]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - drivers/iommu/iommu.c
Merge tag 'drm-next-2019-03-15' of git://anongit.freedesktop.org/drm/drm
[thirdparty/kernel/stable.git] / drivers / iommu / iommu.c
1 /*
2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published
7 * by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
18
19 #define pr_fmt(fmt) "iommu: " fmt
20
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/init.h>
26 #include <linux/export.h>
27 #include <linux/slab.h>
28 #include <linux/errno.h>
29 #include <linux/iommu.h>
30 #include <linux/idr.h>
31 #include <linux/notifier.h>
32 #include <linux/err.h>
33 #include <linux/pci.h>
34 #include <linux/bitops.h>
35 #include <linux/property.h>
36 #include <linux/fsl/mc.h>
37 #include <trace/events/iommu.h>
38
39 static struct kset *iommu_group_kset;
40 static DEFINE_IDA(iommu_group_ida);
41 #ifdef CONFIG_IOMMU_DEFAULT_PASSTHROUGH
42 static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
43 #else
44 static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
45 #endif
46 static bool iommu_dma_strict __read_mostly = true;
47
48 struct iommu_callback_data {
49 const struct iommu_ops *ops;
50 };
51
52 struct iommu_group {
53 struct kobject kobj;
54 struct kobject *devices_kobj;
55 struct list_head devices;
56 struct mutex mutex;
57 struct blocking_notifier_head notifier;
58 void *iommu_data;
59 void (*iommu_data_release)(void *iommu_data);
60 char *name;
61 int id;
62 struct iommu_domain *default_domain;
63 struct iommu_domain *domain;
64 };
65
66 struct group_device {
67 struct list_head list;
68 struct device *dev;
69 char *name;
70 };
71
72 struct iommu_group_attribute {
73 struct attribute attr;
74 ssize_t (*show)(struct iommu_group *group, char *buf);
75 ssize_t (*store)(struct iommu_group *group,
76 const char *buf, size_t count);
77 };
78
79 static const char * const iommu_group_resv_type_string[] = {
80 [IOMMU_RESV_DIRECT] = "direct",
81 [IOMMU_RESV_RESERVED] = "reserved",
82 [IOMMU_RESV_MSI] = "msi",
83 [IOMMU_RESV_SW_MSI] = "msi",
84 };
85
86 #define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
87 struct iommu_group_attribute iommu_group_attr_##_name = \
88 __ATTR(_name, _mode, _show, _store)
89
90 #define to_iommu_group_attr(_attr) \
91 container_of(_attr, struct iommu_group_attribute, attr)
92 #define to_iommu_group(_kobj) \
93 container_of(_kobj, struct iommu_group, kobj)
94
95 static LIST_HEAD(iommu_device_list);
96 static DEFINE_SPINLOCK(iommu_device_lock);
97
98 int iommu_device_register(struct iommu_device *iommu)
99 {
100 spin_lock(&iommu_device_lock);
101 list_add_tail(&iommu->list, &iommu_device_list);
102 spin_unlock(&iommu_device_lock);
103
104 return 0;
105 }
106
107 void iommu_device_unregister(struct iommu_device *iommu)
108 {
109 spin_lock(&iommu_device_lock);
110 list_del(&iommu->list);
111 spin_unlock(&iommu_device_lock);
112 }
113
114 int iommu_probe_device(struct device *dev)
115 {
116 const struct iommu_ops *ops = dev->bus->iommu_ops;
117 int ret = -EINVAL;
118
119 WARN_ON(dev->iommu_group);
120
121 if (ops)
122 ret = ops->add_device(dev);
123
124 return ret;
125 }
126
127 void iommu_release_device(struct device *dev)
128 {
129 const struct iommu_ops *ops = dev->bus->iommu_ops;
130
131 if (dev->iommu_group)
132 ops->remove_device(dev);
133 }
134
135 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
136 unsigned type);
137 static int __iommu_attach_device(struct iommu_domain *domain,
138 struct device *dev);
139 static int __iommu_attach_group(struct iommu_domain *domain,
140 struct iommu_group *group);
141 static void __iommu_detach_group(struct iommu_domain *domain,
142 struct iommu_group *group);
143
144 static int __init iommu_set_def_domain_type(char *str)
145 {
146 bool pt;
147 int ret;
148
149 ret = kstrtobool(str, &pt);
150 if (ret)
151 return ret;
152
153 iommu_def_domain_type = pt ? IOMMU_DOMAIN_IDENTITY : IOMMU_DOMAIN_DMA;
154 return 0;
155 }
156 early_param("iommu.passthrough", iommu_set_def_domain_type);
157
158 static int __init iommu_dma_setup(char *str)
159 {
160 return kstrtobool(str, &iommu_dma_strict);
161 }
162 early_param("iommu.strict", iommu_dma_setup);
163
164 static ssize_t iommu_group_attr_show(struct kobject *kobj,
165 struct attribute *__attr, char *buf)
166 {
167 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
168 struct iommu_group *group = to_iommu_group(kobj);
169 ssize_t ret = -EIO;
170
171 if (attr->show)
172 ret = attr->show(group, buf);
173 return ret;
174 }
175
176 static ssize_t iommu_group_attr_store(struct kobject *kobj,
177 struct attribute *__attr,
178 const char *buf, size_t count)
179 {
180 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
181 struct iommu_group *group = to_iommu_group(kobj);
182 ssize_t ret = -EIO;
183
184 if (attr->store)
185 ret = attr->store(group, buf, count);
186 return ret;
187 }
188
189 static const struct sysfs_ops iommu_group_sysfs_ops = {
190 .show = iommu_group_attr_show,
191 .store = iommu_group_attr_store,
192 };
193
194 static int iommu_group_create_file(struct iommu_group *group,
195 struct iommu_group_attribute *attr)
196 {
197 return sysfs_create_file(&group->kobj, &attr->attr);
198 }
199
200 static void iommu_group_remove_file(struct iommu_group *group,
201 struct iommu_group_attribute *attr)
202 {
203 sysfs_remove_file(&group->kobj, &attr->attr);
204 }
205
206 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
207 {
208 return sprintf(buf, "%s\n", group->name);
209 }
210
211 /**
212 * iommu_insert_resv_region - Insert a new region in the
213 * list of reserved regions.
214 * @new: new region to insert
215 * @regions: list of regions
216 *
217 * The new element is sorted by address with respect to the other
218 * regions of the same type. In case it overlaps with another
219 * region of the same type, regions are merged. In case it
220 * overlaps with another region of different type, regions are
221 * not merged.
