Loading...
1/*
2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
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#include <linux/bug.h>
20#include <linux/types.h>
21#include <linux/module.h>
22#include <linux/slab.h>
23#include <linux/errno.h>
24#include <linux/iommu.h>
25
26static struct iommu_ops *iommu_ops;
27
28void register_iommu(struct iommu_ops *ops)
29{
30 if (iommu_ops)
31 BUG();
32
33 iommu_ops = ops;
34}
35
36bool iommu_found(void)
37{
38 return iommu_ops != NULL;
39}
40EXPORT_SYMBOL_GPL(iommu_found);
41
42struct iommu_domain *iommu_domain_alloc(void)
43{
44 struct iommu_domain *domain;
45 int ret;
46
47 domain = kmalloc(sizeof(*domain), GFP_KERNEL);
48 if (!domain)
49 return NULL;
50
51 ret = iommu_ops->domain_init(domain);
52 if (ret)
53 goto out_free;
54
55 return domain;
56
57out_free:
58 kfree(domain);
59
60 return NULL;
61}
62EXPORT_SYMBOL_GPL(iommu_domain_alloc);
63
64void iommu_domain_free(struct iommu_domain *domain)
65{
66 iommu_ops->domain_destroy(domain);
67 kfree(domain);
68}
69EXPORT_SYMBOL_GPL(iommu_domain_free);
70
71int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
72{
73 return iommu_ops->attach_dev(domain, dev);
74}
75EXPORT_SYMBOL_GPL(iommu_attach_device);
76
77void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
78{
79 iommu_ops->detach_dev(domain, dev);
80}
81EXPORT_SYMBOL_GPL(iommu_detach_device);
82
83phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain,
84 unsigned long iova)
85{
86 return iommu_ops->iova_to_phys(domain, iova);
87}
88EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
89
90int iommu_domain_has_cap(struct iommu_domain *domain,
91 unsigned long cap)
92{
93 return iommu_ops->domain_has_cap(domain, cap);
94}
95EXPORT_SYMBOL_GPL(iommu_domain_has_cap);
96
97int iommu_map(struct iommu_domain *domain, unsigned long iova,
98 phys_addr_t paddr, int gfp_order, int prot)
99{
100 unsigned long invalid_mask;
101 size_t size;
102
103 size = 0x1000UL << gfp_order;
104 invalid_mask = size - 1;
105
106 BUG_ON((iova | paddr) & invalid_mask);
107
108 return iommu_ops->map(domain, iova, paddr, gfp_order, prot);
109}
110EXPORT_SYMBOL_GPL(iommu_map);
111
112int iommu_unmap(struct iommu_domain *domain, unsigned long iova, int gfp_order)
113{
114 unsigned long invalid_mask;
115 size_t size;
116
117 size = 0x1000UL << gfp_order;
118 invalid_mask = size - 1;
119
120 BUG_ON(iova & invalid_mask);
121
122 return iommu_ops->unmap(domain, iova, gfp_order);
123}
124EXPORT_SYMBOL_GPL(iommu_unmap);
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/module.h>
26#include <linux/slab.h>
27#include <linux/errno.h>
28#include <linux/iommu.h>
29#include <linux/idr.h>
30#include <linux/notifier.h>
31#include <linux/err.h>
32#include <linux/pci.h>
33#include <linux/bitops.h>
34#include <linux/property.h>
35#include <trace/events/iommu.h>
36
37static struct kset *iommu_group_kset;
38static DEFINE_IDA(iommu_group_ida);
39static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
40
41struct iommu_callback_data {
42 const struct iommu_ops *ops;
43};
44
45struct iommu_group {
46 struct kobject kobj;
47 struct kobject *devices_kobj;
48 struct list_head devices;
49 struct mutex mutex;
50 struct blocking_notifier_head notifier;
51 void *iommu_data;
52 void (*iommu_data_release)(void *iommu_data);
53 char *name;
54 int id;
55 struct iommu_domain *default_domain;
56 struct iommu_domain *domain;
57};
58
59struct group_device {
60 struct list_head list;
61 struct device *dev;
62 char *name;
63};
64
65struct iommu_group_attribute {
66 struct attribute attr;
67 ssize_t (*show)(struct iommu_group *group, char *buf);
68 ssize_t (*store)(struct iommu_group *group,
69 const char *buf, size_t count);
70};
71
72static const char * const iommu_group_resv_type_string[] = {
73 [IOMMU_RESV_DIRECT] = "direct",
74 [IOMMU_RESV_RESERVED] = "reserved",
75 [IOMMU_RESV_MSI] = "msi",
76 [IOMMU_RESV_SW_MSI] = "msi",
77};
78
79#define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
80struct iommu_group_attribute iommu_group_attr_##_name = \
81 __ATTR(_name, _mode, _show, _store)
82
83#define to_iommu_group_attr(_attr) \
84 container_of(_attr, struct iommu_group_attribute, attr)
85#define to_iommu_group(_kobj) \
86 container_of(_kobj, struct iommu_group, kobj)
87
88static LIST_HEAD(iommu_device_list);
89static DEFINE_SPINLOCK(iommu_device_lock);
90
91int iommu_device_register(struct iommu_device *iommu)
92{
93 spin_lock(&iommu_device_lock);
94 list_add_tail(&iommu->list, &iommu_device_list);
95 spin_unlock(&iommu_device_lock);
96
97 return 0;
98}
99
100void iommu_device_unregister(struct iommu_device *iommu)
101{
102 spin_lock(&iommu_device_lock);
103 list_del(&iommu->list);
104 spin_unlock(&iommu_device_lock);
105}
106
107static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
108 unsigned type);
109static int __iommu_attach_device(struct iommu_domain *domain,
110 struct device *dev);
111static int __iommu_attach_group(struct iommu_domain *domain,
112 struct iommu_group *group);
113static void __iommu_detach_group(struct iommu_domain *domain,
114 struct iommu_group *group);
115
116static int __init iommu_set_def_domain_type(char *str)
117{
118 bool pt;
119
120 if (!str || strtobool(str, &pt))
121 return -EINVAL;
122
123 iommu_def_domain_type = pt ? IOMMU_DOMAIN_IDENTITY : IOMMU_DOMAIN_DMA;
124 return 0;
125}
126early_param("iommu.passthrough", iommu_set_def_domain_type);
127
128static ssize_t iommu_group_attr_show(struct kobject *kobj,
129 struct attribute *__attr, char *buf)
130{
131 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
132 struct iommu_group *group = to_iommu_group(kobj);
133 ssize_t ret = -EIO;
134
135 if (attr->show)
136 ret = attr->show(group, buf);
137 return ret;
138}
139
140static ssize_t iommu_group_attr_store(struct kobject *kobj,
141 struct attribute *__attr,
142 const char *buf, size_t count)
143{
144 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
145 struct iommu_group *group = to_iommu_group(kobj);
146 ssize_t ret = -EIO;
147
148 if (attr->store)
149 ret = attr->store(group, buf, count);
150 return ret;
151}
152
153static const struct sysfs_ops iommu_group_sysfs_ops = {
154 .show = iommu_group_attr_show,
155 .store = iommu_group_attr_store,
156};
157
158static int iommu_group_create_file(struct iommu_group *group,
159 struct iommu_group_attribute *attr)
160{
161 return sysfs_create_file(&group->kobj, &attr->attr);
162}
163
164static void iommu_group_remove_file(struct iommu_group *group,
165 struct iommu_group_attribute *attr)
166{
167 sysfs_remove_file(&group->kobj, &attr->attr);
168}
169
170static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
171{
172 return sprintf(buf, "%s\n", group->name);
173}
174
175/**
176 * iommu_insert_resv_region - Insert a new region in the
177 * list of reserved regions.
