, *top;
296 	LIST_HEAD(stack);
297 
298 	nr = iommu_alloc_resv_region(new->start, new->length,
299 				     new->prot, new->type);
300 	if (!nr)
301 		return -ENOMEM;
302 
303 	/* First add the new element based on start address sorting */
304 	list_for_each_entry(iter, regions, list) {
305 		if (nr->start < iter->start ||
306 		    (nr->start == iter->start && nr->type <= iter->type))
307 			break;
308 	}
309 	list_add_tail(&nr->list, &iter->list);
310 
311 	/* Merge overlapping segments of type nr->type in @regions, if any */
312 	list_for_each_entry_safe(iter, tmp, regions, list) {
313 		phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
314 
315 		/* no merge needed on elements of different types than @nr */
316 		if (iter->type != nr->type) {
317 			list_move_tail(&iter->list, &stack);
318 			continue;
319 		}
320 
321 		/* look for the last stack element of same type as @iter */
322 		list_for_each_entry_reverse(top, &stack, list)
323 			if (top->type == iter->type)
324 				goto check_overlap;
325 
326 		list_move_tail(&iter->list, &stack);
327 		continue;
328 
329 check_overlap:
330 		top_end = top->start + top->length - 1;
331 
332 		if (iter->start > top_end + 1) {
333 			list_move_tail(&iter->list, &stack);
334 		} else {
335 			top->length = max(top_end, iter_end) - top->start + 1;
336 			list_del(&iter->list);
337 			kfree(iter);
338 		}
339 	}
340 	list_splice(&stack, regions);
341 	return 0;
342 }
343 
344 static int
345 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
346 				 struct list_head *group_resv_regions)
347 {
348 	struct iommu_resv_region *entry;
349 	int ret = 0;
350 
351 	list_for_each_entry(entry, dev_resv_regions, list) {
352 		ret = iommu_insert_resv_region(entry, group_resv_regions);
353 		if (ret)
354 			break;
355 	}
356 	return ret;
357 }
358 
359 int iommu_get_group_resv_regions(struct iommu_group *group,
360 				 struct list_head *head)
361 {
362 	struct group_device *device;
363 	int ret = 0;
364 
365 	mutex_lock(&group->mutex);
366 	list_for_each_entry(device, &group->devices, list) {
367 		struct list_head dev_resv_regions;
368 
369 		INIT_LIST_HEAD(&dev_resv_regions);
370 		iommu_get_resv_regions(device->dev, &dev_resv_regions);
371 		ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
372 		iommu_put_resv_regions(device->dev, &dev_resv_regions);
373 		if (ret)
374 			break;
375 	}
376 	mutex_unlock(&group->mutex);
377 	return ret;
378 }
379 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
380 
381 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
382 					     char *buf)
383 {
384 	struct iommu_resv_region *region, *next;
385 	struct list_head group_resv_regions;
386 	char *str = buf;
387 
388 	INIT_LIST_HEAD(&group_resv_regions);
389 	iommu_get_group_resv_regions(group, &group_resv_regions);
390 
391 	list_for_each_entry_safe(region, next, &group_resv_regions, list) {
392 		str += sprintf(str, "0x%016llx 0x%016llx %s\n",
393 			       (long long int)region->start,
394 			       (long long int)(region->start +
395 						region->length - 1),
396 			       iommu_group_resv_type_string[region->type]);
397 		kfree(region);
398 	}
399 
400 	return (str - buf);
401 }
402 
403 static ssize_t iommu_group_show_type(struct iommu_group *group,
404 				     char *buf)
405 {
406 	char *type = "unknown\n";
407 
408 	if (group->default_domain) {
409 		switch (group->default_domain->type) {
410 		case IOMMU_DOMAIN_BLOCKED:
411 			type = "blocked\n";
412 			break;
413 		case IOMMU_DOMAIN_IDENTITY:
414 			type = "identity\n";
415 			break;
416 		case IOMMU_DOMAIN_UNMANAGED:
417 			type = "unmanaged\n";
418 			break;
419 		case IOMMU_DOMAIN_DMA:
420 			type = "DMA\n";
421 			break;
422 		}
423 	}
424 	strcpy(buf, type);
425 
426 	return strlen(type);
427 }
428 
429 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
430 
431 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
432 			iommu_group_show_resv_regions, NULL);
433 
434 static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
435 
436 static void iommu_group_release(struct kobject *kobj)
437 {
438 	struct iommu_group *group = to_iommu_group(kobj);
439 
440 	pr_debug("Releasing group %d\n", group->id);
441 
442 	if (group->iommu_data_release)
443 		group->iommu_data_release(group->iommu_data);
444 
445 	ida_simple_remove(&iommu_group_ida, group->id);
446 
447 	if (group->default_domain)
448 		iommu_domain_free(group->default_domain);
449 
450 	kfree(group->name);
451 	kfree(group);
452 }
453 
454 static struct kobj_type iommu_group_ktype = {
455 	.sysfs_ops = &iommu_group_sysfs_ops,
456 	.release = iommu_group_release,
457 };
458 
459 /**
460  * iommu_group_alloc - Allocate a new group
461  *
462  * This function is called by an iommu driver to allocate a new iommu
463  * group.  The iommu group represents the minimum granularity of the iommu.
464  * Upon successful return, the caller holds a reference to the supplied
465  * group in order to hold the group until devices are added.  Use
466  * iommu_group_put() to release this extra reference count, allowing the
467  * group to be automatically reclaimed once it has no devices or external
468  * references.
469  */
470 struct iommu_group *iommu_group_alloc(void)
471 {
472 	struct iommu_group *group;
473 	int ret;
474 
475 	group = kzalloc(sizeof(*group), GFP_KERNEL);
476 	if (!group)
477 		return ERR_PTR(-ENOMEM);
478 
479 	group->kobj.kset = iommu_group_kset;
480 	mutex_init(&group->mutex);
481 	INIT_LIST_HEAD(&group->devices);
482 	BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
483 
484 	ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
485 	if (ret < 0) {
486 		kfree(group);
487 		return ERR_PTR(ret);
488 	}
489 	group->id = ret;
490 
491 	ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
492 				   NULL, "%d", group->id);
493 	if (ret) {
494 		ida_simple_remove(&iommu_group_ida, group->id);
495 		kfree(group);
496 		return ERR_PTR(ret);
497 	}
498 
499 	group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
500 	if (!group->devices_kobj) {
501 		kobject_put(&group->kobj); /* triggers .release & free */
502 		return ERR_PTR(-ENOMEM);
503 	}
504 
505 	/*
506 	 * The devices_kobj holds a reference on the group kobject, so
507 	 * as long as that exists so will the group.  We can therefore
508 	 * use the devices_kobj for reference counting.
509 	 */
510 	kobject_put(&group->kobj);
511 
512 	ret = iommu_group_create_file(group,
513 				      &iommu_group_attr_reserved_regions);
514 	if (ret)
515 		return ERR_PTR(ret);
516 
517 	ret = iommu_group_create_file(group, &iommu_group_attr_type);
518 	if (ret)
519 		return ERR_PTR(ret);
520 
521 	pr_debug("Allocated group %d\n", group->id);
522 
523 	return group;
524 }
525 EXPORT_SYMBOL_GPL(iommu_group_alloc);
526 
527 struct iommu_group *iommu_group_get_by_id(int id)
528 {
529 	struct kobject *group_kobj;
530 	struct iommu_group *group;
531 	const char *name;
532 
533 	if (!iommu_group_kset)
534 		return NULL;
535 
536 	name = kasprintf(GFP_KERNEL, "%d", id);
537 	if (!name)
538 		return NULL;
539 
540 	group_kobj = kset_find_obj(iommu_group_kset, name);
541 	kfree(name);
542 
543 	if (!group_kobj)
544 		return NULL;
545 
546 	group = container_of(group_kobj, struct iommu_group, kobj);
547 	BUG_ON(group->id != id);
548 
549 	kobject_get(group->devices_kobj);
550 	kobject_put(&group->kobj);
551 
552 	return group;
553 }
554 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
555 
556 /**
557  * iommu_group_get_iommudata - retrieve iommu_data registered for a group
558  * @group: the group
559  *
560  * iommu drivers can store data in the group for use when doing iommu
561  * operations.  This function provides a way to retrieve it.  Caller
562  * should hold a group reference.
563  */
564 void *iommu_group_get_iommudata(struct iommu_group *group)
565 {
566 	return group->iommu_data;
567 }
568 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
569 
570 /**
571  * iommu_group_set_iommudata - set iommu_data for a group
572  * @group: the group
573  * @iommu_data: new data
574  * @release: release function for iommu_data
575  *
576  * iommu drivers can store data in the group for use when doing iommu
577  * operations.  This function provides a way to set the data after
578  * the group has been allocated.  Caller should hold a group reference.