222 */
223 static int iommu_insert_resv_region(struct iommu_resv_region *new,
224 struct list_head *regions)
225 {
226 struct iommu_resv_region *region;
227 phys_addr_t start = new->start;
228 phys_addr_t end = new->start + new->length - 1;
229 struct list_head *pos = regions->next;
230
231 while (pos != regions) {
232 struct iommu_resv_region *entry =
233 list_entry(pos, struct iommu_resv_region, list);
234 phys_addr_t a = entry->start;
235 phys_addr_t b = entry->start + entry->length - 1;
236 int type = entry->type;
237
238 if (end < a) {
239 goto insert;
240 } else if (start > b) {
241 pos = pos->next;
242 } else if ((start >= a) && (end <= b)) {
243 if (new->type == type)
244 goto done;
245 else
246 pos = pos->next;
247 } else {
248 if (new->type == type) {
249 phys_addr_t new_start = min(a, start);
250 phys_addr_t new_end = max(b, end);
251
252 list_del(&entry->list);
253 entry->start = new_start;
254 entry->length = new_end - new_start + 1;
255 iommu_insert_resv_region(entry, regions);
256 } else {
257 pos = pos->next;
258 }
259 }
260 }
261 insert:
262 region = iommu_alloc_resv_region(new->start, new->length,
263 new->prot, new->type);
264 if (!region)
265 return -ENOMEM;
266
267 list_add_tail(&region->list, pos);
268 done:
269 return 0;
270 }
271
272 static int
273 iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
274 struct list_head *group_resv_regions)
275 {
276 struct iommu_resv_region *entry;
277 int ret = 0;
278
279 list_for_each_entry(entry, dev_resv_regions, list) {
280 ret = iommu_insert_resv_region(entry, group_resv_regions);
281 if (ret)
282 break;
283 }
284 return ret;
285 }
286
287 int iommu_get_group_resv_regions(struct iommu_group *group,
288 struct list_head *head)
289 {
290 struct group_device *device;
291 int ret = 0;
292
293 mutex_lock(&group->mutex);
294 list_for_each_entry(device, &group->devices, list) {
295 struct list_head dev_resv_regions;
296
297 INIT_LIST_HEAD(&dev_resv_regions);
298 iommu_get_resv_regions(device->dev, &dev_resv_regions);
299 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
300 iommu_put_resv_regions(device->dev, &dev_resv_regions);
301 if (ret)
302 break;
303 }
304 mutex_unlock(&group->mutex);
305 return ret;
306 }
307 EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
308
309 static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
310 char *buf)
311 {
312 struct iommu_resv_region *region, *next;
313 struct list_head group_resv_regions;
314 char *str = buf;
315
316 INIT_LIST_HEAD(&group_resv_regions);
317 iommu_get_group_resv_regions(group, &group_resv_regions);
318
319 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
320 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
321 (long long int)region->start,
322 (long long int)(region->start +
323 region->length - 1),
324 iommu_group_resv_type_string[region->type]);
325 kfree(region);
326 }
327
328 return (str - buf);
329 }
330
331 static ssize_t iommu_group_show_type(struct iommu_group *group,
332 char *buf)
333 {
334 char *type = "unknown\n";
335
336 if (group->default_domain) {
337 switch (group->default_domain->type) {
338 case IOMMU_DOMAIN_BLOCKED:
339 type = "blocked\n";
340 break;
341 case IOMMU_DOMAIN_IDENTITY:
342 type = "identity\n";
343 break;
344 case IOMMU_DOMAIN_UNMANAGED:
345 type = "unmanaged\n";
346 break;
347 case IOMMU_DOMAIN_DMA:
348 type = "DMA";
349 break;
350 }
351 }
352 strcpy(buf, type);
353
354 return strlen(type);
355 }
356
357 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
358
359 static IOMMU_GROUP_ATTR(reserved_regions, 0444,
360 iommu_group_show_resv_regions, NULL);
361
362 static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
363
364 static void iommu_group_release(struct kobject *kobj)
365 {
366 struct iommu_group *group = to_iommu_group(kobj);
367
368 pr_debug("Releasing group %d\n", group->id);
369
370 if (group->iommu_data_release)
371 group->iommu_data_release(group->iommu_data);
372
373 ida_simple_remove(&iommu_group_ida, group->id);
374
375 if (group->default_domain)
376 iommu_domain_free(group->default_domain);
377
378 kfree(group->name);
379 kfree(group);
380 }
381
382 static struct kobj_type iommu_group_ktype = {
383 .sysfs_ops = &iommu_group_sysfs_ops,
384 .release = iommu_group_release,
385 };
386
387 /**
388 * iommu_group_alloc - Allocate a new group
389 *
390 * This function is called by an iommu driver to allocate a new iommu
391 * group. The iommu group represents the minimum granularity of the iommu.
392 * Upon successful return, the caller holds a reference to the supplied
393 * group in order to hold the group until devices are added. Use
394 * iommu_group_put() to release this extra reference count, allowing the
395 * group to be automatically reclaimed once it has no devices or external
396 * references.
397 */
398 struct iommu_group *iommu_group_alloc(void)
399 {
400 struct iommu_group *group;
401 int ret;
402
403 group = kzalloc(sizeof(*group), GFP_KERNEL);
404 if (!group)
405 return ERR_PTR(-ENOMEM);
406
407 group->kobj.kset = iommu_group_kset;
408 mutex_init(&group->mutex);
409 INIT_LIST_HEAD(&group->devices);
410 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
411
412 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
413 if (ret < 0) {
414 kfree(group);
415 return ERR_PTR(ret);
416 }
417 group->id = ret;
418
419 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
420 NULL, "%d", group->id);
421 if (ret) {
422 ida_simple_remove(&iommu_group_ida, group->id);
423 kfree(group);
424 return ERR_PTR(ret);
425 }
426
427 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
428 if (!group->devices_kobj) {
429 kobject_put(&group->kobj); /* triggers .release & free */
430 return ERR_PTR(-ENOMEM);
431 }
432
433 /*
434 * The devices_kobj holds a reference on the group kobject, so
435 * as long as that exists so will the group. We can therefore
436 * use the devices_kobj for reference counting.
437 */
438 kobject_put(&group->kobj);
439
440 ret = iommu_group_create_file(group,
441 &iommu_group_attr_reserved_regions);
442 if (ret)
443 return ERR_PTR(ret);
444
445 ret = iommu_group_create_file(group, &iommu_group_attr_type);
446 if (ret)
447 return ERR_PTR(ret);
448
449 pr_debug("Allocated group %d\n", group->id);
450
451 return group;
452 }
453 EXPORT_SYMBOL_GPL(iommu_group_alloc);
454
455 struct iommu_group *iommu_group_get_by_id(int id)
456 {
457 struct kobject *group_kobj;
458 struct iommu_group *group;
459 const char *name;
460
461 if (!iommu_group_kset)
462 return NULL;
463
464 name = kasprintf(GFP_KERNEL, "%d", id);
465 if (!name)
466 return NULL;
467
468 group_kobj = kset_find_obj(iommu_group_kset, name);
469 kfree(name);
470
471 if (!group_kobj)
472 return NULL;
473
474 group = container_of(group_kobj, struct iommu_group, kobj);
475 BUG_ON(group->id != id);
476
477 kobject_get(group->devices_kobj);
478 kobject_put(&group->kobj);
479
480 return group;
481 }
482 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
483
484 /**
485 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
486 * @group: the group
487 *
488 * iommu drivers can store data in the group for use when doing iommu
489 * operations. This function provides a way to retrieve it. Caller
490 * should hold a group reference.
491 */
492 void *iommu_group_get_iommudata(struct iommu_group *group)
493 {
494 return group->iommu_data;
495 }
496 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
497
498 /**
499 * iommu_group_set_iommudata - set iommu_data for a group
500 * @group: the group
501 * @iommu_data: new data
502 * @release: release function for iommu_data
503 *
504 * iommu drivers can store data in the group for use when doing iommu
505 * operations. This function provides a way to set the data after
506 * the group has been allocated. Caller should hold a group reference.