178 * @new: new region to insert
179 * @regions: list of regions
180 *
181 * The new element is sorted by address with respect to the other
182 * regions of the same type. In case it overlaps with another
183 * region of the same type, regions are merged. In case it
184 * overlaps with another region of different type, regions are
185 * not merged.
186 */
187static int iommu_insert_resv_region(struct iommu_resv_region *new,
188 struct list_head *regions)
189{
190 struct iommu_resv_region *region;
191 phys_addr_t start = new->start;
192 phys_addr_t end = new->start + new->length - 1;
193 struct list_head *pos = regions->next;
194
195 while (pos != regions) {
196 struct iommu_resv_region *entry =
197 list_entry(pos, struct iommu_resv_region, list);
198 phys_addr_t a = entry->start;
199 phys_addr_t b = entry->start + entry->length - 1;
200 int type = entry->type;
201
202 if (end < a) {
203 goto insert;
204 } else if (start > b) {
205 pos = pos->next;
206 } else if ((start >= a) && (end <= b)) {
207 if (new->type == type)
208 goto done;
209 else
210 pos = pos->next;
211 } else {
212 if (new->type == type) {
213 phys_addr_t new_start = min(a, start);
214 phys_addr_t new_end = max(b, end);
215
216 list_del(&entry->list);
217 entry->start = new_start;
218 entry->length = new_end - new_start + 1;
219 iommu_insert_resv_region(entry, regions);
220 } else {
221 pos = pos->next;
222 }
223 }
224 }
225insert:
226 region = iommu_alloc_resv_region(new->start, new->length,
227 new->prot, new->type);
228 if (!region)
229 return -ENOMEM;
230
231 list_add_tail(®ion->list, pos);
232done:
233 return 0;
234}
235
236static int
237iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
238 struct list_head *group_resv_regions)
239{
240 struct iommu_resv_region *entry;
241 int ret = 0;
242
243 list_for_each_entry(entry, dev_resv_regions, list) {
244 ret = iommu_insert_resv_region(entry, group_resv_regions);
245 if (ret)
246 break;
247 }
248 return ret;
249}
250
251int iommu_get_group_resv_regions(struct iommu_group *group,
252 struct list_head *head)
253{
254 struct group_device *device;
255 int ret = 0;
256
257 mutex_lock(&group->mutex);
258 list_for_each_entry(device, &group->devices, list) {
259 struct list_head dev_resv_regions;
260
261 INIT_LIST_HEAD(&dev_resv_regions);
262 iommu_get_resv_regions(device->dev, &dev_resv_regions);
263 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
264 iommu_put_resv_regions(device->dev, &dev_resv_regions);
265 if (ret)
266 break;
267 }
268 mutex_unlock(&group->mutex);
269 return ret;
270}
271EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
272
273static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
274 char *buf)
275{
276 struct iommu_resv_region *region, *next;
277 struct list_head group_resv_regions;
278 char *str = buf;
279
280 INIT_LIST_HEAD(&group_resv_regions);
281 iommu_get_group_resv_regions(group, &group_resv_regions);
282
283 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
284 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
285 (long long int)region->start,
286 (long long int)(region->start +
287 region->length - 1),
288 iommu_group_resv_type_string[region->type]);
289 kfree(region);
290 }
291
292 return (str - buf);
293}
294
295static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
296
297static IOMMU_GROUP_ATTR(reserved_regions, 0444,
298 iommu_group_show_resv_regions, NULL);
299
300static void iommu_group_release(struct kobject *kobj)
301{
302 struct iommu_group *group = to_iommu_group(kobj);
303
304 pr_debug("Releasing group %d\n", group->id);
305
306 if (group->iommu_data_release)
307 group->iommu_data_release(group->iommu_data);
308
309 ida_simple_remove(&iommu_group_ida, group->id);
310
311 if (group->default_domain)
312 iommu_domain_free(group->default_domain);
313
314 kfree(group->name);
315 kfree(group);
316}
317
318static struct kobj_type iommu_group_ktype = {
319 .sysfs_ops = &iommu_group_sysfs_ops,
320 .release = iommu_group_release,
321};
322
323/**
324 * iommu_group_alloc - Allocate a new group
325 * @name: Optional name to associate with group, visible in sysfs
326 *
327 * This function is called by an iommu driver to allocate a new iommu
328 * group. The iommu group represents the minimum granularity of the iommu.
329 * Upon successful return, the caller holds a reference to the supplied
330 * group in order to hold the group until devices are added. Use
331 * iommu_group_put() to release this extra reference count, allowing the
332 * group to be automatically reclaimed once it has no devices or external
333 * references.
334 */
335struct iommu_group *iommu_group_alloc(void)
336{
337 struct iommu_group *group;
338 int ret;
339
340 group = kzalloc(sizeof(*group), GFP_KERNEL);
341 if (!group)
342 return ERR_PTR(-ENOMEM);
343
344 group->kobj.kset = iommu_group_kset;
345 mutex_init(&group->mutex);
346 INIT_LIST_HEAD(&group->devices);
347 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
348
349 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
350 if (ret < 0) {
351 kfree(group);
352 return ERR_PTR(ret);
353 }
354 group->id = ret;
355
356 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
357 NULL, "%d", group->id);
358 if (ret) {
359 ida_simple_remove(&iommu_group_ida, group->id);
360 kfree(group);
361 return ERR_PTR(ret);
362 }
363
364 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
365 if (!group->devices_kobj) {
366 kobject_put(&group->kobj); /* triggers .release & free */
367 return ERR_PTR(-ENOMEM);
368 }
369
370 /*
371 * The devices_kobj holds a reference on the group kobject, so
372 * as long as that exists so will the group. We can therefore
373 * use the devices_kobj for reference counting.
374 */
375 kobject_put(&group->kobj);
376
377 ret = iommu_group_create_file(group,
378 &iommu_group_attr_reserved_regions);
379 if (ret)
380 return ERR_PTR(ret);
381
382 pr_debug("Allocated group %d\n", group->id);
383
384 return group;
385}
386EXPORT_SYMBOL_GPL(iommu_group_alloc);
387
388struct iommu_group *iommu_group_get_by_id(int id)
389{
390 struct kobject *group_kobj;
391 struct iommu_group *group;
392 const char *name;
393
394 if (!iommu_group_kset)
395 return NULL;
396
397 name = kasprintf(GFP_KERNEL, "%d", id);
398 if (!name)
399 return NULL;
400
401 group_kobj = kset_find_obj(iommu_group_kset, name);
402 kfree(name);
403
404 if (!group_kobj)
405 return NULL;
406
407 group = container_of(group_kobj, struct iommu_group, kobj);
408 BUG_ON(group->id != id);
409
410 kobject_get(group->devices_kobj);
411 kobject_put(&group->kobj);
412
413 return group;
414}
415EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
416
417/**
418 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
419 * @group: the group
420 *
421 * iommu drivers can store data in the group for use when doing iommu
422 * operations. This function provides a way to retrieve it. Caller
423 * should hold a group reference.
