iommu: Rename struct iommu_device
[linux-2.6-block.git] / drivers / iommu / iommu.c
CommitLineData
fc2100eb
JR
1/*
2 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
63ce3ae8 3 * Author: Joerg Roedel <jroedel@suse.de>
fc2100eb
JR
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
92e7066f 19#define pr_fmt(fmt) "iommu: " fmt
7d3002cc 20
905d66c1 21#include <linux/device.h>
40998188 22#include <linux/kernel.h>
fc2100eb
JR
23#include <linux/bug.h>
24#include <linux/types.h>
60db4027
AM
25#include <linux/module.h>
26#include <linux/slab.h>
fc2100eb
JR
27#include <linux/errno.h>
28#include <linux/iommu.h>
d72e31c9
AW
29#include <linux/idr.h>
30#include <linux/notifier.h>
31#include <linux/err.h>
104a1c13 32#include <linux/pci.h>
f096c061 33#include <linux/bitops.h>
57f98d2f 34#include <linux/property.h>
7f6db171 35#include <trace/events/iommu.h>
d72e31c9
AW
36
37static struct kset *iommu_group_kset;
e38d1f13 38static DEFINE_IDA(iommu_group_ida);
d72e31c9 39
b22f6434
TR
40struct iommu_callback_data {
41 const struct iommu_ops *ops;
42};
43
d72e31c9
AW
44struct iommu_group {
45 struct kobject kobj;
46 struct kobject *devices_kobj;
47 struct list_head devices;
48 struct mutex mutex;
49 struct blocking_notifier_head notifier;
50 void *iommu_data;
51 void (*iommu_data_release)(void *iommu_data);
52 char *name;
53 int id;
53723dc5 54 struct iommu_domain *default_domain;
e39cb8a3 55 struct iommu_domain *domain;
d72e31c9
AW
56};
57
c09e22d5 58struct group_device {
d72e31c9
AW
59 struct list_head list;
60 struct device *dev;
61 char *name;
62};
63
64struct iommu_group_attribute {
65 struct attribute attr;
66 ssize_t (*show)(struct iommu_group *group, char *buf);
67 ssize_t (*store)(struct iommu_group *group,
68 const char *buf, size_t count);
69};
70
71#define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
72struct iommu_group_attribute iommu_group_attr_##_name = \
73 __ATTR(_name, _mode, _show, _store)
fc2100eb 74
d72e31c9
AW
75#define to_iommu_group_attr(_attr) \
76 container_of(_attr, struct iommu_group_attribute, attr)
77#define to_iommu_group(_kobj) \
78 container_of(_kobj, struct iommu_group, kobj)
fc2100eb 79
53723dc5
JR
80static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
81 unsigned type);
e39cb8a3
JR
82static int __iommu_attach_device(struct iommu_domain *domain,
83 struct device *dev);
84static int __iommu_attach_group(struct iommu_domain *domain,
85 struct iommu_group *group);
86static void __iommu_detach_group(struct iommu_domain *domain,
87 struct iommu_group *group);
53723dc5 88
d72e31c9
AW
89static ssize_t iommu_group_attr_show(struct kobject *kobj,
90 struct attribute *__attr, char *buf)
1460432c 91{
d72e31c9
AW
92 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
93 struct iommu_group *group = to_iommu_group(kobj);
94 ssize_t ret = -EIO;
1460432c 95
d72e31c9
AW
96 if (attr->show)
97 ret = attr->show(group, buf);
98 return ret;
99}
100
101static ssize_t iommu_group_attr_store(struct kobject *kobj,
102 struct attribute *__attr,
103 const char *buf, size_t count)
104{
105 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
106 struct iommu_group *group = to_iommu_group(kobj);
107 ssize_t ret = -EIO;
1460432c 108
d72e31c9
AW
109 if (attr->store)
110 ret = attr->store(group, buf, count);
111 return ret;
1460432c 112}
1460432c 113
d72e31c9
AW
114static const struct sysfs_ops iommu_group_sysfs_ops = {
115 .show = iommu_group_attr_show,
116 .store = iommu_group_attr_store,
117};
1460432c 118
d72e31c9
AW
119static int iommu_group_create_file(struct iommu_group *group,
120 struct iommu_group_attribute *attr)
121{
122 return sysfs_create_file(&group->kobj, &attr->attr);
1460432c 123}
1460432c 124
d72e31c9
AW
125static void iommu_group_remove_file(struct iommu_group *group,
126 struct iommu_group_attribute *attr)
127{
128 sysfs_remove_file(&group->kobj, &attr->attr);
129}
130
131static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
132{
133 return sprintf(buf, "%s\n", group->name);
134}
135
136static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
137
138static void iommu_group_release(struct kobject *kobj)
139{
140 struct iommu_group *group = to_iommu_group(kobj);
141
269aa808
JR
142 pr_debug("Releasing group %d\n", group->id);
143
d72e31c9
AW
144 if (group->iommu_data_release)
145 group->iommu_data_release(group->iommu_data);
146
feccf398 147 ida_simple_remove(&iommu_group_ida, group->id);
d72e31c9 148
53723dc5
JR
149 if (group->default_domain)
150 iommu_domain_free(group->default_domain);
151
d72e31c9
AW
152 kfree(group->name);
153 kfree(group);
154}
155
156static struct kobj_type iommu_group_ktype = {
157 .sysfs_ops = &iommu_group_sysfs_ops,
158 .release = iommu_group_release,
159};
160
161/**
162 * iommu_group_alloc - Allocate a new group
163 * @name: Optional name to associate with group, visible in sysfs
164 *
165 * This function is called by an iommu driver to allocate a new iommu
166 * group. The iommu group represents the minimum granularity of the iommu.
167 * Upon successful return, the caller holds a reference to the supplied
168 * group in order to hold the group until devices are added. Use
169 * iommu_group_put() to release this extra reference count, allowing the
170 * group to be automatically reclaimed once it has no devices or external
171 * references.
172 */
173struct iommu_group *iommu_group_alloc(void)
1460432c 174{
d72e31c9
AW
175 struct iommu_group *group;
176 int ret;
177
178 group = kzalloc(sizeof(*group), GFP_KERNEL);
179 if (!group)
180 return ERR_PTR(-ENOMEM);
181
182 group->kobj.kset = iommu_group_kset;
183 mutex_init(&group->mutex);
184 INIT_LIST_HEAD(&group->devices);
185 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
186
feccf398
HK
187 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
188 if (ret < 0) {
d72e31c9 189 kfree(group);
feccf398 190 return ERR_PTR(ret);
d72e31c9 191 }
feccf398 192 group->id = ret;
1460432c 193
d72e31c9
AW
194 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
195 NULL, "%d", group->id);
196 if (ret) {
feccf398 197 ida_simple_remove(&iommu_group_ida, group->id);
d72e31c9
AW
198 kfree(group);
199 return ERR_PTR(ret);
200 }
201
202 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
203 if (!group->devices_kobj) {
204 kobject_put(&group->kobj); /* triggers .release & free */
205 return ERR_PTR(-ENOMEM);
206 }
207
208 /*
209 * The devices_kobj holds a reference on the group kobject, so
210 * as long as that exists so will the group. We can therefore
211 * use the devices_kobj for reference counting.