507 */
508 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
509 void (*release)(void *iommu_data))
510 {
511 group->iommu_data = iommu_data;
512 group->iommu_data_release = release;
513 }
514 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
515
516 /**
517 * iommu_group_set_name - set name for a group
518 * @group: the group
519 * @name: name
520 *
521 * Allow iommu driver to set a name for a group. When set it will
522 * appear in a name attribute file under the group in sysfs.
523 */
524 int iommu_group_set_name(struct iommu_group *group, const char *name)
525 {
526 int ret;
527
528 if (group->name) {
529 iommu_group_remove_file(group, &iommu_group_attr_name);
530 kfree(group->name);
531 group->name = NULL;
532 if (!name)
533 return 0;
534 }
535
536 group->name = kstrdup(name, GFP_KERNEL);
537 if (!group->name)
538 return -ENOMEM;
539
540 ret = iommu_group_create_file(group, &iommu_group_attr_name);
541 if (ret) {
542 kfree(group->name);
543 group->name = NULL;
544 return ret;
545 }
546
547 return 0;
548 }
549 EXPORT_SYMBOL_GPL(iommu_group_set_name);
550
551 static int iommu_group_create_direct_mappings(struct iommu_group *group,
552 struct device *dev)
553 {
554 struct iommu_domain *domain = group->default_domain;
555 struct iommu_resv_region *entry;
556 struct list_head mappings;
557 unsigned long pg_size;
558 int ret = 0;
559
560 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
561 return 0;
562
563 BUG_ON(!domain->pgsize_bitmap);
564
565 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
566 INIT_LIST_HEAD(&mappings);
567
568 iommu_get_resv_regions(dev, &mappings);
569
570 /* We need to consider overlapping regions for different devices */
571 list_for_each_entry(entry, &mappings, list) {
572 dma_addr_t start, end, addr;
573
574 if (domain->ops->apply_resv_region)
575 domain->ops->apply_resv_region(dev, domain, entry);
576
577 start = ALIGN(entry->start, pg_size);
578 end = ALIGN(entry->start + entry->length, pg_size);
579
580 if (entry->type != IOMMU_RESV_DIRECT)
581 continue;
582
583 for (addr = start; addr < end; addr += pg_size) {
584 phys_addr_t phys_addr;
585
586 phys_addr = iommu_iova_to_phys(domain, addr);
587 if (phys_addr)
588 continue;
589
590 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
591 if (ret)
592 goto out;
593 }
594
595 }
596
597 iommu_flush_tlb_all(domain);
598
599 out:
600 iommu_put_resv_regions(dev, &mappings);
601
602 return ret;
603 }
604
605 /**
606 * iommu_group_add_device - add a device to an iommu group
607 * @group: the group into which to add the device (reference should be held)
608 * @dev: the device
609 *
610 * This function is called by an iommu driver to add a device into a
611 * group. Adding a device increments the group reference count.
612 */
613 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
614 {
615 int ret, i = 0;
616 struct group_device *device;
617
618 device = kzalloc(sizeof(*device), GFP_KERNEL);
619 if (!device)
620 return -ENOMEM;
621
622 device->dev = dev;
623
624 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
625 if (ret)
626 goto err_free_device;
627
628 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
629 rename:
630 if (!device->name) {
631 ret = -ENOMEM;
632 goto err_remove_link;
633 }
634
635 ret = sysfs_create_link_nowarn(group->devices_kobj,
636 &dev->kobj, device->name);
637 if (ret) {
638 if (ret == -EEXIST && i >= 0) {
639 /*
640 * Account for the slim chance of collision
641 * and append an instance to the name.
642 */
643 kfree(device->name);
644 device->name = kasprintf(GFP_KERNEL, "%s.%d",
645 kobject_name(&dev->kobj), i++);
646 goto rename;
647 }
648 goto err_free_name;
649 }
650
651 kobject_get(group->devices_kobj);
652
653 dev->iommu_group = group;
654
655 iommu_group_create_direct_mappings(group, dev);
656
657 mutex_lock(&group->mutex);
658 list_add_tail(&device->list, &group->devices);
659 if (group->domain)
660 ret = __iommu_attach_device(group->domain, dev);
661 mutex_unlock(&group->mutex);
662 if (ret)
663 goto err_put_group;
664
665 /* Notify any listeners about change to group. */
666 blocking_notifier_call_chain(&group->notifier,
667 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
668
669 trace_add_device_to_group(group->id, dev);
670
671 dev_info(dev, "Adding to iommu group %d\n", group->id);
672
673 return 0;
674
675 err_put_group:
676 mutex_lock(&group->mutex);
677 list_del(&device->list);
678 mutex_unlock(&group->mutex);
679 dev->iommu_group = NULL;
680 kobject_put(group->devices_kobj);
681 err_free_name:
682 kfree(device->name);
683 err_remove_link:
684 sysfs_remove_link(&dev->kobj, "iommu_group");
685 err_free_device:
686 kfree(device);
687 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
688 return ret;
689 }
690 EXPORT_SYMBOL_GPL(iommu_group_add_device);
691
692 /**
693 * iommu_group_remove_device - remove a device from it's current group
694 * @dev: device to be removed
695 *
696 * This function is called by an iommu driver to remove the device from
697 * it's current group. This decrements the iommu group reference count.
698 */
699 void iommu_group_remove_device(struct device *dev)
700 {
701 struct iommu_group *group = dev->iommu_group;
702 struct group_device *tmp_device, *device = NULL;
703
704 dev_info(dev, "Removing from iommu group %d\n", group->id);
705
706 /* Pre-notify listeners that a device is being removed. */
707 blocking_notifier_call_chain(&group->notifier,
708 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
709
710 mutex_lock(&group->mutex);
711 list_for_each_entry(tmp_device, &group->devices, list) {
712 if (tmp_device->dev == dev) {
713 device = tmp_device;
714 list_del(&device->list);
715 break;
716 }
717 }
718 mutex_unlock(&group->mutex);
719
720 if (!device)
721 return;
722
723 sysfs_remove_link(group->devices_kobj, device->name);
724 sysfs_remove_link(&dev->kobj, "iommu_group");
725
726 trace_remove_device_from_group(group->id, dev);
727
728 kfree(device->name);
729 kfree(device);
730 dev->iommu_group = NULL;
731 kobject_put(group->devices_kobj);
732 }
733 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
734
735 static int iommu_group_device_count(struct iommu_group *group)
736 {
737 struct group_device *entry;
738 int ret = 0;
739
740 list_for_each_entry(entry, &group->devices, list)
741 ret++;
742
743 return ret;
744 }
745
746 /**
747 * iommu_group_for_each_dev - iterate over each device in the group
748 * @group: the group
749 * @data: caller opaque data to be passed to callback function
750 * @fn: caller supplied callback function
751 *
752 * This function is called by group users to iterate over group devices.
753 * Callers should hold a reference count to the group during callback.
754 * The group->mutex is held across callbacks, which will block calls to
755 * iommu_group_add/remove_device.