424 */
425void *iommu_group_get_iommudata(struct iommu_group *group)
426{
427 return group->iommu_data;
428}
429EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
430
431/**
432 * iommu_group_set_iommudata - set iommu_data for a group
433 * @group: the group
434 * @iommu_data: new data
435 * @release: release function for iommu_data
436 *
437 * iommu drivers can store data in the group for use when doing iommu
438 * operations. This function provides a way to set the data after
439 * the group has been allocated. Caller should hold a group reference.
440 */
441void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
442 void (*release)(void *iommu_data))
443{
444 group->iommu_data = iommu_data;
445 group->iommu_data_release = release;
446}
447EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
448
449/**
450 * iommu_group_set_name - set name for a group
451 * @group: the group
452 * @name: name
453 *
454 * Allow iommu driver to set a name for a group. When set it will
455 * appear in a name attribute file under the group in sysfs.
456 */
457int iommu_group_set_name(struct iommu_group *group, const char *name)
458{
459 int ret;
460
461 if (group->name) {
462 iommu_group_remove_file(group, &iommu_group_attr_name);
463 kfree(group->name);
464 group->name = NULL;
465 if (!name)
466 return 0;
467 }
468
469 group->name = kstrdup(name, GFP_KERNEL);
470 if (!group->name)
471 return -ENOMEM;
472
473 ret = iommu_group_create_file(group, &iommu_group_attr_name);
474 if (ret) {
475 kfree(group->name);
476 group->name = NULL;
477 return ret;
478 }
479
480 return 0;
481}
482EXPORT_SYMBOL_GPL(iommu_group_set_name);
483
484static int iommu_group_create_direct_mappings(struct iommu_group *group,
485 struct device *dev)
486{
487 struct iommu_domain *domain = group->default_domain;
488 struct iommu_resv_region *entry;
489 struct list_head mappings;
490 unsigned long pg_size;
491 int ret = 0;
492
493 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
494 return 0;
495
496 BUG_ON(!domain->pgsize_bitmap);
497
498 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
499 INIT_LIST_HEAD(&mappings);
500
501 iommu_get_resv_regions(dev, &mappings);
502
503 /* We need to consider overlapping regions for different devices */
504 list_for_each_entry(entry, &mappings, list) {
505 dma_addr_t start, end, addr;
506
507 if (domain->ops->apply_resv_region)
508 domain->ops->apply_resv_region(dev, domain, entry);
509
510 start = ALIGN(entry->start, pg_size);
511 end = ALIGN(entry->start + entry->length, pg_size);
512
513 if (entry->type != IOMMU_RESV_DIRECT)
514 continue;
515
516 for (addr = start; addr < end; addr += pg_size) {
517 phys_addr_t phys_addr;
518
519 phys_addr = iommu_iova_to_phys(domain, addr);
520 if (phys_addr)
521 continue;
522
523 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
524 if (ret)
525 goto out;
526 }
527
528 }
529
530 iommu_flush_tlb_all(domain);
531
532out:
533 iommu_put_resv_regions(dev, &mappings);
534
535 return ret;
536}
537
538/**
539 * iommu_group_add_device - add a device to an iommu group
540 * @group: the group into which to add the device (reference should be held)
541 * @dev: the device
542 *
543 * This function is called by an iommu driver to add a device into a
544 * group. Adding a device increments the group reference count.
545 */
546int iommu_group_add_device(struct iommu_group *group, struct device *dev)
547{
548 int ret, i = 0;
549 struct group_device *device;
550
551 device = kzalloc(sizeof(*device), GFP_KERNEL);
552 if (!device)
553 return -ENOMEM;
554
555 device->dev = dev;
556
557 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
558 if (ret)
559 goto err_free_device;
560
561 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
562rename:
563 if (!device->name) {
564 ret = -ENOMEM;
565 goto err_remove_link;
566 }
567
568 ret = sysfs_create_link_nowarn(group->devices_kobj,
569 &dev->kobj, device->name);
570 if (ret) {
571 if (ret == -EEXIST && i >= 0) {
572 /*
573 * Account for the slim chance of collision
574 * and append an instance to the name.
575 */
576 kfree(device->name);
577 device->name = kasprintf(GFP_KERNEL, "%s.%d",
578 kobject_name(&dev->kobj), i++);
579 goto rename;
580 }
581 goto err_free_name;
582 }
583
584 kobject_get(group->devices_kobj);
585
586 dev->iommu_group = group;
587
588 iommu_group_create_direct_mappings(group, dev);
589
590 mutex_lock(&group->mutex);
591 list_add_tail(&device->list, &group->devices);
592 if (group->domain)
593 ret = __iommu_attach_device(group->domain, dev);
594 mutex_unlock(&group->mutex);
595 if (ret)
596 goto err_put_group;
597
598 /* Notify any listeners about change to group. */
599 blocking_notifier_call_chain(&group->notifier,
600 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
601
602 trace_add_device_to_group(group->id, dev);
603
604 pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
605
606 return 0;
607
608err_put_group:
609 mutex_lock(&group->mutex);
610 list_del(&device->list);
611 mutex_unlock(&group->mutex);
612 dev->iommu_group = NULL;
613 kobject_put(group->devices_kobj);
614err_free_name:
615 kfree(device->name);
616err_remove_link:
617 sysfs_remove_link(&dev->kobj, "iommu_group");
618err_free_device:
619 kfree(device);
620 pr_err("Failed to add device %s to group %d: %d\n", dev_name(dev), group->id, ret);
621 return ret;
622}
623EXPORT_SYMBOL_GPL(iommu_group_add_device);
624
625/**
626 * iommu_group_remove_device - remove a device from it's current group
627 * @dev: device to be removed
628 *
629 * This function is called by an iommu driver to remove the device from
630 * it's current group. This decrements the iommu group reference count.
631 */
632void iommu_group_remove_device(struct device *dev)
633{
634 struct iommu_group *group = dev->iommu_group;
635 struct group_device *tmp_device, *device = NULL;
636
637 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
638
639 /* Pre-notify listeners that a device is being removed. */
640 blocking_notifier_call_chain(&group->notifier,
641 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
642
643 mutex_lock(&group->mutex);
644 list_for_each_entry(tmp_device, &group->devices, list) {
645 if (tmp_device->dev == dev) {
646 device = tmp_device;
647 list_del(&device->list);
648 break;
649 }
650 }
651 mutex_unlock(&group->mutex);
652
653 if (!device)
654 return;
655
656 sysfs_remove_link(group->devices_kobj, device->name);
657 sysfs_remove_link(&dev->kobj, "iommu_group");
658
659 trace_remove_device_from_group(group->id, dev);
660
661 kfree(device->name);
662 kfree(device);
663 dev->iommu_group = NULL;
664 kobject_put(group->devices_kobj);
665}
666EXPORT_SYMBOL_GPL(iommu_group_remove_device);
667
668static int iommu_group_device_count(struct iommu_group *group)
669{
670 struct group_device *entry;
671 int ret = 0;
672
673 list_for_each_entry(entry, &group->devices, list)
674 ret++;
675
676 return ret;
677}
678
679/**
680 * iommu_group_for_each_dev - iterate over each device in the group
681 * @group: the group
682 * @data: caller opaque data to be passed to callback function
683 * @fn: caller supplied callback function
684 *
685 * This function is called by group users to iterate over group devices.