212 */
213 kobject_put(&group->kobj);
214
269aa808
JR
215 pr_debug("Allocated group %d\n", group->id);
216
d72e31c9
AW
217 return group;
218}
219EXPORT_SYMBOL_GPL(iommu_group_alloc);
220
aa16bea9
AK
221struct iommu_group *iommu_group_get_by_id(int id)
222{
223 struct kobject *group_kobj;
224 struct iommu_group *group;
225 const char *name;
226
227 if (!iommu_group_kset)
228 return NULL;
229
230 name = kasprintf(GFP_KERNEL, "%d", id);
231 if (!name)
232 return NULL;
233
234 group_kobj = kset_find_obj(iommu_group_kset, name);
235 kfree(name);
236
237 if (!group_kobj)
238 return NULL;
239
240 group = container_of(group_kobj, struct iommu_group, kobj);
241 BUG_ON(group->id != id);
242
243 kobject_get(group->devices_kobj);
244 kobject_put(&group->kobj);
245
246 return group;
247}
248EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
249
d72e31c9
AW
250/**
251 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
252 * @group: the group
253 *
254 * iommu drivers can store data in the group for use when doing iommu
255 * operations. This function provides a way to retrieve it. Caller
256 * should hold a group reference.
257 */
258void *iommu_group_get_iommudata(struct iommu_group *group)
259{
260 return group->iommu_data;
261}
262EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
263
264/**
265 * iommu_group_set_iommudata - set iommu_data for a group
266 * @group: the group
267 * @iommu_data: new data
268 * @release: release function for iommu_data
269 *
270 * iommu drivers can store data in the group for use when doing iommu
271 * operations. This function provides a way to set the data after
272 * the group has been allocated. Caller should hold a group reference.
273 */
274void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
275 void (*release)(void *iommu_data))
1460432c 276{
d72e31c9
AW
277 group->iommu_data = iommu_data;
278 group->iommu_data_release = release;
279}
280EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
1460432c 281
d72e31c9
AW
282/**
283 * iommu_group_set_name - set name for a group
284 * @group: the group
285 * @name: name
286 *
287 * Allow iommu driver to set a name for a group. When set it will
288 * appear in a name attribute file under the group in sysfs.
289 */
290int iommu_group_set_name(struct iommu_group *group, const char *name)
291{
292 int ret;
293
294 if (group->name) {
295 iommu_group_remove_file(group, &iommu_group_attr_name);
296 kfree(group->name);
297 group->name = NULL;
298 if (!name)
299 return 0;
300 }
301
302 group->name = kstrdup(name, GFP_KERNEL);
303 if (!group->name)
304 return -ENOMEM;
305
306 ret = iommu_group_create_file(group, &iommu_group_attr_name);
307 if (ret) {
308 kfree(group->name);
309 group->name = NULL;
310 return ret;
311 }
1460432c
AW
312
313 return 0;
314}
d72e31c9 315EXPORT_SYMBOL_GPL(iommu_group_set_name);
1460432c 316
beed2821
JR
317static int iommu_group_create_direct_mappings(struct iommu_group *group,
318 struct device *dev)
319{
320 struct iommu_domain *domain = group->default_domain;
321 struct iommu_dm_region *entry;
322 struct list_head mappings;
323 unsigned long pg_size;
324 int ret = 0;
325
326 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
327 return 0;
328
d16e0faa 329 BUG_ON(!domain->pgsize_bitmap);
beed2821 330
d16e0faa 331 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
beed2821
JR
332 INIT_LIST_HEAD(&mappings);
333
334 iommu_get_dm_regions(dev, &mappings);
335
336 /* We need to consider overlapping regions for different devices */
337 list_for_each_entry(entry, &mappings, list) {
338 dma_addr_t start, end, addr;
339
33b21a6b
JR
340 if (domain->ops->apply_dm_region)
341 domain->ops->apply_dm_region(dev, domain, entry);
342
beed2821
JR
343 start = ALIGN(entry->start, pg_size);
344 end = ALIGN(entry->start + entry->length, pg_size);
345
346 for (addr = start; addr < end; addr += pg_size) {
347 phys_addr_t phys_addr;
348
349 phys_addr = iommu_iova_to_phys(domain, addr);
350 if (phys_addr)
351 continue;
352
353 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
354 if (ret)
355 goto out;
356 }
357
358 }
359
360out:
361 iommu_put_dm_regions(dev, &mappings);
362
363 return ret;
364}
365
d72e31c9
AW
366/**
367 * iommu_group_add_device - add a device to an iommu group
368 * @group: the group into which to add the device (reference should be held)
369 * @dev: the device
370 *
371 * This function is called by an iommu driver to add a device into a
372 * group. Adding a device increments the group reference count.
373 */
374int iommu_group_add_device(struct iommu_group *group, struct device *dev)
1460432c 375{
d72e31c9 376 int ret, i = 0;
c09e22d5 377 struct group_device *device;
d72e31c9
AW
378
379 device = kzalloc(sizeof(*device), GFP_KERNEL);
380 if (!device)
381 return -ENOMEM;
382
383 device->dev = dev;
1460432c 384
d72e31c9 385 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
797a8b4d
RM
386 if (ret)
387 goto err_free_device;
d72e31c9
AW
388
389 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
390rename:
391 if (!device->name) {
797a8b4d
RM
392 ret = -ENOMEM;
393 goto err_remove_link;
d72e31c9 394 }
1460432c 395
d72e31c9
AW
396 ret = sysfs_create_link_nowarn(group->devices_kobj,
397 &dev->kobj, device->name);
398 if (ret) {
d72e31c9
AW
399 if (ret == -EEXIST && i >= 0) {
400 /*
401 * Account for the slim chance of collision
402 * and append an instance to the name.
403 */
797a8b4d 404 kfree(device->name);
d72e31c9
AW
405 device->name = kasprintf(GFP_KERNEL, "%s.%d",
406 kobject_name(&dev->kobj), i++);
407 goto rename;
408 }
797a8b4d 409 goto err_free_name;
d72e31c9
AW
410 }
411
412 kobject_get(group->devices_kobj);
413
414 dev->iommu_group = group;
415
beed2821
JR
416 iommu_group_create_direct_mappings(group, dev);
417
d72e31c9
AW
418 mutex_lock(&group->mutex);
419 list_add_tail(&device->list, &group->devices);
e39cb8a3 420 if (group->domain)
797a8b4d 421 ret = __iommu_attach_device(group->domain, dev);
d72e31c9 422 mutex_unlock(&group->mutex);
797a8b4d
RM
423 if (ret)
424 goto err_put_group;
d72e31c9
AW
425
426 /* Notify any listeners about change to group. */
427 blocking_notifier_call_chain(&group->notifier,
428 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
d1cf7e82
SK
429
430 trace_add_device_to_group(group->id, dev);
269aa808
JR
431
432 pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
433
1460432c 434 return 0;
797a8b4d
RM
435
436err_put_group:
437 mutex_lock(&group->mutex);
438 list_del(&device->list);
439 mutex_unlock(&group->mutex);
440 dev->iommu_group = NULL;
441 kobject_put(group->devices_kobj);
442err_free_name:
443 kfree(device->name);
444err_remove_link:
445 sysfs_remove_link(&dev->kobj, "iommu_group");
446err_free_device:
447 kfree(device);
448 pr_err("Failed to add device %s to group %d: %d\n", dev_name(dev), group->id, ret);
449 return ret;
1460432c 450}
d72e31c9 451EXPORT_SYMBOL_GPL(iommu_group_add_device);
1460432c 452
d72e31c9
AW
453/**
454 * iommu_group_remove_device - remove a device from it's current group
455 * @dev: device to be removed
456 *
457 * This function is called by an iommu driver to remove the device from
458 * it's current group. This decrements the iommu group reference count.