579  */
580 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
581 			       void (*release)(void *iommu_data))
582 {
583 	group->iommu_data = iommu_data;
584 	group->iommu_data_release = release;
585 }
586 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
587 
588 /**
589  * iommu_group_set_name - set name for a group
590  * @group: the group
591  * @name: name
592  *
593  * Allow iommu driver to set a name for a group.  When set it will
594  * appear in a name attribute file under the group in sysfs.
595  */
596 int iommu_group_set_name(struct iommu_group *group, const char *name)
597 {
598 	int ret;
599 
600 	if (group->name) {
601 		iommu_group_remove_file(group, &iommu_group_attr_name);
602 		kfree(group->name);
603 		group->name = NULL;
604 		if (!name)
605 			return 0;
606 	}
607 
608 	group->name = kstrdup(name, GFP_KERNEL);
609 	if (!group->name)
610 		return -ENOMEM;
611 
612 	ret = iommu_group_create_file(group, &iommu_group_attr_name);
613 	if (ret) {
614 		kfree(group->name);
615 		group->name = NULL;
616 		return ret;
617 	}
618 
619 	return 0;
620 }
621 EXPORT_SYMBOL_GPL(iommu_group_set_name);
622 
623 static int iommu_group_create_direct_mappings(struct iommu_group *group,
624 					      struct device *dev)
625 {
626 	struct iommu_domain *domain = group->default_domain;
627 	struct iommu_resv_region *entry;
628 	struct list_head mappings;
629 	unsigned long pg_size;
630 	int ret = 0;
631 
632 	if (!domain || domain->type != IOMMU_DOMAIN_DMA)
633 		return 0;
634 
635 	BUG_ON(!domain->pgsize_bitmap);
636 
637 	pg_size = 1UL << __ffs(domain->pgsize_bitmap);
638 	INIT_LIST_HEAD(&mappings);
639 
640 	iommu_get_resv_regions(dev, &mappings);
641 
642 	/* We need to consider overlapping regions for different devices */
643 	list_for_each_entry(entry, &mappings, list) {
644 		dma_addr_t start, end, addr;
645 
646 		if (domain->ops->apply_resv_region)
647 			domain->ops->apply_resv_region(dev, domain, entry);
648 
649 		start = ALIGN(entry->start, pg_size);
650 		end   = ALIGN(entry->start + entry->length, pg_size);
651 
652 		if (entry->type != IOMMU_RESV_DIRECT &&
653 		    entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
654 			continue;
655 
656 		for (addr = start; addr < end; addr += pg_size) {
657 			phys_addr_t phys_addr;
658 
659 			phys_addr = iommu_iova_to_phys(domain, addr);
660 			if (phys_addr)
661 				continue;
662 
663 			ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
664 			if (ret)
665 				goto out;
666 		}
667 
668 	}
669 
670 	iommu_flush_tlb_all(domain);
671 
672 out:
673 	iommu_put_resv_regions(dev, &mappings);
674 
675 	return ret;
676 }
677 
678 /**
679  * iommu_group_add_device - add a device to an iommu group
680  * @group: the group into which to add the device (reference should be held)
681  * @dev: the device
682  *
683  * This function is called by an iommu driver to add a device into a
684  * group.  Adding a device increments the group reference count.
685  */
686 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
687 {
688 	int ret, i = 0;
689 	struct group_device *device;
690 
691 	device = kzalloc(sizeof(*device), GFP_KERNEL);
692 	if (!device)
693 		return -ENOMEM;
694 
695 	device->dev = dev;
696 
697 	ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
698 	if (ret)
699 		goto err_free_device;
700 
701 	device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
702 rename:
703 	if (!device->name) {
704 		ret = -ENOMEM;
705 		goto err_remove_link;
706 	}
707 
708 	ret = sysfs_create_link_nowarn(group->devices_kobj,
709 				       &dev->kobj, device->name);
710 	if (ret) {
711 		if (ret == -EEXIST && i >= 0) {
712 			/*
713 			 * Account for the slim chance of collision
714 			 * and append an instance to the name.
715 			 */
716 			kfree(device->name);
717 			device->name = kasprintf(GFP_KERNEL, "%s.%d",
718 						 kobject_name(&dev->kobj), i++);
719 			goto rename;
720 		}
721 		goto err_free_name;
722 	}
723 
724 	kobject_get(group->devices_kobj);
725 
726 	dev->iommu_group = group;
727 
728 	iommu_group_create_direct_mappings(group, dev);
729 
730 	mutex_lock(&group->mutex);
731 	list_add_tail(&device->list, &group->devices);
732 	if (group->domain)
733 		ret = __iommu_attach_device(group->domain, dev);
734 	mutex_unlock(&group->mutex);
735 	if (ret)
736 		goto err_put_group;
737 
738 	/* Notify any listeners about change to group. */
739 	blocking_notifier_call_chain(&group->notifier,
740 				     IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
741 
742 	trace_add_device_to_group(group->id, dev);
743 
744 	dev_info(dev, "Adding to iommu group %d\n", group->id);
745 
746 	return 0;
747 
748 err_put_group:
749 	mutex_lock(&group->mutex);
750 	list_del(&device->list);
751 	mutex_unlock(&group->mutex);
752 	dev->iommu_group = NULL;
753 	kobject_put(group->devices_kobj);
754 err_free_name:
755 	kfree(device->name);
756 err_remove_link:
757 	sysfs_remove_link(&dev->kobj, "iommu_group");
758 err_free_device:
759 	kfree(device);
760 	dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
761 	return ret;
762 }
763 EXPORT_SYMBOL_GPL(iommu_group_add_device);
764 
765 /**
766  * iommu_group_remove_device - remove a device from it's current group
767  * @dev: device to be removed
768  *
769  * This function is called by an iommu driver to remove the device from
770  * it's current group.  This decrements the iommu group reference count.
771  */
772 void iommu_group_remove_device(struct device *dev)
773 {
774 	struct iommu_group *group = dev->iommu_group;
775 	struct group_device *tmp_device, *device = NULL;
776 
777 	dev_info(dev, "Removing from iommu group %d\n", group->id);
778 
779 	/* Pre-notify listeners that a device is being removed. */
780 	blocking_notifier_call_chain(&group->notifier,
781 				     IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
782 
783 	mutex_lock(&group->mutex);
784 	list_for_each_entry(tmp_device, &group->devices, list) {
785 		if (tmp_device->dev == dev) {
786 			device = tmp_device;
787 			list_del(&device->list);
788 			break;
789 		}
790 	}
791 	mutex_unlock(&group->mutex);
792 
793 	if (!device)
794 		return;
795 
796 	sysfs_remove_link(group->devices_kobj, device->name);
797 	sysfs_remove_link(&dev->kobj, "iommu_group");
798 
799 	trace_remove_device_from_group(group->id, dev);
800 
801 	kfree(device->name);
802 	kfree(device);
803 	dev->iommu_group = NULL;
804 	kobject_put(group->devices_kobj);
805 }
806 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
807 
808 static int iommu_group_device_count(struct iommu_group *group)
809 {
810 	struct group_device *entry;
811 	int ret = 0;
812 
813 	list_for_each_entry(entry, &group->devices, list)
814 		ret++;
815 
816 	return ret;
817 }
818 
819 /**
820  * iommu_group_for_each_dev - iterate over each device in the group
821  * @group: the group
822  * @data: caller opaque data to be passed to callback function
823  * @fn: caller supplied callback function
824  *
825  * This function is called by group users to iterate over group devices.
826  * Callers should hold a reference count to the group during callback.
827  * The group->mutex is held across callbacks, which will block calls to
828  * iommu_group_add/remove_device.
829  */
830 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
831 				      int (*fn)(struct device *, void *))
832 {
833 	struct group_device *device;
834 	int ret = 0;
835 
836 	list_for_each_entry(device, &group->devices, list) {
837 		ret = fn(device->dev, data);
838 		if (ret)
839 			break;
840 	}
841 	return ret;
842 }
843 
844 
845 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
846 			     int (*fn)(struct device *, void *))
847 {
848 	int ret;
849 
850 	mutex_lock(&group->mutex);
851 	ret = __iommu_group_for_each_dev(group, data, fn);
852 	mutex_unlock(&group->mutex);
853 
854 	return ret;
855 }
856 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
857 
858 /**
859  * iommu_group_get - Return the group for a device and increment reference
860  * @dev: get the group that this device belongs to
861  *
862  * This function is called by iommu drivers and users to get the group
863  * for the specified device.  If found, the group is returned and the group
864  * reference in incremented, else NULL.
865  */
866 struct iommu_group *iommu_group_get(struct device *dev)
867 {
868 	struct iommu_group *group = dev->iommu_group;
869 
870 	if (group)
871 		kobject_get(group->devices_kobj);
872 
873 	return group;
874 }
875 EXPORT_SYMBOL_GPL(iommu_group_get);
876 
877 /**
878  * iommu_group_ref_get - Increment reference on a group
879  * @group: the group to use, must not be NULL
880  *
881  * This function is called by iommu drivers to take additional references on an
882  * existing group.  Returns the given group for convenience.