686 * Callers should hold a reference count to the group during callback.
687 * The group->mutex is held across callbacks, which will block calls to
688 * iommu_group_add/remove_device.
689 */
690static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
691 int (*fn)(struct device *, void *))
692{
693 struct group_device *device;
694 int ret = 0;
695
696 list_for_each_entry(device, &group->devices, list) {
697 ret = fn(device->dev, data);
698 if (ret)
699 break;
700 }
701 return ret;
702}
703
704
705int iommu_group_for_each_dev(struct iommu_group *group, void *data,
706 int (*fn)(struct device *, void *))
707{
708 int ret;
709
710 mutex_lock(&group->mutex);
711 ret = __iommu_group_for_each_dev(group, data, fn);
712 mutex_unlock(&group->mutex);
713
714 return ret;
715}
716EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
717
718/**
719 * iommu_group_get - Return the group for a device and increment reference
720 * @dev: get the group that this device belongs to
721 *
722 * This function is called by iommu drivers and users to get the group
723 * for the specified device. If found, the group is returned and the group
724 * reference in incremented, else NULL.
725 */
726struct iommu_group *iommu_group_get(struct device *dev)
727{
728 struct iommu_group *group = dev->iommu_group;
729
730 if (group)
731 kobject_get(group->devices_kobj);
732
733 return group;
734}
735EXPORT_SYMBOL_GPL(iommu_group_get);
736
737/**
738 * iommu_group_ref_get - Increment reference on a group
739 * @group: the group to use, must not be NULL
740 *
741 * This function is called by iommu drivers to take additional references on an
742 * existing group. Returns the given group for convenience.
743 */
744struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
745{
746 kobject_get(group->devices_kobj);
747 return group;
748}
749
750/**
751 * iommu_group_put - Decrement group reference
752 * @group: the group to use
753 *
754 * This function is called by iommu drivers and users to release the
755 * iommu group. Once the reference count is zero, the group is released.
756 */
757void iommu_group_put(struct iommu_group *group)
758{
759 if (group)
760 kobject_put(group->devices_kobj);
761}
762EXPORT_SYMBOL_GPL(iommu_group_put);
763
764/**
765 * iommu_group_register_notifier - Register a notifier for group changes
766 * @group: the group to watch
767 * @nb: notifier block to signal
768 *
769 * This function allows iommu group users to track changes in a group.
770 * See include/linux/iommu.h for actions sent via this notifier. Caller
771 * should hold a reference to the group throughout notifier registration.
772 */
773int iommu_group_register_notifier(struct iommu_group *group,
774 struct notifier_block *nb)
775{
776 return blocking_notifier_chain_register(&group->notifier, nb);
777}
778EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
779
780/**
781 * iommu_group_unregister_notifier - Unregister a notifier
782 * @group: the group to watch
783 * @nb: notifier block to signal
784 *
785 * Unregister a previously registered group notifier block.
786 */
787int iommu_group_unregister_notifier(struct iommu_group *group,
788 struct notifier_block *nb)
789{
790 return blocking_notifier_chain_unregister(&group->notifier, nb);
791}
792EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
793
794/**
795 * iommu_group_id - Return ID for a group
796 * @group: the group to ID
797 *
798 * Return the unique ID for the group matching the sysfs group number.
799 */
800int iommu_group_id(struct iommu_group *group)
801{
802 return group->id;
803}
804EXPORT_SYMBOL_GPL(iommu_group_id);
805
806static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
807 unsigned long *devfns);
808
809/*
810 * To consider a PCI device isolated, we require ACS to support Source
811 * Validation, Request Redirection, Completer Redirection, and Upstream
812 * Forwarding. This effectively means that devices cannot spoof their
813 * requester ID, requests and completions cannot be redirected, and all
814 * transactions are forwarded upstream, even as it passes through a
815 * bridge where the target device is downstream.
816 */
817#define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
818
819/*
820 * For multifunction devices which are not isolated from each other, find
821 * all the other non-isolated functions and look for existing groups. For
822 * each function, we also need to look for aliases to or from other devices
823 * that may already have a group.
824 */
825static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
826 unsigned long *devfns)
827{
828 struct pci_dev *tmp = NULL;
829 struct iommu_group *group;
830
831 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
832 return NULL;
833
834 for_each_pci_dev(tmp) {
835 if (tmp == pdev || tmp->bus != pdev->bus ||
836 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
837 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
838 continue;
839
840 group = get_pci_alias_group(tmp, devfns);
841 if (group) {
842 pci_dev_put(tmp);
843 return group;
844 }
845 }
846
847 return NULL;
848}
849
850/*
851 * Look for aliases to or from the given device for existing groups. DMA
852 * aliases are only supported on the same bus, therefore the search
853 * space is quite small (especially since we're really only looking at pcie
854 * device, and therefore only expect multiple slots on the root complex or
855 * downstream switch ports). It's conceivable though that a pair of
856 * multifunction devices could have aliases between them that would cause a
857 * loop. To prevent this, we use a bitmap to track where we've been.
858 */
859static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
860 unsigned long *devfns)
861{
862 struct pci_dev *tmp = NULL;
863 struct iommu_group *group;
864
865 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
866 return NULL;
867
868 group = iommu_group_get(&pdev->dev);
869 if (group)
870 return group;
871
872 for_each_pci_dev(tmp) {
873 if (tmp == pdev || tmp->bus != pdev->bus)
874 continue;
875
876 /* We alias them or they alias us */
877 if (pci_devs_are_dma_aliases(pdev, tmp)) {
878 group = get_pci_alias_group(tmp, devfns);
879 if (group) {
880 pci_dev_put(tmp);
881 return group;
882 }
883
884 group = get_pci_function_alias_group(tmp, devfns);
885 if (group) {
886 pci_dev_put(tmp);
887 return group;
888 }
889 }
890 }
891
892 return NULL;
893}
894
895struct group_for_pci_data {
896 struct pci_dev *pdev;
897 struct iommu_group *group;
898};
899
900/*
901 * DMA alias iterator callback, return the last seen device. Stop and return
902 * the IOMMU group if we find one along the way.
903 */
904static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
905{
906 struct group_for_pci_data *data = opaque;
907
908 data->pdev = pdev;
909 data->group = iommu_group_get(&pdev->dev);
910
911 return data->group != NULL;
912}
913
914/*
915 * Generic device_group call-back function. It just allocates one
916 * iommu-group per device.
917 */
918struct iommu_group *generic_device_group(struct device *dev)
919{
920 return iommu_group_alloc();
921}
922
923/*
924 * Use standard PCI bus topology, isolation features, and DMA alias quirks
925 * to find or create an IOMMU group for a device.
926 */
927struct iommu_group *pci_device_group(struct device *dev)
928{
929 struct pci_dev *pdev = to_pci_dev(dev);
930 struct group_for_pci_data data;
931 struct pci_bus *bus;
932 struct iommu_group *group = NULL;
933 u64 devfns[4] = { 0 };
934
935 if (WARN_ON(!dev_is_pci(dev)))
936 return ERR_PTR(-EINVAL);
937
938 /*
939 * Find the upstream DMA alias for the device. A device must not
940 * be aliased due to topology in order to have its own IOMMU group.