459 */
460void iommu_group_remove_device(struct device *dev)
461{
462 struct iommu_group *group = dev->iommu_group;
c09e22d5 463 struct group_device *tmp_device, *device = NULL;
d72e31c9 464
269aa808
JR
465 pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
466
d72e31c9
AW
467 /* Pre-notify listeners that a device is being removed. */
468 blocking_notifier_call_chain(&group->notifier,
469 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
470
471 mutex_lock(&group->mutex);
472 list_for_each_entry(tmp_device, &group->devices, list) {
473 if (tmp_device->dev == dev) {
474 device = tmp_device;
475 list_del(&device->list);
476 break;
477 }
478 }
479 mutex_unlock(&group->mutex);
480
481 if (!device)
482 return;
483
484 sysfs_remove_link(group->devices_kobj, device->name);
485 sysfs_remove_link(&dev->kobj, "iommu_group");
486
2e757086
SK
487 trace_remove_device_from_group(group->id, dev);
488
d72e31c9
AW
489 kfree(device->name);
490 kfree(device);
491 dev->iommu_group = NULL;
492 kobject_put(group->devices_kobj);
493}
494EXPORT_SYMBOL_GPL(iommu_group_remove_device);
495
426a2738
JR
496static int iommu_group_device_count(struct iommu_group *group)
497{
c09e22d5 498 struct group_device *entry;
426a2738
JR
499 int ret = 0;
500
501 list_for_each_entry(entry, &group->devices, list)
502 ret++;
503
504 return ret;
505}
506
d72e31c9
AW
507/**
508 * iommu_group_for_each_dev - iterate over each device in the group
509 * @group: the group
510 * @data: caller opaque data to be passed to callback function
511 * @fn: caller supplied callback function
512 *
513 * This function is called by group users to iterate over group devices.
514 * Callers should hold a reference count to the group during callback.
515 * The group->mutex is held across callbacks, which will block calls to
516 * iommu_group_add/remove_device.
517 */
e39cb8a3
JR
518static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
519 int (*fn)(struct device *, void *))
d72e31c9 520{
c09e22d5 521 struct group_device *device;
d72e31c9
AW
522 int ret = 0;
523
d72e31c9
AW
524 list_for_each_entry(device, &group->devices, list) {
525 ret = fn(device->dev, data);
526 if (ret)
527 break;
528 }
e39cb8a3
JR
529 return ret;
530}
531
532
533int iommu_group_for_each_dev(struct iommu_group *group, void *data,
534 int (*fn)(struct device *, void *))
535{
536 int ret;
537
538 mutex_lock(&group->mutex);
539 ret = __iommu_group_for_each_dev(group, data, fn);
d72e31c9 540 mutex_unlock(&group->mutex);
e39cb8a3 541
d72e31c9
AW
542 return ret;
543}
544EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
545
546/**
547 * iommu_group_get - Return the group for a device and increment reference
548 * @dev: get the group that this device belongs to
549 *
550 * This function is called by iommu drivers and users to get the group
551 * for the specified device. If found, the group is returned and the group
552 * reference in incremented, else NULL.
553 */
554struct iommu_group *iommu_group_get(struct device *dev)
555{
556 struct iommu_group *group = dev->iommu_group;
557
558 if (group)
559 kobject_get(group->devices_kobj);
560
561 return group;
562}
563EXPORT_SYMBOL_GPL(iommu_group_get);
564
13f59a78
RM
565/**
566 * iommu_group_ref_get - Increment reference on a group
567 * @group: the group to use, must not be NULL
568 *
569 * This function is called by iommu drivers to take additional references on an
570 * existing group. Returns the given group for convenience.
571 */
572struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
573{
574 kobject_get(group->devices_kobj);
575 return group;
576}
577
d72e31c9
AW
578/**
579 * iommu_group_put - Decrement group reference
580 * @group: the group to use
581 *
582 * This function is called by iommu drivers and users to release the
583 * iommu group. Once the reference count is zero, the group is released.
584 */
585void iommu_group_put(struct iommu_group *group)
586{
587 if (group)
588 kobject_put(group->devices_kobj);
589}
590EXPORT_SYMBOL_GPL(iommu_group_put);
591
592/**
593 * iommu_group_register_notifier - Register a notifier for group changes
594 * @group: the group to watch
595 * @nb: notifier block to signal
596 *
597 * This function allows iommu group users to track changes in a group.
598 * See include/linux/iommu.h for actions sent via this notifier. Caller
599 * should hold a reference to the group throughout notifier registration.
600 */
601int iommu_group_register_notifier(struct iommu_group *group,
602 struct notifier_block *nb)
603{
604 return blocking_notifier_chain_register(&group->notifier, nb);
605}
606EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
607
608/**
609 * iommu_group_unregister_notifier - Unregister a notifier
610 * @group: the group to watch
611 * @nb: notifier block to signal
612 *
613 * Unregister a previously registered group notifier block.
614 */
615int iommu_group_unregister_notifier(struct iommu_group *group,
616 struct notifier_block *nb)
617{
618 return blocking_notifier_chain_unregister(&group->notifier, nb);
619}
620EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
621
622/**
623 * iommu_group_id - Return ID for a group
624 * @group: the group to ID
625 *
626 * Return the unique ID for the group matching the sysfs group number.
627 */
628int iommu_group_id(struct iommu_group *group)
629{
630 return group->id;
631}
632EXPORT_SYMBOL_GPL(iommu_group_id);
1460432c 633
f096c061
AW
634static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
635 unsigned long *devfns);
636
104a1c13
AW
637/*
638 * To consider a PCI device isolated, we require ACS to support Source
639 * Validation, Request Redirection, Completer Redirection, and Upstream
640 * Forwarding. This effectively means that devices cannot spoof their
641 * requester ID, requests and completions cannot be redirected, and all
642 * transactions are forwarded upstream, even as it passes through a
643 * bridge where the target device is downstream.
644 */
645#define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
646
f096c061
AW
647/*
648 * For multifunction devices which are not isolated from each other, find
649 * all the other non-isolated functions and look for existing groups. For
650 * each function, we also need to look for aliases to or from other devices
651 * that may already have a group.