883  */
884 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
885 {
886 	kobject_get(group->devices_kobj);
887 	return group;
888 }
889 
890 /**
891  * iommu_group_put - Decrement group reference
892  * @group: the group to use
893  *
894  * This function is called by iommu drivers and users to release the
895  * iommu group.  Once the reference count is zero, the group is released.
896  */
897 void iommu_group_put(struct iommu_group *group)
898 {
899 	if (group)
900 		kobject_put(group->devices_kobj);
901 }
902 EXPORT_SYMBOL_GPL(iommu_group_put);
903 
904 /**
905  * iommu_group_register_notifier - Register a notifier for group changes
906  * @group: the group to watch
907  * @nb: notifier block to signal
908  *
909  * This function allows iommu group users to track changes in a group.
910  * See include/linux/iommu.h for actions sent via this notifier.  Caller
911  * should hold a reference to the group throughout notifier registration.
912  */
913 int iommu_group_register_notifier(struct iommu_group *group,
914 				  struct notifier_block *nb)
915 {
916 	return blocking_notifier_chain_register(&group->notifier, nb);
917 }
918 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
919 
920 /**
921  * iommu_group_unregister_notifier - Unregister a notifier
922  * @group: the group to watch
923  * @nb: notifier block to signal
924  *
925  * Unregister a previously registered group notifier block.
926  */
927 int iommu_group_unregister_notifier(struct iommu_group *group,
928 				    struct notifier_block *nb)
929 {
930 	return blocking_notifier_chain_unregister(&group->notifier, nb);
931 }
932 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
933 
934 /**
935  * iommu_register_device_fault_handler() - Register a device fault handler
936  * @dev: the device
937  * @handler: the fault handler
938  * @data: private data passed as argument to the handler
939  *
940  * When an IOMMU fault event is received, this handler gets called with the
941  * fault event and data as argument. The handler should return 0 on success. If
942  * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
943  * complete the fault by calling iommu_page_response() with one of the following
944  * response code:
945  * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
946  * - IOMMU_PAGE_RESP_INVALID: terminate the fault
947  * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
948  *   page faults if possible.
949  *
950  * Return 0 if the fault handler was installed successfully, or an error.
951  */
952 int iommu_register_device_fault_handler(struct device *dev,
953 					iommu_dev_fault_handler_t handler,
954 					void *data)
955 {
956 	struct iommu_param *param = dev->iommu_param;
957 	int ret = 0;
958 
959 	if (!param)
960 		return -EINVAL;
961 
962 	mutex_lock(&param->lock);
963 	/* Only allow one fault handler registered for each device */
964 	if (param->fault_param) {
965 		ret = -EBUSY;
966 		goto done_unlock;
967 	}
968 
969 	get_device(dev);
970 	param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
971 	if (!param->fault_param) {
972 		put_device(dev);
973 		ret = -ENOMEM;
974 		goto done_unlock;
975 	}
976 	param->fault_param->handler = handler;
977 	param->fault_param->data = data;
978 	mutex_init(&param->fault_param->lock);
979 	INIT_LIST_HEAD(&param->fault_param->faults);
980 
981 done_unlock:
982 	mutex_unlock(&param->lock);
983 
984 	return ret;
985 }
986 EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
987 
988 /**
989  * iommu_unregister_device_fault_handler() - Unregister the device fault handler
990  * @dev: the device
991  *
992  * Remove the device fault handler installed with
993  * iommu_register_device_fault_handler().
994  *
995  * Return 0 on success, or an error.
996  */
997 int iommu_unregister_device_fault_handler(struct device *dev)
998 {
999 	struct iommu_param *param = dev->iommu_param;
1000 	int ret = 0;
1001 
1002 	if (!param)
1003 		return -EINVAL;
1004 
1005 	mutex_lock(&param->lock);
1006 
1007 	if (!param->fault_param)
1008 		goto unlock;
1009 
1010 	/* we cannot unregister handler if there are pending faults */
1011 	if (!list_empty(&param->fault_param->faults)) {
1012 		ret = -EBUSY;
1013 		goto unlock;
1014 	}
1015 
1016 	kfree(param->fault_param);
1017 	param->fault_param = NULL;
1018 	put_device(dev);
1019 unlock:
1020 	mutex_unlock(&param->lock);
1021 
1022 	return ret;
1023 }
1024 EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1025 
1026 /**
1027  * iommu_report_device_fault() - Report fault event to device driver
1028  * @dev: the device
1029  * @evt: fault event data
1030  *
1031  * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
1032  * handler. When this function fails and the fault is recoverable, it is the
1033  * caller's responsibility to complete the fault.
1034  *
1035  * Return 0 on success, or an error.
1036  */
1037 int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1038 {
1039 	struct iommu_param *param = dev->iommu_param;
1040 	struct iommu_fault_event *evt_pending = NULL;
1041 	struct iommu_fault_param *fparam;
1042 	int ret = 0;
1043 
1044 	if (!param || !evt)
1045 		return -EINVAL;
1046 
1047 	/* we only report device fault if there is a handler registered */
1048 	mutex_lock(&param->lock);
1049 	fparam = param->fault_param;
1050 	if (!fparam || !fparam->handler) {
1051 		ret = -EINVAL;
1052 		goto done_unlock;
1053 	}
1054 
1055 	if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1056 	    (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1057 		evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1058 				      GFP_KERNEL);
1059 		if (!evt_pending) {
1060 			ret = -ENOMEM;
1061 			goto done_unlock;
1062 		}
1063 		mutex_lock(&fparam->lock);
1064 		list_add_tail(&evt_pending->list, &fparam->faults);
1065 		mutex_unlock(&fparam->lock);
1066 	}
1067 
1068 	ret = fparam->handler(&evt->fault, fparam->data);
1069 	if (ret && evt_pending) {
1070 		mutex_lock(&fparam->lock);
1071 		list_del(&evt_pending->list);
1072 		mutex_unlock(&fparam->lock);
1073 		kfree(evt_pending);
1074 	}
1075 done_unlock:
1076 	mutex_unlock(&param->lock);
1077 	return ret;
1078 }
1079 EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1080 
1081 int iommu_page_response(struct device *dev,
1082 			struct iommu_page_response *msg)
1083 {
1084 	bool pasid_valid;
1085 	int ret = -EINVAL;
1086 	struct iommu_fault_event *evt;
1087 	struct iommu_fault_page_request *prm;
1088 	struct iommu_param *param = dev->iommu_param;
1089 	struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1090 
1091 	if (!domain || !domain->ops->page_response)
1092 		return -ENODEV;
1093 
1094 	if (!param || !param->fault_param)
1095 		return -EINVAL;
1096 
1097 	if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1098 	    msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1099 		return -EINVAL;
1100 
1101 	/* Only send response if there is a fault report pending */
1102 	mutex_lock(&param->fault_param->lock);
1103 	if (list_empty(&param->fault_param->faults)) {
1104 		dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1105 		goto done_unlock;
1106 	}
1107 	/*
1108 	 * Check if we have a matching page request pending to respond,
1109 	 * otherwise return -EINVAL
1110 	 */
1111 	list_for_each_entry(evt, &param->fault_param->faults, list) {
1112 		prm = &evt->fault.prm;
1113 		pasid_valid = prm->flags & IOMMU_FAULT_PAGE_REQUEST_PASID_VALID;
1114 
1115 		if ((pasid_valid && prm->pasid != msg->pasid) ||
1116 		    prm->grpid != msg->grpid)
1117 			continue;
1118 
1119 		/* Sanitize the reply */
1120 		msg->flags = pasid_valid ? IOMMU_PAGE_RESP_PASID_VALID : 0;
1121 
1122 		ret = domain->ops->page_response(dev, evt, msg);
1123 		list_del(&evt->list);
1124 		kfree(evt);
1125 		break;
1126 	}
1127 
1128 done_unlock:
1129 	mutex_unlock(&param->fault_param->lock);
1130 	return ret;
1131 }
1132 EXPORT_SYMBOL_GPL(iommu_page_response);
1133 
1134 /**
1135  * iommu_group_id - Return ID for a group
1136  * @group: the group to ID
1137  *
1138  * Return the unique ID for the group matching the sysfs group number.
1139  */
1140 int iommu_group_id(struct iommu_group *group)
1141 {
1142 	return group->id;
1143 }
1144 EXPORT_SYMBOL_GPL(iommu_group_id);
1145 
1146 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1147 					       unsigned long *devfns);
1148 
1149 /*
1150  * To consider a PCI device isolated, we require ACS to support Source
1151  * Validation, Request Redirection, Completer Redirection, and Upstream
1152  * Forwarding.  This effectively means that devices cannot spoof their
1153  * requester ID, requests and completions cannot be redirected, and all
1154  * transactions are forwarded upstream, even as it passes through a
1155  * bridge where the target device is downstream.