941 * If we find an alias along the way that already belongs to a
942 * group, use it.
943 */
944 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
945 return data.group;
946
947 pdev = data.pdev;
948
949 /*
950 * Continue upstream from the point of minimum IOMMU granularity
951 * due to aliases to the point where devices are protected from
952 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
953 * group, use it.
954 */
955 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
956 if (!bus->self)
957 continue;
958
959 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
960 break;
961
962 pdev = bus->self;
963
964 group = iommu_group_get(&pdev->dev);
965 if (group)
966 return group;
967 }
968
969 /*
970 * Look for existing groups on device aliases. If we alias another
971 * device or another device aliases us, use the same group.
972 */
973 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
974 if (group)
975 return group;
976
977 /*
978 * Look for existing groups on non-isolated functions on the same
979 * slot and aliases of those funcions, if any. No need to clear
980 * the search bitmap, the tested devfns are still valid.
981 */
982 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
983 if (group)
984 return group;
985
986 /* No shared group found, allocate new */
987 return iommu_group_alloc();
988}
989
990/**
991 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
992 * @dev: target device
993 *
994 * This function is intended to be called by IOMMU drivers and extended to
995 * support common, bus-defined algorithms when determining or creating the
996 * IOMMU group for a device. On success, the caller will hold a reference
997 * to the returned IOMMU group, which will already include the provided
998 * device. The reference should be released with iommu_group_put().
999 */
1000struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1001{
1002 const struct iommu_ops *ops = dev->bus->iommu_ops;
1003 struct iommu_group *group;
1004 int ret;
1005
1006 group = iommu_group_get(dev);
1007 if (group)
1008 return group;
1009
1010 if (!ops)
1011 return ERR_PTR(-EINVAL);
1012
1013 group = ops->device_group(dev);
1014 if (WARN_ON_ONCE(group == NULL))
1015 return ERR_PTR(-EINVAL);
1016
1017 if (IS_ERR(group))
1018 return group;
1019
1020 /*
1021 * Try to allocate a default domain - needs support from the
1022 * IOMMU driver.
1023 */
1024 if (!group->default_domain) {
1025 struct iommu_domain *dom;
1026
1027 dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1028 if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
1029 dev_warn(dev,
1030 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1031 iommu_def_domain_type);
1032 dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
1033 }
1034
1035 group->default_domain = dom;
1036 if (!group->domain)
1037 group->domain = dom;
1038 }
1039
1040 ret = iommu_group_add_device(group, dev);
1041 if (ret) {
1042 iommu_group_put(group);
1043 return ERR_PTR(ret);
1044 }
1045
1046 return group;
1047}
1048
1049struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1050{
1051 return group->default_domain;
1052}
1053
1054static int add_iommu_group(struct device *dev, void *data)
1055{
1056 struct iommu_callback_data *cb = data;
1057 const struct iommu_ops *ops = cb->ops;
1058 int ret;
1059
1060 if (!ops->add_device)
1061 return 0;
1062
1063 WARN_ON(dev->iommu_group);
1064
1065 ret = ops->add_device(dev);
1066
1067 /*
1068 * We ignore -ENODEV errors for now, as they just mean that the
1069 * device is not translated by an IOMMU. We still care about
1070 * other errors and fail to initialize when they happen.
1071 */
1072 if (ret == -ENODEV)
1073 ret = 0;
1074
1075 return ret;
1076}
1077
1078static int remove_iommu_group(struct device *dev, void *data)
1079{
1080 struct iommu_callback_data *cb = data;
1081 const struct iommu_ops *ops = cb->ops;
1082
1083 if (ops->remove_device && dev->iommu_group)
1084 ops->remove_device(dev);
1085
1086 return 0;
1087}
1088
1089static int iommu_bus_notifier(struct notifier_block *nb,
1090 unsigned long action, void *data)
1091{
1092 struct device *dev = data;
1093 const struct iommu_ops *ops = dev->bus->iommu_ops;
1094 struct iommu_group *group;
1095 unsigned long group_action = 0;
1096
1097 /*
1098 * ADD/DEL call into iommu driver ops if provided, which may
1099 * result in ADD/DEL notifiers to group->notifier
1100 */
1101 if (action == BUS_NOTIFY_ADD_DEVICE) {
1102 if (ops->add_device) {
1103 int ret;
1104
1105 ret = ops->add_device(dev);
1106 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
1107 }
1108 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
1109 if (ops->remove_device && dev->iommu_group) {
1110 ops->remove_device(dev);
1111 return 0;
1112 }
1113 }
1114
1115 /*
1116 * Remaining BUS_NOTIFYs get filtered and republished to the
1117 * group, if anyone is listening
1118 */
1119 group = iommu_group_get(dev);
1120 if (!group)
1121 return 0;
1122
1123 switch (action) {
1124 case BUS_NOTIFY_BIND_DRIVER:
1125 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1126 break;
1127 case BUS_NOTIFY_BOUND_DRIVER:
1128 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1129 break;
1130 case BUS_NOTIFY_UNBIND_DRIVER:
1131 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1132 break;
1133 case BUS_NOTIFY_UNBOUND_DRIVER:
1134 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1135 break;
1136 }
1137
1138 if (group_action)
1139 blocking_notifier_call_chain(&group->notifier,
1140 group_action, dev);
1141
1142 iommu_group_put(group);
1143 return 0;
1144}
1145
1146static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
1147{
1148 int err;
1149 struct notifier_block *nb;
1150 struct iommu_callback_data cb = {
1151 .ops = ops,
1152 };
1153
1154 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1155 if (!nb)
1156 return -ENOMEM;
1157
1158 nb->notifier_call = iommu_bus_notifier;
1159
1160 err = bus_register_notifier(bus, nb);
1161 if (err)
1162 goto out_free;
1163
1164 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
1165 if (err)
1166 goto out_err;
1167
1168
1169 return 0;
1170
1171out_err:
1172 /* Clean up */
1173 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
1174 bus_unregister_notifier(bus, nb);
1175
1176out_free:
1177 kfree(nb);
1178
1179 return err;
1180}
1181
1182/**
1183 * bus_set_iommu - set iommu-callbacks for the bus
1184 * @bus: bus.
1185 * @ops: the callbacks provided by the iommu-driver
1186 *
1187 * This function is called by an iommu driver to set the iommu methods
1188 * used for a particular bus. Drivers for devices on that bus can use
1189 * the iommu-api after these ops are registered.
1190 * This special function is needed because IOMMUs are usually devices on
1191 * the bus itself, so the iommu drivers are not initialized when the bus
1192 * is set up. With this function the iommu-driver can set the iommu-ops
1193 * afterwards.