652 */
653static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
654 unsigned long *devfns)
655{
656 struct pci_dev *tmp = NULL;
657 struct iommu_group *group;
658
659 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
660 return NULL;
661
662 for_each_pci_dev(tmp) {
663 if (tmp == pdev || tmp->bus != pdev->bus ||
664 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
665 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
666 continue;
667
668 group = get_pci_alias_group(tmp, devfns);
669 if (group) {
670 pci_dev_put(tmp);
671 return group;
672 }
673 }
674
675 return NULL;
676}
677
678/*
338c3149
JL
679 * Look for aliases to or from the given device for existing groups. DMA
680 * aliases are only supported on the same bus, therefore the search
f096c061
AW
681 * space is quite small (especially since we're really only looking at pcie
682 * device, and therefore only expect multiple slots on the root complex or
683 * downstream switch ports). It's conceivable though that a pair of
684 * multifunction devices could have aliases between them that would cause a
685 * loop. To prevent this, we use a bitmap to track where we've been.
686 */
687static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
688 unsigned long *devfns)
689{
690 struct pci_dev *tmp = NULL;
691 struct iommu_group *group;
692
693 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
694 return NULL;
695
696 group = iommu_group_get(&pdev->dev);
697 if (group)
698 return group;
699
700 for_each_pci_dev(tmp) {
701 if (tmp == pdev || tmp->bus != pdev->bus)
702 continue;
703
704 /* We alias them or they alias us */
338c3149 705 if (pci_devs_are_dma_aliases(pdev, tmp)) {
f096c061
AW
706 group = get_pci_alias_group(tmp, devfns);
707 if (group) {
708 pci_dev_put(tmp);
709 return group;
710 }
711
712 group = get_pci_function_alias_group(tmp, devfns);
713 if (group) {
714 pci_dev_put(tmp);
715 return group;
716 }
717 }
718 }
719
720 return NULL;
721}
722
104a1c13
AW
723struct group_for_pci_data {
724 struct pci_dev *pdev;
725 struct iommu_group *group;
726};
727
728/*
729 * DMA alias iterator callback, return the last seen device. Stop and return
730 * the IOMMU group if we find one along the way.
731 */
732static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
733{
734 struct group_for_pci_data *data = opaque;
735
736 data->pdev = pdev;
737 data->group = iommu_group_get(&pdev->dev);
738
739 return data->group != NULL;
740}
741
6eab556a
JR
742/*
743 * Generic device_group call-back function. It just allocates one
744 * iommu-group per device.
745 */
746struct iommu_group *generic_device_group(struct device *dev)
747{
748 struct iommu_group *group;
749
750 group = iommu_group_alloc();
751 if (IS_ERR(group))
752 return NULL;
753
754 return group;
755}
756
104a1c13
AW
757/*
758 * Use standard PCI bus topology, isolation features, and DMA alias quirks
759 * to find or create an IOMMU group for a device.
760 */
5e62292b 761struct iommu_group *pci_device_group(struct device *dev)
104a1c13 762{
5e62292b 763 struct pci_dev *pdev = to_pci_dev(dev);
104a1c13
AW
764 struct group_for_pci_data data;
765 struct pci_bus *bus;
766 struct iommu_group *group = NULL;
f096c061 767 u64 devfns[4] = { 0 };
104a1c13 768
5e62292b
JR
769 if (WARN_ON(!dev_is_pci(dev)))
770 return ERR_PTR(-EINVAL);
771
104a1c13
AW
772 /*
773 * Find the upstream DMA alias for the device. A device must not
774 * be aliased due to topology in order to have its own IOMMU group.
775 * If we find an alias along the way that already belongs to a
776 * group, use it.
777 */
778 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
779 return data.group;
780
781 pdev = data.pdev;
782
783 /*
784 * Continue upstream from the point of minimum IOMMU granularity
785 * due to aliases to the point where devices are protected from
786 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
787 * group, use it.
788 */
789 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
790 if (!bus->self)
791 continue;
792
793 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
794 break;
795
796 pdev = bus->self;
797
798 group = iommu_group_get(&pdev->dev);
799 if (group)
800 return group;
801 }
802
803 /*
f096c061
AW
804 * Look for existing groups on device aliases. If we alias another
805 * device or another device aliases us, use the same group.
104a1c13 806 */
f096c061
AW
807 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
808 if (group)
809 return group;
104a1c13
AW
810
811 /*
f096c061
AW
812 * Look for existing groups on non-isolated functions on the same
813 * slot and aliases of those funcions, if any. No need to clear
814 * the search bitmap, the tested devfns are still valid.
104a1c13 815 */
f096c061
AW
816 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
817 if (group)
818 return group;
104a1c13
AW
819
820 /* No shared group found, allocate new */
53723dc5 821 group = iommu_group_alloc();
409e553d
DC
822 if (IS_ERR(group))
823 return NULL;
824
53723dc5 825 return group;
104a1c13
AW
826}
827
828/**
829 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
830 * @dev: target device
831 *
832 * This function is intended to be called by IOMMU drivers and extended to
833 * support common, bus-defined algorithms when determining or creating the
834 * IOMMU group for a device. On success, the caller will hold a reference
835 * to the returned IOMMU group, which will already include the provided
836 * device. The reference should be released with iommu_group_put().
837 */
838struct iommu_group *iommu_group_get_for_dev(struct device *dev)
839{
46c6b2bc 840 const struct iommu_ops *ops = dev->bus->iommu_ops;
c4a783b8 841 struct iommu_group *group;
104a1c13
AW
842 int ret;
843
844 group = iommu_group_get(dev);
845 if (group)
846 return group;
847
46c6b2bc 848 group = ERR_PTR(-EINVAL);
c4a783b8 849
46c6b2bc
JR
850 if (ops && ops->device_group)
851 group = ops->device_group(dev);
104a1c13
AW
852
853 if (IS_ERR(group))
854 return group;
855
1228236d
JR
856 /*
857 * Try to allocate a default domain - needs support from the
858 * IOMMU driver.
859 */
860 if (!group->default_domain) {
861 group->default_domain = __iommu_domain_alloc(dev->bus,
862 IOMMU_DOMAIN_DMA);
eebb8034
JR
863 if (!group->domain)
864 group->domain = group->default_domain;
1228236d
JR
865 }
866
104a1c13
AW
867 ret = iommu_group_add_device(group, dev);
868 if (ret) {
869 iommu_group_put(group);
870 return ERR_PTR(ret);
871 }
872
873 return group;
874}
875
6827ca83
JR
876struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
877{
878 return group->default_domain;
879}
880
d72e31c9 881static int add_iommu_group(struct device *dev, void *data)
1460432c 882{
b22f6434
TR
883 struct iommu_callback_data *cb = data;
884 const struct iommu_ops *ops = cb->ops;
38667f18 885 int ret;
d72e31c9
AW
886
887 if (!ops->add_device)
461bfb3f 888 return 0;
1460432c 889
d72e31c9
AW
890 WARN_ON(dev->iommu_group);
891
38667f18
JR
892 ret = ops->add_device(dev);
893
894 /*
895 * We ignore -ENODEV errors for now, as they just mean that the
896 * device is not translated by an IOMMU. We still care about
897 * other errors and fail to initialize when they happen.