1156  */
1157 #define REQ_ACS_FLAGS   (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1158 
1159 /*
1160  * For multifunction devices which are not isolated from each other, find
1161  * all the other non-isolated functions and look for existing groups.  For
1162  * each function, we also need to look for aliases to or from other devices
1163  * that may already have a group.
1164  */
1165 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1166 							unsigned long *devfns)
1167 {
1168 	struct pci_dev *tmp = NULL;
1169 	struct iommu_group *group;
1170 
1171 	if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1172 		return NULL;
1173 
1174 	for_each_pci_dev(tmp) {
1175 		if (tmp == pdev || tmp->bus != pdev->bus ||
1176 		    PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1177 		    pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1178 			continue;
1179 
1180 		group = get_pci_alias_group(tmp, devfns);
1181 		if (group) {
1182 			pci_dev_put(tmp);
1183 			return group;
1184 		}
1185 	}
1186 
1187 	return NULL;
1188 }
1189 
1190 /*
1191  * Look for aliases to or from the given device for existing groups. DMA
1192  * aliases are only supported on the same bus, therefore the search
1193  * space is quite small (especially since we're really only looking at pcie
1194  * device, and therefore only expect multiple slots on the root complex or
1195  * downstream switch ports).  It's conceivable though that a pair of
1196  * multifunction devices could have aliases between them that would cause a
1197  * loop.  To prevent this, we use a bitmap to track where we've been.
1198  */
1199 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1200 					       unsigned long *devfns)
1201 {
1202 	struct pci_dev *tmp = NULL;
1203 	struct iommu_group *group;
1204 
1205 	if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1206 		return NULL;
1207 
1208 	group = iommu_group_get(&pdev->dev);
1209 	if (group)
1210 		return group;
1211 
1212 	for_each_pci_dev(tmp) {
1213 		if (tmp == pdev || tmp->bus != pdev->bus)
1214 			continue;
1215 
1216 		/* We alias them or they alias us */
1217 		if (pci_devs_are_dma_aliases(pdev, tmp)) {
1218 			group = get_pci_alias_group(tmp, devfns);
1219 			if (group) {
1220 				pci_dev_put(tmp);
1221 				return group;
1222 			}
1223 
1224 			group = get_pci_function_alias_group(tmp, devfns);
1225 			if (group) {
1226 				pci_dev_put(tmp);
1227 				return group;
1228 			}
1229 		}
1230 	}
1231 
1232 	return NULL;
1233 }
1234 
1235 struct group_for_pci_data {
1236 	struct pci_dev *pdev;
1237 	struct iommu_group *group;
1238 };
1239 
1240 /*
1241  * DMA alias iterator callback, return the last seen device.  Stop and return
1242  * the IOMMU group if we find one along the way.
1243  */
1244 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1245 {
1246 	struct group_for_pci_data *data = opaque;
1247 
1248 	data->pdev = pdev;
1249 	data->group = iommu_group_get(&pdev->dev);
1250 
1251 	return data->group != NULL;
1252 }
1253 
1254 /*
1255  * Generic device_group call-back function. It just allocates one
1256  * iommu-group per device.
1257  */
1258 struct iommu_group *generic_device_group(struct device *dev)
1259 {
1260 	return iommu_group_alloc();
1261 }
1262 
1263 /*
1264  * Use standard PCI bus topology, isolation features, and DMA alias quirks
1265  * to find or create an IOMMU group for a device.
1266  */
1267 struct iommu_group *pci_device_group(struct device *dev)
1268 {
1269 	struct pci_dev *pdev = to_pci_dev(dev);
1270 	struct group_for_pci_data data;
1271 	struct pci_bus *bus;
1272 	struct iommu_group *group = NULL;
1273 	u64 devfns[4] = { 0 };
1274 
1275 	if (WARN_ON(!dev_is_pci(dev)))
1276 		return ERR_PTR(-EINVAL);
1277 
1278 	/*
1279 	 * Find the upstream DMA alias for the device.  A device must not
1280 	 * be aliased due to topology in order to have its own IOMMU group.
1281 	 * If we find an alias along the way that already belongs to a
1282 	 * group, use it.
1283 	 */
1284 	if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1285 		return data.group;
1286 
1287 	pdev = data.pdev;
1288 
1289 	/*
1290 	 * Continue upstream from the point of minimum IOMMU granularity
1291 	 * due to aliases to the point where devices are protected from
1292 	 * peer-to-peer DMA by PCI ACS.  Again, if we find an existing
1293 	 * group, use it.
1294 	 */
1295 	for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1296 		if (!bus->self)
1297 			continue;
1298 
1299 		if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1300 			break;
1301 
1302 		pdev = bus->self;
1303 
1304 		group = iommu_group_get(&pdev->dev);
1305 		if (group)
1306 			return group;
1307 	}
1308 
1309 	/*
1310 	 * Look for existing groups on device aliases.  If we alias another
1311 	 * device or another device aliases us, use the same group.
1312 	 */
1313 	group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1314 	if (group)
1315 		return group;
1316 
1317 	/*
1318 	 * Look for existing groups on non-isolated functions on the same
1319 	 * slot and aliases of those funcions, if any.  No need to clear
1320 	 * the search bitmap, the tested devfns are still valid.
1321 	 */
1322 	group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1323 	if (group)
1324 		return group;
1325 
1326 	/* No shared group found, allocate new */
1327 	return iommu_group_alloc();
1328 }
1329 
1330 /* Get the IOMMU group for device on fsl-mc bus */
1331 struct iommu_group *fsl_mc_device_group(struct device *dev)
1332 {
1333 	struct device *cont_dev = fsl_mc_cont_dev(dev);
1334 	struct iommu_group *group;
1335 
1336 	group = iommu_group_get(cont_dev);
1337 	if (!group)
1338 		group = iommu_group_alloc();
1339 	return group;
1340 }
1341 
1342 /**
1343  * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1344  * @dev: target device
1345  *
1346  * This function is intended to be called by IOMMU drivers and extended to
1347  * support common, bus-defined algorithms when determining or creating the
1348  * IOMMU group for a device.  On success, the caller will hold a reference
1349  * to the returned IOMMU group, which will already include the provided
1350  * device.  The reference should be released with iommu_group_put().
1351  */
1352 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1353 {
1354 	const struct iommu_ops *ops = dev->bus->iommu_ops;
1355 	struct iommu_group *group;
1356 	int ret;
1357 
1358 	group = iommu_group_get(dev);
1359 	if (group)
1360 		return group;
1361 
1362 	if (!ops)
1363 		return ERR_PTR(-EINVAL);
1364 
1365 	group = ops->device_group(dev);
1366 	if (WARN_ON_ONCE(group == NULL))
1367 		return ERR_PTR(-EINVAL);
1368 
1369 	if (IS_ERR(group))
1370 		return group;
1371 
1372 	/*
1373 	 * Try to allocate a default domain - needs support from the
1374 	 * IOMMU driver.
1375 	 */
1376 	if (!group->default_domain) {
1377 		struct iommu_domain *dom;
1378 
1379 		dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1380 		if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
1381 			dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
1382 			if (dom) {
1383 				dev_warn(dev,
1384 					 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1385 					 iommu_def_domain_type);
1386 			}
1387 		}
1388 
1389 		group->default_domain = dom;
1390 		if (!group->domain)
1391 			group->domain = dom;
1392 
1393 		if (dom && !iommu_dma_strict) {
1394 			int attr = 1;
1395 			iommu_domain_set_attr(dom,
1396 					      DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1397 					      &attr);
1398 		}
1399 	}
1400 
1401 	ret = iommu_group_add_device(group, dev);
1402 	if (ret) {
1403 		iommu_group_put(group);
1404 		return ERR_PTR(ret);
1405 	}
1406 
1407 	return group;
1408 }
1409 
1410 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1411 {
1412 	return group->default_domain;
1413 }
1414 
1415 static int add_iommu_group(struct device *dev, void *data)
1416 {
1417 	int ret = iommu_probe_device(dev);
1418 
1419 	/*
1420 	 * We ignore -ENODEV errors for now, as they just mean that the
1421 	 * device is not translated by an IOMMU. We still care about
1422 	 * other errors and fail to initialize when they happen.