756 */
757 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
758 int (*fn)(struct device *, void *))
759 {
760 struct group_device *device;
761 int ret = 0;
762
763 list_for_each_entry(device, &group->devices, list) {
764 ret = fn(device->dev, data);
765 if (ret)
766 break;
767 }
768 return ret;
769 }
770
771
772 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
773 int (*fn)(struct device *, void *))
774 {
775 int ret;
776
777 mutex_lock(&group->mutex);
778 ret = __iommu_group_for_each_dev(group, data, fn);
779 mutex_unlock(&group->mutex);
780
781 return ret;
782 }
783 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
784
785 /**
786 * iommu_group_get - Return the group for a device and increment reference
787 * @dev: get the group that this device belongs to
788 *
789 * This function is called by iommu drivers and users to get the group
790 * for the specified device. If found, the group is returned and the group
791 * reference in incremented, else NULL.
792 */
793 struct iommu_group *iommu_group_get(struct device *dev)
794 {
795 struct iommu_group *group = dev->iommu_group;
796
797 if (group)
798 kobject_get(group->devices_kobj);
799
800 return group;
801 }
802 EXPORT_SYMBOL_GPL(iommu_group_get);
803
804 /**
805 * iommu_group_ref_get - Increment reference on a group
806 * @group: the group to use, must not be NULL
807 *
808 * This function is called by iommu drivers to take additional references on an
809 * existing group. Returns the given group for convenience.
810 */
811 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
812 {
813 kobject_get(group->devices_kobj);
814 return group;
815 }
816
817 /**
818 * iommu_group_put - Decrement group reference
819 * @group: the group to use
820 *
821 * This function is called by iommu drivers and users to release the
822 * iommu group. Once the reference count is zero, the group is released.
823 */
824 void iommu_group_put(struct iommu_group *group)
825 {
826 if (group)
827 kobject_put(group->devices_kobj);
828 }
829 EXPORT_SYMBOL_GPL(iommu_group_put);
830
831 /**
832 * iommu_group_register_notifier - Register a notifier for group changes
833 * @group: the group to watch
834 * @nb: notifier block to signal
835 *
836 * This function allows iommu group users to track changes in a group.
837 * See include/linux/iommu.h for actions sent via this notifier. Caller
838 * should hold a reference to the group throughout notifier registration.
839 */
840 int iommu_group_register_notifier(struct iommu_group *group,
841 struct notifier_block *nb)
842 {
843 return blocking_notifier_chain_register(&group->notifier, nb);
844 }
845 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
846
847 /**
848 * iommu_group_unregister_notifier - Unregister a notifier
849 * @group: the group to watch
850 * @nb: notifier block to signal
851 *
852 * Unregister a previously registered group notifier block.
853 */
854 int iommu_group_unregister_notifier(struct iommu_group *group,
855 struct notifier_block *nb)
856 {
857 return blocking_notifier_chain_unregister(&group->notifier, nb);
858 }
859 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
860
861 /**
862 * iommu_group_id - Return ID for a group
863 * @group: the group to ID
864 *
865 * Return the unique ID for the group matching the sysfs group number.
866 */
867 int iommu_group_id(struct iommu_group *group)
868 {
869 return group->id;
870 }
871 EXPORT_SYMBOL_GPL(iommu_group_id);
872
873 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
874 unsigned long *devfns);
875
876 /*
877 * To consider a PCI device isolated, we require ACS to support Source
878 * Validation, Request Redirection, Completer Redirection, and Upstream
879 * Forwarding. This effectively means that devices cannot spoof their
880 * requester ID, requests and completions cannot be redirected, and all
881 * transactions are forwarded upstream, even as it passes through a
882 * bridge where the target device is downstream.
883 */
884 #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
885
886 /*
887 * For multifunction devices which are not isolated from each other, find
888 * all the other non-isolated functions and look for existing groups. For
889 * each function, we also need to look for aliases to or from other devices
890 * that may already have a group.
891 */
892 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
893 unsigned long *devfns)
894 {
895 struct pci_dev *tmp = NULL;
896 struct iommu_group *group;
897
898 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
899 return NULL;
900
901 for_each_pci_dev(tmp) {
902 if (tmp == pdev || tmp->bus != pdev->bus ||
903 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
904 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
905 continue;
906
907 group = get_pci_alias_group(tmp, devfns);
908 if (group) {
909 pci_dev_put(tmp);
910 return group;
911 }
912 }
913
914 return NULL;
915 }
916
917 /*
918 * Look for aliases to or from the given device for existing groups. DMA
919 * aliases are only supported on the same bus, therefore the search
920 * space is quite small (especially since we're really only looking at pcie
921 * device, and therefore only expect multiple slots on the root complex or
922 * downstream switch ports). It's conceivable though that a pair of
923 * multifunction devices could have aliases between them that would cause a
924 * loop. To prevent this, we use a bitmap to track where we've been.
925 */
926 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
927 unsigned long *devfns)
928 {
929 struct pci_dev *tmp = NULL;
930 struct iommu_group *group;
931
932 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
933 return NULL;
934
935 group = iommu_group_get(&pdev->dev);
936 if (group)
937 return group;
938
939 for_each_pci_dev(tmp) {
940 if (tmp == pdev || tmp->bus != pdev->bus)
941 continue;
942
943 /* We alias them or they alias us */
944 if (pci_devs_are_dma_aliases(pdev, tmp)) {
945 group = get_pci_alias_group(tmp, devfns);
946 if (group) {
947 pci_dev_put(tmp);
948 return group;
949 }
950
951 group = get_pci_function_alias_group(tmp, devfns);
952 if (group) {
953 pci_dev_put(tmp);
954 return group;
955 }
956 }
957 }
958
959 return NULL;
960 }
961
962 struct group_for_pci_data {
963 struct pci_dev *pdev;
964 struct iommu_group *group;
965 };
966
967 /*
968 * DMA alias iterator callback, return the last seen device. Stop and return
969 * the IOMMU group if we find one along the way.
970 */
971 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
972 {
973 struct group_for_pci_data *data = opaque;
974
975 data->pdev = pdev;
976 data->group = iommu_group_get(&pdev->dev);
977
978 return data->group != NULL;
979 }
980
981 /*
982 * Generic device_group call-back function. It just allocates one
983 * iommu-group per device.
984 */
985 struct iommu_group *generic_device_group(struct device *dev)
986 {
987 return iommu_group_alloc();
988 }
989
990 /*
991 * Use standard PCI bus topology, isolation features, and DMA alias quirks
992 * to find or create an IOMMU group for a device.
993 */
994 struct iommu_group *pci_device_group(struct device *dev)
995 {
996 struct pci_dev *pdev = to_pci_dev(dev);
997 struct group_for_pci_data data;
998 struct pci_bus *bus;
999 struct iommu_group *group = NULL;
1000 u64 devfns[4] = { 0 };
1001
1002 if (WARN_ON(!dev_is_pci(dev)))
1003 return ERR_PTR(-EINVAL);
1004
1005 /*
1006 * Find the upstream DMA alias for the device. A device must not
1007 * be aliased due to topology in order to have its own IOMMU group.
1008 * If we find an alias along the way that already belongs to a
1009 * group, use it.
1010 */
1011 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1012 return data.group;
1013
1014 pdev = data.pdev;
1015
1016 /*
1017 * Continue upstream from the point of minimum IOMMU granularity
1018 * due to aliases to the point where devices are protected from
1019 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1020 * group, use it.