1194 */
1195int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1196{
1197 int err;
1198
1199 if (bus->iommu_ops != NULL)
1200 return -EBUSY;
1201
1202 bus->iommu_ops = ops;
1203
1204 /* Do IOMMU specific setup for this bus-type */
1205 err = iommu_bus_init(bus, ops);
1206 if (err)
1207 bus->iommu_ops = NULL;
1208
1209 return err;
1210}
1211EXPORT_SYMBOL_GPL(bus_set_iommu);
1212
1213bool iommu_present(struct bus_type *bus)
1214{
1215 return bus->iommu_ops != NULL;
1216}
1217EXPORT_SYMBOL_GPL(iommu_present);
1218
1219bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1220{
1221 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1222 return false;
1223
1224 return bus->iommu_ops->capable(cap);
1225}
1226EXPORT_SYMBOL_GPL(iommu_capable);
1227
1228/**
1229 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1230 * @domain: iommu domain
1231 * @handler: fault handler
1232 * @token: user data, will be passed back to the fault handler
1233 *
1234 * This function should be used by IOMMU users which want to be notified
1235 * whenever an IOMMU fault happens.
1236 *
1237 * The fault handler itself should return 0 on success, and an appropriate
1238 * error code otherwise.
1239 */
1240void iommu_set_fault_handler(struct iommu_domain *domain,
1241 iommu_fault_handler_t handler,
1242 void *token)
1243{
1244 BUG_ON(!domain);
1245
1246 domain->handler = handler;
1247 domain->handler_token = token;
1248}
1249EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1250
1251static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1252 unsigned type)
1253{
1254 struct iommu_domain *domain;
1255
1256 if (bus == NULL || bus->iommu_ops == NULL)
1257 return NULL;
1258
1259 domain = bus->iommu_ops->domain_alloc(type);
1260 if (!domain)
1261 return NULL;
1262
1263 domain->ops = bus->iommu_ops;
1264 domain->type = type;
1265 /* Assume all sizes by default; the driver may override this later */
1266 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
1267
1268 return domain;
1269}
1270
1271struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1272{
1273 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1274}
1275EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1276
1277void iommu_domain_free(struct iommu_domain *domain)
1278{
1279 domain->ops->domain_free(domain);
1280}
1281EXPORT_SYMBOL_GPL(iommu_domain_free);
1282
1283static int __iommu_attach_device(struct iommu_domain *domain,
1284 struct device *dev)
1285{
1286 int ret;
1287 if ((domain->ops->is_attach_deferred != NULL) &&
1288 domain->ops->is_attach_deferred(domain, dev))
1289 return 0;
1290
1291 if (unlikely(domain->ops->attach_dev == NULL))
1292 return -ENODEV;
1293
1294 ret = domain->ops->attach_dev(domain, dev);
1295 if (!ret)
1296 trace_attach_device_to_domain(dev);
1297 return ret;
1298}
1299
1300int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1301{
1302 struct iommu_group *group;
1303 int ret;
1304
1305 group = iommu_group_get(dev);
1306 if (!group)
1307 return -ENODEV;
1308
1309 /*
1310 * Lock the group to make sure the device-count doesn't
1311 * change while we are attaching
1312 */
1313 mutex_lock(&group->mutex);
1314 ret = -EINVAL;
1315 if (iommu_group_device_count(group) != 1)
1316 goto out_unlock;
1317
1318 ret = __iommu_attach_group(domain, group);
1319
1320out_unlock:
1321 mutex_unlock(&group->mutex);
1322 iommu_group_put(group);
1323
1324 return ret;
1325}
1326EXPORT_SYMBOL_GPL(iommu_attach_device);
1327
1328static void __iommu_detach_device(struct iommu_domain *domain,
1329 struct device *dev)
1330{
1331 if ((domain->ops->is_attach_deferred != NULL) &&
1332 domain->ops->is_attach_deferred(domain, dev))
1333 return;
1334
1335 if (unlikely(domain->ops->detach_dev == NULL))
1336 return;
1337
1338 domain->ops->detach_dev(domain, dev);
1339 trace_detach_device_from_domain(dev);
1340}
1341
1342void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1343{
1344 struct iommu_group *group;
1345
1346 group = iommu_group_get(dev);
1347 if (!group)
1348 return;
1349
1350 mutex_lock(&group->mutex);
1351 if (iommu_group_device_count(group) != 1) {
1352 WARN_ON(1);
1353 goto out_unlock;
1354 }
1355
1356 __iommu_detach_group(domain, group);
1357
1358out_unlock:
1359 mutex_unlock(&group->mutex);
1360 iommu_group_put(group);
1361}
1362EXPORT_SYMBOL_GPL(iommu_detach_device);
1363
1364struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1365{
1366 struct iommu_domain *domain;
1367 struct iommu_group *group;
1368
1369 group = iommu_group_get(dev);
1370 if (!group)
1371 return NULL;
1372
1373 domain = group->domain;
1374
1375 iommu_group_put(group);
1376
1377 return domain;
1378}
1379EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1380
1381/*
1382 * IOMMU groups are really the natrual working unit of the IOMMU, but
1383 * the IOMMU API works on domains and devices. Bridge that gap by
1384 * iterating over the devices in a group. Ideally we'd have a single
1385 * device which represents the requestor ID of the group, but we also
1386 * allow IOMMU drivers to create policy defined minimum sets, where
1387 * the physical hardware may be able to distiguish members, but we
1388 * wish to group them at a higher level (ex. untrusted multi-function
1389 * PCI devices). Thus we attach each device.
1390 */
1391static int iommu_group_do_attach_device(struct device *dev, void *data)
1392{
1393 struct iommu_domain *domain = data;
1394
1395 return __iommu_attach_device(domain, dev);
1396}
1397
1398static int __iommu_attach_group(struct iommu_domain *domain,
1399 struct iommu_group *group)
1400{
1401 int ret;
1402
1403 if (group->default_domain && group->domain != group->default_domain)
1404 return -EBUSY;
1405
1406 ret = __iommu_group_for_each_dev(group, domain,
1407 iommu_group_do_attach_device);
1408 if (ret == 0)
1409 group->domain = domain;
1410
1411 return ret;
1412}
1413
1414int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1415{
1416 int ret;
1417
1418 mutex_lock(&group->mutex);
1419 ret = __iommu_attach_group(domain, group);
1420 mutex_unlock(&group->mutex);
1421
1422 return ret;
1423}
1424EXPORT_SYMBOL_GPL(iommu_attach_group);
1425
1426static int iommu_group_do_detach_device(struct device *dev, void *data)
1427{
1428 struct iommu_domain *domain = data;
1429
1430 __iommu_detach_device(domain, dev);
1431
1432 return 0;
1433}
1434
1435static void __iommu_detach_group(struct iommu_domain *domain,
1436 struct iommu_group *group)
1437{
1438 int ret;
1439
1440 if (!group->default_domain) {
1441 __iommu_group_for_each_dev(group, domain,
1442 iommu_group_do_detach_device);
1443 group->domain = NULL;
1444 return;
1445 }
1446
1447 if (group->domain == group->default_domain)
1448 return;
1449
1450 /* Detach by re-attaching to the default domain */
1451 ret = __iommu_group_for_each_dev(group, group->default_domain,
1452 iommu_group_do_attach_device);
1453 if (ret != 0)
1454 WARN_ON(1);
1455 else
1456 group->domain = group->default_domain;
1457}
1458
1459void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1460{
1461 mutex_lock(&group->mutex);
1462 __iommu_detach_group(domain, group);
1463 mutex_unlock(&group->mutex);
1464}
1465EXPORT_SYMBOL_GPL(iommu_detach_group);
1466
1467phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1468{
1469 if (unlikely(domain->ops->iova_to_phys == NULL))
1470 return 0;
1471
1472 return domain->ops->iova_to_phys(domain, iova);
1473}