1423 	 */
1424 	if (ret == -ENODEV)
1425 		ret = 0;
1426 
1427 	return ret;
1428 }
1429 
1430 static int remove_iommu_group(struct device *dev, void *data)
1431 {
1432 	iommu_release_device(dev);
1433 
1434 	return 0;
1435 }
1436 
1437 static int iommu_bus_notifier(struct notifier_block *nb,
1438 			      unsigned long action, void *data)
1439 {
1440 	unsigned long group_action = 0;
1441 	struct device *dev = data;
1442 	struct iommu_group *group;
1443 
1444 	/*
1445 	 * ADD/DEL call into iommu driver ops if provided, which may
1446 	 * result in ADD/DEL notifiers to group->notifier
1447 	 */
1448 	if (action == BUS_NOTIFY_ADD_DEVICE) {
1449 		int ret;
1450 
1451 		ret = iommu_probe_device(dev);
1452 		return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1453 	} else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1454 		iommu_release_device(dev);
1455 		return NOTIFY_OK;
1456 	}
1457 
1458 	/*
1459 	 * Remaining BUS_NOTIFYs get filtered and republished to the
1460 	 * group, if anyone is listening
1461 	 */
1462 	group = iommu_group_get(dev);
1463 	if (!group)
1464 		return 0;
1465 
1466 	switch (action) {
1467 	case BUS_NOTIFY_BIND_DRIVER:
1468 		group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1469 		break;
1470 	case BUS_NOTIFY_BOUND_DRIVER:
1471 		group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1472 		break;
1473 	case BUS_NOTIFY_UNBIND_DRIVER:
1474 		group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1475 		break;
1476 	case BUS_NOTIFY_UNBOUND_DRIVER:
1477 		group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1478 		break;
1479 	}
1480 
1481 	if (group_action)
1482 		blocking_notifier_call_chain(&group->notifier,
1483 					     group_action, dev);
1484 
1485 	iommu_group_put(group);
1486 	return 0;
1487 }
1488 
1489 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1490 {
1491 	int err;
1492 	struct notifier_block *nb;
1493 
1494 	nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1495 	if (!nb)
1496 		return -ENOMEM;
1497 
1498 	nb->notifier_call = iommu_bus_notifier;
1499 
1500 	err = bus_register_notifier(bus, nb);
1501 	if (err)
1502 		goto out_free;
1503 
1504 	err = bus_for_each_dev(bus, NULL, NULL, add_iommu_group);
1505 	if (err)
1506 		goto out_err;
1507 
1508 
1509 	return 0;
1510 
1511 out_err:
1512 	/* Clean up */
1513 	bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
1514 	bus_unregister_notifier(bus, nb);
1515 
1516 out_free:
1517 	kfree(nb);
1518 
1519 	return err;
1520 }
1521 
1522 /**
1523  * bus_set_iommu - set iommu-callbacks for the bus
1524  * @bus: bus.
1525  * @ops: the callbacks provided by the iommu-driver
1526  *
1527  * This function is called by an iommu driver to set the iommu methods
1528  * used for a particular bus. Drivers for devices on that bus can use
1529  * the iommu-api after these ops are registered.
1530  * This special function is needed because IOMMUs are usually devices on
1531  * the bus itself, so the iommu drivers are not initialized when the bus
1532  * is set up. With this function the iommu-driver can set the iommu-ops
1533  * afterwards.
1534  */
1535 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1536 {
1537 	int err;
1538 
1539 	if (bus->iommu_ops != NULL)
1540 		return -EBUSY;
1541 
1542 	bus->iommu_ops = ops;
1543 
1544 	/* Do IOMMU specific setup for this bus-type */
1545 	err = iommu_bus_init(bus, ops);
1546 	if (err)
1547 		bus->iommu_ops = NULL;
1548 
1549 	return err;
1550 }
1551 EXPORT_SYMBOL_GPL(bus_set_iommu);
1552 
1553 bool iommu_present(struct bus_type *bus)
1554 {
1555 	return bus->iommu_ops != NULL;
1556 }
1557 EXPORT_SYMBOL_GPL(iommu_present);
1558 
1559 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1560 {
1561 	if (!bus->iommu_ops || !bus->iommu_ops->capable)
1562 		return false;
1563 
1564 	return bus->iommu_ops->capable(cap);
1565 }
1566 EXPORT_SYMBOL_GPL(iommu_capable);
1567 
1568 /**
1569  * iommu_set_fault_handler() - set a fault handler for an iommu domain
1570  * @domain: iommu domain
1571  * @handler: fault handler
1572  * @token: user data, will be passed back to the fault handler
1573  *
1574  * This function should be used by IOMMU users which want to be notified
1575  * whenever an IOMMU fault happens.
1576  *
1577  * The fault handler itself should return 0 on success, and an appropriate
1578  * error code otherwise.
1579  */
1580 void iommu_set_fault_handler(struct iommu_domain *domain,
1581 					iommu_fault_handler_t handler,
1582 					void *token)
1583 {
1584 	BUG_ON(!domain);
1585 
1586 	domain->handler = handler;
1587 	domain->handler_token = token;
1588 }
1589 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1590 
1591 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1592 						 unsigned type)
1593 {
1594 	struct iommu_domain *domain;
1595 
1596 	if (bus == NULL || bus->iommu_ops == NULL)
1597 		return NULL;
1598 
1599 	domain = bus->iommu_ops->domain_alloc(type);
1600 	if (!domain)
1601 		return NULL;
1602 
1603 	domain->ops  = bus->iommu_ops;
1604 	domain->type = type;
1605 	/* Assume all sizes by default; the driver may override this later */
1606 	domain->pgsize_bitmap  = bus->iommu_ops->pgsize_bitmap;
1607 
1608 	return domain;
1609 }
1610 
1611 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1612 {
1613 	return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1614 }
1615 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1616 
1617 void iommu_domain_free(struct iommu_domain *domain)
1618 {
1619 	domain->ops->domain_free(domain);
1620 }
1621 EXPORT_SYMBOL_GPL(iommu_domain_free);
1622 
1623 static int __iommu_attach_device(struct iommu_domain *domain,
1624 				 struct device *dev)
1625 {
1626 	int ret;
1627 	if ((domain->ops->is_attach_deferred != NULL) &&
1628 	    domain->ops->is_attach_deferred(domain, dev))
1629 		return 0;
1630 
1631 	if (unlikely(domain->ops->attach_dev == NULL))
1632 		return -ENODEV;
1633 
1634 	ret = domain->ops->attach_dev(domain, dev);
1635 	if (!ret)
1636 		trace_attach_device_to_domain(dev);
1637 	return ret;
1638 }
1639 
1640 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1641 {
1642 	struct iommu_group *group;
1643 	int ret;
1644 
1645 	group = iommu_group_get(dev);
1646 	if (!group)
1647 		return -ENODEV;
1648 
1649 	/*
1650 	 * Lock the group to make sure the device-count doesn't
1651 	 * change while we are attaching
1652 	 */
1653 	mutex_lock(&group->mutex);
1654 	ret = -EINVAL;
1655 	if (iommu_group_device_count(group) != 1)
1656 		goto out_unlock;
1657 
1658 	ret = __iommu_attach_group(domain, group);
1659 
1660 out_unlock:
1661 	mutex_unlock(&group->mutex);
1662 	iommu_group_put(group);
1663 
1664 	return ret;
1665 }
1666 EXPORT_SYMBOL_GPL(iommu_attach_device);
1667 
1668 int iommu_cache_invalidate(struct iommu_domain *domain, struct device *dev,
1669 			   struct iommu_cache_invalidate_info *inv_info)
1670 {
1671 	if (unlikely(!domain->ops->cache_invalidate))
1672 		return -ENODEV;
1673 
1674 	return domain->ops->cache_invalidate(domain, dev, inv_info);
1675 }
1676 EXPORT_SYMBOL_GPL(iommu_cache_invalidate);
1677 
1678 int iommu_sva_bind_gpasid(struct iommu_domain *domain,
1679 			   struct device *dev, struct iommu_gpasid_bind_data *data)
1680 {
1681 	if (unlikely(!domain->ops->sva_bind_gpasid))
1682 		return -ENODEV;
1683 
1684 	return domain->ops->sva_bind_gpasid(domain, dev, data);
1685 }
1686 EXPORT_SYMBOL_GPL(iommu_sva_bind_gpasid);
1687 
1688 int iommu_sva_unbind_gpasid(struct iommu_domain *domain, struct device *dev,
1689 			     ioasid_t pasid)
1690 {
1691 	if (unlikely(!domain->ops->sva_unbind_gpasid))
1692 		return -ENODEV;
1693 
1694 	return domain->ops->sva_unbind_gpasid(dev, pasid);
1695 }
1696 EXPORT_SYMBOL_GPL(iommu_sva_unbind_gpasid);
1697 
1698 static void __iommu_detach_device(struct iommu_domain *domain,
1699 				  struct device *dev)
1700 {
1701 	if ((domain->ops->is_attach_deferred != NULL) &&
1702 	    domain->ops->is_attach_deferred(domain, dev))
1703 		return;
1704 
1705 	if (unlikely(domain->ops->detach_dev == NULL))
1706 		return;
1707 
1708 	domain->ops->detach_dev(domain, dev);
1709 	trace_detach_device_from_domain(dev);
1710 }
1711 
1712 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1713 {
1714 	struct iommu_group *group;
1715 
1716 	group = iommu_group_get(dev);
1717 	if (!group)
1718 		return;
1719 
1720 	mutex_lock(&group->mutex);
1721 	if (iommu_group_device_count(group) != 1) {
1722 		WARN_ON(1);
1723 		goto out_unlock;
1724 	}
1725 
1726 	__iommu_detach_group(domain, group);
1727 
1728 out_unlock:
1729 	mutex_unlock(&group->mutex);
1730 	iommu_group_put(group);
1731 }
1732 EXPORT_SYMBOL_GPL(iommu_detach_device);
1733 
1734 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1735 {
1736 	struct iommu_domain *domain;
1737 	struct iommu_group *group;
1738 
1739 	group = iommu_group_get(dev);
1740 	if (!group)
1741 		return NULL;
1742 
1743 	domain = group->domain;
1744 
1745 	iommu_group_put(group);
1746 
1747 	return domain;
1748 }
1749 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1750 
1751 /*
1752  * For IOMMU_DOMAIN_DMA implementations which already provide their own
1753  * guarantees that the group and its default domain are valid and correct.