1021 */
1022 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1023 if (!bus->self)
1024 continue;
1025
1026 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1027 break;
1028
1029 pdev = bus->self;
1030
1031 group = iommu_group_get(&pdev->dev);
1032 if (group)
1033 return group;
1034 }
1035
1036 /*
1037 * Look for existing groups on device aliases. If we alias another
1038 * device or another device aliases us, use the same group.
1039 */
1040 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1041 if (group)
1042 return group;
1043
1044 /*
1045 * Look for existing groups on non-isolated functions on the same
1046 * slot and aliases of those funcions, if any. No need to clear
1047 * the search bitmap, the tested devfns are still valid.
1048 */
1049 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1050 if (group)
1051 return group;
1052
1053 /* No shared group found, allocate new */
1054 return iommu_group_alloc();
1055 }
1056
1057 /* Get the IOMMU group for device on fsl-mc bus */
1058 struct iommu_group *fsl_mc_device_group(struct device *dev)
1059 {
1060 struct device *cont_dev = fsl_mc_cont_dev(dev);
1061 struct iommu_group *group;
1062
1063 group = iommu_group_get(cont_dev);
1064 if (!group)
1065 group = iommu_group_alloc();
1066 return group;
1067 }
1068
1069 /**
1070 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1071 * @dev: target device
1072 *
1073 * This function is intended to be called by IOMMU drivers and extended to
1074 * support common, bus-defined algorithms when determining or creating the
1075 * IOMMU group for a device. On success, the caller will hold a reference
1076 * to the returned IOMMU group, which will already include the provided
1077 * device. The reference should be released with iommu_group_put().
1078 */
1079 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1080 {
1081 const struct iommu_ops *ops = dev->bus->iommu_ops;
1082 struct iommu_group *group;
1083 int ret;
1084
1085 group = iommu_group_get(dev);
1086 if (group)
1087 return group;
1088
1089 if (!ops)
1090 return ERR_PTR(-EINVAL);
1091
1092 group = ops->device_group(dev);
1093 if (WARN_ON_ONCE(group == NULL))
1094 return ERR_PTR(-EINVAL);
1095
1096 if (IS_ERR(group))
1097 return group;
1098
1099 /*
1100 * Try to allocate a default domain - needs support from the
1101 * IOMMU driver.
1102 */
1103 if (!group->default_domain) {
1104 struct iommu_domain *dom;
1105
1106 dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1107 if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
1108 dev_warn(dev,
1109 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1110 iommu_def_domain_type);
1111 dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
1112 }
1113
1114 group->default_domain = dom;
1115 if (!group->domain)
1116 group->domain = dom;
1117
1118 if (dom && !iommu_dma_strict) {
1119 int attr = 1;
1120 iommu_domain_set_attr(dom,
1121 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1122 &attr);
1123 }
1124 }
1125
1126 ret = iommu_group_add_device(group, dev);
1127 if (ret) {
1128 iommu_group_put(group);
1129 return ERR_PTR(ret);
1130 }
1131
1132 return group;
1133 }
1134
1135 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1136 {
1137 return group->default_domain;
1138 }
1139
1140 static int add_iommu_group(struct device *dev, void *data)
1141 {
1142 int ret = iommu_probe_device(dev);
1143
1144 /*
1145 * We ignore -ENODEV errors for now, as they just mean that the
1146 * device is not translated by an IOMMU. We still care about
1147 * other errors and fail to initialize when they happen.
1148 */
1149 if (ret == -ENODEV)
1150 ret = 0;
1151
1152 return ret;
1153 }
1154
1155 static int remove_iommu_group(struct device *dev, void *data)
1156 {
1157 iommu_release_device(dev);
1158
1159 return 0;
1160 }
1161
1162 static int iommu_bus_notifier(struct notifier_block *nb,
1163 unsigned long action, void *data)
1164 {
1165 unsigned long group_action = 0;
1166 struct device *dev = data;
1167 struct iommu_group *group;
1168
1169 /*
1170 * ADD/DEL call into iommu driver ops if provided, which may
1171 * result in ADD/DEL notifiers to group->notifier
1172 */
1173 if (action == BUS_NOTIFY_ADD_DEVICE) {
1174 int ret;
1175
1176 ret = iommu_probe_device(dev);
1177 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1178 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1179 iommu_release_device(dev);
1180 return NOTIFY_OK;
1181 }
1182
1183 /*
1184 * Remaining BUS_NOTIFYs get filtered and republished to the
1185 * group, if anyone is listening
1186 */
1187 group = iommu_group_get(dev);
1188 if (!group)
1189 return 0;
1190
1191 switch (action) {
1192 case BUS_NOTIFY_BIND_DRIVER:
1193 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1194 break;
1195 case BUS_NOTIFY_BOUND_DRIVER:
1196 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1197 break;
1198 case BUS_NOTIFY_UNBIND_DRIVER:
1199 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1200 break;
1201 case BUS_NOTIFY_UNBOUND_DRIVER:
1202 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1203 break;
1204 }
1205
1206 if (group_action)
1207 blocking_notifier_call_chain(&group->notifier,
1208 group_action, dev);
1209
1210 iommu_group_put(group);
1211 return 0;
1212 }
1213
1214 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1215 {
1216 int err;
1217 struct notifier_block *nb;
1218 struct iommu_callback_data cb = {
1219 .ops = ops,
1220 };
1221
1222 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1223 if (!nb)
1224 return -ENOMEM;
1225
1226 nb->notifier_call = iommu_bus_notifier;
1227
1228 err = bus_register_notifier(bus, nb);
1229 if (err)
1230 goto out_free;
1231
1232 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
1233 if (err)
1234 goto out_err;
1235
1236
1237 return 0;
1238
1239 out_err:
1240 /* Clean up */
1241 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
1242 bus_unregister_notifier(bus, nb);
1243
1244 out_free:
1245 kfree(nb);
1246
1247 return err;
1248 }
1249
1250 /**
1251 * bus_set_iommu - set iommu-callbacks for the bus
1252 * @bus: bus.
1253 * @ops: the callbacks provided by the iommu-driver
1254 *
1255 * This function is called by an iommu driver to set the iommu methods
1256 * used for a particular bus. Drivers for devices on that bus can use
1257 * the iommu-api after these ops are registered.
1258 * This special function is needed because IOMMUs are usually devices on
1259 * the bus itself, so the iommu drivers are not initialized when the bus
1260 * is set up. With this function the iommu-driver can set the iommu-ops
1261 * afterwards.
1262 */
1263 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1264 {
1265 int err;
1266
1267 if (bus->iommu_ops != NULL)
1268 return -EBUSY;
1269
1270 bus->iommu_ops = ops;
1271
1272 /* Do IOMMU specific setup for this bus-type */
1273 err = iommu_bus_init(bus, ops);
1274 if (err)
1275 bus->iommu_ops = NULL;
1276
1277 return err;
1278 }
1279 EXPORT_SYMBOL_GPL(bus_set_iommu);
1280
1281 bool iommu_present(struct bus_type *bus)
1282 {
1283 return bus->iommu_ops != NULL;
1284 }
1285 EXPORT_SYMBOL_GPL(iommu_present);
1286
1287 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1288 {
1289 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1290 return false;
1291
1292 return bus->iommu_ops->capable(cap);
1293 }
1294 EXPORT_SYMBOL_GPL(iommu_capable);
1295
1296 /**
1297 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1298 * @domain: iommu domain
1299 * @handler: fault handler
1300 * @token: user data, will be passed back to the fault handler
1301 *
1302 * This function should be used by IOMMU users which want to be notified
1303 * whenever an IOMMU fault happens.