898 */
899 if (ret == -ENODEV)
900 ret = 0;
901
902 return ret;
1460432c
AW
903}
904
8da30142
JR
905static int remove_iommu_group(struct device *dev, void *data)
906{
907 struct iommu_callback_data *cb = data;
908 const struct iommu_ops *ops = cb->ops;
909
910 if (ops->remove_device && dev->iommu_group)
911 ops->remove_device(dev);
1460432c
AW
912
913 return 0;
914}
915
d72e31c9
AW
916static int iommu_bus_notifier(struct notifier_block *nb,
917 unsigned long action, void *data)
1460432c
AW
918{
919 struct device *dev = data;
b22f6434 920 const struct iommu_ops *ops = dev->bus->iommu_ops;
d72e31c9
AW
921 struct iommu_group *group;
922 unsigned long group_action = 0;
923
924 /*
925 * ADD/DEL call into iommu driver ops if provided, which may
926 * result in ADD/DEL notifiers to group->notifier
927 */
928 if (action == BUS_NOTIFY_ADD_DEVICE) {
929 if (ops->add_device)
930 return ops->add_device(dev);
843cb6dc 931 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
d72e31c9
AW
932 if (ops->remove_device && dev->iommu_group) {
933 ops->remove_device(dev);
934 return 0;
935 }
936 }
1460432c 937
d72e31c9
AW
938 /*
939 * Remaining BUS_NOTIFYs get filtered and republished to the
940 * group, if anyone is listening
941 */
942 group = iommu_group_get(dev);
943 if (!group)
944 return 0;
1460432c 945
d72e31c9
AW
946 switch (action) {
947 case BUS_NOTIFY_BIND_DRIVER:
948 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
949 break;
950 case BUS_NOTIFY_BOUND_DRIVER:
951 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
952 break;
953 case BUS_NOTIFY_UNBIND_DRIVER:
954 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
955 break;
956 case BUS_NOTIFY_UNBOUND_DRIVER:
957 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
958 break;
959 }
1460432c 960
d72e31c9
AW
961 if (group_action)
962 blocking_notifier_call_chain(&group->notifier,
963 group_action, dev);
1460432c 964
d72e31c9 965 iommu_group_put(group);
1460432c
AW
966 return 0;
967}
968
fb3e3065 969static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
ff21776d 970{
fb3e3065
MS
971 int err;
972 struct notifier_block *nb;
b22f6434
TR
973 struct iommu_callback_data cb = {
974 .ops = ops,
975 };
976
fb3e3065
MS
977 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
978 if (!nb)
979 return -ENOMEM;
980
981 nb->notifier_call = iommu_bus_notifier;
982
983 err = bus_register_notifier(bus, nb);
8da30142
JR
984 if (err)
985 goto out_free;
d7da6bdc
HS
986
987 err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
8da30142
JR
988 if (err)
989 goto out_err;
990
d7da6bdc
HS
991
992 return 0;
8da30142
JR
993
994out_err:
995 /* Clean up */
996 bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
997 bus_unregister_notifier(bus, nb);
998
999out_free:
1000 kfree(nb);
1001
1002 return err;
ff21776d 1003}
fc2100eb 1004
ff21776d
JR
1005/**
1006 * bus_set_iommu - set iommu-callbacks for the bus
1007 * @bus: bus.
1008 * @ops: the callbacks provided by the iommu-driver
1009 *
1010 * This function is called by an iommu driver to set the iommu methods
1011 * used for a particular bus. Drivers for devices on that bus can use
1012 * the iommu-api after these ops are registered.
1013 * This special function is needed because IOMMUs are usually devices on
1014 * the bus itself, so the iommu drivers are not initialized when the bus
1015 * is set up. With this function the iommu-driver can set the iommu-ops
1016 * afterwards.
1017 */
b22f6434 1018int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
fc2100eb 1019{
d7da6bdc
HS
1020 int err;
1021
ff21776d
JR
1022 if (bus->iommu_ops != NULL)
1023 return -EBUSY;
fc2100eb 1024
ff21776d
JR
1025 bus->iommu_ops = ops;
1026
1027 /* Do IOMMU specific setup for this bus-type */
d7da6bdc
HS
1028 err = iommu_bus_init(bus, ops);
1029 if (err)
1030 bus->iommu_ops = NULL;
1031
1032 return err;
fc2100eb 1033}
ff21776d 1034EXPORT_SYMBOL_GPL(bus_set_iommu);
fc2100eb 1035
a1b60c1c 1036bool iommu_present(struct bus_type *bus)
fc2100eb 1037{
94441c3b 1038 return bus->iommu_ops != NULL;
fc2100eb 1039}
a1b60c1c 1040EXPORT_SYMBOL_GPL(iommu_present);
fc2100eb 1041
3c0e0ca0
JR
1042bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1043{
1044 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1045 return false;
1046
1047 return bus->iommu_ops->capable(cap);
1048}
1049EXPORT_SYMBOL_GPL(iommu_capable);
1050
4f3f8d9d
OBC
1051/**
1052 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1053 * @domain: iommu domain
1054 * @handler: fault handler
77ca2332 1055 * @token: user data, will be passed back to the fault handler
0ed6d2d2
OBC
1056 *
1057 * This function should be used by IOMMU users which want to be notified
1058 * whenever an IOMMU fault happens.
1059 *
1060 * The fault handler itself should return 0 on success, and an appropriate
1061 * error code otherwise.