1754  */
1755 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
1756 {
1757 	return dev->iommu_group->default_domain;
1758 }
1759 
1760 /*
1761  * IOMMU groups are really the natural working unit of the IOMMU, but
1762  * the IOMMU API works on domains and devices.  Bridge that gap by
1763  * iterating over the devices in a group.  Ideally we'd have a single
1764  * device which represents the requestor ID of the group, but we also
1765  * allow IOMMU drivers to create policy defined minimum sets, where
1766  * the physical hardware may be able to distiguish members, but we
1767  * wish to group them at a higher level (ex. untrusted multi-function
1768  * PCI devices).  Thus we attach each device.
1769  */
1770 static int iommu_group_do_attach_device(struct device *dev, void *data)
1771 {
1772 	struct iommu_domain *domain = data;
1773 
1774 	return __iommu_attach_device(domain, dev);
1775 }
1776 
1777 static int __iommu_attach_group(struct iommu_domain *domain,
1778 				struct iommu_group *group)
1779 {
1780 	int ret;
1781 
1782 	if (group->default_domain && group->domain != group->default_domain)
1783 		return -EBUSY;
1784 
1785 	ret = __iommu_group_for_each_dev(group, domain,
1786 					 iommu_group_do_attach_device);
1787 	if (ret == 0)
1788 		group->domain = domain;
1789 
1790 	return ret;
1791 }
1792 
1793 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1794 {
1795 	int ret;
1796 
1797 	mutex_lock(&group->mutex);
1798 	ret = __iommu_attach_group(domain, group);
1799 	mutex_unlock(&group->mutex);
1800 
1801 	return ret;
1802 }
1803 EXPORT_SYMBOL_GPL(iommu_attach_group);
1804 
1805 static int iommu_group_do_detach_device(struct device *dev, void *data)
1806 {
1807 	struct iommu_domain *domain = data;
1808 
1809 	__iommu_detach_device(domain, dev);
1810 
1811 	return 0;
1812 }
1813 
1814 static void __iommu_detach_group(struct iommu_domain *domain,
1815 				 struct iommu_group *group)
1816 {
1817 	int ret;
1818 
1819 	if (!group->default_domain) {
1820 		__iommu_group_for_each_dev(group, domain,
1821 					   iommu_group_do_detach_device);
1822 		group->domain = NULL;
1823 		return;
1824 	}
1825 
1826 	if (group->domain == group->default_domain)
1827 		return;
1828 
1829 	/* Detach by re-attaching to the default domain */
1830 	ret = __iommu_group_for_each_dev(group, group->default_domain,
1831 					 iommu_group_do_attach_device);
1832 	if (ret != 0)
1833 		WARN_ON(1);
1834 	else
1835 		group->domain = group->default_domain;
1836 }
1837 
1838 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1839 {
1840 	mutex_lock(&group->mutex);
1841 	__iommu_detach_group(domain, group);
1842 	mutex_unlock(&group->mutex);
1843 }
1844 EXPORT_SYMBOL_GPL(iommu_detach_group);
1845 
1846 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1847 {
1848 	if (unlikely(domain->ops->iova_to_phys == NULL))
1849 		return 0;
1850 
1851 	return domain->ops->iova_to_phys(domain, iova);
1852 }
1853 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1854 
1855 static size_t iommu_pgsize(struct iommu_domain *domain,
1856 			   unsigned long addr_merge, size_t size)
1857 {
1858 	unsigned int pgsize_idx;
1859 	size_t pgsize;
1860 
1861 	/* Max page size that still fits into 'size' */
1862 	pgsize_idx = __fls(size);
1863 
1864 	/* need to consider alignment requirements ? */
1865 	if (likely(addr_merge)) {
1866 		/* Max page size allowed by address */
1867 		unsigned int align_pgsize_idx = __ffs(addr_merge);
1868 		pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1869 	}
1870 
1871 	/* build a mask of acceptable page sizes */
1872 	pgsize = (1UL << (pgsize_idx + 1)) - 1;
1873 
1874 	/* throw away page sizes not supported by the hardware */
1875 	pgsize &= domain->pgsize_bitmap;
1876 
1877 	/* make sure we're still sane */
1878 	BUG_ON(!pgsize);
1879 
1880 	/* pick the biggest page */
1881 	pgsize_idx = __fls(pgsize);
1882 	pgsize = 1UL << pgsize_idx;
1883 
1884 	return pgsize;
1885 }
1886 
1887 int __iommu_map(struct iommu_domain *domain, unsigned long iova,
1888 	      phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
1889 {
1890 	const struct iommu_ops *ops = domain->ops;
1891 	unsigned long orig_iova = iova;
1892 	unsigned int min_pagesz;
1893 	size_t orig_size = size;
1894 	phys_addr_t orig_paddr = paddr;
1895 	int ret = 0;
1896 
1897 	if (unlikely(ops->map == NULL ||
1898 		     domain->pgsize_bitmap == 0UL))
1899 		return -ENODEV;
1900 
1901 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1902 		return -EINVAL;
1903 
1904 	/* find out the minimum page size supported */
1905 	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1906 
1907 	/*
1908 	 * both the virtual address and the physical one, as well as
1909 	 * the size of the mapping, must be aligned (at least) to the
1910 	 * size of the smallest page supported by the hardware
1911 	 */
1912 	if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1913 		pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1914 		       iova, &paddr, size, min_pagesz);
1915 		return -EINVAL;
1916 	}
1917 
1918 	pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1919 
1920 	while (size) {
1921 		size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1922 
1923 		pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1924 			 iova, &paddr, pgsize);
1925 		ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
1926 
1927 		if (ret)
1928 			break;
1929 
1930 		iova += pgsize;
1931 		paddr += pgsize;
1932 		size -= pgsize;
1933 	}
1934 
1935 	if (ops->iotlb_sync_map)
1936 		ops->iotlb_sync_map(domain);
1937 
1938 	/* unroll mapping in case something went wrong */
1939 	if (ret)
1940 		iommu_unmap(domain, orig_iova, orig_size - size);
1941 	else
1942 		trace_map(orig_iova, orig_paddr, orig_size);
1943 
1944 	return ret;
1945 }
1946 
1947 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1948 	      phys_addr_t paddr, size_t size, int prot)
1949 {
1950 	might_sleep();
1951 	return __iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
1952 }
1953 EXPORT_SYMBOL_GPL(iommu_map);
1954 
1955 int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
1956 	      phys_addr_t paddr, size_t size, int prot)
1957 {
1958 	return __iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
1959 }
1960 EXPORT_SYMBOL_GPL(iommu_map_atomic);
1961 
1962 static size_t __iommu_unmap(struct iommu_domain *domain,
1963 			    unsigned long iova, size_t size,
1964 			    struct iommu_iotlb_gather *iotlb_gather)
1965 {
1966 	const struct iommu_ops *ops = domain->ops;
1967 	size_t unmapped_page, unmapped = 0;
1968 	unsigned long orig_iova = iova;
1969 	unsigned int min_pagesz;
1970 
1971 	if (unlikely(ops->unmap == NULL ||
1972 		     domain->pgsize_bitmap == 0UL))
1973 		return 0;
1974 
1975 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1976 		return 0;
1977 
1978 	/* find out the minimum page size supported */
1979 	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1980 
1981 	/*
1982 	 * The virtual address, as well as the size of the mapping, must be
1983 	 * aligned (at least) to the size of the smallest page supported
1984 	 * by the hardware
1985 	 */
1986 	if (!IS_ALIGNED(iova | size, min_pagesz)) {
1987 		pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1988 		       iova, size, min_pagesz);
1989 		return 0;
1990 	}
1991 
1992 	pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1993 
1994 	/*
1995 	 * Keep iterating until we either unmap 'size' bytes (or more)
1996 	 * or we hit an area that isn't mapped.