1304 *
1305 * The fault handler itself should return 0 on success, and an appropriate
1306 * error code otherwise.
1307 */
1308 void iommu_set_fault_handler(struct iommu_domain *domain,
1309 iommu_fault_handler_t handler,
1310 void *token)
1311 {
1312 BUG_ON(!domain);
1313
1314 domain->handler = handler;
1315 domain->handler_token = token;
1316 }
1317 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1318
1319 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1320 unsigned type)
1321 {
1322 struct iommu_domain *domain;
1323
1324 if (bus == NULL || bus->iommu_ops == NULL)
1325 return NULL;
1326
1327 domain = bus->iommu_ops->domain_alloc(type);
1328 if (!domain)
1329 return NULL;
1330
1331 domain->ops = bus->iommu_ops;
1332 domain->type = type;
1333 /* Assume all sizes by default; the driver may override this later */
1334 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1335
1336 return domain;
1337 }
1338
1339 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1340 {
1341 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1342 }
1343 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1344
1345 void iommu_domain_free(struct iommu_domain *domain)
1346 {
1347 domain->ops->domain_free(domain);
1348 }
1349 EXPORT_SYMBOL_GPL(iommu_domain_free);
1350
1351 static int __iommu_attach_device(struct iommu_domain *domain,
1352 struct device *dev)
1353 {
1354 int ret;
1355 if ((domain->ops->is_attach_deferred != NULL) &&
1356 domain->ops->is_attach_deferred(domain, dev))
1357 return 0;
1358
1359 if (unlikely(domain->ops->attach_dev == NULL))
1360 return -ENODEV;
1361
1362 ret = domain->ops->attach_dev(domain, dev);
1363 if (!ret)
1364 trace_attach_device_to_domain(dev);
1365 return ret;
1366 }
1367
1368 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1369 {
1370 struct iommu_group *group;
1371 int ret;
1372
1373 group = iommu_group_get(dev);
1374 if (!group)
1375 return -ENODEV;
1376
1377 /*
1378 * Lock the group to make sure the device-count doesn't
1379 * change while we are attaching
1380 */
1381 mutex_lock(&group->mutex);
1382 ret = -EINVAL;
1383 if (iommu_group_device_count(group) != 1)
1384 goto out_unlock;
1385
1386 ret = __iommu_attach_group(domain, group);
1387
1388 out_unlock:
1389 mutex_unlock(&group->mutex);
1390 iommu_group_put(group);
1391
1392 return ret;
1393 }
1394 EXPORT_SYMBOL_GPL(iommu_attach_device);
1395
1396 static void __iommu_detach_device(struct iommu_domain *domain,
1397 struct device *dev)
1398 {
1399 if ((domain->ops->is_attach_deferred != NULL) &&
1400 domain->ops->is_attach_deferred(domain, dev))
1401 return;
1402
1403 if (unlikely(domain->ops->detach_dev == NULL))
1404 return;
1405
1406 domain->ops->detach_dev(domain, dev);
1407 trace_detach_device_from_domain(dev);
1408 }
1409
1410 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1411 {
1412 struct iommu_group *group;
1413
1414 group = iommu_group_get(dev);
1415 if (!group)
1416 return;
1417
1418 mutex_lock(&group->mutex);
1419 if (iommu_group_device_count(group) != 1) {
1420 WARN_ON(1);
1421 goto out_unlock;
1422 }
1423
1424 __iommu_detach_group(domain, group);
1425
1426 out_unlock:
1427 mutex_unlock(&group->mutex);
1428 iommu_group_put(group);
1429 }
1430 EXPORT_SYMBOL_GPL(iommu_detach_device);
1431
1432 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1433 {
1434 struct iommu_domain *domain;
1435 struct iommu_group *group;
1436
1437 group = iommu_group_get(dev);
1438 if (!group)
1439 return NULL;
1440
1441 domain = group->domain;
1442
1443 iommu_group_put(group);
1444
1445 return domain;
1446 }
1447 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1448
1449 /*
1450 * For IOMMU_DOMAIN_DMA implementations which already provide their own
1451 * guarantees that the group and its default domain are valid and correct.
1452 */
1453 struct iommu_domain *iommu_get_dma_domain(struct device *dev)
1454 {
1455 return dev->iommu_group->default_domain;
1456 }
1457
1458 /*
1459 * IOMMU groups are really the natural working unit of the IOMMU, but
1460 * the IOMMU API works on domains and devices. Bridge that gap by
1461 * iterating over the devices in a group. Ideally we'd have a single
1462 * device which represents the requestor ID of the group, but we also
1463 * allow IOMMU drivers to create policy defined minimum sets, where
1464 * the physical hardware may be able to distiguish members, but we
1465 * wish to group them at a higher level (ex. untrusted multi-function
1466 * PCI devices). Thus we attach each device.