4f3f8d9d
OBC
1062 */
1063void iommu_set_fault_handler(struct iommu_domain *domain,
77ca2332
OBC
1064 iommu_fault_handler_t handler,
1065 void *token)
4f3f8d9d
OBC
1066{
1067 BUG_ON(!domain);
1068
1069 domain->handler = handler;
77ca2332 1070 domain->handler_token = token;
4f3f8d9d 1071}
30bd918c 1072EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
4f3f8d9d 1073
53723dc5
JR
1074static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1075 unsigned type)
fc2100eb
JR
1076{
1077 struct iommu_domain *domain;
fc2100eb 1078
94441c3b 1079 if (bus == NULL || bus->iommu_ops == NULL)
905d66c1
JR
1080 return NULL;
1081
53723dc5 1082 domain = bus->iommu_ops->domain_alloc(type);
fc2100eb
JR
1083 if (!domain)
1084 return NULL;
1085
8539c7c1 1086 domain->ops = bus->iommu_ops;
53723dc5 1087 domain->type = type;
d16e0faa
RM
1088 /* Assume all sizes by default; the driver may override this later */
1089 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
905d66c1 1090
fc2100eb 1091 return domain;
fc2100eb 1092}
fc2100eb 1093
53723dc5
JR
1094struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1095{
1096 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
fc2100eb
JR
1097}
1098EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1099
1100void iommu_domain_free(struct iommu_domain *domain)
1101{
89be34a1 1102 domain->ops->domain_free(domain);
fc2100eb
JR
1103}
1104EXPORT_SYMBOL_GPL(iommu_domain_free);
1105
426a2738
JR
1106static int __iommu_attach_device(struct iommu_domain *domain,
1107 struct device *dev)
fc2100eb 1108{
b54db778 1109 int ret;
e5aa7f00
JR
1110 if (unlikely(domain->ops->attach_dev == NULL))
1111 return -ENODEV;
1112
b54db778
SK
1113 ret = domain->ops->attach_dev(domain, dev);
1114 if (!ret)
1115 trace_attach_device_to_domain(dev);
1116 return ret;
fc2100eb 1117}
426a2738
JR
1118
1119int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1120{
1121 struct iommu_group *group;
1122 int ret;
1123
1124 group = iommu_group_get(dev);
1125 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1126 if (group == NULL)
1127 return __iommu_attach_device(domain, dev);
1128
1129 /*
1130 * We have a group - lock it to make sure the device-count doesn't
1131 * change while we are attaching
1132 */
1133 mutex_lock(&group->mutex);
1134 ret = -EINVAL;
1135 if (iommu_group_device_count(group) != 1)
1136 goto out_unlock;
1137
e39cb8a3 1138 ret = __iommu_attach_group(domain, group);
426a2738
JR
1139
1140out_unlock:
1141 mutex_unlock(&group->mutex);
1142 iommu_group_put(group);
1143
1144 return ret;
1145}
fc2100eb
JR
1146EXPORT_SYMBOL_GPL(iommu_attach_device);
1147
426a2738
JR
1148static void __iommu_detach_device(struct iommu_domain *domain,
1149 struct device *dev)
fc2100eb 1150{
e5aa7f00
JR
1151 if (unlikely(domain->ops->detach_dev == NULL))
1152 return;
1153
1154 domain->ops->detach_dev(domain, dev);
69980630 1155 trace_detach_device_from_domain(dev);
fc2100eb 1156}
426a2738
JR
1157
1158void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1159{
1160 struct iommu_group *group;
1161
1162 group = iommu_group_get(dev);
1163 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1164 if (group == NULL)
1165 return __iommu_detach_device(domain, dev);
1166
1167 mutex_lock(&group->mutex);
1168 if (iommu_group_device_count(group) != 1) {
1169 WARN_ON(1);
1170 goto out_unlock;
1171 }
1172
e39cb8a3 1173 __iommu_detach_group(domain, group);
426a2738
JR
1174
1175out_unlock:
1176 mutex_unlock(&group->mutex);
1177 iommu_group_put(group);
1178}
fc2100eb
JR
1179EXPORT_SYMBOL_GPL(iommu_detach_device);
1180
2c1296d9
JR
1181struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1182{
1183 struct iommu_domain *domain;
1184 struct iommu_group *group;
1185
1186 group = iommu_group_get(dev);
1187 /* FIXME: Remove this when groups a mandatory for iommu drivers */
1188 if (group == NULL)
1189 return NULL;
1190
1191 domain = group->domain;
1192
1193 iommu_group_put(group);
1194
1195 return domain;
1196}
1197EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
fc2100eb 1198
d72e31c9
AW
1199/*
1200 * IOMMU groups are really the natrual working unit of the IOMMU, but
1201 * the IOMMU API works on domains and devices. Bridge that gap by
1202 * iterating over the devices in a group. Ideally we'd have a single
1203 * device which represents the requestor ID of the group, but we also
1204 * allow IOMMU drivers to create policy defined minimum sets, where
1205 * the physical hardware may be able to distiguish members, but we
1206 * wish to group them at a higher level (ex. untrusted multi-function
1207 * PCI devices). Thus we attach each device.
1208 */
1209static int iommu_group_do_attach_device(struct device *dev, void *data)
1210{
1211 struct iommu_domain *domain = data;
1212
426a2738 1213 return __iommu_attach_device(domain, dev);
d72e31c9
AW
1214}
1215
e39cb8a3
JR
1216static int __iommu_attach_group(struct iommu_domain *domain,
1217 struct iommu_group *group)
1218{
1219 int ret;
1220
1221 if (group->default_domain && group->domain != group->default_domain)
1222 return -EBUSY;
1223
1224 ret = __iommu_group_for_each_dev(group, domain,
1225 iommu_group_do_attach_device);
1226 if (ret == 0)
1227 group->domain = domain;
1228
1229 return ret;
d72e31c9
AW
1230}
1231
1232int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1233{
e39cb8a3
JR
1234 int ret;
1235
1236 mutex_lock(&group->mutex);
1237 ret = __iommu_attach_group(domain, group);
1238 mutex_unlock(&group->mutex);
1239
1240 return ret;
d72e31c9
AW
1241}
1242EXPORT_SYMBOL_GPL(iommu_attach_group);
1243
1244static int iommu_group_do_detach_device(struct device *dev, void *data)
1245{
1246 struct iommu_domain *domain = data;
1247
426a2738 1248 __iommu_detach_device(domain, dev);
d72e31c9
AW
1249
1250 return 0;
1251}
1252
e39cb8a3
JR
1253static void __iommu_detach_group(struct iommu_domain *domain,
1254 struct iommu_group *group)
1255{
1256 int ret;
1257
1258 if (!group->default_domain) {
1259 __iommu_group_for_each_dev(group, domain,
1260 iommu_group_do_detach_device);
1261 group->domain = NULL;
1262 return;
1263 }
1264
1265 if (group->domain == group->default_domain)
1266 return;
1267
1268 /* Detach by re-attaching to the default domain */
1269 ret = __iommu_group_for_each_dev(group, group->default_domain,