1997 	 */
1998 	while (unmapped < size) {
1999 		size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
2000 
2001 		unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
2002 		if (!unmapped_page)
2003 			break;
2004 
2005 		pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
2006 			 iova, unmapped_page);
2007 
2008 		iova += unmapped_page;
2009 		unmapped += unmapped_page;
2010 	}
2011 
2012 	trace_unmap(orig_iova, size, unmapped);
2013 	return unmapped;
2014 }
2015 
2016 size_t iommu_unmap(struct iommu_domain *domain,
2017 		   unsigned long iova, size_t size)
2018 {
2019 	struct iommu_iotlb_gather iotlb_gather;
2020 	size_t ret;
2021 
2022 	iommu_iotlb_gather_init(&iotlb_gather);
2023 	ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
2024 	iommu_tlb_sync(domain, &iotlb_gather);
2025 
2026 	return ret;
2027 }
2028 EXPORT_SYMBOL_GPL(iommu_unmap);
2029 
2030 size_t iommu_unmap_fast(struct iommu_domain *domain,
2031 			unsigned long iova, size_t size,
2032 			struct iommu_iotlb_gather *iotlb_gather)
2033 {
2034 	return __iommu_unmap(domain, iova, size, iotlb_gather);
2035 }
2036 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2037 
2038 size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2039 		    struct scatterlist *sg, unsigned int nents, int prot,
2040 		    gfp_t gfp)
2041 {
2042 	size_t len = 0, mapped = 0;
2043 	phys_addr_t start;
2044 	unsigned int i = 0;
2045 	int ret;
2046 
2047 	while (i <= nents) {
2048 		phys_addr_t s_phys = sg_phys(sg);
2049 
2050 		if (len && s_phys != start + len) {
2051 			ret = __iommu_map(domain, iova + mapped, start,
2052 					len, prot, gfp);
2053 
2054 			if (ret)
2055 				goto out_err;
2056 
2057 			mapped += len;
2058 			len = 0;
2059 		}
2060 
2061 		if (len) {
2062 			len += sg->length;
2063 		} else {
2064 			len = sg->length;
2065 			start = s_phys;
2066 		}
2067 
2068 		if (++i < nents)
2069 			sg = sg_next(sg);
2070 	}
2071 
2072 	return mapped;
2073 
2074 out_err:
2075 	/* undo mappings already done */
2076 	iommu_unmap(domain, iova, mapped);
2077 
2078 	return 0;
2079 
2080 }
2081 
2082 size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2083 		    struct scatterlist *sg, unsigned int nents, int prot)
2084 {
2085 	might_sleep();
2086 	return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2087 }
2088 EXPORT_SYMBOL_GPL(iommu_map_sg);
2089 
2090 size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2091 		    struct scatterlist *sg, unsigned int nents, int prot)
2092 {
2093 	return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2094 }
2095 EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2096 
2097 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
2098 			       phys_addr_t paddr, u64 size, int prot)
2099 {
2100 	if (unlikely(domain->ops->domain_window_enable == NULL))
2101 		return -ENODEV;
2102 
2103 	return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2104 						 prot);
2105 }
2106 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2107 
2108 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2109 {
2110 	if (unlikely(domain->ops->domain_window_disable == NULL))
2111 		return;
2112 
2113 	return domain->ops->domain_window_disable(domain, wnd_nr);
2114 }
2115 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2116 
2117 /**
2118  * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2119  * @domain: the iommu domain where the fault has happened
2120  * @dev: the device where the fault has happened
2121  * @iova: the faulting address
2122  * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2123  *
2124  * This function should be called by the low-level IOMMU implementations
2125  * whenever IOMMU faults happen, to allow high-level users, that are
2126  * interested in such events, to know about them.
2127  *
2128  * This event may be useful for several possible use cases:
2129  * - mere logging of the event
2130  * - dynamic TLB/PTE loading
2131  * - if restarting of the faulting device is required
2132  *
2133  * Returns 0 on success and an appropriate error code otherwise (if dynamic
2134  * PTE/TLB loading will one day be supported, implementations will be able
2135  * to tell whether it succeeded or not according to this return value).
2136  *
2137  * Specifically, -ENOSYS is returned if a fault handler isn't installed
2138  * (though fault handlers can also return -ENOSYS, in case they want to
2139  * elicit the default behavior of the IOMMU drivers).
2140  */
2141 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2142 		       unsigned long iova, int flags)
2143 {
2144 	int ret = -ENOSYS;
2145 
2146 	/*
2147 	 * if upper layers showed interest and installed a fault handler,
2148 	 * invoke it.
2149 	 */
2150 	if (domain->handler)
2151 		ret = domain->handler(domain, dev, iova, flags,
2152 						domain->handler_token);
2153 
2154 	trace_io_page_fault(dev, iova, flags);
2155 	return ret;
2156 }
2157 EXPORT_SYMBOL_GPL(report_iommu_fault);
2158 
2159 static int __init iommu_init(void)
2160 {
2161 	iommu_group_kset = kset_create_and_add("iommu_groups",
2162 					       NULL, kernel_kobj);
2163 	BUG_ON(!iommu_group_kset);
2164 
2165 	iommu_debugfs_setup();
2166 
2167 	return 0;
2168 }
2169 core_initcall(iommu_init);
2170 
2171 int iommu_domain_get_attr(struct iommu_domain *domain,
2172 			  enum iommu_attr attr, void *data)
2173 {
2174 	struct iommu_domain_geometry *geometry;
2175 	bool *paging;
2176 	int ret = 0;
2177 
2178 	switch (attr) {
2179 	case DOMAIN_ATTR_GEOMETRY:
2180 		geometry  = data;
2181 		*geometry = domain->geometry;
2182 
2183 		break;
2184 	case DOMAIN_ATTR_PAGING:
2185 		paging  = data;
2186 		*paging = (domain->pgsize_bitmap != 0UL);
2187 		break;
2188 	default:
2189 		if (!domain->ops->domain_get_attr)
2190 			return -EINVAL;
2191 
2192 		ret = domain->ops->domain_get_attr(domain, attr, data);
2193 	}
2194 
2195 	return ret;
2196 }
2197 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2198 
2199 int iommu_domain_set_attr(struct iommu_domain *domain,
2200 			  enum iommu_attr attr, void *data)
2201 {
2202 	int ret = 0;
2203 
2204 	switch (attr) {
2205 	default:
2206 		if (domain->ops->domain_set_attr == NULL)
2207 			return -EINVAL;
2208 
2209 		ret = domain->ops->domain_set_attr(domain, attr, data);
2210 	}
2211 
2212 	return ret;
2213 }
2214 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
2215 
2216 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
2217 {
2218 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2219 
2220 	if (ops && ops->get_resv_regions)
2221 		ops->get_resv_regions(dev, list);
2222 }
2223 
2224 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
2225 {
2226 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2227 
2228 	if (ops && ops->put_resv_regions)
2229 		ops->put_resv_regions(dev, list);
2230 }
2231 
2232 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
2233 						  size_t length, int prot,
2234 						  enum iommu_resv_type type)
2235 {
2236 	struct iommu_resv_region *region;
2237 
2238 	region = kzalloc(sizeof(*region), GFP_KERNEL);
2239 	if (!region)
2240 		return NULL;
2241 
2242 	INIT_LIST_HEAD(&region->list);
2243 	region->start = start;
2244 	region->length = length;
2245 	region->prot = prot;
2246 	region->type = type;
2247 	return region;
2248 }
2249 
2250 static int
2251 request_default_domain_for_dev(struct device *dev, unsigned long type)
2252 {
2253 	struct iommu_domain *domain;
2254 	struct iommu_group *group;
2255 	int ret;
2256 
2257 	/* Device must already be in a group before calling this function */
2258 	group = iommu_group_get(dev);
2259 	if (!group)
2260 		return -EINVAL;
2261 
2262 	mutex_lock(&group->mutex);
2263 
2264 	ret = 0;
2265 	if (group->default_domain && group->default_domain->type == type)
2266 		goto out;
2267 
2268 	/* Don't change mappings of existing devices */
2269 	ret = -EBUSY;
2270 	if (iommu_group_device_count(group) != 1)
2271 		goto out;
2272 
2273 	ret = -ENOMEM;
2274 	domain = __iommu_domain_alloc(dev->bus, type);
2275 	if (!domain)
2276 		goto out;
2277 
2278 	/* Attach the device to the domain */
2279 	ret = __iommu_attach_group(domain, group);
2280 	if (ret) {
2281 		iommu_domain_free(domain);
2282 		goto out;
2283 	}
2284 
2285 	iommu_group_create_direct_mappings(group, dev);
2286 
2287 	/* Make the domain the default for this group */
2288 	if (group->default_domain)
2289 		iommu_domain_free(group->default_domain);
2290 	group->default_domain = domain;
2291 
2292 	dev_info(dev, "Using iommu %s mapping\n",
2293 		 type == IOMMU_DOMAIN_DMA ? "dma" : "direct");
2294 
2295 	ret = 0;
2296 out:
2297 	mutex_unlock(&group->mutex);
2298 	iommu_group_put(group);
2299 
2300 	return ret;
2301 }
2302 
2303 /* Request that a device is direct mapped by the IOMMU */
2304 int iommu_request_dm_for_dev(struct device *dev)
2305 {
2306 	return request_default_domain_for_dev(dev, IOMMU_DOMAIN_IDENTITY);
2307 }
2308 
2309 /* Request that a device can't be direct mapped by the IOMMU */
2310 int iommu_request_dma_domain_for_dev(struct device *dev)
2311 {