1467 */
1468 static int iommu_group_do_attach_device(struct device *dev, void *data)
1469 {
1470 struct iommu_domain *domain = data;
1471
1472 return __iommu_attach_device(domain, dev);
1473 }
1474
1475 static int __iommu_attach_group(struct iommu_domain *domain,
1476 struct iommu_group *group)
1477 {
1478 int ret;
1479
1480 if (group->default_domain && group->domain != group->default_domain)
1481 return -EBUSY;
1482
1483 ret = __iommu_group_for_each_dev(group, domain,
1484 iommu_group_do_attach_device);
1485 if (ret == 0)
1486 group->domain = domain;
1487
1488 return ret;
1489 }
1490
1491 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1492 {
1493 int ret;
1494
1495 mutex_lock(&group->mutex);
1496 ret = __iommu_attach_group(domain, group);
1497 mutex_unlock(&group->mutex);
1498
1499 return ret;
1500 }
1501 EXPORT_SYMBOL_GPL(iommu_attach_group);
1502
1503 static int iommu_group_do_detach_device(struct device *dev, void *data)
1504 {
1505 struct iommu_domain *domain = data;
1506
1507 __iommu_detach_device(domain, dev);
1508
1509 return 0;
1510 }
1511
1512 static void __iommu_detach_group(struct iommu_domain *domain,
1513 struct iommu_group *group)
1514 {
1515 int ret;
1516
1517 if (!group->default_domain) {
1518 __iommu_group_for_each_dev(group, domain,
1519 iommu_group_do_detach_device);
1520 group->domain = NULL;
1521 return;
1522 }
1523
1524 if (group->domain == group->default_domain)
1525 return;
1526
1527 /* Detach by re-attaching to the default domain */
1528 ret = __iommu_group_for_each_dev(group, group->default_domain,
1529 iommu_group_do_attach_device);
1530 if (ret != 0)
1531 WARN_ON(1);
1532 else
1533 group->domain = group->default_domain;
1534 }
1535
1536 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1537 {
1538 mutex_lock(&group->mutex);
1539 __iommu_detach_group(domain, group);
1540 mutex_unlock(&group->mutex);
1541 }
1542 EXPORT_SYMBOL_GPL(iommu_detach_group);
1543
1544 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1545 {
1546 if (unlikely(domain->ops->iova_to_phys == NULL))
1547 return 0;
1548
1549 return domain->ops->iova_to_phys(domain, iova);
1550 }
1551 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1552
1553 static size_t iommu_pgsize(struct iommu_domain *domain,
1554 unsigned long addr_merge, size_t size)
1555 {
1556 unsigned int pgsize_idx;
1557 size_t pgsize;
1558
1559 /* Max page size that still fits into 'size' */
1560 pgsize_idx = __fls(size);
1561
1562 /* need to consider alignment requirements ? */
1563 if (likely(addr_merge)) {
1564 /* Max page size allowed by address */
1565 unsigned int align_pgsize_idx = __ffs(addr_merge);
1566 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1567 }
1568
1569 /* build a mask of acceptable page sizes */
1570 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1571
1572 /* throw away page sizes not supported by the hardware */
1573 pgsize &= domain->pgsize_bitmap;
1574
1575 /* make sure we're still sane */
1576 BUG_ON(!pgsize);
1577
1578 /* pick the biggest page */
1579 pgsize_idx = __fls(pgsize);
1580 pgsize = 1UL << pgsize_idx;
1581
1582 return pgsize;
1583 }
1584
1585 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1586 phys_addr_t paddr, size_t size, int prot)
1587 {
1588 const struct iommu_ops *ops = domain->ops;
1589 unsigned long orig_iova = iova;
1590 unsigned int min_pagesz;
1591 size_t orig_size = size;
1592 phys_addr_t orig_paddr = paddr;
1593 int ret = 0;
1594
1595 if (unlikely(ops->map == NULL ||
1596 domain->pgsize_bitmap == 0UL))
1597 return -ENODEV;
1598
1599 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1600 return -EINVAL;
1601
1602 /* find out the minimum page size supported */
1603 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1604
1605 /*
1606 * both the virtual address and the physical one, as well as
1607 * the size of the mapping, must be aligned (at least) to the
1608 * size of the smallest page supported by the hardware
1609 */
1610 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1611 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1612 iova, &paddr, size, min_pagesz);
1613 return -EINVAL;
1614 }
1615
1616 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1617
1618 while (size) {
1619 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1620
1621 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1622 iova, &paddr, pgsize);
1623
1624 ret = ops->map(domain, iova, paddr, pgsize, prot);
1625 if (ret)
1626 break;
1627
1628 iova += pgsize;
1629 paddr += pgsize;
1630 size -= pgsize;
1631 }
1632
1633 if (ops->iotlb_sync_map)
1634 ops->iotlb_sync_map(domain);
1635
1636 /* unroll mapping in case something went wrong */
1637 if (ret)
1638 iommu_unmap(domain, orig_iova, orig_size - size);
1639 else
1640 trace_map(orig_iova, orig_paddr, orig_size);
1641
1642 return ret;
1643 }
1644 EXPORT_SYMBOL_GPL(iommu_map);
1645
1646 static size_t __iommu_unmap(struct iommu_domain *domain,
1647 unsigned long iova, size_t size,
1648 bool sync)
1649 {
1650 const struct iommu_ops *ops = domain->ops;
1651 size_t unmapped_page, unmapped = 0;
1652 unsigned long orig_iova = iova;
1653 unsigned int min_pagesz;
1654
1655 if (unlikely(ops->unmap == NULL ||
1656 domain->pgsize_bitmap == 0UL))
1657 return 0;
1658
1659 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1660 return 0;
1661
1662 /* find out the minimum page size supported */
1663 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1664
1665 /*
1666 * The virtual address, as well as the size of the mapping, must be
1667 * aligned (at least) to the size of the smallest page supported
1668 * by the hardware
1669 */
1670 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1671 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1672 iova, size, min_pagesz);
1673 return 0;
1674 }
1675
1676 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1677
1678 /*
1679 * Keep iterating until we either unmap 'size' bytes (or more)
1680 * or we hit an area that isn't mapped.
1681 */
1682 while (unmapped < size) {
1683 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1684
1685 unmapped_page = ops->unmap(domain, iova, pgsize);
1686 if (!unmapped_page)
1687 break;
1688
1689 if (sync && ops->iotlb_range_add)
1690 ops->iotlb_range_add(domain, iova, pgsize);
1691
1692 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1693 iova, unmapped_page);
1694
1695 iova += unmapped_page;
1696 unmapped += unmapped_page;
1697 }
1698
1699 if (sync && ops->iotlb_sync)
1700 ops->iotlb_sync(domain);
1701
1702 trace_unmap(orig_iova, size, unmapped);
1703 return unmapped;
1704 }
1705
1706 size_t iommu_unmap(struct iommu_domain *domain,
1707 unsigned long iova, size_t size)
1708 {
1709 return __iommu_unmap(domain, iova, size, true);
1710 }
1711 EXPORT_SYMBOL_GPL(iommu_unmap);
1712
1713 size_t iommu_unmap_fast(struct iommu_domain *domain,
1714 unsigned long iova, size_t size)
1715 {
1716 return __iommu_unmap(domain, iova, size, false);
1717 }
1718 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
1719
1720 size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1721 struct scatterlist *sg, unsigned int nents, int prot)
1722 {
1723 size_t len = 0, mapped = 0;
1724 phys_addr_t start;
1725 unsigned int i = 0;
1726 int ret;
1727
1728 while (i <= nents) {
1729 phys_addr_t s_phys = sg_phys(sg);
1730
1731 if (len && s_phys != start + len) {
1732 ret = iommu_map(domain, iova + mapped, start, len, prot);
1733 if (ret)
1734 goto out_err;
1735
1736 mapped += len;
1737 len = 0;
1738 }
1739
1740 if (len) {
1741 len += sg->length;
1742 } else {
1743 len = sg->length;
1744 start = s_phys;
1745 }
1746
1747 if (++i < nents)
1748 sg = sg_next(sg);
1749 }
1750
1751 return mapped;
1752
1753 out_err:
1754 /* undo mappings already done */
1755 iommu_unmap(domain, iova, mapped);
1756
1757 return 0;
1758
1759 }
1760 EXPORT_SYMBOL_GPL(iommu_map_sg);
1761
1762 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1763 phys_addr_t paddr, u64 size, int prot)
1764 {
1765 if (unlikely(domain->ops->domain_window_enable == NULL))
1766 return -ENODEV;
1767
1768 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1769 prot);
1770 }
1771 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1772
1773 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1774 {
1775 if (unlikely(domain->ops->domain_window_disable == NULL))
1776 return;
1777
1778 return domain->ops->domain_window_disable(domain, wnd_nr);
1779 }
1780 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1781
1782 /**
1783 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
1784 * @domain: the iommu domain where the fault has happened
1785 * @dev: the device where the fault has happened
1786 * @iova: the faulting address
1787 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
1788 *
1789 * This function should be called by the low-level IOMMU implementations
1790 * whenever IOMMU faults happen, to allow high-level users, that are
1791 * interested in such events, to know about them.