1270 iommu_group_do_attach_device);
1271 if (ret != 0)
1272 WARN_ON(1);
1273 else
1274 group->domain = group->default_domain;
1275}
1276
d72e31c9
AW
1277void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1278{
e39cb8a3
JR
1279 mutex_lock(&group->mutex);
1280 __iommu_detach_group(domain, group);
1281 mutex_unlock(&group->mutex);
d72e31c9
AW
1282}
1283EXPORT_SYMBOL_GPL(iommu_detach_group);
1284
bb5547ac 1285phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
fc2100eb 1286{
e5aa7f00
JR
1287 if (unlikely(domain->ops->iova_to_phys == NULL))
1288 return 0;
1289
1290 return domain->ops->iova_to_phys(domain, iova);
fc2100eb
JR
1291}
1292EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
dbb9fd86 1293
bd13969b
AW
1294static size_t iommu_pgsize(struct iommu_domain *domain,
1295 unsigned long addr_merge, size_t size)
1296{
1297 unsigned int pgsize_idx;
1298 size_t pgsize;
1299
1300 /* Max page size that still fits into 'size' */
1301 pgsize_idx = __fls(size);
1302
1303 /* need to consider alignment requirements ? */
1304 if (likely(addr_merge)) {
1305 /* Max page size allowed by address */
1306 unsigned int align_pgsize_idx = __ffs(addr_merge);
1307 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1308 }
1309
1310 /* build a mask of acceptable page sizes */
1311 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1312
1313 /* throw away page sizes not supported by the hardware */
d16e0faa 1314 pgsize &= domain->pgsize_bitmap;
bd13969b
AW
1315
1316 /* make sure we're still sane */
1317 BUG_ON(!pgsize);
1318
1319 /* pick the biggest page */
1320 pgsize_idx = __fls(pgsize);
1321 pgsize = 1UL << pgsize_idx;
1322
1323 return pgsize;
1324}
1325
cefc53c7 1326int iommu_map(struct iommu_domain *domain, unsigned long iova,
7d3002cc 1327 phys_addr_t paddr, size_t size, int prot)
cefc53c7 1328{
7d3002cc
OBC
1329 unsigned long orig_iova = iova;
1330 unsigned int min_pagesz;
1331 size_t orig_size = size;
06bfcaa9 1332 phys_addr_t orig_paddr = paddr;
7d3002cc 1333 int ret = 0;
cefc53c7 1334
9db4ad91 1335 if (unlikely(domain->ops->map == NULL ||
d16e0faa 1336 domain->pgsize_bitmap == 0UL))
e5aa7f00 1337 return -ENODEV;
cefc53c7 1338
a10315e5
JR
1339 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1340 return -EINVAL;
1341
7d3002cc 1342 /* find out the minimum page size supported */
d16e0faa 1343 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
7d3002cc
OBC
1344
1345 /*
1346 * both the virtual address and the physical one, as well as
1347 * the size of the mapping, must be aligned (at least) to the
1348 * size of the smallest page supported by the hardware
1349 */
1350 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
abedb049 1351 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
6197ca82 1352 iova, &paddr, size, min_pagesz);
7d3002cc
OBC
1353 return -EINVAL;
1354 }
1355
abedb049 1356 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
7d3002cc
OBC
1357
1358 while (size) {
bd13969b 1359 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
7d3002cc 1360
abedb049 1361 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
6197ca82 1362 iova, &paddr, pgsize);
7d3002cc
OBC
1363
1364 ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1365 if (ret)
1366 break;
1367
1368 iova += pgsize;
1369 paddr += pgsize;
1370 size -= pgsize;
1371 }
1372
1373 /* unroll mapping in case something went wrong */
1374 if (ret)
1375 iommu_unmap(domain, orig_iova, orig_size - size);
e0be7c86 1376 else
06bfcaa9 1377 trace_map(orig_iova, orig_paddr, orig_size);
7d3002cc
OBC
1378
1379 return ret;
cefc53c7
JR
1380}
1381EXPORT_SYMBOL_GPL(iommu_map);
1382
7d3002cc 1383size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
cefc53c7 1384{
7d3002cc
OBC
1385 size_t unmapped_page, unmapped = 0;
1386 unsigned int min_pagesz;
6fd492fd 1387 unsigned long orig_iova = iova;
cefc53c7 1388
57886518 1389 if (unlikely(domain->ops->unmap == NULL ||
d16e0faa 1390 domain->pgsize_bitmap == 0UL))
e5aa7f00
JR
1391 return -ENODEV;
1392
a10315e5
JR
1393 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1394 return -EINVAL;
1395
7d3002cc 1396 /* find out the minimum page size supported */
d16e0faa 1397 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
7d3002cc
OBC
1398
1399 /*
1400 * The virtual address, as well as the size of the mapping, must be
1401 * aligned (at least) to the size of the smallest page supported
1402 * by the hardware
1403 */
1404 if (!IS_ALIGNED(iova | size, min_pagesz)) {
6197ca82
JP
1405 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1406 iova, size, min_pagesz);
7d3002cc
OBC
1407 return -EINVAL;
1408 }
1409
6197ca82 1410 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
7d3002cc
OBC
1411
1412 /*
1413 * Keep iterating until we either unmap 'size' bytes (or more)
1414 * or we hit an area that isn't mapped.
1415 */
1416 while (unmapped < size) {
bd13969b 1417 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
7d3002cc 1418
bd13969b 1419 unmapped_page = domain->ops->unmap(domain, iova, pgsize);
7d3002cc
OBC
1420 if (!unmapped_page)
1421 break;
1422
6197ca82
JP
1423 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1424 iova, unmapped_page);
7d3002cc
OBC
1425
1426 iova += unmapped_page;
1427 unmapped += unmapped_page;
1428 }
1429
db8614d3 1430 trace_unmap(orig_iova, size, unmapped);
7d3002cc 1431 return unmapped;
cefc53c7
JR
1432}
1433EXPORT_SYMBOL_GPL(iommu_unmap);
1460432c 1434
315786eb
OH
1435size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1436 struct scatterlist *sg, unsigned int nents, int prot)
1437{
38ec010d 1438 struct scatterlist *s;
315786eb 1439 size_t mapped = 0;
18f23409 1440 unsigned int i, min_pagesz;
38ec010d 1441 int ret;
315786eb 1442
d16e0faa 1443 if (unlikely(domain->pgsize_bitmap == 0UL))
18f23409 1444 return 0;
315786eb 1445
d16e0faa 1446 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
18f23409
RM
1447
1448 for_each_sg(sg, s, nents, i) {
3e6110fd 1449 phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
18f23409
RM
1450
1451 /*
1452 * We are mapping on IOMMU page boundaries, so offset within
1453 * the page must be 0. However, the IOMMU may support pages
1454 * smaller than PAGE_SIZE, so s->offset may still represent
1455 * an offset of that boundary within the CPU page.