1474EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1475
1476static size_t iommu_pgsize(struct iommu_domain *domain,
1477 unsigned long addr_merge, size_t size)
1478{
1479 unsigned int pgsize_idx;
1480 size_t pgsize;
1481
1482 /* Max page size that still fits into 'size' */
1483 pgsize_idx = __fls(size);
1484
1485 /* need to consider alignment requirements ? */
1486 if (likely(addr_merge)) {
1487 /* Max page size allowed by address */
1488 unsigned int align_pgsize_idx = __ffs(addr_merge);
1489 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1490 }
1491
1492 /* build a mask of acceptable page sizes */
1493 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1494
1495 /* throw away page sizes not supported by the hardware */
1496 pgsize &= domain->pgsize_bitmap;
1497
1498 /* make sure we're still sane */
1499 BUG_ON(!pgsize);
1500
1501 /* pick the biggest page */
1502 pgsize_idx = __fls(pgsize);
1503 pgsize = 1UL << pgsize_idx;
1504
1505 return pgsize;
1506}
1507
1508int iommu_map(struct iommu_domain *domain, unsigned long iova,
1509 phys_addr_t paddr, size_t size, int prot)
1510{
1511 unsigned long orig_iova = iova;
1512 unsigned int min_pagesz;
1513 size_t orig_size = size;
1514 phys_addr_t orig_paddr = paddr;
1515 int ret = 0;
1516
1517 if (unlikely(domain->ops->map == NULL ||
1518 domain->pgsize_bitmap == 0UL))
1519 return -ENODEV;
1520
1521 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1522 return -EINVAL;
1523
1524 /* find out the minimum page size supported */
1525 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1526
1527 /*
1528 * both the virtual address and the physical one, as well as
1529 * the size of the mapping, must be aligned (at least) to the
1530 * size of the smallest page supported by the hardware
1531 */
1532 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1533 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1534 iova, &paddr, size, min_pagesz);
1535 return -EINVAL;
1536 }
1537
1538 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1539
1540 while (size) {
1541 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1542
1543 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1544 iova, &paddr, pgsize);
1545
1546 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1547 if (ret)
1548 break;
1549
1550 iova += pgsize;
1551 paddr += pgsize;
1552 size -= pgsize;
1553 }
1554
1555 /* unroll mapping in case something went wrong */
1556 if (ret)
1557 iommu_unmap(domain, orig_iova, orig_size - size);
1558 else
1559 trace_map(orig_iova, orig_paddr, orig_size);
1560
1561 return ret;
1562}
1563EXPORT_SYMBOL_GPL(iommu_map);
1564
1565static size_t __iommu_unmap(struct iommu_domain *domain,
1566 unsigned long iova, size_t size,
1567 bool sync)
1568{
1569 const struct iommu_ops *ops = domain->ops;
1570 size_t unmapped_page, unmapped = 0;
1571 unsigned long orig_iova = iova;
1572 unsigned int min_pagesz;
1573
1574 if (unlikely(ops->unmap == NULL ||
1575 domain->pgsize_bitmap == 0UL))
1576 return 0;
1577
1578 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1579 return 0;
1580
1581 /* find out the minimum page size supported */
1582 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1583
1584 /*
1585 * The virtual address, as well as the size of the mapping, must be
1586 * aligned (at least) to the size of the smallest page supported
1587 * by the hardware
1588 */
1589 if (!IS_ALIGNED(iova | size, min_pagesz)) {
1590 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1591 iova, size, min_pagesz);
1592 return 0;
1593 }
1594
1595 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1596
1597 /*
1598 * Keep iterating until we either unmap 'size' bytes (or more)
1599 * or we hit an area that isn't mapped.
1600 */
1601 while (unmapped < size) {
1602 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1603
1604 unmapped_page = ops->unmap(domain, iova, pgsize);
1605 if (!unmapped_page)
1606 break;
1607
1608 if (sync && ops->iotlb_range_add)
1609 ops->iotlb_range_add(domain, iova, pgsize);
1610
1611 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1612 iova, unmapped_page);
1613
1614 iova += unmapped_page;
1615 unmapped += unmapped_page;
1616 }
1617
1618 if (sync && ops->iotlb_sync)
1619 ops->iotlb_sync(domain);
1620
1621 trace_unmap(orig_iova, size, unmapped);
1622 return unmapped;
1623}
1624
1625size_t iommu_unmap(struct iommu_domain *domain,
1626 unsigned long iova, size_t size)
1627{
1628 return __iommu_unmap(domain, iova, size, true);
1629}
1630EXPORT_SYMBOL_GPL(iommu_unmap);
1631
1632size_t iommu_unmap_fast(struct iommu_domain *domain,
1633 unsigned long iova, size_t size)
1634{
1635 return __iommu_unmap(domain, iova, size, false);
1636}
1637EXPORT_SYMBOL_GPL(iommu_unmap_fast);
1638
1639size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1640 struct scatterlist *sg, unsigned int nents, int prot)
1641{
1642 struct scatterlist *s;
1643 size_t mapped = 0;
1644 unsigned int i, min_pagesz;
1645 int ret;
1646
1647 if (unlikely(domain->pgsize_bitmap == 0UL))
1648 return 0;
1649
1650 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1651
1652 for_each_sg(sg, s, nents, i) {
1653 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1654
1655 /*
1656 * We are mapping on IOMMU page boundaries, so offset within
1657 * the page must be 0. However, the IOMMU may support pages
1658 * smaller than PAGE_SIZE, so s->offset may still represent
1659 * an offset of that boundary within the CPU page.
1660 */
1661 if (!IS_ALIGNED(s->offset, min_pagesz))
1662 goto out_err;
1663
1664 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1665 if (ret)
1666 goto out_err;
1667
1668 mapped += s->length;
1669 }
1670
1671 return mapped;
1672
1673out_err:
1674 /* undo mappings already done */
1675 iommu_unmap(domain, iova, mapped);
1676
1677 return 0;
1678
1679}
1680EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1681
1682int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1683 phys_addr_t paddr, u64 size, int prot)
1684{
1685 if (unlikely(domain->ops->domain_window_enable == NULL))
1686 return -ENODEV;
1687
1688 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1689 prot);
1690}
1691EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1692
1693void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1694{
1695 if (unlikely(domain->ops->domain_window_disable == NULL))
1696 return;
1697
1698 return domain->ops->domain_window_disable(domain, wnd_nr);
1699}
1700EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1701
1702/**
1703 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
1704 * @domain: the iommu domain where the fault has happened
1705 * @dev: the device where the fault has happened
1706 * @iova: the faulting address
1707 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
1708 *
1709 * This function should be called by the low-level IOMMU implementations
1710 * whenever IOMMU faults happen, to allow high-level users, that are
1711 * interested in such events, to know about them.
1712 *
1713 * This event may be useful for several possible use cases:
1714 * - mere logging of the event
1715 * - dynamic TLB/PTE loading
1716 * - if restarting of the faulting device is required
1717 *
1718 * Returns 0 on success and an appropriate error code otherwise (if dynamic
1719 * PTE/TLB loading will one day be supported, implementations will be able
1720 * to tell whether it succeeded or not according to this return value).