2312 	return request_default_domain_for_dev(dev, IOMMU_DOMAIN_DMA);
2313 }
2314 
2315 void iommu_set_default_passthrough(bool cmd_line)
2316 {
2317 	if (cmd_line)
2318 		iommu_set_cmd_line_dma_api();
2319 
2320 	iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2321 }
2322 
2323 void iommu_set_default_translated(bool cmd_line)
2324 {
2325 	if (cmd_line)
2326 		iommu_set_cmd_line_dma_api();
2327 
2328 	iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2329 }
2330 
2331 bool iommu_default_passthrough(void)
2332 {
2333 	return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2334 }
2335 EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2336 
2337 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
2338 {
2339 	const struct iommu_ops *ops = NULL;
2340 	struct iommu_device *iommu;
2341 
2342 	spin_lock(&iommu_device_lock);
2343 	list_for_each_entry(iommu, &iommu_device_list, list)
2344 		if (iommu->fwnode == fwnode) {
2345 			ops = iommu->ops;
2346 			break;
2347 		}
2348 	spin_unlock(&iommu_device_lock);
2349 	return ops;
2350 }
2351 
2352 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2353 		      const struct iommu_ops *ops)
2354 {
2355 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2356 
2357 	if (fwspec)
2358 		return ops == fwspec->ops ? 0 : -EINVAL;
2359 
2360 	fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
2361 	if (!fwspec)
2362 		return -ENOMEM;
2363 
2364 	of_node_get(to_of_node(iommu_fwnode));
2365 	fwspec->iommu_fwnode = iommu_fwnode;
2366 	fwspec->ops = ops;
2367 	dev_iommu_fwspec_set(dev, fwspec);
2368 	return 0;
2369 }
2370 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2371 
2372 void iommu_fwspec_free(struct device *dev)
2373 {
2374 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2375 
2376 	if (fwspec) {
2377 		fwnode_handle_put(fwspec->iommu_fwnode);
2378 		kfree(fwspec);
2379 		dev_iommu_fwspec_set(dev, NULL);
2380 	}
2381 }
2382 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2383 
2384 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2385 {
2386 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2387 	size_t size;
2388 	int i;
2389 
2390 	if (!fwspec)
2391 		return -EINVAL;
2392 
2393 	size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
2394 	if (size > sizeof(*fwspec)) {
2395 		fwspec = krealloc(fwspec, size, GFP_KERNEL);
2396 		if (!fwspec)
2397 			return -ENOMEM;
2398 
2399 		dev_iommu_fwspec_set(dev, fwspec);
2400 	}
2401 
2402 	for (i = 0; i < num_ids; i++)
2403 		fwspec->ids[fwspec->num_ids + i] = ids[i];
2404 
2405 	fwspec->num_ids += num_ids;
2406 	return 0;
2407 }
2408 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
2409 
2410 /*
2411  * Per device IOMMU features.
2412  */
2413 bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2414 {
2415 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2416 
2417 	if (ops && ops->dev_has_feat)
2418 		return ops->dev_has_feat(dev, feat);
2419 
2420 	return false;
2421 }
2422 EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2423 
2424 int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2425 {
2426 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2427 
2428 	if (ops && ops->dev_enable_feat)
2429 		return ops->dev_enable_feat(dev, feat);
2430 
2431 	return -ENODEV;
2432 }
2433 EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2434 
2435 /*
2436  * The device drivers should do the necessary cleanups before calling this.
2437  * For example, before disabling the aux-domain feature, the device driver
2438  * should detach all aux-domains. Otherwise, this will return -EBUSY.
2439  */
2440 int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2441 {
2442 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2443 
2444 	if (ops && ops->dev_disable_feat)
2445 		return ops->dev_disable_feat(dev, feat);
2446 
2447 	return -EBUSY;
2448 }
2449 EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2450 
2451 bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2452 {
2453 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2454 
2455 	if (ops && ops->dev_feat_enabled)
2456 		return ops->dev_feat_enabled(dev, feat);
2457 
2458 	return false;
2459 }
2460 EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2461 
2462 /*
2463  * Aux-domain specific attach/detach.
2464  *
2465  * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2466  * true. Also, as long as domains are attached to a device through this
2467  * interface, any tries to call iommu_attach_device() should fail
2468  * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2469  * This should make us safe against a device being attached to a guest as a
2470  * whole while there are still pasid users on it (aux and sva).
2471  */
2472 int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2473 {
2474 	int ret = -ENODEV;
2475 
2476 	if (domain->ops->aux_attach_dev)
2477 		ret = domain->ops->aux_attach_dev(domain, dev);
2478 
2479 	if (!ret)
2480 		trace_attach_device_to_domain(dev);
2481 
2482 	return ret;
2483 }
2484 EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2485 
2486 void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2487 {
2488 	if (domain->ops->aux_detach_dev) {
2489 		domain->ops->aux_detach_dev(domain, dev);
2490 		trace_detach_device_from_domain(dev);
2491 	}
2492 }
2493 EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2494 
2495 int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2496 {
2497 	int ret = -ENODEV;
2498 
2499 	if (domain->ops->aux_get_pasid)
2500 		ret = domain->ops->aux_get_pasid(domain, dev);
2501 
2502 	return ret;
2503 }
2504 EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
2505 
2506 /**
2507  * iommu_sva_bind_device() - Bind a process address space to a device
2508  * @dev: the device
2509  * @mm: the mm to bind, caller must hold a reference to it
2510  *
2511  * Create a bond between device and address space, allowing the device to access
2512  * the mm using the returned PASID. If a bond already exists between @device and
2513  * @mm, it is returned and an additional reference is taken. Caller must call
2514  * iommu_sva_unbind_device() to release each reference.
2515  *
2516  * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2517  * initialize the required SVA features.
2518  *
2519  * On error, returns an ERR_PTR value.
2520  */
2521 struct iommu_sva *
2522 iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2523 {
2524 	struct iommu_group *group;
2525 	struct iommu_sva *handle = ERR_PTR(-EINVAL);
2526 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2527 
2528 	if (!ops || !ops->sva_bind)
2529 		return ERR_PTR(-ENODEV);
2530 
2531 	group = iommu_group_get(dev);
2532 	if (!group)
2533 		return ERR_PTR(-ENODEV);
2534 
2535 	/* Ensure device count and domain don't change while we're binding */
2536 	mutex_lock(&group->mutex);
2537 
2538 	/*
2539 	 * To keep things simple, SVA currently doesn't support IOMMU groups
2540 	 * with more than one device. Existing SVA-capable systems are not
2541 	 * affected by the problems that required IOMMU groups (lack of ACS
2542 	 * isolation, device ID aliasing and other hardware issues).
2543 	 */
2544 	if (iommu_group_device_count(group) != 1)
2545 		goto out_unlock;
2546 
2547 	handle = ops->sva_bind(dev, mm, drvdata);
2548 
2549 out_unlock:
2550 	mutex_unlock(&group->mutex);
2551 	iommu_group_put(group);
2552 
2553 	return handle;
2554 }
2555 EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
2556 
2557 /**
2558  * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2559  * @handle: the handle returned by iommu_sva_bind_device()
2560  *
2561  * Put reference to a bond between device and address space. The device should
2562  * not be issuing any more transaction for this PASID. All outstanding page
2563  * requests for this PASID must have been flushed to the IOMMU.
2564  *
2565  * Returns 0 on success, or an error value
2566  */
2567 void iommu_sva_unbind_device(struct iommu_sva *handle)
2568 {
2569 	struct iommu_group *group;
2570 	struct device *dev = handle->dev;
2571 	const struct iommu_ops *ops = dev->bus->iommu_ops;
2572 
2573 	if (!ops || !ops->sva_unbind)
2574 		return;
2575 
2576 	group = iommu_group_get(dev);
2577 	if (!group)
2578 		return;
2579 
2580 	mutex_lock(&group->mutex);
2581 	ops->sva_unbind(handle);
2582 	mutex_unlock(&group->mutex);
2583 
2584 	iommu_group_put(group);
2585 }
2586 EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
2587 
2588 int iommu_sva_set_ops(struct iommu_sva *handle,
2589 		      const struct iommu_sva_ops *sva_ops)
2590 {
2591 	if (handle->ops && handle->ops != sva_ops)
2592 		return -EEXIST;
2593 
2594 	handle->ops = sva_ops;
2595 	return 0;
2596 }
2597 EXPORT_SYMBOL_GPL(iommu_sva_set_ops);
2598 
2599 int iommu_sva_get_pasid(struct iommu_sva *handle)
2600 {
2601 	const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
2602 
2603 	if (!ops || !ops->sva_get_pasid)
2604 		return IOMMU_PASID_INVALID;
2605 
2606 	return ops->sva_get_pasid(handle);
2607 }
2608 EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);
2609