1456 */
1457 if (!IS_ALIGNED(s->offset, min_pagesz))
38ec010d
JR
1458 goto out_err;
1459
1460 ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1461 if (ret)
1462 goto out_err;
1463
1464 mapped += s->length;
315786eb
OH
1465 }
1466
1467 return mapped;
38ec010d
JR
1468
1469out_err:
1470 /* undo mappings already done */
1471 iommu_unmap(domain, iova, mapped);
1472
1473 return 0;
1474
315786eb
OH
1475}
1476EXPORT_SYMBOL_GPL(default_iommu_map_sg);
d7787d57
JR
1477
1478int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
80f97f0f 1479 phys_addr_t paddr, u64 size, int prot)
d7787d57
JR
1480{
1481 if (unlikely(domain->ops->domain_window_enable == NULL))
1482 return -ENODEV;
1483
80f97f0f
VS
1484 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1485 prot);
d7787d57
JR
1486}
1487EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1488
1489void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1490{
1491 if (unlikely(domain->ops->domain_window_disable == NULL))
1492 return;
1493
1494 return domain->ops->domain_window_disable(domain, wnd_nr);
1495}
1496EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1497
d72e31c9 1498static int __init iommu_init(void)
1460432c 1499{
d72e31c9
AW
1500 iommu_group_kset = kset_create_and_add("iommu_groups",
1501 NULL, kernel_kobj);
d72e31c9
AW
1502 BUG_ON(!iommu_group_kset);
1503
1504 return 0;
1460432c 1505}
d7ef9995 1506core_initcall(iommu_init);
0cd76dd1
JR
1507
1508int iommu_domain_get_attr(struct iommu_domain *domain,
1509 enum iommu_attr attr, void *data)
1510{
0ff64f80 1511 struct iommu_domain_geometry *geometry;
d2e12160 1512 bool *paging;
0ff64f80 1513 int ret = 0;
69356712 1514 u32 *count;
0ff64f80
JR
1515
1516 switch (attr) {
1517 case DOMAIN_ATTR_GEOMETRY:
1518 geometry = data;
1519 *geometry = domain->geometry;
1520
d2e12160
JR
1521 break;
1522 case DOMAIN_ATTR_PAGING:
1523 paging = data;
d16e0faa 1524 *paging = (domain->pgsize_bitmap != 0UL);
69356712
JR
1525 break;
1526 case DOMAIN_ATTR_WINDOWS:
1527 count = data;
1528
1529 if (domain->ops->domain_get_windows != NULL)
1530 *count = domain->ops->domain_get_windows(domain);
1531 else
1532 ret = -ENODEV;
1533
0ff64f80
JR
1534 break;
1535 default:
1536 if (!domain->ops->domain_get_attr)
1537 return -EINVAL;
0cd76dd1 1538
0ff64f80
JR
1539 ret = domain->ops->domain_get_attr(domain, attr, data);
1540 }
1541
1542 return ret;
0cd76dd1
JR
1543}
1544EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1545
1546int iommu_domain_set_attr(struct iommu_domain *domain,
1547 enum iommu_attr attr, void *data)
1548{
69356712
JR
1549 int ret = 0;
1550 u32 *count;
1551
1552 switch (attr) {
1553 case DOMAIN_ATTR_WINDOWS:
1554 count = data;
1555
1556 if (domain->ops->domain_set_windows != NULL)
1557 ret = domain->ops->domain_set_windows(domain, *count);
1558 else
1559 ret = -ENODEV;
1460432c 1560
69356712
JR
1561 break;
1562 default:
1563 if (domain->ops->domain_set_attr == NULL)
1564 return -EINVAL;
1565
1566 ret = domain->ops->domain_set_attr(domain, attr, data);
1567 }
1568
1569 return ret;
1460432c 1570}
0cd76dd1 1571EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
a1015c2b
JR
1572
1573void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1574{
1575 const struct iommu_ops *ops = dev->bus->iommu_ops;
1576
1577 if (ops && ops->get_dm_regions)
1578 ops->get_dm_regions(dev, list);
1579}
1580
1581void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1582{
1583 const struct iommu_ops *ops = dev->bus->iommu_ops;
1584
1585 if (ops && ops->put_dm_regions)
1586 ops->put_dm_regions(dev, list);
1587}
d290f1e7
JR
1588
1589/* Request that a device is direct mapped by the IOMMU */
1590int iommu_request_dm_for_dev(struct device *dev)
1591{
1592 struct iommu_domain *dm_domain;
1593 struct iommu_group *group;
1594 int ret;
1595
1596 /* Device must already be in a group before calling this function */
1597 group = iommu_group_get_for_dev(dev);
409e553d
DC
1598 if (IS_ERR(group))
1599 return PTR_ERR(group);
d290f1e7
JR
1600
1601 mutex_lock(&group->mutex);
1602
1603 /* Check if the default domain is already direct mapped */
1604 ret = 0;
1605 if (group->default_domain &&
1606 group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1607 goto out;
1608
1609 /* Don't change mappings of existing devices */
1610 ret = -EBUSY;
1611 if (iommu_group_device_count(group) != 1)
1612 goto out;
1613
1614 /* Allocate a direct mapped domain */
1615 ret = -ENOMEM;
1616 dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1617 if (!dm_domain)
1618 goto out;
1619
1620 /* Attach the device to the domain */
1621 ret = __iommu_attach_group(dm_domain, group);
1622 if (ret) {
1623 iommu_domain_free(dm_domain);
1624 goto out;
1625 }
1626
1627 /* Make the direct mapped domain the default for this group */
1628 if (group->default_domain)
1629 iommu_domain_free(group->default_domain);
1630 group->default_domain = dm_domain;
1631
1632 pr_info("Using direct mapping for device %s\n", dev_name(dev));
1633
1634 ret = 0;
1635out:
1636 mutex_unlock(&group->mutex);
1637 iommu_group_put(group);
1638
1639 return ret;
1640}
57f98d2f 1641
e4f10ffe
LP
1642struct iommu_instance {
1643 struct list_head list;
1644 struct fwnode_handle *fwnode;
1645 const struct iommu_ops *ops;
1646};
1647static LIST_HEAD(iommu_instance_list);
1648static DEFINE_SPINLOCK(iommu_instance_lock);
1649
1650void iommu_register_instance(struct fwnode_handle *fwnode,
1651 const struct iommu_ops *ops)
1652{
1653 struct iommu_instance *iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1654
1655 if (WARN_ON(!iommu))
1656 return;
1657
1658 of_node_get(to_of_node(fwnode));
1659 INIT_LIST_HEAD(&iommu->list);
1660 iommu->fwnode = fwnode;
1661 iommu->ops = ops;
1662 spin_lock(&iommu_instance_lock);
1663 list_add_tail(&iommu->list, &iommu_instance_list);
1664 spin_unlock(&iommu_instance_lock);
1665}
1666
534766df 1667const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
e4f10ffe
LP
1668{
1669 struct iommu_instance *instance;
1670 const struct iommu_ops *ops = NULL;
1671
1672 spin_lock(&iommu_instance_lock);
1673 list_for_each_entry(instance, &iommu_instance_list, list)
1674 if (instance->fwnode == fwnode) {
1675 ops = instance->ops;
1676 break;
1677 }
1678 spin_unlock(&iommu_instance_lock);
1679 return ops;
1680}
1681
57f98d2f
RM
1682int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1683 const struct iommu_ops *ops)
1684{
1685 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1686
1687 if (fwspec)
1688 return ops == fwspec->ops ? 0 : -EINVAL;
1689
1690 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1691 if (!fwspec)
1692 return -ENOMEM;
1693
1694 of_node_get(to_of_node(iommu_fwnode));
1695 fwspec->iommu_fwnode = iommu_fwnode;
1696 fwspec->ops = ops;
1697 dev->iommu_fwspec = fwspec;
1698 return 0;
1699}
1700EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1701
1702void iommu_fwspec_free(struct device *dev)
1703{
1704 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1705
1706 if (fwspec) {
1707 fwnode_handle_put(fwspec->iommu_fwnode);
1708 kfree(fwspec);
1709 dev->iommu_fwspec = NULL;
1710 }
1711}
1712EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1713
1714int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1715{
1716 struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1717 size_t size;
1718 int i;
1719
1720 if (!fwspec)
1721 return -EINVAL;
1722
1723 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1724 if (size > sizeof(*fwspec)) {
1725 fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1726 if (!fwspec)
1727 return -ENOMEM;
1728 }
1729
1730 for (i = 0; i < num_ids; i++)
1731 fwspec->ids[fwspec->num_ids + i] = ids[i];
1732
1733 fwspec->num_ids += num_ids;
1734 dev->iommu_fwspec = fwspec;
1735 return 0;
1736}
1737EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);