1721 *
1722 * Specifically, -ENOSYS is returned if a fault handler isn't installed
1723 * (though fault handlers can also return -ENOSYS, in case they want to
1724 * elicit the default behavior of the IOMMU drivers).
1725 */
1726int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
1727 unsigned long iova, int flags)
1728{
1729 int ret = -ENOSYS;
1730
1731 /*
1732 * if upper layers showed interest and installed a fault handler,
1733 * invoke it.
1734 */
1735 if (domain->handler)
1736 ret = domain->handler(domain, dev, iova, flags,
1737 domain->handler_token);
1738
1739 trace_io_page_fault(dev, iova, flags);
1740 return ret;
1741}
1742EXPORT_SYMBOL_GPL(report_iommu_fault);
1743
1744static int __init iommu_init(void)
1745{
1746 iommu_group_kset = kset_create_and_add("iommu_groups",
1747 NULL, kernel_kobj);
1748 BUG_ON(!iommu_group_kset);
1749
1750 return 0;
1751}
1752core_initcall(iommu_init);
1753
1754int iommu_domain_get_attr(struct iommu_domain *domain,
1755 enum iommu_attr attr, void *data)
1756{
1757 struct iommu_domain_geometry *geometry;
1758 bool *paging;
1759 int ret = 0;
1760 u32 *count;
1761
1762 switch (attr) {
1763 case DOMAIN_ATTR_GEOMETRY:
1764 geometry = data;
1765 *geometry = domain->geometry;
1766
1767 break;
1768 case DOMAIN_ATTR_PAGING:
1769 paging = data;
1770 *paging = (domain->pgsize_bitmap != 0UL);
1771 break;
1772 case DOMAIN_ATTR_WINDOWS:
1773 count = data;
1774
1775 if (domain->ops->domain_get_windows != NULL)
1776 *count = domain->ops->domain_get_windows(domain);
1777 else
1778 ret = -ENODEV;
1779
1780 break;
1781 default:
1782 if (!domain->ops->domain_get_attr)
1783 return -EINVAL;
1784
1785 ret = domain->ops->domain_get_attr(domain, attr, data);
1786 }
1787
1788 return ret;
1789}
1790EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1791
1792int iommu_domain_set_attr(struct iommu_domain *domain,
1793 enum iommu_attr attr, void *data)
1794{
1795 int ret = 0;
1796 u32 *count;
1797
1798 switch (attr) {
1799 case DOMAIN_ATTR_WINDOWS:
1800 count = data;
1801
1802 if (domain->ops->domain_set_windows != NULL)
1803 ret = domain->ops->domain_set_windows(domain, *count);
1804 else
1805 ret = -ENODEV;
1806
1807 break;
1808 default:
1809 if (domain->ops->domain_set_attr == NULL)
1810 return -EINVAL;
1811
1812 ret = domain->ops->domain_set_attr(domain, attr, data);
1813 }
1814
1815 return ret;
1816}
1817EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1818
1819void iommu_get_resv_regions(struct device *dev, struct list_head *list)
1820{
1821 const struct iommu_ops *ops = dev->bus->iommu_ops;
1822
1823 if (ops && ops->get_resv_regions)
1824 ops->get_resv_regions(dev, list);
1825}
1826
1827void iommu_put_resv_regions(struct device *dev, struct list_head *list)
1828{
1829 const struct iommu_ops *ops = dev->bus->iommu_ops;
1830
1831 if (ops && ops->put_resv_regions)
1832 ops->put_resv_regions(dev, list);
1833}
1834
1835struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
1836 size_t length, int prot,
1837 enum iommu_resv_type type)
1838{
1839 struct iommu_resv_region *region;
1840
1841 region = kzalloc(sizeof(*region), GFP_KERNEL);
1842 if (!region)
1843 return NULL;
1844
1845 INIT_LIST_HEAD(®ion->list);
1846 region->start = start;
1847 region->length = length;
1848 region->prot = prot;
1849 region->type = type;
1850 return region;
1851}
1852
1853/* Request that a device is direct mapped by the IOMMU */
1854int iommu_request_dm_for_dev(struct device *dev)
1855{
1856 struct iommu_domain *dm_domain;
1857 struct iommu_group *group;
1858 int ret;
1859
1860 /* Device must already be in a group before calling this function */
1861 group = iommu_group_get_for_dev(dev);
1862 if (IS_ERR(group))
1863 return PTR_ERR(group);
1864
1865 mutex_lock(&group->mutex);
1866
1867 /* Check if the default domain is already direct mapped */
1868 ret = 0;
1869 if (group->default_domain &&
1870 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1871 goto out;
1872
1873 /* Don't change mappings of existing devices */
1874 ret = -EBUSY;
1875 if (iommu_group_device_count(group) != 1)
1876 goto out;
1877
1878 /* Allocate a direct mapped domain */
1879 ret = -ENOMEM;
1880 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1881 if (!dm_domain)
1882 goto out;
1883
1884 /* Attach the device to the domain */
1885 ret = __iommu_attach_group(dm_domain, group);
1886 if (ret) {
1887 iommu_domain_free(dm_domain);
1888 goto out;
1889 }
1890
1891 /* Make the direct mapped domain the default for this group */
1892 if (group->default_domain)
1893 iommu_domain_free(group->default_domain);
1894 group->default_domain = dm_domain;
1895
1896 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1897
1898 ret = 0;
1899out:
1900 mutex_unlock(&group->mutex);
1901 iommu_group_put(group);
1902
1903 return ret;
1904}
1905
1906const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
1907{
1908 const struct iommu_ops *ops = NULL;
1909 struct iommu_device *iommu;
1910
1911 spin_lock(&iommu_device_lock);
1912 list_for_each_entry(iommu, &iommu_device_list, list)
1913 if (iommu->fwnode == fwnode) {
1914 ops = iommu->ops;
1915 break;
1916 }
1917 spin_unlock(&iommu_device_lock);
1918 return ops;
1919}
1920
1921int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1922 const struct iommu_ops *ops)
1923{
1924 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1925
1926 if (fwspec)
1927 return ops == fwspec->ops ? 0 : -EINVAL;
1928
1929 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1930 if (!fwspec)
1931 return -ENOMEM;
1932
1933 of_node_get(to_of_node(iommu_fwnode));
1934 fwspec->iommu_fwnode = iommu_fwnode;
1935 fwspec->ops = ops;
1936 dev->iommu_fwspec = fwspec;
1937 return 0;
1938}
1939EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1940
1941void iommu_fwspec_free(struct device *dev)
1942{
1943 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1944
1945 if (fwspec) {
1946 fwnode_handle_put(fwspec->iommu_fwnode);
1947 kfree(fwspec);
1948 dev->iommu_fwspec = NULL;
1949 }
1950}
1951EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1952
1953int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1954{
1955 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1956 size_t size;
1957 int i;
1958
1959 if (!fwspec)
1960 return -EINVAL;
1961
1962 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1963 if (size > sizeof(*fwspec)) {
1964 fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1965 if (!fwspec)
1966 return -ENOMEM;
1967
1968 dev->iommu_fwspec = fwspec;
1969 }
1970
1971 for (i = 0; i < num_ids; i++)
1972 fwspec->ids[fwspec->num_ids + i] = ids[i];
1973
1974 fwspec->num_ids += num_ids;
1975 return 0;
1976}
1977EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);