1792 *
1793 * This event may be useful for several possible use cases:
1794 * - mere logging of the event
1795 * - dynamic TLB/PTE loading
1796 * - if restarting of the faulting device is required
1797 *
1798 * Returns 0 on success and an appropriate error code otherwise (if dynamic
1799 * PTE/TLB loading will one day be supported, implementations will be able
1800 * to tell whether it succeeded or not according to this return value).
1801 *
1802 * Specifically, -ENOSYS is returned if a fault handler isn't installed
1803 * (though fault handlers can also return -ENOSYS, in case they want to
1804 * elicit the default behavior of the IOMMU drivers).
1805 */
1806 int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
1807 unsigned long iova, int flags)
1808 {
1809 int ret = -ENOSYS;
1810
1811 /*
1812 * if upper layers showed interest and installed a fault handler,
1813 * invoke it.
1814 */
1815 if (domain->handler)
1816 ret = domain->handler(domain, dev, iova, flags,
1817 domain->handler_token);
1818
1819 trace_io_page_fault(dev, iova, flags);
1820 return ret;
1821 }
1822 EXPORT_SYMBOL_GPL(report_iommu_fault);
1823
1824 static int __init iommu_init(void)
1825 {
1826 iommu_group_kset = kset_create_and_add("iommu_groups",
1827 NULL, kernel_kobj);
1828 BUG_ON(!iommu_group_kset);
1829
1830 iommu_debugfs_setup();
1831
1832 return 0;
1833 }
1834 core_initcall(iommu_init);
1835
1836 int iommu_domain_get_attr(struct iommu_domain *domain,
1837 enum iommu_attr attr, void *data)
1838 {
1839 struct iommu_domain_geometry *geometry;
1840 bool *paging;
1841 int ret = 0;
1842
1843 switch (attr) {
1844 case DOMAIN_ATTR_GEOMETRY:
1845 geometry = data;
1846 *geometry = domain->geometry;
1847
1848 break;
1849 case DOMAIN_ATTR_PAGING:
1850 paging = data;
1851 *paging = (domain->pgsize_bitmap != 0UL);
1852 break;
1853 default:
1854 if (!domain->ops->domain_get_attr)
1855 return -EINVAL;
1856
1857 ret = domain->ops->domain_get_attr(domain, attr, data);
1858 }
1859
1860 return ret;
1861 }
1862 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1863
1864 int iommu_domain_set_attr(struct iommu_domain *domain,
1865 enum iommu_attr attr, void *data)
1866 {
1867 int ret = 0;
1868
1869 switch (attr) {
1870 default:
1871 if (domain->ops->domain_set_attr == NULL)
1872 return -EINVAL;
1873
1874 ret = domain->ops->domain_set_attr(domain, attr, data);
1875 }
1876
1877 return ret;
1878 }
1879 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1880
1881 void iommu_get_resv_regions(struct device *dev, struct list_head *list)
1882 {
1883 const struct iommu_ops *ops = dev->bus->iommu_ops;
1884
1885 if (ops && ops->get_resv_regions)
1886 ops->get_resv_regions(dev, list);
1887 }
1888
1889 void iommu_put_resv_regions(struct device *dev, struct list_head *list)
1890 {
1891 const struct iommu_ops *ops = dev->bus->iommu_ops;
1892
1893 if (ops && ops->put_resv_regions)
1894 ops->put_resv_regions(dev, list);
1895 }
1896
1897 struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
1898 size_t length, int prot,
1899 enum iommu_resv_type type)
1900 {
1901 struct iommu_resv_region *region;
1902
1903 region = kzalloc(sizeof(*region), GFP_KERNEL);
1904 if (!region)
1905 return NULL;
1906
1907 INIT_LIST_HEAD(&region->list);
1908 region->start = start;
1909 region->length = length;
1910 region->prot = prot;
1911 region->type = type;
1912 return region;
1913 }
1914
1915 /* Request that a device is direct mapped by the IOMMU */
1916 int iommu_request_dm_for_dev(struct device *dev)
1917 {
1918 struct iommu_domain *dm_domain;
1919 struct iommu_group *group;
1920 int ret;
1921
1922 /* Device must already be in a group before calling this function */
1923 group = iommu_group_get_for_dev(dev);
1924 if (IS_ERR(group))
1925 return PTR_ERR(group);
1926
1927 mutex_lock(&group->mutex);
1928
1929 /* Check if the default domain is already direct mapped */
1930 ret = 0;
1931 if (group->default_domain &&
1932 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1933 goto out;
1934
1935 /* Don't change mappings of existing devices */
1936 ret = -EBUSY;
1937 if (iommu_group_device_count(group) != 1)
1938 goto out;
1939
1940 /* Allocate a direct mapped domain */
1941 ret = -ENOMEM;
1942 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1943 if (!dm_domain)
1944 goto out;
1945
1946 /* Attach the device to the domain */
1947 ret = __iommu_attach_group(dm_domain, group);
1948 if (ret) {
1949 iommu_domain_free(dm_domain);
1950 goto out;
1951 }
1952
1953 /* Make the direct mapped domain the default for this group */
1954 if (group->default_domain)
1955 iommu_domain_free(group->default_domain);
1956 group->default_domain = dm_domain;
1957
1958 dev_info(dev, "Using iommu direct mapping\n");
1959
1960 ret = 0;
1961 out:
1962 mutex_unlock(&group->mutex);
1963 iommu_group_put(group);
1964
1965 return ret;
1966 }
1967
1968 const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
1969 {
1970 const struct iommu_ops *ops = NULL;
1971 struct iommu_device *iommu;
1972
1973 spin_lock(&iommu_device_lock);
1974 list_for_each_entry(iommu, &iommu_device_list, list)
1975 if (iommu->fwnode == fwnode) {
1976 ops = iommu->ops;
1977 break;
1978 }
1979 spin_unlock(&iommu_device_lock);
1980 return ops;
1981 }
1982
1983 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1984 const struct iommu_ops *ops)
1985 {
1986 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
1987
1988 if (fwspec)
1989 return ops == fwspec->ops ? 0 : -EINVAL;
1990
1991 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1992 if (!fwspec)
1993 return -ENOMEM;
1994
1995 of_node_get(to_of_node(iommu_fwnode));
1996 fwspec->iommu_fwnode = iommu_fwnode;
1997 fwspec->ops = ops;
1998 dev_iommu_fwspec_set(dev, fwspec);
1999 return 0;
2000 }
2001 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2002
2003 void iommu_fwspec_free(struct device *dev)
2004 {
2005 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2006
2007 if (fwspec) {
2008 fwnode_handle_put(fwspec->iommu_fwnode);
2009 kfree(fwspec);
2010 dev_iommu_fwspec_set(dev, NULL);
2011 }
2012 }
2013 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2014
2015 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2016 {
2017 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
2018 size_t size;
2019 int i;
2020
2021 if (!fwspec)
2022 return -EINVAL;
2023
2024 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
2025 if (size > sizeof(*fwspec)) {
2026 fwspec = krealloc(fwspec, size, GFP_KERNEL);
2027 if (!fwspec)
2028 return -ENOMEM;
2029
2030 dev_iommu_fwspec_set(dev, fwspec);
2031 }
2032
2033 for (i = 0; i < num_ids; i++)
2034 fwspec->ids[fwspec->num_ids + i] = ids[i];
2035
2036 fwspec->num_ids += num_ids;
2037 return 0;
2038 }
2039 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);